1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <linux/kernel.h>
8 #include <linux/buffer_head.h>
9 #include <linux/file.h>
11 #include <linux/pagemap.h>
12 #include <linux/highmem.h>
13 #include <linux/time.h>
14 #include <linux/init.h>
15 #include <linux/string.h>
16 #include <linux/backing-dev.h>
17 #include <linux/writeback.h>
18 #include <linux/compat.h>
19 #include <linux/xattr.h>
20 #include <linux/posix_acl.h>
21 #include <linux/falloc.h>
22 #include <linux/slab.h>
23 #include <linux/ratelimit.h>
24 #include <linux/btrfs.h>
25 #include <linux/blkdev.h>
26 #include <linux/posix_acl_xattr.h>
27 #include <linux/uio.h>
28 #include <linux/magic.h>
29 #include <linux/iversion.h>
30 #include <asm/unaligned.h>
33 #include "transaction.h"
34 #include "btrfs_inode.h"
35 #include "print-tree.h"
36 #include "ordered-data.h"
40 #include "compression.h"
42 #include "free-space-cache.h"
43 #include "inode-map.h"
49 struct btrfs_iget_args {
50 struct btrfs_key *location;
51 struct btrfs_root *root;
54 struct btrfs_dio_data {
56 u64 unsubmitted_oe_range_start;
57 u64 unsubmitted_oe_range_end;
61 static const struct inode_operations btrfs_dir_inode_operations;
62 static const struct inode_operations btrfs_symlink_inode_operations;
63 static const struct inode_operations btrfs_dir_ro_inode_operations;
64 static const struct inode_operations btrfs_special_inode_operations;
65 static const struct inode_operations btrfs_file_inode_operations;
66 static const struct address_space_operations btrfs_aops;
67 static const struct file_operations btrfs_dir_file_operations;
68 static const struct extent_io_ops btrfs_extent_io_ops;
70 static struct kmem_cache *btrfs_inode_cachep;
71 struct kmem_cache *btrfs_trans_handle_cachep;
72 struct kmem_cache *btrfs_path_cachep;
73 struct kmem_cache *btrfs_free_space_cachep;
76 static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
77 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
78 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
79 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
80 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
81 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
82 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
83 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
86 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
87 static int btrfs_truncate(struct inode *inode, bool skip_writeback);
88 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
89 static noinline int cow_file_range(struct inode *inode,
90 struct page *locked_page,
91 u64 start, u64 end, u64 delalloc_end,
92 int *page_started, unsigned long *nr_written,
93 int unlock, struct btrfs_dedupe_hash *hash);
94 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
95 u64 orig_start, u64 block_start,
96 u64 block_len, u64 orig_block_len,
97 u64 ram_bytes, int compress_type,
100 static void __endio_write_update_ordered(struct inode *inode,
101 const u64 offset, const u64 bytes,
102 const bool uptodate);
105 * Cleanup all submitted ordered extents in specified range to handle errors
106 * from the fill_dellaloc() callback.
108 * NOTE: caller must ensure that when an error happens, it can not call
109 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
110 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
111 * to be released, which we want to happen only when finishing the ordered
112 * extent (btrfs_finish_ordered_io()). Also note that the caller of the
113 * fill_delalloc() callback already does proper cleanup for the first page of
114 * the range, that is, it invokes the callback writepage_end_io_hook() for the
115 * range of the first page.
117 static inline void btrfs_cleanup_ordered_extents(struct inode *inode,
121 unsigned long index = offset >> PAGE_SHIFT;
122 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
125 while (index <= end_index) {
126 page = find_get_page(inode->i_mapping, index);
130 ClearPagePrivate2(page);
133 return __endio_write_update_ordered(inode, offset + PAGE_SIZE,
134 bytes - PAGE_SIZE, false);
137 static int btrfs_dirty_inode(struct inode *inode);
139 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
140 void btrfs_test_inode_set_ops(struct inode *inode)
142 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
146 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
147 struct inode *inode, struct inode *dir,
148 const struct qstr *qstr)
152 err = btrfs_init_acl(trans, inode, dir);
154 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
159 * this does all the hard work for inserting an inline extent into
160 * the btree. The caller should have done a btrfs_drop_extents so that
161 * no overlapping inline items exist in the btree
163 static int insert_inline_extent(struct btrfs_trans_handle *trans,
164 struct btrfs_path *path, int extent_inserted,
165 struct btrfs_root *root, struct inode *inode,
166 u64 start, size_t size, size_t compressed_size,
168 struct page **compressed_pages)
170 struct extent_buffer *leaf;
171 struct page *page = NULL;
174 struct btrfs_file_extent_item *ei;
176 size_t cur_size = size;
177 unsigned long offset;
179 if (compressed_size && compressed_pages)
180 cur_size = compressed_size;
182 inode_add_bytes(inode, size);
184 if (!extent_inserted) {
185 struct btrfs_key key;
188 key.objectid = btrfs_ino(BTRFS_I(inode));
190 key.type = BTRFS_EXTENT_DATA_KEY;
192 datasize = btrfs_file_extent_calc_inline_size(cur_size);
193 path->leave_spinning = 1;
194 ret = btrfs_insert_empty_item(trans, root, path, &key,
199 leaf = path->nodes[0];
200 ei = btrfs_item_ptr(leaf, path->slots[0],
201 struct btrfs_file_extent_item);
202 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
203 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
204 btrfs_set_file_extent_encryption(leaf, ei, 0);
205 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
206 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
207 ptr = btrfs_file_extent_inline_start(ei);
209 if (compress_type != BTRFS_COMPRESS_NONE) {
212 while (compressed_size > 0) {
213 cpage = compressed_pages[i];
214 cur_size = min_t(unsigned long, compressed_size,
217 kaddr = kmap_atomic(cpage);
218 write_extent_buffer(leaf, kaddr, ptr, cur_size);
219 kunmap_atomic(kaddr);
223 compressed_size -= cur_size;
225 btrfs_set_file_extent_compression(leaf, ei,
228 page = find_get_page(inode->i_mapping,
229 start >> PAGE_SHIFT);
230 btrfs_set_file_extent_compression(leaf, ei, 0);
231 kaddr = kmap_atomic(page);
232 offset = start & (PAGE_SIZE - 1);
233 write_extent_buffer(leaf, kaddr + offset, ptr, size);
234 kunmap_atomic(kaddr);
237 btrfs_mark_buffer_dirty(leaf);
238 btrfs_release_path(path);
241 * we're an inline extent, so nobody can
242 * extend the file past i_size without locking
243 * a page we already have locked.
245 * We must do any isize and inode updates
246 * before we unlock the pages. Otherwise we
247 * could end up racing with unlink.
249 BTRFS_I(inode)->disk_i_size = inode->i_size;
250 ret = btrfs_update_inode(trans, root, inode);
258 * conditionally insert an inline extent into the file. This
259 * does the checks required to make sure the data is small enough
260 * to fit as an inline extent.
262 static noinline int cow_file_range_inline(struct inode *inode, u64 start,
263 u64 end, size_t compressed_size,
265 struct page **compressed_pages)
267 struct btrfs_root *root = BTRFS_I(inode)->root;
268 struct btrfs_fs_info *fs_info = root->fs_info;
269 struct btrfs_trans_handle *trans;
270 u64 isize = i_size_read(inode);
271 u64 actual_end = min(end + 1, isize);
272 u64 inline_len = actual_end - start;
273 u64 aligned_end = ALIGN(end, fs_info->sectorsize);
274 u64 data_len = inline_len;
276 struct btrfs_path *path;
277 int extent_inserted = 0;
278 u32 extent_item_size;
281 data_len = compressed_size;
284 actual_end > fs_info->sectorsize ||
285 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
287 (actual_end & (fs_info->sectorsize - 1)) == 0) ||
289 data_len > fs_info->max_inline) {
293 path = btrfs_alloc_path();
297 trans = btrfs_join_transaction(root);
299 btrfs_free_path(path);
300 return PTR_ERR(trans);
302 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
304 if (compressed_size && compressed_pages)
305 extent_item_size = btrfs_file_extent_calc_inline_size(
308 extent_item_size = btrfs_file_extent_calc_inline_size(
311 ret = __btrfs_drop_extents(trans, root, inode, path,
312 start, aligned_end, NULL,
313 1, 1, extent_item_size, &extent_inserted);
315 btrfs_abort_transaction(trans, ret);
319 if (isize > actual_end)
320 inline_len = min_t(u64, isize, actual_end);
321 ret = insert_inline_extent(trans, path, extent_inserted,
323 inline_len, compressed_size,
324 compress_type, compressed_pages);
325 if (ret && ret != -ENOSPC) {
326 btrfs_abort_transaction(trans, ret);
328 } else if (ret == -ENOSPC) {
333 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
334 btrfs_drop_extent_cache(BTRFS_I(inode), start, aligned_end - 1, 0);
337 * Don't forget to free the reserved space, as for inlined extent
338 * it won't count as data extent, free them directly here.
339 * And at reserve time, it's always aligned to page size, so
340 * just free one page here.
342 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
343 btrfs_free_path(path);
344 btrfs_end_transaction(trans);
348 struct async_extent {
353 unsigned long nr_pages;
355 struct list_head list;
360 struct btrfs_root *root;
361 struct page *locked_page;
364 unsigned int write_flags;
365 struct list_head extents;
366 struct btrfs_work work;
369 static noinline int add_async_extent(struct async_cow *cow,
370 u64 start, u64 ram_size,
373 unsigned long nr_pages,
376 struct async_extent *async_extent;
378 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
379 BUG_ON(!async_extent); /* -ENOMEM */
380 async_extent->start = start;
381 async_extent->ram_size = ram_size;
382 async_extent->compressed_size = compressed_size;
383 async_extent->pages = pages;
384 async_extent->nr_pages = nr_pages;
385 async_extent->compress_type = compress_type;
386 list_add_tail(&async_extent->list, &cow->extents);
390 static inline int inode_need_compress(struct inode *inode, u64 start, u64 end)
392 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
395 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
398 if (BTRFS_I(inode)->defrag_compress)
400 /* bad compression ratios */
401 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
403 if (btrfs_test_opt(fs_info, COMPRESS) ||
404 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
405 BTRFS_I(inode)->prop_compress)
406 return btrfs_compress_heuristic(inode, start, end);
410 static inline void inode_should_defrag(struct btrfs_inode *inode,
411 u64 start, u64 end, u64 num_bytes, u64 small_write)
413 /* If this is a small write inside eof, kick off a defrag */
414 if (num_bytes < small_write &&
415 (start > 0 || end + 1 < inode->disk_i_size))
416 btrfs_add_inode_defrag(NULL, inode);
420 * we create compressed extents in two phases. The first
421 * phase compresses a range of pages that have already been
422 * locked (both pages and state bits are locked).
424 * This is done inside an ordered work queue, and the compression
425 * is spread across many cpus. The actual IO submission is step
426 * two, and the ordered work queue takes care of making sure that
427 * happens in the same order things were put onto the queue by
428 * writepages and friends.
430 * If this code finds it can't get good compression, it puts an
431 * entry onto the work queue to write the uncompressed bytes. This
432 * makes sure that both compressed inodes and uncompressed inodes
433 * are written in the same order that the flusher thread sent them
436 static noinline void compress_file_range(struct inode *inode,
437 struct page *locked_page,
439 struct async_cow *async_cow,
442 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
443 u64 blocksize = fs_info->sectorsize;
445 u64 isize = i_size_read(inode);
447 struct page **pages = NULL;
448 unsigned long nr_pages;
449 unsigned long total_compressed = 0;
450 unsigned long total_in = 0;
453 int compress_type = fs_info->compress_type;
456 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
459 actual_end = min_t(u64, isize, end + 1);
462 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
463 BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
464 nr_pages = min_t(unsigned long, nr_pages,
465 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
468 * we don't want to send crud past the end of i_size through
469 * compression, that's just a waste of CPU time. So, if the
470 * end of the file is before the start of our current
471 * requested range of bytes, we bail out to the uncompressed
472 * cleanup code that can deal with all of this.
474 * It isn't really the fastest way to fix things, but this is a
475 * very uncommon corner.
477 if (actual_end <= start)
478 goto cleanup_and_bail_uncompressed;
480 total_compressed = actual_end - start;
483 * skip compression for a small file range(<=blocksize) that
484 * isn't an inline extent, since it doesn't save disk space at all.
486 if (total_compressed <= blocksize &&
487 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
488 goto cleanup_and_bail_uncompressed;
490 total_compressed = min_t(unsigned long, total_compressed,
491 BTRFS_MAX_UNCOMPRESSED);
496 * we do compression for mount -o compress and when the
497 * inode has not been flagged as nocompress. This flag can
498 * change at any time if we discover bad compression ratios.
500 if (inode_need_compress(inode, start, end)) {
502 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
504 /* just bail out to the uncompressed code */
509 if (BTRFS_I(inode)->defrag_compress)
510 compress_type = BTRFS_I(inode)->defrag_compress;
511 else if (BTRFS_I(inode)->prop_compress)
512 compress_type = BTRFS_I(inode)->prop_compress;
515 * we need to call clear_page_dirty_for_io on each
516 * page in the range. Otherwise applications with the file
517 * mmap'd can wander in and change the page contents while
518 * we are compressing them.
520 * If the compression fails for any reason, we set the pages
521 * dirty again later on.
523 * Note that the remaining part is redirtied, the start pointer
524 * has moved, the end is the original one.
527 extent_range_clear_dirty_for_io(inode, start, end);
531 /* Compression level is applied here and only here */
532 ret = btrfs_compress_pages(
533 compress_type | (fs_info->compress_level << 4),
534 inode->i_mapping, start,
541 unsigned long offset = total_compressed &
543 struct page *page = pages[nr_pages - 1];
546 /* zero the tail end of the last page, we might be
547 * sending it down to disk
550 kaddr = kmap_atomic(page);
551 memset(kaddr + offset, 0,
553 kunmap_atomic(kaddr);
560 /* lets try to make an inline extent */
561 if (ret || total_in < actual_end) {
562 /* we didn't compress the entire range, try
563 * to make an uncompressed inline extent.
565 ret = cow_file_range_inline(inode, start, end, 0,
566 BTRFS_COMPRESS_NONE, NULL);
568 /* try making a compressed inline extent */
569 ret = cow_file_range_inline(inode, start, end,
571 compress_type, pages);
574 unsigned long clear_flags = EXTENT_DELALLOC |
575 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
576 EXTENT_DO_ACCOUNTING;
577 unsigned long page_error_op;
579 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
582 * inline extent creation worked or returned error,
583 * we don't need to create any more async work items.
584 * Unlock and free up our temp pages.
586 * We use DO_ACCOUNTING here because we need the
587 * delalloc_release_metadata to be done _after_ we drop
588 * our outstanding extent for clearing delalloc for this
591 extent_clear_unlock_delalloc(inode, start, end, end,
604 * we aren't doing an inline extent round the compressed size
605 * up to a block size boundary so the allocator does sane
608 total_compressed = ALIGN(total_compressed, blocksize);
611 * one last check to make sure the compression is really a
612 * win, compare the page count read with the blocks on disk,
613 * compression must free at least one sector size
615 total_in = ALIGN(total_in, PAGE_SIZE);
616 if (total_compressed + blocksize <= total_in) {
620 * The async work queues will take care of doing actual
621 * allocation on disk for these compressed pages, and
622 * will submit them to the elevator.
624 add_async_extent(async_cow, start, total_in,
625 total_compressed, pages, nr_pages,
628 if (start + total_in < end) {
639 * the compression code ran but failed to make things smaller,
640 * free any pages it allocated and our page pointer array
642 for (i = 0; i < nr_pages; i++) {
643 WARN_ON(pages[i]->mapping);
648 total_compressed = 0;
651 /* flag the file so we don't compress in the future */
652 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
653 !(BTRFS_I(inode)->prop_compress)) {
654 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
657 cleanup_and_bail_uncompressed:
659 * No compression, but we still need to write the pages in the file
660 * we've been given so far. redirty the locked page if it corresponds
661 * to our extent and set things up for the async work queue to run
662 * cow_file_range to do the normal delalloc dance.
664 if (page_offset(locked_page) >= start &&
665 page_offset(locked_page) <= end)
666 __set_page_dirty_nobuffers(locked_page);
667 /* unlocked later on in the async handlers */
670 extent_range_redirty_for_io(inode, start, end);
671 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0,
672 BTRFS_COMPRESS_NONE);
678 for (i = 0; i < nr_pages; i++) {
679 WARN_ON(pages[i]->mapping);
685 static void free_async_extent_pages(struct async_extent *async_extent)
689 if (!async_extent->pages)
692 for (i = 0; i < async_extent->nr_pages; i++) {
693 WARN_ON(async_extent->pages[i]->mapping);
694 put_page(async_extent->pages[i]);
696 kfree(async_extent->pages);
697 async_extent->nr_pages = 0;
698 async_extent->pages = NULL;
702 * phase two of compressed writeback. This is the ordered portion
703 * of the code, which only gets called in the order the work was
704 * queued. We walk all the async extents created by compress_file_range
705 * and send them down to the disk.
707 static noinline void submit_compressed_extents(struct inode *inode,
708 struct async_cow *async_cow)
710 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
711 struct async_extent *async_extent;
713 struct btrfs_key ins;
714 struct extent_map *em;
715 struct btrfs_root *root = BTRFS_I(inode)->root;
716 struct extent_io_tree *io_tree;
720 while (!list_empty(&async_cow->extents)) {
721 async_extent = list_entry(async_cow->extents.next,
722 struct async_extent, list);
723 list_del(&async_extent->list);
725 io_tree = &BTRFS_I(inode)->io_tree;
728 /* did the compression code fall back to uncompressed IO? */
729 if (!async_extent->pages) {
730 int page_started = 0;
731 unsigned long nr_written = 0;
733 lock_extent(io_tree, async_extent->start,
734 async_extent->start +
735 async_extent->ram_size - 1);
737 /* allocate blocks */
738 ret = cow_file_range(inode, async_cow->locked_page,
740 async_extent->start +
741 async_extent->ram_size - 1,
742 async_extent->start +
743 async_extent->ram_size - 1,
744 &page_started, &nr_written, 0,
750 * if page_started, cow_file_range inserted an
751 * inline extent and took care of all the unlocking
752 * and IO for us. Otherwise, we need to submit
753 * all those pages down to the drive.
755 if (!page_started && !ret)
756 extent_write_locked_range(inode,
758 async_extent->start +
759 async_extent->ram_size - 1,
762 unlock_page(async_cow->locked_page);
768 lock_extent(io_tree, async_extent->start,
769 async_extent->start + async_extent->ram_size - 1);
771 ret = btrfs_reserve_extent(root, async_extent->ram_size,
772 async_extent->compressed_size,
773 async_extent->compressed_size,
774 0, alloc_hint, &ins, 1, 1);
776 free_async_extent_pages(async_extent);
778 if (ret == -ENOSPC) {
779 unlock_extent(io_tree, async_extent->start,
780 async_extent->start +
781 async_extent->ram_size - 1);
784 * we need to redirty the pages if we decide to
785 * fallback to uncompressed IO, otherwise we
786 * will not submit these pages down to lower
789 extent_range_redirty_for_io(inode,
791 async_extent->start +
792 async_extent->ram_size - 1);
799 * here we're doing allocation and writeback of the
802 em = create_io_em(inode, async_extent->start,
803 async_extent->ram_size, /* len */
804 async_extent->start, /* orig_start */
805 ins.objectid, /* block_start */
806 ins.offset, /* block_len */
807 ins.offset, /* orig_block_len */
808 async_extent->ram_size, /* ram_bytes */
809 async_extent->compress_type,
810 BTRFS_ORDERED_COMPRESSED);
812 /* ret value is not necessary due to void function */
813 goto out_free_reserve;
816 ret = btrfs_add_ordered_extent_compress(inode,
819 async_extent->ram_size,
821 BTRFS_ORDERED_COMPRESSED,
822 async_extent->compress_type);
824 btrfs_drop_extent_cache(BTRFS_I(inode),
826 async_extent->start +
827 async_extent->ram_size - 1, 0);
828 goto out_free_reserve;
830 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
833 * clear dirty, set writeback and unlock the pages.
835 extent_clear_unlock_delalloc(inode, async_extent->start,
836 async_extent->start +
837 async_extent->ram_size - 1,
838 async_extent->start +
839 async_extent->ram_size - 1,
840 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
841 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
843 if (btrfs_submit_compressed_write(inode,
845 async_extent->ram_size,
847 ins.offset, async_extent->pages,
848 async_extent->nr_pages,
849 async_cow->write_flags)) {
850 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
851 struct page *p = async_extent->pages[0];
852 const u64 start = async_extent->start;
853 const u64 end = start + async_extent->ram_size - 1;
855 p->mapping = inode->i_mapping;
856 tree->ops->writepage_end_io_hook(p, start, end,
859 extent_clear_unlock_delalloc(inode, start, end, end,
863 free_async_extent_pages(async_extent);
865 alloc_hint = ins.objectid + ins.offset;
871 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
872 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
874 extent_clear_unlock_delalloc(inode, async_extent->start,
875 async_extent->start +
876 async_extent->ram_size - 1,
877 async_extent->start +
878 async_extent->ram_size - 1,
879 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
880 EXTENT_DELALLOC_NEW |
881 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
882 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
883 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
885 free_async_extent_pages(async_extent);
890 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
893 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
894 struct extent_map *em;
897 read_lock(&em_tree->lock);
898 em = search_extent_mapping(em_tree, start, num_bytes);
901 * if block start isn't an actual block number then find the
902 * first block in this inode and use that as a hint. If that
903 * block is also bogus then just don't worry about it.
905 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
907 em = search_extent_mapping(em_tree, 0, 0);
908 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
909 alloc_hint = em->block_start;
913 alloc_hint = em->block_start;
917 read_unlock(&em_tree->lock);
923 * when extent_io.c finds a delayed allocation range in the file,
924 * the call backs end up in this code. The basic idea is to
925 * allocate extents on disk for the range, and create ordered data structs
926 * in ram to track those extents.
928 * locked_page is the page that writepage had locked already. We use
929 * it to make sure we don't do extra locks or unlocks.
931 * *page_started is set to one if we unlock locked_page and do everything
932 * required to start IO on it. It may be clean and already done with
935 static noinline int cow_file_range(struct inode *inode,
936 struct page *locked_page,
937 u64 start, u64 end, u64 delalloc_end,
938 int *page_started, unsigned long *nr_written,
939 int unlock, struct btrfs_dedupe_hash *hash)
941 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
942 struct btrfs_root *root = BTRFS_I(inode)->root;
945 unsigned long ram_size;
946 u64 cur_alloc_size = 0;
947 u64 blocksize = fs_info->sectorsize;
948 struct btrfs_key ins;
949 struct extent_map *em;
951 unsigned long page_ops;
952 bool extent_reserved = false;
955 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
961 num_bytes = ALIGN(end - start + 1, blocksize);
962 num_bytes = max(blocksize, num_bytes);
963 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
965 inode_should_defrag(BTRFS_I(inode), start, end, num_bytes, SZ_64K);
968 /* lets try to make an inline extent */
969 ret = cow_file_range_inline(inode, start, end, 0,
970 BTRFS_COMPRESS_NONE, NULL);
973 * We use DO_ACCOUNTING here because we need the
974 * delalloc_release_metadata to be run _after_ we drop
975 * our outstanding extent for clearing delalloc for this
978 extent_clear_unlock_delalloc(inode, start, end,
980 EXTENT_LOCKED | EXTENT_DELALLOC |
981 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
982 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
983 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
985 *nr_written = *nr_written +
986 (end - start + PAGE_SIZE) / PAGE_SIZE;
989 } else if (ret < 0) {
994 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
995 btrfs_drop_extent_cache(BTRFS_I(inode), start,
996 start + num_bytes - 1, 0);
998 while (num_bytes > 0) {
999 cur_alloc_size = num_bytes;
1000 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1001 fs_info->sectorsize, 0, alloc_hint,
1005 cur_alloc_size = ins.offset;
1006 extent_reserved = true;
1008 ram_size = ins.offset;
1009 em = create_io_em(inode, start, ins.offset, /* len */
1010 start, /* orig_start */
1011 ins.objectid, /* block_start */
1012 ins.offset, /* block_len */
1013 ins.offset, /* orig_block_len */
1014 ram_size, /* ram_bytes */
1015 BTRFS_COMPRESS_NONE, /* compress_type */
1016 BTRFS_ORDERED_REGULAR /* type */);
1021 free_extent_map(em);
1023 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1024 ram_size, cur_alloc_size, 0);
1026 goto out_drop_extent_cache;
1028 if (root->root_key.objectid ==
1029 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1030 ret = btrfs_reloc_clone_csums(inode, start,
1033 * Only drop cache here, and process as normal.
1035 * We must not allow extent_clear_unlock_delalloc()
1036 * at out_unlock label to free meta of this ordered
1037 * extent, as its meta should be freed by
1038 * btrfs_finish_ordered_io().
1040 * So we must continue until @start is increased to
1041 * skip current ordered extent.
1044 btrfs_drop_extent_cache(BTRFS_I(inode), start,
1045 start + ram_size - 1, 0);
1048 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1050 /* we're not doing compressed IO, don't unlock the first
1051 * page (which the caller expects to stay locked), don't
1052 * clear any dirty bits and don't set any writeback bits
1054 * Do set the Private2 bit so we know this page was properly
1055 * setup for writepage
1057 page_ops = unlock ? PAGE_UNLOCK : 0;
1058 page_ops |= PAGE_SET_PRIVATE2;
1060 extent_clear_unlock_delalloc(inode, start,
1061 start + ram_size - 1,
1062 delalloc_end, locked_page,
1063 EXTENT_LOCKED | EXTENT_DELALLOC,
1065 if (num_bytes < cur_alloc_size)
1068 num_bytes -= cur_alloc_size;
1069 alloc_hint = ins.objectid + ins.offset;
1070 start += cur_alloc_size;
1071 extent_reserved = false;
1074 * btrfs_reloc_clone_csums() error, since start is increased
1075 * extent_clear_unlock_delalloc() at out_unlock label won't
1076 * free metadata of current ordered extent, we're OK to exit.
1084 out_drop_extent_cache:
1085 btrfs_drop_extent_cache(BTRFS_I(inode), start, start + ram_size - 1, 0);
1087 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1088 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1090 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1091 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1092 page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1095 * If we reserved an extent for our delalloc range (or a subrange) and
1096 * failed to create the respective ordered extent, then it means that
1097 * when we reserved the extent we decremented the extent's size from
1098 * the data space_info's bytes_may_use counter and incremented the
1099 * space_info's bytes_reserved counter by the same amount. We must make
1100 * sure extent_clear_unlock_delalloc() does not try to decrement again
1101 * the data space_info's bytes_may_use counter, therefore we do not pass
1102 * it the flag EXTENT_CLEAR_DATA_RESV.
1104 if (extent_reserved) {
1105 extent_clear_unlock_delalloc(inode, start,
1106 start + cur_alloc_size,
1107 start + cur_alloc_size,
1111 start += cur_alloc_size;
1115 extent_clear_unlock_delalloc(inode, start, end, delalloc_end,
1117 clear_bits | EXTENT_CLEAR_DATA_RESV,
1123 * work queue call back to started compression on a file and pages
1125 static noinline void async_cow_start(struct btrfs_work *work)
1127 struct async_cow *async_cow;
1129 async_cow = container_of(work, struct async_cow, work);
1131 compress_file_range(async_cow->inode, async_cow->locked_page,
1132 async_cow->start, async_cow->end, async_cow,
1134 if (num_added == 0) {
1135 btrfs_add_delayed_iput(async_cow->inode);
1136 async_cow->inode = NULL;
1141 * work queue call back to submit previously compressed pages
1143 static noinline void async_cow_submit(struct btrfs_work *work)
1145 struct btrfs_fs_info *fs_info;
1146 struct async_cow *async_cow;
1147 struct btrfs_root *root;
1148 unsigned long nr_pages;
1150 async_cow = container_of(work, struct async_cow, work);
1152 root = async_cow->root;
1153 fs_info = root->fs_info;
1154 nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1157 /* atomic_sub_return implies a barrier */
1158 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1160 cond_wake_up_nomb(&fs_info->async_submit_wait);
1162 if (async_cow->inode)
1163 submit_compressed_extents(async_cow->inode, async_cow);
1166 static noinline void async_cow_free(struct btrfs_work *work)
1168 struct async_cow *async_cow;
1169 async_cow = container_of(work, struct async_cow, work);
1170 if (async_cow->inode)
1171 btrfs_add_delayed_iput(async_cow->inode);
1175 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1176 u64 start, u64 end, int *page_started,
1177 unsigned long *nr_written,
1178 unsigned int write_flags)
1180 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1181 struct async_cow *async_cow;
1182 struct btrfs_root *root = BTRFS_I(inode)->root;
1183 unsigned long nr_pages;
1186 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1188 while (start < end) {
1189 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1190 BUG_ON(!async_cow); /* -ENOMEM */
1191 async_cow->inode = igrab(inode);
1192 async_cow->root = root;
1193 async_cow->locked_page = locked_page;
1194 async_cow->start = start;
1195 async_cow->write_flags = write_flags;
1197 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1198 !btrfs_test_opt(fs_info, FORCE_COMPRESS))
1201 cur_end = min(end, start + SZ_512K - 1);
1203 async_cow->end = cur_end;
1204 INIT_LIST_HEAD(&async_cow->extents);
1206 btrfs_init_work(&async_cow->work,
1207 btrfs_delalloc_helper,
1208 async_cow_start, async_cow_submit,
1211 nr_pages = (cur_end - start + PAGE_SIZE) >>
1213 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1215 btrfs_queue_work(fs_info->delalloc_workers, &async_cow->work);
1217 *nr_written += nr_pages;
1218 start = cur_end + 1;
1224 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1225 u64 bytenr, u64 num_bytes)
1228 struct btrfs_ordered_sum *sums;
1231 ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1232 bytenr + num_bytes - 1, &list, 0);
1233 if (ret == 0 && list_empty(&list))
1236 while (!list_empty(&list)) {
1237 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1238 list_del(&sums->list);
1247 * when nowcow writeback call back. This checks for snapshots or COW copies
1248 * of the extents that exist in the file, and COWs the file as required.
1250 * If no cow copies or snapshots exist, we write directly to the existing
1253 static noinline int run_delalloc_nocow(struct inode *inode,
1254 struct page *locked_page,
1255 u64 start, u64 end, int *page_started, int force,
1256 unsigned long *nr_written)
1258 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1259 struct btrfs_root *root = BTRFS_I(inode)->root;
1260 struct extent_buffer *leaf;
1261 struct btrfs_path *path;
1262 struct btrfs_file_extent_item *fi;
1263 struct btrfs_key found_key;
1264 struct extent_map *em;
1279 u64 ino = btrfs_ino(BTRFS_I(inode));
1281 path = btrfs_alloc_path();
1283 extent_clear_unlock_delalloc(inode, start, end, end,
1285 EXTENT_LOCKED | EXTENT_DELALLOC |
1286 EXTENT_DO_ACCOUNTING |
1287 EXTENT_DEFRAG, PAGE_UNLOCK |
1289 PAGE_SET_WRITEBACK |
1290 PAGE_END_WRITEBACK);
1294 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
1296 cow_start = (u64)-1;
1299 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1303 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1304 leaf = path->nodes[0];
1305 btrfs_item_key_to_cpu(leaf, &found_key,
1306 path->slots[0] - 1);
1307 if (found_key.objectid == ino &&
1308 found_key.type == BTRFS_EXTENT_DATA_KEY)
1313 leaf = path->nodes[0];
1314 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1315 ret = btrfs_next_leaf(root, path);
1317 if (cow_start != (u64)-1)
1318 cur_offset = cow_start;
1323 leaf = path->nodes[0];
1329 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1331 if (found_key.objectid > ino)
1333 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1334 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1338 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1339 found_key.offset > end)
1342 if (found_key.offset > cur_offset) {
1343 extent_end = found_key.offset;
1348 fi = btrfs_item_ptr(leaf, path->slots[0],
1349 struct btrfs_file_extent_item);
1350 extent_type = btrfs_file_extent_type(leaf, fi);
1352 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1353 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1354 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1355 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1356 extent_offset = btrfs_file_extent_offset(leaf, fi);
1357 extent_end = found_key.offset +
1358 btrfs_file_extent_num_bytes(leaf, fi);
1360 btrfs_file_extent_disk_num_bytes(leaf, fi);
1361 if (extent_end <= start) {
1365 if (disk_bytenr == 0)
1367 if (btrfs_file_extent_compression(leaf, fi) ||
1368 btrfs_file_extent_encryption(leaf, fi) ||
1369 btrfs_file_extent_other_encoding(leaf, fi))
1372 * Do the same check as in btrfs_cross_ref_exist but
1373 * without the unnecessary search.
1375 if (btrfs_file_extent_generation(leaf, fi) <=
1376 btrfs_root_last_snapshot(&root->root_item))
1378 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1380 if (btrfs_extent_readonly(fs_info, disk_bytenr))
1382 ret = btrfs_cross_ref_exist(root, ino,
1384 extent_offset, disk_bytenr);
1387 * ret could be -EIO if the above fails to read
1391 if (cow_start != (u64)-1)
1392 cur_offset = cow_start;
1396 WARN_ON_ONCE(nolock);
1399 disk_bytenr += extent_offset;
1400 disk_bytenr += cur_offset - found_key.offset;
1401 num_bytes = min(end + 1, extent_end) - cur_offset;
1403 * if there are pending snapshots for this root,
1404 * we fall into common COW way.
1406 if (!nolock && atomic_read(&root->snapshot_force_cow))
1409 * force cow if csum exists in the range.
1410 * this ensure that csum for a given extent are
1411 * either valid or do not exist.
1413 ret = csum_exist_in_range(fs_info, disk_bytenr,
1417 * ret could be -EIO if the above fails to read
1421 if (cow_start != (u64)-1)
1422 cur_offset = cow_start;
1425 WARN_ON_ONCE(nolock);
1428 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr))
1431 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1432 extent_end = found_key.offset +
1433 btrfs_file_extent_ram_bytes(leaf, fi);
1434 extent_end = ALIGN(extent_end,
1435 fs_info->sectorsize);
1440 if (extent_end <= start) {
1443 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1447 if (cow_start == (u64)-1)
1448 cow_start = cur_offset;
1449 cur_offset = extent_end;
1450 if (cur_offset > end)
1456 btrfs_release_path(path);
1457 if (cow_start != (u64)-1) {
1458 ret = cow_file_range(inode, locked_page,
1459 cow_start, found_key.offset - 1,
1460 end, page_started, nr_written, 1,
1464 btrfs_dec_nocow_writers(fs_info,
1468 cow_start = (u64)-1;
1471 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1472 u64 orig_start = found_key.offset - extent_offset;
1474 em = create_io_em(inode, cur_offset, num_bytes,
1476 disk_bytenr, /* block_start */
1477 num_bytes, /* block_len */
1478 disk_num_bytes, /* orig_block_len */
1479 ram_bytes, BTRFS_COMPRESS_NONE,
1480 BTRFS_ORDERED_PREALLOC);
1483 btrfs_dec_nocow_writers(fs_info,
1488 free_extent_map(em);
1491 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1492 type = BTRFS_ORDERED_PREALLOC;
1494 type = BTRFS_ORDERED_NOCOW;
1497 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1498 num_bytes, num_bytes, type);
1500 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1501 BUG_ON(ret); /* -ENOMEM */
1503 if (root->root_key.objectid ==
1504 BTRFS_DATA_RELOC_TREE_OBJECTID)
1506 * Error handled later, as we must prevent
1507 * extent_clear_unlock_delalloc() in error handler
1508 * from freeing metadata of created ordered extent.
1510 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1513 extent_clear_unlock_delalloc(inode, cur_offset,
1514 cur_offset + num_bytes - 1, end,
1515 locked_page, EXTENT_LOCKED |
1517 EXTENT_CLEAR_DATA_RESV,
1518 PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1520 cur_offset = extent_end;
1523 * btrfs_reloc_clone_csums() error, now we're OK to call error
1524 * handler, as metadata for created ordered extent will only
1525 * be freed by btrfs_finish_ordered_io().
1529 if (cur_offset > end)
1532 btrfs_release_path(path);
1534 if (cur_offset <= end && cow_start == (u64)-1) {
1535 cow_start = cur_offset;
1539 if (cow_start != (u64)-1) {
1540 ret = cow_file_range(inode, locked_page, cow_start, end, end,
1541 page_started, nr_written, 1, NULL);
1547 if (ret && cur_offset < end)
1548 extent_clear_unlock_delalloc(inode, cur_offset, end, end,
1549 locked_page, EXTENT_LOCKED |
1550 EXTENT_DELALLOC | EXTENT_DEFRAG |
1551 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1553 PAGE_SET_WRITEBACK |
1554 PAGE_END_WRITEBACK);
1555 btrfs_free_path(path);
1559 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1562 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1563 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1567 * @defrag_bytes is a hint value, no spinlock held here,
1568 * if is not zero, it means the file is defragging.
1569 * Force cow if given extent needs to be defragged.
1571 if (BTRFS_I(inode)->defrag_bytes &&
1572 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1573 EXTENT_DEFRAG, 0, NULL))
1580 * extent_io.c call back to do delayed allocation processing
1582 static int run_delalloc_range(void *private_data, struct page *locked_page,
1583 u64 start, u64 end, int *page_started,
1584 unsigned long *nr_written,
1585 struct writeback_control *wbc)
1587 struct inode *inode = private_data;
1589 int force_cow = need_force_cow(inode, start, end);
1590 unsigned int write_flags = wbc_to_write_flags(wbc);
1592 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1593 ret = run_delalloc_nocow(inode, locked_page, start, end,
1594 page_started, 1, nr_written);
1595 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1596 ret = run_delalloc_nocow(inode, locked_page, start, end,
1597 page_started, 0, nr_written);
1598 } else if (!inode_need_compress(inode, start, end)) {
1599 ret = cow_file_range(inode, locked_page, start, end, end,
1600 page_started, nr_written, 1, NULL);
1602 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1603 &BTRFS_I(inode)->runtime_flags);
1604 ret = cow_file_range_async(inode, locked_page, start, end,
1605 page_started, nr_written,
1609 btrfs_cleanup_ordered_extents(inode, start, end - start + 1);
1613 static void btrfs_split_extent_hook(void *private_data,
1614 struct extent_state *orig, u64 split)
1616 struct inode *inode = private_data;
1619 /* not delalloc, ignore it */
1620 if (!(orig->state & EXTENT_DELALLOC))
1623 size = orig->end - orig->start + 1;
1624 if (size > BTRFS_MAX_EXTENT_SIZE) {
1629 * See the explanation in btrfs_merge_extent_hook, the same
1630 * applies here, just in reverse.
1632 new_size = orig->end - split + 1;
1633 num_extents = count_max_extents(new_size);
1634 new_size = split - orig->start;
1635 num_extents += count_max_extents(new_size);
1636 if (count_max_extents(size) >= num_extents)
1640 spin_lock(&BTRFS_I(inode)->lock);
1641 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1642 spin_unlock(&BTRFS_I(inode)->lock);
1646 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1647 * extents so we can keep track of new extents that are just merged onto old
1648 * extents, such as when we are doing sequential writes, so we can properly
1649 * account for the metadata space we'll need.
1651 static void btrfs_merge_extent_hook(void *private_data,
1652 struct extent_state *new,
1653 struct extent_state *other)
1655 struct inode *inode = private_data;
1656 u64 new_size, old_size;
1659 /* not delalloc, ignore it */
1660 if (!(other->state & EXTENT_DELALLOC))
1663 if (new->start > other->start)
1664 new_size = new->end - other->start + 1;
1666 new_size = other->end - new->start + 1;
1668 /* we're not bigger than the max, unreserve the space and go */
1669 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1670 spin_lock(&BTRFS_I(inode)->lock);
1671 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1672 spin_unlock(&BTRFS_I(inode)->lock);
1677 * We have to add up either side to figure out how many extents were
1678 * accounted for before we merged into one big extent. If the number of
1679 * extents we accounted for is <= the amount we need for the new range
1680 * then we can return, otherwise drop. Think of it like this
1684 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1685 * need 2 outstanding extents, on one side we have 1 and the other side
1686 * we have 1 so they are == and we can return. But in this case
1688 * [MAX_SIZE+4k][MAX_SIZE+4k]
1690 * Each range on their own accounts for 2 extents, but merged together
1691 * they are only 3 extents worth of accounting, so we need to drop in
1694 old_size = other->end - other->start + 1;
1695 num_extents = count_max_extents(old_size);
1696 old_size = new->end - new->start + 1;
1697 num_extents += count_max_extents(old_size);
1698 if (count_max_extents(new_size) >= num_extents)
1701 spin_lock(&BTRFS_I(inode)->lock);
1702 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1703 spin_unlock(&BTRFS_I(inode)->lock);
1706 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1707 struct inode *inode)
1709 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1711 spin_lock(&root->delalloc_lock);
1712 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1713 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1714 &root->delalloc_inodes);
1715 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1716 &BTRFS_I(inode)->runtime_flags);
1717 root->nr_delalloc_inodes++;
1718 if (root->nr_delalloc_inodes == 1) {
1719 spin_lock(&fs_info->delalloc_root_lock);
1720 BUG_ON(!list_empty(&root->delalloc_root));
1721 list_add_tail(&root->delalloc_root,
1722 &fs_info->delalloc_roots);
1723 spin_unlock(&fs_info->delalloc_root_lock);
1726 spin_unlock(&root->delalloc_lock);
1730 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
1731 struct btrfs_inode *inode)
1733 struct btrfs_fs_info *fs_info = root->fs_info;
1735 if (!list_empty(&inode->delalloc_inodes)) {
1736 list_del_init(&inode->delalloc_inodes);
1737 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1738 &inode->runtime_flags);
1739 root->nr_delalloc_inodes--;
1740 if (!root->nr_delalloc_inodes) {
1741 ASSERT(list_empty(&root->delalloc_inodes));
1742 spin_lock(&fs_info->delalloc_root_lock);
1743 BUG_ON(list_empty(&root->delalloc_root));
1744 list_del_init(&root->delalloc_root);
1745 spin_unlock(&fs_info->delalloc_root_lock);
1750 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1751 struct btrfs_inode *inode)
1753 spin_lock(&root->delalloc_lock);
1754 __btrfs_del_delalloc_inode(root, inode);
1755 spin_unlock(&root->delalloc_lock);
1759 * extent_io.c set_bit_hook, used to track delayed allocation
1760 * bytes in this file, and to maintain the list of inodes that
1761 * have pending delalloc work to be done.
1763 static void btrfs_set_bit_hook(void *private_data,
1764 struct extent_state *state, unsigned *bits)
1766 struct inode *inode = private_data;
1768 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1770 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1773 * set_bit and clear bit hooks normally require _irqsave/restore
1774 * but in this case, we are only testing for the DELALLOC
1775 * bit, which is only set or cleared with irqs on
1777 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1778 struct btrfs_root *root = BTRFS_I(inode)->root;
1779 u64 len = state->end + 1 - state->start;
1780 u32 num_extents = count_max_extents(len);
1781 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
1783 spin_lock(&BTRFS_I(inode)->lock);
1784 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
1785 spin_unlock(&BTRFS_I(inode)->lock);
1787 /* For sanity tests */
1788 if (btrfs_is_testing(fs_info))
1791 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
1792 fs_info->delalloc_batch);
1793 spin_lock(&BTRFS_I(inode)->lock);
1794 BTRFS_I(inode)->delalloc_bytes += len;
1795 if (*bits & EXTENT_DEFRAG)
1796 BTRFS_I(inode)->defrag_bytes += len;
1797 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1798 &BTRFS_I(inode)->runtime_flags))
1799 btrfs_add_delalloc_inodes(root, inode);
1800 spin_unlock(&BTRFS_I(inode)->lock);
1803 if (!(state->state & EXTENT_DELALLOC_NEW) &&
1804 (*bits & EXTENT_DELALLOC_NEW)) {
1805 spin_lock(&BTRFS_I(inode)->lock);
1806 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
1808 spin_unlock(&BTRFS_I(inode)->lock);
1813 * extent_io.c clear_bit_hook, see set_bit_hook for why
1815 static void btrfs_clear_bit_hook(void *private_data,
1816 struct extent_state *state,
1819 struct btrfs_inode *inode = BTRFS_I((struct inode *)private_data);
1820 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1821 u64 len = state->end + 1 - state->start;
1822 u32 num_extents = count_max_extents(len);
1824 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
1825 spin_lock(&inode->lock);
1826 inode->defrag_bytes -= len;
1827 spin_unlock(&inode->lock);
1831 * set_bit and clear bit hooks normally require _irqsave/restore
1832 * but in this case, we are only testing for the DELALLOC
1833 * bit, which is only set or cleared with irqs on
1835 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1836 struct btrfs_root *root = inode->root;
1837 bool do_list = !btrfs_is_free_space_inode(inode);
1839 spin_lock(&inode->lock);
1840 btrfs_mod_outstanding_extents(inode, -num_extents);
1841 spin_unlock(&inode->lock);
1844 * We don't reserve metadata space for space cache inodes so we
1845 * don't need to call dellalloc_release_metadata if there is an
1848 if (*bits & EXTENT_CLEAR_META_RESV &&
1849 root != fs_info->tree_root)
1850 btrfs_delalloc_release_metadata(inode, len, false);
1852 /* For sanity tests. */
1853 if (btrfs_is_testing(fs_info))
1856 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
1857 do_list && !(state->state & EXTENT_NORESERVE) &&
1858 (*bits & EXTENT_CLEAR_DATA_RESV))
1859 btrfs_free_reserved_data_space_noquota(
1863 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
1864 fs_info->delalloc_batch);
1865 spin_lock(&inode->lock);
1866 inode->delalloc_bytes -= len;
1867 if (do_list && inode->delalloc_bytes == 0 &&
1868 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1869 &inode->runtime_flags))
1870 btrfs_del_delalloc_inode(root, inode);
1871 spin_unlock(&inode->lock);
1874 if ((state->state & EXTENT_DELALLOC_NEW) &&
1875 (*bits & EXTENT_DELALLOC_NEW)) {
1876 spin_lock(&inode->lock);
1877 ASSERT(inode->new_delalloc_bytes >= len);
1878 inode->new_delalloc_bytes -= len;
1879 spin_unlock(&inode->lock);
1884 * Merge bio hook, this must check the chunk tree to make sure we don't create
1885 * bios that span stripes or chunks
1887 * return 1 if page cannot be merged to bio
1888 * return 0 if page can be merged to bio
1889 * return error otherwise
1891 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1892 size_t size, struct bio *bio,
1893 unsigned long bio_flags)
1895 struct inode *inode = page->mapping->host;
1896 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1897 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1902 if (bio_flags & EXTENT_BIO_COMPRESSED)
1905 length = bio->bi_iter.bi_size;
1906 map_length = length;
1907 ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
1911 if (map_length < length + size)
1917 * in order to insert checksums into the metadata in large chunks,
1918 * we wait until bio submission time. All the pages in the bio are
1919 * checksummed and sums are attached onto the ordered extent record.
1921 * At IO completion time the cums attached on the ordered extent record
1922 * are inserted into the btree
1924 static blk_status_t btrfs_submit_bio_start(void *private_data, struct bio *bio,
1927 struct inode *inode = private_data;
1928 blk_status_t ret = 0;
1930 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1931 BUG_ON(ret); /* -ENOMEM */
1936 * in order to insert checksums into the metadata in large chunks,
1937 * we wait until bio submission time. All the pages in the bio are
1938 * checksummed and sums are attached onto the ordered extent record.
1940 * At IO completion time the cums attached on the ordered extent record
1941 * are inserted into the btree
1943 blk_status_t btrfs_submit_bio_done(void *private_data, struct bio *bio,
1946 struct inode *inode = private_data;
1947 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1950 ret = btrfs_map_bio(fs_info, bio, mirror_num, 1);
1952 bio->bi_status = ret;
1959 * extent_io.c submission hook. This does the right thing for csum calculation
1960 * on write, or reading the csums from the tree before a read.
1962 * Rules about async/sync submit,
1963 * a) read: sync submit
1965 * b) write without checksum: sync submit
1967 * c) write with checksum:
1968 * c-1) if bio is issued by fsync: sync submit
1969 * (sync_writers != 0)
1971 * c-2) if root is reloc root: sync submit
1972 * (only in case of buffered IO)
1974 * c-3) otherwise: async submit
1976 static blk_status_t btrfs_submit_bio_hook(void *private_data, struct bio *bio,
1977 int mirror_num, unsigned long bio_flags,
1980 struct inode *inode = private_data;
1981 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1982 struct btrfs_root *root = BTRFS_I(inode)->root;
1983 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1984 blk_status_t ret = 0;
1986 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1988 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1990 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
1991 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1993 if (bio_op(bio) != REQ_OP_WRITE) {
1994 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
1998 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1999 ret = btrfs_submit_compressed_read(inode, bio,
2003 } else if (!skip_sum) {
2004 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
2009 } else if (async && !skip_sum) {
2010 /* csum items have already been cloned */
2011 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2013 /* we're doing a write, do the async checksumming */
2014 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags,
2016 btrfs_submit_bio_start);
2018 } else if (!skip_sum) {
2019 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
2025 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
2029 bio->bi_status = ret;
2036 * given a list of ordered sums record them in the inode. This happens
2037 * at IO completion time based on sums calculated at bio submission time.
2039 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
2040 struct inode *inode, struct list_head *list)
2042 struct btrfs_ordered_sum *sum;
2045 list_for_each_entry(sum, list, list) {
2046 trans->adding_csums = true;
2047 ret = btrfs_csum_file_blocks(trans,
2048 BTRFS_I(inode)->root->fs_info->csum_root, sum);
2049 trans->adding_csums = false;
2056 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
2057 unsigned int extra_bits,
2058 struct extent_state **cached_state, int dedupe)
2060 WARN_ON((end & (PAGE_SIZE - 1)) == 0);
2061 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2062 extra_bits, cached_state);
2065 /* see btrfs_writepage_start_hook for details on why this is required */
2066 struct btrfs_writepage_fixup {
2068 struct btrfs_work work;
2071 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2073 struct btrfs_writepage_fixup *fixup;
2074 struct btrfs_ordered_extent *ordered;
2075 struct extent_state *cached_state = NULL;
2076 struct extent_changeset *data_reserved = NULL;
2078 struct inode *inode;
2083 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2087 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2088 ClearPageChecked(page);
2092 inode = page->mapping->host;
2093 page_start = page_offset(page);
2094 page_end = page_offset(page) + PAGE_SIZE - 1;
2096 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2099 /* already ordered? We're done */
2100 if (PagePrivate2(page))
2103 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
2106 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2107 page_end, &cached_state);
2109 btrfs_start_ordered_extent(inode, ordered, 1);
2110 btrfs_put_ordered_extent(ordered);
2114 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2117 mapping_set_error(page->mapping, ret);
2118 end_extent_writepage(page, ret, page_start, page_end);
2119 ClearPageChecked(page);
2123 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2126 mapping_set_error(page->mapping, ret);
2127 end_extent_writepage(page, ret, page_start, page_end);
2128 ClearPageChecked(page);
2132 ClearPageChecked(page);
2133 set_page_dirty(page);
2134 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, false);
2136 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2142 extent_changeset_free(data_reserved);
2146 * There are a few paths in the higher layers of the kernel that directly
2147 * set the page dirty bit without asking the filesystem if it is a
2148 * good idea. This causes problems because we want to make sure COW
2149 * properly happens and the data=ordered rules are followed.
2151 * In our case any range that doesn't have the ORDERED bit set
2152 * hasn't been properly setup for IO. We kick off an async process
2153 * to fix it up. The async helper will wait for ordered extents, set
2154 * the delalloc bit and make it safe to write the page.
2156 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2158 struct inode *inode = page->mapping->host;
2159 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2160 struct btrfs_writepage_fixup *fixup;
2162 /* this page is properly in the ordered list */
2163 if (TestClearPagePrivate2(page))
2166 if (PageChecked(page))
2169 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2173 SetPageChecked(page);
2175 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2176 btrfs_writepage_fixup_worker, NULL, NULL);
2178 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2182 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2183 struct inode *inode, u64 file_pos,
2184 u64 disk_bytenr, u64 disk_num_bytes,
2185 u64 num_bytes, u64 ram_bytes,
2186 u8 compression, u8 encryption,
2187 u16 other_encoding, int extent_type)
2189 struct btrfs_root *root = BTRFS_I(inode)->root;
2190 struct btrfs_file_extent_item *fi;
2191 struct btrfs_path *path;
2192 struct extent_buffer *leaf;
2193 struct btrfs_key ins;
2195 int extent_inserted = 0;
2198 path = btrfs_alloc_path();
2203 * we may be replacing one extent in the tree with another.
2204 * The new extent is pinned in the extent map, and we don't want
2205 * to drop it from the cache until it is completely in the btree.
2207 * So, tell btrfs_drop_extents to leave this extent in the cache.
2208 * the caller is expected to unpin it and allow it to be merged
2211 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2212 file_pos + num_bytes, NULL, 0,
2213 1, sizeof(*fi), &extent_inserted);
2217 if (!extent_inserted) {
2218 ins.objectid = btrfs_ino(BTRFS_I(inode));
2219 ins.offset = file_pos;
2220 ins.type = BTRFS_EXTENT_DATA_KEY;
2222 path->leave_spinning = 1;
2223 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2228 leaf = path->nodes[0];
2229 fi = btrfs_item_ptr(leaf, path->slots[0],
2230 struct btrfs_file_extent_item);
2231 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2232 btrfs_set_file_extent_type(leaf, fi, extent_type);
2233 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2234 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2235 btrfs_set_file_extent_offset(leaf, fi, 0);
2236 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2237 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2238 btrfs_set_file_extent_compression(leaf, fi, compression);
2239 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2240 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2242 btrfs_mark_buffer_dirty(leaf);
2243 btrfs_release_path(path);
2245 inode_add_bytes(inode, num_bytes);
2247 ins.objectid = disk_bytenr;
2248 ins.offset = disk_num_bytes;
2249 ins.type = BTRFS_EXTENT_ITEM_KEY;
2252 * Release the reserved range from inode dirty range map, as it is
2253 * already moved into delayed_ref_head
2255 ret = btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2259 ret = btrfs_alloc_reserved_file_extent(trans, root,
2260 btrfs_ino(BTRFS_I(inode)),
2261 file_pos, qg_released, &ins);
2263 btrfs_free_path(path);
2268 /* snapshot-aware defrag */
2269 struct sa_defrag_extent_backref {
2270 struct rb_node node;
2271 struct old_sa_defrag_extent *old;
2280 struct old_sa_defrag_extent {
2281 struct list_head list;
2282 struct new_sa_defrag_extent *new;
2291 struct new_sa_defrag_extent {
2292 struct rb_root root;
2293 struct list_head head;
2294 struct btrfs_path *path;
2295 struct inode *inode;
2303 static int backref_comp(struct sa_defrag_extent_backref *b1,
2304 struct sa_defrag_extent_backref *b2)
2306 if (b1->root_id < b2->root_id)
2308 else if (b1->root_id > b2->root_id)
2311 if (b1->inum < b2->inum)
2313 else if (b1->inum > b2->inum)
2316 if (b1->file_pos < b2->file_pos)
2318 else if (b1->file_pos > b2->file_pos)
2322 * [------------------------------] ===> (a range of space)
2323 * |<--->| |<---->| =============> (fs/file tree A)
2324 * |<---------------------------->| ===> (fs/file tree B)
2326 * A range of space can refer to two file extents in one tree while
2327 * refer to only one file extent in another tree.
2329 * So we may process a disk offset more than one time(two extents in A)
2330 * and locate at the same extent(one extent in B), then insert two same
2331 * backrefs(both refer to the extent in B).
2336 static void backref_insert(struct rb_root *root,
2337 struct sa_defrag_extent_backref *backref)
2339 struct rb_node **p = &root->rb_node;
2340 struct rb_node *parent = NULL;
2341 struct sa_defrag_extent_backref *entry;
2346 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2348 ret = backref_comp(backref, entry);
2352 p = &(*p)->rb_right;
2355 rb_link_node(&backref->node, parent, p);
2356 rb_insert_color(&backref->node, root);
2360 * Note the backref might has changed, and in this case we just return 0.
2362 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2365 struct btrfs_file_extent_item *extent;
2366 struct old_sa_defrag_extent *old = ctx;
2367 struct new_sa_defrag_extent *new = old->new;
2368 struct btrfs_path *path = new->path;
2369 struct btrfs_key key;
2370 struct btrfs_root *root;
2371 struct sa_defrag_extent_backref *backref;
2372 struct extent_buffer *leaf;
2373 struct inode *inode = new->inode;
2374 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2380 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2381 inum == btrfs_ino(BTRFS_I(inode)))
2384 key.objectid = root_id;
2385 key.type = BTRFS_ROOT_ITEM_KEY;
2386 key.offset = (u64)-1;
2388 root = btrfs_read_fs_root_no_name(fs_info, &key);
2390 if (PTR_ERR(root) == -ENOENT)
2393 btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu",
2394 inum, offset, root_id);
2395 return PTR_ERR(root);
2398 key.objectid = inum;
2399 key.type = BTRFS_EXTENT_DATA_KEY;
2400 if (offset > (u64)-1 << 32)
2403 key.offset = offset;
2405 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2406 if (WARN_ON(ret < 0))
2413 leaf = path->nodes[0];
2414 slot = path->slots[0];
2416 if (slot >= btrfs_header_nritems(leaf)) {
2417 ret = btrfs_next_leaf(root, path);
2420 } else if (ret > 0) {
2429 btrfs_item_key_to_cpu(leaf, &key, slot);
2431 if (key.objectid > inum)
2434 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2437 extent = btrfs_item_ptr(leaf, slot,
2438 struct btrfs_file_extent_item);
2440 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2444 * 'offset' refers to the exact key.offset,
2445 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2446 * (key.offset - extent_offset).
2448 if (key.offset != offset)
2451 extent_offset = btrfs_file_extent_offset(leaf, extent);
2452 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2454 if (extent_offset >= old->extent_offset + old->offset +
2455 old->len || extent_offset + num_bytes <=
2456 old->extent_offset + old->offset)
2461 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2467 backref->root_id = root_id;
2468 backref->inum = inum;
2469 backref->file_pos = offset;
2470 backref->num_bytes = num_bytes;
2471 backref->extent_offset = extent_offset;
2472 backref->generation = btrfs_file_extent_generation(leaf, extent);
2474 backref_insert(&new->root, backref);
2477 btrfs_release_path(path);
2482 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2483 struct new_sa_defrag_extent *new)
2485 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2486 struct old_sa_defrag_extent *old, *tmp;
2491 list_for_each_entry_safe(old, tmp, &new->head, list) {
2492 ret = iterate_inodes_from_logical(old->bytenr +
2493 old->extent_offset, fs_info,
2494 path, record_one_backref,
2496 if (ret < 0 && ret != -ENOENT)
2499 /* no backref to be processed for this extent */
2501 list_del(&old->list);
2506 if (list_empty(&new->head))
2512 static int relink_is_mergable(struct extent_buffer *leaf,
2513 struct btrfs_file_extent_item *fi,
2514 struct new_sa_defrag_extent *new)
2516 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2519 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2522 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2525 if (btrfs_file_extent_encryption(leaf, fi) ||
2526 btrfs_file_extent_other_encoding(leaf, fi))
2533 * Note the backref might has changed, and in this case we just return 0.
2535 static noinline int relink_extent_backref(struct btrfs_path *path,
2536 struct sa_defrag_extent_backref *prev,
2537 struct sa_defrag_extent_backref *backref)
2539 struct btrfs_file_extent_item *extent;
2540 struct btrfs_file_extent_item *item;
2541 struct btrfs_ordered_extent *ordered;
2542 struct btrfs_trans_handle *trans;
2543 struct btrfs_root *root;
2544 struct btrfs_key key;
2545 struct extent_buffer *leaf;
2546 struct old_sa_defrag_extent *old = backref->old;
2547 struct new_sa_defrag_extent *new = old->new;
2548 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2549 struct inode *inode;
2550 struct extent_state *cached = NULL;
2559 if (prev && prev->root_id == backref->root_id &&
2560 prev->inum == backref->inum &&
2561 prev->file_pos + prev->num_bytes == backref->file_pos)
2564 /* step 1: get root */
2565 key.objectid = backref->root_id;
2566 key.type = BTRFS_ROOT_ITEM_KEY;
2567 key.offset = (u64)-1;
2569 index = srcu_read_lock(&fs_info->subvol_srcu);
2571 root = btrfs_read_fs_root_no_name(fs_info, &key);
2573 srcu_read_unlock(&fs_info->subvol_srcu, index);
2574 if (PTR_ERR(root) == -ENOENT)
2576 return PTR_ERR(root);
2579 if (btrfs_root_readonly(root)) {
2580 srcu_read_unlock(&fs_info->subvol_srcu, index);
2584 /* step 2: get inode */
2585 key.objectid = backref->inum;
2586 key.type = BTRFS_INODE_ITEM_KEY;
2589 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2590 if (IS_ERR(inode)) {
2591 srcu_read_unlock(&fs_info->subvol_srcu, index);
2595 srcu_read_unlock(&fs_info->subvol_srcu, index);
2597 /* step 3: relink backref */
2598 lock_start = backref->file_pos;
2599 lock_end = backref->file_pos + backref->num_bytes - 1;
2600 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2603 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2605 btrfs_put_ordered_extent(ordered);
2609 trans = btrfs_join_transaction(root);
2610 if (IS_ERR(trans)) {
2611 ret = PTR_ERR(trans);
2615 key.objectid = backref->inum;
2616 key.type = BTRFS_EXTENT_DATA_KEY;
2617 key.offset = backref->file_pos;
2619 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2622 } else if (ret > 0) {
2627 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2628 struct btrfs_file_extent_item);
2630 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2631 backref->generation)
2634 btrfs_release_path(path);
2636 start = backref->file_pos;
2637 if (backref->extent_offset < old->extent_offset + old->offset)
2638 start += old->extent_offset + old->offset -
2639 backref->extent_offset;
2641 len = min(backref->extent_offset + backref->num_bytes,
2642 old->extent_offset + old->offset + old->len);
2643 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2645 ret = btrfs_drop_extents(trans, root, inode, start,
2650 key.objectid = btrfs_ino(BTRFS_I(inode));
2651 key.type = BTRFS_EXTENT_DATA_KEY;
2654 path->leave_spinning = 1;
2656 struct btrfs_file_extent_item *fi;
2658 struct btrfs_key found_key;
2660 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2665 leaf = path->nodes[0];
2666 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2668 fi = btrfs_item_ptr(leaf, path->slots[0],
2669 struct btrfs_file_extent_item);
2670 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2672 if (extent_len + found_key.offset == start &&
2673 relink_is_mergable(leaf, fi, new)) {
2674 btrfs_set_file_extent_num_bytes(leaf, fi,
2676 btrfs_mark_buffer_dirty(leaf);
2677 inode_add_bytes(inode, len);
2683 btrfs_release_path(path);
2688 ret = btrfs_insert_empty_item(trans, root, path, &key,
2691 btrfs_abort_transaction(trans, ret);
2695 leaf = path->nodes[0];
2696 item = btrfs_item_ptr(leaf, path->slots[0],
2697 struct btrfs_file_extent_item);
2698 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2699 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2700 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2701 btrfs_set_file_extent_num_bytes(leaf, item, len);
2702 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2703 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2704 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2705 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2706 btrfs_set_file_extent_encryption(leaf, item, 0);
2707 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2709 btrfs_mark_buffer_dirty(leaf);
2710 inode_add_bytes(inode, len);
2711 btrfs_release_path(path);
2713 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2715 backref->root_id, backref->inum,
2716 new->file_pos); /* start - extent_offset */
2718 btrfs_abort_transaction(trans, ret);
2724 btrfs_release_path(path);
2725 path->leave_spinning = 0;
2726 btrfs_end_transaction(trans);
2728 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2734 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2736 struct old_sa_defrag_extent *old, *tmp;
2741 list_for_each_entry_safe(old, tmp, &new->head, list) {
2747 static void relink_file_extents(struct new_sa_defrag_extent *new)
2749 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2750 struct btrfs_path *path;
2751 struct sa_defrag_extent_backref *backref;
2752 struct sa_defrag_extent_backref *prev = NULL;
2753 struct rb_node *node;
2756 path = btrfs_alloc_path();
2760 if (!record_extent_backrefs(path, new)) {
2761 btrfs_free_path(path);
2764 btrfs_release_path(path);
2767 node = rb_first(&new->root);
2770 rb_erase(node, &new->root);
2772 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2774 ret = relink_extent_backref(path, prev, backref);
2787 btrfs_free_path(path);
2789 free_sa_defrag_extent(new);
2791 atomic_dec(&fs_info->defrag_running);
2792 wake_up(&fs_info->transaction_wait);
2795 static struct new_sa_defrag_extent *
2796 record_old_file_extents(struct inode *inode,
2797 struct btrfs_ordered_extent *ordered)
2799 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2800 struct btrfs_root *root = BTRFS_I(inode)->root;
2801 struct btrfs_path *path;
2802 struct btrfs_key key;
2803 struct old_sa_defrag_extent *old;
2804 struct new_sa_defrag_extent *new;
2807 new = kmalloc(sizeof(*new), GFP_NOFS);
2812 new->file_pos = ordered->file_offset;
2813 new->len = ordered->len;
2814 new->bytenr = ordered->start;
2815 new->disk_len = ordered->disk_len;
2816 new->compress_type = ordered->compress_type;
2817 new->root = RB_ROOT;
2818 INIT_LIST_HEAD(&new->head);
2820 path = btrfs_alloc_path();
2824 key.objectid = btrfs_ino(BTRFS_I(inode));
2825 key.type = BTRFS_EXTENT_DATA_KEY;
2826 key.offset = new->file_pos;
2828 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2831 if (ret > 0 && path->slots[0] > 0)
2834 /* find out all the old extents for the file range */
2836 struct btrfs_file_extent_item *extent;
2837 struct extent_buffer *l;
2846 slot = path->slots[0];
2848 if (slot >= btrfs_header_nritems(l)) {
2849 ret = btrfs_next_leaf(root, path);
2857 btrfs_item_key_to_cpu(l, &key, slot);
2859 if (key.objectid != btrfs_ino(BTRFS_I(inode)))
2861 if (key.type != BTRFS_EXTENT_DATA_KEY)
2863 if (key.offset >= new->file_pos + new->len)
2866 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2868 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2869 if (key.offset + num_bytes < new->file_pos)
2872 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2876 extent_offset = btrfs_file_extent_offset(l, extent);
2878 old = kmalloc(sizeof(*old), GFP_NOFS);
2882 offset = max(new->file_pos, key.offset);
2883 end = min(new->file_pos + new->len, key.offset + num_bytes);
2885 old->bytenr = disk_bytenr;
2886 old->extent_offset = extent_offset;
2887 old->offset = offset - key.offset;
2888 old->len = end - offset;
2891 list_add_tail(&old->list, &new->head);
2897 btrfs_free_path(path);
2898 atomic_inc(&fs_info->defrag_running);
2903 btrfs_free_path(path);
2905 free_sa_defrag_extent(new);
2909 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2912 struct btrfs_block_group_cache *cache;
2914 cache = btrfs_lookup_block_group(fs_info, start);
2917 spin_lock(&cache->lock);
2918 cache->delalloc_bytes -= len;
2919 spin_unlock(&cache->lock);
2921 btrfs_put_block_group(cache);
2924 /* as ordered data IO finishes, this gets called so we can finish
2925 * an ordered extent if the range of bytes in the file it covers are
2928 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2930 struct inode *inode = ordered_extent->inode;
2931 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2932 struct btrfs_root *root = BTRFS_I(inode)->root;
2933 struct btrfs_trans_handle *trans = NULL;
2934 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2935 struct extent_state *cached_state = NULL;
2936 struct new_sa_defrag_extent *new = NULL;
2937 int compress_type = 0;
2939 u64 logical_len = ordered_extent->len;
2941 bool truncated = false;
2942 bool range_locked = false;
2943 bool clear_new_delalloc_bytes = false;
2944 bool clear_reserved_extent = true;
2946 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2947 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2948 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2949 clear_new_delalloc_bytes = true;
2951 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
2953 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2958 btrfs_free_io_failure_record(BTRFS_I(inode),
2959 ordered_extent->file_offset,
2960 ordered_extent->file_offset +
2961 ordered_extent->len - 1);
2963 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2965 logical_len = ordered_extent->truncated_len;
2966 /* Truncated the entire extent, don't bother adding */
2971 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2972 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2975 * For mwrite(mmap + memset to write) case, we still reserve
2976 * space for NOCOW range.
2977 * As NOCOW won't cause a new delayed ref, just free the space
2979 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
2980 ordered_extent->len);
2981 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2983 trans = btrfs_join_transaction_nolock(root);
2985 trans = btrfs_join_transaction(root);
2986 if (IS_ERR(trans)) {
2987 ret = PTR_ERR(trans);
2991 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
2992 ret = btrfs_update_inode_fallback(trans, root, inode);
2993 if (ret) /* -ENOMEM or corruption */
2994 btrfs_abort_transaction(trans, ret);
2998 range_locked = true;
2999 lock_extent_bits(io_tree, ordered_extent->file_offset,
3000 ordered_extent->file_offset + ordered_extent->len - 1,
3003 ret = test_range_bit(io_tree, ordered_extent->file_offset,
3004 ordered_extent->file_offset + ordered_extent->len - 1,
3005 EXTENT_DEFRAG, 0, cached_state);
3007 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
3008 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
3009 /* the inode is shared */
3010 new = record_old_file_extents(inode, ordered_extent);
3012 clear_extent_bit(io_tree, ordered_extent->file_offset,
3013 ordered_extent->file_offset + ordered_extent->len - 1,
3014 EXTENT_DEFRAG, 0, 0, &cached_state);
3018 trans = btrfs_join_transaction_nolock(root);
3020 trans = btrfs_join_transaction(root);
3021 if (IS_ERR(trans)) {
3022 ret = PTR_ERR(trans);
3027 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
3029 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3030 compress_type = ordered_extent->compress_type;
3031 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3032 BUG_ON(compress_type);
3033 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
3034 ordered_extent->len);
3035 ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
3036 ordered_extent->file_offset,
3037 ordered_extent->file_offset +
3040 BUG_ON(root == fs_info->tree_root);
3041 ret = insert_reserved_file_extent(trans, inode,
3042 ordered_extent->file_offset,
3043 ordered_extent->start,
3044 ordered_extent->disk_len,
3045 logical_len, logical_len,
3046 compress_type, 0, 0,
3047 BTRFS_FILE_EXTENT_REG);
3049 clear_reserved_extent = false;
3050 btrfs_release_delalloc_bytes(fs_info,
3051 ordered_extent->start,
3052 ordered_extent->disk_len);
3055 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
3056 ordered_extent->file_offset, ordered_extent->len,
3059 btrfs_abort_transaction(trans, ret);
3063 ret = add_pending_csums(trans, inode, &ordered_extent->list);
3065 btrfs_abort_transaction(trans, ret);
3069 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
3070 ret = btrfs_update_inode_fallback(trans, root, inode);
3071 if (ret) { /* -ENOMEM or corruption */
3072 btrfs_abort_transaction(trans, ret);
3077 if (range_locked || clear_new_delalloc_bytes) {
3078 unsigned int clear_bits = 0;
3081 clear_bits |= EXTENT_LOCKED;
3082 if (clear_new_delalloc_bytes)
3083 clear_bits |= EXTENT_DELALLOC_NEW;
3084 clear_extent_bit(&BTRFS_I(inode)->io_tree,
3085 ordered_extent->file_offset,
3086 ordered_extent->file_offset +
3087 ordered_extent->len - 1,
3089 (clear_bits & EXTENT_LOCKED) ? 1 : 0,
3094 btrfs_end_transaction(trans);
3096 if (ret || truncated) {
3100 start = ordered_extent->file_offset + logical_len;
3102 start = ordered_extent->file_offset;
3103 end = ordered_extent->file_offset + ordered_extent->len - 1;
3104 clear_extent_uptodate(io_tree, start, end, NULL);
3106 /* Drop the cache for the part of the extent we didn't write. */
3107 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
3110 * If the ordered extent had an IOERR or something else went
3111 * wrong we need to return the space for this ordered extent
3112 * back to the allocator. We only free the extent in the
3113 * truncated case if we didn't write out the extent at all.
3115 * If we made it past insert_reserved_file_extent before we
3116 * errored out then we don't need to do this as the accounting
3117 * has already been done.
3119 if ((ret || !logical_len) &&
3120 clear_reserved_extent &&
3121 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3122 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
3123 btrfs_free_reserved_extent(fs_info,
3124 ordered_extent->start,
3125 ordered_extent->disk_len, 1);
3130 * This needs to be done to make sure anybody waiting knows we are done
3131 * updating everything for this ordered extent.
3133 btrfs_remove_ordered_extent(inode, ordered_extent);
3135 /* for snapshot-aware defrag */
3138 free_sa_defrag_extent(new);
3139 atomic_dec(&fs_info->defrag_running);
3141 relink_file_extents(new);
3146 btrfs_put_ordered_extent(ordered_extent);
3147 /* once for the tree */
3148 btrfs_put_ordered_extent(ordered_extent);
3150 /* Try to release some metadata so we don't get an OOM but don't wait */
3151 btrfs_btree_balance_dirty_nodelay(fs_info);
3156 static void finish_ordered_fn(struct btrfs_work *work)
3158 struct btrfs_ordered_extent *ordered_extent;
3159 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3160 btrfs_finish_ordered_io(ordered_extent);
3163 static void btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3164 struct extent_state *state, int uptodate)
3166 struct inode *inode = page->mapping->host;
3167 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3168 struct btrfs_ordered_extent *ordered_extent = NULL;
3169 struct btrfs_workqueue *wq;
3170 btrfs_work_func_t func;
3172 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3174 ClearPagePrivate2(page);
3175 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3176 end - start + 1, uptodate))
3179 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
3180 wq = fs_info->endio_freespace_worker;
3181 func = btrfs_freespace_write_helper;
3183 wq = fs_info->endio_write_workers;
3184 func = btrfs_endio_write_helper;
3187 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3189 btrfs_queue_work(wq, &ordered_extent->work);
3192 static int __readpage_endio_check(struct inode *inode,
3193 struct btrfs_io_bio *io_bio,
3194 int icsum, struct page *page,
3195 int pgoff, u64 start, size_t len)
3201 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3203 kaddr = kmap_atomic(page);
3204 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3205 btrfs_csum_final(csum, (u8 *)&csum);
3206 if (csum != csum_expected)
3209 kunmap_atomic(kaddr);
3212 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3213 io_bio->mirror_num);
3214 memset(kaddr + pgoff, 1, len);
3215 flush_dcache_page(page);
3216 kunmap_atomic(kaddr);
3221 * when reads are done, we need to check csums to verify the data is correct
3222 * if there's a match, we allow the bio to finish. If not, the code in
3223 * extent_io.c will try to find good copies for us.
3225 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3226 u64 phy_offset, struct page *page,
3227 u64 start, u64 end, int mirror)
3229 size_t offset = start - page_offset(page);
3230 struct inode *inode = page->mapping->host;
3231 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3232 struct btrfs_root *root = BTRFS_I(inode)->root;
3234 if (PageChecked(page)) {
3235 ClearPageChecked(page);
3239 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3242 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3243 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3244 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3248 phy_offset >>= inode->i_sb->s_blocksize_bits;
3249 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3250 start, (size_t)(end - start + 1));
3254 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3256 * @inode: The inode we want to perform iput on
3258 * This function uses the generic vfs_inode::i_count to track whether we should
3259 * just decrement it (in case it's > 1) or if this is the last iput then link
3260 * the inode to the delayed iput machinery. Delayed iputs are processed at
3261 * transaction commit time/superblock commit/cleaner kthread.
3263 void btrfs_add_delayed_iput(struct inode *inode)
3265 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3266 struct btrfs_inode *binode = BTRFS_I(inode);
3268 if (atomic_add_unless(&inode->i_count, -1, 1))
3271 spin_lock(&fs_info->delayed_iput_lock);
3272 ASSERT(list_empty(&binode->delayed_iput));
3273 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3274 spin_unlock(&fs_info->delayed_iput_lock);
3277 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3280 spin_lock(&fs_info->delayed_iput_lock);
3281 while (!list_empty(&fs_info->delayed_iputs)) {
3282 struct btrfs_inode *inode;
3284 inode = list_first_entry(&fs_info->delayed_iputs,
3285 struct btrfs_inode, delayed_iput);
3286 list_del_init(&inode->delayed_iput);
3287 spin_unlock(&fs_info->delayed_iput_lock);
3288 iput(&inode->vfs_inode);
3289 spin_lock(&fs_info->delayed_iput_lock);
3291 spin_unlock(&fs_info->delayed_iput_lock);
3295 * This creates an orphan entry for the given inode in case something goes wrong
3296 * in the middle of an unlink.
3298 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3299 struct btrfs_inode *inode)
3303 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3304 if (ret && ret != -EEXIST) {
3305 btrfs_abort_transaction(trans, ret);
3313 * We have done the delete so we can go ahead and remove the orphan item for
3314 * this particular inode.
3316 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3317 struct btrfs_inode *inode)
3319 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3323 * this cleans up any orphans that may be left on the list from the last use
3326 int btrfs_orphan_cleanup(struct btrfs_root *root)
3328 struct btrfs_fs_info *fs_info = root->fs_info;
3329 struct btrfs_path *path;
3330 struct extent_buffer *leaf;
3331 struct btrfs_key key, found_key;
3332 struct btrfs_trans_handle *trans;
3333 struct inode *inode;
3334 u64 last_objectid = 0;
3335 int ret = 0, nr_unlink = 0;
3337 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3340 path = btrfs_alloc_path();
3345 path->reada = READA_BACK;
3347 key.objectid = BTRFS_ORPHAN_OBJECTID;
3348 key.type = BTRFS_ORPHAN_ITEM_KEY;
3349 key.offset = (u64)-1;
3352 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3357 * if ret == 0 means we found what we were searching for, which
3358 * is weird, but possible, so only screw with path if we didn't
3359 * find the key and see if we have stuff that matches
3363 if (path->slots[0] == 0)
3368 /* pull out the item */
3369 leaf = path->nodes[0];
3370 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3372 /* make sure the item matches what we want */
3373 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3375 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3378 /* release the path since we're done with it */
3379 btrfs_release_path(path);
3382 * this is where we are basically btrfs_lookup, without the
3383 * crossing root thing. we store the inode number in the
3384 * offset of the orphan item.
3387 if (found_key.offset == last_objectid) {
3389 "Error removing orphan entry, stopping orphan cleanup");
3394 last_objectid = found_key.offset;
3396 found_key.objectid = found_key.offset;
3397 found_key.type = BTRFS_INODE_ITEM_KEY;
3398 found_key.offset = 0;
3399 inode = btrfs_iget(fs_info->sb, &found_key, root, NULL);
3400 ret = PTR_ERR_OR_ZERO(inode);
3401 if (ret && ret != -ENOENT)
3404 if (ret == -ENOENT && root == fs_info->tree_root) {
3405 struct btrfs_root *dead_root;
3406 struct btrfs_fs_info *fs_info = root->fs_info;
3407 int is_dead_root = 0;
3410 * this is an orphan in the tree root. Currently these
3411 * could come from 2 sources:
3412 * a) a snapshot deletion in progress
3413 * b) a free space cache inode
3414 * We need to distinguish those two, as the snapshot
3415 * orphan must not get deleted.
3416 * find_dead_roots already ran before us, so if this
3417 * is a snapshot deletion, we should find the root
3418 * in the dead_roots list
3420 spin_lock(&fs_info->trans_lock);
3421 list_for_each_entry(dead_root, &fs_info->dead_roots,
3423 if (dead_root->root_key.objectid ==
3424 found_key.objectid) {
3429 spin_unlock(&fs_info->trans_lock);
3431 /* prevent this orphan from being found again */
3432 key.offset = found_key.objectid - 1;
3439 * If we have an inode with links, there are a couple of
3440 * possibilities. Old kernels (before v3.12) used to create an
3441 * orphan item for truncate indicating that there were possibly
3442 * extent items past i_size that needed to be deleted. In v3.12,
3443 * truncate was changed to update i_size in sync with the extent
3444 * items, but the (useless) orphan item was still created. Since
3445 * v4.18, we don't create the orphan item for truncate at all.
3447 * So, this item could mean that we need to do a truncate, but
3448 * only if this filesystem was last used on a pre-v3.12 kernel
3449 * and was not cleanly unmounted. The odds of that are quite
3450 * slim, and it's a pain to do the truncate now, so just delete
3453 * It's also possible that this orphan item was supposed to be
3454 * deleted but wasn't. The inode number may have been reused,
3455 * but either way, we can delete the orphan item.
3457 if (ret == -ENOENT || inode->i_nlink) {
3460 trans = btrfs_start_transaction(root, 1);
3461 if (IS_ERR(trans)) {
3462 ret = PTR_ERR(trans);
3465 btrfs_debug(fs_info, "auto deleting %Lu",
3466 found_key.objectid);
3467 ret = btrfs_del_orphan_item(trans, root,
3468 found_key.objectid);
3469 btrfs_end_transaction(trans);
3477 /* this will do delete_inode and everything for us */
3480 /* release the path since we're done with it */
3481 btrfs_release_path(path);
3483 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3485 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3486 trans = btrfs_join_transaction(root);
3488 btrfs_end_transaction(trans);
3492 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3496 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3497 btrfs_free_path(path);
3502 * very simple check to peek ahead in the leaf looking for xattrs. If we
3503 * don't find any xattrs, we know there can't be any acls.
3505 * slot is the slot the inode is in, objectid is the objectid of the inode
3507 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3508 int slot, u64 objectid,
3509 int *first_xattr_slot)
3511 u32 nritems = btrfs_header_nritems(leaf);
3512 struct btrfs_key found_key;
3513 static u64 xattr_access = 0;
3514 static u64 xattr_default = 0;
3517 if (!xattr_access) {
3518 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3519 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3520 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3521 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3525 *first_xattr_slot = -1;
3526 while (slot < nritems) {
3527 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3529 /* we found a different objectid, there must not be acls */
3530 if (found_key.objectid != objectid)
3533 /* we found an xattr, assume we've got an acl */
3534 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3535 if (*first_xattr_slot == -1)
3536 *first_xattr_slot = slot;
3537 if (found_key.offset == xattr_access ||
3538 found_key.offset == xattr_default)
3543 * we found a key greater than an xattr key, there can't
3544 * be any acls later on
3546 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3553 * it goes inode, inode backrefs, xattrs, extents,
3554 * so if there are a ton of hard links to an inode there can
3555 * be a lot of backrefs. Don't waste time searching too hard,
3556 * this is just an optimization
3561 /* we hit the end of the leaf before we found an xattr or
3562 * something larger than an xattr. We have to assume the inode
3565 if (*first_xattr_slot == -1)
3566 *first_xattr_slot = slot;
3571 * read an inode from the btree into the in-memory inode
3573 static int btrfs_read_locked_inode(struct inode *inode)
3575 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3576 struct btrfs_path *path;
3577 struct extent_buffer *leaf;
3578 struct btrfs_inode_item *inode_item;
3579 struct btrfs_root *root = BTRFS_I(inode)->root;
3580 struct btrfs_key location;
3585 bool filled = false;
3586 int first_xattr_slot;
3588 ret = btrfs_fill_inode(inode, &rdev);
3592 path = btrfs_alloc_path();
3596 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3598 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3600 btrfs_free_path(path);
3604 leaf = path->nodes[0];
3609 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3610 struct btrfs_inode_item);
3611 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3612 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3613 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3614 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3615 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3617 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3618 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3620 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3621 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3623 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3624 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3626 BTRFS_I(inode)->i_otime.tv_sec =
3627 btrfs_timespec_sec(leaf, &inode_item->otime);
3628 BTRFS_I(inode)->i_otime.tv_nsec =
3629 btrfs_timespec_nsec(leaf, &inode_item->otime);
3631 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3632 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3633 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3635 inode_set_iversion_queried(inode,
3636 btrfs_inode_sequence(leaf, inode_item));
3637 inode->i_generation = BTRFS_I(inode)->generation;
3639 rdev = btrfs_inode_rdev(leaf, inode_item);
3641 BTRFS_I(inode)->index_cnt = (u64)-1;
3642 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3646 * If we were modified in the current generation and evicted from memory
3647 * and then re-read we need to do a full sync since we don't have any
3648 * idea about which extents were modified before we were evicted from
3651 * This is required for both inode re-read from disk and delayed inode
3652 * in delayed_nodes_tree.
3654 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3655 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3656 &BTRFS_I(inode)->runtime_flags);
3659 * We don't persist the id of the transaction where an unlink operation
3660 * against the inode was last made. So here we assume the inode might
3661 * have been evicted, and therefore the exact value of last_unlink_trans
3662 * lost, and set it to last_trans to avoid metadata inconsistencies
3663 * between the inode and its parent if the inode is fsync'ed and the log
3664 * replayed. For example, in the scenario:
3667 * ln mydir/foo mydir/bar
3670 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3671 * xfs_io -c fsync mydir/foo
3673 * mount fs, triggers fsync log replay
3675 * We must make sure that when we fsync our inode foo we also log its
3676 * parent inode, otherwise after log replay the parent still has the
3677 * dentry with the "bar" name but our inode foo has a link count of 1
3678 * and doesn't have an inode ref with the name "bar" anymore.
3680 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3681 * but it guarantees correctness at the expense of occasional full
3682 * transaction commits on fsync if our inode is a directory, or if our
3683 * inode is not a directory, logging its parent unnecessarily.
3685 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3688 if (inode->i_nlink != 1 ||
3689 path->slots[0] >= btrfs_header_nritems(leaf))
3692 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3693 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3696 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3697 if (location.type == BTRFS_INODE_REF_KEY) {
3698 struct btrfs_inode_ref *ref;
3700 ref = (struct btrfs_inode_ref *)ptr;
3701 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3702 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3703 struct btrfs_inode_extref *extref;
3705 extref = (struct btrfs_inode_extref *)ptr;
3706 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3711 * try to precache a NULL acl entry for files that don't have
3712 * any xattrs or acls
3714 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3715 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3716 if (first_xattr_slot != -1) {
3717 path->slots[0] = first_xattr_slot;
3718 ret = btrfs_load_inode_props(inode, path);
3721 "error loading props for ino %llu (root %llu): %d",
3722 btrfs_ino(BTRFS_I(inode)),
3723 root->root_key.objectid, ret);
3725 btrfs_free_path(path);
3728 cache_no_acl(inode);
3730 switch (inode->i_mode & S_IFMT) {
3732 inode->i_mapping->a_ops = &btrfs_aops;
3733 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3734 inode->i_fop = &btrfs_file_operations;
3735 inode->i_op = &btrfs_file_inode_operations;
3738 inode->i_fop = &btrfs_dir_file_operations;
3739 inode->i_op = &btrfs_dir_inode_operations;
3742 inode->i_op = &btrfs_symlink_inode_operations;
3743 inode_nohighmem(inode);
3744 inode->i_mapping->a_ops = &btrfs_aops;
3747 inode->i_op = &btrfs_special_inode_operations;
3748 init_special_inode(inode, inode->i_mode, rdev);
3752 btrfs_sync_inode_flags_to_i_flags(inode);
3757 * given a leaf and an inode, copy the inode fields into the leaf
3759 static void fill_inode_item(struct btrfs_trans_handle *trans,
3760 struct extent_buffer *leaf,
3761 struct btrfs_inode_item *item,
3762 struct inode *inode)
3764 struct btrfs_map_token token;
3766 btrfs_init_map_token(&token);
3768 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3769 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3770 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3772 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3773 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3775 btrfs_set_token_timespec_sec(leaf, &item->atime,
3776 inode->i_atime.tv_sec, &token);
3777 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3778 inode->i_atime.tv_nsec, &token);
3780 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3781 inode->i_mtime.tv_sec, &token);
3782 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3783 inode->i_mtime.tv_nsec, &token);
3785 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3786 inode->i_ctime.tv_sec, &token);
3787 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3788 inode->i_ctime.tv_nsec, &token);
3790 btrfs_set_token_timespec_sec(leaf, &item->otime,
3791 BTRFS_I(inode)->i_otime.tv_sec, &token);
3792 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3793 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3795 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3797 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3799 btrfs_set_token_inode_sequence(leaf, item, inode_peek_iversion(inode),
3801 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3802 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3803 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3804 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3808 * copy everything in the in-memory inode into the btree.
3810 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3811 struct btrfs_root *root, struct inode *inode)
3813 struct btrfs_inode_item *inode_item;
3814 struct btrfs_path *path;
3815 struct extent_buffer *leaf;
3818 path = btrfs_alloc_path();
3822 path->leave_spinning = 1;
3823 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3831 leaf = path->nodes[0];
3832 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3833 struct btrfs_inode_item);
3835 fill_inode_item(trans, leaf, inode_item, inode);
3836 btrfs_mark_buffer_dirty(leaf);
3837 btrfs_set_inode_last_trans(trans, inode);
3840 btrfs_free_path(path);
3845 * copy everything in the in-memory inode into the btree.
3847 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3848 struct btrfs_root *root, struct inode *inode)
3850 struct btrfs_fs_info *fs_info = root->fs_info;
3854 * If the inode is a free space inode, we can deadlock during commit
3855 * if we put it into the delayed code.
3857 * The data relocation inode should also be directly updated
3860 if (!btrfs_is_free_space_inode(BTRFS_I(inode))
3861 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3862 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
3863 btrfs_update_root_times(trans, root);
3865 ret = btrfs_delayed_update_inode(trans, root, inode);
3867 btrfs_set_inode_last_trans(trans, inode);
3871 return btrfs_update_inode_item(trans, root, inode);
3874 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3875 struct btrfs_root *root,
3876 struct inode *inode)
3880 ret = btrfs_update_inode(trans, root, inode);
3882 return btrfs_update_inode_item(trans, root, inode);
3887 * unlink helper that gets used here in inode.c and in the tree logging
3888 * recovery code. It remove a link in a directory with a given name, and
3889 * also drops the back refs in the inode to the directory
3891 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3892 struct btrfs_root *root,
3893 struct btrfs_inode *dir,
3894 struct btrfs_inode *inode,
3895 const char *name, int name_len)
3897 struct btrfs_fs_info *fs_info = root->fs_info;
3898 struct btrfs_path *path;
3900 struct extent_buffer *leaf;
3901 struct btrfs_dir_item *di;
3902 struct btrfs_key key;
3904 u64 ino = btrfs_ino(inode);
3905 u64 dir_ino = btrfs_ino(dir);
3907 path = btrfs_alloc_path();
3913 path->leave_spinning = 1;
3914 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3915 name, name_len, -1);
3916 if (IS_ERR_OR_NULL(di)) {
3917 ret = di ? PTR_ERR(di) : -ENOENT;
3920 leaf = path->nodes[0];
3921 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3922 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3925 btrfs_release_path(path);
3928 * If we don't have dir index, we have to get it by looking up
3929 * the inode ref, since we get the inode ref, remove it directly,
3930 * it is unnecessary to do delayed deletion.
3932 * But if we have dir index, needn't search inode ref to get it.
3933 * Since the inode ref is close to the inode item, it is better
3934 * that we delay to delete it, and just do this deletion when
3935 * we update the inode item.
3937 if (inode->dir_index) {
3938 ret = btrfs_delayed_delete_inode_ref(inode);
3940 index = inode->dir_index;
3945 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3949 "failed to delete reference to %.*s, inode %llu parent %llu",
3950 name_len, name, ino, dir_ino);
3951 btrfs_abort_transaction(trans, ret);
3955 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
3957 btrfs_abort_transaction(trans, ret);
3961 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
3963 if (ret != 0 && ret != -ENOENT) {
3964 btrfs_abort_transaction(trans, ret);
3968 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
3973 btrfs_abort_transaction(trans, ret);
3975 btrfs_free_path(path);
3979 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
3980 inode_inc_iversion(&inode->vfs_inode);
3981 inode_inc_iversion(&dir->vfs_inode);
3982 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
3983 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
3984 ret = btrfs_update_inode(trans, root, &dir->vfs_inode);
3989 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3990 struct btrfs_root *root,
3991 struct btrfs_inode *dir, struct btrfs_inode *inode,
3992 const char *name, int name_len)
3995 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3997 drop_nlink(&inode->vfs_inode);
3998 ret = btrfs_update_inode(trans, root, &inode->vfs_inode);
4004 * helper to start transaction for unlink and rmdir.
4006 * unlink and rmdir are special in btrfs, they do not always free space, so
4007 * if we cannot make our reservations the normal way try and see if there is
4008 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4009 * allow the unlink to occur.
4011 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4013 struct btrfs_root *root = BTRFS_I(dir)->root;
4016 * 1 for the possible orphan item
4017 * 1 for the dir item
4018 * 1 for the dir index
4019 * 1 for the inode ref
4022 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4025 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4027 struct btrfs_root *root = BTRFS_I(dir)->root;
4028 struct btrfs_trans_handle *trans;
4029 struct inode *inode = d_inode(dentry);
4032 trans = __unlink_start_trans(dir);
4034 return PTR_ERR(trans);
4036 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4039 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4040 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4041 dentry->d_name.len);
4045 if (inode->i_nlink == 0) {
4046 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4052 btrfs_end_transaction(trans);
4053 btrfs_btree_balance_dirty(root->fs_info);
4057 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4058 struct inode *dir, u64 objectid,
4059 const char *name, int name_len)
4061 struct btrfs_root *root = BTRFS_I(dir)->root;
4062 struct btrfs_path *path;
4063 struct extent_buffer *leaf;
4064 struct btrfs_dir_item *di;
4065 struct btrfs_key key;
4068 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4070 path = btrfs_alloc_path();
4074 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4075 name, name_len, -1);
4076 if (IS_ERR_OR_NULL(di)) {
4077 ret = di ? PTR_ERR(di) : -ENOENT;
4081 leaf = path->nodes[0];
4082 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4083 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4084 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4086 btrfs_abort_transaction(trans, ret);
4089 btrfs_release_path(path);
4091 ret = btrfs_del_root_ref(trans, objectid, root->root_key.objectid,
4092 dir_ino, &index, name, name_len);
4094 if (ret != -ENOENT) {
4095 btrfs_abort_transaction(trans, ret);
4098 di = btrfs_search_dir_index_item(root, path, dir_ino,
4100 if (IS_ERR_OR_NULL(di)) {
4105 btrfs_abort_transaction(trans, ret);
4109 leaf = path->nodes[0];
4110 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4113 btrfs_release_path(path);
4115 ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index);
4117 btrfs_abort_transaction(trans, ret);
4121 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4122 inode_inc_iversion(dir);
4123 dir->i_mtime = dir->i_ctime = current_time(dir);
4124 ret = btrfs_update_inode_fallback(trans, root, dir);
4126 btrfs_abort_transaction(trans, ret);
4128 btrfs_free_path(path);
4133 * Helper to check if the subvolume references other subvolumes or if it's
4136 static noinline int may_destroy_subvol(struct btrfs_root *root)
4138 struct btrfs_fs_info *fs_info = root->fs_info;
4139 struct btrfs_path *path;
4140 struct btrfs_dir_item *di;
4141 struct btrfs_key key;
4145 path = btrfs_alloc_path();
4149 /* Make sure this root isn't set as the default subvol */
4150 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4151 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4152 dir_id, "default", 7, 0);
4153 if (di && !IS_ERR(di)) {
4154 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4155 if (key.objectid == root->root_key.objectid) {
4158 "deleting default subvolume %llu is not allowed",
4162 btrfs_release_path(path);
4165 key.objectid = root->root_key.objectid;
4166 key.type = BTRFS_ROOT_REF_KEY;
4167 key.offset = (u64)-1;
4169 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4175 if (path->slots[0] > 0) {
4177 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4178 if (key.objectid == root->root_key.objectid &&
4179 key.type == BTRFS_ROOT_REF_KEY)
4183 btrfs_free_path(path);
4187 /* Delete all dentries for inodes belonging to the root */
4188 static void btrfs_prune_dentries(struct btrfs_root *root)
4190 struct btrfs_fs_info *fs_info = root->fs_info;
4191 struct rb_node *node;
4192 struct rb_node *prev;
4193 struct btrfs_inode *entry;
4194 struct inode *inode;
4197 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
4198 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4200 spin_lock(&root->inode_lock);
4202 node = root->inode_tree.rb_node;
4206 entry = rb_entry(node, struct btrfs_inode, rb_node);
4208 if (objectid < btrfs_ino(entry))
4209 node = node->rb_left;
4210 else if (objectid > btrfs_ino(entry))
4211 node = node->rb_right;
4217 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4218 if (objectid <= btrfs_ino(entry)) {
4222 prev = rb_next(prev);
4226 entry = rb_entry(node, struct btrfs_inode, rb_node);
4227 objectid = btrfs_ino(entry) + 1;
4228 inode = igrab(&entry->vfs_inode);
4230 spin_unlock(&root->inode_lock);
4231 if (atomic_read(&inode->i_count) > 1)
4232 d_prune_aliases(inode);
4234 * btrfs_drop_inode will have it removed from the inode
4235 * cache when its usage count hits zero.
4239 spin_lock(&root->inode_lock);
4243 if (cond_resched_lock(&root->inode_lock))
4246 node = rb_next(node);
4248 spin_unlock(&root->inode_lock);
4251 int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
4253 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4254 struct btrfs_root *root = BTRFS_I(dir)->root;
4255 struct inode *inode = d_inode(dentry);
4256 struct btrfs_root *dest = BTRFS_I(inode)->root;
4257 struct btrfs_trans_handle *trans;
4258 struct btrfs_block_rsv block_rsv;
4264 * Don't allow to delete a subvolume with send in progress. This is
4265 * inside the inode lock so the error handling that has to drop the bit
4266 * again is not run concurrently.
4268 spin_lock(&dest->root_item_lock);
4269 if (dest->send_in_progress) {
4270 spin_unlock(&dest->root_item_lock);
4272 "attempt to delete subvolume %llu during send",
4273 dest->root_key.objectid);
4276 root_flags = btrfs_root_flags(&dest->root_item);
4277 btrfs_set_root_flags(&dest->root_item,
4278 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4279 spin_unlock(&dest->root_item_lock);
4281 down_write(&fs_info->subvol_sem);
4283 err = may_destroy_subvol(dest);
4287 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4289 * One for dir inode,
4290 * two for dir entries,
4291 * two for root ref/backref.
4293 err = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4297 trans = btrfs_start_transaction(root, 0);
4298 if (IS_ERR(trans)) {
4299 err = PTR_ERR(trans);
4302 trans->block_rsv = &block_rsv;
4303 trans->bytes_reserved = block_rsv.size;
4305 btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
4307 ret = btrfs_unlink_subvol(trans, dir, dest->root_key.objectid,
4308 dentry->d_name.name, dentry->d_name.len);
4311 btrfs_abort_transaction(trans, ret);
4315 btrfs_record_root_in_trans(trans, dest);
4317 memset(&dest->root_item.drop_progress, 0,
4318 sizeof(dest->root_item.drop_progress));
4319 dest->root_item.drop_level = 0;
4320 btrfs_set_root_refs(&dest->root_item, 0);
4322 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4323 ret = btrfs_insert_orphan_item(trans,
4325 dest->root_key.objectid);
4327 btrfs_abort_transaction(trans, ret);
4333 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4334 BTRFS_UUID_KEY_SUBVOL,
4335 dest->root_key.objectid);
4336 if (ret && ret != -ENOENT) {
4337 btrfs_abort_transaction(trans, ret);
4341 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4342 ret = btrfs_uuid_tree_remove(trans,
4343 dest->root_item.received_uuid,
4344 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4345 dest->root_key.objectid);
4346 if (ret && ret != -ENOENT) {
4347 btrfs_abort_transaction(trans, ret);
4354 trans->block_rsv = NULL;
4355 trans->bytes_reserved = 0;
4356 ret = btrfs_end_transaction(trans);
4359 inode->i_flags |= S_DEAD;
4361 btrfs_subvolume_release_metadata(fs_info, &block_rsv);
4363 up_write(&fs_info->subvol_sem);
4365 spin_lock(&dest->root_item_lock);
4366 root_flags = btrfs_root_flags(&dest->root_item);
4367 btrfs_set_root_flags(&dest->root_item,
4368 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4369 spin_unlock(&dest->root_item_lock);
4371 d_invalidate(dentry);
4372 btrfs_prune_dentries(dest);
4373 ASSERT(dest->send_in_progress == 0);
4376 if (dest->ino_cache_inode) {
4377 iput(dest->ino_cache_inode);
4378 dest->ino_cache_inode = NULL;
4385 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4387 struct inode *inode = d_inode(dentry);
4389 struct btrfs_root *root = BTRFS_I(dir)->root;
4390 struct btrfs_trans_handle *trans;
4391 u64 last_unlink_trans;
4393 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4395 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4396 return btrfs_delete_subvolume(dir, dentry);
4398 trans = __unlink_start_trans(dir);
4400 return PTR_ERR(trans);
4402 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4403 err = btrfs_unlink_subvol(trans, dir,
4404 BTRFS_I(inode)->location.objectid,
4405 dentry->d_name.name,
4406 dentry->d_name.len);
4410 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4414 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4416 /* now the directory is empty */
4417 err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4418 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4419 dentry->d_name.len);
4421 btrfs_i_size_write(BTRFS_I(inode), 0);
4423 * Propagate the last_unlink_trans value of the deleted dir to
4424 * its parent directory. This is to prevent an unrecoverable
4425 * log tree in the case we do something like this:
4427 * 2) create snapshot under dir foo
4428 * 3) delete the snapshot
4431 * 6) fsync foo or some file inside foo
4433 if (last_unlink_trans >= trans->transid)
4434 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4437 btrfs_end_transaction(trans);
4438 btrfs_btree_balance_dirty(root->fs_info);
4443 static int truncate_space_check(struct btrfs_trans_handle *trans,
4444 struct btrfs_root *root,
4447 struct btrfs_fs_info *fs_info = root->fs_info;
4451 * This is only used to apply pressure to the enospc system, we don't
4452 * intend to use this reservation at all.
4454 bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted);
4455 bytes_deleted *= fs_info->nodesize;
4456 ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
4457 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4459 trace_btrfs_space_reservation(fs_info, "transaction",
4462 trans->bytes_reserved += bytes_deleted;
4469 * Return this if we need to call truncate_block for the last bit of the
4472 #define NEED_TRUNCATE_BLOCK 1
4475 * this can truncate away extent items, csum items and directory items.
4476 * It starts at a high offset and removes keys until it can't find
4477 * any higher than new_size
4479 * csum items that cross the new i_size are truncated to the new size
4482 * min_type is the minimum key type to truncate down to. If set to 0, this
4483 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4485 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4486 struct btrfs_root *root,
4487 struct inode *inode,
4488 u64 new_size, u32 min_type)
4490 struct btrfs_fs_info *fs_info = root->fs_info;
4491 struct btrfs_path *path;
4492 struct extent_buffer *leaf;
4493 struct btrfs_file_extent_item *fi;
4494 struct btrfs_key key;
4495 struct btrfs_key found_key;
4496 u64 extent_start = 0;
4497 u64 extent_num_bytes = 0;
4498 u64 extent_offset = 0;
4500 u64 last_size = new_size;
4501 u32 found_type = (u8)-1;
4504 int pending_del_nr = 0;
4505 int pending_del_slot = 0;
4506 int extent_type = -1;
4508 u64 ino = btrfs_ino(BTRFS_I(inode));
4509 u64 bytes_deleted = 0;
4510 bool be_nice = false;
4511 bool should_throttle = false;
4512 bool should_end = false;
4514 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4517 * for non-free space inodes and ref cows, we want to back off from
4520 if (!btrfs_is_free_space_inode(BTRFS_I(inode)) &&
4521 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4524 path = btrfs_alloc_path();
4527 path->reada = READA_BACK;
4530 * We want to drop from the next block forward in case this new size is
4531 * not block aligned since we will be keeping the last block of the
4532 * extent just the way it is.
4534 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4535 root == fs_info->tree_root)
4536 btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size,
4537 fs_info->sectorsize),
4541 * This function is also used to drop the items in the log tree before
4542 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4543 * it is used to drop the loged items. So we shouldn't kill the delayed
4546 if (min_type == 0 && root == BTRFS_I(inode)->root)
4547 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
4550 key.offset = (u64)-1;
4555 * with a 16K leaf size and 128MB extents, you can actually queue
4556 * up a huge file in a single leaf. Most of the time that
4557 * bytes_deleted is > 0, it will be huge by the time we get here
4559 if (be_nice && bytes_deleted > SZ_32M &&
4560 btrfs_should_end_transaction(trans)) {
4565 path->leave_spinning = 1;
4566 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4572 /* there are no items in the tree for us to truncate, we're
4575 if (path->slots[0] == 0)
4582 leaf = path->nodes[0];
4583 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4584 found_type = found_key.type;
4586 if (found_key.objectid != ino)
4589 if (found_type < min_type)
4592 item_end = found_key.offset;
4593 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4594 fi = btrfs_item_ptr(leaf, path->slots[0],
4595 struct btrfs_file_extent_item);
4596 extent_type = btrfs_file_extent_type(leaf, fi);
4597 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4599 btrfs_file_extent_num_bytes(leaf, fi);
4601 trace_btrfs_truncate_show_fi_regular(
4602 BTRFS_I(inode), leaf, fi,
4604 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4605 item_end += btrfs_file_extent_ram_bytes(leaf,
4608 trace_btrfs_truncate_show_fi_inline(
4609 BTRFS_I(inode), leaf, fi, path->slots[0],
4614 if (found_type > min_type) {
4617 if (item_end < new_size)
4619 if (found_key.offset >= new_size)
4625 /* FIXME, shrink the extent if the ref count is only 1 */
4626 if (found_type != BTRFS_EXTENT_DATA_KEY)
4629 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4631 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4633 u64 orig_num_bytes =
4634 btrfs_file_extent_num_bytes(leaf, fi);
4635 extent_num_bytes = ALIGN(new_size -
4637 fs_info->sectorsize);
4638 btrfs_set_file_extent_num_bytes(leaf, fi,
4640 num_dec = (orig_num_bytes -
4642 if (test_bit(BTRFS_ROOT_REF_COWS,
4645 inode_sub_bytes(inode, num_dec);
4646 btrfs_mark_buffer_dirty(leaf);
4649 btrfs_file_extent_disk_num_bytes(leaf,
4651 extent_offset = found_key.offset -
4652 btrfs_file_extent_offset(leaf, fi);
4654 /* FIXME blocksize != 4096 */
4655 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4656 if (extent_start != 0) {
4658 if (test_bit(BTRFS_ROOT_REF_COWS,
4660 inode_sub_bytes(inode, num_dec);
4663 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4665 * we can't truncate inline items that have had
4669 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4670 btrfs_file_extent_other_encoding(leaf, fi) == 0 &&
4671 btrfs_file_extent_compression(leaf, fi) == 0) {
4672 u32 size = (u32)(new_size - found_key.offset);
4674 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4675 size = btrfs_file_extent_calc_inline_size(size);
4676 btrfs_truncate_item(root->fs_info, path, size, 1);
4677 } else if (!del_item) {
4679 * We have to bail so the last_size is set to
4680 * just before this extent.
4682 ret = NEED_TRUNCATE_BLOCK;
4686 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4687 inode_sub_bytes(inode, item_end + 1 - new_size);
4691 last_size = found_key.offset;
4693 last_size = new_size;
4695 if (!pending_del_nr) {
4696 /* no pending yet, add ourselves */
4697 pending_del_slot = path->slots[0];
4699 } else if (pending_del_nr &&
4700 path->slots[0] + 1 == pending_del_slot) {
4701 /* hop on the pending chunk */
4703 pending_del_slot = path->slots[0];
4710 should_throttle = false;
4713 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4714 root == fs_info->tree_root)) {
4715 btrfs_set_path_blocking(path);
4716 bytes_deleted += extent_num_bytes;
4717 ret = btrfs_free_extent(trans, root, extent_start,
4718 extent_num_bytes, 0,
4719 btrfs_header_owner(leaf),
4720 ino, extent_offset);
4722 btrfs_abort_transaction(trans, ret);
4725 if (btrfs_should_throttle_delayed_refs(trans))
4726 btrfs_async_run_delayed_refs(fs_info,
4727 trans->delayed_ref_updates * 2,
4730 if (truncate_space_check(trans, root,
4731 extent_num_bytes)) {
4734 if (btrfs_should_throttle_delayed_refs(trans))
4735 should_throttle = true;
4739 if (found_type == BTRFS_INODE_ITEM_KEY)
4742 if (path->slots[0] == 0 ||
4743 path->slots[0] != pending_del_slot ||
4744 should_throttle || should_end) {
4745 if (pending_del_nr) {
4746 ret = btrfs_del_items(trans, root, path,
4750 btrfs_abort_transaction(trans, ret);
4755 btrfs_release_path(path);
4756 if (should_throttle) {
4757 unsigned long updates = trans->delayed_ref_updates;
4759 trans->delayed_ref_updates = 0;
4760 ret = btrfs_run_delayed_refs(trans,
4767 * if we failed to refill our space rsv, bail out
4768 * and let the transaction restart
4780 if (ret >= 0 && pending_del_nr) {
4783 err = btrfs_del_items(trans, root, path, pending_del_slot,
4786 btrfs_abort_transaction(trans, err);
4790 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4791 ASSERT(last_size >= new_size);
4792 if (!ret && last_size > new_size)
4793 last_size = new_size;
4794 btrfs_ordered_update_i_size(inode, last_size, NULL);
4797 btrfs_free_path(path);
4799 if (be_nice && bytes_deleted > SZ_32M && (ret >= 0 || ret == -EAGAIN)) {
4800 unsigned long updates = trans->delayed_ref_updates;
4804 trans->delayed_ref_updates = 0;
4805 err = btrfs_run_delayed_refs(trans, updates * 2);
4814 * btrfs_truncate_block - read, zero a chunk and write a block
4815 * @inode - inode that we're zeroing
4816 * @from - the offset to start zeroing
4817 * @len - the length to zero, 0 to zero the entire range respective to the
4819 * @front - zero up to the offset instead of from the offset on
4821 * This will find the block for the "from" offset and cow the block and zero the
4822 * part we want to zero. This is used with truncate and hole punching.
4824 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4827 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4828 struct address_space *mapping = inode->i_mapping;
4829 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4830 struct btrfs_ordered_extent *ordered;
4831 struct extent_state *cached_state = NULL;
4832 struct extent_changeset *data_reserved = NULL;
4834 u32 blocksize = fs_info->sectorsize;
4835 pgoff_t index = from >> PAGE_SHIFT;
4836 unsigned offset = from & (blocksize - 1);
4838 gfp_t mask = btrfs_alloc_write_mask(mapping);
4843 if (IS_ALIGNED(offset, blocksize) &&
4844 (!len || IS_ALIGNED(len, blocksize)))
4847 block_start = round_down(from, blocksize);
4848 block_end = block_start + blocksize - 1;
4850 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
4851 block_start, blocksize);
4856 page = find_or_create_page(mapping, index, mask);
4858 btrfs_delalloc_release_space(inode, data_reserved,
4859 block_start, blocksize, true);
4860 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize, true);
4865 if (!PageUptodate(page)) {
4866 ret = btrfs_readpage(NULL, page);
4868 if (page->mapping != mapping) {
4873 if (!PageUptodate(page)) {
4878 wait_on_page_writeback(page);
4880 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4881 set_page_extent_mapped(page);
4883 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4885 unlock_extent_cached(io_tree, block_start, block_end,
4889 btrfs_start_ordered_extent(inode, ordered, 1);
4890 btrfs_put_ordered_extent(ordered);
4894 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4895 EXTENT_DIRTY | EXTENT_DELALLOC |
4896 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4897 0, 0, &cached_state);
4899 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4902 unlock_extent_cached(io_tree, block_start, block_end,
4907 if (offset != blocksize) {
4909 len = blocksize - offset;
4912 memset(kaddr + (block_start - page_offset(page)),
4915 memset(kaddr + (block_start - page_offset(page)) + offset,
4917 flush_dcache_page(page);
4920 ClearPageChecked(page);
4921 set_page_dirty(page);
4922 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4926 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4928 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize, (ret != 0));
4932 extent_changeset_free(data_reserved);
4936 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4937 u64 offset, u64 len)
4939 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4940 struct btrfs_trans_handle *trans;
4944 * Still need to make sure the inode looks like it's been updated so
4945 * that any holes get logged if we fsync.
4947 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4948 BTRFS_I(inode)->last_trans = fs_info->generation;
4949 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4950 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4955 * 1 - for the one we're dropping
4956 * 1 - for the one we're adding
4957 * 1 - for updating the inode.
4959 trans = btrfs_start_transaction(root, 3);
4961 return PTR_ERR(trans);
4963 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4965 btrfs_abort_transaction(trans, ret);
4966 btrfs_end_transaction(trans);
4970 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)),
4971 offset, 0, 0, len, 0, len, 0, 0, 0);
4973 btrfs_abort_transaction(trans, ret);
4975 btrfs_update_inode(trans, root, inode);
4976 btrfs_end_transaction(trans);
4981 * This function puts in dummy file extents for the area we're creating a hole
4982 * for. So if we are truncating this file to a larger size we need to insert
4983 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4984 * the range between oldsize and size
4986 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4988 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4989 struct btrfs_root *root = BTRFS_I(inode)->root;
4990 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4991 struct extent_map *em = NULL;
4992 struct extent_state *cached_state = NULL;
4993 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4994 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4995 u64 block_end = ALIGN(size, fs_info->sectorsize);
5002 * If our size started in the middle of a block we need to zero out the
5003 * rest of the block before we expand the i_size, otherwise we could
5004 * expose stale data.
5006 err = btrfs_truncate_block(inode, oldsize, 0, 0);
5010 if (size <= hole_start)
5014 struct btrfs_ordered_extent *ordered;
5016 lock_extent_bits(io_tree, hole_start, block_end - 1,
5018 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), hole_start,
5019 block_end - hole_start);
5022 unlock_extent_cached(io_tree, hole_start, block_end - 1,
5024 btrfs_start_ordered_extent(inode, ordered, 1);
5025 btrfs_put_ordered_extent(ordered);
5028 cur_offset = hole_start;
5030 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
5031 block_end - cur_offset, 0);
5037 last_byte = min(extent_map_end(em), block_end);
5038 last_byte = ALIGN(last_byte, fs_info->sectorsize);
5039 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5040 struct extent_map *hole_em;
5041 hole_size = last_byte - cur_offset;
5043 err = maybe_insert_hole(root, inode, cur_offset,
5047 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
5048 cur_offset + hole_size - 1, 0);
5049 hole_em = alloc_extent_map();
5051 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5052 &BTRFS_I(inode)->runtime_flags);
5055 hole_em->start = cur_offset;
5056 hole_em->len = hole_size;
5057 hole_em->orig_start = cur_offset;
5059 hole_em->block_start = EXTENT_MAP_HOLE;
5060 hole_em->block_len = 0;
5061 hole_em->orig_block_len = 0;
5062 hole_em->ram_bytes = hole_size;
5063 hole_em->bdev = fs_info->fs_devices->latest_bdev;
5064 hole_em->compress_type = BTRFS_COMPRESS_NONE;
5065 hole_em->generation = fs_info->generation;
5068 write_lock(&em_tree->lock);
5069 err = add_extent_mapping(em_tree, hole_em, 1);
5070 write_unlock(&em_tree->lock);
5073 btrfs_drop_extent_cache(BTRFS_I(inode),
5078 free_extent_map(hole_em);
5081 free_extent_map(em);
5083 cur_offset = last_byte;
5084 if (cur_offset >= block_end)
5087 free_extent_map(em);
5088 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
5092 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5094 struct btrfs_root *root = BTRFS_I(inode)->root;
5095 struct btrfs_trans_handle *trans;
5096 loff_t oldsize = i_size_read(inode);
5097 loff_t newsize = attr->ia_size;
5098 int mask = attr->ia_valid;
5102 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5103 * special case where we need to update the times despite not having
5104 * these flags set. For all other operations the VFS set these flags
5105 * explicitly if it wants a timestamp update.
5107 if (newsize != oldsize) {
5108 inode_inc_iversion(inode);
5109 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5110 inode->i_ctime = inode->i_mtime =
5111 current_time(inode);
5114 if (newsize > oldsize) {
5116 * Don't do an expanding truncate while snapshotting is ongoing.
5117 * This is to ensure the snapshot captures a fully consistent
5118 * state of this file - if the snapshot captures this expanding
5119 * truncation, it must capture all writes that happened before
5122 btrfs_wait_for_snapshot_creation(root);
5123 ret = btrfs_cont_expand(inode, oldsize, newsize);
5125 btrfs_end_write_no_snapshotting(root);
5129 trans = btrfs_start_transaction(root, 1);
5130 if (IS_ERR(trans)) {
5131 btrfs_end_write_no_snapshotting(root);
5132 return PTR_ERR(trans);
5135 i_size_write(inode, newsize);
5136 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
5137 pagecache_isize_extended(inode, oldsize, newsize);
5138 ret = btrfs_update_inode(trans, root, inode);
5139 btrfs_end_write_no_snapshotting(root);
5140 btrfs_end_transaction(trans);
5144 * We're truncating a file that used to have good data down to
5145 * zero. Make sure it gets into the ordered flush list so that
5146 * any new writes get down to disk quickly.
5149 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
5150 &BTRFS_I(inode)->runtime_flags);
5152 truncate_setsize(inode, newsize);
5154 /* Disable nonlocked read DIO to avoid the end less truncate */
5155 btrfs_inode_block_unlocked_dio(BTRFS_I(inode));
5156 inode_dio_wait(inode);
5157 btrfs_inode_resume_unlocked_dio(BTRFS_I(inode));
5159 ret = btrfs_truncate(inode, newsize == oldsize);
5160 if (ret && inode->i_nlink) {
5164 * Truncate failed, so fix up the in-memory size. We
5165 * adjusted disk_i_size down as we removed extents, so
5166 * wait for disk_i_size to be stable and then update the
5167 * in-memory size to match.
5169 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5172 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5179 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5181 struct inode *inode = d_inode(dentry);
5182 struct btrfs_root *root = BTRFS_I(inode)->root;
5185 if (btrfs_root_readonly(root))
5188 err = setattr_prepare(dentry, attr);
5192 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5193 err = btrfs_setsize(inode, attr);
5198 if (attr->ia_valid) {
5199 setattr_copy(inode, attr);
5200 inode_inc_iversion(inode);
5201 err = btrfs_dirty_inode(inode);
5203 if (!err && attr->ia_valid & ATTR_MODE)
5204 err = posix_acl_chmod(inode, inode->i_mode);
5211 * While truncating the inode pages during eviction, we get the VFS calling
5212 * btrfs_invalidatepage() against each page of the inode. This is slow because
5213 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5214 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5215 * extent_state structures over and over, wasting lots of time.
5217 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5218 * those expensive operations on a per page basis and do only the ordered io
5219 * finishing, while we release here the extent_map and extent_state structures,
5220 * without the excessive merging and splitting.
5222 static void evict_inode_truncate_pages(struct inode *inode)
5224 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5225 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5226 struct rb_node *node;
5228 ASSERT(inode->i_state & I_FREEING);
5229 truncate_inode_pages_final(&inode->i_data);
5231 write_lock(&map_tree->lock);
5232 while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) {
5233 struct extent_map *em;
5235 node = rb_first_cached(&map_tree->map);
5236 em = rb_entry(node, struct extent_map, rb_node);
5237 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5238 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5239 remove_extent_mapping(map_tree, em);
5240 free_extent_map(em);
5241 if (need_resched()) {
5242 write_unlock(&map_tree->lock);
5244 write_lock(&map_tree->lock);
5247 write_unlock(&map_tree->lock);
5250 * Keep looping until we have no more ranges in the io tree.
5251 * We can have ongoing bios started by readpages (called from readahead)
5252 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5253 * still in progress (unlocked the pages in the bio but did not yet
5254 * unlocked the ranges in the io tree). Therefore this means some
5255 * ranges can still be locked and eviction started because before
5256 * submitting those bios, which are executed by a separate task (work
5257 * queue kthread), inode references (inode->i_count) were not taken
5258 * (which would be dropped in the end io callback of each bio).
5259 * Therefore here we effectively end up waiting for those bios and
5260 * anyone else holding locked ranges without having bumped the inode's
5261 * reference count - if we don't do it, when they access the inode's
5262 * io_tree to unlock a range it may be too late, leading to an
5263 * use-after-free issue.
5265 spin_lock(&io_tree->lock);
5266 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5267 struct extent_state *state;
5268 struct extent_state *cached_state = NULL;
5271 unsigned state_flags;
5273 node = rb_first(&io_tree->state);
5274 state = rb_entry(node, struct extent_state, rb_node);
5275 start = state->start;
5277 state_flags = state->state;
5278 spin_unlock(&io_tree->lock);
5280 lock_extent_bits(io_tree, start, end, &cached_state);
5283 * If still has DELALLOC flag, the extent didn't reach disk,
5284 * and its reserved space won't be freed by delayed_ref.
5285 * So we need to free its reserved space here.
5286 * (Refer to comment in btrfs_invalidatepage, case 2)
5288 * Note, end is the bytenr of last byte, so we need + 1 here.
5290 if (state_flags & EXTENT_DELALLOC)
5291 btrfs_qgroup_free_data(inode, NULL, start, end - start + 1);
5293 clear_extent_bit(io_tree, start, end,
5294 EXTENT_LOCKED | EXTENT_DIRTY |
5295 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5296 EXTENT_DEFRAG, 1, 1, &cached_state);
5299 spin_lock(&io_tree->lock);
5301 spin_unlock(&io_tree->lock);
5304 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5305 struct btrfs_block_rsv *rsv)
5307 struct btrfs_fs_info *fs_info = root->fs_info;
5308 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5312 struct btrfs_trans_handle *trans;
5315 ret = btrfs_block_rsv_refill(root, rsv, rsv->size,
5316 BTRFS_RESERVE_FLUSH_LIMIT);
5318 if (ret && ++failures > 2) {
5320 "could not allocate space for a delete; will truncate on mount");
5321 return ERR_PTR(-ENOSPC);
5324 trans = btrfs_join_transaction(root);
5325 if (IS_ERR(trans) || !ret)
5329 * Try to steal from the global reserve if there is space for
5332 if (!btrfs_check_space_for_delayed_refs(trans) &&
5333 !btrfs_block_rsv_migrate(global_rsv, rsv, rsv->size, false))
5336 /* If not, commit and try again. */
5337 ret = btrfs_commit_transaction(trans);
5339 return ERR_PTR(ret);
5343 void btrfs_evict_inode(struct inode *inode)
5345 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5346 struct btrfs_trans_handle *trans;
5347 struct btrfs_root *root = BTRFS_I(inode)->root;
5348 struct btrfs_block_rsv *rsv;
5351 trace_btrfs_inode_evict(inode);
5358 evict_inode_truncate_pages(inode);
5360 if (inode->i_nlink &&
5361 ((btrfs_root_refs(&root->root_item) != 0 &&
5362 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5363 btrfs_is_free_space_inode(BTRFS_I(inode))))
5366 if (is_bad_inode(inode))
5369 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5371 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5374 if (inode->i_nlink > 0) {
5375 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5376 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5380 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5384 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5387 rsv->size = btrfs_calc_trunc_metadata_size(fs_info, 1);
5390 btrfs_i_size_write(BTRFS_I(inode), 0);
5393 trans = evict_refill_and_join(root, rsv);
5397 trans->block_rsv = rsv;
5399 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5400 trans->block_rsv = &fs_info->trans_block_rsv;
5401 btrfs_end_transaction(trans);
5402 btrfs_btree_balance_dirty(fs_info);
5403 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5410 * Errors here aren't a big deal, it just means we leave orphan items in
5411 * the tree. They will be cleaned up on the next mount. If the inode
5412 * number gets reused, cleanup deletes the orphan item without doing
5413 * anything, and unlink reuses the existing orphan item.
5415 * If it turns out that we are dropping too many of these, we might want
5416 * to add a mechanism for retrying these after a commit.
5418 trans = evict_refill_and_join(root, rsv);
5419 if (!IS_ERR(trans)) {
5420 trans->block_rsv = rsv;
5421 btrfs_orphan_del(trans, BTRFS_I(inode));
5422 trans->block_rsv = &fs_info->trans_block_rsv;
5423 btrfs_end_transaction(trans);
5426 if (!(root == fs_info->tree_root ||
5427 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5428 btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
5431 btrfs_free_block_rsv(fs_info, rsv);
5434 * If we didn't successfully delete, the orphan item will still be in
5435 * the tree and we'll retry on the next mount. Again, we might also want
5436 * to retry these periodically in the future.
5438 btrfs_remove_delayed_node(BTRFS_I(inode));
5443 * this returns the key found in the dir entry in the location pointer.
5444 * If no dir entries were found, returns -ENOENT.
5445 * If found a corrupted location in dir entry, returns -EUCLEAN.
5447 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5448 struct btrfs_key *location)
5450 const char *name = dentry->d_name.name;
5451 int namelen = dentry->d_name.len;
5452 struct btrfs_dir_item *di;
5453 struct btrfs_path *path;
5454 struct btrfs_root *root = BTRFS_I(dir)->root;
5457 path = btrfs_alloc_path();
5461 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5463 if (IS_ERR_OR_NULL(di)) {
5464 ret = di ? PTR_ERR(di) : -ENOENT;
5468 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5469 if (location->type != BTRFS_INODE_ITEM_KEY &&
5470 location->type != BTRFS_ROOT_ITEM_KEY) {
5472 btrfs_warn(root->fs_info,
5473 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5474 __func__, name, btrfs_ino(BTRFS_I(dir)),
5475 location->objectid, location->type, location->offset);
5478 btrfs_free_path(path);
5483 * when we hit a tree root in a directory, the btrfs part of the inode
5484 * needs to be changed to reflect the root directory of the tree root. This
5485 * is kind of like crossing a mount point.
5487 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5489 struct dentry *dentry,
5490 struct btrfs_key *location,
5491 struct btrfs_root **sub_root)
5493 struct btrfs_path *path;
5494 struct btrfs_root *new_root;
5495 struct btrfs_root_ref *ref;
5496 struct extent_buffer *leaf;
5497 struct btrfs_key key;
5501 path = btrfs_alloc_path();
5508 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5509 key.type = BTRFS_ROOT_REF_KEY;
5510 key.offset = location->objectid;
5512 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5519 leaf = path->nodes[0];
5520 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5521 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5522 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5525 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5526 (unsigned long)(ref + 1),
5527 dentry->d_name.len);
5531 btrfs_release_path(path);
5533 new_root = btrfs_read_fs_root_no_name(fs_info, location);
5534 if (IS_ERR(new_root)) {
5535 err = PTR_ERR(new_root);
5539 *sub_root = new_root;
5540 location->objectid = btrfs_root_dirid(&new_root->root_item);
5541 location->type = BTRFS_INODE_ITEM_KEY;
5542 location->offset = 0;
5545 btrfs_free_path(path);
5549 static void inode_tree_add(struct inode *inode)
5551 struct btrfs_root *root = BTRFS_I(inode)->root;
5552 struct btrfs_inode *entry;
5554 struct rb_node *parent;
5555 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5556 u64 ino = btrfs_ino(BTRFS_I(inode));
5558 if (inode_unhashed(inode))
5561 spin_lock(&root->inode_lock);
5562 p = &root->inode_tree.rb_node;
5565 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5567 if (ino < btrfs_ino(entry))
5568 p = &parent->rb_left;
5569 else if (ino > btrfs_ino(entry))
5570 p = &parent->rb_right;
5572 WARN_ON(!(entry->vfs_inode.i_state &
5573 (I_WILL_FREE | I_FREEING)));
5574 rb_replace_node(parent, new, &root->inode_tree);
5575 RB_CLEAR_NODE(parent);
5576 spin_unlock(&root->inode_lock);
5580 rb_link_node(new, parent, p);
5581 rb_insert_color(new, &root->inode_tree);
5582 spin_unlock(&root->inode_lock);
5585 static void inode_tree_del(struct inode *inode)
5587 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5588 struct btrfs_root *root = BTRFS_I(inode)->root;
5591 spin_lock(&root->inode_lock);
5592 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5593 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5594 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5595 empty = RB_EMPTY_ROOT(&root->inode_tree);
5597 spin_unlock(&root->inode_lock);
5599 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5600 synchronize_srcu(&fs_info->subvol_srcu);
5601 spin_lock(&root->inode_lock);
5602 empty = RB_EMPTY_ROOT(&root->inode_tree);
5603 spin_unlock(&root->inode_lock);
5605 btrfs_add_dead_root(root);
5610 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5612 struct btrfs_iget_args *args = p;
5613 inode->i_ino = args->location->objectid;
5614 memcpy(&BTRFS_I(inode)->location, args->location,
5615 sizeof(*args->location));
5616 BTRFS_I(inode)->root = args->root;
5620 static int btrfs_find_actor(struct inode *inode, void *opaque)
5622 struct btrfs_iget_args *args = opaque;
5623 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5624 args->root == BTRFS_I(inode)->root;
5627 static struct inode *btrfs_iget_locked(struct super_block *s,
5628 struct btrfs_key *location,
5629 struct btrfs_root *root)
5631 struct inode *inode;
5632 struct btrfs_iget_args args;
5633 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5635 args.location = location;
5638 inode = iget5_locked(s, hashval, btrfs_find_actor,
5639 btrfs_init_locked_inode,
5644 /* Get an inode object given its location and corresponding root.
5645 * Returns in *is_new if the inode was read from disk
5647 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5648 struct btrfs_root *root, int *new)
5650 struct inode *inode;
5652 inode = btrfs_iget_locked(s, location, root);
5654 return ERR_PTR(-ENOMEM);
5656 if (inode->i_state & I_NEW) {
5659 ret = btrfs_read_locked_inode(inode);
5661 inode_tree_add(inode);
5662 unlock_new_inode(inode);
5668 * ret > 0 can come from btrfs_search_slot called by
5669 * btrfs_read_locked_inode, this means the inode item
5674 inode = ERR_PTR(ret);
5681 static struct inode *new_simple_dir(struct super_block *s,
5682 struct btrfs_key *key,
5683 struct btrfs_root *root)
5685 struct inode *inode = new_inode(s);
5688 return ERR_PTR(-ENOMEM);
5690 BTRFS_I(inode)->root = root;
5691 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5692 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5694 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5695 inode->i_op = &btrfs_dir_ro_inode_operations;
5696 inode->i_opflags &= ~IOP_XATTR;
5697 inode->i_fop = &simple_dir_operations;
5698 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5699 inode->i_mtime = current_time(inode);
5700 inode->i_atime = inode->i_mtime;
5701 inode->i_ctime = inode->i_mtime;
5702 BTRFS_I(inode)->i_otime = inode->i_mtime;
5707 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5709 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5710 struct inode *inode;
5711 struct btrfs_root *root = BTRFS_I(dir)->root;
5712 struct btrfs_root *sub_root = root;
5713 struct btrfs_key location;
5717 if (dentry->d_name.len > BTRFS_NAME_LEN)
5718 return ERR_PTR(-ENAMETOOLONG);
5720 ret = btrfs_inode_by_name(dir, dentry, &location);
5722 return ERR_PTR(ret);
5724 if (location.type == BTRFS_INODE_ITEM_KEY) {
5725 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5729 index = srcu_read_lock(&fs_info->subvol_srcu);
5730 ret = fixup_tree_root_location(fs_info, dir, dentry,
5731 &location, &sub_root);
5734 inode = ERR_PTR(ret);
5736 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5738 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5740 srcu_read_unlock(&fs_info->subvol_srcu, index);
5742 if (!IS_ERR(inode) && root != sub_root) {
5743 down_read(&fs_info->cleanup_work_sem);
5744 if (!sb_rdonly(inode->i_sb))
5745 ret = btrfs_orphan_cleanup(sub_root);
5746 up_read(&fs_info->cleanup_work_sem);
5749 inode = ERR_PTR(ret);
5756 static int btrfs_dentry_delete(const struct dentry *dentry)
5758 struct btrfs_root *root;
5759 struct inode *inode = d_inode(dentry);
5761 if (!inode && !IS_ROOT(dentry))
5762 inode = d_inode(dentry->d_parent);
5765 root = BTRFS_I(inode)->root;
5766 if (btrfs_root_refs(&root->root_item) == 0)
5769 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5775 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5778 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5780 if (inode == ERR_PTR(-ENOENT))
5782 return d_splice_alias(inode, dentry);
5785 unsigned char btrfs_filetype_table[] = {
5786 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5790 * All this infrastructure exists because dir_emit can fault, and we are holding
5791 * the tree lock when doing readdir. For now just allocate a buffer and copy
5792 * our information into that, and then dir_emit from the buffer. This is
5793 * similar to what NFS does, only we don't keep the buffer around in pagecache
5794 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5795 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5798 static int btrfs_opendir(struct inode *inode, struct file *file)
5800 struct btrfs_file_private *private;
5802 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5805 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5806 if (!private->filldir_buf) {
5810 file->private_data = private;
5821 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5824 struct dir_entry *entry = addr;
5825 char *name = (char *)(entry + 1);
5827 ctx->pos = get_unaligned(&entry->offset);
5828 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5829 get_unaligned(&entry->ino),
5830 get_unaligned(&entry->type)))
5832 addr += sizeof(struct dir_entry) +
5833 get_unaligned(&entry->name_len);
5839 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5841 struct inode *inode = file_inode(file);
5842 struct btrfs_root *root = BTRFS_I(inode)->root;
5843 struct btrfs_file_private *private = file->private_data;
5844 struct btrfs_dir_item *di;
5845 struct btrfs_key key;
5846 struct btrfs_key found_key;
5847 struct btrfs_path *path;
5849 struct list_head ins_list;
5850 struct list_head del_list;
5852 struct extent_buffer *leaf;
5859 struct btrfs_key location;
5861 if (!dir_emit_dots(file, ctx))
5864 path = btrfs_alloc_path();
5868 addr = private->filldir_buf;
5869 path->reada = READA_FORWARD;
5871 INIT_LIST_HEAD(&ins_list);
5872 INIT_LIST_HEAD(&del_list);
5873 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5876 key.type = BTRFS_DIR_INDEX_KEY;
5877 key.offset = ctx->pos;
5878 key.objectid = btrfs_ino(BTRFS_I(inode));
5880 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5885 struct dir_entry *entry;
5887 leaf = path->nodes[0];
5888 slot = path->slots[0];
5889 if (slot >= btrfs_header_nritems(leaf)) {
5890 ret = btrfs_next_leaf(root, path);
5898 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5900 if (found_key.objectid != key.objectid)
5902 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5904 if (found_key.offset < ctx->pos)
5906 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5908 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5909 name_len = btrfs_dir_name_len(leaf, di);
5910 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5912 btrfs_release_path(path);
5913 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5916 addr = private->filldir_buf;
5923 put_unaligned(name_len, &entry->name_len);
5924 name_ptr = (char *)(entry + 1);
5925 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5927 put_unaligned(btrfs_filetype_table[btrfs_dir_type(leaf, di)],
5929 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5930 put_unaligned(location.objectid, &entry->ino);
5931 put_unaligned(found_key.offset, &entry->offset);
5933 addr += sizeof(struct dir_entry) + name_len;
5934 total_len += sizeof(struct dir_entry) + name_len;
5938 btrfs_release_path(path);
5940 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5944 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5949 * Stop new entries from being returned after we return the last
5952 * New directory entries are assigned a strictly increasing
5953 * offset. This means that new entries created during readdir
5954 * are *guaranteed* to be seen in the future by that readdir.
5955 * This has broken buggy programs which operate on names as
5956 * they're returned by readdir. Until we re-use freed offsets
5957 * we have this hack to stop new entries from being returned
5958 * under the assumption that they'll never reach this huge
5961 * This is being careful not to overflow 32bit loff_t unless the
5962 * last entry requires it because doing so has broken 32bit apps
5965 if (ctx->pos >= INT_MAX)
5966 ctx->pos = LLONG_MAX;
5973 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5974 btrfs_free_path(path);
5979 * This is somewhat expensive, updating the tree every time the
5980 * inode changes. But, it is most likely to find the inode in cache.
5981 * FIXME, needs more benchmarking...there are no reasons other than performance
5982 * to keep or drop this code.
5984 static int btrfs_dirty_inode(struct inode *inode)
5986 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5987 struct btrfs_root *root = BTRFS_I(inode)->root;
5988 struct btrfs_trans_handle *trans;
5991 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5994 trans = btrfs_join_transaction(root);
5996 return PTR_ERR(trans);
5998 ret = btrfs_update_inode(trans, root, inode);
5999 if (ret && ret == -ENOSPC) {
6000 /* whoops, lets try again with the full transaction */
6001 btrfs_end_transaction(trans);
6002 trans = btrfs_start_transaction(root, 1);
6004 return PTR_ERR(trans);
6006 ret = btrfs_update_inode(trans, root, inode);
6008 btrfs_end_transaction(trans);
6009 if (BTRFS_I(inode)->delayed_node)
6010 btrfs_balance_delayed_items(fs_info);
6016 * This is a copy of file_update_time. We need this so we can return error on
6017 * ENOSPC for updating the inode in the case of file write and mmap writes.
6019 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
6022 struct btrfs_root *root = BTRFS_I(inode)->root;
6023 bool dirty = flags & ~S_VERSION;
6025 if (btrfs_root_readonly(root))
6028 if (flags & S_VERSION)
6029 dirty |= inode_maybe_inc_iversion(inode, dirty);
6030 if (flags & S_CTIME)
6031 inode->i_ctime = *now;
6032 if (flags & S_MTIME)
6033 inode->i_mtime = *now;
6034 if (flags & S_ATIME)
6035 inode->i_atime = *now;
6036 return dirty ? btrfs_dirty_inode(inode) : 0;
6040 * find the highest existing sequence number in a directory
6041 * and then set the in-memory index_cnt variable to reflect
6042 * free sequence numbers
6044 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6046 struct btrfs_root *root = inode->root;
6047 struct btrfs_key key, found_key;
6048 struct btrfs_path *path;
6049 struct extent_buffer *leaf;
6052 key.objectid = btrfs_ino(inode);
6053 key.type = BTRFS_DIR_INDEX_KEY;
6054 key.offset = (u64)-1;
6056 path = btrfs_alloc_path();
6060 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6063 /* FIXME: we should be able to handle this */
6069 * MAGIC NUMBER EXPLANATION:
6070 * since we search a directory based on f_pos we have to start at 2
6071 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6072 * else has to start at 2
6074 if (path->slots[0] == 0) {
6075 inode->index_cnt = 2;
6081 leaf = path->nodes[0];
6082 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6084 if (found_key.objectid != btrfs_ino(inode) ||
6085 found_key.type != BTRFS_DIR_INDEX_KEY) {
6086 inode->index_cnt = 2;
6090 inode->index_cnt = found_key.offset + 1;
6092 btrfs_free_path(path);
6097 * helper to find a free sequence number in a given directory. This current
6098 * code is very simple, later versions will do smarter things in the btree
6100 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6104 if (dir->index_cnt == (u64)-1) {
6105 ret = btrfs_inode_delayed_dir_index_count(dir);
6107 ret = btrfs_set_inode_index_count(dir);
6113 *index = dir->index_cnt;
6119 static int btrfs_insert_inode_locked(struct inode *inode)
6121 struct btrfs_iget_args args;
6122 args.location = &BTRFS_I(inode)->location;
6123 args.root = BTRFS_I(inode)->root;
6125 return insert_inode_locked4(inode,
6126 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6127 btrfs_find_actor, &args);
6131 * Inherit flags from the parent inode.
6133 * Currently only the compression flags and the cow flags are inherited.
6135 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6142 flags = BTRFS_I(dir)->flags;
6144 if (flags & BTRFS_INODE_NOCOMPRESS) {
6145 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6146 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6147 } else if (flags & BTRFS_INODE_COMPRESS) {
6148 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6149 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6152 if (flags & BTRFS_INODE_NODATACOW) {
6153 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6154 if (S_ISREG(inode->i_mode))
6155 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6158 btrfs_sync_inode_flags_to_i_flags(inode);
6161 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6162 struct btrfs_root *root,
6164 const char *name, int name_len,
6165 u64 ref_objectid, u64 objectid,
6166 umode_t mode, u64 *index)
6168 struct btrfs_fs_info *fs_info = root->fs_info;
6169 struct inode *inode;
6170 struct btrfs_inode_item *inode_item;
6171 struct btrfs_key *location;
6172 struct btrfs_path *path;
6173 struct btrfs_inode_ref *ref;
6174 struct btrfs_key key[2];
6176 int nitems = name ? 2 : 1;
6180 path = btrfs_alloc_path();
6182 return ERR_PTR(-ENOMEM);
6184 inode = new_inode(fs_info->sb);
6186 btrfs_free_path(path);
6187 return ERR_PTR(-ENOMEM);
6191 * O_TMPFILE, set link count to 0, so that after this point,
6192 * we fill in an inode item with the correct link count.
6195 set_nlink(inode, 0);
6198 * we have to initialize this early, so we can reclaim the inode
6199 * number if we fail afterwards in this function.
6201 inode->i_ino = objectid;
6204 trace_btrfs_inode_request(dir);
6206 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6208 btrfs_free_path(path);
6210 return ERR_PTR(ret);
6216 * index_cnt is ignored for everything but a dir,
6217 * btrfs_set_inode_index_count has an explanation for the magic
6220 BTRFS_I(inode)->index_cnt = 2;
6221 BTRFS_I(inode)->dir_index = *index;
6222 BTRFS_I(inode)->root = root;
6223 BTRFS_I(inode)->generation = trans->transid;
6224 inode->i_generation = BTRFS_I(inode)->generation;
6227 * We could have gotten an inode number from somebody who was fsynced
6228 * and then removed in this same transaction, so let's just set full
6229 * sync since it will be a full sync anyway and this will blow away the
6230 * old info in the log.
6232 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6234 key[0].objectid = objectid;
6235 key[0].type = BTRFS_INODE_ITEM_KEY;
6238 sizes[0] = sizeof(struct btrfs_inode_item);
6242 * Start new inodes with an inode_ref. This is slightly more
6243 * efficient for small numbers of hard links since they will
6244 * be packed into one item. Extended refs will kick in if we
6245 * add more hard links than can fit in the ref item.
6247 key[1].objectid = objectid;
6248 key[1].type = BTRFS_INODE_REF_KEY;
6249 key[1].offset = ref_objectid;
6251 sizes[1] = name_len + sizeof(*ref);
6254 location = &BTRFS_I(inode)->location;
6255 location->objectid = objectid;
6256 location->offset = 0;
6257 location->type = BTRFS_INODE_ITEM_KEY;
6259 ret = btrfs_insert_inode_locked(inode);
6265 path->leave_spinning = 1;
6266 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6270 inode_init_owner(inode, dir, mode);
6271 inode_set_bytes(inode, 0);
6273 inode->i_mtime = current_time(inode);
6274 inode->i_atime = inode->i_mtime;
6275 inode->i_ctime = inode->i_mtime;
6276 BTRFS_I(inode)->i_otime = inode->i_mtime;
6278 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6279 struct btrfs_inode_item);
6280 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6281 sizeof(*inode_item));
6282 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6285 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6286 struct btrfs_inode_ref);
6287 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6288 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6289 ptr = (unsigned long)(ref + 1);
6290 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6293 btrfs_mark_buffer_dirty(path->nodes[0]);
6294 btrfs_free_path(path);
6296 btrfs_inherit_iflags(inode, dir);
6298 if (S_ISREG(mode)) {
6299 if (btrfs_test_opt(fs_info, NODATASUM))
6300 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6301 if (btrfs_test_opt(fs_info, NODATACOW))
6302 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6303 BTRFS_INODE_NODATASUM;
6306 inode_tree_add(inode);
6308 trace_btrfs_inode_new(inode);
6309 btrfs_set_inode_last_trans(trans, inode);
6311 btrfs_update_root_times(trans, root);
6313 ret = btrfs_inode_inherit_props(trans, inode, dir);
6316 "error inheriting props for ino %llu (root %llu): %d",
6317 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6322 discard_new_inode(inode);
6325 BTRFS_I(dir)->index_cnt--;
6326 btrfs_free_path(path);
6327 return ERR_PTR(ret);
6330 static inline u8 btrfs_inode_type(struct inode *inode)
6332 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6336 * utility function to add 'inode' into 'parent_inode' with
6337 * a give name and a given sequence number.
6338 * if 'add_backref' is true, also insert a backref from the
6339 * inode to the parent directory.
6341 int btrfs_add_link(struct btrfs_trans_handle *trans,
6342 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6343 const char *name, int name_len, int add_backref, u64 index)
6346 struct btrfs_key key;
6347 struct btrfs_root *root = parent_inode->root;
6348 u64 ino = btrfs_ino(inode);
6349 u64 parent_ino = btrfs_ino(parent_inode);
6351 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6352 memcpy(&key, &inode->root->root_key, sizeof(key));
6355 key.type = BTRFS_INODE_ITEM_KEY;
6359 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6360 ret = btrfs_add_root_ref(trans, key.objectid,
6361 root->root_key.objectid, parent_ino,
6362 index, name, name_len);
6363 } else if (add_backref) {
6364 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6368 /* Nothing to clean up yet */
6372 ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key,
6373 btrfs_inode_type(&inode->vfs_inode), index);
6374 if (ret == -EEXIST || ret == -EOVERFLOW)
6377 btrfs_abort_transaction(trans, ret);
6381 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6383 inode_inc_iversion(&parent_inode->vfs_inode);
6384 parent_inode->vfs_inode.i_mtime = parent_inode->vfs_inode.i_ctime =
6385 current_time(&parent_inode->vfs_inode);
6386 ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode);
6388 btrfs_abort_transaction(trans, ret);
6392 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6395 err = btrfs_del_root_ref(trans, key.objectid,
6396 root->root_key.objectid, parent_ino,
6397 &local_index, name, name_len);
6399 } else if (add_backref) {
6403 err = btrfs_del_inode_ref(trans, root, name, name_len,
6404 ino, parent_ino, &local_index);
6409 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6410 struct btrfs_inode *dir, struct dentry *dentry,
6411 struct btrfs_inode *inode, int backref, u64 index)
6413 int err = btrfs_add_link(trans, dir, inode,
6414 dentry->d_name.name, dentry->d_name.len,
6421 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6422 umode_t mode, dev_t rdev)
6424 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6425 struct btrfs_trans_handle *trans;
6426 struct btrfs_root *root = BTRFS_I(dir)->root;
6427 struct inode *inode = NULL;
6433 * 2 for inode item and ref
6435 * 1 for xattr if selinux is on
6437 trans = btrfs_start_transaction(root, 5);
6439 return PTR_ERR(trans);
6441 err = btrfs_find_free_ino(root, &objectid);
6445 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6446 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6448 if (IS_ERR(inode)) {
6449 err = PTR_ERR(inode);
6455 * If the active LSM wants to access the inode during
6456 * d_instantiate it needs these. Smack checks to see
6457 * if the filesystem supports xattrs by looking at the
6460 inode->i_op = &btrfs_special_inode_operations;
6461 init_special_inode(inode, inode->i_mode, rdev);
6463 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6467 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6472 btrfs_update_inode(trans, root, inode);
6473 d_instantiate_new(dentry, inode);
6476 btrfs_end_transaction(trans);
6477 btrfs_btree_balance_dirty(fs_info);
6479 inode_dec_link_count(inode);
6480 discard_new_inode(inode);
6485 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6486 umode_t mode, bool excl)
6488 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6489 struct btrfs_trans_handle *trans;
6490 struct btrfs_root *root = BTRFS_I(dir)->root;
6491 struct inode *inode = NULL;
6497 * 2 for inode item and ref
6499 * 1 for xattr if selinux is on
6501 trans = btrfs_start_transaction(root, 5);
6503 return PTR_ERR(trans);
6505 err = btrfs_find_free_ino(root, &objectid);
6509 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6510 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6512 if (IS_ERR(inode)) {
6513 err = PTR_ERR(inode);
6518 * If the active LSM wants to access the inode during
6519 * d_instantiate it needs these. Smack checks to see
6520 * if the filesystem supports xattrs by looking at the
6523 inode->i_fop = &btrfs_file_operations;
6524 inode->i_op = &btrfs_file_inode_operations;
6525 inode->i_mapping->a_ops = &btrfs_aops;
6527 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6531 err = btrfs_update_inode(trans, root, inode);
6535 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6540 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6541 d_instantiate_new(dentry, inode);
6544 btrfs_end_transaction(trans);
6546 inode_dec_link_count(inode);
6547 discard_new_inode(inode);
6549 btrfs_btree_balance_dirty(fs_info);
6553 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6554 struct dentry *dentry)
6556 struct btrfs_trans_handle *trans = NULL;
6557 struct btrfs_root *root = BTRFS_I(dir)->root;
6558 struct inode *inode = d_inode(old_dentry);
6559 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6564 /* do not allow sys_link's with other subvols of the same device */
6565 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6568 if (inode->i_nlink >= BTRFS_LINK_MAX)
6571 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6576 * 2 items for inode and inode ref
6577 * 2 items for dir items
6578 * 1 item for parent inode
6579 * 1 item for orphan item deletion if O_TMPFILE
6581 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6582 if (IS_ERR(trans)) {
6583 err = PTR_ERR(trans);
6588 /* There are several dir indexes for this inode, clear the cache. */
6589 BTRFS_I(inode)->dir_index = 0ULL;
6591 inode_inc_iversion(inode);
6592 inode->i_ctime = current_time(inode);
6594 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6596 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6602 struct dentry *parent = dentry->d_parent;
6605 err = btrfs_update_inode(trans, root, inode);
6608 if (inode->i_nlink == 1) {
6610 * If new hard link count is 1, it's a file created
6611 * with open(2) O_TMPFILE flag.
6613 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6617 d_instantiate(dentry, inode);
6618 ret = btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent,
6620 if (ret == BTRFS_NEED_TRANS_COMMIT) {
6621 err = btrfs_commit_transaction(trans);
6628 btrfs_end_transaction(trans);
6630 inode_dec_link_count(inode);
6633 btrfs_btree_balance_dirty(fs_info);
6637 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6639 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6640 struct inode *inode = NULL;
6641 struct btrfs_trans_handle *trans;
6642 struct btrfs_root *root = BTRFS_I(dir)->root;
6644 int drop_on_err = 0;
6649 * 2 items for inode and ref
6650 * 2 items for dir items
6651 * 1 for xattr if selinux is on
6653 trans = btrfs_start_transaction(root, 5);
6655 return PTR_ERR(trans);
6657 err = btrfs_find_free_ino(root, &objectid);
6661 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6662 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6663 S_IFDIR | mode, &index);
6664 if (IS_ERR(inode)) {
6665 err = PTR_ERR(inode);
6671 /* these must be set before we unlock the inode */
6672 inode->i_op = &btrfs_dir_inode_operations;
6673 inode->i_fop = &btrfs_dir_file_operations;
6675 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6679 btrfs_i_size_write(BTRFS_I(inode), 0);
6680 err = btrfs_update_inode(trans, root, inode);
6684 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6685 dentry->d_name.name,
6686 dentry->d_name.len, 0, index);
6690 d_instantiate_new(dentry, inode);
6694 btrfs_end_transaction(trans);
6696 inode_dec_link_count(inode);
6697 discard_new_inode(inode);
6699 btrfs_btree_balance_dirty(fs_info);
6703 static noinline int uncompress_inline(struct btrfs_path *path,
6705 size_t pg_offset, u64 extent_offset,
6706 struct btrfs_file_extent_item *item)
6709 struct extent_buffer *leaf = path->nodes[0];
6712 unsigned long inline_size;
6716 WARN_ON(pg_offset != 0);
6717 compress_type = btrfs_file_extent_compression(leaf, item);
6718 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6719 inline_size = btrfs_file_extent_inline_item_len(leaf,
6720 btrfs_item_nr(path->slots[0]));
6721 tmp = kmalloc(inline_size, GFP_NOFS);
6724 ptr = btrfs_file_extent_inline_start(item);
6726 read_extent_buffer(leaf, tmp, ptr, inline_size);
6728 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6729 ret = btrfs_decompress(compress_type, tmp, page,
6730 extent_offset, inline_size, max_size);
6733 * decompression code contains a memset to fill in any space between the end
6734 * of the uncompressed data and the end of max_size in case the decompressed
6735 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6736 * the end of an inline extent and the beginning of the next block, so we
6737 * cover that region here.
6740 if (max_size + pg_offset < PAGE_SIZE) {
6741 char *map = kmap(page);
6742 memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6750 * a bit scary, this does extent mapping from logical file offset to the disk.
6751 * the ugly parts come from merging extents from the disk with the in-ram
6752 * representation. This gets more complex because of the data=ordered code,
6753 * where the in-ram extents might be locked pending data=ordered completion.
6755 * This also copies inline extents directly into the page.
6757 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6759 size_t pg_offset, u64 start, u64 len,
6762 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6765 u64 extent_start = 0;
6767 u64 objectid = btrfs_ino(inode);
6769 struct btrfs_path *path = NULL;
6770 struct btrfs_root *root = inode->root;
6771 struct btrfs_file_extent_item *item;
6772 struct extent_buffer *leaf;
6773 struct btrfs_key found_key;
6774 struct extent_map *em = NULL;
6775 struct extent_map_tree *em_tree = &inode->extent_tree;
6776 struct extent_io_tree *io_tree = &inode->io_tree;
6777 const bool new_inline = !page || create;
6779 read_lock(&em_tree->lock);
6780 em = lookup_extent_mapping(em_tree, start, len);
6782 em->bdev = fs_info->fs_devices->latest_bdev;
6783 read_unlock(&em_tree->lock);
6786 if (em->start > start || em->start + em->len <= start)
6787 free_extent_map(em);
6788 else if (em->block_start == EXTENT_MAP_INLINE && page)
6789 free_extent_map(em);
6793 em = alloc_extent_map();
6798 em->bdev = fs_info->fs_devices->latest_bdev;
6799 em->start = EXTENT_MAP_HOLE;
6800 em->orig_start = EXTENT_MAP_HOLE;
6802 em->block_len = (u64)-1;
6804 path = btrfs_alloc_path();
6810 /* Chances are we'll be called again, so go ahead and do readahead */
6811 path->reada = READA_FORWARD;
6814 * Unless we're going to uncompress the inline extent, no sleep would
6817 path->leave_spinning = 1;
6819 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6826 if (path->slots[0] == 0)
6831 leaf = path->nodes[0];
6832 item = btrfs_item_ptr(leaf, path->slots[0],
6833 struct btrfs_file_extent_item);
6834 /* are we inside the extent that was found? */
6835 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6836 found_type = found_key.type;
6837 if (found_key.objectid != objectid ||
6838 found_type != BTRFS_EXTENT_DATA_KEY) {
6840 * If we backup past the first extent we want to move forward
6841 * and see if there is an extent in front of us, otherwise we'll
6842 * say there is a hole for our whole search range which can
6849 found_type = btrfs_file_extent_type(leaf, item);
6850 extent_start = found_key.offset;
6851 if (found_type == BTRFS_FILE_EXTENT_REG ||
6852 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6853 extent_end = extent_start +
6854 btrfs_file_extent_num_bytes(leaf, item);
6856 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6858 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6861 size = btrfs_file_extent_ram_bytes(leaf, item);
6862 extent_end = ALIGN(extent_start + size,
6863 fs_info->sectorsize);
6865 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6870 if (start >= extent_end) {
6872 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6873 ret = btrfs_next_leaf(root, path);
6880 leaf = path->nodes[0];
6882 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6883 if (found_key.objectid != objectid ||
6884 found_key.type != BTRFS_EXTENT_DATA_KEY)
6886 if (start + len <= found_key.offset)
6888 if (start > found_key.offset)
6891 em->orig_start = start;
6892 em->len = found_key.offset - start;
6896 btrfs_extent_item_to_extent_map(inode, path, item,
6899 if (found_type == BTRFS_FILE_EXTENT_REG ||
6900 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6902 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6906 size_t extent_offset;
6912 size = btrfs_file_extent_ram_bytes(leaf, item);
6913 extent_offset = page_offset(page) + pg_offset - extent_start;
6914 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6915 size - extent_offset);
6916 em->start = extent_start + extent_offset;
6917 em->len = ALIGN(copy_size, fs_info->sectorsize);
6918 em->orig_block_len = em->len;
6919 em->orig_start = em->start;
6920 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6922 btrfs_set_path_blocking(path);
6923 if (!PageUptodate(page)) {
6924 if (btrfs_file_extent_compression(leaf, item) !=
6925 BTRFS_COMPRESS_NONE) {
6926 ret = uncompress_inline(path, page, pg_offset,
6927 extent_offset, item);
6934 read_extent_buffer(leaf, map + pg_offset, ptr,
6936 if (pg_offset + copy_size < PAGE_SIZE) {
6937 memset(map + pg_offset + copy_size, 0,
6938 PAGE_SIZE - pg_offset -
6943 flush_dcache_page(page);
6945 set_extent_uptodate(io_tree, em->start,
6946 extent_map_end(em) - 1, NULL, GFP_NOFS);
6951 em->orig_start = start;
6954 em->block_start = EXTENT_MAP_HOLE;
6956 btrfs_release_path(path);
6957 if (em->start > start || extent_map_end(em) <= start) {
6959 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6960 em->start, em->len, start, len);
6966 write_lock(&em_tree->lock);
6967 err = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6968 write_unlock(&em_tree->lock);
6970 btrfs_free_path(path);
6972 trace_btrfs_get_extent(root, inode, em);
6975 free_extent_map(em);
6976 return ERR_PTR(err);
6978 BUG_ON(!em); /* Error is always set */
6982 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
6984 size_t pg_offset, u64 start, u64 len,
6987 struct extent_map *em;
6988 struct extent_map *hole_em = NULL;
6989 u64 range_start = start;
6995 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6999 * If our em maps to:
7001 * - a pre-alloc extent,
7002 * there might actually be delalloc bytes behind it.
7004 if (em->block_start != EXTENT_MAP_HOLE &&
7005 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7010 /* check to see if we've wrapped (len == -1 or similar) */
7019 /* ok, we didn't find anything, lets look for delalloc */
7020 found = count_range_bits(&inode->io_tree, &range_start,
7021 end, len, EXTENT_DELALLOC, 1);
7022 found_end = range_start + found;
7023 if (found_end < range_start)
7024 found_end = (u64)-1;
7027 * we didn't find anything useful, return
7028 * the original results from get_extent()
7030 if (range_start > end || found_end <= start) {
7036 /* adjust the range_start to make sure it doesn't
7037 * go backwards from the start they passed in
7039 range_start = max(start, range_start);
7040 found = found_end - range_start;
7043 u64 hole_start = start;
7046 em = alloc_extent_map();
7052 * when btrfs_get_extent can't find anything it
7053 * returns one huge hole
7055 * make sure what it found really fits our range, and
7056 * adjust to make sure it is based on the start from
7060 u64 calc_end = extent_map_end(hole_em);
7062 if (calc_end <= start || (hole_em->start > end)) {
7063 free_extent_map(hole_em);
7066 hole_start = max(hole_em->start, start);
7067 hole_len = calc_end - hole_start;
7071 if (hole_em && range_start > hole_start) {
7072 /* our hole starts before our delalloc, so we
7073 * have to return just the parts of the hole
7074 * that go until the delalloc starts
7076 em->len = min(hole_len,
7077 range_start - hole_start);
7078 em->start = hole_start;
7079 em->orig_start = hole_start;
7081 * don't adjust block start at all,
7082 * it is fixed at EXTENT_MAP_HOLE
7084 em->block_start = hole_em->block_start;
7085 em->block_len = hole_len;
7086 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7087 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7089 em->start = range_start;
7091 em->orig_start = range_start;
7092 em->block_start = EXTENT_MAP_DELALLOC;
7093 em->block_len = found;
7100 free_extent_map(hole_em);
7102 free_extent_map(em);
7103 return ERR_PTR(err);
7108 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7111 const u64 orig_start,
7112 const u64 block_start,
7113 const u64 block_len,
7114 const u64 orig_block_len,
7115 const u64 ram_bytes,
7118 struct extent_map *em = NULL;
7121 if (type != BTRFS_ORDERED_NOCOW) {
7122 em = create_io_em(inode, start, len, orig_start,
7123 block_start, block_len, orig_block_len,
7125 BTRFS_COMPRESS_NONE, /* compress_type */
7130 ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7131 len, block_len, type);
7134 free_extent_map(em);
7135 btrfs_drop_extent_cache(BTRFS_I(inode), start,
7136 start + len - 1, 0);
7145 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7148 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7149 struct btrfs_root *root = BTRFS_I(inode)->root;
7150 struct extent_map *em;
7151 struct btrfs_key ins;
7155 alloc_hint = get_extent_allocation_hint(inode, start, len);
7156 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7157 0, alloc_hint, &ins, 1, 1);
7159 return ERR_PTR(ret);
7161 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7162 ins.objectid, ins.offset, ins.offset,
7163 ins.offset, BTRFS_ORDERED_REGULAR);
7164 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7166 btrfs_free_reserved_extent(fs_info, ins.objectid,
7173 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7174 * block must be cow'd
7176 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7177 u64 *orig_start, u64 *orig_block_len,
7180 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7181 struct btrfs_path *path;
7183 struct extent_buffer *leaf;
7184 struct btrfs_root *root = BTRFS_I(inode)->root;
7185 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7186 struct btrfs_file_extent_item *fi;
7187 struct btrfs_key key;
7194 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7196 path = btrfs_alloc_path();
7200 ret = btrfs_lookup_file_extent(NULL, root, path,
7201 btrfs_ino(BTRFS_I(inode)), offset, 0);
7205 slot = path->slots[0];
7208 /* can't find the item, must cow */
7215 leaf = path->nodes[0];
7216 btrfs_item_key_to_cpu(leaf, &key, slot);
7217 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7218 key.type != BTRFS_EXTENT_DATA_KEY) {
7219 /* not our file or wrong item type, must cow */
7223 if (key.offset > offset) {
7224 /* Wrong offset, must cow */
7228 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7229 found_type = btrfs_file_extent_type(leaf, fi);
7230 if (found_type != BTRFS_FILE_EXTENT_REG &&
7231 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7232 /* not a regular extent, must cow */
7236 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7239 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7240 if (extent_end <= offset)
7243 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7244 if (disk_bytenr == 0)
7247 if (btrfs_file_extent_compression(leaf, fi) ||
7248 btrfs_file_extent_encryption(leaf, fi) ||
7249 btrfs_file_extent_other_encoding(leaf, fi))
7253 * Do the same check as in btrfs_cross_ref_exist but without the
7254 * unnecessary search.
7256 if (btrfs_file_extent_generation(leaf, fi) <=
7257 btrfs_root_last_snapshot(&root->root_item))
7260 backref_offset = btrfs_file_extent_offset(leaf, fi);
7263 *orig_start = key.offset - backref_offset;
7264 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7265 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7268 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7271 num_bytes = min(offset + *len, extent_end) - offset;
7272 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7275 range_end = round_up(offset + num_bytes,
7276 root->fs_info->sectorsize) - 1;
7277 ret = test_range_bit(io_tree, offset, range_end,
7278 EXTENT_DELALLOC, 0, NULL);
7285 btrfs_release_path(path);
7288 * look for other files referencing this extent, if we
7289 * find any we must cow
7292 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7293 key.offset - backref_offset, disk_bytenr);
7300 * adjust disk_bytenr and num_bytes to cover just the bytes
7301 * in this extent we are about to write. If there
7302 * are any csums in that range we have to cow in order
7303 * to keep the csums correct
7305 disk_bytenr += backref_offset;
7306 disk_bytenr += offset - key.offset;
7307 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7310 * all of the above have passed, it is safe to overwrite this extent
7316 btrfs_free_path(path);
7320 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7321 struct extent_state **cached_state, int writing)
7323 struct btrfs_ordered_extent *ordered;
7327 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7330 * We're concerned with the entire range that we're going to be
7331 * doing DIO to, so we need to make sure there's no ordered
7332 * extents in this range.
7334 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7335 lockend - lockstart + 1);
7338 * We need to make sure there are no buffered pages in this
7339 * range either, we could have raced between the invalidate in
7340 * generic_file_direct_write and locking the extent. The
7341 * invalidate needs to happen so that reads after a write do not
7345 (!writing || !filemap_range_has_page(inode->i_mapping,
7346 lockstart, lockend)))
7349 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7354 * If we are doing a DIO read and the ordered extent we
7355 * found is for a buffered write, we can not wait for it
7356 * to complete and retry, because if we do so we can
7357 * deadlock with concurrent buffered writes on page
7358 * locks. This happens only if our DIO read covers more
7359 * than one extent map, if at this point has already
7360 * created an ordered extent for a previous extent map
7361 * and locked its range in the inode's io tree, and a
7362 * concurrent write against that previous extent map's
7363 * range and this range started (we unlock the ranges
7364 * in the io tree only when the bios complete and
7365 * buffered writes always lock pages before attempting
7366 * to lock range in the io tree).
7369 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7370 btrfs_start_ordered_extent(inode, ordered, 1);
7373 btrfs_put_ordered_extent(ordered);
7376 * We could trigger writeback for this range (and wait
7377 * for it to complete) and then invalidate the pages for
7378 * this range (through invalidate_inode_pages2_range()),
7379 * but that can lead us to a deadlock with a concurrent
7380 * call to readpages() (a buffered read or a defrag call
7381 * triggered a readahead) on a page lock due to an
7382 * ordered dio extent we created before but did not have
7383 * yet a corresponding bio submitted (whence it can not
7384 * complete), which makes readpages() wait for that
7385 * ordered extent to complete while holding a lock on
7400 /* The callers of this must take lock_extent() */
7401 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
7402 u64 orig_start, u64 block_start,
7403 u64 block_len, u64 orig_block_len,
7404 u64 ram_bytes, int compress_type,
7407 struct extent_map_tree *em_tree;
7408 struct extent_map *em;
7409 struct btrfs_root *root = BTRFS_I(inode)->root;
7412 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7413 type == BTRFS_ORDERED_COMPRESSED ||
7414 type == BTRFS_ORDERED_NOCOW ||
7415 type == BTRFS_ORDERED_REGULAR);
7417 em_tree = &BTRFS_I(inode)->extent_tree;
7418 em = alloc_extent_map();
7420 return ERR_PTR(-ENOMEM);
7423 em->orig_start = orig_start;
7425 em->block_len = block_len;
7426 em->block_start = block_start;
7427 em->bdev = root->fs_info->fs_devices->latest_bdev;
7428 em->orig_block_len = orig_block_len;
7429 em->ram_bytes = ram_bytes;
7430 em->generation = -1;
7431 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7432 if (type == BTRFS_ORDERED_PREALLOC) {
7433 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7434 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7435 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7436 em->compress_type = compress_type;
7440 btrfs_drop_extent_cache(BTRFS_I(inode), em->start,
7441 em->start + em->len - 1, 0);
7442 write_lock(&em_tree->lock);
7443 ret = add_extent_mapping(em_tree, em, 1);
7444 write_unlock(&em_tree->lock);
7446 * The caller has taken lock_extent(), who could race with us
7449 } while (ret == -EEXIST);
7452 free_extent_map(em);
7453 return ERR_PTR(ret);
7456 /* em got 2 refs now, callers needs to do free_extent_map once. */
7461 static int btrfs_get_blocks_direct_read(struct extent_map *em,
7462 struct buffer_head *bh_result,
7463 struct inode *inode,
7466 if (em->block_start == EXTENT_MAP_HOLE ||
7467 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7470 len = min(len, em->len - (start - em->start));
7472 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7474 bh_result->b_size = len;
7475 bh_result->b_bdev = em->bdev;
7476 set_buffer_mapped(bh_result);
7481 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7482 struct buffer_head *bh_result,
7483 struct inode *inode,
7484 struct btrfs_dio_data *dio_data,
7487 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7488 struct extent_map *em = *map;
7492 * We don't allocate a new extent in the following cases
7494 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7496 * 2) The extent is marked as PREALLOC. We're good to go here and can
7497 * just use the extent.
7500 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7501 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7502 em->block_start != EXTENT_MAP_HOLE)) {
7504 u64 block_start, orig_start, orig_block_len, ram_bytes;
7506 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7507 type = BTRFS_ORDERED_PREALLOC;
7509 type = BTRFS_ORDERED_NOCOW;
7510 len = min(len, em->len - (start - em->start));
7511 block_start = em->block_start + (start - em->start);
7513 if (can_nocow_extent(inode, start, &len, &orig_start,
7514 &orig_block_len, &ram_bytes) == 1 &&
7515 btrfs_inc_nocow_writers(fs_info, block_start)) {
7516 struct extent_map *em2;
7518 em2 = btrfs_create_dio_extent(inode, start, len,
7519 orig_start, block_start,
7520 len, orig_block_len,
7522 btrfs_dec_nocow_writers(fs_info, block_start);
7523 if (type == BTRFS_ORDERED_PREALLOC) {
7524 free_extent_map(em);
7528 if (em2 && IS_ERR(em2)) {
7533 * For inode marked NODATACOW or extent marked PREALLOC,
7534 * use the existing or preallocated extent, so does not
7535 * need to adjust btrfs_space_info's bytes_may_use.
7537 btrfs_free_reserved_data_space_noquota(inode, start,
7543 /* this will cow the extent */
7544 len = bh_result->b_size;
7545 free_extent_map(em);
7546 *map = em = btrfs_new_extent_direct(inode, start, len);
7552 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);
7561 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7562 set_buffer_new(bh_result);
7565 * Need to update the i_size under the extent lock so buffered
7566 * readers will get the updated i_size when we unlock.
7568 if (!dio_data->overwrite && start + len > i_size_read(inode))
7569 i_size_write(inode, start + len);
7571 WARN_ON(dio_data->reserve < len);
7572 dio_data->reserve -= len;
7573 dio_data->unsubmitted_oe_range_end = start + len;
7574 current->journal_info = dio_data;
7579 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7580 struct buffer_head *bh_result, int create)
7582 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7583 struct extent_map *em;
7584 struct extent_state *cached_state = NULL;
7585 struct btrfs_dio_data *dio_data = NULL;
7586 u64 start = iblock << inode->i_blkbits;
7587 u64 lockstart, lockend;
7588 u64 len = bh_result->b_size;
7589 int unlock_bits = EXTENT_LOCKED;
7593 unlock_bits |= EXTENT_DIRTY;
7595 len = min_t(u64, len, fs_info->sectorsize);
7598 lockend = start + len - 1;
7600 if (current->journal_info) {
7602 * Need to pull our outstanding extents and set journal_info to NULL so
7603 * that anything that needs to check if there's a transaction doesn't get
7606 dio_data = current->journal_info;
7607 current->journal_info = NULL;
7611 * If this errors out it's because we couldn't invalidate pagecache for
7612 * this range and we need to fallback to buffered.
7614 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7620 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
7627 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7628 * io. INLINE is special, and we could probably kludge it in here, but
7629 * it's still buffered so for safety lets just fall back to the generic
7632 * For COMPRESSED we _have_ to read the entire extent in so we can
7633 * decompress it, so there will be buffering required no matter what we
7634 * do, so go ahead and fallback to buffered.
7636 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7637 * to buffered IO. Don't blame me, this is the price we pay for using
7640 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7641 em->block_start == EXTENT_MAP_INLINE) {
7642 free_extent_map(em);
7648 ret = btrfs_get_blocks_direct_write(&em, bh_result, inode,
7649 dio_data, start, len);
7653 /* clear and unlock the entire range */
7654 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7655 unlock_bits, 1, 0, &cached_state);
7657 ret = btrfs_get_blocks_direct_read(em, bh_result, inode,
7659 /* Can be negative only if we read from a hole */
7662 free_extent_map(em);
7666 * We need to unlock only the end area that we aren't using.
7667 * The rest is going to be unlocked by the endio routine.
7669 lockstart = start + bh_result->b_size;
7670 if (lockstart < lockend) {
7671 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7672 lockend, unlock_bits, 1, 0,
7675 free_extent_state(cached_state);
7679 free_extent_map(em);
7684 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7685 unlock_bits, 1, 0, &cached_state);
7688 current->journal_info = dio_data;
7692 static inline blk_status_t submit_dio_repair_bio(struct inode *inode,
7696 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7699 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7701 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR);
7705 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
7710 static int btrfs_check_dio_repairable(struct inode *inode,
7711 struct bio *failed_bio,
7712 struct io_failure_record *failrec,
7715 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7718 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
7719 if (num_copies == 1) {
7721 * we only have a single copy of the data, so don't bother with
7722 * all the retry and error correction code that follows. no
7723 * matter what the error is, it is very likely to persist.
7725 btrfs_debug(fs_info,
7726 "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
7727 num_copies, failrec->this_mirror, failed_mirror);
7731 failrec->failed_mirror = failed_mirror;
7732 failrec->this_mirror++;
7733 if (failrec->this_mirror == failed_mirror)
7734 failrec->this_mirror++;
7736 if (failrec->this_mirror > num_copies) {
7737 btrfs_debug(fs_info,
7738 "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
7739 num_copies, failrec->this_mirror, failed_mirror);
7746 static blk_status_t dio_read_error(struct inode *inode, struct bio *failed_bio,
7747 struct page *page, unsigned int pgoff,
7748 u64 start, u64 end, int failed_mirror,
7749 bio_end_io_t *repair_endio, void *repair_arg)
7751 struct io_failure_record *failrec;
7752 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7753 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7756 unsigned int read_mode = 0;
7759 blk_status_t status;
7760 struct bio_vec bvec;
7762 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
7764 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7766 return errno_to_blk_status(ret);
7768 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7771 free_io_failure(failure_tree, io_tree, failrec);
7772 return BLK_STS_IOERR;
7775 segs = bio_segments(failed_bio);
7776 bio_get_first_bvec(failed_bio, &bvec);
7778 (bvec.bv_len > btrfs_inode_sectorsize(inode)))
7779 read_mode |= REQ_FAILFAST_DEV;
7781 isector = start - btrfs_io_bio(failed_bio)->logical;
7782 isector >>= inode->i_sb->s_blocksize_bits;
7783 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7784 pgoff, isector, repair_endio, repair_arg);
7785 bio->bi_opf = REQ_OP_READ | read_mode;
7787 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7788 "repair DIO read error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d",
7789 read_mode, failrec->this_mirror, failrec->in_validation);
7791 status = submit_dio_repair_bio(inode, bio, failrec->this_mirror);
7793 free_io_failure(failure_tree, io_tree, failrec);
7800 struct btrfs_retry_complete {
7801 struct completion done;
7802 struct inode *inode;
7807 static void btrfs_retry_endio_nocsum(struct bio *bio)
7809 struct btrfs_retry_complete *done = bio->bi_private;
7810 struct inode *inode = done->inode;
7811 struct bio_vec *bvec;
7812 struct extent_io_tree *io_tree, *failure_tree;
7818 ASSERT(bio->bi_vcnt == 1);
7819 io_tree = &BTRFS_I(inode)->io_tree;
7820 failure_tree = &BTRFS_I(inode)->io_failure_tree;
7821 ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(inode));
7824 ASSERT(!bio_flagged(bio, BIO_CLONED));
7825 bio_for_each_segment_all(bvec, bio, i)
7826 clean_io_failure(BTRFS_I(inode)->root->fs_info, failure_tree,
7827 io_tree, done->start, bvec->bv_page,
7828 btrfs_ino(BTRFS_I(inode)), 0);
7830 complete(&done->done);
7834 static blk_status_t __btrfs_correct_data_nocsum(struct inode *inode,
7835 struct btrfs_io_bio *io_bio)
7837 struct btrfs_fs_info *fs_info;
7838 struct bio_vec bvec;
7839 struct bvec_iter iter;
7840 struct btrfs_retry_complete done;
7846 blk_status_t err = BLK_STS_OK;
7848 fs_info = BTRFS_I(inode)->root->fs_info;
7849 sectorsize = fs_info->sectorsize;
7851 start = io_bio->logical;
7853 io_bio->bio.bi_iter = io_bio->iter;
7855 bio_for_each_segment(bvec, &io_bio->bio, iter) {
7856 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7857 pgoff = bvec.bv_offset;
7859 next_block_or_try_again:
7862 init_completion(&done.done);
7864 ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
7865 pgoff, start, start + sectorsize - 1,
7867 btrfs_retry_endio_nocsum, &done);
7873 wait_for_completion_io(&done.done);
7875 if (!done.uptodate) {
7876 /* We might have another mirror, so try again */
7877 goto next_block_or_try_again;
7881 start += sectorsize;
7885 pgoff += sectorsize;
7886 ASSERT(pgoff < PAGE_SIZE);
7887 goto next_block_or_try_again;
7894 static void btrfs_retry_endio(struct bio *bio)
7896 struct btrfs_retry_complete *done = bio->bi_private;
7897 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7898 struct extent_io_tree *io_tree, *failure_tree;
7899 struct inode *inode = done->inode;
7900 struct bio_vec *bvec;
7910 ASSERT(bio->bi_vcnt == 1);
7911 ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(done->inode));
7913 io_tree = &BTRFS_I(inode)->io_tree;
7914 failure_tree = &BTRFS_I(inode)->io_failure_tree;
7916 ASSERT(!bio_flagged(bio, BIO_CLONED));
7917 bio_for_each_segment_all(bvec, bio, i) {
7918 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
7919 bvec->bv_offset, done->start,
7922 clean_io_failure(BTRFS_I(inode)->root->fs_info,
7923 failure_tree, io_tree, done->start,
7925 btrfs_ino(BTRFS_I(inode)),
7931 done->uptodate = uptodate;
7933 complete(&done->done);
7937 static blk_status_t __btrfs_subio_endio_read(struct inode *inode,
7938 struct btrfs_io_bio *io_bio, blk_status_t err)
7940 struct btrfs_fs_info *fs_info;
7941 struct bio_vec bvec;
7942 struct bvec_iter iter;
7943 struct btrfs_retry_complete done;
7950 bool uptodate = (err == 0);
7952 blk_status_t status;
7954 fs_info = BTRFS_I(inode)->root->fs_info;
7955 sectorsize = fs_info->sectorsize;
7958 start = io_bio->logical;
7960 io_bio->bio.bi_iter = io_bio->iter;
7962 bio_for_each_segment(bvec, &io_bio->bio, iter) {
7963 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7965 pgoff = bvec.bv_offset;
7968 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
7969 ret = __readpage_endio_check(inode, io_bio, csum_pos,
7970 bvec.bv_page, pgoff, start, sectorsize);
7977 init_completion(&done.done);
7979 status = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
7980 pgoff, start, start + sectorsize - 1,
7981 io_bio->mirror_num, btrfs_retry_endio,
7988 wait_for_completion_io(&done.done);
7990 if (!done.uptodate) {
7991 /* We might have another mirror, so try again */
7995 offset += sectorsize;
7996 start += sectorsize;
8002 pgoff += sectorsize;
8003 ASSERT(pgoff < PAGE_SIZE);
8011 static blk_status_t btrfs_subio_endio_read(struct inode *inode,
8012 struct btrfs_io_bio *io_bio, blk_status_t err)
8014 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8018 return __btrfs_correct_data_nocsum(inode, io_bio);
8022 return __btrfs_subio_endio_read(inode, io_bio, err);
8026 static void btrfs_endio_direct_read(struct bio *bio)
8028 struct btrfs_dio_private *dip = bio->bi_private;
8029 struct inode *inode = dip->inode;
8030 struct bio *dio_bio;
8031 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8032 blk_status_t err = bio->bi_status;
8034 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8035 err = btrfs_subio_endio_read(inode, io_bio, err);
8037 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8038 dip->logical_offset + dip->bytes - 1);
8039 dio_bio = dip->dio_bio;
8043 dio_bio->bi_status = err;
8044 dio_end_io(dio_bio);
8047 io_bio->end_io(io_bio, blk_status_to_errno(err));
8051 static void __endio_write_update_ordered(struct inode *inode,
8052 const u64 offset, const u64 bytes,
8053 const bool uptodate)
8055 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8056 struct btrfs_ordered_extent *ordered = NULL;
8057 struct btrfs_workqueue *wq;
8058 btrfs_work_func_t func;
8059 u64 ordered_offset = offset;
8060 u64 ordered_bytes = bytes;
8063 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
8064 wq = fs_info->endio_freespace_worker;
8065 func = btrfs_freespace_write_helper;
8067 wq = fs_info->endio_write_workers;
8068 func = btrfs_endio_write_helper;
8071 while (ordered_offset < offset + bytes) {
8072 last_offset = ordered_offset;
8073 if (btrfs_dec_test_first_ordered_pending(inode, &ordered,
8077 btrfs_init_work(&ordered->work, func,
8080 btrfs_queue_work(wq, &ordered->work);
8083 * If btrfs_dec_test_ordered_pending does not find any ordered
8084 * extent in the range, we can exit.
8086 if (ordered_offset == last_offset)
8089 * Our bio might span multiple ordered extents. In this case
8090 * we keep goin until we have accounted the whole dio.
8092 if (ordered_offset < offset + bytes) {
8093 ordered_bytes = offset + bytes - ordered_offset;
8099 static void btrfs_endio_direct_write(struct bio *bio)
8101 struct btrfs_dio_private *dip = bio->bi_private;
8102 struct bio *dio_bio = dip->dio_bio;
8104 __endio_write_update_ordered(dip->inode, dip->logical_offset,
8105 dip->bytes, !bio->bi_status);
8109 dio_bio->bi_status = bio->bi_status;
8110 dio_end_io(dio_bio);
8114 static blk_status_t btrfs_submit_bio_start_direct_io(void *private_data,
8115 struct bio *bio, u64 offset)
8117 struct inode *inode = private_data;
8119 ret = btrfs_csum_one_bio(inode, bio, offset, 1);
8120 BUG_ON(ret); /* -ENOMEM */
8124 static void btrfs_end_dio_bio(struct bio *bio)
8126 struct btrfs_dio_private *dip = bio->bi_private;
8127 blk_status_t err = bio->bi_status;
8130 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8131 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8132 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8134 (unsigned long long)bio->bi_iter.bi_sector,
8135 bio->bi_iter.bi_size, err);
8137 if (dip->subio_endio)
8138 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8142 * We want to perceive the errors flag being set before
8143 * decrementing the reference count. We don't need a barrier
8144 * since atomic operations with a return value are fully
8145 * ordered as per atomic_t.txt
8150 /* if there are more bios still pending for this dio, just exit */
8151 if (!atomic_dec_and_test(&dip->pending_bios))
8155 bio_io_error(dip->orig_bio);
8157 dip->dio_bio->bi_status = BLK_STS_OK;
8158 bio_endio(dip->orig_bio);
8164 static inline blk_status_t btrfs_lookup_and_bind_dio_csum(struct inode *inode,
8165 struct btrfs_dio_private *dip,
8169 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8170 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8174 * We load all the csum data we need when we submit
8175 * the first bio to reduce the csum tree search and
8178 if (dip->logical_offset == file_offset) {
8179 ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio,
8185 if (bio == dip->orig_bio)
8188 file_offset -= dip->logical_offset;
8189 file_offset >>= inode->i_sb->s_blocksize_bits;
8190 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8195 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
8196 struct inode *inode, u64 file_offset, int async_submit)
8198 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8199 struct btrfs_dio_private *dip = bio->bi_private;
8200 bool write = bio_op(bio) == REQ_OP_WRITE;
8203 /* Check btrfs_submit_bio_hook() for rules about async submit. */
8205 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8208 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8213 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
8216 if (write && async_submit) {
8217 ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0,
8219 btrfs_submit_bio_start_direct_io);
8223 * If we aren't doing async submit, calculate the csum of the
8226 ret = btrfs_csum_one_bio(inode, bio, file_offset, 1);
8230 ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio,
8236 ret = btrfs_map_bio(fs_info, bio, 0, 0);
8241 static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip)
8243 struct inode *inode = dip->inode;
8244 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8246 struct bio *orig_bio = dip->orig_bio;
8247 u64 start_sector = orig_bio->bi_iter.bi_sector;
8248 u64 file_offset = dip->logical_offset;
8250 int async_submit = 0;
8252 int clone_offset = 0;
8255 blk_status_t status;
8257 map_length = orig_bio->bi_iter.bi_size;
8258 submit_len = map_length;
8259 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9,
8260 &map_length, NULL, 0);
8264 if (map_length >= submit_len) {
8266 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8270 /* async crcs make it difficult to collect full stripe writes. */
8271 if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8277 ASSERT(map_length <= INT_MAX);
8278 atomic_inc(&dip->pending_bios);
8280 clone_len = min_t(int, submit_len, map_length);
8283 * This will never fail as it's passing GPF_NOFS and
8284 * the allocation is backed by btrfs_bioset.
8286 bio = btrfs_bio_clone_partial(orig_bio, clone_offset,
8288 bio->bi_private = dip;
8289 bio->bi_end_io = btrfs_end_dio_bio;
8290 btrfs_io_bio(bio)->logical = file_offset;
8292 ASSERT(submit_len >= clone_len);
8293 submit_len -= clone_len;
8294 if (submit_len == 0)
8298 * Increase the count before we submit the bio so we know
8299 * the end IO handler won't happen before we increase the
8300 * count. Otherwise, the dip might get freed before we're
8301 * done setting it up.
8303 atomic_inc(&dip->pending_bios);
8305 status = btrfs_submit_dio_bio(bio, inode, file_offset,
8309 atomic_dec(&dip->pending_bios);
8313 clone_offset += clone_len;
8314 start_sector += clone_len >> 9;
8315 file_offset += clone_len;
8317 map_length = submit_len;
8318 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio),
8319 start_sector << 9, &map_length, NULL, 0);
8322 } while (submit_len > 0);
8325 status = btrfs_submit_dio_bio(bio, inode, file_offset, async_submit);
8333 * Before atomic variable goto zero, we must make sure dip->errors is
8334 * perceived to be set. This ordering is ensured by the fact that an
8335 * atomic operations with a return value are fully ordered as per
8338 if (atomic_dec_and_test(&dip->pending_bios))
8339 bio_io_error(dip->orig_bio);
8341 /* bio_end_io() will handle error, so we needn't return it */
8345 static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode,
8348 struct btrfs_dio_private *dip = NULL;
8349 struct bio *bio = NULL;
8350 struct btrfs_io_bio *io_bio;
8351 bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
8354 bio = btrfs_bio_clone(dio_bio);
8356 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8362 dip->private = dio_bio->bi_private;
8364 dip->logical_offset = file_offset;
8365 dip->bytes = dio_bio->bi_iter.bi_size;
8366 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8367 bio->bi_private = dip;
8368 dip->orig_bio = bio;
8369 dip->dio_bio = dio_bio;
8370 atomic_set(&dip->pending_bios, 0);
8371 io_bio = btrfs_io_bio(bio);
8372 io_bio->logical = file_offset;
8375 bio->bi_end_io = btrfs_endio_direct_write;
8377 bio->bi_end_io = btrfs_endio_direct_read;
8378 dip->subio_endio = btrfs_subio_endio_read;
8382 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8383 * even if we fail to submit a bio, because in such case we do the
8384 * corresponding error handling below and it must not be done a second
8385 * time by btrfs_direct_IO().
8388 struct btrfs_dio_data *dio_data = current->journal_info;
8390 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8392 dio_data->unsubmitted_oe_range_start =
8393 dio_data->unsubmitted_oe_range_end;
8396 ret = btrfs_submit_direct_hook(dip);
8401 io_bio->end_io(io_bio, ret);
8405 * If we arrived here it means either we failed to submit the dip
8406 * or we either failed to clone the dio_bio or failed to allocate the
8407 * dip. If we cloned the dio_bio and allocated the dip, we can just
8408 * call bio_endio against our io_bio so that we get proper resource
8409 * cleanup if we fail to submit the dip, otherwise, we must do the
8410 * same as btrfs_endio_direct_[write|read] because we can't call these
8411 * callbacks - they require an allocated dip and a clone of dio_bio.
8416 * The end io callbacks free our dip, do the final put on bio
8417 * and all the cleanup and final put for dio_bio (through
8424 __endio_write_update_ordered(inode,
8426 dio_bio->bi_iter.bi_size,
8429 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8430 file_offset + dio_bio->bi_iter.bi_size - 1);
8432 dio_bio->bi_status = BLK_STS_IOERR;
8434 * Releases and cleans up our dio_bio, no need to bio_put()
8435 * nor bio_endio()/bio_io_error() against dio_bio.
8437 dio_end_io(dio_bio);
8444 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
8445 const struct iov_iter *iter, loff_t offset)
8449 unsigned int blocksize_mask = fs_info->sectorsize - 1;
8450 ssize_t retval = -EINVAL;
8452 if (offset & blocksize_mask)
8455 if (iov_iter_alignment(iter) & blocksize_mask)
8458 /* If this is a write we don't need to check anymore */
8459 if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter))
8462 * Check to make sure we don't have duplicate iov_base's in this
8463 * iovec, if so return EINVAL, otherwise we'll get csum errors
8464 * when reading back.
8466 for (seg = 0; seg < iter->nr_segs; seg++) {
8467 for (i = seg + 1; i < iter->nr_segs; i++) {
8468 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8477 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8479 struct file *file = iocb->ki_filp;
8480 struct inode *inode = file->f_mapping->host;
8481 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8482 struct btrfs_dio_data dio_data = { 0 };
8483 struct extent_changeset *data_reserved = NULL;
8484 loff_t offset = iocb->ki_pos;
8488 bool relock = false;
8491 if (check_direct_IO(fs_info, iter, offset))
8494 inode_dio_begin(inode);
8497 * The generic stuff only does filemap_write_and_wait_range, which
8498 * isn't enough if we've written compressed pages to this area, so
8499 * we need to flush the dirty pages again to make absolutely sure
8500 * that any outstanding dirty pages are on disk.
8502 count = iov_iter_count(iter);
8503 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8504 &BTRFS_I(inode)->runtime_flags))
8505 filemap_fdatawrite_range(inode->i_mapping, offset,
8506 offset + count - 1);
8508 if (iov_iter_rw(iter) == WRITE) {
8510 * If the write DIO is beyond the EOF, we need update
8511 * the isize, but it is protected by i_mutex. So we can
8512 * not unlock the i_mutex at this case.
8514 if (offset + count <= inode->i_size) {
8515 dio_data.overwrite = 1;
8516 inode_unlock(inode);
8518 } else if (iocb->ki_flags & IOCB_NOWAIT) {
8522 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
8528 * We need to know how many extents we reserved so that we can
8529 * do the accounting properly if we go over the number we
8530 * originally calculated. Abuse current->journal_info for this.
8532 dio_data.reserve = round_up(count,
8533 fs_info->sectorsize);
8534 dio_data.unsubmitted_oe_range_start = (u64)offset;
8535 dio_data.unsubmitted_oe_range_end = (u64)offset;
8536 current->journal_info = &dio_data;
8537 down_read(&BTRFS_I(inode)->dio_sem);
8538 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8539 &BTRFS_I(inode)->runtime_flags)) {
8540 inode_dio_end(inode);
8541 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8545 ret = __blockdev_direct_IO(iocb, inode,
8546 fs_info->fs_devices->latest_bdev,
8547 iter, btrfs_get_blocks_direct, NULL,
8548 btrfs_submit_direct, flags);
8549 if (iov_iter_rw(iter) == WRITE) {
8550 up_read(&BTRFS_I(inode)->dio_sem);
8551 current->journal_info = NULL;
8552 if (ret < 0 && ret != -EIOCBQUEUED) {
8553 if (dio_data.reserve)
8554 btrfs_delalloc_release_space(inode, data_reserved,
8555 offset, dio_data.reserve, true);
8557 * On error we might have left some ordered extents
8558 * without submitting corresponding bios for them, so
8559 * cleanup them up to avoid other tasks getting them
8560 * and waiting for them to complete forever.
8562 if (dio_data.unsubmitted_oe_range_start <
8563 dio_data.unsubmitted_oe_range_end)
8564 __endio_write_update_ordered(inode,
8565 dio_data.unsubmitted_oe_range_start,
8566 dio_data.unsubmitted_oe_range_end -
8567 dio_data.unsubmitted_oe_range_start,
8569 } else if (ret >= 0 && (size_t)ret < count)
8570 btrfs_delalloc_release_space(inode, data_reserved,
8571 offset, count - (size_t)ret, true);
8572 btrfs_delalloc_release_extents(BTRFS_I(inode), count, false);
8576 inode_dio_end(inode);
8580 extent_changeset_free(data_reserved);
8584 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8586 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8587 __u64 start, __u64 len)
8591 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8595 return extent_fiemap(inode, fieinfo, start, len);
8598 int btrfs_readpage(struct file *file, struct page *page)
8600 struct extent_io_tree *tree;
8601 tree = &BTRFS_I(page->mapping->host)->io_tree;
8602 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8605 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8607 struct inode *inode = page->mapping->host;
8610 if (current->flags & PF_MEMALLOC) {
8611 redirty_page_for_writepage(wbc, page);
8617 * If we are under memory pressure we will call this directly from the
8618 * VM, we need to make sure we have the inode referenced for the ordered
8619 * extent. If not just return like we didn't do anything.
8621 if (!igrab(inode)) {
8622 redirty_page_for_writepage(wbc, page);
8623 return AOP_WRITEPAGE_ACTIVATE;
8625 ret = extent_write_full_page(page, wbc);
8626 btrfs_add_delayed_iput(inode);
8630 static int btrfs_writepages(struct address_space *mapping,
8631 struct writeback_control *wbc)
8633 return extent_writepages(mapping, wbc);
8637 btrfs_readpages(struct file *file, struct address_space *mapping,
8638 struct list_head *pages, unsigned nr_pages)
8640 return extent_readpages(mapping, pages, nr_pages);
8643 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8645 int ret = try_release_extent_mapping(page, gfp_flags);
8647 ClearPagePrivate(page);
8648 set_page_private(page, 0);
8654 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8656 if (PageWriteback(page) || PageDirty(page))
8658 return __btrfs_releasepage(page, gfp_flags);
8661 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8662 unsigned int length)
8664 struct inode *inode = page->mapping->host;
8665 struct extent_io_tree *tree;
8666 struct btrfs_ordered_extent *ordered;
8667 struct extent_state *cached_state = NULL;
8668 u64 page_start = page_offset(page);
8669 u64 page_end = page_start + PAGE_SIZE - 1;
8672 int inode_evicting = inode->i_state & I_FREEING;
8675 * we have the page locked, so new writeback can't start,
8676 * and the dirty bit won't be cleared while we are here.
8678 * Wait for IO on this page so that we can safely clear
8679 * the PagePrivate2 bit and do ordered accounting
8681 wait_on_page_writeback(page);
8683 tree = &BTRFS_I(inode)->io_tree;
8685 btrfs_releasepage(page, GFP_NOFS);
8689 if (!inode_evicting)
8690 lock_extent_bits(tree, page_start, page_end, &cached_state);
8693 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
8694 page_end - start + 1);
8696 end = min(page_end, ordered->file_offset + ordered->len - 1);
8698 * IO on this page will never be started, so we need
8699 * to account for any ordered extents now
8701 if (!inode_evicting)
8702 clear_extent_bit(tree, start, end,
8703 EXTENT_DIRTY | EXTENT_DELALLOC |
8704 EXTENT_DELALLOC_NEW |
8705 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8706 EXTENT_DEFRAG, 1, 0, &cached_state);
8708 * whoever cleared the private bit is responsible
8709 * for the finish_ordered_io
8711 if (TestClearPagePrivate2(page)) {
8712 struct btrfs_ordered_inode_tree *tree;
8715 tree = &BTRFS_I(inode)->ordered_tree;
8717 spin_lock_irq(&tree->lock);
8718 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8719 new_len = start - ordered->file_offset;
8720 if (new_len < ordered->truncated_len)
8721 ordered->truncated_len = new_len;
8722 spin_unlock_irq(&tree->lock);
8724 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8726 end - start + 1, 1))
8727 btrfs_finish_ordered_io(ordered);
8729 btrfs_put_ordered_extent(ordered);
8730 if (!inode_evicting) {
8731 cached_state = NULL;
8732 lock_extent_bits(tree, start, end,
8737 if (start < page_end)
8742 * Qgroup reserved space handler
8743 * Page here will be either
8744 * 1) Already written to disk
8745 * In this case, its reserved space is released from data rsv map
8746 * and will be freed by delayed_ref handler finally.
8747 * So even we call qgroup_free_data(), it won't decrease reserved
8749 * 2) Not written to disk
8750 * This means the reserved space should be freed here. However,
8751 * if a truncate invalidates the page (by clearing PageDirty)
8752 * and the page is accounted for while allocating extent
8753 * in btrfs_check_data_free_space() we let delayed_ref to
8754 * free the entire extent.
8756 if (PageDirty(page))
8757 btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
8758 if (!inode_evicting) {
8759 clear_extent_bit(tree, page_start, page_end,
8760 EXTENT_LOCKED | EXTENT_DIRTY |
8761 EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
8762 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
8765 __btrfs_releasepage(page, GFP_NOFS);
8768 ClearPageChecked(page);
8769 if (PagePrivate(page)) {
8770 ClearPagePrivate(page);
8771 set_page_private(page, 0);
8777 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8778 * called from a page fault handler when a page is first dirtied. Hence we must
8779 * be careful to check for EOF conditions here. We set the page up correctly
8780 * for a written page which means we get ENOSPC checking when writing into
8781 * holes and correct delalloc and unwritten extent mapping on filesystems that
8782 * support these features.
8784 * We are not allowed to take the i_mutex here so we have to play games to
8785 * protect against truncate races as the page could now be beyond EOF. Because
8786 * truncate_setsize() writes the inode size before removing pages, once we have
8787 * the page lock we can determine safely if the page is beyond EOF. If it is not
8788 * beyond EOF, then the page is guaranteed safe against truncation until we
8791 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8793 struct page *page = vmf->page;
8794 struct inode *inode = file_inode(vmf->vma->vm_file);
8795 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8796 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8797 struct btrfs_ordered_extent *ordered;
8798 struct extent_state *cached_state = NULL;
8799 struct extent_changeset *data_reserved = NULL;
8801 unsigned long zero_start;
8811 reserved_space = PAGE_SIZE;
8813 sb_start_pagefault(inode->i_sb);
8814 page_start = page_offset(page);
8815 page_end = page_start + PAGE_SIZE - 1;
8819 * Reserving delalloc space after obtaining the page lock can lead to
8820 * deadlock. For example, if a dirty page is locked by this function
8821 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8822 * dirty page write out, then the btrfs_writepage() function could
8823 * end up waiting indefinitely to get a lock on the page currently
8824 * being processed by btrfs_page_mkwrite() function.
8826 ret2 = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
8829 ret2 = file_update_time(vmf->vma->vm_file);
8833 ret = vmf_error(ret2);
8839 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8842 size = i_size_read(inode);
8844 if ((page->mapping != inode->i_mapping) ||
8845 (page_start >= size)) {
8846 /* page got truncated out from underneath us */
8849 wait_on_page_writeback(page);
8851 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8852 set_page_extent_mapped(page);
8855 * we can't set the delalloc bits if there are pending ordered
8856 * extents. Drop our locks and wait for them to finish
8858 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8861 unlock_extent_cached(io_tree, page_start, page_end,
8864 btrfs_start_ordered_extent(inode, ordered, 1);
8865 btrfs_put_ordered_extent(ordered);
8869 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8870 reserved_space = round_up(size - page_start,
8871 fs_info->sectorsize);
8872 if (reserved_space < PAGE_SIZE) {
8873 end = page_start + reserved_space - 1;
8874 btrfs_delalloc_release_space(inode, data_reserved,
8875 page_start, PAGE_SIZE - reserved_space,
8881 * page_mkwrite gets called when the page is firstly dirtied after it's
8882 * faulted in, but write(2) could also dirty a page and set delalloc
8883 * bits, thus in this case for space account reason, we still need to
8884 * clear any delalloc bits within this page range since we have to
8885 * reserve data&meta space before lock_page() (see above comments).
8887 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8888 EXTENT_DIRTY | EXTENT_DELALLOC |
8889 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8890 0, 0, &cached_state);
8892 ret2 = btrfs_set_extent_delalloc(inode, page_start, end, 0,
8895 unlock_extent_cached(io_tree, page_start, page_end,
8897 ret = VM_FAULT_SIGBUS;
8902 /* page is wholly or partially inside EOF */
8903 if (page_start + PAGE_SIZE > size)
8904 zero_start = size & ~PAGE_MASK;
8906 zero_start = PAGE_SIZE;
8908 if (zero_start != PAGE_SIZE) {
8910 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
8911 flush_dcache_page(page);
8914 ClearPageChecked(page);
8915 set_page_dirty(page);
8916 SetPageUptodate(page);
8918 BTRFS_I(inode)->last_trans = fs_info->generation;
8919 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8920 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8922 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8925 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, true);
8926 sb_end_pagefault(inode->i_sb);
8927 extent_changeset_free(data_reserved);
8928 return VM_FAULT_LOCKED;
8934 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, (ret != 0));
8935 btrfs_delalloc_release_space(inode, data_reserved, page_start,
8936 reserved_space, (ret != 0));
8938 sb_end_pagefault(inode->i_sb);
8939 extent_changeset_free(data_reserved);
8943 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
8945 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8946 struct btrfs_root *root = BTRFS_I(inode)->root;
8947 struct btrfs_block_rsv *rsv;
8949 struct btrfs_trans_handle *trans;
8950 u64 mask = fs_info->sectorsize - 1;
8951 u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
8953 if (!skip_writeback) {
8954 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8961 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8962 * things going on here:
8964 * 1) We need to reserve space to update our inode.
8966 * 2) We need to have something to cache all the space that is going to
8967 * be free'd up by the truncate operation, but also have some slack
8968 * space reserved in case it uses space during the truncate (thank you
8969 * very much snapshotting).
8971 * And we need these to be separate. The fact is we can use a lot of
8972 * space doing the truncate, and we have no earthly idea how much space
8973 * we will use, so we need the truncate reservation to be separate so it
8974 * doesn't end up using space reserved for updating the inode. We also
8975 * need to be able to stop the transaction and start a new one, which
8976 * means we need to be able to update the inode several times, and we
8977 * have no idea of knowing how many times that will be, so we can't just
8978 * reserve 1 item for the entirety of the operation, so that has to be
8979 * done separately as well.
8981 * So that leaves us with
8983 * 1) rsv - for the truncate reservation, which we will steal from the
8984 * transaction reservation.
8985 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8986 * updating the inode.
8988 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8991 rsv->size = min_size;
8995 * 1 for the truncate slack space
8996 * 1 for updating the inode.
8998 trans = btrfs_start_transaction(root, 2);
8999 if (IS_ERR(trans)) {
9000 ret = PTR_ERR(trans);
9004 /* Migrate the slack space for the truncate to our reserve */
9005 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
9010 * So if we truncate and then write and fsync we normally would just
9011 * write the extents that changed, which is a problem if we need to
9012 * first truncate that entire inode. So set this flag so we write out
9013 * all of the extents in the inode to the sync log so we're completely
9016 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9017 trans->block_rsv = rsv;
9020 ret = btrfs_truncate_inode_items(trans, root, inode,
9022 BTRFS_EXTENT_DATA_KEY);
9023 trans->block_rsv = &fs_info->trans_block_rsv;
9024 if (ret != -ENOSPC && ret != -EAGAIN)
9027 ret = btrfs_update_inode(trans, root, inode);
9031 btrfs_end_transaction(trans);
9032 btrfs_btree_balance_dirty(fs_info);
9034 trans = btrfs_start_transaction(root, 2);
9035 if (IS_ERR(trans)) {
9036 ret = PTR_ERR(trans);
9041 btrfs_block_rsv_release(fs_info, rsv, -1);
9042 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
9043 rsv, min_size, false);
9044 BUG_ON(ret); /* shouldn't happen */
9045 trans->block_rsv = rsv;
9049 * We can't call btrfs_truncate_block inside a trans handle as we could
9050 * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
9051 * we've truncated everything except the last little bit, and can do
9052 * btrfs_truncate_block and then update the disk_i_size.
9054 if (ret == NEED_TRUNCATE_BLOCK) {
9055 btrfs_end_transaction(trans);
9056 btrfs_btree_balance_dirty(fs_info);
9058 ret = btrfs_truncate_block(inode, inode->i_size, 0, 0);
9061 trans = btrfs_start_transaction(root, 1);
9062 if (IS_ERR(trans)) {
9063 ret = PTR_ERR(trans);
9066 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
9072 trans->block_rsv = &fs_info->trans_block_rsv;
9073 ret2 = btrfs_update_inode(trans, root, inode);
9077 ret2 = btrfs_end_transaction(trans);
9080 btrfs_btree_balance_dirty(fs_info);
9083 btrfs_free_block_rsv(fs_info, rsv);
9089 * create a new subvolume directory/inode (helper for the ioctl).
9091 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9092 struct btrfs_root *new_root,
9093 struct btrfs_root *parent_root,
9096 struct inode *inode;
9100 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9101 new_dirid, new_dirid,
9102 S_IFDIR | (~current_umask() & S_IRWXUGO),
9105 return PTR_ERR(inode);
9106 inode->i_op = &btrfs_dir_inode_operations;
9107 inode->i_fop = &btrfs_dir_file_operations;
9109 set_nlink(inode, 1);
9110 btrfs_i_size_write(BTRFS_I(inode), 0);
9111 unlock_new_inode(inode);
9113 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9115 btrfs_err(new_root->fs_info,
9116 "error inheriting subvolume %llu properties: %d",
9117 new_root->root_key.objectid, err);
9119 err = btrfs_update_inode(trans, new_root, inode);
9125 struct inode *btrfs_alloc_inode(struct super_block *sb)
9127 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
9128 struct btrfs_inode *ei;
9129 struct inode *inode;
9131 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL);
9138 ei->last_sub_trans = 0;
9139 ei->logged_trans = 0;
9140 ei->delalloc_bytes = 0;
9141 ei->new_delalloc_bytes = 0;
9142 ei->defrag_bytes = 0;
9143 ei->disk_i_size = 0;
9146 ei->index_cnt = (u64)-1;
9148 ei->last_unlink_trans = 0;
9149 ei->last_log_commit = 0;
9151 spin_lock_init(&ei->lock);
9152 ei->outstanding_extents = 0;
9153 if (sb->s_magic != BTRFS_TEST_MAGIC)
9154 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
9155 BTRFS_BLOCK_RSV_DELALLOC);
9156 ei->runtime_flags = 0;
9157 ei->prop_compress = BTRFS_COMPRESS_NONE;
9158 ei->defrag_compress = BTRFS_COMPRESS_NONE;
9160 ei->delayed_node = NULL;
9162 ei->i_otime.tv_sec = 0;
9163 ei->i_otime.tv_nsec = 0;
9165 inode = &ei->vfs_inode;
9166 extent_map_tree_init(&ei->extent_tree);
9167 extent_io_tree_init(&ei->io_tree, inode);
9168 extent_io_tree_init(&ei->io_failure_tree, inode);
9169 ei->io_tree.track_uptodate = 1;
9170 ei->io_failure_tree.track_uptodate = 1;
9171 atomic_set(&ei->sync_writers, 0);
9172 mutex_init(&ei->log_mutex);
9173 mutex_init(&ei->delalloc_mutex);
9174 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9175 INIT_LIST_HEAD(&ei->delalloc_inodes);
9176 INIT_LIST_HEAD(&ei->delayed_iput);
9177 RB_CLEAR_NODE(&ei->rb_node);
9178 init_rwsem(&ei->dio_sem);
9183 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9184 void btrfs_test_destroy_inode(struct inode *inode)
9186 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9187 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9191 static void btrfs_i_callback(struct rcu_head *head)
9193 struct inode *inode = container_of(head, struct inode, i_rcu);
9194 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9197 void btrfs_destroy_inode(struct inode *inode)
9199 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9200 struct btrfs_ordered_extent *ordered;
9201 struct btrfs_root *root = BTRFS_I(inode)->root;
9203 WARN_ON(!hlist_empty(&inode->i_dentry));
9204 WARN_ON(inode->i_data.nrpages);
9205 WARN_ON(BTRFS_I(inode)->block_rsv.reserved);
9206 WARN_ON(BTRFS_I(inode)->block_rsv.size);
9207 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9208 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9209 WARN_ON(BTRFS_I(inode)->new_delalloc_bytes);
9210 WARN_ON(BTRFS_I(inode)->csum_bytes);
9211 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9214 * This can happen where we create an inode, but somebody else also
9215 * created the same inode and we need to destroy the one we already
9222 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9227 "found ordered extent %llu %llu on inode cleanup",
9228 ordered->file_offset, ordered->len);
9229 btrfs_remove_ordered_extent(inode, ordered);
9230 btrfs_put_ordered_extent(ordered);
9231 btrfs_put_ordered_extent(ordered);
9234 btrfs_qgroup_check_reserved_leak(inode);
9235 inode_tree_del(inode);
9236 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9238 call_rcu(&inode->i_rcu, btrfs_i_callback);
9241 int btrfs_drop_inode(struct inode *inode)
9243 struct btrfs_root *root = BTRFS_I(inode)->root;
9248 /* the snap/subvol tree is on deleting */
9249 if (btrfs_root_refs(&root->root_item) == 0)
9252 return generic_drop_inode(inode);
9255 static void init_once(void *foo)
9257 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9259 inode_init_once(&ei->vfs_inode);
9262 void __cold btrfs_destroy_cachep(void)
9265 * Make sure all delayed rcu free inodes are flushed before we
9269 kmem_cache_destroy(btrfs_inode_cachep);
9270 kmem_cache_destroy(btrfs_trans_handle_cachep);
9271 kmem_cache_destroy(btrfs_path_cachep);
9272 kmem_cache_destroy(btrfs_free_space_cachep);
9275 int __init btrfs_init_cachep(void)
9277 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9278 sizeof(struct btrfs_inode), 0,
9279 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9281 if (!btrfs_inode_cachep)
9284 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9285 sizeof(struct btrfs_trans_handle), 0,
9286 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9287 if (!btrfs_trans_handle_cachep)
9290 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9291 sizeof(struct btrfs_path), 0,
9292 SLAB_MEM_SPREAD, NULL);
9293 if (!btrfs_path_cachep)
9296 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9297 sizeof(struct btrfs_free_space), 0,
9298 SLAB_MEM_SPREAD, NULL);
9299 if (!btrfs_free_space_cachep)
9304 btrfs_destroy_cachep();
9308 static int btrfs_getattr(const struct path *path, struct kstat *stat,
9309 u32 request_mask, unsigned int flags)
9312 struct inode *inode = d_inode(path->dentry);
9313 u32 blocksize = inode->i_sb->s_blocksize;
9314 u32 bi_flags = BTRFS_I(inode)->flags;
9316 stat->result_mask |= STATX_BTIME;
9317 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9318 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9319 if (bi_flags & BTRFS_INODE_APPEND)
9320 stat->attributes |= STATX_ATTR_APPEND;
9321 if (bi_flags & BTRFS_INODE_COMPRESS)
9322 stat->attributes |= STATX_ATTR_COMPRESSED;
9323 if (bi_flags & BTRFS_INODE_IMMUTABLE)
9324 stat->attributes |= STATX_ATTR_IMMUTABLE;
9325 if (bi_flags & BTRFS_INODE_NODUMP)
9326 stat->attributes |= STATX_ATTR_NODUMP;
9328 stat->attributes_mask |= (STATX_ATTR_APPEND |
9329 STATX_ATTR_COMPRESSED |
9330 STATX_ATTR_IMMUTABLE |
9333 generic_fillattr(inode, stat);
9334 stat->dev = BTRFS_I(inode)->root->anon_dev;
9336 spin_lock(&BTRFS_I(inode)->lock);
9337 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9338 spin_unlock(&BTRFS_I(inode)->lock);
9339 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9340 ALIGN(delalloc_bytes, blocksize)) >> 9;
9344 static int btrfs_rename_exchange(struct inode *old_dir,
9345 struct dentry *old_dentry,
9346 struct inode *new_dir,
9347 struct dentry *new_dentry)
9349 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9350 struct btrfs_trans_handle *trans;
9351 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9352 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9353 struct inode *new_inode = new_dentry->d_inode;
9354 struct inode *old_inode = old_dentry->d_inode;
9355 struct timespec64 ctime = current_time(old_inode);
9356 struct dentry *parent;
9357 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9358 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9363 bool root_log_pinned = false;
9364 bool dest_log_pinned = false;
9365 struct btrfs_log_ctx ctx_root;
9366 struct btrfs_log_ctx ctx_dest;
9367 bool sync_log_root = false;
9368 bool sync_log_dest = false;
9369 bool commit_transaction = false;
9371 /* we only allow rename subvolume link between subvolumes */
9372 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9375 btrfs_init_log_ctx(&ctx_root, old_inode);
9376 btrfs_init_log_ctx(&ctx_dest, new_inode);
9378 /* close the race window with snapshot create/destroy ioctl */
9379 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9380 down_read(&fs_info->subvol_sem);
9381 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9382 down_read(&fs_info->subvol_sem);
9385 * We want to reserve the absolute worst case amount of items. So if
9386 * both inodes are subvols and we need to unlink them then that would
9387 * require 4 item modifications, but if they are both normal inodes it
9388 * would require 5 item modifications, so we'll assume their normal
9389 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9390 * should cover the worst case number of items we'll modify.
9392 trans = btrfs_start_transaction(root, 12);
9393 if (IS_ERR(trans)) {
9394 ret = PTR_ERR(trans);
9399 * We need to find a free sequence number both in the source and
9400 * in the destination directory for the exchange.
9402 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9405 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9409 BTRFS_I(old_inode)->dir_index = 0ULL;
9410 BTRFS_I(new_inode)->dir_index = 0ULL;
9412 /* Reference for the source. */
9413 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9414 /* force full log commit if subvolume involved. */
9415 btrfs_set_log_full_commit(fs_info, trans);
9417 btrfs_pin_log_trans(root);
9418 root_log_pinned = true;
9419 ret = btrfs_insert_inode_ref(trans, dest,
9420 new_dentry->d_name.name,
9421 new_dentry->d_name.len,
9423 btrfs_ino(BTRFS_I(new_dir)),
9429 /* And now for the dest. */
9430 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9431 /* force full log commit if subvolume involved. */
9432 btrfs_set_log_full_commit(fs_info, trans);
9434 btrfs_pin_log_trans(dest);
9435 dest_log_pinned = true;
9436 ret = btrfs_insert_inode_ref(trans, root,
9437 old_dentry->d_name.name,
9438 old_dentry->d_name.len,
9440 btrfs_ino(BTRFS_I(old_dir)),
9446 /* Update inode version and ctime/mtime. */
9447 inode_inc_iversion(old_dir);
9448 inode_inc_iversion(new_dir);
9449 inode_inc_iversion(old_inode);
9450 inode_inc_iversion(new_inode);
9451 old_dir->i_ctime = old_dir->i_mtime = ctime;
9452 new_dir->i_ctime = new_dir->i_mtime = ctime;
9453 old_inode->i_ctime = ctime;
9454 new_inode->i_ctime = ctime;
9456 if (old_dentry->d_parent != new_dentry->d_parent) {
9457 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9458 BTRFS_I(old_inode), 1);
9459 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9460 BTRFS_I(new_inode), 1);
9463 /* src is a subvolume */
9464 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9465 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9466 ret = btrfs_unlink_subvol(trans, old_dir, root_objectid,
9467 old_dentry->d_name.name,
9468 old_dentry->d_name.len);
9469 } else { /* src is an inode */
9470 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9471 BTRFS_I(old_dentry->d_inode),
9472 old_dentry->d_name.name,
9473 old_dentry->d_name.len);
9475 ret = btrfs_update_inode(trans, root, old_inode);
9478 btrfs_abort_transaction(trans, ret);
9482 /* dest is a subvolume */
9483 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9484 root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9485 ret = btrfs_unlink_subvol(trans, new_dir, root_objectid,
9486 new_dentry->d_name.name,
9487 new_dentry->d_name.len);
9488 } else { /* dest is an inode */
9489 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9490 BTRFS_I(new_dentry->d_inode),
9491 new_dentry->d_name.name,
9492 new_dentry->d_name.len);
9494 ret = btrfs_update_inode(trans, dest, new_inode);
9497 btrfs_abort_transaction(trans, ret);
9501 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9502 new_dentry->d_name.name,
9503 new_dentry->d_name.len, 0, old_idx);
9505 btrfs_abort_transaction(trans, ret);
9509 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9510 old_dentry->d_name.name,
9511 old_dentry->d_name.len, 0, new_idx);
9513 btrfs_abort_transaction(trans, ret);
9517 if (old_inode->i_nlink == 1)
9518 BTRFS_I(old_inode)->dir_index = old_idx;
9519 if (new_inode->i_nlink == 1)
9520 BTRFS_I(new_inode)->dir_index = new_idx;
9522 if (root_log_pinned) {
9523 parent = new_dentry->d_parent;
9524 ret = btrfs_log_new_name(trans, BTRFS_I(old_inode),
9525 BTRFS_I(old_dir), parent,
9527 if (ret == BTRFS_NEED_LOG_SYNC)
9528 sync_log_root = true;
9529 else if (ret == BTRFS_NEED_TRANS_COMMIT)
9530 commit_transaction = true;
9532 btrfs_end_log_trans(root);
9533 root_log_pinned = false;
9535 if (dest_log_pinned) {
9536 if (!commit_transaction) {
9537 parent = old_dentry->d_parent;
9538 ret = btrfs_log_new_name(trans, BTRFS_I(new_inode),
9539 BTRFS_I(new_dir), parent,
9541 if (ret == BTRFS_NEED_LOG_SYNC)
9542 sync_log_dest = true;
9543 else if (ret == BTRFS_NEED_TRANS_COMMIT)
9544 commit_transaction = true;
9547 btrfs_end_log_trans(dest);
9548 dest_log_pinned = false;
9552 * If we have pinned a log and an error happened, we unpin tasks
9553 * trying to sync the log and force them to fallback to a transaction
9554 * commit if the log currently contains any of the inodes involved in
9555 * this rename operation (to ensure we do not persist a log with an
9556 * inconsistent state for any of these inodes or leading to any
9557 * inconsistencies when replayed). If the transaction was aborted, the
9558 * abortion reason is propagated to userspace when attempting to commit
9559 * the transaction. If the log does not contain any of these inodes, we
9560 * allow the tasks to sync it.
9562 if (ret && (root_log_pinned || dest_log_pinned)) {
9563 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9564 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9565 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9567 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9568 btrfs_set_log_full_commit(fs_info, trans);
9570 if (root_log_pinned) {
9571 btrfs_end_log_trans(root);
9572 root_log_pinned = false;
9574 if (dest_log_pinned) {
9575 btrfs_end_log_trans(dest);
9576 dest_log_pinned = false;
9579 if (!ret && sync_log_root && !commit_transaction) {
9580 ret = btrfs_sync_log(trans, BTRFS_I(old_inode)->root,
9583 commit_transaction = true;
9585 if (!ret && sync_log_dest && !commit_transaction) {
9586 ret = btrfs_sync_log(trans, BTRFS_I(new_inode)->root,
9589 commit_transaction = true;
9591 if (commit_transaction) {
9592 ret = btrfs_commit_transaction(trans);
9596 ret2 = btrfs_end_transaction(trans);
9597 ret = ret ? ret : ret2;
9600 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9601 up_read(&fs_info->subvol_sem);
9602 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9603 up_read(&fs_info->subvol_sem);
9608 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9609 struct btrfs_root *root,
9611 struct dentry *dentry)
9614 struct inode *inode;
9618 ret = btrfs_find_free_ino(root, &objectid);
9622 inode = btrfs_new_inode(trans, root, dir,
9623 dentry->d_name.name,
9625 btrfs_ino(BTRFS_I(dir)),
9627 S_IFCHR | WHITEOUT_MODE,
9630 if (IS_ERR(inode)) {
9631 ret = PTR_ERR(inode);
9635 inode->i_op = &btrfs_special_inode_operations;
9636 init_special_inode(inode, inode->i_mode,
9639 ret = btrfs_init_inode_security(trans, inode, dir,
9644 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9645 BTRFS_I(inode), 0, index);
9649 ret = btrfs_update_inode(trans, root, inode);
9651 unlock_new_inode(inode);
9653 inode_dec_link_count(inode);
9659 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9660 struct inode *new_dir, struct dentry *new_dentry,
9663 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9664 struct btrfs_trans_handle *trans;
9665 unsigned int trans_num_items;
9666 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9667 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9668 struct inode *new_inode = d_inode(new_dentry);
9669 struct inode *old_inode = d_inode(old_dentry);
9673 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9674 bool log_pinned = false;
9675 struct btrfs_log_ctx ctx;
9676 bool sync_log = false;
9677 bool commit_transaction = false;
9679 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9682 /* we only allow rename subvolume link between subvolumes */
9683 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9686 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9687 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9690 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9691 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9695 /* check for collisions, even if the name isn't there */
9696 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9697 new_dentry->d_name.name,
9698 new_dentry->d_name.len);
9701 if (ret == -EEXIST) {
9703 * eexist without a new_inode */
9704 if (WARN_ON(!new_inode)) {
9708 /* maybe -EOVERFLOW */
9715 * we're using rename to replace one file with another. Start IO on it
9716 * now so we don't add too much work to the end of the transaction
9718 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9719 filemap_flush(old_inode->i_mapping);
9721 /* close the racy window with snapshot create/destroy ioctl */
9722 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9723 down_read(&fs_info->subvol_sem);
9725 * We want to reserve the absolute worst case amount of items. So if
9726 * both inodes are subvols and we need to unlink them then that would
9727 * require 4 item modifications, but if they are both normal inodes it
9728 * would require 5 item modifications, so we'll assume they are normal
9729 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9730 * should cover the worst case number of items we'll modify.
9731 * If our rename has the whiteout flag, we need more 5 units for the
9732 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9733 * when selinux is enabled).
9735 trans_num_items = 11;
9736 if (flags & RENAME_WHITEOUT)
9737 trans_num_items += 5;
9738 trans = btrfs_start_transaction(root, trans_num_items);
9739 if (IS_ERR(trans)) {
9740 ret = PTR_ERR(trans);
9745 btrfs_record_root_in_trans(trans, dest);
9747 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9751 BTRFS_I(old_inode)->dir_index = 0ULL;
9752 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9753 /* force full log commit if subvolume involved. */
9754 btrfs_set_log_full_commit(fs_info, trans);
9756 btrfs_pin_log_trans(root);
9758 ret = btrfs_insert_inode_ref(trans, dest,
9759 new_dentry->d_name.name,
9760 new_dentry->d_name.len,
9762 btrfs_ino(BTRFS_I(new_dir)), index);
9767 inode_inc_iversion(old_dir);
9768 inode_inc_iversion(new_dir);
9769 inode_inc_iversion(old_inode);
9770 old_dir->i_ctime = old_dir->i_mtime =
9771 new_dir->i_ctime = new_dir->i_mtime =
9772 old_inode->i_ctime = current_time(old_dir);
9774 if (old_dentry->d_parent != new_dentry->d_parent)
9775 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9776 BTRFS_I(old_inode), 1);
9778 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9779 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9780 ret = btrfs_unlink_subvol(trans, old_dir, root_objectid,
9781 old_dentry->d_name.name,
9782 old_dentry->d_name.len);
9784 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9785 BTRFS_I(d_inode(old_dentry)),
9786 old_dentry->d_name.name,
9787 old_dentry->d_name.len);
9789 ret = btrfs_update_inode(trans, root, old_inode);
9792 btrfs_abort_transaction(trans, ret);
9797 inode_inc_iversion(new_inode);
9798 new_inode->i_ctime = current_time(new_inode);
9799 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9800 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9801 root_objectid = BTRFS_I(new_inode)->location.objectid;
9802 ret = btrfs_unlink_subvol(trans, new_dir, root_objectid,
9803 new_dentry->d_name.name,
9804 new_dentry->d_name.len);
9805 BUG_ON(new_inode->i_nlink == 0);
9807 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9808 BTRFS_I(d_inode(new_dentry)),
9809 new_dentry->d_name.name,
9810 new_dentry->d_name.len);
9812 if (!ret && new_inode->i_nlink == 0)
9813 ret = btrfs_orphan_add(trans,
9814 BTRFS_I(d_inode(new_dentry)));
9816 btrfs_abort_transaction(trans, ret);
9821 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9822 new_dentry->d_name.name,
9823 new_dentry->d_name.len, 0, index);
9825 btrfs_abort_transaction(trans, ret);
9829 if (old_inode->i_nlink == 1)
9830 BTRFS_I(old_inode)->dir_index = index;
9833 struct dentry *parent = new_dentry->d_parent;
9835 btrfs_init_log_ctx(&ctx, old_inode);
9836 ret = btrfs_log_new_name(trans, BTRFS_I(old_inode),
9837 BTRFS_I(old_dir), parent,
9839 if (ret == BTRFS_NEED_LOG_SYNC)
9841 else if (ret == BTRFS_NEED_TRANS_COMMIT)
9842 commit_transaction = true;
9844 btrfs_end_log_trans(root);
9848 if (flags & RENAME_WHITEOUT) {
9849 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9853 btrfs_abort_transaction(trans, ret);
9859 * If we have pinned the log and an error happened, we unpin tasks
9860 * trying to sync the log and force them to fallback to a transaction
9861 * commit if the log currently contains any of the inodes involved in
9862 * this rename operation (to ensure we do not persist a log with an
9863 * inconsistent state for any of these inodes or leading to any
9864 * inconsistencies when replayed). If the transaction was aborted, the
9865 * abortion reason is propagated to userspace when attempting to commit
9866 * the transaction. If the log does not contain any of these inodes, we
9867 * allow the tasks to sync it.
9869 if (ret && log_pinned) {
9870 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9871 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9872 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9874 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9875 btrfs_set_log_full_commit(fs_info, trans);
9877 btrfs_end_log_trans(root);
9880 if (!ret && sync_log) {
9881 ret = btrfs_sync_log(trans, BTRFS_I(old_inode)->root, &ctx);
9883 commit_transaction = true;
9885 if (commit_transaction) {
9886 ret = btrfs_commit_transaction(trans);
9890 ret2 = btrfs_end_transaction(trans);
9891 ret = ret ? ret : ret2;
9894 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9895 up_read(&fs_info->subvol_sem);
9900 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9901 struct inode *new_dir, struct dentry *new_dentry,
9904 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9907 if (flags & RENAME_EXCHANGE)
9908 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9911 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9914 struct btrfs_delalloc_work {
9915 struct inode *inode;
9916 struct completion completion;
9917 struct list_head list;
9918 struct btrfs_work work;
9921 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9923 struct btrfs_delalloc_work *delalloc_work;
9924 struct inode *inode;
9926 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9928 inode = delalloc_work->inode;
9929 filemap_flush(inode->i_mapping);
9930 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9931 &BTRFS_I(inode)->runtime_flags))
9932 filemap_flush(inode->i_mapping);
9935 complete(&delalloc_work->completion);
9938 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9940 struct btrfs_delalloc_work *work;
9942 work = kmalloc(sizeof(*work), GFP_NOFS);
9946 init_completion(&work->completion);
9947 INIT_LIST_HEAD(&work->list);
9948 work->inode = inode;
9949 WARN_ON_ONCE(!inode);
9950 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9951 btrfs_run_delalloc_work, NULL, NULL);
9957 * some fairly slow code that needs optimization. This walks the list
9958 * of all the inodes with pending delalloc and forces them to disk.
9960 static int start_delalloc_inodes(struct btrfs_root *root, int nr)
9962 struct btrfs_inode *binode;
9963 struct inode *inode;
9964 struct btrfs_delalloc_work *work, *next;
9965 struct list_head works;
9966 struct list_head splice;
9969 INIT_LIST_HEAD(&works);
9970 INIT_LIST_HEAD(&splice);
9972 mutex_lock(&root->delalloc_mutex);
9973 spin_lock(&root->delalloc_lock);
9974 list_splice_init(&root->delalloc_inodes, &splice);
9975 while (!list_empty(&splice)) {
9976 binode = list_entry(splice.next, struct btrfs_inode,
9979 list_move_tail(&binode->delalloc_inodes,
9980 &root->delalloc_inodes);
9981 inode = igrab(&binode->vfs_inode);
9983 cond_resched_lock(&root->delalloc_lock);
9986 spin_unlock(&root->delalloc_lock);
9988 work = btrfs_alloc_delalloc_work(inode);
9994 list_add_tail(&work->list, &works);
9995 btrfs_queue_work(root->fs_info->flush_workers,
9998 if (nr != -1 && ret >= nr)
10001 spin_lock(&root->delalloc_lock);
10003 spin_unlock(&root->delalloc_lock);
10006 list_for_each_entry_safe(work, next, &works, list) {
10007 list_del_init(&work->list);
10008 wait_for_completion(&work->completion);
10012 if (!list_empty(&splice)) {
10013 spin_lock(&root->delalloc_lock);
10014 list_splice_tail(&splice, &root->delalloc_inodes);
10015 spin_unlock(&root->delalloc_lock);
10017 mutex_unlock(&root->delalloc_mutex);
10021 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
10023 struct btrfs_fs_info *fs_info = root->fs_info;
10026 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10029 ret = start_delalloc_inodes(root, -1);
10035 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int nr)
10037 struct btrfs_root *root;
10038 struct list_head splice;
10041 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10044 INIT_LIST_HEAD(&splice);
10046 mutex_lock(&fs_info->delalloc_root_mutex);
10047 spin_lock(&fs_info->delalloc_root_lock);
10048 list_splice_init(&fs_info->delalloc_roots, &splice);
10049 while (!list_empty(&splice) && nr) {
10050 root = list_first_entry(&splice, struct btrfs_root,
10052 root = btrfs_grab_fs_root(root);
10054 list_move_tail(&root->delalloc_root,
10055 &fs_info->delalloc_roots);
10056 spin_unlock(&fs_info->delalloc_root_lock);
10058 ret = start_delalloc_inodes(root, nr);
10059 btrfs_put_fs_root(root);
10067 spin_lock(&fs_info->delalloc_root_lock);
10069 spin_unlock(&fs_info->delalloc_root_lock);
10073 if (!list_empty(&splice)) {
10074 spin_lock(&fs_info->delalloc_root_lock);
10075 list_splice_tail(&splice, &fs_info->delalloc_roots);
10076 spin_unlock(&fs_info->delalloc_root_lock);
10078 mutex_unlock(&fs_info->delalloc_root_mutex);
10082 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10083 const char *symname)
10085 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10086 struct btrfs_trans_handle *trans;
10087 struct btrfs_root *root = BTRFS_I(dir)->root;
10088 struct btrfs_path *path;
10089 struct btrfs_key key;
10090 struct inode *inode = NULL;
10097 struct btrfs_file_extent_item *ei;
10098 struct extent_buffer *leaf;
10100 name_len = strlen(symname);
10101 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
10102 return -ENAMETOOLONG;
10105 * 2 items for inode item and ref
10106 * 2 items for dir items
10107 * 1 item for updating parent inode item
10108 * 1 item for the inline extent item
10109 * 1 item for xattr if selinux is on
10111 trans = btrfs_start_transaction(root, 7);
10113 return PTR_ERR(trans);
10115 err = btrfs_find_free_ino(root, &objectid);
10119 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10120 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
10121 objectid, S_IFLNK|S_IRWXUGO, &index);
10122 if (IS_ERR(inode)) {
10123 err = PTR_ERR(inode);
10129 * If the active LSM wants to access the inode during
10130 * d_instantiate it needs these. Smack checks to see
10131 * if the filesystem supports xattrs by looking at the
10134 inode->i_fop = &btrfs_file_operations;
10135 inode->i_op = &btrfs_file_inode_operations;
10136 inode->i_mapping->a_ops = &btrfs_aops;
10137 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10139 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10143 path = btrfs_alloc_path();
10148 key.objectid = btrfs_ino(BTRFS_I(inode));
10150 key.type = BTRFS_EXTENT_DATA_KEY;
10151 datasize = btrfs_file_extent_calc_inline_size(name_len);
10152 err = btrfs_insert_empty_item(trans, root, path, &key,
10155 btrfs_free_path(path);
10158 leaf = path->nodes[0];
10159 ei = btrfs_item_ptr(leaf, path->slots[0],
10160 struct btrfs_file_extent_item);
10161 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10162 btrfs_set_file_extent_type(leaf, ei,
10163 BTRFS_FILE_EXTENT_INLINE);
10164 btrfs_set_file_extent_encryption(leaf, ei, 0);
10165 btrfs_set_file_extent_compression(leaf, ei, 0);
10166 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10167 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10169 ptr = btrfs_file_extent_inline_start(ei);
10170 write_extent_buffer(leaf, symname, ptr, name_len);
10171 btrfs_mark_buffer_dirty(leaf);
10172 btrfs_free_path(path);
10174 inode->i_op = &btrfs_symlink_inode_operations;
10175 inode_nohighmem(inode);
10176 inode->i_mapping->a_ops = &btrfs_aops;
10177 inode_set_bytes(inode, name_len);
10178 btrfs_i_size_write(BTRFS_I(inode), name_len);
10179 err = btrfs_update_inode(trans, root, inode);
10181 * Last step, add directory indexes for our symlink inode. This is the
10182 * last step to avoid extra cleanup of these indexes if an error happens
10186 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
10187 BTRFS_I(inode), 0, index);
10191 d_instantiate_new(dentry, inode);
10194 btrfs_end_transaction(trans);
10195 if (err && inode) {
10196 inode_dec_link_count(inode);
10197 discard_new_inode(inode);
10199 btrfs_btree_balance_dirty(fs_info);
10203 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10204 u64 start, u64 num_bytes, u64 min_size,
10205 loff_t actual_len, u64 *alloc_hint,
10206 struct btrfs_trans_handle *trans)
10208 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
10209 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10210 struct extent_map *em;
10211 struct btrfs_root *root = BTRFS_I(inode)->root;
10212 struct btrfs_key ins;
10213 u64 cur_offset = start;
10216 u64 last_alloc = (u64)-1;
10218 bool own_trans = true;
10219 u64 end = start + num_bytes - 1;
10223 while (num_bytes > 0) {
10225 trans = btrfs_start_transaction(root, 3);
10226 if (IS_ERR(trans)) {
10227 ret = PTR_ERR(trans);
10232 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10233 cur_bytes = max(cur_bytes, min_size);
10235 * If we are severely fragmented we could end up with really
10236 * small allocations, so if the allocator is returning small
10237 * chunks lets make its job easier by only searching for those
10240 cur_bytes = min(cur_bytes, last_alloc);
10241 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10242 min_size, 0, *alloc_hint, &ins, 1, 0);
10245 btrfs_end_transaction(trans);
10248 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10250 last_alloc = ins.offset;
10251 ret = insert_reserved_file_extent(trans, inode,
10252 cur_offset, ins.objectid,
10253 ins.offset, ins.offset,
10254 ins.offset, 0, 0, 0,
10255 BTRFS_FILE_EXTENT_PREALLOC);
10257 btrfs_free_reserved_extent(fs_info, ins.objectid,
10259 btrfs_abort_transaction(trans, ret);
10261 btrfs_end_transaction(trans);
10265 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10266 cur_offset + ins.offset -1, 0);
10268 em = alloc_extent_map();
10270 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10271 &BTRFS_I(inode)->runtime_flags);
10275 em->start = cur_offset;
10276 em->orig_start = cur_offset;
10277 em->len = ins.offset;
10278 em->block_start = ins.objectid;
10279 em->block_len = ins.offset;
10280 em->orig_block_len = ins.offset;
10281 em->ram_bytes = ins.offset;
10282 em->bdev = fs_info->fs_devices->latest_bdev;
10283 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10284 em->generation = trans->transid;
10287 write_lock(&em_tree->lock);
10288 ret = add_extent_mapping(em_tree, em, 1);
10289 write_unlock(&em_tree->lock);
10290 if (ret != -EEXIST)
10292 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10293 cur_offset + ins.offset - 1,
10296 free_extent_map(em);
10298 num_bytes -= ins.offset;
10299 cur_offset += ins.offset;
10300 *alloc_hint = ins.objectid + ins.offset;
10302 inode_inc_iversion(inode);
10303 inode->i_ctime = current_time(inode);
10304 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10305 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10306 (actual_len > inode->i_size) &&
10307 (cur_offset > inode->i_size)) {
10308 if (cur_offset > actual_len)
10309 i_size = actual_len;
10311 i_size = cur_offset;
10312 i_size_write(inode, i_size);
10313 btrfs_ordered_update_i_size(inode, i_size, NULL);
10316 ret = btrfs_update_inode(trans, root, inode);
10319 btrfs_abort_transaction(trans, ret);
10321 btrfs_end_transaction(trans);
10326 btrfs_end_transaction(trans);
10328 if (cur_offset < end)
10329 btrfs_free_reserved_data_space(inode, NULL, cur_offset,
10330 end - cur_offset + 1);
10334 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10335 u64 start, u64 num_bytes, u64 min_size,
10336 loff_t actual_len, u64 *alloc_hint)
10338 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10339 min_size, actual_len, alloc_hint,
10343 int btrfs_prealloc_file_range_trans(struct inode *inode,
10344 struct btrfs_trans_handle *trans, int mode,
10345 u64 start, u64 num_bytes, u64 min_size,
10346 loff_t actual_len, u64 *alloc_hint)
10348 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10349 min_size, actual_len, alloc_hint, trans);
10352 static int btrfs_set_page_dirty(struct page *page)
10354 return __set_page_dirty_nobuffers(page);
10357 static int btrfs_permission(struct inode *inode, int mask)
10359 struct btrfs_root *root = BTRFS_I(inode)->root;
10360 umode_t mode = inode->i_mode;
10362 if (mask & MAY_WRITE &&
10363 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10364 if (btrfs_root_readonly(root))
10366 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10369 return generic_permission(inode, mask);
10372 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10374 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10375 struct btrfs_trans_handle *trans;
10376 struct btrfs_root *root = BTRFS_I(dir)->root;
10377 struct inode *inode = NULL;
10383 * 5 units required for adding orphan entry
10385 trans = btrfs_start_transaction(root, 5);
10387 return PTR_ERR(trans);
10389 ret = btrfs_find_free_ino(root, &objectid);
10393 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10394 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10395 if (IS_ERR(inode)) {
10396 ret = PTR_ERR(inode);
10401 inode->i_fop = &btrfs_file_operations;
10402 inode->i_op = &btrfs_file_inode_operations;
10404 inode->i_mapping->a_ops = &btrfs_aops;
10405 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10407 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10411 ret = btrfs_update_inode(trans, root, inode);
10414 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10419 * We set number of links to 0 in btrfs_new_inode(), and here we set
10420 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10423 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10425 set_nlink(inode, 1);
10426 d_tmpfile(dentry, inode);
10427 unlock_new_inode(inode);
10428 mark_inode_dirty(inode);
10430 btrfs_end_transaction(trans);
10432 discard_new_inode(inode);
10433 btrfs_btree_balance_dirty(fs_info);
10437 __attribute__((const))
10438 static int btrfs_readpage_io_failed_hook(struct page *page, int failed_mirror)
10443 static void btrfs_check_extent_io_range(void *private_data, const char *caller,
10444 u64 start, u64 end)
10446 struct inode *inode = private_data;
10449 isize = i_size_read(inode);
10450 if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
10451 btrfs_debug_rl(BTRFS_I(inode)->root->fs_info,
10452 "%s: ino %llu isize %llu odd range [%llu,%llu]",
10453 caller, btrfs_ino(BTRFS_I(inode)), isize, start, end);
10457 void btrfs_set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
10459 struct inode *inode = tree->private_data;
10460 unsigned long index = start >> PAGE_SHIFT;
10461 unsigned long end_index = end >> PAGE_SHIFT;
10464 while (index <= end_index) {
10465 page = find_get_page(inode->i_mapping, index);
10466 ASSERT(page); /* Pages should be in the extent_io_tree */
10467 set_page_writeback(page);
10473 static const struct inode_operations btrfs_dir_inode_operations = {
10474 .getattr = btrfs_getattr,
10475 .lookup = btrfs_lookup,
10476 .create = btrfs_create,
10477 .unlink = btrfs_unlink,
10478 .link = btrfs_link,
10479 .mkdir = btrfs_mkdir,
10480 .rmdir = btrfs_rmdir,
10481 .rename = btrfs_rename2,
10482 .symlink = btrfs_symlink,
10483 .setattr = btrfs_setattr,
10484 .mknod = btrfs_mknod,
10485 .listxattr = btrfs_listxattr,
10486 .permission = btrfs_permission,
10487 .get_acl = btrfs_get_acl,
10488 .set_acl = btrfs_set_acl,
10489 .update_time = btrfs_update_time,
10490 .tmpfile = btrfs_tmpfile,
10492 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10493 .lookup = btrfs_lookup,
10494 .permission = btrfs_permission,
10495 .update_time = btrfs_update_time,
10498 static const struct file_operations btrfs_dir_file_operations = {
10499 .llseek = generic_file_llseek,
10500 .read = generic_read_dir,
10501 .iterate_shared = btrfs_real_readdir,
10502 .open = btrfs_opendir,
10503 .unlocked_ioctl = btrfs_ioctl,
10504 #ifdef CONFIG_COMPAT
10505 .compat_ioctl = btrfs_compat_ioctl,
10507 .release = btrfs_release_file,
10508 .fsync = btrfs_sync_file,
10511 static const struct extent_io_ops btrfs_extent_io_ops = {
10512 /* mandatory callbacks */
10513 .submit_bio_hook = btrfs_submit_bio_hook,
10514 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10515 .readpage_io_failed_hook = btrfs_readpage_io_failed_hook,
10517 /* optional callbacks */
10518 .fill_delalloc = run_delalloc_range,
10519 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10520 .writepage_start_hook = btrfs_writepage_start_hook,
10521 .set_bit_hook = btrfs_set_bit_hook,
10522 .clear_bit_hook = btrfs_clear_bit_hook,
10523 .merge_extent_hook = btrfs_merge_extent_hook,
10524 .split_extent_hook = btrfs_split_extent_hook,
10525 .check_extent_io_range = btrfs_check_extent_io_range,
10529 * btrfs doesn't support the bmap operation because swapfiles
10530 * use bmap to make a mapping of extents in the file. They assume
10531 * these extents won't change over the life of the file and they
10532 * use the bmap result to do IO directly to the drive.
10534 * the btrfs bmap call would return logical addresses that aren't
10535 * suitable for IO and they also will change frequently as COW
10536 * operations happen. So, swapfile + btrfs == corruption.
10538 * For now we're avoiding this by dropping bmap.
10540 static const struct address_space_operations btrfs_aops = {
10541 .readpage = btrfs_readpage,
10542 .writepage = btrfs_writepage,
10543 .writepages = btrfs_writepages,
10544 .readpages = btrfs_readpages,
10545 .direct_IO = btrfs_direct_IO,
10546 .invalidatepage = btrfs_invalidatepage,
10547 .releasepage = btrfs_releasepage,
10548 .set_page_dirty = btrfs_set_page_dirty,
10549 .error_remove_page = generic_error_remove_page,
10552 static const struct inode_operations btrfs_file_inode_operations = {
10553 .getattr = btrfs_getattr,
10554 .setattr = btrfs_setattr,
10555 .listxattr = btrfs_listxattr,
10556 .permission = btrfs_permission,
10557 .fiemap = btrfs_fiemap,
10558 .get_acl = btrfs_get_acl,
10559 .set_acl = btrfs_set_acl,
10560 .update_time = btrfs_update_time,
10562 static const struct inode_operations btrfs_special_inode_operations = {
10563 .getattr = btrfs_getattr,
10564 .setattr = btrfs_setattr,
10565 .permission = btrfs_permission,
10566 .listxattr = btrfs_listxattr,
10567 .get_acl = btrfs_get_acl,
10568 .set_acl = btrfs_set_acl,
10569 .update_time = btrfs_update_time,
10571 static const struct inode_operations btrfs_symlink_inode_operations = {
10572 .get_link = page_get_link,
10573 .getattr = btrfs_getattr,
10574 .setattr = btrfs_setattr,
10575 .permission = btrfs_permission,
10576 .listxattr = btrfs_listxattr,
10577 .update_time = btrfs_update_time,
10580 const struct dentry_operations btrfs_dentry_operations = {
10581 .d_delete = btrfs_dentry_delete,