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
113 * btrfs_run_delalloc_range 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 page *p = async_extent->pages[0];
851 const u64 start = async_extent->start;
852 const u64 end = start + async_extent->ram_size - 1;
854 p->mapping = inode->i_mapping;
855 btrfs_writepage_endio_finish_ordered(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;
1537 if (cow_start != (u64)-1) {
1539 ret = cow_file_range(inode, locked_page, cow_start, end, end,
1540 page_started, nr_written, 1, NULL);
1546 if (ret && cur_offset < end)
1547 extent_clear_unlock_delalloc(inode, cur_offset, end, end,
1548 locked_page, EXTENT_LOCKED |
1549 EXTENT_DELALLOC | EXTENT_DEFRAG |
1550 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1552 PAGE_SET_WRITEBACK |
1553 PAGE_END_WRITEBACK);
1554 btrfs_free_path(path);
1558 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1561 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1562 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1566 * @defrag_bytes is a hint value, no spinlock held here,
1567 * if is not zero, it means the file is defragging.
1568 * Force cow if given extent needs to be defragged.
1570 if (BTRFS_I(inode)->defrag_bytes &&
1571 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1572 EXTENT_DEFRAG, 0, NULL))
1579 * Function to process delayed allocation (create CoW) for ranges which are
1580 * being touched for the first time.
1582 int btrfs_run_delalloc_range(void *private_data, struct page *locked_page,
1583 u64 start, u64 end, int *page_started, unsigned long *nr_written,
1584 struct writeback_control *wbc)
1586 struct inode *inode = private_data;
1588 int force_cow = need_force_cow(inode, start, end);
1589 unsigned int write_flags = wbc_to_write_flags(wbc);
1591 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1592 ret = run_delalloc_nocow(inode, locked_page, start, end,
1593 page_started, 1, nr_written);
1594 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1595 ret = run_delalloc_nocow(inode, locked_page, start, end,
1596 page_started, 0, nr_written);
1597 } else if (!inode_need_compress(inode, start, end)) {
1598 ret = cow_file_range(inode, locked_page, start, end, end,
1599 page_started, nr_written, 1, NULL);
1601 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1602 &BTRFS_I(inode)->runtime_flags);
1603 ret = cow_file_range_async(inode, locked_page, start, end,
1604 page_started, nr_written,
1608 btrfs_cleanup_ordered_extents(inode, start, end - start + 1);
1612 static void btrfs_split_extent_hook(void *private_data,
1613 struct extent_state *orig, u64 split)
1615 struct inode *inode = private_data;
1618 /* not delalloc, ignore it */
1619 if (!(orig->state & EXTENT_DELALLOC))
1622 size = orig->end - orig->start + 1;
1623 if (size > BTRFS_MAX_EXTENT_SIZE) {
1628 * See the explanation in btrfs_merge_delalloc_extent, the same
1629 * applies here, just in reverse.
1631 new_size = orig->end - split + 1;
1632 num_extents = count_max_extents(new_size);
1633 new_size = split - orig->start;
1634 num_extents += count_max_extents(new_size);
1635 if (count_max_extents(size) >= num_extents)
1639 spin_lock(&BTRFS_I(inode)->lock);
1640 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1641 spin_unlock(&BTRFS_I(inode)->lock);
1645 * Handle merged delayed allocation extents so we can keep track of new extents
1646 * that are just merged onto old extents, such as when we are doing sequential
1647 * writes, so we can properly account for the metadata space we'll need.
1649 void btrfs_merge_delalloc_extent(struct inode *inode, struct extent_state *new,
1650 struct extent_state *other)
1652 u64 new_size, old_size;
1655 /* not delalloc, ignore it */
1656 if (!(other->state & EXTENT_DELALLOC))
1659 if (new->start > other->start)
1660 new_size = new->end - other->start + 1;
1662 new_size = other->end - new->start + 1;
1664 /* we're not bigger than the max, unreserve the space and go */
1665 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1666 spin_lock(&BTRFS_I(inode)->lock);
1667 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1668 spin_unlock(&BTRFS_I(inode)->lock);
1673 * We have to add up either side to figure out how many extents were
1674 * accounted for before we merged into one big extent. If the number of
1675 * extents we accounted for is <= the amount we need for the new range
1676 * then we can return, otherwise drop. Think of it like this
1680 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1681 * need 2 outstanding extents, on one side we have 1 and the other side
1682 * we have 1 so they are == and we can return. But in this case
1684 * [MAX_SIZE+4k][MAX_SIZE+4k]
1686 * Each range on their own accounts for 2 extents, but merged together
1687 * they are only 3 extents worth of accounting, so we need to drop in
1690 old_size = other->end - other->start + 1;
1691 num_extents = count_max_extents(old_size);
1692 old_size = new->end - new->start + 1;
1693 num_extents += count_max_extents(old_size);
1694 if (count_max_extents(new_size) >= num_extents)
1697 spin_lock(&BTRFS_I(inode)->lock);
1698 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1699 spin_unlock(&BTRFS_I(inode)->lock);
1702 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1703 struct inode *inode)
1705 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1707 spin_lock(&root->delalloc_lock);
1708 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1709 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1710 &root->delalloc_inodes);
1711 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1712 &BTRFS_I(inode)->runtime_flags);
1713 root->nr_delalloc_inodes++;
1714 if (root->nr_delalloc_inodes == 1) {
1715 spin_lock(&fs_info->delalloc_root_lock);
1716 BUG_ON(!list_empty(&root->delalloc_root));
1717 list_add_tail(&root->delalloc_root,
1718 &fs_info->delalloc_roots);
1719 spin_unlock(&fs_info->delalloc_root_lock);
1722 spin_unlock(&root->delalloc_lock);
1726 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
1727 struct btrfs_inode *inode)
1729 struct btrfs_fs_info *fs_info = root->fs_info;
1731 if (!list_empty(&inode->delalloc_inodes)) {
1732 list_del_init(&inode->delalloc_inodes);
1733 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1734 &inode->runtime_flags);
1735 root->nr_delalloc_inodes--;
1736 if (!root->nr_delalloc_inodes) {
1737 ASSERT(list_empty(&root->delalloc_inodes));
1738 spin_lock(&fs_info->delalloc_root_lock);
1739 BUG_ON(list_empty(&root->delalloc_root));
1740 list_del_init(&root->delalloc_root);
1741 spin_unlock(&fs_info->delalloc_root_lock);
1746 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1747 struct btrfs_inode *inode)
1749 spin_lock(&root->delalloc_lock);
1750 __btrfs_del_delalloc_inode(root, inode);
1751 spin_unlock(&root->delalloc_lock);
1755 * Properly track delayed allocation bytes in the inode and to maintain the
1756 * list of inodes that have pending delalloc work to be done.
1758 void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state,
1761 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1763 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1766 * set_bit and clear bit hooks normally require _irqsave/restore
1767 * but in this case, we are only testing for the DELALLOC
1768 * bit, which is only set or cleared with irqs on
1770 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1771 struct btrfs_root *root = BTRFS_I(inode)->root;
1772 u64 len = state->end + 1 - state->start;
1773 u32 num_extents = count_max_extents(len);
1774 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
1776 spin_lock(&BTRFS_I(inode)->lock);
1777 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
1778 spin_unlock(&BTRFS_I(inode)->lock);
1780 /* For sanity tests */
1781 if (btrfs_is_testing(fs_info))
1784 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
1785 fs_info->delalloc_batch);
1786 spin_lock(&BTRFS_I(inode)->lock);
1787 BTRFS_I(inode)->delalloc_bytes += len;
1788 if (*bits & EXTENT_DEFRAG)
1789 BTRFS_I(inode)->defrag_bytes += len;
1790 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1791 &BTRFS_I(inode)->runtime_flags))
1792 btrfs_add_delalloc_inodes(root, inode);
1793 spin_unlock(&BTRFS_I(inode)->lock);
1796 if (!(state->state & EXTENT_DELALLOC_NEW) &&
1797 (*bits & EXTENT_DELALLOC_NEW)) {
1798 spin_lock(&BTRFS_I(inode)->lock);
1799 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
1801 spin_unlock(&BTRFS_I(inode)->lock);
1806 * Once a range is no longer delalloc this function ensures that proper
1807 * accounting happens.
1809 void btrfs_clear_delalloc_extent(struct inode *vfs_inode,
1810 struct extent_state *state, unsigned *bits)
1812 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
1813 struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb);
1814 u64 len = state->end + 1 - state->start;
1815 u32 num_extents = count_max_extents(len);
1817 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
1818 spin_lock(&inode->lock);
1819 inode->defrag_bytes -= len;
1820 spin_unlock(&inode->lock);
1824 * set_bit and clear bit hooks normally require _irqsave/restore
1825 * but in this case, we are only testing for the DELALLOC
1826 * bit, which is only set or cleared with irqs on
1828 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1829 struct btrfs_root *root = inode->root;
1830 bool do_list = !btrfs_is_free_space_inode(inode);
1832 spin_lock(&inode->lock);
1833 btrfs_mod_outstanding_extents(inode, -num_extents);
1834 spin_unlock(&inode->lock);
1837 * We don't reserve metadata space for space cache inodes so we
1838 * don't need to call dellalloc_release_metadata if there is an
1841 if (*bits & EXTENT_CLEAR_META_RESV &&
1842 root != fs_info->tree_root)
1843 btrfs_delalloc_release_metadata(inode, len, false);
1845 /* For sanity tests. */
1846 if (btrfs_is_testing(fs_info))
1849 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
1850 do_list && !(state->state & EXTENT_NORESERVE) &&
1851 (*bits & EXTENT_CLEAR_DATA_RESV))
1852 btrfs_free_reserved_data_space_noquota(
1856 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
1857 fs_info->delalloc_batch);
1858 spin_lock(&inode->lock);
1859 inode->delalloc_bytes -= len;
1860 if (do_list && inode->delalloc_bytes == 0 &&
1861 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1862 &inode->runtime_flags))
1863 btrfs_del_delalloc_inode(root, inode);
1864 spin_unlock(&inode->lock);
1867 if ((state->state & EXTENT_DELALLOC_NEW) &&
1868 (*bits & EXTENT_DELALLOC_NEW)) {
1869 spin_lock(&inode->lock);
1870 ASSERT(inode->new_delalloc_bytes >= len);
1871 inode->new_delalloc_bytes -= len;
1872 spin_unlock(&inode->lock);
1877 * Merge bio hook, this must check the chunk tree to make sure we don't create
1878 * bios that span stripes or chunks
1880 * return 1 if page cannot be merged to bio
1881 * return 0 if page can be merged to bio
1882 * return error otherwise
1884 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1885 size_t size, struct bio *bio,
1886 unsigned long bio_flags)
1888 struct inode *inode = page->mapping->host;
1889 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1890 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1895 if (bio_flags & EXTENT_BIO_COMPRESSED)
1898 length = bio->bi_iter.bi_size;
1899 map_length = length;
1900 ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
1904 if (map_length < length + size)
1910 * in order to insert checksums into the metadata in large chunks,
1911 * we wait until bio submission time. All the pages in the bio are
1912 * checksummed and sums are attached onto the ordered extent record.
1914 * At IO completion time the cums attached on the ordered extent record
1915 * are inserted into the btree
1917 static blk_status_t btrfs_submit_bio_start(void *private_data, struct bio *bio,
1920 struct inode *inode = private_data;
1921 blk_status_t ret = 0;
1923 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1924 BUG_ON(ret); /* -ENOMEM */
1929 * in order to insert checksums into the metadata in large chunks,
1930 * we wait until bio submission time. All the pages in the bio are
1931 * checksummed and sums are attached onto the ordered extent record.
1933 * At IO completion time the cums attached on the ordered extent record
1934 * are inserted into the btree
1936 blk_status_t btrfs_submit_bio_done(void *private_data, struct bio *bio,
1939 struct inode *inode = private_data;
1940 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1943 ret = btrfs_map_bio(fs_info, bio, mirror_num, 1);
1945 bio->bi_status = ret;
1952 * extent_io.c submission hook. This does the right thing for csum calculation
1953 * on write, or reading the csums from the tree before a read.
1955 * Rules about async/sync submit,
1956 * a) read: sync submit
1958 * b) write without checksum: sync submit
1960 * c) write with checksum:
1961 * c-1) if bio is issued by fsync: sync submit
1962 * (sync_writers != 0)
1964 * c-2) if root is reloc root: sync submit
1965 * (only in case of buffered IO)
1967 * c-3) otherwise: async submit
1969 static blk_status_t btrfs_submit_bio_hook(void *private_data, struct bio *bio,
1970 int mirror_num, unsigned long bio_flags,
1973 struct inode *inode = private_data;
1974 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1975 struct btrfs_root *root = BTRFS_I(inode)->root;
1976 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1977 blk_status_t ret = 0;
1979 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1981 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1983 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
1984 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1986 if (bio_op(bio) != REQ_OP_WRITE) {
1987 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
1991 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1992 ret = btrfs_submit_compressed_read(inode, bio,
1996 } else if (!skip_sum) {
1997 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
2002 } else if (async && !skip_sum) {
2003 /* csum items have already been cloned */
2004 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2006 /* we're doing a write, do the async checksumming */
2007 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags,
2009 btrfs_submit_bio_start);
2011 } else if (!skip_sum) {
2012 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
2018 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
2022 bio->bi_status = ret;
2029 * given a list of ordered sums record them in the inode. This happens
2030 * at IO completion time based on sums calculated at bio submission time.
2032 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
2033 struct inode *inode, struct list_head *list)
2035 struct btrfs_ordered_sum *sum;
2038 list_for_each_entry(sum, list, list) {
2039 trans->adding_csums = true;
2040 ret = btrfs_csum_file_blocks(trans,
2041 BTRFS_I(inode)->root->fs_info->csum_root, sum);
2042 trans->adding_csums = false;
2049 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
2050 unsigned int extra_bits,
2051 struct extent_state **cached_state, int dedupe)
2053 WARN_ON((end & (PAGE_SIZE - 1)) == 0);
2054 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2055 extra_bits, cached_state);
2058 /* see btrfs_writepage_start_hook for details on why this is required */
2059 struct btrfs_writepage_fixup {
2061 struct btrfs_work work;
2064 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2066 struct btrfs_writepage_fixup *fixup;
2067 struct btrfs_ordered_extent *ordered;
2068 struct extent_state *cached_state = NULL;
2069 struct extent_changeset *data_reserved = NULL;
2071 struct inode *inode;
2076 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2080 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2081 ClearPageChecked(page);
2085 inode = page->mapping->host;
2086 page_start = page_offset(page);
2087 page_end = page_offset(page) + PAGE_SIZE - 1;
2089 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2092 /* already ordered? We're done */
2093 if (PagePrivate2(page))
2096 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
2099 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2100 page_end, &cached_state);
2102 btrfs_start_ordered_extent(inode, ordered, 1);
2103 btrfs_put_ordered_extent(ordered);
2107 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2110 mapping_set_error(page->mapping, ret);
2111 end_extent_writepage(page, ret, page_start, page_end);
2112 ClearPageChecked(page);
2116 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2119 mapping_set_error(page->mapping, ret);
2120 end_extent_writepage(page, ret, page_start, page_end);
2121 ClearPageChecked(page);
2125 ClearPageChecked(page);
2126 set_page_dirty(page);
2127 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, false);
2129 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2135 extent_changeset_free(data_reserved);
2139 * There are a few paths in the higher layers of the kernel that directly
2140 * set the page dirty bit without asking the filesystem if it is a
2141 * good idea. This causes problems because we want to make sure COW
2142 * properly happens and the data=ordered rules are followed.
2144 * In our case any range that doesn't have the ORDERED bit set
2145 * hasn't been properly setup for IO. We kick off an async process
2146 * to fix it up. The async helper will wait for ordered extents, set
2147 * the delalloc bit and make it safe to write the page.
2149 int btrfs_writepage_cow_fixup(struct page *page, u64 start, u64 end)
2151 struct inode *inode = page->mapping->host;
2152 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2153 struct btrfs_writepage_fixup *fixup;
2155 /* this page is properly in the ordered list */
2156 if (TestClearPagePrivate2(page))
2159 if (PageChecked(page))
2162 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2166 SetPageChecked(page);
2168 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2169 btrfs_writepage_fixup_worker, NULL, NULL);
2171 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2175 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2176 struct inode *inode, u64 file_pos,
2177 u64 disk_bytenr, u64 disk_num_bytes,
2178 u64 num_bytes, u64 ram_bytes,
2179 u8 compression, u8 encryption,
2180 u16 other_encoding, int extent_type)
2182 struct btrfs_root *root = BTRFS_I(inode)->root;
2183 struct btrfs_file_extent_item *fi;
2184 struct btrfs_path *path;
2185 struct extent_buffer *leaf;
2186 struct btrfs_key ins;
2188 int extent_inserted = 0;
2191 path = btrfs_alloc_path();
2196 * we may be replacing one extent in the tree with another.
2197 * The new extent is pinned in the extent map, and we don't want
2198 * to drop it from the cache until it is completely in the btree.
2200 * So, tell btrfs_drop_extents to leave this extent in the cache.
2201 * the caller is expected to unpin it and allow it to be merged
2204 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2205 file_pos + num_bytes, NULL, 0,
2206 1, sizeof(*fi), &extent_inserted);
2210 if (!extent_inserted) {
2211 ins.objectid = btrfs_ino(BTRFS_I(inode));
2212 ins.offset = file_pos;
2213 ins.type = BTRFS_EXTENT_DATA_KEY;
2215 path->leave_spinning = 1;
2216 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2221 leaf = path->nodes[0];
2222 fi = btrfs_item_ptr(leaf, path->slots[0],
2223 struct btrfs_file_extent_item);
2224 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2225 btrfs_set_file_extent_type(leaf, fi, extent_type);
2226 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2227 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2228 btrfs_set_file_extent_offset(leaf, fi, 0);
2229 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2230 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2231 btrfs_set_file_extent_compression(leaf, fi, compression);
2232 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2233 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2235 btrfs_mark_buffer_dirty(leaf);
2236 btrfs_release_path(path);
2238 inode_add_bytes(inode, num_bytes);
2240 ins.objectid = disk_bytenr;
2241 ins.offset = disk_num_bytes;
2242 ins.type = BTRFS_EXTENT_ITEM_KEY;
2245 * Release the reserved range from inode dirty range map, as it is
2246 * already moved into delayed_ref_head
2248 ret = btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2252 ret = btrfs_alloc_reserved_file_extent(trans, root,
2253 btrfs_ino(BTRFS_I(inode)),
2254 file_pos, qg_released, &ins);
2256 btrfs_free_path(path);
2261 /* snapshot-aware defrag */
2262 struct sa_defrag_extent_backref {
2263 struct rb_node node;
2264 struct old_sa_defrag_extent *old;
2273 struct old_sa_defrag_extent {
2274 struct list_head list;
2275 struct new_sa_defrag_extent *new;
2284 struct new_sa_defrag_extent {
2285 struct rb_root root;
2286 struct list_head head;
2287 struct btrfs_path *path;
2288 struct inode *inode;
2296 static int backref_comp(struct sa_defrag_extent_backref *b1,
2297 struct sa_defrag_extent_backref *b2)
2299 if (b1->root_id < b2->root_id)
2301 else if (b1->root_id > b2->root_id)
2304 if (b1->inum < b2->inum)
2306 else if (b1->inum > b2->inum)
2309 if (b1->file_pos < b2->file_pos)
2311 else if (b1->file_pos > b2->file_pos)
2315 * [------------------------------] ===> (a range of space)
2316 * |<--->| |<---->| =============> (fs/file tree A)
2317 * |<---------------------------->| ===> (fs/file tree B)
2319 * A range of space can refer to two file extents in one tree while
2320 * refer to only one file extent in another tree.
2322 * So we may process a disk offset more than one time(two extents in A)
2323 * and locate at the same extent(one extent in B), then insert two same
2324 * backrefs(both refer to the extent in B).
2329 static void backref_insert(struct rb_root *root,
2330 struct sa_defrag_extent_backref *backref)
2332 struct rb_node **p = &root->rb_node;
2333 struct rb_node *parent = NULL;
2334 struct sa_defrag_extent_backref *entry;
2339 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2341 ret = backref_comp(backref, entry);
2345 p = &(*p)->rb_right;
2348 rb_link_node(&backref->node, parent, p);
2349 rb_insert_color(&backref->node, root);
2353 * Note the backref might has changed, and in this case we just return 0.
2355 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2358 struct btrfs_file_extent_item *extent;
2359 struct old_sa_defrag_extent *old = ctx;
2360 struct new_sa_defrag_extent *new = old->new;
2361 struct btrfs_path *path = new->path;
2362 struct btrfs_key key;
2363 struct btrfs_root *root;
2364 struct sa_defrag_extent_backref *backref;
2365 struct extent_buffer *leaf;
2366 struct inode *inode = new->inode;
2367 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2373 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2374 inum == btrfs_ino(BTRFS_I(inode)))
2377 key.objectid = root_id;
2378 key.type = BTRFS_ROOT_ITEM_KEY;
2379 key.offset = (u64)-1;
2381 root = btrfs_read_fs_root_no_name(fs_info, &key);
2383 if (PTR_ERR(root) == -ENOENT)
2386 btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu",
2387 inum, offset, root_id);
2388 return PTR_ERR(root);
2391 key.objectid = inum;
2392 key.type = BTRFS_EXTENT_DATA_KEY;
2393 if (offset > (u64)-1 << 32)
2396 key.offset = offset;
2398 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2399 if (WARN_ON(ret < 0))
2406 leaf = path->nodes[0];
2407 slot = path->slots[0];
2409 if (slot >= btrfs_header_nritems(leaf)) {
2410 ret = btrfs_next_leaf(root, path);
2413 } else if (ret > 0) {
2422 btrfs_item_key_to_cpu(leaf, &key, slot);
2424 if (key.objectid > inum)
2427 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2430 extent = btrfs_item_ptr(leaf, slot,
2431 struct btrfs_file_extent_item);
2433 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2437 * 'offset' refers to the exact key.offset,
2438 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2439 * (key.offset - extent_offset).
2441 if (key.offset != offset)
2444 extent_offset = btrfs_file_extent_offset(leaf, extent);
2445 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2447 if (extent_offset >= old->extent_offset + old->offset +
2448 old->len || extent_offset + num_bytes <=
2449 old->extent_offset + old->offset)
2454 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2460 backref->root_id = root_id;
2461 backref->inum = inum;
2462 backref->file_pos = offset;
2463 backref->num_bytes = num_bytes;
2464 backref->extent_offset = extent_offset;
2465 backref->generation = btrfs_file_extent_generation(leaf, extent);
2467 backref_insert(&new->root, backref);
2470 btrfs_release_path(path);
2475 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2476 struct new_sa_defrag_extent *new)
2478 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2479 struct old_sa_defrag_extent *old, *tmp;
2484 list_for_each_entry_safe(old, tmp, &new->head, list) {
2485 ret = iterate_inodes_from_logical(old->bytenr +
2486 old->extent_offset, fs_info,
2487 path, record_one_backref,
2489 if (ret < 0 && ret != -ENOENT)
2492 /* no backref to be processed for this extent */
2494 list_del(&old->list);
2499 if (list_empty(&new->head))
2505 static int relink_is_mergable(struct extent_buffer *leaf,
2506 struct btrfs_file_extent_item *fi,
2507 struct new_sa_defrag_extent *new)
2509 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2512 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2515 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2518 if (btrfs_file_extent_encryption(leaf, fi) ||
2519 btrfs_file_extent_other_encoding(leaf, fi))
2526 * Note the backref might has changed, and in this case we just return 0.
2528 static noinline int relink_extent_backref(struct btrfs_path *path,
2529 struct sa_defrag_extent_backref *prev,
2530 struct sa_defrag_extent_backref *backref)
2532 struct btrfs_file_extent_item *extent;
2533 struct btrfs_file_extent_item *item;
2534 struct btrfs_ordered_extent *ordered;
2535 struct btrfs_trans_handle *trans;
2536 struct btrfs_root *root;
2537 struct btrfs_key key;
2538 struct extent_buffer *leaf;
2539 struct old_sa_defrag_extent *old = backref->old;
2540 struct new_sa_defrag_extent *new = old->new;
2541 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2542 struct inode *inode;
2543 struct extent_state *cached = NULL;
2552 if (prev && prev->root_id == backref->root_id &&
2553 prev->inum == backref->inum &&
2554 prev->file_pos + prev->num_bytes == backref->file_pos)
2557 /* step 1: get root */
2558 key.objectid = backref->root_id;
2559 key.type = BTRFS_ROOT_ITEM_KEY;
2560 key.offset = (u64)-1;
2562 index = srcu_read_lock(&fs_info->subvol_srcu);
2564 root = btrfs_read_fs_root_no_name(fs_info, &key);
2566 srcu_read_unlock(&fs_info->subvol_srcu, index);
2567 if (PTR_ERR(root) == -ENOENT)
2569 return PTR_ERR(root);
2572 if (btrfs_root_readonly(root)) {
2573 srcu_read_unlock(&fs_info->subvol_srcu, index);
2577 /* step 2: get inode */
2578 key.objectid = backref->inum;
2579 key.type = BTRFS_INODE_ITEM_KEY;
2582 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2583 if (IS_ERR(inode)) {
2584 srcu_read_unlock(&fs_info->subvol_srcu, index);
2588 srcu_read_unlock(&fs_info->subvol_srcu, index);
2590 /* step 3: relink backref */
2591 lock_start = backref->file_pos;
2592 lock_end = backref->file_pos + backref->num_bytes - 1;
2593 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2596 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2598 btrfs_put_ordered_extent(ordered);
2602 trans = btrfs_join_transaction(root);
2603 if (IS_ERR(trans)) {
2604 ret = PTR_ERR(trans);
2608 key.objectid = backref->inum;
2609 key.type = BTRFS_EXTENT_DATA_KEY;
2610 key.offset = backref->file_pos;
2612 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2615 } else if (ret > 0) {
2620 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2621 struct btrfs_file_extent_item);
2623 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2624 backref->generation)
2627 btrfs_release_path(path);
2629 start = backref->file_pos;
2630 if (backref->extent_offset < old->extent_offset + old->offset)
2631 start += old->extent_offset + old->offset -
2632 backref->extent_offset;
2634 len = min(backref->extent_offset + backref->num_bytes,
2635 old->extent_offset + old->offset + old->len);
2636 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2638 ret = btrfs_drop_extents(trans, root, inode, start,
2643 key.objectid = btrfs_ino(BTRFS_I(inode));
2644 key.type = BTRFS_EXTENT_DATA_KEY;
2647 path->leave_spinning = 1;
2649 struct btrfs_file_extent_item *fi;
2651 struct btrfs_key found_key;
2653 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2658 leaf = path->nodes[0];
2659 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2661 fi = btrfs_item_ptr(leaf, path->slots[0],
2662 struct btrfs_file_extent_item);
2663 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2665 if (extent_len + found_key.offset == start &&
2666 relink_is_mergable(leaf, fi, new)) {
2667 btrfs_set_file_extent_num_bytes(leaf, fi,
2669 btrfs_mark_buffer_dirty(leaf);
2670 inode_add_bytes(inode, len);
2676 btrfs_release_path(path);
2681 ret = btrfs_insert_empty_item(trans, root, path, &key,
2684 btrfs_abort_transaction(trans, ret);
2688 leaf = path->nodes[0];
2689 item = btrfs_item_ptr(leaf, path->slots[0],
2690 struct btrfs_file_extent_item);
2691 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2692 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2693 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2694 btrfs_set_file_extent_num_bytes(leaf, item, len);
2695 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2696 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2697 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2698 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2699 btrfs_set_file_extent_encryption(leaf, item, 0);
2700 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2702 btrfs_mark_buffer_dirty(leaf);
2703 inode_add_bytes(inode, len);
2704 btrfs_release_path(path);
2706 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2708 backref->root_id, backref->inum,
2709 new->file_pos); /* start - extent_offset */
2711 btrfs_abort_transaction(trans, ret);
2717 btrfs_release_path(path);
2718 path->leave_spinning = 0;
2719 btrfs_end_transaction(trans);
2721 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2727 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2729 struct old_sa_defrag_extent *old, *tmp;
2734 list_for_each_entry_safe(old, tmp, &new->head, list) {
2740 static void relink_file_extents(struct new_sa_defrag_extent *new)
2742 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2743 struct btrfs_path *path;
2744 struct sa_defrag_extent_backref *backref;
2745 struct sa_defrag_extent_backref *prev = NULL;
2746 struct rb_node *node;
2749 path = btrfs_alloc_path();
2753 if (!record_extent_backrefs(path, new)) {
2754 btrfs_free_path(path);
2757 btrfs_release_path(path);
2760 node = rb_first(&new->root);
2763 rb_erase(node, &new->root);
2765 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2767 ret = relink_extent_backref(path, prev, backref);
2780 btrfs_free_path(path);
2782 free_sa_defrag_extent(new);
2784 atomic_dec(&fs_info->defrag_running);
2785 wake_up(&fs_info->transaction_wait);
2788 static struct new_sa_defrag_extent *
2789 record_old_file_extents(struct inode *inode,
2790 struct btrfs_ordered_extent *ordered)
2792 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2793 struct btrfs_root *root = BTRFS_I(inode)->root;
2794 struct btrfs_path *path;
2795 struct btrfs_key key;
2796 struct old_sa_defrag_extent *old;
2797 struct new_sa_defrag_extent *new;
2800 new = kmalloc(sizeof(*new), GFP_NOFS);
2805 new->file_pos = ordered->file_offset;
2806 new->len = ordered->len;
2807 new->bytenr = ordered->start;
2808 new->disk_len = ordered->disk_len;
2809 new->compress_type = ordered->compress_type;
2810 new->root = RB_ROOT;
2811 INIT_LIST_HEAD(&new->head);
2813 path = btrfs_alloc_path();
2817 key.objectid = btrfs_ino(BTRFS_I(inode));
2818 key.type = BTRFS_EXTENT_DATA_KEY;
2819 key.offset = new->file_pos;
2821 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2824 if (ret > 0 && path->slots[0] > 0)
2827 /* find out all the old extents for the file range */
2829 struct btrfs_file_extent_item *extent;
2830 struct extent_buffer *l;
2839 slot = path->slots[0];
2841 if (slot >= btrfs_header_nritems(l)) {
2842 ret = btrfs_next_leaf(root, path);
2850 btrfs_item_key_to_cpu(l, &key, slot);
2852 if (key.objectid != btrfs_ino(BTRFS_I(inode)))
2854 if (key.type != BTRFS_EXTENT_DATA_KEY)
2856 if (key.offset >= new->file_pos + new->len)
2859 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2861 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2862 if (key.offset + num_bytes < new->file_pos)
2865 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2869 extent_offset = btrfs_file_extent_offset(l, extent);
2871 old = kmalloc(sizeof(*old), GFP_NOFS);
2875 offset = max(new->file_pos, key.offset);
2876 end = min(new->file_pos + new->len, key.offset + num_bytes);
2878 old->bytenr = disk_bytenr;
2879 old->extent_offset = extent_offset;
2880 old->offset = offset - key.offset;
2881 old->len = end - offset;
2884 list_add_tail(&old->list, &new->head);
2890 btrfs_free_path(path);
2891 atomic_inc(&fs_info->defrag_running);
2896 btrfs_free_path(path);
2898 free_sa_defrag_extent(new);
2902 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2905 struct btrfs_block_group_cache *cache;
2907 cache = btrfs_lookup_block_group(fs_info, start);
2910 spin_lock(&cache->lock);
2911 cache->delalloc_bytes -= len;
2912 spin_unlock(&cache->lock);
2914 btrfs_put_block_group(cache);
2917 /* as ordered data IO finishes, this gets called so we can finish
2918 * an ordered extent if the range of bytes in the file it covers are
2921 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2923 struct inode *inode = ordered_extent->inode;
2924 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2925 struct btrfs_root *root = BTRFS_I(inode)->root;
2926 struct btrfs_trans_handle *trans = NULL;
2927 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2928 struct extent_state *cached_state = NULL;
2929 struct new_sa_defrag_extent *new = NULL;
2930 int compress_type = 0;
2932 u64 logical_len = ordered_extent->len;
2934 bool truncated = false;
2935 bool range_locked = false;
2936 bool clear_new_delalloc_bytes = false;
2937 bool clear_reserved_extent = true;
2939 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2940 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2941 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2942 clear_new_delalloc_bytes = true;
2944 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
2946 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2951 btrfs_free_io_failure_record(BTRFS_I(inode),
2952 ordered_extent->file_offset,
2953 ordered_extent->file_offset +
2954 ordered_extent->len - 1);
2956 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2958 logical_len = ordered_extent->truncated_len;
2959 /* Truncated the entire extent, don't bother adding */
2964 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2965 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2968 * For mwrite(mmap + memset to write) case, we still reserve
2969 * space for NOCOW range.
2970 * As NOCOW won't cause a new delayed ref, just free the space
2972 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
2973 ordered_extent->len);
2974 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2976 trans = btrfs_join_transaction_nolock(root);
2978 trans = btrfs_join_transaction(root);
2979 if (IS_ERR(trans)) {
2980 ret = PTR_ERR(trans);
2984 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
2985 ret = btrfs_update_inode_fallback(trans, root, inode);
2986 if (ret) /* -ENOMEM or corruption */
2987 btrfs_abort_transaction(trans, ret);
2991 range_locked = true;
2992 lock_extent_bits(io_tree, ordered_extent->file_offset,
2993 ordered_extent->file_offset + ordered_extent->len - 1,
2996 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2997 ordered_extent->file_offset + ordered_extent->len - 1,
2998 EXTENT_DEFRAG, 0, cached_state);
3000 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
3001 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
3002 /* the inode is shared */
3003 new = record_old_file_extents(inode, ordered_extent);
3005 clear_extent_bit(io_tree, ordered_extent->file_offset,
3006 ordered_extent->file_offset + ordered_extent->len - 1,
3007 EXTENT_DEFRAG, 0, 0, &cached_state);
3011 trans = btrfs_join_transaction_nolock(root);
3013 trans = btrfs_join_transaction(root);
3014 if (IS_ERR(trans)) {
3015 ret = PTR_ERR(trans);
3020 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
3022 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3023 compress_type = ordered_extent->compress_type;
3024 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3025 BUG_ON(compress_type);
3026 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
3027 ordered_extent->len);
3028 ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
3029 ordered_extent->file_offset,
3030 ordered_extent->file_offset +
3033 BUG_ON(root == fs_info->tree_root);
3034 ret = insert_reserved_file_extent(trans, inode,
3035 ordered_extent->file_offset,
3036 ordered_extent->start,
3037 ordered_extent->disk_len,
3038 logical_len, logical_len,
3039 compress_type, 0, 0,
3040 BTRFS_FILE_EXTENT_REG);
3042 clear_reserved_extent = false;
3043 btrfs_release_delalloc_bytes(fs_info,
3044 ordered_extent->start,
3045 ordered_extent->disk_len);
3048 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
3049 ordered_extent->file_offset, ordered_extent->len,
3052 btrfs_abort_transaction(trans, ret);
3056 ret = add_pending_csums(trans, inode, &ordered_extent->list);
3058 btrfs_abort_transaction(trans, ret);
3062 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
3063 ret = btrfs_update_inode_fallback(trans, root, inode);
3064 if (ret) { /* -ENOMEM or corruption */
3065 btrfs_abort_transaction(trans, ret);
3070 if (range_locked || clear_new_delalloc_bytes) {
3071 unsigned int clear_bits = 0;
3074 clear_bits |= EXTENT_LOCKED;
3075 if (clear_new_delalloc_bytes)
3076 clear_bits |= EXTENT_DELALLOC_NEW;
3077 clear_extent_bit(&BTRFS_I(inode)->io_tree,
3078 ordered_extent->file_offset,
3079 ordered_extent->file_offset +
3080 ordered_extent->len - 1,
3082 (clear_bits & EXTENT_LOCKED) ? 1 : 0,
3087 btrfs_end_transaction(trans);
3089 if (ret || truncated) {
3093 start = ordered_extent->file_offset + logical_len;
3095 start = ordered_extent->file_offset;
3096 end = ordered_extent->file_offset + ordered_extent->len - 1;
3097 clear_extent_uptodate(io_tree, start, end, NULL);
3099 /* Drop the cache for the part of the extent we didn't write. */
3100 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
3103 * If the ordered extent had an IOERR or something else went
3104 * wrong we need to return the space for this ordered extent
3105 * back to the allocator. We only free the extent in the
3106 * truncated case if we didn't write out the extent at all.
3108 * If we made it past insert_reserved_file_extent before we
3109 * errored out then we don't need to do this as the accounting
3110 * has already been done.
3112 if ((ret || !logical_len) &&
3113 clear_reserved_extent &&
3114 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3115 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
3116 btrfs_free_reserved_extent(fs_info,
3117 ordered_extent->start,
3118 ordered_extent->disk_len, 1);
3123 * This needs to be done to make sure anybody waiting knows we are done
3124 * updating everything for this ordered extent.
3126 btrfs_remove_ordered_extent(inode, ordered_extent);
3128 /* for snapshot-aware defrag */
3131 free_sa_defrag_extent(new);
3132 atomic_dec(&fs_info->defrag_running);
3134 relink_file_extents(new);
3139 btrfs_put_ordered_extent(ordered_extent);
3140 /* once for the tree */
3141 btrfs_put_ordered_extent(ordered_extent);
3143 /* Try to release some metadata so we don't get an OOM but don't wait */
3144 btrfs_btree_balance_dirty_nodelay(fs_info);
3149 static void finish_ordered_fn(struct btrfs_work *work)
3151 struct btrfs_ordered_extent *ordered_extent;
3152 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3153 btrfs_finish_ordered_io(ordered_extent);
3156 void btrfs_writepage_endio_finish_ordered(struct page *page, u64 start, u64 end,
3157 struct extent_state *state, int uptodate)
3159 struct inode *inode = page->mapping->host;
3160 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3161 struct btrfs_ordered_extent *ordered_extent = NULL;
3162 struct btrfs_workqueue *wq;
3163 btrfs_work_func_t func;
3165 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3167 ClearPagePrivate2(page);
3168 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3169 end - start + 1, uptodate))
3172 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
3173 wq = fs_info->endio_freespace_worker;
3174 func = btrfs_freespace_write_helper;
3176 wq = fs_info->endio_write_workers;
3177 func = btrfs_endio_write_helper;
3180 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3182 btrfs_queue_work(wq, &ordered_extent->work);
3185 static int __readpage_endio_check(struct inode *inode,
3186 struct btrfs_io_bio *io_bio,
3187 int icsum, struct page *page,
3188 int pgoff, u64 start, size_t len)
3194 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3196 kaddr = kmap_atomic(page);
3197 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3198 btrfs_csum_final(csum, (u8 *)&csum);
3199 if (csum != csum_expected)
3202 kunmap_atomic(kaddr);
3205 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3206 io_bio->mirror_num);
3207 memset(kaddr + pgoff, 1, len);
3208 flush_dcache_page(page);
3209 kunmap_atomic(kaddr);
3214 * when reads are done, we need to check csums to verify the data is correct
3215 * if there's a match, we allow the bio to finish. If not, the code in
3216 * extent_io.c will try to find good copies for us.
3218 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3219 u64 phy_offset, struct page *page,
3220 u64 start, u64 end, int mirror)
3222 size_t offset = start - page_offset(page);
3223 struct inode *inode = page->mapping->host;
3224 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3225 struct btrfs_root *root = BTRFS_I(inode)->root;
3227 if (PageChecked(page)) {
3228 ClearPageChecked(page);
3232 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3235 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3236 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3237 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3241 phy_offset >>= inode->i_sb->s_blocksize_bits;
3242 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3243 start, (size_t)(end - start + 1));
3247 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3249 * @inode: The inode we want to perform iput on
3251 * This function uses the generic vfs_inode::i_count to track whether we should
3252 * just decrement it (in case it's > 1) or if this is the last iput then link
3253 * the inode to the delayed iput machinery. Delayed iputs are processed at
3254 * transaction commit time/superblock commit/cleaner kthread.
3256 void btrfs_add_delayed_iput(struct inode *inode)
3258 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3259 struct btrfs_inode *binode = BTRFS_I(inode);
3261 if (atomic_add_unless(&inode->i_count, -1, 1))
3264 spin_lock(&fs_info->delayed_iput_lock);
3265 ASSERT(list_empty(&binode->delayed_iput));
3266 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3267 spin_unlock(&fs_info->delayed_iput_lock);
3270 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3273 spin_lock(&fs_info->delayed_iput_lock);
3274 while (!list_empty(&fs_info->delayed_iputs)) {
3275 struct btrfs_inode *inode;
3277 inode = list_first_entry(&fs_info->delayed_iputs,
3278 struct btrfs_inode, delayed_iput);
3279 list_del_init(&inode->delayed_iput);
3280 spin_unlock(&fs_info->delayed_iput_lock);
3281 iput(&inode->vfs_inode);
3282 spin_lock(&fs_info->delayed_iput_lock);
3284 spin_unlock(&fs_info->delayed_iput_lock);
3288 * This creates an orphan entry for the given inode in case something goes wrong
3289 * in the middle of an unlink.
3291 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3292 struct btrfs_inode *inode)
3296 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3297 if (ret && ret != -EEXIST) {
3298 btrfs_abort_transaction(trans, ret);
3306 * We have done the delete so we can go ahead and remove the orphan item for
3307 * this particular inode.
3309 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3310 struct btrfs_inode *inode)
3312 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3316 * this cleans up any orphans that may be left on the list from the last use
3319 int btrfs_orphan_cleanup(struct btrfs_root *root)
3321 struct btrfs_fs_info *fs_info = root->fs_info;
3322 struct btrfs_path *path;
3323 struct extent_buffer *leaf;
3324 struct btrfs_key key, found_key;
3325 struct btrfs_trans_handle *trans;
3326 struct inode *inode;
3327 u64 last_objectid = 0;
3328 int ret = 0, nr_unlink = 0;
3330 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3333 path = btrfs_alloc_path();
3338 path->reada = READA_BACK;
3340 key.objectid = BTRFS_ORPHAN_OBJECTID;
3341 key.type = BTRFS_ORPHAN_ITEM_KEY;
3342 key.offset = (u64)-1;
3345 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3350 * if ret == 0 means we found what we were searching for, which
3351 * is weird, but possible, so only screw with path if we didn't
3352 * find the key and see if we have stuff that matches
3356 if (path->slots[0] == 0)
3361 /* pull out the item */
3362 leaf = path->nodes[0];
3363 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3365 /* make sure the item matches what we want */
3366 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3368 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3371 /* release the path since we're done with it */
3372 btrfs_release_path(path);
3375 * this is where we are basically btrfs_lookup, without the
3376 * crossing root thing. we store the inode number in the
3377 * offset of the orphan item.
3380 if (found_key.offset == last_objectid) {
3382 "Error removing orphan entry, stopping orphan cleanup");
3387 last_objectid = found_key.offset;
3389 found_key.objectid = found_key.offset;
3390 found_key.type = BTRFS_INODE_ITEM_KEY;
3391 found_key.offset = 0;
3392 inode = btrfs_iget(fs_info->sb, &found_key, root, NULL);
3393 ret = PTR_ERR_OR_ZERO(inode);
3394 if (ret && ret != -ENOENT)
3397 if (ret == -ENOENT && root == fs_info->tree_root) {
3398 struct btrfs_root *dead_root;
3399 struct btrfs_fs_info *fs_info = root->fs_info;
3400 int is_dead_root = 0;
3403 * this is an orphan in the tree root. Currently these
3404 * could come from 2 sources:
3405 * a) a snapshot deletion in progress
3406 * b) a free space cache inode
3407 * We need to distinguish those two, as the snapshot
3408 * orphan must not get deleted.
3409 * find_dead_roots already ran before us, so if this
3410 * is a snapshot deletion, we should find the root
3411 * in the dead_roots list
3413 spin_lock(&fs_info->trans_lock);
3414 list_for_each_entry(dead_root, &fs_info->dead_roots,
3416 if (dead_root->root_key.objectid ==
3417 found_key.objectid) {
3422 spin_unlock(&fs_info->trans_lock);
3424 /* prevent this orphan from being found again */
3425 key.offset = found_key.objectid - 1;
3432 * If we have an inode with links, there are a couple of
3433 * possibilities. Old kernels (before v3.12) used to create an
3434 * orphan item for truncate indicating that there were possibly
3435 * extent items past i_size that needed to be deleted. In v3.12,
3436 * truncate was changed to update i_size in sync with the extent
3437 * items, but the (useless) orphan item was still created. Since
3438 * v4.18, we don't create the orphan item for truncate at all.
3440 * So, this item could mean that we need to do a truncate, but
3441 * only if this filesystem was last used on a pre-v3.12 kernel
3442 * and was not cleanly unmounted. The odds of that are quite
3443 * slim, and it's a pain to do the truncate now, so just delete
3446 * It's also possible that this orphan item was supposed to be
3447 * deleted but wasn't. The inode number may have been reused,
3448 * but either way, we can delete the orphan item.
3450 if (ret == -ENOENT || inode->i_nlink) {
3453 trans = btrfs_start_transaction(root, 1);
3454 if (IS_ERR(trans)) {
3455 ret = PTR_ERR(trans);
3458 btrfs_debug(fs_info, "auto deleting %Lu",
3459 found_key.objectid);
3460 ret = btrfs_del_orphan_item(trans, root,
3461 found_key.objectid);
3462 btrfs_end_transaction(trans);
3470 /* this will do delete_inode and everything for us */
3473 /* release the path since we're done with it */
3474 btrfs_release_path(path);
3476 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3478 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3479 trans = btrfs_join_transaction(root);
3481 btrfs_end_transaction(trans);
3485 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3489 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3490 btrfs_free_path(path);
3495 * very simple check to peek ahead in the leaf looking for xattrs. If we
3496 * don't find any xattrs, we know there can't be any acls.
3498 * slot is the slot the inode is in, objectid is the objectid of the inode
3500 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3501 int slot, u64 objectid,
3502 int *first_xattr_slot)
3504 u32 nritems = btrfs_header_nritems(leaf);
3505 struct btrfs_key found_key;
3506 static u64 xattr_access = 0;
3507 static u64 xattr_default = 0;
3510 if (!xattr_access) {
3511 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3512 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3513 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3514 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3518 *first_xattr_slot = -1;
3519 while (slot < nritems) {
3520 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3522 /* we found a different objectid, there must not be acls */
3523 if (found_key.objectid != objectid)
3526 /* we found an xattr, assume we've got an acl */
3527 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3528 if (*first_xattr_slot == -1)
3529 *first_xattr_slot = slot;
3530 if (found_key.offset == xattr_access ||
3531 found_key.offset == xattr_default)
3536 * we found a key greater than an xattr key, there can't
3537 * be any acls later on
3539 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3546 * it goes inode, inode backrefs, xattrs, extents,
3547 * so if there are a ton of hard links to an inode there can
3548 * be a lot of backrefs. Don't waste time searching too hard,
3549 * this is just an optimization
3554 /* we hit the end of the leaf before we found an xattr or
3555 * something larger than an xattr. We have to assume the inode
3558 if (*first_xattr_slot == -1)
3559 *first_xattr_slot = slot;
3564 * read an inode from the btree into the in-memory inode
3566 static int btrfs_read_locked_inode(struct inode *inode,
3567 struct btrfs_path *in_path)
3569 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3570 struct btrfs_path *path = in_path;
3571 struct extent_buffer *leaf;
3572 struct btrfs_inode_item *inode_item;
3573 struct btrfs_root *root = BTRFS_I(inode)->root;
3574 struct btrfs_key location;
3579 bool filled = false;
3580 int first_xattr_slot;
3582 ret = btrfs_fill_inode(inode, &rdev);
3587 path = btrfs_alloc_path();
3592 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3594 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3596 if (path != in_path)
3597 btrfs_free_path(path);
3601 leaf = path->nodes[0];
3606 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3607 struct btrfs_inode_item);
3608 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3609 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3610 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3611 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3612 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3614 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3615 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3617 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3618 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3620 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3621 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3623 BTRFS_I(inode)->i_otime.tv_sec =
3624 btrfs_timespec_sec(leaf, &inode_item->otime);
3625 BTRFS_I(inode)->i_otime.tv_nsec =
3626 btrfs_timespec_nsec(leaf, &inode_item->otime);
3628 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3629 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3630 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3632 inode_set_iversion_queried(inode,
3633 btrfs_inode_sequence(leaf, inode_item));
3634 inode->i_generation = BTRFS_I(inode)->generation;
3636 rdev = btrfs_inode_rdev(leaf, inode_item);
3638 BTRFS_I(inode)->index_cnt = (u64)-1;
3639 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3643 * If we were modified in the current generation and evicted from memory
3644 * and then re-read we need to do a full sync since we don't have any
3645 * idea about which extents were modified before we were evicted from
3648 * This is required for both inode re-read from disk and delayed inode
3649 * in delayed_nodes_tree.
3651 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3652 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3653 &BTRFS_I(inode)->runtime_flags);
3656 * We don't persist the id of the transaction where an unlink operation
3657 * against the inode was last made. So here we assume the inode might
3658 * have been evicted, and therefore the exact value of last_unlink_trans
3659 * lost, and set it to last_trans to avoid metadata inconsistencies
3660 * between the inode and its parent if the inode is fsync'ed and the log
3661 * replayed. For example, in the scenario:
3664 * ln mydir/foo mydir/bar
3667 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3668 * xfs_io -c fsync mydir/foo
3670 * mount fs, triggers fsync log replay
3672 * We must make sure that when we fsync our inode foo we also log its
3673 * parent inode, otherwise after log replay the parent still has the
3674 * dentry with the "bar" name but our inode foo has a link count of 1
3675 * and doesn't have an inode ref with the name "bar" anymore.
3677 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3678 * but it guarantees correctness at the expense of occasional full
3679 * transaction commits on fsync if our inode is a directory, or if our
3680 * inode is not a directory, logging its parent unnecessarily.
3682 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3685 if (inode->i_nlink != 1 ||
3686 path->slots[0] >= btrfs_header_nritems(leaf))
3689 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3690 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3693 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3694 if (location.type == BTRFS_INODE_REF_KEY) {
3695 struct btrfs_inode_ref *ref;
3697 ref = (struct btrfs_inode_ref *)ptr;
3698 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3699 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3700 struct btrfs_inode_extref *extref;
3702 extref = (struct btrfs_inode_extref *)ptr;
3703 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3708 * try to precache a NULL acl entry for files that don't have
3709 * any xattrs or acls
3711 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3712 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3713 if (first_xattr_slot != -1) {
3714 path->slots[0] = first_xattr_slot;
3715 ret = btrfs_load_inode_props(inode, path);
3718 "error loading props for ino %llu (root %llu): %d",
3719 btrfs_ino(BTRFS_I(inode)),
3720 root->root_key.objectid, ret);
3722 if (path != in_path)
3723 btrfs_free_path(path);
3726 cache_no_acl(inode);
3728 switch (inode->i_mode & S_IFMT) {
3730 inode->i_mapping->a_ops = &btrfs_aops;
3731 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3732 inode->i_fop = &btrfs_file_operations;
3733 inode->i_op = &btrfs_file_inode_operations;
3736 inode->i_fop = &btrfs_dir_file_operations;
3737 inode->i_op = &btrfs_dir_inode_operations;
3740 inode->i_op = &btrfs_symlink_inode_operations;
3741 inode_nohighmem(inode);
3742 inode->i_mapping->a_ops = &btrfs_aops;
3745 inode->i_op = &btrfs_special_inode_operations;
3746 init_special_inode(inode, inode->i_mode, rdev);
3750 btrfs_sync_inode_flags_to_i_flags(inode);
3755 * given a leaf and an inode, copy the inode fields into the leaf
3757 static void fill_inode_item(struct btrfs_trans_handle *trans,
3758 struct extent_buffer *leaf,
3759 struct btrfs_inode_item *item,
3760 struct inode *inode)
3762 struct btrfs_map_token token;
3764 btrfs_init_map_token(&token);
3766 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3767 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3768 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3770 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3771 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3773 btrfs_set_token_timespec_sec(leaf, &item->atime,
3774 inode->i_atime.tv_sec, &token);
3775 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3776 inode->i_atime.tv_nsec, &token);
3778 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3779 inode->i_mtime.tv_sec, &token);
3780 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3781 inode->i_mtime.tv_nsec, &token);
3783 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3784 inode->i_ctime.tv_sec, &token);
3785 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3786 inode->i_ctime.tv_nsec, &token);
3788 btrfs_set_token_timespec_sec(leaf, &item->otime,
3789 BTRFS_I(inode)->i_otime.tv_sec, &token);
3790 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3791 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3793 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3795 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3797 btrfs_set_token_inode_sequence(leaf, item, inode_peek_iversion(inode),
3799 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3800 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3801 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3802 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3806 * copy everything in the in-memory inode into the btree.
3808 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3809 struct btrfs_root *root, struct inode *inode)
3811 struct btrfs_inode_item *inode_item;
3812 struct btrfs_path *path;
3813 struct extent_buffer *leaf;
3816 path = btrfs_alloc_path();
3820 path->leave_spinning = 1;
3821 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3829 leaf = path->nodes[0];
3830 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3831 struct btrfs_inode_item);
3833 fill_inode_item(trans, leaf, inode_item, inode);
3834 btrfs_mark_buffer_dirty(leaf);
3835 btrfs_set_inode_last_trans(trans, inode);
3838 btrfs_free_path(path);
3843 * copy everything in the in-memory inode into the btree.
3845 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3846 struct btrfs_root *root, struct inode *inode)
3848 struct btrfs_fs_info *fs_info = root->fs_info;
3852 * If the inode is a free space inode, we can deadlock during commit
3853 * if we put it into the delayed code.
3855 * The data relocation inode should also be directly updated
3858 if (!btrfs_is_free_space_inode(BTRFS_I(inode))
3859 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3860 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
3861 btrfs_update_root_times(trans, root);
3863 ret = btrfs_delayed_update_inode(trans, root, inode);
3865 btrfs_set_inode_last_trans(trans, inode);
3869 return btrfs_update_inode_item(trans, root, inode);
3872 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3873 struct btrfs_root *root,
3874 struct inode *inode)
3878 ret = btrfs_update_inode(trans, root, inode);
3880 return btrfs_update_inode_item(trans, root, inode);
3885 * unlink helper that gets used here in inode.c and in the tree logging
3886 * recovery code. It remove a link in a directory with a given name, and
3887 * also drops the back refs in the inode to the directory
3889 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3890 struct btrfs_root *root,
3891 struct btrfs_inode *dir,
3892 struct btrfs_inode *inode,
3893 const char *name, int name_len)
3895 struct btrfs_fs_info *fs_info = root->fs_info;
3896 struct btrfs_path *path;
3898 struct extent_buffer *leaf;
3899 struct btrfs_dir_item *di;
3900 struct btrfs_key key;
3902 u64 ino = btrfs_ino(inode);
3903 u64 dir_ino = btrfs_ino(dir);
3905 path = btrfs_alloc_path();
3911 path->leave_spinning = 1;
3912 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3913 name, name_len, -1);
3914 if (IS_ERR_OR_NULL(di)) {
3915 ret = di ? PTR_ERR(di) : -ENOENT;
3918 leaf = path->nodes[0];
3919 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3920 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3923 btrfs_release_path(path);
3926 * If we don't have dir index, we have to get it by looking up
3927 * the inode ref, since we get the inode ref, remove it directly,
3928 * it is unnecessary to do delayed deletion.
3930 * But if we have dir index, needn't search inode ref to get it.
3931 * Since the inode ref is close to the inode item, it is better
3932 * that we delay to delete it, and just do this deletion when
3933 * we update the inode item.
3935 if (inode->dir_index) {
3936 ret = btrfs_delayed_delete_inode_ref(inode);
3938 index = inode->dir_index;
3943 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3947 "failed to delete reference to %.*s, inode %llu parent %llu",
3948 name_len, name, ino, dir_ino);
3949 btrfs_abort_transaction(trans, ret);
3953 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
3955 btrfs_abort_transaction(trans, ret);
3959 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
3961 if (ret != 0 && ret != -ENOENT) {
3962 btrfs_abort_transaction(trans, ret);
3966 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
3971 btrfs_abort_transaction(trans, ret);
3973 btrfs_free_path(path);
3977 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
3978 inode_inc_iversion(&inode->vfs_inode);
3979 inode_inc_iversion(&dir->vfs_inode);
3980 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
3981 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
3982 ret = btrfs_update_inode(trans, root, &dir->vfs_inode);
3987 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3988 struct btrfs_root *root,
3989 struct btrfs_inode *dir, struct btrfs_inode *inode,
3990 const char *name, int name_len)
3993 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3995 drop_nlink(&inode->vfs_inode);
3996 ret = btrfs_update_inode(trans, root, &inode->vfs_inode);
4002 * helper to start transaction for unlink and rmdir.
4004 * unlink and rmdir are special in btrfs, they do not always free space, so
4005 * if we cannot make our reservations the normal way try and see if there is
4006 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4007 * allow the unlink to occur.
4009 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4011 struct btrfs_root *root = BTRFS_I(dir)->root;
4014 * 1 for the possible orphan item
4015 * 1 for the dir item
4016 * 1 for the dir index
4017 * 1 for the inode ref
4020 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4023 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4025 struct btrfs_root *root = BTRFS_I(dir)->root;
4026 struct btrfs_trans_handle *trans;
4027 struct inode *inode = d_inode(dentry);
4030 trans = __unlink_start_trans(dir);
4032 return PTR_ERR(trans);
4034 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4037 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4038 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4039 dentry->d_name.len);
4043 if (inode->i_nlink == 0) {
4044 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4050 btrfs_end_transaction(trans);
4051 btrfs_btree_balance_dirty(root->fs_info);
4055 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4056 struct inode *dir, u64 objectid,
4057 const char *name, int name_len)
4059 struct btrfs_root *root = BTRFS_I(dir)->root;
4060 struct btrfs_path *path;
4061 struct extent_buffer *leaf;
4062 struct btrfs_dir_item *di;
4063 struct btrfs_key key;
4066 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4068 path = btrfs_alloc_path();
4072 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4073 name, name_len, -1);
4074 if (IS_ERR_OR_NULL(di)) {
4075 ret = di ? PTR_ERR(di) : -ENOENT;
4079 leaf = path->nodes[0];
4080 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4081 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4082 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4084 btrfs_abort_transaction(trans, ret);
4087 btrfs_release_path(path);
4089 ret = btrfs_del_root_ref(trans, objectid, root->root_key.objectid,
4090 dir_ino, &index, name, name_len);
4092 if (ret != -ENOENT) {
4093 btrfs_abort_transaction(trans, ret);
4096 di = btrfs_search_dir_index_item(root, path, dir_ino,
4098 if (IS_ERR_OR_NULL(di)) {
4103 btrfs_abort_transaction(trans, ret);
4107 leaf = path->nodes[0];
4108 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4111 btrfs_release_path(path);
4113 ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index);
4115 btrfs_abort_transaction(trans, ret);
4119 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4120 inode_inc_iversion(dir);
4121 dir->i_mtime = dir->i_ctime = current_time(dir);
4122 ret = btrfs_update_inode_fallback(trans, root, dir);
4124 btrfs_abort_transaction(trans, ret);
4126 btrfs_free_path(path);
4131 * Helper to check if the subvolume references other subvolumes or if it's
4134 static noinline int may_destroy_subvol(struct btrfs_root *root)
4136 struct btrfs_fs_info *fs_info = root->fs_info;
4137 struct btrfs_path *path;
4138 struct btrfs_dir_item *di;
4139 struct btrfs_key key;
4143 path = btrfs_alloc_path();
4147 /* Make sure this root isn't set as the default subvol */
4148 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4149 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4150 dir_id, "default", 7, 0);
4151 if (di && !IS_ERR(di)) {
4152 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4153 if (key.objectid == root->root_key.objectid) {
4156 "deleting default subvolume %llu is not allowed",
4160 btrfs_release_path(path);
4163 key.objectid = root->root_key.objectid;
4164 key.type = BTRFS_ROOT_REF_KEY;
4165 key.offset = (u64)-1;
4167 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4173 if (path->slots[0] > 0) {
4175 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4176 if (key.objectid == root->root_key.objectid &&
4177 key.type == BTRFS_ROOT_REF_KEY)
4181 btrfs_free_path(path);
4185 /* Delete all dentries for inodes belonging to the root */
4186 static void btrfs_prune_dentries(struct btrfs_root *root)
4188 struct btrfs_fs_info *fs_info = root->fs_info;
4189 struct rb_node *node;
4190 struct rb_node *prev;
4191 struct btrfs_inode *entry;
4192 struct inode *inode;
4195 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
4196 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4198 spin_lock(&root->inode_lock);
4200 node = root->inode_tree.rb_node;
4204 entry = rb_entry(node, struct btrfs_inode, rb_node);
4206 if (objectid < btrfs_ino(entry))
4207 node = node->rb_left;
4208 else if (objectid > btrfs_ino(entry))
4209 node = node->rb_right;
4215 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4216 if (objectid <= btrfs_ino(entry)) {
4220 prev = rb_next(prev);
4224 entry = rb_entry(node, struct btrfs_inode, rb_node);
4225 objectid = btrfs_ino(entry) + 1;
4226 inode = igrab(&entry->vfs_inode);
4228 spin_unlock(&root->inode_lock);
4229 if (atomic_read(&inode->i_count) > 1)
4230 d_prune_aliases(inode);
4232 * btrfs_drop_inode will have it removed from the inode
4233 * cache when its usage count hits zero.
4237 spin_lock(&root->inode_lock);
4241 if (cond_resched_lock(&root->inode_lock))
4244 node = rb_next(node);
4246 spin_unlock(&root->inode_lock);
4249 int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
4251 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4252 struct btrfs_root *root = BTRFS_I(dir)->root;
4253 struct inode *inode = d_inode(dentry);
4254 struct btrfs_root *dest = BTRFS_I(inode)->root;
4255 struct btrfs_trans_handle *trans;
4256 struct btrfs_block_rsv block_rsv;
4262 * Don't allow to delete a subvolume with send in progress. This is
4263 * inside the inode lock so the error handling that has to drop the bit
4264 * again is not run concurrently.
4266 spin_lock(&dest->root_item_lock);
4267 if (dest->send_in_progress) {
4268 spin_unlock(&dest->root_item_lock);
4270 "attempt to delete subvolume %llu during send",
4271 dest->root_key.objectid);
4274 root_flags = btrfs_root_flags(&dest->root_item);
4275 btrfs_set_root_flags(&dest->root_item,
4276 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4277 spin_unlock(&dest->root_item_lock);
4279 down_write(&fs_info->subvol_sem);
4281 err = may_destroy_subvol(dest);
4285 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4287 * One for dir inode,
4288 * two for dir entries,
4289 * two for root ref/backref.
4291 err = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4295 trans = btrfs_start_transaction(root, 0);
4296 if (IS_ERR(trans)) {
4297 err = PTR_ERR(trans);
4300 trans->block_rsv = &block_rsv;
4301 trans->bytes_reserved = block_rsv.size;
4303 btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
4305 ret = btrfs_unlink_subvol(trans, dir, dest->root_key.objectid,
4306 dentry->d_name.name, dentry->d_name.len);
4309 btrfs_abort_transaction(trans, ret);
4313 btrfs_record_root_in_trans(trans, dest);
4315 memset(&dest->root_item.drop_progress, 0,
4316 sizeof(dest->root_item.drop_progress));
4317 dest->root_item.drop_level = 0;
4318 btrfs_set_root_refs(&dest->root_item, 0);
4320 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4321 ret = btrfs_insert_orphan_item(trans,
4323 dest->root_key.objectid);
4325 btrfs_abort_transaction(trans, ret);
4331 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4332 BTRFS_UUID_KEY_SUBVOL,
4333 dest->root_key.objectid);
4334 if (ret && ret != -ENOENT) {
4335 btrfs_abort_transaction(trans, ret);
4339 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4340 ret = btrfs_uuid_tree_remove(trans,
4341 dest->root_item.received_uuid,
4342 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4343 dest->root_key.objectid);
4344 if (ret && ret != -ENOENT) {
4345 btrfs_abort_transaction(trans, ret);
4352 trans->block_rsv = NULL;
4353 trans->bytes_reserved = 0;
4354 ret = btrfs_end_transaction(trans);
4357 inode->i_flags |= S_DEAD;
4359 btrfs_subvolume_release_metadata(fs_info, &block_rsv);
4361 up_write(&fs_info->subvol_sem);
4363 spin_lock(&dest->root_item_lock);
4364 root_flags = btrfs_root_flags(&dest->root_item);
4365 btrfs_set_root_flags(&dest->root_item,
4366 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4367 spin_unlock(&dest->root_item_lock);
4369 d_invalidate(dentry);
4370 btrfs_prune_dentries(dest);
4371 ASSERT(dest->send_in_progress == 0);
4374 if (dest->ino_cache_inode) {
4375 iput(dest->ino_cache_inode);
4376 dest->ino_cache_inode = NULL;
4383 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4385 struct inode *inode = d_inode(dentry);
4387 struct btrfs_root *root = BTRFS_I(dir)->root;
4388 struct btrfs_trans_handle *trans;
4389 u64 last_unlink_trans;
4391 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4393 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4394 return btrfs_delete_subvolume(dir, dentry);
4396 trans = __unlink_start_trans(dir);
4398 return PTR_ERR(trans);
4400 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4401 err = btrfs_unlink_subvol(trans, dir,
4402 BTRFS_I(inode)->location.objectid,
4403 dentry->d_name.name,
4404 dentry->d_name.len);
4408 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4412 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4414 /* now the directory is empty */
4415 err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4416 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4417 dentry->d_name.len);
4419 btrfs_i_size_write(BTRFS_I(inode), 0);
4421 * Propagate the last_unlink_trans value of the deleted dir to
4422 * its parent directory. This is to prevent an unrecoverable
4423 * log tree in the case we do something like this:
4425 * 2) create snapshot under dir foo
4426 * 3) delete the snapshot
4429 * 6) fsync foo or some file inside foo
4431 if (last_unlink_trans >= trans->transid)
4432 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4435 btrfs_end_transaction(trans);
4436 btrfs_btree_balance_dirty(root->fs_info);
4441 static int truncate_space_check(struct btrfs_trans_handle *trans,
4442 struct btrfs_root *root,
4445 struct btrfs_fs_info *fs_info = root->fs_info;
4449 * This is only used to apply pressure to the enospc system, we don't
4450 * intend to use this reservation at all.
4452 bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted);
4453 bytes_deleted *= fs_info->nodesize;
4454 ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
4455 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4457 trace_btrfs_space_reservation(fs_info, "transaction",
4460 trans->bytes_reserved += bytes_deleted;
4467 * Return this if we need to call truncate_block for the last bit of the
4470 #define NEED_TRUNCATE_BLOCK 1
4473 * this can truncate away extent items, csum items and directory items.
4474 * It starts at a high offset and removes keys until it can't find
4475 * any higher than new_size
4477 * csum items that cross the new i_size are truncated to the new size
4480 * min_type is the minimum key type to truncate down to. If set to 0, this
4481 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4483 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4484 struct btrfs_root *root,
4485 struct inode *inode,
4486 u64 new_size, u32 min_type)
4488 struct btrfs_fs_info *fs_info = root->fs_info;
4489 struct btrfs_path *path;
4490 struct extent_buffer *leaf;
4491 struct btrfs_file_extent_item *fi;
4492 struct btrfs_key key;
4493 struct btrfs_key found_key;
4494 u64 extent_start = 0;
4495 u64 extent_num_bytes = 0;
4496 u64 extent_offset = 0;
4498 u64 last_size = new_size;
4499 u32 found_type = (u8)-1;
4502 int pending_del_nr = 0;
4503 int pending_del_slot = 0;
4504 int extent_type = -1;
4506 u64 ino = btrfs_ino(BTRFS_I(inode));
4507 u64 bytes_deleted = 0;
4508 bool be_nice = false;
4509 bool should_throttle = false;
4510 bool should_end = false;
4512 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4515 * for non-free space inodes and ref cows, we want to back off from
4518 if (!btrfs_is_free_space_inode(BTRFS_I(inode)) &&
4519 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4522 path = btrfs_alloc_path();
4525 path->reada = READA_BACK;
4528 * We want to drop from the next block forward in case this new size is
4529 * not block aligned since we will be keeping the last block of the
4530 * extent just the way it is.
4532 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4533 root == fs_info->tree_root)
4534 btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size,
4535 fs_info->sectorsize),
4539 * This function is also used to drop the items in the log tree before
4540 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4541 * it is used to drop the loged items. So we shouldn't kill the delayed
4544 if (min_type == 0 && root == BTRFS_I(inode)->root)
4545 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
4548 key.offset = (u64)-1;
4553 * with a 16K leaf size and 128MB extents, you can actually queue
4554 * up a huge file in a single leaf. Most of the time that
4555 * bytes_deleted is > 0, it will be huge by the time we get here
4557 if (be_nice && bytes_deleted > SZ_32M &&
4558 btrfs_should_end_transaction(trans)) {
4563 path->leave_spinning = 1;
4564 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4570 /* there are no items in the tree for us to truncate, we're
4573 if (path->slots[0] == 0)
4580 leaf = path->nodes[0];
4581 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4582 found_type = found_key.type;
4584 if (found_key.objectid != ino)
4587 if (found_type < min_type)
4590 item_end = found_key.offset;
4591 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4592 fi = btrfs_item_ptr(leaf, path->slots[0],
4593 struct btrfs_file_extent_item);
4594 extent_type = btrfs_file_extent_type(leaf, fi);
4595 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4597 btrfs_file_extent_num_bytes(leaf, fi);
4599 trace_btrfs_truncate_show_fi_regular(
4600 BTRFS_I(inode), leaf, fi,
4602 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4603 item_end += btrfs_file_extent_ram_bytes(leaf,
4606 trace_btrfs_truncate_show_fi_inline(
4607 BTRFS_I(inode), leaf, fi, path->slots[0],
4612 if (found_type > min_type) {
4615 if (item_end < new_size)
4617 if (found_key.offset >= new_size)
4623 /* FIXME, shrink the extent if the ref count is only 1 */
4624 if (found_type != BTRFS_EXTENT_DATA_KEY)
4627 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4629 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4631 u64 orig_num_bytes =
4632 btrfs_file_extent_num_bytes(leaf, fi);
4633 extent_num_bytes = ALIGN(new_size -
4635 fs_info->sectorsize);
4636 btrfs_set_file_extent_num_bytes(leaf, fi,
4638 num_dec = (orig_num_bytes -
4640 if (test_bit(BTRFS_ROOT_REF_COWS,
4643 inode_sub_bytes(inode, num_dec);
4644 btrfs_mark_buffer_dirty(leaf);
4647 btrfs_file_extent_disk_num_bytes(leaf,
4649 extent_offset = found_key.offset -
4650 btrfs_file_extent_offset(leaf, fi);
4652 /* FIXME blocksize != 4096 */
4653 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4654 if (extent_start != 0) {
4656 if (test_bit(BTRFS_ROOT_REF_COWS,
4658 inode_sub_bytes(inode, num_dec);
4661 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4663 * we can't truncate inline items that have had
4667 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4668 btrfs_file_extent_other_encoding(leaf, fi) == 0 &&
4669 btrfs_file_extent_compression(leaf, fi) == 0) {
4670 u32 size = (u32)(new_size - found_key.offset);
4672 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4673 size = btrfs_file_extent_calc_inline_size(size);
4674 btrfs_truncate_item(root->fs_info, path, size, 1);
4675 } else if (!del_item) {
4677 * We have to bail so the last_size is set to
4678 * just before this extent.
4680 ret = NEED_TRUNCATE_BLOCK;
4684 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4685 inode_sub_bytes(inode, item_end + 1 - new_size);
4689 last_size = found_key.offset;
4691 last_size = new_size;
4693 if (!pending_del_nr) {
4694 /* no pending yet, add ourselves */
4695 pending_del_slot = path->slots[0];
4697 } else if (pending_del_nr &&
4698 path->slots[0] + 1 == pending_del_slot) {
4699 /* hop on the pending chunk */
4701 pending_del_slot = path->slots[0];
4708 should_throttle = false;
4711 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4712 root == fs_info->tree_root)) {
4713 btrfs_set_path_blocking(path);
4714 bytes_deleted += extent_num_bytes;
4715 ret = btrfs_free_extent(trans, root, extent_start,
4716 extent_num_bytes, 0,
4717 btrfs_header_owner(leaf),
4718 ino, extent_offset);
4720 btrfs_abort_transaction(trans, ret);
4723 if (btrfs_should_throttle_delayed_refs(trans))
4724 btrfs_async_run_delayed_refs(fs_info,
4725 trans->delayed_ref_updates * 2,
4728 if (truncate_space_check(trans, root,
4729 extent_num_bytes)) {
4732 if (btrfs_should_throttle_delayed_refs(trans))
4733 should_throttle = true;
4737 if (found_type == BTRFS_INODE_ITEM_KEY)
4740 if (path->slots[0] == 0 ||
4741 path->slots[0] != pending_del_slot ||
4742 should_throttle || should_end) {
4743 if (pending_del_nr) {
4744 ret = btrfs_del_items(trans, root, path,
4748 btrfs_abort_transaction(trans, ret);
4753 btrfs_release_path(path);
4754 if (should_throttle) {
4755 unsigned long updates = trans->delayed_ref_updates;
4757 trans->delayed_ref_updates = 0;
4758 ret = btrfs_run_delayed_refs(trans,
4765 * if we failed to refill our space rsv, bail out
4766 * and let the transaction restart
4778 if (ret >= 0 && pending_del_nr) {
4781 err = btrfs_del_items(trans, root, path, pending_del_slot,
4784 btrfs_abort_transaction(trans, err);
4788 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4789 ASSERT(last_size >= new_size);
4790 if (!ret && last_size > new_size)
4791 last_size = new_size;
4792 btrfs_ordered_update_i_size(inode, last_size, NULL);
4795 btrfs_free_path(path);
4797 if (be_nice && bytes_deleted > SZ_32M && (ret >= 0 || ret == -EAGAIN)) {
4798 unsigned long updates = trans->delayed_ref_updates;
4802 trans->delayed_ref_updates = 0;
4803 err = btrfs_run_delayed_refs(trans, updates * 2);
4812 * btrfs_truncate_block - read, zero a chunk and write a block
4813 * @inode - inode that we're zeroing
4814 * @from - the offset to start zeroing
4815 * @len - the length to zero, 0 to zero the entire range respective to the
4817 * @front - zero up to the offset instead of from the offset on
4819 * This will find the block for the "from" offset and cow the block and zero the
4820 * part we want to zero. This is used with truncate and hole punching.
4822 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4825 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4826 struct address_space *mapping = inode->i_mapping;
4827 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4828 struct btrfs_ordered_extent *ordered;
4829 struct extent_state *cached_state = NULL;
4830 struct extent_changeset *data_reserved = NULL;
4832 u32 blocksize = fs_info->sectorsize;
4833 pgoff_t index = from >> PAGE_SHIFT;
4834 unsigned offset = from & (blocksize - 1);
4836 gfp_t mask = btrfs_alloc_write_mask(mapping);
4841 if (IS_ALIGNED(offset, blocksize) &&
4842 (!len || IS_ALIGNED(len, blocksize)))
4845 block_start = round_down(from, blocksize);
4846 block_end = block_start + blocksize - 1;
4848 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
4849 block_start, blocksize);
4854 page = find_or_create_page(mapping, index, mask);
4856 btrfs_delalloc_release_space(inode, data_reserved,
4857 block_start, blocksize, true);
4858 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize, true);
4863 if (!PageUptodate(page)) {
4864 ret = btrfs_readpage(NULL, page);
4866 if (page->mapping != mapping) {
4871 if (!PageUptodate(page)) {
4876 wait_on_page_writeback(page);
4878 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4879 set_page_extent_mapped(page);
4881 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4883 unlock_extent_cached(io_tree, block_start, block_end,
4887 btrfs_start_ordered_extent(inode, ordered, 1);
4888 btrfs_put_ordered_extent(ordered);
4892 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4893 EXTENT_DIRTY | EXTENT_DELALLOC |
4894 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4895 0, 0, &cached_state);
4897 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4900 unlock_extent_cached(io_tree, block_start, block_end,
4905 if (offset != blocksize) {
4907 len = blocksize - offset;
4910 memset(kaddr + (block_start - page_offset(page)),
4913 memset(kaddr + (block_start - page_offset(page)) + offset,
4915 flush_dcache_page(page);
4918 ClearPageChecked(page);
4919 set_page_dirty(page);
4920 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4924 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4926 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize, (ret != 0));
4930 extent_changeset_free(data_reserved);
4934 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4935 u64 offset, u64 len)
4937 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4938 struct btrfs_trans_handle *trans;
4942 * Still need to make sure the inode looks like it's been updated so
4943 * that any holes get logged if we fsync.
4945 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4946 BTRFS_I(inode)->last_trans = fs_info->generation;
4947 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4948 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4953 * 1 - for the one we're dropping
4954 * 1 - for the one we're adding
4955 * 1 - for updating the inode.
4957 trans = btrfs_start_transaction(root, 3);
4959 return PTR_ERR(trans);
4961 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4963 btrfs_abort_transaction(trans, ret);
4964 btrfs_end_transaction(trans);
4968 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)),
4969 offset, 0, 0, len, 0, len, 0, 0, 0);
4971 btrfs_abort_transaction(trans, ret);
4973 btrfs_update_inode(trans, root, inode);
4974 btrfs_end_transaction(trans);
4979 * This function puts in dummy file extents for the area we're creating a hole
4980 * for. So if we are truncating this file to a larger size we need to insert
4981 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4982 * the range between oldsize and size
4984 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4986 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4987 struct btrfs_root *root = BTRFS_I(inode)->root;
4988 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4989 struct extent_map *em = NULL;
4990 struct extent_state *cached_state = NULL;
4991 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4992 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4993 u64 block_end = ALIGN(size, fs_info->sectorsize);
5000 * If our size started in the middle of a block we need to zero out the
5001 * rest of the block before we expand the i_size, otherwise we could
5002 * expose stale data.
5004 err = btrfs_truncate_block(inode, oldsize, 0, 0);
5008 if (size <= hole_start)
5012 struct btrfs_ordered_extent *ordered;
5014 lock_extent_bits(io_tree, hole_start, block_end - 1,
5016 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), hole_start,
5017 block_end - hole_start);
5020 unlock_extent_cached(io_tree, hole_start, block_end - 1,
5022 btrfs_start_ordered_extent(inode, ordered, 1);
5023 btrfs_put_ordered_extent(ordered);
5026 cur_offset = hole_start;
5028 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
5029 block_end - cur_offset, 0);
5035 last_byte = min(extent_map_end(em), block_end);
5036 last_byte = ALIGN(last_byte, fs_info->sectorsize);
5037 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5038 struct extent_map *hole_em;
5039 hole_size = last_byte - cur_offset;
5041 err = maybe_insert_hole(root, inode, cur_offset,
5045 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
5046 cur_offset + hole_size - 1, 0);
5047 hole_em = alloc_extent_map();
5049 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5050 &BTRFS_I(inode)->runtime_flags);
5053 hole_em->start = cur_offset;
5054 hole_em->len = hole_size;
5055 hole_em->orig_start = cur_offset;
5057 hole_em->block_start = EXTENT_MAP_HOLE;
5058 hole_em->block_len = 0;
5059 hole_em->orig_block_len = 0;
5060 hole_em->ram_bytes = hole_size;
5061 hole_em->bdev = fs_info->fs_devices->latest_bdev;
5062 hole_em->compress_type = BTRFS_COMPRESS_NONE;
5063 hole_em->generation = fs_info->generation;
5066 write_lock(&em_tree->lock);
5067 err = add_extent_mapping(em_tree, hole_em, 1);
5068 write_unlock(&em_tree->lock);
5071 btrfs_drop_extent_cache(BTRFS_I(inode),
5076 free_extent_map(hole_em);
5079 free_extent_map(em);
5081 cur_offset = last_byte;
5082 if (cur_offset >= block_end)
5085 free_extent_map(em);
5086 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
5090 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5092 struct btrfs_root *root = BTRFS_I(inode)->root;
5093 struct btrfs_trans_handle *trans;
5094 loff_t oldsize = i_size_read(inode);
5095 loff_t newsize = attr->ia_size;
5096 int mask = attr->ia_valid;
5100 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5101 * special case where we need to update the times despite not having
5102 * these flags set. For all other operations the VFS set these flags
5103 * explicitly if it wants a timestamp update.
5105 if (newsize != oldsize) {
5106 inode_inc_iversion(inode);
5107 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5108 inode->i_ctime = inode->i_mtime =
5109 current_time(inode);
5112 if (newsize > oldsize) {
5114 * Don't do an expanding truncate while snapshotting is ongoing.
5115 * This is to ensure the snapshot captures a fully consistent
5116 * state of this file - if the snapshot captures this expanding
5117 * truncation, it must capture all writes that happened before
5120 btrfs_wait_for_snapshot_creation(root);
5121 ret = btrfs_cont_expand(inode, oldsize, newsize);
5123 btrfs_end_write_no_snapshotting(root);
5127 trans = btrfs_start_transaction(root, 1);
5128 if (IS_ERR(trans)) {
5129 btrfs_end_write_no_snapshotting(root);
5130 return PTR_ERR(trans);
5133 i_size_write(inode, newsize);
5134 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
5135 pagecache_isize_extended(inode, oldsize, newsize);
5136 ret = btrfs_update_inode(trans, root, inode);
5137 btrfs_end_write_no_snapshotting(root);
5138 btrfs_end_transaction(trans);
5142 * We're truncating a file that used to have good data down to
5143 * zero. Make sure it gets into the ordered flush list so that
5144 * any new writes get down to disk quickly.
5147 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
5148 &BTRFS_I(inode)->runtime_flags);
5150 truncate_setsize(inode, newsize);
5152 /* Disable nonlocked read DIO to avoid the end less truncate */
5153 btrfs_inode_block_unlocked_dio(BTRFS_I(inode));
5154 inode_dio_wait(inode);
5155 btrfs_inode_resume_unlocked_dio(BTRFS_I(inode));
5157 ret = btrfs_truncate(inode, newsize == oldsize);
5158 if (ret && inode->i_nlink) {
5162 * Truncate failed, so fix up the in-memory size. We
5163 * adjusted disk_i_size down as we removed extents, so
5164 * wait for disk_i_size to be stable and then update the
5165 * in-memory size to match.
5167 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5170 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5177 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5179 struct inode *inode = d_inode(dentry);
5180 struct btrfs_root *root = BTRFS_I(inode)->root;
5183 if (btrfs_root_readonly(root))
5186 err = setattr_prepare(dentry, attr);
5190 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5191 err = btrfs_setsize(inode, attr);
5196 if (attr->ia_valid) {
5197 setattr_copy(inode, attr);
5198 inode_inc_iversion(inode);
5199 err = btrfs_dirty_inode(inode);
5201 if (!err && attr->ia_valid & ATTR_MODE)
5202 err = posix_acl_chmod(inode, inode->i_mode);
5209 * While truncating the inode pages during eviction, we get the VFS calling
5210 * btrfs_invalidatepage() against each page of the inode. This is slow because
5211 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5212 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5213 * extent_state structures over and over, wasting lots of time.
5215 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5216 * those expensive operations on a per page basis and do only the ordered io
5217 * finishing, while we release here the extent_map and extent_state structures,
5218 * without the excessive merging and splitting.
5220 static void evict_inode_truncate_pages(struct inode *inode)
5222 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5223 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5224 struct rb_node *node;
5226 ASSERT(inode->i_state & I_FREEING);
5227 truncate_inode_pages_final(&inode->i_data);
5229 write_lock(&map_tree->lock);
5230 while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) {
5231 struct extent_map *em;
5233 node = rb_first_cached(&map_tree->map);
5234 em = rb_entry(node, struct extent_map, rb_node);
5235 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5236 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5237 remove_extent_mapping(map_tree, em);
5238 free_extent_map(em);
5239 if (need_resched()) {
5240 write_unlock(&map_tree->lock);
5242 write_lock(&map_tree->lock);
5245 write_unlock(&map_tree->lock);
5248 * Keep looping until we have no more ranges in the io tree.
5249 * We can have ongoing bios started by readpages (called from readahead)
5250 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5251 * still in progress (unlocked the pages in the bio but did not yet
5252 * unlocked the ranges in the io tree). Therefore this means some
5253 * ranges can still be locked and eviction started because before
5254 * submitting those bios, which are executed by a separate task (work
5255 * queue kthread), inode references (inode->i_count) were not taken
5256 * (which would be dropped in the end io callback of each bio).
5257 * Therefore here we effectively end up waiting for those bios and
5258 * anyone else holding locked ranges without having bumped the inode's
5259 * reference count - if we don't do it, when they access the inode's
5260 * io_tree to unlock a range it may be too late, leading to an
5261 * use-after-free issue.
5263 spin_lock(&io_tree->lock);
5264 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5265 struct extent_state *state;
5266 struct extent_state *cached_state = NULL;
5269 unsigned state_flags;
5271 node = rb_first(&io_tree->state);
5272 state = rb_entry(node, struct extent_state, rb_node);
5273 start = state->start;
5275 state_flags = state->state;
5276 spin_unlock(&io_tree->lock);
5278 lock_extent_bits(io_tree, start, end, &cached_state);
5281 * If still has DELALLOC flag, the extent didn't reach disk,
5282 * and its reserved space won't be freed by delayed_ref.
5283 * So we need to free its reserved space here.
5284 * (Refer to comment in btrfs_invalidatepage, case 2)
5286 * Note, end is the bytenr of last byte, so we need + 1 here.
5288 if (state_flags & EXTENT_DELALLOC)
5289 btrfs_qgroup_free_data(inode, NULL, start, end - start + 1);
5291 clear_extent_bit(io_tree, start, end,
5292 EXTENT_LOCKED | EXTENT_DIRTY |
5293 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5294 EXTENT_DEFRAG, 1, 1, &cached_state);
5297 spin_lock(&io_tree->lock);
5299 spin_unlock(&io_tree->lock);
5302 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5303 struct btrfs_block_rsv *rsv)
5305 struct btrfs_fs_info *fs_info = root->fs_info;
5306 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5310 struct btrfs_trans_handle *trans;
5313 ret = btrfs_block_rsv_refill(root, rsv, rsv->size,
5314 BTRFS_RESERVE_FLUSH_LIMIT);
5316 if (ret && ++failures > 2) {
5318 "could not allocate space for a delete; will truncate on mount");
5319 return ERR_PTR(-ENOSPC);
5322 trans = btrfs_join_transaction(root);
5323 if (IS_ERR(trans) || !ret)
5327 * Try to steal from the global reserve if there is space for
5330 if (!btrfs_check_space_for_delayed_refs(trans) &&
5331 !btrfs_block_rsv_migrate(global_rsv, rsv, rsv->size, false))
5334 /* If not, commit and try again. */
5335 ret = btrfs_commit_transaction(trans);
5337 return ERR_PTR(ret);
5341 void btrfs_evict_inode(struct inode *inode)
5343 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5344 struct btrfs_trans_handle *trans;
5345 struct btrfs_root *root = BTRFS_I(inode)->root;
5346 struct btrfs_block_rsv *rsv;
5349 trace_btrfs_inode_evict(inode);
5356 evict_inode_truncate_pages(inode);
5358 if (inode->i_nlink &&
5359 ((btrfs_root_refs(&root->root_item) != 0 &&
5360 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5361 btrfs_is_free_space_inode(BTRFS_I(inode))))
5364 if (is_bad_inode(inode))
5367 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5369 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5372 if (inode->i_nlink > 0) {
5373 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5374 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5378 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5382 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5385 rsv->size = btrfs_calc_trunc_metadata_size(fs_info, 1);
5388 btrfs_i_size_write(BTRFS_I(inode), 0);
5391 trans = evict_refill_and_join(root, rsv);
5395 trans->block_rsv = rsv;
5397 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5398 trans->block_rsv = &fs_info->trans_block_rsv;
5399 btrfs_end_transaction(trans);
5400 btrfs_btree_balance_dirty(fs_info);
5401 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5408 * Errors here aren't a big deal, it just means we leave orphan items in
5409 * the tree. They will be cleaned up on the next mount. If the inode
5410 * number gets reused, cleanup deletes the orphan item without doing
5411 * anything, and unlink reuses the existing orphan item.
5413 * If it turns out that we are dropping too many of these, we might want
5414 * to add a mechanism for retrying these after a commit.
5416 trans = evict_refill_and_join(root, rsv);
5417 if (!IS_ERR(trans)) {
5418 trans->block_rsv = rsv;
5419 btrfs_orphan_del(trans, BTRFS_I(inode));
5420 trans->block_rsv = &fs_info->trans_block_rsv;
5421 btrfs_end_transaction(trans);
5424 if (!(root == fs_info->tree_root ||
5425 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5426 btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
5429 btrfs_free_block_rsv(fs_info, rsv);
5432 * If we didn't successfully delete, the orphan item will still be in
5433 * the tree and we'll retry on the next mount. Again, we might also want
5434 * to retry these periodically in the future.
5436 btrfs_remove_delayed_node(BTRFS_I(inode));
5441 * this returns the key found in the dir entry in the location pointer.
5442 * If no dir entries were found, returns -ENOENT.
5443 * If found a corrupted location in dir entry, returns -EUCLEAN.
5445 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5446 struct btrfs_key *location)
5448 const char *name = dentry->d_name.name;
5449 int namelen = dentry->d_name.len;
5450 struct btrfs_dir_item *di;
5451 struct btrfs_path *path;
5452 struct btrfs_root *root = BTRFS_I(dir)->root;
5455 path = btrfs_alloc_path();
5459 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5461 if (IS_ERR_OR_NULL(di)) {
5462 ret = di ? PTR_ERR(di) : -ENOENT;
5466 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5467 if (location->type != BTRFS_INODE_ITEM_KEY &&
5468 location->type != BTRFS_ROOT_ITEM_KEY) {
5470 btrfs_warn(root->fs_info,
5471 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5472 __func__, name, btrfs_ino(BTRFS_I(dir)),
5473 location->objectid, location->type, location->offset);
5476 btrfs_free_path(path);
5481 * when we hit a tree root in a directory, the btrfs part of the inode
5482 * needs to be changed to reflect the root directory of the tree root. This
5483 * is kind of like crossing a mount point.
5485 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5487 struct dentry *dentry,
5488 struct btrfs_key *location,
5489 struct btrfs_root **sub_root)
5491 struct btrfs_path *path;
5492 struct btrfs_root *new_root;
5493 struct btrfs_root_ref *ref;
5494 struct extent_buffer *leaf;
5495 struct btrfs_key key;
5499 path = btrfs_alloc_path();
5506 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5507 key.type = BTRFS_ROOT_REF_KEY;
5508 key.offset = location->objectid;
5510 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5517 leaf = path->nodes[0];
5518 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5519 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5520 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5523 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5524 (unsigned long)(ref + 1),
5525 dentry->d_name.len);
5529 btrfs_release_path(path);
5531 new_root = btrfs_read_fs_root_no_name(fs_info, location);
5532 if (IS_ERR(new_root)) {
5533 err = PTR_ERR(new_root);
5537 *sub_root = new_root;
5538 location->objectid = btrfs_root_dirid(&new_root->root_item);
5539 location->type = BTRFS_INODE_ITEM_KEY;
5540 location->offset = 0;
5543 btrfs_free_path(path);
5547 static void inode_tree_add(struct inode *inode)
5549 struct btrfs_root *root = BTRFS_I(inode)->root;
5550 struct btrfs_inode *entry;
5552 struct rb_node *parent;
5553 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5554 u64 ino = btrfs_ino(BTRFS_I(inode));
5556 if (inode_unhashed(inode))
5559 spin_lock(&root->inode_lock);
5560 p = &root->inode_tree.rb_node;
5563 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5565 if (ino < btrfs_ino(entry))
5566 p = &parent->rb_left;
5567 else if (ino > btrfs_ino(entry))
5568 p = &parent->rb_right;
5570 WARN_ON(!(entry->vfs_inode.i_state &
5571 (I_WILL_FREE | I_FREEING)));
5572 rb_replace_node(parent, new, &root->inode_tree);
5573 RB_CLEAR_NODE(parent);
5574 spin_unlock(&root->inode_lock);
5578 rb_link_node(new, parent, p);
5579 rb_insert_color(new, &root->inode_tree);
5580 spin_unlock(&root->inode_lock);
5583 static void inode_tree_del(struct inode *inode)
5585 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5586 struct btrfs_root *root = BTRFS_I(inode)->root;
5589 spin_lock(&root->inode_lock);
5590 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5591 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5592 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5593 empty = RB_EMPTY_ROOT(&root->inode_tree);
5595 spin_unlock(&root->inode_lock);
5597 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5598 synchronize_srcu(&fs_info->subvol_srcu);
5599 spin_lock(&root->inode_lock);
5600 empty = RB_EMPTY_ROOT(&root->inode_tree);
5601 spin_unlock(&root->inode_lock);
5603 btrfs_add_dead_root(root);
5608 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5610 struct btrfs_iget_args *args = p;
5611 inode->i_ino = args->location->objectid;
5612 memcpy(&BTRFS_I(inode)->location, args->location,
5613 sizeof(*args->location));
5614 BTRFS_I(inode)->root = args->root;
5618 static int btrfs_find_actor(struct inode *inode, void *opaque)
5620 struct btrfs_iget_args *args = opaque;
5621 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5622 args->root == BTRFS_I(inode)->root;
5625 static struct inode *btrfs_iget_locked(struct super_block *s,
5626 struct btrfs_key *location,
5627 struct btrfs_root *root)
5629 struct inode *inode;
5630 struct btrfs_iget_args args;
5631 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5633 args.location = location;
5636 inode = iget5_locked(s, hashval, btrfs_find_actor,
5637 btrfs_init_locked_inode,
5642 /* Get an inode object given its location and corresponding root.
5643 * Returns in *is_new if the inode was read from disk
5645 struct inode *btrfs_iget_path(struct super_block *s, struct btrfs_key *location,
5646 struct btrfs_root *root, int *new,
5647 struct btrfs_path *path)
5649 struct inode *inode;
5651 inode = btrfs_iget_locked(s, location, root);
5653 return ERR_PTR(-ENOMEM);
5655 if (inode->i_state & I_NEW) {
5658 ret = btrfs_read_locked_inode(inode, path);
5660 inode_tree_add(inode);
5661 unlock_new_inode(inode);
5667 * ret > 0 can come from btrfs_search_slot called by
5668 * btrfs_read_locked_inode, this means the inode item
5673 inode = ERR_PTR(ret);
5680 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5681 struct btrfs_root *root, int *new)
5683 return btrfs_iget_path(s, location, root, new, NULL);
5686 static struct inode *new_simple_dir(struct super_block *s,
5687 struct btrfs_key *key,
5688 struct btrfs_root *root)
5690 struct inode *inode = new_inode(s);
5693 return ERR_PTR(-ENOMEM);
5695 BTRFS_I(inode)->root = root;
5696 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5697 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5699 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5700 inode->i_op = &btrfs_dir_ro_inode_operations;
5701 inode->i_opflags &= ~IOP_XATTR;
5702 inode->i_fop = &simple_dir_operations;
5703 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5704 inode->i_mtime = current_time(inode);
5705 inode->i_atime = inode->i_mtime;
5706 inode->i_ctime = inode->i_mtime;
5707 BTRFS_I(inode)->i_otime = inode->i_mtime;
5712 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5714 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5715 struct inode *inode;
5716 struct btrfs_root *root = BTRFS_I(dir)->root;
5717 struct btrfs_root *sub_root = root;
5718 struct btrfs_key location;
5722 if (dentry->d_name.len > BTRFS_NAME_LEN)
5723 return ERR_PTR(-ENAMETOOLONG);
5725 ret = btrfs_inode_by_name(dir, dentry, &location);
5727 return ERR_PTR(ret);
5729 if (location.type == BTRFS_INODE_ITEM_KEY) {
5730 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5734 index = srcu_read_lock(&fs_info->subvol_srcu);
5735 ret = fixup_tree_root_location(fs_info, dir, dentry,
5736 &location, &sub_root);
5739 inode = ERR_PTR(ret);
5741 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5743 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5745 srcu_read_unlock(&fs_info->subvol_srcu, index);
5747 if (!IS_ERR(inode) && root != sub_root) {
5748 down_read(&fs_info->cleanup_work_sem);
5749 if (!sb_rdonly(inode->i_sb))
5750 ret = btrfs_orphan_cleanup(sub_root);
5751 up_read(&fs_info->cleanup_work_sem);
5754 inode = ERR_PTR(ret);
5761 static int btrfs_dentry_delete(const struct dentry *dentry)
5763 struct btrfs_root *root;
5764 struct inode *inode = d_inode(dentry);
5766 if (!inode && !IS_ROOT(dentry))
5767 inode = d_inode(dentry->d_parent);
5770 root = BTRFS_I(inode)->root;
5771 if (btrfs_root_refs(&root->root_item) == 0)
5774 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5780 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5783 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5785 if (inode == ERR_PTR(-ENOENT))
5787 return d_splice_alias(inode, dentry);
5790 unsigned char btrfs_filetype_table[] = {
5791 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5795 * All this infrastructure exists because dir_emit can fault, and we are holding
5796 * the tree lock when doing readdir. For now just allocate a buffer and copy
5797 * our information into that, and then dir_emit from the buffer. This is
5798 * similar to what NFS does, only we don't keep the buffer around in pagecache
5799 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5800 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5803 static int btrfs_opendir(struct inode *inode, struct file *file)
5805 struct btrfs_file_private *private;
5807 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5810 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5811 if (!private->filldir_buf) {
5815 file->private_data = private;
5826 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5829 struct dir_entry *entry = addr;
5830 char *name = (char *)(entry + 1);
5832 ctx->pos = get_unaligned(&entry->offset);
5833 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5834 get_unaligned(&entry->ino),
5835 get_unaligned(&entry->type)))
5837 addr += sizeof(struct dir_entry) +
5838 get_unaligned(&entry->name_len);
5844 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5846 struct inode *inode = file_inode(file);
5847 struct btrfs_root *root = BTRFS_I(inode)->root;
5848 struct btrfs_file_private *private = file->private_data;
5849 struct btrfs_dir_item *di;
5850 struct btrfs_key key;
5851 struct btrfs_key found_key;
5852 struct btrfs_path *path;
5854 struct list_head ins_list;
5855 struct list_head del_list;
5857 struct extent_buffer *leaf;
5864 struct btrfs_key location;
5866 if (!dir_emit_dots(file, ctx))
5869 path = btrfs_alloc_path();
5873 addr = private->filldir_buf;
5874 path->reada = READA_FORWARD;
5876 INIT_LIST_HEAD(&ins_list);
5877 INIT_LIST_HEAD(&del_list);
5878 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5881 key.type = BTRFS_DIR_INDEX_KEY;
5882 key.offset = ctx->pos;
5883 key.objectid = btrfs_ino(BTRFS_I(inode));
5885 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5890 struct dir_entry *entry;
5892 leaf = path->nodes[0];
5893 slot = path->slots[0];
5894 if (slot >= btrfs_header_nritems(leaf)) {
5895 ret = btrfs_next_leaf(root, path);
5903 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5905 if (found_key.objectid != key.objectid)
5907 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5909 if (found_key.offset < ctx->pos)
5911 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5913 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5914 name_len = btrfs_dir_name_len(leaf, di);
5915 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5917 btrfs_release_path(path);
5918 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5921 addr = private->filldir_buf;
5928 put_unaligned(name_len, &entry->name_len);
5929 name_ptr = (char *)(entry + 1);
5930 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5932 put_unaligned(btrfs_filetype_table[btrfs_dir_type(leaf, di)],
5934 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5935 put_unaligned(location.objectid, &entry->ino);
5936 put_unaligned(found_key.offset, &entry->offset);
5938 addr += sizeof(struct dir_entry) + name_len;
5939 total_len += sizeof(struct dir_entry) + name_len;
5943 btrfs_release_path(path);
5945 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5949 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5954 * Stop new entries from being returned after we return the last
5957 * New directory entries are assigned a strictly increasing
5958 * offset. This means that new entries created during readdir
5959 * are *guaranteed* to be seen in the future by that readdir.
5960 * This has broken buggy programs which operate on names as
5961 * they're returned by readdir. Until we re-use freed offsets
5962 * we have this hack to stop new entries from being returned
5963 * under the assumption that they'll never reach this huge
5966 * This is being careful not to overflow 32bit loff_t unless the
5967 * last entry requires it because doing so has broken 32bit apps
5970 if (ctx->pos >= INT_MAX)
5971 ctx->pos = LLONG_MAX;
5978 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5979 btrfs_free_path(path);
5984 * This is somewhat expensive, updating the tree every time the
5985 * inode changes. But, it is most likely to find the inode in cache.
5986 * FIXME, needs more benchmarking...there are no reasons other than performance
5987 * to keep or drop this code.
5989 static int btrfs_dirty_inode(struct inode *inode)
5991 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5992 struct btrfs_root *root = BTRFS_I(inode)->root;
5993 struct btrfs_trans_handle *trans;
5996 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5999 trans = btrfs_join_transaction(root);
6001 return PTR_ERR(trans);
6003 ret = btrfs_update_inode(trans, root, inode);
6004 if (ret && ret == -ENOSPC) {
6005 /* whoops, lets try again with the full transaction */
6006 btrfs_end_transaction(trans);
6007 trans = btrfs_start_transaction(root, 1);
6009 return PTR_ERR(trans);
6011 ret = btrfs_update_inode(trans, root, inode);
6013 btrfs_end_transaction(trans);
6014 if (BTRFS_I(inode)->delayed_node)
6015 btrfs_balance_delayed_items(fs_info);
6021 * This is a copy of file_update_time. We need this so we can return error on
6022 * ENOSPC for updating the inode in the case of file write and mmap writes.
6024 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
6027 struct btrfs_root *root = BTRFS_I(inode)->root;
6028 bool dirty = flags & ~S_VERSION;
6030 if (btrfs_root_readonly(root))
6033 if (flags & S_VERSION)
6034 dirty |= inode_maybe_inc_iversion(inode, dirty);
6035 if (flags & S_CTIME)
6036 inode->i_ctime = *now;
6037 if (flags & S_MTIME)
6038 inode->i_mtime = *now;
6039 if (flags & S_ATIME)
6040 inode->i_atime = *now;
6041 return dirty ? btrfs_dirty_inode(inode) : 0;
6045 * find the highest existing sequence number in a directory
6046 * and then set the in-memory index_cnt variable to reflect
6047 * free sequence numbers
6049 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6051 struct btrfs_root *root = inode->root;
6052 struct btrfs_key key, found_key;
6053 struct btrfs_path *path;
6054 struct extent_buffer *leaf;
6057 key.objectid = btrfs_ino(inode);
6058 key.type = BTRFS_DIR_INDEX_KEY;
6059 key.offset = (u64)-1;
6061 path = btrfs_alloc_path();
6065 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6068 /* FIXME: we should be able to handle this */
6074 * MAGIC NUMBER EXPLANATION:
6075 * since we search a directory based on f_pos we have to start at 2
6076 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6077 * else has to start at 2
6079 if (path->slots[0] == 0) {
6080 inode->index_cnt = 2;
6086 leaf = path->nodes[0];
6087 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6089 if (found_key.objectid != btrfs_ino(inode) ||
6090 found_key.type != BTRFS_DIR_INDEX_KEY) {
6091 inode->index_cnt = 2;
6095 inode->index_cnt = found_key.offset + 1;
6097 btrfs_free_path(path);
6102 * helper to find a free sequence number in a given directory. This current
6103 * code is very simple, later versions will do smarter things in the btree
6105 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6109 if (dir->index_cnt == (u64)-1) {
6110 ret = btrfs_inode_delayed_dir_index_count(dir);
6112 ret = btrfs_set_inode_index_count(dir);
6118 *index = dir->index_cnt;
6124 static int btrfs_insert_inode_locked(struct inode *inode)
6126 struct btrfs_iget_args args;
6127 args.location = &BTRFS_I(inode)->location;
6128 args.root = BTRFS_I(inode)->root;
6130 return insert_inode_locked4(inode,
6131 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6132 btrfs_find_actor, &args);
6136 * Inherit flags from the parent inode.
6138 * Currently only the compression flags and the cow flags are inherited.
6140 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6147 flags = BTRFS_I(dir)->flags;
6149 if (flags & BTRFS_INODE_NOCOMPRESS) {
6150 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6151 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6152 } else if (flags & BTRFS_INODE_COMPRESS) {
6153 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6154 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6157 if (flags & BTRFS_INODE_NODATACOW) {
6158 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6159 if (S_ISREG(inode->i_mode))
6160 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6163 btrfs_sync_inode_flags_to_i_flags(inode);
6166 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6167 struct btrfs_root *root,
6169 const char *name, int name_len,
6170 u64 ref_objectid, u64 objectid,
6171 umode_t mode, u64 *index)
6173 struct btrfs_fs_info *fs_info = root->fs_info;
6174 struct inode *inode;
6175 struct btrfs_inode_item *inode_item;
6176 struct btrfs_key *location;
6177 struct btrfs_path *path;
6178 struct btrfs_inode_ref *ref;
6179 struct btrfs_key key[2];
6181 int nitems = name ? 2 : 1;
6185 path = btrfs_alloc_path();
6187 return ERR_PTR(-ENOMEM);
6189 inode = new_inode(fs_info->sb);
6191 btrfs_free_path(path);
6192 return ERR_PTR(-ENOMEM);
6196 * O_TMPFILE, set link count to 0, so that after this point,
6197 * we fill in an inode item with the correct link count.
6200 set_nlink(inode, 0);
6203 * we have to initialize this early, so we can reclaim the inode
6204 * number if we fail afterwards in this function.
6206 inode->i_ino = objectid;
6209 trace_btrfs_inode_request(dir);
6211 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6213 btrfs_free_path(path);
6215 return ERR_PTR(ret);
6221 * index_cnt is ignored for everything but a dir,
6222 * btrfs_set_inode_index_count has an explanation for the magic
6225 BTRFS_I(inode)->index_cnt = 2;
6226 BTRFS_I(inode)->dir_index = *index;
6227 BTRFS_I(inode)->root = root;
6228 BTRFS_I(inode)->generation = trans->transid;
6229 inode->i_generation = BTRFS_I(inode)->generation;
6232 * We could have gotten an inode number from somebody who was fsynced
6233 * and then removed in this same transaction, so let's just set full
6234 * sync since it will be a full sync anyway and this will blow away the
6235 * old info in the log.
6237 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6239 key[0].objectid = objectid;
6240 key[0].type = BTRFS_INODE_ITEM_KEY;
6243 sizes[0] = sizeof(struct btrfs_inode_item);
6247 * Start new inodes with an inode_ref. This is slightly more
6248 * efficient for small numbers of hard links since they will
6249 * be packed into one item. Extended refs will kick in if we
6250 * add more hard links than can fit in the ref item.
6252 key[1].objectid = objectid;
6253 key[1].type = BTRFS_INODE_REF_KEY;
6254 key[1].offset = ref_objectid;
6256 sizes[1] = name_len + sizeof(*ref);
6259 location = &BTRFS_I(inode)->location;
6260 location->objectid = objectid;
6261 location->offset = 0;
6262 location->type = BTRFS_INODE_ITEM_KEY;
6264 ret = btrfs_insert_inode_locked(inode);
6270 path->leave_spinning = 1;
6271 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6275 inode_init_owner(inode, dir, mode);
6276 inode_set_bytes(inode, 0);
6278 inode->i_mtime = current_time(inode);
6279 inode->i_atime = inode->i_mtime;
6280 inode->i_ctime = inode->i_mtime;
6281 BTRFS_I(inode)->i_otime = inode->i_mtime;
6283 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6284 struct btrfs_inode_item);
6285 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6286 sizeof(*inode_item));
6287 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6290 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6291 struct btrfs_inode_ref);
6292 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6293 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6294 ptr = (unsigned long)(ref + 1);
6295 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6298 btrfs_mark_buffer_dirty(path->nodes[0]);
6299 btrfs_free_path(path);
6301 btrfs_inherit_iflags(inode, dir);
6303 if (S_ISREG(mode)) {
6304 if (btrfs_test_opt(fs_info, NODATASUM))
6305 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6306 if (btrfs_test_opt(fs_info, NODATACOW))
6307 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6308 BTRFS_INODE_NODATASUM;
6311 inode_tree_add(inode);
6313 trace_btrfs_inode_new(inode);
6314 btrfs_set_inode_last_trans(trans, inode);
6316 btrfs_update_root_times(trans, root);
6318 ret = btrfs_inode_inherit_props(trans, inode, dir);
6321 "error inheriting props for ino %llu (root %llu): %d",
6322 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6327 discard_new_inode(inode);
6330 BTRFS_I(dir)->index_cnt--;
6331 btrfs_free_path(path);
6332 return ERR_PTR(ret);
6335 static inline u8 btrfs_inode_type(struct inode *inode)
6337 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6341 * utility function to add 'inode' into 'parent_inode' with
6342 * a give name and a given sequence number.
6343 * if 'add_backref' is true, also insert a backref from the
6344 * inode to the parent directory.
6346 int btrfs_add_link(struct btrfs_trans_handle *trans,
6347 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6348 const char *name, int name_len, int add_backref, u64 index)
6351 struct btrfs_key key;
6352 struct btrfs_root *root = parent_inode->root;
6353 u64 ino = btrfs_ino(inode);
6354 u64 parent_ino = btrfs_ino(parent_inode);
6356 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6357 memcpy(&key, &inode->root->root_key, sizeof(key));
6360 key.type = BTRFS_INODE_ITEM_KEY;
6364 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6365 ret = btrfs_add_root_ref(trans, key.objectid,
6366 root->root_key.objectid, parent_ino,
6367 index, name, name_len);
6368 } else if (add_backref) {
6369 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6373 /* Nothing to clean up yet */
6377 ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key,
6378 btrfs_inode_type(&inode->vfs_inode), index);
6379 if (ret == -EEXIST || ret == -EOVERFLOW)
6382 btrfs_abort_transaction(trans, ret);
6386 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6388 inode_inc_iversion(&parent_inode->vfs_inode);
6389 parent_inode->vfs_inode.i_mtime = parent_inode->vfs_inode.i_ctime =
6390 current_time(&parent_inode->vfs_inode);
6391 ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode);
6393 btrfs_abort_transaction(trans, ret);
6397 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6400 err = btrfs_del_root_ref(trans, key.objectid,
6401 root->root_key.objectid, parent_ino,
6402 &local_index, name, name_len);
6404 } else if (add_backref) {
6408 err = btrfs_del_inode_ref(trans, root, name, name_len,
6409 ino, parent_ino, &local_index);
6414 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6415 struct btrfs_inode *dir, struct dentry *dentry,
6416 struct btrfs_inode *inode, int backref, u64 index)
6418 int err = btrfs_add_link(trans, dir, inode,
6419 dentry->d_name.name, dentry->d_name.len,
6426 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6427 umode_t mode, dev_t rdev)
6429 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6430 struct btrfs_trans_handle *trans;
6431 struct btrfs_root *root = BTRFS_I(dir)->root;
6432 struct inode *inode = NULL;
6438 * 2 for inode item and ref
6440 * 1 for xattr if selinux is on
6442 trans = btrfs_start_transaction(root, 5);
6444 return PTR_ERR(trans);
6446 err = btrfs_find_free_ino(root, &objectid);
6450 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6451 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6453 if (IS_ERR(inode)) {
6454 err = PTR_ERR(inode);
6460 * If the active LSM wants to access the inode during
6461 * d_instantiate it needs these. Smack checks to see
6462 * if the filesystem supports xattrs by looking at the
6465 inode->i_op = &btrfs_special_inode_operations;
6466 init_special_inode(inode, inode->i_mode, rdev);
6468 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6472 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6477 btrfs_update_inode(trans, root, inode);
6478 d_instantiate_new(dentry, inode);
6481 btrfs_end_transaction(trans);
6482 btrfs_btree_balance_dirty(fs_info);
6484 inode_dec_link_count(inode);
6485 discard_new_inode(inode);
6490 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6491 umode_t mode, bool excl)
6493 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6494 struct btrfs_trans_handle *trans;
6495 struct btrfs_root *root = BTRFS_I(dir)->root;
6496 struct inode *inode = NULL;
6502 * 2 for inode item and ref
6504 * 1 for xattr if selinux is on
6506 trans = btrfs_start_transaction(root, 5);
6508 return PTR_ERR(trans);
6510 err = btrfs_find_free_ino(root, &objectid);
6514 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6515 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6517 if (IS_ERR(inode)) {
6518 err = PTR_ERR(inode);
6523 * If the active LSM wants to access the inode during
6524 * d_instantiate it needs these. Smack checks to see
6525 * if the filesystem supports xattrs by looking at the
6528 inode->i_fop = &btrfs_file_operations;
6529 inode->i_op = &btrfs_file_inode_operations;
6530 inode->i_mapping->a_ops = &btrfs_aops;
6532 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6536 err = btrfs_update_inode(trans, root, inode);
6540 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6545 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6546 d_instantiate_new(dentry, inode);
6549 btrfs_end_transaction(trans);
6551 inode_dec_link_count(inode);
6552 discard_new_inode(inode);
6554 btrfs_btree_balance_dirty(fs_info);
6558 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6559 struct dentry *dentry)
6561 struct btrfs_trans_handle *trans = NULL;
6562 struct btrfs_root *root = BTRFS_I(dir)->root;
6563 struct inode *inode = d_inode(old_dentry);
6564 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6569 /* do not allow sys_link's with other subvols of the same device */
6570 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6573 if (inode->i_nlink >= BTRFS_LINK_MAX)
6576 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6581 * 2 items for inode and inode ref
6582 * 2 items for dir items
6583 * 1 item for parent inode
6584 * 1 item for orphan item deletion if O_TMPFILE
6586 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6587 if (IS_ERR(trans)) {
6588 err = PTR_ERR(trans);
6593 /* There are several dir indexes for this inode, clear the cache. */
6594 BTRFS_I(inode)->dir_index = 0ULL;
6596 inode_inc_iversion(inode);
6597 inode->i_ctime = current_time(inode);
6599 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6601 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6607 struct dentry *parent = dentry->d_parent;
6610 err = btrfs_update_inode(trans, root, inode);
6613 if (inode->i_nlink == 1) {
6615 * If new hard link count is 1, it's a file created
6616 * with open(2) O_TMPFILE flag.
6618 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6622 d_instantiate(dentry, inode);
6623 ret = btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent,
6625 if (ret == BTRFS_NEED_TRANS_COMMIT) {
6626 err = btrfs_commit_transaction(trans);
6633 btrfs_end_transaction(trans);
6635 inode_dec_link_count(inode);
6638 btrfs_btree_balance_dirty(fs_info);
6642 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6644 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6645 struct inode *inode = NULL;
6646 struct btrfs_trans_handle *trans;
6647 struct btrfs_root *root = BTRFS_I(dir)->root;
6649 int drop_on_err = 0;
6654 * 2 items for inode and ref
6655 * 2 items for dir items
6656 * 1 for xattr if selinux is on
6658 trans = btrfs_start_transaction(root, 5);
6660 return PTR_ERR(trans);
6662 err = btrfs_find_free_ino(root, &objectid);
6666 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6667 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6668 S_IFDIR | mode, &index);
6669 if (IS_ERR(inode)) {
6670 err = PTR_ERR(inode);
6676 /* these must be set before we unlock the inode */
6677 inode->i_op = &btrfs_dir_inode_operations;
6678 inode->i_fop = &btrfs_dir_file_operations;
6680 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6684 btrfs_i_size_write(BTRFS_I(inode), 0);
6685 err = btrfs_update_inode(trans, root, inode);
6689 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6690 dentry->d_name.name,
6691 dentry->d_name.len, 0, index);
6695 d_instantiate_new(dentry, inode);
6699 btrfs_end_transaction(trans);
6701 inode_dec_link_count(inode);
6702 discard_new_inode(inode);
6704 btrfs_btree_balance_dirty(fs_info);
6708 static noinline int uncompress_inline(struct btrfs_path *path,
6710 size_t pg_offset, u64 extent_offset,
6711 struct btrfs_file_extent_item *item)
6714 struct extent_buffer *leaf = path->nodes[0];
6717 unsigned long inline_size;
6721 WARN_ON(pg_offset != 0);
6722 compress_type = btrfs_file_extent_compression(leaf, item);
6723 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6724 inline_size = btrfs_file_extent_inline_item_len(leaf,
6725 btrfs_item_nr(path->slots[0]));
6726 tmp = kmalloc(inline_size, GFP_NOFS);
6729 ptr = btrfs_file_extent_inline_start(item);
6731 read_extent_buffer(leaf, tmp, ptr, inline_size);
6733 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6734 ret = btrfs_decompress(compress_type, tmp, page,
6735 extent_offset, inline_size, max_size);
6738 * decompression code contains a memset to fill in any space between the end
6739 * of the uncompressed data and the end of max_size in case the decompressed
6740 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6741 * the end of an inline extent and the beginning of the next block, so we
6742 * cover that region here.
6745 if (max_size + pg_offset < PAGE_SIZE) {
6746 char *map = kmap(page);
6747 memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6755 * a bit scary, this does extent mapping from logical file offset to the disk.
6756 * the ugly parts come from merging extents from the disk with the in-ram
6757 * representation. This gets more complex because of the data=ordered code,
6758 * where the in-ram extents might be locked pending data=ordered completion.
6760 * This also copies inline extents directly into the page.
6762 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6764 size_t pg_offset, u64 start, u64 len,
6767 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6770 u64 extent_start = 0;
6772 u64 objectid = btrfs_ino(inode);
6774 struct btrfs_path *path = NULL;
6775 struct btrfs_root *root = inode->root;
6776 struct btrfs_file_extent_item *item;
6777 struct extent_buffer *leaf;
6778 struct btrfs_key found_key;
6779 struct extent_map *em = NULL;
6780 struct extent_map_tree *em_tree = &inode->extent_tree;
6781 struct extent_io_tree *io_tree = &inode->io_tree;
6782 const bool new_inline = !page || create;
6784 read_lock(&em_tree->lock);
6785 em = lookup_extent_mapping(em_tree, start, len);
6787 em->bdev = fs_info->fs_devices->latest_bdev;
6788 read_unlock(&em_tree->lock);
6791 if (em->start > start || em->start + em->len <= start)
6792 free_extent_map(em);
6793 else if (em->block_start == EXTENT_MAP_INLINE && page)
6794 free_extent_map(em);
6798 em = alloc_extent_map();
6803 em->bdev = fs_info->fs_devices->latest_bdev;
6804 em->start = EXTENT_MAP_HOLE;
6805 em->orig_start = EXTENT_MAP_HOLE;
6807 em->block_len = (u64)-1;
6809 path = btrfs_alloc_path();
6815 /* Chances are we'll be called again, so go ahead and do readahead */
6816 path->reada = READA_FORWARD;
6819 * Unless we're going to uncompress the inline extent, no sleep would
6822 path->leave_spinning = 1;
6824 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6831 if (path->slots[0] == 0)
6836 leaf = path->nodes[0];
6837 item = btrfs_item_ptr(leaf, path->slots[0],
6838 struct btrfs_file_extent_item);
6839 /* are we inside the extent that was found? */
6840 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6841 found_type = found_key.type;
6842 if (found_key.objectid != objectid ||
6843 found_type != BTRFS_EXTENT_DATA_KEY) {
6845 * If we backup past the first extent we want to move forward
6846 * and see if there is an extent in front of us, otherwise we'll
6847 * say there is a hole for our whole search range which can
6854 found_type = btrfs_file_extent_type(leaf, item);
6855 extent_start = found_key.offset;
6856 if (found_type == BTRFS_FILE_EXTENT_REG ||
6857 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6858 extent_end = extent_start +
6859 btrfs_file_extent_num_bytes(leaf, item);
6861 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6863 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6866 size = btrfs_file_extent_ram_bytes(leaf, item);
6867 extent_end = ALIGN(extent_start + size,
6868 fs_info->sectorsize);
6870 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6875 if (start >= extent_end) {
6877 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6878 ret = btrfs_next_leaf(root, path);
6885 leaf = path->nodes[0];
6887 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6888 if (found_key.objectid != objectid ||
6889 found_key.type != BTRFS_EXTENT_DATA_KEY)
6891 if (start + len <= found_key.offset)
6893 if (start > found_key.offset)
6896 em->orig_start = start;
6897 em->len = found_key.offset - start;
6901 btrfs_extent_item_to_extent_map(inode, path, item,
6904 if (found_type == BTRFS_FILE_EXTENT_REG ||
6905 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6907 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6911 size_t extent_offset;
6917 size = btrfs_file_extent_ram_bytes(leaf, item);
6918 extent_offset = page_offset(page) + pg_offset - extent_start;
6919 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6920 size - extent_offset);
6921 em->start = extent_start + extent_offset;
6922 em->len = ALIGN(copy_size, fs_info->sectorsize);
6923 em->orig_block_len = em->len;
6924 em->orig_start = em->start;
6925 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6927 btrfs_set_path_blocking(path);
6928 if (!PageUptodate(page)) {
6929 if (btrfs_file_extent_compression(leaf, item) !=
6930 BTRFS_COMPRESS_NONE) {
6931 ret = uncompress_inline(path, page, pg_offset,
6932 extent_offset, item);
6939 read_extent_buffer(leaf, map + pg_offset, ptr,
6941 if (pg_offset + copy_size < PAGE_SIZE) {
6942 memset(map + pg_offset + copy_size, 0,
6943 PAGE_SIZE - pg_offset -
6948 flush_dcache_page(page);
6950 set_extent_uptodate(io_tree, em->start,
6951 extent_map_end(em) - 1, NULL, GFP_NOFS);
6956 em->orig_start = start;
6959 em->block_start = EXTENT_MAP_HOLE;
6961 btrfs_release_path(path);
6962 if (em->start > start || extent_map_end(em) <= start) {
6964 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6965 em->start, em->len, start, len);
6971 write_lock(&em_tree->lock);
6972 err = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6973 write_unlock(&em_tree->lock);
6975 btrfs_free_path(path);
6977 trace_btrfs_get_extent(root, inode, em);
6980 free_extent_map(em);
6981 return ERR_PTR(err);
6983 BUG_ON(!em); /* Error is always set */
6987 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
6989 size_t pg_offset, u64 start, u64 len,
6992 struct extent_map *em;
6993 struct extent_map *hole_em = NULL;
6994 u64 range_start = start;
7000 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7004 * If our em maps to:
7006 * - a pre-alloc extent,
7007 * there might actually be delalloc bytes behind it.
7009 if (em->block_start != EXTENT_MAP_HOLE &&
7010 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7015 /* check to see if we've wrapped (len == -1 or similar) */
7024 /* ok, we didn't find anything, lets look for delalloc */
7025 found = count_range_bits(&inode->io_tree, &range_start,
7026 end, len, EXTENT_DELALLOC, 1);
7027 found_end = range_start + found;
7028 if (found_end < range_start)
7029 found_end = (u64)-1;
7032 * we didn't find anything useful, return
7033 * the original results from get_extent()
7035 if (range_start > end || found_end <= start) {
7041 /* adjust the range_start to make sure it doesn't
7042 * go backwards from the start they passed in
7044 range_start = max(start, range_start);
7045 found = found_end - range_start;
7048 u64 hole_start = start;
7051 em = alloc_extent_map();
7057 * when btrfs_get_extent can't find anything it
7058 * returns one huge hole
7060 * make sure what it found really fits our range, and
7061 * adjust to make sure it is based on the start from
7065 u64 calc_end = extent_map_end(hole_em);
7067 if (calc_end <= start || (hole_em->start > end)) {
7068 free_extent_map(hole_em);
7071 hole_start = max(hole_em->start, start);
7072 hole_len = calc_end - hole_start;
7076 if (hole_em && range_start > hole_start) {
7077 /* our hole starts before our delalloc, so we
7078 * have to return just the parts of the hole
7079 * that go until the delalloc starts
7081 em->len = min(hole_len,
7082 range_start - hole_start);
7083 em->start = hole_start;
7084 em->orig_start = hole_start;
7086 * don't adjust block start at all,
7087 * it is fixed at EXTENT_MAP_HOLE
7089 em->block_start = hole_em->block_start;
7090 em->block_len = hole_len;
7091 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7092 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7094 em->start = range_start;
7096 em->orig_start = range_start;
7097 em->block_start = EXTENT_MAP_DELALLOC;
7098 em->block_len = found;
7105 free_extent_map(hole_em);
7107 free_extent_map(em);
7108 return ERR_PTR(err);
7113 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7116 const u64 orig_start,
7117 const u64 block_start,
7118 const u64 block_len,
7119 const u64 orig_block_len,
7120 const u64 ram_bytes,
7123 struct extent_map *em = NULL;
7126 if (type != BTRFS_ORDERED_NOCOW) {
7127 em = create_io_em(inode, start, len, orig_start,
7128 block_start, block_len, orig_block_len,
7130 BTRFS_COMPRESS_NONE, /* compress_type */
7135 ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7136 len, block_len, type);
7139 free_extent_map(em);
7140 btrfs_drop_extent_cache(BTRFS_I(inode), start,
7141 start + len - 1, 0);
7150 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7153 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7154 struct btrfs_root *root = BTRFS_I(inode)->root;
7155 struct extent_map *em;
7156 struct btrfs_key ins;
7160 alloc_hint = get_extent_allocation_hint(inode, start, len);
7161 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7162 0, alloc_hint, &ins, 1, 1);
7164 return ERR_PTR(ret);
7166 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7167 ins.objectid, ins.offset, ins.offset,
7168 ins.offset, BTRFS_ORDERED_REGULAR);
7169 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7171 btrfs_free_reserved_extent(fs_info, ins.objectid,
7178 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7179 * block must be cow'd
7181 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7182 u64 *orig_start, u64 *orig_block_len,
7185 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7186 struct btrfs_path *path;
7188 struct extent_buffer *leaf;
7189 struct btrfs_root *root = BTRFS_I(inode)->root;
7190 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7191 struct btrfs_file_extent_item *fi;
7192 struct btrfs_key key;
7199 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7201 path = btrfs_alloc_path();
7205 ret = btrfs_lookup_file_extent(NULL, root, path,
7206 btrfs_ino(BTRFS_I(inode)), offset, 0);
7210 slot = path->slots[0];
7213 /* can't find the item, must cow */
7220 leaf = path->nodes[0];
7221 btrfs_item_key_to_cpu(leaf, &key, slot);
7222 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7223 key.type != BTRFS_EXTENT_DATA_KEY) {
7224 /* not our file or wrong item type, must cow */
7228 if (key.offset > offset) {
7229 /* Wrong offset, must cow */
7233 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7234 found_type = btrfs_file_extent_type(leaf, fi);
7235 if (found_type != BTRFS_FILE_EXTENT_REG &&
7236 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7237 /* not a regular extent, must cow */
7241 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7244 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7245 if (extent_end <= offset)
7248 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7249 if (disk_bytenr == 0)
7252 if (btrfs_file_extent_compression(leaf, fi) ||
7253 btrfs_file_extent_encryption(leaf, fi) ||
7254 btrfs_file_extent_other_encoding(leaf, fi))
7258 * Do the same check as in btrfs_cross_ref_exist but without the
7259 * unnecessary search.
7261 if (btrfs_file_extent_generation(leaf, fi) <=
7262 btrfs_root_last_snapshot(&root->root_item))
7265 backref_offset = btrfs_file_extent_offset(leaf, fi);
7268 *orig_start = key.offset - backref_offset;
7269 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7270 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7273 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7276 num_bytes = min(offset + *len, extent_end) - offset;
7277 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7280 range_end = round_up(offset + num_bytes,
7281 root->fs_info->sectorsize) - 1;
7282 ret = test_range_bit(io_tree, offset, range_end,
7283 EXTENT_DELALLOC, 0, NULL);
7290 btrfs_release_path(path);
7293 * look for other files referencing this extent, if we
7294 * find any we must cow
7297 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7298 key.offset - backref_offset, disk_bytenr);
7305 * adjust disk_bytenr and num_bytes to cover just the bytes
7306 * in this extent we are about to write. If there
7307 * are any csums in that range we have to cow in order
7308 * to keep the csums correct
7310 disk_bytenr += backref_offset;
7311 disk_bytenr += offset - key.offset;
7312 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7315 * all of the above have passed, it is safe to overwrite this extent
7321 btrfs_free_path(path);
7325 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7326 struct extent_state **cached_state, int writing)
7328 struct btrfs_ordered_extent *ordered;
7332 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7335 * We're concerned with the entire range that we're going to be
7336 * doing DIO to, so we need to make sure there's no ordered
7337 * extents in this range.
7339 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7340 lockend - lockstart + 1);
7343 * We need to make sure there are no buffered pages in this
7344 * range either, we could have raced between the invalidate in
7345 * generic_file_direct_write and locking the extent. The
7346 * invalidate needs to happen so that reads after a write do not
7350 (!writing || !filemap_range_has_page(inode->i_mapping,
7351 lockstart, lockend)))
7354 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7359 * If we are doing a DIO read and the ordered extent we
7360 * found is for a buffered write, we can not wait for it
7361 * to complete and retry, because if we do so we can
7362 * deadlock with concurrent buffered writes on page
7363 * locks. This happens only if our DIO read covers more
7364 * than one extent map, if at this point has already
7365 * created an ordered extent for a previous extent map
7366 * and locked its range in the inode's io tree, and a
7367 * concurrent write against that previous extent map's
7368 * range and this range started (we unlock the ranges
7369 * in the io tree only when the bios complete and
7370 * buffered writes always lock pages before attempting
7371 * to lock range in the io tree).
7374 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7375 btrfs_start_ordered_extent(inode, ordered, 1);
7378 btrfs_put_ordered_extent(ordered);
7381 * We could trigger writeback for this range (and wait
7382 * for it to complete) and then invalidate the pages for
7383 * this range (through invalidate_inode_pages2_range()),
7384 * but that can lead us to a deadlock with a concurrent
7385 * call to readpages() (a buffered read or a defrag call
7386 * triggered a readahead) on a page lock due to an
7387 * ordered dio extent we created before but did not have
7388 * yet a corresponding bio submitted (whence it can not
7389 * complete), which makes readpages() wait for that
7390 * ordered extent to complete while holding a lock on
7405 /* The callers of this must take lock_extent() */
7406 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
7407 u64 orig_start, u64 block_start,
7408 u64 block_len, u64 orig_block_len,
7409 u64 ram_bytes, int compress_type,
7412 struct extent_map_tree *em_tree;
7413 struct extent_map *em;
7414 struct btrfs_root *root = BTRFS_I(inode)->root;
7417 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7418 type == BTRFS_ORDERED_COMPRESSED ||
7419 type == BTRFS_ORDERED_NOCOW ||
7420 type == BTRFS_ORDERED_REGULAR);
7422 em_tree = &BTRFS_I(inode)->extent_tree;
7423 em = alloc_extent_map();
7425 return ERR_PTR(-ENOMEM);
7428 em->orig_start = orig_start;
7430 em->block_len = block_len;
7431 em->block_start = block_start;
7432 em->bdev = root->fs_info->fs_devices->latest_bdev;
7433 em->orig_block_len = orig_block_len;
7434 em->ram_bytes = ram_bytes;
7435 em->generation = -1;
7436 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7437 if (type == BTRFS_ORDERED_PREALLOC) {
7438 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7439 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7440 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7441 em->compress_type = compress_type;
7445 btrfs_drop_extent_cache(BTRFS_I(inode), em->start,
7446 em->start + em->len - 1, 0);
7447 write_lock(&em_tree->lock);
7448 ret = add_extent_mapping(em_tree, em, 1);
7449 write_unlock(&em_tree->lock);
7451 * The caller has taken lock_extent(), who could race with us
7454 } while (ret == -EEXIST);
7457 free_extent_map(em);
7458 return ERR_PTR(ret);
7461 /* em got 2 refs now, callers needs to do free_extent_map once. */
7466 static int btrfs_get_blocks_direct_read(struct extent_map *em,
7467 struct buffer_head *bh_result,
7468 struct inode *inode,
7471 if (em->block_start == EXTENT_MAP_HOLE ||
7472 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7475 len = min(len, em->len - (start - em->start));
7477 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7479 bh_result->b_size = len;
7480 bh_result->b_bdev = em->bdev;
7481 set_buffer_mapped(bh_result);
7486 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7487 struct buffer_head *bh_result,
7488 struct inode *inode,
7489 struct btrfs_dio_data *dio_data,
7492 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7493 struct extent_map *em = *map;
7497 * We don't allocate a new extent in the following cases
7499 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7501 * 2) The extent is marked as PREALLOC. We're good to go here and can
7502 * just use the extent.
7505 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7506 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7507 em->block_start != EXTENT_MAP_HOLE)) {
7509 u64 block_start, orig_start, orig_block_len, ram_bytes;
7511 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7512 type = BTRFS_ORDERED_PREALLOC;
7514 type = BTRFS_ORDERED_NOCOW;
7515 len = min(len, em->len - (start - em->start));
7516 block_start = em->block_start + (start - em->start);
7518 if (can_nocow_extent(inode, start, &len, &orig_start,
7519 &orig_block_len, &ram_bytes) == 1 &&
7520 btrfs_inc_nocow_writers(fs_info, block_start)) {
7521 struct extent_map *em2;
7523 em2 = btrfs_create_dio_extent(inode, start, len,
7524 orig_start, block_start,
7525 len, orig_block_len,
7527 btrfs_dec_nocow_writers(fs_info, block_start);
7528 if (type == BTRFS_ORDERED_PREALLOC) {
7529 free_extent_map(em);
7533 if (em2 && IS_ERR(em2)) {
7538 * For inode marked NODATACOW or extent marked PREALLOC,
7539 * use the existing or preallocated extent, so does not
7540 * need to adjust btrfs_space_info's bytes_may_use.
7542 btrfs_free_reserved_data_space_noquota(inode, start,
7548 /* this will cow the extent */
7549 len = bh_result->b_size;
7550 free_extent_map(em);
7551 *map = em = btrfs_new_extent_direct(inode, start, len);
7557 len = min(len, em->len - (start - em->start));
7560 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7562 bh_result->b_size = len;
7563 bh_result->b_bdev = em->bdev;
7564 set_buffer_mapped(bh_result);
7566 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7567 set_buffer_new(bh_result);
7570 * Need to update the i_size under the extent lock so buffered
7571 * readers will get the updated i_size when we unlock.
7573 if (!dio_data->overwrite && start + len > i_size_read(inode))
7574 i_size_write(inode, start + len);
7576 WARN_ON(dio_data->reserve < len);
7577 dio_data->reserve -= len;
7578 dio_data->unsubmitted_oe_range_end = start + len;
7579 current->journal_info = dio_data;
7584 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7585 struct buffer_head *bh_result, int create)
7587 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7588 struct extent_map *em;
7589 struct extent_state *cached_state = NULL;
7590 struct btrfs_dio_data *dio_data = NULL;
7591 u64 start = iblock << inode->i_blkbits;
7592 u64 lockstart, lockend;
7593 u64 len = bh_result->b_size;
7594 int unlock_bits = EXTENT_LOCKED;
7598 unlock_bits |= EXTENT_DIRTY;
7600 len = min_t(u64, len, fs_info->sectorsize);
7603 lockend = start + len - 1;
7605 if (current->journal_info) {
7607 * Need to pull our outstanding extents and set journal_info to NULL so
7608 * that anything that needs to check if there's a transaction doesn't get
7611 dio_data = current->journal_info;
7612 current->journal_info = NULL;
7616 * If this errors out it's because we couldn't invalidate pagecache for
7617 * this range and we need to fallback to buffered.
7619 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7625 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
7632 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7633 * io. INLINE is special, and we could probably kludge it in here, but
7634 * it's still buffered so for safety lets just fall back to the generic
7637 * For COMPRESSED we _have_ to read the entire extent in so we can
7638 * decompress it, so there will be buffering required no matter what we
7639 * do, so go ahead and fallback to buffered.
7641 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7642 * to buffered IO. Don't blame me, this is the price we pay for using
7645 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7646 em->block_start == EXTENT_MAP_INLINE) {
7647 free_extent_map(em);
7653 ret = btrfs_get_blocks_direct_write(&em, bh_result, inode,
7654 dio_data, start, len);
7658 /* clear and unlock the entire range */
7659 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7660 unlock_bits, 1, 0, &cached_state);
7662 ret = btrfs_get_blocks_direct_read(em, bh_result, inode,
7664 /* Can be negative only if we read from a hole */
7667 free_extent_map(em);
7671 * We need to unlock only the end area that we aren't using.
7672 * The rest is going to be unlocked by the endio routine.
7674 lockstart = start + bh_result->b_size;
7675 if (lockstart < lockend) {
7676 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7677 lockend, unlock_bits, 1, 0,
7680 free_extent_state(cached_state);
7684 free_extent_map(em);
7689 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7690 unlock_bits, 1, 0, &cached_state);
7693 current->journal_info = dio_data;
7697 static inline blk_status_t submit_dio_repair_bio(struct inode *inode,
7701 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7704 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7706 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR);
7710 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
7715 static int btrfs_check_dio_repairable(struct inode *inode,
7716 struct bio *failed_bio,
7717 struct io_failure_record *failrec,
7720 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7723 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
7724 if (num_copies == 1) {
7726 * we only have a single copy of the data, so don't bother with
7727 * all the retry and error correction code that follows. no
7728 * matter what the error is, it is very likely to persist.
7730 btrfs_debug(fs_info,
7731 "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
7732 num_copies, failrec->this_mirror, failed_mirror);
7736 failrec->failed_mirror = failed_mirror;
7737 failrec->this_mirror++;
7738 if (failrec->this_mirror == failed_mirror)
7739 failrec->this_mirror++;
7741 if (failrec->this_mirror > num_copies) {
7742 btrfs_debug(fs_info,
7743 "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
7744 num_copies, failrec->this_mirror, failed_mirror);
7751 static blk_status_t dio_read_error(struct inode *inode, struct bio *failed_bio,
7752 struct page *page, unsigned int pgoff,
7753 u64 start, u64 end, int failed_mirror,
7754 bio_end_io_t *repair_endio, void *repair_arg)
7756 struct io_failure_record *failrec;
7757 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7758 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7761 unsigned int read_mode = 0;
7764 blk_status_t status;
7765 struct bio_vec bvec;
7767 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
7769 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7771 return errno_to_blk_status(ret);
7773 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7776 free_io_failure(failure_tree, io_tree, failrec);
7777 return BLK_STS_IOERR;
7780 segs = bio_segments(failed_bio);
7781 bio_get_first_bvec(failed_bio, &bvec);
7783 (bvec.bv_len > btrfs_inode_sectorsize(inode)))
7784 read_mode |= REQ_FAILFAST_DEV;
7786 isector = start - btrfs_io_bio(failed_bio)->logical;
7787 isector >>= inode->i_sb->s_blocksize_bits;
7788 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7789 pgoff, isector, repair_endio, repair_arg);
7790 bio->bi_opf = REQ_OP_READ | read_mode;
7792 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7793 "repair DIO read error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d",
7794 read_mode, failrec->this_mirror, failrec->in_validation);
7796 status = submit_dio_repair_bio(inode, bio, failrec->this_mirror);
7798 free_io_failure(failure_tree, io_tree, failrec);
7805 struct btrfs_retry_complete {
7806 struct completion done;
7807 struct inode *inode;
7812 static void btrfs_retry_endio_nocsum(struct bio *bio)
7814 struct btrfs_retry_complete *done = bio->bi_private;
7815 struct inode *inode = done->inode;
7816 struct bio_vec *bvec;
7817 struct extent_io_tree *io_tree, *failure_tree;
7823 ASSERT(bio->bi_vcnt == 1);
7824 io_tree = &BTRFS_I(inode)->io_tree;
7825 failure_tree = &BTRFS_I(inode)->io_failure_tree;
7826 ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(inode));
7829 ASSERT(!bio_flagged(bio, BIO_CLONED));
7830 bio_for_each_segment_all(bvec, bio, i)
7831 clean_io_failure(BTRFS_I(inode)->root->fs_info, failure_tree,
7832 io_tree, done->start, bvec->bv_page,
7833 btrfs_ino(BTRFS_I(inode)), 0);
7835 complete(&done->done);
7839 static blk_status_t __btrfs_correct_data_nocsum(struct inode *inode,
7840 struct btrfs_io_bio *io_bio)
7842 struct btrfs_fs_info *fs_info;
7843 struct bio_vec bvec;
7844 struct bvec_iter iter;
7845 struct btrfs_retry_complete done;
7851 blk_status_t err = BLK_STS_OK;
7853 fs_info = BTRFS_I(inode)->root->fs_info;
7854 sectorsize = fs_info->sectorsize;
7856 start = io_bio->logical;
7858 io_bio->bio.bi_iter = io_bio->iter;
7860 bio_for_each_segment(bvec, &io_bio->bio, iter) {
7861 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7862 pgoff = bvec.bv_offset;
7864 next_block_or_try_again:
7867 init_completion(&done.done);
7869 ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
7870 pgoff, start, start + sectorsize - 1,
7872 btrfs_retry_endio_nocsum, &done);
7878 wait_for_completion_io(&done.done);
7880 if (!done.uptodate) {
7881 /* We might have another mirror, so try again */
7882 goto next_block_or_try_again;
7886 start += sectorsize;
7890 pgoff += sectorsize;
7891 ASSERT(pgoff < PAGE_SIZE);
7892 goto next_block_or_try_again;
7899 static void btrfs_retry_endio(struct bio *bio)
7901 struct btrfs_retry_complete *done = bio->bi_private;
7902 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7903 struct extent_io_tree *io_tree, *failure_tree;
7904 struct inode *inode = done->inode;
7905 struct bio_vec *bvec;
7915 ASSERT(bio->bi_vcnt == 1);
7916 ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(done->inode));
7918 io_tree = &BTRFS_I(inode)->io_tree;
7919 failure_tree = &BTRFS_I(inode)->io_failure_tree;
7921 ASSERT(!bio_flagged(bio, BIO_CLONED));
7922 bio_for_each_segment_all(bvec, bio, i) {
7923 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
7924 bvec->bv_offset, done->start,
7927 clean_io_failure(BTRFS_I(inode)->root->fs_info,
7928 failure_tree, io_tree, done->start,
7930 btrfs_ino(BTRFS_I(inode)),
7936 done->uptodate = uptodate;
7938 complete(&done->done);
7942 static blk_status_t __btrfs_subio_endio_read(struct inode *inode,
7943 struct btrfs_io_bio *io_bio, blk_status_t err)
7945 struct btrfs_fs_info *fs_info;
7946 struct bio_vec bvec;
7947 struct bvec_iter iter;
7948 struct btrfs_retry_complete done;
7955 bool uptodate = (err == 0);
7957 blk_status_t status;
7959 fs_info = BTRFS_I(inode)->root->fs_info;
7960 sectorsize = fs_info->sectorsize;
7963 start = io_bio->logical;
7965 io_bio->bio.bi_iter = io_bio->iter;
7967 bio_for_each_segment(bvec, &io_bio->bio, iter) {
7968 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7970 pgoff = bvec.bv_offset;
7973 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
7974 ret = __readpage_endio_check(inode, io_bio, csum_pos,
7975 bvec.bv_page, pgoff, start, sectorsize);
7982 init_completion(&done.done);
7984 status = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
7985 pgoff, start, start + sectorsize - 1,
7986 io_bio->mirror_num, btrfs_retry_endio,
7993 wait_for_completion_io(&done.done);
7995 if (!done.uptodate) {
7996 /* We might have another mirror, so try again */
8000 offset += sectorsize;
8001 start += sectorsize;
8007 pgoff += sectorsize;
8008 ASSERT(pgoff < PAGE_SIZE);
8016 static blk_status_t btrfs_subio_endio_read(struct inode *inode,
8017 struct btrfs_io_bio *io_bio, blk_status_t err)
8019 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8023 return __btrfs_correct_data_nocsum(inode, io_bio);
8027 return __btrfs_subio_endio_read(inode, io_bio, err);
8031 static void btrfs_endio_direct_read(struct bio *bio)
8033 struct btrfs_dio_private *dip = bio->bi_private;
8034 struct inode *inode = dip->inode;
8035 struct bio *dio_bio;
8036 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8037 blk_status_t err = bio->bi_status;
8039 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8040 err = btrfs_subio_endio_read(inode, io_bio, err);
8042 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8043 dip->logical_offset + dip->bytes - 1);
8044 dio_bio = dip->dio_bio;
8048 dio_bio->bi_status = err;
8049 dio_end_io(dio_bio);
8052 io_bio->end_io(io_bio, blk_status_to_errno(err));
8056 static void __endio_write_update_ordered(struct inode *inode,
8057 const u64 offset, const u64 bytes,
8058 const bool uptodate)
8060 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8061 struct btrfs_ordered_extent *ordered = NULL;
8062 struct btrfs_workqueue *wq;
8063 btrfs_work_func_t func;
8064 u64 ordered_offset = offset;
8065 u64 ordered_bytes = bytes;
8068 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
8069 wq = fs_info->endio_freespace_worker;
8070 func = btrfs_freespace_write_helper;
8072 wq = fs_info->endio_write_workers;
8073 func = btrfs_endio_write_helper;
8076 while (ordered_offset < offset + bytes) {
8077 last_offset = ordered_offset;
8078 if (btrfs_dec_test_first_ordered_pending(inode, &ordered,
8082 btrfs_init_work(&ordered->work, func,
8085 btrfs_queue_work(wq, &ordered->work);
8088 * If btrfs_dec_test_ordered_pending does not find any ordered
8089 * extent in the range, we can exit.
8091 if (ordered_offset == last_offset)
8094 * Our bio might span multiple ordered extents. In this case
8095 * we keep goin until we have accounted the whole dio.
8097 if (ordered_offset < offset + bytes) {
8098 ordered_bytes = offset + bytes - ordered_offset;
8104 static void btrfs_endio_direct_write(struct bio *bio)
8106 struct btrfs_dio_private *dip = bio->bi_private;
8107 struct bio *dio_bio = dip->dio_bio;
8109 __endio_write_update_ordered(dip->inode, dip->logical_offset,
8110 dip->bytes, !bio->bi_status);
8114 dio_bio->bi_status = bio->bi_status;
8115 dio_end_io(dio_bio);
8119 static blk_status_t btrfs_submit_bio_start_direct_io(void *private_data,
8120 struct bio *bio, u64 offset)
8122 struct inode *inode = private_data;
8124 ret = btrfs_csum_one_bio(inode, bio, offset, 1);
8125 BUG_ON(ret); /* -ENOMEM */
8129 static void btrfs_end_dio_bio(struct bio *bio)
8131 struct btrfs_dio_private *dip = bio->bi_private;
8132 blk_status_t err = bio->bi_status;
8135 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8136 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8137 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8139 (unsigned long long)bio->bi_iter.bi_sector,
8140 bio->bi_iter.bi_size, err);
8142 if (dip->subio_endio)
8143 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8147 * We want to perceive the errors flag being set before
8148 * decrementing the reference count. We don't need a barrier
8149 * since atomic operations with a return value are fully
8150 * ordered as per atomic_t.txt
8155 /* if there are more bios still pending for this dio, just exit */
8156 if (!atomic_dec_and_test(&dip->pending_bios))
8160 bio_io_error(dip->orig_bio);
8162 dip->dio_bio->bi_status = BLK_STS_OK;
8163 bio_endio(dip->orig_bio);
8169 static inline blk_status_t btrfs_lookup_and_bind_dio_csum(struct inode *inode,
8170 struct btrfs_dio_private *dip,
8174 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8175 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8179 * We load all the csum data we need when we submit
8180 * the first bio to reduce the csum tree search and
8183 if (dip->logical_offset == file_offset) {
8184 ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio,
8190 if (bio == dip->orig_bio)
8193 file_offset -= dip->logical_offset;
8194 file_offset >>= inode->i_sb->s_blocksize_bits;
8195 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8200 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
8201 struct inode *inode, u64 file_offset, int async_submit)
8203 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8204 struct btrfs_dio_private *dip = bio->bi_private;
8205 bool write = bio_op(bio) == REQ_OP_WRITE;
8208 /* Check btrfs_submit_bio_hook() for rules about async submit. */
8210 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8213 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8218 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
8221 if (write && async_submit) {
8222 ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0,
8224 btrfs_submit_bio_start_direct_io);
8228 * If we aren't doing async submit, calculate the csum of the
8231 ret = btrfs_csum_one_bio(inode, bio, file_offset, 1);
8235 ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio,
8241 ret = btrfs_map_bio(fs_info, bio, 0, 0);
8246 static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip)
8248 struct inode *inode = dip->inode;
8249 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8251 struct bio *orig_bio = dip->orig_bio;
8252 u64 start_sector = orig_bio->bi_iter.bi_sector;
8253 u64 file_offset = dip->logical_offset;
8255 int async_submit = 0;
8257 int clone_offset = 0;
8260 blk_status_t status;
8262 map_length = orig_bio->bi_iter.bi_size;
8263 submit_len = map_length;
8264 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9,
8265 &map_length, NULL, 0);
8269 if (map_length >= submit_len) {
8271 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8275 /* async crcs make it difficult to collect full stripe writes. */
8276 if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8282 ASSERT(map_length <= INT_MAX);
8283 atomic_inc(&dip->pending_bios);
8285 clone_len = min_t(int, submit_len, map_length);
8288 * This will never fail as it's passing GPF_NOFS and
8289 * the allocation is backed by btrfs_bioset.
8291 bio = btrfs_bio_clone_partial(orig_bio, clone_offset,
8293 bio->bi_private = dip;
8294 bio->bi_end_io = btrfs_end_dio_bio;
8295 btrfs_io_bio(bio)->logical = file_offset;
8297 ASSERT(submit_len >= clone_len);
8298 submit_len -= clone_len;
8299 if (submit_len == 0)
8303 * Increase the count before we submit the bio so we know
8304 * the end IO handler won't happen before we increase the
8305 * count. Otherwise, the dip might get freed before we're
8306 * done setting it up.
8308 atomic_inc(&dip->pending_bios);
8310 status = btrfs_submit_dio_bio(bio, inode, file_offset,
8314 atomic_dec(&dip->pending_bios);
8318 clone_offset += clone_len;
8319 start_sector += clone_len >> 9;
8320 file_offset += clone_len;
8322 map_length = submit_len;
8323 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio),
8324 start_sector << 9, &map_length, NULL, 0);
8327 } while (submit_len > 0);
8330 status = btrfs_submit_dio_bio(bio, inode, file_offset, async_submit);
8338 * Before atomic variable goto zero, we must make sure dip->errors is
8339 * perceived to be set. This ordering is ensured by the fact that an
8340 * atomic operations with a return value are fully ordered as per
8343 if (atomic_dec_and_test(&dip->pending_bios))
8344 bio_io_error(dip->orig_bio);
8346 /* bio_end_io() will handle error, so we needn't return it */
8350 static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode,
8353 struct btrfs_dio_private *dip = NULL;
8354 struct bio *bio = NULL;
8355 struct btrfs_io_bio *io_bio;
8356 bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
8359 bio = btrfs_bio_clone(dio_bio);
8361 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8367 dip->private = dio_bio->bi_private;
8369 dip->logical_offset = file_offset;
8370 dip->bytes = dio_bio->bi_iter.bi_size;
8371 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8372 bio->bi_private = dip;
8373 dip->orig_bio = bio;
8374 dip->dio_bio = dio_bio;
8375 atomic_set(&dip->pending_bios, 0);
8376 io_bio = btrfs_io_bio(bio);
8377 io_bio->logical = file_offset;
8380 bio->bi_end_io = btrfs_endio_direct_write;
8382 bio->bi_end_io = btrfs_endio_direct_read;
8383 dip->subio_endio = btrfs_subio_endio_read;
8387 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8388 * even if we fail to submit a bio, because in such case we do the
8389 * corresponding error handling below and it must not be done a second
8390 * time by btrfs_direct_IO().
8393 struct btrfs_dio_data *dio_data = current->journal_info;
8395 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8397 dio_data->unsubmitted_oe_range_start =
8398 dio_data->unsubmitted_oe_range_end;
8401 ret = btrfs_submit_direct_hook(dip);
8406 io_bio->end_io(io_bio, ret);
8410 * If we arrived here it means either we failed to submit the dip
8411 * or we either failed to clone the dio_bio or failed to allocate the
8412 * dip. If we cloned the dio_bio and allocated the dip, we can just
8413 * call bio_endio against our io_bio so that we get proper resource
8414 * cleanup if we fail to submit the dip, otherwise, we must do the
8415 * same as btrfs_endio_direct_[write|read] because we can't call these
8416 * callbacks - they require an allocated dip and a clone of dio_bio.
8421 * The end io callbacks free our dip, do the final put on bio
8422 * and all the cleanup and final put for dio_bio (through
8429 __endio_write_update_ordered(inode,
8431 dio_bio->bi_iter.bi_size,
8434 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8435 file_offset + dio_bio->bi_iter.bi_size - 1);
8437 dio_bio->bi_status = BLK_STS_IOERR;
8439 * Releases and cleans up our dio_bio, no need to bio_put()
8440 * nor bio_endio()/bio_io_error() against dio_bio.
8442 dio_end_io(dio_bio);
8449 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
8450 const struct iov_iter *iter, loff_t offset)
8454 unsigned int blocksize_mask = fs_info->sectorsize - 1;
8455 ssize_t retval = -EINVAL;
8457 if (offset & blocksize_mask)
8460 if (iov_iter_alignment(iter) & blocksize_mask)
8463 /* If this is a write we don't need to check anymore */
8464 if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter))
8467 * Check to make sure we don't have duplicate iov_base's in this
8468 * iovec, if so return EINVAL, otherwise we'll get csum errors
8469 * when reading back.
8471 for (seg = 0; seg < iter->nr_segs; seg++) {
8472 for (i = seg + 1; i < iter->nr_segs; i++) {
8473 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8482 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8484 struct file *file = iocb->ki_filp;
8485 struct inode *inode = file->f_mapping->host;
8486 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8487 struct btrfs_dio_data dio_data = { 0 };
8488 struct extent_changeset *data_reserved = NULL;
8489 loff_t offset = iocb->ki_pos;
8493 bool relock = false;
8496 if (check_direct_IO(fs_info, iter, offset))
8499 inode_dio_begin(inode);
8502 * The generic stuff only does filemap_write_and_wait_range, which
8503 * isn't enough if we've written compressed pages to this area, so
8504 * we need to flush the dirty pages again to make absolutely sure
8505 * that any outstanding dirty pages are on disk.
8507 count = iov_iter_count(iter);
8508 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8509 &BTRFS_I(inode)->runtime_flags))
8510 filemap_fdatawrite_range(inode->i_mapping, offset,
8511 offset + count - 1);
8513 if (iov_iter_rw(iter) == WRITE) {
8515 * If the write DIO is beyond the EOF, we need update
8516 * the isize, but it is protected by i_mutex. So we can
8517 * not unlock the i_mutex at this case.
8519 if (offset + count <= inode->i_size) {
8520 dio_data.overwrite = 1;
8521 inode_unlock(inode);
8523 } else if (iocb->ki_flags & IOCB_NOWAIT) {
8527 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
8533 * We need to know how many extents we reserved so that we can
8534 * do the accounting properly if we go over the number we
8535 * originally calculated. Abuse current->journal_info for this.
8537 dio_data.reserve = round_up(count,
8538 fs_info->sectorsize);
8539 dio_data.unsubmitted_oe_range_start = (u64)offset;
8540 dio_data.unsubmitted_oe_range_end = (u64)offset;
8541 current->journal_info = &dio_data;
8542 down_read(&BTRFS_I(inode)->dio_sem);
8543 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8544 &BTRFS_I(inode)->runtime_flags)) {
8545 inode_dio_end(inode);
8546 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8550 ret = __blockdev_direct_IO(iocb, inode,
8551 fs_info->fs_devices->latest_bdev,
8552 iter, btrfs_get_blocks_direct, NULL,
8553 btrfs_submit_direct, flags);
8554 if (iov_iter_rw(iter) == WRITE) {
8555 up_read(&BTRFS_I(inode)->dio_sem);
8556 current->journal_info = NULL;
8557 if (ret < 0 && ret != -EIOCBQUEUED) {
8558 if (dio_data.reserve)
8559 btrfs_delalloc_release_space(inode, data_reserved,
8560 offset, dio_data.reserve, true);
8562 * On error we might have left some ordered extents
8563 * without submitting corresponding bios for them, so
8564 * cleanup them up to avoid other tasks getting them
8565 * and waiting for them to complete forever.
8567 if (dio_data.unsubmitted_oe_range_start <
8568 dio_data.unsubmitted_oe_range_end)
8569 __endio_write_update_ordered(inode,
8570 dio_data.unsubmitted_oe_range_start,
8571 dio_data.unsubmitted_oe_range_end -
8572 dio_data.unsubmitted_oe_range_start,
8574 } else if (ret >= 0 && (size_t)ret < count)
8575 btrfs_delalloc_release_space(inode, data_reserved,
8576 offset, count - (size_t)ret, true);
8577 btrfs_delalloc_release_extents(BTRFS_I(inode), count, false);
8581 inode_dio_end(inode);
8585 extent_changeset_free(data_reserved);
8589 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8591 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8592 __u64 start, __u64 len)
8596 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8600 return extent_fiemap(inode, fieinfo, start, len);
8603 int btrfs_readpage(struct file *file, struct page *page)
8605 struct extent_io_tree *tree;
8606 tree = &BTRFS_I(page->mapping->host)->io_tree;
8607 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8610 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8612 struct inode *inode = page->mapping->host;
8615 if (current->flags & PF_MEMALLOC) {
8616 redirty_page_for_writepage(wbc, page);
8622 * If we are under memory pressure we will call this directly from the
8623 * VM, we need to make sure we have the inode referenced for the ordered
8624 * extent. If not just return like we didn't do anything.
8626 if (!igrab(inode)) {
8627 redirty_page_for_writepage(wbc, page);
8628 return AOP_WRITEPAGE_ACTIVATE;
8630 ret = extent_write_full_page(page, wbc);
8631 btrfs_add_delayed_iput(inode);
8635 static int btrfs_writepages(struct address_space *mapping,
8636 struct writeback_control *wbc)
8638 return extent_writepages(mapping, wbc);
8642 btrfs_readpages(struct file *file, struct address_space *mapping,
8643 struct list_head *pages, unsigned nr_pages)
8645 return extent_readpages(mapping, pages, nr_pages);
8648 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8650 int ret = try_release_extent_mapping(page, gfp_flags);
8652 ClearPagePrivate(page);
8653 set_page_private(page, 0);
8659 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8661 if (PageWriteback(page) || PageDirty(page))
8663 return __btrfs_releasepage(page, gfp_flags);
8666 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8667 unsigned int length)
8669 struct inode *inode = page->mapping->host;
8670 struct extent_io_tree *tree;
8671 struct btrfs_ordered_extent *ordered;
8672 struct extent_state *cached_state = NULL;
8673 u64 page_start = page_offset(page);
8674 u64 page_end = page_start + PAGE_SIZE - 1;
8677 int inode_evicting = inode->i_state & I_FREEING;
8680 * we have the page locked, so new writeback can't start,
8681 * and the dirty bit won't be cleared while we are here.
8683 * Wait for IO on this page so that we can safely clear
8684 * the PagePrivate2 bit and do ordered accounting
8686 wait_on_page_writeback(page);
8688 tree = &BTRFS_I(inode)->io_tree;
8690 btrfs_releasepage(page, GFP_NOFS);
8694 if (!inode_evicting)
8695 lock_extent_bits(tree, page_start, page_end, &cached_state);
8698 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
8699 page_end - start + 1);
8701 end = min(page_end, ordered->file_offset + ordered->len - 1);
8703 * IO on this page will never be started, so we need
8704 * to account for any ordered extents now
8706 if (!inode_evicting)
8707 clear_extent_bit(tree, start, end,
8708 EXTENT_DIRTY | EXTENT_DELALLOC |
8709 EXTENT_DELALLOC_NEW |
8710 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8711 EXTENT_DEFRAG, 1, 0, &cached_state);
8713 * whoever cleared the private bit is responsible
8714 * for the finish_ordered_io
8716 if (TestClearPagePrivate2(page)) {
8717 struct btrfs_ordered_inode_tree *tree;
8720 tree = &BTRFS_I(inode)->ordered_tree;
8722 spin_lock_irq(&tree->lock);
8723 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8724 new_len = start - ordered->file_offset;
8725 if (new_len < ordered->truncated_len)
8726 ordered->truncated_len = new_len;
8727 spin_unlock_irq(&tree->lock);
8729 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8731 end - start + 1, 1))
8732 btrfs_finish_ordered_io(ordered);
8734 btrfs_put_ordered_extent(ordered);
8735 if (!inode_evicting) {
8736 cached_state = NULL;
8737 lock_extent_bits(tree, start, end,
8742 if (start < page_end)
8747 * Qgroup reserved space handler
8748 * Page here will be either
8749 * 1) Already written to disk
8750 * In this case, its reserved space is released from data rsv map
8751 * and will be freed by delayed_ref handler finally.
8752 * So even we call qgroup_free_data(), it won't decrease reserved
8754 * 2) Not written to disk
8755 * This means the reserved space should be freed here. However,
8756 * if a truncate invalidates the page (by clearing PageDirty)
8757 * and the page is accounted for while allocating extent
8758 * in btrfs_check_data_free_space() we let delayed_ref to
8759 * free the entire extent.
8761 if (PageDirty(page))
8762 btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
8763 if (!inode_evicting) {
8764 clear_extent_bit(tree, page_start, page_end,
8765 EXTENT_LOCKED | EXTENT_DIRTY |
8766 EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
8767 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
8770 __btrfs_releasepage(page, GFP_NOFS);
8773 ClearPageChecked(page);
8774 if (PagePrivate(page)) {
8775 ClearPagePrivate(page);
8776 set_page_private(page, 0);
8782 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8783 * called from a page fault handler when a page is first dirtied. Hence we must
8784 * be careful to check for EOF conditions here. We set the page up correctly
8785 * for a written page which means we get ENOSPC checking when writing into
8786 * holes and correct delalloc and unwritten extent mapping on filesystems that
8787 * support these features.
8789 * We are not allowed to take the i_mutex here so we have to play games to
8790 * protect against truncate races as the page could now be beyond EOF. Because
8791 * truncate_setsize() writes the inode size before removing pages, once we have
8792 * the page lock we can determine safely if the page is beyond EOF. If it is not
8793 * beyond EOF, then the page is guaranteed safe against truncation until we
8796 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8798 struct page *page = vmf->page;
8799 struct inode *inode = file_inode(vmf->vma->vm_file);
8800 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8801 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8802 struct btrfs_ordered_extent *ordered;
8803 struct extent_state *cached_state = NULL;
8804 struct extent_changeset *data_reserved = NULL;
8806 unsigned long zero_start;
8816 reserved_space = PAGE_SIZE;
8818 sb_start_pagefault(inode->i_sb);
8819 page_start = page_offset(page);
8820 page_end = page_start + PAGE_SIZE - 1;
8824 * Reserving delalloc space after obtaining the page lock can lead to
8825 * deadlock. For example, if a dirty page is locked by this function
8826 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8827 * dirty page write out, then the btrfs_writepage() function could
8828 * end up waiting indefinitely to get a lock on the page currently
8829 * being processed by btrfs_page_mkwrite() function.
8831 ret2 = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
8834 ret2 = file_update_time(vmf->vma->vm_file);
8838 ret = vmf_error(ret2);
8844 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8847 size = i_size_read(inode);
8849 if ((page->mapping != inode->i_mapping) ||
8850 (page_start >= size)) {
8851 /* page got truncated out from underneath us */
8854 wait_on_page_writeback(page);
8856 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8857 set_page_extent_mapped(page);
8860 * we can't set the delalloc bits if there are pending ordered
8861 * extents. Drop our locks and wait for them to finish
8863 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8866 unlock_extent_cached(io_tree, page_start, page_end,
8869 btrfs_start_ordered_extent(inode, ordered, 1);
8870 btrfs_put_ordered_extent(ordered);
8874 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8875 reserved_space = round_up(size - page_start,
8876 fs_info->sectorsize);
8877 if (reserved_space < PAGE_SIZE) {
8878 end = page_start + reserved_space - 1;
8879 btrfs_delalloc_release_space(inode, data_reserved,
8880 page_start, PAGE_SIZE - reserved_space,
8886 * page_mkwrite gets called when the page is firstly dirtied after it's
8887 * faulted in, but write(2) could also dirty a page and set delalloc
8888 * bits, thus in this case for space account reason, we still need to
8889 * clear any delalloc bits within this page range since we have to
8890 * reserve data&meta space before lock_page() (see above comments).
8892 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8893 EXTENT_DIRTY | EXTENT_DELALLOC |
8894 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8895 0, 0, &cached_state);
8897 ret2 = btrfs_set_extent_delalloc(inode, page_start, end, 0,
8900 unlock_extent_cached(io_tree, page_start, page_end,
8902 ret = VM_FAULT_SIGBUS;
8907 /* page is wholly or partially inside EOF */
8908 if (page_start + PAGE_SIZE > size)
8909 zero_start = size & ~PAGE_MASK;
8911 zero_start = PAGE_SIZE;
8913 if (zero_start != PAGE_SIZE) {
8915 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
8916 flush_dcache_page(page);
8919 ClearPageChecked(page);
8920 set_page_dirty(page);
8921 SetPageUptodate(page);
8923 BTRFS_I(inode)->last_trans = fs_info->generation;
8924 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8925 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8927 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8930 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, true);
8931 sb_end_pagefault(inode->i_sb);
8932 extent_changeset_free(data_reserved);
8933 return VM_FAULT_LOCKED;
8939 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, (ret != 0));
8940 btrfs_delalloc_release_space(inode, data_reserved, page_start,
8941 reserved_space, (ret != 0));
8943 sb_end_pagefault(inode->i_sb);
8944 extent_changeset_free(data_reserved);
8948 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
8950 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8951 struct btrfs_root *root = BTRFS_I(inode)->root;
8952 struct btrfs_block_rsv *rsv;
8954 struct btrfs_trans_handle *trans;
8955 u64 mask = fs_info->sectorsize - 1;
8956 u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
8958 if (!skip_writeback) {
8959 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8966 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8967 * things going on here:
8969 * 1) We need to reserve space to update our inode.
8971 * 2) We need to have something to cache all the space that is going to
8972 * be free'd up by the truncate operation, but also have some slack
8973 * space reserved in case it uses space during the truncate (thank you
8974 * very much snapshotting).
8976 * And we need these to be separate. The fact is we can use a lot of
8977 * space doing the truncate, and we have no earthly idea how much space
8978 * we will use, so we need the truncate reservation to be separate so it
8979 * doesn't end up using space reserved for updating the inode. We also
8980 * need to be able to stop the transaction and start a new one, which
8981 * means we need to be able to update the inode several times, and we
8982 * have no idea of knowing how many times that will be, so we can't just
8983 * reserve 1 item for the entirety of the operation, so that has to be
8984 * done separately as well.
8986 * So that leaves us with
8988 * 1) rsv - for the truncate reservation, which we will steal from the
8989 * transaction reservation.
8990 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8991 * updating the inode.
8993 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8996 rsv->size = min_size;
9000 * 1 for the truncate slack space
9001 * 1 for updating the inode.
9003 trans = btrfs_start_transaction(root, 2);
9004 if (IS_ERR(trans)) {
9005 ret = PTR_ERR(trans);
9009 /* Migrate the slack space for the truncate to our reserve */
9010 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
9015 * So if we truncate and then write and fsync we normally would just
9016 * write the extents that changed, which is a problem if we need to
9017 * first truncate that entire inode. So set this flag so we write out
9018 * all of the extents in the inode to the sync log so we're completely
9021 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9022 trans->block_rsv = rsv;
9025 ret = btrfs_truncate_inode_items(trans, root, inode,
9027 BTRFS_EXTENT_DATA_KEY);
9028 trans->block_rsv = &fs_info->trans_block_rsv;
9029 if (ret != -ENOSPC && ret != -EAGAIN)
9032 ret = btrfs_update_inode(trans, root, inode);
9036 btrfs_end_transaction(trans);
9037 btrfs_btree_balance_dirty(fs_info);
9039 trans = btrfs_start_transaction(root, 2);
9040 if (IS_ERR(trans)) {
9041 ret = PTR_ERR(trans);
9046 btrfs_block_rsv_release(fs_info, rsv, -1);
9047 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
9048 rsv, min_size, false);
9049 BUG_ON(ret); /* shouldn't happen */
9050 trans->block_rsv = rsv;
9054 * We can't call btrfs_truncate_block inside a trans handle as we could
9055 * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
9056 * we've truncated everything except the last little bit, and can do
9057 * btrfs_truncate_block and then update the disk_i_size.
9059 if (ret == NEED_TRUNCATE_BLOCK) {
9060 btrfs_end_transaction(trans);
9061 btrfs_btree_balance_dirty(fs_info);
9063 ret = btrfs_truncate_block(inode, inode->i_size, 0, 0);
9066 trans = btrfs_start_transaction(root, 1);
9067 if (IS_ERR(trans)) {
9068 ret = PTR_ERR(trans);
9071 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
9077 trans->block_rsv = &fs_info->trans_block_rsv;
9078 ret2 = btrfs_update_inode(trans, root, inode);
9082 ret2 = btrfs_end_transaction(trans);
9085 btrfs_btree_balance_dirty(fs_info);
9088 btrfs_free_block_rsv(fs_info, rsv);
9094 * create a new subvolume directory/inode (helper for the ioctl).
9096 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9097 struct btrfs_root *new_root,
9098 struct btrfs_root *parent_root,
9101 struct inode *inode;
9105 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9106 new_dirid, new_dirid,
9107 S_IFDIR | (~current_umask() & S_IRWXUGO),
9110 return PTR_ERR(inode);
9111 inode->i_op = &btrfs_dir_inode_operations;
9112 inode->i_fop = &btrfs_dir_file_operations;
9114 set_nlink(inode, 1);
9115 btrfs_i_size_write(BTRFS_I(inode), 0);
9116 unlock_new_inode(inode);
9118 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9120 btrfs_err(new_root->fs_info,
9121 "error inheriting subvolume %llu properties: %d",
9122 new_root->root_key.objectid, err);
9124 err = btrfs_update_inode(trans, new_root, inode);
9130 struct inode *btrfs_alloc_inode(struct super_block *sb)
9132 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
9133 struct btrfs_inode *ei;
9134 struct inode *inode;
9136 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL);
9143 ei->last_sub_trans = 0;
9144 ei->logged_trans = 0;
9145 ei->delalloc_bytes = 0;
9146 ei->new_delalloc_bytes = 0;
9147 ei->defrag_bytes = 0;
9148 ei->disk_i_size = 0;
9151 ei->index_cnt = (u64)-1;
9153 ei->last_unlink_trans = 0;
9154 ei->last_log_commit = 0;
9156 spin_lock_init(&ei->lock);
9157 ei->outstanding_extents = 0;
9158 if (sb->s_magic != BTRFS_TEST_MAGIC)
9159 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
9160 BTRFS_BLOCK_RSV_DELALLOC);
9161 ei->runtime_flags = 0;
9162 ei->prop_compress = BTRFS_COMPRESS_NONE;
9163 ei->defrag_compress = BTRFS_COMPRESS_NONE;
9165 ei->delayed_node = NULL;
9167 ei->i_otime.tv_sec = 0;
9168 ei->i_otime.tv_nsec = 0;
9170 inode = &ei->vfs_inode;
9171 extent_map_tree_init(&ei->extent_tree);
9172 extent_io_tree_init(&ei->io_tree, inode);
9173 extent_io_tree_init(&ei->io_failure_tree, inode);
9174 ei->io_tree.track_uptodate = 1;
9175 ei->io_failure_tree.track_uptodate = 1;
9176 atomic_set(&ei->sync_writers, 0);
9177 mutex_init(&ei->log_mutex);
9178 mutex_init(&ei->delalloc_mutex);
9179 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9180 INIT_LIST_HEAD(&ei->delalloc_inodes);
9181 INIT_LIST_HEAD(&ei->delayed_iput);
9182 RB_CLEAR_NODE(&ei->rb_node);
9183 init_rwsem(&ei->dio_sem);
9188 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9189 void btrfs_test_destroy_inode(struct inode *inode)
9191 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9192 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9196 static void btrfs_i_callback(struct rcu_head *head)
9198 struct inode *inode = container_of(head, struct inode, i_rcu);
9199 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9202 void btrfs_destroy_inode(struct inode *inode)
9204 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9205 struct btrfs_ordered_extent *ordered;
9206 struct btrfs_root *root = BTRFS_I(inode)->root;
9208 WARN_ON(!hlist_empty(&inode->i_dentry));
9209 WARN_ON(inode->i_data.nrpages);
9210 WARN_ON(BTRFS_I(inode)->block_rsv.reserved);
9211 WARN_ON(BTRFS_I(inode)->block_rsv.size);
9212 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9213 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9214 WARN_ON(BTRFS_I(inode)->new_delalloc_bytes);
9215 WARN_ON(BTRFS_I(inode)->csum_bytes);
9216 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9219 * This can happen where we create an inode, but somebody else also
9220 * created the same inode and we need to destroy the one we already
9227 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9232 "found ordered extent %llu %llu on inode cleanup",
9233 ordered->file_offset, ordered->len);
9234 btrfs_remove_ordered_extent(inode, ordered);
9235 btrfs_put_ordered_extent(ordered);
9236 btrfs_put_ordered_extent(ordered);
9239 btrfs_qgroup_check_reserved_leak(inode);
9240 inode_tree_del(inode);
9241 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9243 call_rcu(&inode->i_rcu, btrfs_i_callback);
9246 int btrfs_drop_inode(struct inode *inode)
9248 struct btrfs_root *root = BTRFS_I(inode)->root;
9253 /* the snap/subvol tree is on deleting */
9254 if (btrfs_root_refs(&root->root_item) == 0)
9257 return generic_drop_inode(inode);
9260 static void init_once(void *foo)
9262 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9264 inode_init_once(&ei->vfs_inode);
9267 void __cold btrfs_destroy_cachep(void)
9270 * Make sure all delayed rcu free inodes are flushed before we
9274 kmem_cache_destroy(btrfs_inode_cachep);
9275 kmem_cache_destroy(btrfs_trans_handle_cachep);
9276 kmem_cache_destroy(btrfs_path_cachep);
9277 kmem_cache_destroy(btrfs_free_space_cachep);
9280 int __init btrfs_init_cachep(void)
9282 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9283 sizeof(struct btrfs_inode), 0,
9284 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9286 if (!btrfs_inode_cachep)
9289 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9290 sizeof(struct btrfs_trans_handle), 0,
9291 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9292 if (!btrfs_trans_handle_cachep)
9295 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9296 sizeof(struct btrfs_path), 0,
9297 SLAB_MEM_SPREAD, NULL);
9298 if (!btrfs_path_cachep)
9301 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9302 sizeof(struct btrfs_free_space), 0,
9303 SLAB_MEM_SPREAD, NULL);
9304 if (!btrfs_free_space_cachep)
9309 btrfs_destroy_cachep();
9313 static int btrfs_getattr(const struct path *path, struct kstat *stat,
9314 u32 request_mask, unsigned int flags)
9317 struct inode *inode = d_inode(path->dentry);
9318 u32 blocksize = inode->i_sb->s_blocksize;
9319 u32 bi_flags = BTRFS_I(inode)->flags;
9321 stat->result_mask |= STATX_BTIME;
9322 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9323 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9324 if (bi_flags & BTRFS_INODE_APPEND)
9325 stat->attributes |= STATX_ATTR_APPEND;
9326 if (bi_flags & BTRFS_INODE_COMPRESS)
9327 stat->attributes |= STATX_ATTR_COMPRESSED;
9328 if (bi_flags & BTRFS_INODE_IMMUTABLE)
9329 stat->attributes |= STATX_ATTR_IMMUTABLE;
9330 if (bi_flags & BTRFS_INODE_NODUMP)
9331 stat->attributes |= STATX_ATTR_NODUMP;
9333 stat->attributes_mask |= (STATX_ATTR_APPEND |
9334 STATX_ATTR_COMPRESSED |
9335 STATX_ATTR_IMMUTABLE |
9338 generic_fillattr(inode, stat);
9339 stat->dev = BTRFS_I(inode)->root->anon_dev;
9341 spin_lock(&BTRFS_I(inode)->lock);
9342 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9343 spin_unlock(&BTRFS_I(inode)->lock);
9344 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9345 ALIGN(delalloc_bytes, blocksize)) >> 9;
9349 static int btrfs_rename_exchange(struct inode *old_dir,
9350 struct dentry *old_dentry,
9351 struct inode *new_dir,
9352 struct dentry *new_dentry)
9354 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9355 struct btrfs_trans_handle *trans;
9356 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9357 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9358 struct inode *new_inode = new_dentry->d_inode;
9359 struct inode *old_inode = old_dentry->d_inode;
9360 struct timespec64 ctime = current_time(old_inode);
9361 struct dentry *parent;
9362 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9363 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9368 bool root_log_pinned = false;
9369 bool dest_log_pinned = false;
9370 struct btrfs_log_ctx ctx_root;
9371 struct btrfs_log_ctx ctx_dest;
9372 bool sync_log_root = false;
9373 bool sync_log_dest = false;
9374 bool commit_transaction = false;
9376 /* we only allow rename subvolume link between subvolumes */
9377 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9380 btrfs_init_log_ctx(&ctx_root, old_inode);
9381 btrfs_init_log_ctx(&ctx_dest, new_inode);
9383 /* close the race window with snapshot create/destroy ioctl */
9384 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9385 down_read(&fs_info->subvol_sem);
9386 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9387 down_read(&fs_info->subvol_sem);
9390 * We want to reserve the absolute worst case amount of items. So if
9391 * both inodes are subvols and we need to unlink them then that would
9392 * require 4 item modifications, but if they are both normal inodes it
9393 * would require 5 item modifications, so we'll assume their normal
9394 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9395 * should cover the worst case number of items we'll modify.
9397 trans = btrfs_start_transaction(root, 12);
9398 if (IS_ERR(trans)) {
9399 ret = PTR_ERR(trans);
9404 * We need to find a free sequence number both in the source and
9405 * in the destination directory for the exchange.
9407 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9410 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9414 BTRFS_I(old_inode)->dir_index = 0ULL;
9415 BTRFS_I(new_inode)->dir_index = 0ULL;
9417 /* Reference for the source. */
9418 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9419 /* force full log commit if subvolume involved. */
9420 btrfs_set_log_full_commit(fs_info, trans);
9422 btrfs_pin_log_trans(root);
9423 root_log_pinned = true;
9424 ret = btrfs_insert_inode_ref(trans, dest,
9425 new_dentry->d_name.name,
9426 new_dentry->d_name.len,
9428 btrfs_ino(BTRFS_I(new_dir)),
9434 /* And now for the dest. */
9435 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9436 /* force full log commit if subvolume involved. */
9437 btrfs_set_log_full_commit(fs_info, trans);
9439 btrfs_pin_log_trans(dest);
9440 dest_log_pinned = true;
9441 ret = btrfs_insert_inode_ref(trans, root,
9442 old_dentry->d_name.name,
9443 old_dentry->d_name.len,
9445 btrfs_ino(BTRFS_I(old_dir)),
9451 /* Update inode version and ctime/mtime. */
9452 inode_inc_iversion(old_dir);
9453 inode_inc_iversion(new_dir);
9454 inode_inc_iversion(old_inode);
9455 inode_inc_iversion(new_inode);
9456 old_dir->i_ctime = old_dir->i_mtime = ctime;
9457 new_dir->i_ctime = new_dir->i_mtime = ctime;
9458 old_inode->i_ctime = ctime;
9459 new_inode->i_ctime = ctime;
9461 if (old_dentry->d_parent != new_dentry->d_parent) {
9462 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9463 BTRFS_I(old_inode), 1);
9464 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9465 BTRFS_I(new_inode), 1);
9468 /* src is a subvolume */
9469 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9470 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9471 ret = btrfs_unlink_subvol(trans, old_dir, root_objectid,
9472 old_dentry->d_name.name,
9473 old_dentry->d_name.len);
9474 } else { /* src is an inode */
9475 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9476 BTRFS_I(old_dentry->d_inode),
9477 old_dentry->d_name.name,
9478 old_dentry->d_name.len);
9480 ret = btrfs_update_inode(trans, root, old_inode);
9483 btrfs_abort_transaction(trans, ret);
9487 /* dest is a subvolume */
9488 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9489 root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9490 ret = btrfs_unlink_subvol(trans, new_dir, root_objectid,
9491 new_dentry->d_name.name,
9492 new_dentry->d_name.len);
9493 } else { /* dest is an inode */
9494 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9495 BTRFS_I(new_dentry->d_inode),
9496 new_dentry->d_name.name,
9497 new_dentry->d_name.len);
9499 ret = btrfs_update_inode(trans, dest, new_inode);
9502 btrfs_abort_transaction(trans, ret);
9506 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9507 new_dentry->d_name.name,
9508 new_dentry->d_name.len, 0, old_idx);
9510 btrfs_abort_transaction(trans, ret);
9514 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9515 old_dentry->d_name.name,
9516 old_dentry->d_name.len, 0, new_idx);
9518 btrfs_abort_transaction(trans, ret);
9522 if (old_inode->i_nlink == 1)
9523 BTRFS_I(old_inode)->dir_index = old_idx;
9524 if (new_inode->i_nlink == 1)
9525 BTRFS_I(new_inode)->dir_index = new_idx;
9527 if (root_log_pinned) {
9528 parent = new_dentry->d_parent;
9529 ret = btrfs_log_new_name(trans, BTRFS_I(old_inode),
9530 BTRFS_I(old_dir), parent,
9532 if (ret == BTRFS_NEED_LOG_SYNC)
9533 sync_log_root = true;
9534 else if (ret == BTRFS_NEED_TRANS_COMMIT)
9535 commit_transaction = true;
9537 btrfs_end_log_trans(root);
9538 root_log_pinned = false;
9540 if (dest_log_pinned) {
9541 if (!commit_transaction) {
9542 parent = old_dentry->d_parent;
9543 ret = btrfs_log_new_name(trans, BTRFS_I(new_inode),
9544 BTRFS_I(new_dir), parent,
9546 if (ret == BTRFS_NEED_LOG_SYNC)
9547 sync_log_dest = true;
9548 else if (ret == BTRFS_NEED_TRANS_COMMIT)
9549 commit_transaction = true;
9552 btrfs_end_log_trans(dest);
9553 dest_log_pinned = false;
9557 * If we have pinned a log and an error happened, we unpin tasks
9558 * trying to sync the log and force them to fallback to a transaction
9559 * commit if the log currently contains any of the inodes involved in
9560 * this rename operation (to ensure we do not persist a log with an
9561 * inconsistent state for any of these inodes or leading to any
9562 * inconsistencies when replayed). If the transaction was aborted, the
9563 * abortion reason is propagated to userspace when attempting to commit
9564 * the transaction. If the log does not contain any of these inodes, we
9565 * allow the tasks to sync it.
9567 if (ret && (root_log_pinned || dest_log_pinned)) {
9568 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9569 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9570 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9572 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9573 btrfs_set_log_full_commit(fs_info, trans);
9575 if (root_log_pinned) {
9576 btrfs_end_log_trans(root);
9577 root_log_pinned = false;
9579 if (dest_log_pinned) {
9580 btrfs_end_log_trans(dest);
9581 dest_log_pinned = false;
9584 if (!ret && sync_log_root && !commit_transaction) {
9585 ret = btrfs_sync_log(trans, BTRFS_I(old_inode)->root,
9588 commit_transaction = true;
9590 if (!ret && sync_log_dest && !commit_transaction) {
9591 ret = btrfs_sync_log(trans, BTRFS_I(new_inode)->root,
9594 commit_transaction = true;
9596 if (commit_transaction) {
9597 ret = btrfs_commit_transaction(trans);
9601 ret2 = btrfs_end_transaction(trans);
9602 ret = ret ? ret : ret2;
9605 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9606 up_read(&fs_info->subvol_sem);
9607 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9608 up_read(&fs_info->subvol_sem);
9613 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9614 struct btrfs_root *root,
9616 struct dentry *dentry)
9619 struct inode *inode;
9623 ret = btrfs_find_free_ino(root, &objectid);
9627 inode = btrfs_new_inode(trans, root, dir,
9628 dentry->d_name.name,
9630 btrfs_ino(BTRFS_I(dir)),
9632 S_IFCHR | WHITEOUT_MODE,
9635 if (IS_ERR(inode)) {
9636 ret = PTR_ERR(inode);
9640 inode->i_op = &btrfs_special_inode_operations;
9641 init_special_inode(inode, inode->i_mode,
9644 ret = btrfs_init_inode_security(trans, inode, dir,
9649 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9650 BTRFS_I(inode), 0, index);
9654 ret = btrfs_update_inode(trans, root, inode);
9656 unlock_new_inode(inode);
9658 inode_dec_link_count(inode);
9664 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9665 struct inode *new_dir, struct dentry *new_dentry,
9668 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9669 struct btrfs_trans_handle *trans;
9670 unsigned int trans_num_items;
9671 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9672 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9673 struct inode *new_inode = d_inode(new_dentry);
9674 struct inode *old_inode = d_inode(old_dentry);
9678 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9679 bool log_pinned = false;
9680 struct btrfs_log_ctx ctx;
9681 bool sync_log = false;
9682 bool commit_transaction = false;
9684 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9687 /* we only allow rename subvolume link between subvolumes */
9688 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9691 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9692 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9695 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9696 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9700 /* check for collisions, even if the name isn't there */
9701 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9702 new_dentry->d_name.name,
9703 new_dentry->d_name.len);
9706 if (ret == -EEXIST) {
9708 * eexist without a new_inode */
9709 if (WARN_ON(!new_inode)) {
9713 /* maybe -EOVERFLOW */
9720 * we're using rename to replace one file with another. Start IO on it
9721 * now so we don't add too much work to the end of the transaction
9723 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9724 filemap_flush(old_inode->i_mapping);
9726 /* close the racy window with snapshot create/destroy ioctl */
9727 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9728 down_read(&fs_info->subvol_sem);
9730 * We want to reserve the absolute worst case amount of items. So if
9731 * both inodes are subvols and we need to unlink them then that would
9732 * require 4 item modifications, but if they are both normal inodes it
9733 * would require 5 item modifications, so we'll assume they are normal
9734 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9735 * should cover the worst case number of items we'll modify.
9736 * If our rename has the whiteout flag, we need more 5 units for the
9737 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9738 * when selinux is enabled).
9740 trans_num_items = 11;
9741 if (flags & RENAME_WHITEOUT)
9742 trans_num_items += 5;
9743 trans = btrfs_start_transaction(root, trans_num_items);
9744 if (IS_ERR(trans)) {
9745 ret = PTR_ERR(trans);
9750 btrfs_record_root_in_trans(trans, dest);
9752 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9756 BTRFS_I(old_inode)->dir_index = 0ULL;
9757 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9758 /* force full log commit if subvolume involved. */
9759 btrfs_set_log_full_commit(fs_info, trans);
9761 btrfs_pin_log_trans(root);
9763 ret = btrfs_insert_inode_ref(trans, dest,
9764 new_dentry->d_name.name,
9765 new_dentry->d_name.len,
9767 btrfs_ino(BTRFS_I(new_dir)), index);
9772 inode_inc_iversion(old_dir);
9773 inode_inc_iversion(new_dir);
9774 inode_inc_iversion(old_inode);
9775 old_dir->i_ctime = old_dir->i_mtime =
9776 new_dir->i_ctime = new_dir->i_mtime =
9777 old_inode->i_ctime = current_time(old_dir);
9779 if (old_dentry->d_parent != new_dentry->d_parent)
9780 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9781 BTRFS_I(old_inode), 1);
9783 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9784 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9785 ret = btrfs_unlink_subvol(trans, old_dir, root_objectid,
9786 old_dentry->d_name.name,
9787 old_dentry->d_name.len);
9789 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9790 BTRFS_I(d_inode(old_dentry)),
9791 old_dentry->d_name.name,
9792 old_dentry->d_name.len);
9794 ret = btrfs_update_inode(trans, root, old_inode);
9797 btrfs_abort_transaction(trans, ret);
9802 inode_inc_iversion(new_inode);
9803 new_inode->i_ctime = current_time(new_inode);
9804 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9805 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9806 root_objectid = BTRFS_I(new_inode)->location.objectid;
9807 ret = btrfs_unlink_subvol(trans, new_dir, root_objectid,
9808 new_dentry->d_name.name,
9809 new_dentry->d_name.len);
9810 BUG_ON(new_inode->i_nlink == 0);
9812 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9813 BTRFS_I(d_inode(new_dentry)),
9814 new_dentry->d_name.name,
9815 new_dentry->d_name.len);
9817 if (!ret && new_inode->i_nlink == 0)
9818 ret = btrfs_orphan_add(trans,
9819 BTRFS_I(d_inode(new_dentry)));
9821 btrfs_abort_transaction(trans, ret);
9826 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9827 new_dentry->d_name.name,
9828 new_dentry->d_name.len, 0, index);
9830 btrfs_abort_transaction(trans, ret);
9834 if (old_inode->i_nlink == 1)
9835 BTRFS_I(old_inode)->dir_index = index;
9838 struct dentry *parent = new_dentry->d_parent;
9840 btrfs_init_log_ctx(&ctx, old_inode);
9841 ret = btrfs_log_new_name(trans, BTRFS_I(old_inode),
9842 BTRFS_I(old_dir), parent,
9844 if (ret == BTRFS_NEED_LOG_SYNC)
9846 else if (ret == BTRFS_NEED_TRANS_COMMIT)
9847 commit_transaction = true;
9849 btrfs_end_log_trans(root);
9853 if (flags & RENAME_WHITEOUT) {
9854 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9858 btrfs_abort_transaction(trans, ret);
9864 * If we have pinned the log and an error happened, we unpin tasks
9865 * trying to sync the log and force them to fallback to a transaction
9866 * commit if the log currently contains any of the inodes involved in
9867 * this rename operation (to ensure we do not persist a log with an
9868 * inconsistent state for any of these inodes or leading to any
9869 * inconsistencies when replayed). If the transaction was aborted, the
9870 * abortion reason is propagated to userspace when attempting to commit
9871 * the transaction. If the log does not contain any of these inodes, we
9872 * allow the tasks to sync it.
9874 if (ret && log_pinned) {
9875 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9876 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9877 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9879 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9880 btrfs_set_log_full_commit(fs_info, trans);
9882 btrfs_end_log_trans(root);
9885 if (!ret && sync_log) {
9886 ret = btrfs_sync_log(trans, BTRFS_I(old_inode)->root, &ctx);
9888 commit_transaction = true;
9890 if (commit_transaction) {
9891 ret = btrfs_commit_transaction(trans);
9895 ret2 = btrfs_end_transaction(trans);
9896 ret = ret ? ret : ret2;
9899 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9900 up_read(&fs_info->subvol_sem);
9905 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9906 struct inode *new_dir, struct dentry *new_dentry,
9909 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9912 if (flags & RENAME_EXCHANGE)
9913 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9916 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9919 struct btrfs_delalloc_work {
9920 struct inode *inode;
9921 struct completion completion;
9922 struct list_head list;
9923 struct btrfs_work work;
9926 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9928 struct btrfs_delalloc_work *delalloc_work;
9929 struct inode *inode;
9931 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9933 inode = delalloc_work->inode;
9934 filemap_flush(inode->i_mapping);
9935 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9936 &BTRFS_I(inode)->runtime_flags))
9937 filemap_flush(inode->i_mapping);
9940 complete(&delalloc_work->completion);
9943 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9945 struct btrfs_delalloc_work *work;
9947 work = kmalloc(sizeof(*work), GFP_NOFS);
9951 init_completion(&work->completion);
9952 INIT_LIST_HEAD(&work->list);
9953 work->inode = inode;
9954 WARN_ON_ONCE(!inode);
9955 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9956 btrfs_run_delalloc_work, NULL, NULL);
9962 * some fairly slow code that needs optimization. This walks the list
9963 * of all the inodes with pending delalloc and forces them to disk.
9965 static int start_delalloc_inodes(struct btrfs_root *root, int nr)
9967 struct btrfs_inode *binode;
9968 struct inode *inode;
9969 struct btrfs_delalloc_work *work, *next;
9970 struct list_head works;
9971 struct list_head splice;
9974 INIT_LIST_HEAD(&works);
9975 INIT_LIST_HEAD(&splice);
9977 mutex_lock(&root->delalloc_mutex);
9978 spin_lock(&root->delalloc_lock);
9979 list_splice_init(&root->delalloc_inodes, &splice);
9980 while (!list_empty(&splice)) {
9981 binode = list_entry(splice.next, struct btrfs_inode,
9984 list_move_tail(&binode->delalloc_inodes,
9985 &root->delalloc_inodes);
9986 inode = igrab(&binode->vfs_inode);
9988 cond_resched_lock(&root->delalloc_lock);
9991 spin_unlock(&root->delalloc_lock);
9993 work = btrfs_alloc_delalloc_work(inode);
9999 list_add_tail(&work->list, &works);
10000 btrfs_queue_work(root->fs_info->flush_workers,
10003 if (nr != -1 && ret >= nr)
10006 spin_lock(&root->delalloc_lock);
10008 spin_unlock(&root->delalloc_lock);
10011 list_for_each_entry_safe(work, next, &works, list) {
10012 list_del_init(&work->list);
10013 wait_for_completion(&work->completion);
10017 if (!list_empty(&splice)) {
10018 spin_lock(&root->delalloc_lock);
10019 list_splice_tail(&splice, &root->delalloc_inodes);
10020 spin_unlock(&root->delalloc_lock);
10022 mutex_unlock(&root->delalloc_mutex);
10026 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
10028 struct btrfs_fs_info *fs_info = root->fs_info;
10031 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10034 ret = start_delalloc_inodes(root, -1);
10040 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int nr)
10042 struct btrfs_root *root;
10043 struct list_head splice;
10046 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10049 INIT_LIST_HEAD(&splice);
10051 mutex_lock(&fs_info->delalloc_root_mutex);
10052 spin_lock(&fs_info->delalloc_root_lock);
10053 list_splice_init(&fs_info->delalloc_roots, &splice);
10054 while (!list_empty(&splice) && nr) {
10055 root = list_first_entry(&splice, struct btrfs_root,
10057 root = btrfs_grab_fs_root(root);
10059 list_move_tail(&root->delalloc_root,
10060 &fs_info->delalloc_roots);
10061 spin_unlock(&fs_info->delalloc_root_lock);
10063 ret = start_delalloc_inodes(root, nr);
10064 btrfs_put_fs_root(root);
10072 spin_lock(&fs_info->delalloc_root_lock);
10074 spin_unlock(&fs_info->delalloc_root_lock);
10078 if (!list_empty(&splice)) {
10079 spin_lock(&fs_info->delalloc_root_lock);
10080 list_splice_tail(&splice, &fs_info->delalloc_roots);
10081 spin_unlock(&fs_info->delalloc_root_lock);
10083 mutex_unlock(&fs_info->delalloc_root_mutex);
10087 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10088 const char *symname)
10090 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10091 struct btrfs_trans_handle *trans;
10092 struct btrfs_root *root = BTRFS_I(dir)->root;
10093 struct btrfs_path *path;
10094 struct btrfs_key key;
10095 struct inode *inode = NULL;
10102 struct btrfs_file_extent_item *ei;
10103 struct extent_buffer *leaf;
10105 name_len = strlen(symname);
10106 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
10107 return -ENAMETOOLONG;
10110 * 2 items for inode item and ref
10111 * 2 items for dir items
10112 * 1 item for updating parent inode item
10113 * 1 item for the inline extent item
10114 * 1 item for xattr if selinux is on
10116 trans = btrfs_start_transaction(root, 7);
10118 return PTR_ERR(trans);
10120 err = btrfs_find_free_ino(root, &objectid);
10124 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10125 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
10126 objectid, S_IFLNK|S_IRWXUGO, &index);
10127 if (IS_ERR(inode)) {
10128 err = PTR_ERR(inode);
10134 * If the active LSM wants to access the inode during
10135 * d_instantiate it needs these. Smack checks to see
10136 * if the filesystem supports xattrs by looking at the
10139 inode->i_fop = &btrfs_file_operations;
10140 inode->i_op = &btrfs_file_inode_operations;
10141 inode->i_mapping->a_ops = &btrfs_aops;
10142 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10144 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10148 path = btrfs_alloc_path();
10153 key.objectid = btrfs_ino(BTRFS_I(inode));
10155 key.type = BTRFS_EXTENT_DATA_KEY;
10156 datasize = btrfs_file_extent_calc_inline_size(name_len);
10157 err = btrfs_insert_empty_item(trans, root, path, &key,
10160 btrfs_free_path(path);
10163 leaf = path->nodes[0];
10164 ei = btrfs_item_ptr(leaf, path->slots[0],
10165 struct btrfs_file_extent_item);
10166 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10167 btrfs_set_file_extent_type(leaf, ei,
10168 BTRFS_FILE_EXTENT_INLINE);
10169 btrfs_set_file_extent_encryption(leaf, ei, 0);
10170 btrfs_set_file_extent_compression(leaf, ei, 0);
10171 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10172 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10174 ptr = btrfs_file_extent_inline_start(ei);
10175 write_extent_buffer(leaf, symname, ptr, name_len);
10176 btrfs_mark_buffer_dirty(leaf);
10177 btrfs_free_path(path);
10179 inode->i_op = &btrfs_symlink_inode_operations;
10180 inode_nohighmem(inode);
10181 inode->i_mapping->a_ops = &btrfs_aops;
10182 inode_set_bytes(inode, name_len);
10183 btrfs_i_size_write(BTRFS_I(inode), name_len);
10184 err = btrfs_update_inode(trans, root, inode);
10186 * Last step, add directory indexes for our symlink inode. This is the
10187 * last step to avoid extra cleanup of these indexes if an error happens
10191 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
10192 BTRFS_I(inode), 0, index);
10196 d_instantiate_new(dentry, inode);
10199 btrfs_end_transaction(trans);
10200 if (err && inode) {
10201 inode_dec_link_count(inode);
10202 discard_new_inode(inode);
10204 btrfs_btree_balance_dirty(fs_info);
10208 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10209 u64 start, u64 num_bytes, u64 min_size,
10210 loff_t actual_len, u64 *alloc_hint,
10211 struct btrfs_trans_handle *trans)
10213 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
10214 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10215 struct extent_map *em;
10216 struct btrfs_root *root = BTRFS_I(inode)->root;
10217 struct btrfs_key ins;
10218 u64 cur_offset = start;
10221 u64 last_alloc = (u64)-1;
10223 bool own_trans = true;
10224 u64 end = start + num_bytes - 1;
10228 while (num_bytes > 0) {
10230 trans = btrfs_start_transaction(root, 3);
10231 if (IS_ERR(trans)) {
10232 ret = PTR_ERR(trans);
10237 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10238 cur_bytes = max(cur_bytes, min_size);
10240 * If we are severely fragmented we could end up with really
10241 * small allocations, so if the allocator is returning small
10242 * chunks lets make its job easier by only searching for those
10245 cur_bytes = min(cur_bytes, last_alloc);
10246 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10247 min_size, 0, *alloc_hint, &ins, 1, 0);
10250 btrfs_end_transaction(trans);
10253 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10255 last_alloc = ins.offset;
10256 ret = insert_reserved_file_extent(trans, inode,
10257 cur_offset, ins.objectid,
10258 ins.offset, ins.offset,
10259 ins.offset, 0, 0, 0,
10260 BTRFS_FILE_EXTENT_PREALLOC);
10262 btrfs_free_reserved_extent(fs_info, ins.objectid,
10264 btrfs_abort_transaction(trans, ret);
10266 btrfs_end_transaction(trans);
10270 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10271 cur_offset + ins.offset -1, 0);
10273 em = alloc_extent_map();
10275 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10276 &BTRFS_I(inode)->runtime_flags);
10280 em->start = cur_offset;
10281 em->orig_start = cur_offset;
10282 em->len = ins.offset;
10283 em->block_start = ins.objectid;
10284 em->block_len = ins.offset;
10285 em->orig_block_len = ins.offset;
10286 em->ram_bytes = ins.offset;
10287 em->bdev = fs_info->fs_devices->latest_bdev;
10288 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10289 em->generation = trans->transid;
10292 write_lock(&em_tree->lock);
10293 ret = add_extent_mapping(em_tree, em, 1);
10294 write_unlock(&em_tree->lock);
10295 if (ret != -EEXIST)
10297 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10298 cur_offset + ins.offset - 1,
10301 free_extent_map(em);
10303 num_bytes -= ins.offset;
10304 cur_offset += ins.offset;
10305 *alloc_hint = ins.objectid + ins.offset;
10307 inode_inc_iversion(inode);
10308 inode->i_ctime = current_time(inode);
10309 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10310 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10311 (actual_len > inode->i_size) &&
10312 (cur_offset > inode->i_size)) {
10313 if (cur_offset > actual_len)
10314 i_size = actual_len;
10316 i_size = cur_offset;
10317 i_size_write(inode, i_size);
10318 btrfs_ordered_update_i_size(inode, i_size, NULL);
10321 ret = btrfs_update_inode(trans, root, inode);
10324 btrfs_abort_transaction(trans, ret);
10326 btrfs_end_transaction(trans);
10331 btrfs_end_transaction(trans);
10333 if (cur_offset < end)
10334 btrfs_free_reserved_data_space(inode, NULL, cur_offset,
10335 end - cur_offset + 1);
10339 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10340 u64 start, u64 num_bytes, u64 min_size,
10341 loff_t actual_len, u64 *alloc_hint)
10343 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10344 min_size, actual_len, alloc_hint,
10348 int btrfs_prealloc_file_range_trans(struct inode *inode,
10349 struct btrfs_trans_handle *trans, int mode,
10350 u64 start, u64 num_bytes, u64 min_size,
10351 loff_t actual_len, u64 *alloc_hint)
10353 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10354 min_size, actual_len, alloc_hint, trans);
10357 static int btrfs_set_page_dirty(struct page *page)
10359 return __set_page_dirty_nobuffers(page);
10362 static int btrfs_permission(struct inode *inode, int mask)
10364 struct btrfs_root *root = BTRFS_I(inode)->root;
10365 umode_t mode = inode->i_mode;
10367 if (mask & MAY_WRITE &&
10368 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10369 if (btrfs_root_readonly(root))
10371 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10374 return generic_permission(inode, mask);
10377 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10379 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10380 struct btrfs_trans_handle *trans;
10381 struct btrfs_root *root = BTRFS_I(dir)->root;
10382 struct inode *inode = NULL;
10388 * 5 units required for adding orphan entry
10390 trans = btrfs_start_transaction(root, 5);
10392 return PTR_ERR(trans);
10394 ret = btrfs_find_free_ino(root, &objectid);
10398 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10399 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10400 if (IS_ERR(inode)) {
10401 ret = PTR_ERR(inode);
10406 inode->i_fop = &btrfs_file_operations;
10407 inode->i_op = &btrfs_file_inode_operations;
10409 inode->i_mapping->a_ops = &btrfs_aops;
10410 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10412 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10416 ret = btrfs_update_inode(trans, root, inode);
10419 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10424 * We set number of links to 0 in btrfs_new_inode(), and here we set
10425 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10428 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10430 set_nlink(inode, 1);
10431 d_tmpfile(dentry, inode);
10432 unlock_new_inode(inode);
10433 mark_inode_dirty(inode);
10435 btrfs_end_transaction(trans);
10437 discard_new_inode(inode);
10438 btrfs_btree_balance_dirty(fs_info);
10442 __attribute__((const))
10443 static int btrfs_readpage_io_failed_hook(struct page *page, int failed_mirror)
10448 void btrfs_set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
10450 struct inode *inode = tree->private_data;
10451 unsigned long index = start >> PAGE_SHIFT;
10452 unsigned long end_index = end >> PAGE_SHIFT;
10455 while (index <= end_index) {
10456 page = find_get_page(inode->i_mapping, index);
10457 ASSERT(page); /* Pages should be in the extent_io_tree */
10458 set_page_writeback(page);
10464 static const struct inode_operations btrfs_dir_inode_operations = {
10465 .getattr = btrfs_getattr,
10466 .lookup = btrfs_lookup,
10467 .create = btrfs_create,
10468 .unlink = btrfs_unlink,
10469 .link = btrfs_link,
10470 .mkdir = btrfs_mkdir,
10471 .rmdir = btrfs_rmdir,
10472 .rename = btrfs_rename2,
10473 .symlink = btrfs_symlink,
10474 .setattr = btrfs_setattr,
10475 .mknod = btrfs_mknod,
10476 .listxattr = btrfs_listxattr,
10477 .permission = btrfs_permission,
10478 .get_acl = btrfs_get_acl,
10479 .set_acl = btrfs_set_acl,
10480 .update_time = btrfs_update_time,
10481 .tmpfile = btrfs_tmpfile,
10483 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10484 .lookup = btrfs_lookup,
10485 .permission = btrfs_permission,
10486 .update_time = btrfs_update_time,
10489 static const struct file_operations btrfs_dir_file_operations = {
10490 .llseek = generic_file_llseek,
10491 .read = generic_read_dir,
10492 .iterate_shared = btrfs_real_readdir,
10493 .open = btrfs_opendir,
10494 .unlocked_ioctl = btrfs_ioctl,
10495 #ifdef CONFIG_COMPAT
10496 .compat_ioctl = btrfs_compat_ioctl,
10498 .release = btrfs_release_file,
10499 .fsync = btrfs_sync_file,
10502 static const struct extent_io_ops btrfs_extent_io_ops = {
10503 /* mandatory callbacks */
10504 .submit_bio_hook = btrfs_submit_bio_hook,
10505 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10506 .readpage_io_failed_hook = btrfs_readpage_io_failed_hook,
10508 /* optional callbacks */
10509 .split_extent_hook = btrfs_split_extent_hook,
10513 * btrfs doesn't support the bmap operation because swapfiles
10514 * use bmap to make a mapping of extents in the file. They assume
10515 * these extents won't change over the life of the file and they
10516 * use the bmap result to do IO directly to the drive.
10518 * the btrfs bmap call would return logical addresses that aren't
10519 * suitable for IO and they also will change frequently as COW
10520 * operations happen. So, swapfile + btrfs == corruption.
10522 * For now we're avoiding this by dropping bmap.
10524 static const struct address_space_operations btrfs_aops = {
10525 .readpage = btrfs_readpage,
10526 .writepage = btrfs_writepage,
10527 .writepages = btrfs_writepages,
10528 .readpages = btrfs_readpages,
10529 .direct_IO = btrfs_direct_IO,
10530 .invalidatepage = btrfs_invalidatepage,
10531 .releasepage = btrfs_releasepage,
10532 .set_page_dirty = btrfs_set_page_dirty,
10533 .error_remove_page = generic_error_remove_page,
10536 static const struct inode_operations btrfs_file_inode_operations = {
10537 .getattr = btrfs_getattr,
10538 .setattr = btrfs_setattr,
10539 .listxattr = btrfs_listxattr,
10540 .permission = btrfs_permission,
10541 .fiemap = btrfs_fiemap,
10542 .get_acl = btrfs_get_acl,
10543 .set_acl = btrfs_set_acl,
10544 .update_time = btrfs_update_time,
10546 static const struct inode_operations btrfs_special_inode_operations = {
10547 .getattr = btrfs_getattr,
10548 .setattr = btrfs_setattr,
10549 .permission = btrfs_permission,
10550 .listxattr = btrfs_listxattr,
10551 .get_acl = btrfs_get_acl,
10552 .set_acl = btrfs_set_acl,
10553 .update_time = btrfs_update_time,
10555 static const struct inode_operations btrfs_symlink_inode_operations = {
10556 .get_link = page_get_link,
10557 .getattr = btrfs_getattr,
10558 .setattr = btrfs_setattr,
10559 .permission = btrfs_permission,
10560 .listxattr = btrfs_listxattr,
10561 .update_time = btrfs_update_time,
10564 const struct dentry_operations btrfs_dentry_operations = {
10565 .d_delete = btrfs_dentry_delete,