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_extent_hook, the same
1629 * applies here, just in reverse.
1631 new_size = orig->end - split + 1;
1632 num_extents = count_max_extents(new_size);
1633 new_size = split - orig->start;
1634 num_extents += count_max_extents(new_size);
1635 if (count_max_extents(size) >= num_extents)
1639 spin_lock(&BTRFS_I(inode)->lock);
1640 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1641 spin_unlock(&BTRFS_I(inode)->lock);
1645 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1646 * extents so we can keep track of new extents that are just merged onto old
1647 * extents, such as when we are doing sequential writes, so we can properly
1648 * account for the metadata space we'll need.
1650 static void btrfs_merge_extent_hook(void *private_data,
1651 struct extent_state *new,
1652 struct extent_state *other)
1654 struct inode *inode = private_data;
1655 u64 new_size, old_size;
1658 /* not delalloc, ignore it */
1659 if (!(other->state & EXTENT_DELALLOC))
1662 if (new->start > other->start)
1663 new_size = new->end - other->start + 1;
1665 new_size = other->end - new->start + 1;
1667 /* we're not bigger than the max, unreserve the space and go */
1668 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1669 spin_lock(&BTRFS_I(inode)->lock);
1670 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1671 spin_unlock(&BTRFS_I(inode)->lock);
1676 * We have to add up either side to figure out how many extents were
1677 * accounted for before we merged into one big extent. If the number of
1678 * extents we accounted for is <= the amount we need for the new range
1679 * then we can return, otherwise drop. Think of it like this
1683 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1684 * need 2 outstanding extents, on one side we have 1 and the other side
1685 * we have 1 so they are == and we can return. But in this case
1687 * [MAX_SIZE+4k][MAX_SIZE+4k]
1689 * Each range on their own accounts for 2 extents, but merged together
1690 * they are only 3 extents worth of accounting, so we need to drop in
1693 old_size = other->end - other->start + 1;
1694 num_extents = count_max_extents(old_size);
1695 old_size = new->end - new->start + 1;
1696 num_extents += count_max_extents(old_size);
1697 if (count_max_extents(new_size) >= num_extents)
1700 spin_lock(&BTRFS_I(inode)->lock);
1701 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1702 spin_unlock(&BTRFS_I(inode)->lock);
1705 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1706 struct inode *inode)
1708 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1710 spin_lock(&root->delalloc_lock);
1711 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1712 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1713 &root->delalloc_inodes);
1714 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1715 &BTRFS_I(inode)->runtime_flags);
1716 root->nr_delalloc_inodes++;
1717 if (root->nr_delalloc_inodes == 1) {
1718 spin_lock(&fs_info->delalloc_root_lock);
1719 BUG_ON(!list_empty(&root->delalloc_root));
1720 list_add_tail(&root->delalloc_root,
1721 &fs_info->delalloc_roots);
1722 spin_unlock(&fs_info->delalloc_root_lock);
1725 spin_unlock(&root->delalloc_lock);
1729 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
1730 struct btrfs_inode *inode)
1732 struct btrfs_fs_info *fs_info = root->fs_info;
1734 if (!list_empty(&inode->delalloc_inodes)) {
1735 list_del_init(&inode->delalloc_inodes);
1736 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1737 &inode->runtime_flags);
1738 root->nr_delalloc_inodes--;
1739 if (!root->nr_delalloc_inodes) {
1740 ASSERT(list_empty(&root->delalloc_inodes));
1741 spin_lock(&fs_info->delalloc_root_lock);
1742 BUG_ON(list_empty(&root->delalloc_root));
1743 list_del_init(&root->delalloc_root);
1744 spin_unlock(&fs_info->delalloc_root_lock);
1749 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1750 struct btrfs_inode *inode)
1752 spin_lock(&root->delalloc_lock);
1753 __btrfs_del_delalloc_inode(root, inode);
1754 spin_unlock(&root->delalloc_lock);
1758 * Properly track delayed allocation bytes in the inode and to maintain the
1759 * list of inodes that have pending delalloc work to be done.
1761 void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state,
1764 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1766 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1769 * set_bit and clear bit hooks normally require _irqsave/restore
1770 * but in this case, we are only testing for the DELALLOC
1771 * bit, which is only set or cleared with irqs on
1773 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1774 struct btrfs_root *root = BTRFS_I(inode)->root;
1775 u64 len = state->end + 1 - state->start;
1776 u32 num_extents = count_max_extents(len);
1777 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
1779 spin_lock(&BTRFS_I(inode)->lock);
1780 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
1781 spin_unlock(&BTRFS_I(inode)->lock);
1783 /* For sanity tests */
1784 if (btrfs_is_testing(fs_info))
1787 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
1788 fs_info->delalloc_batch);
1789 spin_lock(&BTRFS_I(inode)->lock);
1790 BTRFS_I(inode)->delalloc_bytes += len;
1791 if (*bits & EXTENT_DEFRAG)
1792 BTRFS_I(inode)->defrag_bytes += len;
1793 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1794 &BTRFS_I(inode)->runtime_flags))
1795 btrfs_add_delalloc_inodes(root, inode);
1796 spin_unlock(&BTRFS_I(inode)->lock);
1799 if (!(state->state & EXTENT_DELALLOC_NEW) &&
1800 (*bits & EXTENT_DELALLOC_NEW)) {
1801 spin_lock(&BTRFS_I(inode)->lock);
1802 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
1804 spin_unlock(&BTRFS_I(inode)->lock);
1809 * Once a range is no longer delalloc this function ensures that proper
1810 * accounting happens.
1812 void btrfs_clear_delalloc_extent(struct inode *vfs_inode,
1813 struct extent_state *state, unsigned *bits)
1815 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
1816 struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb);
1817 u64 len = state->end + 1 - state->start;
1818 u32 num_extents = count_max_extents(len);
1820 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
1821 spin_lock(&inode->lock);
1822 inode->defrag_bytes -= len;
1823 spin_unlock(&inode->lock);
1827 * set_bit and clear bit hooks normally require _irqsave/restore
1828 * but in this case, we are only testing for the DELALLOC
1829 * bit, which is only set or cleared with irqs on
1831 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1832 struct btrfs_root *root = inode->root;
1833 bool do_list = !btrfs_is_free_space_inode(inode);
1835 spin_lock(&inode->lock);
1836 btrfs_mod_outstanding_extents(inode, -num_extents);
1837 spin_unlock(&inode->lock);
1840 * We don't reserve metadata space for space cache inodes so we
1841 * don't need to call dellalloc_release_metadata if there is an
1844 if (*bits & EXTENT_CLEAR_META_RESV &&
1845 root != fs_info->tree_root)
1846 btrfs_delalloc_release_metadata(inode, len, false);
1848 /* For sanity tests. */
1849 if (btrfs_is_testing(fs_info))
1852 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
1853 do_list && !(state->state & EXTENT_NORESERVE) &&
1854 (*bits & EXTENT_CLEAR_DATA_RESV))
1855 btrfs_free_reserved_data_space_noquota(
1859 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
1860 fs_info->delalloc_batch);
1861 spin_lock(&inode->lock);
1862 inode->delalloc_bytes -= len;
1863 if (do_list && inode->delalloc_bytes == 0 &&
1864 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1865 &inode->runtime_flags))
1866 btrfs_del_delalloc_inode(root, inode);
1867 spin_unlock(&inode->lock);
1870 if ((state->state & EXTENT_DELALLOC_NEW) &&
1871 (*bits & EXTENT_DELALLOC_NEW)) {
1872 spin_lock(&inode->lock);
1873 ASSERT(inode->new_delalloc_bytes >= len);
1874 inode->new_delalloc_bytes -= len;
1875 spin_unlock(&inode->lock);
1880 * Merge bio hook, this must check the chunk tree to make sure we don't create
1881 * bios that span stripes or chunks
1883 * return 1 if page cannot be merged to bio
1884 * return 0 if page can be merged to bio
1885 * return error otherwise
1887 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1888 size_t size, struct bio *bio,
1889 unsigned long bio_flags)
1891 struct inode *inode = page->mapping->host;
1892 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1893 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1898 if (bio_flags & EXTENT_BIO_COMPRESSED)
1901 length = bio->bi_iter.bi_size;
1902 map_length = length;
1903 ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
1907 if (map_length < length + size)
1913 * in order to insert checksums into the metadata in large chunks,
1914 * we wait until bio submission time. All the pages in the bio are
1915 * checksummed and sums are attached onto the ordered extent record.
1917 * At IO completion time the cums attached on the ordered extent record
1918 * are inserted into the btree
1920 static blk_status_t btrfs_submit_bio_start(void *private_data, struct bio *bio,
1923 struct inode *inode = private_data;
1924 blk_status_t ret = 0;
1926 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1927 BUG_ON(ret); /* -ENOMEM */
1932 * in order to insert checksums into the metadata in large chunks,
1933 * we wait until bio submission time. All the pages in the bio are
1934 * checksummed and sums are attached onto the ordered extent record.
1936 * At IO completion time the cums attached on the ordered extent record
1937 * are inserted into the btree
1939 blk_status_t btrfs_submit_bio_done(void *private_data, struct bio *bio,
1942 struct inode *inode = private_data;
1943 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1946 ret = btrfs_map_bio(fs_info, bio, mirror_num, 1);
1948 bio->bi_status = ret;
1955 * extent_io.c submission hook. This does the right thing for csum calculation
1956 * on write, or reading the csums from the tree before a read.
1958 * Rules about async/sync submit,
1959 * a) read: sync submit
1961 * b) write without checksum: sync submit
1963 * c) write with checksum:
1964 * c-1) if bio is issued by fsync: sync submit
1965 * (sync_writers != 0)
1967 * c-2) if root is reloc root: sync submit
1968 * (only in case of buffered IO)
1970 * c-3) otherwise: async submit
1972 static blk_status_t btrfs_submit_bio_hook(void *private_data, struct bio *bio,
1973 int mirror_num, unsigned long bio_flags,
1976 struct inode *inode = private_data;
1977 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1978 struct btrfs_root *root = BTRFS_I(inode)->root;
1979 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1980 blk_status_t ret = 0;
1982 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1984 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1986 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
1987 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1989 if (bio_op(bio) != REQ_OP_WRITE) {
1990 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
1994 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1995 ret = btrfs_submit_compressed_read(inode, bio,
1999 } else if (!skip_sum) {
2000 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
2005 } else if (async && !skip_sum) {
2006 /* csum items have already been cloned */
2007 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2009 /* we're doing a write, do the async checksumming */
2010 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags,
2012 btrfs_submit_bio_start);
2014 } else if (!skip_sum) {
2015 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
2021 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
2025 bio->bi_status = ret;
2032 * given a list of ordered sums record them in the inode. This happens
2033 * at IO completion time based on sums calculated at bio submission time.
2035 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
2036 struct inode *inode, struct list_head *list)
2038 struct btrfs_ordered_sum *sum;
2041 list_for_each_entry(sum, list, list) {
2042 trans->adding_csums = true;
2043 ret = btrfs_csum_file_blocks(trans,
2044 BTRFS_I(inode)->root->fs_info->csum_root, sum);
2045 trans->adding_csums = false;
2052 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
2053 unsigned int extra_bits,
2054 struct extent_state **cached_state, int dedupe)
2056 WARN_ON((end & (PAGE_SIZE - 1)) == 0);
2057 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2058 extra_bits, cached_state);
2061 /* see btrfs_writepage_start_hook for details on why this is required */
2062 struct btrfs_writepage_fixup {
2064 struct btrfs_work work;
2067 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2069 struct btrfs_writepage_fixup *fixup;
2070 struct btrfs_ordered_extent *ordered;
2071 struct extent_state *cached_state = NULL;
2072 struct extent_changeset *data_reserved = NULL;
2074 struct inode *inode;
2079 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2083 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2084 ClearPageChecked(page);
2088 inode = page->mapping->host;
2089 page_start = page_offset(page);
2090 page_end = page_offset(page) + PAGE_SIZE - 1;
2092 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2095 /* already ordered? We're done */
2096 if (PagePrivate2(page))
2099 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
2102 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2103 page_end, &cached_state);
2105 btrfs_start_ordered_extent(inode, ordered, 1);
2106 btrfs_put_ordered_extent(ordered);
2110 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2113 mapping_set_error(page->mapping, ret);
2114 end_extent_writepage(page, ret, page_start, page_end);
2115 ClearPageChecked(page);
2119 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2122 mapping_set_error(page->mapping, ret);
2123 end_extent_writepage(page, ret, page_start, page_end);
2124 ClearPageChecked(page);
2128 ClearPageChecked(page);
2129 set_page_dirty(page);
2130 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, false);
2132 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2138 extent_changeset_free(data_reserved);
2142 * There are a few paths in the higher layers of the kernel that directly
2143 * set the page dirty bit without asking the filesystem if it is a
2144 * good idea. This causes problems because we want to make sure COW
2145 * properly happens and the data=ordered rules are followed.
2147 * In our case any range that doesn't have the ORDERED bit set
2148 * hasn't been properly setup for IO. We kick off an async process
2149 * to fix it up. The async helper will wait for ordered extents, set
2150 * the delalloc bit and make it safe to write the page.
2152 int btrfs_writepage_cow_fixup(struct page *page, u64 start, u64 end)
2154 struct inode *inode = page->mapping->host;
2155 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2156 struct btrfs_writepage_fixup *fixup;
2158 /* this page is properly in the ordered list */
2159 if (TestClearPagePrivate2(page))
2162 if (PageChecked(page))
2165 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2169 SetPageChecked(page);
2171 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2172 btrfs_writepage_fixup_worker, NULL, NULL);
2174 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2178 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2179 struct inode *inode, u64 file_pos,
2180 u64 disk_bytenr, u64 disk_num_bytes,
2181 u64 num_bytes, u64 ram_bytes,
2182 u8 compression, u8 encryption,
2183 u16 other_encoding, int extent_type)
2185 struct btrfs_root *root = BTRFS_I(inode)->root;
2186 struct btrfs_file_extent_item *fi;
2187 struct btrfs_path *path;
2188 struct extent_buffer *leaf;
2189 struct btrfs_key ins;
2191 int extent_inserted = 0;
2194 path = btrfs_alloc_path();
2199 * we may be replacing one extent in the tree with another.
2200 * The new extent is pinned in the extent map, and we don't want
2201 * to drop it from the cache until it is completely in the btree.
2203 * So, tell btrfs_drop_extents to leave this extent in the cache.
2204 * the caller is expected to unpin it and allow it to be merged
2207 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2208 file_pos + num_bytes, NULL, 0,
2209 1, sizeof(*fi), &extent_inserted);
2213 if (!extent_inserted) {
2214 ins.objectid = btrfs_ino(BTRFS_I(inode));
2215 ins.offset = file_pos;
2216 ins.type = BTRFS_EXTENT_DATA_KEY;
2218 path->leave_spinning = 1;
2219 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2224 leaf = path->nodes[0];
2225 fi = btrfs_item_ptr(leaf, path->slots[0],
2226 struct btrfs_file_extent_item);
2227 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2228 btrfs_set_file_extent_type(leaf, fi, extent_type);
2229 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2230 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2231 btrfs_set_file_extent_offset(leaf, fi, 0);
2232 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2233 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2234 btrfs_set_file_extent_compression(leaf, fi, compression);
2235 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2236 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2238 btrfs_mark_buffer_dirty(leaf);
2239 btrfs_release_path(path);
2241 inode_add_bytes(inode, num_bytes);
2243 ins.objectid = disk_bytenr;
2244 ins.offset = disk_num_bytes;
2245 ins.type = BTRFS_EXTENT_ITEM_KEY;
2248 * Release the reserved range from inode dirty range map, as it is
2249 * already moved into delayed_ref_head
2251 ret = btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2255 ret = btrfs_alloc_reserved_file_extent(trans, root,
2256 btrfs_ino(BTRFS_I(inode)),
2257 file_pos, qg_released, &ins);
2259 btrfs_free_path(path);
2264 /* snapshot-aware defrag */
2265 struct sa_defrag_extent_backref {
2266 struct rb_node node;
2267 struct old_sa_defrag_extent *old;
2276 struct old_sa_defrag_extent {
2277 struct list_head list;
2278 struct new_sa_defrag_extent *new;
2287 struct new_sa_defrag_extent {
2288 struct rb_root root;
2289 struct list_head head;
2290 struct btrfs_path *path;
2291 struct inode *inode;
2299 static int backref_comp(struct sa_defrag_extent_backref *b1,
2300 struct sa_defrag_extent_backref *b2)
2302 if (b1->root_id < b2->root_id)
2304 else if (b1->root_id > b2->root_id)
2307 if (b1->inum < b2->inum)
2309 else if (b1->inum > b2->inum)
2312 if (b1->file_pos < b2->file_pos)
2314 else if (b1->file_pos > b2->file_pos)
2318 * [------------------------------] ===> (a range of space)
2319 * |<--->| |<---->| =============> (fs/file tree A)
2320 * |<---------------------------->| ===> (fs/file tree B)
2322 * A range of space can refer to two file extents in one tree while
2323 * refer to only one file extent in another tree.
2325 * So we may process a disk offset more than one time(two extents in A)
2326 * and locate at the same extent(one extent in B), then insert two same
2327 * backrefs(both refer to the extent in B).
2332 static void backref_insert(struct rb_root *root,
2333 struct sa_defrag_extent_backref *backref)
2335 struct rb_node **p = &root->rb_node;
2336 struct rb_node *parent = NULL;
2337 struct sa_defrag_extent_backref *entry;
2342 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2344 ret = backref_comp(backref, entry);
2348 p = &(*p)->rb_right;
2351 rb_link_node(&backref->node, parent, p);
2352 rb_insert_color(&backref->node, root);
2356 * Note the backref might has changed, and in this case we just return 0.
2358 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2361 struct btrfs_file_extent_item *extent;
2362 struct old_sa_defrag_extent *old = ctx;
2363 struct new_sa_defrag_extent *new = old->new;
2364 struct btrfs_path *path = new->path;
2365 struct btrfs_key key;
2366 struct btrfs_root *root;
2367 struct sa_defrag_extent_backref *backref;
2368 struct extent_buffer *leaf;
2369 struct inode *inode = new->inode;
2370 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2376 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2377 inum == btrfs_ino(BTRFS_I(inode)))
2380 key.objectid = root_id;
2381 key.type = BTRFS_ROOT_ITEM_KEY;
2382 key.offset = (u64)-1;
2384 root = btrfs_read_fs_root_no_name(fs_info, &key);
2386 if (PTR_ERR(root) == -ENOENT)
2389 btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu",
2390 inum, offset, root_id);
2391 return PTR_ERR(root);
2394 key.objectid = inum;
2395 key.type = BTRFS_EXTENT_DATA_KEY;
2396 if (offset > (u64)-1 << 32)
2399 key.offset = offset;
2401 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2402 if (WARN_ON(ret < 0))
2409 leaf = path->nodes[0];
2410 slot = path->slots[0];
2412 if (slot >= btrfs_header_nritems(leaf)) {
2413 ret = btrfs_next_leaf(root, path);
2416 } else if (ret > 0) {
2425 btrfs_item_key_to_cpu(leaf, &key, slot);
2427 if (key.objectid > inum)
2430 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2433 extent = btrfs_item_ptr(leaf, slot,
2434 struct btrfs_file_extent_item);
2436 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2440 * 'offset' refers to the exact key.offset,
2441 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2442 * (key.offset - extent_offset).
2444 if (key.offset != offset)
2447 extent_offset = btrfs_file_extent_offset(leaf, extent);
2448 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2450 if (extent_offset >= old->extent_offset + old->offset +
2451 old->len || extent_offset + num_bytes <=
2452 old->extent_offset + old->offset)
2457 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2463 backref->root_id = root_id;
2464 backref->inum = inum;
2465 backref->file_pos = offset;
2466 backref->num_bytes = num_bytes;
2467 backref->extent_offset = extent_offset;
2468 backref->generation = btrfs_file_extent_generation(leaf, extent);
2470 backref_insert(&new->root, backref);
2473 btrfs_release_path(path);
2478 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2479 struct new_sa_defrag_extent *new)
2481 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2482 struct old_sa_defrag_extent *old, *tmp;
2487 list_for_each_entry_safe(old, tmp, &new->head, list) {
2488 ret = iterate_inodes_from_logical(old->bytenr +
2489 old->extent_offset, fs_info,
2490 path, record_one_backref,
2492 if (ret < 0 && ret != -ENOENT)
2495 /* no backref to be processed for this extent */
2497 list_del(&old->list);
2502 if (list_empty(&new->head))
2508 static int relink_is_mergable(struct extent_buffer *leaf,
2509 struct btrfs_file_extent_item *fi,
2510 struct new_sa_defrag_extent *new)
2512 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2515 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2518 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2521 if (btrfs_file_extent_encryption(leaf, fi) ||
2522 btrfs_file_extent_other_encoding(leaf, fi))
2529 * Note the backref might has changed, and in this case we just return 0.
2531 static noinline int relink_extent_backref(struct btrfs_path *path,
2532 struct sa_defrag_extent_backref *prev,
2533 struct sa_defrag_extent_backref *backref)
2535 struct btrfs_file_extent_item *extent;
2536 struct btrfs_file_extent_item *item;
2537 struct btrfs_ordered_extent *ordered;
2538 struct btrfs_trans_handle *trans;
2539 struct btrfs_root *root;
2540 struct btrfs_key key;
2541 struct extent_buffer *leaf;
2542 struct old_sa_defrag_extent *old = backref->old;
2543 struct new_sa_defrag_extent *new = old->new;
2544 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2545 struct inode *inode;
2546 struct extent_state *cached = NULL;
2555 if (prev && prev->root_id == backref->root_id &&
2556 prev->inum == backref->inum &&
2557 prev->file_pos + prev->num_bytes == backref->file_pos)
2560 /* step 1: get root */
2561 key.objectid = backref->root_id;
2562 key.type = BTRFS_ROOT_ITEM_KEY;
2563 key.offset = (u64)-1;
2565 index = srcu_read_lock(&fs_info->subvol_srcu);
2567 root = btrfs_read_fs_root_no_name(fs_info, &key);
2569 srcu_read_unlock(&fs_info->subvol_srcu, index);
2570 if (PTR_ERR(root) == -ENOENT)
2572 return PTR_ERR(root);
2575 if (btrfs_root_readonly(root)) {
2576 srcu_read_unlock(&fs_info->subvol_srcu, index);
2580 /* step 2: get inode */
2581 key.objectid = backref->inum;
2582 key.type = BTRFS_INODE_ITEM_KEY;
2585 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2586 if (IS_ERR(inode)) {
2587 srcu_read_unlock(&fs_info->subvol_srcu, index);
2591 srcu_read_unlock(&fs_info->subvol_srcu, index);
2593 /* step 3: relink backref */
2594 lock_start = backref->file_pos;
2595 lock_end = backref->file_pos + backref->num_bytes - 1;
2596 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2599 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2601 btrfs_put_ordered_extent(ordered);
2605 trans = btrfs_join_transaction(root);
2606 if (IS_ERR(trans)) {
2607 ret = PTR_ERR(trans);
2611 key.objectid = backref->inum;
2612 key.type = BTRFS_EXTENT_DATA_KEY;
2613 key.offset = backref->file_pos;
2615 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2618 } else if (ret > 0) {
2623 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2624 struct btrfs_file_extent_item);
2626 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2627 backref->generation)
2630 btrfs_release_path(path);
2632 start = backref->file_pos;
2633 if (backref->extent_offset < old->extent_offset + old->offset)
2634 start += old->extent_offset + old->offset -
2635 backref->extent_offset;
2637 len = min(backref->extent_offset + backref->num_bytes,
2638 old->extent_offset + old->offset + old->len);
2639 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2641 ret = btrfs_drop_extents(trans, root, inode, start,
2646 key.objectid = btrfs_ino(BTRFS_I(inode));
2647 key.type = BTRFS_EXTENT_DATA_KEY;
2650 path->leave_spinning = 1;
2652 struct btrfs_file_extent_item *fi;
2654 struct btrfs_key found_key;
2656 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2661 leaf = path->nodes[0];
2662 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2664 fi = btrfs_item_ptr(leaf, path->slots[0],
2665 struct btrfs_file_extent_item);
2666 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2668 if (extent_len + found_key.offset == start &&
2669 relink_is_mergable(leaf, fi, new)) {
2670 btrfs_set_file_extent_num_bytes(leaf, fi,
2672 btrfs_mark_buffer_dirty(leaf);
2673 inode_add_bytes(inode, len);
2679 btrfs_release_path(path);
2684 ret = btrfs_insert_empty_item(trans, root, path, &key,
2687 btrfs_abort_transaction(trans, ret);
2691 leaf = path->nodes[0];
2692 item = btrfs_item_ptr(leaf, path->slots[0],
2693 struct btrfs_file_extent_item);
2694 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2695 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2696 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2697 btrfs_set_file_extent_num_bytes(leaf, item, len);
2698 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2699 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2700 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2701 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2702 btrfs_set_file_extent_encryption(leaf, item, 0);
2703 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2705 btrfs_mark_buffer_dirty(leaf);
2706 inode_add_bytes(inode, len);
2707 btrfs_release_path(path);
2709 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2711 backref->root_id, backref->inum,
2712 new->file_pos); /* start - extent_offset */
2714 btrfs_abort_transaction(trans, ret);
2720 btrfs_release_path(path);
2721 path->leave_spinning = 0;
2722 btrfs_end_transaction(trans);
2724 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2730 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2732 struct old_sa_defrag_extent *old, *tmp;
2737 list_for_each_entry_safe(old, tmp, &new->head, list) {
2743 static void relink_file_extents(struct new_sa_defrag_extent *new)
2745 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2746 struct btrfs_path *path;
2747 struct sa_defrag_extent_backref *backref;
2748 struct sa_defrag_extent_backref *prev = NULL;
2749 struct rb_node *node;
2752 path = btrfs_alloc_path();
2756 if (!record_extent_backrefs(path, new)) {
2757 btrfs_free_path(path);
2760 btrfs_release_path(path);
2763 node = rb_first(&new->root);
2766 rb_erase(node, &new->root);
2768 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2770 ret = relink_extent_backref(path, prev, backref);
2783 btrfs_free_path(path);
2785 free_sa_defrag_extent(new);
2787 atomic_dec(&fs_info->defrag_running);
2788 wake_up(&fs_info->transaction_wait);
2791 static struct new_sa_defrag_extent *
2792 record_old_file_extents(struct inode *inode,
2793 struct btrfs_ordered_extent *ordered)
2795 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2796 struct btrfs_root *root = BTRFS_I(inode)->root;
2797 struct btrfs_path *path;
2798 struct btrfs_key key;
2799 struct old_sa_defrag_extent *old;
2800 struct new_sa_defrag_extent *new;
2803 new = kmalloc(sizeof(*new), GFP_NOFS);
2808 new->file_pos = ordered->file_offset;
2809 new->len = ordered->len;
2810 new->bytenr = ordered->start;
2811 new->disk_len = ordered->disk_len;
2812 new->compress_type = ordered->compress_type;
2813 new->root = RB_ROOT;
2814 INIT_LIST_HEAD(&new->head);
2816 path = btrfs_alloc_path();
2820 key.objectid = btrfs_ino(BTRFS_I(inode));
2821 key.type = BTRFS_EXTENT_DATA_KEY;
2822 key.offset = new->file_pos;
2824 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2827 if (ret > 0 && path->slots[0] > 0)
2830 /* find out all the old extents for the file range */
2832 struct btrfs_file_extent_item *extent;
2833 struct extent_buffer *l;
2842 slot = path->slots[0];
2844 if (slot >= btrfs_header_nritems(l)) {
2845 ret = btrfs_next_leaf(root, path);
2853 btrfs_item_key_to_cpu(l, &key, slot);
2855 if (key.objectid != btrfs_ino(BTRFS_I(inode)))
2857 if (key.type != BTRFS_EXTENT_DATA_KEY)
2859 if (key.offset >= new->file_pos + new->len)
2862 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2864 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2865 if (key.offset + num_bytes < new->file_pos)
2868 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2872 extent_offset = btrfs_file_extent_offset(l, extent);
2874 old = kmalloc(sizeof(*old), GFP_NOFS);
2878 offset = max(new->file_pos, key.offset);
2879 end = min(new->file_pos + new->len, key.offset + num_bytes);
2881 old->bytenr = disk_bytenr;
2882 old->extent_offset = extent_offset;
2883 old->offset = offset - key.offset;
2884 old->len = end - offset;
2887 list_add_tail(&old->list, &new->head);
2893 btrfs_free_path(path);
2894 atomic_inc(&fs_info->defrag_running);
2899 btrfs_free_path(path);
2901 free_sa_defrag_extent(new);
2905 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2908 struct btrfs_block_group_cache *cache;
2910 cache = btrfs_lookup_block_group(fs_info, start);
2913 spin_lock(&cache->lock);
2914 cache->delalloc_bytes -= len;
2915 spin_unlock(&cache->lock);
2917 btrfs_put_block_group(cache);
2920 /* as ordered data IO finishes, this gets called so we can finish
2921 * an ordered extent if the range of bytes in the file it covers are
2924 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2926 struct inode *inode = ordered_extent->inode;
2927 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2928 struct btrfs_root *root = BTRFS_I(inode)->root;
2929 struct btrfs_trans_handle *trans = NULL;
2930 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2931 struct extent_state *cached_state = NULL;
2932 struct new_sa_defrag_extent *new = NULL;
2933 int compress_type = 0;
2935 u64 logical_len = ordered_extent->len;
2937 bool truncated = false;
2938 bool range_locked = false;
2939 bool clear_new_delalloc_bytes = false;
2940 bool clear_reserved_extent = true;
2942 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2943 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2944 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2945 clear_new_delalloc_bytes = true;
2947 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
2949 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2954 btrfs_free_io_failure_record(BTRFS_I(inode),
2955 ordered_extent->file_offset,
2956 ordered_extent->file_offset +
2957 ordered_extent->len - 1);
2959 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2961 logical_len = ordered_extent->truncated_len;
2962 /* Truncated the entire extent, don't bother adding */
2967 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2968 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2971 * For mwrite(mmap + memset to write) case, we still reserve
2972 * space for NOCOW range.
2973 * As NOCOW won't cause a new delayed ref, just free the space
2975 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
2976 ordered_extent->len);
2977 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2979 trans = btrfs_join_transaction_nolock(root);
2981 trans = btrfs_join_transaction(root);
2982 if (IS_ERR(trans)) {
2983 ret = PTR_ERR(trans);
2987 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
2988 ret = btrfs_update_inode_fallback(trans, root, inode);
2989 if (ret) /* -ENOMEM or corruption */
2990 btrfs_abort_transaction(trans, ret);
2994 range_locked = true;
2995 lock_extent_bits(io_tree, ordered_extent->file_offset,
2996 ordered_extent->file_offset + ordered_extent->len - 1,
2999 ret = test_range_bit(io_tree, ordered_extent->file_offset,
3000 ordered_extent->file_offset + ordered_extent->len - 1,
3001 EXTENT_DEFRAG, 0, cached_state);
3003 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
3004 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
3005 /* the inode is shared */
3006 new = record_old_file_extents(inode, ordered_extent);
3008 clear_extent_bit(io_tree, ordered_extent->file_offset,
3009 ordered_extent->file_offset + ordered_extent->len - 1,
3010 EXTENT_DEFRAG, 0, 0, &cached_state);
3014 trans = btrfs_join_transaction_nolock(root);
3016 trans = btrfs_join_transaction(root);
3017 if (IS_ERR(trans)) {
3018 ret = PTR_ERR(trans);
3023 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
3025 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3026 compress_type = ordered_extent->compress_type;
3027 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3028 BUG_ON(compress_type);
3029 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
3030 ordered_extent->len);
3031 ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
3032 ordered_extent->file_offset,
3033 ordered_extent->file_offset +
3036 BUG_ON(root == fs_info->tree_root);
3037 ret = insert_reserved_file_extent(trans, inode,
3038 ordered_extent->file_offset,
3039 ordered_extent->start,
3040 ordered_extent->disk_len,
3041 logical_len, logical_len,
3042 compress_type, 0, 0,
3043 BTRFS_FILE_EXTENT_REG);
3045 clear_reserved_extent = false;
3046 btrfs_release_delalloc_bytes(fs_info,
3047 ordered_extent->start,
3048 ordered_extent->disk_len);
3051 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
3052 ordered_extent->file_offset, ordered_extent->len,
3055 btrfs_abort_transaction(trans, ret);
3059 ret = add_pending_csums(trans, inode, &ordered_extent->list);
3061 btrfs_abort_transaction(trans, ret);
3065 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
3066 ret = btrfs_update_inode_fallback(trans, root, inode);
3067 if (ret) { /* -ENOMEM or corruption */
3068 btrfs_abort_transaction(trans, ret);
3073 if (range_locked || clear_new_delalloc_bytes) {
3074 unsigned int clear_bits = 0;
3077 clear_bits |= EXTENT_LOCKED;
3078 if (clear_new_delalloc_bytes)
3079 clear_bits |= EXTENT_DELALLOC_NEW;
3080 clear_extent_bit(&BTRFS_I(inode)->io_tree,
3081 ordered_extent->file_offset,
3082 ordered_extent->file_offset +
3083 ordered_extent->len - 1,
3085 (clear_bits & EXTENT_LOCKED) ? 1 : 0,
3090 btrfs_end_transaction(trans);
3092 if (ret || truncated) {
3096 start = ordered_extent->file_offset + logical_len;
3098 start = ordered_extent->file_offset;
3099 end = ordered_extent->file_offset + ordered_extent->len - 1;
3100 clear_extent_uptodate(io_tree, start, end, NULL);
3102 /* Drop the cache for the part of the extent we didn't write. */
3103 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
3106 * If the ordered extent had an IOERR or something else went
3107 * wrong we need to return the space for this ordered extent
3108 * back to the allocator. We only free the extent in the
3109 * truncated case if we didn't write out the extent at all.
3111 * If we made it past insert_reserved_file_extent before we
3112 * errored out then we don't need to do this as the accounting
3113 * has already been done.
3115 if ((ret || !logical_len) &&
3116 clear_reserved_extent &&
3117 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3118 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
3119 btrfs_free_reserved_extent(fs_info,
3120 ordered_extent->start,
3121 ordered_extent->disk_len, 1);
3126 * This needs to be done to make sure anybody waiting knows we are done
3127 * updating everything for this ordered extent.
3129 btrfs_remove_ordered_extent(inode, ordered_extent);
3131 /* for snapshot-aware defrag */
3134 free_sa_defrag_extent(new);
3135 atomic_dec(&fs_info->defrag_running);
3137 relink_file_extents(new);
3142 btrfs_put_ordered_extent(ordered_extent);
3143 /* once for the tree */
3144 btrfs_put_ordered_extent(ordered_extent);
3146 /* Try to release some metadata so we don't get an OOM but don't wait */
3147 btrfs_btree_balance_dirty_nodelay(fs_info);
3152 static void finish_ordered_fn(struct btrfs_work *work)
3154 struct btrfs_ordered_extent *ordered_extent;
3155 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3156 btrfs_finish_ordered_io(ordered_extent);
3159 void btrfs_writepage_endio_finish_ordered(struct page *page, u64 start, u64 end,
3160 struct extent_state *state, int uptodate)
3162 struct inode *inode = page->mapping->host;
3163 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3164 struct btrfs_ordered_extent *ordered_extent = NULL;
3165 struct btrfs_workqueue *wq;
3166 btrfs_work_func_t func;
3168 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3170 ClearPagePrivate2(page);
3171 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3172 end - start + 1, uptodate))
3175 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
3176 wq = fs_info->endio_freespace_worker;
3177 func = btrfs_freespace_write_helper;
3179 wq = fs_info->endio_write_workers;
3180 func = btrfs_endio_write_helper;
3183 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3185 btrfs_queue_work(wq, &ordered_extent->work);
3188 static int __readpage_endio_check(struct inode *inode,
3189 struct btrfs_io_bio *io_bio,
3190 int icsum, struct page *page,
3191 int pgoff, u64 start, size_t len)
3197 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3199 kaddr = kmap_atomic(page);
3200 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3201 btrfs_csum_final(csum, (u8 *)&csum);
3202 if (csum != csum_expected)
3205 kunmap_atomic(kaddr);
3208 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3209 io_bio->mirror_num);
3210 memset(kaddr + pgoff, 1, len);
3211 flush_dcache_page(page);
3212 kunmap_atomic(kaddr);
3217 * when reads are done, we need to check csums to verify the data is correct
3218 * if there's a match, we allow the bio to finish. If not, the code in
3219 * extent_io.c will try to find good copies for us.
3221 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3222 u64 phy_offset, struct page *page,
3223 u64 start, u64 end, int mirror)
3225 size_t offset = start - page_offset(page);
3226 struct inode *inode = page->mapping->host;
3227 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3228 struct btrfs_root *root = BTRFS_I(inode)->root;
3230 if (PageChecked(page)) {
3231 ClearPageChecked(page);
3235 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3238 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3239 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3240 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3244 phy_offset >>= inode->i_sb->s_blocksize_bits;
3245 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3246 start, (size_t)(end - start + 1));
3250 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3252 * @inode: The inode we want to perform iput on
3254 * This function uses the generic vfs_inode::i_count to track whether we should
3255 * just decrement it (in case it's > 1) or if this is the last iput then link
3256 * the inode to the delayed iput machinery. Delayed iputs are processed at
3257 * transaction commit time/superblock commit/cleaner kthread.
3259 void btrfs_add_delayed_iput(struct inode *inode)
3261 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3262 struct btrfs_inode *binode = BTRFS_I(inode);
3264 if (atomic_add_unless(&inode->i_count, -1, 1))
3267 spin_lock(&fs_info->delayed_iput_lock);
3268 ASSERT(list_empty(&binode->delayed_iput));
3269 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3270 spin_unlock(&fs_info->delayed_iput_lock);
3273 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3276 spin_lock(&fs_info->delayed_iput_lock);
3277 while (!list_empty(&fs_info->delayed_iputs)) {
3278 struct btrfs_inode *inode;
3280 inode = list_first_entry(&fs_info->delayed_iputs,
3281 struct btrfs_inode, delayed_iput);
3282 list_del_init(&inode->delayed_iput);
3283 spin_unlock(&fs_info->delayed_iput_lock);
3284 iput(&inode->vfs_inode);
3285 spin_lock(&fs_info->delayed_iput_lock);
3287 spin_unlock(&fs_info->delayed_iput_lock);
3291 * This creates an orphan entry for the given inode in case something goes wrong
3292 * in the middle of an unlink.
3294 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3295 struct btrfs_inode *inode)
3299 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3300 if (ret && ret != -EEXIST) {
3301 btrfs_abort_transaction(trans, ret);
3309 * We have done the delete so we can go ahead and remove the orphan item for
3310 * this particular inode.
3312 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3313 struct btrfs_inode *inode)
3315 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3319 * this cleans up any orphans that may be left on the list from the last use
3322 int btrfs_orphan_cleanup(struct btrfs_root *root)
3324 struct btrfs_fs_info *fs_info = root->fs_info;
3325 struct btrfs_path *path;
3326 struct extent_buffer *leaf;
3327 struct btrfs_key key, found_key;
3328 struct btrfs_trans_handle *trans;
3329 struct inode *inode;
3330 u64 last_objectid = 0;
3331 int ret = 0, nr_unlink = 0;
3333 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3336 path = btrfs_alloc_path();
3341 path->reada = READA_BACK;
3343 key.objectid = BTRFS_ORPHAN_OBJECTID;
3344 key.type = BTRFS_ORPHAN_ITEM_KEY;
3345 key.offset = (u64)-1;
3348 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3353 * if ret == 0 means we found what we were searching for, which
3354 * is weird, but possible, so only screw with path if we didn't
3355 * find the key and see if we have stuff that matches
3359 if (path->slots[0] == 0)
3364 /* pull out the item */
3365 leaf = path->nodes[0];
3366 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3368 /* make sure the item matches what we want */
3369 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3371 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3374 /* release the path since we're done with it */
3375 btrfs_release_path(path);
3378 * this is where we are basically btrfs_lookup, without the
3379 * crossing root thing. we store the inode number in the
3380 * offset of the orphan item.
3383 if (found_key.offset == last_objectid) {
3385 "Error removing orphan entry, stopping orphan cleanup");
3390 last_objectid = found_key.offset;
3392 found_key.objectid = found_key.offset;
3393 found_key.type = BTRFS_INODE_ITEM_KEY;
3394 found_key.offset = 0;
3395 inode = btrfs_iget(fs_info->sb, &found_key, root, NULL);
3396 ret = PTR_ERR_OR_ZERO(inode);
3397 if (ret && ret != -ENOENT)
3400 if (ret == -ENOENT && root == fs_info->tree_root) {
3401 struct btrfs_root *dead_root;
3402 struct btrfs_fs_info *fs_info = root->fs_info;
3403 int is_dead_root = 0;
3406 * this is an orphan in the tree root. Currently these
3407 * could come from 2 sources:
3408 * a) a snapshot deletion in progress
3409 * b) a free space cache inode
3410 * We need to distinguish those two, as the snapshot
3411 * orphan must not get deleted.
3412 * find_dead_roots already ran before us, so if this
3413 * is a snapshot deletion, we should find the root
3414 * in the dead_roots list
3416 spin_lock(&fs_info->trans_lock);
3417 list_for_each_entry(dead_root, &fs_info->dead_roots,
3419 if (dead_root->root_key.objectid ==
3420 found_key.objectid) {
3425 spin_unlock(&fs_info->trans_lock);
3427 /* prevent this orphan from being found again */
3428 key.offset = found_key.objectid - 1;
3435 * If we have an inode with links, there are a couple of
3436 * possibilities. Old kernels (before v3.12) used to create an
3437 * orphan item for truncate indicating that there were possibly
3438 * extent items past i_size that needed to be deleted. In v3.12,
3439 * truncate was changed to update i_size in sync with the extent
3440 * items, but the (useless) orphan item was still created. Since
3441 * v4.18, we don't create the orphan item for truncate at all.
3443 * So, this item could mean that we need to do a truncate, but
3444 * only if this filesystem was last used on a pre-v3.12 kernel
3445 * and was not cleanly unmounted. The odds of that are quite
3446 * slim, and it's a pain to do the truncate now, so just delete
3449 * It's also possible that this orphan item was supposed to be
3450 * deleted but wasn't. The inode number may have been reused,
3451 * but either way, we can delete the orphan item.
3453 if (ret == -ENOENT || inode->i_nlink) {
3456 trans = btrfs_start_transaction(root, 1);
3457 if (IS_ERR(trans)) {
3458 ret = PTR_ERR(trans);
3461 btrfs_debug(fs_info, "auto deleting %Lu",
3462 found_key.objectid);
3463 ret = btrfs_del_orphan_item(trans, root,
3464 found_key.objectid);
3465 btrfs_end_transaction(trans);
3473 /* this will do delete_inode and everything for us */
3476 /* release the path since we're done with it */
3477 btrfs_release_path(path);
3479 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3481 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3482 trans = btrfs_join_transaction(root);
3484 btrfs_end_transaction(trans);
3488 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3492 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3493 btrfs_free_path(path);
3498 * very simple check to peek ahead in the leaf looking for xattrs. If we
3499 * don't find any xattrs, we know there can't be any acls.
3501 * slot is the slot the inode is in, objectid is the objectid of the inode
3503 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3504 int slot, u64 objectid,
3505 int *first_xattr_slot)
3507 u32 nritems = btrfs_header_nritems(leaf);
3508 struct btrfs_key found_key;
3509 static u64 xattr_access = 0;
3510 static u64 xattr_default = 0;
3513 if (!xattr_access) {
3514 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3515 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3516 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3517 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3521 *first_xattr_slot = -1;
3522 while (slot < nritems) {
3523 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3525 /* we found a different objectid, there must not be acls */
3526 if (found_key.objectid != objectid)
3529 /* we found an xattr, assume we've got an acl */
3530 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3531 if (*first_xattr_slot == -1)
3532 *first_xattr_slot = slot;
3533 if (found_key.offset == xattr_access ||
3534 found_key.offset == xattr_default)
3539 * we found a key greater than an xattr key, there can't
3540 * be any acls later on
3542 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3549 * it goes inode, inode backrefs, xattrs, extents,
3550 * so if there are a ton of hard links to an inode there can
3551 * be a lot of backrefs. Don't waste time searching too hard,
3552 * this is just an optimization
3557 /* we hit the end of the leaf before we found an xattr or
3558 * something larger than an xattr. We have to assume the inode
3561 if (*first_xattr_slot == -1)
3562 *first_xattr_slot = slot;
3567 * read an inode from the btree into the in-memory inode
3569 static int btrfs_read_locked_inode(struct inode *inode,
3570 struct btrfs_path *in_path)
3572 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3573 struct btrfs_path *path = in_path;
3574 struct extent_buffer *leaf;
3575 struct btrfs_inode_item *inode_item;
3576 struct btrfs_root *root = BTRFS_I(inode)->root;
3577 struct btrfs_key location;
3582 bool filled = false;
3583 int first_xattr_slot;
3585 ret = btrfs_fill_inode(inode, &rdev);
3590 path = btrfs_alloc_path();
3595 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3597 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3599 if (path != in_path)
3600 btrfs_free_path(path);
3604 leaf = path->nodes[0];
3609 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3610 struct btrfs_inode_item);
3611 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3612 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3613 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3614 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3615 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3617 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3618 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3620 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3621 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3623 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3624 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3626 BTRFS_I(inode)->i_otime.tv_sec =
3627 btrfs_timespec_sec(leaf, &inode_item->otime);
3628 BTRFS_I(inode)->i_otime.tv_nsec =
3629 btrfs_timespec_nsec(leaf, &inode_item->otime);
3631 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3632 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3633 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3635 inode_set_iversion_queried(inode,
3636 btrfs_inode_sequence(leaf, inode_item));
3637 inode->i_generation = BTRFS_I(inode)->generation;
3639 rdev = btrfs_inode_rdev(leaf, inode_item);
3641 BTRFS_I(inode)->index_cnt = (u64)-1;
3642 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3646 * If we were modified in the current generation and evicted from memory
3647 * and then re-read we need to do a full sync since we don't have any
3648 * idea about which extents were modified before we were evicted from
3651 * This is required for both inode re-read from disk and delayed inode
3652 * in delayed_nodes_tree.
3654 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3655 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3656 &BTRFS_I(inode)->runtime_flags);
3659 * We don't persist the id of the transaction where an unlink operation
3660 * against the inode was last made. So here we assume the inode might
3661 * have been evicted, and therefore the exact value of last_unlink_trans
3662 * lost, and set it to last_trans to avoid metadata inconsistencies
3663 * between the inode and its parent if the inode is fsync'ed and the log
3664 * replayed. For example, in the scenario:
3667 * ln mydir/foo mydir/bar
3670 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3671 * xfs_io -c fsync mydir/foo
3673 * mount fs, triggers fsync log replay
3675 * We must make sure that when we fsync our inode foo we also log its
3676 * parent inode, otherwise after log replay the parent still has the
3677 * dentry with the "bar" name but our inode foo has a link count of 1
3678 * and doesn't have an inode ref with the name "bar" anymore.
3680 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3681 * but it guarantees correctness at the expense of occasional full
3682 * transaction commits on fsync if our inode is a directory, or if our
3683 * inode is not a directory, logging its parent unnecessarily.
3685 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3688 if (inode->i_nlink != 1 ||
3689 path->slots[0] >= btrfs_header_nritems(leaf))
3692 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3693 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3696 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3697 if (location.type == BTRFS_INODE_REF_KEY) {
3698 struct btrfs_inode_ref *ref;
3700 ref = (struct btrfs_inode_ref *)ptr;
3701 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3702 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3703 struct btrfs_inode_extref *extref;
3705 extref = (struct btrfs_inode_extref *)ptr;
3706 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3711 * try to precache a NULL acl entry for files that don't have
3712 * any xattrs or acls
3714 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3715 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3716 if (first_xattr_slot != -1) {
3717 path->slots[0] = first_xattr_slot;
3718 ret = btrfs_load_inode_props(inode, path);
3721 "error loading props for ino %llu (root %llu): %d",
3722 btrfs_ino(BTRFS_I(inode)),
3723 root->root_key.objectid, ret);
3725 if (path != in_path)
3726 btrfs_free_path(path);
3729 cache_no_acl(inode);
3731 switch (inode->i_mode & S_IFMT) {
3733 inode->i_mapping->a_ops = &btrfs_aops;
3734 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3735 inode->i_fop = &btrfs_file_operations;
3736 inode->i_op = &btrfs_file_inode_operations;
3739 inode->i_fop = &btrfs_dir_file_operations;
3740 inode->i_op = &btrfs_dir_inode_operations;
3743 inode->i_op = &btrfs_symlink_inode_operations;
3744 inode_nohighmem(inode);
3745 inode->i_mapping->a_ops = &btrfs_aops;
3748 inode->i_op = &btrfs_special_inode_operations;
3749 init_special_inode(inode, inode->i_mode, rdev);
3753 btrfs_sync_inode_flags_to_i_flags(inode);
3758 * given a leaf and an inode, copy the inode fields into the leaf
3760 static void fill_inode_item(struct btrfs_trans_handle *trans,
3761 struct extent_buffer *leaf,
3762 struct btrfs_inode_item *item,
3763 struct inode *inode)
3765 struct btrfs_map_token token;
3767 btrfs_init_map_token(&token);
3769 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3770 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3771 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3773 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3774 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3776 btrfs_set_token_timespec_sec(leaf, &item->atime,
3777 inode->i_atime.tv_sec, &token);
3778 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3779 inode->i_atime.tv_nsec, &token);
3781 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3782 inode->i_mtime.tv_sec, &token);
3783 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3784 inode->i_mtime.tv_nsec, &token);
3786 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3787 inode->i_ctime.tv_sec, &token);
3788 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3789 inode->i_ctime.tv_nsec, &token);
3791 btrfs_set_token_timespec_sec(leaf, &item->otime,
3792 BTRFS_I(inode)->i_otime.tv_sec, &token);
3793 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3794 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3796 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3798 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3800 btrfs_set_token_inode_sequence(leaf, item, inode_peek_iversion(inode),
3802 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3803 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3804 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3805 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3809 * copy everything in the in-memory inode into the btree.
3811 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3812 struct btrfs_root *root, struct inode *inode)
3814 struct btrfs_inode_item *inode_item;
3815 struct btrfs_path *path;
3816 struct extent_buffer *leaf;
3819 path = btrfs_alloc_path();
3823 path->leave_spinning = 1;
3824 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3832 leaf = path->nodes[0];
3833 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3834 struct btrfs_inode_item);
3836 fill_inode_item(trans, leaf, inode_item, inode);
3837 btrfs_mark_buffer_dirty(leaf);
3838 btrfs_set_inode_last_trans(trans, inode);
3841 btrfs_free_path(path);
3846 * copy everything in the in-memory inode into the btree.
3848 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3849 struct btrfs_root *root, struct inode *inode)
3851 struct btrfs_fs_info *fs_info = root->fs_info;
3855 * If the inode is a free space inode, we can deadlock during commit
3856 * if we put it into the delayed code.
3858 * The data relocation inode should also be directly updated
3861 if (!btrfs_is_free_space_inode(BTRFS_I(inode))
3862 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3863 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
3864 btrfs_update_root_times(trans, root);
3866 ret = btrfs_delayed_update_inode(trans, root, inode);
3868 btrfs_set_inode_last_trans(trans, inode);
3872 return btrfs_update_inode_item(trans, root, inode);
3875 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3876 struct btrfs_root *root,
3877 struct inode *inode)
3881 ret = btrfs_update_inode(trans, root, inode);
3883 return btrfs_update_inode_item(trans, root, inode);
3888 * unlink helper that gets used here in inode.c and in the tree logging
3889 * recovery code. It remove a link in a directory with a given name, and
3890 * also drops the back refs in the inode to the directory
3892 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3893 struct btrfs_root *root,
3894 struct btrfs_inode *dir,
3895 struct btrfs_inode *inode,
3896 const char *name, int name_len)
3898 struct btrfs_fs_info *fs_info = root->fs_info;
3899 struct btrfs_path *path;
3901 struct extent_buffer *leaf;
3902 struct btrfs_dir_item *di;
3903 struct btrfs_key key;
3905 u64 ino = btrfs_ino(inode);
3906 u64 dir_ino = btrfs_ino(dir);
3908 path = btrfs_alloc_path();
3914 path->leave_spinning = 1;
3915 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3916 name, name_len, -1);
3917 if (IS_ERR_OR_NULL(di)) {
3918 ret = di ? PTR_ERR(di) : -ENOENT;
3921 leaf = path->nodes[0];
3922 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3923 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3926 btrfs_release_path(path);
3929 * If we don't have dir index, we have to get it by looking up
3930 * the inode ref, since we get the inode ref, remove it directly,
3931 * it is unnecessary to do delayed deletion.
3933 * But if we have dir index, needn't search inode ref to get it.
3934 * Since the inode ref is close to the inode item, it is better
3935 * that we delay to delete it, and just do this deletion when
3936 * we update the inode item.
3938 if (inode->dir_index) {
3939 ret = btrfs_delayed_delete_inode_ref(inode);
3941 index = inode->dir_index;
3946 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3950 "failed to delete reference to %.*s, inode %llu parent %llu",
3951 name_len, name, ino, dir_ino);
3952 btrfs_abort_transaction(trans, ret);
3956 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
3958 btrfs_abort_transaction(trans, ret);
3962 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
3964 if (ret != 0 && ret != -ENOENT) {
3965 btrfs_abort_transaction(trans, ret);
3969 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
3974 btrfs_abort_transaction(trans, ret);
3976 btrfs_free_path(path);
3980 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
3981 inode_inc_iversion(&inode->vfs_inode);
3982 inode_inc_iversion(&dir->vfs_inode);
3983 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
3984 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
3985 ret = btrfs_update_inode(trans, root, &dir->vfs_inode);
3990 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3991 struct btrfs_root *root,
3992 struct btrfs_inode *dir, struct btrfs_inode *inode,
3993 const char *name, int name_len)
3996 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3998 drop_nlink(&inode->vfs_inode);
3999 ret = btrfs_update_inode(trans, root, &inode->vfs_inode);
4005 * helper to start transaction for unlink and rmdir.
4007 * unlink and rmdir are special in btrfs, they do not always free space, so
4008 * if we cannot make our reservations the normal way try and see if there is
4009 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4010 * allow the unlink to occur.
4012 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4014 struct btrfs_root *root = BTRFS_I(dir)->root;
4017 * 1 for the possible orphan item
4018 * 1 for the dir item
4019 * 1 for the dir index
4020 * 1 for the inode ref
4023 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4026 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4028 struct btrfs_root *root = BTRFS_I(dir)->root;
4029 struct btrfs_trans_handle *trans;
4030 struct inode *inode = d_inode(dentry);
4033 trans = __unlink_start_trans(dir);
4035 return PTR_ERR(trans);
4037 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4040 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4041 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4042 dentry->d_name.len);
4046 if (inode->i_nlink == 0) {
4047 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4053 btrfs_end_transaction(trans);
4054 btrfs_btree_balance_dirty(root->fs_info);
4058 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4059 struct inode *dir, u64 objectid,
4060 const char *name, int name_len)
4062 struct btrfs_root *root = BTRFS_I(dir)->root;
4063 struct btrfs_path *path;
4064 struct extent_buffer *leaf;
4065 struct btrfs_dir_item *di;
4066 struct btrfs_key key;
4069 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4071 path = btrfs_alloc_path();
4075 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4076 name, name_len, -1);
4077 if (IS_ERR_OR_NULL(di)) {
4078 ret = di ? PTR_ERR(di) : -ENOENT;
4082 leaf = path->nodes[0];
4083 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4084 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4085 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4087 btrfs_abort_transaction(trans, ret);
4090 btrfs_release_path(path);
4092 ret = btrfs_del_root_ref(trans, objectid, root->root_key.objectid,
4093 dir_ino, &index, name, name_len);
4095 if (ret != -ENOENT) {
4096 btrfs_abort_transaction(trans, ret);
4099 di = btrfs_search_dir_index_item(root, path, dir_ino,
4101 if (IS_ERR_OR_NULL(di)) {
4106 btrfs_abort_transaction(trans, ret);
4110 leaf = path->nodes[0];
4111 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4114 btrfs_release_path(path);
4116 ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index);
4118 btrfs_abort_transaction(trans, ret);
4122 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4123 inode_inc_iversion(dir);
4124 dir->i_mtime = dir->i_ctime = current_time(dir);
4125 ret = btrfs_update_inode_fallback(trans, root, dir);
4127 btrfs_abort_transaction(trans, ret);
4129 btrfs_free_path(path);
4134 * Helper to check if the subvolume references other subvolumes or if it's
4137 static noinline int may_destroy_subvol(struct btrfs_root *root)
4139 struct btrfs_fs_info *fs_info = root->fs_info;
4140 struct btrfs_path *path;
4141 struct btrfs_dir_item *di;
4142 struct btrfs_key key;
4146 path = btrfs_alloc_path();
4150 /* Make sure this root isn't set as the default subvol */
4151 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4152 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4153 dir_id, "default", 7, 0);
4154 if (di && !IS_ERR(di)) {
4155 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4156 if (key.objectid == root->root_key.objectid) {
4159 "deleting default subvolume %llu is not allowed",
4163 btrfs_release_path(path);
4166 key.objectid = root->root_key.objectid;
4167 key.type = BTRFS_ROOT_REF_KEY;
4168 key.offset = (u64)-1;
4170 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4176 if (path->slots[0] > 0) {
4178 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4179 if (key.objectid == root->root_key.objectid &&
4180 key.type == BTRFS_ROOT_REF_KEY)
4184 btrfs_free_path(path);
4188 /* Delete all dentries for inodes belonging to the root */
4189 static void btrfs_prune_dentries(struct btrfs_root *root)
4191 struct btrfs_fs_info *fs_info = root->fs_info;
4192 struct rb_node *node;
4193 struct rb_node *prev;
4194 struct btrfs_inode *entry;
4195 struct inode *inode;
4198 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
4199 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4201 spin_lock(&root->inode_lock);
4203 node = root->inode_tree.rb_node;
4207 entry = rb_entry(node, struct btrfs_inode, rb_node);
4209 if (objectid < btrfs_ino(entry))
4210 node = node->rb_left;
4211 else if (objectid > btrfs_ino(entry))
4212 node = node->rb_right;
4218 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4219 if (objectid <= btrfs_ino(entry)) {
4223 prev = rb_next(prev);
4227 entry = rb_entry(node, struct btrfs_inode, rb_node);
4228 objectid = btrfs_ino(entry) + 1;
4229 inode = igrab(&entry->vfs_inode);
4231 spin_unlock(&root->inode_lock);
4232 if (atomic_read(&inode->i_count) > 1)
4233 d_prune_aliases(inode);
4235 * btrfs_drop_inode will have it removed from the inode
4236 * cache when its usage count hits zero.
4240 spin_lock(&root->inode_lock);
4244 if (cond_resched_lock(&root->inode_lock))
4247 node = rb_next(node);
4249 spin_unlock(&root->inode_lock);
4252 int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
4254 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4255 struct btrfs_root *root = BTRFS_I(dir)->root;
4256 struct inode *inode = d_inode(dentry);
4257 struct btrfs_root *dest = BTRFS_I(inode)->root;
4258 struct btrfs_trans_handle *trans;
4259 struct btrfs_block_rsv block_rsv;
4265 * Don't allow to delete a subvolume with send in progress. This is
4266 * inside the inode lock so the error handling that has to drop the bit
4267 * again is not run concurrently.
4269 spin_lock(&dest->root_item_lock);
4270 if (dest->send_in_progress) {
4271 spin_unlock(&dest->root_item_lock);
4273 "attempt to delete subvolume %llu during send",
4274 dest->root_key.objectid);
4277 root_flags = btrfs_root_flags(&dest->root_item);
4278 btrfs_set_root_flags(&dest->root_item,
4279 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4280 spin_unlock(&dest->root_item_lock);
4282 down_write(&fs_info->subvol_sem);
4284 err = may_destroy_subvol(dest);
4288 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4290 * One for dir inode,
4291 * two for dir entries,
4292 * two for root ref/backref.
4294 err = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4298 trans = btrfs_start_transaction(root, 0);
4299 if (IS_ERR(trans)) {
4300 err = PTR_ERR(trans);
4303 trans->block_rsv = &block_rsv;
4304 trans->bytes_reserved = block_rsv.size;
4306 btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
4308 ret = btrfs_unlink_subvol(trans, dir, dest->root_key.objectid,
4309 dentry->d_name.name, dentry->d_name.len);
4312 btrfs_abort_transaction(trans, ret);
4316 btrfs_record_root_in_trans(trans, dest);
4318 memset(&dest->root_item.drop_progress, 0,
4319 sizeof(dest->root_item.drop_progress));
4320 dest->root_item.drop_level = 0;
4321 btrfs_set_root_refs(&dest->root_item, 0);
4323 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4324 ret = btrfs_insert_orphan_item(trans,
4326 dest->root_key.objectid);
4328 btrfs_abort_transaction(trans, ret);
4334 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4335 BTRFS_UUID_KEY_SUBVOL,
4336 dest->root_key.objectid);
4337 if (ret && ret != -ENOENT) {
4338 btrfs_abort_transaction(trans, ret);
4342 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4343 ret = btrfs_uuid_tree_remove(trans,
4344 dest->root_item.received_uuid,
4345 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4346 dest->root_key.objectid);
4347 if (ret && ret != -ENOENT) {
4348 btrfs_abort_transaction(trans, ret);
4355 trans->block_rsv = NULL;
4356 trans->bytes_reserved = 0;
4357 ret = btrfs_end_transaction(trans);
4360 inode->i_flags |= S_DEAD;
4362 btrfs_subvolume_release_metadata(fs_info, &block_rsv);
4364 up_write(&fs_info->subvol_sem);
4366 spin_lock(&dest->root_item_lock);
4367 root_flags = btrfs_root_flags(&dest->root_item);
4368 btrfs_set_root_flags(&dest->root_item,
4369 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4370 spin_unlock(&dest->root_item_lock);
4372 d_invalidate(dentry);
4373 btrfs_prune_dentries(dest);
4374 ASSERT(dest->send_in_progress == 0);
4377 if (dest->ino_cache_inode) {
4378 iput(dest->ino_cache_inode);
4379 dest->ino_cache_inode = NULL;
4386 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4388 struct inode *inode = d_inode(dentry);
4390 struct btrfs_root *root = BTRFS_I(dir)->root;
4391 struct btrfs_trans_handle *trans;
4392 u64 last_unlink_trans;
4394 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4396 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4397 return btrfs_delete_subvolume(dir, dentry);
4399 trans = __unlink_start_trans(dir);
4401 return PTR_ERR(trans);
4403 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4404 err = btrfs_unlink_subvol(trans, dir,
4405 BTRFS_I(inode)->location.objectid,
4406 dentry->d_name.name,
4407 dentry->d_name.len);
4411 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4415 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4417 /* now the directory is empty */
4418 err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4419 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4420 dentry->d_name.len);
4422 btrfs_i_size_write(BTRFS_I(inode), 0);
4424 * Propagate the last_unlink_trans value of the deleted dir to
4425 * its parent directory. This is to prevent an unrecoverable
4426 * log tree in the case we do something like this:
4428 * 2) create snapshot under dir foo
4429 * 3) delete the snapshot
4432 * 6) fsync foo or some file inside foo
4434 if (last_unlink_trans >= trans->transid)
4435 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4438 btrfs_end_transaction(trans);
4439 btrfs_btree_balance_dirty(root->fs_info);
4444 static int truncate_space_check(struct btrfs_trans_handle *trans,
4445 struct btrfs_root *root,
4448 struct btrfs_fs_info *fs_info = root->fs_info;
4452 * This is only used to apply pressure to the enospc system, we don't
4453 * intend to use this reservation at all.
4455 bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted);
4456 bytes_deleted *= fs_info->nodesize;
4457 ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
4458 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4460 trace_btrfs_space_reservation(fs_info, "transaction",
4463 trans->bytes_reserved += bytes_deleted;
4470 * Return this if we need to call truncate_block for the last bit of the
4473 #define NEED_TRUNCATE_BLOCK 1
4476 * this can truncate away extent items, csum items and directory items.
4477 * It starts at a high offset and removes keys until it can't find
4478 * any higher than new_size
4480 * csum items that cross the new i_size are truncated to the new size
4483 * min_type is the minimum key type to truncate down to. If set to 0, this
4484 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4486 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4487 struct btrfs_root *root,
4488 struct inode *inode,
4489 u64 new_size, u32 min_type)
4491 struct btrfs_fs_info *fs_info = root->fs_info;
4492 struct btrfs_path *path;
4493 struct extent_buffer *leaf;
4494 struct btrfs_file_extent_item *fi;
4495 struct btrfs_key key;
4496 struct btrfs_key found_key;
4497 u64 extent_start = 0;
4498 u64 extent_num_bytes = 0;
4499 u64 extent_offset = 0;
4501 u64 last_size = new_size;
4502 u32 found_type = (u8)-1;
4505 int pending_del_nr = 0;
4506 int pending_del_slot = 0;
4507 int extent_type = -1;
4509 u64 ino = btrfs_ino(BTRFS_I(inode));
4510 u64 bytes_deleted = 0;
4511 bool be_nice = false;
4512 bool should_throttle = false;
4513 bool should_end = false;
4515 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4518 * for non-free space inodes and ref cows, we want to back off from
4521 if (!btrfs_is_free_space_inode(BTRFS_I(inode)) &&
4522 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4525 path = btrfs_alloc_path();
4528 path->reada = READA_BACK;
4531 * We want to drop from the next block forward in case this new size is
4532 * not block aligned since we will be keeping the last block of the
4533 * extent just the way it is.
4535 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4536 root == fs_info->tree_root)
4537 btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size,
4538 fs_info->sectorsize),
4542 * This function is also used to drop the items in the log tree before
4543 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4544 * it is used to drop the loged items. So we shouldn't kill the delayed
4547 if (min_type == 0 && root == BTRFS_I(inode)->root)
4548 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
4551 key.offset = (u64)-1;
4556 * with a 16K leaf size and 128MB extents, you can actually queue
4557 * up a huge file in a single leaf. Most of the time that
4558 * bytes_deleted is > 0, it will be huge by the time we get here
4560 if (be_nice && bytes_deleted > SZ_32M &&
4561 btrfs_should_end_transaction(trans)) {
4566 path->leave_spinning = 1;
4567 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4573 /* there are no items in the tree for us to truncate, we're
4576 if (path->slots[0] == 0)
4583 leaf = path->nodes[0];
4584 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4585 found_type = found_key.type;
4587 if (found_key.objectid != ino)
4590 if (found_type < min_type)
4593 item_end = found_key.offset;
4594 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4595 fi = btrfs_item_ptr(leaf, path->slots[0],
4596 struct btrfs_file_extent_item);
4597 extent_type = btrfs_file_extent_type(leaf, fi);
4598 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4600 btrfs_file_extent_num_bytes(leaf, fi);
4602 trace_btrfs_truncate_show_fi_regular(
4603 BTRFS_I(inode), leaf, fi,
4605 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4606 item_end += btrfs_file_extent_ram_bytes(leaf,
4609 trace_btrfs_truncate_show_fi_inline(
4610 BTRFS_I(inode), leaf, fi, path->slots[0],
4615 if (found_type > min_type) {
4618 if (item_end < new_size)
4620 if (found_key.offset >= new_size)
4626 /* FIXME, shrink the extent if the ref count is only 1 */
4627 if (found_type != BTRFS_EXTENT_DATA_KEY)
4630 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4632 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4634 u64 orig_num_bytes =
4635 btrfs_file_extent_num_bytes(leaf, fi);
4636 extent_num_bytes = ALIGN(new_size -
4638 fs_info->sectorsize);
4639 btrfs_set_file_extent_num_bytes(leaf, fi,
4641 num_dec = (orig_num_bytes -
4643 if (test_bit(BTRFS_ROOT_REF_COWS,
4646 inode_sub_bytes(inode, num_dec);
4647 btrfs_mark_buffer_dirty(leaf);
4650 btrfs_file_extent_disk_num_bytes(leaf,
4652 extent_offset = found_key.offset -
4653 btrfs_file_extent_offset(leaf, fi);
4655 /* FIXME blocksize != 4096 */
4656 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4657 if (extent_start != 0) {
4659 if (test_bit(BTRFS_ROOT_REF_COWS,
4661 inode_sub_bytes(inode, num_dec);
4664 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4666 * we can't truncate inline items that have had
4670 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4671 btrfs_file_extent_other_encoding(leaf, fi) == 0 &&
4672 btrfs_file_extent_compression(leaf, fi) == 0) {
4673 u32 size = (u32)(new_size - found_key.offset);
4675 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4676 size = btrfs_file_extent_calc_inline_size(size);
4677 btrfs_truncate_item(root->fs_info, path, size, 1);
4678 } else if (!del_item) {
4680 * We have to bail so the last_size is set to
4681 * just before this extent.
4683 ret = NEED_TRUNCATE_BLOCK;
4687 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4688 inode_sub_bytes(inode, item_end + 1 - new_size);
4692 last_size = found_key.offset;
4694 last_size = new_size;
4696 if (!pending_del_nr) {
4697 /* no pending yet, add ourselves */
4698 pending_del_slot = path->slots[0];
4700 } else if (pending_del_nr &&
4701 path->slots[0] + 1 == pending_del_slot) {
4702 /* hop on the pending chunk */
4704 pending_del_slot = path->slots[0];
4711 should_throttle = false;
4714 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4715 root == fs_info->tree_root)) {
4716 btrfs_set_path_blocking(path);
4717 bytes_deleted += extent_num_bytes;
4718 ret = btrfs_free_extent(trans, root, extent_start,
4719 extent_num_bytes, 0,
4720 btrfs_header_owner(leaf),
4721 ino, extent_offset);
4723 btrfs_abort_transaction(trans, ret);
4726 if (btrfs_should_throttle_delayed_refs(trans))
4727 btrfs_async_run_delayed_refs(fs_info,
4728 trans->delayed_ref_updates * 2,
4731 if (truncate_space_check(trans, root,
4732 extent_num_bytes)) {
4735 if (btrfs_should_throttle_delayed_refs(trans))
4736 should_throttle = true;
4740 if (found_type == BTRFS_INODE_ITEM_KEY)
4743 if (path->slots[0] == 0 ||
4744 path->slots[0] != pending_del_slot ||
4745 should_throttle || should_end) {
4746 if (pending_del_nr) {
4747 ret = btrfs_del_items(trans, root, path,
4751 btrfs_abort_transaction(trans, ret);
4756 btrfs_release_path(path);
4757 if (should_throttle) {
4758 unsigned long updates = trans->delayed_ref_updates;
4760 trans->delayed_ref_updates = 0;
4761 ret = btrfs_run_delayed_refs(trans,
4768 * if we failed to refill our space rsv, bail out
4769 * and let the transaction restart
4781 if (ret >= 0 && pending_del_nr) {
4784 err = btrfs_del_items(trans, root, path, pending_del_slot,
4787 btrfs_abort_transaction(trans, err);
4791 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4792 ASSERT(last_size >= new_size);
4793 if (!ret && last_size > new_size)
4794 last_size = new_size;
4795 btrfs_ordered_update_i_size(inode, last_size, NULL);
4798 btrfs_free_path(path);
4800 if (be_nice && bytes_deleted > SZ_32M && (ret >= 0 || ret == -EAGAIN)) {
4801 unsigned long updates = trans->delayed_ref_updates;
4805 trans->delayed_ref_updates = 0;
4806 err = btrfs_run_delayed_refs(trans, updates * 2);
4815 * btrfs_truncate_block - read, zero a chunk and write a block
4816 * @inode - inode that we're zeroing
4817 * @from - the offset to start zeroing
4818 * @len - the length to zero, 0 to zero the entire range respective to the
4820 * @front - zero up to the offset instead of from the offset on
4822 * This will find the block for the "from" offset and cow the block and zero the
4823 * part we want to zero. This is used with truncate and hole punching.
4825 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4828 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4829 struct address_space *mapping = inode->i_mapping;
4830 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4831 struct btrfs_ordered_extent *ordered;
4832 struct extent_state *cached_state = NULL;
4833 struct extent_changeset *data_reserved = NULL;
4835 u32 blocksize = fs_info->sectorsize;
4836 pgoff_t index = from >> PAGE_SHIFT;
4837 unsigned offset = from & (blocksize - 1);
4839 gfp_t mask = btrfs_alloc_write_mask(mapping);
4844 if (IS_ALIGNED(offset, blocksize) &&
4845 (!len || IS_ALIGNED(len, blocksize)))
4848 block_start = round_down(from, blocksize);
4849 block_end = block_start + blocksize - 1;
4851 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
4852 block_start, blocksize);
4857 page = find_or_create_page(mapping, index, mask);
4859 btrfs_delalloc_release_space(inode, data_reserved,
4860 block_start, blocksize, true);
4861 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize, true);
4866 if (!PageUptodate(page)) {
4867 ret = btrfs_readpage(NULL, page);
4869 if (page->mapping != mapping) {
4874 if (!PageUptodate(page)) {
4879 wait_on_page_writeback(page);
4881 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4882 set_page_extent_mapped(page);
4884 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4886 unlock_extent_cached(io_tree, block_start, block_end,
4890 btrfs_start_ordered_extent(inode, ordered, 1);
4891 btrfs_put_ordered_extent(ordered);
4895 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4896 EXTENT_DIRTY | EXTENT_DELALLOC |
4897 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4898 0, 0, &cached_state);
4900 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4903 unlock_extent_cached(io_tree, block_start, block_end,
4908 if (offset != blocksize) {
4910 len = blocksize - offset;
4913 memset(kaddr + (block_start - page_offset(page)),
4916 memset(kaddr + (block_start - page_offset(page)) + offset,
4918 flush_dcache_page(page);
4921 ClearPageChecked(page);
4922 set_page_dirty(page);
4923 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4927 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4929 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize, (ret != 0));
4933 extent_changeset_free(data_reserved);
4937 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4938 u64 offset, u64 len)
4940 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4941 struct btrfs_trans_handle *trans;
4945 * Still need to make sure the inode looks like it's been updated so
4946 * that any holes get logged if we fsync.
4948 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4949 BTRFS_I(inode)->last_trans = fs_info->generation;
4950 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4951 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4956 * 1 - for the one we're dropping
4957 * 1 - for the one we're adding
4958 * 1 - for updating the inode.
4960 trans = btrfs_start_transaction(root, 3);
4962 return PTR_ERR(trans);
4964 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4966 btrfs_abort_transaction(trans, ret);
4967 btrfs_end_transaction(trans);
4971 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)),
4972 offset, 0, 0, len, 0, len, 0, 0, 0);
4974 btrfs_abort_transaction(trans, ret);
4976 btrfs_update_inode(trans, root, inode);
4977 btrfs_end_transaction(trans);
4982 * This function puts in dummy file extents for the area we're creating a hole
4983 * for. So if we are truncating this file to a larger size we need to insert
4984 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4985 * the range between oldsize and size
4987 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4989 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4990 struct btrfs_root *root = BTRFS_I(inode)->root;
4991 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4992 struct extent_map *em = NULL;
4993 struct extent_state *cached_state = NULL;
4994 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4995 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4996 u64 block_end = ALIGN(size, fs_info->sectorsize);
5003 * If our size started in the middle of a block we need to zero out the
5004 * rest of the block before we expand the i_size, otherwise we could
5005 * expose stale data.
5007 err = btrfs_truncate_block(inode, oldsize, 0, 0);
5011 if (size <= hole_start)
5015 struct btrfs_ordered_extent *ordered;
5017 lock_extent_bits(io_tree, hole_start, block_end - 1,
5019 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), hole_start,
5020 block_end - hole_start);
5023 unlock_extent_cached(io_tree, hole_start, block_end - 1,
5025 btrfs_start_ordered_extent(inode, ordered, 1);
5026 btrfs_put_ordered_extent(ordered);
5029 cur_offset = hole_start;
5031 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
5032 block_end - cur_offset, 0);
5038 last_byte = min(extent_map_end(em), block_end);
5039 last_byte = ALIGN(last_byte, fs_info->sectorsize);
5040 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5041 struct extent_map *hole_em;
5042 hole_size = last_byte - cur_offset;
5044 err = maybe_insert_hole(root, inode, cur_offset,
5048 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
5049 cur_offset + hole_size - 1, 0);
5050 hole_em = alloc_extent_map();
5052 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5053 &BTRFS_I(inode)->runtime_flags);
5056 hole_em->start = cur_offset;
5057 hole_em->len = hole_size;
5058 hole_em->orig_start = cur_offset;
5060 hole_em->block_start = EXTENT_MAP_HOLE;
5061 hole_em->block_len = 0;
5062 hole_em->orig_block_len = 0;
5063 hole_em->ram_bytes = hole_size;
5064 hole_em->bdev = fs_info->fs_devices->latest_bdev;
5065 hole_em->compress_type = BTRFS_COMPRESS_NONE;
5066 hole_em->generation = fs_info->generation;
5069 write_lock(&em_tree->lock);
5070 err = add_extent_mapping(em_tree, hole_em, 1);
5071 write_unlock(&em_tree->lock);
5074 btrfs_drop_extent_cache(BTRFS_I(inode),
5079 free_extent_map(hole_em);
5082 free_extent_map(em);
5084 cur_offset = last_byte;
5085 if (cur_offset >= block_end)
5088 free_extent_map(em);
5089 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
5093 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5095 struct btrfs_root *root = BTRFS_I(inode)->root;
5096 struct btrfs_trans_handle *trans;
5097 loff_t oldsize = i_size_read(inode);
5098 loff_t newsize = attr->ia_size;
5099 int mask = attr->ia_valid;
5103 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5104 * special case where we need to update the times despite not having
5105 * these flags set. For all other operations the VFS set these flags
5106 * explicitly if it wants a timestamp update.
5108 if (newsize != oldsize) {
5109 inode_inc_iversion(inode);
5110 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5111 inode->i_ctime = inode->i_mtime =
5112 current_time(inode);
5115 if (newsize > oldsize) {
5117 * Don't do an expanding truncate while snapshotting is ongoing.
5118 * This is to ensure the snapshot captures a fully consistent
5119 * state of this file - if the snapshot captures this expanding
5120 * truncation, it must capture all writes that happened before
5123 btrfs_wait_for_snapshot_creation(root);
5124 ret = btrfs_cont_expand(inode, oldsize, newsize);
5126 btrfs_end_write_no_snapshotting(root);
5130 trans = btrfs_start_transaction(root, 1);
5131 if (IS_ERR(trans)) {
5132 btrfs_end_write_no_snapshotting(root);
5133 return PTR_ERR(trans);
5136 i_size_write(inode, newsize);
5137 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
5138 pagecache_isize_extended(inode, oldsize, newsize);
5139 ret = btrfs_update_inode(trans, root, inode);
5140 btrfs_end_write_no_snapshotting(root);
5141 btrfs_end_transaction(trans);
5145 * We're truncating a file that used to have good data down to
5146 * zero. Make sure it gets into the ordered flush list so that
5147 * any new writes get down to disk quickly.
5150 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
5151 &BTRFS_I(inode)->runtime_flags);
5153 truncate_setsize(inode, newsize);
5155 /* Disable nonlocked read DIO to avoid the end less truncate */
5156 btrfs_inode_block_unlocked_dio(BTRFS_I(inode));
5157 inode_dio_wait(inode);
5158 btrfs_inode_resume_unlocked_dio(BTRFS_I(inode));
5160 ret = btrfs_truncate(inode, newsize == oldsize);
5161 if (ret && inode->i_nlink) {
5165 * Truncate failed, so fix up the in-memory size. We
5166 * adjusted disk_i_size down as we removed extents, so
5167 * wait for disk_i_size to be stable and then update the
5168 * in-memory size to match.
5170 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5173 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5180 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5182 struct inode *inode = d_inode(dentry);
5183 struct btrfs_root *root = BTRFS_I(inode)->root;
5186 if (btrfs_root_readonly(root))
5189 err = setattr_prepare(dentry, attr);
5193 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5194 err = btrfs_setsize(inode, attr);
5199 if (attr->ia_valid) {
5200 setattr_copy(inode, attr);
5201 inode_inc_iversion(inode);
5202 err = btrfs_dirty_inode(inode);
5204 if (!err && attr->ia_valid & ATTR_MODE)
5205 err = posix_acl_chmod(inode, inode->i_mode);
5212 * While truncating the inode pages during eviction, we get the VFS calling
5213 * btrfs_invalidatepage() against each page of the inode. This is slow because
5214 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5215 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5216 * extent_state structures over and over, wasting lots of time.
5218 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5219 * those expensive operations on a per page basis and do only the ordered io
5220 * finishing, while we release here the extent_map and extent_state structures,
5221 * without the excessive merging and splitting.
5223 static void evict_inode_truncate_pages(struct inode *inode)
5225 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5226 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5227 struct rb_node *node;
5229 ASSERT(inode->i_state & I_FREEING);
5230 truncate_inode_pages_final(&inode->i_data);
5232 write_lock(&map_tree->lock);
5233 while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) {
5234 struct extent_map *em;
5236 node = rb_first_cached(&map_tree->map);
5237 em = rb_entry(node, struct extent_map, rb_node);
5238 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5239 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5240 remove_extent_mapping(map_tree, em);
5241 free_extent_map(em);
5242 if (need_resched()) {
5243 write_unlock(&map_tree->lock);
5245 write_lock(&map_tree->lock);
5248 write_unlock(&map_tree->lock);
5251 * Keep looping until we have no more ranges in the io tree.
5252 * We can have ongoing bios started by readpages (called from readahead)
5253 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5254 * still in progress (unlocked the pages in the bio but did not yet
5255 * unlocked the ranges in the io tree). Therefore this means some
5256 * ranges can still be locked and eviction started because before
5257 * submitting those bios, which are executed by a separate task (work
5258 * queue kthread), inode references (inode->i_count) were not taken
5259 * (which would be dropped in the end io callback of each bio).
5260 * Therefore here we effectively end up waiting for those bios and
5261 * anyone else holding locked ranges without having bumped the inode's
5262 * reference count - if we don't do it, when they access the inode's
5263 * io_tree to unlock a range it may be too late, leading to an
5264 * use-after-free issue.
5266 spin_lock(&io_tree->lock);
5267 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5268 struct extent_state *state;
5269 struct extent_state *cached_state = NULL;
5272 unsigned state_flags;
5274 node = rb_first(&io_tree->state);
5275 state = rb_entry(node, struct extent_state, rb_node);
5276 start = state->start;
5278 state_flags = state->state;
5279 spin_unlock(&io_tree->lock);
5281 lock_extent_bits(io_tree, start, end, &cached_state);
5284 * If still has DELALLOC flag, the extent didn't reach disk,
5285 * and its reserved space won't be freed by delayed_ref.
5286 * So we need to free its reserved space here.
5287 * (Refer to comment in btrfs_invalidatepage, case 2)
5289 * Note, end is the bytenr of last byte, so we need + 1 here.
5291 if (state_flags & EXTENT_DELALLOC)
5292 btrfs_qgroup_free_data(inode, NULL, start, end - start + 1);
5294 clear_extent_bit(io_tree, start, end,
5295 EXTENT_LOCKED | EXTENT_DIRTY |
5296 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5297 EXTENT_DEFRAG, 1, 1, &cached_state);
5300 spin_lock(&io_tree->lock);
5302 spin_unlock(&io_tree->lock);
5305 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5306 struct btrfs_block_rsv *rsv)
5308 struct btrfs_fs_info *fs_info = root->fs_info;
5309 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5313 struct btrfs_trans_handle *trans;
5316 ret = btrfs_block_rsv_refill(root, rsv, rsv->size,
5317 BTRFS_RESERVE_FLUSH_LIMIT);
5319 if (ret && ++failures > 2) {
5321 "could not allocate space for a delete; will truncate on mount");
5322 return ERR_PTR(-ENOSPC);
5325 trans = btrfs_join_transaction(root);
5326 if (IS_ERR(trans) || !ret)
5330 * Try to steal from the global reserve if there is space for
5333 if (!btrfs_check_space_for_delayed_refs(trans) &&
5334 !btrfs_block_rsv_migrate(global_rsv, rsv, rsv->size, false))
5337 /* If not, commit and try again. */
5338 ret = btrfs_commit_transaction(trans);
5340 return ERR_PTR(ret);
5344 void btrfs_evict_inode(struct inode *inode)
5346 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5347 struct btrfs_trans_handle *trans;
5348 struct btrfs_root *root = BTRFS_I(inode)->root;
5349 struct btrfs_block_rsv *rsv;
5352 trace_btrfs_inode_evict(inode);
5359 evict_inode_truncate_pages(inode);
5361 if (inode->i_nlink &&
5362 ((btrfs_root_refs(&root->root_item) != 0 &&
5363 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5364 btrfs_is_free_space_inode(BTRFS_I(inode))))
5367 if (is_bad_inode(inode))
5370 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5372 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5375 if (inode->i_nlink > 0) {
5376 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5377 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5381 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5385 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5388 rsv->size = btrfs_calc_trunc_metadata_size(fs_info, 1);
5391 btrfs_i_size_write(BTRFS_I(inode), 0);
5394 trans = evict_refill_and_join(root, rsv);
5398 trans->block_rsv = rsv;
5400 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5401 trans->block_rsv = &fs_info->trans_block_rsv;
5402 btrfs_end_transaction(trans);
5403 btrfs_btree_balance_dirty(fs_info);
5404 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5411 * Errors here aren't a big deal, it just means we leave orphan items in
5412 * the tree. They will be cleaned up on the next mount. If the inode
5413 * number gets reused, cleanup deletes the orphan item without doing
5414 * anything, and unlink reuses the existing orphan item.
5416 * If it turns out that we are dropping too many of these, we might want
5417 * to add a mechanism for retrying these after a commit.
5419 trans = evict_refill_and_join(root, rsv);
5420 if (!IS_ERR(trans)) {
5421 trans->block_rsv = rsv;
5422 btrfs_orphan_del(trans, BTRFS_I(inode));
5423 trans->block_rsv = &fs_info->trans_block_rsv;
5424 btrfs_end_transaction(trans);
5427 if (!(root == fs_info->tree_root ||
5428 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5429 btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
5432 btrfs_free_block_rsv(fs_info, rsv);
5435 * If we didn't successfully delete, the orphan item will still be in
5436 * the tree and we'll retry on the next mount. Again, we might also want
5437 * to retry these periodically in the future.
5439 btrfs_remove_delayed_node(BTRFS_I(inode));
5444 * this returns the key found in the dir entry in the location pointer.
5445 * If no dir entries were found, returns -ENOENT.
5446 * If found a corrupted location in dir entry, returns -EUCLEAN.
5448 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5449 struct btrfs_key *location)
5451 const char *name = dentry->d_name.name;
5452 int namelen = dentry->d_name.len;
5453 struct btrfs_dir_item *di;
5454 struct btrfs_path *path;
5455 struct btrfs_root *root = BTRFS_I(dir)->root;
5458 path = btrfs_alloc_path();
5462 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5464 if (IS_ERR_OR_NULL(di)) {
5465 ret = di ? PTR_ERR(di) : -ENOENT;
5469 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5470 if (location->type != BTRFS_INODE_ITEM_KEY &&
5471 location->type != BTRFS_ROOT_ITEM_KEY) {
5473 btrfs_warn(root->fs_info,
5474 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5475 __func__, name, btrfs_ino(BTRFS_I(dir)),
5476 location->objectid, location->type, location->offset);
5479 btrfs_free_path(path);
5484 * when we hit a tree root in a directory, the btrfs part of the inode
5485 * needs to be changed to reflect the root directory of the tree root. This
5486 * is kind of like crossing a mount point.
5488 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5490 struct dentry *dentry,
5491 struct btrfs_key *location,
5492 struct btrfs_root **sub_root)
5494 struct btrfs_path *path;
5495 struct btrfs_root *new_root;
5496 struct btrfs_root_ref *ref;
5497 struct extent_buffer *leaf;
5498 struct btrfs_key key;
5502 path = btrfs_alloc_path();
5509 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5510 key.type = BTRFS_ROOT_REF_KEY;
5511 key.offset = location->objectid;
5513 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5520 leaf = path->nodes[0];
5521 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5522 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5523 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5526 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5527 (unsigned long)(ref + 1),
5528 dentry->d_name.len);
5532 btrfs_release_path(path);
5534 new_root = btrfs_read_fs_root_no_name(fs_info, location);
5535 if (IS_ERR(new_root)) {
5536 err = PTR_ERR(new_root);
5540 *sub_root = new_root;
5541 location->objectid = btrfs_root_dirid(&new_root->root_item);
5542 location->type = BTRFS_INODE_ITEM_KEY;
5543 location->offset = 0;
5546 btrfs_free_path(path);
5550 static void inode_tree_add(struct inode *inode)
5552 struct btrfs_root *root = BTRFS_I(inode)->root;
5553 struct btrfs_inode *entry;
5555 struct rb_node *parent;
5556 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5557 u64 ino = btrfs_ino(BTRFS_I(inode));
5559 if (inode_unhashed(inode))
5562 spin_lock(&root->inode_lock);
5563 p = &root->inode_tree.rb_node;
5566 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5568 if (ino < btrfs_ino(entry))
5569 p = &parent->rb_left;
5570 else if (ino > btrfs_ino(entry))
5571 p = &parent->rb_right;
5573 WARN_ON(!(entry->vfs_inode.i_state &
5574 (I_WILL_FREE | I_FREEING)));
5575 rb_replace_node(parent, new, &root->inode_tree);
5576 RB_CLEAR_NODE(parent);
5577 spin_unlock(&root->inode_lock);
5581 rb_link_node(new, parent, p);
5582 rb_insert_color(new, &root->inode_tree);
5583 spin_unlock(&root->inode_lock);
5586 static void inode_tree_del(struct inode *inode)
5588 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5589 struct btrfs_root *root = BTRFS_I(inode)->root;
5592 spin_lock(&root->inode_lock);
5593 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5594 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5595 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5596 empty = RB_EMPTY_ROOT(&root->inode_tree);
5598 spin_unlock(&root->inode_lock);
5600 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5601 synchronize_srcu(&fs_info->subvol_srcu);
5602 spin_lock(&root->inode_lock);
5603 empty = RB_EMPTY_ROOT(&root->inode_tree);
5604 spin_unlock(&root->inode_lock);
5606 btrfs_add_dead_root(root);
5611 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5613 struct btrfs_iget_args *args = p;
5614 inode->i_ino = args->location->objectid;
5615 memcpy(&BTRFS_I(inode)->location, args->location,
5616 sizeof(*args->location));
5617 BTRFS_I(inode)->root = args->root;
5621 static int btrfs_find_actor(struct inode *inode, void *opaque)
5623 struct btrfs_iget_args *args = opaque;
5624 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5625 args->root == BTRFS_I(inode)->root;
5628 static struct inode *btrfs_iget_locked(struct super_block *s,
5629 struct btrfs_key *location,
5630 struct btrfs_root *root)
5632 struct inode *inode;
5633 struct btrfs_iget_args args;
5634 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5636 args.location = location;
5639 inode = iget5_locked(s, hashval, btrfs_find_actor,
5640 btrfs_init_locked_inode,
5645 /* Get an inode object given its location and corresponding root.
5646 * Returns in *is_new if the inode was read from disk
5648 struct inode *btrfs_iget_path(struct super_block *s, struct btrfs_key *location,
5649 struct btrfs_root *root, int *new,
5650 struct btrfs_path *path)
5652 struct inode *inode;
5654 inode = btrfs_iget_locked(s, location, root);
5656 return ERR_PTR(-ENOMEM);
5658 if (inode->i_state & I_NEW) {
5661 ret = btrfs_read_locked_inode(inode, path);
5663 inode_tree_add(inode);
5664 unlock_new_inode(inode);
5670 * ret > 0 can come from btrfs_search_slot called by
5671 * btrfs_read_locked_inode, this means the inode item
5676 inode = ERR_PTR(ret);
5683 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5684 struct btrfs_root *root, int *new)
5686 return btrfs_iget_path(s, location, root, new, NULL);
5689 static struct inode *new_simple_dir(struct super_block *s,
5690 struct btrfs_key *key,
5691 struct btrfs_root *root)
5693 struct inode *inode = new_inode(s);
5696 return ERR_PTR(-ENOMEM);
5698 BTRFS_I(inode)->root = root;
5699 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5700 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5702 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5703 inode->i_op = &btrfs_dir_ro_inode_operations;
5704 inode->i_opflags &= ~IOP_XATTR;
5705 inode->i_fop = &simple_dir_operations;
5706 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5707 inode->i_mtime = current_time(inode);
5708 inode->i_atime = inode->i_mtime;
5709 inode->i_ctime = inode->i_mtime;
5710 BTRFS_I(inode)->i_otime = inode->i_mtime;
5715 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5717 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5718 struct inode *inode;
5719 struct btrfs_root *root = BTRFS_I(dir)->root;
5720 struct btrfs_root *sub_root = root;
5721 struct btrfs_key location;
5725 if (dentry->d_name.len > BTRFS_NAME_LEN)
5726 return ERR_PTR(-ENAMETOOLONG);
5728 ret = btrfs_inode_by_name(dir, dentry, &location);
5730 return ERR_PTR(ret);
5732 if (location.type == BTRFS_INODE_ITEM_KEY) {
5733 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5737 index = srcu_read_lock(&fs_info->subvol_srcu);
5738 ret = fixup_tree_root_location(fs_info, dir, dentry,
5739 &location, &sub_root);
5742 inode = ERR_PTR(ret);
5744 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5746 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5748 srcu_read_unlock(&fs_info->subvol_srcu, index);
5750 if (!IS_ERR(inode) && root != sub_root) {
5751 down_read(&fs_info->cleanup_work_sem);
5752 if (!sb_rdonly(inode->i_sb))
5753 ret = btrfs_orphan_cleanup(sub_root);
5754 up_read(&fs_info->cleanup_work_sem);
5757 inode = ERR_PTR(ret);
5764 static int btrfs_dentry_delete(const struct dentry *dentry)
5766 struct btrfs_root *root;
5767 struct inode *inode = d_inode(dentry);
5769 if (!inode && !IS_ROOT(dentry))
5770 inode = d_inode(dentry->d_parent);
5773 root = BTRFS_I(inode)->root;
5774 if (btrfs_root_refs(&root->root_item) == 0)
5777 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5783 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5786 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5788 if (inode == ERR_PTR(-ENOENT))
5790 return d_splice_alias(inode, dentry);
5793 unsigned char btrfs_filetype_table[] = {
5794 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5798 * All this infrastructure exists because dir_emit can fault, and we are holding
5799 * the tree lock when doing readdir. For now just allocate a buffer and copy
5800 * our information into that, and then dir_emit from the buffer. This is
5801 * similar to what NFS does, only we don't keep the buffer around in pagecache
5802 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5803 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5806 static int btrfs_opendir(struct inode *inode, struct file *file)
5808 struct btrfs_file_private *private;
5810 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5813 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5814 if (!private->filldir_buf) {
5818 file->private_data = private;
5829 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5832 struct dir_entry *entry = addr;
5833 char *name = (char *)(entry + 1);
5835 ctx->pos = get_unaligned(&entry->offset);
5836 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5837 get_unaligned(&entry->ino),
5838 get_unaligned(&entry->type)))
5840 addr += sizeof(struct dir_entry) +
5841 get_unaligned(&entry->name_len);
5847 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5849 struct inode *inode = file_inode(file);
5850 struct btrfs_root *root = BTRFS_I(inode)->root;
5851 struct btrfs_file_private *private = file->private_data;
5852 struct btrfs_dir_item *di;
5853 struct btrfs_key key;
5854 struct btrfs_key found_key;
5855 struct btrfs_path *path;
5857 struct list_head ins_list;
5858 struct list_head del_list;
5860 struct extent_buffer *leaf;
5867 struct btrfs_key location;
5869 if (!dir_emit_dots(file, ctx))
5872 path = btrfs_alloc_path();
5876 addr = private->filldir_buf;
5877 path->reada = READA_FORWARD;
5879 INIT_LIST_HEAD(&ins_list);
5880 INIT_LIST_HEAD(&del_list);
5881 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5884 key.type = BTRFS_DIR_INDEX_KEY;
5885 key.offset = ctx->pos;
5886 key.objectid = btrfs_ino(BTRFS_I(inode));
5888 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5893 struct dir_entry *entry;
5895 leaf = path->nodes[0];
5896 slot = path->slots[0];
5897 if (slot >= btrfs_header_nritems(leaf)) {
5898 ret = btrfs_next_leaf(root, path);
5906 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5908 if (found_key.objectid != key.objectid)
5910 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5912 if (found_key.offset < ctx->pos)
5914 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5916 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5917 name_len = btrfs_dir_name_len(leaf, di);
5918 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5920 btrfs_release_path(path);
5921 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5924 addr = private->filldir_buf;
5931 put_unaligned(name_len, &entry->name_len);
5932 name_ptr = (char *)(entry + 1);
5933 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5935 put_unaligned(btrfs_filetype_table[btrfs_dir_type(leaf, di)],
5937 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5938 put_unaligned(location.objectid, &entry->ino);
5939 put_unaligned(found_key.offset, &entry->offset);
5941 addr += sizeof(struct dir_entry) + name_len;
5942 total_len += sizeof(struct dir_entry) + name_len;
5946 btrfs_release_path(path);
5948 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5952 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5957 * Stop new entries from being returned after we return the last
5960 * New directory entries are assigned a strictly increasing
5961 * offset. This means that new entries created during readdir
5962 * are *guaranteed* to be seen in the future by that readdir.
5963 * This has broken buggy programs which operate on names as
5964 * they're returned by readdir. Until we re-use freed offsets
5965 * we have this hack to stop new entries from being returned
5966 * under the assumption that they'll never reach this huge
5969 * This is being careful not to overflow 32bit loff_t unless the
5970 * last entry requires it because doing so has broken 32bit apps
5973 if (ctx->pos >= INT_MAX)
5974 ctx->pos = LLONG_MAX;
5981 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5982 btrfs_free_path(path);
5987 * This is somewhat expensive, updating the tree every time the
5988 * inode changes. But, it is most likely to find the inode in cache.
5989 * FIXME, needs more benchmarking...there are no reasons other than performance
5990 * to keep or drop this code.
5992 static int btrfs_dirty_inode(struct inode *inode)
5994 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5995 struct btrfs_root *root = BTRFS_I(inode)->root;
5996 struct btrfs_trans_handle *trans;
5999 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6002 trans = btrfs_join_transaction(root);
6004 return PTR_ERR(trans);
6006 ret = btrfs_update_inode(trans, root, inode);
6007 if (ret && ret == -ENOSPC) {
6008 /* whoops, lets try again with the full transaction */
6009 btrfs_end_transaction(trans);
6010 trans = btrfs_start_transaction(root, 1);
6012 return PTR_ERR(trans);
6014 ret = btrfs_update_inode(trans, root, inode);
6016 btrfs_end_transaction(trans);
6017 if (BTRFS_I(inode)->delayed_node)
6018 btrfs_balance_delayed_items(fs_info);
6024 * This is a copy of file_update_time. We need this so we can return error on
6025 * ENOSPC for updating the inode in the case of file write and mmap writes.
6027 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
6030 struct btrfs_root *root = BTRFS_I(inode)->root;
6031 bool dirty = flags & ~S_VERSION;
6033 if (btrfs_root_readonly(root))
6036 if (flags & S_VERSION)
6037 dirty |= inode_maybe_inc_iversion(inode, dirty);
6038 if (flags & S_CTIME)
6039 inode->i_ctime = *now;
6040 if (flags & S_MTIME)
6041 inode->i_mtime = *now;
6042 if (flags & S_ATIME)
6043 inode->i_atime = *now;
6044 return dirty ? btrfs_dirty_inode(inode) : 0;
6048 * find the highest existing sequence number in a directory
6049 * and then set the in-memory index_cnt variable to reflect
6050 * free sequence numbers
6052 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6054 struct btrfs_root *root = inode->root;
6055 struct btrfs_key key, found_key;
6056 struct btrfs_path *path;
6057 struct extent_buffer *leaf;
6060 key.objectid = btrfs_ino(inode);
6061 key.type = BTRFS_DIR_INDEX_KEY;
6062 key.offset = (u64)-1;
6064 path = btrfs_alloc_path();
6068 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6071 /* FIXME: we should be able to handle this */
6077 * MAGIC NUMBER EXPLANATION:
6078 * since we search a directory based on f_pos we have to start at 2
6079 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6080 * else has to start at 2
6082 if (path->slots[0] == 0) {
6083 inode->index_cnt = 2;
6089 leaf = path->nodes[0];
6090 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6092 if (found_key.objectid != btrfs_ino(inode) ||
6093 found_key.type != BTRFS_DIR_INDEX_KEY) {
6094 inode->index_cnt = 2;
6098 inode->index_cnt = found_key.offset + 1;
6100 btrfs_free_path(path);
6105 * helper to find a free sequence number in a given directory. This current
6106 * code is very simple, later versions will do smarter things in the btree
6108 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6112 if (dir->index_cnt == (u64)-1) {
6113 ret = btrfs_inode_delayed_dir_index_count(dir);
6115 ret = btrfs_set_inode_index_count(dir);
6121 *index = dir->index_cnt;
6127 static int btrfs_insert_inode_locked(struct inode *inode)
6129 struct btrfs_iget_args args;
6130 args.location = &BTRFS_I(inode)->location;
6131 args.root = BTRFS_I(inode)->root;
6133 return insert_inode_locked4(inode,
6134 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6135 btrfs_find_actor, &args);
6139 * Inherit flags from the parent inode.
6141 * Currently only the compression flags and the cow flags are inherited.
6143 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6150 flags = BTRFS_I(dir)->flags;
6152 if (flags & BTRFS_INODE_NOCOMPRESS) {
6153 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6154 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6155 } else if (flags & BTRFS_INODE_COMPRESS) {
6156 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6157 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6160 if (flags & BTRFS_INODE_NODATACOW) {
6161 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6162 if (S_ISREG(inode->i_mode))
6163 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6166 btrfs_sync_inode_flags_to_i_flags(inode);
6169 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6170 struct btrfs_root *root,
6172 const char *name, int name_len,
6173 u64 ref_objectid, u64 objectid,
6174 umode_t mode, u64 *index)
6176 struct btrfs_fs_info *fs_info = root->fs_info;
6177 struct inode *inode;
6178 struct btrfs_inode_item *inode_item;
6179 struct btrfs_key *location;
6180 struct btrfs_path *path;
6181 struct btrfs_inode_ref *ref;
6182 struct btrfs_key key[2];
6184 int nitems = name ? 2 : 1;
6188 path = btrfs_alloc_path();
6190 return ERR_PTR(-ENOMEM);
6192 inode = new_inode(fs_info->sb);
6194 btrfs_free_path(path);
6195 return ERR_PTR(-ENOMEM);
6199 * O_TMPFILE, set link count to 0, so that after this point,
6200 * we fill in an inode item with the correct link count.
6203 set_nlink(inode, 0);
6206 * we have to initialize this early, so we can reclaim the inode
6207 * number if we fail afterwards in this function.
6209 inode->i_ino = objectid;
6212 trace_btrfs_inode_request(dir);
6214 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6216 btrfs_free_path(path);
6218 return ERR_PTR(ret);
6224 * index_cnt is ignored for everything but a dir,
6225 * btrfs_set_inode_index_count has an explanation for the magic
6228 BTRFS_I(inode)->index_cnt = 2;
6229 BTRFS_I(inode)->dir_index = *index;
6230 BTRFS_I(inode)->root = root;
6231 BTRFS_I(inode)->generation = trans->transid;
6232 inode->i_generation = BTRFS_I(inode)->generation;
6235 * We could have gotten an inode number from somebody who was fsynced
6236 * and then removed in this same transaction, so let's just set full
6237 * sync since it will be a full sync anyway and this will blow away the
6238 * old info in the log.
6240 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6242 key[0].objectid = objectid;
6243 key[0].type = BTRFS_INODE_ITEM_KEY;
6246 sizes[0] = sizeof(struct btrfs_inode_item);
6250 * Start new inodes with an inode_ref. This is slightly more
6251 * efficient for small numbers of hard links since they will
6252 * be packed into one item. Extended refs will kick in if we
6253 * add more hard links than can fit in the ref item.
6255 key[1].objectid = objectid;
6256 key[1].type = BTRFS_INODE_REF_KEY;
6257 key[1].offset = ref_objectid;
6259 sizes[1] = name_len + sizeof(*ref);
6262 location = &BTRFS_I(inode)->location;
6263 location->objectid = objectid;
6264 location->offset = 0;
6265 location->type = BTRFS_INODE_ITEM_KEY;
6267 ret = btrfs_insert_inode_locked(inode);
6273 path->leave_spinning = 1;
6274 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6278 inode_init_owner(inode, dir, mode);
6279 inode_set_bytes(inode, 0);
6281 inode->i_mtime = current_time(inode);
6282 inode->i_atime = inode->i_mtime;
6283 inode->i_ctime = inode->i_mtime;
6284 BTRFS_I(inode)->i_otime = inode->i_mtime;
6286 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6287 struct btrfs_inode_item);
6288 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6289 sizeof(*inode_item));
6290 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6293 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6294 struct btrfs_inode_ref);
6295 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6296 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6297 ptr = (unsigned long)(ref + 1);
6298 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6301 btrfs_mark_buffer_dirty(path->nodes[0]);
6302 btrfs_free_path(path);
6304 btrfs_inherit_iflags(inode, dir);
6306 if (S_ISREG(mode)) {
6307 if (btrfs_test_opt(fs_info, NODATASUM))
6308 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6309 if (btrfs_test_opt(fs_info, NODATACOW))
6310 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6311 BTRFS_INODE_NODATASUM;
6314 inode_tree_add(inode);
6316 trace_btrfs_inode_new(inode);
6317 btrfs_set_inode_last_trans(trans, inode);
6319 btrfs_update_root_times(trans, root);
6321 ret = btrfs_inode_inherit_props(trans, inode, dir);
6324 "error inheriting props for ino %llu (root %llu): %d",
6325 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6330 discard_new_inode(inode);
6333 BTRFS_I(dir)->index_cnt--;
6334 btrfs_free_path(path);
6335 return ERR_PTR(ret);
6338 static inline u8 btrfs_inode_type(struct inode *inode)
6340 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6344 * utility function to add 'inode' into 'parent_inode' with
6345 * a give name and a given sequence number.
6346 * if 'add_backref' is true, also insert a backref from the
6347 * inode to the parent directory.
6349 int btrfs_add_link(struct btrfs_trans_handle *trans,
6350 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6351 const char *name, int name_len, int add_backref, u64 index)
6354 struct btrfs_key key;
6355 struct btrfs_root *root = parent_inode->root;
6356 u64 ino = btrfs_ino(inode);
6357 u64 parent_ino = btrfs_ino(parent_inode);
6359 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6360 memcpy(&key, &inode->root->root_key, sizeof(key));
6363 key.type = BTRFS_INODE_ITEM_KEY;
6367 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6368 ret = btrfs_add_root_ref(trans, key.objectid,
6369 root->root_key.objectid, parent_ino,
6370 index, name, name_len);
6371 } else if (add_backref) {
6372 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6376 /* Nothing to clean up yet */
6380 ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key,
6381 btrfs_inode_type(&inode->vfs_inode), index);
6382 if (ret == -EEXIST || ret == -EOVERFLOW)
6385 btrfs_abort_transaction(trans, ret);
6389 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6391 inode_inc_iversion(&parent_inode->vfs_inode);
6392 parent_inode->vfs_inode.i_mtime = parent_inode->vfs_inode.i_ctime =
6393 current_time(&parent_inode->vfs_inode);
6394 ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode);
6396 btrfs_abort_transaction(trans, ret);
6400 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6403 err = btrfs_del_root_ref(trans, key.objectid,
6404 root->root_key.objectid, parent_ino,
6405 &local_index, name, name_len);
6407 } else if (add_backref) {
6411 err = btrfs_del_inode_ref(trans, root, name, name_len,
6412 ino, parent_ino, &local_index);
6417 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6418 struct btrfs_inode *dir, struct dentry *dentry,
6419 struct btrfs_inode *inode, int backref, u64 index)
6421 int err = btrfs_add_link(trans, dir, inode,
6422 dentry->d_name.name, dentry->d_name.len,
6429 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6430 umode_t mode, dev_t rdev)
6432 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6433 struct btrfs_trans_handle *trans;
6434 struct btrfs_root *root = BTRFS_I(dir)->root;
6435 struct inode *inode = NULL;
6441 * 2 for inode item and ref
6443 * 1 for xattr if selinux is on
6445 trans = btrfs_start_transaction(root, 5);
6447 return PTR_ERR(trans);
6449 err = btrfs_find_free_ino(root, &objectid);
6453 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6454 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6456 if (IS_ERR(inode)) {
6457 err = PTR_ERR(inode);
6463 * If the active LSM wants to access the inode during
6464 * d_instantiate it needs these. Smack checks to see
6465 * if the filesystem supports xattrs by looking at the
6468 inode->i_op = &btrfs_special_inode_operations;
6469 init_special_inode(inode, inode->i_mode, rdev);
6471 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6475 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6480 btrfs_update_inode(trans, root, inode);
6481 d_instantiate_new(dentry, inode);
6484 btrfs_end_transaction(trans);
6485 btrfs_btree_balance_dirty(fs_info);
6487 inode_dec_link_count(inode);
6488 discard_new_inode(inode);
6493 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6494 umode_t mode, bool excl)
6496 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6497 struct btrfs_trans_handle *trans;
6498 struct btrfs_root *root = BTRFS_I(dir)->root;
6499 struct inode *inode = NULL;
6505 * 2 for inode item and ref
6507 * 1 for xattr if selinux is on
6509 trans = btrfs_start_transaction(root, 5);
6511 return PTR_ERR(trans);
6513 err = btrfs_find_free_ino(root, &objectid);
6517 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6518 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6520 if (IS_ERR(inode)) {
6521 err = PTR_ERR(inode);
6526 * If the active LSM wants to access the inode during
6527 * d_instantiate it needs these. Smack checks to see
6528 * if the filesystem supports xattrs by looking at the
6531 inode->i_fop = &btrfs_file_operations;
6532 inode->i_op = &btrfs_file_inode_operations;
6533 inode->i_mapping->a_ops = &btrfs_aops;
6535 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6539 err = btrfs_update_inode(trans, root, inode);
6543 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6548 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6549 d_instantiate_new(dentry, inode);
6552 btrfs_end_transaction(trans);
6554 inode_dec_link_count(inode);
6555 discard_new_inode(inode);
6557 btrfs_btree_balance_dirty(fs_info);
6561 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6562 struct dentry *dentry)
6564 struct btrfs_trans_handle *trans = NULL;
6565 struct btrfs_root *root = BTRFS_I(dir)->root;
6566 struct inode *inode = d_inode(old_dentry);
6567 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6572 /* do not allow sys_link's with other subvols of the same device */
6573 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6576 if (inode->i_nlink >= BTRFS_LINK_MAX)
6579 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6584 * 2 items for inode and inode ref
6585 * 2 items for dir items
6586 * 1 item for parent inode
6587 * 1 item for orphan item deletion if O_TMPFILE
6589 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6590 if (IS_ERR(trans)) {
6591 err = PTR_ERR(trans);
6596 /* There are several dir indexes for this inode, clear the cache. */
6597 BTRFS_I(inode)->dir_index = 0ULL;
6599 inode_inc_iversion(inode);
6600 inode->i_ctime = current_time(inode);
6602 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6604 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6610 struct dentry *parent = dentry->d_parent;
6613 err = btrfs_update_inode(trans, root, inode);
6616 if (inode->i_nlink == 1) {
6618 * If new hard link count is 1, it's a file created
6619 * with open(2) O_TMPFILE flag.
6621 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6625 d_instantiate(dentry, inode);
6626 ret = btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent,
6628 if (ret == BTRFS_NEED_TRANS_COMMIT) {
6629 err = btrfs_commit_transaction(trans);
6636 btrfs_end_transaction(trans);
6638 inode_dec_link_count(inode);
6641 btrfs_btree_balance_dirty(fs_info);
6645 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6647 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6648 struct inode *inode = NULL;
6649 struct btrfs_trans_handle *trans;
6650 struct btrfs_root *root = BTRFS_I(dir)->root;
6652 int drop_on_err = 0;
6657 * 2 items for inode and ref
6658 * 2 items for dir items
6659 * 1 for xattr if selinux is on
6661 trans = btrfs_start_transaction(root, 5);
6663 return PTR_ERR(trans);
6665 err = btrfs_find_free_ino(root, &objectid);
6669 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6670 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6671 S_IFDIR | mode, &index);
6672 if (IS_ERR(inode)) {
6673 err = PTR_ERR(inode);
6679 /* these must be set before we unlock the inode */
6680 inode->i_op = &btrfs_dir_inode_operations;
6681 inode->i_fop = &btrfs_dir_file_operations;
6683 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6687 btrfs_i_size_write(BTRFS_I(inode), 0);
6688 err = btrfs_update_inode(trans, root, inode);
6692 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6693 dentry->d_name.name,
6694 dentry->d_name.len, 0, index);
6698 d_instantiate_new(dentry, inode);
6702 btrfs_end_transaction(trans);
6704 inode_dec_link_count(inode);
6705 discard_new_inode(inode);
6707 btrfs_btree_balance_dirty(fs_info);
6711 static noinline int uncompress_inline(struct btrfs_path *path,
6713 size_t pg_offset, u64 extent_offset,
6714 struct btrfs_file_extent_item *item)
6717 struct extent_buffer *leaf = path->nodes[0];
6720 unsigned long inline_size;
6724 WARN_ON(pg_offset != 0);
6725 compress_type = btrfs_file_extent_compression(leaf, item);
6726 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6727 inline_size = btrfs_file_extent_inline_item_len(leaf,
6728 btrfs_item_nr(path->slots[0]));
6729 tmp = kmalloc(inline_size, GFP_NOFS);
6732 ptr = btrfs_file_extent_inline_start(item);
6734 read_extent_buffer(leaf, tmp, ptr, inline_size);
6736 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6737 ret = btrfs_decompress(compress_type, tmp, page,
6738 extent_offset, inline_size, max_size);
6741 * decompression code contains a memset to fill in any space between the end
6742 * of the uncompressed data and the end of max_size in case the decompressed
6743 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6744 * the end of an inline extent and the beginning of the next block, so we
6745 * cover that region here.
6748 if (max_size + pg_offset < PAGE_SIZE) {
6749 char *map = kmap(page);
6750 memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6758 * a bit scary, this does extent mapping from logical file offset to the disk.
6759 * the ugly parts come from merging extents from the disk with the in-ram
6760 * representation. This gets more complex because of the data=ordered code,
6761 * where the in-ram extents might be locked pending data=ordered completion.
6763 * This also copies inline extents directly into the page.
6765 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6767 size_t pg_offset, u64 start, u64 len,
6770 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6773 u64 extent_start = 0;
6775 u64 objectid = btrfs_ino(inode);
6777 struct btrfs_path *path = NULL;
6778 struct btrfs_root *root = inode->root;
6779 struct btrfs_file_extent_item *item;
6780 struct extent_buffer *leaf;
6781 struct btrfs_key found_key;
6782 struct extent_map *em = NULL;
6783 struct extent_map_tree *em_tree = &inode->extent_tree;
6784 struct extent_io_tree *io_tree = &inode->io_tree;
6785 const bool new_inline = !page || create;
6787 read_lock(&em_tree->lock);
6788 em = lookup_extent_mapping(em_tree, start, len);
6790 em->bdev = fs_info->fs_devices->latest_bdev;
6791 read_unlock(&em_tree->lock);
6794 if (em->start > start || em->start + em->len <= start)
6795 free_extent_map(em);
6796 else if (em->block_start == EXTENT_MAP_INLINE && page)
6797 free_extent_map(em);
6801 em = alloc_extent_map();
6806 em->bdev = fs_info->fs_devices->latest_bdev;
6807 em->start = EXTENT_MAP_HOLE;
6808 em->orig_start = EXTENT_MAP_HOLE;
6810 em->block_len = (u64)-1;
6812 path = btrfs_alloc_path();
6818 /* Chances are we'll be called again, so go ahead and do readahead */
6819 path->reada = READA_FORWARD;
6822 * Unless we're going to uncompress the inline extent, no sleep would
6825 path->leave_spinning = 1;
6827 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6834 if (path->slots[0] == 0)
6839 leaf = path->nodes[0];
6840 item = btrfs_item_ptr(leaf, path->slots[0],
6841 struct btrfs_file_extent_item);
6842 /* are we inside the extent that was found? */
6843 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6844 found_type = found_key.type;
6845 if (found_key.objectid != objectid ||
6846 found_type != BTRFS_EXTENT_DATA_KEY) {
6848 * If we backup past the first extent we want to move forward
6849 * and see if there is an extent in front of us, otherwise we'll
6850 * say there is a hole for our whole search range which can
6857 found_type = btrfs_file_extent_type(leaf, item);
6858 extent_start = found_key.offset;
6859 if (found_type == BTRFS_FILE_EXTENT_REG ||
6860 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6861 extent_end = extent_start +
6862 btrfs_file_extent_num_bytes(leaf, item);
6864 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6866 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6869 size = btrfs_file_extent_ram_bytes(leaf, item);
6870 extent_end = ALIGN(extent_start + size,
6871 fs_info->sectorsize);
6873 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6878 if (start >= extent_end) {
6880 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6881 ret = btrfs_next_leaf(root, path);
6888 leaf = path->nodes[0];
6890 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6891 if (found_key.objectid != objectid ||
6892 found_key.type != BTRFS_EXTENT_DATA_KEY)
6894 if (start + len <= found_key.offset)
6896 if (start > found_key.offset)
6899 em->orig_start = start;
6900 em->len = found_key.offset - start;
6904 btrfs_extent_item_to_extent_map(inode, path, item,
6907 if (found_type == BTRFS_FILE_EXTENT_REG ||
6908 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6910 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6914 size_t extent_offset;
6920 size = btrfs_file_extent_ram_bytes(leaf, item);
6921 extent_offset = page_offset(page) + pg_offset - extent_start;
6922 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6923 size - extent_offset);
6924 em->start = extent_start + extent_offset;
6925 em->len = ALIGN(copy_size, fs_info->sectorsize);
6926 em->orig_block_len = em->len;
6927 em->orig_start = em->start;
6928 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6930 btrfs_set_path_blocking(path);
6931 if (!PageUptodate(page)) {
6932 if (btrfs_file_extent_compression(leaf, item) !=
6933 BTRFS_COMPRESS_NONE) {
6934 ret = uncompress_inline(path, page, pg_offset,
6935 extent_offset, item);
6942 read_extent_buffer(leaf, map + pg_offset, ptr,
6944 if (pg_offset + copy_size < PAGE_SIZE) {
6945 memset(map + pg_offset + copy_size, 0,
6946 PAGE_SIZE - pg_offset -
6951 flush_dcache_page(page);
6953 set_extent_uptodate(io_tree, em->start,
6954 extent_map_end(em) - 1, NULL, GFP_NOFS);
6959 em->orig_start = start;
6962 em->block_start = EXTENT_MAP_HOLE;
6964 btrfs_release_path(path);
6965 if (em->start > start || extent_map_end(em) <= start) {
6967 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6968 em->start, em->len, start, len);
6974 write_lock(&em_tree->lock);
6975 err = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6976 write_unlock(&em_tree->lock);
6978 btrfs_free_path(path);
6980 trace_btrfs_get_extent(root, inode, em);
6983 free_extent_map(em);
6984 return ERR_PTR(err);
6986 BUG_ON(!em); /* Error is always set */
6990 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
6992 size_t pg_offset, u64 start, u64 len,
6995 struct extent_map *em;
6996 struct extent_map *hole_em = NULL;
6997 u64 range_start = start;
7003 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7007 * If our em maps to:
7009 * - a pre-alloc extent,
7010 * there might actually be delalloc bytes behind it.
7012 if (em->block_start != EXTENT_MAP_HOLE &&
7013 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7018 /* check to see if we've wrapped (len == -1 or similar) */
7027 /* ok, we didn't find anything, lets look for delalloc */
7028 found = count_range_bits(&inode->io_tree, &range_start,
7029 end, len, EXTENT_DELALLOC, 1);
7030 found_end = range_start + found;
7031 if (found_end < range_start)
7032 found_end = (u64)-1;
7035 * we didn't find anything useful, return
7036 * the original results from get_extent()
7038 if (range_start > end || found_end <= start) {
7044 /* adjust the range_start to make sure it doesn't
7045 * go backwards from the start they passed in
7047 range_start = max(start, range_start);
7048 found = found_end - range_start;
7051 u64 hole_start = start;
7054 em = alloc_extent_map();
7060 * when btrfs_get_extent can't find anything it
7061 * returns one huge hole
7063 * make sure what it found really fits our range, and
7064 * adjust to make sure it is based on the start from
7068 u64 calc_end = extent_map_end(hole_em);
7070 if (calc_end <= start || (hole_em->start > end)) {
7071 free_extent_map(hole_em);
7074 hole_start = max(hole_em->start, start);
7075 hole_len = calc_end - hole_start;
7079 if (hole_em && range_start > hole_start) {
7080 /* our hole starts before our delalloc, so we
7081 * have to return just the parts of the hole
7082 * that go until the delalloc starts
7084 em->len = min(hole_len,
7085 range_start - hole_start);
7086 em->start = hole_start;
7087 em->orig_start = hole_start;
7089 * don't adjust block start at all,
7090 * it is fixed at EXTENT_MAP_HOLE
7092 em->block_start = hole_em->block_start;
7093 em->block_len = hole_len;
7094 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7095 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7097 em->start = range_start;
7099 em->orig_start = range_start;
7100 em->block_start = EXTENT_MAP_DELALLOC;
7101 em->block_len = found;
7108 free_extent_map(hole_em);
7110 free_extent_map(em);
7111 return ERR_PTR(err);
7116 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7119 const u64 orig_start,
7120 const u64 block_start,
7121 const u64 block_len,
7122 const u64 orig_block_len,
7123 const u64 ram_bytes,
7126 struct extent_map *em = NULL;
7129 if (type != BTRFS_ORDERED_NOCOW) {
7130 em = create_io_em(inode, start, len, orig_start,
7131 block_start, block_len, orig_block_len,
7133 BTRFS_COMPRESS_NONE, /* compress_type */
7138 ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7139 len, block_len, type);
7142 free_extent_map(em);
7143 btrfs_drop_extent_cache(BTRFS_I(inode), start,
7144 start + len - 1, 0);
7153 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7156 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7157 struct btrfs_root *root = BTRFS_I(inode)->root;
7158 struct extent_map *em;
7159 struct btrfs_key ins;
7163 alloc_hint = get_extent_allocation_hint(inode, start, len);
7164 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7165 0, alloc_hint, &ins, 1, 1);
7167 return ERR_PTR(ret);
7169 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7170 ins.objectid, ins.offset, ins.offset,
7171 ins.offset, BTRFS_ORDERED_REGULAR);
7172 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7174 btrfs_free_reserved_extent(fs_info, ins.objectid,
7181 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7182 * block must be cow'd
7184 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7185 u64 *orig_start, u64 *orig_block_len,
7188 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7189 struct btrfs_path *path;
7191 struct extent_buffer *leaf;
7192 struct btrfs_root *root = BTRFS_I(inode)->root;
7193 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7194 struct btrfs_file_extent_item *fi;
7195 struct btrfs_key key;
7202 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7204 path = btrfs_alloc_path();
7208 ret = btrfs_lookup_file_extent(NULL, root, path,
7209 btrfs_ino(BTRFS_I(inode)), offset, 0);
7213 slot = path->slots[0];
7216 /* can't find the item, must cow */
7223 leaf = path->nodes[0];
7224 btrfs_item_key_to_cpu(leaf, &key, slot);
7225 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7226 key.type != BTRFS_EXTENT_DATA_KEY) {
7227 /* not our file or wrong item type, must cow */
7231 if (key.offset > offset) {
7232 /* Wrong offset, must cow */
7236 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7237 found_type = btrfs_file_extent_type(leaf, fi);
7238 if (found_type != BTRFS_FILE_EXTENT_REG &&
7239 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7240 /* not a regular extent, must cow */
7244 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7247 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7248 if (extent_end <= offset)
7251 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7252 if (disk_bytenr == 0)
7255 if (btrfs_file_extent_compression(leaf, fi) ||
7256 btrfs_file_extent_encryption(leaf, fi) ||
7257 btrfs_file_extent_other_encoding(leaf, fi))
7261 * Do the same check as in btrfs_cross_ref_exist but without the
7262 * unnecessary search.
7264 if (btrfs_file_extent_generation(leaf, fi) <=
7265 btrfs_root_last_snapshot(&root->root_item))
7268 backref_offset = btrfs_file_extent_offset(leaf, fi);
7271 *orig_start = key.offset - backref_offset;
7272 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7273 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7276 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7279 num_bytes = min(offset + *len, extent_end) - offset;
7280 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7283 range_end = round_up(offset + num_bytes,
7284 root->fs_info->sectorsize) - 1;
7285 ret = test_range_bit(io_tree, offset, range_end,
7286 EXTENT_DELALLOC, 0, NULL);
7293 btrfs_release_path(path);
7296 * look for other files referencing this extent, if we
7297 * find any we must cow
7300 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7301 key.offset - backref_offset, disk_bytenr);
7308 * adjust disk_bytenr and num_bytes to cover just the bytes
7309 * in this extent we are about to write. If there
7310 * are any csums in that range we have to cow in order
7311 * to keep the csums correct
7313 disk_bytenr += backref_offset;
7314 disk_bytenr += offset - key.offset;
7315 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7318 * all of the above have passed, it is safe to overwrite this extent
7324 btrfs_free_path(path);
7328 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7329 struct extent_state **cached_state, int writing)
7331 struct btrfs_ordered_extent *ordered;
7335 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7338 * We're concerned with the entire range that we're going to be
7339 * doing DIO to, so we need to make sure there's no ordered
7340 * extents in this range.
7342 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7343 lockend - lockstart + 1);
7346 * We need to make sure there are no buffered pages in this
7347 * range either, we could have raced between the invalidate in
7348 * generic_file_direct_write and locking the extent. The
7349 * invalidate needs to happen so that reads after a write do not
7353 (!writing || !filemap_range_has_page(inode->i_mapping,
7354 lockstart, lockend)))
7357 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7362 * If we are doing a DIO read and the ordered extent we
7363 * found is for a buffered write, we can not wait for it
7364 * to complete and retry, because if we do so we can
7365 * deadlock with concurrent buffered writes on page
7366 * locks. This happens only if our DIO read covers more
7367 * than one extent map, if at this point has already
7368 * created an ordered extent for a previous extent map
7369 * and locked its range in the inode's io tree, and a
7370 * concurrent write against that previous extent map's
7371 * range and this range started (we unlock the ranges
7372 * in the io tree only when the bios complete and
7373 * buffered writes always lock pages before attempting
7374 * to lock range in the io tree).
7377 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7378 btrfs_start_ordered_extent(inode, ordered, 1);
7381 btrfs_put_ordered_extent(ordered);
7384 * We could trigger writeback for this range (and wait
7385 * for it to complete) and then invalidate the pages for
7386 * this range (through invalidate_inode_pages2_range()),
7387 * but that can lead us to a deadlock with a concurrent
7388 * call to readpages() (a buffered read or a defrag call
7389 * triggered a readahead) on a page lock due to an
7390 * ordered dio extent we created before but did not have
7391 * yet a corresponding bio submitted (whence it can not
7392 * complete), which makes readpages() wait for that
7393 * ordered extent to complete while holding a lock on
7408 /* The callers of this must take lock_extent() */
7409 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
7410 u64 orig_start, u64 block_start,
7411 u64 block_len, u64 orig_block_len,
7412 u64 ram_bytes, int compress_type,
7415 struct extent_map_tree *em_tree;
7416 struct extent_map *em;
7417 struct btrfs_root *root = BTRFS_I(inode)->root;
7420 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7421 type == BTRFS_ORDERED_COMPRESSED ||
7422 type == BTRFS_ORDERED_NOCOW ||
7423 type == BTRFS_ORDERED_REGULAR);
7425 em_tree = &BTRFS_I(inode)->extent_tree;
7426 em = alloc_extent_map();
7428 return ERR_PTR(-ENOMEM);
7431 em->orig_start = orig_start;
7433 em->block_len = block_len;
7434 em->block_start = block_start;
7435 em->bdev = root->fs_info->fs_devices->latest_bdev;
7436 em->orig_block_len = orig_block_len;
7437 em->ram_bytes = ram_bytes;
7438 em->generation = -1;
7439 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7440 if (type == BTRFS_ORDERED_PREALLOC) {
7441 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7442 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7443 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7444 em->compress_type = compress_type;
7448 btrfs_drop_extent_cache(BTRFS_I(inode), em->start,
7449 em->start + em->len - 1, 0);
7450 write_lock(&em_tree->lock);
7451 ret = add_extent_mapping(em_tree, em, 1);
7452 write_unlock(&em_tree->lock);
7454 * The caller has taken lock_extent(), who could race with us
7457 } while (ret == -EEXIST);
7460 free_extent_map(em);
7461 return ERR_PTR(ret);
7464 /* em got 2 refs now, callers needs to do free_extent_map once. */
7469 static int btrfs_get_blocks_direct_read(struct extent_map *em,
7470 struct buffer_head *bh_result,
7471 struct inode *inode,
7474 if (em->block_start == EXTENT_MAP_HOLE ||
7475 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7478 len = min(len, em->len - (start - em->start));
7480 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7482 bh_result->b_size = len;
7483 bh_result->b_bdev = em->bdev;
7484 set_buffer_mapped(bh_result);
7489 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7490 struct buffer_head *bh_result,
7491 struct inode *inode,
7492 struct btrfs_dio_data *dio_data,
7495 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7496 struct extent_map *em = *map;
7500 * We don't allocate a new extent in the following cases
7502 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7504 * 2) The extent is marked as PREALLOC. We're good to go here and can
7505 * just use the extent.
7508 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7509 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7510 em->block_start != EXTENT_MAP_HOLE)) {
7512 u64 block_start, orig_start, orig_block_len, ram_bytes;
7514 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7515 type = BTRFS_ORDERED_PREALLOC;
7517 type = BTRFS_ORDERED_NOCOW;
7518 len = min(len, em->len - (start - em->start));
7519 block_start = em->block_start + (start - em->start);
7521 if (can_nocow_extent(inode, start, &len, &orig_start,
7522 &orig_block_len, &ram_bytes) == 1 &&
7523 btrfs_inc_nocow_writers(fs_info, block_start)) {
7524 struct extent_map *em2;
7526 em2 = btrfs_create_dio_extent(inode, start, len,
7527 orig_start, block_start,
7528 len, orig_block_len,
7530 btrfs_dec_nocow_writers(fs_info, block_start);
7531 if (type == BTRFS_ORDERED_PREALLOC) {
7532 free_extent_map(em);
7536 if (em2 && IS_ERR(em2)) {
7541 * For inode marked NODATACOW or extent marked PREALLOC,
7542 * use the existing or preallocated extent, so does not
7543 * need to adjust btrfs_space_info's bytes_may_use.
7545 btrfs_free_reserved_data_space_noquota(inode, start,
7551 /* this will cow the extent */
7552 len = bh_result->b_size;
7553 free_extent_map(em);
7554 *map = em = btrfs_new_extent_direct(inode, start, len);
7560 len = min(len, em->len - (start - em->start));
7563 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7565 bh_result->b_size = len;
7566 bh_result->b_bdev = em->bdev;
7567 set_buffer_mapped(bh_result);
7569 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7570 set_buffer_new(bh_result);
7573 * Need to update the i_size under the extent lock so buffered
7574 * readers will get the updated i_size when we unlock.
7576 if (!dio_data->overwrite && start + len > i_size_read(inode))
7577 i_size_write(inode, start + len);
7579 WARN_ON(dio_data->reserve < len);
7580 dio_data->reserve -= len;
7581 dio_data->unsubmitted_oe_range_end = start + len;
7582 current->journal_info = dio_data;
7587 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7588 struct buffer_head *bh_result, int create)
7590 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7591 struct extent_map *em;
7592 struct extent_state *cached_state = NULL;
7593 struct btrfs_dio_data *dio_data = NULL;
7594 u64 start = iblock << inode->i_blkbits;
7595 u64 lockstart, lockend;
7596 u64 len = bh_result->b_size;
7597 int unlock_bits = EXTENT_LOCKED;
7601 unlock_bits |= EXTENT_DIRTY;
7603 len = min_t(u64, len, fs_info->sectorsize);
7606 lockend = start + len - 1;
7608 if (current->journal_info) {
7610 * Need to pull our outstanding extents and set journal_info to NULL so
7611 * that anything that needs to check if there's a transaction doesn't get
7614 dio_data = current->journal_info;
7615 current->journal_info = NULL;
7619 * If this errors out it's because we couldn't invalidate pagecache for
7620 * this range and we need to fallback to buffered.
7622 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7628 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
7635 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7636 * io. INLINE is special, and we could probably kludge it in here, but
7637 * it's still buffered so for safety lets just fall back to the generic
7640 * For COMPRESSED we _have_ to read the entire extent in so we can
7641 * decompress it, so there will be buffering required no matter what we
7642 * do, so go ahead and fallback to buffered.
7644 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7645 * to buffered IO. Don't blame me, this is the price we pay for using
7648 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7649 em->block_start == EXTENT_MAP_INLINE) {
7650 free_extent_map(em);
7656 ret = btrfs_get_blocks_direct_write(&em, bh_result, inode,
7657 dio_data, start, len);
7661 /* clear and unlock the entire range */
7662 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7663 unlock_bits, 1, 0, &cached_state);
7665 ret = btrfs_get_blocks_direct_read(em, bh_result, inode,
7667 /* Can be negative only if we read from a hole */
7670 free_extent_map(em);
7674 * We need to unlock only the end area that we aren't using.
7675 * The rest is going to be unlocked by the endio routine.
7677 lockstart = start + bh_result->b_size;
7678 if (lockstart < lockend) {
7679 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7680 lockend, unlock_bits, 1, 0,
7683 free_extent_state(cached_state);
7687 free_extent_map(em);
7692 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7693 unlock_bits, 1, 0, &cached_state);
7696 current->journal_info = dio_data;
7700 static inline blk_status_t submit_dio_repair_bio(struct inode *inode,
7704 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7707 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7709 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR);
7713 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
7718 static int btrfs_check_dio_repairable(struct inode *inode,
7719 struct bio *failed_bio,
7720 struct io_failure_record *failrec,
7723 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7726 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
7727 if (num_copies == 1) {
7729 * we only have a single copy of the data, so don't bother with
7730 * all the retry and error correction code that follows. no
7731 * matter what the error is, it is very likely to persist.
7733 btrfs_debug(fs_info,
7734 "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
7735 num_copies, failrec->this_mirror, failed_mirror);
7739 failrec->failed_mirror = failed_mirror;
7740 failrec->this_mirror++;
7741 if (failrec->this_mirror == failed_mirror)
7742 failrec->this_mirror++;
7744 if (failrec->this_mirror > num_copies) {
7745 btrfs_debug(fs_info,
7746 "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
7747 num_copies, failrec->this_mirror, failed_mirror);
7754 static blk_status_t dio_read_error(struct inode *inode, struct bio *failed_bio,
7755 struct page *page, unsigned int pgoff,
7756 u64 start, u64 end, int failed_mirror,
7757 bio_end_io_t *repair_endio, void *repair_arg)
7759 struct io_failure_record *failrec;
7760 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7761 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7764 unsigned int read_mode = 0;
7767 blk_status_t status;
7768 struct bio_vec bvec;
7770 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
7772 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7774 return errno_to_blk_status(ret);
7776 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7779 free_io_failure(failure_tree, io_tree, failrec);
7780 return BLK_STS_IOERR;
7783 segs = bio_segments(failed_bio);
7784 bio_get_first_bvec(failed_bio, &bvec);
7786 (bvec.bv_len > btrfs_inode_sectorsize(inode)))
7787 read_mode |= REQ_FAILFAST_DEV;
7789 isector = start - btrfs_io_bio(failed_bio)->logical;
7790 isector >>= inode->i_sb->s_blocksize_bits;
7791 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7792 pgoff, isector, repair_endio, repair_arg);
7793 bio->bi_opf = REQ_OP_READ | read_mode;
7795 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7796 "repair DIO read error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d",
7797 read_mode, failrec->this_mirror, failrec->in_validation);
7799 status = submit_dio_repair_bio(inode, bio, failrec->this_mirror);
7801 free_io_failure(failure_tree, io_tree, failrec);
7808 struct btrfs_retry_complete {
7809 struct completion done;
7810 struct inode *inode;
7815 static void btrfs_retry_endio_nocsum(struct bio *bio)
7817 struct btrfs_retry_complete *done = bio->bi_private;
7818 struct inode *inode = done->inode;
7819 struct bio_vec *bvec;
7820 struct extent_io_tree *io_tree, *failure_tree;
7826 ASSERT(bio->bi_vcnt == 1);
7827 io_tree = &BTRFS_I(inode)->io_tree;
7828 failure_tree = &BTRFS_I(inode)->io_failure_tree;
7829 ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(inode));
7832 ASSERT(!bio_flagged(bio, BIO_CLONED));
7833 bio_for_each_segment_all(bvec, bio, i)
7834 clean_io_failure(BTRFS_I(inode)->root->fs_info, failure_tree,
7835 io_tree, done->start, bvec->bv_page,
7836 btrfs_ino(BTRFS_I(inode)), 0);
7838 complete(&done->done);
7842 static blk_status_t __btrfs_correct_data_nocsum(struct inode *inode,
7843 struct btrfs_io_bio *io_bio)
7845 struct btrfs_fs_info *fs_info;
7846 struct bio_vec bvec;
7847 struct bvec_iter iter;
7848 struct btrfs_retry_complete done;
7854 blk_status_t err = BLK_STS_OK;
7856 fs_info = BTRFS_I(inode)->root->fs_info;
7857 sectorsize = fs_info->sectorsize;
7859 start = io_bio->logical;
7861 io_bio->bio.bi_iter = io_bio->iter;
7863 bio_for_each_segment(bvec, &io_bio->bio, iter) {
7864 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7865 pgoff = bvec.bv_offset;
7867 next_block_or_try_again:
7870 init_completion(&done.done);
7872 ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
7873 pgoff, start, start + sectorsize - 1,
7875 btrfs_retry_endio_nocsum, &done);
7881 wait_for_completion_io(&done.done);
7883 if (!done.uptodate) {
7884 /* We might have another mirror, so try again */
7885 goto next_block_or_try_again;
7889 start += sectorsize;
7893 pgoff += sectorsize;
7894 ASSERT(pgoff < PAGE_SIZE);
7895 goto next_block_or_try_again;
7902 static void btrfs_retry_endio(struct bio *bio)
7904 struct btrfs_retry_complete *done = bio->bi_private;
7905 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7906 struct extent_io_tree *io_tree, *failure_tree;
7907 struct inode *inode = done->inode;
7908 struct bio_vec *bvec;
7918 ASSERT(bio->bi_vcnt == 1);
7919 ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(done->inode));
7921 io_tree = &BTRFS_I(inode)->io_tree;
7922 failure_tree = &BTRFS_I(inode)->io_failure_tree;
7924 ASSERT(!bio_flagged(bio, BIO_CLONED));
7925 bio_for_each_segment_all(bvec, bio, i) {
7926 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
7927 bvec->bv_offset, done->start,
7930 clean_io_failure(BTRFS_I(inode)->root->fs_info,
7931 failure_tree, io_tree, done->start,
7933 btrfs_ino(BTRFS_I(inode)),
7939 done->uptodate = uptodate;
7941 complete(&done->done);
7945 static blk_status_t __btrfs_subio_endio_read(struct inode *inode,
7946 struct btrfs_io_bio *io_bio, blk_status_t err)
7948 struct btrfs_fs_info *fs_info;
7949 struct bio_vec bvec;
7950 struct bvec_iter iter;
7951 struct btrfs_retry_complete done;
7958 bool uptodate = (err == 0);
7960 blk_status_t status;
7962 fs_info = BTRFS_I(inode)->root->fs_info;
7963 sectorsize = fs_info->sectorsize;
7966 start = io_bio->logical;
7968 io_bio->bio.bi_iter = io_bio->iter;
7970 bio_for_each_segment(bvec, &io_bio->bio, iter) {
7971 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7973 pgoff = bvec.bv_offset;
7976 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
7977 ret = __readpage_endio_check(inode, io_bio, csum_pos,
7978 bvec.bv_page, pgoff, start, sectorsize);
7985 init_completion(&done.done);
7987 status = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
7988 pgoff, start, start + sectorsize - 1,
7989 io_bio->mirror_num, btrfs_retry_endio,
7996 wait_for_completion_io(&done.done);
7998 if (!done.uptodate) {
7999 /* We might have another mirror, so try again */
8003 offset += sectorsize;
8004 start += sectorsize;
8010 pgoff += sectorsize;
8011 ASSERT(pgoff < PAGE_SIZE);
8019 static blk_status_t btrfs_subio_endio_read(struct inode *inode,
8020 struct btrfs_io_bio *io_bio, blk_status_t err)
8022 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8026 return __btrfs_correct_data_nocsum(inode, io_bio);
8030 return __btrfs_subio_endio_read(inode, io_bio, err);
8034 static void btrfs_endio_direct_read(struct bio *bio)
8036 struct btrfs_dio_private *dip = bio->bi_private;
8037 struct inode *inode = dip->inode;
8038 struct bio *dio_bio;
8039 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8040 blk_status_t err = bio->bi_status;
8042 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8043 err = btrfs_subio_endio_read(inode, io_bio, err);
8045 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8046 dip->logical_offset + dip->bytes - 1);
8047 dio_bio = dip->dio_bio;
8051 dio_bio->bi_status = err;
8052 dio_end_io(dio_bio);
8055 io_bio->end_io(io_bio, blk_status_to_errno(err));
8059 static void __endio_write_update_ordered(struct inode *inode,
8060 const u64 offset, const u64 bytes,
8061 const bool uptodate)
8063 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8064 struct btrfs_ordered_extent *ordered = NULL;
8065 struct btrfs_workqueue *wq;
8066 btrfs_work_func_t func;
8067 u64 ordered_offset = offset;
8068 u64 ordered_bytes = bytes;
8071 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
8072 wq = fs_info->endio_freespace_worker;
8073 func = btrfs_freespace_write_helper;
8075 wq = fs_info->endio_write_workers;
8076 func = btrfs_endio_write_helper;
8079 while (ordered_offset < offset + bytes) {
8080 last_offset = ordered_offset;
8081 if (btrfs_dec_test_first_ordered_pending(inode, &ordered,
8085 btrfs_init_work(&ordered->work, func,
8088 btrfs_queue_work(wq, &ordered->work);
8091 * If btrfs_dec_test_ordered_pending does not find any ordered
8092 * extent in the range, we can exit.
8094 if (ordered_offset == last_offset)
8097 * Our bio might span multiple ordered extents. In this case
8098 * we keep goin until we have accounted the whole dio.
8100 if (ordered_offset < offset + bytes) {
8101 ordered_bytes = offset + bytes - ordered_offset;
8107 static void btrfs_endio_direct_write(struct bio *bio)
8109 struct btrfs_dio_private *dip = bio->bi_private;
8110 struct bio *dio_bio = dip->dio_bio;
8112 __endio_write_update_ordered(dip->inode, dip->logical_offset,
8113 dip->bytes, !bio->bi_status);
8117 dio_bio->bi_status = bio->bi_status;
8118 dio_end_io(dio_bio);
8122 static blk_status_t btrfs_submit_bio_start_direct_io(void *private_data,
8123 struct bio *bio, u64 offset)
8125 struct inode *inode = private_data;
8127 ret = btrfs_csum_one_bio(inode, bio, offset, 1);
8128 BUG_ON(ret); /* -ENOMEM */
8132 static void btrfs_end_dio_bio(struct bio *bio)
8134 struct btrfs_dio_private *dip = bio->bi_private;
8135 blk_status_t err = bio->bi_status;
8138 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8139 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8140 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8142 (unsigned long long)bio->bi_iter.bi_sector,
8143 bio->bi_iter.bi_size, err);
8145 if (dip->subio_endio)
8146 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8150 * We want to perceive the errors flag being set before
8151 * decrementing the reference count. We don't need a barrier
8152 * since atomic operations with a return value are fully
8153 * ordered as per atomic_t.txt
8158 /* if there are more bios still pending for this dio, just exit */
8159 if (!atomic_dec_and_test(&dip->pending_bios))
8163 bio_io_error(dip->orig_bio);
8165 dip->dio_bio->bi_status = BLK_STS_OK;
8166 bio_endio(dip->orig_bio);
8172 static inline blk_status_t btrfs_lookup_and_bind_dio_csum(struct inode *inode,
8173 struct btrfs_dio_private *dip,
8177 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8178 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8182 * We load all the csum data we need when we submit
8183 * the first bio to reduce the csum tree search and
8186 if (dip->logical_offset == file_offset) {
8187 ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio,
8193 if (bio == dip->orig_bio)
8196 file_offset -= dip->logical_offset;
8197 file_offset >>= inode->i_sb->s_blocksize_bits;
8198 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8203 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
8204 struct inode *inode, u64 file_offset, int async_submit)
8206 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8207 struct btrfs_dio_private *dip = bio->bi_private;
8208 bool write = bio_op(bio) == REQ_OP_WRITE;
8211 /* Check btrfs_submit_bio_hook() for rules about async submit. */
8213 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8216 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8221 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
8224 if (write && async_submit) {
8225 ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0,
8227 btrfs_submit_bio_start_direct_io);
8231 * If we aren't doing async submit, calculate the csum of the
8234 ret = btrfs_csum_one_bio(inode, bio, file_offset, 1);
8238 ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio,
8244 ret = btrfs_map_bio(fs_info, bio, 0, 0);
8249 static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip)
8251 struct inode *inode = dip->inode;
8252 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8254 struct bio *orig_bio = dip->orig_bio;
8255 u64 start_sector = orig_bio->bi_iter.bi_sector;
8256 u64 file_offset = dip->logical_offset;
8258 int async_submit = 0;
8260 int clone_offset = 0;
8263 blk_status_t status;
8265 map_length = orig_bio->bi_iter.bi_size;
8266 submit_len = map_length;
8267 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9,
8268 &map_length, NULL, 0);
8272 if (map_length >= submit_len) {
8274 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8278 /* async crcs make it difficult to collect full stripe writes. */
8279 if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8285 ASSERT(map_length <= INT_MAX);
8286 atomic_inc(&dip->pending_bios);
8288 clone_len = min_t(int, submit_len, map_length);
8291 * This will never fail as it's passing GPF_NOFS and
8292 * the allocation is backed by btrfs_bioset.
8294 bio = btrfs_bio_clone_partial(orig_bio, clone_offset,
8296 bio->bi_private = dip;
8297 bio->bi_end_io = btrfs_end_dio_bio;
8298 btrfs_io_bio(bio)->logical = file_offset;
8300 ASSERT(submit_len >= clone_len);
8301 submit_len -= clone_len;
8302 if (submit_len == 0)
8306 * Increase the count before we submit the bio so we know
8307 * the end IO handler won't happen before we increase the
8308 * count. Otherwise, the dip might get freed before we're
8309 * done setting it up.
8311 atomic_inc(&dip->pending_bios);
8313 status = btrfs_submit_dio_bio(bio, inode, file_offset,
8317 atomic_dec(&dip->pending_bios);
8321 clone_offset += clone_len;
8322 start_sector += clone_len >> 9;
8323 file_offset += clone_len;
8325 map_length = submit_len;
8326 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio),
8327 start_sector << 9, &map_length, NULL, 0);
8330 } while (submit_len > 0);
8333 status = btrfs_submit_dio_bio(bio, inode, file_offset, async_submit);
8341 * Before atomic variable goto zero, we must make sure dip->errors is
8342 * perceived to be set. This ordering is ensured by the fact that an
8343 * atomic operations with a return value are fully ordered as per
8346 if (atomic_dec_and_test(&dip->pending_bios))
8347 bio_io_error(dip->orig_bio);
8349 /* bio_end_io() will handle error, so we needn't return it */
8353 static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode,
8356 struct btrfs_dio_private *dip = NULL;
8357 struct bio *bio = NULL;
8358 struct btrfs_io_bio *io_bio;
8359 bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
8362 bio = btrfs_bio_clone(dio_bio);
8364 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8370 dip->private = dio_bio->bi_private;
8372 dip->logical_offset = file_offset;
8373 dip->bytes = dio_bio->bi_iter.bi_size;
8374 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8375 bio->bi_private = dip;
8376 dip->orig_bio = bio;
8377 dip->dio_bio = dio_bio;
8378 atomic_set(&dip->pending_bios, 0);
8379 io_bio = btrfs_io_bio(bio);
8380 io_bio->logical = file_offset;
8383 bio->bi_end_io = btrfs_endio_direct_write;
8385 bio->bi_end_io = btrfs_endio_direct_read;
8386 dip->subio_endio = btrfs_subio_endio_read;
8390 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8391 * even if we fail to submit a bio, because in such case we do the
8392 * corresponding error handling below and it must not be done a second
8393 * time by btrfs_direct_IO().
8396 struct btrfs_dio_data *dio_data = current->journal_info;
8398 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8400 dio_data->unsubmitted_oe_range_start =
8401 dio_data->unsubmitted_oe_range_end;
8404 ret = btrfs_submit_direct_hook(dip);
8409 io_bio->end_io(io_bio, ret);
8413 * If we arrived here it means either we failed to submit the dip
8414 * or we either failed to clone the dio_bio or failed to allocate the
8415 * dip. If we cloned the dio_bio and allocated the dip, we can just
8416 * call bio_endio against our io_bio so that we get proper resource
8417 * cleanup if we fail to submit the dip, otherwise, we must do the
8418 * same as btrfs_endio_direct_[write|read] because we can't call these
8419 * callbacks - they require an allocated dip and a clone of dio_bio.
8424 * The end io callbacks free our dip, do the final put on bio
8425 * and all the cleanup and final put for dio_bio (through
8432 __endio_write_update_ordered(inode,
8434 dio_bio->bi_iter.bi_size,
8437 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8438 file_offset + dio_bio->bi_iter.bi_size - 1);
8440 dio_bio->bi_status = BLK_STS_IOERR;
8442 * Releases and cleans up our dio_bio, no need to bio_put()
8443 * nor bio_endio()/bio_io_error() against dio_bio.
8445 dio_end_io(dio_bio);
8452 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
8453 const struct iov_iter *iter, loff_t offset)
8457 unsigned int blocksize_mask = fs_info->sectorsize - 1;
8458 ssize_t retval = -EINVAL;
8460 if (offset & blocksize_mask)
8463 if (iov_iter_alignment(iter) & blocksize_mask)
8466 /* If this is a write we don't need to check anymore */
8467 if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter))
8470 * Check to make sure we don't have duplicate iov_base's in this
8471 * iovec, if so return EINVAL, otherwise we'll get csum errors
8472 * when reading back.
8474 for (seg = 0; seg < iter->nr_segs; seg++) {
8475 for (i = seg + 1; i < iter->nr_segs; i++) {
8476 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8485 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8487 struct file *file = iocb->ki_filp;
8488 struct inode *inode = file->f_mapping->host;
8489 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8490 struct btrfs_dio_data dio_data = { 0 };
8491 struct extent_changeset *data_reserved = NULL;
8492 loff_t offset = iocb->ki_pos;
8496 bool relock = false;
8499 if (check_direct_IO(fs_info, iter, offset))
8502 inode_dio_begin(inode);
8505 * The generic stuff only does filemap_write_and_wait_range, which
8506 * isn't enough if we've written compressed pages to this area, so
8507 * we need to flush the dirty pages again to make absolutely sure
8508 * that any outstanding dirty pages are on disk.
8510 count = iov_iter_count(iter);
8511 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8512 &BTRFS_I(inode)->runtime_flags))
8513 filemap_fdatawrite_range(inode->i_mapping, offset,
8514 offset + count - 1);
8516 if (iov_iter_rw(iter) == WRITE) {
8518 * If the write DIO is beyond the EOF, we need update
8519 * the isize, but it is protected by i_mutex. So we can
8520 * not unlock the i_mutex at this case.
8522 if (offset + count <= inode->i_size) {
8523 dio_data.overwrite = 1;
8524 inode_unlock(inode);
8526 } else if (iocb->ki_flags & IOCB_NOWAIT) {
8530 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
8536 * We need to know how many extents we reserved so that we can
8537 * do the accounting properly if we go over the number we
8538 * originally calculated. Abuse current->journal_info for this.
8540 dio_data.reserve = round_up(count,
8541 fs_info->sectorsize);
8542 dio_data.unsubmitted_oe_range_start = (u64)offset;
8543 dio_data.unsubmitted_oe_range_end = (u64)offset;
8544 current->journal_info = &dio_data;
8545 down_read(&BTRFS_I(inode)->dio_sem);
8546 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8547 &BTRFS_I(inode)->runtime_flags)) {
8548 inode_dio_end(inode);
8549 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8553 ret = __blockdev_direct_IO(iocb, inode,
8554 fs_info->fs_devices->latest_bdev,
8555 iter, btrfs_get_blocks_direct, NULL,
8556 btrfs_submit_direct, flags);
8557 if (iov_iter_rw(iter) == WRITE) {
8558 up_read(&BTRFS_I(inode)->dio_sem);
8559 current->journal_info = NULL;
8560 if (ret < 0 && ret != -EIOCBQUEUED) {
8561 if (dio_data.reserve)
8562 btrfs_delalloc_release_space(inode, data_reserved,
8563 offset, dio_data.reserve, true);
8565 * On error we might have left some ordered extents
8566 * without submitting corresponding bios for them, so
8567 * cleanup them up to avoid other tasks getting them
8568 * and waiting for them to complete forever.
8570 if (dio_data.unsubmitted_oe_range_start <
8571 dio_data.unsubmitted_oe_range_end)
8572 __endio_write_update_ordered(inode,
8573 dio_data.unsubmitted_oe_range_start,
8574 dio_data.unsubmitted_oe_range_end -
8575 dio_data.unsubmitted_oe_range_start,
8577 } else if (ret >= 0 && (size_t)ret < count)
8578 btrfs_delalloc_release_space(inode, data_reserved,
8579 offset, count - (size_t)ret, true);
8580 btrfs_delalloc_release_extents(BTRFS_I(inode), count, false);
8584 inode_dio_end(inode);
8588 extent_changeset_free(data_reserved);
8592 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8594 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8595 __u64 start, __u64 len)
8599 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8603 return extent_fiemap(inode, fieinfo, start, len);
8606 int btrfs_readpage(struct file *file, struct page *page)
8608 struct extent_io_tree *tree;
8609 tree = &BTRFS_I(page->mapping->host)->io_tree;
8610 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8613 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8615 struct inode *inode = page->mapping->host;
8618 if (current->flags & PF_MEMALLOC) {
8619 redirty_page_for_writepage(wbc, page);
8625 * If we are under memory pressure we will call this directly from the
8626 * VM, we need to make sure we have the inode referenced for the ordered
8627 * extent. If not just return like we didn't do anything.
8629 if (!igrab(inode)) {
8630 redirty_page_for_writepage(wbc, page);
8631 return AOP_WRITEPAGE_ACTIVATE;
8633 ret = extent_write_full_page(page, wbc);
8634 btrfs_add_delayed_iput(inode);
8638 static int btrfs_writepages(struct address_space *mapping,
8639 struct writeback_control *wbc)
8641 return extent_writepages(mapping, wbc);
8645 btrfs_readpages(struct file *file, struct address_space *mapping,
8646 struct list_head *pages, unsigned nr_pages)
8648 return extent_readpages(mapping, pages, nr_pages);
8651 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8653 int ret = try_release_extent_mapping(page, gfp_flags);
8655 ClearPagePrivate(page);
8656 set_page_private(page, 0);
8662 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8664 if (PageWriteback(page) || PageDirty(page))
8666 return __btrfs_releasepage(page, gfp_flags);
8669 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8670 unsigned int length)
8672 struct inode *inode = page->mapping->host;
8673 struct extent_io_tree *tree;
8674 struct btrfs_ordered_extent *ordered;
8675 struct extent_state *cached_state = NULL;
8676 u64 page_start = page_offset(page);
8677 u64 page_end = page_start + PAGE_SIZE - 1;
8680 int inode_evicting = inode->i_state & I_FREEING;
8683 * we have the page locked, so new writeback can't start,
8684 * and the dirty bit won't be cleared while we are here.
8686 * Wait for IO on this page so that we can safely clear
8687 * the PagePrivate2 bit and do ordered accounting
8689 wait_on_page_writeback(page);
8691 tree = &BTRFS_I(inode)->io_tree;
8693 btrfs_releasepage(page, GFP_NOFS);
8697 if (!inode_evicting)
8698 lock_extent_bits(tree, page_start, page_end, &cached_state);
8701 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
8702 page_end - start + 1);
8704 end = min(page_end, ordered->file_offset + ordered->len - 1);
8706 * IO on this page will never be started, so we need
8707 * to account for any ordered extents now
8709 if (!inode_evicting)
8710 clear_extent_bit(tree, start, end,
8711 EXTENT_DIRTY | EXTENT_DELALLOC |
8712 EXTENT_DELALLOC_NEW |
8713 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8714 EXTENT_DEFRAG, 1, 0, &cached_state);
8716 * whoever cleared the private bit is responsible
8717 * for the finish_ordered_io
8719 if (TestClearPagePrivate2(page)) {
8720 struct btrfs_ordered_inode_tree *tree;
8723 tree = &BTRFS_I(inode)->ordered_tree;
8725 spin_lock_irq(&tree->lock);
8726 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8727 new_len = start - ordered->file_offset;
8728 if (new_len < ordered->truncated_len)
8729 ordered->truncated_len = new_len;
8730 spin_unlock_irq(&tree->lock);
8732 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8734 end - start + 1, 1))
8735 btrfs_finish_ordered_io(ordered);
8737 btrfs_put_ordered_extent(ordered);
8738 if (!inode_evicting) {
8739 cached_state = NULL;
8740 lock_extent_bits(tree, start, end,
8745 if (start < page_end)
8750 * Qgroup reserved space handler
8751 * Page here will be either
8752 * 1) Already written to disk
8753 * In this case, its reserved space is released from data rsv map
8754 * and will be freed by delayed_ref handler finally.
8755 * So even we call qgroup_free_data(), it won't decrease reserved
8757 * 2) Not written to disk
8758 * This means the reserved space should be freed here. However,
8759 * if a truncate invalidates the page (by clearing PageDirty)
8760 * and the page is accounted for while allocating extent
8761 * in btrfs_check_data_free_space() we let delayed_ref to
8762 * free the entire extent.
8764 if (PageDirty(page))
8765 btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
8766 if (!inode_evicting) {
8767 clear_extent_bit(tree, page_start, page_end,
8768 EXTENT_LOCKED | EXTENT_DIRTY |
8769 EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
8770 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
8773 __btrfs_releasepage(page, GFP_NOFS);
8776 ClearPageChecked(page);
8777 if (PagePrivate(page)) {
8778 ClearPagePrivate(page);
8779 set_page_private(page, 0);
8785 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8786 * called from a page fault handler when a page is first dirtied. Hence we must
8787 * be careful to check for EOF conditions here. We set the page up correctly
8788 * for a written page which means we get ENOSPC checking when writing into
8789 * holes and correct delalloc and unwritten extent mapping on filesystems that
8790 * support these features.
8792 * We are not allowed to take the i_mutex here so we have to play games to
8793 * protect against truncate races as the page could now be beyond EOF. Because
8794 * truncate_setsize() writes the inode size before removing pages, once we have
8795 * the page lock we can determine safely if the page is beyond EOF. If it is not
8796 * beyond EOF, then the page is guaranteed safe against truncation until we
8799 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8801 struct page *page = vmf->page;
8802 struct inode *inode = file_inode(vmf->vma->vm_file);
8803 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8804 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8805 struct btrfs_ordered_extent *ordered;
8806 struct extent_state *cached_state = NULL;
8807 struct extent_changeset *data_reserved = NULL;
8809 unsigned long zero_start;
8819 reserved_space = PAGE_SIZE;
8821 sb_start_pagefault(inode->i_sb);
8822 page_start = page_offset(page);
8823 page_end = page_start + PAGE_SIZE - 1;
8827 * Reserving delalloc space after obtaining the page lock can lead to
8828 * deadlock. For example, if a dirty page is locked by this function
8829 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8830 * dirty page write out, then the btrfs_writepage() function could
8831 * end up waiting indefinitely to get a lock on the page currently
8832 * being processed by btrfs_page_mkwrite() function.
8834 ret2 = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
8837 ret2 = file_update_time(vmf->vma->vm_file);
8841 ret = vmf_error(ret2);
8847 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8850 size = i_size_read(inode);
8852 if ((page->mapping != inode->i_mapping) ||
8853 (page_start >= size)) {
8854 /* page got truncated out from underneath us */
8857 wait_on_page_writeback(page);
8859 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8860 set_page_extent_mapped(page);
8863 * we can't set the delalloc bits if there are pending ordered
8864 * extents. Drop our locks and wait for them to finish
8866 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8869 unlock_extent_cached(io_tree, page_start, page_end,
8872 btrfs_start_ordered_extent(inode, ordered, 1);
8873 btrfs_put_ordered_extent(ordered);
8877 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8878 reserved_space = round_up(size - page_start,
8879 fs_info->sectorsize);
8880 if (reserved_space < PAGE_SIZE) {
8881 end = page_start + reserved_space - 1;
8882 btrfs_delalloc_release_space(inode, data_reserved,
8883 page_start, PAGE_SIZE - reserved_space,
8889 * page_mkwrite gets called when the page is firstly dirtied after it's
8890 * faulted in, but write(2) could also dirty a page and set delalloc
8891 * bits, thus in this case for space account reason, we still need to
8892 * clear any delalloc bits within this page range since we have to
8893 * reserve data&meta space before lock_page() (see above comments).
8895 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8896 EXTENT_DIRTY | EXTENT_DELALLOC |
8897 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8898 0, 0, &cached_state);
8900 ret2 = btrfs_set_extent_delalloc(inode, page_start, end, 0,
8903 unlock_extent_cached(io_tree, page_start, page_end,
8905 ret = VM_FAULT_SIGBUS;
8910 /* page is wholly or partially inside EOF */
8911 if (page_start + PAGE_SIZE > size)
8912 zero_start = size & ~PAGE_MASK;
8914 zero_start = PAGE_SIZE;
8916 if (zero_start != PAGE_SIZE) {
8918 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
8919 flush_dcache_page(page);
8922 ClearPageChecked(page);
8923 set_page_dirty(page);
8924 SetPageUptodate(page);
8926 BTRFS_I(inode)->last_trans = fs_info->generation;
8927 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8928 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8930 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8933 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, true);
8934 sb_end_pagefault(inode->i_sb);
8935 extent_changeset_free(data_reserved);
8936 return VM_FAULT_LOCKED;
8942 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, (ret != 0));
8943 btrfs_delalloc_release_space(inode, data_reserved, page_start,
8944 reserved_space, (ret != 0));
8946 sb_end_pagefault(inode->i_sb);
8947 extent_changeset_free(data_reserved);
8951 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
8953 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8954 struct btrfs_root *root = BTRFS_I(inode)->root;
8955 struct btrfs_block_rsv *rsv;
8957 struct btrfs_trans_handle *trans;
8958 u64 mask = fs_info->sectorsize - 1;
8959 u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
8961 if (!skip_writeback) {
8962 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8969 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8970 * things going on here:
8972 * 1) We need to reserve space to update our inode.
8974 * 2) We need to have something to cache all the space that is going to
8975 * be free'd up by the truncate operation, but also have some slack
8976 * space reserved in case it uses space during the truncate (thank you
8977 * very much snapshotting).
8979 * And we need these to be separate. The fact is we can use a lot of
8980 * space doing the truncate, and we have no earthly idea how much space
8981 * we will use, so we need the truncate reservation to be separate so it
8982 * doesn't end up using space reserved for updating the inode. We also
8983 * need to be able to stop the transaction and start a new one, which
8984 * means we need to be able to update the inode several times, and we
8985 * have no idea of knowing how many times that will be, so we can't just
8986 * reserve 1 item for the entirety of the operation, so that has to be
8987 * done separately as well.
8989 * So that leaves us with
8991 * 1) rsv - for the truncate reservation, which we will steal from the
8992 * transaction reservation.
8993 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8994 * updating the inode.
8996 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8999 rsv->size = min_size;
9003 * 1 for the truncate slack space
9004 * 1 for updating the inode.
9006 trans = btrfs_start_transaction(root, 2);
9007 if (IS_ERR(trans)) {
9008 ret = PTR_ERR(trans);
9012 /* Migrate the slack space for the truncate to our reserve */
9013 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
9018 * So if we truncate and then write and fsync we normally would just
9019 * write the extents that changed, which is a problem if we need to
9020 * first truncate that entire inode. So set this flag so we write out
9021 * all of the extents in the inode to the sync log so we're completely
9024 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9025 trans->block_rsv = rsv;
9028 ret = btrfs_truncate_inode_items(trans, root, inode,
9030 BTRFS_EXTENT_DATA_KEY);
9031 trans->block_rsv = &fs_info->trans_block_rsv;
9032 if (ret != -ENOSPC && ret != -EAGAIN)
9035 ret = btrfs_update_inode(trans, root, inode);
9039 btrfs_end_transaction(trans);
9040 btrfs_btree_balance_dirty(fs_info);
9042 trans = btrfs_start_transaction(root, 2);
9043 if (IS_ERR(trans)) {
9044 ret = PTR_ERR(trans);
9049 btrfs_block_rsv_release(fs_info, rsv, -1);
9050 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
9051 rsv, min_size, false);
9052 BUG_ON(ret); /* shouldn't happen */
9053 trans->block_rsv = rsv;
9057 * We can't call btrfs_truncate_block inside a trans handle as we could
9058 * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
9059 * we've truncated everything except the last little bit, and can do
9060 * btrfs_truncate_block and then update the disk_i_size.
9062 if (ret == NEED_TRUNCATE_BLOCK) {
9063 btrfs_end_transaction(trans);
9064 btrfs_btree_balance_dirty(fs_info);
9066 ret = btrfs_truncate_block(inode, inode->i_size, 0, 0);
9069 trans = btrfs_start_transaction(root, 1);
9070 if (IS_ERR(trans)) {
9071 ret = PTR_ERR(trans);
9074 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
9080 trans->block_rsv = &fs_info->trans_block_rsv;
9081 ret2 = btrfs_update_inode(trans, root, inode);
9085 ret2 = btrfs_end_transaction(trans);
9088 btrfs_btree_balance_dirty(fs_info);
9091 btrfs_free_block_rsv(fs_info, rsv);
9097 * create a new subvolume directory/inode (helper for the ioctl).
9099 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9100 struct btrfs_root *new_root,
9101 struct btrfs_root *parent_root,
9104 struct inode *inode;
9108 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9109 new_dirid, new_dirid,
9110 S_IFDIR | (~current_umask() & S_IRWXUGO),
9113 return PTR_ERR(inode);
9114 inode->i_op = &btrfs_dir_inode_operations;
9115 inode->i_fop = &btrfs_dir_file_operations;
9117 set_nlink(inode, 1);
9118 btrfs_i_size_write(BTRFS_I(inode), 0);
9119 unlock_new_inode(inode);
9121 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9123 btrfs_err(new_root->fs_info,
9124 "error inheriting subvolume %llu properties: %d",
9125 new_root->root_key.objectid, err);
9127 err = btrfs_update_inode(trans, new_root, inode);
9133 struct inode *btrfs_alloc_inode(struct super_block *sb)
9135 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
9136 struct btrfs_inode *ei;
9137 struct inode *inode;
9139 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL);
9146 ei->last_sub_trans = 0;
9147 ei->logged_trans = 0;
9148 ei->delalloc_bytes = 0;
9149 ei->new_delalloc_bytes = 0;
9150 ei->defrag_bytes = 0;
9151 ei->disk_i_size = 0;
9154 ei->index_cnt = (u64)-1;
9156 ei->last_unlink_trans = 0;
9157 ei->last_log_commit = 0;
9159 spin_lock_init(&ei->lock);
9160 ei->outstanding_extents = 0;
9161 if (sb->s_magic != BTRFS_TEST_MAGIC)
9162 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
9163 BTRFS_BLOCK_RSV_DELALLOC);
9164 ei->runtime_flags = 0;
9165 ei->prop_compress = BTRFS_COMPRESS_NONE;
9166 ei->defrag_compress = BTRFS_COMPRESS_NONE;
9168 ei->delayed_node = NULL;
9170 ei->i_otime.tv_sec = 0;
9171 ei->i_otime.tv_nsec = 0;
9173 inode = &ei->vfs_inode;
9174 extent_map_tree_init(&ei->extent_tree);
9175 extent_io_tree_init(&ei->io_tree, inode);
9176 extent_io_tree_init(&ei->io_failure_tree, inode);
9177 ei->io_tree.track_uptodate = 1;
9178 ei->io_failure_tree.track_uptodate = 1;
9179 atomic_set(&ei->sync_writers, 0);
9180 mutex_init(&ei->log_mutex);
9181 mutex_init(&ei->delalloc_mutex);
9182 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9183 INIT_LIST_HEAD(&ei->delalloc_inodes);
9184 INIT_LIST_HEAD(&ei->delayed_iput);
9185 RB_CLEAR_NODE(&ei->rb_node);
9186 init_rwsem(&ei->dio_sem);
9191 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9192 void btrfs_test_destroy_inode(struct inode *inode)
9194 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9195 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9199 static void btrfs_i_callback(struct rcu_head *head)
9201 struct inode *inode = container_of(head, struct inode, i_rcu);
9202 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9205 void btrfs_destroy_inode(struct inode *inode)
9207 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9208 struct btrfs_ordered_extent *ordered;
9209 struct btrfs_root *root = BTRFS_I(inode)->root;
9211 WARN_ON(!hlist_empty(&inode->i_dentry));
9212 WARN_ON(inode->i_data.nrpages);
9213 WARN_ON(BTRFS_I(inode)->block_rsv.reserved);
9214 WARN_ON(BTRFS_I(inode)->block_rsv.size);
9215 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9216 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9217 WARN_ON(BTRFS_I(inode)->new_delalloc_bytes);
9218 WARN_ON(BTRFS_I(inode)->csum_bytes);
9219 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9222 * This can happen where we create an inode, but somebody else also
9223 * created the same inode and we need to destroy the one we already
9230 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9235 "found ordered extent %llu %llu on inode cleanup",
9236 ordered->file_offset, ordered->len);
9237 btrfs_remove_ordered_extent(inode, ordered);
9238 btrfs_put_ordered_extent(ordered);
9239 btrfs_put_ordered_extent(ordered);
9242 btrfs_qgroup_check_reserved_leak(inode);
9243 inode_tree_del(inode);
9244 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9246 call_rcu(&inode->i_rcu, btrfs_i_callback);
9249 int btrfs_drop_inode(struct inode *inode)
9251 struct btrfs_root *root = BTRFS_I(inode)->root;
9256 /* the snap/subvol tree is on deleting */
9257 if (btrfs_root_refs(&root->root_item) == 0)
9260 return generic_drop_inode(inode);
9263 static void init_once(void *foo)
9265 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9267 inode_init_once(&ei->vfs_inode);
9270 void __cold btrfs_destroy_cachep(void)
9273 * Make sure all delayed rcu free inodes are flushed before we
9277 kmem_cache_destroy(btrfs_inode_cachep);
9278 kmem_cache_destroy(btrfs_trans_handle_cachep);
9279 kmem_cache_destroy(btrfs_path_cachep);
9280 kmem_cache_destroy(btrfs_free_space_cachep);
9283 int __init btrfs_init_cachep(void)
9285 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9286 sizeof(struct btrfs_inode), 0,
9287 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9289 if (!btrfs_inode_cachep)
9292 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9293 sizeof(struct btrfs_trans_handle), 0,
9294 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9295 if (!btrfs_trans_handle_cachep)
9298 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9299 sizeof(struct btrfs_path), 0,
9300 SLAB_MEM_SPREAD, NULL);
9301 if (!btrfs_path_cachep)
9304 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9305 sizeof(struct btrfs_free_space), 0,
9306 SLAB_MEM_SPREAD, NULL);
9307 if (!btrfs_free_space_cachep)
9312 btrfs_destroy_cachep();
9316 static int btrfs_getattr(const struct path *path, struct kstat *stat,
9317 u32 request_mask, unsigned int flags)
9320 struct inode *inode = d_inode(path->dentry);
9321 u32 blocksize = inode->i_sb->s_blocksize;
9322 u32 bi_flags = BTRFS_I(inode)->flags;
9324 stat->result_mask |= STATX_BTIME;
9325 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9326 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9327 if (bi_flags & BTRFS_INODE_APPEND)
9328 stat->attributes |= STATX_ATTR_APPEND;
9329 if (bi_flags & BTRFS_INODE_COMPRESS)
9330 stat->attributes |= STATX_ATTR_COMPRESSED;
9331 if (bi_flags & BTRFS_INODE_IMMUTABLE)
9332 stat->attributes |= STATX_ATTR_IMMUTABLE;
9333 if (bi_flags & BTRFS_INODE_NODUMP)
9334 stat->attributes |= STATX_ATTR_NODUMP;
9336 stat->attributes_mask |= (STATX_ATTR_APPEND |
9337 STATX_ATTR_COMPRESSED |
9338 STATX_ATTR_IMMUTABLE |
9341 generic_fillattr(inode, stat);
9342 stat->dev = BTRFS_I(inode)->root->anon_dev;
9344 spin_lock(&BTRFS_I(inode)->lock);
9345 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9346 spin_unlock(&BTRFS_I(inode)->lock);
9347 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9348 ALIGN(delalloc_bytes, blocksize)) >> 9;
9352 static int btrfs_rename_exchange(struct inode *old_dir,
9353 struct dentry *old_dentry,
9354 struct inode *new_dir,
9355 struct dentry *new_dentry)
9357 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9358 struct btrfs_trans_handle *trans;
9359 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9360 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9361 struct inode *new_inode = new_dentry->d_inode;
9362 struct inode *old_inode = old_dentry->d_inode;
9363 struct timespec64 ctime = current_time(old_inode);
9364 struct dentry *parent;
9365 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9366 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9371 bool root_log_pinned = false;
9372 bool dest_log_pinned = false;
9373 struct btrfs_log_ctx ctx_root;
9374 struct btrfs_log_ctx ctx_dest;
9375 bool sync_log_root = false;
9376 bool sync_log_dest = false;
9377 bool commit_transaction = false;
9379 /* we only allow rename subvolume link between subvolumes */
9380 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9383 btrfs_init_log_ctx(&ctx_root, old_inode);
9384 btrfs_init_log_ctx(&ctx_dest, new_inode);
9386 /* close the race window with snapshot create/destroy ioctl */
9387 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9388 down_read(&fs_info->subvol_sem);
9389 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9390 down_read(&fs_info->subvol_sem);
9393 * We want to reserve the absolute worst case amount of items. So if
9394 * both inodes are subvols and we need to unlink them then that would
9395 * require 4 item modifications, but if they are both normal inodes it
9396 * would require 5 item modifications, so we'll assume their normal
9397 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9398 * should cover the worst case number of items we'll modify.
9400 trans = btrfs_start_transaction(root, 12);
9401 if (IS_ERR(trans)) {
9402 ret = PTR_ERR(trans);
9407 * We need to find a free sequence number both in the source and
9408 * in the destination directory for the exchange.
9410 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9413 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9417 BTRFS_I(old_inode)->dir_index = 0ULL;
9418 BTRFS_I(new_inode)->dir_index = 0ULL;
9420 /* Reference for the source. */
9421 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9422 /* force full log commit if subvolume involved. */
9423 btrfs_set_log_full_commit(fs_info, trans);
9425 btrfs_pin_log_trans(root);
9426 root_log_pinned = true;
9427 ret = btrfs_insert_inode_ref(trans, dest,
9428 new_dentry->d_name.name,
9429 new_dentry->d_name.len,
9431 btrfs_ino(BTRFS_I(new_dir)),
9437 /* And now for the dest. */
9438 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9439 /* force full log commit if subvolume involved. */
9440 btrfs_set_log_full_commit(fs_info, trans);
9442 btrfs_pin_log_trans(dest);
9443 dest_log_pinned = true;
9444 ret = btrfs_insert_inode_ref(trans, root,
9445 old_dentry->d_name.name,
9446 old_dentry->d_name.len,
9448 btrfs_ino(BTRFS_I(old_dir)),
9454 /* Update inode version and ctime/mtime. */
9455 inode_inc_iversion(old_dir);
9456 inode_inc_iversion(new_dir);
9457 inode_inc_iversion(old_inode);
9458 inode_inc_iversion(new_inode);
9459 old_dir->i_ctime = old_dir->i_mtime = ctime;
9460 new_dir->i_ctime = new_dir->i_mtime = ctime;
9461 old_inode->i_ctime = ctime;
9462 new_inode->i_ctime = ctime;
9464 if (old_dentry->d_parent != new_dentry->d_parent) {
9465 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9466 BTRFS_I(old_inode), 1);
9467 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9468 BTRFS_I(new_inode), 1);
9471 /* src is a subvolume */
9472 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9473 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9474 ret = btrfs_unlink_subvol(trans, old_dir, root_objectid,
9475 old_dentry->d_name.name,
9476 old_dentry->d_name.len);
9477 } else { /* src is an inode */
9478 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9479 BTRFS_I(old_dentry->d_inode),
9480 old_dentry->d_name.name,
9481 old_dentry->d_name.len);
9483 ret = btrfs_update_inode(trans, root, old_inode);
9486 btrfs_abort_transaction(trans, ret);
9490 /* dest is a subvolume */
9491 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9492 root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9493 ret = btrfs_unlink_subvol(trans, new_dir, root_objectid,
9494 new_dentry->d_name.name,
9495 new_dentry->d_name.len);
9496 } else { /* dest is an inode */
9497 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9498 BTRFS_I(new_dentry->d_inode),
9499 new_dentry->d_name.name,
9500 new_dentry->d_name.len);
9502 ret = btrfs_update_inode(trans, dest, new_inode);
9505 btrfs_abort_transaction(trans, ret);
9509 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9510 new_dentry->d_name.name,
9511 new_dentry->d_name.len, 0, old_idx);
9513 btrfs_abort_transaction(trans, ret);
9517 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9518 old_dentry->d_name.name,
9519 old_dentry->d_name.len, 0, new_idx);
9521 btrfs_abort_transaction(trans, ret);
9525 if (old_inode->i_nlink == 1)
9526 BTRFS_I(old_inode)->dir_index = old_idx;
9527 if (new_inode->i_nlink == 1)
9528 BTRFS_I(new_inode)->dir_index = new_idx;
9530 if (root_log_pinned) {
9531 parent = new_dentry->d_parent;
9532 ret = btrfs_log_new_name(trans, BTRFS_I(old_inode),
9533 BTRFS_I(old_dir), parent,
9535 if (ret == BTRFS_NEED_LOG_SYNC)
9536 sync_log_root = true;
9537 else if (ret == BTRFS_NEED_TRANS_COMMIT)
9538 commit_transaction = true;
9540 btrfs_end_log_trans(root);
9541 root_log_pinned = false;
9543 if (dest_log_pinned) {
9544 if (!commit_transaction) {
9545 parent = old_dentry->d_parent;
9546 ret = btrfs_log_new_name(trans, BTRFS_I(new_inode),
9547 BTRFS_I(new_dir), parent,
9549 if (ret == BTRFS_NEED_LOG_SYNC)
9550 sync_log_dest = true;
9551 else if (ret == BTRFS_NEED_TRANS_COMMIT)
9552 commit_transaction = true;
9555 btrfs_end_log_trans(dest);
9556 dest_log_pinned = false;
9560 * If we have pinned a log and an error happened, we unpin tasks
9561 * trying to sync the log and force them to fallback to a transaction
9562 * commit if the log currently contains any of the inodes involved in
9563 * this rename operation (to ensure we do not persist a log with an
9564 * inconsistent state for any of these inodes or leading to any
9565 * inconsistencies when replayed). If the transaction was aborted, the
9566 * abortion reason is propagated to userspace when attempting to commit
9567 * the transaction. If the log does not contain any of these inodes, we
9568 * allow the tasks to sync it.
9570 if (ret && (root_log_pinned || dest_log_pinned)) {
9571 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9572 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9573 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9575 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9576 btrfs_set_log_full_commit(fs_info, trans);
9578 if (root_log_pinned) {
9579 btrfs_end_log_trans(root);
9580 root_log_pinned = false;
9582 if (dest_log_pinned) {
9583 btrfs_end_log_trans(dest);
9584 dest_log_pinned = false;
9587 if (!ret && sync_log_root && !commit_transaction) {
9588 ret = btrfs_sync_log(trans, BTRFS_I(old_inode)->root,
9591 commit_transaction = true;
9593 if (!ret && sync_log_dest && !commit_transaction) {
9594 ret = btrfs_sync_log(trans, BTRFS_I(new_inode)->root,
9597 commit_transaction = true;
9599 if (commit_transaction) {
9600 ret = btrfs_commit_transaction(trans);
9604 ret2 = btrfs_end_transaction(trans);
9605 ret = ret ? ret : ret2;
9608 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9609 up_read(&fs_info->subvol_sem);
9610 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9611 up_read(&fs_info->subvol_sem);
9616 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9617 struct btrfs_root *root,
9619 struct dentry *dentry)
9622 struct inode *inode;
9626 ret = btrfs_find_free_ino(root, &objectid);
9630 inode = btrfs_new_inode(trans, root, dir,
9631 dentry->d_name.name,
9633 btrfs_ino(BTRFS_I(dir)),
9635 S_IFCHR | WHITEOUT_MODE,
9638 if (IS_ERR(inode)) {
9639 ret = PTR_ERR(inode);
9643 inode->i_op = &btrfs_special_inode_operations;
9644 init_special_inode(inode, inode->i_mode,
9647 ret = btrfs_init_inode_security(trans, inode, dir,
9652 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9653 BTRFS_I(inode), 0, index);
9657 ret = btrfs_update_inode(trans, root, inode);
9659 unlock_new_inode(inode);
9661 inode_dec_link_count(inode);
9667 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9668 struct inode *new_dir, struct dentry *new_dentry,
9671 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9672 struct btrfs_trans_handle *trans;
9673 unsigned int trans_num_items;
9674 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9675 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9676 struct inode *new_inode = d_inode(new_dentry);
9677 struct inode *old_inode = d_inode(old_dentry);
9681 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9682 bool log_pinned = false;
9683 struct btrfs_log_ctx ctx;
9684 bool sync_log = false;
9685 bool commit_transaction = false;
9687 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9690 /* we only allow rename subvolume link between subvolumes */
9691 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9694 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9695 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9698 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9699 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9703 /* check for collisions, even if the name isn't there */
9704 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9705 new_dentry->d_name.name,
9706 new_dentry->d_name.len);
9709 if (ret == -EEXIST) {
9711 * eexist without a new_inode */
9712 if (WARN_ON(!new_inode)) {
9716 /* maybe -EOVERFLOW */
9723 * we're using rename to replace one file with another. Start IO on it
9724 * now so we don't add too much work to the end of the transaction
9726 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9727 filemap_flush(old_inode->i_mapping);
9729 /* close the racy window with snapshot create/destroy ioctl */
9730 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9731 down_read(&fs_info->subvol_sem);
9733 * We want to reserve the absolute worst case amount of items. So if
9734 * both inodes are subvols and we need to unlink them then that would
9735 * require 4 item modifications, but if they are both normal inodes it
9736 * would require 5 item modifications, so we'll assume they are normal
9737 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9738 * should cover the worst case number of items we'll modify.
9739 * If our rename has the whiteout flag, we need more 5 units for the
9740 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9741 * when selinux is enabled).
9743 trans_num_items = 11;
9744 if (flags & RENAME_WHITEOUT)
9745 trans_num_items += 5;
9746 trans = btrfs_start_transaction(root, trans_num_items);
9747 if (IS_ERR(trans)) {
9748 ret = PTR_ERR(trans);
9753 btrfs_record_root_in_trans(trans, dest);
9755 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9759 BTRFS_I(old_inode)->dir_index = 0ULL;
9760 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9761 /* force full log commit if subvolume involved. */
9762 btrfs_set_log_full_commit(fs_info, trans);
9764 btrfs_pin_log_trans(root);
9766 ret = btrfs_insert_inode_ref(trans, dest,
9767 new_dentry->d_name.name,
9768 new_dentry->d_name.len,
9770 btrfs_ino(BTRFS_I(new_dir)), index);
9775 inode_inc_iversion(old_dir);
9776 inode_inc_iversion(new_dir);
9777 inode_inc_iversion(old_inode);
9778 old_dir->i_ctime = old_dir->i_mtime =
9779 new_dir->i_ctime = new_dir->i_mtime =
9780 old_inode->i_ctime = current_time(old_dir);
9782 if (old_dentry->d_parent != new_dentry->d_parent)
9783 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9784 BTRFS_I(old_inode), 1);
9786 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9787 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9788 ret = btrfs_unlink_subvol(trans, old_dir, root_objectid,
9789 old_dentry->d_name.name,
9790 old_dentry->d_name.len);
9792 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9793 BTRFS_I(d_inode(old_dentry)),
9794 old_dentry->d_name.name,
9795 old_dentry->d_name.len);
9797 ret = btrfs_update_inode(trans, root, old_inode);
9800 btrfs_abort_transaction(trans, ret);
9805 inode_inc_iversion(new_inode);
9806 new_inode->i_ctime = current_time(new_inode);
9807 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9808 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9809 root_objectid = BTRFS_I(new_inode)->location.objectid;
9810 ret = btrfs_unlink_subvol(trans, new_dir, root_objectid,
9811 new_dentry->d_name.name,
9812 new_dentry->d_name.len);
9813 BUG_ON(new_inode->i_nlink == 0);
9815 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9816 BTRFS_I(d_inode(new_dentry)),
9817 new_dentry->d_name.name,
9818 new_dentry->d_name.len);
9820 if (!ret && new_inode->i_nlink == 0)
9821 ret = btrfs_orphan_add(trans,
9822 BTRFS_I(d_inode(new_dentry)));
9824 btrfs_abort_transaction(trans, ret);
9829 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9830 new_dentry->d_name.name,
9831 new_dentry->d_name.len, 0, index);
9833 btrfs_abort_transaction(trans, ret);
9837 if (old_inode->i_nlink == 1)
9838 BTRFS_I(old_inode)->dir_index = index;
9841 struct dentry *parent = new_dentry->d_parent;
9843 btrfs_init_log_ctx(&ctx, old_inode);
9844 ret = btrfs_log_new_name(trans, BTRFS_I(old_inode),
9845 BTRFS_I(old_dir), parent,
9847 if (ret == BTRFS_NEED_LOG_SYNC)
9849 else if (ret == BTRFS_NEED_TRANS_COMMIT)
9850 commit_transaction = true;
9852 btrfs_end_log_trans(root);
9856 if (flags & RENAME_WHITEOUT) {
9857 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9861 btrfs_abort_transaction(trans, ret);
9867 * If we have pinned the log and an error happened, we unpin tasks
9868 * trying to sync the log and force them to fallback to a transaction
9869 * commit if the log currently contains any of the inodes involved in
9870 * this rename operation (to ensure we do not persist a log with an
9871 * inconsistent state for any of these inodes or leading to any
9872 * inconsistencies when replayed). If the transaction was aborted, the
9873 * abortion reason is propagated to userspace when attempting to commit
9874 * the transaction. If the log does not contain any of these inodes, we
9875 * allow the tasks to sync it.
9877 if (ret && log_pinned) {
9878 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9879 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9880 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9882 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9883 btrfs_set_log_full_commit(fs_info, trans);
9885 btrfs_end_log_trans(root);
9888 if (!ret && sync_log) {
9889 ret = btrfs_sync_log(trans, BTRFS_I(old_inode)->root, &ctx);
9891 commit_transaction = true;
9893 if (commit_transaction) {
9894 ret = btrfs_commit_transaction(trans);
9898 ret2 = btrfs_end_transaction(trans);
9899 ret = ret ? ret : ret2;
9902 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9903 up_read(&fs_info->subvol_sem);
9908 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9909 struct inode *new_dir, struct dentry *new_dentry,
9912 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9915 if (flags & RENAME_EXCHANGE)
9916 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9919 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9922 struct btrfs_delalloc_work {
9923 struct inode *inode;
9924 struct completion completion;
9925 struct list_head list;
9926 struct btrfs_work work;
9929 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9931 struct btrfs_delalloc_work *delalloc_work;
9932 struct inode *inode;
9934 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9936 inode = delalloc_work->inode;
9937 filemap_flush(inode->i_mapping);
9938 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9939 &BTRFS_I(inode)->runtime_flags))
9940 filemap_flush(inode->i_mapping);
9943 complete(&delalloc_work->completion);
9946 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9948 struct btrfs_delalloc_work *work;
9950 work = kmalloc(sizeof(*work), GFP_NOFS);
9954 init_completion(&work->completion);
9955 INIT_LIST_HEAD(&work->list);
9956 work->inode = inode;
9957 WARN_ON_ONCE(!inode);
9958 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9959 btrfs_run_delalloc_work, NULL, NULL);
9965 * some fairly slow code that needs optimization. This walks the list
9966 * of all the inodes with pending delalloc and forces them to disk.
9968 static int start_delalloc_inodes(struct btrfs_root *root, int nr)
9970 struct btrfs_inode *binode;
9971 struct inode *inode;
9972 struct btrfs_delalloc_work *work, *next;
9973 struct list_head works;
9974 struct list_head splice;
9977 INIT_LIST_HEAD(&works);
9978 INIT_LIST_HEAD(&splice);
9980 mutex_lock(&root->delalloc_mutex);
9981 spin_lock(&root->delalloc_lock);
9982 list_splice_init(&root->delalloc_inodes, &splice);
9983 while (!list_empty(&splice)) {
9984 binode = list_entry(splice.next, struct btrfs_inode,
9987 list_move_tail(&binode->delalloc_inodes,
9988 &root->delalloc_inodes);
9989 inode = igrab(&binode->vfs_inode);
9991 cond_resched_lock(&root->delalloc_lock);
9994 spin_unlock(&root->delalloc_lock);
9996 work = btrfs_alloc_delalloc_work(inode);
10002 list_add_tail(&work->list, &works);
10003 btrfs_queue_work(root->fs_info->flush_workers,
10006 if (nr != -1 && ret >= nr)
10009 spin_lock(&root->delalloc_lock);
10011 spin_unlock(&root->delalloc_lock);
10014 list_for_each_entry_safe(work, next, &works, list) {
10015 list_del_init(&work->list);
10016 wait_for_completion(&work->completion);
10020 if (!list_empty(&splice)) {
10021 spin_lock(&root->delalloc_lock);
10022 list_splice_tail(&splice, &root->delalloc_inodes);
10023 spin_unlock(&root->delalloc_lock);
10025 mutex_unlock(&root->delalloc_mutex);
10029 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
10031 struct btrfs_fs_info *fs_info = root->fs_info;
10034 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10037 ret = start_delalloc_inodes(root, -1);
10043 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int nr)
10045 struct btrfs_root *root;
10046 struct list_head splice;
10049 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10052 INIT_LIST_HEAD(&splice);
10054 mutex_lock(&fs_info->delalloc_root_mutex);
10055 spin_lock(&fs_info->delalloc_root_lock);
10056 list_splice_init(&fs_info->delalloc_roots, &splice);
10057 while (!list_empty(&splice) && nr) {
10058 root = list_first_entry(&splice, struct btrfs_root,
10060 root = btrfs_grab_fs_root(root);
10062 list_move_tail(&root->delalloc_root,
10063 &fs_info->delalloc_roots);
10064 spin_unlock(&fs_info->delalloc_root_lock);
10066 ret = start_delalloc_inodes(root, nr);
10067 btrfs_put_fs_root(root);
10075 spin_lock(&fs_info->delalloc_root_lock);
10077 spin_unlock(&fs_info->delalloc_root_lock);
10081 if (!list_empty(&splice)) {
10082 spin_lock(&fs_info->delalloc_root_lock);
10083 list_splice_tail(&splice, &fs_info->delalloc_roots);
10084 spin_unlock(&fs_info->delalloc_root_lock);
10086 mutex_unlock(&fs_info->delalloc_root_mutex);
10090 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10091 const char *symname)
10093 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10094 struct btrfs_trans_handle *trans;
10095 struct btrfs_root *root = BTRFS_I(dir)->root;
10096 struct btrfs_path *path;
10097 struct btrfs_key key;
10098 struct inode *inode = NULL;
10105 struct btrfs_file_extent_item *ei;
10106 struct extent_buffer *leaf;
10108 name_len = strlen(symname);
10109 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
10110 return -ENAMETOOLONG;
10113 * 2 items for inode item and ref
10114 * 2 items for dir items
10115 * 1 item for updating parent inode item
10116 * 1 item for the inline extent item
10117 * 1 item for xattr if selinux is on
10119 trans = btrfs_start_transaction(root, 7);
10121 return PTR_ERR(trans);
10123 err = btrfs_find_free_ino(root, &objectid);
10127 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10128 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
10129 objectid, S_IFLNK|S_IRWXUGO, &index);
10130 if (IS_ERR(inode)) {
10131 err = PTR_ERR(inode);
10137 * If the active LSM wants to access the inode during
10138 * d_instantiate it needs these. Smack checks to see
10139 * if the filesystem supports xattrs by looking at the
10142 inode->i_fop = &btrfs_file_operations;
10143 inode->i_op = &btrfs_file_inode_operations;
10144 inode->i_mapping->a_ops = &btrfs_aops;
10145 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10147 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10151 path = btrfs_alloc_path();
10156 key.objectid = btrfs_ino(BTRFS_I(inode));
10158 key.type = BTRFS_EXTENT_DATA_KEY;
10159 datasize = btrfs_file_extent_calc_inline_size(name_len);
10160 err = btrfs_insert_empty_item(trans, root, path, &key,
10163 btrfs_free_path(path);
10166 leaf = path->nodes[0];
10167 ei = btrfs_item_ptr(leaf, path->slots[0],
10168 struct btrfs_file_extent_item);
10169 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10170 btrfs_set_file_extent_type(leaf, ei,
10171 BTRFS_FILE_EXTENT_INLINE);
10172 btrfs_set_file_extent_encryption(leaf, ei, 0);
10173 btrfs_set_file_extent_compression(leaf, ei, 0);
10174 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10175 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10177 ptr = btrfs_file_extent_inline_start(ei);
10178 write_extent_buffer(leaf, symname, ptr, name_len);
10179 btrfs_mark_buffer_dirty(leaf);
10180 btrfs_free_path(path);
10182 inode->i_op = &btrfs_symlink_inode_operations;
10183 inode_nohighmem(inode);
10184 inode->i_mapping->a_ops = &btrfs_aops;
10185 inode_set_bytes(inode, name_len);
10186 btrfs_i_size_write(BTRFS_I(inode), name_len);
10187 err = btrfs_update_inode(trans, root, inode);
10189 * Last step, add directory indexes for our symlink inode. This is the
10190 * last step to avoid extra cleanup of these indexes if an error happens
10194 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
10195 BTRFS_I(inode), 0, index);
10199 d_instantiate_new(dentry, inode);
10202 btrfs_end_transaction(trans);
10203 if (err && inode) {
10204 inode_dec_link_count(inode);
10205 discard_new_inode(inode);
10207 btrfs_btree_balance_dirty(fs_info);
10211 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10212 u64 start, u64 num_bytes, u64 min_size,
10213 loff_t actual_len, u64 *alloc_hint,
10214 struct btrfs_trans_handle *trans)
10216 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
10217 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10218 struct extent_map *em;
10219 struct btrfs_root *root = BTRFS_I(inode)->root;
10220 struct btrfs_key ins;
10221 u64 cur_offset = start;
10224 u64 last_alloc = (u64)-1;
10226 bool own_trans = true;
10227 u64 end = start + num_bytes - 1;
10231 while (num_bytes > 0) {
10233 trans = btrfs_start_transaction(root, 3);
10234 if (IS_ERR(trans)) {
10235 ret = PTR_ERR(trans);
10240 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10241 cur_bytes = max(cur_bytes, min_size);
10243 * If we are severely fragmented we could end up with really
10244 * small allocations, so if the allocator is returning small
10245 * chunks lets make its job easier by only searching for those
10248 cur_bytes = min(cur_bytes, last_alloc);
10249 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10250 min_size, 0, *alloc_hint, &ins, 1, 0);
10253 btrfs_end_transaction(trans);
10256 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10258 last_alloc = ins.offset;
10259 ret = insert_reserved_file_extent(trans, inode,
10260 cur_offset, ins.objectid,
10261 ins.offset, ins.offset,
10262 ins.offset, 0, 0, 0,
10263 BTRFS_FILE_EXTENT_PREALLOC);
10265 btrfs_free_reserved_extent(fs_info, ins.objectid,
10267 btrfs_abort_transaction(trans, ret);
10269 btrfs_end_transaction(trans);
10273 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10274 cur_offset + ins.offset -1, 0);
10276 em = alloc_extent_map();
10278 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10279 &BTRFS_I(inode)->runtime_flags);
10283 em->start = cur_offset;
10284 em->orig_start = cur_offset;
10285 em->len = ins.offset;
10286 em->block_start = ins.objectid;
10287 em->block_len = ins.offset;
10288 em->orig_block_len = ins.offset;
10289 em->ram_bytes = ins.offset;
10290 em->bdev = fs_info->fs_devices->latest_bdev;
10291 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10292 em->generation = trans->transid;
10295 write_lock(&em_tree->lock);
10296 ret = add_extent_mapping(em_tree, em, 1);
10297 write_unlock(&em_tree->lock);
10298 if (ret != -EEXIST)
10300 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10301 cur_offset + ins.offset - 1,
10304 free_extent_map(em);
10306 num_bytes -= ins.offset;
10307 cur_offset += ins.offset;
10308 *alloc_hint = ins.objectid + ins.offset;
10310 inode_inc_iversion(inode);
10311 inode->i_ctime = current_time(inode);
10312 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10313 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10314 (actual_len > inode->i_size) &&
10315 (cur_offset > inode->i_size)) {
10316 if (cur_offset > actual_len)
10317 i_size = actual_len;
10319 i_size = cur_offset;
10320 i_size_write(inode, i_size);
10321 btrfs_ordered_update_i_size(inode, i_size, NULL);
10324 ret = btrfs_update_inode(trans, root, inode);
10327 btrfs_abort_transaction(trans, ret);
10329 btrfs_end_transaction(trans);
10334 btrfs_end_transaction(trans);
10336 if (cur_offset < end)
10337 btrfs_free_reserved_data_space(inode, NULL, cur_offset,
10338 end - cur_offset + 1);
10342 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10343 u64 start, u64 num_bytes, u64 min_size,
10344 loff_t actual_len, u64 *alloc_hint)
10346 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10347 min_size, actual_len, alloc_hint,
10351 int btrfs_prealloc_file_range_trans(struct inode *inode,
10352 struct btrfs_trans_handle *trans, int mode,
10353 u64 start, u64 num_bytes, u64 min_size,
10354 loff_t actual_len, u64 *alloc_hint)
10356 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10357 min_size, actual_len, alloc_hint, trans);
10360 static int btrfs_set_page_dirty(struct page *page)
10362 return __set_page_dirty_nobuffers(page);
10365 static int btrfs_permission(struct inode *inode, int mask)
10367 struct btrfs_root *root = BTRFS_I(inode)->root;
10368 umode_t mode = inode->i_mode;
10370 if (mask & MAY_WRITE &&
10371 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10372 if (btrfs_root_readonly(root))
10374 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10377 return generic_permission(inode, mask);
10380 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10382 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10383 struct btrfs_trans_handle *trans;
10384 struct btrfs_root *root = BTRFS_I(dir)->root;
10385 struct inode *inode = NULL;
10391 * 5 units required for adding orphan entry
10393 trans = btrfs_start_transaction(root, 5);
10395 return PTR_ERR(trans);
10397 ret = btrfs_find_free_ino(root, &objectid);
10401 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10402 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10403 if (IS_ERR(inode)) {
10404 ret = PTR_ERR(inode);
10409 inode->i_fop = &btrfs_file_operations;
10410 inode->i_op = &btrfs_file_inode_operations;
10412 inode->i_mapping->a_ops = &btrfs_aops;
10413 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10415 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10419 ret = btrfs_update_inode(trans, root, inode);
10422 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10427 * We set number of links to 0 in btrfs_new_inode(), and here we set
10428 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10431 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10433 set_nlink(inode, 1);
10434 d_tmpfile(dentry, inode);
10435 unlock_new_inode(inode);
10436 mark_inode_dirty(inode);
10438 btrfs_end_transaction(trans);
10440 discard_new_inode(inode);
10441 btrfs_btree_balance_dirty(fs_info);
10445 __attribute__((const))
10446 static int btrfs_readpage_io_failed_hook(struct page *page, int failed_mirror)
10451 void btrfs_set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
10453 struct inode *inode = tree->private_data;
10454 unsigned long index = start >> PAGE_SHIFT;
10455 unsigned long end_index = end >> PAGE_SHIFT;
10458 while (index <= end_index) {
10459 page = find_get_page(inode->i_mapping, index);
10460 ASSERT(page); /* Pages should be in the extent_io_tree */
10461 set_page_writeback(page);
10467 static const struct inode_operations btrfs_dir_inode_operations = {
10468 .getattr = btrfs_getattr,
10469 .lookup = btrfs_lookup,
10470 .create = btrfs_create,
10471 .unlink = btrfs_unlink,
10472 .link = btrfs_link,
10473 .mkdir = btrfs_mkdir,
10474 .rmdir = btrfs_rmdir,
10475 .rename = btrfs_rename2,
10476 .symlink = btrfs_symlink,
10477 .setattr = btrfs_setattr,
10478 .mknod = btrfs_mknod,
10479 .listxattr = btrfs_listxattr,
10480 .permission = btrfs_permission,
10481 .get_acl = btrfs_get_acl,
10482 .set_acl = btrfs_set_acl,
10483 .update_time = btrfs_update_time,
10484 .tmpfile = btrfs_tmpfile,
10486 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10487 .lookup = btrfs_lookup,
10488 .permission = btrfs_permission,
10489 .update_time = btrfs_update_time,
10492 static const struct file_operations btrfs_dir_file_operations = {
10493 .llseek = generic_file_llseek,
10494 .read = generic_read_dir,
10495 .iterate_shared = btrfs_real_readdir,
10496 .open = btrfs_opendir,
10497 .unlocked_ioctl = btrfs_ioctl,
10498 #ifdef CONFIG_COMPAT
10499 .compat_ioctl = btrfs_compat_ioctl,
10501 .release = btrfs_release_file,
10502 .fsync = btrfs_sync_file,
10505 static const struct extent_io_ops btrfs_extent_io_ops = {
10506 /* mandatory callbacks */
10507 .submit_bio_hook = btrfs_submit_bio_hook,
10508 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10509 .readpage_io_failed_hook = btrfs_readpage_io_failed_hook,
10511 /* optional callbacks */
10512 .merge_extent_hook = btrfs_merge_extent_hook,
10513 .split_extent_hook = btrfs_split_extent_hook,
10517 * btrfs doesn't support the bmap operation because swapfiles
10518 * use bmap to make a mapping of extents in the file. They assume
10519 * these extents won't change over the life of the file and they
10520 * use the bmap result to do IO directly to the drive.
10522 * the btrfs bmap call would return logical addresses that aren't
10523 * suitable for IO and they also will change frequently as COW
10524 * operations happen. So, swapfile + btrfs == corruption.
10526 * For now we're avoiding this by dropping bmap.
10528 static const struct address_space_operations btrfs_aops = {
10529 .readpage = btrfs_readpage,
10530 .writepage = btrfs_writepage,
10531 .writepages = btrfs_writepages,
10532 .readpages = btrfs_readpages,
10533 .direct_IO = btrfs_direct_IO,
10534 .invalidatepage = btrfs_invalidatepage,
10535 .releasepage = btrfs_releasepage,
10536 .set_page_dirty = btrfs_set_page_dirty,
10537 .error_remove_page = generic_error_remove_page,
10540 static const struct inode_operations btrfs_file_inode_operations = {
10541 .getattr = btrfs_getattr,
10542 .setattr = btrfs_setattr,
10543 .listxattr = btrfs_listxattr,
10544 .permission = btrfs_permission,
10545 .fiemap = btrfs_fiemap,
10546 .get_acl = btrfs_get_acl,
10547 .set_acl = btrfs_set_acl,
10548 .update_time = btrfs_update_time,
10550 static const struct inode_operations btrfs_special_inode_operations = {
10551 .getattr = btrfs_getattr,
10552 .setattr = btrfs_setattr,
10553 .permission = btrfs_permission,
10554 .listxattr = btrfs_listxattr,
10555 .get_acl = btrfs_get_acl,
10556 .set_acl = btrfs_set_acl,
10557 .update_time = btrfs_update_time,
10559 static const struct inode_operations btrfs_symlink_inode_operations = {
10560 .get_link = page_get_link,
10561 .getattr = btrfs_getattr,
10562 .setattr = btrfs_setattr,
10563 .permission = btrfs_permission,
10564 .listxattr = btrfs_listxattr,
10565 .update_time = btrfs_update_time,
10568 const struct dentry_operations btrfs_dentry_operations = {
10569 .d_delete = btrfs_dentry_delete,