1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <linux/kernel.h>
8 #include <linux/buffer_head.h>
9 #include <linux/file.h>
11 #include <linux/pagemap.h>
12 #include <linux/highmem.h>
13 #include <linux/time.h>
14 #include <linux/init.h>
15 #include <linux/string.h>
16 #include <linux/backing-dev.h>
17 #include <linux/mpage.h>
18 #include <linux/swap.h>
19 #include <linux/writeback.h>
20 #include <linux/compat.h>
21 #include <linux/bit_spinlock.h>
22 #include <linux/xattr.h>
23 #include <linux/posix_acl.h>
24 #include <linux/falloc.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
27 #include <linux/mount.h>
28 #include <linux/btrfs.h>
29 #include <linux/blkdev.h>
30 #include <linux/posix_acl_xattr.h>
31 #include <linux/uio.h>
32 #include <linux/magic.h>
33 #include <linux/iversion.h>
34 #include <asm/unaligned.h>
37 #include "transaction.h"
38 #include "btrfs_inode.h"
39 #include "print-tree.h"
40 #include "ordered-data.h"
44 #include "compression.h"
46 #include "free-space-cache.h"
47 #include "inode-map.h"
53 struct btrfs_iget_args {
54 struct btrfs_key *location;
55 struct btrfs_root *root;
58 struct btrfs_dio_data {
60 u64 unsubmitted_oe_range_start;
61 u64 unsubmitted_oe_range_end;
65 static const struct inode_operations btrfs_dir_inode_operations;
66 static const struct inode_operations btrfs_symlink_inode_operations;
67 static const struct inode_operations btrfs_dir_ro_inode_operations;
68 static const struct inode_operations btrfs_special_inode_operations;
69 static const struct inode_operations btrfs_file_inode_operations;
70 static const struct address_space_operations btrfs_aops;
71 static const struct address_space_operations btrfs_symlink_aops;
72 static const struct file_operations btrfs_dir_file_operations;
73 static const struct extent_io_ops btrfs_extent_io_ops;
75 static struct kmem_cache *btrfs_inode_cachep;
76 struct kmem_cache *btrfs_trans_handle_cachep;
77 struct kmem_cache *btrfs_path_cachep;
78 struct kmem_cache *btrfs_free_space_cachep;
81 static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
82 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
83 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
84 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
85 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
86 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
87 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
88 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
91 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
92 static int btrfs_truncate(struct inode *inode, bool skip_writeback);
93 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
94 static noinline int cow_file_range(struct inode *inode,
95 struct page *locked_page,
96 u64 start, u64 end, u64 delalloc_end,
97 int *page_started, unsigned long *nr_written,
98 int unlock, struct btrfs_dedupe_hash *hash);
99 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
100 u64 orig_start, u64 block_start,
101 u64 block_len, u64 orig_block_len,
102 u64 ram_bytes, int compress_type,
105 static void __endio_write_update_ordered(struct inode *inode,
106 const u64 offset, const u64 bytes,
107 const bool uptodate);
110 * Cleanup all submitted ordered extents in specified range to handle errors
111 * from the fill_dellaloc() callback.
113 * NOTE: caller must ensure that when an error happens, it can not call
114 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
115 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
116 * to be released, which we want to happen only when finishing the ordered
117 * extent (btrfs_finish_ordered_io()). Also note that the caller of the
118 * fill_delalloc() callback already does proper cleanup for the first page of
119 * the range, that is, it invokes the callback writepage_end_io_hook() for the
120 * range of the first page.
122 static inline void btrfs_cleanup_ordered_extents(struct inode *inode,
126 unsigned long index = offset >> PAGE_SHIFT;
127 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
130 while (index <= end_index) {
131 page = find_get_page(inode->i_mapping, index);
135 ClearPagePrivate2(page);
138 return __endio_write_update_ordered(inode, offset + PAGE_SIZE,
139 bytes - PAGE_SIZE, false);
142 static int btrfs_dirty_inode(struct inode *inode);
144 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
145 void btrfs_test_inode_set_ops(struct inode *inode)
147 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
151 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
152 struct inode *inode, struct inode *dir,
153 const struct qstr *qstr)
157 err = btrfs_init_acl(trans, inode, dir);
159 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
164 * this does all the hard work for inserting an inline extent into
165 * the btree. The caller should have done a btrfs_drop_extents so that
166 * no overlapping inline items exist in the btree
168 static int insert_inline_extent(struct btrfs_trans_handle *trans,
169 struct btrfs_path *path, int extent_inserted,
170 struct btrfs_root *root, struct inode *inode,
171 u64 start, size_t size, size_t compressed_size,
173 struct page **compressed_pages)
175 struct extent_buffer *leaf;
176 struct page *page = NULL;
179 struct btrfs_file_extent_item *ei;
181 size_t cur_size = size;
182 unsigned long offset;
184 if (compressed_size && compressed_pages)
185 cur_size = compressed_size;
187 inode_add_bytes(inode, size);
189 if (!extent_inserted) {
190 struct btrfs_key key;
193 key.objectid = btrfs_ino(BTRFS_I(inode));
195 key.type = BTRFS_EXTENT_DATA_KEY;
197 datasize = btrfs_file_extent_calc_inline_size(cur_size);
198 path->leave_spinning = 1;
199 ret = btrfs_insert_empty_item(trans, root, path, &key,
204 leaf = path->nodes[0];
205 ei = btrfs_item_ptr(leaf, path->slots[0],
206 struct btrfs_file_extent_item);
207 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
208 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
209 btrfs_set_file_extent_encryption(leaf, ei, 0);
210 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
211 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
212 ptr = btrfs_file_extent_inline_start(ei);
214 if (compress_type != BTRFS_COMPRESS_NONE) {
217 while (compressed_size > 0) {
218 cpage = compressed_pages[i];
219 cur_size = min_t(unsigned long, compressed_size,
222 kaddr = kmap_atomic(cpage);
223 write_extent_buffer(leaf, kaddr, ptr, cur_size);
224 kunmap_atomic(kaddr);
228 compressed_size -= cur_size;
230 btrfs_set_file_extent_compression(leaf, ei,
233 page = find_get_page(inode->i_mapping,
234 start >> PAGE_SHIFT);
235 btrfs_set_file_extent_compression(leaf, ei, 0);
236 kaddr = kmap_atomic(page);
237 offset = start & (PAGE_SIZE - 1);
238 write_extent_buffer(leaf, kaddr + offset, ptr, size);
239 kunmap_atomic(kaddr);
242 btrfs_mark_buffer_dirty(leaf);
243 btrfs_release_path(path);
246 * we're an inline extent, so nobody can
247 * extend the file past i_size without locking
248 * a page we already have locked.
250 * We must do any isize and inode updates
251 * before we unlock the pages. Otherwise we
252 * could end up racing with unlink.
254 BTRFS_I(inode)->disk_i_size = inode->i_size;
255 ret = btrfs_update_inode(trans, root, inode);
263 * conditionally insert an inline extent into the file. This
264 * does the checks required to make sure the data is small enough
265 * to fit as an inline extent.
267 static noinline int cow_file_range_inline(struct inode *inode, u64 start,
268 u64 end, size_t compressed_size,
270 struct page **compressed_pages)
272 struct btrfs_root *root = BTRFS_I(inode)->root;
273 struct btrfs_fs_info *fs_info = root->fs_info;
274 struct btrfs_trans_handle *trans;
275 u64 isize = i_size_read(inode);
276 u64 actual_end = min(end + 1, isize);
277 u64 inline_len = actual_end - start;
278 u64 aligned_end = ALIGN(end, fs_info->sectorsize);
279 u64 data_len = inline_len;
281 struct btrfs_path *path;
282 int extent_inserted = 0;
283 u32 extent_item_size;
286 data_len = compressed_size;
289 actual_end > fs_info->sectorsize ||
290 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
292 (actual_end & (fs_info->sectorsize - 1)) == 0) ||
294 data_len > fs_info->max_inline) {
298 path = btrfs_alloc_path();
302 trans = btrfs_join_transaction(root);
304 btrfs_free_path(path);
305 return PTR_ERR(trans);
307 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
309 if (compressed_size && compressed_pages)
310 extent_item_size = btrfs_file_extent_calc_inline_size(
313 extent_item_size = btrfs_file_extent_calc_inline_size(
316 ret = __btrfs_drop_extents(trans, root, inode, path,
317 start, aligned_end, NULL,
318 1, 1, extent_item_size, &extent_inserted);
320 btrfs_abort_transaction(trans, ret);
324 if (isize > actual_end)
325 inline_len = min_t(u64, isize, actual_end);
326 ret = insert_inline_extent(trans, path, extent_inserted,
328 inline_len, compressed_size,
329 compress_type, compressed_pages);
330 if (ret && ret != -ENOSPC) {
331 btrfs_abort_transaction(trans, ret);
333 } else if (ret == -ENOSPC) {
338 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
339 btrfs_drop_extent_cache(BTRFS_I(inode), start, aligned_end - 1, 0);
342 * Don't forget to free the reserved space, as for inlined extent
343 * it won't count as data extent, free them directly here.
344 * And at reserve time, it's always aligned to page size, so
345 * just free one page here.
347 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
348 btrfs_free_path(path);
349 btrfs_end_transaction(trans);
353 struct async_extent {
358 unsigned long nr_pages;
360 struct list_head list;
365 struct btrfs_root *root;
366 struct page *locked_page;
369 unsigned int write_flags;
370 struct list_head extents;
371 struct btrfs_work work;
374 static noinline int add_async_extent(struct async_cow *cow,
375 u64 start, u64 ram_size,
378 unsigned long nr_pages,
381 struct async_extent *async_extent;
383 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
384 BUG_ON(!async_extent); /* -ENOMEM */
385 async_extent->start = start;
386 async_extent->ram_size = ram_size;
387 async_extent->compressed_size = compressed_size;
388 async_extent->pages = pages;
389 async_extent->nr_pages = nr_pages;
390 async_extent->compress_type = compress_type;
391 list_add_tail(&async_extent->list, &cow->extents);
395 static inline int inode_need_compress(struct inode *inode, u64 start, u64 end)
397 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
400 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
403 if (BTRFS_I(inode)->defrag_compress)
405 /* bad compression ratios */
406 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
408 if (btrfs_test_opt(fs_info, COMPRESS) ||
409 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
410 BTRFS_I(inode)->prop_compress)
411 return btrfs_compress_heuristic(inode, start, end);
415 static inline void inode_should_defrag(struct btrfs_inode *inode,
416 u64 start, u64 end, u64 num_bytes, u64 small_write)
418 /* If this is a small write inside eof, kick off a defrag */
419 if (num_bytes < small_write &&
420 (start > 0 || end + 1 < inode->disk_i_size))
421 btrfs_add_inode_defrag(NULL, inode);
425 * we create compressed extents in two phases. The first
426 * phase compresses a range of pages that have already been
427 * locked (both pages and state bits are locked).
429 * This is done inside an ordered work queue, and the compression
430 * is spread across many cpus. The actual IO submission is step
431 * two, and the ordered work queue takes care of making sure that
432 * happens in the same order things were put onto the queue by
433 * writepages and friends.
435 * If this code finds it can't get good compression, it puts an
436 * entry onto the work queue to write the uncompressed bytes. This
437 * makes sure that both compressed inodes and uncompressed inodes
438 * are written in the same order that the flusher thread sent them
441 static noinline void compress_file_range(struct inode *inode,
442 struct page *locked_page,
444 struct async_cow *async_cow,
447 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
448 u64 blocksize = fs_info->sectorsize;
450 u64 isize = i_size_read(inode);
452 struct page **pages = NULL;
453 unsigned long nr_pages;
454 unsigned long total_compressed = 0;
455 unsigned long total_in = 0;
458 int compress_type = fs_info->compress_type;
461 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
464 actual_end = min_t(u64, isize, end + 1);
467 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
468 BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
469 nr_pages = min_t(unsigned long, nr_pages,
470 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
473 * we don't want to send crud past the end of i_size through
474 * compression, that's just a waste of CPU time. So, if the
475 * end of the file is before the start of our current
476 * requested range of bytes, we bail out to the uncompressed
477 * cleanup code that can deal with all of this.
479 * It isn't really the fastest way to fix things, but this is a
480 * very uncommon corner.
482 if (actual_end <= start)
483 goto cleanup_and_bail_uncompressed;
485 total_compressed = actual_end - start;
488 * skip compression for a small file range(<=blocksize) that
489 * isn't an inline extent, since it doesn't save disk space at all.
491 if (total_compressed <= blocksize &&
492 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
493 goto cleanup_and_bail_uncompressed;
495 total_compressed = min_t(unsigned long, total_compressed,
496 BTRFS_MAX_UNCOMPRESSED);
501 * we do compression for mount -o compress and when the
502 * inode has not been flagged as nocompress. This flag can
503 * change at any time if we discover bad compression ratios.
505 if (inode_need_compress(inode, start, end)) {
507 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
509 /* just bail out to the uncompressed code */
513 if (BTRFS_I(inode)->defrag_compress)
514 compress_type = BTRFS_I(inode)->defrag_compress;
515 else if (BTRFS_I(inode)->prop_compress)
516 compress_type = BTRFS_I(inode)->prop_compress;
519 * we need to call clear_page_dirty_for_io on each
520 * page in the range. Otherwise applications with the file
521 * mmap'd can wander in and change the page contents while
522 * we are compressing them.
524 * If the compression fails for any reason, we set the pages
525 * dirty again later on.
527 * Note that the remaining part is redirtied, the start pointer
528 * has moved, the end is the original one.
531 extent_range_clear_dirty_for_io(inode, start, end);
535 /* Compression level is applied here and only here */
536 ret = btrfs_compress_pages(
537 compress_type | (fs_info->compress_level << 4),
538 inode->i_mapping, start,
545 unsigned long offset = total_compressed &
547 struct page *page = pages[nr_pages - 1];
550 /* zero the tail end of the last page, we might be
551 * sending it down to disk
554 kaddr = kmap_atomic(page);
555 memset(kaddr + offset, 0,
557 kunmap_atomic(kaddr);
564 /* lets try to make an inline extent */
565 if (ret || total_in < actual_end) {
566 /* we didn't compress the entire range, try
567 * to make an uncompressed inline extent.
569 ret = cow_file_range_inline(inode, start, end, 0,
570 BTRFS_COMPRESS_NONE, NULL);
572 /* try making a compressed inline extent */
573 ret = cow_file_range_inline(inode, start, end,
575 compress_type, pages);
578 unsigned long clear_flags = EXTENT_DELALLOC |
579 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
580 EXTENT_DO_ACCOUNTING;
581 unsigned long page_error_op;
583 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
586 * inline extent creation worked or returned error,
587 * we don't need to create any more async work items.
588 * Unlock and free up our temp pages.
590 * We use DO_ACCOUNTING here because we need the
591 * delalloc_release_metadata to be done _after_ we drop
592 * our outstanding extent for clearing delalloc for this
595 extent_clear_unlock_delalloc(inode, start, end, end,
608 * we aren't doing an inline extent round the compressed size
609 * up to a block size boundary so the allocator does sane
612 total_compressed = ALIGN(total_compressed, blocksize);
615 * one last check to make sure the compression is really a
616 * win, compare the page count read with the blocks on disk,
617 * compression must free at least one sector size
619 total_in = ALIGN(total_in, PAGE_SIZE);
620 if (total_compressed + blocksize <= total_in) {
624 * The async work queues will take care of doing actual
625 * allocation on disk for these compressed pages, and
626 * will submit them to the elevator.
628 add_async_extent(async_cow, start, total_in,
629 total_compressed, pages, nr_pages,
632 if (start + total_in < end) {
643 * the compression code ran but failed to make things smaller,
644 * free any pages it allocated and our page pointer array
646 for (i = 0; i < nr_pages; i++) {
647 WARN_ON(pages[i]->mapping);
652 total_compressed = 0;
655 /* flag the file so we don't compress in the future */
656 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
657 !(BTRFS_I(inode)->prop_compress)) {
658 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
661 cleanup_and_bail_uncompressed:
663 * No compression, but we still need to write the pages in the file
664 * we've been given so far. redirty the locked page if it corresponds
665 * to our extent and set things up for the async work queue to run
666 * cow_file_range to do the normal delalloc dance.
668 if (page_offset(locked_page) >= start &&
669 page_offset(locked_page) <= end)
670 __set_page_dirty_nobuffers(locked_page);
671 /* unlocked later on in the async handlers */
674 extent_range_redirty_for_io(inode, start, end);
675 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0,
676 BTRFS_COMPRESS_NONE);
682 for (i = 0; i < nr_pages; i++) {
683 WARN_ON(pages[i]->mapping);
689 static void free_async_extent_pages(struct async_extent *async_extent)
693 if (!async_extent->pages)
696 for (i = 0; i < async_extent->nr_pages; i++) {
697 WARN_ON(async_extent->pages[i]->mapping);
698 put_page(async_extent->pages[i]);
700 kfree(async_extent->pages);
701 async_extent->nr_pages = 0;
702 async_extent->pages = NULL;
706 * phase two of compressed writeback. This is the ordered portion
707 * of the code, which only gets called in the order the work was
708 * queued. We walk all the async extents created by compress_file_range
709 * and send them down to the disk.
711 static noinline void submit_compressed_extents(struct inode *inode,
712 struct async_cow *async_cow)
714 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
715 struct async_extent *async_extent;
717 struct btrfs_key ins;
718 struct extent_map *em;
719 struct btrfs_root *root = BTRFS_I(inode)->root;
720 struct extent_io_tree *io_tree;
724 while (!list_empty(&async_cow->extents)) {
725 async_extent = list_entry(async_cow->extents.next,
726 struct async_extent, list);
727 list_del(&async_extent->list);
729 io_tree = &BTRFS_I(inode)->io_tree;
732 /* did the compression code fall back to uncompressed IO? */
733 if (!async_extent->pages) {
734 int page_started = 0;
735 unsigned long nr_written = 0;
737 lock_extent(io_tree, async_extent->start,
738 async_extent->start +
739 async_extent->ram_size - 1);
741 /* allocate blocks */
742 ret = cow_file_range(inode, async_cow->locked_page,
744 async_extent->start +
745 async_extent->ram_size - 1,
746 async_extent->start +
747 async_extent->ram_size - 1,
748 &page_started, &nr_written, 0,
754 * if page_started, cow_file_range inserted an
755 * inline extent and took care of all the unlocking
756 * and IO for us. Otherwise, we need to submit
757 * all those pages down to the drive.
759 if (!page_started && !ret)
760 extent_write_locked_range(inode,
762 async_extent->start +
763 async_extent->ram_size - 1,
766 unlock_page(async_cow->locked_page);
772 lock_extent(io_tree, async_extent->start,
773 async_extent->start + async_extent->ram_size - 1);
775 ret = btrfs_reserve_extent(root, async_extent->ram_size,
776 async_extent->compressed_size,
777 async_extent->compressed_size,
778 0, alloc_hint, &ins, 1, 1);
780 free_async_extent_pages(async_extent);
782 if (ret == -ENOSPC) {
783 unlock_extent(io_tree, async_extent->start,
784 async_extent->start +
785 async_extent->ram_size - 1);
788 * we need to redirty the pages if we decide to
789 * fallback to uncompressed IO, otherwise we
790 * will not submit these pages down to lower
793 extent_range_redirty_for_io(inode,
795 async_extent->start +
796 async_extent->ram_size - 1);
803 * here we're doing allocation and writeback of the
806 em = create_io_em(inode, async_extent->start,
807 async_extent->ram_size, /* len */
808 async_extent->start, /* orig_start */
809 ins.objectid, /* block_start */
810 ins.offset, /* block_len */
811 ins.offset, /* orig_block_len */
812 async_extent->ram_size, /* ram_bytes */
813 async_extent->compress_type,
814 BTRFS_ORDERED_COMPRESSED);
816 /* ret value is not necessary due to void function */
817 goto out_free_reserve;
820 ret = btrfs_add_ordered_extent_compress(inode,
823 async_extent->ram_size,
825 BTRFS_ORDERED_COMPRESSED,
826 async_extent->compress_type);
828 btrfs_drop_extent_cache(BTRFS_I(inode),
830 async_extent->start +
831 async_extent->ram_size - 1, 0);
832 goto out_free_reserve;
834 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
837 * clear dirty, set writeback and unlock the pages.
839 extent_clear_unlock_delalloc(inode, async_extent->start,
840 async_extent->start +
841 async_extent->ram_size - 1,
842 async_extent->start +
843 async_extent->ram_size - 1,
844 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
845 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
847 if (btrfs_submit_compressed_write(inode,
849 async_extent->ram_size,
851 ins.offset, async_extent->pages,
852 async_extent->nr_pages,
853 async_cow->write_flags)) {
854 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
855 struct page *p = async_extent->pages[0];
856 const u64 start = async_extent->start;
857 const u64 end = start + async_extent->ram_size - 1;
859 p->mapping = inode->i_mapping;
860 tree->ops->writepage_end_io_hook(p, start, end,
863 extent_clear_unlock_delalloc(inode, start, end, end,
867 free_async_extent_pages(async_extent);
869 alloc_hint = ins.objectid + ins.offset;
875 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
876 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
878 extent_clear_unlock_delalloc(inode, async_extent->start,
879 async_extent->start +
880 async_extent->ram_size - 1,
881 async_extent->start +
882 async_extent->ram_size - 1,
883 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
884 EXTENT_DELALLOC_NEW |
885 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
886 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
887 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
889 free_async_extent_pages(async_extent);
894 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
897 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
898 struct extent_map *em;
901 read_lock(&em_tree->lock);
902 em = search_extent_mapping(em_tree, start, num_bytes);
905 * if block start isn't an actual block number then find the
906 * first block in this inode and use that as a hint. If that
907 * block is also bogus then just don't worry about it.
909 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
911 em = search_extent_mapping(em_tree, 0, 0);
912 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
913 alloc_hint = em->block_start;
917 alloc_hint = em->block_start;
921 read_unlock(&em_tree->lock);
927 * when extent_io.c finds a delayed allocation range in the file,
928 * the call backs end up in this code. The basic idea is to
929 * allocate extents on disk for the range, and create ordered data structs
930 * in ram to track those extents.
932 * locked_page is the page that writepage had locked already. We use
933 * it to make sure we don't do extra locks or unlocks.
935 * *page_started is set to one if we unlock locked_page and do everything
936 * required to start IO on it. It may be clean and already done with
939 static noinline int cow_file_range(struct inode *inode,
940 struct page *locked_page,
941 u64 start, u64 end, u64 delalloc_end,
942 int *page_started, unsigned long *nr_written,
943 int unlock, struct btrfs_dedupe_hash *hash)
945 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
946 struct btrfs_root *root = BTRFS_I(inode)->root;
949 unsigned long ram_size;
950 u64 cur_alloc_size = 0;
951 u64 blocksize = fs_info->sectorsize;
952 struct btrfs_key ins;
953 struct extent_map *em;
955 unsigned long page_ops;
956 bool extent_reserved = false;
959 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
965 num_bytes = ALIGN(end - start + 1, blocksize);
966 num_bytes = max(blocksize, num_bytes);
967 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
969 inode_should_defrag(BTRFS_I(inode), start, end, num_bytes, SZ_64K);
972 /* lets try to make an inline extent */
973 ret = cow_file_range_inline(inode, start, end, 0,
974 BTRFS_COMPRESS_NONE, NULL);
977 * We use DO_ACCOUNTING here because we need the
978 * delalloc_release_metadata to be run _after_ we drop
979 * our outstanding extent for clearing delalloc for this
982 extent_clear_unlock_delalloc(inode, start, end,
984 EXTENT_LOCKED | EXTENT_DELALLOC |
985 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
986 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
987 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
989 *nr_written = *nr_written +
990 (end - start + PAGE_SIZE) / PAGE_SIZE;
993 } else if (ret < 0) {
998 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
999 btrfs_drop_extent_cache(BTRFS_I(inode), start,
1000 start + num_bytes - 1, 0);
1002 while (num_bytes > 0) {
1003 cur_alloc_size = num_bytes;
1004 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1005 fs_info->sectorsize, 0, alloc_hint,
1009 cur_alloc_size = ins.offset;
1010 extent_reserved = true;
1012 ram_size = ins.offset;
1013 em = create_io_em(inode, start, ins.offset, /* len */
1014 start, /* orig_start */
1015 ins.objectid, /* block_start */
1016 ins.offset, /* block_len */
1017 ins.offset, /* orig_block_len */
1018 ram_size, /* ram_bytes */
1019 BTRFS_COMPRESS_NONE, /* compress_type */
1020 BTRFS_ORDERED_REGULAR /* type */);
1023 free_extent_map(em);
1025 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1026 ram_size, cur_alloc_size, 0);
1028 goto out_drop_extent_cache;
1030 if (root->root_key.objectid ==
1031 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1032 ret = btrfs_reloc_clone_csums(inode, start,
1035 * Only drop cache here, and process as normal.
1037 * We must not allow extent_clear_unlock_delalloc()
1038 * at out_unlock label to free meta of this ordered
1039 * extent, as its meta should be freed by
1040 * btrfs_finish_ordered_io().
1042 * So we must continue until @start is increased to
1043 * skip current ordered extent.
1046 btrfs_drop_extent_cache(BTRFS_I(inode), start,
1047 start + ram_size - 1, 0);
1050 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1052 /* we're not doing compressed IO, don't unlock the first
1053 * page (which the caller expects to stay locked), don't
1054 * clear any dirty bits and don't set any writeback bits
1056 * Do set the Private2 bit so we know this page was properly
1057 * setup for writepage
1059 page_ops = unlock ? PAGE_UNLOCK : 0;
1060 page_ops |= PAGE_SET_PRIVATE2;
1062 extent_clear_unlock_delalloc(inode, start,
1063 start + ram_size - 1,
1064 delalloc_end, locked_page,
1065 EXTENT_LOCKED | EXTENT_DELALLOC,
1067 if (num_bytes < cur_alloc_size)
1070 num_bytes -= cur_alloc_size;
1071 alloc_hint = ins.objectid + ins.offset;
1072 start += cur_alloc_size;
1073 extent_reserved = false;
1076 * btrfs_reloc_clone_csums() error, since start is increased
1077 * extent_clear_unlock_delalloc() at out_unlock label won't
1078 * free metadata of current ordered extent, we're OK to exit.
1086 out_drop_extent_cache:
1087 btrfs_drop_extent_cache(BTRFS_I(inode), start, start + ram_size - 1, 0);
1089 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1090 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1092 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1093 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1094 page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1097 * If we reserved an extent for our delalloc range (or a subrange) and
1098 * failed to create the respective ordered extent, then it means that
1099 * when we reserved the extent we decremented the extent's size from
1100 * the data space_info's bytes_may_use counter and incremented the
1101 * space_info's bytes_reserved counter by the same amount. We must make
1102 * sure extent_clear_unlock_delalloc() does not try to decrement again
1103 * the data space_info's bytes_may_use counter, therefore we do not pass
1104 * it the flag EXTENT_CLEAR_DATA_RESV.
1106 if (extent_reserved) {
1107 extent_clear_unlock_delalloc(inode, start,
1108 start + cur_alloc_size,
1109 start + cur_alloc_size,
1113 start += cur_alloc_size;
1117 extent_clear_unlock_delalloc(inode, start, end, delalloc_end,
1119 clear_bits | EXTENT_CLEAR_DATA_RESV,
1125 * work queue call back to started compression on a file and pages
1127 static noinline void async_cow_start(struct btrfs_work *work)
1129 struct async_cow *async_cow;
1131 async_cow = container_of(work, struct async_cow, work);
1133 compress_file_range(async_cow->inode, async_cow->locked_page,
1134 async_cow->start, async_cow->end, async_cow,
1136 if (num_added == 0) {
1137 btrfs_add_delayed_iput(async_cow->inode);
1138 async_cow->inode = NULL;
1143 * work queue call back to submit previously compressed pages
1145 static noinline void async_cow_submit(struct btrfs_work *work)
1147 struct btrfs_fs_info *fs_info;
1148 struct async_cow *async_cow;
1149 struct btrfs_root *root;
1150 unsigned long nr_pages;
1152 async_cow = container_of(work, struct async_cow, work);
1154 root = async_cow->root;
1155 fs_info = root->fs_info;
1156 nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1160 * atomic_sub_return implies a barrier for waitqueue_active
1162 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1164 waitqueue_active(&fs_info->async_submit_wait))
1165 wake_up(&fs_info->async_submit_wait);
1167 if (async_cow->inode)
1168 submit_compressed_extents(async_cow->inode, async_cow);
1171 static noinline void async_cow_free(struct btrfs_work *work)
1173 struct async_cow *async_cow;
1174 async_cow = container_of(work, struct async_cow, work);
1175 if (async_cow->inode)
1176 btrfs_add_delayed_iput(async_cow->inode);
1180 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1181 u64 start, u64 end, int *page_started,
1182 unsigned long *nr_written,
1183 unsigned int write_flags)
1185 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1186 struct async_cow *async_cow;
1187 struct btrfs_root *root = BTRFS_I(inode)->root;
1188 unsigned long nr_pages;
1191 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1193 while (start < end) {
1194 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1195 BUG_ON(!async_cow); /* -ENOMEM */
1196 async_cow->inode = igrab(inode);
1197 async_cow->root = root;
1198 async_cow->locked_page = locked_page;
1199 async_cow->start = start;
1200 async_cow->write_flags = write_flags;
1202 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1203 !btrfs_test_opt(fs_info, FORCE_COMPRESS))
1206 cur_end = min(end, start + SZ_512K - 1);
1208 async_cow->end = cur_end;
1209 INIT_LIST_HEAD(&async_cow->extents);
1211 btrfs_init_work(&async_cow->work,
1212 btrfs_delalloc_helper,
1213 async_cow_start, async_cow_submit,
1216 nr_pages = (cur_end - start + PAGE_SIZE) >>
1218 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1220 btrfs_queue_work(fs_info->delalloc_workers, &async_cow->work);
1222 *nr_written += nr_pages;
1223 start = cur_end + 1;
1229 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1230 u64 bytenr, u64 num_bytes)
1233 struct btrfs_ordered_sum *sums;
1236 ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1237 bytenr + num_bytes - 1, &list, 0);
1238 if (ret == 0 && list_empty(&list))
1241 while (!list_empty(&list)) {
1242 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1243 list_del(&sums->list);
1252 * when nowcow writeback call back. This checks for snapshots or COW copies
1253 * of the extents that exist in the file, and COWs the file as required.
1255 * If no cow copies or snapshots exist, we write directly to the existing
1258 static noinline int run_delalloc_nocow(struct inode *inode,
1259 struct page *locked_page,
1260 u64 start, u64 end, int *page_started, int force,
1261 unsigned long *nr_written)
1263 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1264 struct btrfs_root *root = BTRFS_I(inode)->root;
1265 struct extent_buffer *leaf;
1266 struct btrfs_path *path;
1267 struct btrfs_file_extent_item *fi;
1268 struct btrfs_key found_key;
1269 struct extent_map *em;
1284 u64 ino = btrfs_ino(BTRFS_I(inode));
1286 path = btrfs_alloc_path();
1288 extent_clear_unlock_delalloc(inode, start, end, end,
1290 EXTENT_LOCKED | EXTENT_DELALLOC |
1291 EXTENT_DO_ACCOUNTING |
1292 EXTENT_DEFRAG, PAGE_UNLOCK |
1294 PAGE_SET_WRITEBACK |
1295 PAGE_END_WRITEBACK);
1299 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
1301 cow_start = (u64)-1;
1304 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1308 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1309 leaf = path->nodes[0];
1310 btrfs_item_key_to_cpu(leaf, &found_key,
1311 path->slots[0] - 1);
1312 if (found_key.objectid == ino &&
1313 found_key.type == BTRFS_EXTENT_DATA_KEY)
1318 leaf = path->nodes[0];
1319 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1320 ret = btrfs_next_leaf(root, path);
1322 if (cow_start != (u64)-1)
1323 cur_offset = cow_start;
1328 leaf = path->nodes[0];
1334 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1336 if (found_key.objectid > ino)
1338 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1339 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1343 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1344 found_key.offset > end)
1347 if (found_key.offset > cur_offset) {
1348 extent_end = found_key.offset;
1353 fi = btrfs_item_ptr(leaf, path->slots[0],
1354 struct btrfs_file_extent_item);
1355 extent_type = btrfs_file_extent_type(leaf, fi);
1357 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1358 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1359 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1360 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1361 extent_offset = btrfs_file_extent_offset(leaf, fi);
1362 extent_end = found_key.offset +
1363 btrfs_file_extent_num_bytes(leaf, fi);
1365 btrfs_file_extent_disk_num_bytes(leaf, fi);
1366 if (extent_end <= start) {
1370 if (disk_bytenr == 0)
1372 if (btrfs_file_extent_compression(leaf, fi) ||
1373 btrfs_file_extent_encryption(leaf, fi) ||
1374 btrfs_file_extent_other_encoding(leaf, fi))
1376 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1378 if (btrfs_extent_readonly(fs_info, disk_bytenr))
1380 ret = btrfs_cross_ref_exist(root, ino,
1382 extent_offset, disk_bytenr);
1385 * ret could be -EIO if the above fails to read
1389 if (cow_start != (u64)-1)
1390 cur_offset = cow_start;
1394 WARN_ON_ONCE(nolock);
1397 disk_bytenr += extent_offset;
1398 disk_bytenr += cur_offset - found_key.offset;
1399 num_bytes = min(end + 1, extent_end) - cur_offset;
1401 * if there are pending snapshots for this root,
1402 * we fall into common COW way.
1405 err = btrfs_start_write_no_snapshotting(root);
1410 * force cow if csum exists in the range.
1411 * this ensure that csum for a given extent are
1412 * either valid or do not exist.
1414 ret = csum_exist_in_range(fs_info, disk_bytenr,
1418 btrfs_end_write_no_snapshotting(root);
1421 * ret could be -EIO if the above fails to read
1425 if (cow_start != (u64)-1)
1426 cur_offset = cow_start;
1429 WARN_ON_ONCE(nolock);
1432 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr)) {
1434 btrfs_end_write_no_snapshotting(root);
1438 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1439 extent_end = found_key.offset +
1440 btrfs_file_extent_inline_len(leaf,
1441 path->slots[0], fi);
1442 extent_end = ALIGN(extent_end,
1443 fs_info->sectorsize);
1448 if (extent_end <= start) {
1450 if (!nolock && nocow)
1451 btrfs_end_write_no_snapshotting(root);
1453 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1457 if (cow_start == (u64)-1)
1458 cow_start = cur_offset;
1459 cur_offset = extent_end;
1460 if (cur_offset > end)
1466 btrfs_release_path(path);
1467 if (cow_start != (u64)-1) {
1468 ret = cow_file_range(inode, locked_page,
1469 cow_start, found_key.offset - 1,
1470 end, page_started, nr_written, 1,
1473 if (!nolock && nocow)
1474 btrfs_end_write_no_snapshotting(root);
1476 btrfs_dec_nocow_writers(fs_info,
1480 cow_start = (u64)-1;
1483 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1484 u64 orig_start = found_key.offset - extent_offset;
1486 em = create_io_em(inode, cur_offset, num_bytes,
1488 disk_bytenr, /* block_start */
1489 num_bytes, /* block_len */
1490 disk_num_bytes, /* orig_block_len */
1491 ram_bytes, BTRFS_COMPRESS_NONE,
1492 BTRFS_ORDERED_PREALLOC);
1494 if (!nolock && nocow)
1495 btrfs_end_write_no_snapshotting(root);
1497 btrfs_dec_nocow_writers(fs_info,
1502 free_extent_map(em);
1505 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1506 type = BTRFS_ORDERED_PREALLOC;
1508 type = BTRFS_ORDERED_NOCOW;
1511 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1512 num_bytes, num_bytes, type);
1514 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1515 BUG_ON(ret); /* -ENOMEM */
1517 if (root->root_key.objectid ==
1518 BTRFS_DATA_RELOC_TREE_OBJECTID)
1520 * Error handled later, as we must prevent
1521 * extent_clear_unlock_delalloc() in error handler
1522 * from freeing metadata of created ordered extent.
1524 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1527 extent_clear_unlock_delalloc(inode, cur_offset,
1528 cur_offset + num_bytes - 1, end,
1529 locked_page, EXTENT_LOCKED |
1531 EXTENT_CLEAR_DATA_RESV,
1532 PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1534 if (!nolock && nocow)
1535 btrfs_end_write_no_snapshotting(root);
1536 cur_offset = extent_end;
1539 * btrfs_reloc_clone_csums() error, now we're OK to call error
1540 * handler, as metadata for created ordered extent will only
1541 * be freed by btrfs_finish_ordered_io().
1545 if (cur_offset > end)
1548 btrfs_release_path(path);
1550 if (cur_offset <= end && cow_start == (u64)-1) {
1551 cow_start = cur_offset;
1555 if (cow_start != (u64)-1) {
1556 ret = cow_file_range(inode, locked_page, cow_start, end, end,
1557 page_started, nr_written, 1, NULL);
1563 if (ret && cur_offset < end)
1564 extent_clear_unlock_delalloc(inode, cur_offset, end, end,
1565 locked_page, EXTENT_LOCKED |
1566 EXTENT_DELALLOC | EXTENT_DEFRAG |
1567 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1569 PAGE_SET_WRITEBACK |
1570 PAGE_END_WRITEBACK);
1571 btrfs_free_path(path);
1575 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1578 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1579 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1583 * @defrag_bytes is a hint value, no spinlock held here,
1584 * if is not zero, it means the file is defragging.
1585 * Force cow if given extent needs to be defragged.
1587 if (BTRFS_I(inode)->defrag_bytes &&
1588 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1589 EXTENT_DEFRAG, 0, NULL))
1596 * extent_io.c call back to do delayed allocation processing
1598 static int run_delalloc_range(void *private_data, struct page *locked_page,
1599 u64 start, u64 end, int *page_started,
1600 unsigned long *nr_written,
1601 struct writeback_control *wbc)
1603 struct inode *inode = private_data;
1605 int force_cow = need_force_cow(inode, start, end);
1606 unsigned int write_flags = wbc_to_write_flags(wbc);
1608 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1609 ret = run_delalloc_nocow(inode, locked_page, start, end,
1610 page_started, 1, nr_written);
1611 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1612 ret = run_delalloc_nocow(inode, locked_page, start, end,
1613 page_started, 0, nr_written);
1614 } else if (!inode_need_compress(inode, start, end)) {
1615 ret = cow_file_range(inode, locked_page, start, end, end,
1616 page_started, nr_written, 1, NULL);
1618 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1619 &BTRFS_I(inode)->runtime_flags);
1620 ret = cow_file_range_async(inode, locked_page, start, end,
1621 page_started, nr_written,
1625 btrfs_cleanup_ordered_extents(inode, start, end - start + 1);
1629 static void btrfs_split_extent_hook(void *private_data,
1630 struct extent_state *orig, u64 split)
1632 struct inode *inode = private_data;
1635 /* not delalloc, ignore it */
1636 if (!(orig->state & EXTENT_DELALLOC))
1639 size = orig->end - orig->start + 1;
1640 if (size > BTRFS_MAX_EXTENT_SIZE) {
1645 * See the explanation in btrfs_merge_extent_hook, the same
1646 * applies here, just in reverse.
1648 new_size = orig->end - split + 1;
1649 num_extents = count_max_extents(new_size);
1650 new_size = split - orig->start;
1651 num_extents += count_max_extents(new_size);
1652 if (count_max_extents(size) >= num_extents)
1656 spin_lock(&BTRFS_I(inode)->lock);
1657 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1658 spin_unlock(&BTRFS_I(inode)->lock);
1662 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1663 * extents so we can keep track of new extents that are just merged onto old
1664 * extents, such as when we are doing sequential writes, so we can properly
1665 * account for the metadata space we'll need.
1667 static void btrfs_merge_extent_hook(void *private_data,
1668 struct extent_state *new,
1669 struct extent_state *other)
1671 struct inode *inode = private_data;
1672 u64 new_size, old_size;
1675 /* not delalloc, ignore it */
1676 if (!(other->state & EXTENT_DELALLOC))
1679 if (new->start > other->start)
1680 new_size = new->end - other->start + 1;
1682 new_size = other->end - new->start + 1;
1684 /* we're not bigger than the max, unreserve the space and go */
1685 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1686 spin_lock(&BTRFS_I(inode)->lock);
1687 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1688 spin_unlock(&BTRFS_I(inode)->lock);
1693 * We have to add up either side to figure out how many extents were
1694 * accounted for before we merged into one big extent. If the number of
1695 * extents we accounted for is <= the amount we need for the new range
1696 * then we can return, otherwise drop. Think of it like this
1700 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1701 * need 2 outstanding extents, on one side we have 1 and the other side
1702 * we have 1 so they are == and we can return. But in this case
1704 * [MAX_SIZE+4k][MAX_SIZE+4k]
1706 * Each range on their own accounts for 2 extents, but merged together
1707 * they are only 3 extents worth of accounting, so we need to drop in
1710 old_size = other->end - other->start + 1;
1711 num_extents = count_max_extents(old_size);
1712 old_size = new->end - new->start + 1;
1713 num_extents += count_max_extents(old_size);
1714 if (count_max_extents(new_size) >= num_extents)
1717 spin_lock(&BTRFS_I(inode)->lock);
1718 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1719 spin_unlock(&BTRFS_I(inode)->lock);
1722 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1723 struct inode *inode)
1725 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1727 spin_lock(&root->delalloc_lock);
1728 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1729 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1730 &root->delalloc_inodes);
1731 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1732 &BTRFS_I(inode)->runtime_flags);
1733 root->nr_delalloc_inodes++;
1734 if (root->nr_delalloc_inodes == 1) {
1735 spin_lock(&fs_info->delalloc_root_lock);
1736 BUG_ON(!list_empty(&root->delalloc_root));
1737 list_add_tail(&root->delalloc_root,
1738 &fs_info->delalloc_roots);
1739 spin_unlock(&fs_info->delalloc_root_lock);
1742 spin_unlock(&root->delalloc_lock);
1746 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
1747 struct btrfs_inode *inode)
1749 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1751 if (!list_empty(&inode->delalloc_inodes)) {
1752 list_del_init(&inode->delalloc_inodes);
1753 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1754 &inode->runtime_flags);
1755 root->nr_delalloc_inodes--;
1756 if (!root->nr_delalloc_inodes) {
1757 spin_lock(&fs_info->delalloc_root_lock);
1758 BUG_ON(list_empty(&root->delalloc_root));
1759 list_del_init(&root->delalloc_root);
1760 spin_unlock(&fs_info->delalloc_root_lock);
1765 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1766 struct btrfs_inode *inode)
1768 spin_lock(&root->delalloc_lock);
1769 __btrfs_del_delalloc_inode(root, inode);
1770 spin_unlock(&root->delalloc_lock);
1774 * extent_io.c set_bit_hook, used to track delayed allocation
1775 * bytes in this file, and to maintain the list of inodes that
1776 * have pending delalloc work to be done.
1778 static void btrfs_set_bit_hook(void *private_data,
1779 struct extent_state *state, unsigned *bits)
1781 struct inode *inode = private_data;
1783 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1785 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1788 * set_bit and clear bit hooks normally require _irqsave/restore
1789 * but in this case, we are only testing for the DELALLOC
1790 * bit, which is only set or cleared with irqs on
1792 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1793 struct btrfs_root *root = BTRFS_I(inode)->root;
1794 u64 len = state->end + 1 - state->start;
1795 u32 num_extents = count_max_extents(len);
1796 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
1798 spin_lock(&BTRFS_I(inode)->lock);
1799 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
1800 spin_unlock(&BTRFS_I(inode)->lock);
1802 /* For sanity tests */
1803 if (btrfs_is_testing(fs_info))
1806 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
1807 fs_info->delalloc_batch);
1808 spin_lock(&BTRFS_I(inode)->lock);
1809 BTRFS_I(inode)->delalloc_bytes += len;
1810 if (*bits & EXTENT_DEFRAG)
1811 BTRFS_I(inode)->defrag_bytes += len;
1812 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1813 &BTRFS_I(inode)->runtime_flags))
1814 btrfs_add_delalloc_inodes(root, inode);
1815 spin_unlock(&BTRFS_I(inode)->lock);
1818 if (!(state->state & EXTENT_DELALLOC_NEW) &&
1819 (*bits & EXTENT_DELALLOC_NEW)) {
1820 spin_lock(&BTRFS_I(inode)->lock);
1821 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
1823 spin_unlock(&BTRFS_I(inode)->lock);
1828 * extent_io.c clear_bit_hook, see set_bit_hook for why
1830 static void btrfs_clear_bit_hook(void *private_data,
1831 struct extent_state *state,
1834 struct btrfs_inode *inode = BTRFS_I((struct inode *)private_data);
1835 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1836 u64 len = state->end + 1 - state->start;
1837 u32 num_extents = count_max_extents(len);
1839 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
1840 spin_lock(&inode->lock);
1841 inode->defrag_bytes -= len;
1842 spin_unlock(&inode->lock);
1846 * set_bit and clear bit hooks normally require _irqsave/restore
1847 * but in this case, we are only testing for the DELALLOC
1848 * bit, which is only set or cleared with irqs on
1850 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1851 struct btrfs_root *root = inode->root;
1852 bool do_list = !btrfs_is_free_space_inode(inode);
1854 spin_lock(&inode->lock);
1855 btrfs_mod_outstanding_extents(inode, -num_extents);
1856 spin_unlock(&inode->lock);
1859 * We don't reserve metadata space for space cache inodes so we
1860 * don't need to call dellalloc_release_metadata if there is an
1863 if (*bits & EXTENT_CLEAR_META_RESV &&
1864 root != fs_info->tree_root)
1865 btrfs_delalloc_release_metadata(inode, len, false);
1867 /* For sanity tests. */
1868 if (btrfs_is_testing(fs_info))
1871 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
1872 do_list && !(state->state & EXTENT_NORESERVE) &&
1873 (*bits & EXTENT_CLEAR_DATA_RESV))
1874 btrfs_free_reserved_data_space_noquota(
1878 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
1879 fs_info->delalloc_batch);
1880 spin_lock(&inode->lock);
1881 inode->delalloc_bytes -= len;
1882 if (do_list && inode->delalloc_bytes == 0 &&
1883 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1884 &inode->runtime_flags))
1885 btrfs_del_delalloc_inode(root, inode);
1886 spin_unlock(&inode->lock);
1889 if ((state->state & EXTENT_DELALLOC_NEW) &&
1890 (*bits & EXTENT_DELALLOC_NEW)) {
1891 spin_lock(&inode->lock);
1892 ASSERT(inode->new_delalloc_bytes >= len);
1893 inode->new_delalloc_bytes -= len;
1894 spin_unlock(&inode->lock);
1899 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1900 * we don't create bios that span stripes or chunks
1902 * return 1 if page cannot be merged to bio
1903 * return 0 if page can be merged to bio
1904 * return error otherwise
1906 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1907 size_t size, struct bio *bio,
1908 unsigned long bio_flags)
1910 struct inode *inode = page->mapping->host;
1911 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1912 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1917 if (bio_flags & EXTENT_BIO_COMPRESSED)
1920 length = bio->bi_iter.bi_size;
1921 map_length = length;
1922 ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
1926 if (map_length < length + size)
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 static blk_status_t btrfs_submit_bio_start(void *private_data, struct bio *bio,
1942 struct inode *inode = private_data;
1943 blk_status_t ret = 0;
1945 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1946 BUG_ON(ret); /* -ENOMEM */
1951 * in order to insert checksums into the metadata in large chunks,
1952 * we wait until bio submission time. All the pages in the bio are
1953 * checksummed and sums are attached onto the ordered extent record.
1955 * At IO completion time the cums attached on the ordered extent record
1956 * are inserted into the btree
1958 static blk_status_t btrfs_submit_bio_done(void *private_data, struct bio *bio,
1961 struct inode *inode = private_data;
1962 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1965 ret = btrfs_map_bio(fs_info, bio, mirror_num, 1);
1967 bio->bi_status = ret;
1974 * extent_io.c submission hook. This does the right thing for csum calculation
1975 * on write, or reading the csums from the tree before a read.
1977 * Rules about async/sync submit,
1978 * a) read: sync submit
1980 * b) write without checksum: sync submit
1982 * c) write with checksum:
1983 * c-1) if bio is issued by fsync: sync submit
1984 * (sync_writers != 0)
1986 * c-2) if root is reloc root: sync submit
1987 * (only in case of buffered IO)
1989 * c-3) otherwise: async submit
1991 static blk_status_t btrfs_submit_bio_hook(void *private_data, struct bio *bio,
1992 int mirror_num, unsigned long bio_flags,
1995 struct inode *inode = private_data;
1996 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1997 struct btrfs_root *root = BTRFS_I(inode)->root;
1998 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1999 blk_status_t ret = 0;
2001 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
2003 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
2005 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
2006 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
2008 if (bio_op(bio) != REQ_OP_WRITE) {
2009 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
2013 if (bio_flags & EXTENT_BIO_COMPRESSED) {
2014 ret = btrfs_submit_compressed_read(inode, bio,
2018 } else if (!skip_sum) {
2019 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
2024 } else if (async && !skip_sum) {
2025 /* csum items have already been cloned */
2026 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2028 /* we're doing a write, do the async checksumming */
2029 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags,
2031 btrfs_submit_bio_start,
2032 btrfs_submit_bio_done);
2034 } else if (!skip_sum) {
2035 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
2041 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
2045 bio->bi_status = ret;
2052 * given a list of ordered sums record them in the inode. This happens
2053 * at IO completion time based on sums calculated at bio submission time.
2055 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
2056 struct inode *inode, struct list_head *list)
2058 struct btrfs_ordered_sum *sum;
2061 list_for_each_entry(sum, list, list) {
2062 trans->adding_csums = true;
2063 ret = btrfs_csum_file_blocks(trans,
2064 BTRFS_I(inode)->root->fs_info->csum_root, sum);
2065 trans->adding_csums = false;
2072 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
2073 unsigned int extra_bits,
2074 struct extent_state **cached_state, int dedupe)
2076 WARN_ON((end & (PAGE_SIZE - 1)) == 0);
2077 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2078 extra_bits, cached_state);
2081 /* see btrfs_writepage_start_hook for details on why this is required */
2082 struct btrfs_writepage_fixup {
2084 struct btrfs_work work;
2087 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2089 struct btrfs_writepage_fixup *fixup;
2090 struct btrfs_ordered_extent *ordered;
2091 struct extent_state *cached_state = NULL;
2092 struct extent_changeset *data_reserved = NULL;
2094 struct inode *inode;
2099 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2103 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2104 ClearPageChecked(page);
2108 inode = page->mapping->host;
2109 page_start = page_offset(page);
2110 page_end = page_offset(page) + PAGE_SIZE - 1;
2112 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2115 /* already ordered? We're done */
2116 if (PagePrivate2(page))
2119 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
2122 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2123 page_end, &cached_state);
2125 btrfs_start_ordered_extent(inode, ordered, 1);
2126 btrfs_put_ordered_extent(ordered);
2130 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2133 mapping_set_error(page->mapping, ret);
2134 end_extent_writepage(page, ret, page_start, page_end);
2135 ClearPageChecked(page);
2139 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2142 mapping_set_error(page->mapping, ret);
2143 end_extent_writepage(page, ret, page_start, page_end);
2144 ClearPageChecked(page);
2148 ClearPageChecked(page);
2149 set_page_dirty(page);
2150 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, false);
2152 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2158 extent_changeset_free(data_reserved);
2162 * There are a few paths in the higher layers of the kernel that directly
2163 * set the page dirty bit without asking the filesystem if it is a
2164 * good idea. This causes problems because we want to make sure COW
2165 * properly happens and the data=ordered rules are followed.
2167 * In our case any range that doesn't have the ORDERED bit set
2168 * hasn't been properly setup for IO. We kick off an async process
2169 * to fix it up. The async helper will wait for ordered extents, set
2170 * the delalloc bit and make it safe to write the page.
2172 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2174 struct inode *inode = page->mapping->host;
2175 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2176 struct btrfs_writepage_fixup *fixup;
2178 /* this page is properly in the ordered list */
2179 if (TestClearPagePrivate2(page))
2182 if (PageChecked(page))
2185 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2189 SetPageChecked(page);
2191 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2192 btrfs_writepage_fixup_worker, NULL, NULL);
2194 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2198 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2199 struct inode *inode, u64 file_pos,
2200 u64 disk_bytenr, u64 disk_num_bytes,
2201 u64 num_bytes, u64 ram_bytes,
2202 u8 compression, u8 encryption,
2203 u16 other_encoding, int extent_type)
2205 struct btrfs_root *root = BTRFS_I(inode)->root;
2206 struct btrfs_file_extent_item *fi;
2207 struct btrfs_path *path;
2208 struct extent_buffer *leaf;
2209 struct btrfs_key ins;
2211 int extent_inserted = 0;
2214 path = btrfs_alloc_path();
2219 * we may be replacing one extent in the tree with another.
2220 * The new extent is pinned in the extent map, and we don't want
2221 * to drop it from the cache until it is completely in the btree.
2223 * So, tell btrfs_drop_extents to leave this extent in the cache.
2224 * the caller is expected to unpin it and allow it to be merged
2227 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2228 file_pos + num_bytes, NULL, 0,
2229 1, sizeof(*fi), &extent_inserted);
2233 if (!extent_inserted) {
2234 ins.objectid = btrfs_ino(BTRFS_I(inode));
2235 ins.offset = file_pos;
2236 ins.type = BTRFS_EXTENT_DATA_KEY;
2238 path->leave_spinning = 1;
2239 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2244 leaf = path->nodes[0];
2245 fi = btrfs_item_ptr(leaf, path->slots[0],
2246 struct btrfs_file_extent_item);
2247 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2248 btrfs_set_file_extent_type(leaf, fi, extent_type);
2249 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2250 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2251 btrfs_set_file_extent_offset(leaf, fi, 0);
2252 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2253 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2254 btrfs_set_file_extent_compression(leaf, fi, compression);
2255 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2256 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2258 btrfs_mark_buffer_dirty(leaf);
2259 btrfs_release_path(path);
2261 inode_add_bytes(inode, num_bytes);
2263 ins.objectid = disk_bytenr;
2264 ins.offset = disk_num_bytes;
2265 ins.type = BTRFS_EXTENT_ITEM_KEY;
2268 * Release the reserved range from inode dirty range map, as it is
2269 * already moved into delayed_ref_head
2271 ret = btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2275 ret = btrfs_alloc_reserved_file_extent(trans, root,
2276 btrfs_ino(BTRFS_I(inode)),
2277 file_pos, qg_released, &ins);
2279 btrfs_free_path(path);
2284 /* snapshot-aware defrag */
2285 struct sa_defrag_extent_backref {
2286 struct rb_node node;
2287 struct old_sa_defrag_extent *old;
2296 struct old_sa_defrag_extent {
2297 struct list_head list;
2298 struct new_sa_defrag_extent *new;
2307 struct new_sa_defrag_extent {
2308 struct rb_root root;
2309 struct list_head head;
2310 struct btrfs_path *path;
2311 struct inode *inode;
2319 static int backref_comp(struct sa_defrag_extent_backref *b1,
2320 struct sa_defrag_extent_backref *b2)
2322 if (b1->root_id < b2->root_id)
2324 else if (b1->root_id > b2->root_id)
2327 if (b1->inum < b2->inum)
2329 else if (b1->inum > b2->inum)
2332 if (b1->file_pos < b2->file_pos)
2334 else if (b1->file_pos > b2->file_pos)
2338 * [------------------------------] ===> (a range of space)
2339 * |<--->| |<---->| =============> (fs/file tree A)
2340 * |<---------------------------->| ===> (fs/file tree B)
2342 * A range of space can refer to two file extents in one tree while
2343 * refer to only one file extent in another tree.
2345 * So we may process a disk offset more than one time(two extents in A)
2346 * and locate at the same extent(one extent in B), then insert two same
2347 * backrefs(both refer to the extent in B).
2352 static void backref_insert(struct rb_root *root,
2353 struct sa_defrag_extent_backref *backref)
2355 struct rb_node **p = &root->rb_node;
2356 struct rb_node *parent = NULL;
2357 struct sa_defrag_extent_backref *entry;
2362 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2364 ret = backref_comp(backref, entry);
2368 p = &(*p)->rb_right;
2371 rb_link_node(&backref->node, parent, p);
2372 rb_insert_color(&backref->node, root);
2376 * Note the backref might has changed, and in this case we just return 0.
2378 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2381 struct btrfs_file_extent_item *extent;
2382 struct old_sa_defrag_extent *old = ctx;
2383 struct new_sa_defrag_extent *new = old->new;
2384 struct btrfs_path *path = new->path;
2385 struct btrfs_key key;
2386 struct btrfs_root *root;
2387 struct sa_defrag_extent_backref *backref;
2388 struct extent_buffer *leaf;
2389 struct inode *inode = new->inode;
2390 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2396 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2397 inum == btrfs_ino(BTRFS_I(inode)))
2400 key.objectid = root_id;
2401 key.type = BTRFS_ROOT_ITEM_KEY;
2402 key.offset = (u64)-1;
2404 root = btrfs_read_fs_root_no_name(fs_info, &key);
2406 if (PTR_ERR(root) == -ENOENT)
2409 btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu",
2410 inum, offset, root_id);
2411 return PTR_ERR(root);
2414 key.objectid = inum;
2415 key.type = BTRFS_EXTENT_DATA_KEY;
2416 if (offset > (u64)-1 << 32)
2419 key.offset = offset;
2421 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2422 if (WARN_ON(ret < 0))
2429 leaf = path->nodes[0];
2430 slot = path->slots[0];
2432 if (slot >= btrfs_header_nritems(leaf)) {
2433 ret = btrfs_next_leaf(root, path);
2436 } else if (ret > 0) {
2445 btrfs_item_key_to_cpu(leaf, &key, slot);
2447 if (key.objectid > inum)
2450 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2453 extent = btrfs_item_ptr(leaf, slot,
2454 struct btrfs_file_extent_item);
2456 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2460 * 'offset' refers to the exact key.offset,
2461 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2462 * (key.offset - extent_offset).
2464 if (key.offset != offset)
2467 extent_offset = btrfs_file_extent_offset(leaf, extent);
2468 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2470 if (extent_offset >= old->extent_offset + old->offset +
2471 old->len || extent_offset + num_bytes <=
2472 old->extent_offset + old->offset)
2477 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2483 backref->root_id = root_id;
2484 backref->inum = inum;
2485 backref->file_pos = offset;
2486 backref->num_bytes = num_bytes;
2487 backref->extent_offset = extent_offset;
2488 backref->generation = btrfs_file_extent_generation(leaf, extent);
2490 backref_insert(&new->root, backref);
2493 btrfs_release_path(path);
2498 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2499 struct new_sa_defrag_extent *new)
2501 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2502 struct old_sa_defrag_extent *old, *tmp;
2507 list_for_each_entry_safe(old, tmp, &new->head, list) {
2508 ret = iterate_inodes_from_logical(old->bytenr +
2509 old->extent_offset, fs_info,
2510 path, record_one_backref,
2512 if (ret < 0 && ret != -ENOENT)
2515 /* no backref to be processed for this extent */
2517 list_del(&old->list);
2522 if (list_empty(&new->head))
2528 static int relink_is_mergable(struct extent_buffer *leaf,
2529 struct btrfs_file_extent_item *fi,
2530 struct new_sa_defrag_extent *new)
2532 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2535 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2538 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2541 if (btrfs_file_extent_encryption(leaf, fi) ||
2542 btrfs_file_extent_other_encoding(leaf, fi))
2549 * Note the backref might has changed, and in this case we just return 0.
2551 static noinline int relink_extent_backref(struct btrfs_path *path,
2552 struct sa_defrag_extent_backref *prev,
2553 struct sa_defrag_extent_backref *backref)
2555 struct btrfs_file_extent_item *extent;
2556 struct btrfs_file_extent_item *item;
2557 struct btrfs_ordered_extent *ordered;
2558 struct btrfs_trans_handle *trans;
2559 struct btrfs_root *root;
2560 struct btrfs_key key;
2561 struct extent_buffer *leaf;
2562 struct old_sa_defrag_extent *old = backref->old;
2563 struct new_sa_defrag_extent *new = old->new;
2564 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2565 struct inode *inode;
2566 struct extent_state *cached = NULL;
2575 if (prev && prev->root_id == backref->root_id &&
2576 prev->inum == backref->inum &&
2577 prev->file_pos + prev->num_bytes == backref->file_pos)
2580 /* step 1: get root */
2581 key.objectid = backref->root_id;
2582 key.type = BTRFS_ROOT_ITEM_KEY;
2583 key.offset = (u64)-1;
2585 index = srcu_read_lock(&fs_info->subvol_srcu);
2587 root = btrfs_read_fs_root_no_name(fs_info, &key);
2589 srcu_read_unlock(&fs_info->subvol_srcu, index);
2590 if (PTR_ERR(root) == -ENOENT)
2592 return PTR_ERR(root);
2595 if (btrfs_root_readonly(root)) {
2596 srcu_read_unlock(&fs_info->subvol_srcu, index);
2600 /* step 2: get inode */
2601 key.objectid = backref->inum;
2602 key.type = BTRFS_INODE_ITEM_KEY;
2605 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2606 if (IS_ERR(inode)) {
2607 srcu_read_unlock(&fs_info->subvol_srcu, index);
2611 srcu_read_unlock(&fs_info->subvol_srcu, index);
2613 /* step 3: relink backref */
2614 lock_start = backref->file_pos;
2615 lock_end = backref->file_pos + backref->num_bytes - 1;
2616 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2619 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2621 btrfs_put_ordered_extent(ordered);
2625 trans = btrfs_join_transaction(root);
2626 if (IS_ERR(trans)) {
2627 ret = PTR_ERR(trans);
2631 key.objectid = backref->inum;
2632 key.type = BTRFS_EXTENT_DATA_KEY;
2633 key.offset = backref->file_pos;
2635 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2638 } else if (ret > 0) {
2643 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2644 struct btrfs_file_extent_item);
2646 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2647 backref->generation)
2650 btrfs_release_path(path);
2652 start = backref->file_pos;
2653 if (backref->extent_offset < old->extent_offset + old->offset)
2654 start += old->extent_offset + old->offset -
2655 backref->extent_offset;
2657 len = min(backref->extent_offset + backref->num_bytes,
2658 old->extent_offset + old->offset + old->len);
2659 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2661 ret = btrfs_drop_extents(trans, root, inode, start,
2666 key.objectid = btrfs_ino(BTRFS_I(inode));
2667 key.type = BTRFS_EXTENT_DATA_KEY;
2670 path->leave_spinning = 1;
2672 struct btrfs_file_extent_item *fi;
2674 struct btrfs_key found_key;
2676 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2681 leaf = path->nodes[0];
2682 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2684 fi = btrfs_item_ptr(leaf, path->slots[0],
2685 struct btrfs_file_extent_item);
2686 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2688 if (extent_len + found_key.offset == start &&
2689 relink_is_mergable(leaf, fi, new)) {
2690 btrfs_set_file_extent_num_bytes(leaf, fi,
2692 btrfs_mark_buffer_dirty(leaf);
2693 inode_add_bytes(inode, len);
2699 btrfs_release_path(path);
2704 ret = btrfs_insert_empty_item(trans, root, path, &key,
2707 btrfs_abort_transaction(trans, ret);
2711 leaf = path->nodes[0];
2712 item = btrfs_item_ptr(leaf, path->slots[0],
2713 struct btrfs_file_extent_item);
2714 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2715 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2716 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2717 btrfs_set_file_extent_num_bytes(leaf, item, len);
2718 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2719 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2720 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2721 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2722 btrfs_set_file_extent_encryption(leaf, item, 0);
2723 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2725 btrfs_mark_buffer_dirty(leaf);
2726 inode_add_bytes(inode, len);
2727 btrfs_release_path(path);
2729 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2731 backref->root_id, backref->inum,
2732 new->file_pos); /* start - extent_offset */
2734 btrfs_abort_transaction(trans, ret);
2740 btrfs_release_path(path);
2741 path->leave_spinning = 0;
2742 btrfs_end_transaction(trans);
2744 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2750 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2752 struct old_sa_defrag_extent *old, *tmp;
2757 list_for_each_entry_safe(old, tmp, &new->head, list) {
2763 static void relink_file_extents(struct new_sa_defrag_extent *new)
2765 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2766 struct btrfs_path *path;
2767 struct sa_defrag_extent_backref *backref;
2768 struct sa_defrag_extent_backref *prev = NULL;
2769 struct inode *inode;
2770 struct rb_node *node;
2775 path = btrfs_alloc_path();
2779 if (!record_extent_backrefs(path, new)) {
2780 btrfs_free_path(path);
2783 btrfs_release_path(path);
2786 node = rb_first(&new->root);
2789 rb_erase(node, &new->root);
2791 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2793 ret = relink_extent_backref(path, prev, backref);
2806 btrfs_free_path(path);
2808 free_sa_defrag_extent(new);
2810 atomic_dec(&fs_info->defrag_running);
2811 wake_up(&fs_info->transaction_wait);
2814 static struct new_sa_defrag_extent *
2815 record_old_file_extents(struct inode *inode,
2816 struct btrfs_ordered_extent *ordered)
2818 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2819 struct btrfs_root *root = BTRFS_I(inode)->root;
2820 struct btrfs_path *path;
2821 struct btrfs_key key;
2822 struct old_sa_defrag_extent *old;
2823 struct new_sa_defrag_extent *new;
2826 new = kmalloc(sizeof(*new), GFP_NOFS);
2831 new->file_pos = ordered->file_offset;
2832 new->len = ordered->len;
2833 new->bytenr = ordered->start;
2834 new->disk_len = ordered->disk_len;
2835 new->compress_type = ordered->compress_type;
2836 new->root = RB_ROOT;
2837 INIT_LIST_HEAD(&new->head);
2839 path = btrfs_alloc_path();
2843 key.objectid = btrfs_ino(BTRFS_I(inode));
2844 key.type = BTRFS_EXTENT_DATA_KEY;
2845 key.offset = new->file_pos;
2847 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2850 if (ret > 0 && path->slots[0] > 0)
2853 /* find out all the old extents for the file range */
2855 struct btrfs_file_extent_item *extent;
2856 struct extent_buffer *l;
2865 slot = path->slots[0];
2867 if (slot >= btrfs_header_nritems(l)) {
2868 ret = btrfs_next_leaf(root, path);
2876 btrfs_item_key_to_cpu(l, &key, slot);
2878 if (key.objectid != btrfs_ino(BTRFS_I(inode)))
2880 if (key.type != BTRFS_EXTENT_DATA_KEY)
2882 if (key.offset >= new->file_pos + new->len)
2885 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2887 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2888 if (key.offset + num_bytes < new->file_pos)
2891 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2895 extent_offset = btrfs_file_extent_offset(l, extent);
2897 old = kmalloc(sizeof(*old), GFP_NOFS);
2901 offset = max(new->file_pos, key.offset);
2902 end = min(new->file_pos + new->len, key.offset + num_bytes);
2904 old->bytenr = disk_bytenr;
2905 old->extent_offset = extent_offset;
2906 old->offset = offset - key.offset;
2907 old->len = end - offset;
2910 list_add_tail(&old->list, &new->head);
2916 btrfs_free_path(path);
2917 atomic_inc(&fs_info->defrag_running);
2922 btrfs_free_path(path);
2924 free_sa_defrag_extent(new);
2928 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2931 struct btrfs_block_group_cache *cache;
2933 cache = btrfs_lookup_block_group(fs_info, start);
2936 spin_lock(&cache->lock);
2937 cache->delalloc_bytes -= len;
2938 spin_unlock(&cache->lock);
2940 btrfs_put_block_group(cache);
2943 /* as ordered data IO finishes, this gets called so we can finish
2944 * an ordered extent if the range of bytes in the file it covers are
2947 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2949 struct inode *inode = ordered_extent->inode;
2950 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2951 struct btrfs_root *root = BTRFS_I(inode)->root;
2952 struct btrfs_trans_handle *trans = NULL;
2953 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2954 struct extent_state *cached_state = NULL;
2955 struct new_sa_defrag_extent *new = NULL;
2956 int compress_type = 0;
2958 u64 logical_len = ordered_extent->len;
2960 bool truncated = false;
2961 bool range_locked = false;
2962 bool clear_new_delalloc_bytes = false;
2964 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2965 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2966 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2967 clear_new_delalloc_bytes = true;
2969 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
2971 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2976 btrfs_free_io_failure_record(BTRFS_I(inode),
2977 ordered_extent->file_offset,
2978 ordered_extent->file_offset +
2979 ordered_extent->len - 1);
2981 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2983 logical_len = ordered_extent->truncated_len;
2984 /* Truncated the entire extent, don't bother adding */
2989 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2990 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2993 * For mwrite(mmap + memset to write) case, we still reserve
2994 * space for NOCOW range.
2995 * As NOCOW won't cause a new delayed ref, just free the space
2997 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
2998 ordered_extent->len);
2999 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
3001 trans = btrfs_join_transaction_nolock(root);
3003 trans = btrfs_join_transaction(root);
3004 if (IS_ERR(trans)) {
3005 ret = PTR_ERR(trans);
3009 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
3010 ret = btrfs_update_inode_fallback(trans, root, inode);
3011 if (ret) /* -ENOMEM or corruption */
3012 btrfs_abort_transaction(trans, ret);
3016 range_locked = true;
3017 lock_extent_bits(io_tree, ordered_extent->file_offset,
3018 ordered_extent->file_offset + ordered_extent->len - 1,
3021 ret = test_range_bit(io_tree, ordered_extent->file_offset,
3022 ordered_extent->file_offset + ordered_extent->len - 1,
3023 EXTENT_DEFRAG, 0, cached_state);
3025 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
3026 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
3027 /* the inode is shared */
3028 new = record_old_file_extents(inode, ordered_extent);
3030 clear_extent_bit(io_tree, ordered_extent->file_offset,
3031 ordered_extent->file_offset + ordered_extent->len - 1,
3032 EXTENT_DEFRAG, 0, 0, &cached_state);
3036 trans = btrfs_join_transaction_nolock(root);
3038 trans = btrfs_join_transaction(root);
3039 if (IS_ERR(trans)) {
3040 ret = PTR_ERR(trans);
3045 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
3047 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3048 compress_type = ordered_extent->compress_type;
3049 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3050 BUG_ON(compress_type);
3051 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
3052 ordered_extent->len);
3053 ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
3054 ordered_extent->file_offset,
3055 ordered_extent->file_offset +
3058 BUG_ON(root == fs_info->tree_root);
3059 ret = insert_reserved_file_extent(trans, inode,
3060 ordered_extent->file_offset,
3061 ordered_extent->start,
3062 ordered_extent->disk_len,
3063 logical_len, logical_len,
3064 compress_type, 0, 0,
3065 BTRFS_FILE_EXTENT_REG);
3067 btrfs_release_delalloc_bytes(fs_info,
3068 ordered_extent->start,
3069 ordered_extent->disk_len);
3071 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
3072 ordered_extent->file_offset, ordered_extent->len,
3075 btrfs_abort_transaction(trans, ret);
3079 ret = add_pending_csums(trans, inode, &ordered_extent->list);
3081 btrfs_abort_transaction(trans, ret);
3085 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
3086 ret = btrfs_update_inode_fallback(trans, root, inode);
3087 if (ret) { /* -ENOMEM or corruption */
3088 btrfs_abort_transaction(trans, ret);
3093 if (range_locked || clear_new_delalloc_bytes) {
3094 unsigned int clear_bits = 0;
3097 clear_bits |= EXTENT_LOCKED;
3098 if (clear_new_delalloc_bytes)
3099 clear_bits |= EXTENT_DELALLOC_NEW;
3100 clear_extent_bit(&BTRFS_I(inode)->io_tree,
3101 ordered_extent->file_offset,
3102 ordered_extent->file_offset +
3103 ordered_extent->len - 1,
3105 (clear_bits & EXTENT_LOCKED) ? 1 : 0,
3110 btrfs_end_transaction(trans);
3112 if (ret || truncated) {
3116 start = ordered_extent->file_offset + logical_len;
3118 start = ordered_extent->file_offset;
3119 end = ordered_extent->file_offset + ordered_extent->len - 1;
3120 clear_extent_uptodate(io_tree, start, end, NULL);
3122 /* Drop the cache for the part of the extent we didn't write. */
3123 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
3126 * If the ordered extent had an IOERR or something else went
3127 * wrong we need to return the space for this ordered extent
3128 * back to the allocator. We only free the extent in the
3129 * truncated case if we didn't write out the extent at all.
3131 if ((ret || !logical_len) &&
3132 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3133 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
3134 btrfs_free_reserved_extent(fs_info,
3135 ordered_extent->start,
3136 ordered_extent->disk_len, 1);
3141 * This needs to be done to make sure anybody waiting knows we are done
3142 * updating everything for this ordered extent.
3144 btrfs_remove_ordered_extent(inode, ordered_extent);
3146 /* for snapshot-aware defrag */
3149 free_sa_defrag_extent(new);
3150 atomic_dec(&fs_info->defrag_running);
3152 relink_file_extents(new);
3157 btrfs_put_ordered_extent(ordered_extent);
3158 /* once for the tree */
3159 btrfs_put_ordered_extent(ordered_extent);
3164 static void finish_ordered_fn(struct btrfs_work *work)
3166 struct btrfs_ordered_extent *ordered_extent;
3167 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3168 btrfs_finish_ordered_io(ordered_extent);
3171 static void btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3172 struct extent_state *state, int uptodate)
3174 struct inode *inode = page->mapping->host;
3175 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3176 struct btrfs_ordered_extent *ordered_extent = NULL;
3177 struct btrfs_workqueue *wq;
3178 btrfs_work_func_t func;
3180 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3182 ClearPagePrivate2(page);
3183 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3184 end - start + 1, uptodate))
3187 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
3188 wq = fs_info->endio_freespace_worker;
3189 func = btrfs_freespace_write_helper;
3191 wq = fs_info->endio_write_workers;
3192 func = btrfs_endio_write_helper;
3195 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3197 btrfs_queue_work(wq, &ordered_extent->work);
3200 static int __readpage_endio_check(struct inode *inode,
3201 struct btrfs_io_bio *io_bio,
3202 int icsum, struct page *page,
3203 int pgoff, u64 start, size_t len)
3209 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3211 kaddr = kmap_atomic(page);
3212 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3213 btrfs_csum_final(csum, (u8 *)&csum);
3214 if (csum != csum_expected)
3217 kunmap_atomic(kaddr);
3220 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3221 io_bio->mirror_num);
3222 memset(kaddr + pgoff, 1, len);
3223 flush_dcache_page(page);
3224 kunmap_atomic(kaddr);
3229 * when reads are done, we need to check csums to verify the data is correct
3230 * if there's a match, we allow the bio to finish. If not, the code in
3231 * extent_io.c will try to find good copies for us.
3233 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3234 u64 phy_offset, struct page *page,
3235 u64 start, u64 end, int mirror)
3237 size_t offset = start - page_offset(page);
3238 struct inode *inode = page->mapping->host;
3239 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3240 struct btrfs_root *root = BTRFS_I(inode)->root;
3242 if (PageChecked(page)) {
3243 ClearPageChecked(page);
3247 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3250 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3251 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3252 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3256 phy_offset >>= inode->i_sb->s_blocksize_bits;
3257 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3258 start, (size_t)(end - start + 1));
3262 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3264 * @inode: The inode we want to perform iput on
3266 * This function uses the generic vfs_inode::i_count to track whether we should
3267 * just decrement it (in case it's > 1) or if this is the last iput then link
3268 * the inode to the delayed iput machinery. Delayed iputs are processed at
3269 * transaction commit time/superblock commit/cleaner kthread.
3271 void btrfs_add_delayed_iput(struct inode *inode)
3273 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3274 struct btrfs_inode *binode = BTRFS_I(inode);
3276 if (atomic_add_unless(&inode->i_count, -1, 1))
3279 spin_lock(&fs_info->delayed_iput_lock);
3280 ASSERT(list_empty(&binode->delayed_iput));
3281 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3282 spin_unlock(&fs_info->delayed_iput_lock);
3285 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3288 spin_lock(&fs_info->delayed_iput_lock);
3289 while (!list_empty(&fs_info->delayed_iputs)) {
3290 struct btrfs_inode *inode;
3292 inode = list_first_entry(&fs_info->delayed_iputs,
3293 struct btrfs_inode, delayed_iput);
3294 list_del_init(&inode->delayed_iput);
3295 spin_unlock(&fs_info->delayed_iput_lock);
3296 iput(&inode->vfs_inode);
3297 spin_lock(&fs_info->delayed_iput_lock);
3299 spin_unlock(&fs_info->delayed_iput_lock);
3303 * This is called in transaction commit time. If there are no orphan
3304 * files in the subvolume, it removes orphan item and frees block_rsv
3307 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3308 struct btrfs_root *root)
3310 struct btrfs_fs_info *fs_info = root->fs_info;
3311 struct btrfs_block_rsv *block_rsv;
3314 if (atomic_read(&root->orphan_inodes) ||
3315 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3318 spin_lock(&root->orphan_lock);
3319 if (atomic_read(&root->orphan_inodes)) {
3320 spin_unlock(&root->orphan_lock);
3324 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3325 spin_unlock(&root->orphan_lock);
3329 block_rsv = root->orphan_block_rsv;
3330 root->orphan_block_rsv = NULL;
3331 spin_unlock(&root->orphan_lock);
3333 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3334 btrfs_root_refs(&root->root_item) > 0) {
3335 ret = btrfs_del_orphan_item(trans, fs_info->tree_root,
3336 root->root_key.objectid);
3338 btrfs_abort_transaction(trans, ret);
3340 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3345 WARN_ON(block_rsv->size > 0);
3346 btrfs_free_block_rsv(fs_info, block_rsv);
3351 * This creates an orphan entry for the given inode in case something goes
3352 * wrong in the middle of an unlink/truncate.
3354 * NOTE: caller of this function should reserve 5 units of metadata for
3357 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3358 struct btrfs_inode *inode)
3360 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
3361 struct btrfs_root *root = inode->root;
3362 struct btrfs_block_rsv *block_rsv = NULL;
3364 bool insert = false;
3367 if (!root->orphan_block_rsv) {
3368 block_rsv = btrfs_alloc_block_rsv(fs_info,
3369 BTRFS_BLOCK_RSV_TEMP);
3374 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3375 &inode->runtime_flags))
3378 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3379 &inode->runtime_flags))
3382 spin_lock(&root->orphan_lock);
3383 /* If someone has created ->orphan_block_rsv, be happy to use it. */
3384 if (!root->orphan_block_rsv) {
3385 root->orphan_block_rsv = block_rsv;
3386 } else if (block_rsv) {
3387 btrfs_free_block_rsv(fs_info, block_rsv);
3392 atomic_inc(&root->orphan_inodes);
3393 spin_unlock(&root->orphan_lock);
3395 /* grab metadata reservation from transaction handle */
3397 ret = btrfs_orphan_reserve_metadata(trans, inode);
3401 * dec doesn't need spin_lock as ->orphan_block_rsv
3402 * would be released only if ->orphan_inodes is
3405 atomic_dec(&root->orphan_inodes);
3406 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3407 &inode->runtime_flags);
3409 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3410 &inode->runtime_flags);
3415 /* insert an orphan item to track this unlinked/truncated file */
3417 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3420 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3421 &inode->runtime_flags);
3422 btrfs_orphan_release_metadata(inode);
3425 * btrfs_orphan_commit_root may race with us and set
3426 * ->orphan_block_rsv to zero, in order to avoid that,
3427 * decrease ->orphan_inodes after everything is done.
3429 atomic_dec(&root->orphan_inodes);
3430 if (ret != -EEXIST) {
3431 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3432 &inode->runtime_flags);
3433 btrfs_abort_transaction(trans, ret);
3444 * We have done the truncate/delete so we can go ahead and remove the orphan
3445 * item for this particular inode.
3447 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3448 struct btrfs_inode *inode)
3450 struct btrfs_root *root = inode->root;
3451 int delete_item = 0;
3454 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3455 &inode->runtime_flags))
3458 if (delete_item && trans)
3459 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
3461 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3462 &inode->runtime_flags))
3463 btrfs_orphan_release_metadata(inode);
3466 * btrfs_orphan_commit_root may race with us and set ->orphan_block_rsv
3467 * to zero, in order to avoid that, decrease ->orphan_inodes after
3468 * everything is done.
3471 atomic_dec(&root->orphan_inodes);
3477 * this cleans up any orphans that may be left on the list from the last use
3480 int btrfs_orphan_cleanup(struct btrfs_root *root)
3482 struct btrfs_fs_info *fs_info = root->fs_info;
3483 struct btrfs_path *path;
3484 struct extent_buffer *leaf;
3485 struct btrfs_key key, found_key;
3486 struct btrfs_trans_handle *trans;
3487 struct inode *inode;
3488 u64 last_objectid = 0;
3489 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3491 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3494 path = btrfs_alloc_path();
3499 path->reada = READA_BACK;
3501 key.objectid = BTRFS_ORPHAN_OBJECTID;
3502 key.type = BTRFS_ORPHAN_ITEM_KEY;
3503 key.offset = (u64)-1;
3506 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3511 * if ret == 0 means we found what we were searching for, which
3512 * is weird, but possible, so only screw with path if we didn't
3513 * find the key and see if we have stuff that matches
3517 if (path->slots[0] == 0)
3522 /* pull out the item */
3523 leaf = path->nodes[0];
3524 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3526 /* make sure the item matches what we want */
3527 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3529 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3532 /* release the path since we're done with it */
3533 btrfs_release_path(path);
3536 * this is where we are basically btrfs_lookup, without the
3537 * crossing root thing. we store the inode number in the
3538 * offset of the orphan item.
3541 if (found_key.offset == last_objectid) {
3543 "Error removing orphan entry, stopping orphan cleanup");
3548 last_objectid = found_key.offset;
3550 found_key.objectid = found_key.offset;
3551 found_key.type = BTRFS_INODE_ITEM_KEY;
3552 found_key.offset = 0;
3553 inode = btrfs_iget(fs_info->sb, &found_key, root, NULL);
3554 ret = PTR_ERR_OR_ZERO(inode);
3555 if (ret && ret != -ENOENT)
3558 if (ret == -ENOENT && root == fs_info->tree_root) {
3559 struct btrfs_root *dead_root;
3560 struct btrfs_fs_info *fs_info = root->fs_info;
3561 int is_dead_root = 0;
3564 * this is an orphan in the tree root. Currently these
3565 * could come from 2 sources:
3566 * a) a snapshot deletion in progress
3567 * b) a free space cache inode
3568 * We need to distinguish those two, as the snapshot
3569 * orphan must not get deleted.
3570 * find_dead_roots already ran before us, so if this
3571 * is a snapshot deletion, we should find the root
3572 * in the dead_roots list
3574 spin_lock(&fs_info->trans_lock);
3575 list_for_each_entry(dead_root, &fs_info->dead_roots,
3577 if (dead_root->root_key.objectid ==
3578 found_key.objectid) {
3583 spin_unlock(&fs_info->trans_lock);
3585 /* prevent this orphan from being found again */
3586 key.offset = found_key.objectid - 1;
3591 * Inode is already gone but the orphan item is still there,
3592 * kill the orphan item.
3594 if (ret == -ENOENT) {
3595 trans = btrfs_start_transaction(root, 1);
3596 if (IS_ERR(trans)) {
3597 ret = PTR_ERR(trans);
3600 btrfs_debug(fs_info, "auto deleting %Lu",
3601 found_key.objectid);
3602 ret = btrfs_del_orphan_item(trans, root,
3603 found_key.objectid);
3604 btrfs_end_transaction(trans);
3611 * add this inode to the orphan list so btrfs_orphan_del does
3612 * the proper thing when we hit it
3614 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3615 &BTRFS_I(inode)->runtime_flags);
3616 atomic_inc(&root->orphan_inodes);
3618 /* if we have links, this was a truncate, lets do that */
3619 if (inode->i_nlink) {
3620 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3626 /* 1 for the orphan item deletion. */
3627 trans = btrfs_start_transaction(root, 1);
3628 if (IS_ERR(trans)) {
3630 ret = PTR_ERR(trans);
3633 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
3634 btrfs_end_transaction(trans);
3640 ret = btrfs_truncate(inode, false);
3642 btrfs_orphan_del(NULL, BTRFS_I(inode));
3647 /* this will do delete_inode and everything for us */
3652 /* release the path since we're done with it */
3653 btrfs_release_path(path);
3655 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3657 if (root->orphan_block_rsv)
3658 btrfs_block_rsv_release(fs_info, root->orphan_block_rsv,
3661 if (root->orphan_block_rsv ||
3662 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3663 trans = btrfs_join_transaction(root);
3665 btrfs_end_transaction(trans);
3669 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3671 btrfs_debug(fs_info, "truncated %d orphans", nr_truncate);
3675 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3676 btrfs_free_path(path);
3681 * very simple check to peek ahead in the leaf looking for xattrs. If we
3682 * don't find any xattrs, we know there can't be any acls.
3684 * slot is the slot the inode is in, objectid is the objectid of the inode
3686 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3687 int slot, u64 objectid,
3688 int *first_xattr_slot)
3690 u32 nritems = btrfs_header_nritems(leaf);
3691 struct btrfs_key found_key;
3692 static u64 xattr_access = 0;
3693 static u64 xattr_default = 0;
3696 if (!xattr_access) {
3697 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3698 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3699 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3700 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3704 *first_xattr_slot = -1;
3705 while (slot < nritems) {
3706 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3708 /* we found a different objectid, there must not be acls */
3709 if (found_key.objectid != objectid)
3712 /* we found an xattr, assume we've got an acl */
3713 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3714 if (*first_xattr_slot == -1)
3715 *first_xattr_slot = slot;
3716 if (found_key.offset == xattr_access ||
3717 found_key.offset == xattr_default)
3722 * we found a key greater than an xattr key, there can't
3723 * be any acls later on
3725 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3732 * it goes inode, inode backrefs, xattrs, extents,
3733 * so if there are a ton of hard links to an inode there can
3734 * be a lot of backrefs. Don't waste time searching too hard,
3735 * this is just an optimization
3740 /* we hit the end of the leaf before we found an xattr or
3741 * something larger than an xattr. We have to assume the inode
3744 if (*first_xattr_slot == -1)
3745 *first_xattr_slot = slot;
3750 * read an inode from the btree into the in-memory inode
3752 static int btrfs_read_locked_inode(struct inode *inode)
3754 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3755 struct btrfs_path *path;
3756 struct extent_buffer *leaf;
3757 struct btrfs_inode_item *inode_item;
3758 struct btrfs_root *root = BTRFS_I(inode)->root;
3759 struct btrfs_key location;
3764 bool filled = false;
3765 int first_xattr_slot;
3767 ret = btrfs_fill_inode(inode, &rdev);
3771 path = btrfs_alloc_path();
3777 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3779 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3786 leaf = path->nodes[0];
3791 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3792 struct btrfs_inode_item);
3793 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3794 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3795 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3796 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3797 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3799 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3800 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3802 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3803 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3805 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3806 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3808 BTRFS_I(inode)->i_otime.tv_sec =
3809 btrfs_timespec_sec(leaf, &inode_item->otime);
3810 BTRFS_I(inode)->i_otime.tv_nsec =
3811 btrfs_timespec_nsec(leaf, &inode_item->otime);
3813 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3814 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3815 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3817 inode_set_iversion_queried(inode,
3818 btrfs_inode_sequence(leaf, inode_item));
3819 inode->i_generation = BTRFS_I(inode)->generation;
3821 rdev = btrfs_inode_rdev(leaf, inode_item);
3823 BTRFS_I(inode)->index_cnt = (u64)-1;
3824 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3828 * If we were modified in the current generation and evicted from memory
3829 * and then re-read we need to do a full sync since we don't have any
3830 * idea about which extents were modified before we were evicted from
3833 * This is required for both inode re-read from disk and delayed inode
3834 * in delayed_nodes_tree.
3836 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3837 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3838 &BTRFS_I(inode)->runtime_flags);
3841 * We don't persist the id of the transaction where an unlink operation
3842 * against the inode was last made. So here we assume the inode might
3843 * have been evicted, and therefore the exact value of last_unlink_trans
3844 * lost, and set it to last_trans to avoid metadata inconsistencies
3845 * between the inode and its parent if the inode is fsync'ed and the log
3846 * replayed. For example, in the scenario:
3849 * ln mydir/foo mydir/bar
3852 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3853 * xfs_io -c fsync mydir/foo
3855 * mount fs, triggers fsync log replay
3857 * We must make sure that when we fsync our inode foo we also log its
3858 * parent inode, otherwise after log replay the parent still has the
3859 * dentry with the "bar" name but our inode foo has a link count of 1
3860 * and doesn't have an inode ref with the name "bar" anymore.
3862 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3863 * but it guarantees correctness at the expense of occasional full
3864 * transaction commits on fsync if our inode is a directory, or if our
3865 * inode is not a directory, logging its parent unnecessarily.
3867 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3870 if (inode->i_nlink != 1 ||
3871 path->slots[0] >= btrfs_header_nritems(leaf))
3874 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3875 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3878 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3879 if (location.type == BTRFS_INODE_REF_KEY) {
3880 struct btrfs_inode_ref *ref;
3882 ref = (struct btrfs_inode_ref *)ptr;
3883 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3884 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3885 struct btrfs_inode_extref *extref;
3887 extref = (struct btrfs_inode_extref *)ptr;
3888 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3893 * try to precache a NULL acl entry for files that don't have
3894 * any xattrs or acls
3896 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3897 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3898 if (first_xattr_slot != -1) {
3899 path->slots[0] = first_xattr_slot;
3900 ret = btrfs_load_inode_props(inode, path);
3903 "error loading props for ino %llu (root %llu): %d",
3904 btrfs_ino(BTRFS_I(inode)),
3905 root->root_key.objectid, ret);
3907 btrfs_free_path(path);
3910 cache_no_acl(inode);
3912 switch (inode->i_mode & S_IFMT) {
3914 inode->i_mapping->a_ops = &btrfs_aops;
3915 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3916 inode->i_fop = &btrfs_file_operations;
3917 inode->i_op = &btrfs_file_inode_operations;
3920 inode->i_fop = &btrfs_dir_file_operations;
3921 inode->i_op = &btrfs_dir_inode_operations;
3924 inode->i_op = &btrfs_symlink_inode_operations;
3925 inode_nohighmem(inode);
3926 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3929 inode->i_op = &btrfs_special_inode_operations;
3930 init_special_inode(inode, inode->i_mode, rdev);
3934 btrfs_update_iflags(inode);
3938 btrfs_free_path(path);
3939 make_bad_inode(inode);
3944 * given a leaf and an inode, copy the inode fields into the leaf
3946 static void fill_inode_item(struct btrfs_trans_handle *trans,
3947 struct extent_buffer *leaf,
3948 struct btrfs_inode_item *item,
3949 struct inode *inode)
3951 struct btrfs_map_token token;
3953 btrfs_init_map_token(&token);
3955 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3956 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3957 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3959 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3960 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3962 btrfs_set_token_timespec_sec(leaf, &item->atime,
3963 inode->i_atime.tv_sec, &token);
3964 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3965 inode->i_atime.tv_nsec, &token);
3967 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3968 inode->i_mtime.tv_sec, &token);
3969 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3970 inode->i_mtime.tv_nsec, &token);
3972 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3973 inode->i_ctime.tv_sec, &token);
3974 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3975 inode->i_ctime.tv_nsec, &token);
3977 btrfs_set_token_timespec_sec(leaf, &item->otime,
3978 BTRFS_I(inode)->i_otime.tv_sec, &token);
3979 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3980 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3982 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3984 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3986 btrfs_set_token_inode_sequence(leaf, item, inode_peek_iversion(inode),
3988 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3989 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3990 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3991 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3995 * copy everything in the in-memory inode into the btree.
3997 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3998 struct btrfs_root *root, struct inode *inode)
4000 struct btrfs_inode_item *inode_item;
4001 struct btrfs_path *path;
4002 struct extent_buffer *leaf;
4005 path = btrfs_alloc_path();
4009 path->leave_spinning = 1;
4010 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
4018 leaf = path->nodes[0];
4019 inode_item = btrfs_item_ptr(leaf, path->slots[0],
4020 struct btrfs_inode_item);
4022 fill_inode_item(trans, leaf, inode_item, inode);
4023 btrfs_mark_buffer_dirty(leaf);
4024 btrfs_set_inode_last_trans(trans, inode);
4027 btrfs_free_path(path);
4032 * copy everything in the in-memory inode into the btree.
4034 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
4035 struct btrfs_root *root, struct inode *inode)
4037 struct btrfs_fs_info *fs_info = root->fs_info;
4041 * If the inode is a free space inode, we can deadlock during commit
4042 * if we put it into the delayed code.
4044 * The data relocation inode should also be directly updated
4047 if (!btrfs_is_free_space_inode(BTRFS_I(inode))
4048 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
4049 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4050 btrfs_update_root_times(trans, root);
4052 ret = btrfs_delayed_update_inode(trans, root, inode);
4054 btrfs_set_inode_last_trans(trans, inode);
4058 return btrfs_update_inode_item(trans, root, inode);
4061 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4062 struct btrfs_root *root,
4063 struct inode *inode)
4067 ret = btrfs_update_inode(trans, root, inode);
4069 return btrfs_update_inode_item(trans, root, inode);
4074 * unlink helper that gets used here in inode.c and in the tree logging
4075 * recovery code. It remove a link in a directory with a given name, and
4076 * also drops the back refs in the inode to the directory
4078 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4079 struct btrfs_root *root,
4080 struct btrfs_inode *dir,
4081 struct btrfs_inode *inode,
4082 const char *name, int name_len)
4084 struct btrfs_fs_info *fs_info = root->fs_info;
4085 struct btrfs_path *path;
4087 struct extent_buffer *leaf;
4088 struct btrfs_dir_item *di;
4089 struct btrfs_key key;
4091 u64 ino = btrfs_ino(inode);
4092 u64 dir_ino = btrfs_ino(dir);
4094 path = btrfs_alloc_path();
4100 path->leave_spinning = 1;
4101 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4102 name, name_len, -1);
4111 leaf = path->nodes[0];
4112 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4113 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4116 btrfs_release_path(path);
4119 * If we don't have dir index, we have to get it by looking up
4120 * the inode ref, since we get the inode ref, remove it directly,
4121 * it is unnecessary to do delayed deletion.
4123 * But if we have dir index, needn't search inode ref to get it.
4124 * Since the inode ref is close to the inode item, it is better
4125 * that we delay to delete it, and just do this deletion when
4126 * we update the inode item.
4128 if (inode->dir_index) {
4129 ret = btrfs_delayed_delete_inode_ref(inode);
4131 index = inode->dir_index;
4136 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4140 "failed to delete reference to %.*s, inode %llu parent %llu",
4141 name_len, name, ino, dir_ino);
4142 btrfs_abort_transaction(trans, ret);
4146 ret = btrfs_delete_delayed_dir_index(trans, fs_info, dir, index);
4148 btrfs_abort_transaction(trans, ret);
4152 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
4154 if (ret != 0 && ret != -ENOENT) {
4155 btrfs_abort_transaction(trans, ret);
4159 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
4164 btrfs_abort_transaction(trans, ret);
4166 btrfs_free_path(path);
4170 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
4171 inode_inc_iversion(&inode->vfs_inode);
4172 inode_inc_iversion(&dir->vfs_inode);
4173 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
4174 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4175 ret = btrfs_update_inode(trans, root, &dir->vfs_inode);
4180 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4181 struct btrfs_root *root,
4182 struct btrfs_inode *dir, struct btrfs_inode *inode,
4183 const char *name, int name_len)
4186 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4188 drop_nlink(&inode->vfs_inode);
4189 ret = btrfs_update_inode(trans, root, &inode->vfs_inode);
4195 * helper to start transaction for unlink and rmdir.
4197 * unlink and rmdir are special in btrfs, they do not always free space, so
4198 * if we cannot make our reservations the normal way try and see if there is
4199 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4200 * allow the unlink to occur.
4202 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4204 struct btrfs_root *root = BTRFS_I(dir)->root;
4207 * 1 for the possible orphan item
4208 * 1 for the dir item
4209 * 1 for the dir index
4210 * 1 for the inode ref
4213 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4216 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4218 struct btrfs_root *root = BTRFS_I(dir)->root;
4219 struct btrfs_trans_handle *trans;
4220 struct inode *inode = d_inode(dentry);
4223 trans = __unlink_start_trans(dir);
4225 return PTR_ERR(trans);
4227 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4230 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4231 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4232 dentry->d_name.len);
4236 if (inode->i_nlink == 0) {
4237 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4243 btrfs_end_transaction(trans);
4244 btrfs_btree_balance_dirty(root->fs_info);
4248 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4249 struct btrfs_root *root,
4250 struct inode *dir, u64 objectid,
4251 const char *name, int name_len)
4253 struct btrfs_fs_info *fs_info = root->fs_info;
4254 struct btrfs_path *path;
4255 struct extent_buffer *leaf;
4256 struct btrfs_dir_item *di;
4257 struct btrfs_key key;
4260 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4262 path = btrfs_alloc_path();
4266 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4267 name, name_len, -1);
4268 if (IS_ERR_OR_NULL(di)) {
4276 leaf = path->nodes[0];
4277 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4278 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4279 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4281 btrfs_abort_transaction(trans, ret);
4284 btrfs_release_path(path);
4286 ret = btrfs_del_root_ref(trans, fs_info, objectid,
4287 root->root_key.objectid, dir_ino,
4288 &index, name, name_len);
4290 if (ret != -ENOENT) {
4291 btrfs_abort_transaction(trans, ret);
4294 di = btrfs_search_dir_index_item(root, path, dir_ino,
4296 if (IS_ERR_OR_NULL(di)) {
4301 btrfs_abort_transaction(trans, ret);
4305 leaf = path->nodes[0];
4306 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4307 btrfs_release_path(path);
4310 btrfs_release_path(path);
4312 ret = btrfs_delete_delayed_dir_index(trans, fs_info, BTRFS_I(dir), index);
4314 btrfs_abort_transaction(trans, ret);
4318 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4319 inode_inc_iversion(dir);
4320 dir->i_mtime = dir->i_ctime = current_time(dir);
4321 ret = btrfs_update_inode_fallback(trans, root, dir);
4323 btrfs_abort_transaction(trans, ret);
4325 btrfs_free_path(path);
4330 * Helper to check if the subvolume references other subvolumes or if it's
4333 static noinline int may_destroy_subvol(struct btrfs_root *root)
4335 struct btrfs_fs_info *fs_info = root->fs_info;
4336 struct btrfs_path *path;
4337 struct btrfs_dir_item *di;
4338 struct btrfs_key key;
4342 path = btrfs_alloc_path();
4346 /* Make sure this root isn't set as the default subvol */
4347 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4348 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4349 dir_id, "default", 7, 0);
4350 if (di && !IS_ERR(di)) {
4351 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4352 if (key.objectid == root->root_key.objectid) {
4355 "deleting default subvolume %llu is not allowed",
4359 btrfs_release_path(path);
4362 key.objectid = root->root_key.objectid;
4363 key.type = BTRFS_ROOT_REF_KEY;
4364 key.offset = (u64)-1;
4366 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4372 if (path->slots[0] > 0) {
4374 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4375 if (key.objectid == root->root_key.objectid &&
4376 key.type == BTRFS_ROOT_REF_KEY)
4380 btrfs_free_path(path);
4384 int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
4386 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4387 struct btrfs_root *root = BTRFS_I(dir)->root;
4388 struct inode *inode = d_inode(dentry);
4389 struct btrfs_root *dest = BTRFS_I(inode)->root;
4390 struct btrfs_trans_handle *trans;
4391 struct btrfs_block_rsv block_rsv;
4393 u64 qgroup_reserved;
4398 * Don't allow to delete a subvolume with send in progress. This is
4399 * inside the inode lock so the error handling that has to drop the bit
4400 * again is not run concurrently.
4402 spin_lock(&dest->root_item_lock);
4403 root_flags = btrfs_root_flags(&dest->root_item);
4404 if (dest->send_in_progress == 0) {
4405 btrfs_set_root_flags(&dest->root_item,
4406 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4407 spin_unlock(&dest->root_item_lock);
4409 spin_unlock(&dest->root_item_lock);
4411 "attempt to delete subvolume %llu during send",
4412 dest->root_key.objectid);
4416 down_write(&fs_info->subvol_sem);
4418 err = may_destroy_subvol(dest);
4422 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4424 * One for dir inode,
4425 * two for dir entries,
4426 * two for root ref/backref.
4428 err = btrfs_subvolume_reserve_metadata(root, &block_rsv,
4429 5, &qgroup_reserved, true);
4433 trans = btrfs_start_transaction(root, 0);
4434 if (IS_ERR(trans)) {
4435 err = PTR_ERR(trans);
4438 trans->block_rsv = &block_rsv;
4439 trans->bytes_reserved = block_rsv.size;
4441 btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
4443 ret = btrfs_unlink_subvol(trans, root, dir,
4444 dest->root_key.objectid,
4445 dentry->d_name.name,
4446 dentry->d_name.len);
4449 btrfs_abort_transaction(trans, ret);
4453 btrfs_record_root_in_trans(trans, dest);
4455 memset(&dest->root_item.drop_progress, 0,
4456 sizeof(dest->root_item.drop_progress));
4457 dest->root_item.drop_level = 0;
4458 btrfs_set_root_refs(&dest->root_item, 0);
4460 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4461 ret = btrfs_insert_orphan_item(trans,
4463 dest->root_key.objectid);
4465 btrfs_abort_transaction(trans, ret);
4471 ret = btrfs_uuid_tree_rem(trans, fs_info, dest->root_item.uuid,
4472 BTRFS_UUID_KEY_SUBVOL,
4473 dest->root_key.objectid);
4474 if (ret && ret != -ENOENT) {
4475 btrfs_abort_transaction(trans, ret);
4479 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4480 ret = btrfs_uuid_tree_rem(trans, fs_info,
4481 dest->root_item.received_uuid,
4482 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4483 dest->root_key.objectid);
4484 if (ret && ret != -ENOENT) {
4485 btrfs_abort_transaction(trans, ret);
4492 trans->block_rsv = NULL;
4493 trans->bytes_reserved = 0;
4494 ret = btrfs_end_transaction(trans);
4497 inode->i_flags |= S_DEAD;
4499 btrfs_subvolume_release_metadata(fs_info, &block_rsv);
4501 up_write(&fs_info->subvol_sem);
4503 spin_lock(&dest->root_item_lock);
4504 root_flags = btrfs_root_flags(&dest->root_item);
4505 btrfs_set_root_flags(&dest->root_item,
4506 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4507 spin_unlock(&dest->root_item_lock);
4509 d_invalidate(dentry);
4510 btrfs_invalidate_inodes(dest);
4511 ASSERT(dest->send_in_progress == 0);
4514 if (dest->ino_cache_inode) {
4515 iput(dest->ino_cache_inode);
4516 dest->ino_cache_inode = NULL;
4523 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4525 struct inode *inode = d_inode(dentry);
4527 struct btrfs_root *root = BTRFS_I(dir)->root;
4528 struct btrfs_trans_handle *trans;
4529 u64 last_unlink_trans;
4531 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4533 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4534 return btrfs_delete_subvolume(dir, dentry);
4536 trans = __unlink_start_trans(dir);
4538 return PTR_ERR(trans);
4540 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4541 err = btrfs_unlink_subvol(trans, root, dir,
4542 BTRFS_I(inode)->location.objectid,
4543 dentry->d_name.name,
4544 dentry->d_name.len);
4548 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4552 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4554 /* now the directory is empty */
4555 err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4556 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4557 dentry->d_name.len);
4559 btrfs_i_size_write(BTRFS_I(inode), 0);
4561 * Propagate the last_unlink_trans value of the deleted dir to
4562 * its parent directory. This is to prevent an unrecoverable
4563 * log tree in the case we do something like this:
4565 * 2) create snapshot under dir foo
4566 * 3) delete the snapshot
4569 * 6) fsync foo or some file inside foo
4571 if (last_unlink_trans >= trans->transid)
4572 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4575 btrfs_end_transaction(trans);
4576 btrfs_btree_balance_dirty(root->fs_info);
4581 static int truncate_space_check(struct btrfs_trans_handle *trans,
4582 struct btrfs_root *root,
4585 struct btrfs_fs_info *fs_info = root->fs_info;
4589 * This is only used to apply pressure to the enospc system, we don't
4590 * intend to use this reservation at all.
4592 bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted);
4593 bytes_deleted *= fs_info->nodesize;
4594 ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
4595 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4597 trace_btrfs_space_reservation(fs_info, "transaction",
4600 trans->bytes_reserved += bytes_deleted;
4607 * Return this if we need to call truncate_block for the last bit of the
4610 #define NEED_TRUNCATE_BLOCK 1
4613 * this can truncate away extent items, csum items and directory items.
4614 * It starts at a high offset and removes keys until it can't find
4615 * any higher than new_size
4617 * csum items that cross the new i_size are truncated to the new size
4620 * min_type is the minimum key type to truncate down to. If set to 0, this
4621 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4623 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4624 struct btrfs_root *root,
4625 struct inode *inode,
4626 u64 new_size, u32 min_type)
4628 struct btrfs_fs_info *fs_info = root->fs_info;
4629 struct btrfs_path *path;
4630 struct extent_buffer *leaf;
4631 struct btrfs_file_extent_item *fi;
4632 struct btrfs_key key;
4633 struct btrfs_key found_key;
4634 u64 extent_start = 0;
4635 u64 extent_num_bytes = 0;
4636 u64 extent_offset = 0;
4638 u64 last_size = new_size;
4639 u32 found_type = (u8)-1;
4642 int pending_del_nr = 0;
4643 int pending_del_slot = 0;
4644 int extent_type = -1;
4647 u64 ino = btrfs_ino(BTRFS_I(inode));
4648 u64 bytes_deleted = 0;
4649 bool be_nice = false;
4650 bool should_throttle = false;
4651 bool should_end = false;
4653 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4656 * for non-free space inodes and ref cows, we want to back off from
4659 if (!btrfs_is_free_space_inode(BTRFS_I(inode)) &&
4660 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4663 path = btrfs_alloc_path();
4666 path->reada = READA_BACK;
4669 * We want to drop from the next block forward in case this new size is
4670 * not block aligned since we will be keeping the last block of the
4671 * extent just the way it is.
4673 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4674 root == fs_info->tree_root)
4675 btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size,
4676 fs_info->sectorsize),
4680 * This function is also used to drop the items in the log tree before
4681 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4682 * it is used to drop the loged items. So we shouldn't kill the delayed
4685 if (min_type == 0 && root == BTRFS_I(inode)->root)
4686 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
4689 key.offset = (u64)-1;
4694 * with a 16K leaf size and 128MB extents, you can actually queue
4695 * up a huge file in a single leaf. Most of the time that
4696 * bytes_deleted is > 0, it will be huge by the time we get here
4698 if (be_nice && bytes_deleted > SZ_32M) {
4699 if (btrfs_should_end_transaction(trans)) {
4706 path->leave_spinning = 1;
4707 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4714 /* there are no items in the tree for us to truncate, we're
4717 if (path->slots[0] == 0)
4724 leaf = path->nodes[0];
4725 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4726 found_type = found_key.type;
4728 if (found_key.objectid != ino)
4731 if (found_type < min_type)
4734 item_end = found_key.offset;
4735 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4736 fi = btrfs_item_ptr(leaf, path->slots[0],
4737 struct btrfs_file_extent_item);
4738 extent_type = btrfs_file_extent_type(leaf, fi);
4739 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4741 btrfs_file_extent_num_bytes(leaf, fi);
4743 trace_btrfs_truncate_show_fi_regular(
4744 BTRFS_I(inode), leaf, fi,
4746 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4747 item_end += btrfs_file_extent_inline_len(leaf,
4748 path->slots[0], fi);
4750 trace_btrfs_truncate_show_fi_inline(
4751 BTRFS_I(inode), leaf, fi, path->slots[0],
4756 if (found_type > min_type) {
4759 if (item_end < new_size)
4761 if (found_key.offset >= new_size)
4767 /* FIXME, shrink the extent if the ref count is only 1 */
4768 if (found_type != BTRFS_EXTENT_DATA_KEY)
4771 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4773 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4775 u64 orig_num_bytes =
4776 btrfs_file_extent_num_bytes(leaf, fi);
4777 extent_num_bytes = ALIGN(new_size -
4779 fs_info->sectorsize);
4780 btrfs_set_file_extent_num_bytes(leaf, fi,
4782 num_dec = (orig_num_bytes -
4784 if (test_bit(BTRFS_ROOT_REF_COWS,
4787 inode_sub_bytes(inode, num_dec);
4788 btrfs_mark_buffer_dirty(leaf);
4791 btrfs_file_extent_disk_num_bytes(leaf,
4793 extent_offset = found_key.offset -
4794 btrfs_file_extent_offset(leaf, fi);
4796 /* FIXME blocksize != 4096 */
4797 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4798 if (extent_start != 0) {
4800 if (test_bit(BTRFS_ROOT_REF_COWS,
4802 inode_sub_bytes(inode, num_dec);
4805 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4807 * we can't truncate inline items that have had
4811 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4812 btrfs_file_extent_other_encoding(leaf, fi) == 0 &&
4813 btrfs_file_extent_compression(leaf, fi) == 0) {
4814 u32 size = (u32)(new_size - found_key.offset);
4816 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4817 size = btrfs_file_extent_calc_inline_size(size);
4818 btrfs_truncate_item(root->fs_info, path, size, 1);
4819 } else if (!del_item) {
4821 * We have to bail so the last_size is set to
4822 * just before this extent.
4824 err = NEED_TRUNCATE_BLOCK;
4828 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4829 inode_sub_bytes(inode, item_end + 1 - new_size);
4833 last_size = found_key.offset;
4835 last_size = new_size;
4837 if (!pending_del_nr) {
4838 /* no pending yet, add ourselves */
4839 pending_del_slot = path->slots[0];
4841 } else if (pending_del_nr &&
4842 path->slots[0] + 1 == pending_del_slot) {
4843 /* hop on the pending chunk */
4845 pending_del_slot = path->slots[0];
4852 should_throttle = false;
4855 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4856 root == fs_info->tree_root)) {
4857 btrfs_set_path_blocking(path);
4858 bytes_deleted += extent_num_bytes;
4859 ret = btrfs_free_extent(trans, root, extent_start,
4860 extent_num_bytes, 0,
4861 btrfs_header_owner(leaf),
4862 ino, extent_offset);
4864 if (btrfs_should_throttle_delayed_refs(trans, fs_info))
4865 btrfs_async_run_delayed_refs(fs_info,
4866 trans->delayed_ref_updates * 2,
4869 if (truncate_space_check(trans, root,
4870 extent_num_bytes)) {
4873 if (btrfs_should_throttle_delayed_refs(trans,
4875 should_throttle = true;
4879 if (found_type == BTRFS_INODE_ITEM_KEY)
4882 if (path->slots[0] == 0 ||
4883 path->slots[0] != pending_del_slot ||
4884 should_throttle || should_end) {
4885 if (pending_del_nr) {
4886 ret = btrfs_del_items(trans, root, path,
4890 btrfs_abort_transaction(trans, ret);
4895 btrfs_release_path(path);
4896 if (should_throttle) {
4897 unsigned long updates = trans->delayed_ref_updates;
4899 trans->delayed_ref_updates = 0;
4900 ret = btrfs_run_delayed_refs(trans,
4907 * if we failed to refill our space rsv, bail out
4908 * and let the transaction restart
4920 if (pending_del_nr) {
4921 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4924 btrfs_abort_transaction(trans, ret);
4927 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4928 ASSERT(last_size >= new_size);
4929 if (!err && last_size > new_size)
4930 last_size = new_size;
4931 btrfs_ordered_update_i_size(inode, last_size, NULL);
4934 btrfs_free_path(path);
4936 if (be_nice && bytes_deleted > SZ_32M) {
4937 unsigned long updates = trans->delayed_ref_updates;
4939 trans->delayed_ref_updates = 0;
4940 ret = btrfs_run_delayed_refs(trans, updates * 2);
4949 * btrfs_truncate_block - read, zero a chunk and write a block
4950 * @inode - inode that we're zeroing
4951 * @from - the offset to start zeroing
4952 * @len - the length to zero, 0 to zero the entire range respective to the
4954 * @front - zero up to the offset instead of from the offset on
4956 * This will find the block for the "from" offset and cow the block and zero the
4957 * part we want to zero. This is used with truncate and hole punching.
4959 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4962 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4963 struct address_space *mapping = inode->i_mapping;
4964 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4965 struct btrfs_ordered_extent *ordered;
4966 struct extent_state *cached_state = NULL;
4967 struct extent_changeset *data_reserved = NULL;
4969 u32 blocksize = fs_info->sectorsize;
4970 pgoff_t index = from >> PAGE_SHIFT;
4971 unsigned offset = from & (blocksize - 1);
4973 gfp_t mask = btrfs_alloc_write_mask(mapping);
4978 if (IS_ALIGNED(offset, blocksize) &&
4979 (!len || IS_ALIGNED(len, blocksize)))
4982 block_start = round_down(from, blocksize);
4983 block_end = block_start + blocksize - 1;
4985 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
4986 block_start, blocksize);
4991 page = find_or_create_page(mapping, index, mask);
4993 btrfs_delalloc_release_space(inode, data_reserved,
4994 block_start, blocksize, true);
4995 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize, true);
5000 if (!PageUptodate(page)) {
5001 ret = btrfs_readpage(NULL, page);
5003 if (page->mapping != mapping) {
5008 if (!PageUptodate(page)) {
5013 wait_on_page_writeback(page);
5015 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
5016 set_page_extent_mapped(page);
5018 ordered = btrfs_lookup_ordered_extent(inode, block_start);
5020 unlock_extent_cached(io_tree, block_start, block_end,
5024 btrfs_start_ordered_extent(inode, ordered, 1);
5025 btrfs_put_ordered_extent(ordered);
5029 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
5030 EXTENT_DIRTY | EXTENT_DELALLOC |
5031 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
5032 0, 0, &cached_state);
5034 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
5037 unlock_extent_cached(io_tree, block_start, block_end,
5042 if (offset != blocksize) {
5044 len = blocksize - offset;
5047 memset(kaddr + (block_start - page_offset(page)),
5050 memset(kaddr + (block_start - page_offset(page)) + offset,
5052 flush_dcache_page(page);
5055 ClearPageChecked(page);
5056 set_page_dirty(page);
5057 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
5061 btrfs_delalloc_release_space(inode, data_reserved, block_start,
5063 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize, (ret != 0));
5067 extent_changeset_free(data_reserved);
5071 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
5072 u64 offset, u64 len)
5074 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5075 struct btrfs_trans_handle *trans;
5079 * Still need to make sure the inode looks like it's been updated so
5080 * that any holes get logged if we fsync.
5082 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
5083 BTRFS_I(inode)->last_trans = fs_info->generation;
5084 BTRFS_I(inode)->last_sub_trans = root->log_transid;
5085 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
5090 * 1 - for the one we're dropping
5091 * 1 - for the one we're adding
5092 * 1 - for updating the inode.
5094 trans = btrfs_start_transaction(root, 3);
5096 return PTR_ERR(trans);
5098 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
5100 btrfs_abort_transaction(trans, ret);
5101 btrfs_end_transaction(trans);
5105 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)),
5106 offset, 0, 0, len, 0, len, 0, 0, 0);
5108 btrfs_abort_transaction(trans, ret);
5110 btrfs_update_inode(trans, root, inode);
5111 btrfs_end_transaction(trans);
5116 * This function puts in dummy file extents for the area we're creating a hole
5117 * for. So if we are truncating this file to a larger size we need to insert
5118 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
5119 * the range between oldsize and size
5121 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
5123 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5124 struct btrfs_root *root = BTRFS_I(inode)->root;
5125 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5126 struct extent_map *em = NULL;
5127 struct extent_state *cached_state = NULL;
5128 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5129 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
5130 u64 block_end = ALIGN(size, fs_info->sectorsize);
5137 * If our size started in the middle of a block we need to zero out the
5138 * rest of the block before we expand the i_size, otherwise we could
5139 * expose stale data.
5141 err = btrfs_truncate_block(inode, oldsize, 0, 0);
5145 if (size <= hole_start)
5149 struct btrfs_ordered_extent *ordered;
5151 lock_extent_bits(io_tree, hole_start, block_end - 1,
5153 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), hole_start,
5154 block_end - hole_start);
5157 unlock_extent_cached(io_tree, hole_start, block_end - 1,
5159 btrfs_start_ordered_extent(inode, ordered, 1);
5160 btrfs_put_ordered_extent(ordered);
5163 cur_offset = hole_start;
5165 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
5166 block_end - cur_offset, 0);
5172 last_byte = min(extent_map_end(em), block_end);
5173 last_byte = ALIGN(last_byte, fs_info->sectorsize);
5174 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5175 struct extent_map *hole_em;
5176 hole_size = last_byte - cur_offset;
5178 err = maybe_insert_hole(root, inode, cur_offset,
5182 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
5183 cur_offset + hole_size - 1, 0);
5184 hole_em = alloc_extent_map();
5186 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5187 &BTRFS_I(inode)->runtime_flags);
5190 hole_em->start = cur_offset;
5191 hole_em->len = hole_size;
5192 hole_em->orig_start = cur_offset;
5194 hole_em->block_start = EXTENT_MAP_HOLE;
5195 hole_em->block_len = 0;
5196 hole_em->orig_block_len = 0;
5197 hole_em->ram_bytes = hole_size;
5198 hole_em->bdev = fs_info->fs_devices->latest_bdev;
5199 hole_em->compress_type = BTRFS_COMPRESS_NONE;
5200 hole_em->generation = fs_info->generation;
5203 write_lock(&em_tree->lock);
5204 err = add_extent_mapping(em_tree, hole_em, 1);
5205 write_unlock(&em_tree->lock);
5208 btrfs_drop_extent_cache(BTRFS_I(inode),
5213 free_extent_map(hole_em);
5216 free_extent_map(em);
5218 cur_offset = last_byte;
5219 if (cur_offset >= block_end)
5222 free_extent_map(em);
5223 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
5227 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5229 struct btrfs_root *root = BTRFS_I(inode)->root;
5230 struct btrfs_trans_handle *trans;
5231 loff_t oldsize = i_size_read(inode);
5232 loff_t newsize = attr->ia_size;
5233 int mask = attr->ia_valid;
5237 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5238 * special case where we need to update the times despite not having
5239 * these flags set. For all other operations the VFS set these flags
5240 * explicitly if it wants a timestamp update.
5242 if (newsize != oldsize) {
5243 inode_inc_iversion(inode);
5244 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5245 inode->i_ctime = inode->i_mtime =
5246 current_time(inode);
5249 if (newsize > oldsize) {
5251 * Don't do an expanding truncate while snapshotting is ongoing.
5252 * This is to ensure the snapshot captures a fully consistent
5253 * state of this file - if the snapshot captures this expanding
5254 * truncation, it must capture all writes that happened before
5257 btrfs_wait_for_snapshot_creation(root);
5258 ret = btrfs_cont_expand(inode, oldsize, newsize);
5260 btrfs_end_write_no_snapshotting(root);
5264 trans = btrfs_start_transaction(root, 1);
5265 if (IS_ERR(trans)) {
5266 btrfs_end_write_no_snapshotting(root);
5267 return PTR_ERR(trans);
5270 i_size_write(inode, newsize);
5271 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
5272 pagecache_isize_extended(inode, oldsize, newsize);
5273 ret = btrfs_update_inode(trans, root, inode);
5274 btrfs_end_write_no_snapshotting(root);
5275 btrfs_end_transaction(trans);
5279 * We're truncating a file that used to have good data down to
5280 * zero. Make sure it gets into the ordered flush list so that
5281 * any new writes get down to disk quickly.
5284 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
5285 &BTRFS_I(inode)->runtime_flags);
5288 * 1 for the orphan item we're going to add
5289 * 1 for the orphan item deletion.
5291 trans = btrfs_start_transaction(root, 2);
5293 return PTR_ERR(trans);
5296 * We need to do this in case we fail at _any_ point during the
5297 * actual truncate. Once we do the truncate_setsize we could
5298 * invalidate pages which forces any outstanding ordered io to
5299 * be instantly completed which will give us extents that need
5300 * to be truncated. If we fail to get an orphan inode down we
5301 * could have left over extents that were never meant to live,
5302 * so we need to guarantee from this point on that everything
5303 * will be consistent.
5305 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
5306 btrfs_end_transaction(trans);
5310 /* we don't support swapfiles, so vmtruncate shouldn't fail */
5311 truncate_setsize(inode, newsize);
5313 /* Disable nonlocked read DIO to avoid the end less truncate */
5314 btrfs_inode_block_unlocked_dio(BTRFS_I(inode));
5315 inode_dio_wait(inode);
5316 btrfs_inode_resume_unlocked_dio(BTRFS_I(inode));
5318 ret = btrfs_truncate(inode, newsize == oldsize);
5319 if (ret && inode->i_nlink) {
5322 /* To get a stable disk_i_size */
5323 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5325 btrfs_orphan_del(NULL, BTRFS_I(inode));
5330 * failed to truncate, disk_i_size is only adjusted down
5331 * as we remove extents, so it should represent the true
5332 * size of the inode, so reset the in memory size and
5333 * delete our orphan entry.
5335 trans = btrfs_join_transaction(root);
5336 if (IS_ERR(trans)) {
5337 btrfs_orphan_del(NULL, BTRFS_I(inode));
5340 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5341 err = btrfs_orphan_del(trans, BTRFS_I(inode));
5343 btrfs_abort_transaction(trans, err);
5344 btrfs_end_transaction(trans);
5351 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5353 struct inode *inode = d_inode(dentry);
5354 struct btrfs_root *root = BTRFS_I(inode)->root;
5357 if (btrfs_root_readonly(root))
5360 err = setattr_prepare(dentry, attr);
5364 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5365 err = btrfs_setsize(inode, attr);
5370 if (attr->ia_valid) {
5371 setattr_copy(inode, attr);
5372 inode_inc_iversion(inode);
5373 err = btrfs_dirty_inode(inode);
5375 if (!err && attr->ia_valid & ATTR_MODE)
5376 err = posix_acl_chmod(inode, inode->i_mode);
5383 * While truncating the inode pages during eviction, we get the VFS calling
5384 * btrfs_invalidatepage() against each page of the inode. This is slow because
5385 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5386 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5387 * extent_state structures over and over, wasting lots of time.
5389 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5390 * those expensive operations on a per page basis and do only the ordered io
5391 * finishing, while we release here the extent_map and extent_state structures,
5392 * without the excessive merging and splitting.
5394 static void evict_inode_truncate_pages(struct inode *inode)
5396 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5397 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5398 struct rb_node *node;
5400 ASSERT(inode->i_state & I_FREEING);
5401 truncate_inode_pages_final(&inode->i_data);
5403 write_lock(&map_tree->lock);
5404 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5405 struct extent_map *em;
5407 node = rb_first(&map_tree->map);
5408 em = rb_entry(node, struct extent_map, rb_node);
5409 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5410 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5411 remove_extent_mapping(map_tree, em);
5412 free_extent_map(em);
5413 if (need_resched()) {
5414 write_unlock(&map_tree->lock);
5416 write_lock(&map_tree->lock);
5419 write_unlock(&map_tree->lock);
5422 * Keep looping until we have no more ranges in the io tree.
5423 * We can have ongoing bios started by readpages (called from readahead)
5424 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5425 * still in progress (unlocked the pages in the bio but did not yet
5426 * unlocked the ranges in the io tree). Therefore this means some
5427 * ranges can still be locked and eviction started because before
5428 * submitting those bios, which are executed by a separate task (work
5429 * queue kthread), inode references (inode->i_count) were not taken
5430 * (which would be dropped in the end io callback of each bio).
5431 * Therefore here we effectively end up waiting for those bios and
5432 * anyone else holding locked ranges without having bumped the inode's
5433 * reference count - if we don't do it, when they access the inode's
5434 * io_tree to unlock a range it may be too late, leading to an
5435 * use-after-free issue.
5437 spin_lock(&io_tree->lock);
5438 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5439 struct extent_state *state;
5440 struct extent_state *cached_state = NULL;
5444 node = rb_first(&io_tree->state);
5445 state = rb_entry(node, struct extent_state, rb_node);
5446 start = state->start;
5448 spin_unlock(&io_tree->lock);
5450 lock_extent_bits(io_tree, start, end, &cached_state);
5453 * If still has DELALLOC flag, the extent didn't reach disk,
5454 * and its reserved space won't be freed by delayed_ref.
5455 * So we need to free its reserved space here.
5456 * (Refer to comment in btrfs_invalidatepage, case 2)
5458 * Note, end is the bytenr of last byte, so we need + 1 here.
5460 if (state->state & EXTENT_DELALLOC)
5461 btrfs_qgroup_free_data(inode, NULL, start, end - start + 1);
5463 clear_extent_bit(io_tree, start, end,
5464 EXTENT_LOCKED | EXTENT_DIRTY |
5465 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5466 EXTENT_DEFRAG, 1, 1, &cached_state);
5469 spin_lock(&io_tree->lock);
5471 spin_unlock(&io_tree->lock);
5474 void btrfs_evict_inode(struct inode *inode)
5476 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5477 struct btrfs_trans_handle *trans;
5478 struct btrfs_root *root = BTRFS_I(inode)->root;
5479 struct btrfs_block_rsv *rsv, *global_rsv;
5480 int steal_from_global = 0;
5484 trace_btrfs_inode_evict(inode);
5491 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
5493 evict_inode_truncate_pages(inode);
5495 if (inode->i_nlink &&
5496 ((btrfs_root_refs(&root->root_item) != 0 &&
5497 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5498 btrfs_is_free_space_inode(BTRFS_I(inode))))
5501 if (is_bad_inode(inode)) {
5502 btrfs_orphan_del(NULL, BTRFS_I(inode));
5505 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5506 if (!special_file(inode->i_mode))
5507 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5509 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5511 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
5512 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5513 &BTRFS_I(inode)->runtime_flags));
5517 if (inode->i_nlink > 0) {
5518 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5519 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5523 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5525 btrfs_orphan_del(NULL, BTRFS_I(inode));
5529 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5531 btrfs_orphan_del(NULL, BTRFS_I(inode));
5534 rsv->size = min_size;
5536 global_rsv = &fs_info->global_block_rsv;
5538 btrfs_i_size_write(BTRFS_I(inode), 0);
5541 * This is a bit simpler than btrfs_truncate since we've already
5542 * reserved our space for our orphan item in the unlink, so we just
5543 * need to reserve some slack space in case we add bytes and update
5544 * inode item when doing the truncate.
5547 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5548 BTRFS_RESERVE_FLUSH_LIMIT);
5551 * Try and steal from the global reserve since we will
5552 * likely not use this space anyway, we want to try as
5553 * hard as possible to get this to work.
5556 steal_from_global++;
5558 steal_from_global = 0;
5562 * steal_from_global == 0: we reserved stuff, hooray!
5563 * steal_from_global == 1: we didn't reserve stuff, boo!
5564 * steal_from_global == 2: we've committed, still not a lot of
5565 * room but maybe we'll have room in the global reserve this
5567 * steal_from_global == 3: abandon all hope!
5569 if (steal_from_global > 2) {
5571 "Could not get space for a delete, will truncate on mount %d",
5573 btrfs_orphan_del(NULL, BTRFS_I(inode));
5574 btrfs_free_block_rsv(fs_info, rsv);
5578 trans = btrfs_join_transaction(root);
5579 if (IS_ERR(trans)) {
5580 btrfs_orphan_del(NULL, BTRFS_I(inode));
5581 btrfs_free_block_rsv(fs_info, rsv);
5586 * We can't just steal from the global reserve, we need to make
5587 * sure there is room to do it, if not we need to commit and try
5590 if (steal_from_global) {
5591 if (!btrfs_check_space_for_delayed_refs(trans, fs_info))
5592 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5599 * Couldn't steal from the global reserve, we have too much
5600 * pending stuff built up, commit the transaction and try it
5604 ret = btrfs_commit_transaction(trans);
5606 btrfs_orphan_del(NULL, BTRFS_I(inode));
5607 btrfs_free_block_rsv(fs_info, rsv);
5612 steal_from_global = 0;
5615 trans->block_rsv = rsv;
5617 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5618 if (ret != -ENOSPC && ret != -EAGAIN)
5621 trans->block_rsv = &fs_info->trans_block_rsv;
5622 btrfs_end_transaction(trans);
5624 btrfs_btree_balance_dirty(fs_info);
5627 btrfs_free_block_rsv(fs_info, rsv);
5630 * Errors here aren't a big deal, it just means we leave orphan items
5631 * in the tree. They will be cleaned up on the next mount.
5634 trans->block_rsv = root->orphan_block_rsv;
5635 btrfs_orphan_del(trans, BTRFS_I(inode));
5637 btrfs_orphan_del(NULL, BTRFS_I(inode));
5640 trans->block_rsv = &fs_info->trans_block_rsv;
5641 if (!(root == fs_info->tree_root ||
5642 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5643 btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
5645 btrfs_end_transaction(trans);
5646 btrfs_btree_balance_dirty(fs_info);
5648 btrfs_remove_delayed_node(BTRFS_I(inode));
5653 * this returns the key found in the dir entry in the location pointer.
5654 * If no dir entries were found, returns -ENOENT.
5655 * If found a corrupted location in dir entry, returns -EUCLEAN.
5657 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5658 struct btrfs_key *location)
5660 const char *name = dentry->d_name.name;
5661 int namelen = dentry->d_name.len;
5662 struct btrfs_dir_item *di;
5663 struct btrfs_path *path;
5664 struct btrfs_root *root = BTRFS_I(dir)->root;
5667 path = btrfs_alloc_path();
5671 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5682 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5683 if (location->type != BTRFS_INODE_ITEM_KEY &&
5684 location->type != BTRFS_ROOT_ITEM_KEY) {
5686 btrfs_warn(root->fs_info,
5687 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5688 __func__, name, btrfs_ino(BTRFS_I(dir)),
5689 location->objectid, location->type, location->offset);
5692 btrfs_free_path(path);
5697 * when we hit a tree root in a directory, the btrfs part of the inode
5698 * needs to be changed to reflect the root directory of the tree root. This
5699 * is kind of like crossing a mount point.
5701 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5703 struct dentry *dentry,
5704 struct btrfs_key *location,
5705 struct btrfs_root **sub_root)
5707 struct btrfs_path *path;
5708 struct btrfs_root *new_root;
5709 struct btrfs_root_ref *ref;
5710 struct extent_buffer *leaf;
5711 struct btrfs_key key;
5715 path = btrfs_alloc_path();
5722 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5723 key.type = BTRFS_ROOT_REF_KEY;
5724 key.offset = location->objectid;
5726 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5733 leaf = path->nodes[0];
5734 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5735 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5736 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5739 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5740 (unsigned long)(ref + 1),
5741 dentry->d_name.len);
5745 btrfs_release_path(path);
5747 new_root = btrfs_read_fs_root_no_name(fs_info, location);
5748 if (IS_ERR(new_root)) {
5749 err = PTR_ERR(new_root);
5753 *sub_root = new_root;
5754 location->objectid = btrfs_root_dirid(&new_root->root_item);
5755 location->type = BTRFS_INODE_ITEM_KEY;
5756 location->offset = 0;
5759 btrfs_free_path(path);
5763 static void inode_tree_add(struct inode *inode)
5765 struct btrfs_root *root = BTRFS_I(inode)->root;
5766 struct btrfs_inode *entry;
5768 struct rb_node *parent;
5769 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5770 u64 ino = btrfs_ino(BTRFS_I(inode));
5772 if (inode_unhashed(inode))
5775 spin_lock(&root->inode_lock);
5776 p = &root->inode_tree.rb_node;
5779 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5781 if (ino < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5782 p = &parent->rb_left;
5783 else if (ino > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5784 p = &parent->rb_right;
5786 WARN_ON(!(entry->vfs_inode.i_state &
5787 (I_WILL_FREE | I_FREEING)));
5788 rb_replace_node(parent, new, &root->inode_tree);
5789 RB_CLEAR_NODE(parent);
5790 spin_unlock(&root->inode_lock);
5794 rb_link_node(new, parent, p);
5795 rb_insert_color(new, &root->inode_tree);
5796 spin_unlock(&root->inode_lock);
5799 static void inode_tree_del(struct inode *inode)
5801 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5802 struct btrfs_root *root = BTRFS_I(inode)->root;
5805 spin_lock(&root->inode_lock);
5806 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5807 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5808 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5809 empty = RB_EMPTY_ROOT(&root->inode_tree);
5811 spin_unlock(&root->inode_lock);
5813 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5814 synchronize_srcu(&fs_info->subvol_srcu);
5815 spin_lock(&root->inode_lock);
5816 empty = RB_EMPTY_ROOT(&root->inode_tree);
5817 spin_unlock(&root->inode_lock);
5819 btrfs_add_dead_root(root);
5823 void btrfs_invalidate_inodes(struct btrfs_root *root)
5825 struct btrfs_fs_info *fs_info = root->fs_info;
5826 struct rb_node *node;
5827 struct rb_node *prev;
5828 struct btrfs_inode *entry;
5829 struct inode *inode;
5832 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
5833 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5835 spin_lock(&root->inode_lock);
5837 node = root->inode_tree.rb_node;
5841 entry = rb_entry(node, struct btrfs_inode, rb_node);
5843 if (objectid < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5844 node = node->rb_left;
5845 else if (objectid > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5846 node = node->rb_right;
5852 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5853 if (objectid <= btrfs_ino(BTRFS_I(&entry->vfs_inode))) {
5857 prev = rb_next(prev);
5861 entry = rb_entry(node, struct btrfs_inode, rb_node);
5862 objectid = btrfs_ino(BTRFS_I(&entry->vfs_inode)) + 1;
5863 inode = igrab(&entry->vfs_inode);
5865 spin_unlock(&root->inode_lock);
5866 if (atomic_read(&inode->i_count) > 1)
5867 d_prune_aliases(inode);
5869 * btrfs_drop_inode will have it removed from
5870 * the inode cache when its usage count
5875 spin_lock(&root->inode_lock);
5879 if (cond_resched_lock(&root->inode_lock))
5882 node = rb_next(node);
5884 spin_unlock(&root->inode_lock);
5887 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5889 struct btrfs_iget_args *args = p;
5890 inode->i_ino = args->location->objectid;
5891 memcpy(&BTRFS_I(inode)->location, args->location,
5892 sizeof(*args->location));
5893 BTRFS_I(inode)->root = args->root;
5897 static int btrfs_find_actor(struct inode *inode, void *opaque)
5899 struct btrfs_iget_args *args = opaque;
5900 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5901 args->root == BTRFS_I(inode)->root;
5904 static struct inode *btrfs_iget_locked(struct super_block *s,
5905 struct btrfs_key *location,
5906 struct btrfs_root *root)
5908 struct inode *inode;
5909 struct btrfs_iget_args args;
5910 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5912 args.location = location;
5915 inode = iget5_locked(s, hashval, btrfs_find_actor,
5916 btrfs_init_locked_inode,
5921 /* Get an inode object given its location and corresponding root.
5922 * Returns in *is_new if the inode was read from disk
5924 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5925 struct btrfs_root *root, int *new)
5927 struct inode *inode;
5929 inode = btrfs_iget_locked(s, location, root);
5931 return ERR_PTR(-ENOMEM);
5933 if (inode->i_state & I_NEW) {
5936 ret = btrfs_read_locked_inode(inode);
5937 if (!is_bad_inode(inode)) {
5938 inode_tree_add(inode);
5939 unlock_new_inode(inode);
5943 unlock_new_inode(inode);
5946 inode = ERR_PTR(ret < 0 ? ret : -ESTALE);
5953 static struct inode *new_simple_dir(struct super_block *s,
5954 struct btrfs_key *key,
5955 struct btrfs_root *root)
5957 struct inode *inode = new_inode(s);
5960 return ERR_PTR(-ENOMEM);
5962 BTRFS_I(inode)->root = root;
5963 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5964 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5966 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5967 inode->i_op = &btrfs_dir_ro_inode_operations;
5968 inode->i_opflags &= ~IOP_XATTR;
5969 inode->i_fop = &simple_dir_operations;
5970 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5971 inode->i_mtime = current_time(inode);
5972 inode->i_atime = inode->i_mtime;
5973 inode->i_ctime = inode->i_mtime;
5974 BTRFS_I(inode)->i_otime = inode->i_mtime;
5979 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5981 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5982 struct inode *inode;
5983 struct btrfs_root *root = BTRFS_I(dir)->root;
5984 struct btrfs_root *sub_root = root;
5985 struct btrfs_key location;
5989 if (dentry->d_name.len > BTRFS_NAME_LEN)
5990 return ERR_PTR(-ENAMETOOLONG);
5992 ret = btrfs_inode_by_name(dir, dentry, &location);
5994 return ERR_PTR(ret);
5996 if (location.type == BTRFS_INODE_ITEM_KEY) {
5997 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
6001 index = srcu_read_lock(&fs_info->subvol_srcu);
6002 ret = fixup_tree_root_location(fs_info, dir, dentry,
6003 &location, &sub_root);
6006 inode = ERR_PTR(ret);
6008 inode = new_simple_dir(dir->i_sb, &location, sub_root);
6010 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
6012 srcu_read_unlock(&fs_info->subvol_srcu, index);
6014 if (!IS_ERR(inode) && root != sub_root) {
6015 down_read(&fs_info->cleanup_work_sem);
6016 if (!sb_rdonly(inode->i_sb))
6017 ret = btrfs_orphan_cleanup(sub_root);
6018 up_read(&fs_info->cleanup_work_sem);
6021 inode = ERR_PTR(ret);
6028 static int btrfs_dentry_delete(const struct dentry *dentry)
6030 struct btrfs_root *root;
6031 struct inode *inode = d_inode(dentry);
6033 if (!inode && !IS_ROOT(dentry))
6034 inode = d_inode(dentry->d_parent);
6037 root = BTRFS_I(inode)->root;
6038 if (btrfs_root_refs(&root->root_item) == 0)
6041 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6047 static void btrfs_dentry_release(struct dentry *dentry)
6049 kfree(dentry->d_fsdata);
6052 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
6055 struct inode *inode;
6057 inode = btrfs_lookup_dentry(dir, dentry);
6058 if (IS_ERR(inode)) {
6059 if (PTR_ERR(inode) == -ENOENT)
6062 return ERR_CAST(inode);
6065 return d_splice_alias(inode, dentry);
6068 unsigned char btrfs_filetype_table[] = {
6069 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
6073 * All this infrastructure exists because dir_emit can fault, and we are holding
6074 * the tree lock when doing readdir. For now just allocate a buffer and copy
6075 * our information into that, and then dir_emit from the buffer. This is
6076 * similar to what NFS does, only we don't keep the buffer around in pagecache
6077 * because I'm afraid I'll mess that up. Long term we need to make filldir do
6078 * copy_to_user_inatomic so we don't have to worry about page faulting under the
6081 static int btrfs_opendir(struct inode *inode, struct file *file)
6083 struct btrfs_file_private *private;
6085 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
6088 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
6089 if (!private->filldir_buf) {
6093 file->private_data = private;
6104 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
6107 struct dir_entry *entry = addr;
6108 char *name = (char *)(entry + 1);
6110 ctx->pos = get_unaligned(&entry->offset);
6111 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
6112 get_unaligned(&entry->ino),
6113 get_unaligned(&entry->type)))
6115 addr += sizeof(struct dir_entry) +
6116 get_unaligned(&entry->name_len);
6122 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
6124 struct inode *inode = file_inode(file);
6125 struct btrfs_root *root = BTRFS_I(inode)->root;
6126 struct btrfs_file_private *private = file->private_data;
6127 struct btrfs_dir_item *di;
6128 struct btrfs_key key;
6129 struct btrfs_key found_key;
6130 struct btrfs_path *path;
6132 struct list_head ins_list;
6133 struct list_head del_list;
6135 struct extent_buffer *leaf;
6142 struct btrfs_key location;
6144 if (!dir_emit_dots(file, ctx))
6147 path = btrfs_alloc_path();
6151 addr = private->filldir_buf;
6152 path->reada = READA_FORWARD;
6154 INIT_LIST_HEAD(&ins_list);
6155 INIT_LIST_HEAD(&del_list);
6156 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
6159 key.type = BTRFS_DIR_INDEX_KEY;
6160 key.offset = ctx->pos;
6161 key.objectid = btrfs_ino(BTRFS_I(inode));
6163 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6168 struct dir_entry *entry;
6170 leaf = path->nodes[0];
6171 slot = path->slots[0];
6172 if (slot >= btrfs_header_nritems(leaf)) {
6173 ret = btrfs_next_leaf(root, path);
6181 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6183 if (found_key.objectid != key.objectid)
6185 if (found_key.type != BTRFS_DIR_INDEX_KEY)
6187 if (found_key.offset < ctx->pos)
6189 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
6191 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
6192 name_len = btrfs_dir_name_len(leaf, di);
6193 if ((total_len + sizeof(struct dir_entry) + name_len) >=
6195 btrfs_release_path(path);
6196 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6199 addr = private->filldir_buf;
6206 put_unaligned(name_len, &entry->name_len);
6207 name_ptr = (char *)(entry + 1);
6208 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
6210 put_unaligned(btrfs_filetype_table[btrfs_dir_type(leaf, di)],
6212 btrfs_dir_item_key_to_cpu(leaf, di, &location);
6213 put_unaligned(location.objectid, &entry->ino);
6214 put_unaligned(found_key.offset, &entry->offset);
6216 addr += sizeof(struct dir_entry) + name_len;
6217 total_len += sizeof(struct dir_entry) + name_len;
6221 btrfs_release_path(path);
6223 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6227 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
6232 * Stop new entries from being returned after we return the last
6235 * New directory entries are assigned a strictly increasing
6236 * offset. This means that new entries created during readdir
6237 * are *guaranteed* to be seen in the future by that readdir.
6238 * This has broken buggy programs which operate on names as
6239 * they're returned by readdir. Until we re-use freed offsets
6240 * we have this hack to stop new entries from being returned
6241 * under the assumption that they'll never reach this huge
6244 * This is being careful not to overflow 32bit loff_t unless the
6245 * last entry requires it because doing so has broken 32bit apps
6248 if (ctx->pos >= INT_MAX)
6249 ctx->pos = LLONG_MAX;
6256 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6257 btrfs_free_path(path);
6261 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
6263 struct btrfs_root *root = BTRFS_I(inode)->root;
6264 struct btrfs_trans_handle *trans;
6266 bool nolock = false;
6268 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6271 if (btrfs_fs_closing(root->fs_info) &&
6272 btrfs_is_free_space_inode(BTRFS_I(inode)))
6275 if (wbc->sync_mode == WB_SYNC_ALL) {
6277 trans = btrfs_join_transaction_nolock(root);
6279 trans = btrfs_join_transaction(root);
6281 return PTR_ERR(trans);
6282 ret = btrfs_commit_transaction(trans);
6288 * This is somewhat expensive, updating the tree every time the
6289 * inode changes. But, it is most likely to find the inode in cache.
6290 * FIXME, needs more benchmarking...there are no reasons other than performance
6291 * to keep or drop this code.
6293 static int btrfs_dirty_inode(struct inode *inode)
6295 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6296 struct btrfs_root *root = BTRFS_I(inode)->root;
6297 struct btrfs_trans_handle *trans;
6300 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6303 trans = btrfs_join_transaction(root);
6305 return PTR_ERR(trans);
6307 ret = btrfs_update_inode(trans, root, inode);
6308 if (ret && ret == -ENOSPC) {
6309 /* whoops, lets try again with the full transaction */
6310 btrfs_end_transaction(trans);
6311 trans = btrfs_start_transaction(root, 1);
6313 return PTR_ERR(trans);
6315 ret = btrfs_update_inode(trans, root, inode);
6317 btrfs_end_transaction(trans);
6318 if (BTRFS_I(inode)->delayed_node)
6319 btrfs_balance_delayed_items(fs_info);
6325 * This is a copy of file_update_time. We need this so we can return error on
6326 * ENOSPC for updating the inode in the case of file write and mmap writes.
6328 static int btrfs_update_time(struct inode *inode, struct timespec *now,
6331 struct btrfs_root *root = BTRFS_I(inode)->root;
6332 bool dirty = flags & ~S_VERSION;
6334 if (btrfs_root_readonly(root))
6337 if (flags & S_VERSION)
6338 dirty |= inode_maybe_inc_iversion(inode, dirty);
6339 if (flags & S_CTIME)
6340 inode->i_ctime = *now;
6341 if (flags & S_MTIME)
6342 inode->i_mtime = *now;
6343 if (flags & S_ATIME)
6344 inode->i_atime = *now;
6345 return dirty ? btrfs_dirty_inode(inode) : 0;
6349 * find the highest existing sequence number in a directory
6350 * and then set the in-memory index_cnt variable to reflect
6351 * free sequence numbers
6353 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6355 struct btrfs_root *root = inode->root;
6356 struct btrfs_key key, found_key;
6357 struct btrfs_path *path;
6358 struct extent_buffer *leaf;
6361 key.objectid = btrfs_ino(inode);
6362 key.type = BTRFS_DIR_INDEX_KEY;
6363 key.offset = (u64)-1;
6365 path = btrfs_alloc_path();
6369 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6372 /* FIXME: we should be able to handle this */
6378 * MAGIC NUMBER EXPLANATION:
6379 * since we search a directory based on f_pos we have to start at 2
6380 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6381 * else has to start at 2
6383 if (path->slots[0] == 0) {
6384 inode->index_cnt = 2;
6390 leaf = path->nodes[0];
6391 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6393 if (found_key.objectid != btrfs_ino(inode) ||
6394 found_key.type != BTRFS_DIR_INDEX_KEY) {
6395 inode->index_cnt = 2;
6399 inode->index_cnt = found_key.offset + 1;
6401 btrfs_free_path(path);
6406 * helper to find a free sequence number in a given directory. This current
6407 * code is very simple, later versions will do smarter things in the btree
6409 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6413 if (dir->index_cnt == (u64)-1) {
6414 ret = btrfs_inode_delayed_dir_index_count(dir);
6416 ret = btrfs_set_inode_index_count(dir);
6422 *index = dir->index_cnt;
6428 static int btrfs_insert_inode_locked(struct inode *inode)
6430 struct btrfs_iget_args args;
6431 args.location = &BTRFS_I(inode)->location;
6432 args.root = BTRFS_I(inode)->root;
6434 return insert_inode_locked4(inode,
6435 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6436 btrfs_find_actor, &args);
6440 * Inherit flags from the parent inode.
6442 * Currently only the compression flags and the cow flags are inherited.
6444 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6451 flags = BTRFS_I(dir)->flags;
6453 if (flags & BTRFS_INODE_NOCOMPRESS) {
6454 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6455 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6456 } else if (flags & BTRFS_INODE_COMPRESS) {
6457 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6458 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6461 if (flags & BTRFS_INODE_NODATACOW) {
6462 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6463 if (S_ISREG(inode->i_mode))
6464 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6467 btrfs_update_iflags(inode);
6470 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6471 struct btrfs_root *root,
6473 const char *name, int name_len,
6474 u64 ref_objectid, u64 objectid,
6475 umode_t mode, u64 *index)
6477 struct btrfs_fs_info *fs_info = root->fs_info;
6478 struct inode *inode;
6479 struct btrfs_inode_item *inode_item;
6480 struct btrfs_key *location;
6481 struct btrfs_path *path;
6482 struct btrfs_inode_ref *ref;
6483 struct btrfs_key key[2];
6485 int nitems = name ? 2 : 1;
6489 path = btrfs_alloc_path();
6491 return ERR_PTR(-ENOMEM);
6493 inode = new_inode(fs_info->sb);
6495 btrfs_free_path(path);
6496 return ERR_PTR(-ENOMEM);
6500 * O_TMPFILE, set link count to 0, so that after this point,
6501 * we fill in an inode item with the correct link count.
6504 set_nlink(inode, 0);
6507 * we have to initialize this early, so we can reclaim the inode
6508 * number if we fail afterwards in this function.
6510 inode->i_ino = objectid;
6513 trace_btrfs_inode_request(dir);
6515 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6517 btrfs_free_path(path);
6519 return ERR_PTR(ret);
6525 * index_cnt is ignored for everything but a dir,
6526 * btrfs_set_inode_index_count has an explanation for the magic
6529 BTRFS_I(inode)->index_cnt = 2;
6530 BTRFS_I(inode)->dir_index = *index;
6531 BTRFS_I(inode)->root = root;
6532 BTRFS_I(inode)->generation = trans->transid;
6533 inode->i_generation = BTRFS_I(inode)->generation;
6536 * We could have gotten an inode number from somebody who was fsynced
6537 * and then removed in this same transaction, so let's just set full
6538 * sync since it will be a full sync anyway and this will blow away the
6539 * old info in the log.
6541 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6543 key[0].objectid = objectid;
6544 key[0].type = BTRFS_INODE_ITEM_KEY;
6547 sizes[0] = sizeof(struct btrfs_inode_item);
6551 * Start new inodes with an inode_ref. This is slightly more
6552 * efficient for small numbers of hard links since they will
6553 * be packed into one item. Extended refs will kick in if we
6554 * add more hard links than can fit in the ref item.
6556 key[1].objectid = objectid;
6557 key[1].type = BTRFS_INODE_REF_KEY;
6558 key[1].offset = ref_objectid;
6560 sizes[1] = name_len + sizeof(*ref);
6563 location = &BTRFS_I(inode)->location;
6564 location->objectid = objectid;
6565 location->offset = 0;
6566 location->type = BTRFS_INODE_ITEM_KEY;
6568 ret = btrfs_insert_inode_locked(inode);
6572 path->leave_spinning = 1;
6573 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6577 inode_init_owner(inode, dir, mode);
6578 inode_set_bytes(inode, 0);
6580 inode->i_mtime = current_time(inode);
6581 inode->i_atime = inode->i_mtime;
6582 inode->i_ctime = inode->i_mtime;
6583 BTRFS_I(inode)->i_otime = inode->i_mtime;
6585 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6586 struct btrfs_inode_item);
6587 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6588 sizeof(*inode_item));
6589 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6592 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6593 struct btrfs_inode_ref);
6594 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6595 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6596 ptr = (unsigned long)(ref + 1);
6597 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6600 btrfs_mark_buffer_dirty(path->nodes[0]);
6601 btrfs_free_path(path);
6603 btrfs_inherit_iflags(inode, dir);
6605 if (S_ISREG(mode)) {
6606 if (btrfs_test_opt(fs_info, NODATASUM))
6607 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6608 if (btrfs_test_opt(fs_info, NODATACOW))
6609 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6610 BTRFS_INODE_NODATASUM;
6613 inode_tree_add(inode);
6615 trace_btrfs_inode_new(inode);
6616 btrfs_set_inode_last_trans(trans, inode);
6618 btrfs_update_root_times(trans, root);
6620 ret = btrfs_inode_inherit_props(trans, inode, dir);
6623 "error inheriting props for ino %llu (root %llu): %d",
6624 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6629 unlock_new_inode(inode);
6632 BTRFS_I(dir)->index_cnt--;
6633 btrfs_free_path(path);
6635 return ERR_PTR(ret);
6638 static inline u8 btrfs_inode_type(struct inode *inode)
6640 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6644 * utility function to add 'inode' into 'parent_inode' with
6645 * a give name and a given sequence number.
6646 * if 'add_backref' is true, also insert a backref from the
6647 * inode to the parent directory.
6649 int btrfs_add_link(struct btrfs_trans_handle *trans,
6650 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6651 const char *name, int name_len, int add_backref, u64 index)
6653 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6655 struct btrfs_key key;
6656 struct btrfs_root *root = parent_inode->root;
6657 u64 ino = btrfs_ino(inode);
6658 u64 parent_ino = btrfs_ino(parent_inode);
6660 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6661 memcpy(&key, &inode->root->root_key, sizeof(key));
6664 key.type = BTRFS_INODE_ITEM_KEY;
6668 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6669 ret = btrfs_add_root_ref(trans, fs_info, key.objectid,
6670 root->root_key.objectid, parent_ino,
6671 index, name, name_len);
6672 } else if (add_backref) {
6673 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6677 /* Nothing to clean up yet */
6681 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6683 btrfs_inode_type(&inode->vfs_inode), index);
6684 if (ret == -EEXIST || ret == -EOVERFLOW)
6687 btrfs_abort_transaction(trans, ret);
6691 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6693 inode_inc_iversion(&parent_inode->vfs_inode);
6694 parent_inode->vfs_inode.i_mtime = parent_inode->vfs_inode.i_ctime =
6695 current_time(&parent_inode->vfs_inode);
6696 ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode);
6698 btrfs_abort_transaction(trans, ret);
6702 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6705 err = btrfs_del_root_ref(trans, fs_info, key.objectid,
6706 root->root_key.objectid, parent_ino,
6707 &local_index, name, name_len);
6709 } else if (add_backref) {
6713 err = btrfs_del_inode_ref(trans, root, name, name_len,
6714 ino, parent_ino, &local_index);
6719 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6720 struct btrfs_inode *dir, struct dentry *dentry,
6721 struct btrfs_inode *inode, int backref, u64 index)
6723 int err = btrfs_add_link(trans, dir, inode,
6724 dentry->d_name.name, dentry->d_name.len,
6731 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6732 umode_t mode, dev_t rdev)
6734 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6735 struct btrfs_trans_handle *trans;
6736 struct btrfs_root *root = BTRFS_I(dir)->root;
6737 struct inode *inode = NULL;
6744 * 2 for inode item and ref
6746 * 1 for xattr if selinux is on
6748 trans = btrfs_start_transaction(root, 5);
6750 return PTR_ERR(trans);
6752 err = btrfs_find_free_ino(root, &objectid);
6756 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6757 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6759 if (IS_ERR(inode)) {
6760 err = PTR_ERR(inode);
6765 * If the active LSM wants to access the inode during
6766 * d_instantiate it needs these. Smack checks to see
6767 * if the filesystem supports xattrs by looking at the
6770 inode->i_op = &btrfs_special_inode_operations;
6771 init_special_inode(inode, inode->i_mode, rdev);
6773 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6775 goto out_unlock_inode;
6777 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6780 goto out_unlock_inode;
6782 btrfs_update_inode(trans, root, inode);
6783 d_instantiate_new(dentry, inode);
6787 btrfs_end_transaction(trans);
6788 btrfs_btree_balance_dirty(fs_info);
6790 inode_dec_link_count(inode);
6797 unlock_new_inode(inode);
6802 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6803 umode_t mode, bool excl)
6805 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6806 struct btrfs_trans_handle *trans;
6807 struct btrfs_root *root = BTRFS_I(dir)->root;
6808 struct inode *inode = NULL;
6809 int drop_inode_on_err = 0;
6815 * 2 for inode item and ref
6817 * 1 for xattr if selinux is on
6819 trans = btrfs_start_transaction(root, 5);
6821 return PTR_ERR(trans);
6823 err = btrfs_find_free_ino(root, &objectid);
6827 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6828 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6830 if (IS_ERR(inode)) {
6831 err = PTR_ERR(inode);
6834 drop_inode_on_err = 1;
6836 * If the active LSM wants to access the inode during
6837 * d_instantiate it needs these. Smack checks to see
6838 * if the filesystem supports xattrs by looking at the
6841 inode->i_fop = &btrfs_file_operations;
6842 inode->i_op = &btrfs_file_inode_operations;
6843 inode->i_mapping->a_ops = &btrfs_aops;
6845 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6847 goto out_unlock_inode;
6849 err = btrfs_update_inode(trans, root, inode);
6851 goto out_unlock_inode;
6853 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6856 goto out_unlock_inode;
6858 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6859 d_instantiate_new(dentry, inode);
6862 btrfs_end_transaction(trans);
6863 if (err && drop_inode_on_err) {
6864 inode_dec_link_count(inode);
6867 btrfs_btree_balance_dirty(fs_info);
6871 unlock_new_inode(inode);
6876 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6877 struct dentry *dentry)
6879 struct btrfs_trans_handle *trans = NULL;
6880 struct btrfs_root *root = BTRFS_I(dir)->root;
6881 struct inode *inode = d_inode(old_dentry);
6882 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6887 /* do not allow sys_link's with other subvols of the same device */
6888 if (root->objectid != BTRFS_I(inode)->root->objectid)
6891 if (inode->i_nlink >= BTRFS_LINK_MAX)
6894 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6899 * 2 items for inode and inode ref
6900 * 2 items for dir items
6901 * 1 item for parent inode
6903 trans = btrfs_start_transaction(root, 5);
6904 if (IS_ERR(trans)) {
6905 err = PTR_ERR(trans);
6910 /* There are several dir indexes for this inode, clear the cache. */
6911 BTRFS_I(inode)->dir_index = 0ULL;
6913 inode_inc_iversion(inode);
6914 inode->i_ctime = current_time(inode);
6916 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6918 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6924 struct dentry *parent = dentry->d_parent;
6925 err = btrfs_update_inode(trans, root, inode);
6928 if (inode->i_nlink == 1) {
6930 * If new hard link count is 1, it's a file created
6931 * with open(2) O_TMPFILE flag.
6933 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6937 d_instantiate(dentry, inode);
6938 btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent);
6943 btrfs_end_transaction(trans);
6945 inode_dec_link_count(inode);
6948 btrfs_btree_balance_dirty(fs_info);
6952 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6954 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6955 struct inode *inode = NULL;
6956 struct btrfs_trans_handle *trans;
6957 struct btrfs_root *root = BTRFS_I(dir)->root;
6959 int drop_on_err = 0;
6964 * 2 items for inode and ref
6965 * 2 items for dir items
6966 * 1 for xattr if selinux is on
6968 trans = btrfs_start_transaction(root, 5);
6970 return PTR_ERR(trans);
6972 err = btrfs_find_free_ino(root, &objectid);
6976 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6977 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6978 S_IFDIR | mode, &index);
6979 if (IS_ERR(inode)) {
6980 err = PTR_ERR(inode);
6985 /* these must be set before we unlock the inode */
6986 inode->i_op = &btrfs_dir_inode_operations;
6987 inode->i_fop = &btrfs_dir_file_operations;
6989 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6991 goto out_fail_inode;
6993 btrfs_i_size_write(BTRFS_I(inode), 0);
6994 err = btrfs_update_inode(trans, root, inode);
6996 goto out_fail_inode;
6998 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6999 dentry->d_name.name,
7000 dentry->d_name.len, 0, index);
7002 goto out_fail_inode;
7004 d_instantiate_new(dentry, inode);
7008 btrfs_end_transaction(trans);
7010 inode_dec_link_count(inode);
7013 btrfs_btree_balance_dirty(fs_info);
7017 unlock_new_inode(inode);
7021 static noinline int uncompress_inline(struct btrfs_path *path,
7023 size_t pg_offset, u64 extent_offset,
7024 struct btrfs_file_extent_item *item)
7027 struct extent_buffer *leaf = path->nodes[0];
7030 unsigned long inline_size;
7034 WARN_ON(pg_offset != 0);
7035 compress_type = btrfs_file_extent_compression(leaf, item);
7036 max_size = btrfs_file_extent_ram_bytes(leaf, item);
7037 inline_size = btrfs_file_extent_inline_item_len(leaf,
7038 btrfs_item_nr(path->slots[0]));
7039 tmp = kmalloc(inline_size, GFP_NOFS);
7042 ptr = btrfs_file_extent_inline_start(item);
7044 read_extent_buffer(leaf, tmp, ptr, inline_size);
7046 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
7047 ret = btrfs_decompress(compress_type, tmp, page,
7048 extent_offset, inline_size, max_size);
7051 * decompression code contains a memset to fill in any space between the end
7052 * of the uncompressed data and the end of max_size in case the decompressed
7053 * data ends up shorter than ram_bytes. That doesn't cover the hole between
7054 * the end of an inline extent and the beginning of the next block, so we
7055 * cover that region here.
7058 if (max_size + pg_offset < PAGE_SIZE) {
7059 char *map = kmap(page);
7060 memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
7068 * a bit scary, this does extent mapping from logical file offset to the disk.
7069 * the ugly parts come from merging extents from the disk with the in-ram
7070 * representation. This gets more complex because of the data=ordered code,
7071 * where the in-ram extents might be locked pending data=ordered completion.
7073 * This also copies inline extents directly into the page.
7075 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
7077 size_t pg_offset, u64 start, u64 len,
7080 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
7083 u64 extent_start = 0;
7085 u64 objectid = btrfs_ino(inode);
7087 struct btrfs_path *path = NULL;
7088 struct btrfs_root *root = inode->root;
7089 struct btrfs_file_extent_item *item;
7090 struct extent_buffer *leaf;
7091 struct btrfs_key found_key;
7092 struct extent_map *em = NULL;
7093 struct extent_map_tree *em_tree = &inode->extent_tree;
7094 struct extent_io_tree *io_tree = &inode->io_tree;
7095 const bool new_inline = !page || create;
7097 read_lock(&em_tree->lock);
7098 em = lookup_extent_mapping(em_tree, start, len);
7100 em->bdev = fs_info->fs_devices->latest_bdev;
7101 read_unlock(&em_tree->lock);
7104 if (em->start > start || em->start + em->len <= start)
7105 free_extent_map(em);
7106 else if (em->block_start == EXTENT_MAP_INLINE && page)
7107 free_extent_map(em);
7111 em = alloc_extent_map();
7116 em->bdev = fs_info->fs_devices->latest_bdev;
7117 em->start = EXTENT_MAP_HOLE;
7118 em->orig_start = EXTENT_MAP_HOLE;
7120 em->block_len = (u64)-1;
7123 path = btrfs_alloc_path();
7129 * Chances are we'll be called again, so go ahead and do
7132 path->reada = READA_FORWARD;
7135 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
7142 if (path->slots[0] == 0)
7147 leaf = path->nodes[0];
7148 item = btrfs_item_ptr(leaf, path->slots[0],
7149 struct btrfs_file_extent_item);
7150 /* are we inside the extent that was found? */
7151 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7152 found_type = found_key.type;
7153 if (found_key.objectid != objectid ||
7154 found_type != BTRFS_EXTENT_DATA_KEY) {
7156 * If we backup past the first extent we want to move forward
7157 * and see if there is an extent in front of us, otherwise we'll
7158 * say there is a hole for our whole search range which can
7165 found_type = btrfs_file_extent_type(leaf, item);
7166 extent_start = found_key.offset;
7167 if (found_type == BTRFS_FILE_EXTENT_REG ||
7168 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7169 extent_end = extent_start +
7170 btrfs_file_extent_num_bytes(leaf, item);
7172 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
7174 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
7176 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
7177 extent_end = ALIGN(extent_start + size,
7178 fs_info->sectorsize);
7180 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
7185 if (start >= extent_end) {
7187 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
7188 ret = btrfs_next_leaf(root, path);
7195 leaf = path->nodes[0];
7197 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7198 if (found_key.objectid != objectid ||
7199 found_key.type != BTRFS_EXTENT_DATA_KEY)
7201 if (start + len <= found_key.offset)
7203 if (start > found_key.offset)
7206 em->orig_start = start;
7207 em->len = found_key.offset - start;
7211 btrfs_extent_item_to_extent_map(inode, path, item,
7214 if (found_type == BTRFS_FILE_EXTENT_REG ||
7215 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7217 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
7221 size_t extent_offset;
7227 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
7228 extent_offset = page_offset(page) + pg_offset - extent_start;
7229 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
7230 size - extent_offset);
7231 em->start = extent_start + extent_offset;
7232 em->len = ALIGN(copy_size, fs_info->sectorsize);
7233 em->orig_block_len = em->len;
7234 em->orig_start = em->start;
7235 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
7236 if (!PageUptodate(page)) {
7237 if (btrfs_file_extent_compression(leaf, item) !=
7238 BTRFS_COMPRESS_NONE) {
7239 ret = uncompress_inline(path, page, pg_offset,
7240 extent_offset, item);
7247 read_extent_buffer(leaf, map + pg_offset, ptr,
7249 if (pg_offset + copy_size < PAGE_SIZE) {
7250 memset(map + pg_offset + copy_size, 0,
7251 PAGE_SIZE - pg_offset -
7256 flush_dcache_page(page);
7258 set_extent_uptodate(io_tree, em->start,
7259 extent_map_end(em) - 1, NULL, GFP_NOFS);
7264 em->orig_start = start;
7267 em->block_start = EXTENT_MAP_HOLE;
7269 btrfs_release_path(path);
7270 if (em->start > start || extent_map_end(em) <= start) {
7272 "bad extent! em: [%llu %llu] passed [%llu %llu]",
7273 em->start, em->len, start, len);
7279 write_lock(&em_tree->lock);
7280 err = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
7281 write_unlock(&em_tree->lock);
7284 trace_btrfs_get_extent(root, inode, em);
7286 btrfs_free_path(path);
7288 free_extent_map(em);
7289 return ERR_PTR(err);
7291 BUG_ON(!em); /* Error is always set */
7295 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
7297 size_t pg_offset, u64 start, u64 len,
7300 struct extent_map *em;
7301 struct extent_map *hole_em = NULL;
7302 u64 range_start = start;
7308 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7312 * If our em maps to:
7314 * - a pre-alloc extent,
7315 * there might actually be delalloc bytes behind it.
7317 if (em->block_start != EXTENT_MAP_HOLE &&
7318 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7323 /* check to see if we've wrapped (len == -1 or similar) */
7332 /* ok, we didn't find anything, lets look for delalloc */
7333 found = count_range_bits(&inode->io_tree, &range_start,
7334 end, len, EXTENT_DELALLOC, 1);
7335 found_end = range_start + found;
7336 if (found_end < range_start)
7337 found_end = (u64)-1;
7340 * we didn't find anything useful, return
7341 * the original results from get_extent()
7343 if (range_start > end || found_end <= start) {
7349 /* adjust the range_start to make sure it doesn't
7350 * go backwards from the start they passed in
7352 range_start = max(start, range_start);
7353 found = found_end - range_start;
7356 u64 hole_start = start;
7359 em = alloc_extent_map();
7365 * when btrfs_get_extent can't find anything it
7366 * returns one huge hole
7368 * make sure what it found really fits our range, and
7369 * adjust to make sure it is based on the start from
7373 u64 calc_end = extent_map_end(hole_em);
7375 if (calc_end <= start || (hole_em->start > end)) {
7376 free_extent_map(hole_em);
7379 hole_start = max(hole_em->start, start);
7380 hole_len = calc_end - hole_start;
7384 if (hole_em && range_start > hole_start) {
7385 /* our hole starts before our delalloc, so we
7386 * have to return just the parts of the hole
7387 * that go until the delalloc starts
7389 em->len = min(hole_len,
7390 range_start - hole_start);
7391 em->start = hole_start;
7392 em->orig_start = hole_start;
7394 * don't adjust block start at all,
7395 * it is fixed at EXTENT_MAP_HOLE
7397 em->block_start = hole_em->block_start;
7398 em->block_len = hole_len;
7399 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7400 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7402 em->start = range_start;
7404 em->orig_start = range_start;
7405 em->block_start = EXTENT_MAP_DELALLOC;
7406 em->block_len = found;
7413 free_extent_map(hole_em);
7415 free_extent_map(em);
7416 return ERR_PTR(err);
7421 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7424 const u64 orig_start,
7425 const u64 block_start,
7426 const u64 block_len,
7427 const u64 orig_block_len,
7428 const u64 ram_bytes,
7431 struct extent_map *em = NULL;
7434 if (type != BTRFS_ORDERED_NOCOW) {
7435 em = create_io_em(inode, start, len, orig_start,
7436 block_start, block_len, orig_block_len,
7438 BTRFS_COMPRESS_NONE, /* compress_type */
7443 ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7444 len, block_len, type);
7447 free_extent_map(em);
7448 btrfs_drop_extent_cache(BTRFS_I(inode), start,
7449 start + len - 1, 0);
7458 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7461 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7462 struct btrfs_root *root = BTRFS_I(inode)->root;
7463 struct extent_map *em;
7464 struct btrfs_key ins;
7468 alloc_hint = get_extent_allocation_hint(inode, start, len);
7469 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7470 0, alloc_hint, &ins, 1, 1);
7472 return ERR_PTR(ret);
7474 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7475 ins.objectid, ins.offset, ins.offset,
7476 ins.offset, BTRFS_ORDERED_REGULAR);
7477 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7479 btrfs_free_reserved_extent(fs_info, ins.objectid,
7486 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7487 * block must be cow'd
7489 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7490 u64 *orig_start, u64 *orig_block_len,
7493 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7494 struct btrfs_path *path;
7496 struct extent_buffer *leaf;
7497 struct btrfs_root *root = BTRFS_I(inode)->root;
7498 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7499 struct btrfs_file_extent_item *fi;
7500 struct btrfs_key key;
7507 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7509 path = btrfs_alloc_path();
7513 ret = btrfs_lookup_file_extent(NULL, root, path,
7514 btrfs_ino(BTRFS_I(inode)), offset, 0);
7518 slot = path->slots[0];
7521 /* can't find the item, must cow */
7528 leaf = path->nodes[0];
7529 btrfs_item_key_to_cpu(leaf, &key, slot);
7530 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7531 key.type != BTRFS_EXTENT_DATA_KEY) {
7532 /* not our file or wrong item type, must cow */
7536 if (key.offset > offset) {
7537 /* Wrong offset, must cow */
7541 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7542 found_type = btrfs_file_extent_type(leaf, fi);
7543 if (found_type != BTRFS_FILE_EXTENT_REG &&
7544 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7545 /* not a regular extent, must cow */
7549 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7552 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7553 if (extent_end <= offset)
7556 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7557 if (disk_bytenr == 0)
7560 if (btrfs_file_extent_compression(leaf, fi) ||
7561 btrfs_file_extent_encryption(leaf, fi) ||
7562 btrfs_file_extent_other_encoding(leaf, fi))
7565 backref_offset = btrfs_file_extent_offset(leaf, fi);
7568 *orig_start = key.offset - backref_offset;
7569 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7570 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7573 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7576 num_bytes = min(offset + *len, extent_end) - offset;
7577 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7580 range_end = round_up(offset + num_bytes,
7581 root->fs_info->sectorsize) - 1;
7582 ret = test_range_bit(io_tree, offset, range_end,
7583 EXTENT_DELALLOC, 0, NULL);
7590 btrfs_release_path(path);
7593 * look for other files referencing this extent, if we
7594 * find any we must cow
7597 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7598 key.offset - backref_offset, disk_bytenr);
7605 * adjust disk_bytenr and num_bytes to cover just the bytes
7606 * in this extent we are about to write. If there
7607 * are any csums in that range we have to cow in order
7608 * to keep the csums correct
7610 disk_bytenr += backref_offset;
7611 disk_bytenr += offset - key.offset;
7612 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7615 * all of the above have passed, it is safe to overwrite this extent
7621 btrfs_free_path(path);
7625 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7626 struct extent_state **cached_state, int writing)
7628 struct btrfs_ordered_extent *ordered;
7632 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7635 * We're concerned with the entire range that we're going to be
7636 * doing DIO to, so we need to make sure there's no ordered
7637 * extents in this range.
7639 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7640 lockend - lockstart + 1);
7643 * We need to make sure there are no buffered pages in this
7644 * range either, we could have raced between the invalidate in
7645 * generic_file_direct_write and locking the extent. The
7646 * invalidate needs to happen so that reads after a write do not
7650 (!writing || !filemap_range_has_page(inode->i_mapping,
7651 lockstart, lockend)))
7654 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7659 * If we are doing a DIO read and the ordered extent we
7660 * found is for a buffered write, we can not wait for it
7661 * to complete and retry, because if we do so we can
7662 * deadlock with concurrent buffered writes on page
7663 * locks. This happens only if our DIO read covers more
7664 * than one extent map, if at this point has already
7665 * created an ordered extent for a previous extent map
7666 * and locked its range in the inode's io tree, and a
7667 * concurrent write against that previous extent map's
7668 * range and this range started (we unlock the ranges
7669 * in the io tree only when the bios complete and
7670 * buffered writes always lock pages before attempting
7671 * to lock range in the io tree).
7674 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7675 btrfs_start_ordered_extent(inode, ordered, 1);
7678 btrfs_put_ordered_extent(ordered);
7681 * We could trigger writeback for this range (and wait
7682 * for it to complete) and then invalidate the pages for
7683 * this range (through invalidate_inode_pages2_range()),
7684 * but that can lead us to a deadlock with a concurrent
7685 * call to readpages() (a buffered read or a defrag call
7686 * triggered a readahead) on a page lock due to an
7687 * ordered dio extent we created before but did not have
7688 * yet a corresponding bio submitted (whence it can not
7689 * complete), which makes readpages() wait for that
7690 * ordered extent to complete while holding a lock on
7705 /* The callers of this must take lock_extent() */
7706 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
7707 u64 orig_start, u64 block_start,
7708 u64 block_len, u64 orig_block_len,
7709 u64 ram_bytes, int compress_type,
7712 struct extent_map_tree *em_tree;
7713 struct extent_map *em;
7714 struct btrfs_root *root = BTRFS_I(inode)->root;
7717 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7718 type == BTRFS_ORDERED_COMPRESSED ||
7719 type == BTRFS_ORDERED_NOCOW ||
7720 type == BTRFS_ORDERED_REGULAR);
7722 em_tree = &BTRFS_I(inode)->extent_tree;
7723 em = alloc_extent_map();
7725 return ERR_PTR(-ENOMEM);
7728 em->orig_start = orig_start;
7730 em->block_len = block_len;
7731 em->block_start = block_start;
7732 em->bdev = root->fs_info->fs_devices->latest_bdev;
7733 em->orig_block_len = orig_block_len;
7734 em->ram_bytes = ram_bytes;
7735 em->generation = -1;
7736 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7737 if (type == BTRFS_ORDERED_PREALLOC) {
7738 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7739 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7740 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7741 em->compress_type = compress_type;
7745 btrfs_drop_extent_cache(BTRFS_I(inode), em->start,
7746 em->start + em->len - 1, 0);
7747 write_lock(&em_tree->lock);
7748 ret = add_extent_mapping(em_tree, em, 1);
7749 write_unlock(&em_tree->lock);
7751 * The caller has taken lock_extent(), who could race with us
7754 } while (ret == -EEXIST);
7757 free_extent_map(em);
7758 return ERR_PTR(ret);
7761 /* em got 2 refs now, callers needs to do free_extent_map once. */
7765 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7766 struct buffer_head *bh_result, int create)
7768 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7769 struct extent_map *em;
7770 struct extent_state *cached_state = NULL;
7771 struct btrfs_dio_data *dio_data = NULL;
7772 u64 start = iblock << inode->i_blkbits;
7773 u64 lockstart, lockend;
7774 u64 len = bh_result->b_size;
7775 int unlock_bits = EXTENT_LOCKED;
7779 unlock_bits |= EXTENT_DIRTY;
7781 len = min_t(u64, len, fs_info->sectorsize);
7784 lockend = start + len - 1;
7786 if (current->journal_info) {
7788 * Need to pull our outstanding extents and set journal_info to NULL so
7789 * that anything that needs to check if there's a transaction doesn't get
7792 dio_data = current->journal_info;
7793 current->journal_info = NULL;
7797 * If this errors out it's because we couldn't invalidate pagecache for
7798 * this range and we need to fallback to buffered.
7800 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7806 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
7813 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7814 * io. INLINE is special, and we could probably kludge it in here, but
7815 * it's still buffered so for safety lets just fall back to the generic
7818 * For COMPRESSED we _have_ to read the entire extent in so we can
7819 * decompress it, so there will be buffering required no matter what we
7820 * do, so go ahead and fallback to buffered.
7822 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7823 * to buffered IO. Don't blame me, this is the price we pay for using
7826 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7827 em->block_start == EXTENT_MAP_INLINE) {
7828 free_extent_map(em);
7833 /* Just a good old fashioned hole, return */
7834 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7835 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7836 free_extent_map(em);
7841 * We don't allocate a new extent in the following cases
7843 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7845 * 2) The extent is marked as PREALLOC. We're good to go here and can
7846 * just use the extent.
7850 len = min(len, em->len - (start - em->start));
7851 lockstart = start + len;
7855 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7856 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7857 em->block_start != EXTENT_MAP_HOLE)) {
7859 u64 block_start, orig_start, orig_block_len, ram_bytes;
7861 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7862 type = BTRFS_ORDERED_PREALLOC;
7864 type = BTRFS_ORDERED_NOCOW;
7865 len = min(len, em->len - (start - em->start));
7866 block_start = em->block_start + (start - em->start);
7868 if (can_nocow_extent(inode, start, &len, &orig_start,
7869 &orig_block_len, &ram_bytes) == 1 &&
7870 btrfs_inc_nocow_writers(fs_info, block_start)) {
7871 struct extent_map *em2;
7873 em2 = btrfs_create_dio_extent(inode, start, len,
7874 orig_start, block_start,
7875 len, orig_block_len,
7877 btrfs_dec_nocow_writers(fs_info, block_start);
7878 if (type == BTRFS_ORDERED_PREALLOC) {
7879 free_extent_map(em);
7882 if (em2 && IS_ERR(em2)) {
7887 * For inode marked NODATACOW or extent marked PREALLOC,
7888 * use the existing or preallocated extent, so does not
7889 * need to adjust btrfs_space_info's bytes_may_use.
7891 btrfs_free_reserved_data_space_noquota(inode,
7898 * this will cow the extent, reset the len in case we changed
7901 len = bh_result->b_size;
7902 free_extent_map(em);
7903 em = btrfs_new_extent_direct(inode, start, len);
7908 len = min(len, em->len - (start - em->start));
7910 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7912 bh_result->b_size = len;
7913 bh_result->b_bdev = em->bdev;
7914 set_buffer_mapped(bh_result);
7916 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7917 set_buffer_new(bh_result);
7920 * Need to update the i_size under the extent lock so buffered
7921 * readers will get the updated i_size when we unlock.
7923 if (!dio_data->overwrite && start + len > i_size_read(inode))
7924 i_size_write(inode, start + len);
7926 WARN_ON(dio_data->reserve < len);
7927 dio_data->reserve -= len;
7928 dio_data->unsubmitted_oe_range_end = start + len;
7929 current->journal_info = dio_data;
7933 * In the case of write we need to clear and unlock the entire range,
7934 * in the case of read we need to unlock only the end area that we
7935 * aren't using if there is any left over space.
7937 if (lockstart < lockend) {
7938 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7939 lockend, unlock_bits, 1, 0,
7942 free_extent_state(cached_state);
7945 free_extent_map(em);
7950 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7951 unlock_bits, 1, 0, &cached_state);
7954 current->journal_info = dio_data;
7958 static inline blk_status_t submit_dio_repair_bio(struct inode *inode,
7962 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7965 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7967 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR);
7971 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
7976 static int btrfs_check_dio_repairable(struct inode *inode,
7977 struct bio *failed_bio,
7978 struct io_failure_record *failrec,
7981 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7984 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
7985 if (num_copies == 1) {
7987 * we only have a single copy of the data, so don't bother with
7988 * all the retry and error correction code that follows. no
7989 * matter what the error is, it is very likely to persist.
7991 btrfs_debug(fs_info,
7992 "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
7993 num_copies, failrec->this_mirror, failed_mirror);
7997 failrec->failed_mirror = failed_mirror;
7998 failrec->this_mirror++;
7999 if (failrec->this_mirror == failed_mirror)
8000 failrec->this_mirror++;
8002 if (failrec->this_mirror > num_copies) {
8003 btrfs_debug(fs_info,
8004 "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
8005 num_copies, failrec->this_mirror, failed_mirror);
8012 static blk_status_t dio_read_error(struct inode *inode, struct bio *failed_bio,
8013 struct page *page, unsigned int pgoff,
8014 u64 start, u64 end, int failed_mirror,
8015 bio_end_io_t *repair_endio, void *repair_arg)
8017 struct io_failure_record *failrec;
8018 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8019 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
8022 unsigned int read_mode = 0;
8025 blk_status_t status;
8026 struct bio_vec bvec;
8028 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
8030 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
8032 return errno_to_blk_status(ret);
8034 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
8037 free_io_failure(failure_tree, io_tree, failrec);
8038 return BLK_STS_IOERR;
8041 segs = bio_segments(failed_bio);
8042 bio_get_first_bvec(failed_bio, &bvec);
8044 (bvec.bv_len > btrfs_inode_sectorsize(inode)))
8045 read_mode |= REQ_FAILFAST_DEV;
8047 isector = start - btrfs_io_bio(failed_bio)->logical;
8048 isector >>= inode->i_sb->s_blocksize_bits;
8049 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
8050 pgoff, isector, repair_endio, repair_arg);
8051 bio_set_op_attrs(bio, REQ_OP_READ, read_mode);
8053 btrfs_debug(BTRFS_I(inode)->root->fs_info,
8054 "repair DIO read error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d",
8055 read_mode, failrec->this_mirror, failrec->in_validation);
8057 status = submit_dio_repair_bio(inode, bio, failrec->this_mirror);
8059 free_io_failure(failure_tree, io_tree, failrec);
8066 struct btrfs_retry_complete {
8067 struct completion done;
8068 struct inode *inode;
8073 static void btrfs_retry_endio_nocsum(struct bio *bio)
8075 struct btrfs_retry_complete *done = bio->bi_private;
8076 struct inode *inode = done->inode;
8077 struct bio_vec *bvec;
8078 struct extent_io_tree *io_tree, *failure_tree;
8084 ASSERT(bio->bi_vcnt == 1);
8085 io_tree = &BTRFS_I(inode)->io_tree;
8086 failure_tree = &BTRFS_I(inode)->io_failure_tree;
8087 ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(inode));
8090 ASSERT(!bio_flagged(bio, BIO_CLONED));
8091 bio_for_each_segment_all(bvec, bio, i)
8092 clean_io_failure(BTRFS_I(inode)->root->fs_info, failure_tree,
8093 io_tree, done->start, bvec->bv_page,
8094 btrfs_ino(BTRFS_I(inode)), 0);
8096 complete(&done->done);
8100 static blk_status_t __btrfs_correct_data_nocsum(struct inode *inode,
8101 struct btrfs_io_bio *io_bio)
8103 struct btrfs_fs_info *fs_info;
8104 struct bio_vec bvec;
8105 struct bvec_iter iter;
8106 struct btrfs_retry_complete done;
8112 blk_status_t err = BLK_STS_OK;
8114 fs_info = BTRFS_I(inode)->root->fs_info;
8115 sectorsize = fs_info->sectorsize;
8117 start = io_bio->logical;
8119 io_bio->bio.bi_iter = io_bio->iter;
8121 bio_for_each_segment(bvec, &io_bio->bio, iter) {
8122 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
8123 pgoff = bvec.bv_offset;
8125 next_block_or_try_again:
8128 init_completion(&done.done);
8130 ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
8131 pgoff, start, start + sectorsize - 1,
8133 btrfs_retry_endio_nocsum, &done);
8139 wait_for_completion_io(&done.done);
8141 if (!done.uptodate) {
8142 /* We might have another mirror, so try again */
8143 goto next_block_or_try_again;
8147 start += sectorsize;
8151 pgoff += sectorsize;
8152 ASSERT(pgoff < PAGE_SIZE);
8153 goto next_block_or_try_again;
8160 static void btrfs_retry_endio(struct bio *bio)
8162 struct btrfs_retry_complete *done = bio->bi_private;
8163 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8164 struct extent_io_tree *io_tree, *failure_tree;
8165 struct inode *inode = done->inode;
8166 struct bio_vec *bvec;
8176 ASSERT(bio->bi_vcnt == 1);
8177 ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(done->inode));
8179 io_tree = &BTRFS_I(inode)->io_tree;
8180 failure_tree = &BTRFS_I(inode)->io_failure_tree;
8182 ASSERT(!bio_flagged(bio, BIO_CLONED));
8183 bio_for_each_segment_all(bvec, bio, i) {
8184 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
8185 bvec->bv_offset, done->start,
8188 clean_io_failure(BTRFS_I(inode)->root->fs_info,
8189 failure_tree, io_tree, done->start,
8191 btrfs_ino(BTRFS_I(inode)),
8197 done->uptodate = uptodate;
8199 complete(&done->done);
8203 static blk_status_t __btrfs_subio_endio_read(struct inode *inode,
8204 struct btrfs_io_bio *io_bio, blk_status_t err)
8206 struct btrfs_fs_info *fs_info;
8207 struct bio_vec bvec;
8208 struct bvec_iter iter;
8209 struct btrfs_retry_complete done;
8216 bool uptodate = (err == 0);
8218 blk_status_t status;
8220 fs_info = BTRFS_I(inode)->root->fs_info;
8221 sectorsize = fs_info->sectorsize;
8224 start = io_bio->logical;
8226 io_bio->bio.bi_iter = io_bio->iter;
8228 bio_for_each_segment(bvec, &io_bio->bio, iter) {
8229 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
8231 pgoff = bvec.bv_offset;
8234 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
8235 ret = __readpage_endio_check(inode, io_bio, csum_pos,
8236 bvec.bv_page, pgoff, start, sectorsize);
8243 init_completion(&done.done);
8245 status = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
8246 pgoff, start, start + sectorsize - 1,
8247 io_bio->mirror_num, btrfs_retry_endio,
8254 wait_for_completion_io(&done.done);
8256 if (!done.uptodate) {
8257 /* We might have another mirror, so try again */
8261 offset += sectorsize;
8262 start += sectorsize;
8268 pgoff += sectorsize;
8269 ASSERT(pgoff < PAGE_SIZE);
8277 static blk_status_t btrfs_subio_endio_read(struct inode *inode,
8278 struct btrfs_io_bio *io_bio, blk_status_t err)
8280 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8284 return __btrfs_correct_data_nocsum(inode, io_bio);
8288 return __btrfs_subio_endio_read(inode, io_bio, err);
8292 static void btrfs_endio_direct_read(struct bio *bio)
8294 struct btrfs_dio_private *dip = bio->bi_private;
8295 struct inode *inode = dip->inode;
8296 struct bio *dio_bio;
8297 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8298 blk_status_t err = bio->bi_status;
8300 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8301 err = btrfs_subio_endio_read(inode, io_bio, err);
8303 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8304 dip->logical_offset + dip->bytes - 1);
8305 dio_bio = dip->dio_bio;
8309 dio_bio->bi_status = err;
8310 dio_end_io(dio_bio);
8313 io_bio->end_io(io_bio, blk_status_to_errno(err));
8317 static void __endio_write_update_ordered(struct inode *inode,
8318 const u64 offset, const u64 bytes,
8319 const bool uptodate)
8321 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8322 struct btrfs_ordered_extent *ordered = NULL;
8323 struct btrfs_workqueue *wq;
8324 btrfs_work_func_t func;
8325 u64 ordered_offset = offset;
8326 u64 ordered_bytes = bytes;
8329 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
8330 wq = fs_info->endio_freespace_worker;
8331 func = btrfs_freespace_write_helper;
8333 wq = fs_info->endio_write_workers;
8334 func = btrfs_endio_write_helper;
8337 while (ordered_offset < offset + bytes) {
8338 last_offset = ordered_offset;
8339 if (btrfs_dec_test_first_ordered_pending(inode, &ordered,
8343 btrfs_init_work(&ordered->work, func,
8346 btrfs_queue_work(wq, &ordered->work);
8349 * If btrfs_dec_test_ordered_pending does not find any ordered
8350 * extent in the range, we can exit.
8352 if (ordered_offset == last_offset)
8355 * Our bio might span multiple ordered extents. In this case
8356 * we keep goin until we have accounted the whole dio.
8358 if (ordered_offset < offset + bytes) {
8359 ordered_bytes = offset + bytes - ordered_offset;
8365 static void btrfs_endio_direct_write(struct bio *bio)
8367 struct btrfs_dio_private *dip = bio->bi_private;
8368 struct bio *dio_bio = dip->dio_bio;
8370 __endio_write_update_ordered(dip->inode, dip->logical_offset,
8371 dip->bytes, !bio->bi_status);
8375 dio_bio->bi_status = bio->bi_status;
8376 dio_end_io(dio_bio);
8380 static blk_status_t btrfs_submit_bio_start_direct_io(void *private_data,
8381 struct bio *bio, u64 offset)
8383 struct inode *inode = private_data;
8385 ret = btrfs_csum_one_bio(inode, bio, offset, 1);
8386 BUG_ON(ret); /* -ENOMEM */
8390 static void btrfs_end_dio_bio(struct bio *bio)
8392 struct btrfs_dio_private *dip = bio->bi_private;
8393 blk_status_t err = bio->bi_status;
8396 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8397 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8398 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8400 (unsigned long long)bio->bi_iter.bi_sector,
8401 bio->bi_iter.bi_size, err);
8403 if (dip->subio_endio)
8404 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8408 * We want to perceive the errors flag being set before
8409 * decrementing the reference count. We don't need a barrier
8410 * since atomic operations with a return value are fully
8411 * ordered as per atomic_t.txt
8416 /* if there are more bios still pending for this dio, just exit */
8417 if (!atomic_dec_and_test(&dip->pending_bios))
8421 bio_io_error(dip->orig_bio);
8423 dip->dio_bio->bi_status = BLK_STS_OK;
8424 bio_endio(dip->orig_bio);
8430 static inline blk_status_t btrfs_lookup_and_bind_dio_csum(struct inode *inode,
8431 struct btrfs_dio_private *dip,
8435 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8436 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8440 * We load all the csum data we need when we submit
8441 * the first bio to reduce the csum tree search and
8444 if (dip->logical_offset == file_offset) {
8445 ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio,
8451 if (bio == dip->orig_bio)
8454 file_offset -= dip->logical_offset;
8455 file_offset >>= inode->i_sb->s_blocksize_bits;
8456 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8461 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
8462 struct inode *inode, u64 file_offset, int async_submit)
8464 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8465 struct btrfs_dio_private *dip = bio->bi_private;
8466 bool write = bio_op(bio) == REQ_OP_WRITE;
8469 /* Check btrfs_submit_bio_hook() for rules about async submit. */
8471 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8474 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8479 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
8482 if (write && async_submit) {
8483 ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0,
8485 btrfs_submit_bio_start_direct_io,
8486 btrfs_submit_bio_done);
8490 * If we aren't doing async submit, calculate the csum of the
8493 ret = btrfs_csum_one_bio(inode, bio, file_offset, 1);
8497 ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio,
8503 ret = btrfs_map_bio(fs_info, bio, 0, 0);
8508 static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip)
8510 struct inode *inode = dip->inode;
8511 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8513 struct bio *orig_bio = dip->orig_bio;
8514 u64 start_sector = orig_bio->bi_iter.bi_sector;
8515 u64 file_offset = dip->logical_offset;
8517 int async_submit = 0;
8519 int clone_offset = 0;
8522 blk_status_t status;
8524 map_length = orig_bio->bi_iter.bi_size;
8525 submit_len = map_length;
8526 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9,
8527 &map_length, NULL, 0);
8531 if (map_length >= submit_len) {
8533 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8537 /* async crcs make it difficult to collect full stripe writes. */
8538 if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8544 ASSERT(map_length <= INT_MAX);
8545 atomic_inc(&dip->pending_bios);
8547 clone_len = min_t(int, submit_len, map_length);
8550 * This will never fail as it's passing GPF_NOFS and
8551 * the allocation is backed by btrfs_bioset.
8553 bio = btrfs_bio_clone_partial(orig_bio, clone_offset,
8555 bio->bi_private = dip;
8556 bio->bi_end_io = btrfs_end_dio_bio;
8557 btrfs_io_bio(bio)->logical = file_offset;
8559 ASSERT(submit_len >= clone_len);
8560 submit_len -= clone_len;
8561 if (submit_len == 0)
8565 * Increase the count before we submit the bio so we know
8566 * the end IO handler won't happen before we increase the
8567 * count. Otherwise, the dip might get freed before we're
8568 * done setting it up.
8570 atomic_inc(&dip->pending_bios);
8572 status = btrfs_submit_dio_bio(bio, inode, file_offset,
8576 atomic_dec(&dip->pending_bios);
8580 clone_offset += clone_len;
8581 start_sector += clone_len >> 9;
8582 file_offset += clone_len;
8584 map_length = submit_len;
8585 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio),
8586 start_sector << 9, &map_length, NULL, 0);
8589 } while (submit_len > 0);
8592 status = btrfs_submit_dio_bio(bio, inode, file_offset, async_submit);
8600 * Before atomic variable goto zero, we must make sure dip->errors is
8601 * perceived to be set. This ordering is ensured by the fact that an
8602 * atomic operations with a return value are fully ordered as per
8605 if (atomic_dec_and_test(&dip->pending_bios))
8606 bio_io_error(dip->orig_bio);
8608 /* bio_end_io() will handle error, so we needn't return it */
8612 static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode,
8615 struct btrfs_dio_private *dip = NULL;
8616 struct bio *bio = NULL;
8617 struct btrfs_io_bio *io_bio;
8618 bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
8621 bio = btrfs_bio_clone(dio_bio);
8623 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8629 dip->private = dio_bio->bi_private;
8631 dip->logical_offset = file_offset;
8632 dip->bytes = dio_bio->bi_iter.bi_size;
8633 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8634 bio->bi_private = dip;
8635 dip->orig_bio = bio;
8636 dip->dio_bio = dio_bio;
8637 atomic_set(&dip->pending_bios, 0);
8638 io_bio = btrfs_io_bio(bio);
8639 io_bio->logical = file_offset;
8642 bio->bi_end_io = btrfs_endio_direct_write;
8644 bio->bi_end_io = btrfs_endio_direct_read;
8645 dip->subio_endio = btrfs_subio_endio_read;
8649 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8650 * even if we fail to submit a bio, because in such case we do the
8651 * corresponding error handling below and it must not be done a second
8652 * time by btrfs_direct_IO().
8655 struct btrfs_dio_data *dio_data = current->journal_info;
8657 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8659 dio_data->unsubmitted_oe_range_start =
8660 dio_data->unsubmitted_oe_range_end;
8663 ret = btrfs_submit_direct_hook(dip);
8668 io_bio->end_io(io_bio, ret);
8672 * If we arrived here it means either we failed to submit the dip
8673 * or we either failed to clone the dio_bio or failed to allocate the
8674 * dip. If we cloned the dio_bio and allocated the dip, we can just
8675 * call bio_endio against our io_bio so that we get proper resource
8676 * cleanup if we fail to submit the dip, otherwise, we must do the
8677 * same as btrfs_endio_direct_[write|read] because we can't call these
8678 * callbacks - they require an allocated dip and a clone of dio_bio.
8683 * The end io callbacks free our dip, do the final put on bio
8684 * and all the cleanup and final put for dio_bio (through
8691 __endio_write_update_ordered(inode,
8693 dio_bio->bi_iter.bi_size,
8696 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8697 file_offset + dio_bio->bi_iter.bi_size - 1);
8699 dio_bio->bi_status = BLK_STS_IOERR;
8701 * Releases and cleans up our dio_bio, no need to bio_put()
8702 * nor bio_endio()/bio_io_error() against dio_bio.
8704 dio_end_io(dio_bio);
8711 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
8712 const struct iov_iter *iter, loff_t offset)
8716 unsigned int blocksize_mask = fs_info->sectorsize - 1;
8717 ssize_t retval = -EINVAL;
8719 if (offset & blocksize_mask)
8722 if (iov_iter_alignment(iter) & blocksize_mask)
8725 /* If this is a write we don't need to check anymore */
8726 if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter))
8729 * Check to make sure we don't have duplicate iov_base's in this
8730 * iovec, if so return EINVAL, otherwise we'll get csum errors
8731 * when reading back.
8733 for (seg = 0; seg < iter->nr_segs; seg++) {
8734 for (i = seg + 1; i < iter->nr_segs; i++) {
8735 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8744 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8746 struct file *file = iocb->ki_filp;
8747 struct inode *inode = file->f_mapping->host;
8748 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8749 struct btrfs_dio_data dio_data = { 0 };
8750 struct extent_changeset *data_reserved = NULL;
8751 loff_t offset = iocb->ki_pos;
8755 bool relock = false;
8758 if (check_direct_IO(fs_info, iter, offset))
8761 inode_dio_begin(inode);
8764 * The generic stuff only does filemap_write_and_wait_range, which
8765 * isn't enough if we've written compressed pages to this area, so
8766 * we need to flush the dirty pages again to make absolutely sure
8767 * that any outstanding dirty pages are on disk.
8769 count = iov_iter_count(iter);
8770 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8771 &BTRFS_I(inode)->runtime_flags))
8772 filemap_fdatawrite_range(inode->i_mapping, offset,
8773 offset + count - 1);
8775 if (iov_iter_rw(iter) == WRITE) {
8777 * If the write DIO is beyond the EOF, we need update
8778 * the isize, but it is protected by i_mutex. So we can
8779 * not unlock the i_mutex at this case.
8781 if (offset + count <= inode->i_size) {
8782 dio_data.overwrite = 1;
8783 inode_unlock(inode);
8785 } else if (iocb->ki_flags & IOCB_NOWAIT) {
8789 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
8795 * We need to know how many extents we reserved so that we can
8796 * do the accounting properly if we go over the number we
8797 * originally calculated. Abuse current->journal_info for this.
8799 dio_data.reserve = round_up(count,
8800 fs_info->sectorsize);
8801 dio_data.unsubmitted_oe_range_start = (u64)offset;
8802 dio_data.unsubmitted_oe_range_end = (u64)offset;
8803 current->journal_info = &dio_data;
8804 down_read(&BTRFS_I(inode)->dio_sem);
8805 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8806 &BTRFS_I(inode)->runtime_flags)) {
8807 inode_dio_end(inode);
8808 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8812 ret = __blockdev_direct_IO(iocb, inode,
8813 fs_info->fs_devices->latest_bdev,
8814 iter, btrfs_get_blocks_direct, NULL,
8815 btrfs_submit_direct, flags);
8816 if (iov_iter_rw(iter) == WRITE) {
8817 up_read(&BTRFS_I(inode)->dio_sem);
8818 current->journal_info = NULL;
8819 if (ret < 0 && ret != -EIOCBQUEUED) {
8820 if (dio_data.reserve)
8821 btrfs_delalloc_release_space(inode, data_reserved,
8822 offset, dio_data.reserve, true);
8824 * On error we might have left some ordered extents
8825 * without submitting corresponding bios for them, so
8826 * cleanup them up to avoid other tasks getting them
8827 * and waiting for them to complete forever.
8829 if (dio_data.unsubmitted_oe_range_start <
8830 dio_data.unsubmitted_oe_range_end)
8831 __endio_write_update_ordered(inode,
8832 dio_data.unsubmitted_oe_range_start,
8833 dio_data.unsubmitted_oe_range_end -
8834 dio_data.unsubmitted_oe_range_start,
8836 } else if (ret >= 0 && (size_t)ret < count)
8837 btrfs_delalloc_release_space(inode, data_reserved,
8838 offset, count - (size_t)ret, true);
8839 btrfs_delalloc_release_extents(BTRFS_I(inode), count, false);
8843 inode_dio_end(inode);
8847 extent_changeset_free(data_reserved);
8851 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8853 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8854 __u64 start, __u64 len)
8858 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8862 return extent_fiemap(inode, fieinfo, start, len);
8865 int btrfs_readpage(struct file *file, struct page *page)
8867 struct extent_io_tree *tree;
8868 tree = &BTRFS_I(page->mapping->host)->io_tree;
8869 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8872 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8874 struct inode *inode = page->mapping->host;
8877 if (current->flags & PF_MEMALLOC) {
8878 redirty_page_for_writepage(wbc, page);
8884 * If we are under memory pressure we will call this directly from the
8885 * VM, we need to make sure we have the inode referenced for the ordered
8886 * extent. If not just return like we didn't do anything.
8888 if (!igrab(inode)) {
8889 redirty_page_for_writepage(wbc, page);
8890 return AOP_WRITEPAGE_ACTIVATE;
8892 ret = extent_write_full_page(page, wbc);
8893 btrfs_add_delayed_iput(inode);
8897 static int btrfs_writepages(struct address_space *mapping,
8898 struct writeback_control *wbc)
8900 return extent_writepages(mapping, wbc);
8904 btrfs_readpages(struct file *file, struct address_space *mapping,
8905 struct list_head *pages, unsigned nr_pages)
8907 return extent_readpages(mapping, pages, nr_pages);
8910 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8912 int ret = try_release_extent_mapping(page, gfp_flags);
8914 ClearPagePrivate(page);
8915 set_page_private(page, 0);
8921 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8923 if (PageWriteback(page) || PageDirty(page))
8925 return __btrfs_releasepage(page, gfp_flags);
8928 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8929 unsigned int length)
8931 struct inode *inode = page->mapping->host;
8932 struct extent_io_tree *tree;
8933 struct btrfs_ordered_extent *ordered;
8934 struct extent_state *cached_state = NULL;
8935 u64 page_start = page_offset(page);
8936 u64 page_end = page_start + PAGE_SIZE - 1;
8939 int inode_evicting = inode->i_state & I_FREEING;
8942 * we have the page locked, so new writeback can't start,
8943 * and the dirty bit won't be cleared while we are here.
8945 * Wait for IO on this page so that we can safely clear
8946 * the PagePrivate2 bit and do ordered accounting
8948 wait_on_page_writeback(page);
8950 tree = &BTRFS_I(inode)->io_tree;
8952 btrfs_releasepage(page, GFP_NOFS);
8956 if (!inode_evicting)
8957 lock_extent_bits(tree, page_start, page_end, &cached_state);
8960 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
8961 page_end - start + 1);
8963 end = min(page_end, ordered->file_offset + ordered->len - 1);
8965 * IO on this page will never be started, so we need
8966 * to account for any ordered extents now
8968 if (!inode_evicting)
8969 clear_extent_bit(tree, start, end,
8970 EXTENT_DIRTY | EXTENT_DELALLOC |
8971 EXTENT_DELALLOC_NEW |
8972 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8973 EXTENT_DEFRAG, 1, 0, &cached_state);
8975 * whoever cleared the private bit is responsible
8976 * for the finish_ordered_io
8978 if (TestClearPagePrivate2(page)) {
8979 struct btrfs_ordered_inode_tree *tree;
8982 tree = &BTRFS_I(inode)->ordered_tree;
8984 spin_lock_irq(&tree->lock);
8985 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8986 new_len = start - ordered->file_offset;
8987 if (new_len < ordered->truncated_len)
8988 ordered->truncated_len = new_len;
8989 spin_unlock_irq(&tree->lock);
8991 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8993 end - start + 1, 1))
8994 btrfs_finish_ordered_io(ordered);
8996 btrfs_put_ordered_extent(ordered);
8997 if (!inode_evicting) {
8998 cached_state = NULL;
8999 lock_extent_bits(tree, start, end,
9004 if (start < page_end)
9009 * Qgroup reserved space handler
9010 * Page here will be either
9011 * 1) Already written to disk
9012 * In this case, its reserved space is released from data rsv map
9013 * and will be freed by delayed_ref handler finally.
9014 * So even we call qgroup_free_data(), it won't decrease reserved
9016 * 2) Not written to disk
9017 * This means the reserved space should be freed here. However,
9018 * if a truncate invalidates the page (by clearing PageDirty)
9019 * and the page is accounted for while allocating extent
9020 * in btrfs_check_data_free_space() we let delayed_ref to
9021 * free the entire extent.
9023 if (PageDirty(page))
9024 btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
9025 if (!inode_evicting) {
9026 clear_extent_bit(tree, page_start, page_end,
9027 EXTENT_LOCKED | EXTENT_DIRTY |
9028 EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
9029 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
9032 __btrfs_releasepage(page, GFP_NOFS);
9035 ClearPageChecked(page);
9036 if (PagePrivate(page)) {
9037 ClearPagePrivate(page);
9038 set_page_private(page, 0);
9044 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
9045 * called from a page fault handler when a page is first dirtied. Hence we must
9046 * be careful to check for EOF conditions here. We set the page up correctly
9047 * for a written page which means we get ENOSPC checking when writing into
9048 * holes and correct delalloc and unwritten extent mapping on filesystems that
9049 * support these features.
9051 * We are not allowed to take the i_mutex here so we have to play games to
9052 * protect against truncate races as the page could now be beyond EOF. Because
9053 * vmtruncate() writes the inode size before removing pages, once we have the
9054 * page lock we can determine safely if the page is beyond EOF. If it is not
9055 * beyond EOF, then the page is guaranteed safe against truncation until we
9058 int btrfs_page_mkwrite(struct vm_fault *vmf)
9060 struct page *page = vmf->page;
9061 struct inode *inode = file_inode(vmf->vma->vm_file);
9062 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9063 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
9064 struct btrfs_ordered_extent *ordered;
9065 struct extent_state *cached_state = NULL;
9066 struct extent_changeset *data_reserved = NULL;
9068 unsigned long zero_start;
9077 reserved_space = PAGE_SIZE;
9079 sb_start_pagefault(inode->i_sb);
9080 page_start = page_offset(page);
9081 page_end = page_start + PAGE_SIZE - 1;
9085 * Reserving delalloc space after obtaining the page lock can lead to
9086 * deadlock. For example, if a dirty page is locked by this function
9087 * and the call to btrfs_delalloc_reserve_space() ends up triggering
9088 * dirty page write out, then the btrfs_writepage() function could
9089 * end up waiting indefinitely to get a lock on the page currently
9090 * being processed by btrfs_page_mkwrite() function.
9092 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
9095 ret = file_update_time(vmf->vma->vm_file);
9101 else /* -ENOSPC, -EIO, etc */
9102 ret = VM_FAULT_SIGBUS;
9108 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
9111 size = i_size_read(inode);
9113 if ((page->mapping != inode->i_mapping) ||
9114 (page_start >= size)) {
9115 /* page got truncated out from underneath us */
9118 wait_on_page_writeback(page);
9120 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
9121 set_page_extent_mapped(page);
9124 * we can't set the delalloc bits if there are pending ordered
9125 * extents. Drop our locks and wait for them to finish
9127 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
9130 unlock_extent_cached(io_tree, page_start, page_end,
9133 btrfs_start_ordered_extent(inode, ordered, 1);
9134 btrfs_put_ordered_extent(ordered);
9138 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
9139 reserved_space = round_up(size - page_start,
9140 fs_info->sectorsize);
9141 if (reserved_space < PAGE_SIZE) {
9142 end = page_start + reserved_space - 1;
9143 btrfs_delalloc_release_space(inode, data_reserved,
9144 page_start, PAGE_SIZE - reserved_space,
9150 * page_mkwrite gets called when the page is firstly dirtied after it's
9151 * faulted in, but write(2) could also dirty a page and set delalloc
9152 * bits, thus in this case for space account reason, we still need to
9153 * clear any delalloc bits within this page range since we have to
9154 * reserve data&meta space before lock_page() (see above comments).
9156 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
9157 EXTENT_DIRTY | EXTENT_DELALLOC |
9158 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
9159 0, 0, &cached_state);
9161 ret = btrfs_set_extent_delalloc(inode, page_start, end, 0,
9164 unlock_extent_cached(io_tree, page_start, page_end,
9166 ret = VM_FAULT_SIGBUS;
9171 /* page is wholly or partially inside EOF */
9172 if (page_start + PAGE_SIZE > size)
9173 zero_start = size & ~PAGE_MASK;
9175 zero_start = PAGE_SIZE;
9177 if (zero_start != PAGE_SIZE) {
9179 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
9180 flush_dcache_page(page);
9183 ClearPageChecked(page);
9184 set_page_dirty(page);
9185 SetPageUptodate(page);
9187 BTRFS_I(inode)->last_trans = fs_info->generation;
9188 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
9189 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
9191 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
9195 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, true);
9196 sb_end_pagefault(inode->i_sb);
9197 extent_changeset_free(data_reserved);
9198 return VM_FAULT_LOCKED;
9202 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, (ret != 0));
9203 btrfs_delalloc_release_space(inode, data_reserved, page_start,
9204 reserved_space, (ret != 0));
9206 sb_end_pagefault(inode->i_sb);
9207 extent_changeset_free(data_reserved);
9211 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
9213 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9214 struct btrfs_root *root = BTRFS_I(inode)->root;
9215 struct btrfs_block_rsv *rsv;
9218 struct btrfs_trans_handle *trans;
9219 u64 mask = fs_info->sectorsize - 1;
9220 u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
9222 if (!skip_writeback) {
9223 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
9230 * Yes ladies and gentlemen, this is indeed ugly. The fact is we have
9231 * 3 things going on here
9233 * 1) We need to reserve space for our orphan item and the space to
9234 * delete our orphan item. Lord knows we don't want to have a dangling
9235 * orphan item because we didn't reserve space to remove it.
9237 * 2) We need to reserve space to update our inode.
9239 * 3) We need to have something to cache all the space that is going to
9240 * be free'd up by the truncate operation, but also have some slack
9241 * space reserved in case it uses space during the truncate (thank you
9242 * very much snapshotting).
9244 * And we need these to all be separate. The fact is we can use a lot of
9245 * space doing the truncate, and we have no earthly idea how much space
9246 * we will use, so we need the truncate reservation to be separate so it
9247 * doesn't end up using space reserved for updating the inode or
9248 * removing the orphan item. We also need to be able to stop the
9249 * transaction and start a new one, which means we need to be able to
9250 * update the inode several times, and we have no idea of knowing how
9251 * many times that will be, so we can't just reserve 1 item for the
9252 * entirety of the operation, so that has to be done separately as well.
9253 * Then there is the orphan item, which does indeed need to be held on
9254 * to for the whole operation, and we need nobody to touch this reserved
9255 * space except the orphan code.
9257 * So that leaves us with
9259 * 1) root->orphan_block_rsv - for the orphan deletion.
9260 * 2) rsv - for the truncate reservation, which we will steal from the
9261 * transaction reservation.
9262 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9263 * updating the inode.
9265 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
9268 rsv->size = min_size;
9272 * 1 for the truncate slack space
9273 * 1 for updating the inode.
9275 trans = btrfs_start_transaction(root, 2);
9276 if (IS_ERR(trans)) {
9277 err = PTR_ERR(trans);
9281 /* Migrate the slack space for the truncate to our reserve */
9282 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
9287 * So if we truncate and then write and fsync we normally would just
9288 * write the extents that changed, which is a problem if we need to
9289 * first truncate that entire inode. So set this flag so we write out
9290 * all of the extents in the inode to the sync log so we're completely
9293 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9294 trans->block_rsv = rsv;
9297 ret = btrfs_truncate_inode_items(trans, root, inode,
9299 BTRFS_EXTENT_DATA_KEY);
9300 trans->block_rsv = &fs_info->trans_block_rsv;
9301 if (ret != -ENOSPC && ret != -EAGAIN) {
9307 ret = btrfs_update_inode(trans, root, inode);
9313 btrfs_end_transaction(trans);
9314 btrfs_btree_balance_dirty(fs_info);
9316 trans = btrfs_start_transaction(root, 2);
9317 if (IS_ERR(trans)) {
9318 ret = err = PTR_ERR(trans);
9323 btrfs_block_rsv_release(fs_info, rsv, -1);
9324 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
9326 BUG_ON(ret); /* shouldn't happen */
9327 trans->block_rsv = rsv;
9331 * We can't call btrfs_truncate_block inside a trans handle as we could
9332 * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
9333 * we've truncated everything except the last little bit, and can do
9334 * btrfs_truncate_block and then update the disk_i_size.
9336 if (ret == NEED_TRUNCATE_BLOCK) {
9337 btrfs_end_transaction(trans);
9338 btrfs_btree_balance_dirty(fs_info);
9340 ret = btrfs_truncate_block(inode, inode->i_size, 0, 0);
9343 trans = btrfs_start_transaction(root, 1);
9344 if (IS_ERR(trans)) {
9345 ret = PTR_ERR(trans);
9348 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
9351 if (ret == 0 && inode->i_nlink > 0) {
9352 trans->block_rsv = root->orphan_block_rsv;
9353 ret = btrfs_orphan_del(trans, BTRFS_I(inode));
9359 trans->block_rsv = &fs_info->trans_block_rsv;
9360 ret = btrfs_update_inode(trans, root, inode);
9364 ret = btrfs_end_transaction(trans);
9365 btrfs_btree_balance_dirty(fs_info);
9368 btrfs_free_block_rsv(fs_info, rsv);
9377 * create a new subvolume directory/inode (helper for the ioctl).
9379 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9380 struct btrfs_root *new_root,
9381 struct btrfs_root *parent_root,
9384 struct inode *inode;
9388 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9389 new_dirid, new_dirid,
9390 S_IFDIR | (~current_umask() & S_IRWXUGO),
9393 return PTR_ERR(inode);
9394 inode->i_op = &btrfs_dir_inode_operations;
9395 inode->i_fop = &btrfs_dir_file_operations;
9397 set_nlink(inode, 1);
9398 btrfs_i_size_write(BTRFS_I(inode), 0);
9399 unlock_new_inode(inode);
9401 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9403 btrfs_err(new_root->fs_info,
9404 "error inheriting subvolume %llu properties: %d",
9405 new_root->root_key.objectid, err);
9407 err = btrfs_update_inode(trans, new_root, inode);
9413 struct inode *btrfs_alloc_inode(struct super_block *sb)
9415 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
9416 struct btrfs_inode *ei;
9417 struct inode *inode;
9419 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL);
9426 ei->last_sub_trans = 0;
9427 ei->logged_trans = 0;
9428 ei->delalloc_bytes = 0;
9429 ei->new_delalloc_bytes = 0;
9430 ei->defrag_bytes = 0;
9431 ei->disk_i_size = 0;
9434 ei->index_cnt = (u64)-1;
9436 ei->last_unlink_trans = 0;
9437 ei->last_log_commit = 0;
9439 spin_lock_init(&ei->lock);
9440 ei->outstanding_extents = 0;
9441 if (sb->s_magic != BTRFS_TEST_MAGIC)
9442 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
9443 BTRFS_BLOCK_RSV_DELALLOC);
9444 ei->runtime_flags = 0;
9445 ei->prop_compress = BTRFS_COMPRESS_NONE;
9446 ei->defrag_compress = BTRFS_COMPRESS_NONE;
9448 ei->delayed_node = NULL;
9450 ei->i_otime.tv_sec = 0;
9451 ei->i_otime.tv_nsec = 0;
9453 inode = &ei->vfs_inode;
9454 extent_map_tree_init(&ei->extent_tree);
9455 extent_io_tree_init(&ei->io_tree, inode);
9456 extent_io_tree_init(&ei->io_failure_tree, inode);
9457 ei->io_tree.track_uptodate = 1;
9458 ei->io_failure_tree.track_uptodate = 1;
9459 atomic_set(&ei->sync_writers, 0);
9460 mutex_init(&ei->log_mutex);
9461 mutex_init(&ei->delalloc_mutex);
9462 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9463 INIT_LIST_HEAD(&ei->delalloc_inodes);
9464 INIT_LIST_HEAD(&ei->delayed_iput);
9465 RB_CLEAR_NODE(&ei->rb_node);
9466 init_rwsem(&ei->dio_sem);
9471 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9472 void btrfs_test_destroy_inode(struct inode *inode)
9474 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9475 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9479 static void btrfs_i_callback(struct rcu_head *head)
9481 struct inode *inode = container_of(head, struct inode, i_rcu);
9482 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9485 void btrfs_destroy_inode(struct inode *inode)
9487 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9488 struct btrfs_ordered_extent *ordered;
9489 struct btrfs_root *root = BTRFS_I(inode)->root;
9491 WARN_ON(!hlist_empty(&inode->i_dentry));
9492 WARN_ON(inode->i_data.nrpages);
9493 WARN_ON(BTRFS_I(inode)->block_rsv.reserved);
9494 WARN_ON(BTRFS_I(inode)->block_rsv.size);
9495 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9496 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9497 WARN_ON(BTRFS_I(inode)->new_delalloc_bytes);
9498 WARN_ON(BTRFS_I(inode)->csum_bytes);
9499 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9502 * This can happen where we create an inode, but somebody else also
9503 * created the same inode and we need to destroy the one we already
9509 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9510 &BTRFS_I(inode)->runtime_flags)) {
9511 btrfs_info(fs_info, "inode %llu still on the orphan list",
9512 btrfs_ino(BTRFS_I(inode)));
9513 atomic_dec(&root->orphan_inodes);
9517 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9522 "found ordered extent %llu %llu on inode cleanup",
9523 ordered->file_offset, ordered->len);
9524 btrfs_remove_ordered_extent(inode, ordered);
9525 btrfs_put_ordered_extent(ordered);
9526 btrfs_put_ordered_extent(ordered);
9529 btrfs_qgroup_check_reserved_leak(inode);
9530 inode_tree_del(inode);
9531 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9533 call_rcu(&inode->i_rcu, btrfs_i_callback);
9536 int btrfs_drop_inode(struct inode *inode)
9538 struct btrfs_root *root = BTRFS_I(inode)->root;
9543 /* the snap/subvol tree is on deleting */
9544 if (btrfs_root_refs(&root->root_item) == 0)
9547 return generic_drop_inode(inode);
9550 static void init_once(void *foo)
9552 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9554 inode_init_once(&ei->vfs_inode);
9557 void __cold btrfs_destroy_cachep(void)
9560 * Make sure all delayed rcu free inodes are flushed before we
9564 kmem_cache_destroy(btrfs_inode_cachep);
9565 kmem_cache_destroy(btrfs_trans_handle_cachep);
9566 kmem_cache_destroy(btrfs_path_cachep);
9567 kmem_cache_destroy(btrfs_free_space_cachep);
9570 int __init btrfs_init_cachep(void)
9572 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9573 sizeof(struct btrfs_inode), 0,
9574 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9576 if (!btrfs_inode_cachep)
9579 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9580 sizeof(struct btrfs_trans_handle), 0,
9581 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9582 if (!btrfs_trans_handle_cachep)
9585 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9586 sizeof(struct btrfs_path), 0,
9587 SLAB_MEM_SPREAD, NULL);
9588 if (!btrfs_path_cachep)
9591 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9592 sizeof(struct btrfs_free_space), 0,
9593 SLAB_MEM_SPREAD, NULL);
9594 if (!btrfs_free_space_cachep)
9599 btrfs_destroy_cachep();
9603 static int btrfs_getattr(const struct path *path, struct kstat *stat,
9604 u32 request_mask, unsigned int flags)
9607 struct inode *inode = d_inode(path->dentry);
9608 u32 blocksize = inode->i_sb->s_blocksize;
9609 u32 bi_flags = BTRFS_I(inode)->flags;
9611 stat->result_mask |= STATX_BTIME;
9612 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9613 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9614 if (bi_flags & BTRFS_INODE_APPEND)
9615 stat->attributes |= STATX_ATTR_APPEND;
9616 if (bi_flags & BTRFS_INODE_COMPRESS)
9617 stat->attributes |= STATX_ATTR_COMPRESSED;
9618 if (bi_flags & BTRFS_INODE_IMMUTABLE)
9619 stat->attributes |= STATX_ATTR_IMMUTABLE;
9620 if (bi_flags & BTRFS_INODE_NODUMP)
9621 stat->attributes |= STATX_ATTR_NODUMP;
9623 stat->attributes_mask |= (STATX_ATTR_APPEND |
9624 STATX_ATTR_COMPRESSED |
9625 STATX_ATTR_IMMUTABLE |
9628 generic_fillattr(inode, stat);
9629 stat->dev = BTRFS_I(inode)->root->anon_dev;
9631 spin_lock(&BTRFS_I(inode)->lock);
9632 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9633 spin_unlock(&BTRFS_I(inode)->lock);
9634 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9635 ALIGN(delalloc_bytes, blocksize)) >> 9;
9639 static int btrfs_rename_exchange(struct inode *old_dir,
9640 struct dentry *old_dentry,
9641 struct inode *new_dir,
9642 struct dentry *new_dentry)
9644 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9645 struct btrfs_trans_handle *trans;
9646 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9647 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9648 struct inode *new_inode = new_dentry->d_inode;
9649 struct inode *old_inode = old_dentry->d_inode;
9650 struct timespec ctime = current_time(old_inode);
9651 struct dentry *parent;
9652 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9653 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9658 bool root_log_pinned = false;
9659 bool dest_log_pinned = false;
9661 /* we only allow rename subvolume link between subvolumes */
9662 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9665 /* close the race window with snapshot create/destroy ioctl */
9666 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9667 down_read(&fs_info->subvol_sem);
9668 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9669 down_read(&fs_info->subvol_sem);
9672 * We want to reserve the absolute worst case amount of items. So if
9673 * both inodes are subvols and we need to unlink them then that would
9674 * require 4 item modifications, but if they are both normal inodes it
9675 * would require 5 item modifications, so we'll assume their normal
9676 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9677 * should cover the worst case number of items we'll modify.
9679 trans = btrfs_start_transaction(root, 12);
9680 if (IS_ERR(trans)) {
9681 ret = PTR_ERR(trans);
9686 * We need to find a free sequence number both in the source and
9687 * in the destination directory for the exchange.
9689 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9692 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9696 BTRFS_I(old_inode)->dir_index = 0ULL;
9697 BTRFS_I(new_inode)->dir_index = 0ULL;
9699 /* Reference for the source. */
9700 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9701 /* force full log commit if subvolume involved. */
9702 btrfs_set_log_full_commit(fs_info, trans);
9704 btrfs_pin_log_trans(root);
9705 root_log_pinned = true;
9706 ret = btrfs_insert_inode_ref(trans, dest,
9707 new_dentry->d_name.name,
9708 new_dentry->d_name.len,
9710 btrfs_ino(BTRFS_I(new_dir)),
9716 /* And now for the dest. */
9717 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9718 /* force full log commit if subvolume involved. */
9719 btrfs_set_log_full_commit(fs_info, trans);
9721 btrfs_pin_log_trans(dest);
9722 dest_log_pinned = true;
9723 ret = btrfs_insert_inode_ref(trans, root,
9724 old_dentry->d_name.name,
9725 old_dentry->d_name.len,
9727 btrfs_ino(BTRFS_I(old_dir)),
9733 /* Update inode version and ctime/mtime. */
9734 inode_inc_iversion(old_dir);
9735 inode_inc_iversion(new_dir);
9736 inode_inc_iversion(old_inode);
9737 inode_inc_iversion(new_inode);
9738 old_dir->i_ctime = old_dir->i_mtime = ctime;
9739 new_dir->i_ctime = new_dir->i_mtime = ctime;
9740 old_inode->i_ctime = ctime;
9741 new_inode->i_ctime = ctime;
9743 if (old_dentry->d_parent != new_dentry->d_parent) {
9744 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9745 BTRFS_I(old_inode), 1);
9746 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9747 BTRFS_I(new_inode), 1);
9750 /* src is a subvolume */
9751 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9752 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9753 ret = btrfs_unlink_subvol(trans, root, old_dir,
9755 old_dentry->d_name.name,
9756 old_dentry->d_name.len);
9757 } else { /* src is an inode */
9758 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9759 BTRFS_I(old_dentry->d_inode),
9760 old_dentry->d_name.name,
9761 old_dentry->d_name.len);
9763 ret = btrfs_update_inode(trans, root, old_inode);
9766 btrfs_abort_transaction(trans, ret);
9770 /* dest is a subvolume */
9771 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9772 root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9773 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9775 new_dentry->d_name.name,
9776 new_dentry->d_name.len);
9777 } else { /* dest is an inode */
9778 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9779 BTRFS_I(new_dentry->d_inode),
9780 new_dentry->d_name.name,
9781 new_dentry->d_name.len);
9783 ret = btrfs_update_inode(trans, dest, new_inode);
9786 btrfs_abort_transaction(trans, ret);
9790 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9791 new_dentry->d_name.name,
9792 new_dentry->d_name.len, 0, old_idx);
9794 btrfs_abort_transaction(trans, ret);
9798 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9799 old_dentry->d_name.name,
9800 old_dentry->d_name.len, 0, new_idx);
9802 btrfs_abort_transaction(trans, ret);
9806 if (old_inode->i_nlink == 1)
9807 BTRFS_I(old_inode)->dir_index = old_idx;
9808 if (new_inode->i_nlink == 1)
9809 BTRFS_I(new_inode)->dir_index = new_idx;
9811 if (root_log_pinned) {
9812 parent = new_dentry->d_parent;
9813 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9815 btrfs_end_log_trans(root);
9816 root_log_pinned = false;
9818 if (dest_log_pinned) {
9819 parent = old_dentry->d_parent;
9820 btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir),
9822 btrfs_end_log_trans(dest);
9823 dest_log_pinned = false;
9827 * If we have pinned a log and an error happened, we unpin tasks
9828 * trying to sync the log and force them to fallback to a transaction
9829 * commit if the log currently contains any of the inodes involved in
9830 * this rename operation (to ensure we do not persist a log with an
9831 * inconsistent state for any of these inodes or leading to any
9832 * inconsistencies when replayed). If the transaction was aborted, the
9833 * abortion reason is propagated to userspace when attempting to commit
9834 * the transaction. If the log does not contain any of these inodes, we
9835 * allow the tasks to sync it.
9837 if (ret && (root_log_pinned || dest_log_pinned)) {
9838 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9839 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9840 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9842 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9843 btrfs_set_log_full_commit(fs_info, trans);
9845 if (root_log_pinned) {
9846 btrfs_end_log_trans(root);
9847 root_log_pinned = false;
9849 if (dest_log_pinned) {
9850 btrfs_end_log_trans(dest);
9851 dest_log_pinned = false;
9854 ret = btrfs_end_transaction(trans);
9856 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9857 up_read(&fs_info->subvol_sem);
9858 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9859 up_read(&fs_info->subvol_sem);
9864 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9865 struct btrfs_root *root,
9867 struct dentry *dentry)
9870 struct inode *inode;
9874 ret = btrfs_find_free_ino(root, &objectid);
9878 inode = btrfs_new_inode(trans, root, dir,
9879 dentry->d_name.name,
9881 btrfs_ino(BTRFS_I(dir)),
9883 S_IFCHR | WHITEOUT_MODE,
9886 if (IS_ERR(inode)) {
9887 ret = PTR_ERR(inode);
9891 inode->i_op = &btrfs_special_inode_operations;
9892 init_special_inode(inode, inode->i_mode,
9895 ret = btrfs_init_inode_security(trans, inode, dir,
9900 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9901 BTRFS_I(inode), 0, index);
9905 ret = btrfs_update_inode(trans, root, inode);
9907 unlock_new_inode(inode);
9909 inode_dec_link_count(inode);
9915 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9916 struct inode *new_dir, struct dentry *new_dentry,
9919 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9920 struct btrfs_trans_handle *trans;
9921 unsigned int trans_num_items;
9922 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9923 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9924 struct inode *new_inode = d_inode(new_dentry);
9925 struct inode *old_inode = d_inode(old_dentry);
9929 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9930 bool log_pinned = false;
9932 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9935 /* we only allow rename subvolume link between subvolumes */
9936 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9939 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9940 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9943 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9944 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9948 /* check for collisions, even if the name isn't there */
9949 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9950 new_dentry->d_name.name,
9951 new_dentry->d_name.len);
9954 if (ret == -EEXIST) {
9956 * eexist without a new_inode */
9957 if (WARN_ON(!new_inode)) {
9961 /* maybe -EOVERFLOW */
9968 * we're using rename to replace one file with another. Start IO on it
9969 * now so we don't add too much work to the end of the transaction
9971 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9972 filemap_flush(old_inode->i_mapping);
9974 /* close the racy window with snapshot create/destroy ioctl */
9975 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9976 down_read(&fs_info->subvol_sem);
9978 * We want to reserve the absolute worst case amount of items. So if
9979 * both inodes are subvols and we need to unlink them then that would
9980 * require 4 item modifications, but if they are both normal inodes it
9981 * would require 5 item modifications, so we'll assume they are normal
9982 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9983 * should cover the worst case number of items we'll modify.
9984 * If our rename has the whiteout flag, we need more 5 units for the
9985 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9986 * when selinux is enabled).
9988 trans_num_items = 11;
9989 if (flags & RENAME_WHITEOUT)
9990 trans_num_items += 5;
9991 trans = btrfs_start_transaction(root, trans_num_items);
9992 if (IS_ERR(trans)) {
9993 ret = PTR_ERR(trans);
9998 btrfs_record_root_in_trans(trans, dest);
10000 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
10004 BTRFS_I(old_inode)->dir_index = 0ULL;
10005 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
10006 /* force full log commit if subvolume involved. */
10007 btrfs_set_log_full_commit(fs_info, trans);
10009 btrfs_pin_log_trans(root);
10011 ret = btrfs_insert_inode_ref(trans, dest,
10012 new_dentry->d_name.name,
10013 new_dentry->d_name.len,
10015 btrfs_ino(BTRFS_I(new_dir)), index);
10020 inode_inc_iversion(old_dir);
10021 inode_inc_iversion(new_dir);
10022 inode_inc_iversion(old_inode);
10023 old_dir->i_ctime = old_dir->i_mtime =
10024 new_dir->i_ctime = new_dir->i_mtime =
10025 old_inode->i_ctime = current_time(old_dir);
10027 if (old_dentry->d_parent != new_dentry->d_parent)
10028 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
10029 BTRFS_I(old_inode), 1);
10031 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
10032 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
10033 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
10034 old_dentry->d_name.name,
10035 old_dentry->d_name.len);
10037 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
10038 BTRFS_I(d_inode(old_dentry)),
10039 old_dentry->d_name.name,
10040 old_dentry->d_name.len);
10042 ret = btrfs_update_inode(trans, root, old_inode);
10045 btrfs_abort_transaction(trans, ret);
10050 inode_inc_iversion(new_inode);
10051 new_inode->i_ctime = current_time(new_inode);
10052 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
10053 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
10054 root_objectid = BTRFS_I(new_inode)->location.objectid;
10055 ret = btrfs_unlink_subvol(trans, dest, new_dir,
10057 new_dentry->d_name.name,
10058 new_dentry->d_name.len);
10059 BUG_ON(new_inode->i_nlink == 0);
10061 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
10062 BTRFS_I(d_inode(new_dentry)),
10063 new_dentry->d_name.name,
10064 new_dentry->d_name.len);
10066 if (!ret && new_inode->i_nlink == 0)
10067 ret = btrfs_orphan_add(trans,
10068 BTRFS_I(d_inode(new_dentry)));
10070 btrfs_abort_transaction(trans, ret);
10075 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
10076 new_dentry->d_name.name,
10077 new_dentry->d_name.len, 0, index);
10079 btrfs_abort_transaction(trans, ret);
10083 if (old_inode->i_nlink == 1)
10084 BTRFS_I(old_inode)->dir_index = index;
10087 struct dentry *parent = new_dentry->d_parent;
10089 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
10091 btrfs_end_log_trans(root);
10092 log_pinned = false;
10095 if (flags & RENAME_WHITEOUT) {
10096 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
10100 btrfs_abort_transaction(trans, ret);
10106 * If we have pinned the log and an error happened, we unpin tasks
10107 * trying to sync the log and force them to fallback to a transaction
10108 * commit if the log currently contains any of the inodes involved in
10109 * this rename operation (to ensure we do not persist a log with an
10110 * inconsistent state for any of these inodes or leading to any
10111 * inconsistencies when replayed). If the transaction was aborted, the
10112 * abortion reason is propagated to userspace when attempting to commit
10113 * the transaction. If the log does not contain any of these inodes, we
10114 * allow the tasks to sync it.
10116 if (ret && log_pinned) {
10117 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
10118 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
10119 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
10121 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
10122 btrfs_set_log_full_commit(fs_info, trans);
10124 btrfs_end_log_trans(root);
10125 log_pinned = false;
10127 btrfs_end_transaction(trans);
10129 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
10130 up_read(&fs_info->subvol_sem);
10135 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
10136 struct inode *new_dir, struct dentry *new_dentry,
10137 unsigned int flags)
10139 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
10142 if (flags & RENAME_EXCHANGE)
10143 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
10146 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
10149 static void btrfs_run_delalloc_work(struct btrfs_work *work)
10151 struct btrfs_delalloc_work *delalloc_work;
10152 struct inode *inode;
10154 delalloc_work = container_of(work, struct btrfs_delalloc_work,
10156 inode = delalloc_work->inode;
10157 filemap_flush(inode->i_mapping);
10158 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
10159 &BTRFS_I(inode)->runtime_flags))
10160 filemap_flush(inode->i_mapping);
10162 if (delalloc_work->delay_iput)
10163 btrfs_add_delayed_iput(inode);
10166 complete(&delalloc_work->completion);
10169 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
10172 struct btrfs_delalloc_work *work;
10174 work = kmalloc(sizeof(*work), GFP_NOFS);
10178 init_completion(&work->completion);
10179 INIT_LIST_HEAD(&work->list);
10180 work->inode = inode;
10181 work->delay_iput = delay_iput;
10182 WARN_ON_ONCE(!inode);
10183 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
10184 btrfs_run_delalloc_work, NULL, NULL);
10190 * some fairly slow code that needs optimization. This walks the list
10191 * of all the inodes with pending delalloc and forces them to disk.
10193 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
10196 struct btrfs_inode *binode;
10197 struct inode *inode;
10198 struct btrfs_delalloc_work *work, *next;
10199 struct list_head works;
10200 struct list_head splice;
10203 INIT_LIST_HEAD(&works);
10204 INIT_LIST_HEAD(&splice);
10206 mutex_lock(&root->delalloc_mutex);
10207 spin_lock(&root->delalloc_lock);
10208 list_splice_init(&root->delalloc_inodes, &splice);
10209 while (!list_empty(&splice)) {
10210 binode = list_entry(splice.next, struct btrfs_inode,
10213 list_move_tail(&binode->delalloc_inodes,
10214 &root->delalloc_inodes);
10215 inode = igrab(&binode->vfs_inode);
10217 cond_resched_lock(&root->delalloc_lock);
10220 spin_unlock(&root->delalloc_lock);
10222 work = btrfs_alloc_delalloc_work(inode, delay_iput);
10225 btrfs_add_delayed_iput(inode);
10231 list_add_tail(&work->list, &works);
10232 btrfs_queue_work(root->fs_info->flush_workers,
10235 if (nr != -1 && ret >= nr)
10238 spin_lock(&root->delalloc_lock);
10240 spin_unlock(&root->delalloc_lock);
10243 list_for_each_entry_safe(work, next, &works, list) {
10244 list_del_init(&work->list);
10245 wait_for_completion(&work->completion);
10249 if (!list_empty(&splice)) {
10250 spin_lock(&root->delalloc_lock);
10251 list_splice_tail(&splice, &root->delalloc_inodes);
10252 spin_unlock(&root->delalloc_lock);
10254 mutex_unlock(&root->delalloc_mutex);
10258 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
10260 struct btrfs_fs_info *fs_info = root->fs_info;
10263 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10266 ret = __start_delalloc_inodes(root, 0, -1);
10272 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int nr)
10274 struct btrfs_root *root;
10275 struct list_head splice;
10278 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10281 INIT_LIST_HEAD(&splice);
10283 mutex_lock(&fs_info->delalloc_root_mutex);
10284 spin_lock(&fs_info->delalloc_root_lock);
10285 list_splice_init(&fs_info->delalloc_roots, &splice);
10286 while (!list_empty(&splice) && nr) {
10287 root = list_first_entry(&splice, struct btrfs_root,
10289 root = btrfs_grab_fs_root(root);
10291 list_move_tail(&root->delalloc_root,
10292 &fs_info->delalloc_roots);
10293 spin_unlock(&fs_info->delalloc_root_lock);
10295 ret = __start_delalloc_inodes(root, 0, nr);
10296 btrfs_put_fs_root(root);
10304 spin_lock(&fs_info->delalloc_root_lock);
10306 spin_unlock(&fs_info->delalloc_root_lock);
10310 if (!list_empty(&splice)) {
10311 spin_lock(&fs_info->delalloc_root_lock);
10312 list_splice_tail(&splice, &fs_info->delalloc_roots);
10313 spin_unlock(&fs_info->delalloc_root_lock);
10315 mutex_unlock(&fs_info->delalloc_root_mutex);
10319 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10320 const char *symname)
10322 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10323 struct btrfs_trans_handle *trans;
10324 struct btrfs_root *root = BTRFS_I(dir)->root;
10325 struct btrfs_path *path;
10326 struct btrfs_key key;
10327 struct inode *inode = NULL;
10329 int drop_inode = 0;
10335 struct btrfs_file_extent_item *ei;
10336 struct extent_buffer *leaf;
10338 name_len = strlen(symname);
10339 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
10340 return -ENAMETOOLONG;
10343 * 2 items for inode item and ref
10344 * 2 items for dir items
10345 * 1 item for updating parent inode item
10346 * 1 item for the inline extent item
10347 * 1 item for xattr if selinux is on
10349 trans = btrfs_start_transaction(root, 7);
10351 return PTR_ERR(trans);
10353 err = btrfs_find_free_ino(root, &objectid);
10357 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10358 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
10359 objectid, S_IFLNK|S_IRWXUGO, &index);
10360 if (IS_ERR(inode)) {
10361 err = PTR_ERR(inode);
10366 * If the active LSM wants to access the inode during
10367 * d_instantiate it needs these. Smack checks to see
10368 * if the filesystem supports xattrs by looking at the
10371 inode->i_fop = &btrfs_file_operations;
10372 inode->i_op = &btrfs_file_inode_operations;
10373 inode->i_mapping->a_ops = &btrfs_aops;
10374 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10376 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10378 goto out_unlock_inode;
10380 path = btrfs_alloc_path();
10383 goto out_unlock_inode;
10385 key.objectid = btrfs_ino(BTRFS_I(inode));
10387 key.type = BTRFS_EXTENT_DATA_KEY;
10388 datasize = btrfs_file_extent_calc_inline_size(name_len);
10389 err = btrfs_insert_empty_item(trans, root, path, &key,
10392 btrfs_free_path(path);
10393 goto out_unlock_inode;
10395 leaf = path->nodes[0];
10396 ei = btrfs_item_ptr(leaf, path->slots[0],
10397 struct btrfs_file_extent_item);
10398 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10399 btrfs_set_file_extent_type(leaf, ei,
10400 BTRFS_FILE_EXTENT_INLINE);
10401 btrfs_set_file_extent_encryption(leaf, ei, 0);
10402 btrfs_set_file_extent_compression(leaf, ei, 0);
10403 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10404 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10406 ptr = btrfs_file_extent_inline_start(ei);
10407 write_extent_buffer(leaf, symname, ptr, name_len);
10408 btrfs_mark_buffer_dirty(leaf);
10409 btrfs_free_path(path);
10411 inode->i_op = &btrfs_symlink_inode_operations;
10412 inode_nohighmem(inode);
10413 inode->i_mapping->a_ops = &btrfs_symlink_aops;
10414 inode_set_bytes(inode, name_len);
10415 btrfs_i_size_write(BTRFS_I(inode), name_len);
10416 err = btrfs_update_inode(trans, root, inode);
10418 * Last step, add directory indexes for our symlink inode. This is the
10419 * last step to avoid extra cleanup of these indexes if an error happens
10423 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
10424 BTRFS_I(inode), 0, index);
10427 goto out_unlock_inode;
10430 d_instantiate_new(dentry, inode);
10433 btrfs_end_transaction(trans);
10435 inode_dec_link_count(inode);
10438 btrfs_btree_balance_dirty(fs_info);
10443 unlock_new_inode(inode);
10447 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10448 u64 start, u64 num_bytes, u64 min_size,
10449 loff_t actual_len, u64 *alloc_hint,
10450 struct btrfs_trans_handle *trans)
10452 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
10453 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10454 struct extent_map *em;
10455 struct btrfs_root *root = BTRFS_I(inode)->root;
10456 struct btrfs_key ins;
10457 u64 cur_offset = start;
10460 u64 last_alloc = (u64)-1;
10462 bool own_trans = true;
10463 u64 end = start + num_bytes - 1;
10467 while (num_bytes > 0) {
10469 trans = btrfs_start_transaction(root, 3);
10470 if (IS_ERR(trans)) {
10471 ret = PTR_ERR(trans);
10476 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10477 cur_bytes = max(cur_bytes, min_size);
10479 * If we are severely fragmented we could end up with really
10480 * small allocations, so if the allocator is returning small
10481 * chunks lets make its job easier by only searching for those
10484 cur_bytes = min(cur_bytes, last_alloc);
10485 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10486 min_size, 0, *alloc_hint, &ins, 1, 0);
10489 btrfs_end_transaction(trans);
10492 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10494 last_alloc = ins.offset;
10495 ret = insert_reserved_file_extent(trans, inode,
10496 cur_offset, ins.objectid,
10497 ins.offset, ins.offset,
10498 ins.offset, 0, 0, 0,
10499 BTRFS_FILE_EXTENT_PREALLOC);
10501 btrfs_free_reserved_extent(fs_info, ins.objectid,
10503 btrfs_abort_transaction(trans, ret);
10505 btrfs_end_transaction(trans);
10509 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10510 cur_offset + ins.offset -1, 0);
10512 em = alloc_extent_map();
10514 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10515 &BTRFS_I(inode)->runtime_flags);
10519 em->start = cur_offset;
10520 em->orig_start = cur_offset;
10521 em->len = ins.offset;
10522 em->block_start = ins.objectid;
10523 em->block_len = ins.offset;
10524 em->orig_block_len = ins.offset;
10525 em->ram_bytes = ins.offset;
10526 em->bdev = fs_info->fs_devices->latest_bdev;
10527 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10528 em->generation = trans->transid;
10531 write_lock(&em_tree->lock);
10532 ret = add_extent_mapping(em_tree, em, 1);
10533 write_unlock(&em_tree->lock);
10534 if (ret != -EEXIST)
10536 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10537 cur_offset + ins.offset - 1,
10540 free_extent_map(em);
10542 num_bytes -= ins.offset;
10543 cur_offset += ins.offset;
10544 *alloc_hint = ins.objectid + ins.offset;
10546 inode_inc_iversion(inode);
10547 inode->i_ctime = current_time(inode);
10548 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10549 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10550 (actual_len > inode->i_size) &&
10551 (cur_offset > inode->i_size)) {
10552 if (cur_offset > actual_len)
10553 i_size = actual_len;
10555 i_size = cur_offset;
10556 i_size_write(inode, i_size);
10557 btrfs_ordered_update_i_size(inode, i_size, NULL);
10560 ret = btrfs_update_inode(trans, root, inode);
10563 btrfs_abort_transaction(trans, ret);
10565 btrfs_end_transaction(trans);
10570 btrfs_end_transaction(trans);
10572 if (cur_offset < end)
10573 btrfs_free_reserved_data_space(inode, NULL, cur_offset,
10574 end - cur_offset + 1);
10578 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10579 u64 start, u64 num_bytes, u64 min_size,
10580 loff_t actual_len, u64 *alloc_hint)
10582 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10583 min_size, actual_len, alloc_hint,
10587 int btrfs_prealloc_file_range_trans(struct inode *inode,
10588 struct btrfs_trans_handle *trans, int mode,
10589 u64 start, u64 num_bytes, u64 min_size,
10590 loff_t actual_len, u64 *alloc_hint)
10592 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10593 min_size, actual_len, alloc_hint, trans);
10596 static int btrfs_set_page_dirty(struct page *page)
10598 return __set_page_dirty_nobuffers(page);
10601 static int btrfs_permission(struct inode *inode, int mask)
10603 struct btrfs_root *root = BTRFS_I(inode)->root;
10604 umode_t mode = inode->i_mode;
10606 if (mask & MAY_WRITE &&
10607 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10608 if (btrfs_root_readonly(root))
10610 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10613 return generic_permission(inode, mask);
10616 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10618 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10619 struct btrfs_trans_handle *trans;
10620 struct btrfs_root *root = BTRFS_I(dir)->root;
10621 struct inode *inode = NULL;
10627 * 5 units required for adding orphan entry
10629 trans = btrfs_start_transaction(root, 5);
10631 return PTR_ERR(trans);
10633 ret = btrfs_find_free_ino(root, &objectid);
10637 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10638 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10639 if (IS_ERR(inode)) {
10640 ret = PTR_ERR(inode);
10645 inode->i_fop = &btrfs_file_operations;
10646 inode->i_op = &btrfs_file_inode_operations;
10648 inode->i_mapping->a_ops = &btrfs_aops;
10649 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10651 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10655 ret = btrfs_update_inode(trans, root, inode);
10658 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10663 * We set number of links to 0 in btrfs_new_inode(), and here we set
10664 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10667 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10669 set_nlink(inode, 1);
10670 unlock_new_inode(inode);
10671 d_tmpfile(dentry, inode);
10672 mark_inode_dirty(inode);
10675 btrfs_end_transaction(trans);
10678 btrfs_btree_balance_dirty(fs_info);
10682 unlock_new_inode(inode);
10687 __attribute__((const))
10688 static int btrfs_readpage_io_failed_hook(struct page *page, int failed_mirror)
10693 static struct btrfs_fs_info *iotree_fs_info(void *private_data)
10695 struct inode *inode = private_data;
10696 return btrfs_sb(inode->i_sb);
10699 static void btrfs_check_extent_io_range(void *private_data, const char *caller,
10700 u64 start, u64 end)
10702 struct inode *inode = private_data;
10705 isize = i_size_read(inode);
10706 if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
10707 btrfs_debug_rl(BTRFS_I(inode)->root->fs_info,
10708 "%s: ino %llu isize %llu odd range [%llu,%llu]",
10709 caller, btrfs_ino(BTRFS_I(inode)), isize, start, end);
10713 void btrfs_set_range_writeback(void *private_data, u64 start, u64 end)
10715 struct inode *inode = private_data;
10716 unsigned long index = start >> PAGE_SHIFT;
10717 unsigned long end_index = end >> PAGE_SHIFT;
10720 while (index <= end_index) {
10721 page = find_get_page(inode->i_mapping, index);
10722 ASSERT(page); /* Pages should be in the extent_io_tree */
10723 set_page_writeback(page);
10729 static const struct inode_operations btrfs_dir_inode_operations = {
10730 .getattr = btrfs_getattr,
10731 .lookup = btrfs_lookup,
10732 .create = btrfs_create,
10733 .unlink = btrfs_unlink,
10734 .link = btrfs_link,
10735 .mkdir = btrfs_mkdir,
10736 .rmdir = btrfs_rmdir,
10737 .rename = btrfs_rename2,
10738 .symlink = btrfs_symlink,
10739 .setattr = btrfs_setattr,
10740 .mknod = btrfs_mknod,
10741 .listxattr = btrfs_listxattr,
10742 .permission = btrfs_permission,
10743 .get_acl = btrfs_get_acl,
10744 .set_acl = btrfs_set_acl,
10745 .update_time = btrfs_update_time,
10746 .tmpfile = btrfs_tmpfile,
10748 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10749 .lookup = btrfs_lookup,
10750 .permission = btrfs_permission,
10751 .update_time = btrfs_update_time,
10754 static const struct file_operations btrfs_dir_file_operations = {
10755 .llseek = generic_file_llseek,
10756 .read = generic_read_dir,
10757 .iterate_shared = btrfs_real_readdir,
10758 .open = btrfs_opendir,
10759 .unlocked_ioctl = btrfs_ioctl,
10760 #ifdef CONFIG_COMPAT
10761 .compat_ioctl = btrfs_compat_ioctl,
10763 .release = btrfs_release_file,
10764 .fsync = btrfs_sync_file,
10767 static const struct extent_io_ops btrfs_extent_io_ops = {
10768 /* mandatory callbacks */
10769 .submit_bio_hook = btrfs_submit_bio_hook,
10770 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10771 .merge_bio_hook = btrfs_merge_bio_hook,
10772 .readpage_io_failed_hook = btrfs_readpage_io_failed_hook,
10773 .tree_fs_info = iotree_fs_info,
10774 .set_range_writeback = btrfs_set_range_writeback,
10776 /* optional callbacks */
10777 .fill_delalloc = run_delalloc_range,
10778 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10779 .writepage_start_hook = btrfs_writepage_start_hook,
10780 .set_bit_hook = btrfs_set_bit_hook,
10781 .clear_bit_hook = btrfs_clear_bit_hook,
10782 .merge_extent_hook = btrfs_merge_extent_hook,
10783 .split_extent_hook = btrfs_split_extent_hook,
10784 .check_extent_io_range = btrfs_check_extent_io_range,
10788 * btrfs doesn't support the bmap operation because swapfiles
10789 * use bmap to make a mapping of extents in the file. They assume
10790 * these extents won't change over the life of the file and they
10791 * use the bmap result to do IO directly to the drive.
10793 * the btrfs bmap call would return logical addresses that aren't
10794 * suitable for IO and they also will change frequently as COW
10795 * operations happen. So, swapfile + btrfs == corruption.
10797 * For now we're avoiding this by dropping bmap.
10799 static const struct address_space_operations btrfs_aops = {
10800 .readpage = btrfs_readpage,
10801 .writepage = btrfs_writepage,
10802 .writepages = btrfs_writepages,
10803 .readpages = btrfs_readpages,
10804 .direct_IO = btrfs_direct_IO,
10805 .invalidatepage = btrfs_invalidatepage,
10806 .releasepage = btrfs_releasepage,
10807 .set_page_dirty = btrfs_set_page_dirty,
10808 .error_remove_page = generic_error_remove_page,
10811 static const struct address_space_operations btrfs_symlink_aops = {
10812 .readpage = btrfs_readpage,
10813 .writepage = btrfs_writepage,
10814 .invalidatepage = btrfs_invalidatepage,
10815 .releasepage = btrfs_releasepage,
10818 static const struct inode_operations btrfs_file_inode_operations = {
10819 .getattr = btrfs_getattr,
10820 .setattr = btrfs_setattr,
10821 .listxattr = btrfs_listxattr,
10822 .permission = btrfs_permission,
10823 .fiemap = btrfs_fiemap,
10824 .get_acl = btrfs_get_acl,
10825 .set_acl = btrfs_set_acl,
10826 .update_time = btrfs_update_time,
10828 static const struct inode_operations btrfs_special_inode_operations = {
10829 .getattr = btrfs_getattr,
10830 .setattr = btrfs_setattr,
10831 .permission = btrfs_permission,
10832 .listxattr = btrfs_listxattr,
10833 .get_acl = btrfs_get_acl,
10834 .set_acl = btrfs_set_acl,
10835 .update_time = btrfs_update_time,
10837 static const struct inode_operations btrfs_symlink_inode_operations = {
10838 .get_link = page_get_link,
10839 .getattr = btrfs_getattr,
10840 .setattr = btrfs_setattr,
10841 .permission = btrfs_permission,
10842 .listxattr = btrfs_listxattr,
10843 .update_time = btrfs_update_time,
10846 const struct dentry_operations btrfs_dentry_operations = {
10847 .d_delete = btrfs_dentry_delete,
10848 .d_release = btrfs_dentry_release,