1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <linux/kernel.h>
8 #include <linux/buffer_head.h>
9 #include <linux/file.h>
11 #include <linux/pagemap.h>
12 #include <linux/highmem.h>
13 #include <linux/time.h>
14 #include <linux/init.h>
15 #include <linux/string.h>
16 #include <linux/backing-dev.h>
17 #include <linux/writeback.h>
18 #include <linux/compat.h>
19 #include <linux/xattr.h>
20 #include <linux/posix_acl.h>
21 #include <linux/falloc.h>
22 #include <linux/slab.h>
23 #include <linux/ratelimit.h>
24 #include <linux/btrfs.h>
25 #include <linux/blkdev.h>
26 #include <linux/posix_acl_xattr.h>
27 #include <linux/uio.h>
28 #include <linux/magic.h>
29 #include <linux/iversion.h>
30 #include <linux/swap.h>
31 #include <linux/migrate.h>
32 #include <linux/sched/mm.h>
33 #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"
50 #include "delalloc-space.h"
51 #include "block-group.h"
52 #include "space-info.h"
54 struct btrfs_iget_args {
56 struct btrfs_root *root;
59 struct btrfs_dio_data {
61 u64 unsubmitted_oe_range_start;
62 u64 unsubmitted_oe_range_end;
66 static const struct inode_operations btrfs_dir_inode_operations;
67 static const struct inode_operations btrfs_symlink_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 file_operations btrfs_dir_file_operations;
72 static const struct extent_io_ops btrfs_extent_io_ops;
74 static struct kmem_cache *btrfs_inode_cachep;
75 struct kmem_cache *btrfs_trans_handle_cachep;
76 struct kmem_cache *btrfs_path_cachep;
77 struct kmem_cache *btrfs_free_space_cachep;
78 struct kmem_cache *btrfs_free_space_bitmap_cachep;
80 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
81 static int btrfs_truncate(struct inode *inode, bool skip_writeback);
82 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
83 static noinline int cow_file_range(struct inode *inode,
84 struct page *locked_page,
85 u64 start, u64 end, int *page_started,
86 unsigned long *nr_written, int unlock);
87 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
88 u64 orig_start, u64 block_start,
89 u64 block_len, u64 orig_block_len,
90 u64 ram_bytes, int compress_type,
93 static void __endio_write_update_ordered(struct inode *inode,
94 const u64 offset, const u64 bytes,
98 * Cleanup all submitted ordered extents in specified range to handle errors
99 * from the btrfs_run_delalloc_range() callback.
101 * NOTE: caller must ensure that when an error happens, it can not call
102 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
103 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
104 * to be released, which we want to happen only when finishing the ordered
105 * extent (btrfs_finish_ordered_io()).
107 static inline void btrfs_cleanup_ordered_extents(struct inode *inode,
108 struct page *locked_page,
109 u64 offset, u64 bytes)
111 unsigned long index = offset >> PAGE_SHIFT;
112 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
113 u64 page_start = page_offset(locked_page);
114 u64 page_end = page_start + PAGE_SIZE - 1;
118 while (index <= end_index) {
119 page = find_get_page(inode->i_mapping, index);
123 ClearPagePrivate2(page);
128 * In case this page belongs to the delalloc range being instantiated
129 * then skip it, since the first page of a range is going to be
130 * properly cleaned up by the caller of run_delalloc_range
132 if (page_start >= offset && page_end <= (offset + bytes - 1)) {
137 return __endio_write_update_ordered(inode, offset, bytes, false);
140 static int btrfs_dirty_inode(struct inode *inode);
142 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
143 void btrfs_test_inode_set_ops(struct inode *inode)
145 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
149 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
150 struct inode *inode, struct inode *dir,
151 const struct qstr *qstr)
155 err = btrfs_init_acl(trans, inode, dir);
157 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
162 * this does all the hard work for inserting an inline extent into
163 * the btree. The caller should have done a btrfs_drop_extents so that
164 * no overlapping inline items exist in the btree
166 static int insert_inline_extent(struct btrfs_trans_handle *trans,
167 struct btrfs_path *path, int extent_inserted,
168 struct btrfs_root *root, struct inode *inode,
169 u64 start, size_t size, size_t compressed_size,
171 struct page **compressed_pages)
173 struct extent_buffer *leaf;
174 struct page *page = NULL;
177 struct btrfs_file_extent_item *ei;
179 size_t cur_size = size;
180 unsigned long offset;
182 ASSERT((compressed_size > 0 && compressed_pages) ||
183 (compressed_size == 0 && !compressed_pages));
185 if (compressed_size && compressed_pages)
186 cur_size = compressed_size;
188 inode_add_bytes(inode, size);
190 if (!extent_inserted) {
191 struct btrfs_key key;
194 key.objectid = btrfs_ino(BTRFS_I(inode));
196 key.type = BTRFS_EXTENT_DATA_KEY;
198 datasize = btrfs_file_extent_calc_inline_size(cur_size);
199 path->leave_spinning = 1;
200 ret = btrfs_insert_empty_item(trans, root, path, &key,
205 leaf = path->nodes[0];
206 ei = btrfs_item_ptr(leaf, path->slots[0],
207 struct btrfs_file_extent_item);
208 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
209 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
210 btrfs_set_file_extent_encryption(leaf, ei, 0);
211 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
212 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
213 ptr = btrfs_file_extent_inline_start(ei);
215 if (compress_type != BTRFS_COMPRESS_NONE) {
218 while (compressed_size > 0) {
219 cpage = compressed_pages[i];
220 cur_size = min_t(unsigned long, compressed_size,
223 kaddr = kmap_atomic(cpage);
224 write_extent_buffer(leaf, kaddr, ptr, cur_size);
225 kunmap_atomic(kaddr);
229 compressed_size -= cur_size;
231 btrfs_set_file_extent_compression(leaf, ei,
234 page = find_get_page(inode->i_mapping,
235 start >> PAGE_SHIFT);
236 btrfs_set_file_extent_compression(leaf, ei, 0);
237 kaddr = kmap_atomic(page);
238 offset = offset_in_page(start);
239 write_extent_buffer(leaf, kaddr + offset, ptr, size);
240 kunmap_atomic(kaddr);
243 btrfs_mark_buffer_dirty(leaf);
244 btrfs_release_path(path);
247 * We align size to sectorsize for inline extents just for simplicity
250 size = ALIGN(size, root->fs_info->sectorsize);
251 ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start, size);
256 * we're an inline extent, so nobody can
257 * extend the file past i_size without locking
258 * a page we already have locked.
260 * We must do any isize and inode updates
261 * before we unlock the pages. Otherwise we
262 * could end up racing with unlink.
264 BTRFS_I(inode)->disk_i_size = inode->i_size;
265 ret = btrfs_update_inode(trans, root, inode);
273 * conditionally insert an inline extent into the file. This
274 * does the checks required to make sure the data is small enough
275 * to fit as an inline extent.
277 static noinline int cow_file_range_inline(struct inode *inode, u64 start,
278 u64 end, size_t compressed_size,
280 struct page **compressed_pages)
282 struct btrfs_root *root = BTRFS_I(inode)->root;
283 struct btrfs_fs_info *fs_info = root->fs_info;
284 struct btrfs_trans_handle *trans;
285 u64 isize = i_size_read(inode);
286 u64 actual_end = min(end + 1, isize);
287 u64 inline_len = actual_end - start;
288 u64 aligned_end = ALIGN(end, fs_info->sectorsize);
289 u64 data_len = inline_len;
291 struct btrfs_path *path;
292 int extent_inserted = 0;
293 u32 extent_item_size;
296 data_len = compressed_size;
299 actual_end > fs_info->sectorsize ||
300 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
302 (actual_end & (fs_info->sectorsize - 1)) == 0) ||
304 data_len > fs_info->max_inline) {
308 path = btrfs_alloc_path();
312 trans = btrfs_join_transaction(root);
314 btrfs_free_path(path);
315 return PTR_ERR(trans);
317 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
319 if (compressed_size && compressed_pages)
320 extent_item_size = btrfs_file_extent_calc_inline_size(
323 extent_item_size = btrfs_file_extent_calc_inline_size(
326 ret = __btrfs_drop_extents(trans, root, inode, path,
327 start, aligned_end, NULL,
328 1, 1, extent_item_size, &extent_inserted);
330 btrfs_abort_transaction(trans, ret);
334 if (isize > actual_end)
335 inline_len = min_t(u64, isize, actual_end);
336 ret = insert_inline_extent(trans, path, extent_inserted,
338 inline_len, compressed_size,
339 compress_type, compressed_pages);
340 if (ret && ret != -ENOSPC) {
341 btrfs_abort_transaction(trans, ret);
343 } else if (ret == -ENOSPC) {
348 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
349 btrfs_drop_extent_cache(BTRFS_I(inode), start, aligned_end - 1, 0);
352 * Don't forget to free the reserved space, as for inlined extent
353 * it won't count as data extent, free them directly here.
354 * And at reserve time, it's always aligned to page size, so
355 * just free one page here.
357 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
358 btrfs_free_path(path);
359 btrfs_end_transaction(trans);
363 struct async_extent {
368 unsigned long nr_pages;
370 struct list_head list;
375 struct page *locked_page;
378 unsigned int write_flags;
379 struct list_head extents;
380 struct cgroup_subsys_state *blkcg_css;
381 struct btrfs_work work;
386 /* Number of chunks in flight; must be first in the structure */
388 struct async_chunk chunks[];
391 static noinline int add_async_extent(struct async_chunk *cow,
392 u64 start, u64 ram_size,
395 unsigned long nr_pages,
398 struct async_extent *async_extent;
400 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
401 BUG_ON(!async_extent); /* -ENOMEM */
402 async_extent->start = start;
403 async_extent->ram_size = ram_size;
404 async_extent->compressed_size = compressed_size;
405 async_extent->pages = pages;
406 async_extent->nr_pages = nr_pages;
407 async_extent->compress_type = compress_type;
408 list_add_tail(&async_extent->list, &cow->extents);
413 * Check if the inode has flags compatible with compression
415 static inline bool inode_can_compress(struct inode *inode)
417 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW ||
418 BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
424 * Check if the inode needs to be submitted to compression, based on mount
425 * options, defragmentation, properties or heuristics.
427 static inline int inode_need_compress(struct inode *inode, u64 start, u64 end)
429 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
431 if (!inode_can_compress(inode)) {
432 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
433 KERN_ERR "BTRFS: unexpected compression for ino %llu\n",
434 btrfs_ino(BTRFS_I(inode)));
438 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
441 if (BTRFS_I(inode)->defrag_compress)
443 /* bad compression ratios */
444 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
446 if (btrfs_test_opt(fs_info, COMPRESS) ||
447 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
448 BTRFS_I(inode)->prop_compress)
449 return btrfs_compress_heuristic(inode, start, end);
453 static inline void inode_should_defrag(struct btrfs_inode *inode,
454 u64 start, u64 end, u64 num_bytes, u64 small_write)
456 /* If this is a small write inside eof, kick off a defrag */
457 if (num_bytes < small_write &&
458 (start > 0 || end + 1 < inode->disk_i_size))
459 btrfs_add_inode_defrag(NULL, inode);
463 * we create compressed extents in two phases. The first
464 * phase compresses a range of pages that have already been
465 * locked (both pages and state bits are locked).
467 * This is done inside an ordered work queue, and the compression
468 * is spread across many cpus. The actual IO submission is step
469 * two, and the ordered work queue takes care of making sure that
470 * happens in the same order things were put onto the queue by
471 * writepages and friends.
473 * If this code finds it can't get good compression, it puts an
474 * entry onto the work queue to write the uncompressed bytes. This
475 * makes sure that both compressed inodes and uncompressed inodes
476 * are written in the same order that the flusher thread sent them
479 static noinline int compress_file_range(struct async_chunk *async_chunk)
481 struct inode *inode = async_chunk->inode;
482 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
483 u64 blocksize = fs_info->sectorsize;
484 u64 start = async_chunk->start;
485 u64 end = async_chunk->end;
489 struct page **pages = NULL;
490 unsigned long nr_pages;
491 unsigned long total_compressed = 0;
492 unsigned long total_in = 0;
495 int compress_type = fs_info->compress_type;
496 int compressed_extents = 0;
499 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
503 * We need to save i_size before now because it could change in between
504 * us evaluating the size and assigning it. This is because we lock and
505 * unlock the page in truncate and fallocate, and then modify the i_size
508 * The barriers are to emulate READ_ONCE, remove that once i_size_read
512 i_size = i_size_read(inode);
514 actual_end = min_t(u64, i_size, end + 1);
517 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
518 BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
519 nr_pages = min_t(unsigned long, nr_pages,
520 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
523 * we don't want to send crud past the end of i_size through
524 * compression, that's just a waste of CPU time. So, if the
525 * end of the file is before the start of our current
526 * requested range of bytes, we bail out to the uncompressed
527 * cleanup code that can deal with all of this.
529 * It isn't really the fastest way to fix things, but this is a
530 * very uncommon corner.
532 if (actual_end <= start)
533 goto cleanup_and_bail_uncompressed;
535 total_compressed = actual_end - start;
538 * skip compression for a small file range(<=blocksize) that
539 * isn't an inline extent, since it doesn't save disk space at all.
541 if (total_compressed <= blocksize &&
542 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
543 goto cleanup_and_bail_uncompressed;
545 total_compressed = min_t(unsigned long, total_compressed,
546 BTRFS_MAX_UNCOMPRESSED);
551 * we do compression for mount -o compress and when the
552 * inode has not been flagged as nocompress. This flag can
553 * change at any time if we discover bad compression ratios.
555 if (inode_need_compress(inode, start, end)) {
557 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
559 /* just bail out to the uncompressed code */
564 if (BTRFS_I(inode)->defrag_compress)
565 compress_type = BTRFS_I(inode)->defrag_compress;
566 else if (BTRFS_I(inode)->prop_compress)
567 compress_type = BTRFS_I(inode)->prop_compress;
570 * we need to call clear_page_dirty_for_io on each
571 * page in the range. Otherwise applications with the file
572 * mmap'd can wander in and change the page contents while
573 * we are compressing them.
575 * If the compression fails for any reason, we set the pages
576 * dirty again later on.
578 * Note that the remaining part is redirtied, the start pointer
579 * has moved, the end is the original one.
582 extent_range_clear_dirty_for_io(inode, start, end);
586 /* Compression level is applied here and only here */
587 ret = btrfs_compress_pages(
588 compress_type | (fs_info->compress_level << 4),
589 inode->i_mapping, start,
596 unsigned long offset = offset_in_page(total_compressed);
597 struct page *page = pages[nr_pages - 1];
600 /* zero the tail end of the last page, we might be
601 * sending it down to disk
604 kaddr = kmap_atomic(page);
605 memset(kaddr + offset, 0,
607 kunmap_atomic(kaddr);
614 /* lets try to make an inline extent */
615 if (ret || total_in < actual_end) {
616 /* we didn't compress the entire range, try
617 * to make an uncompressed inline extent.
619 ret = cow_file_range_inline(inode, start, end, 0,
620 BTRFS_COMPRESS_NONE, NULL);
622 /* try making a compressed inline extent */
623 ret = cow_file_range_inline(inode, start, end,
625 compress_type, pages);
628 unsigned long clear_flags = EXTENT_DELALLOC |
629 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
630 EXTENT_DO_ACCOUNTING;
631 unsigned long page_error_op;
633 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
636 * inline extent creation worked or returned error,
637 * we don't need to create any more async work items.
638 * Unlock and free up our temp pages.
640 * We use DO_ACCOUNTING here because we need the
641 * delalloc_release_metadata to be done _after_ we drop
642 * our outstanding extent for clearing delalloc for this
645 extent_clear_unlock_delalloc(inode, start, end, NULL,
653 for (i = 0; i < nr_pages; i++) {
654 WARN_ON(pages[i]->mapping);
665 * we aren't doing an inline extent round the compressed size
666 * up to a block size boundary so the allocator does sane
669 total_compressed = ALIGN(total_compressed, blocksize);
672 * one last check to make sure the compression is really a
673 * win, compare the page count read with the blocks on disk,
674 * compression must free at least one sector size
676 total_in = ALIGN(total_in, PAGE_SIZE);
677 if (total_compressed + blocksize <= total_in) {
678 compressed_extents++;
681 * The async work queues will take care of doing actual
682 * allocation on disk for these compressed pages, and
683 * will submit them to the elevator.
685 add_async_extent(async_chunk, start, total_in,
686 total_compressed, pages, nr_pages,
689 if (start + total_in < end) {
695 return compressed_extents;
700 * the compression code ran but failed to make things smaller,
701 * free any pages it allocated and our page pointer array
703 for (i = 0; i < nr_pages; i++) {
704 WARN_ON(pages[i]->mapping);
709 total_compressed = 0;
712 /* flag the file so we don't compress in the future */
713 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
714 !(BTRFS_I(inode)->prop_compress)) {
715 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
718 cleanup_and_bail_uncompressed:
720 * No compression, but we still need to write the pages in the file
721 * we've been given so far. redirty the locked page if it corresponds
722 * to our extent and set things up for the async work queue to run
723 * cow_file_range to do the normal delalloc dance.
725 if (async_chunk->locked_page &&
726 (page_offset(async_chunk->locked_page) >= start &&
727 page_offset(async_chunk->locked_page)) <= end) {
728 __set_page_dirty_nobuffers(async_chunk->locked_page);
729 /* unlocked later on in the async handlers */
733 extent_range_redirty_for_io(inode, start, end);
734 add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
735 BTRFS_COMPRESS_NONE);
736 compressed_extents++;
738 return compressed_extents;
741 static void free_async_extent_pages(struct async_extent *async_extent)
745 if (!async_extent->pages)
748 for (i = 0; i < async_extent->nr_pages; i++) {
749 WARN_ON(async_extent->pages[i]->mapping);
750 put_page(async_extent->pages[i]);
752 kfree(async_extent->pages);
753 async_extent->nr_pages = 0;
754 async_extent->pages = NULL;
758 * phase two of compressed writeback. This is the ordered portion
759 * of the code, which only gets called in the order the work was
760 * queued. We walk all the async extents created by compress_file_range
761 * and send them down to the disk.
763 static noinline void submit_compressed_extents(struct async_chunk *async_chunk)
765 struct inode *inode = async_chunk->inode;
766 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
767 struct async_extent *async_extent;
769 struct btrfs_key ins;
770 struct extent_map *em;
771 struct btrfs_root *root = BTRFS_I(inode)->root;
772 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
776 while (!list_empty(&async_chunk->extents)) {
777 async_extent = list_entry(async_chunk->extents.next,
778 struct async_extent, list);
779 list_del(&async_extent->list);
782 lock_extent(io_tree, async_extent->start,
783 async_extent->start + async_extent->ram_size - 1);
784 /* did the compression code fall back to uncompressed IO? */
785 if (!async_extent->pages) {
786 int page_started = 0;
787 unsigned long nr_written = 0;
789 /* allocate blocks */
790 ret = cow_file_range(inode, async_chunk->locked_page,
792 async_extent->start +
793 async_extent->ram_size - 1,
794 &page_started, &nr_written, 0);
799 * if page_started, cow_file_range inserted an
800 * inline extent and took care of all the unlocking
801 * and IO for us. Otherwise, we need to submit
802 * all those pages down to the drive.
804 if (!page_started && !ret)
805 extent_write_locked_range(inode,
807 async_extent->start +
808 async_extent->ram_size - 1,
810 else if (ret && async_chunk->locked_page)
811 unlock_page(async_chunk->locked_page);
817 ret = btrfs_reserve_extent(root, async_extent->ram_size,
818 async_extent->compressed_size,
819 async_extent->compressed_size,
820 0, alloc_hint, &ins, 1, 1);
822 free_async_extent_pages(async_extent);
824 if (ret == -ENOSPC) {
825 unlock_extent(io_tree, async_extent->start,
826 async_extent->start +
827 async_extent->ram_size - 1);
830 * we need to redirty the pages if we decide to
831 * fallback to uncompressed IO, otherwise we
832 * will not submit these pages down to lower
835 extent_range_redirty_for_io(inode,
837 async_extent->start +
838 async_extent->ram_size - 1);
845 * here we're doing allocation and writeback of the
848 em = create_io_em(inode, async_extent->start,
849 async_extent->ram_size, /* len */
850 async_extent->start, /* orig_start */
851 ins.objectid, /* block_start */
852 ins.offset, /* block_len */
853 ins.offset, /* orig_block_len */
854 async_extent->ram_size, /* ram_bytes */
855 async_extent->compress_type,
856 BTRFS_ORDERED_COMPRESSED);
858 /* ret value is not necessary due to void function */
859 goto out_free_reserve;
862 ret = btrfs_add_ordered_extent_compress(inode,
865 async_extent->ram_size,
867 BTRFS_ORDERED_COMPRESSED,
868 async_extent->compress_type);
870 btrfs_drop_extent_cache(BTRFS_I(inode),
872 async_extent->start +
873 async_extent->ram_size - 1, 0);
874 goto out_free_reserve;
876 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
879 * clear dirty, set writeback and unlock the pages.
881 extent_clear_unlock_delalloc(inode, async_extent->start,
882 async_extent->start +
883 async_extent->ram_size - 1,
884 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
885 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
887 if (btrfs_submit_compressed_write(inode,
889 async_extent->ram_size,
891 ins.offset, async_extent->pages,
892 async_extent->nr_pages,
893 async_chunk->write_flags,
894 async_chunk->blkcg_css)) {
895 struct page *p = async_extent->pages[0];
896 const u64 start = async_extent->start;
897 const u64 end = start + async_extent->ram_size - 1;
899 p->mapping = inode->i_mapping;
900 btrfs_writepage_endio_finish_ordered(p, start, end, 0);
903 extent_clear_unlock_delalloc(inode, start, end,
907 free_async_extent_pages(async_extent);
909 alloc_hint = ins.objectid + ins.offset;
915 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
916 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
918 extent_clear_unlock_delalloc(inode, async_extent->start,
919 async_extent->start +
920 async_extent->ram_size - 1,
921 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
922 EXTENT_DELALLOC_NEW |
923 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
924 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
925 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
927 free_async_extent_pages(async_extent);
932 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
935 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
936 struct extent_map *em;
939 read_lock(&em_tree->lock);
940 em = search_extent_mapping(em_tree, start, num_bytes);
943 * if block start isn't an actual block number then find the
944 * first block in this inode and use that as a hint. If that
945 * block is also bogus then just don't worry about it.
947 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
949 em = search_extent_mapping(em_tree, 0, 0);
950 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
951 alloc_hint = em->block_start;
955 alloc_hint = em->block_start;
959 read_unlock(&em_tree->lock);
965 * when extent_io.c finds a delayed allocation range in the file,
966 * the call backs end up in this code. The basic idea is to
967 * allocate extents on disk for the range, and create ordered data structs
968 * in ram to track those extents.
970 * locked_page is the page that writepage had locked already. We use
971 * it to make sure we don't do extra locks or unlocks.
973 * *page_started is set to one if we unlock locked_page and do everything
974 * required to start IO on it. It may be clean and already done with
977 static noinline int cow_file_range(struct inode *inode,
978 struct page *locked_page,
979 u64 start, u64 end, int *page_started,
980 unsigned long *nr_written, int unlock)
982 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
983 struct btrfs_root *root = BTRFS_I(inode)->root;
986 unsigned long ram_size;
987 u64 cur_alloc_size = 0;
989 u64 blocksize = fs_info->sectorsize;
990 struct btrfs_key ins;
991 struct extent_map *em;
993 unsigned long page_ops;
994 bool extent_reserved = false;
997 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
1003 num_bytes = ALIGN(end - start + 1, blocksize);
1004 num_bytes = max(blocksize, num_bytes);
1005 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1007 inode_should_defrag(BTRFS_I(inode), start, end, num_bytes, SZ_64K);
1010 /* lets try to make an inline extent */
1011 ret = cow_file_range_inline(inode, start, end, 0,
1012 BTRFS_COMPRESS_NONE, NULL);
1015 * We use DO_ACCOUNTING here because we need the
1016 * delalloc_release_metadata to be run _after_ we drop
1017 * our outstanding extent for clearing delalloc for this
1020 extent_clear_unlock_delalloc(inode, start, end, NULL,
1021 EXTENT_LOCKED | EXTENT_DELALLOC |
1022 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1023 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1024 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1025 PAGE_END_WRITEBACK);
1026 *nr_written = *nr_written +
1027 (end - start + PAGE_SIZE) / PAGE_SIZE;
1030 } else if (ret < 0) {
1035 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1036 btrfs_drop_extent_cache(BTRFS_I(inode), start,
1037 start + num_bytes - 1, 0);
1040 * Relocation relies on the relocated extents to have exactly the same
1041 * size as the original extents. Normally writeback for relocation data
1042 * extents follows a NOCOW path because relocation preallocates the
1043 * extents. However, due to an operation such as scrub turning a block
1044 * group to RO mode, it may fallback to COW mode, so we must make sure
1045 * an extent allocated during COW has exactly the requested size and can
1046 * not be split into smaller extents, otherwise relocation breaks and
1047 * fails during the stage where it updates the bytenr of file extent
1050 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1051 min_alloc_size = num_bytes;
1053 min_alloc_size = fs_info->sectorsize;
1055 while (num_bytes > 0) {
1056 cur_alloc_size = num_bytes;
1057 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1058 min_alloc_size, 0, alloc_hint,
1062 cur_alloc_size = ins.offset;
1063 extent_reserved = true;
1065 ram_size = ins.offset;
1066 em = create_io_em(inode, start, ins.offset, /* len */
1067 start, /* orig_start */
1068 ins.objectid, /* block_start */
1069 ins.offset, /* block_len */
1070 ins.offset, /* orig_block_len */
1071 ram_size, /* ram_bytes */
1072 BTRFS_COMPRESS_NONE, /* compress_type */
1073 BTRFS_ORDERED_REGULAR /* type */);
1078 free_extent_map(em);
1080 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1081 ram_size, cur_alloc_size, 0);
1083 goto out_drop_extent_cache;
1085 if (root->root_key.objectid ==
1086 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1087 ret = btrfs_reloc_clone_csums(inode, start,
1090 * Only drop cache here, and process as normal.
1092 * We must not allow extent_clear_unlock_delalloc()
1093 * at out_unlock label to free meta of this ordered
1094 * extent, as its meta should be freed by
1095 * btrfs_finish_ordered_io().
1097 * So we must continue until @start is increased to
1098 * skip current ordered extent.
1101 btrfs_drop_extent_cache(BTRFS_I(inode), start,
1102 start + ram_size - 1, 0);
1105 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1107 /* we're not doing compressed IO, don't unlock the first
1108 * page (which the caller expects to stay locked), don't
1109 * clear any dirty bits and don't set any writeback bits
1111 * Do set the Private2 bit so we know this page was properly
1112 * setup for writepage
1114 page_ops = unlock ? PAGE_UNLOCK : 0;
1115 page_ops |= PAGE_SET_PRIVATE2;
1117 extent_clear_unlock_delalloc(inode, start,
1118 start + ram_size - 1,
1120 EXTENT_LOCKED | EXTENT_DELALLOC,
1122 if (num_bytes < cur_alloc_size)
1125 num_bytes -= cur_alloc_size;
1126 alloc_hint = ins.objectid + ins.offset;
1127 start += cur_alloc_size;
1128 extent_reserved = false;
1131 * btrfs_reloc_clone_csums() error, since start is increased
1132 * extent_clear_unlock_delalloc() at out_unlock label won't
1133 * free metadata of current ordered extent, we're OK to exit.
1141 out_drop_extent_cache:
1142 btrfs_drop_extent_cache(BTRFS_I(inode), start, start + ram_size - 1, 0);
1144 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1145 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1147 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1148 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1149 page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1152 * If we reserved an extent for our delalloc range (or a subrange) and
1153 * failed to create the respective ordered extent, then it means that
1154 * when we reserved the extent we decremented the extent's size from
1155 * the data space_info's bytes_may_use counter and incremented the
1156 * space_info's bytes_reserved counter by the same amount. We must make
1157 * sure extent_clear_unlock_delalloc() does not try to decrement again
1158 * the data space_info's bytes_may_use counter, therefore we do not pass
1159 * it the flag EXTENT_CLEAR_DATA_RESV.
1161 if (extent_reserved) {
1162 extent_clear_unlock_delalloc(inode, start,
1163 start + cur_alloc_size - 1,
1167 start += cur_alloc_size;
1171 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1172 clear_bits | EXTENT_CLEAR_DATA_RESV,
1178 * work queue call back to started compression on a file and pages
1180 static noinline void async_cow_start(struct btrfs_work *work)
1182 struct async_chunk *async_chunk;
1183 int compressed_extents;
1185 async_chunk = container_of(work, struct async_chunk, work);
1187 compressed_extents = compress_file_range(async_chunk);
1188 if (compressed_extents == 0) {
1189 btrfs_add_delayed_iput(async_chunk->inode);
1190 async_chunk->inode = NULL;
1195 * work queue call back to submit previously compressed pages
1197 static noinline void async_cow_submit(struct btrfs_work *work)
1199 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1201 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1202 unsigned long nr_pages;
1204 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1207 /* atomic_sub_return implies a barrier */
1208 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1210 cond_wake_up_nomb(&fs_info->async_submit_wait);
1213 * ->inode could be NULL if async_chunk_start has failed to compress,
1214 * in which case we don't have anything to submit, yet we need to
1215 * always adjust ->async_delalloc_pages as its paired with the init
1216 * happening in cow_file_range_async
1218 if (async_chunk->inode)
1219 submit_compressed_extents(async_chunk);
1222 static noinline void async_cow_free(struct btrfs_work *work)
1224 struct async_chunk *async_chunk;
1226 async_chunk = container_of(work, struct async_chunk, work);
1227 if (async_chunk->inode)
1228 btrfs_add_delayed_iput(async_chunk->inode);
1229 if (async_chunk->blkcg_css)
1230 css_put(async_chunk->blkcg_css);
1232 * Since the pointer to 'pending' is at the beginning of the array of
1233 * async_chunk's, freeing it ensures the whole array has been freed.
1235 if (atomic_dec_and_test(async_chunk->pending))
1236 kvfree(async_chunk->pending);
1239 static int cow_file_range_async(struct inode *inode,
1240 struct writeback_control *wbc,
1241 struct page *locked_page,
1242 u64 start, u64 end, int *page_started,
1243 unsigned long *nr_written)
1245 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1246 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1247 struct async_cow *ctx;
1248 struct async_chunk *async_chunk;
1249 unsigned long nr_pages;
1251 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1253 bool should_compress;
1255 const unsigned int write_flags = wbc_to_write_flags(wbc);
1257 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
1259 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1260 !btrfs_test_opt(fs_info, FORCE_COMPRESS)) {
1262 should_compress = false;
1264 should_compress = true;
1267 nofs_flag = memalloc_nofs_save();
1268 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1269 memalloc_nofs_restore(nofs_flag);
1272 unsigned clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC |
1273 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1274 EXTENT_DO_ACCOUNTING;
1275 unsigned long page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1276 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
1279 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1280 clear_bits, page_ops);
1284 async_chunk = ctx->chunks;
1285 atomic_set(&ctx->num_chunks, num_chunks);
1287 for (i = 0; i < num_chunks; i++) {
1288 if (should_compress)
1289 cur_end = min(end, start + SZ_512K - 1);
1294 * igrab is called higher up in the call chain, take only the
1295 * lightweight reference for the callback lifetime
1298 async_chunk[i].pending = &ctx->num_chunks;
1299 async_chunk[i].inode = inode;
1300 async_chunk[i].start = start;
1301 async_chunk[i].end = cur_end;
1302 async_chunk[i].write_flags = write_flags;
1303 INIT_LIST_HEAD(&async_chunk[i].extents);
1306 * The locked_page comes all the way from writepage and its
1307 * the original page we were actually given. As we spread
1308 * this large delalloc region across multiple async_chunk
1309 * structs, only the first struct needs a pointer to locked_page
1311 * This way we don't need racey decisions about who is supposed
1316 * Depending on the compressibility, the pages might or
1317 * might not go through async. We want all of them to
1318 * be accounted against wbc once. Let's do it here
1319 * before the paths diverge. wbc accounting is used
1320 * only for foreign writeback detection and doesn't
1321 * need full accuracy. Just account the whole thing
1322 * against the first page.
1324 wbc_account_cgroup_owner(wbc, locked_page,
1326 async_chunk[i].locked_page = locked_page;
1329 async_chunk[i].locked_page = NULL;
1332 if (blkcg_css != blkcg_root_css) {
1334 async_chunk[i].blkcg_css = blkcg_css;
1336 async_chunk[i].blkcg_css = NULL;
1339 btrfs_init_work(&async_chunk[i].work, async_cow_start,
1340 async_cow_submit, async_cow_free);
1342 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1343 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1345 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1347 *nr_written += nr_pages;
1348 start = cur_end + 1;
1354 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1355 u64 bytenr, u64 num_bytes)
1358 struct btrfs_ordered_sum *sums;
1361 ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1362 bytenr + num_bytes - 1, &list, 0);
1363 if (ret == 0 && list_empty(&list))
1366 while (!list_empty(&list)) {
1367 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1368 list_del(&sums->list);
1376 static int fallback_to_cow(struct inode *inode, struct page *locked_page,
1377 const u64 start, const u64 end,
1378 int *page_started, unsigned long *nr_written)
1380 const bool is_space_ino = btrfs_is_free_space_inode(BTRFS_I(inode));
1381 const bool is_reloc_ino = (BTRFS_I(inode)->root->root_key.objectid ==
1382 BTRFS_DATA_RELOC_TREE_OBJECTID);
1383 const u64 range_bytes = end + 1 - start;
1384 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1385 u64 range_start = start;
1389 * If EXTENT_NORESERVE is set it means that when the buffered write was
1390 * made we had not enough available data space and therefore we did not
1391 * reserve data space for it, since we though we could do NOCOW for the
1392 * respective file range (either there is prealloc extent or the inode
1393 * has the NOCOW bit set).
1395 * However when we need to fallback to COW mode (because for example the
1396 * block group for the corresponding extent was turned to RO mode by a
1397 * scrub or relocation) we need to do the following:
1399 * 1) We increment the bytes_may_use counter of the data space info.
1400 * If COW succeeds, it allocates a new data extent and after doing
1401 * that it decrements the space info's bytes_may_use counter and
1402 * increments its bytes_reserved counter by the same amount (we do
1403 * this at btrfs_add_reserved_bytes()). So we need to increment the
1404 * bytes_may_use counter to compensate (when space is reserved at
1405 * buffered write time, the bytes_may_use counter is incremented);
1407 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1408 * that if the COW path fails for any reason, it decrements (through
1409 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1410 * data space info, which we incremented in the step above.
1412 * If we need to fallback to cow and the inode corresponds to a free
1413 * space cache inode or an inode of the data relocation tree, we must
1414 * also increment bytes_may_use of the data space_info for the same
1415 * reason. Space caches and relocated data extents always get a prealloc
1416 * extent for them, however scrub or balance may have set the block
1417 * group that contains that extent to RO mode and therefore force COW
1418 * when starting writeback.
1420 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1421 EXTENT_NORESERVE, 0);
1422 if (count > 0 || is_space_ino || is_reloc_ino) {
1424 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1425 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1427 if (is_space_ino || is_reloc_ino)
1428 bytes = range_bytes;
1430 spin_lock(&sinfo->lock);
1431 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1432 spin_unlock(&sinfo->lock);
1435 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1439 return cow_file_range(inode, locked_page, start, end, page_started,
1444 * when nowcow writeback call back. This checks for snapshots or COW copies
1445 * of the extents that exist in the file, and COWs the file as required.
1447 * If no cow copies or snapshots exist, we write directly to the existing
1450 static noinline int run_delalloc_nocow(struct inode *inode,
1451 struct page *locked_page,
1452 const u64 start, const u64 end,
1453 int *page_started, int force,
1454 unsigned long *nr_written)
1456 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1457 struct btrfs_root *root = BTRFS_I(inode)->root;
1458 struct btrfs_path *path;
1459 u64 cow_start = (u64)-1;
1460 u64 cur_offset = start;
1462 bool check_prev = true;
1463 const bool freespace_inode = btrfs_is_free_space_inode(BTRFS_I(inode));
1464 u64 ino = btrfs_ino(BTRFS_I(inode));
1466 u64 disk_bytenr = 0;
1468 path = btrfs_alloc_path();
1470 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1471 EXTENT_LOCKED | EXTENT_DELALLOC |
1472 EXTENT_DO_ACCOUNTING |
1473 EXTENT_DEFRAG, PAGE_UNLOCK |
1475 PAGE_SET_WRITEBACK |
1476 PAGE_END_WRITEBACK);
1481 struct btrfs_key found_key;
1482 struct btrfs_file_extent_item *fi;
1483 struct extent_buffer *leaf;
1493 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1499 * If there is no extent for our range when doing the initial
1500 * search, then go back to the previous slot as it will be the
1501 * one containing the search offset
1503 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1504 leaf = path->nodes[0];
1505 btrfs_item_key_to_cpu(leaf, &found_key,
1506 path->slots[0] - 1);
1507 if (found_key.objectid == ino &&
1508 found_key.type == BTRFS_EXTENT_DATA_KEY)
1513 /* Go to next leaf if we have exhausted the current one */
1514 leaf = path->nodes[0];
1515 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1516 ret = btrfs_next_leaf(root, path);
1518 if (cow_start != (u64)-1)
1519 cur_offset = cow_start;
1524 leaf = path->nodes[0];
1527 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1529 /* Didn't find anything for our INO */
1530 if (found_key.objectid > ino)
1533 * Keep searching until we find an EXTENT_ITEM or there are no
1534 * more extents for this inode
1536 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1537 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1542 /* Found key is not EXTENT_DATA_KEY or starts after req range */
1543 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1544 found_key.offset > end)
1548 * If the found extent starts after requested offset, then
1549 * adjust extent_end to be right before this extent begins
1551 if (found_key.offset > cur_offset) {
1552 extent_end = found_key.offset;
1558 * Found extent which begins before our range and potentially
1561 fi = btrfs_item_ptr(leaf, path->slots[0],
1562 struct btrfs_file_extent_item);
1563 extent_type = btrfs_file_extent_type(leaf, fi);
1565 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1566 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1567 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1568 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1569 extent_offset = btrfs_file_extent_offset(leaf, fi);
1570 extent_end = found_key.offset +
1571 btrfs_file_extent_num_bytes(leaf, fi);
1573 btrfs_file_extent_disk_num_bytes(leaf, fi);
1575 * If the extent we got ends before our current offset,
1576 * skip to the next extent.
1578 if (extent_end <= cur_offset) {
1583 if (disk_bytenr == 0)
1585 /* Skip compressed/encrypted/encoded extents */
1586 if (btrfs_file_extent_compression(leaf, fi) ||
1587 btrfs_file_extent_encryption(leaf, fi) ||
1588 btrfs_file_extent_other_encoding(leaf, fi))
1591 * If extent is created before the last volume's snapshot
1592 * this implies the extent is shared, hence we can't do
1593 * nocow. This is the same check as in
1594 * btrfs_cross_ref_exist but without calling
1595 * btrfs_search_slot.
1597 if (!freespace_inode &&
1598 btrfs_file_extent_generation(leaf, fi) <=
1599 btrfs_root_last_snapshot(&root->root_item))
1601 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1603 /* If extent is RO, we must COW it */
1604 if (btrfs_extent_readonly(fs_info, disk_bytenr))
1606 ret = btrfs_cross_ref_exist(root, ino,
1608 extent_offset, disk_bytenr);
1611 * ret could be -EIO if the above fails to read
1615 if (cow_start != (u64)-1)
1616 cur_offset = cow_start;
1620 WARN_ON_ONCE(freespace_inode);
1623 disk_bytenr += extent_offset;
1624 disk_bytenr += cur_offset - found_key.offset;
1625 num_bytes = min(end + 1, extent_end) - cur_offset;
1627 * If there are pending snapshots for this root, we
1628 * fall into common COW way
1630 if (!freespace_inode && atomic_read(&root->snapshot_force_cow))
1633 * force cow if csum exists in the range.
1634 * this ensure that csum for a given extent are
1635 * either valid or do not exist.
1637 ret = csum_exist_in_range(fs_info, disk_bytenr,
1641 * ret could be -EIO if the above fails to read
1645 if (cow_start != (u64)-1)
1646 cur_offset = cow_start;
1649 WARN_ON_ONCE(freespace_inode);
1652 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr))
1655 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1656 extent_end = found_key.offset + ram_bytes;
1657 extent_end = ALIGN(extent_end, fs_info->sectorsize);
1658 /* Skip extents outside of our requested range */
1659 if (extent_end <= start) {
1664 /* If this triggers then we have a memory corruption */
1669 * If nocow is false then record the beginning of the range
1670 * that needs to be COWed
1673 if (cow_start == (u64)-1)
1674 cow_start = cur_offset;
1675 cur_offset = extent_end;
1676 if (cur_offset > end)
1682 btrfs_release_path(path);
1685 * COW range from cow_start to found_key.offset - 1. As the key
1686 * will contain the beginning of the first extent that can be
1687 * NOCOW, following one which needs to be COW'ed
1689 if (cow_start != (u64)-1) {
1690 ret = fallback_to_cow(inode, locked_page, cow_start,
1691 found_key.offset - 1,
1692 page_started, nr_written);
1695 btrfs_dec_nocow_writers(fs_info,
1699 cow_start = (u64)-1;
1702 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1703 u64 orig_start = found_key.offset - extent_offset;
1704 struct extent_map *em;
1706 em = create_io_em(inode, cur_offset, num_bytes,
1708 disk_bytenr, /* block_start */
1709 num_bytes, /* block_len */
1710 disk_num_bytes, /* orig_block_len */
1711 ram_bytes, BTRFS_COMPRESS_NONE,
1712 BTRFS_ORDERED_PREALLOC);
1715 btrfs_dec_nocow_writers(fs_info,
1720 free_extent_map(em);
1721 ret = btrfs_add_ordered_extent(inode, cur_offset,
1722 disk_bytenr, num_bytes,
1724 BTRFS_ORDERED_PREALLOC);
1726 btrfs_drop_extent_cache(BTRFS_I(inode),
1728 cur_offset + num_bytes - 1,
1733 ret = btrfs_add_ordered_extent(inode, cur_offset,
1734 disk_bytenr, num_bytes,
1736 BTRFS_ORDERED_NOCOW);
1742 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1745 if (root->root_key.objectid ==
1746 BTRFS_DATA_RELOC_TREE_OBJECTID)
1748 * Error handled later, as we must prevent
1749 * extent_clear_unlock_delalloc() in error handler
1750 * from freeing metadata of created ordered extent.
1752 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1755 extent_clear_unlock_delalloc(inode, cur_offset,
1756 cur_offset + num_bytes - 1,
1757 locked_page, EXTENT_LOCKED |
1759 EXTENT_CLEAR_DATA_RESV,
1760 PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1762 cur_offset = extent_end;
1765 * btrfs_reloc_clone_csums() error, now we're OK to call error
1766 * handler, as metadata for created ordered extent will only
1767 * be freed by btrfs_finish_ordered_io().
1771 if (cur_offset > end)
1774 btrfs_release_path(path);
1776 if (cur_offset <= end && cow_start == (u64)-1)
1777 cow_start = cur_offset;
1779 if (cow_start != (u64)-1) {
1781 ret = fallback_to_cow(inode, locked_page, cow_start, end,
1782 page_started, nr_written);
1789 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1791 if (ret && cur_offset < end)
1792 extent_clear_unlock_delalloc(inode, cur_offset, end,
1793 locked_page, EXTENT_LOCKED |
1794 EXTENT_DELALLOC | EXTENT_DEFRAG |
1795 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1797 PAGE_SET_WRITEBACK |
1798 PAGE_END_WRITEBACK);
1799 btrfs_free_path(path);
1803 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1806 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1807 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1811 * @defrag_bytes is a hint value, no spinlock held here,
1812 * if is not zero, it means the file is defragging.
1813 * Force cow if given extent needs to be defragged.
1815 if (BTRFS_I(inode)->defrag_bytes &&
1816 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1817 EXTENT_DEFRAG, 0, NULL))
1824 * Function to process delayed allocation (create CoW) for ranges which are
1825 * being touched for the first time.
1827 int btrfs_run_delalloc_range(struct inode *inode, struct page *locked_page,
1828 u64 start, u64 end, int *page_started, unsigned long *nr_written,
1829 struct writeback_control *wbc)
1832 int force_cow = need_force_cow(inode, start, end);
1834 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1835 ret = run_delalloc_nocow(inode, locked_page, start, end,
1836 page_started, 1, nr_written);
1837 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1838 ret = run_delalloc_nocow(inode, locked_page, start, end,
1839 page_started, 0, nr_written);
1840 } else if (!inode_can_compress(inode) ||
1841 !inode_need_compress(inode, start, end)) {
1842 ret = cow_file_range(inode, locked_page, start, end,
1843 page_started, nr_written, 1);
1845 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1846 &BTRFS_I(inode)->runtime_flags);
1847 ret = cow_file_range_async(inode, wbc, locked_page, start, end,
1848 page_started, nr_written);
1851 btrfs_cleanup_ordered_extents(inode, locked_page, start,
1856 void btrfs_split_delalloc_extent(struct inode *inode,
1857 struct extent_state *orig, u64 split)
1861 /* not delalloc, ignore it */
1862 if (!(orig->state & EXTENT_DELALLOC))
1865 size = orig->end - orig->start + 1;
1866 if (size > BTRFS_MAX_EXTENT_SIZE) {
1871 * See the explanation in btrfs_merge_delalloc_extent, the same
1872 * applies here, just in reverse.
1874 new_size = orig->end - split + 1;
1875 num_extents = count_max_extents(new_size);
1876 new_size = split - orig->start;
1877 num_extents += count_max_extents(new_size);
1878 if (count_max_extents(size) >= num_extents)
1882 spin_lock(&BTRFS_I(inode)->lock);
1883 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1884 spin_unlock(&BTRFS_I(inode)->lock);
1888 * Handle merged delayed allocation extents so we can keep track of new extents
1889 * that are just merged onto old extents, such as when we are doing sequential
1890 * writes, so we can properly account for the metadata space we'll need.
1892 void btrfs_merge_delalloc_extent(struct inode *inode, struct extent_state *new,
1893 struct extent_state *other)
1895 u64 new_size, old_size;
1898 /* not delalloc, ignore it */
1899 if (!(other->state & EXTENT_DELALLOC))
1902 if (new->start > other->start)
1903 new_size = new->end - other->start + 1;
1905 new_size = other->end - new->start + 1;
1907 /* we're not bigger than the max, unreserve the space and go */
1908 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1909 spin_lock(&BTRFS_I(inode)->lock);
1910 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1911 spin_unlock(&BTRFS_I(inode)->lock);
1916 * We have to add up either side to figure out how many extents were
1917 * accounted for before we merged into one big extent. If the number of
1918 * extents we accounted for is <= the amount we need for the new range
1919 * then we can return, otherwise drop. Think of it like this
1923 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1924 * need 2 outstanding extents, on one side we have 1 and the other side
1925 * we have 1 so they are == and we can return. But in this case
1927 * [MAX_SIZE+4k][MAX_SIZE+4k]
1929 * Each range on their own accounts for 2 extents, but merged together
1930 * they are only 3 extents worth of accounting, so we need to drop in
1933 old_size = other->end - other->start + 1;
1934 num_extents = count_max_extents(old_size);
1935 old_size = new->end - new->start + 1;
1936 num_extents += count_max_extents(old_size);
1937 if (count_max_extents(new_size) >= num_extents)
1940 spin_lock(&BTRFS_I(inode)->lock);
1941 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1942 spin_unlock(&BTRFS_I(inode)->lock);
1945 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1946 struct inode *inode)
1948 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1950 spin_lock(&root->delalloc_lock);
1951 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1952 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1953 &root->delalloc_inodes);
1954 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1955 &BTRFS_I(inode)->runtime_flags);
1956 root->nr_delalloc_inodes++;
1957 if (root->nr_delalloc_inodes == 1) {
1958 spin_lock(&fs_info->delalloc_root_lock);
1959 BUG_ON(!list_empty(&root->delalloc_root));
1960 list_add_tail(&root->delalloc_root,
1961 &fs_info->delalloc_roots);
1962 spin_unlock(&fs_info->delalloc_root_lock);
1965 spin_unlock(&root->delalloc_lock);
1969 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
1970 struct btrfs_inode *inode)
1972 struct btrfs_fs_info *fs_info = root->fs_info;
1974 if (!list_empty(&inode->delalloc_inodes)) {
1975 list_del_init(&inode->delalloc_inodes);
1976 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1977 &inode->runtime_flags);
1978 root->nr_delalloc_inodes--;
1979 if (!root->nr_delalloc_inodes) {
1980 ASSERT(list_empty(&root->delalloc_inodes));
1981 spin_lock(&fs_info->delalloc_root_lock);
1982 BUG_ON(list_empty(&root->delalloc_root));
1983 list_del_init(&root->delalloc_root);
1984 spin_unlock(&fs_info->delalloc_root_lock);
1989 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1990 struct btrfs_inode *inode)
1992 spin_lock(&root->delalloc_lock);
1993 __btrfs_del_delalloc_inode(root, inode);
1994 spin_unlock(&root->delalloc_lock);
1998 * Properly track delayed allocation bytes in the inode and to maintain the
1999 * list of inodes that have pending delalloc work to be done.
2001 void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state,
2004 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2006 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
2009 * set_bit and clear bit hooks normally require _irqsave/restore
2010 * but in this case, we are only testing for the DELALLOC
2011 * bit, which is only set or cleared with irqs on
2013 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2014 struct btrfs_root *root = BTRFS_I(inode)->root;
2015 u64 len = state->end + 1 - state->start;
2016 u32 num_extents = count_max_extents(len);
2017 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
2019 spin_lock(&BTRFS_I(inode)->lock);
2020 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
2021 spin_unlock(&BTRFS_I(inode)->lock);
2023 /* For sanity tests */
2024 if (btrfs_is_testing(fs_info))
2027 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2028 fs_info->delalloc_batch);
2029 spin_lock(&BTRFS_I(inode)->lock);
2030 BTRFS_I(inode)->delalloc_bytes += len;
2031 if (*bits & EXTENT_DEFRAG)
2032 BTRFS_I(inode)->defrag_bytes += len;
2033 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2034 &BTRFS_I(inode)->runtime_flags))
2035 btrfs_add_delalloc_inodes(root, inode);
2036 spin_unlock(&BTRFS_I(inode)->lock);
2039 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2040 (*bits & EXTENT_DELALLOC_NEW)) {
2041 spin_lock(&BTRFS_I(inode)->lock);
2042 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
2044 spin_unlock(&BTRFS_I(inode)->lock);
2049 * Once a range is no longer delalloc this function ensures that proper
2050 * accounting happens.
2052 void btrfs_clear_delalloc_extent(struct inode *vfs_inode,
2053 struct extent_state *state, unsigned *bits)
2055 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
2056 struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb);
2057 u64 len = state->end + 1 - state->start;
2058 u32 num_extents = count_max_extents(len);
2060 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
2061 spin_lock(&inode->lock);
2062 inode->defrag_bytes -= len;
2063 spin_unlock(&inode->lock);
2067 * set_bit and clear bit hooks normally require _irqsave/restore
2068 * but in this case, we are only testing for the DELALLOC
2069 * bit, which is only set or cleared with irqs on
2071 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2072 struct btrfs_root *root = inode->root;
2073 bool do_list = !btrfs_is_free_space_inode(inode);
2075 spin_lock(&inode->lock);
2076 btrfs_mod_outstanding_extents(inode, -num_extents);
2077 spin_unlock(&inode->lock);
2080 * We don't reserve metadata space for space cache inodes so we
2081 * don't need to call delalloc_release_metadata if there is an
2084 if (*bits & EXTENT_CLEAR_META_RESV &&
2085 root != fs_info->tree_root)
2086 btrfs_delalloc_release_metadata(inode, len, false);
2088 /* For sanity tests. */
2089 if (btrfs_is_testing(fs_info))
2092 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
2093 do_list && !(state->state & EXTENT_NORESERVE) &&
2094 (*bits & EXTENT_CLEAR_DATA_RESV))
2095 btrfs_free_reserved_data_space_noquota(
2099 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2100 fs_info->delalloc_batch);
2101 spin_lock(&inode->lock);
2102 inode->delalloc_bytes -= len;
2103 if (do_list && inode->delalloc_bytes == 0 &&
2104 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2105 &inode->runtime_flags))
2106 btrfs_del_delalloc_inode(root, inode);
2107 spin_unlock(&inode->lock);
2110 if ((state->state & EXTENT_DELALLOC_NEW) &&
2111 (*bits & EXTENT_DELALLOC_NEW)) {
2112 spin_lock(&inode->lock);
2113 ASSERT(inode->new_delalloc_bytes >= len);
2114 inode->new_delalloc_bytes -= len;
2115 spin_unlock(&inode->lock);
2120 * btrfs_bio_fits_in_stripe - Checks whether the size of the given bio will fit
2121 * in a chunk's stripe. This function ensures that bios do not span a
2124 * @page - The page we are about to add to the bio
2125 * @size - size we want to add to the bio
2126 * @bio - bio we want to ensure is smaller than a stripe
2127 * @bio_flags - flags of the bio
2129 * return 1 if page cannot be added to the bio
2130 * return 0 if page can be added to the bio
2131 * return error otherwise
2133 int btrfs_bio_fits_in_stripe(struct page *page, size_t size, struct bio *bio,
2134 unsigned long bio_flags)
2136 struct inode *inode = page->mapping->host;
2137 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2138 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
2142 struct btrfs_io_geometry geom;
2144 if (bio_flags & EXTENT_BIO_COMPRESSED)
2147 length = bio->bi_iter.bi_size;
2148 map_length = length;
2149 ret = btrfs_get_io_geometry(fs_info, btrfs_op(bio), logical, map_length,
2154 if (geom.len < length + size)
2160 * in order to insert checksums into the metadata in large chunks,
2161 * we wait until bio submission time. All the pages in the bio are
2162 * checksummed and sums are attached onto the ordered extent record.
2164 * At IO completion time the cums attached on the ordered extent record
2165 * are inserted into the btree
2167 static blk_status_t btrfs_submit_bio_start(void *private_data, struct bio *bio,
2170 struct inode *inode = private_data;
2171 blk_status_t ret = 0;
2173 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
2174 BUG_ON(ret); /* -ENOMEM */
2179 * extent_io.c submission hook. This does the right thing for csum calculation
2180 * on write, or reading the csums from the tree before a read.
2182 * Rules about async/sync submit,
2183 * a) read: sync submit
2185 * b) write without checksum: sync submit
2187 * c) write with checksum:
2188 * c-1) if bio is issued by fsync: sync submit
2189 * (sync_writers != 0)
2191 * c-2) if root is reloc root: sync submit
2192 * (only in case of buffered IO)
2194 * c-3) otherwise: async submit
2196 static blk_status_t btrfs_submit_bio_hook(struct inode *inode, struct bio *bio,
2198 unsigned long bio_flags)
2201 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2202 struct btrfs_root *root = BTRFS_I(inode)->root;
2203 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
2204 blk_status_t ret = 0;
2206 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
2208 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
2210 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
2211 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
2213 if (bio_op(bio) != REQ_OP_WRITE) {
2214 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
2218 if (bio_flags & EXTENT_BIO_COMPRESSED) {
2219 ret = btrfs_submit_compressed_read(inode, bio,
2223 } else if (!skip_sum) {
2224 ret = btrfs_lookup_bio_sums(inode, bio, (u64)-1, NULL);
2229 } else if (async && !skip_sum) {
2230 /* csum items have already been cloned */
2231 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2233 /* we're doing a write, do the async checksumming */
2234 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags,
2235 0, inode, btrfs_submit_bio_start);
2237 } else if (!skip_sum) {
2238 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
2244 ret = btrfs_map_bio(fs_info, bio, mirror_num);
2248 bio->bi_status = ret;
2255 * given a list of ordered sums record them in the inode. This happens
2256 * at IO completion time based on sums calculated at bio submission time.
2258 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
2259 struct inode *inode, struct list_head *list)
2261 struct btrfs_ordered_sum *sum;
2264 list_for_each_entry(sum, list, list) {
2265 trans->adding_csums = true;
2266 ret = btrfs_csum_file_blocks(trans,
2267 BTRFS_I(inode)->root->fs_info->csum_root, sum);
2268 trans->adding_csums = false;
2275 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
2276 unsigned int extra_bits,
2277 struct extent_state **cached_state)
2279 WARN_ON(PAGE_ALIGNED(end));
2280 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2281 extra_bits, cached_state);
2284 /* see btrfs_writepage_start_hook for details on why this is required */
2285 struct btrfs_writepage_fixup {
2287 struct inode *inode;
2288 struct btrfs_work work;
2291 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2293 struct btrfs_writepage_fixup *fixup;
2294 struct btrfs_ordered_extent *ordered;
2295 struct extent_state *cached_state = NULL;
2296 struct extent_changeset *data_reserved = NULL;
2298 struct inode *inode;
2302 bool free_delalloc_space = true;
2304 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2306 inode = fixup->inode;
2307 page_start = page_offset(page);
2308 page_end = page_offset(page) + PAGE_SIZE - 1;
2311 * This is similar to page_mkwrite, we need to reserve the space before
2312 * we take the page lock.
2314 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2320 * Before we queued this fixup, we took a reference on the page.
2321 * page->mapping may go NULL, but it shouldn't be moved to a different
2324 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2326 * Unfortunately this is a little tricky, either
2328 * 1) We got here and our page had already been dealt with and
2329 * we reserved our space, thus ret == 0, so we need to just
2330 * drop our space reservation and bail. This can happen the
2331 * first time we come into the fixup worker, or could happen
2332 * while waiting for the ordered extent.
2333 * 2) Our page was already dealt with, but we happened to get an
2334 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2335 * this case we obviously don't have anything to release, but
2336 * because the page was already dealt with we don't want to
2337 * mark the page with an error, so make sure we're resetting
2338 * ret to 0. This is why we have this check _before_ the ret
2339 * check, because we do not want to have a surprise ENOSPC
2340 * when the page was already properly dealt with.
2343 btrfs_delalloc_release_extents(BTRFS_I(inode),
2345 btrfs_delalloc_release_space(inode, data_reserved,
2346 page_start, PAGE_SIZE,
2354 * We can't mess with the page state unless it is locked, so now that
2355 * it is locked bail if we failed to make our space reservation.
2360 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2363 /* already ordered? We're done */
2364 if (PagePrivate2(page))
2367 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
2370 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2371 page_end, &cached_state);
2373 btrfs_start_ordered_extent(inode, ordered, 1);
2374 btrfs_put_ordered_extent(ordered);
2378 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2384 * Everything went as planned, we're now the owner of a dirty page with
2385 * delayed allocation bits set and space reserved for our COW
2388 * The page was dirty when we started, nothing should have cleaned it.
2390 BUG_ON(!PageDirty(page));
2391 free_delalloc_space = false;
2393 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
2394 if (free_delalloc_space)
2395 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2397 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2402 * We hit ENOSPC or other errors. Update the mapping and page
2403 * to reflect the errors and clean the page.
2405 mapping_set_error(page->mapping, ret);
2406 end_extent_writepage(page, ret, page_start, page_end);
2407 clear_page_dirty_for_io(page);
2410 ClearPageChecked(page);
2414 extent_changeset_free(data_reserved);
2416 * As a precaution, do a delayed iput in case it would be the last iput
2417 * that could need flushing space. Recursing back to fixup worker would
2420 btrfs_add_delayed_iput(inode);
2424 * There are a few paths in the higher layers of the kernel that directly
2425 * set the page dirty bit without asking the filesystem if it is a
2426 * good idea. This causes problems because we want to make sure COW
2427 * properly happens and the data=ordered rules are followed.
2429 * In our case any range that doesn't have the ORDERED bit set
2430 * hasn't been properly setup for IO. We kick off an async process
2431 * to fix it up. The async helper will wait for ordered extents, set
2432 * the delalloc bit and make it safe to write the page.
2434 int btrfs_writepage_cow_fixup(struct page *page, u64 start, u64 end)
2436 struct inode *inode = page->mapping->host;
2437 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2438 struct btrfs_writepage_fixup *fixup;
2440 /* this page is properly in the ordered list */
2441 if (TestClearPagePrivate2(page))
2445 * PageChecked is set below when we create a fixup worker for this page,
2446 * don't try to create another one if we're already PageChecked()
2448 * The extent_io writepage code will redirty the page if we send back
2451 if (PageChecked(page))
2454 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2459 * We are already holding a reference to this inode from
2460 * write_cache_pages. We need to hold it because the space reservation
2461 * takes place outside of the page lock, and we can't trust
2462 * page->mapping outside of the page lock.
2465 SetPageChecked(page);
2467 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
2469 fixup->inode = inode;
2470 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2475 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2476 struct inode *inode, u64 file_pos,
2477 u64 disk_bytenr, u64 disk_num_bytes,
2478 u64 num_bytes, u64 ram_bytes,
2479 u8 compression, u8 encryption,
2480 u16 other_encoding, int extent_type)
2482 struct btrfs_root *root = BTRFS_I(inode)->root;
2483 struct btrfs_file_extent_item *fi;
2484 struct btrfs_path *path;
2485 struct extent_buffer *leaf;
2486 struct btrfs_key ins;
2488 int extent_inserted = 0;
2491 path = btrfs_alloc_path();
2496 * we may be replacing one extent in the tree with another.
2497 * The new extent is pinned in the extent map, and we don't want
2498 * to drop it from the cache until it is completely in the btree.
2500 * So, tell btrfs_drop_extents to leave this extent in the cache.
2501 * the caller is expected to unpin it and allow it to be merged
2504 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2505 file_pos + num_bytes, NULL, 0,
2506 1, sizeof(*fi), &extent_inserted);
2510 if (!extent_inserted) {
2511 ins.objectid = btrfs_ino(BTRFS_I(inode));
2512 ins.offset = file_pos;
2513 ins.type = BTRFS_EXTENT_DATA_KEY;
2515 path->leave_spinning = 1;
2516 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2521 leaf = path->nodes[0];
2522 fi = btrfs_item_ptr(leaf, path->slots[0],
2523 struct btrfs_file_extent_item);
2524 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2525 btrfs_set_file_extent_type(leaf, fi, extent_type);
2526 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2527 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2528 btrfs_set_file_extent_offset(leaf, fi, 0);
2529 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2530 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2531 btrfs_set_file_extent_compression(leaf, fi, compression);
2532 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2533 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2535 btrfs_mark_buffer_dirty(leaf);
2536 btrfs_release_path(path);
2538 inode_add_bytes(inode, num_bytes);
2540 ins.objectid = disk_bytenr;
2541 ins.offset = disk_num_bytes;
2542 ins.type = BTRFS_EXTENT_ITEM_KEY;
2544 ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), file_pos,
2550 * Release the reserved range from inode dirty range map, as it is
2551 * already moved into delayed_ref_head
2553 ret = btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2557 ret = btrfs_alloc_reserved_file_extent(trans, root,
2558 btrfs_ino(BTRFS_I(inode)),
2559 file_pos, qg_released, &ins);
2561 btrfs_free_path(path);
2566 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2569 struct btrfs_block_group *cache;
2571 cache = btrfs_lookup_block_group(fs_info, start);
2574 spin_lock(&cache->lock);
2575 cache->delalloc_bytes -= len;
2576 spin_unlock(&cache->lock);
2578 btrfs_put_block_group(cache);
2581 /* as ordered data IO finishes, this gets called so we can finish
2582 * an ordered extent if the range of bytes in the file it covers are
2585 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2587 struct inode *inode = ordered_extent->inode;
2588 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2589 struct btrfs_root *root = BTRFS_I(inode)->root;
2590 struct btrfs_trans_handle *trans = NULL;
2591 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2592 struct extent_state *cached_state = NULL;
2594 int compress_type = 0;
2596 u64 logical_len = ordered_extent->num_bytes;
2597 bool freespace_inode;
2598 bool truncated = false;
2599 bool range_locked = false;
2600 bool clear_new_delalloc_bytes = false;
2601 bool clear_reserved_extent = true;
2602 unsigned int clear_bits;
2604 start = ordered_extent->file_offset;
2605 end = start + ordered_extent->num_bytes - 1;
2607 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2608 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2609 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2610 clear_new_delalloc_bytes = true;
2612 freespace_inode = btrfs_is_free_space_inode(BTRFS_I(inode));
2614 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2619 btrfs_free_io_failure_record(BTRFS_I(inode), start, end);
2621 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2623 logical_len = ordered_extent->truncated_len;
2624 /* Truncated the entire extent, don't bother adding */
2629 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2630 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2633 * For mwrite(mmap + memset to write) case, we still reserve
2634 * space for NOCOW range.
2635 * As NOCOW won't cause a new delayed ref, just free the space
2637 btrfs_qgroup_free_data(inode, NULL, start,
2638 ordered_extent->num_bytes);
2639 btrfs_inode_safe_disk_i_size_write(inode, 0);
2640 if (freespace_inode)
2641 trans = btrfs_join_transaction_spacecache(root);
2643 trans = btrfs_join_transaction(root);
2644 if (IS_ERR(trans)) {
2645 ret = PTR_ERR(trans);
2649 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
2650 ret = btrfs_update_inode_fallback(trans, root, inode);
2651 if (ret) /* -ENOMEM or corruption */
2652 btrfs_abort_transaction(trans, ret);
2656 range_locked = true;
2657 lock_extent_bits(io_tree, start, end, &cached_state);
2659 if (freespace_inode)
2660 trans = btrfs_join_transaction_spacecache(root);
2662 trans = btrfs_join_transaction(root);
2663 if (IS_ERR(trans)) {
2664 ret = PTR_ERR(trans);
2669 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
2671 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2672 compress_type = ordered_extent->compress_type;
2673 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2674 BUG_ON(compress_type);
2675 btrfs_qgroup_free_data(inode, NULL, start,
2676 ordered_extent->num_bytes);
2677 ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
2678 ordered_extent->file_offset,
2679 ordered_extent->file_offset +
2682 BUG_ON(root == fs_info->tree_root);
2683 ret = insert_reserved_file_extent(trans, inode, start,
2684 ordered_extent->disk_bytenr,
2685 ordered_extent->disk_num_bytes,
2686 logical_len, logical_len,
2687 compress_type, 0, 0,
2688 BTRFS_FILE_EXTENT_REG);
2690 clear_reserved_extent = false;
2691 btrfs_release_delalloc_bytes(fs_info,
2692 ordered_extent->disk_bytenr,
2693 ordered_extent->disk_num_bytes);
2696 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2697 ordered_extent->file_offset,
2698 ordered_extent->num_bytes, trans->transid);
2700 btrfs_abort_transaction(trans, ret);
2704 ret = add_pending_csums(trans, inode, &ordered_extent->list);
2706 btrfs_abort_transaction(trans, ret);
2710 btrfs_inode_safe_disk_i_size_write(inode, 0);
2711 ret = btrfs_update_inode_fallback(trans, root, inode);
2712 if (ret) { /* -ENOMEM or corruption */
2713 btrfs_abort_transaction(trans, ret);
2718 clear_bits = EXTENT_DEFRAG;
2720 clear_bits |= EXTENT_LOCKED;
2721 if (clear_new_delalloc_bytes)
2722 clear_bits |= EXTENT_DELALLOC_NEW;
2723 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, clear_bits,
2724 (clear_bits & EXTENT_LOCKED) ? 1 : 0, 0,
2728 btrfs_end_transaction(trans);
2730 if (ret || truncated) {
2731 u64 unwritten_start = start;
2734 unwritten_start += logical_len;
2735 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
2737 /* Drop the cache for the part of the extent we didn't write. */
2738 btrfs_drop_extent_cache(BTRFS_I(inode), unwritten_start, end, 0);
2741 * If the ordered extent had an IOERR or something else went
2742 * wrong we need to return the space for this ordered extent
2743 * back to the allocator. We only free the extent in the
2744 * truncated case if we didn't write out the extent at all.
2746 * If we made it past insert_reserved_file_extent before we
2747 * errored out then we don't need to do this as the accounting
2748 * has already been done.
2750 if ((ret || !logical_len) &&
2751 clear_reserved_extent &&
2752 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2753 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2755 * Discard the range before returning it back to the
2758 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
2759 btrfs_discard_extent(fs_info,
2760 ordered_extent->disk_bytenr,
2761 ordered_extent->disk_num_bytes,
2763 btrfs_free_reserved_extent(fs_info,
2764 ordered_extent->disk_bytenr,
2765 ordered_extent->disk_num_bytes, 1);
2770 * This needs to be done to make sure anybody waiting knows we are done
2771 * updating everything for this ordered extent.
2773 btrfs_remove_ordered_extent(inode, ordered_extent);
2776 btrfs_put_ordered_extent(ordered_extent);
2777 /* once for the tree */
2778 btrfs_put_ordered_extent(ordered_extent);
2783 static void finish_ordered_fn(struct btrfs_work *work)
2785 struct btrfs_ordered_extent *ordered_extent;
2786 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2787 btrfs_finish_ordered_io(ordered_extent);
2790 void btrfs_writepage_endio_finish_ordered(struct page *page, u64 start,
2791 u64 end, int uptodate)
2793 struct inode *inode = page->mapping->host;
2794 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2795 struct btrfs_ordered_extent *ordered_extent = NULL;
2796 struct btrfs_workqueue *wq;
2798 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2800 ClearPagePrivate2(page);
2801 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2802 end - start + 1, uptodate))
2805 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
2806 wq = fs_info->endio_freespace_worker;
2808 wq = fs_info->endio_write_workers;
2810 btrfs_init_work(&ordered_extent->work, finish_ordered_fn, NULL, NULL);
2811 btrfs_queue_work(wq, &ordered_extent->work);
2814 static int check_data_csum(struct inode *inode, struct btrfs_io_bio *io_bio,
2815 int icsum, struct page *page, int pgoff, u64 start,
2818 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2819 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
2821 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2823 u8 csum[BTRFS_CSUM_SIZE];
2825 csum_expected = ((u8 *)io_bio->csum) + icsum * csum_size;
2827 kaddr = kmap_atomic(page);
2828 shash->tfm = fs_info->csum_shash;
2830 crypto_shash_digest(shash, kaddr + pgoff, len, csum);
2832 if (memcmp(csum, csum_expected, csum_size))
2835 kunmap_atomic(kaddr);
2838 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
2839 io_bio->mirror_num);
2840 memset(kaddr + pgoff, 1, len);
2841 flush_dcache_page(page);
2842 kunmap_atomic(kaddr);
2847 * when reads are done, we need to check csums to verify the data is correct
2848 * if there's a match, we allow the bio to finish. If not, the code in
2849 * extent_io.c will try to find good copies for us.
2851 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
2852 u64 phy_offset, struct page *page,
2853 u64 start, u64 end, int mirror)
2855 size_t offset = start - page_offset(page);
2856 struct inode *inode = page->mapping->host;
2857 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2858 struct btrfs_root *root = BTRFS_I(inode)->root;
2860 if (PageChecked(page)) {
2861 ClearPageChecked(page);
2865 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2868 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2869 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2870 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
2874 phy_offset >>= inode->i_sb->s_blocksize_bits;
2875 return check_data_csum(inode, io_bio, phy_offset, page, offset, start,
2876 (size_t)(end - start + 1));
2880 * btrfs_add_delayed_iput - perform a delayed iput on @inode
2882 * @inode: The inode we want to perform iput on
2884 * This function uses the generic vfs_inode::i_count to track whether we should
2885 * just decrement it (in case it's > 1) or if this is the last iput then link
2886 * the inode to the delayed iput machinery. Delayed iputs are processed at
2887 * transaction commit time/superblock commit/cleaner kthread.
2889 void btrfs_add_delayed_iput(struct inode *inode)
2891 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2892 struct btrfs_inode *binode = BTRFS_I(inode);
2894 if (atomic_add_unless(&inode->i_count, -1, 1))
2897 atomic_inc(&fs_info->nr_delayed_iputs);
2898 spin_lock(&fs_info->delayed_iput_lock);
2899 ASSERT(list_empty(&binode->delayed_iput));
2900 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
2901 spin_unlock(&fs_info->delayed_iput_lock);
2902 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
2903 wake_up_process(fs_info->cleaner_kthread);
2906 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
2907 struct btrfs_inode *inode)
2909 list_del_init(&inode->delayed_iput);
2910 spin_unlock(&fs_info->delayed_iput_lock);
2911 iput(&inode->vfs_inode);
2912 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
2913 wake_up(&fs_info->delayed_iputs_wait);
2914 spin_lock(&fs_info->delayed_iput_lock);
2917 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
2918 struct btrfs_inode *inode)
2920 if (!list_empty(&inode->delayed_iput)) {
2921 spin_lock(&fs_info->delayed_iput_lock);
2922 if (!list_empty(&inode->delayed_iput))
2923 run_delayed_iput_locked(fs_info, inode);
2924 spin_unlock(&fs_info->delayed_iput_lock);
2928 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
2931 spin_lock(&fs_info->delayed_iput_lock);
2932 while (!list_empty(&fs_info->delayed_iputs)) {
2933 struct btrfs_inode *inode;
2935 inode = list_first_entry(&fs_info->delayed_iputs,
2936 struct btrfs_inode, delayed_iput);
2937 run_delayed_iput_locked(fs_info, inode);
2939 spin_unlock(&fs_info->delayed_iput_lock);
2943 * btrfs_wait_on_delayed_iputs - wait on the delayed iputs to be done running
2944 * @fs_info - the fs_info for this fs
2945 * @return - EINTR if we were killed, 0 if nothing's pending
2947 * This will wait on any delayed iputs that are currently running with KILLABLE
2948 * set. Once they are all done running we will return, unless we are killed in
2949 * which case we return EINTR. This helps in user operations like fallocate etc
2950 * that might get blocked on the iputs.
2952 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
2954 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
2955 atomic_read(&fs_info->nr_delayed_iputs) == 0);
2962 * This creates an orphan entry for the given inode in case something goes wrong
2963 * in the middle of an unlink.
2965 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
2966 struct btrfs_inode *inode)
2970 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
2971 if (ret && ret != -EEXIST) {
2972 btrfs_abort_transaction(trans, ret);
2980 * We have done the delete so we can go ahead and remove the orphan item for
2981 * this particular inode.
2983 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
2984 struct btrfs_inode *inode)
2986 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
2990 * this cleans up any orphans that may be left on the list from the last use
2993 int btrfs_orphan_cleanup(struct btrfs_root *root)
2995 struct btrfs_fs_info *fs_info = root->fs_info;
2996 struct btrfs_path *path;
2997 struct extent_buffer *leaf;
2998 struct btrfs_key key, found_key;
2999 struct btrfs_trans_handle *trans;
3000 struct inode *inode;
3001 u64 last_objectid = 0;
3002 int ret = 0, nr_unlink = 0;
3004 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3007 path = btrfs_alloc_path();
3012 path->reada = READA_BACK;
3014 key.objectid = BTRFS_ORPHAN_OBJECTID;
3015 key.type = BTRFS_ORPHAN_ITEM_KEY;
3016 key.offset = (u64)-1;
3019 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3024 * if ret == 0 means we found what we were searching for, which
3025 * is weird, but possible, so only screw with path if we didn't
3026 * find the key and see if we have stuff that matches
3030 if (path->slots[0] == 0)
3035 /* pull out the item */
3036 leaf = path->nodes[0];
3037 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3039 /* make sure the item matches what we want */
3040 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3042 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3045 /* release the path since we're done with it */
3046 btrfs_release_path(path);
3049 * this is where we are basically btrfs_lookup, without the
3050 * crossing root thing. we store the inode number in the
3051 * offset of the orphan item.
3054 if (found_key.offset == last_objectid) {
3056 "Error removing orphan entry, stopping orphan cleanup");
3061 last_objectid = found_key.offset;
3063 found_key.objectid = found_key.offset;
3064 found_key.type = BTRFS_INODE_ITEM_KEY;
3065 found_key.offset = 0;
3066 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3067 ret = PTR_ERR_OR_ZERO(inode);
3068 if (ret && ret != -ENOENT)
3071 if (ret == -ENOENT && root == fs_info->tree_root) {
3072 struct btrfs_root *dead_root;
3073 struct btrfs_fs_info *fs_info = root->fs_info;
3074 int is_dead_root = 0;
3077 * this is an orphan in the tree root. Currently these
3078 * could come from 2 sources:
3079 * a) a snapshot deletion in progress
3080 * b) a free space cache inode
3081 * We need to distinguish those two, as the snapshot
3082 * orphan must not get deleted.
3083 * find_dead_roots already ran before us, so if this
3084 * is a snapshot deletion, we should find the root
3085 * in the fs_roots radix tree.
3088 spin_lock(&fs_info->fs_roots_radix_lock);
3089 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3090 (unsigned long)found_key.objectid);
3091 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3093 spin_unlock(&fs_info->fs_roots_radix_lock);
3096 /* prevent this orphan from being found again */
3097 key.offset = found_key.objectid - 1;
3104 * If we have an inode with links, there are a couple of
3105 * possibilities. Old kernels (before v3.12) used to create an
3106 * orphan item for truncate indicating that there were possibly
3107 * extent items past i_size that needed to be deleted. In v3.12,
3108 * truncate was changed to update i_size in sync with the extent
3109 * items, but the (useless) orphan item was still created. Since
3110 * v4.18, we don't create the orphan item for truncate at all.
3112 * So, this item could mean that we need to do a truncate, but
3113 * only if this filesystem was last used on a pre-v3.12 kernel
3114 * and was not cleanly unmounted. The odds of that are quite
3115 * slim, and it's a pain to do the truncate now, so just delete
3118 * It's also possible that this orphan item was supposed to be
3119 * deleted but wasn't. The inode number may have been reused,
3120 * but either way, we can delete the orphan item.
3122 if (ret == -ENOENT || inode->i_nlink) {
3125 trans = btrfs_start_transaction(root, 1);
3126 if (IS_ERR(trans)) {
3127 ret = PTR_ERR(trans);
3130 btrfs_debug(fs_info, "auto deleting %Lu",
3131 found_key.objectid);
3132 ret = btrfs_del_orphan_item(trans, root,
3133 found_key.objectid);
3134 btrfs_end_transaction(trans);
3142 /* this will do delete_inode and everything for us */
3145 /* release the path since we're done with it */
3146 btrfs_release_path(path);
3148 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3150 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3151 trans = btrfs_join_transaction(root);
3153 btrfs_end_transaction(trans);
3157 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3161 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3162 btrfs_free_path(path);
3167 * very simple check to peek ahead in the leaf looking for xattrs. If we
3168 * don't find any xattrs, we know there can't be any acls.
3170 * slot is the slot the inode is in, objectid is the objectid of the inode
3172 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3173 int slot, u64 objectid,
3174 int *first_xattr_slot)
3176 u32 nritems = btrfs_header_nritems(leaf);
3177 struct btrfs_key found_key;
3178 static u64 xattr_access = 0;
3179 static u64 xattr_default = 0;
3182 if (!xattr_access) {
3183 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3184 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3185 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3186 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3190 *first_xattr_slot = -1;
3191 while (slot < nritems) {
3192 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3194 /* we found a different objectid, there must not be acls */
3195 if (found_key.objectid != objectid)
3198 /* we found an xattr, assume we've got an acl */
3199 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3200 if (*first_xattr_slot == -1)
3201 *first_xattr_slot = slot;
3202 if (found_key.offset == xattr_access ||
3203 found_key.offset == xattr_default)
3208 * we found a key greater than an xattr key, there can't
3209 * be any acls later on
3211 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3218 * it goes inode, inode backrefs, xattrs, extents,
3219 * so if there are a ton of hard links to an inode there can
3220 * be a lot of backrefs. Don't waste time searching too hard,
3221 * this is just an optimization
3226 /* we hit the end of the leaf before we found an xattr or
3227 * something larger than an xattr. We have to assume the inode
3230 if (*first_xattr_slot == -1)
3231 *first_xattr_slot = slot;
3236 * read an inode from the btree into the in-memory inode
3238 static int btrfs_read_locked_inode(struct inode *inode,
3239 struct btrfs_path *in_path)
3241 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3242 struct btrfs_path *path = in_path;
3243 struct extent_buffer *leaf;
3244 struct btrfs_inode_item *inode_item;
3245 struct btrfs_root *root = BTRFS_I(inode)->root;
3246 struct btrfs_key location;
3251 bool filled = false;
3252 int first_xattr_slot;
3254 ret = btrfs_fill_inode(inode, &rdev);
3259 path = btrfs_alloc_path();
3264 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3266 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3268 if (path != in_path)
3269 btrfs_free_path(path);
3273 leaf = path->nodes[0];
3278 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3279 struct btrfs_inode_item);
3280 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3281 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3282 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3283 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3284 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3285 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3286 round_up(i_size_read(inode), fs_info->sectorsize));
3288 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3289 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3291 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3292 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3294 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3295 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3297 BTRFS_I(inode)->i_otime.tv_sec =
3298 btrfs_timespec_sec(leaf, &inode_item->otime);
3299 BTRFS_I(inode)->i_otime.tv_nsec =
3300 btrfs_timespec_nsec(leaf, &inode_item->otime);
3302 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3303 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3304 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3306 inode_set_iversion_queried(inode,
3307 btrfs_inode_sequence(leaf, inode_item));
3308 inode->i_generation = BTRFS_I(inode)->generation;
3310 rdev = btrfs_inode_rdev(leaf, inode_item);
3312 BTRFS_I(inode)->index_cnt = (u64)-1;
3313 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3317 * If we were modified in the current generation and evicted from memory
3318 * and then re-read we need to do a full sync since we don't have any
3319 * idea about which extents were modified before we were evicted from
3322 * This is required for both inode re-read from disk and delayed inode
3323 * in delayed_nodes_tree.
3325 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3326 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3327 &BTRFS_I(inode)->runtime_flags);
3330 * We don't persist the id of the transaction where an unlink operation
3331 * against the inode was last made. So here we assume the inode might
3332 * have been evicted, and therefore the exact value of last_unlink_trans
3333 * lost, and set it to last_trans to avoid metadata inconsistencies
3334 * between the inode and its parent if the inode is fsync'ed and the log
3335 * replayed. For example, in the scenario:
3338 * ln mydir/foo mydir/bar
3341 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3342 * xfs_io -c fsync mydir/foo
3344 * mount fs, triggers fsync log replay
3346 * We must make sure that when we fsync our inode foo we also log its
3347 * parent inode, otherwise after log replay the parent still has the
3348 * dentry with the "bar" name but our inode foo has a link count of 1
3349 * and doesn't have an inode ref with the name "bar" anymore.
3351 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3352 * but it guarantees correctness at the expense of occasional full
3353 * transaction commits on fsync if our inode is a directory, or if our
3354 * inode is not a directory, logging its parent unnecessarily.
3356 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3359 if (inode->i_nlink != 1 ||
3360 path->slots[0] >= btrfs_header_nritems(leaf))
3363 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3364 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3367 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3368 if (location.type == BTRFS_INODE_REF_KEY) {
3369 struct btrfs_inode_ref *ref;
3371 ref = (struct btrfs_inode_ref *)ptr;
3372 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3373 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3374 struct btrfs_inode_extref *extref;
3376 extref = (struct btrfs_inode_extref *)ptr;
3377 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3382 * try to precache a NULL acl entry for files that don't have
3383 * any xattrs or acls
3385 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3386 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3387 if (first_xattr_slot != -1) {
3388 path->slots[0] = first_xattr_slot;
3389 ret = btrfs_load_inode_props(inode, path);
3392 "error loading props for ino %llu (root %llu): %d",
3393 btrfs_ino(BTRFS_I(inode)),
3394 root->root_key.objectid, ret);
3396 if (path != in_path)
3397 btrfs_free_path(path);
3400 cache_no_acl(inode);
3402 switch (inode->i_mode & S_IFMT) {
3404 inode->i_mapping->a_ops = &btrfs_aops;
3405 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3406 inode->i_fop = &btrfs_file_operations;
3407 inode->i_op = &btrfs_file_inode_operations;
3410 inode->i_fop = &btrfs_dir_file_operations;
3411 inode->i_op = &btrfs_dir_inode_operations;
3414 inode->i_op = &btrfs_symlink_inode_operations;
3415 inode_nohighmem(inode);
3416 inode->i_mapping->a_ops = &btrfs_aops;
3419 inode->i_op = &btrfs_special_inode_operations;
3420 init_special_inode(inode, inode->i_mode, rdev);
3424 btrfs_sync_inode_flags_to_i_flags(inode);
3429 * given a leaf and an inode, copy the inode fields into the leaf
3431 static void fill_inode_item(struct btrfs_trans_handle *trans,
3432 struct extent_buffer *leaf,
3433 struct btrfs_inode_item *item,
3434 struct inode *inode)
3436 struct btrfs_map_token token;
3438 btrfs_init_map_token(&token, leaf);
3440 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3441 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3442 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
3443 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3444 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3446 btrfs_set_token_timespec_sec(&token, &item->atime,
3447 inode->i_atime.tv_sec);
3448 btrfs_set_token_timespec_nsec(&token, &item->atime,
3449 inode->i_atime.tv_nsec);
3451 btrfs_set_token_timespec_sec(&token, &item->mtime,
3452 inode->i_mtime.tv_sec);
3453 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3454 inode->i_mtime.tv_nsec);
3456 btrfs_set_token_timespec_sec(&token, &item->ctime,
3457 inode->i_ctime.tv_sec);
3458 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3459 inode->i_ctime.tv_nsec);
3461 btrfs_set_token_timespec_sec(&token, &item->otime,
3462 BTRFS_I(inode)->i_otime.tv_sec);
3463 btrfs_set_token_timespec_nsec(&token, &item->otime,
3464 BTRFS_I(inode)->i_otime.tv_nsec);
3466 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
3467 btrfs_set_token_inode_generation(&token, item,
3468 BTRFS_I(inode)->generation);
3469 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
3470 btrfs_set_token_inode_transid(&token, item, trans->transid);
3471 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
3472 btrfs_set_token_inode_flags(&token, item, BTRFS_I(inode)->flags);
3473 btrfs_set_token_inode_block_group(&token, item, 0);
3477 * copy everything in the in-memory inode into the btree.
3479 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3480 struct btrfs_root *root, struct inode *inode)
3482 struct btrfs_inode_item *inode_item;
3483 struct btrfs_path *path;
3484 struct extent_buffer *leaf;
3487 path = btrfs_alloc_path();
3491 path->leave_spinning = 1;
3492 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3500 leaf = path->nodes[0];
3501 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3502 struct btrfs_inode_item);
3504 fill_inode_item(trans, leaf, inode_item, inode);
3505 btrfs_mark_buffer_dirty(leaf);
3506 btrfs_set_inode_last_trans(trans, inode);
3509 btrfs_free_path(path);
3514 * copy everything in the in-memory inode into the btree.
3516 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3517 struct btrfs_root *root, struct inode *inode)
3519 struct btrfs_fs_info *fs_info = root->fs_info;
3523 * If the inode is a free space inode, we can deadlock during commit
3524 * if we put it into the delayed code.
3526 * The data relocation inode should also be directly updated
3529 if (!btrfs_is_free_space_inode(BTRFS_I(inode))
3530 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3531 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
3532 btrfs_update_root_times(trans, root);
3534 ret = btrfs_delayed_update_inode(trans, root, inode);
3536 btrfs_set_inode_last_trans(trans, inode);
3540 return btrfs_update_inode_item(trans, root, inode);
3543 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3544 struct btrfs_root *root,
3545 struct inode *inode)
3549 ret = btrfs_update_inode(trans, root, inode);
3551 return btrfs_update_inode_item(trans, root, inode);
3556 * unlink helper that gets used here in inode.c and in the tree logging
3557 * recovery code. It remove a link in a directory with a given name, and
3558 * also drops the back refs in the inode to the directory
3560 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3561 struct btrfs_root *root,
3562 struct btrfs_inode *dir,
3563 struct btrfs_inode *inode,
3564 const char *name, int name_len)
3566 struct btrfs_fs_info *fs_info = root->fs_info;
3567 struct btrfs_path *path;
3569 struct btrfs_dir_item *di;
3571 u64 ino = btrfs_ino(inode);
3572 u64 dir_ino = btrfs_ino(dir);
3574 path = btrfs_alloc_path();
3580 path->leave_spinning = 1;
3581 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3582 name, name_len, -1);
3583 if (IS_ERR_OR_NULL(di)) {
3584 ret = di ? PTR_ERR(di) : -ENOENT;
3587 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3590 btrfs_release_path(path);
3593 * If we don't have dir index, we have to get it by looking up
3594 * the inode ref, since we get the inode ref, remove it directly,
3595 * it is unnecessary to do delayed deletion.
3597 * But if we have dir index, needn't search inode ref to get it.
3598 * Since the inode ref is close to the inode item, it is better
3599 * that we delay to delete it, and just do this deletion when
3600 * we update the inode item.
3602 if (inode->dir_index) {
3603 ret = btrfs_delayed_delete_inode_ref(inode);
3605 index = inode->dir_index;
3610 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3614 "failed to delete reference to %.*s, inode %llu parent %llu",
3615 name_len, name, ino, dir_ino);
3616 btrfs_abort_transaction(trans, ret);
3620 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
3622 btrfs_abort_transaction(trans, ret);
3626 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
3628 if (ret != 0 && ret != -ENOENT) {
3629 btrfs_abort_transaction(trans, ret);
3633 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
3638 btrfs_abort_transaction(trans, ret);
3641 * If we have a pending delayed iput we could end up with the final iput
3642 * being run in btrfs-cleaner context. If we have enough of these built
3643 * up we can end up burning a lot of time in btrfs-cleaner without any
3644 * way to throttle the unlinks. Since we're currently holding a ref on
3645 * the inode we can run the delayed iput here without any issues as the
3646 * final iput won't be done until after we drop the ref we're currently
3649 btrfs_run_delayed_iput(fs_info, inode);
3651 btrfs_free_path(path);
3655 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
3656 inode_inc_iversion(&inode->vfs_inode);
3657 inode_inc_iversion(&dir->vfs_inode);
3658 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
3659 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
3660 ret = btrfs_update_inode(trans, root, &dir->vfs_inode);
3665 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3666 struct btrfs_root *root,
3667 struct btrfs_inode *dir, struct btrfs_inode *inode,
3668 const char *name, int name_len)
3671 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3673 drop_nlink(&inode->vfs_inode);
3674 ret = btrfs_update_inode(trans, root, &inode->vfs_inode);
3680 * helper to start transaction for unlink and rmdir.
3682 * unlink and rmdir are special in btrfs, they do not always free space, so
3683 * if we cannot make our reservations the normal way try and see if there is
3684 * plenty of slack room in the global reserve to migrate, otherwise we cannot
3685 * allow the unlink to occur.
3687 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3689 struct btrfs_root *root = BTRFS_I(dir)->root;
3692 * 1 for the possible orphan item
3693 * 1 for the dir item
3694 * 1 for the dir index
3695 * 1 for the inode ref
3698 return btrfs_start_transaction_fallback_global_rsv(root, 5);
3701 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3703 struct btrfs_root *root = BTRFS_I(dir)->root;
3704 struct btrfs_trans_handle *trans;
3705 struct inode *inode = d_inode(dentry);
3708 trans = __unlink_start_trans(dir);
3710 return PTR_ERR(trans);
3712 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
3715 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
3716 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
3717 dentry->d_name.len);
3721 if (inode->i_nlink == 0) {
3722 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
3728 btrfs_end_transaction(trans);
3729 btrfs_btree_balance_dirty(root->fs_info);
3733 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3734 struct inode *dir, struct dentry *dentry)
3736 struct btrfs_root *root = BTRFS_I(dir)->root;
3737 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
3738 struct btrfs_path *path;
3739 struct extent_buffer *leaf;
3740 struct btrfs_dir_item *di;
3741 struct btrfs_key key;
3742 const char *name = dentry->d_name.name;
3743 int name_len = dentry->d_name.len;
3747 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
3749 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
3750 objectid = inode->root->root_key.objectid;
3751 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
3752 objectid = inode->location.objectid;
3758 path = btrfs_alloc_path();
3762 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3763 name, name_len, -1);
3764 if (IS_ERR_OR_NULL(di)) {
3765 ret = di ? PTR_ERR(di) : -ENOENT;
3769 leaf = path->nodes[0];
3770 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3771 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3772 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3774 btrfs_abort_transaction(trans, ret);
3777 btrfs_release_path(path);
3780 * This is a placeholder inode for a subvolume we didn't have a
3781 * reference to at the time of the snapshot creation. In the meantime
3782 * we could have renamed the real subvol link into our snapshot, so
3783 * depending on btrfs_del_root_ref to return -ENOENT here is incorret.
3784 * Instead simply lookup the dir_index_item for this entry so we can
3785 * remove it. Otherwise we know we have a ref to the root and we can
3786 * call btrfs_del_root_ref, and it _shouldn't_ fail.
3788 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
3789 di = btrfs_search_dir_index_item(root, path, dir_ino,
3791 if (IS_ERR_OR_NULL(di)) {
3796 btrfs_abort_transaction(trans, ret);
3800 leaf = path->nodes[0];
3801 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3803 btrfs_release_path(path);
3805 ret = btrfs_del_root_ref(trans, objectid,
3806 root->root_key.objectid, dir_ino,
3807 &index, name, name_len);
3809 btrfs_abort_transaction(trans, ret);
3814 ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index);
3816 btrfs_abort_transaction(trans, ret);
3820 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
3821 inode_inc_iversion(dir);
3822 dir->i_mtime = dir->i_ctime = current_time(dir);
3823 ret = btrfs_update_inode_fallback(trans, root, dir);
3825 btrfs_abort_transaction(trans, ret);
3827 btrfs_free_path(path);
3832 * Helper to check if the subvolume references other subvolumes or if it's
3835 static noinline int may_destroy_subvol(struct btrfs_root *root)
3837 struct btrfs_fs_info *fs_info = root->fs_info;
3838 struct btrfs_path *path;
3839 struct btrfs_dir_item *di;
3840 struct btrfs_key key;
3844 path = btrfs_alloc_path();
3848 /* Make sure this root isn't set as the default subvol */
3849 dir_id = btrfs_super_root_dir(fs_info->super_copy);
3850 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
3851 dir_id, "default", 7, 0);
3852 if (di && !IS_ERR(di)) {
3853 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
3854 if (key.objectid == root->root_key.objectid) {
3857 "deleting default subvolume %llu is not allowed",
3861 btrfs_release_path(path);
3864 key.objectid = root->root_key.objectid;
3865 key.type = BTRFS_ROOT_REF_KEY;
3866 key.offset = (u64)-1;
3868 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3874 if (path->slots[0] > 0) {
3876 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3877 if (key.objectid == root->root_key.objectid &&
3878 key.type == BTRFS_ROOT_REF_KEY)
3882 btrfs_free_path(path);
3886 /* Delete all dentries for inodes belonging to the root */
3887 static void btrfs_prune_dentries(struct btrfs_root *root)
3889 struct btrfs_fs_info *fs_info = root->fs_info;
3890 struct rb_node *node;
3891 struct rb_node *prev;
3892 struct btrfs_inode *entry;
3893 struct inode *inode;
3896 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3897 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3899 spin_lock(&root->inode_lock);
3901 node = root->inode_tree.rb_node;
3905 entry = rb_entry(node, struct btrfs_inode, rb_node);
3907 if (objectid < btrfs_ino(entry))
3908 node = node->rb_left;
3909 else if (objectid > btrfs_ino(entry))
3910 node = node->rb_right;
3916 entry = rb_entry(prev, struct btrfs_inode, rb_node);
3917 if (objectid <= btrfs_ino(entry)) {
3921 prev = rb_next(prev);
3925 entry = rb_entry(node, struct btrfs_inode, rb_node);
3926 objectid = btrfs_ino(entry) + 1;
3927 inode = igrab(&entry->vfs_inode);
3929 spin_unlock(&root->inode_lock);
3930 if (atomic_read(&inode->i_count) > 1)
3931 d_prune_aliases(inode);
3933 * btrfs_drop_inode will have it removed from the inode
3934 * cache when its usage count hits zero.
3938 spin_lock(&root->inode_lock);
3942 if (cond_resched_lock(&root->inode_lock))
3945 node = rb_next(node);
3947 spin_unlock(&root->inode_lock);
3950 int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
3952 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
3953 struct btrfs_root *root = BTRFS_I(dir)->root;
3954 struct inode *inode = d_inode(dentry);
3955 struct btrfs_root *dest = BTRFS_I(inode)->root;
3956 struct btrfs_trans_handle *trans;
3957 struct btrfs_block_rsv block_rsv;
3963 * Don't allow to delete a subvolume with send in progress. This is
3964 * inside the inode lock so the error handling that has to drop the bit
3965 * again is not run concurrently.
3967 spin_lock(&dest->root_item_lock);
3968 if (dest->send_in_progress) {
3969 spin_unlock(&dest->root_item_lock);
3971 "attempt to delete subvolume %llu during send",
3972 dest->root_key.objectid);
3975 root_flags = btrfs_root_flags(&dest->root_item);
3976 btrfs_set_root_flags(&dest->root_item,
3977 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
3978 spin_unlock(&dest->root_item_lock);
3980 down_write(&fs_info->subvol_sem);
3982 err = may_destroy_subvol(dest);
3986 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
3988 * One for dir inode,
3989 * two for dir entries,
3990 * two for root ref/backref.
3992 err = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
3996 trans = btrfs_start_transaction(root, 0);
3997 if (IS_ERR(trans)) {
3998 err = PTR_ERR(trans);
4001 trans->block_rsv = &block_rsv;
4002 trans->bytes_reserved = block_rsv.size;
4004 btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
4006 ret = btrfs_unlink_subvol(trans, dir, dentry);
4009 btrfs_abort_transaction(trans, ret);
4013 btrfs_record_root_in_trans(trans, dest);
4015 memset(&dest->root_item.drop_progress, 0,
4016 sizeof(dest->root_item.drop_progress));
4017 dest->root_item.drop_level = 0;
4018 btrfs_set_root_refs(&dest->root_item, 0);
4020 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4021 ret = btrfs_insert_orphan_item(trans,
4023 dest->root_key.objectid);
4025 btrfs_abort_transaction(trans, ret);
4031 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4032 BTRFS_UUID_KEY_SUBVOL,
4033 dest->root_key.objectid);
4034 if (ret && ret != -ENOENT) {
4035 btrfs_abort_transaction(trans, ret);
4039 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4040 ret = btrfs_uuid_tree_remove(trans,
4041 dest->root_item.received_uuid,
4042 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4043 dest->root_key.objectid);
4044 if (ret && ret != -ENOENT) {
4045 btrfs_abort_transaction(trans, ret);
4052 trans->block_rsv = NULL;
4053 trans->bytes_reserved = 0;
4054 ret = btrfs_end_transaction(trans);
4057 inode->i_flags |= S_DEAD;
4059 btrfs_subvolume_release_metadata(fs_info, &block_rsv);
4061 up_write(&fs_info->subvol_sem);
4063 spin_lock(&dest->root_item_lock);
4064 root_flags = btrfs_root_flags(&dest->root_item);
4065 btrfs_set_root_flags(&dest->root_item,
4066 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4067 spin_unlock(&dest->root_item_lock);
4069 d_invalidate(dentry);
4070 btrfs_prune_dentries(dest);
4071 ASSERT(dest->send_in_progress == 0);
4074 if (dest->ino_cache_inode) {
4075 iput(dest->ino_cache_inode);
4076 dest->ino_cache_inode = NULL;
4083 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4085 struct inode *inode = d_inode(dentry);
4087 struct btrfs_root *root = BTRFS_I(dir)->root;
4088 struct btrfs_trans_handle *trans;
4089 u64 last_unlink_trans;
4091 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4093 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4094 return btrfs_delete_subvolume(dir, dentry);
4096 trans = __unlink_start_trans(dir);
4098 return PTR_ERR(trans);
4100 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4101 err = btrfs_unlink_subvol(trans, dir, dentry);
4105 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4109 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4111 /* now the directory is empty */
4112 err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4113 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4114 dentry->d_name.len);
4116 btrfs_i_size_write(BTRFS_I(inode), 0);
4118 * Propagate the last_unlink_trans value of the deleted dir to
4119 * its parent directory. This is to prevent an unrecoverable
4120 * log tree in the case we do something like this:
4122 * 2) create snapshot under dir foo
4123 * 3) delete the snapshot
4126 * 6) fsync foo or some file inside foo
4128 if (last_unlink_trans >= trans->transid)
4129 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4132 btrfs_end_transaction(trans);
4133 btrfs_btree_balance_dirty(root->fs_info);
4139 * Return this if we need to call truncate_block for the last bit of the
4142 #define NEED_TRUNCATE_BLOCK 1
4145 * this can truncate away extent items, csum items and directory items.
4146 * It starts at a high offset and removes keys until it can't find
4147 * any higher than new_size
4149 * csum items that cross the new i_size are truncated to the new size
4152 * min_type is the minimum key type to truncate down to. If set to 0, this
4153 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4155 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4156 struct btrfs_root *root,
4157 struct inode *inode,
4158 u64 new_size, u32 min_type)
4160 struct btrfs_fs_info *fs_info = root->fs_info;
4161 struct btrfs_path *path;
4162 struct extent_buffer *leaf;
4163 struct btrfs_file_extent_item *fi;
4164 struct btrfs_key key;
4165 struct btrfs_key found_key;
4166 u64 extent_start = 0;
4167 u64 extent_num_bytes = 0;
4168 u64 extent_offset = 0;
4170 u64 last_size = new_size;
4171 u32 found_type = (u8)-1;
4174 int pending_del_nr = 0;
4175 int pending_del_slot = 0;
4176 int extent_type = -1;
4178 u64 ino = btrfs_ino(BTRFS_I(inode));
4179 u64 bytes_deleted = 0;
4180 bool be_nice = false;
4181 bool should_throttle = false;
4182 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
4183 struct extent_state *cached_state = NULL;
4185 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4188 * For non-free space inodes and non-shareable roots, we want to back
4189 * off from time to time. This means all inodes in subvolume roots,
4190 * reloc roots, and data reloc roots.
4192 if (!btrfs_is_free_space_inode(BTRFS_I(inode)) &&
4193 test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
4196 path = btrfs_alloc_path();
4199 path->reada = READA_BACK;
4201 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4202 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, (u64)-1,
4206 * We want to drop from the next block forward in case this
4207 * new size is not block aligned since we will be keeping the
4208 * last block of the extent just the way it is.
4210 btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size,
4211 fs_info->sectorsize),
4216 * This function is also used to drop the items in the log tree before
4217 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4218 * it is used to drop the logged items. So we shouldn't kill the delayed
4221 if (min_type == 0 && root == BTRFS_I(inode)->root)
4222 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
4225 key.offset = (u64)-1;
4230 * with a 16K leaf size and 128MB extents, you can actually queue
4231 * up a huge file in a single leaf. Most of the time that
4232 * bytes_deleted is > 0, it will be huge by the time we get here
4234 if (be_nice && bytes_deleted > SZ_32M &&
4235 btrfs_should_end_transaction(trans)) {
4240 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4246 /* there are no items in the tree for us to truncate, we're
4249 if (path->slots[0] == 0)
4255 u64 clear_start = 0, clear_len = 0;
4258 leaf = path->nodes[0];
4259 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4260 found_type = found_key.type;
4262 if (found_key.objectid != ino)
4265 if (found_type < min_type)
4268 item_end = found_key.offset;
4269 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4270 fi = btrfs_item_ptr(leaf, path->slots[0],
4271 struct btrfs_file_extent_item);
4272 extent_type = btrfs_file_extent_type(leaf, fi);
4273 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4275 btrfs_file_extent_num_bytes(leaf, fi);
4277 trace_btrfs_truncate_show_fi_regular(
4278 BTRFS_I(inode), leaf, fi,
4280 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4281 item_end += btrfs_file_extent_ram_bytes(leaf,
4284 trace_btrfs_truncate_show_fi_inline(
4285 BTRFS_I(inode), leaf, fi, path->slots[0],
4290 if (found_type > min_type) {
4293 if (item_end < new_size)
4295 if (found_key.offset >= new_size)
4301 /* FIXME, shrink the extent if the ref count is only 1 */
4302 if (found_type != BTRFS_EXTENT_DATA_KEY)
4305 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4308 clear_start = found_key.offset;
4309 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4311 u64 orig_num_bytes =
4312 btrfs_file_extent_num_bytes(leaf, fi);
4313 extent_num_bytes = ALIGN(new_size -
4315 fs_info->sectorsize);
4316 clear_start = ALIGN(new_size, fs_info->sectorsize);
4317 btrfs_set_file_extent_num_bytes(leaf, fi,
4319 num_dec = (orig_num_bytes -
4321 if (test_bit(BTRFS_ROOT_SHAREABLE,
4324 inode_sub_bytes(inode, num_dec);
4325 btrfs_mark_buffer_dirty(leaf);
4328 btrfs_file_extent_disk_num_bytes(leaf,
4330 extent_offset = found_key.offset -
4331 btrfs_file_extent_offset(leaf, fi);
4333 /* FIXME blocksize != 4096 */
4334 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4335 if (extent_start != 0) {
4337 if (test_bit(BTRFS_ROOT_SHAREABLE,
4339 inode_sub_bytes(inode, num_dec);
4342 clear_len = num_dec;
4343 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4345 * we can't truncate inline items that have had
4349 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4350 btrfs_file_extent_other_encoding(leaf, fi) == 0 &&
4351 btrfs_file_extent_compression(leaf, fi) == 0) {
4352 u32 size = (u32)(new_size - found_key.offset);
4354 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4355 size = btrfs_file_extent_calc_inline_size(size);
4356 btrfs_truncate_item(path, size, 1);
4357 } else if (!del_item) {
4359 * We have to bail so the last_size is set to
4360 * just before this extent.
4362 ret = NEED_TRUNCATE_BLOCK;
4366 * Inline extents are special, we just treat
4367 * them as a full sector worth in the file
4368 * extent tree just for simplicity sake.
4370 clear_len = fs_info->sectorsize;
4373 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
4374 inode_sub_bytes(inode, item_end + 1 - new_size);
4378 * We use btrfs_truncate_inode_items() to clean up log trees for
4379 * multiple fsyncs, and in this case we don't want to clear the
4380 * file extent range because it's just the log.
4382 if (root == BTRFS_I(inode)->root) {
4383 ret = btrfs_inode_clear_file_extent_range(BTRFS_I(inode),
4384 clear_start, clear_len);
4386 btrfs_abort_transaction(trans, ret);
4392 last_size = found_key.offset;
4394 last_size = new_size;
4396 if (!pending_del_nr) {
4397 /* no pending yet, add ourselves */
4398 pending_del_slot = path->slots[0];
4400 } else if (pending_del_nr &&
4401 path->slots[0] + 1 == pending_del_slot) {
4402 /* hop on the pending chunk */
4404 pending_del_slot = path->slots[0];
4411 should_throttle = false;
4414 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4415 struct btrfs_ref ref = { 0 };
4417 bytes_deleted += extent_num_bytes;
4419 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF,
4420 extent_start, extent_num_bytes, 0);
4421 ref.real_root = root->root_key.objectid;
4422 btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
4423 ino, extent_offset);
4424 ret = btrfs_free_extent(trans, &ref);
4426 btrfs_abort_transaction(trans, ret);
4430 if (btrfs_should_throttle_delayed_refs(trans))
4431 should_throttle = true;
4435 if (found_type == BTRFS_INODE_ITEM_KEY)
4438 if (path->slots[0] == 0 ||
4439 path->slots[0] != pending_del_slot ||
4441 if (pending_del_nr) {
4442 ret = btrfs_del_items(trans, root, path,
4446 btrfs_abort_transaction(trans, ret);
4451 btrfs_release_path(path);
4454 * We can generate a lot of delayed refs, so we need to
4455 * throttle every once and a while and make sure we're
4456 * adding enough space to keep up with the work we are
4457 * generating. Since we hold a transaction here we
4458 * can't flush, and we don't want to FLUSH_LIMIT because
4459 * we could have generated too many delayed refs to
4460 * actually allocate, so just bail if we're short and
4461 * let the normal reservation dance happen higher up.
4463 if (should_throttle) {
4464 ret = btrfs_delayed_refs_rsv_refill(fs_info,
4465 BTRFS_RESERVE_NO_FLUSH);
4477 if (ret >= 0 && pending_del_nr) {
4480 err = btrfs_del_items(trans, root, path, pending_del_slot,
4483 btrfs_abort_transaction(trans, err);
4487 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4488 ASSERT(last_size >= new_size);
4489 if (!ret && last_size > new_size)
4490 last_size = new_size;
4491 btrfs_inode_safe_disk_i_size_write(inode, last_size);
4492 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start,
4493 (u64)-1, &cached_state);
4496 btrfs_free_path(path);
4501 * btrfs_truncate_block - read, zero a chunk and write a block
4502 * @inode - inode that we're zeroing
4503 * @from - the offset to start zeroing
4504 * @len - the length to zero, 0 to zero the entire range respective to the
4506 * @front - zero up to the offset instead of from the offset on
4508 * This will find the block for the "from" offset and cow the block and zero the
4509 * part we want to zero. This is used with truncate and hole punching.
4511 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4514 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4515 struct address_space *mapping = inode->i_mapping;
4516 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4517 struct btrfs_ordered_extent *ordered;
4518 struct extent_state *cached_state = NULL;
4519 struct extent_changeset *data_reserved = NULL;
4521 u32 blocksize = fs_info->sectorsize;
4522 pgoff_t index = from >> PAGE_SHIFT;
4523 unsigned offset = from & (blocksize - 1);
4525 gfp_t mask = btrfs_alloc_write_mask(mapping);
4530 if (IS_ALIGNED(offset, blocksize) &&
4531 (!len || IS_ALIGNED(len, blocksize)))
4534 block_start = round_down(from, blocksize);
4535 block_end = block_start + blocksize - 1;
4537 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
4538 block_start, blocksize);
4543 page = find_or_create_page(mapping, index, mask);
4545 btrfs_delalloc_release_space(inode, data_reserved,
4546 block_start, blocksize, true);
4547 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize);
4552 if (!PageUptodate(page)) {
4553 ret = btrfs_readpage(NULL, page);
4555 if (page->mapping != mapping) {
4560 if (!PageUptodate(page)) {
4565 wait_on_page_writeback(page);
4567 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4568 set_page_extent_mapped(page);
4570 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4572 unlock_extent_cached(io_tree, block_start, block_end,
4576 btrfs_start_ordered_extent(inode, ordered, 1);
4577 btrfs_put_ordered_extent(ordered);
4581 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4582 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4583 0, 0, &cached_state);
4585 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4588 unlock_extent_cached(io_tree, block_start, block_end,
4593 if (offset != blocksize) {
4595 len = blocksize - offset;
4598 memset(kaddr + (block_start - page_offset(page)),
4601 memset(kaddr + (block_start - page_offset(page)) + offset,
4603 flush_dcache_page(page);
4606 ClearPageChecked(page);
4607 set_page_dirty(page);
4608 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4612 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4614 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize);
4618 extent_changeset_free(data_reserved);
4622 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4623 u64 offset, u64 len)
4625 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4626 struct btrfs_trans_handle *trans;
4630 * Still need to make sure the inode looks like it's been updated so
4631 * that any holes get logged if we fsync.
4633 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4634 BTRFS_I(inode)->last_trans = fs_info->generation;
4635 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4636 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4641 * 1 - for the one we're dropping
4642 * 1 - for the one we're adding
4643 * 1 - for updating the inode.
4645 trans = btrfs_start_transaction(root, 3);
4647 return PTR_ERR(trans);
4649 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4651 btrfs_abort_transaction(trans, ret);
4652 btrfs_end_transaction(trans);
4656 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)),
4657 offset, 0, 0, len, 0, len, 0, 0, 0);
4659 btrfs_abort_transaction(trans, ret);
4661 btrfs_update_inode(trans, root, inode);
4662 btrfs_end_transaction(trans);
4667 * This function puts in dummy file extents for the area we're creating a hole
4668 * for. So if we are truncating this file to a larger size we need to insert
4669 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4670 * the range between oldsize and size
4672 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4674 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4675 struct btrfs_root *root = BTRFS_I(inode)->root;
4676 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4677 struct extent_map *em = NULL;
4678 struct extent_state *cached_state = NULL;
4679 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4680 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4681 u64 block_end = ALIGN(size, fs_info->sectorsize);
4688 * If our size started in the middle of a block we need to zero out the
4689 * rest of the block before we expand the i_size, otherwise we could
4690 * expose stale data.
4692 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4696 if (size <= hole_start)
4699 btrfs_lock_and_flush_ordered_range(BTRFS_I(inode), hole_start,
4700 block_end - 1, &cached_state);
4701 cur_offset = hole_start;
4703 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
4704 block_end - cur_offset);
4710 last_byte = min(extent_map_end(em), block_end);
4711 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4712 hole_size = last_byte - cur_offset;
4714 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4715 struct extent_map *hole_em;
4717 err = maybe_insert_hole(root, inode, cur_offset,
4722 err = btrfs_inode_set_file_extent_range(BTRFS_I(inode),
4723 cur_offset, hole_size);
4727 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
4728 cur_offset + hole_size - 1, 0);
4729 hole_em = alloc_extent_map();
4731 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4732 &BTRFS_I(inode)->runtime_flags);
4735 hole_em->start = cur_offset;
4736 hole_em->len = hole_size;
4737 hole_em->orig_start = cur_offset;
4739 hole_em->block_start = EXTENT_MAP_HOLE;
4740 hole_em->block_len = 0;
4741 hole_em->orig_block_len = 0;
4742 hole_em->ram_bytes = hole_size;
4743 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4744 hole_em->generation = fs_info->generation;
4747 write_lock(&em_tree->lock);
4748 err = add_extent_mapping(em_tree, hole_em, 1);
4749 write_unlock(&em_tree->lock);
4752 btrfs_drop_extent_cache(BTRFS_I(inode),
4757 free_extent_map(hole_em);
4759 err = btrfs_inode_set_file_extent_range(BTRFS_I(inode),
4760 cur_offset, hole_size);
4765 free_extent_map(em);
4767 cur_offset = last_byte;
4768 if (cur_offset >= block_end)
4771 free_extent_map(em);
4772 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
4776 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4778 struct btrfs_root *root = BTRFS_I(inode)->root;
4779 struct btrfs_trans_handle *trans;
4780 loff_t oldsize = i_size_read(inode);
4781 loff_t newsize = attr->ia_size;
4782 int mask = attr->ia_valid;
4786 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4787 * special case where we need to update the times despite not having
4788 * these flags set. For all other operations the VFS set these flags
4789 * explicitly if it wants a timestamp update.
4791 if (newsize != oldsize) {
4792 inode_inc_iversion(inode);
4793 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4794 inode->i_ctime = inode->i_mtime =
4795 current_time(inode);
4798 if (newsize > oldsize) {
4800 * Don't do an expanding truncate while snapshotting is ongoing.
4801 * This is to ensure the snapshot captures a fully consistent
4802 * state of this file - if the snapshot captures this expanding
4803 * truncation, it must capture all writes that happened before
4806 btrfs_drew_write_lock(&root->snapshot_lock);
4807 ret = btrfs_cont_expand(inode, oldsize, newsize);
4809 btrfs_drew_write_unlock(&root->snapshot_lock);
4813 trans = btrfs_start_transaction(root, 1);
4814 if (IS_ERR(trans)) {
4815 btrfs_drew_write_unlock(&root->snapshot_lock);
4816 return PTR_ERR(trans);
4819 i_size_write(inode, newsize);
4820 btrfs_inode_safe_disk_i_size_write(inode, 0);
4821 pagecache_isize_extended(inode, oldsize, newsize);
4822 ret = btrfs_update_inode(trans, root, inode);
4823 btrfs_drew_write_unlock(&root->snapshot_lock);
4824 btrfs_end_transaction(trans);
4828 * We're truncating a file that used to have good data down to
4829 * zero. Make sure it gets into the ordered flush list so that
4830 * any new writes get down to disk quickly.
4833 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4834 &BTRFS_I(inode)->runtime_flags);
4836 truncate_setsize(inode, newsize);
4838 /* Disable nonlocked read DIO to avoid the endless truncate */
4839 btrfs_inode_block_unlocked_dio(BTRFS_I(inode));
4840 inode_dio_wait(inode);
4841 btrfs_inode_resume_unlocked_dio(BTRFS_I(inode));
4843 ret = btrfs_truncate(inode, newsize == oldsize);
4844 if (ret && inode->i_nlink) {
4848 * Truncate failed, so fix up the in-memory size. We
4849 * adjusted disk_i_size down as we removed extents, so
4850 * wait for disk_i_size to be stable and then update the
4851 * in-memory size to match.
4853 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
4856 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4863 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4865 struct inode *inode = d_inode(dentry);
4866 struct btrfs_root *root = BTRFS_I(inode)->root;
4869 if (btrfs_root_readonly(root))
4872 err = setattr_prepare(dentry, attr);
4876 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4877 err = btrfs_setsize(inode, attr);
4882 if (attr->ia_valid) {
4883 setattr_copy(inode, attr);
4884 inode_inc_iversion(inode);
4885 err = btrfs_dirty_inode(inode);
4887 if (!err && attr->ia_valid & ATTR_MODE)
4888 err = posix_acl_chmod(inode, inode->i_mode);
4895 * While truncating the inode pages during eviction, we get the VFS calling
4896 * btrfs_invalidatepage() against each page of the inode. This is slow because
4897 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4898 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4899 * extent_state structures over and over, wasting lots of time.
4901 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4902 * those expensive operations on a per page basis and do only the ordered io
4903 * finishing, while we release here the extent_map and extent_state structures,
4904 * without the excessive merging and splitting.
4906 static void evict_inode_truncate_pages(struct inode *inode)
4908 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4909 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
4910 struct rb_node *node;
4912 ASSERT(inode->i_state & I_FREEING);
4913 truncate_inode_pages_final(&inode->i_data);
4915 write_lock(&map_tree->lock);
4916 while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) {
4917 struct extent_map *em;
4919 node = rb_first_cached(&map_tree->map);
4920 em = rb_entry(node, struct extent_map, rb_node);
4921 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
4922 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
4923 remove_extent_mapping(map_tree, em);
4924 free_extent_map(em);
4925 if (need_resched()) {
4926 write_unlock(&map_tree->lock);
4928 write_lock(&map_tree->lock);
4931 write_unlock(&map_tree->lock);
4934 * Keep looping until we have no more ranges in the io tree.
4935 * We can have ongoing bios started by readahead that have
4936 * their endio callback (extent_io.c:end_bio_extent_readpage)
4937 * still in progress (unlocked the pages in the bio but did not yet
4938 * unlocked the ranges in the io tree). Therefore this means some
4939 * ranges can still be locked and eviction started because before
4940 * submitting those bios, which are executed by a separate task (work
4941 * queue kthread), inode references (inode->i_count) were not taken
4942 * (which would be dropped in the end io callback of each bio).
4943 * Therefore here we effectively end up waiting for those bios and
4944 * anyone else holding locked ranges without having bumped the inode's
4945 * reference count - if we don't do it, when they access the inode's
4946 * io_tree to unlock a range it may be too late, leading to an
4947 * use-after-free issue.
4949 spin_lock(&io_tree->lock);
4950 while (!RB_EMPTY_ROOT(&io_tree->state)) {
4951 struct extent_state *state;
4952 struct extent_state *cached_state = NULL;
4955 unsigned state_flags;
4957 node = rb_first(&io_tree->state);
4958 state = rb_entry(node, struct extent_state, rb_node);
4959 start = state->start;
4961 state_flags = state->state;
4962 spin_unlock(&io_tree->lock);
4964 lock_extent_bits(io_tree, start, end, &cached_state);
4967 * If still has DELALLOC flag, the extent didn't reach disk,
4968 * and its reserved space won't be freed by delayed_ref.
4969 * So we need to free its reserved space here.
4970 * (Refer to comment in btrfs_invalidatepage, case 2)
4972 * Note, end is the bytenr of last byte, so we need + 1 here.
4974 if (state_flags & EXTENT_DELALLOC)
4975 btrfs_qgroup_free_data(inode, NULL, start, end - start + 1);
4977 clear_extent_bit(io_tree, start, end,
4978 EXTENT_LOCKED | EXTENT_DELALLOC |
4979 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
4983 spin_lock(&io_tree->lock);
4985 spin_unlock(&io_tree->lock);
4988 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
4989 struct btrfs_block_rsv *rsv)
4991 struct btrfs_fs_info *fs_info = root->fs_info;
4992 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4993 struct btrfs_trans_handle *trans;
4994 u64 delayed_refs_extra = btrfs_calc_insert_metadata_size(fs_info, 1);
4998 * Eviction should be taking place at some place safe because of our
4999 * delayed iputs. However the normal flushing code will run delayed
5000 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5002 * We reserve the delayed_refs_extra here again because we can't use
5003 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5004 * above. We reserve our extra bit here because we generate a ton of
5005 * delayed refs activity by truncating.
5007 * If we cannot make our reservation we'll attempt to steal from the
5008 * global reserve, because we really want to be able to free up space.
5010 ret = btrfs_block_rsv_refill(root, rsv, rsv->size + delayed_refs_extra,
5011 BTRFS_RESERVE_FLUSH_EVICT);
5014 * Try to steal from the global reserve if there is space for
5017 if (btrfs_check_space_for_delayed_refs(fs_info) ||
5018 btrfs_block_rsv_migrate(global_rsv, rsv, rsv->size, 0)) {
5020 "could not allocate space for delete; will truncate on mount");
5021 return ERR_PTR(-ENOSPC);
5023 delayed_refs_extra = 0;
5026 trans = btrfs_join_transaction(root);
5030 if (delayed_refs_extra) {
5031 trans->block_rsv = &fs_info->trans_block_rsv;
5032 trans->bytes_reserved = delayed_refs_extra;
5033 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5034 delayed_refs_extra, 1);
5039 void btrfs_evict_inode(struct inode *inode)
5041 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5042 struct btrfs_trans_handle *trans;
5043 struct btrfs_root *root = BTRFS_I(inode)->root;
5044 struct btrfs_block_rsv *rsv;
5047 trace_btrfs_inode_evict(inode);
5054 evict_inode_truncate_pages(inode);
5056 if (inode->i_nlink &&
5057 ((btrfs_root_refs(&root->root_item) != 0 &&
5058 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5059 btrfs_is_free_space_inode(BTRFS_I(inode))))
5062 if (is_bad_inode(inode))
5065 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5067 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5070 if (inode->i_nlink > 0) {
5071 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5072 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5076 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5080 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5083 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5086 btrfs_i_size_write(BTRFS_I(inode), 0);
5089 trans = evict_refill_and_join(root, rsv);
5093 trans->block_rsv = rsv;
5095 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5096 trans->block_rsv = &fs_info->trans_block_rsv;
5097 btrfs_end_transaction(trans);
5098 btrfs_btree_balance_dirty(fs_info);
5099 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5106 * Errors here aren't a big deal, it just means we leave orphan items in
5107 * the tree. They will be cleaned up on the next mount. If the inode
5108 * number gets reused, cleanup deletes the orphan item without doing
5109 * anything, and unlink reuses the existing orphan item.
5111 * If it turns out that we are dropping too many of these, we might want
5112 * to add a mechanism for retrying these after a commit.
5114 trans = evict_refill_and_join(root, rsv);
5115 if (!IS_ERR(trans)) {
5116 trans->block_rsv = rsv;
5117 btrfs_orphan_del(trans, BTRFS_I(inode));
5118 trans->block_rsv = &fs_info->trans_block_rsv;
5119 btrfs_end_transaction(trans);
5122 if (!(root == fs_info->tree_root ||
5123 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5124 btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
5127 btrfs_free_block_rsv(fs_info, rsv);
5130 * If we didn't successfully delete, the orphan item will still be in
5131 * the tree and we'll retry on the next mount. Again, we might also want
5132 * to retry these periodically in the future.
5134 btrfs_remove_delayed_node(BTRFS_I(inode));
5139 * Return the key found in the dir entry in the location pointer, fill @type
5140 * with BTRFS_FT_*, and return 0.
5142 * If no dir entries were found, returns -ENOENT.
5143 * If found a corrupted location in dir entry, returns -EUCLEAN.
5145 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5146 struct btrfs_key *location, u8 *type)
5148 const char *name = dentry->d_name.name;
5149 int namelen = dentry->d_name.len;
5150 struct btrfs_dir_item *di;
5151 struct btrfs_path *path;
5152 struct btrfs_root *root = BTRFS_I(dir)->root;
5155 path = btrfs_alloc_path();
5159 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5161 if (IS_ERR_OR_NULL(di)) {
5162 ret = di ? PTR_ERR(di) : -ENOENT;
5166 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5167 if (location->type != BTRFS_INODE_ITEM_KEY &&
5168 location->type != BTRFS_ROOT_ITEM_KEY) {
5170 btrfs_warn(root->fs_info,
5171 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5172 __func__, name, btrfs_ino(BTRFS_I(dir)),
5173 location->objectid, location->type, location->offset);
5176 *type = btrfs_dir_type(path->nodes[0], di);
5178 btrfs_free_path(path);
5183 * when we hit a tree root in a directory, the btrfs part of the inode
5184 * needs to be changed to reflect the root directory of the tree root. This
5185 * is kind of like crossing a mount point.
5187 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5189 struct dentry *dentry,
5190 struct btrfs_key *location,
5191 struct btrfs_root **sub_root)
5193 struct btrfs_path *path;
5194 struct btrfs_root *new_root;
5195 struct btrfs_root_ref *ref;
5196 struct extent_buffer *leaf;
5197 struct btrfs_key key;
5201 path = btrfs_alloc_path();
5208 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5209 key.type = BTRFS_ROOT_REF_KEY;
5210 key.offset = location->objectid;
5212 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5219 leaf = path->nodes[0];
5220 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5221 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5222 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5225 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5226 (unsigned long)(ref + 1),
5227 dentry->d_name.len);
5231 btrfs_release_path(path);
5233 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5234 if (IS_ERR(new_root)) {
5235 err = PTR_ERR(new_root);
5239 *sub_root = new_root;
5240 location->objectid = btrfs_root_dirid(&new_root->root_item);
5241 location->type = BTRFS_INODE_ITEM_KEY;
5242 location->offset = 0;
5245 btrfs_free_path(path);
5249 static void inode_tree_add(struct inode *inode)
5251 struct btrfs_root *root = BTRFS_I(inode)->root;
5252 struct btrfs_inode *entry;
5254 struct rb_node *parent;
5255 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5256 u64 ino = btrfs_ino(BTRFS_I(inode));
5258 if (inode_unhashed(inode))
5261 spin_lock(&root->inode_lock);
5262 p = &root->inode_tree.rb_node;
5265 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5267 if (ino < btrfs_ino(entry))
5268 p = &parent->rb_left;
5269 else if (ino > btrfs_ino(entry))
5270 p = &parent->rb_right;
5272 WARN_ON(!(entry->vfs_inode.i_state &
5273 (I_WILL_FREE | I_FREEING)));
5274 rb_replace_node(parent, new, &root->inode_tree);
5275 RB_CLEAR_NODE(parent);
5276 spin_unlock(&root->inode_lock);
5280 rb_link_node(new, parent, p);
5281 rb_insert_color(new, &root->inode_tree);
5282 spin_unlock(&root->inode_lock);
5285 static void inode_tree_del(struct inode *inode)
5287 struct btrfs_root *root = BTRFS_I(inode)->root;
5290 spin_lock(&root->inode_lock);
5291 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5292 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5293 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5294 empty = RB_EMPTY_ROOT(&root->inode_tree);
5296 spin_unlock(&root->inode_lock);
5298 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5299 spin_lock(&root->inode_lock);
5300 empty = RB_EMPTY_ROOT(&root->inode_tree);
5301 spin_unlock(&root->inode_lock);
5303 btrfs_add_dead_root(root);
5308 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5310 struct btrfs_iget_args *args = p;
5312 inode->i_ino = args->ino;
5313 BTRFS_I(inode)->location.objectid = args->ino;
5314 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5315 BTRFS_I(inode)->location.offset = 0;
5316 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5317 BUG_ON(args->root && !BTRFS_I(inode)->root);
5321 static int btrfs_find_actor(struct inode *inode, void *opaque)
5323 struct btrfs_iget_args *args = opaque;
5325 return args->ino == BTRFS_I(inode)->location.objectid &&
5326 args->root == BTRFS_I(inode)->root;
5329 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5330 struct btrfs_root *root)
5332 struct inode *inode;
5333 struct btrfs_iget_args args;
5334 unsigned long hashval = btrfs_inode_hash(ino, root);
5339 inode = iget5_locked(s, hashval, btrfs_find_actor,
5340 btrfs_init_locked_inode,
5346 * Get an inode object given its inode number and corresponding root.
5347 * Path can be preallocated to prevent recursing back to iget through
5348 * allocator. NULL is also valid but may require an additional allocation
5351 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5352 struct btrfs_root *root, struct btrfs_path *path)
5354 struct inode *inode;
5356 inode = btrfs_iget_locked(s, ino, root);
5358 return ERR_PTR(-ENOMEM);
5360 if (inode->i_state & I_NEW) {
5363 ret = btrfs_read_locked_inode(inode, path);
5365 inode_tree_add(inode);
5366 unlock_new_inode(inode);
5370 * ret > 0 can come from btrfs_search_slot called by
5371 * btrfs_read_locked_inode, this means the inode item
5376 inode = ERR_PTR(ret);
5383 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5385 return btrfs_iget_path(s, ino, root, NULL);
5388 static struct inode *new_simple_dir(struct super_block *s,
5389 struct btrfs_key *key,
5390 struct btrfs_root *root)
5392 struct inode *inode = new_inode(s);
5395 return ERR_PTR(-ENOMEM);
5397 BTRFS_I(inode)->root = btrfs_grab_root(root);
5398 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5399 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5401 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5403 * We only need lookup, the rest is read-only and there's no inode
5404 * associated with the dentry
5406 inode->i_op = &simple_dir_inode_operations;
5407 inode->i_opflags &= ~IOP_XATTR;
5408 inode->i_fop = &simple_dir_operations;
5409 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5410 inode->i_mtime = current_time(inode);
5411 inode->i_atime = inode->i_mtime;
5412 inode->i_ctime = inode->i_mtime;
5413 BTRFS_I(inode)->i_otime = inode->i_mtime;
5418 static inline u8 btrfs_inode_type(struct inode *inode)
5421 * Compile-time asserts that generic FT_* types still match
5424 BUILD_BUG_ON(BTRFS_FT_UNKNOWN != FT_UNKNOWN);
5425 BUILD_BUG_ON(BTRFS_FT_REG_FILE != FT_REG_FILE);
5426 BUILD_BUG_ON(BTRFS_FT_DIR != FT_DIR);
5427 BUILD_BUG_ON(BTRFS_FT_CHRDEV != FT_CHRDEV);
5428 BUILD_BUG_ON(BTRFS_FT_BLKDEV != FT_BLKDEV);
5429 BUILD_BUG_ON(BTRFS_FT_FIFO != FT_FIFO);
5430 BUILD_BUG_ON(BTRFS_FT_SOCK != FT_SOCK);
5431 BUILD_BUG_ON(BTRFS_FT_SYMLINK != FT_SYMLINK);
5433 return fs_umode_to_ftype(inode->i_mode);
5436 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5438 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5439 struct inode *inode;
5440 struct btrfs_root *root = BTRFS_I(dir)->root;
5441 struct btrfs_root *sub_root = root;
5442 struct btrfs_key location;
5446 if (dentry->d_name.len > BTRFS_NAME_LEN)
5447 return ERR_PTR(-ENAMETOOLONG);
5449 ret = btrfs_inode_by_name(dir, dentry, &location, &di_type);
5451 return ERR_PTR(ret);
5453 if (location.type == BTRFS_INODE_ITEM_KEY) {
5454 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5458 /* Do extra check against inode mode with di_type */
5459 if (btrfs_inode_type(inode) != di_type) {
5461 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5462 inode->i_mode, btrfs_inode_type(inode),
5465 return ERR_PTR(-EUCLEAN);
5470 ret = fixup_tree_root_location(fs_info, dir, dentry,
5471 &location, &sub_root);
5474 inode = ERR_PTR(ret);
5476 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5478 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5480 if (root != sub_root)
5481 btrfs_put_root(sub_root);
5483 if (!IS_ERR(inode) && root != sub_root) {
5484 down_read(&fs_info->cleanup_work_sem);
5485 if (!sb_rdonly(inode->i_sb))
5486 ret = btrfs_orphan_cleanup(sub_root);
5487 up_read(&fs_info->cleanup_work_sem);
5490 inode = ERR_PTR(ret);
5497 static int btrfs_dentry_delete(const struct dentry *dentry)
5499 struct btrfs_root *root;
5500 struct inode *inode = d_inode(dentry);
5502 if (!inode && !IS_ROOT(dentry))
5503 inode = d_inode(dentry->d_parent);
5506 root = BTRFS_I(inode)->root;
5507 if (btrfs_root_refs(&root->root_item) == 0)
5510 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5516 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5519 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5521 if (inode == ERR_PTR(-ENOENT))
5523 return d_splice_alias(inode, dentry);
5527 * All this infrastructure exists because dir_emit can fault, and we are holding
5528 * the tree lock when doing readdir. For now just allocate a buffer and copy
5529 * our information into that, and then dir_emit from the buffer. This is
5530 * similar to what NFS does, only we don't keep the buffer around in pagecache
5531 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5532 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5535 static int btrfs_opendir(struct inode *inode, struct file *file)
5537 struct btrfs_file_private *private;
5539 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5542 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5543 if (!private->filldir_buf) {
5547 file->private_data = private;
5558 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5561 struct dir_entry *entry = addr;
5562 char *name = (char *)(entry + 1);
5564 ctx->pos = get_unaligned(&entry->offset);
5565 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5566 get_unaligned(&entry->ino),
5567 get_unaligned(&entry->type)))
5569 addr += sizeof(struct dir_entry) +
5570 get_unaligned(&entry->name_len);
5576 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5578 struct inode *inode = file_inode(file);
5579 struct btrfs_root *root = BTRFS_I(inode)->root;
5580 struct btrfs_file_private *private = file->private_data;
5581 struct btrfs_dir_item *di;
5582 struct btrfs_key key;
5583 struct btrfs_key found_key;
5584 struct btrfs_path *path;
5586 struct list_head ins_list;
5587 struct list_head del_list;
5589 struct extent_buffer *leaf;
5596 struct btrfs_key location;
5598 if (!dir_emit_dots(file, ctx))
5601 path = btrfs_alloc_path();
5605 addr = private->filldir_buf;
5606 path->reada = READA_FORWARD;
5608 INIT_LIST_HEAD(&ins_list);
5609 INIT_LIST_HEAD(&del_list);
5610 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5613 key.type = BTRFS_DIR_INDEX_KEY;
5614 key.offset = ctx->pos;
5615 key.objectid = btrfs_ino(BTRFS_I(inode));
5617 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5622 struct dir_entry *entry;
5624 leaf = path->nodes[0];
5625 slot = path->slots[0];
5626 if (slot >= btrfs_header_nritems(leaf)) {
5627 ret = btrfs_next_leaf(root, path);
5635 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5637 if (found_key.objectid != key.objectid)
5639 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5641 if (found_key.offset < ctx->pos)
5643 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5645 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5646 name_len = btrfs_dir_name_len(leaf, di);
5647 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5649 btrfs_release_path(path);
5650 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5653 addr = private->filldir_buf;
5660 put_unaligned(name_len, &entry->name_len);
5661 name_ptr = (char *)(entry + 1);
5662 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5664 put_unaligned(fs_ftype_to_dtype(btrfs_dir_type(leaf, di)),
5666 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5667 put_unaligned(location.objectid, &entry->ino);
5668 put_unaligned(found_key.offset, &entry->offset);
5670 addr += sizeof(struct dir_entry) + name_len;
5671 total_len += sizeof(struct dir_entry) + name_len;
5675 btrfs_release_path(path);
5677 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5681 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5686 * Stop new entries from being returned after we return the last
5689 * New directory entries are assigned a strictly increasing
5690 * offset. This means that new entries created during readdir
5691 * are *guaranteed* to be seen in the future by that readdir.
5692 * This has broken buggy programs which operate on names as
5693 * they're returned by readdir. Until we re-use freed offsets
5694 * we have this hack to stop new entries from being returned
5695 * under the assumption that they'll never reach this huge
5698 * This is being careful not to overflow 32bit loff_t unless the
5699 * last entry requires it because doing so has broken 32bit apps
5702 if (ctx->pos >= INT_MAX)
5703 ctx->pos = LLONG_MAX;
5710 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5711 btrfs_free_path(path);
5716 * This is somewhat expensive, updating the tree every time the
5717 * inode changes. But, it is most likely to find the inode in cache.
5718 * FIXME, needs more benchmarking...there are no reasons other than performance
5719 * to keep or drop this code.
5721 static int btrfs_dirty_inode(struct inode *inode)
5723 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5724 struct btrfs_root *root = BTRFS_I(inode)->root;
5725 struct btrfs_trans_handle *trans;
5728 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5731 trans = btrfs_join_transaction(root);
5733 return PTR_ERR(trans);
5735 ret = btrfs_update_inode(trans, root, inode);
5736 if (ret && ret == -ENOSPC) {
5737 /* whoops, lets try again with the full transaction */
5738 btrfs_end_transaction(trans);
5739 trans = btrfs_start_transaction(root, 1);
5741 return PTR_ERR(trans);
5743 ret = btrfs_update_inode(trans, root, inode);
5745 btrfs_end_transaction(trans);
5746 if (BTRFS_I(inode)->delayed_node)
5747 btrfs_balance_delayed_items(fs_info);
5753 * This is a copy of file_update_time. We need this so we can return error on
5754 * ENOSPC for updating the inode in the case of file write and mmap writes.
5756 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
5759 struct btrfs_root *root = BTRFS_I(inode)->root;
5760 bool dirty = flags & ~S_VERSION;
5762 if (btrfs_root_readonly(root))
5765 if (flags & S_VERSION)
5766 dirty |= inode_maybe_inc_iversion(inode, dirty);
5767 if (flags & S_CTIME)
5768 inode->i_ctime = *now;
5769 if (flags & S_MTIME)
5770 inode->i_mtime = *now;
5771 if (flags & S_ATIME)
5772 inode->i_atime = *now;
5773 return dirty ? btrfs_dirty_inode(inode) : 0;
5777 * find the highest existing sequence number in a directory
5778 * and then set the in-memory index_cnt variable to reflect
5779 * free sequence numbers
5781 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
5783 struct btrfs_root *root = inode->root;
5784 struct btrfs_key key, found_key;
5785 struct btrfs_path *path;
5786 struct extent_buffer *leaf;
5789 key.objectid = btrfs_ino(inode);
5790 key.type = BTRFS_DIR_INDEX_KEY;
5791 key.offset = (u64)-1;
5793 path = btrfs_alloc_path();
5797 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5800 /* FIXME: we should be able to handle this */
5806 * MAGIC NUMBER EXPLANATION:
5807 * since we search a directory based on f_pos we have to start at 2
5808 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5809 * else has to start at 2
5811 if (path->slots[0] == 0) {
5812 inode->index_cnt = 2;
5818 leaf = path->nodes[0];
5819 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5821 if (found_key.objectid != btrfs_ino(inode) ||
5822 found_key.type != BTRFS_DIR_INDEX_KEY) {
5823 inode->index_cnt = 2;
5827 inode->index_cnt = found_key.offset + 1;
5829 btrfs_free_path(path);
5834 * helper to find a free sequence number in a given directory. This current
5835 * code is very simple, later versions will do smarter things in the btree
5837 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
5841 if (dir->index_cnt == (u64)-1) {
5842 ret = btrfs_inode_delayed_dir_index_count(dir);
5844 ret = btrfs_set_inode_index_count(dir);
5850 *index = dir->index_cnt;
5856 static int btrfs_insert_inode_locked(struct inode *inode)
5858 struct btrfs_iget_args args;
5860 args.ino = BTRFS_I(inode)->location.objectid;
5861 args.root = BTRFS_I(inode)->root;
5863 return insert_inode_locked4(inode,
5864 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
5865 btrfs_find_actor, &args);
5869 * Inherit flags from the parent inode.
5871 * Currently only the compression flags and the cow flags are inherited.
5873 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
5880 flags = BTRFS_I(dir)->flags;
5882 if (flags & BTRFS_INODE_NOCOMPRESS) {
5883 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
5884 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
5885 } else if (flags & BTRFS_INODE_COMPRESS) {
5886 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
5887 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
5890 if (flags & BTRFS_INODE_NODATACOW) {
5891 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
5892 if (S_ISREG(inode->i_mode))
5893 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5896 btrfs_sync_inode_flags_to_i_flags(inode);
5899 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5900 struct btrfs_root *root,
5902 const char *name, int name_len,
5903 u64 ref_objectid, u64 objectid,
5904 umode_t mode, u64 *index)
5906 struct btrfs_fs_info *fs_info = root->fs_info;
5907 struct inode *inode;
5908 struct btrfs_inode_item *inode_item;
5909 struct btrfs_key *location;
5910 struct btrfs_path *path;
5911 struct btrfs_inode_ref *ref;
5912 struct btrfs_key key[2];
5914 int nitems = name ? 2 : 1;
5916 unsigned int nofs_flag;
5919 path = btrfs_alloc_path();
5921 return ERR_PTR(-ENOMEM);
5923 nofs_flag = memalloc_nofs_save();
5924 inode = new_inode(fs_info->sb);
5925 memalloc_nofs_restore(nofs_flag);
5927 btrfs_free_path(path);
5928 return ERR_PTR(-ENOMEM);
5932 * O_TMPFILE, set link count to 0, so that after this point,
5933 * we fill in an inode item with the correct link count.
5936 set_nlink(inode, 0);
5939 * we have to initialize this early, so we can reclaim the inode
5940 * number if we fail afterwards in this function.
5942 inode->i_ino = objectid;
5945 trace_btrfs_inode_request(dir);
5947 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
5949 btrfs_free_path(path);
5951 return ERR_PTR(ret);
5957 * index_cnt is ignored for everything but a dir,
5958 * btrfs_set_inode_index_count has an explanation for the magic
5961 BTRFS_I(inode)->index_cnt = 2;
5962 BTRFS_I(inode)->dir_index = *index;
5963 BTRFS_I(inode)->root = btrfs_grab_root(root);
5964 BTRFS_I(inode)->generation = trans->transid;
5965 inode->i_generation = BTRFS_I(inode)->generation;
5968 * We could have gotten an inode number from somebody who was fsynced
5969 * and then removed in this same transaction, so let's just set full
5970 * sync since it will be a full sync anyway and this will blow away the
5971 * old info in the log.
5973 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5975 key[0].objectid = objectid;
5976 key[0].type = BTRFS_INODE_ITEM_KEY;
5979 sizes[0] = sizeof(struct btrfs_inode_item);
5983 * Start new inodes with an inode_ref. This is slightly more
5984 * efficient for small numbers of hard links since they will
5985 * be packed into one item. Extended refs will kick in if we
5986 * add more hard links than can fit in the ref item.
5988 key[1].objectid = objectid;
5989 key[1].type = BTRFS_INODE_REF_KEY;
5990 key[1].offset = ref_objectid;
5992 sizes[1] = name_len + sizeof(*ref);
5995 location = &BTRFS_I(inode)->location;
5996 location->objectid = objectid;
5997 location->offset = 0;
5998 location->type = BTRFS_INODE_ITEM_KEY;
6000 ret = btrfs_insert_inode_locked(inode);
6006 path->leave_spinning = 1;
6007 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6011 inode_init_owner(inode, dir, mode);
6012 inode_set_bytes(inode, 0);
6014 inode->i_mtime = current_time(inode);
6015 inode->i_atime = inode->i_mtime;
6016 inode->i_ctime = inode->i_mtime;
6017 BTRFS_I(inode)->i_otime = inode->i_mtime;
6019 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6020 struct btrfs_inode_item);
6021 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6022 sizeof(*inode_item));
6023 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6026 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6027 struct btrfs_inode_ref);
6028 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6029 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6030 ptr = (unsigned long)(ref + 1);
6031 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6034 btrfs_mark_buffer_dirty(path->nodes[0]);
6035 btrfs_free_path(path);
6037 btrfs_inherit_iflags(inode, dir);
6039 if (S_ISREG(mode)) {
6040 if (btrfs_test_opt(fs_info, NODATASUM))
6041 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6042 if (btrfs_test_opt(fs_info, NODATACOW))
6043 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6044 BTRFS_INODE_NODATASUM;
6047 inode_tree_add(inode);
6049 trace_btrfs_inode_new(inode);
6050 btrfs_set_inode_last_trans(trans, inode);
6052 btrfs_update_root_times(trans, root);
6054 ret = btrfs_inode_inherit_props(trans, inode, dir);
6057 "error inheriting props for ino %llu (root %llu): %d",
6058 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6063 discard_new_inode(inode);
6066 BTRFS_I(dir)->index_cnt--;
6067 btrfs_free_path(path);
6068 return ERR_PTR(ret);
6072 * utility function to add 'inode' into 'parent_inode' with
6073 * a give name and a given sequence number.
6074 * if 'add_backref' is true, also insert a backref from the
6075 * inode to the parent directory.
6077 int btrfs_add_link(struct btrfs_trans_handle *trans,
6078 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6079 const char *name, int name_len, int add_backref, u64 index)
6082 struct btrfs_key key;
6083 struct btrfs_root *root = parent_inode->root;
6084 u64 ino = btrfs_ino(inode);
6085 u64 parent_ino = btrfs_ino(parent_inode);
6087 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6088 memcpy(&key, &inode->root->root_key, sizeof(key));
6091 key.type = BTRFS_INODE_ITEM_KEY;
6095 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6096 ret = btrfs_add_root_ref(trans, key.objectid,
6097 root->root_key.objectid, parent_ino,
6098 index, name, name_len);
6099 } else if (add_backref) {
6100 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6104 /* Nothing to clean up yet */
6108 ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key,
6109 btrfs_inode_type(&inode->vfs_inode), index);
6110 if (ret == -EEXIST || ret == -EOVERFLOW)
6113 btrfs_abort_transaction(trans, ret);
6117 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6119 inode_inc_iversion(&parent_inode->vfs_inode);
6121 * If we are replaying a log tree, we do not want to update the mtime
6122 * and ctime of the parent directory with the current time, since the
6123 * log replay procedure is responsible for setting them to their correct
6124 * values (the ones it had when the fsync was done).
6126 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
6127 struct timespec64 now = current_time(&parent_inode->vfs_inode);
6129 parent_inode->vfs_inode.i_mtime = now;
6130 parent_inode->vfs_inode.i_ctime = now;
6132 ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode);
6134 btrfs_abort_transaction(trans, ret);
6138 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6141 err = btrfs_del_root_ref(trans, key.objectid,
6142 root->root_key.objectid, parent_ino,
6143 &local_index, name, name_len);
6145 btrfs_abort_transaction(trans, err);
6146 } else if (add_backref) {
6150 err = btrfs_del_inode_ref(trans, root, name, name_len,
6151 ino, parent_ino, &local_index);
6153 btrfs_abort_transaction(trans, err);
6156 /* Return the original error code */
6160 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6161 struct btrfs_inode *dir, struct dentry *dentry,
6162 struct btrfs_inode *inode, int backref, u64 index)
6164 int err = btrfs_add_link(trans, dir, inode,
6165 dentry->d_name.name, dentry->d_name.len,
6172 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6173 umode_t mode, dev_t rdev)
6175 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6176 struct btrfs_trans_handle *trans;
6177 struct btrfs_root *root = BTRFS_I(dir)->root;
6178 struct inode *inode = NULL;
6184 * 2 for inode item and ref
6186 * 1 for xattr if selinux is on
6188 trans = btrfs_start_transaction(root, 5);
6190 return PTR_ERR(trans);
6192 err = btrfs_find_free_ino(root, &objectid);
6196 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6197 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6199 if (IS_ERR(inode)) {
6200 err = PTR_ERR(inode);
6206 * If the active LSM wants to access the inode during
6207 * d_instantiate it needs these. Smack checks to see
6208 * if the filesystem supports xattrs by looking at the
6211 inode->i_op = &btrfs_special_inode_operations;
6212 init_special_inode(inode, inode->i_mode, rdev);
6214 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6218 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6223 btrfs_update_inode(trans, root, inode);
6224 d_instantiate_new(dentry, inode);
6227 btrfs_end_transaction(trans);
6228 btrfs_btree_balance_dirty(fs_info);
6230 inode_dec_link_count(inode);
6231 discard_new_inode(inode);
6236 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6237 umode_t mode, bool excl)
6239 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6240 struct btrfs_trans_handle *trans;
6241 struct btrfs_root *root = BTRFS_I(dir)->root;
6242 struct inode *inode = NULL;
6248 * 2 for inode item and ref
6250 * 1 for xattr if selinux is on
6252 trans = btrfs_start_transaction(root, 5);
6254 return PTR_ERR(trans);
6256 err = btrfs_find_free_ino(root, &objectid);
6260 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6261 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6263 if (IS_ERR(inode)) {
6264 err = PTR_ERR(inode);
6269 * If the active LSM wants to access the inode during
6270 * d_instantiate it needs these. Smack checks to see
6271 * if the filesystem supports xattrs by looking at the
6274 inode->i_fop = &btrfs_file_operations;
6275 inode->i_op = &btrfs_file_inode_operations;
6276 inode->i_mapping->a_ops = &btrfs_aops;
6278 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6282 err = btrfs_update_inode(trans, root, inode);
6286 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6291 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6292 d_instantiate_new(dentry, inode);
6295 btrfs_end_transaction(trans);
6297 inode_dec_link_count(inode);
6298 discard_new_inode(inode);
6300 btrfs_btree_balance_dirty(fs_info);
6304 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6305 struct dentry *dentry)
6307 struct btrfs_trans_handle *trans = NULL;
6308 struct btrfs_root *root = BTRFS_I(dir)->root;
6309 struct inode *inode = d_inode(old_dentry);
6310 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6315 /* do not allow sys_link's with other subvols of the same device */
6316 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6319 if (inode->i_nlink >= BTRFS_LINK_MAX)
6322 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6327 * 2 items for inode and inode ref
6328 * 2 items for dir items
6329 * 1 item for parent inode
6330 * 1 item for orphan item deletion if O_TMPFILE
6332 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6333 if (IS_ERR(trans)) {
6334 err = PTR_ERR(trans);
6339 /* There are several dir indexes for this inode, clear the cache. */
6340 BTRFS_I(inode)->dir_index = 0ULL;
6342 inode_inc_iversion(inode);
6343 inode->i_ctime = current_time(inode);
6345 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6347 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6353 struct dentry *parent = dentry->d_parent;
6356 err = btrfs_update_inode(trans, root, inode);
6359 if (inode->i_nlink == 1) {
6361 * If new hard link count is 1, it's a file created
6362 * with open(2) O_TMPFILE flag.
6364 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6368 d_instantiate(dentry, inode);
6369 ret = btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent,
6371 if (ret == BTRFS_NEED_TRANS_COMMIT) {
6372 err = btrfs_commit_transaction(trans);
6379 btrfs_end_transaction(trans);
6381 inode_dec_link_count(inode);
6384 btrfs_btree_balance_dirty(fs_info);
6388 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6390 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6391 struct inode *inode = NULL;
6392 struct btrfs_trans_handle *trans;
6393 struct btrfs_root *root = BTRFS_I(dir)->root;
6399 * 2 items for inode and ref
6400 * 2 items for dir items
6401 * 1 for xattr if selinux is on
6403 trans = btrfs_start_transaction(root, 5);
6405 return PTR_ERR(trans);
6407 err = btrfs_find_free_ino(root, &objectid);
6411 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6412 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6413 S_IFDIR | mode, &index);
6414 if (IS_ERR(inode)) {
6415 err = PTR_ERR(inode);
6420 /* these must be set before we unlock the inode */
6421 inode->i_op = &btrfs_dir_inode_operations;
6422 inode->i_fop = &btrfs_dir_file_operations;
6424 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6428 btrfs_i_size_write(BTRFS_I(inode), 0);
6429 err = btrfs_update_inode(trans, root, inode);
6433 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6434 dentry->d_name.name,
6435 dentry->d_name.len, 0, index);
6439 d_instantiate_new(dentry, inode);
6442 btrfs_end_transaction(trans);
6444 inode_dec_link_count(inode);
6445 discard_new_inode(inode);
6447 btrfs_btree_balance_dirty(fs_info);
6451 static noinline int uncompress_inline(struct btrfs_path *path,
6453 size_t pg_offset, u64 extent_offset,
6454 struct btrfs_file_extent_item *item)
6457 struct extent_buffer *leaf = path->nodes[0];
6460 unsigned long inline_size;
6464 WARN_ON(pg_offset != 0);
6465 compress_type = btrfs_file_extent_compression(leaf, item);
6466 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6467 inline_size = btrfs_file_extent_inline_item_len(leaf,
6468 btrfs_item_nr(path->slots[0]));
6469 tmp = kmalloc(inline_size, GFP_NOFS);
6472 ptr = btrfs_file_extent_inline_start(item);
6474 read_extent_buffer(leaf, tmp, ptr, inline_size);
6476 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6477 ret = btrfs_decompress(compress_type, tmp, page,
6478 extent_offset, inline_size, max_size);
6481 * decompression code contains a memset to fill in any space between the end
6482 * of the uncompressed data and the end of max_size in case the decompressed
6483 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6484 * the end of an inline extent and the beginning of the next block, so we
6485 * cover that region here.
6488 if (max_size + pg_offset < PAGE_SIZE) {
6489 char *map = kmap(page);
6490 memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6498 * btrfs_get_extent - Lookup the first extent overlapping a range in a file.
6499 * @inode: file to search in
6500 * @page: page to read extent data into if the extent is inline
6501 * @pg_offset: offset into @page to copy to
6502 * @start: file offset
6503 * @len: length of range starting at @start
6505 * This returns the first &struct extent_map which overlaps with the given
6506 * range, reading it from the B-tree and caching it if necessary. Note that
6507 * there may be more extents which overlap the given range after the returned
6510 * If @page is not NULL and the extent is inline, this also reads the extent
6511 * data directly into the page and marks the extent up to date in the io_tree.
6513 * Return: ERR_PTR on error, non-NULL extent_map on success.
6515 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6516 struct page *page, size_t pg_offset,
6519 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6522 u64 extent_start = 0;
6524 u64 objectid = btrfs_ino(inode);
6525 int extent_type = -1;
6526 struct btrfs_path *path = NULL;
6527 struct btrfs_root *root = inode->root;
6528 struct btrfs_file_extent_item *item;
6529 struct extent_buffer *leaf;
6530 struct btrfs_key found_key;
6531 struct extent_map *em = NULL;
6532 struct extent_map_tree *em_tree = &inode->extent_tree;
6533 struct extent_io_tree *io_tree = &inode->io_tree;
6535 read_lock(&em_tree->lock);
6536 em = lookup_extent_mapping(em_tree, start, len);
6537 read_unlock(&em_tree->lock);
6540 if (em->start > start || em->start + em->len <= start)
6541 free_extent_map(em);
6542 else if (em->block_start == EXTENT_MAP_INLINE && page)
6543 free_extent_map(em);
6547 em = alloc_extent_map();
6552 em->start = EXTENT_MAP_HOLE;
6553 em->orig_start = EXTENT_MAP_HOLE;
6555 em->block_len = (u64)-1;
6557 path = btrfs_alloc_path();
6563 /* Chances are we'll be called again, so go ahead and do readahead */
6564 path->reada = READA_FORWARD;
6567 * Unless we're going to uncompress the inline extent, no sleep would
6570 path->leave_spinning = 1;
6572 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6576 } else if (ret > 0) {
6577 if (path->slots[0] == 0)
6582 leaf = path->nodes[0];
6583 item = btrfs_item_ptr(leaf, path->slots[0],
6584 struct btrfs_file_extent_item);
6585 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6586 if (found_key.objectid != objectid ||
6587 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6589 * If we backup past the first extent we want to move forward
6590 * and see if there is an extent in front of us, otherwise we'll
6591 * say there is a hole for our whole search range which can
6598 extent_type = btrfs_file_extent_type(leaf, item);
6599 extent_start = found_key.offset;
6600 extent_end = btrfs_file_extent_end(path);
6601 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6602 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6603 /* Only regular file could have regular/prealloc extent */
6604 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6607 "regular/prealloc extent found for non-regular inode %llu",
6611 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6613 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6614 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6619 if (start >= extent_end) {
6621 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6622 ret = btrfs_next_leaf(root, path);
6626 } else if (ret > 0) {
6629 leaf = path->nodes[0];
6631 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6632 if (found_key.objectid != objectid ||
6633 found_key.type != BTRFS_EXTENT_DATA_KEY)
6635 if (start + len <= found_key.offset)
6637 if (start > found_key.offset)
6640 /* New extent overlaps with existing one */
6642 em->orig_start = start;
6643 em->len = found_key.offset - start;
6644 em->block_start = EXTENT_MAP_HOLE;
6648 btrfs_extent_item_to_extent_map(inode, path, item, !page, em);
6650 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6651 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6653 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6657 size_t extent_offset;
6663 size = btrfs_file_extent_ram_bytes(leaf, item);
6664 extent_offset = page_offset(page) + pg_offset - extent_start;
6665 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6666 size - extent_offset);
6667 em->start = extent_start + extent_offset;
6668 em->len = ALIGN(copy_size, fs_info->sectorsize);
6669 em->orig_block_len = em->len;
6670 em->orig_start = em->start;
6671 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6673 btrfs_set_path_blocking(path);
6674 if (!PageUptodate(page)) {
6675 if (btrfs_file_extent_compression(leaf, item) !=
6676 BTRFS_COMPRESS_NONE) {
6677 ret = uncompress_inline(path, page, pg_offset,
6678 extent_offset, item);
6685 read_extent_buffer(leaf, map + pg_offset, ptr,
6687 if (pg_offset + copy_size < PAGE_SIZE) {
6688 memset(map + pg_offset + copy_size, 0,
6689 PAGE_SIZE - pg_offset -
6694 flush_dcache_page(page);
6696 set_extent_uptodate(io_tree, em->start,
6697 extent_map_end(em) - 1, NULL, GFP_NOFS);
6702 em->orig_start = start;
6704 em->block_start = EXTENT_MAP_HOLE;
6706 btrfs_release_path(path);
6707 if (em->start > start || extent_map_end(em) <= start) {
6709 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6710 em->start, em->len, start, len);
6716 write_lock(&em_tree->lock);
6717 err = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6718 write_unlock(&em_tree->lock);
6720 btrfs_free_path(path);
6722 trace_btrfs_get_extent(root, inode, em);
6725 free_extent_map(em);
6726 return ERR_PTR(err);
6728 BUG_ON(!em); /* Error is always set */
6732 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
6735 struct extent_map *em;
6736 struct extent_map *hole_em = NULL;
6737 u64 delalloc_start = start;
6743 em = btrfs_get_extent(inode, NULL, 0, start, len);
6747 * If our em maps to:
6749 * - a pre-alloc extent,
6750 * there might actually be delalloc bytes behind it.
6752 if (em->block_start != EXTENT_MAP_HOLE &&
6753 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6758 /* check to see if we've wrapped (len == -1 or similar) */
6767 /* ok, we didn't find anything, lets look for delalloc */
6768 delalloc_len = count_range_bits(&inode->io_tree, &delalloc_start,
6769 end, len, EXTENT_DELALLOC, 1);
6770 delalloc_end = delalloc_start + delalloc_len;
6771 if (delalloc_end < delalloc_start)
6772 delalloc_end = (u64)-1;
6775 * We didn't find anything useful, return the original results from
6778 if (delalloc_start > end || delalloc_end <= start) {
6785 * Adjust the delalloc_start to make sure it doesn't go backwards from
6786 * the start they passed in
6788 delalloc_start = max(start, delalloc_start);
6789 delalloc_len = delalloc_end - delalloc_start;
6791 if (delalloc_len > 0) {
6794 const u64 hole_end = extent_map_end(hole_em);
6796 em = alloc_extent_map();
6804 * When btrfs_get_extent can't find anything it returns one
6807 * Make sure what it found really fits our range, and adjust to
6808 * make sure it is based on the start from the caller
6810 if (hole_end <= start || hole_em->start > end) {
6811 free_extent_map(hole_em);
6814 hole_start = max(hole_em->start, start);
6815 hole_len = hole_end - hole_start;
6818 if (hole_em && delalloc_start > hole_start) {
6820 * Our hole starts before our delalloc, so we have to
6821 * return just the parts of the hole that go until the
6824 em->len = min(hole_len, delalloc_start - hole_start);
6825 em->start = hole_start;
6826 em->orig_start = hole_start;
6828 * Don't adjust block start at all, it is fixed at
6831 em->block_start = hole_em->block_start;
6832 em->block_len = hole_len;
6833 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6834 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6837 * Hole is out of passed range or it starts after
6840 em->start = delalloc_start;
6841 em->len = delalloc_len;
6842 em->orig_start = delalloc_start;
6843 em->block_start = EXTENT_MAP_DELALLOC;
6844 em->block_len = delalloc_len;
6851 free_extent_map(hole_em);
6853 free_extent_map(em);
6854 return ERR_PTR(err);
6859 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
6862 const u64 orig_start,
6863 const u64 block_start,
6864 const u64 block_len,
6865 const u64 orig_block_len,
6866 const u64 ram_bytes,
6869 struct extent_map *em = NULL;
6872 if (type != BTRFS_ORDERED_NOCOW) {
6873 em = create_io_em(inode, start, len, orig_start,
6874 block_start, block_len, orig_block_len,
6876 BTRFS_COMPRESS_NONE, /* compress_type */
6881 ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
6882 len, block_len, type);
6885 free_extent_map(em);
6886 btrfs_drop_extent_cache(BTRFS_I(inode), start,
6887 start + len - 1, 0);
6896 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6899 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6900 struct btrfs_root *root = BTRFS_I(inode)->root;
6901 struct extent_map *em;
6902 struct btrfs_key ins;
6906 alloc_hint = get_extent_allocation_hint(inode, start, len);
6907 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
6908 0, alloc_hint, &ins, 1, 1);
6910 return ERR_PTR(ret);
6912 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
6913 ins.objectid, ins.offset, ins.offset,
6914 ins.offset, BTRFS_ORDERED_REGULAR);
6915 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
6917 btrfs_free_reserved_extent(fs_info, ins.objectid,
6924 * returns 1 when the nocow is safe, < 1 on error, 0 if the
6925 * block must be cow'd
6927 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
6928 u64 *orig_start, u64 *orig_block_len,
6931 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6932 struct btrfs_path *path;
6934 struct extent_buffer *leaf;
6935 struct btrfs_root *root = BTRFS_I(inode)->root;
6936 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6937 struct btrfs_file_extent_item *fi;
6938 struct btrfs_key key;
6945 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
6947 path = btrfs_alloc_path();
6951 ret = btrfs_lookup_file_extent(NULL, root, path,
6952 btrfs_ino(BTRFS_I(inode)), offset, 0);
6956 slot = path->slots[0];
6959 /* can't find the item, must cow */
6966 leaf = path->nodes[0];
6967 btrfs_item_key_to_cpu(leaf, &key, slot);
6968 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
6969 key.type != BTRFS_EXTENT_DATA_KEY) {
6970 /* not our file or wrong item type, must cow */
6974 if (key.offset > offset) {
6975 /* Wrong offset, must cow */
6979 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6980 found_type = btrfs_file_extent_type(leaf, fi);
6981 if (found_type != BTRFS_FILE_EXTENT_REG &&
6982 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6983 /* not a regular extent, must cow */
6987 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
6990 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6991 if (extent_end <= offset)
6994 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6995 if (disk_bytenr == 0)
6998 if (btrfs_file_extent_compression(leaf, fi) ||
6999 btrfs_file_extent_encryption(leaf, fi) ||
7000 btrfs_file_extent_other_encoding(leaf, fi))
7004 * Do the same check as in btrfs_cross_ref_exist but without the
7005 * unnecessary search.
7007 if (btrfs_file_extent_generation(leaf, fi) <=
7008 btrfs_root_last_snapshot(&root->root_item))
7011 backref_offset = btrfs_file_extent_offset(leaf, fi);
7014 *orig_start = key.offset - backref_offset;
7015 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7016 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7019 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7022 num_bytes = min(offset + *len, extent_end) - offset;
7023 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7026 range_end = round_up(offset + num_bytes,
7027 root->fs_info->sectorsize) - 1;
7028 ret = test_range_bit(io_tree, offset, range_end,
7029 EXTENT_DELALLOC, 0, NULL);
7036 btrfs_release_path(path);
7039 * look for other files referencing this extent, if we
7040 * find any we must cow
7043 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7044 key.offset - backref_offset, disk_bytenr);
7051 * adjust disk_bytenr and num_bytes to cover just the bytes
7052 * in this extent we are about to write. If there
7053 * are any csums in that range we have to cow in order
7054 * to keep the csums correct
7056 disk_bytenr += backref_offset;
7057 disk_bytenr += offset - key.offset;
7058 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7061 * all of the above have passed, it is safe to overwrite this extent
7067 btrfs_free_path(path);
7071 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7072 struct extent_state **cached_state, int writing)
7074 struct btrfs_ordered_extent *ordered;
7078 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7081 * We're concerned with the entire range that we're going to be
7082 * doing DIO to, so we need to make sure there's no ordered
7083 * extents in this range.
7085 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7086 lockend - lockstart + 1);
7089 * We need to make sure there are no buffered pages in this
7090 * range either, we could have raced between the invalidate in
7091 * generic_file_direct_write and locking the extent. The
7092 * invalidate needs to happen so that reads after a write do not
7096 (!writing || !filemap_range_has_page(inode->i_mapping,
7097 lockstart, lockend)))
7100 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7105 * If we are doing a DIO read and the ordered extent we
7106 * found is for a buffered write, we can not wait for it
7107 * to complete and retry, because if we do so we can
7108 * deadlock with concurrent buffered writes on page
7109 * locks. This happens only if our DIO read covers more
7110 * than one extent map, if at this point has already
7111 * created an ordered extent for a previous extent map
7112 * and locked its range in the inode's io tree, and a
7113 * concurrent write against that previous extent map's
7114 * range and this range started (we unlock the ranges
7115 * in the io tree only when the bios complete and
7116 * buffered writes always lock pages before attempting
7117 * to lock range in the io tree).
7120 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7121 btrfs_start_ordered_extent(inode, ordered, 1);
7124 btrfs_put_ordered_extent(ordered);
7127 * We could trigger writeback for this range (and wait
7128 * for it to complete) and then invalidate the pages for
7129 * this range (through invalidate_inode_pages2_range()),
7130 * but that can lead us to a deadlock with a concurrent
7131 * call to readahead (a buffered read or a defrag call
7132 * triggered a readahead) on a page lock due to an
7133 * ordered dio extent we created before but did not have
7134 * yet a corresponding bio submitted (whence it can not
7135 * complete), which makes readahead wait for that
7136 * ordered extent to complete while holding a lock on
7151 /* The callers of this must take lock_extent() */
7152 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
7153 u64 orig_start, u64 block_start,
7154 u64 block_len, u64 orig_block_len,
7155 u64 ram_bytes, int compress_type,
7158 struct extent_map_tree *em_tree;
7159 struct extent_map *em;
7162 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7163 type == BTRFS_ORDERED_COMPRESSED ||
7164 type == BTRFS_ORDERED_NOCOW ||
7165 type == BTRFS_ORDERED_REGULAR);
7167 em_tree = &BTRFS_I(inode)->extent_tree;
7168 em = alloc_extent_map();
7170 return ERR_PTR(-ENOMEM);
7173 em->orig_start = orig_start;
7175 em->block_len = block_len;
7176 em->block_start = block_start;
7177 em->orig_block_len = orig_block_len;
7178 em->ram_bytes = ram_bytes;
7179 em->generation = -1;
7180 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7181 if (type == BTRFS_ORDERED_PREALLOC) {
7182 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7183 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7184 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7185 em->compress_type = compress_type;
7189 btrfs_drop_extent_cache(BTRFS_I(inode), em->start,
7190 em->start + em->len - 1, 0);
7191 write_lock(&em_tree->lock);
7192 ret = add_extent_mapping(em_tree, em, 1);
7193 write_unlock(&em_tree->lock);
7195 * The caller has taken lock_extent(), who could race with us
7198 } while (ret == -EEXIST);
7201 free_extent_map(em);
7202 return ERR_PTR(ret);
7205 /* em got 2 refs now, callers needs to do free_extent_map once. */
7210 static int btrfs_get_blocks_direct_read(struct extent_map *em,
7211 struct buffer_head *bh_result,
7212 struct inode *inode,
7215 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7217 if (em->block_start == EXTENT_MAP_HOLE ||
7218 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7221 len = min(len, em->len - (start - em->start));
7223 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7225 bh_result->b_size = len;
7226 bh_result->b_bdev = fs_info->fs_devices->latest_bdev;
7227 set_buffer_mapped(bh_result);
7232 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7233 struct buffer_head *bh_result,
7234 struct inode *inode,
7235 struct btrfs_dio_data *dio_data,
7238 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7239 struct extent_map *em = *map;
7243 * We don't allocate a new extent in the following cases
7245 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7247 * 2) The extent is marked as PREALLOC. We're good to go here and can
7248 * just use the extent.
7251 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7252 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7253 em->block_start != EXTENT_MAP_HOLE)) {
7255 u64 block_start, orig_start, orig_block_len, ram_bytes;
7257 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7258 type = BTRFS_ORDERED_PREALLOC;
7260 type = BTRFS_ORDERED_NOCOW;
7261 len = min(len, em->len - (start - em->start));
7262 block_start = em->block_start + (start - em->start);
7264 if (can_nocow_extent(inode, start, &len, &orig_start,
7265 &orig_block_len, &ram_bytes) == 1 &&
7266 btrfs_inc_nocow_writers(fs_info, block_start)) {
7267 struct extent_map *em2;
7269 em2 = btrfs_create_dio_extent(inode, start, len,
7270 orig_start, block_start,
7271 len, orig_block_len,
7273 btrfs_dec_nocow_writers(fs_info, block_start);
7274 if (type == BTRFS_ORDERED_PREALLOC) {
7275 free_extent_map(em);
7279 if (em2 && IS_ERR(em2)) {
7284 * For inode marked NODATACOW or extent marked PREALLOC,
7285 * use the existing or preallocated extent, so does not
7286 * need to adjust btrfs_space_info's bytes_may_use.
7288 btrfs_free_reserved_data_space_noquota(inode, start,
7294 /* this will cow the extent */
7295 len = bh_result->b_size;
7296 free_extent_map(em);
7297 *map = em = btrfs_new_extent_direct(inode, start, len);
7303 len = min(len, em->len - (start - em->start));
7306 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7308 bh_result->b_size = len;
7309 bh_result->b_bdev = fs_info->fs_devices->latest_bdev;
7310 set_buffer_mapped(bh_result);
7312 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7313 set_buffer_new(bh_result);
7316 * Need to update the i_size under the extent lock so buffered
7317 * readers will get the updated i_size when we unlock.
7319 if (!dio_data->overwrite && start + len > i_size_read(inode))
7320 i_size_write(inode, start + len);
7322 WARN_ON(dio_data->reserve < len);
7323 dio_data->reserve -= len;
7324 dio_data->unsubmitted_oe_range_end = start + len;
7325 current->journal_info = dio_data;
7330 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7331 struct buffer_head *bh_result, int create)
7333 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7334 struct extent_map *em;
7335 struct extent_state *cached_state = NULL;
7336 struct btrfs_dio_data *dio_data = NULL;
7337 u64 start = iblock << inode->i_blkbits;
7338 u64 lockstart, lockend;
7339 u64 len = bh_result->b_size;
7343 len = min_t(u64, len, fs_info->sectorsize);
7346 lockend = start + len - 1;
7348 if (current->journal_info) {
7350 * Need to pull our outstanding extents and set journal_info to NULL so
7351 * that anything that needs to check if there's a transaction doesn't get
7354 dio_data = current->journal_info;
7355 current->journal_info = NULL;
7359 * If this errors out it's because we couldn't invalidate pagecache for
7360 * this range and we need to fallback to buffered.
7362 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7368 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7375 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7376 * io. INLINE is special, and we could probably kludge it in here, but
7377 * it's still buffered so for safety lets just fall back to the generic
7380 * For COMPRESSED we _have_ to read the entire extent in so we can
7381 * decompress it, so there will be buffering required no matter what we
7382 * do, so go ahead and fallback to buffered.
7384 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7385 * to buffered IO. Don't blame me, this is the price we pay for using
7388 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7389 em->block_start == EXTENT_MAP_INLINE) {
7390 free_extent_map(em);
7396 ret = btrfs_get_blocks_direct_write(&em, bh_result, inode,
7397 dio_data, start, len);
7401 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
7402 lockend, &cached_state);
7404 ret = btrfs_get_blocks_direct_read(em, bh_result, inode,
7406 /* Can be negative only if we read from a hole */
7409 free_extent_map(em);
7413 * We need to unlock only the end area that we aren't using.
7414 * The rest is going to be unlocked by the endio routine.
7416 lockstart = start + bh_result->b_size;
7417 if (lockstart < lockend) {
7418 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
7419 lockstart, lockend, &cached_state);
7421 free_extent_state(cached_state);
7425 free_extent_map(em);
7430 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7434 current->journal_info = dio_data;
7438 static void btrfs_dio_private_put(struct btrfs_dio_private *dip)
7441 * This implies a barrier so that stores to dio_bio->bi_status before
7442 * this and loads of dio_bio->bi_status after this are fully ordered.
7444 if (!refcount_dec_and_test(&dip->refs))
7447 if (bio_op(dip->dio_bio) == REQ_OP_WRITE) {
7448 __endio_write_update_ordered(dip->inode, dip->logical_offset,
7450 !dip->dio_bio->bi_status);
7452 unlock_extent(&BTRFS_I(dip->inode)->io_tree,
7453 dip->logical_offset,
7454 dip->logical_offset + dip->bytes - 1);
7457 dio_end_io(dip->dio_bio);
7461 static blk_status_t submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7463 unsigned long bio_flags)
7465 struct btrfs_dio_private *dip = bio->bi_private;
7466 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7469 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7471 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
7475 refcount_inc(&dip->refs);
7476 ret = btrfs_map_bio(fs_info, bio, mirror_num);
7478 refcount_dec(&dip->refs);
7482 static blk_status_t btrfs_check_read_dio_bio(struct inode *inode,
7483 struct btrfs_io_bio *io_bio,
7484 const bool uptodate)
7486 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
7487 const u32 sectorsize = fs_info->sectorsize;
7488 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7489 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7490 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7491 struct bio_vec bvec;
7492 struct bvec_iter iter;
7493 u64 start = io_bio->logical;
7495 blk_status_t err = BLK_STS_OK;
7497 __bio_for_each_segment(bvec, &io_bio->bio, iter, io_bio->iter) {
7498 unsigned int i, nr_sectors, pgoff;
7500 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7501 pgoff = bvec.bv_offset;
7502 for (i = 0; i < nr_sectors; i++) {
7503 ASSERT(pgoff < PAGE_SIZE);
7505 (!csum || !check_data_csum(inode, io_bio, icsum,
7506 bvec.bv_page, pgoff,
7507 start, sectorsize))) {
7508 clean_io_failure(fs_info, failure_tree, io_tree,
7509 start, bvec.bv_page,
7510 btrfs_ino(BTRFS_I(inode)),
7513 blk_status_t status;
7515 status = btrfs_submit_read_repair(inode,
7517 start - io_bio->logical,
7518 bvec.bv_page, pgoff,
7520 start + sectorsize - 1,
7522 submit_dio_repair_bio);
7526 start += sectorsize;
7528 pgoff += sectorsize;
7534 static void __endio_write_update_ordered(struct inode *inode,
7535 const u64 offset, const u64 bytes,
7536 const bool uptodate)
7538 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7539 struct btrfs_ordered_extent *ordered = NULL;
7540 struct btrfs_workqueue *wq;
7541 u64 ordered_offset = offset;
7542 u64 ordered_bytes = bytes;
7545 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
7546 wq = fs_info->endio_freespace_worker;
7548 wq = fs_info->endio_write_workers;
7550 while (ordered_offset < offset + bytes) {
7551 last_offset = ordered_offset;
7552 if (btrfs_dec_test_first_ordered_pending(inode, &ordered,
7556 btrfs_init_work(&ordered->work, finish_ordered_fn, NULL,
7558 btrfs_queue_work(wq, &ordered->work);
7561 * If btrfs_dec_test_ordered_pending does not find any ordered
7562 * extent in the range, we can exit.
7564 if (ordered_offset == last_offset)
7567 * Our bio might span multiple ordered extents. In this case
7568 * we keep going until we have accounted the whole dio.
7570 if (ordered_offset < offset + bytes) {
7571 ordered_bytes = offset + bytes - ordered_offset;
7577 static blk_status_t btrfs_submit_bio_start_direct_io(void *private_data,
7578 struct bio *bio, u64 offset)
7580 struct inode *inode = private_data;
7582 ret = btrfs_csum_one_bio(inode, bio, offset, 1);
7583 BUG_ON(ret); /* -ENOMEM */
7587 static void btrfs_end_dio_bio(struct bio *bio)
7589 struct btrfs_dio_private *dip = bio->bi_private;
7590 blk_status_t err = bio->bi_status;
7593 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
7594 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
7595 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
7597 (unsigned long long)bio->bi_iter.bi_sector,
7598 bio->bi_iter.bi_size, err);
7600 if (bio_op(bio) == REQ_OP_READ) {
7601 err = btrfs_check_read_dio_bio(dip->inode, btrfs_io_bio(bio),
7606 dip->dio_bio->bi_status = err;
7609 btrfs_dio_private_put(dip);
7612 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
7613 struct inode *inode, u64 file_offset, int async_submit)
7615 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7616 struct btrfs_dio_private *dip = bio->bi_private;
7617 bool write = bio_op(bio) == REQ_OP_WRITE;
7620 /* Check btrfs_submit_bio_hook() for rules about async submit. */
7622 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7625 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
7630 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
7633 if (write && async_submit) {
7634 ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0,
7636 btrfs_submit_bio_start_direct_io);
7640 * If we aren't doing async submit, calculate the csum of the
7643 ret = btrfs_csum_one_bio(inode, bio, file_offset, 1);
7649 csum_offset = file_offset - dip->logical_offset;
7650 csum_offset >>= inode->i_sb->s_blocksize_bits;
7651 csum_offset *= btrfs_super_csum_size(fs_info->super_copy);
7652 btrfs_io_bio(bio)->csum = dip->csums + csum_offset;
7655 ret = btrfs_map_bio(fs_info, bio, 0);
7661 * If this succeeds, the btrfs_dio_private is responsible for cleaning up locked
7662 * or ordered extents whether or not we submit any bios.
7664 static struct btrfs_dio_private *btrfs_create_dio_private(struct bio *dio_bio,
7665 struct inode *inode,
7668 const bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
7669 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7671 struct btrfs_dio_private *dip;
7673 dip_size = sizeof(*dip);
7674 if (!write && csum) {
7675 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7676 const u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
7679 nblocks = dio_bio->bi_iter.bi_size >> inode->i_sb->s_blocksize_bits;
7680 dip_size += csum_size * nblocks;
7683 dip = kzalloc(dip_size, GFP_NOFS);
7688 dip->logical_offset = file_offset;
7689 dip->bytes = dio_bio->bi_iter.bi_size;
7690 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
7691 dip->dio_bio = dio_bio;
7692 refcount_set(&dip->refs, 1);
7695 struct btrfs_dio_data *dio_data = current->journal_info;
7698 * Setting range start and end to the same value means that
7699 * no cleanup will happen in btrfs_direct_IO
7701 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
7703 dio_data->unsubmitted_oe_range_start =
7704 dio_data->unsubmitted_oe_range_end;
7709 static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode,
7712 const bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
7713 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7714 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7715 const bool raid56 = (btrfs_data_alloc_profile(fs_info) &
7716 BTRFS_BLOCK_GROUP_RAID56_MASK);
7717 struct btrfs_dio_private *dip;
7720 int async_submit = 0;
7722 int clone_offset = 0;
7725 blk_status_t status;
7726 struct btrfs_io_geometry geom;
7728 dip = btrfs_create_dio_private(dio_bio, inode, file_offset);
7731 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
7732 file_offset + dio_bio->bi_iter.bi_size - 1);
7734 dio_bio->bi_status = BLK_STS_RESOURCE;
7735 dio_end_io(dio_bio);
7739 if (!write && csum) {
7741 * Load the csums up front to reduce csum tree searches and
7742 * contention when submitting bios.
7744 status = btrfs_lookup_bio_sums(inode, dio_bio, file_offset,
7746 if (status != BLK_STS_OK)
7750 start_sector = dio_bio->bi_iter.bi_sector;
7751 submit_len = dio_bio->bi_iter.bi_size;
7754 ret = btrfs_get_io_geometry(fs_info, btrfs_op(dio_bio),
7755 start_sector << 9, submit_len,
7758 status = errno_to_blk_status(ret);
7761 ASSERT(geom.len <= INT_MAX);
7763 clone_len = min_t(int, submit_len, geom.len);
7766 * This will never fail as it's passing GPF_NOFS and
7767 * the allocation is backed by btrfs_bioset.
7769 bio = btrfs_bio_clone_partial(dio_bio, clone_offset, clone_len);
7770 bio->bi_private = dip;
7771 bio->bi_end_io = btrfs_end_dio_bio;
7772 btrfs_io_bio(bio)->logical = file_offset;
7774 ASSERT(submit_len >= clone_len);
7775 submit_len -= clone_len;
7778 * Increase the count before we submit the bio so we know
7779 * the end IO handler won't happen before we increase the
7780 * count. Otherwise, the dip might get freed before we're
7781 * done setting it up.
7783 * We transfer the initial reference to the last bio, so we
7784 * don't need to increment the reference count for the last one.
7786 if (submit_len > 0) {
7787 refcount_inc(&dip->refs);
7789 * If we are submitting more than one bio, submit them
7790 * all asynchronously. The exception is RAID 5 or 6, as
7791 * asynchronous checksums make it difficult to collect
7792 * full stripe writes.
7798 status = btrfs_submit_dio_bio(bio, inode, file_offset,
7803 refcount_dec(&dip->refs);
7807 clone_offset += clone_len;
7808 start_sector += clone_len >> 9;
7809 file_offset += clone_len;
7810 } while (submit_len > 0);
7814 dip->dio_bio->bi_status = status;
7815 btrfs_dio_private_put(dip);
7818 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
7819 const struct iov_iter *iter, loff_t offset)
7823 unsigned int blocksize_mask = fs_info->sectorsize - 1;
7824 ssize_t retval = -EINVAL;
7826 if (offset & blocksize_mask)
7829 if (iov_iter_alignment(iter) & blocksize_mask)
7832 /* If this is a write we don't need to check anymore */
7833 if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter))
7836 * Check to make sure we don't have duplicate iov_base's in this
7837 * iovec, if so return EINVAL, otherwise we'll get csum errors
7838 * when reading back.
7840 for (seg = 0; seg < iter->nr_segs; seg++) {
7841 for (i = seg + 1; i < iter->nr_segs; i++) {
7842 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
7851 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
7853 struct file *file = iocb->ki_filp;
7854 struct inode *inode = file->f_mapping->host;
7855 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7856 struct btrfs_dio_data dio_data = { 0 };
7857 struct extent_changeset *data_reserved = NULL;
7858 loff_t offset = iocb->ki_pos;
7862 bool relock = false;
7865 if (check_direct_IO(fs_info, iter, offset))
7868 inode_dio_begin(inode);
7871 * The generic stuff only does filemap_write_and_wait_range, which
7872 * isn't enough if we've written compressed pages to this area, so
7873 * we need to flush the dirty pages again to make absolutely sure
7874 * that any outstanding dirty pages are on disk.
7876 count = iov_iter_count(iter);
7877 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7878 &BTRFS_I(inode)->runtime_flags))
7879 filemap_fdatawrite_range(inode->i_mapping, offset,
7880 offset + count - 1);
7882 if (iov_iter_rw(iter) == WRITE) {
7884 * If the write DIO is beyond the EOF, we need update
7885 * the isize, but it is protected by i_mutex. So we can
7886 * not unlock the i_mutex at this case.
7888 if (offset + count <= inode->i_size) {
7889 dio_data.overwrite = 1;
7890 inode_unlock(inode);
7893 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
7899 * We need to know how many extents we reserved so that we can
7900 * do the accounting properly if we go over the number we
7901 * originally calculated. Abuse current->journal_info for this.
7903 dio_data.reserve = round_up(count,
7904 fs_info->sectorsize);
7905 dio_data.unsubmitted_oe_range_start = (u64)offset;
7906 dio_data.unsubmitted_oe_range_end = (u64)offset;
7907 current->journal_info = &dio_data;
7908 down_read(&BTRFS_I(inode)->dio_sem);
7909 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7910 &BTRFS_I(inode)->runtime_flags)) {
7911 inode_dio_end(inode);
7912 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7916 ret = __blockdev_direct_IO(iocb, inode,
7917 fs_info->fs_devices->latest_bdev,
7918 iter, btrfs_get_blocks_direct, NULL,
7919 btrfs_submit_direct, flags);
7920 if (iov_iter_rw(iter) == WRITE) {
7921 up_read(&BTRFS_I(inode)->dio_sem);
7922 current->journal_info = NULL;
7923 if (ret < 0 && ret != -EIOCBQUEUED) {
7924 if (dio_data.reserve)
7925 btrfs_delalloc_release_space(inode, data_reserved,
7926 offset, dio_data.reserve, true);
7928 * On error we might have left some ordered extents
7929 * without submitting corresponding bios for them, so
7930 * cleanup them up to avoid other tasks getting them
7931 * and waiting for them to complete forever.
7933 if (dio_data.unsubmitted_oe_range_start <
7934 dio_data.unsubmitted_oe_range_end)
7935 __endio_write_update_ordered(inode,
7936 dio_data.unsubmitted_oe_range_start,
7937 dio_data.unsubmitted_oe_range_end -
7938 dio_data.unsubmitted_oe_range_start,
7940 } else if (ret >= 0 && (size_t)ret < count)
7941 btrfs_delalloc_release_space(inode, data_reserved,
7942 offset, count - (size_t)ret, true);
7943 btrfs_delalloc_release_extents(BTRFS_I(inode), count);
7947 inode_dio_end(inode);
7951 extent_changeset_free(data_reserved);
7955 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7956 __u64 start, __u64 len)
7960 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
7964 return extent_fiemap(inode, fieinfo, start, len);
7967 int btrfs_readpage(struct file *file, struct page *page)
7969 return extent_read_full_page(page, btrfs_get_extent, 0);
7972 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
7974 struct inode *inode = page->mapping->host;
7977 if (current->flags & PF_MEMALLOC) {
7978 redirty_page_for_writepage(wbc, page);
7984 * If we are under memory pressure we will call this directly from the
7985 * VM, we need to make sure we have the inode referenced for the ordered
7986 * extent. If not just return like we didn't do anything.
7988 if (!igrab(inode)) {
7989 redirty_page_for_writepage(wbc, page);
7990 return AOP_WRITEPAGE_ACTIVATE;
7992 ret = extent_write_full_page(page, wbc);
7993 btrfs_add_delayed_iput(inode);
7997 static int btrfs_writepages(struct address_space *mapping,
7998 struct writeback_control *wbc)
8000 return extent_writepages(mapping, wbc);
8003 static void btrfs_readahead(struct readahead_control *rac)
8005 extent_readahead(rac);
8008 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8010 int ret = try_release_extent_mapping(page, gfp_flags);
8012 detach_page_private(page);
8016 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8018 if (PageWriteback(page) || PageDirty(page))
8020 return __btrfs_releasepage(page, gfp_flags);
8023 #ifdef CONFIG_MIGRATION
8024 static int btrfs_migratepage(struct address_space *mapping,
8025 struct page *newpage, struct page *page,
8026 enum migrate_mode mode)
8030 ret = migrate_page_move_mapping(mapping, newpage, page, 0);
8031 if (ret != MIGRATEPAGE_SUCCESS)
8034 if (page_has_private(page))
8035 attach_page_private(newpage, detach_page_private(page));
8037 if (PagePrivate2(page)) {
8038 ClearPagePrivate2(page);
8039 SetPagePrivate2(newpage);
8042 if (mode != MIGRATE_SYNC_NO_COPY)
8043 migrate_page_copy(newpage, page);
8045 migrate_page_states(newpage, page);
8046 return MIGRATEPAGE_SUCCESS;
8050 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8051 unsigned int length)
8053 struct inode *inode = page->mapping->host;
8054 struct extent_io_tree *tree;
8055 struct btrfs_ordered_extent *ordered;
8056 struct extent_state *cached_state = NULL;
8057 u64 page_start = page_offset(page);
8058 u64 page_end = page_start + PAGE_SIZE - 1;
8061 int inode_evicting = inode->i_state & I_FREEING;
8064 * we have the page locked, so new writeback can't start,
8065 * and the dirty bit won't be cleared while we are here.
8067 * Wait for IO on this page so that we can safely clear
8068 * the PagePrivate2 bit and do ordered accounting
8070 wait_on_page_writeback(page);
8072 tree = &BTRFS_I(inode)->io_tree;
8074 btrfs_releasepage(page, GFP_NOFS);
8078 if (!inode_evicting)
8079 lock_extent_bits(tree, page_start, page_end, &cached_state);
8082 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
8083 page_end - start + 1);
8086 ordered->file_offset + ordered->num_bytes - 1);
8088 * IO on this page will never be started, so we need
8089 * to account for any ordered extents now
8091 if (!inode_evicting)
8092 clear_extent_bit(tree, start, end,
8093 EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
8094 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8095 EXTENT_DEFRAG, 1, 0, &cached_state);
8097 * whoever cleared the private bit is responsible
8098 * for the finish_ordered_io
8100 if (TestClearPagePrivate2(page)) {
8101 struct btrfs_ordered_inode_tree *tree;
8104 tree = &BTRFS_I(inode)->ordered_tree;
8106 spin_lock_irq(&tree->lock);
8107 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8108 new_len = start - ordered->file_offset;
8109 if (new_len < ordered->truncated_len)
8110 ordered->truncated_len = new_len;
8111 spin_unlock_irq(&tree->lock);
8113 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8115 end - start + 1, 1))
8116 btrfs_finish_ordered_io(ordered);
8118 btrfs_put_ordered_extent(ordered);
8119 if (!inode_evicting) {
8120 cached_state = NULL;
8121 lock_extent_bits(tree, start, end,
8126 if (start < page_end)
8131 * Qgroup reserved space handler
8132 * Page here will be either
8133 * 1) Already written to disk
8134 * In this case, its reserved space is released from data rsv map
8135 * and will be freed by delayed_ref handler finally.
8136 * So even we call qgroup_free_data(), it won't decrease reserved
8138 * 2) Not written to disk
8139 * This means the reserved space should be freed here. However,
8140 * if a truncate invalidates the page (by clearing PageDirty)
8141 * and the page is accounted for while allocating extent
8142 * in btrfs_check_data_free_space() we let delayed_ref to
8143 * free the entire extent.
8145 if (PageDirty(page))
8146 btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
8147 if (!inode_evicting) {
8148 clear_extent_bit(tree, page_start, page_end, EXTENT_LOCKED |
8149 EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
8150 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
8153 __btrfs_releasepage(page, GFP_NOFS);
8156 ClearPageChecked(page);
8157 detach_page_private(page);
8161 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8162 * called from a page fault handler when a page is first dirtied. Hence we must
8163 * be careful to check for EOF conditions here. We set the page up correctly
8164 * for a written page which means we get ENOSPC checking when writing into
8165 * holes and correct delalloc and unwritten extent mapping on filesystems that
8166 * support these features.
8168 * We are not allowed to take the i_mutex here so we have to play games to
8169 * protect against truncate races as the page could now be beyond EOF. Because
8170 * truncate_setsize() writes the inode size before removing pages, once we have
8171 * the page lock we can determine safely if the page is beyond EOF. If it is not
8172 * beyond EOF, then the page is guaranteed safe against truncation until we
8175 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8177 struct page *page = vmf->page;
8178 struct inode *inode = file_inode(vmf->vma->vm_file);
8179 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8180 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8181 struct btrfs_ordered_extent *ordered;
8182 struct extent_state *cached_state = NULL;
8183 struct extent_changeset *data_reserved = NULL;
8185 unsigned long zero_start;
8195 reserved_space = PAGE_SIZE;
8197 sb_start_pagefault(inode->i_sb);
8198 page_start = page_offset(page);
8199 page_end = page_start + PAGE_SIZE - 1;
8203 * Reserving delalloc space after obtaining the page lock can lead to
8204 * deadlock. For example, if a dirty page is locked by this function
8205 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8206 * dirty page write out, then the btrfs_writepage() function could
8207 * end up waiting indefinitely to get a lock on the page currently
8208 * being processed by btrfs_page_mkwrite() function.
8210 ret2 = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
8213 ret2 = file_update_time(vmf->vma->vm_file);
8217 ret = vmf_error(ret2);
8223 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8226 size = i_size_read(inode);
8228 if ((page->mapping != inode->i_mapping) ||
8229 (page_start >= size)) {
8230 /* page got truncated out from underneath us */
8233 wait_on_page_writeback(page);
8235 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8236 set_page_extent_mapped(page);
8239 * we can't set the delalloc bits if there are pending ordered
8240 * extents. Drop our locks and wait for them to finish
8242 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8245 unlock_extent_cached(io_tree, page_start, page_end,
8248 btrfs_start_ordered_extent(inode, ordered, 1);
8249 btrfs_put_ordered_extent(ordered);
8253 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8254 reserved_space = round_up(size - page_start,
8255 fs_info->sectorsize);
8256 if (reserved_space < PAGE_SIZE) {
8257 end = page_start + reserved_space - 1;
8258 btrfs_delalloc_release_space(inode, data_reserved,
8259 page_start, PAGE_SIZE - reserved_space,
8265 * page_mkwrite gets called when the page is firstly dirtied after it's
8266 * faulted in, but write(2) could also dirty a page and set delalloc
8267 * bits, thus in this case for space account reason, we still need to
8268 * clear any delalloc bits within this page range since we have to
8269 * reserve data&meta space before lock_page() (see above comments).
8271 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8272 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8273 EXTENT_DEFRAG, 0, 0, &cached_state);
8275 ret2 = btrfs_set_extent_delalloc(inode, page_start, end, 0,
8278 unlock_extent_cached(io_tree, page_start, page_end,
8280 ret = VM_FAULT_SIGBUS;
8284 /* page is wholly or partially inside EOF */
8285 if (page_start + PAGE_SIZE > size)
8286 zero_start = offset_in_page(size);
8288 zero_start = PAGE_SIZE;
8290 if (zero_start != PAGE_SIZE) {
8292 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
8293 flush_dcache_page(page);
8296 ClearPageChecked(page);
8297 set_page_dirty(page);
8298 SetPageUptodate(page);
8300 BTRFS_I(inode)->last_trans = fs_info->generation;
8301 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8302 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8304 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8306 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8307 sb_end_pagefault(inode->i_sb);
8308 extent_changeset_free(data_reserved);
8309 return VM_FAULT_LOCKED;
8314 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8315 btrfs_delalloc_release_space(inode, data_reserved, page_start,
8316 reserved_space, (ret != 0));
8318 sb_end_pagefault(inode->i_sb);
8319 extent_changeset_free(data_reserved);
8323 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
8325 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8326 struct btrfs_root *root = BTRFS_I(inode)->root;
8327 struct btrfs_block_rsv *rsv;
8329 struct btrfs_trans_handle *trans;
8330 u64 mask = fs_info->sectorsize - 1;
8331 u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8333 if (!skip_writeback) {
8334 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8341 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8342 * things going on here:
8344 * 1) We need to reserve space to update our inode.
8346 * 2) We need to have something to cache all the space that is going to
8347 * be free'd up by the truncate operation, but also have some slack
8348 * space reserved in case it uses space during the truncate (thank you
8349 * very much snapshotting).
8351 * And we need these to be separate. The fact is we can use a lot of
8352 * space doing the truncate, and we have no earthly idea how much space
8353 * we will use, so we need the truncate reservation to be separate so it
8354 * doesn't end up using space reserved for updating the inode. We also
8355 * need to be able to stop the transaction and start a new one, which
8356 * means we need to be able to update the inode several times, and we
8357 * have no idea of knowing how many times that will be, so we can't just
8358 * reserve 1 item for the entirety of the operation, so that has to be
8359 * done separately as well.
8361 * So that leaves us with
8363 * 1) rsv - for the truncate reservation, which we will steal from the
8364 * transaction reservation.
8365 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8366 * updating the inode.
8368 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8371 rsv->size = min_size;
8375 * 1 for the truncate slack space
8376 * 1 for updating the inode.
8378 trans = btrfs_start_transaction(root, 2);
8379 if (IS_ERR(trans)) {
8380 ret = PTR_ERR(trans);
8384 /* Migrate the slack space for the truncate to our reserve */
8385 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8390 * So if we truncate and then write and fsync we normally would just
8391 * write the extents that changed, which is a problem if we need to
8392 * first truncate that entire inode. So set this flag so we write out
8393 * all of the extents in the inode to the sync log so we're completely
8396 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
8397 trans->block_rsv = rsv;
8400 ret = btrfs_truncate_inode_items(trans, root, inode,
8402 BTRFS_EXTENT_DATA_KEY);
8403 trans->block_rsv = &fs_info->trans_block_rsv;
8404 if (ret != -ENOSPC && ret != -EAGAIN)
8407 ret = btrfs_update_inode(trans, root, inode);
8411 btrfs_end_transaction(trans);
8412 btrfs_btree_balance_dirty(fs_info);
8414 trans = btrfs_start_transaction(root, 2);
8415 if (IS_ERR(trans)) {
8416 ret = PTR_ERR(trans);
8421 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8422 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8423 rsv, min_size, false);
8424 BUG_ON(ret); /* shouldn't happen */
8425 trans->block_rsv = rsv;
8429 * We can't call btrfs_truncate_block inside a trans handle as we could
8430 * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
8431 * we've truncated everything except the last little bit, and can do
8432 * btrfs_truncate_block and then update the disk_i_size.
8434 if (ret == NEED_TRUNCATE_BLOCK) {
8435 btrfs_end_transaction(trans);
8436 btrfs_btree_balance_dirty(fs_info);
8438 ret = btrfs_truncate_block(inode, inode->i_size, 0, 0);
8441 trans = btrfs_start_transaction(root, 1);
8442 if (IS_ERR(trans)) {
8443 ret = PTR_ERR(trans);
8446 btrfs_inode_safe_disk_i_size_write(inode, 0);
8452 trans->block_rsv = &fs_info->trans_block_rsv;
8453 ret2 = btrfs_update_inode(trans, root, inode);
8457 ret2 = btrfs_end_transaction(trans);
8460 btrfs_btree_balance_dirty(fs_info);
8463 btrfs_free_block_rsv(fs_info, rsv);
8469 * create a new subvolume directory/inode (helper for the ioctl).
8471 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8472 struct btrfs_root *new_root,
8473 struct btrfs_root *parent_root,
8476 struct inode *inode;
8480 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
8481 new_dirid, new_dirid,
8482 S_IFDIR | (~current_umask() & S_IRWXUGO),
8485 return PTR_ERR(inode);
8486 inode->i_op = &btrfs_dir_inode_operations;
8487 inode->i_fop = &btrfs_dir_file_operations;
8489 set_nlink(inode, 1);
8490 btrfs_i_size_write(BTRFS_I(inode), 0);
8491 unlock_new_inode(inode);
8493 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8495 btrfs_err(new_root->fs_info,
8496 "error inheriting subvolume %llu properties: %d",
8497 new_root->root_key.objectid, err);
8499 err = btrfs_update_inode(trans, new_root, inode);
8505 struct inode *btrfs_alloc_inode(struct super_block *sb)
8507 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8508 struct btrfs_inode *ei;
8509 struct inode *inode;
8511 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL);
8518 ei->last_sub_trans = 0;
8519 ei->logged_trans = 0;
8520 ei->delalloc_bytes = 0;
8521 ei->new_delalloc_bytes = 0;
8522 ei->defrag_bytes = 0;
8523 ei->disk_i_size = 0;
8526 ei->index_cnt = (u64)-1;
8528 ei->last_unlink_trans = 0;
8529 ei->last_log_commit = 0;
8531 spin_lock_init(&ei->lock);
8532 ei->outstanding_extents = 0;
8533 if (sb->s_magic != BTRFS_TEST_MAGIC)
8534 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8535 BTRFS_BLOCK_RSV_DELALLOC);
8536 ei->runtime_flags = 0;
8537 ei->prop_compress = BTRFS_COMPRESS_NONE;
8538 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8540 ei->delayed_node = NULL;
8542 ei->i_otime.tv_sec = 0;
8543 ei->i_otime.tv_nsec = 0;
8545 inode = &ei->vfs_inode;
8546 extent_map_tree_init(&ei->extent_tree);
8547 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO, inode);
8548 extent_io_tree_init(fs_info, &ei->io_failure_tree,
8549 IO_TREE_INODE_IO_FAILURE, inode);
8550 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8551 IO_TREE_INODE_FILE_EXTENT, inode);
8552 ei->io_tree.track_uptodate = true;
8553 ei->io_failure_tree.track_uptodate = true;
8554 atomic_set(&ei->sync_writers, 0);
8555 mutex_init(&ei->log_mutex);
8556 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8557 INIT_LIST_HEAD(&ei->delalloc_inodes);
8558 INIT_LIST_HEAD(&ei->delayed_iput);
8559 RB_CLEAR_NODE(&ei->rb_node);
8560 init_rwsem(&ei->dio_sem);
8565 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8566 void btrfs_test_destroy_inode(struct inode *inode)
8568 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
8569 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8573 void btrfs_free_inode(struct inode *inode)
8575 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8578 void btrfs_destroy_inode(struct inode *inode)
8580 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8581 struct btrfs_ordered_extent *ordered;
8582 struct btrfs_root *root = BTRFS_I(inode)->root;
8584 WARN_ON(!hlist_empty(&inode->i_dentry));
8585 WARN_ON(inode->i_data.nrpages);
8586 WARN_ON(BTRFS_I(inode)->block_rsv.reserved);
8587 WARN_ON(BTRFS_I(inode)->block_rsv.size);
8588 WARN_ON(BTRFS_I(inode)->outstanding_extents);
8589 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
8590 WARN_ON(BTRFS_I(inode)->new_delalloc_bytes);
8591 WARN_ON(BTRFS_I(inode)->csum_bytes);
8592 WARN_ON(BTRFS_I(inode)->defrag_bytes);
8595 * This can happen where we create an inode, but somebody else also
8596 * created the same inode and we need to destroy the one we already
8603 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8608 "found ordered extent %llu %llu on inode cleanup",
8609 ordered->file_offset, ordered->num_bytes);
8610 btrfs_remove_ordered_extent(inode, ordered);
8611 btrfs_put_ordered_extent(ordered);
8612 btrfs_put_ordered_extent(ordered);
8615 btrfs_qgroup_check_reserved_leak(inode);
8616 inode_tree_del(inode);
8617 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
8618 btrfs_inode_clear_file_extent_range(BTRFS_I(inode), 0, (u64)-1);
8619 btrfs_put_root(BTRFS_I(inode)->root);
8622 int btrfs_drop_inode(struct inode *inode)
8624 struct btrfs_root *root = BTRFS_I(inode)->root;
8629 /* the snap/subvol tree is on deleting */
8630 if (btrfs_root_refs(&root->root_item) == 0)
8633 return generic_drop_inode(inode);
8636 static void init_once(void *foo)
8638 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8640 inode_init_once(&ei->vfs_inode);
8643 void __cold btrfs_destroy_cachep(void)
8646 * Make sure all delayed rcu free inodes are flushed before we
8650 kmem_cache_destroy(btrfs_inode_cachep);
8651 kmem_cache_destroy(btrfs_trans_handle_cachep);
8652 kmem_cache_destroy(btrfs_path_cachep);
8653 kmem_cache_destroy(btrfs_free_space_cachep);
8654 kmem_cache_destroy(btrfs_free_space_bitmap_cachep);
8657 int __init btrfs_init_cachep(void)
8659 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8660 sizeof(struct btrfs_inode), 0,
8661 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
8663 if (!btrfs_inode_cachep)
8666 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8667 sizeof(struct btrfs_trans_handle), 0,
8668 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
8669 if (!btrfs_trans_handle_cachep)
8672 btrfs_path_cachep = kmem_cache_create("btrfs_path",
8673 sizeof(struct btrfs_path), 0,
8674 SLAB_MEM_SPREAD, NULL);
8675 if (!btrfs_path_cachep)
8678 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8679 sizeof(struct btrfs_free_space), 0,
8680 SLAB_MEM_SPREAD, NULL);
8681 if (!btrfs_free_space_cachep)
8684 btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap",
8685 PAGE_SIZE, PAGE_SIZE,
8686 SLAB_RED_ZONE, NULL);
8687 if (!btrfs_free_space_bitmap_cachep)
8692 btrfs_destroy_cachep();
8696 static int btrfs_getattr(const struct path *path, struct kstat *stat,
8697 u32 request_mask, unsigned int flags)
8700 struct inode *inode = d_inode(path->dentry);
8701 u32 blocksize = inode->i_sb->s_blocksize;
8702 u32 bi_flags = BTRFS_I(inode)->flags;
8704 stat->result_mask |= STATX_BTIME;
8705 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
8706 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
8707 if (bi_flags & BTRFS_INODE_APPEND)
8708 stat->attributes |= STATX_ATTR_APPEND;
8709 if (bi_flags & BTRFS_INODE_COMPRESS)
8710 stat->attributes |= STATX_ATTR_COMPRESSED;
8711 if (bi_flags & BTRFS_INODE_IMMUTABLE)
8712 stat->attributes |= STATX_ATTR_IMMUTABLE;
8713 if (bi_flags & BTRFS_INODE_NODUMP)
8714 stat->attributes |= STATX_ATTR_NODUMP;
8716 stat->attributes_mask |= (STATX_ATTR_APPEND |
8717 STATX_ATTR_COMPRESSED |
8718 STATX_ATTR_IMMUTABLE |
8721 generic_fillattr(inode, stat);
8722 stat->dev = BTRFS_I(inode)->root->anon_dev;
8724 spin_lock(&BTRFS_I(inode)->lock);
8725 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
8726 spin_unlock(&BTRFS_I(inode)->lock);
8727 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8728 ALIGN(delalloc_bytes, blocksize)) >> 9;
8732 static int btrfs_rename_exchange(struct inode *old_dir,
8733 struct dentry *old_dentry,
8734 struct inode *new_dir,
8735 struct dentry *new_dentry)
8737 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
8738 struct btrfs_trans_handle *trans;
8739 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8740 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8741 struct inode *new_inode = new_dentry->d_inode;
8742 struct inode *old_inode = old_dentry->d_inode;
8743 struct timespec64 ctime = current_time(old_inode);
8744 struct dentry *parent;
8745 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
8746 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
8750 bool root_log_pinned = false;
8751 bool dest_log_pinned = false;
8752 struct btrfs_log_ctx ctx_root;
8753 struct btrfs_log_ctx ctx_dest;
8754 bool sync_log_root = false;
8755 bool sync_log_dest = false;
8756 bool commit_transaction = false;
8758 /* we only allow rename subvolume link between subvolumes */
8759 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8762 btrfs_init_log_ctx(&ctx_root, old_inode);
8763 btrfs_init_log_ctx(&ctx_dest, new_inode);
8765 /* close the race window with snapshot create/destroy ioctl */
8766 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
8767 new_ino == BTRFS_FIRST_FREE_OBJECTID)
8768 down_read(&fs_info->subvol_sem);
8771 * We want to reserve the absolute worst case amount of items. So if
8772 * both inodes are subvols and we need to unlink them then that would
8773 * require 4 item modifications, but if they are both normal inodes it
8774 * would require 5 item modifications, so we'll assume their normal
8775 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
8776 * should cover the worst case number of items we'll modify.
8778 trans = btrfs_start_transaction(root, 12);
8779 if (IS_ERR(trans)) {
8780 ret = PTR_ERR(trans);
8785 btrfs_record_root_in_trans(trans, dest);
8788 * We need to find a free sequence number both in the source and
8789 * in the destination directory for the exchange.
8791 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
8794 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
8798 BTRFS_I(old_inode)->dir_index = 0ULL;
8799 BTRFS_I(new_inode)->dir_index = 0ULL;
8801 /* Reference for the source. */
8802 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8803 /* force full log commit if subvolume involved. */
8804 btrfs_set_log_full_commit(trans);
8806 btrfs_pin_log_trans(root);
8807 root_log_pinned = true;
8808 ret = btrfs_insert_inode_ref(trans, dest,
8809 new_dentry->d_name.name,
8810 new_dentry->d_name.len,
8812 btrfs_ino(BTRFS_I(new_dir)),
8818 /* And now for the dest. */
8819 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8820 /* force full log commit if subvolume involved. */
8821 btrfs_set_log_full_commit(trans);
8823 btrfs_pin_log_trans(dest);
8824 dest_log_pinned = true;
8825 ret = btrfs_insert_inode_ref(trans, root,
8826 old_dentry->d_name.name,
8827 old_dentry->d_name.len,
8829 btrfs_ino(BTRFS_I(old_dir)),
8835 /* Update inode version and ctime/mtime. */
8836 inode_inc_iversion(old_dir);
8837 inode_inc_iversion(new_dir);
8838 inode_inc_iversion(old_inode);
8839 inode_inc_iversion(new_inode);
8840 old_dir->i_ctime = old_dir->i_mtime = ctime;
8841 new_dir->i_ctime = new_dir->i_mtime = ctime;
8842 old_inode->i_ctime = ctime;
8843 new_inode->i_ctime = ctime;
8845 if (old_dentry->d_parent != new_dentry->d_parent) {
8846 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
8847 BTRFS_I(old_inode), 1);
8848 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
8849 BTRFS_I(new_inode), 1);
8852 /* src is a subvolume */
8853 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8854 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
8855 } else { /* src is an inode */
8856 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
8857 BTRFS_I(old_dentry->d_inode),
8858 old_dentry->d_name.name,
8859 old_dentry->d_name.len);
8861 ret = btrfs_update_inode(trans, root, old_inode);
8864 btrfs_abort_transaction(trans, ret);
8868 /* dest is a subvolume */
8869 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8870 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
8871 } else { /* dest is an inode */
8872 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
8873 BTRFS_I(new_dentry->d_inode),
8874 new_dentry->d_name.name,
8875 new_dentry->d_name.len);
8877 ret = btrfs_update_inode(trans, dest, new_inode);
8880 btrfs_abort_transaction(trans, ret);
8884 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
8885 new_dentry->d_name.name,
8886 new_dentry->d_name.len, 0, old_idx);
8888 btrfs_abort_transaction(trans, ret);
8892 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
8893 old_dentry->d_name.name,
8894 old_dentry->d_name.len, 0, new_idx);
8896 btrfs_abort_transaction(trans, ret);
8900 if (old_inode->i_nlink == 1)
8901 BTRFS_I(old_inode)->dir_index = old_idx;
8902 if (new_inode->i_nlink == 1)
8903 BTRFS_I(new_inode)->dir_index = new_idx;
8905 if (root_log_pinned) {
8906 parent = new_dentry->d_parent;
8907 ret = btrfs_log_new_name(trans, BTRFS_I(old_inode),
8908 BTRFS_I(old_dir), parent,
8910 if (ret == BTRFS_NEED_LOG_SYNC)
8911 sync_log_root = true;
8912 else if (ret == BTRFS_NEED_TRANS_COMMIT)
8913 commit_transaction = true;
8915 btrfs_end_log_trans(root);
8916 root_log_pinned = false;
8918 if (dest_log_pinned) {
8919 if (!commit_transaction) {
8920 parent = old_dentry->d_parent;
8921 ret = btrfs_log_new_name(trans, BTRFS_I(new_inode),
8922 BTRFS_I(new_dir), parent,
8924 if (ret == BTRFS_NEED_LOG_SYNC)
8925 sync_log_dest = true;
8926 else if (ret == BTRFS_NEED_TRANS_COMMIT)
8927 commit_transaction = true;
8930 btrfs_end_log_trans(dest);
8931 dest_log_pinned = false;
8935 * If we have pinned a log and an error happened, we unpin tasks
8936 * trying to sync the log and force them to fallback to a transaction
8937 * commit if the log currently contains any of the inodes involved in
8938 * this rename operation (to ensure we do not persist a log with an
8939 * inconsistent state for any of these inodes or leading to any
8940 * inconsistencies when replayed). If the transaction was aborted, the
8941 * abortion reason is propagated to userspace when attempting to commit
8942 * the transaction. If the log does not contain any of these inodes, we
8943 * allow the tasks to sync it.
8945 if (ret && (root_log_pinned || dest_log_pinned)) {
8946 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
8947 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
8948 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
8950 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
8951 btrfs_set_log_full_commit(trans);
8953 if (root_log_pinned) {
8954 btrfs_end_log_trans(root);
8955 root_log_pinned = false;
8957 if (dest_log_pinned) {
8958 btrfs_end_log_trans(dest);
8959 dest_log_pinned = false;
8962 if (!ret && sync_log_root && !commit_transaction) {
8963 ret = btrfs_sync_log(trans, BTRFS_I(old_inode)->root,
8966 commit_transaction = true;
8968 if (!ret && sync_log_dest && !commit_transaction) {
8969 ret = btrfs_sync_log(trans, BTRFS_I(new_inode)->root,
8972 commit_transaction = true;
8974 if (commit_transaction) {
8976 * We may have set commit_transaction when logging the new name
8977 * in the destination root, in which case we left the source
8978 * root context in the list of log contextes. So make sure we
8979 * remove it to avoid invalid memory accesses, since the context
8980 * was allocated in our stack frame.
8982 if (sync_log_root) {
8983 mutex_lock(&root->log_mutex);
8984 list_del_init(&ctx_root.list);
8985 mutex_unlock(&root->log_mutex);
8987 ret = btrfs_commit_transaction(trans);
8991 ret2 = btrfs_end_transaction(trans);
8992 ret = ret ? ret : ret2;
8995 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
8996 old_ino == BTRFS_FIRST_FREE_OBJECTID)
8997 up_read(&fs_info->subvol_sem);
8999 ASSERT(list_empty(&ctx_root.list));
9000 ASSERT(list_empty(&ctx_dest.list));
9005 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9006 struct btrfs_root *root,
9008 struct dentry *dentry)
9011 struct inode *inode;
9015 ret = btrfs_find_free_ino(root, &objectid);
9019 inode = btrfs_new_inode(trans, root, dir,
9020 dentry->d_name.name,
9022 btrfs_ino(BTRFS_I(dir)),
9024 S_IFCHR | WHITEOUT_MODE,
9027 if (IS_ERR(inode)) {
9028 ret = PTR_ERR(inode);
9032 inode->i_op = &btrfs_special_inode_operations;
9033 init_special_inode(inode, inode->i_mode,
9036 ret = btrfs_init_inode_security(trans, inode, dir,
9041 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9042 BTRFS_I(inode), 0, index);
9046 ret = btrfs_update_inode(trans, root, inode);
9048 unlock_new_inode(inode);
9050 inode_dec_link_count(inode);
9056 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9057 struct inode *new_dir, struct dentry *new_dentry,
9060 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9061 struct btrfs_trans_handle *trans;
9062 unsigned int trans_num_items;
9063 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9064 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9065 struct inode *new_inode = d_inode(new_dentry);
9066 struct inode *old_inode = d_inode(old_dentry);
9069 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9070 bool log_pinned = false;
9071 struct btrfs_log_ctx ctx;
9072 bool sync_log = false;
9073 bool commit_transaction = false;
9075 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9078 /* we only allow rename subvolume link between subvolumes */
9079 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9082 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9083 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9086 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9087 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9091 /* check for collisions, even if the name isn't there */
9092 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9093 new_dentry->d_name.name,
9094 new_dentry->d_name.len);
9097 if (ret == -EEXIST) {
9099 * eexist without a new_inode */
9100 if (WARN_ON(!new_inode)) {
9104 /* maybe -EOVERFLOW */
9111 * we're using rename to replace one file with another. Start IO on it
9112 * now so we don't add too much work to the end of the transaction
9114 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9115 filemap_flush(old_inode->i_mapping);
9117 /* close the racy window with snapshot create/destroy ioctl */
9118 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9119 down_read(&fs_info->subvol_sem);
9121 * We want to reserve the absolute worst case amount of items. So if
9122 * both inodes are subvols and we need to unlink them then that would
9123 * require 4 item modifications, but if they are both normal inodes it
9124 * would require 5 item modifications, so we'll assume they are normal
9125 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9126 * should cover the worst case number of items we'll modify.
9127 * If our rename has the whiteout flag, we need more 5 units for the
9128 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9129 * when selinux is enabled).
9131 trans_num_items = 11;
9132 if (flags & RENAME_WHITEOUT)
9133 trans_num_items += 5;
9134 trans = btrfs_start_transaction(root, trans_num_items);
9135 if (IS_ERR(trans)) {
9136 ret = PTR_ERR(trans);
9141 btrfs_record_root_in_trans(trans, dest);
9143 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9147 BTRFS_I(old_inode)->dir_index = 0ULL;
9148 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9149 /* force full log commit if subvolume involved. */
9150 btrfs_set_log_full_commit(trans);
9152 btrfs_pin_log_trans(root);
9154 ret = btrfs_insert_inode_ref(trans, dest,
9155 new_dentry->d_name.name,
9156 new_dentry->d_name.len,
9158 btrfs_ino(BTRFS_I(new_dir)), index);
9163 inode_inc_iversion(old_dir);
9164 inode_inc_iversion(new_dir);
9165 inode_inc_iversion(old_inode);
9166 old_dir->i_ctime = old_dir->i_mtime =
9167 new_dir->i_ctime = new_dir->i_mtime =
9168 old_inode->i_ctime = current_time(old_dir);
9170 if (old_dentry->d_parent != new_dentry->d_parent)
9171 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9172 BTRFS_I(old_inode), 1);
9174 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9175 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9177 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9178 BTRFS_I(d_inode(old_dentry)),
9179 old_dentry->d_name.name,
9180 old_dentry->d_name.len);
9182 ret = btrfs_update_inode(trans, root, old_inode);
9185 btrfs_abort_transaction(trans, ret);
9190 inode_inc_iversion(new_inode);
9191 new_inode->i_ctime = current_time(new_inode);
9192 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9193 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9194 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9195 BUG_ON(new_inode->i_nlink == 0);
9197 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9198 BTRFS_I(d_inode(new_dentry)),
9199 new_dentry->d_name.name,
9200 new_dentry->d_name.len);
9202 if (!ret && new_inode->i_nlink == 0)
9203 ret = btrfs_orphan_add(trans,
9204 BTRFS_I(d_inode(new_dentry)));
9206 btrfs_abort_transaction(trans, ret);
9211 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9212 new_dentry->d_name.name,
9213 new_dentry->d_name.len, 0, index);
9215 btrfs_abort_transaction(trans, ret);
9219 if (old_inode->i_nlink == 1)
9220 BTRFS_I(old_inode)->dir_index = index;
9223 struct dentry *parent = new_dentry->d_parent;
9225 btrfs_init_log_ctx(&ctx, old_inode);
9226 ret = btrfs_log_new_name(trans, BTRFS_I(old_inode),
9227 BTRFS_I(old_dir), parent,
9229 if (ret == BTRFS_NEED_LOG_SYNC)
9231 else if (ret == BTRFS_NEED_TRANS_COMMIT)
9232 commit_transaction = true;
9234 btrfs_end_log_trans(root);
9238 if (flags & RENAME_WHITEOUT) {
9239 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9243 btrfs_abort_transaction(trans, ret);
9249 * If we have pinned the log and an error happened, we unpin tasks
9250 * trying to sync the log and force them to fallback to a transaction
9251 * commit if the log currently contains any of the inodes involved in
9252 * this rename operation (to ensure we do not persist a log with an
9253 * inconsistent state for any of these inodes or leading to any
9254 * inconsistencies when replayed). If the transaction was aborted, the
9255 * abortion reason is propagated to userspace when attempting to commit
9256 * the transaction. If the log does not contain any of these inodes, we
9257 * allow the tasks to sync it.
9259 if (ret && log_pinned) {
9260 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9261 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9262 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9264 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9265 btrfs_set_log_full_commit(trans);
9267 btrfs_end_log_trans(root);
9270 if (!ret && sync_log) {
9271 ret = btrfs_sync_log(trans, BTRFS_I(old_inode)->root, &ctx);
9273 commit_transaction = true;
9274 } else if (sync_log) {
9275 mutex_lock(&root->log_mutex);
9276 list_del(&ctx.list);
9277 mutex_unlock(&root->log_mutex);
9279 if (commit_transaction) {
9280 ret = btrfs_commit_transaction(trans);
9284 ret2 = btrfs_end_transaction(trans);
9285 ret = ret ? ret : ret2;
9288 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9289 up_read(&fs_info->subvol_sem);
9294 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9295 struct inode *new_dir, struct dentry *new_dentry,
9298 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9301 if (flags & RENAME_EXCHANGE)
9302 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9305 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9308 struct btrfs_delalloc_work {
9309 struct inode *inode;
9310 struct completion completion;
9311 struct list_head list;
9312 struct btrfs_work work;
9315 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9317 struct btrfs_delalloc_work *delalloc_work;
9318 struct inode *inode;
9320 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9322 inode = delalloc_work->inode;
9323 filemap_flush(inode->i_mapping);
9324 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9325 &BTRFS_I(inode)->runtime_flags))
9326 filemap_flush(inode->i_mapping);
9329 complete(&delalloc_work->completion);
9332 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9334 struct btrfs_delalloc_work *work;
9336 work = kmalloc(sizeof(*work), GFP_NOFS);
9340 init_completion(&work->completion);
9341 INIT_LIST_HEAD(&work->list);
9342 work->inode = inode;
9343 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9349 * some fairly slow code that needs optimization. This walks the list
9350 * of all the inodes with pending delalloc and forces them to disk.
9352 static int start_delalloc_inodes(struct btrfs_root *root, int nr, bool snapshot)
9354 struct btrfs_inode *binode;
9355 struct inode *inode;
9356 struct btrfs_delalloc_work *work, *next;
9357 struct list_head works;
9358 struct list_head splice;
9361 INIT_LIST_HEAD(&works);
9362 INIT_LIST_HEAD(&splice);
9364 mutex_lock(&root->delalloc_mutex);
9365 spin_lock(&root->delalloc_lock);
9366 list_splice_init(&root->delalloc_inodes, &splice);
9367 while (!list_empty(&splice)) {
9368 binode = list_entry(splice.next, struct btrfs_inode,
9371 list_move_tail(&binode->delalloc_inodes,
9372 &root->delalloc_inodes);
9373 inode = igrab(&binode->vfs_inode);
9375 cond_resched_lock(&root->delalloc_lock);
9378 spin_unlock(&root->delalloc_lock);
9381 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9382 &binode->runtime_flags);
9383 work = btrfs_alloc_delalloc_work(inode);
9389 list_add_tail(&work->list, &works);
9390 btrfs_queue_work(root->fs_info->flush_workers,
9393 if (nr != -1 && ret >= nr)
9396 spin_lock(&root->delalloc_lock);
9398 spin_unlock(&root->delalloc_lock);
9401 list_for_each_entry_safe(work, next, &works, list) {
9402 list_del_init(&work->list);
9403 wait_for_completion(&work->completion);
9407 if (!list_empty(&splice)) {
9408 spin_lock(&root->delalloc_lock);
9409 list_splice_tail(&splice, &root->delalloc_inodes);
9410 spin_unlock(&root->delalloc_lock);
9412 mutex_unlock(&root->delalloc_mutex);
9416 int btrfs_start_delalloc_snapshot(struct btrfs_root *root)
9418 struct btrfs_fs_info *fs_info = root->fs_info;
9421 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9424 ret = start_delalloc_inodes(root, -1, true);
9430 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int nr)
9432 struct btrfs_root *root;
9433 struct list_head splice;
9436 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9439 INIT_LIST_HEAD(&splice);
9441 mutex_lock(&fs_info->delalloc_root_mutex);
9442 spin_lock(&fs_info->delalloc_root_lock);
9443 list_splice_init(&fs_info->delalloc_roots, &splice);
9444 while (!list_empty(&splice) && nr) {
9445 root = list_first_entry(&splice, struct btrfs_root,
9447 root = btrfs_grab_root(root);
9449 list_move_tail(&root->delalloc_root,
9450 &fs_info->delalloc_roots);
9451 spin_unlock(&fs_info->delalloc_root_lock);
9453 ret = start_delalloc_inodes(root, nr, false);
9454 btrfs_put_root(root);
9462 spin_lock(&fs_info->delalloc_root_lock);
9464 spin_unlock(&fs_info->delalloc_root_lock);
9468 if (!list_empty(&splice)) {
9469 spin_lock(&fs_info->delalloc_root_lock);
9470 list_splice_tail(&splice, &fs_info->delalloc_roots);
9471 spin_unlock(&fs_info->delalloc_root_lock);
9473 mutex_unlock(&fs_info->delalloc_root_mutex);
9477 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9478 const char *symname)
9480 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9481 struct btrfs_trans_handle *trans;
9482 struct btrfs_root *root = BTRFS_I(dir)->root;
9483 struct btrfs_path *path;
9484 struct btrfs_key key;
9485 struct inode *inode = NULL;
9492 struct btrfs_file_extent_item *ei;
9493 struct extent_buffer *leaf;
9495 name_len = strlen(symname);
9496 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9497 return -ENAMETOOLONG;
9500 * 2 items for inode item and ref
9501 * 2 items for dir items
9502 * 1 item for updating parent inode item
9503 * 1 item for the inline extent item
9504 * 1 item for xattr if selinux is on
9506 trans = btrfs_start_transaction(root, 7);
9508 return PTR_ERR(trans);
9510 err = btrfs_find_free_ino(root, &objectid);
9514 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9515 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
9516 objectid, S_IFLNK|S_IRWXUGO, &index);
9517 if (IS_ERR(inode)) {
9518 err = PTR_ERR(inode);
9524 * If the active LSM wants to access the inode during
9525 * d_instantiate it needs these. Smack checks to see
9526 * if the filesystem supports xattrs by looking at the
9529 inode->i_fop = &btrfs_file_operations;
9530 inode->i_op = &btrfs_file_inode_operations;
9531 inode->i_mapping->a_ops = &btrfs_aops;
9532 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9534 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9538 path = btrfs_alloc_path();
9543 key.objectid = btrfs_ino(BTRFS_I(inode));
9545 key.type = BTRFS_EXTENT_DATA_KEY;
9546 datasize = btrfs_file_extent_calc_inline_size(name_len);
9547 err = btrfs_insert_empty_item(trans, root, path, &key,
9550 btrfs_free_path(path);
9553 leaf = path->nodes[0];
9554 ei = btrfs_item_ptr(leaf, path->slots[0],
9555 struct btrfs_file_extent_item);
9556 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9557 btrfs_set_file_extent_type(leaf, ei,
9558 BTRFS_FILE_EXTENT_INLINE);
9559 btrfs_set_file_extent_encryption(leaf, ei, 0);
9560 btrfs_set_file_extent_compression(leaf, ei, 0);
9561 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9562 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9564 ptr = btrfs_file_extent_inline_start(ei);
9565 write_extent_buffer(leaf, symname, ptr, name_len);
9566 btrfs_mark_buffer_dirty(leaf);
9567 btrfs_free_path(path);
9569 inode->i_op = &btrfs_symlink_inode_operations;
9570 inode_nohighmem(inode);
9571 inode_set_bytes(inode, name_len);
9572 btrfs_i_size_write(BTRFS_I(inode), name_len);
9573 err = btrfs_update_inode(trans, root, inode);
9575 * Last step, add directory indexes for our symlink inode. This is the
9576 * last step to avoid extra cleanup of these indexes if an error happens
9580 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9581 BTRFS_I(inode), 0, index);
9585 d_instantiate_new(dentry, inode);
9588 btrfs_end_transaction(trans);
9590 inode_dec_link_count(inode);
9591 discard_new_inode(inode);
9593 btrfs_btree_balance_dirty(fs_info);
9597 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9598 u64 start, u64 num_bytes, u64 min_size,
9599 loff_t actual_len, u64 *alloc_hint,
9600 struct btrfs_trans_handle *trans)
9602 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9603 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9604 struct extent_map *em;
9605 struct btrfs_root *root = BTRFS_I(inode)->root;
9606 struct btrfs_key ins;
9607 u64 cur_offset = start;
9608 u64 clear_offset = start;
9611 u64 last_alloc = (u64)-1;
9613 bool own_trans = true;
9614 u64 end = start + num_bytes - 1;
9618 while (num_bytes > 0) {
9620 trans = btrfs_start_transaction(root, 3);
9621 if (IS_ERR(trans)) {
9622 ret = PTR_ERR(trans);
9627 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9628 cur_bytes = max(cur_bytes, min_size);
9630 * If we are severely fragmented we could end up with really
9631 * small allocations, so if the allocator is returning small
9632 * chunks lets make its job easier by only searching for those
9635 cur_bytes = min(cur_bytes, last_alloc);
9636 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9637 min_size, 0, *alloc_hint, &ins, 1, 0);
9640 btrfs_end_transaction(trans);
9645 * We've reserved this space, and thus converted it from
9646 * ->bytes_may_use to ->bytes_reserved. Any error that happens
9647 * from here on out we will only need to clear our reservation
9648 * for the remaining unreserved area, so advance our
9649 * clear_offset by our extent size.
9651 clear_offset += ins.offset;
9652 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
9654 last_alloc = ins.offset;
9655 ret = insert_reserved_file_extent(trans, inode,
9656 cur_offset, ins.objectid,
9657 ins.offset, ins.offset,
9658 ins.offset, 0, 0, 0,
9659 BTRFS_FILE_EXTENT_PREALLOC);
9661 btrfs_free_reserved_extent(fs_info, ins.objectid,
9663 btrfs_abort_transaction(trans, ret);
9665 btrfs_end_transaction(trans);
9669 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
9670 cur_offset + ins.offset -1, 0);
9672 em = alloc_extent_map();
9674 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9675 &BTRFS_I(inode)->runtime_flags);
9679 em->start = cur_offset;
9680 em->orig_start = cur_offset;
9681 em->len = ins.offset;
9682 em->block_start = ins.objectid;
9683 em->block_len = ins.offset;
9684 em->orig_block_len = ins.offset;
9685 em->ram_bytes = ins.offset;
9686 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9687 em->generation = trans->transid;
9690 write_lock(&em_tree->lock);
9691 ret = add_extent_mapping(em_tree, em, 1);
9692 write_unlock(&em_tree->lock);
9695 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
9696 cur_offset + ins.offset - 1,
9699 free_extent_map(em);
9701 num_bytes -= ins.offset;
9702 cur_offset += ins.offset;
9703 *alloc_hint = ins.objectid + ins.offset;
9705 inode_inc_iversion(inode);
9706 inode->i_ctime = current_time(inode);
9707 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9708 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9709 (actual_len > inode->i_size) &&
9710 (cur_offset > inode->i_size)) {
9711 if (cur_offset > actual_len)
9712 i_size = actual_len;
9714 i_size = cur_offset;
9715 i_size_write(inode, i_size);
9716 btrfs_inode_safe_disk_i_size_write(inode, 0);
9719 ret = btrfs_update_inode(trans, root, inode);
9722 btrfs_abort_transaction(trans, ret);
9724 btrfs_end_transaction(trans);
9729 btrfs_end_transaction(trans);
9731 if (clear_offset < end)
9732 btrfs_free_reserved_data_space(inode, NULL, clear_offset,
9733 end - clear_offset + 1);
9737 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9738 u64 start, u64 num_bytes, u64 min_size,
9739 loff_t actual_len, u64 *alloc_hint)
9741 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9742 min_size, actual_len, alloc_hint,
9746 int btrfs_prealloc_file_range_trans(struct inode *inode,
9747 struct btrfs_trans_handle *trans, int mode,
9748 u64 start, u64 num_bytes, u64 min_size,
9749 loff_t actual_len, u64 *alloc_hint)
9751 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9752 min_size, actual_len, alloc_hint, trans);
9755 static int btrfs_set_page_dirty(struct page *page)
9757 return __set_page_dirty_nobuffers(page);
9760 static int btrfs_permission(struct inode *inode, int mask)
9762 struct btrfs_root *root = BTRFS_I(inode)->root;
9763 umode_t mode = inode->i_mode;
9765 if (mask & MAY_WRITE &&
9766 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9767 if (btrfs_root_readonly(root))
9769 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9772 return generic_permission(inode, mask);
9775 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
9777 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9778 struct btrfs_trans_handle *trans;
9779 struct btrfs_root *root = BTRFS_I(dir)->root;
9780 struct inode *inode = NULL;
9786 * 5 units required for adding orphan entry
9788 trans = btrfs_start_transaction(root, 5);
9790 return PTR_ERR(trans);
9792 ret = btrfs_find_free_ino(root, &objectid);
9796 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
9797 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
9798 if (IS_ERR(inode)) {
9799 ret = PTR_ERR(inode);
9804 inode->i_fop = &btrfs_file_operations;
9805 inode->i_op = &btrfs_file_inode_operations;
9807 inode->i_mapping->a_ops = &btrfs_aops;
9808 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9810 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
9814 ret = btrfs_update_inode(trans, root, inode);
9817 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
9822 * We set number of links to 0 in btrfs_new_inode(), and here we set
9823 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9826 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9828 set_nlink(inode, 1);
9829 d_tmpfile(dentry, inode);
9830 unlock_new_inode(inode);
9831 mark_inode_dirty(inode);
9833 btrfs_end_transaction(trans);
9835 discard_new_inode(inode);
9836 btrfs_btree_balance_dirty(fs_info);
9840 void btrfs_set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
9842 struct inode *inode = tree->private_data;
9843 unsigned long index = start >> PAGE_SHIFT;
9844 unsigned long end_index = end >> PAGE_SHIFT;
9847 while (index <= end_index) {
9848 page = find_get_page(inode->i_mapping, index);
9849 ASSERT(page); /* Pages should be in the extent_io_tree */
9850 set_page_writeback(page);
9858 * Add an entry indicating a block group or device which is pinned by a
9859 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
9860 * negative errno on failure.
9862 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
9863 bool is_block_group)
9865 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
9866 struct btrfs_swapfile_pin *sp, *entry;
9868 struct rb_node *parent = NULL;
9870 sp = kmalloc(sizeof(*sp), GFP_NOFS);
9875 sp->is_block_group = is_block_group;
9877 spin_lock(&fs_info->swapfile_pins_lock);
9878 p = &fs_info->swapfile_pins.rb_node;
9881 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
9882 if (sp->ptr < entry->ptr ||
9883 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
9885 } else if (sp->ptr > entry->ptr ||
9886 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
9887 p = &(*p)->rb_right;
9889 spin_unlock(&fs_info->swapfile_pins_lock);
9894 rb_link_node(&sp->node, parent, p);
9895 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
9896 spin_unlock(&fs_info->swapfile_pins_lock);
9900 /* Free all of the entries pinned by this swapfile. */
9901 static void btrfs_free_swapfile_pins(struct inode *inode)
9903 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
9904 struct btrfs_swapfile_pin *sp;
9905 struct rb_node *node, *next;
9907 spin_lock(&fs_info->swapfile_pins_lock);
9908 node = rb_first(&fs_info->swapfile_pins);
9910 next = rb_next(node);
9911 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
9912 if (sp->inode == inode) {
9913 rb_erase(&sp->node, &fs_info->swapfile_pins);
9914 if (sp->is_block_group)
9915 btrfs_put_block_group(sp->ptr);
9920 spin_unlock(&fs_info->swapfile_pins_lock);
9923 struct btrfs_swap_info {
9929 unsigned long nr_pages;
9933 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
9934 struct btrfs_swap_info *bsi)
9936 unsigned long nr_pages;
9937 u64 first_ppage, first_ppage_reported, next_ppage;
9940 first_ppage = ALIGN(bsi->block_start, PAGE_SIZE) >> PAGE_SHIFT;
9941 next_ppage = ALIGN_DOWN(bsi->block_start + bsi->block_len,
9942 PAGE_SIZE) >> PAGE_SHIFT;
9944 if (first_ppage >= next_ppage)
9946 nr_pages = next_ppage - first_ppage;
9948 first_ppage_reported = first_ppage;
9949 if (bsi->start == 0)
9950 first_ppage_reported++;
9951 if (bsi->lowest_ppage > first_ppage_reported)
9952 bsi->lowest_ppage = first_ppage_reported;
9953 if (bsi->highest_ppage < (next_ppage - 1))
9954 bsi->highest_ppage = next_ppage - 1;
9956 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
9959 bsi->nr_extents += ret;
9960 bsi->nr_pages += nr_pages;
9964 static void btrfs_swap_deactivate(struct file *file)
9966 struct inode *inode = file_inode(file);
9968 btrfs_free_swapfile_pins(inode);
9969 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
9972 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
9975 struct inode *inode = file_inode(file);
9976 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
9977 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
9978 struct extent_state *cached_state = NULL;
9979 struct extent_map *em = NULL;
9980 struct btrfs_device *device = NULL;
9981 struct btrfs_swap_info bsi = {
9982 .lowest_ppage = (sector_t)-1ULL,
9989 * If the swap file was just created, make sure delalloc is done. If the
9990 * file changes again after this, the user is doing something stupid and
9991 * we don't really care.
9993 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
9998 * The inode is locked, so these flags won't change after we check them.
10000 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10001 btrfs_warn(fs_info, "swapfile must not be compressed");
10004 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10005 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10008 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10009 btrfs_warn(fs_info, "swapfile must not be checksummed");
10014 * Balance or device remove/replace/resize can move stuff around from
10015 * under us. The EXCL_OP flag makes sure they aren't running/won't run
10016 * concurrently while we are mapping the swap extents, and
10017 * fs_info->swapfile_pins prevents them from running while the swap file
10018 * is active and moving the extents. Note that this also prevents a
10019 * concurrent device add which isn't actually necessary, but it's not
10020 * really worth the trouble to allow it.
10022 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
10023 btrfs_warn(fs_info,
10024 "cannot activate swapfile while exclusive operation is running");
10028 * Snapshots can create extents which require COW even if NODATACOW is
10029 * set. We use this counter to prevent snapshots. We must increment it
10030 * before walking the extents because we don't want a concurrent
10031 * snapshot to run after we've already checked the extents.
10033 atomic_inc(&BTRFS_I(inode)->root->nr_swapfiles);
10035 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10037 lock_extent_bits(io_tree, 0, isize - 1, &cached_state);
10039 while (start < isize) {
10040 u64 logical_block_start, physical_block_start;
10041 struct btrfs_block_group *bg;
10042 u64 len = isize - start;
10044 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
10050 if (em->block_start == EXTENT_MAP_HOLE) {
10051 btrfs_warn(fs_info, "swapfile must not have holes");
10055 if (em->block_start == EXTENT_MAP_INLINE) {
10057 * It's unlikely we'll ever actually find ourselves
10058 * here, as a file small enough to fit inline won't be
10059 * big enough to store more than the swap header, but in
10060 * case something changes in the future, let's catch it
10061 * here rather than later.
10063 btrfs_warn(fs_info, "swapfile must not be inline");
10067 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10068 btrfs_warn(fs_info, "swapfile must not be compressed");
10073 logical_block_start = em->block_start + (start - em->start);
10074 len = min(len, em->len - (start - em->start));
10075 free_extent_map(em);
10078 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL);
10084 btrfs_warn(fs_info,
10085 "swapfile must not be copy-on-write");
10090 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10096 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10097 btrfs_warn(fs_info,
10098 "swapfile must have single data profile");
10103 if (device == NULL) {
10104 device = em->map_lookup->stripes[0].dev;
10105 ret = btrfs_add_swapfile_pin(inode, device, false);
10110 } else if (device != em->map_lookup->stripes[0].dev) {
10111 btrfs_warn(fs_info, "swapfile must be on one device");
10116 physical_block_start = (em->map_lookup->stripes[0].physical +
10117 (logical_block_start - em->start));
10118 len = min(len, em->len - (logical_block_start - em->start));
10119 free_extent_map(em);
10122 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10124 btrfs_warn(fs_info,
10125 "could not find block group containing swapfile");
10130 ret = btrfs_add_swapfile_pin(inode, bg, true);
10132 btrfs_put_block_group(bg);
10139 if (bsi.block_len &&
10140 bsi.block_start + bsi.block_len == physical_block_start) {
10141 bsi.block_len += len;
10143 if (bsi.block_len) {
10144 ret = btrfs_add_swap_extent(sis, &bsi);
10149 bsi.block_start = physical_block_start;
10150 bsi.block_len = len;
10157 ret = btrfs_add_swap_extent(sis, &bsi);
10160 if (!IS_ERR_OR_NULL(em))
10161 free_extent_map(em);
10163 unlock_extent_cached(io_tree, 0, isize - 1, &cached_state);
10166 btrfs_swap_deactivate(file);
10168 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
10174 sis->bdev = device->bdev;
10175 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10176 sis->max = bsi.nr_pages;
10177 sis->pages = bsi.nr_pages - 1;
10178 sis->highest_bit = bsi.nr_pages - 1;
10179 return bsi.nr_extents;
10182 static void btrfs_swap_deactivate(struct file *file)
10186 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10189 return -EOPNOTSUPP;
10193 static const struct inode_operations btrfs_dir_inode_operations = {
10194 .getattr = btrfs_getattr,
10195 .lookup = btrfs_lookup,
10196 .create = btrfs_create,
10197 .unlink = btrfs_unlink,
10198 .link = btrfs_link,
10199 .mkdir = btrfs_mkdir,
10200 .rmdir = btrfs_rmdir,
10201 .rename = btrfs_rename2,
10202 .symlink = btrfs_symlink,
10203 .setattr = btrfs_setattr,
10204 .mknod = btrfs_mknod,
10205 .listxattr = btrfs_listxattr,
10206 .permission = btrfs_permission,
10207 .get_acl = btrfs_get_acl,
10208 .set_acl = btrfs_set_acl,
10209 .update_time = btrfs_update_time,
10210 .tmpfile = btrfs_tmpfile,
10213 static const struct file_operations btrfs_dir_file_operations = {
10214 .llseek = generic_file_llseek,
10215 .read = generic_read_dir,
10216 .iterate_shared = btrfs_real_readdir,
10217 .open = btrfs_opendir,
10218 .unlocked_ioctl = btrfs_ioctl,
10219 #ifdef CONFIG_COMPAT
10220 .compat_ioctl = btrfs_compat_ioctl,
10222 .release = btrfs_release_file,
10223 .fsync = btrfs_sync_file,
10226 static const struct extent_io_ops btrfs_extent_io_ops = {
10227 /* mandatory callbacks */
10228 .submit_bio_hook = btrfs_submit_bio_hook,
10229 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10233 * btrfs doesn't support the bmap operation because swapfiles
10234 * use bmap to make a mapping of extents in the file. They assume
10235 * these extents won't change over the life of the file and they
10236 * use the bmap result to do IO directly to the drive.
10238 * the btrfs bmap call would return logical addresses that aren't
10239 * suitable for IO and they also will change frequently as COW
10240 * operations happen. So, swapfile + btrfs == corruption.
10242 * For now we're avoiding this by dropping bmap.
10244 static const struct address_space_operations btrfs_aops = {
10245 .readpage = btrfs_readpage,
10246 .writepage = btrfs_writepage,
10247 .writepages = btrfs_writepages,
10248 .readahead = btrfs_readahead,
10249 .direct_IO = btrfs_direct_IO,
10250 .invalidatepage = btrfs_invalidatepage,
10251 .releasepage = btrfs_releasepage,
10252 #ifdef CONFIG_MIGRATION
10253 .migratepage = btrfs_migratepage,
10255 .set_page_dirty = btrfs_set_page_dirty,
10256 .error_remove_page = generic_error_remove_page,
10257 .swap_activate = btrfs_swap_activate,
10258 .swap_deactivate = btrfs_swap_deactivate,
10261 static const struct inode_operations btrfs_file_inode_operations = {
10262 .getattr = btrfs_getattr,
10263 .setattr = btrfs_setattr,
10264 .listxattr = btrfs_listxattr,
10265 .permission = btrfs_permission,
10266 .fiemap = btrfs_fiemap,
10267 .get_acl = btrfs_get_acl,
10268 .set_acl = btrfs_set_acl,
10269 .update_time = btrfs_update_time,
10271 static const struct inode_operations btrfs_special_inode_operations = {
10272 .getattr = btrfs_getattr,
10273 .setattr = btrfs_setattr,
10274 .permission = btrfs_permission,
10275 .listxattr = btrfs_listxattr,
10276 .get_acl = btrfs_get_acl,
10277 .set_acl = btrfs_set_acl,
10278 .update_time = btrfs_update_time,
10280 static const struct inode_operations btrfs_symlink_inode_operations = {
10281 .get_link = page_get_link,
10282 .getattr = btrfs_getattr,
10283 .setattr = btrfs_setattr,
10284 .permission = btrfs_permission,
10285 .listxattr = btrfs_listxattr,
10286 .update_time = btrfs_update_time,
10289 const struct dentry_operations btrfs_dentry_operations = {
10290 .d_delete = btrfs_dentry_delete,