1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <crypto/hash.h>
7 #include <linux/kernel.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 <linux/iomap.h>
34 #include <asm/unaligned.h>
38 #include "transaction.h"
39 #include "btrfs_inode.h"
40 #include "print-tree.h"
41 #include "ordered-data.h"
45 #include "compression.h"
47 #include "free-space-cache.h"
50 #include "delalloc-space.h"
51 #include "block-group.h"
52 #include "space-info.h"
56 struct btrfs_iget_args {
58 struct btrfs_root *root;
61 struct btrfs_dio_data {
65 struct extent_changeset *data_reserved;
68 static const struct inode_operations btrfs_dir_inode_operations;
69 static const struct inode_operations btrfs_symlink_inode_operations;
70 static const struct inode_operations btrfs_special_inode_operations;
71 static const struct inode_operations btrfs_file_inode_operations;
72 static const struct address_space_operations btrfs_aops;
73 static const struct file_operations btrfs_dir_file_operations;
75 static struct kmem_cache *btrfs_inode_cachep;
76 struct kmem_cache *btrfs_trans_handle_cachep;
77 struct kmem_cache *btrfs_path_cachep;
78 struct kmem_cache *btrfs_free_space_cachep;
79 struct kmem_cache *btrfs_free_space_bitmap_cachep;
81 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
82 static int btrfs_truncate(struct inode *inode, bool skip_writeback);
83 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
84 static noinline int cow_file_range(struct btrfs_inode *inode,
85 struct page *locked_page,
86 u64 start, u64 end, int *page_started,
87 unsigned long *nr_written, int unlock);
88 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
89 u64 len, u64 orig_start, u64 block_start,
90 u64 block_len, u64 orig_block_len,
91 u64 ram_bytes, int compress_type,
94 static void __endio_write_update_ordered(struct btrfs_inode *inode,
95 const u64 offset, const u64 bytes,
99 * btrfs_inode_lock - lock inode i_rwsem based on arguments passed
101 * ilock_flags can have the following bit set:
103 * BTRFS_ILOCK_SHARED - acquire a shared lock on the inode
104 * BTRFS_ILOCK_TRY - try to acquire the lock, if fails on first attempt
106 * BTRFS_ILOCK_MMAP - acquire a write lock on the i_mmap_lock
108 int btrfs_inode_lock(struct inode *inode, unsigned int ilock_flags)
110 if (ilock_flags & BTRFS_ILOCK_SHARED) {
111 if (ilock_flags & BTRFS_ILOCK_TRY) {
112 if (!inode_trylock_shared(inode))
117 inode_lock_shared(inode);
119 if (ilock_flags & BTRFS_ILOCK_TRY) {
120 if (!inode_trylock(inode))
127 if (ilock_flags & BTRFS_ILOCK_MMAP)
128 down_write(&BTRFS_I(inode)->i_mmap_lock);
133 * btrfs_inode_unlock - unock inode i_rwsem
135 * ilock_flags should contain the same bits set as passed to btrfs_inode_lock()
136 * to decide whether the lock acquired is shared or exclusive.
138 void btrfs_inode_unlock(struct inode *inode, unsigned int ilock_flags)
140 if (ilock_flags & BTRFS_ILOCK_MMAP)
141 up_write(&BTRFS_I(inode)->i_mmap_lock);
142 if (ilock_flags & BTRFS_ILOCK_SHARED)
143 inode_unlock_shared(inode);
149 * Cleanup all submitted ordered extents in specified range to handle errors
150 * from the btrfs_run_delalloc_range() callback.
152 * NOTE: caller must ensure that when an error happens, it can not call
153 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
154 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
155 * to be released, which we want to happen only when finishing the ordered
156 * extent (btrfs_finish_ordered_io()).
158 static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode,
159 struct page *locked_page,
160 u64 offset, u64 bytes)
162 unsigned long index = offset >> PAGE_SHIFT;
163 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
164 u64 page_start = page_offset(locked_page);
165 u64 page_end = page_start + PAGE_SIZE - 1;
169 while (index <= end_index) {
171 * For locked page, we will call end_extent_writepage() on it
172 * in run_delalloc_range() for the error handling. That
173 * end_extent_writepage() function will call
174 * btrfs_mark_ordered_io_finished() to clear page Ordered and
175 * run the ordered extent accounting.
177 * Here we can't just clear the Ordered bit, or
178 * btrfs_mark_ordered_io_finished() would skip the accounting
179 * for the page range, and the ordered extent will never finish.
181 if (index == (page_offset(locked_page) >> PAGE_SHIFT)) {
185 page = find_get_page(inode->vfs_inode.i_mapping, index);
191 * Here we just clear all Ordered bits for every page in the
192 * range, then __endio_write_update_ordered() will handle
193 * the ordered extent accounting for the range.
195 btrfs_page_clamp_clear_ordered(inode->root->fs_info, page,
200 /* The locked page covers the full range, nothing needs to be done */
201 if (bytes + offset <= page_offset(locked_page) + PAGE_SIZE)
204 * In case this page belongs to the delalloc range being instantiated
205 * then skip it, since the first page of a range is going to be
206 * properly cleaned up by the caller of run_delalloc_range
208 if (page_start >= offset && page_end <= (offset + bytes - 1)) {
209 bytes = offset + bytes - page_offset(locked_page) - PAGE_SIZE;
210 offset = page_offset(locked_page) + PAGE_SIZE;
213 return __endio_write_update_ordered(inode, offset, bytes, false);
216 static int btrfs_dirty_inode(struct inode *inode);
218 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
219 struct inode *inode, struct inode *dir,
220 const struct qstr *qstr)
224 err = btrfs_init_acl(trans, inode, dir);
226 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
231 * this does all the hard work for inserting an inline extent into
232 * the btree. The caller should have done a btrfs_drop_extents so that
233 * no overlapping inline items exist in the btree
235 static int insert_inline_extent(struct btrfs_trans_handle *trans,
236 struct btrfs_path *path, bool extent_inserted,
237 struct btrfs_root *root, struct inode *inode,
238 u64 start, size_t size, size_t compressed_size,
240 struct page **compressed_pages)
242 struct extent_buffer *leaf;
243 struct page *page = NULL;
246 struct btrfs_file_extent_item *ei;
248 size_t cur_size = size;
249 unsigned long offset;
251 ASSERT((compressed_size > 0 && compressed_pages) ||
252 (compressed_size == 0 && !compressed_pages));
254 if (compressed_size && compressed_pages)
255 cur_size = compressed_size;
257 if (!extent_inserted) {
258 struct btrfs_key key;
261 key.objectid = btrfs_ino(BTRFS_I(inode));
263 key.type = BTRFS_EXTENT_DATA_KEY;
265 datasize = btrfs_file_extent_calc_inline_size(cur_size);
266 ret = btrfs_insert_empty_item(trans, root, path, &key,
271 leaf = path->nodes[0];
272 ei = btrfs_item_ptr(leaf, path->slots[0],
273 struct btrfs_file_extent_item);
274 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
275 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
276 btrfs_set_file_extent_encryption(leaf, ei, 0);
277 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
278 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
279 ptr = btrfs_file_extent_inline_start(ei);
281 if (compress_type != BTRFS_COMPRESS_NONE) {
284 while (compressed_size > 0) {
285 cpage = compressed_pages[i];
286 cur_size = min_t(unsigned long, compressed_size,
289 kaddr = kmap_atomic(cpage);
290 write_extent_buffer(leaf, kaddr, ptr, cur_size);
291 kunmap_atomic(kaddr);
295 compressed_size -= cur_size;
297 btrfs_set_file_extent_compression(leaf, ei,
300 page = find_get_page(inode->i_mapping,
301 start >> PAGE_SHIFT);
302 btrfs_set_file_extent_compression(leaf, ei, 0);
303 kaddr = kmap_atomic(page);
304 offset = offset_in_page(start);
305 write_extent_buffer(leaf, kaddr + offset, ptr, size);
306 kunmap_atomic(kaddr);
309 btrfs_mark_buffer_dirty(leaf);
310 btrfs_release_path(path);
313 * We align size to sectorsize for inline extents just for simplicity
316 size = ALIGN(size, root->fs_info->sectorsize);
317 ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start, size);
322 * we're an inline extent, so nobody can
323 * extend the file past i_size without locking
324 * a page we already have locked.
326 * We must do any isize and inode updates
327 * before we unlock the pages. Otherwise we
328 * could end up racing with unlink.
330 BTRFS_I(inode)->disk_i_size = inode->i_size;
337 * conditionally insert an inline extent into the file. This
338 * does the checks required to make sure the data is small enough
339 * to fit as an inline extent.
341 static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 start,
342 u64 end, size_t compressed_size,
344 struct page **compressed_pages)
346 struct btrfs_drop_extents_args drop_args = { 0 };
347 struct btrfs_root *root = inode->root;
348 struct btrfs_fs_info *fs_info = root->fs_info;
349 struct btrfs_trans_handle *trans;
350 u64 isize = i_size_read(&inode->vfs_inode);
351 u64 actual_end = min(end + 1, isize);
352 u64 inline_len = actual_end - start;
353 u64 aligned_end = ALIGN(end, fs_info->sectorsize);
354 u64 data_len = inline_len;
356 struct btrfs_path *path;
359 data_len = compressed_size;
362 actual_end > fs_info->sectorsize ||
363 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
365 (actual_end & (fs_info->sectorsize - 1)) == 0) ||
367 data_len > fs_info->max_inline) {
371 path = btrfs_alloc_path();
375 trans = btrfs_join_transaction(root);
377 btrfs_free_path(path);
378 return PTR_ERR(trans);
380 trans->block_rsv = &inode->block_rsv;
382 drop_args.path = path;
383 drop_args.start = start;
384 drop_args.end = aligned_end;
385 drop_args.drop_cache = true;
386 drop_args.replace_extent = true;
388 if (compressed_size && compressed_pages)
389 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(
392 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(
395 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
397 btrfs_abort_transaction(trans, ret);
401 if (isize > actual_end)
402 inline_len = min_t(u64, isize, actual_end);
403 ret = insert_inline_extent(trans, path, drop_args.extent_inserted,
404 root, &inode->vfs_inode, start,
405 inline_len, compressed_size,
406 compress_type, compressed_pages);
407 if (ret && ret != -ENOSPC) {
408 btrfs_abort_transaction(trans, ret);
410 } else if (ret == -ENOSPC) {
415 btrfs_update_inode_bytes(inode, inline_len, drop_args.bytes_found);
416 ret = btrfs_update_inode(trans, root, inode);
417 if (ret && ret != -ENOSPC) {
418 btrfs_abort_transaction(trans, ret);
420 } else if (ret == -ENOSPC) {
425 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
428 * Don't forget to free the reserved space, as for inlined extent
429 * it won't count as data extent, free them directly here.
430 * And at reserve time, it's always aligned to page size, so
431 * just free one page here.
433 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
434 btrfs_free_path(path);
435 btrfs_end_transaction(trans);
439 struct async_extent {
444 unsigned long nr_pages;
446 struct list_head list;
451 struct page *locked_page;
454 unsigned int write_flags;
455 struct list_head extents;
456 struct cgroup_subsys_state *blkcg_css;
457 struct btrfs_work work;
462 /* Number of chunks in flight; must be first in the structure */
464 struct async_chunk chunks[];
467 static noinline int add_async_extent(struct async_chunk *cow,
468 u64 start, u64 ram_size,
471 unsigned long nr_pages,
474 struct async_extent *async_extent;
476 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
477 BUG_ON(!async_extent); /* -ENOMEM */
478 async_extent->start = start;
479 async_extent->ram_size = ram_size;
480 async_extent->compressed_size = compressed_size;
481 async_extent->pages = pages;
482 async_extent->nr_pages = nr_pages;
483 async_extent->compress_type = compress_type;
484 list_add_tail(&async_extent->list, &cow->extents);
489 * Check if the inode has flags compatible with compression
491 static inline bool inode_can_compress(struct btrfs_inode *inode)
493 if (inode->flags & BTRFS_INODE_NODATACOW ||
494 inode->flags & BTRFS_INODE_NODATASUM)
500 * Check if the inode needs to be submitted to compression, based on mount
501 * options, defragmentation, properties or heuristics.
503 static inline int inode_need_compress(struct btrfs_inode *inode, u64 start,
506 struct btrfs_fs_info *fs_info = inode->root->fs_info;
508 if (!inode_can_compress(inode)) {
509 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
510 KERN_ERR "BTRFS: unexpected compression for ino %llu\n",
515 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
518 if (inode->defrag_compress)
520 /* bad compression ratios */
521 if (inode->flags & BTRFS_INODE_NOCOMPRESS)
523 if (btrfs_test_opt(fs_info, COMPRESS) ||
524 inode->flags & BTRFS_INODE_COMPRESS ||
525 inode->prop_compress)
526 return btrfs_compress_heuristic(&inode->vfs_inode, start, end);
530 static inline void inode_should_defrag(struct btrfs_inode *inode,
531 u64 start, u64 end, u64 num_bytes, u64 small_write)
533 /* If this is a small write inside eof, kick off a defrag */
534 if (num_bytes < small_write &&
535 (start > 0 || end + 1 < inode->disk_i_size))
536 btrfs_add_inode_defrag(NULL, inode);
540 * we create compressed extents in two phases. The first
541 * phase compresses a range of pages that have already been
542 * locked (both pages and state bits are locked).
544 * This is done inside an ordered work queue, and the compression
545 * is spread across many cpus. The actual IO submission is step
546 * two, and the ordered work queue takes care of making sure that
547 * happens in the same order things were put onto the queue by
548 * writepages and friends.
550 * If this code finds it can't get good compression, it puts an
551 * entry onto the work queue to write the uncompressed bytes. This
552 * makes sure that both compressed inodes and uncompressed inodes
553 * are written in the same order that the flusher thread sent them
556 static noinline int compress_file_range(struct async_chunk *async_chunk)
558 struct inode *inode = async_chunk->inode;
559 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
560 u64 blocksize = fs_info->sectorsize;
561 u64 start = async_chunk->start;
562 u64 end = async_chunk->end;
566 struct page **pages = NULL;
567 unsigned long nr_pages;
568 unsigned long total_compressed = 0;
569 unsigned long total_in = 0;
572 int compress_type = fs_info->compress_type;
573 int compressed_extents = 0;
576 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
580 * We need to save i_size before now because it could change in between
581 * us evaluating the size and assigning it. This is because we lock and
582 * unlock the page in truncate and fallocate, and then modify the i_size
585 * The barriers are to emulate READ_ONCE, remove that once i_size_read
589 i_size = i_size_read(inode);
591 actual_end = min_t(u64, i_size, end + 1);
594 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
595 BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
596 nr_pages = min_t(unsigned long, nr_pages,
597 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
600 * we don't want to send crud past the end of i_size through
601 * compression, that's just a waste of CPU time. So, if the
602 * end of the file is before the start of our current
603 * requested range of bytes, we bail out to the uncompressed
604 * cleanup code that can deal with all of this.
606 * It isn't really the fastest way to fix things, but this is a
607 * very uncommon corner.
609 if (actual_end <= start)
610 goto cleanup_and_bail_uncompressed;
612 total_compressed = actual_end - start;
615 * skip compression for a small file range(<=blocksize) that
616 * isn't an inline extent, since it doesn't save disk space at all.
618 if (total_compressed <= blocksize &&
619 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
620 goto cleanup_and_bail_uncompressed;
622 total_compressed = min_t(unsigned long, total_compressed,
623 BTRFS_MAX_UNCOMPRESSED);
628 * we do compression for mount -o compress and when the
629 * inode has not been flagged as nocompress. This flag can
630 * change at any time if we discover bad compression ratios.
632 if (nr_pages > 1 && inode_need_compress(BTRFS_I(inode), start, end)) {
634 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
636 /* just bail out to the uncompressed code */
641 if (BTRFS_I(inode)->defrag_compress)
642 compress_type = BTRFS_I(inode)->defrag_compress;
643 else if (BTRFS_I(inode)->prop_compress)
644 compress_type = BTRFS_I(inode)->prop_compress;
647 * we need to call clear_page_dirty_for_io on each
648 * page in the range. Otherwise applications with the file
649 * mmap'd can wander in and change the page contents while
650 * we are compressing them.
652 * If the compression fails for any reason, we set the pages
653 * dirty again later on.
655 * Note that the remaining part is redirtied, the start pointer
656 * has moved, the end is the original one.
659 extent_range_clear_dirty_for_io(inode, start, end);
663 /* Compression level is applied here and only here */
664 ret = btrfs_compress_pages(
665 compress_type | (fs_info->compress_level << 4),
666 inode->i_mapping, start,
673 unsigned long offset = offset_in_page(total_compressed);
674 struct page *page = pages[nr_pages - 1];
676 /* zero the tail end of the last page, we might be
677 * sending it down to disk
680 memzero_page(page, offset, PAGE_SIZE - offset);
686 /* lets try to make an inline extent */
687 if (ret || total_in < actual_end) {
688 /* we didn't compress the entire range, try
689 * to make an uncompressed inline extent.
691 ret = cow_file_range_inline(BTRFS_I(inode), start, end,
692 0, BTRFS_COMPRESS_NONE,
695 /* try making a compressed inline extent */
696 ret = cow_file_range_inline(BTRFS_I(inode), start, end,
698 compress_type, pages);
701 unsigned long clear_flags = EXTENT_DELALLOC |
702 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
703 EXTENT_DO_ACCOUNTING;
704 unsigned long page_error_op;
706 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
709 * inline extent creation worked or returned error,
710 * we don't need to create any more async work items.
711 * Unlock and free up our temp pages.
713 * We use DO_ACCOUNTING here because we need the
714 * delalloc_release_metadata to be done _after_ we drop
715 * our outstanding extent for clearing delalloc for this
718 extent_clear_unlock_delalloc(BTRFS_I(inode), start, end,
722 PAGE_START_WRITEBACK |
727 * Ensure we only free the compressed pages if we have
728 * them allocated, as we can still reach here with
729 * inode_need_compress() == false.
732 for (i = 0; i < nr_pages; i++) {
733 WARN_ON(pages[i]->mapping);
744 * we aren't doing an inline extent round the compressed size
745 * up to a block size boundary so the allocator does sane
748 total_compressed = ALIGN(total_compressed, blocksize);
751 * one last check to make sure the compression is really a
752 * win, compare the page count read with the blocks on disk,
753 * compression must free at least one sector size
755 total_in = ALIGN(total_in, PAGE_SIZE);
756 if (total_compressed + blocksize <= total_in) {
757 compressed_extents++;
760 * The async work queues will take care of doing actual
761 * allocation on disk for these compressed pages, and
762 * will submit them to the elevator.
764 add_async_extent(async_chunk, start, total_in,
765 total_compressed, pages, nr_pages,
768 if (start + total_in < end) {
774 return compressed_extents;
779 * the compression code ran but failed to make things smaller,
780 * free any pages it allocated and our page pointer array
782 for (i = 0; i < nr_pages; i++) {
783 WARN_ON(pages[i]->mapping);
788 total_compressed = 0;
791 /* flag the file so we don't compress in the future */
792 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
793 !(BTRFS_I(inode)->prop_compress)) {
794 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
797 cleanup_and_bail_uncompressed:
799 * No compression, but we still need to write the pages in the file
800 * we've been given so far. redirty the locked page if it corresponds
801 * to our extent and set things up for the async work queue to run
802 * cow_file_range to do the normal delalloc dance.
804 if (async_chunk->locked_page &&
805 (page_offset(async_chunk->locked_page) >= start &&
806 page_offset(async_chunk->locked_page)) <= end) {
807 __set_page_dirty_nobuffers(async_chunk->locked_page);
808 /* unlocked later on in the async handlers */
812 extent_range_redirty_for_io(inode, start, end);
813 add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
814 BTRFS_COMPRESS_NONE);
815 compressed_extents++;
817 return compressed_extents;
820 static void free_async_extent_pages(struct async_extent *async_extent)
824 if (!async_extent->pages)
827 for (i = 0; i < async_extent->nr_pages; i++) {
828 WARN_ON(async_extent->pages[i]->mapping);
829 put_page(async_extent->pages[i]);
831 kfree(async_extent->pages);
832 async_extent->nr_pages = 0;
833 async_extent->pages = NULL;
837 * phase two of compressed writeback. This is the ordered portion
838 * of the code, which only gets called in the order the work was
839 * queued. We walk all the async extents created by compress_file_range
840 * and send them down to the disk.
842 static noinline void submit_compressed_extents(struct async_chunk *async_chunk)
844 struct btrfs_inode *inode = BTRFS_I(async_chunk->inode);
845 struct btrfs_fs_info *fs_info = inode->root->fs_info;
846 struct async_extent *async_extent;
848 struct btrfs_key ins;
849 struct extent_map *em;
850 struct btrfs_root *root = inode->root;
851 struct extent_io_tree *io_tree = &inode->io_tree;
855 while (!list_empty(&async_chunk->extents)) {
856 async_extent = list_entry(async_chunk->extents.next,
857 struct async_extent, list);
858 list_del(&async_extent->list);
861 lock_extent(io_tree, async_extent->start,
862 async_extent->start + async_extent->ram_size - 1);
863 /* did the compression code fall back to uncompressed IO? */
864 if (!async_extent->pages) {
865 int page_started = 0;
866 unsigned long nr_written = 0;
868 /* allocate blocks */
869 ret = cow_file_range(inode, async_chunk->locked_page,
871 async_extent->start +
872 async_extent->ram_size - 1,
873 &page_started, &nr_written, 0);
878 * if page_started, cow_file_range inserted an
879 * inline extent and took care of all the unlocking
880 * and IO for us. Otherwise, we need to submit
881 * all those pages down to the drive.
883 if (!page_started && !ret)
884 extent_write_locked_range(&inode->vfs_inode,
886 async_extent->start +
887 async_extent->ram_size - 1,
889 else if (ret && async_chunk->locked_page)
890 unlock_page(async_chunk->locked_page);
896 ret = btrfs_reserve_extent(root, async_extent->ram_size,
897 async_extent->compressed_size,
898 async_extent->compressed_size,
899 0, alloc_hint, &ins, 1, 1);
901 free_async_extent_pages(async_extent);
903 if (ret == -ENOSPC) {
904 unlock_extent(io_tree, async_extent->start,
905 async_extent->start +
906 async_extent->ram_size - 1);
909 * we need to redirty the pages if we decide to
910 * fallback to uncompressed IO, otherwise we
911 * will not submit these pages down to lower
914 extent_range_redirty_for_io(&inode->vfs_inode,
916 async_extent->start +
917 async_extent->ram_size - 1);
924 * here we're doing allocation and writeback of the
927 em = create_io_em(inode, async_extent->start,
928 async_extent->ram_size, /* len */
929 async_extent->start, /* orig_start */
930 ins.objectid, /* block_start */
931 ins.offset, /* block_len */
932 ins.offset, /* orig_block_len */
933 async_extent->ram_size, /* ram_bytes */
934 async_extent->compress_type,
935 BTRFS_ORDERED_COMPRESSED);
937 /* ret value is not necessary due to void function */
938 goto out_free_reserve;
941 ret = btrfs_add_ordered_extent_compress(inode,
944 async_extent->ram_size,
946 async_extent->compress_type);
948 btrfs_drop_extent_cache(inode, async_extent->start,
949 async_extent->start +
950 async_extent->ram_size - 1, 0);
951 goto out_free_reserve;
953 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
956 * clear dirty, set writeback and unlock the pages.
958 extent_clear_unlock_delalloc(inode, async_extent->start,
959 async_extent->start +
960 async_extent->ram_size - 1,
961 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
962 PAGE_UNLOCK | PAGE_START_WRITEBACK);
963 if (btrfs_submit_compressed_write(inode, async_extent->start,
964 async_extent->ram_size,
966 ins.offset, async_extent->pages,
967 async_extent->nr_pages,
968 async_chunk->write_flags,
969 async_chunk->blkcg_css)) {
970 struct page *p = async_extent->pages[0];
971 const u64 start = async_extent->start;
972 const u64 end = start + async_extent->ram_size - 1;
974 p->mapping = inode->vfs_inode.i_mapping;
975 btrfs_writepage_endio_finish_ordered(inode, p, start,
979 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
982 free_async_extent_pages(async_extent);
984 alloc_hint = ins.objectid + ins.offset;
990 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
991 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
993 extent_clear_unlock_delalloc(inode, async_extent->start,
994 async_extent->start +
995 async_extent->ram_size - 1,
996 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
997 EXTENT_DELALLOC_NEW |
998 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
999 PAGE_UNLOCK | PAGE_START_WRITEBACK |
1000 PAGE_END_WRITEBACK | PAGE_SET_ERROR);
1001 free_async_extent_pages(async_extent);
1002 kfree(async_extent);
1006 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
1009 struct extent_map_tree *em_tree = &inode->extent_tree;
1010 struct extent_map *em;
1013 read_lock(&em_tree->lock);
1014 em = search_extent_mapping(em_tree, start, num_bytes);
1017 * if block start isn't an actual block number then find the
1018 * first block in this inode and use that as a hint. If that
1019 * block is also bogus then just don't worry about it.
1021 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1022 free_extent_map(em);
1023 em = search_extent_mapping(em_tree, 0, 0);
1024 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
1025 alloc_hint = em->block_start;
1027 free_extent_map(em);
1029 alloc_hint = em->block_start;
1030 free_extent_map(em);
1033 read_unlock(&em_tree->lock);
1039 * when extent_io.c finds a delayed allocation range in the file,
1040 * the call backs end up in this code. The basic idea is to
1041 * allocate extents on disk for the range, and create ordered data structs
1042 * in ram to track those extents.
1044 * locked_page is the page that writepage had locked already. We use
1045 * it to make sure we don't do extra locks or unlocks.
1047 * *page_started is set to one if we unlock locked_page and do everything
1048 * required to start IO on it. It may be clean and already done with
1049 * IO when we return.
1051 static noinline int cow_file_range(struct btrfs_inode *inode,
1052 struct page *locked_page,
1053 u64 start, u64 end, int *page_started,
1054 unsigned long *nr_written, int unlock)
1056 struct btrfs_root *root = inode->root;
1057 struct btrfs_fs_info *fs_info = root->fs_info;
1060 unsigned long ram_size;
1061 u64 cur_alloc_size = 0;
1063 u64 blocksize = fs_info->sectorsize;
1064 struct btrfs_key ins;
1065 struct extent_map *em;
1066 unsigned clear_bits;
1067 unsigned long page_ops;
1068 bool extent_reserved = false;
1071 if (btrfs_is_free_space_inode(inode)) {
1077 num_bytes = ALIGN(end - start + 1, blocksize);
1078 num_bytes = max(blocksize, num_bytes);
1079 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1081 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1084 /* lets try to make an inline extent */
1085 ret = cow_file_range_inline(inode, start, end, 0,
1086 BTRFS_COMPRESS_NONE, NULL);
1089 * We use DO_ACCOUNTING here because we need the
1090 * delalloc_release_metadata to be run _after_ we drop
1091 * our outstanding extent for clearing delalloc for this
1094 extent_clear_unlock_delalloc(inode, start, end,
1096 EXTENT_LOCKED | EXTENT_DELALLOC |
1097 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1098 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1099 PAGE_START_WRITEBACK | PAGE_END_WRITEBACK);
1100 *nr_written = *nr_written +
1101 (end - start + PAGE_SIZE) / PAGE_SIZE;
1104 * locked_page is locked by the caller of
1105 * writepage_delalloc(), not locked by
1106 * __process_pages_contig().
1108 * We can't let __process_pages_contig() to unlock it,
1109 * as it doesn't have any subpage::writers recorded.
1111 * Here we manually unlock the page, since the caller
1112 * can't use page_started to determine if it's an
1113 * inline extent or a compressed extent.
1115 unlock_page(locked_page);
1117 } else if (ret < 0) {
1122 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1123 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
1126 * Relocation relies on the relocated extents to have exactly the same
1127 * size as the original extents. Normally writeback for relocation data
1128 * extents follows a NOCOW path because relocation preallocates the
1129 * extents. However, due to an operation such as scrub turning a block
1130 * group to RO mode, it may fallback to COW mode, so we must make sure
1131 * an extent allocated during COW has exactly the requested size and can
1132 * not be split into smaller extents, otherwise relocation breaks and
1133 * fails during the stage where it updates the bytenr of file extent
1136 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1137 min_alloc_size = num_bytes;
1139 min_alloc_size = fs_info->sectorsize;
1141 while (num_bytes > 0) {
1142 cur_alloc_size = num_bytes;
1143 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1144 min_alloc_size, 0, alloc_hint,
1148 cur_alloc_size = ins.offset;
1149 extent_reserved = true;
1151 ram_size = ins.offset;
1152 em = create_io_em(inode, start, ins.offset, /* len */
1153 start, /* orig_start */
1154 ins.objectid, /* block_start */
1155 ins.offset, /* block_len */
1156 ins.offset, /* orig_block_len */
1157 ram_size, /* ram_bytes */
1158 BTRFS_COMPRESS_NONE, /* compress_type */
1159 BTRFS_ORDERED_REGULAR /* type */);
1164 free_extent_map(em);
1166 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1167 ram_size, cur_alloc_size,
1168 BTRFS_ORDERED_REGULAR);
1170 goto out_drop_extent_cache;
1172 if (root->root_key.objectid ==
1173 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1174 ret = btrfs_reloc_clone_csums(inode, start,
1177 * Only drop cache here, and process as normal.
1179 * We must not allow extent_clear_unlock_delalloc()
1180 * at out_unlock label to free meta of this ordered
1181 * extent, as its meta should be freed by
1182 * btrfs_finish_ordered_io().
1184 * So we must continue until @start is increased to
1185 * skip current ordered extent.
1188 btrfs_drop_extent_cache(inode, start,
1189 start + ram_size - 1, 0);
1192 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1195 * We're not doing compressed IO, don't unlock the first page
1196 * (which the caller expects to stay locked), don't clear any
1197 * dirty bits and don't set any writeback bits
1199 * Do set the Ordered (Private2) bit so we know this page was
1200 * properly setup for writepage.
1202 page_ops = unlock ? PAGE_UNLOCK : 0;
1203 page_ops |= PAGE_SET_ORDERED;
1205 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1207 EXTENT_LOCKED | EXTENT_DELALLOC,
1209 if (num_bytes < cur_alloc_size)
1212 num_bytes -= cur_alloc_size;
1213 alloc_hint = ins.objectid + ins.offset;
1214 start += cur_alloc_size;
1215 extent_reserved = false;
1218 * btrfs_reloc_clone_csums() error, since start is increased
1219 * extent_clear_unlock_delalloc() at out_unlock label won't
1220 * free metadata of current ordered extent, we're OK to exit.
1228 out_drop_extent_cache:
1229 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1231 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1232 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1234 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1235 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1236 page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK;
1238 * If we reserved an extent for our delalloc range (or a subrange) and
1239 * failed to create the respective ordered extent, then it means that
1240 * when we reserved the extent we decremented the extent's size from
1241 * the data space_info's bytes_may_use counter and incremented the
1242 * space_info's bytes_reserved counter by the same amount. We must make
1243 * sure extent_clear_unlock_delalloc() does not try to decrement again
1244 * the data space_info's bytes_may_use counter, therefore we do not pass
1245 * it the flag EXTENT_CLEAR_DATA_RESV.
1247 if (extent_reserved) {
1248 extent_clear_unlock_delalloc(inode, start,
1249 start + cur_alloc_size - 1,
1253 start += cur_alloc_size;
1257 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1258 clear_bits | EXTENT_CLEAR_DATA_RESV,
1264 * work queue call back to started compression on a file and pages
1266 static noinline void async_cow_start(struct btrfs_work *work)
1268 struct async_chunk *async_chunk;
1269 int compressed_extents;
1271 async_chunk = container_of(work, struct async_chunk, work);
1273 compressed_extents = compress_file_range(async_chunk);
1274 if (compressed_extents == 0) {
1275 btrfs_add_delayed_iput(async_chunk->inode);
1276 async_chunk->inode = NULL;
1281 * work queue call back to submit previously compressed pages
1283 static noinline void async_cow_submit(struct btrfs_work *work)
1285 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1287 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1288 unsigned long nr_pages;
1290 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1293 /* atomic_sub_return implies a barrier */
1294 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1296 cond_wake_up_nomb(&fs_info->async_submit_wait);
1299 * ->inode could be NULL if async_chunk_start has failed to compress,
1300 * in which case we don't have anything to submit, yet we need to
1301 * always adjust ->async_delalloc_pages as its paired with the init
1302 * happening in cow_file_range_async
1304 if (async_chunk->inode)
1305 submit_compressed_extents(async_chunk);
1308 static noinline void async_cow_free(struct btrfs_work *work)
1310 struct async_chunk *async_chunk;
1312 async_chunk = container_of(work, struct async_chunk, work);
1313 if (async_chunk->inode)
1314 btrfs_add_delayed_iput(async_chunk->inode);
1315 if (async_chunk->blkcg_css)
1316 css_put(async_chunk->blkcg_css);
1318 * Since the pointer to 'pending' is at the beginning of the array of
1319 * async_chunk's, freeing it ensures the whole array has been freed.
1321 if (atomic_dec_and_test(async_chunk->pending))
1322 kvfree(async_chunk->pending);
1325 static int cow_file_range_async(struct btrfs_inode *inode,
1326 struct writeback_control *wbc,
1327 struct page *locked_page,
1328 u64 start, u64 end, int *page_started,
1329 unsigned long *nr_written)
1331 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1332 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1333 struct async_cow *ctx;
1334 struct async_chunk *async_chunk;
1335 unsigned long nr_pages;
1337 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1339 bool should_compress;
1341 const unsigned int write_flags = wbc_to_write_flags(wbc);
1343 unlock_extent(&inode->io_tree, start, end);
1345 if (inode->flags & BTRFS_INODE_NOCOMPRESS &&
1346 !btrfs_test_opt(fs_info, FORCE_COMPRESS)) {
1348 should_compress = false;
1350 should_compress = true;
1353 nofs_flag = memalloc_nofs_save();
1354 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1355 memalloc_nofs_restore(nofs_flag);
1358 unsigned clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC |
1359 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1360 EXTENT_DO_ACCOUNTING;
1361 unsigned long page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK |
1362 PAGE_END_WRITEBACK | PAGE_SET_ERROR;
1364 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1365 clear_bits, page_ops);
1369 async_chunk = ctx->chunks;
1370 atomic_set(&ctx->num_chunks, num_chunks);
1372 for (i = 0; i < num_chunks; i++) {
1373 if (should_compress)
1374 cur_end = min(end, start + SZ_512K - 1);
1379 * igrab is called higher up in the call chain, take only the
1380 * lightweight reference for the callback lifetime
1382 ihold(&inode->vfs_inode);
1383 async_chunk[i].pending = &ctx->num_chunks;
1384 async_chunk[i].inode = &inode->vfs_inode;
1385 async_chunk[i].start = start;
1386 async_chunk[i].end = cur_end;
1387 async_chunk[i].write_flags = write_flags;
1388 INIT_LIST_HEAD(&async_chunk[i].extents);
1391 * The locked_page comes all the way from writepage and its
1392 * the original page we were actually given. As we spread
1393 * this large delalloc region across multiple async_chunk
1394 * structs, only the first struct needs a pointer to locked_page
1396 * This way we don't need racey decisions about who is supposed
1401 * Depending on the compressibility, the pages might or
1402 * might not go through async. We want all of them to
1403 * be accounted against wbc once. Let's do it here
1404 * before the paths diverge. wbc accounting is used
1405 * only for foreign writeback detection and doesn't
1406 * need full accuracy. Just account the whole thing
1407 * against the first page.
1409 wbc_account_cgroup_owner(wbc, locked_page,
1411 async_chunk[i].locked_page = locked_page;
1414 async_chunk[i].locked_page = NULL;
1417 if (blkcg_css != blkcg_root_css) {
1419 async_chunk[i].blkcg_css = blkcg_css;
1421 async_chunk[i].blkcg_css = NULL;
1424 btrfs_init_work(&async_chunk[i].work, async_cow_start,
1425 async_cow_submit, async_cow_free);
1427 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1428 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1430 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1432 *nr_written += nr_pages;
1433 start = cur_end + 1;
1439 static noinline int run_delalloc_zoned(struct btrfs_inode *inode,
1440 struct page *locked_page, u64 start,
1441 u64 end, int *page_started,
1442 unsigned long *nr_written)
1446 ret = cow_file_range(inode, locked_page, start, end, page_started,
1454 __set_page_dirty_nobuffers(locked_page);
1455 account_page_redirty(locked_page);
1456 extent_write_locked_range(&inode->vfs_inode, start, end, WB_SYNC_ALL);
1462 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1463 u64 bytenr, u64 num_bytes)
1466 struct btrfs_ordered_sum *sums;
1469 ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1470 bytenr + num_bytes - 1, &list, 0);
1471 if (ret == 0 && list_empty(&list))
1474 while (!list_empty(&list)) {
1475 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1476 list_del(&sums->list);
1484 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1485 const u64 start, const u64 end,
1486 int *page_started, unsigned long *nr_written)
1488 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1489 const bool is_reloc_ino = (inode->root->root_key.objectid ==
1490 BTRFS_DATA_RELOC_TREE_OBJECTID);
1491 const u64 range_bytes = end + 1 - start;
1492 struct extent_io_tree *io_tree = &inode->io_tree;
1493 u64 range_start = start;
1497 * If EXTENT_NORESERVE is set it means that when the buffered write was
1498 * made we had not enough available data space and therefore we did not
1499 * reserve data space for it, since we though we could do NOCOW for the
1500 * respective file range (either there is prealloc extent or the inode
1501 * has the NOCOW bit set).
1503 * However when we need to fallback to COW mode (because for example the
1504 * block group for the corresponding extent was turned to RO mode by a
1505 * scrub or relocation) we need to do the following:
1507 * 1) We increment the bytes_may_use counter of the data space info.
1508 * If COW succeeds, it allocates a new data extent and after doing
1509 * that it decrements the space info's bytes_may_use counter and
1510 * increments its bytes_reserved counter by the same amount (we do
1511 * this at btrfs_add_reserved_bytes()). So we need to increment the
1512 * bytes_may_use counter to compensate (when space is reserved at
1513 * buffered write time, the bytes_may_use counter is incremented);
1515 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1516 * that if the COW path fails for any reason, it decrements (through
1517 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1518 * data space info, which we incremented in the step above.
1520 * If we need to fallback to cow and the inode corresponds to a free
1521 * space cache inode or an inode of the data relocation tree, we must
1522 * also increment bytes_may_use of the data space_info for the same
1523 * reason. Space caches and relocated data extents always get a prealloc
1524 * extent for them, however scrub or balance may have set the block
1525 * group that contains that extent to RO mode and therefore force COW
1526 * when starting writeback.
1528 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1529 EXTENT_NORESERVE, 0);
1530 if (count > 0 || is_space_ino || is_reloc_ino) {
1532 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1533 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1535 if (is_space_ino || is_reloc_ino)
1536 bytes = range_bytes;
1538 spin_lock(&sinfo->lock);
1539 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1540 spin_unlock(&sinfo->lock);
1543 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1547 return cow_file_range(inode, locked_page, start, end, page_started,
1552 * when nowcow writeback call back. This checks for snapshots or COW copies
1553 * of the extents that exist in the file, and COWs the file as required.
1555 * If no cow copies or snapshots exist, we write directly to the existing
1558 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
1559 struct page *locked_page,
1560 const u64 start, const u64 end,
1562 unsigned long *nr_written)
1564 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1565 struct btrfs_root *root = inode->root;
1566 struct btrfs_path *path;
1567 u64 cow_start = (u64)-1;
1568 u64 cur_offset = start;
1570 bool check_prev = true;
1571 const bool freespace_inode = btrfs_is_free_space_inode(inode);
1572 u64 ino = btrfs_ino(inode);
1574 u64 disk_bytenr = 0;
1575 const bool force = inode->flags & BTRFS_INODE_NODATACOW;
1577 path = btrfs_alloc_path();
1579 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1580 EXTENT_LOCKED | EXTENT_DELALLOC |
1581 EXTENT_DO_ACCOUNTING |
1582 EXTENT_DEFRAG, PAGE_UNLOCK |
1583 PAGE_START_WRITEBACK |
1584 PAGE_END_WRITEBACK);
1589 struct btrfs_key found_key;
1590 struct btrfs_file_extent_item *fi;
1591 struct extent_buffer *leaf;
1601 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1607 * If there is no extent for our range when doing the initial
1608 * search, then go back to the previous slot as it will be the
1609 * one containing the search offset
1611 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1612 leaf = path->nodes[0];
1613 btrfs_item_key_to_cpu(leaf, &found_key,
1614 path->slots[0] - 1);
1615 if (found_key.objectid == ino &&
1616 found_key.type == BTRFS_EXTENT_DATA_KEY)
1621 /* Go to next leaf if we have exhausted the current one */
1622 leaf = path->nodes[0];
1623 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1624 ret = btrfs_next_leaf(root, path);
1626 if (cow_start != (u64)-1)
1627 cur_offset = cow_start;
1632 leaf = path->nodes[0];
1635 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1637 /* Didn't find anything for our INO */
1638 if (found_key.objectid > ino)
1641 * Keep searching until we find an EXTENT_ITEM or there are no
1642 * more extents for this inode
1644 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1645 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1650 /* Found key is not EXTENT_DATA_KEY or starts after req range */
1651 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1652 found_key.offset > end)
1656 * If the found extent starts after requested offset, then
1657 * adjust extent_end to be right before this extent begins
1659 if (found_key.offset > cur_offset) {
1660 extent_end = found_key.offset;
1666 * Found extent which begins before our range and potentially
1669 fi = btrfs_item_ptr(leaf, path->slots[0],
1670 struct btrfs_file_extent_item);
1671 extent_type = btrfs_file_extent_type(leaf, fi);
1673 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1674 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1675 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1676 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1677 extent_offset = btrfs_file_extent_offset(leaf, fi);
1678 extent_end = found_key.offset +
1679 btrfs_file_extent_num_bytes(leaf, fi);
1681 btrfs_file_extent_disk_num_bytes(leaf, fi);
1683 * If the extent we got ends before our current offset,
1684 * skip to the next extent.
1686 if (extent_end <= cur_offset) {
1691 if (disk_bytenr == 0)
1693 /* Skip compressed/encrypted/encoded extents */
1694 if (btrfs_file_extent_compression(leaf, fi) ||
1695 btrfs_file_extent_encryption(leaf, fi) ||
1696 btrfs_file_extent_other_encoding(leaf, fi))
1699 * If extent is created before the last volume's snapshot
1700 * this implies the extent is shared, hence we can't do
1701 * nocow. This is the same check as in
1702 * btrfs_cross_ref_exist but without calling
1703 * btrfs_search_slot.
1705 if (!freespace_inode &&
1706 btrfs_file_extent_generation(leaf, fi) <=
1707 btrfs_root_last_snapshot(&root->root_item))
1709 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1713 * The following checks can be expensive, as they need to
1714 * take other locks and do btree or rbtree searches, so
1715 * release the path to avoid blocking other tasks for too
1718 btrfs_release_path(path);
1720 ret = btrfs_cross_ref_exist(root, ino,
1722 extent_offset, disk_bytenr, false);
1725 * ret could be -EIO if the above fails to read
1729 if (cow_start != (u64)-1)
1730 cur_offset = cow_start;
1734 WARN_ON_ONCE(freespace_inode);
1737 disk_bytenr += extent_offset;
1738 disk_bytenr += cur_offset - found_key.offset;
1739 num_bytes = min(end + 1, extent_end) - cur_offset;
1741 * If there are pending snapshots for this root, we
1742 * fall into common COW way
1744 if (!freespace_inode && atomic_read(&root->snapshot_force_cow))
1747 * force cow if csum exists in the range.
1748 * this ensure that csum for a given extent are
1749 * either valid or do not exist.
1751 ret = csum_exist_in_range(fs_info, disk_bytenr,
1755 * ret could be -EIO if the above fails to read
1759 if (cow_start != (u64)-1)
1760 cur_offset = cow_start;
1763 WARN_ON_ONCE(freespace_inode);
1766 /* If the extent's block group is RO, we must COW */
1767 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr))
1770 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1771 extent_end = found_key.offset + ram_bytes;
1772 extent_end = ALIGN(extent_end, fs_info->sectorsize);
1773 /* Skip extents outside of our requested range */
1774 if (extent_end <= start) {
1779 /* If this triggers then we have a memory corruption */
1784 * If nocow is false then record the beginning of the range
1785 * that needs to be COWed
1788 if (cow_start == (u64)-1)
1789 cow_start = cur_offset;
1790 cur_offset = extent_end;
1791 if (cur_offset > end)
1793 if (!path->nodes[0])
1800 * COW range from cow_start to found_key.offset - 1. As the key
1801 * will contain the beginning of the first extent that can be
1802 * NOCOW, following one which needs to be COW'ed
1804 if (cow_start != (u64)-1) {
1805 ret = fallback_to_cow(inode, locked_page,
1806 cow_start, found_key.offset - 1,
1807 page_started, nr_written);
1810 cow_start = (u64)-1;
1813 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1814 u64 orig_start = found_key.offset - extent_offset;
1815 struct extent_map *em;
1817 em = create_io_em(inode, cur_offset, num_bytes,
1819 disk_bytenr, /* block_start */
1820 num_bytes, /* block_len */
1821 disk_num_bytes, /* orig_block_len */
1822 ram_bytes, BTRFS_COMPRESS_NONE,
1823 BTRFS_ORDERED_PREALLOC);
1828 free_extent_map(em);
1829 ret = btrfs_add_ordered_extent(inode, cur_offset,
1830 disk_bytenr, num_bytes,
1832 BTRFS_ORDERED_PREALLOC);
1834 btrfs_drop_extent_cache(inode, cur_offset,
1835 cur_offset + num_bytes - 1,
1840 ret = btrfs_add_ordered_extent(inode, cur_offset,
1841 disk_bytenr, num_bytes,
1843 BTRFS_ORDERED_NOCOW);
1849 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1852 if (root->root_key.objectid ==
1853 BTRFS_DATA_RELOC_TREE_OBJECTID)
1855 * Error handled later, as we must prevent
1856 * extent_clear_unlock_delalloc() in error handler
1857 * from freeing metadata of created ordered extent.
1859 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1862 extent_clear_unlock_delalloc(inode, cur_offset,
1863 cur_offset + num_bytes - 1,
1864 locked_page, EXTENT_LOCKED |
1866 EXTENT_CLEAR_DATA_RESV,
1867 PAGE_UNLOCK | PAGE_SET_ORDERED);
1869 cur_offset = extent_end;
1872 * btrfs_reloc_clone_csums() error, now we're OK to call error
1873 * handler, as metadata for created ordered extent will only
1874 * be freed by btrfs_finish_ordered_io().
1878 if (cur_offset > end)
1881 btrfs_release_path(path);
1883 if (cur_offset <= end && cow_start == (u64)-1)
1884 cow_start = cur_offset;
1886 if (cow_start != (u64)-1) {
1888 ret = fallback_to_cow(inode, locked_page, cow_start, end,
1889 page_started, nr_written);
1896 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1898 if (ret && cur_offset < end)
1899 extent_clear_unlock_delalloc(inode, cur_offset, end,
1900 locked_page, EXTENT_LOCKED |
1901 EXTENT_DELALLOC | EXTENT_DEFRAG |
1902 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1903 PAGE_START_WRITEBACK |
1904 PAGE_END_WRITEBACK);
1905 btrfs_free_path(path);
1909 static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end)
1911 if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) {
1912 if (inode->defrag_bytes &&
1913 test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG,
1922 * Function to process delayed allocation (create CoW) for ranges which are
1923 * being touched for the first time.
1925 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
1926 u64 start, u64 end, int *page_started, unsigned long *nr_written,
1927 struct writeback_control *wbc)
1930 const bool zoned = btrfs_is_zoned(inode->root->fs_info);
1932 if (should_nocow(inode, start, end)) {
1934 ret = run_delalloc_nocow(inode, locked_page, start, end,
1935 page_started, nr_written);
1936 } else if (!inode_can_compress(inode) ||
1937 !inode_need_compress(inode, start, end)) {
1939 ret = run_delalloc_zoned(inode, locked_page, start, end,
1940 page_started, nr_written);
1942 ret = cow_file_range(inode, locked_page, start, end,
1943 page_started, nr_written, 1);
1945 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
1946 ret = cow_file_range_async(inode, wbc, locked_page, start, end,
1947 page_started, nr_written);
1950 btrfs_cleanup_ordered_extents(inode, locked_page, start,
1955 void btrfs_split_delalloc_extent(struct inode *inode,
1956 struct extent_state *orig, u64 split)
1960 /* not delalloc, ignore it */
1961 if (!(orig->state & EXTENT_DELALLOC))
1964 size = orig->end - orig->start + 1;
1965 if (size > BTRFS_MAX_EXTENT_SIZE) {
1970 * See the explanation in btrfs_merge_delalloc_extent, the same
1971 * applies here, just in reverse.
1973 new_size = orig->end - split + 1;
1974 num_extents = count_max_extents(new_size);
1975 new_size = split - orig->start;
1976 num_extents += count_max_extents(new_size);
1977 if (count_max_extents(size) >= num_extents)
1981 spin_lock(&BTRFS_I(inode)->lock);
1982 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1983 spin_unlock(&BTRFS_I(inode)->lock);
1987 * Handle merged delayed allocation extents so we can keep track of new extents
1988 * that are just merged onto old extents, such as when we are doing sequential
1989 * writes, so we can properly account for the metadata space we'll need.
1991 void btrfs_merge_delalloc_extent(struct inode *inode, struct extent_state *new,
1992 struct extent_state *other)
1994 u64 new_size, old_size;
1997 /* not delalloc, ignore it */
1998 if (!(other->state & EXTENT_DELALLOC))
2001 if (new->start > other->start)
2002 new_size = new->end - other->start + 1;
2004 new_size = other->end - new->start + 1;
2006 /* we're not bigger than the max, unreserve the space and go */
2007 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
2008 spin_lock(&BTRFS_I(inode)->lock);
2009 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
2010 spin_unlock(&BTRFS_I(inode)->lock);
2015 * We have to add up either side to figure out how many extents were
2016 * accounted for before we merged into one big extent. If the number of
2017 * extents we accounted for is <= the amount we need for the new range
2018 * then we can return, otherwise drop. Think of it like this
2022 * So we've grown the extent by a MAX_SIZE extent, this would mean we
2023 * need 2 outstanding extents, on one side we have 1 and the other side
2024 * we have 1 so they are == and we can return. But in this case
2026 * [MAX_SIZE+4k][MAX_SIZE+4k]
2028 * Each range on their own accounts for 2 extents, but merged together
2029 * they are only 3 extents worth of accounting, so we need to drop in
2032 old_size = other->end - other->start + 1;
2033 num_extents = count_max_extents(old_size);
2034 old_size = new->end - new->start + 1;
2035 num_extents += count_max_extents(old_size);
2036 if (count_max_extents(new_size) >= num_extents)
2039 spin_lock(&BTRFS_I(inode)->lock);
2040 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
2041 spin_unlock(&BTRFS_I(inode)->lock);
2044 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
2045 struct inode *inode)
2047 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2049 spin_lock(&root->delalloc_lock);
2050 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
2051 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
2052 &root->delalloc_inodes);
2053 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2054 &BTRFS_I(inode)->runtime_flags);
2055 root->nr_delalloc_inodes++;
2056 if (root->nr_delalloc_inodes == 1) {
2057 spin_lock(&fs_info->delalloc_root_lock);
2058 BUG_ON(!list_empty(&root->delalloc_root));
2059 list_add_tail(&root->delalloc_root,
2060 &fs_info->delalloc_roots);
2061 spin_unlock(&fs_info->delalloc_root_lock);
2064 spin_unlock(&root->delalloc_lock);
2068 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
2069 struct btrfs_inode *inode)
2071 struct btrfs_fs_info *fs_info = root->fs_info;
2073 if (!list_empty(&inode->delalloc_inodes)) {
2074 list_del_init(&inode->delalloc_inodes);
2075 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2076 &inode->runtime_flags);
2077 root->nr_delalloc_inodes--;
2078 if (!root->nr_delalloc_inodes) {
2079 ASSERT(list_empty(&root->delalloc_inodes));
2080 spin_lock(&fs_info->delalloc_root_lock);
2081 BUG_ON(list_empty(&root->delalloc_root));
2082 list_del_init(&root->delalloc_root);
2083 spin_unlock(&fs_info->delalloc_root_lock);
2088 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
2089 struct btrfs_inode *inode)
2091 spin_lock(&root->delalloc_lock);
2092 __btrfs_del_delalloc_inode(root, inode);
2093 spin_unlock(&root->delalloc_lock);
2097 * Properly track delayed allocation bytes in the inode and to maintain the
2098 * list of inodes that have pending delalloc work to be done.
2100 void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state,
2103 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2105 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
2108 * set_bit and clear bit hooks normally require _irqsave/restore
2109 * but in this case, we are only testing for the DELALLOC
2110 * bit, which is only set or cleared with irqs on
2112 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2113 struct btrfs_root *root = BTRFS_I(inode)->root;
2114 u64 len = state->end + 1 - state->start;
2115 u32 num_extents = count_max_extents(len);
2116 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
2118 spin_lock(&BTRFS_I(inode)->lock);
2119 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
2120 spin_unlock(&BTRFS_I(inode)->lock);
2122 /* For sanity tests */
2123 if (btrfs_is_testing(fs_info))
2126 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2127 fs_info->delalloc_batch);
2128 spin_lock(&BTRFS_I(inode)->lock);
2129 BTRFS_I(inode)->delalloc_bytes += len;
2130 if (*bits & EXTENT_DEFRAG)
2131 BTRFS_I(inode)->defrag_bytes += len;
2132 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2133 &BTRFS_I(inode)->runtime_flags))
2134 btrfs_add_delalloc_inodes(root, inode);
2135 spin_unlock(&BTRFS_I(inode)->lock);
2138 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2139 (*bits & EXTENT_DELALLOC_NEW)) {
2140 spin_lock(&BTRFS_I(inode)->lock);
2141 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
2143 spin_unlock(&BTRFS_I(inode)->lock);
2148 * Once a range is no longer delalloc this function ensures that proper
2149 * accounting happens.
2151 void btrfs_clear_delalloc_extent(struct inode *vfs_inode,
2152 struct extent_state *state, unsigned *bits)
2154 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
2155 struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb);
2156 u64 len = state->end + 1 - state->start;
2157 u32 num_extents = count_max_extents(len);
2159 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
2160 spin_lock(&inode->lock);
2161 inode->defrag_bytes -= len;
2162 spin_unlock(&inode->lock);
2166 * set_bit and clear bit hooks normally require _irqsave/restore
2167 * but in this case, we are only testing for the DELALLOC
2168 * bit, which is only set or cleared with irqs on
2170 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2171 struct btrfs_root *root = inode->root;
2172 bool do_list = !btrfs_is_free_space_inode(inode);
2174 spin_lock(&inode->lock);
2175 btrfs_mod_outstanding_extents(inode, -num_extents);
2176 spin_unlock(&inode->lock);
2179 * We don't reserve metadata space for space cache inodes so we
2180 * don't need to call delalloc_release_metadata if there is an
2183 if (*bits & EXTENT_CLEAR_META_RESV &&
2184 root != fs_info->tree_root)
2185 btrfs_delalloc_release_metadata(inode, len, false);
2187 /* For sanity tests. */
2188 if (btrfs_is_testing(fs_info))
2191 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
2192 do_list && !(state->state & EXTENT_NORESERVE) &&
2193 (*bits & EXTENT_CLEAR_DATA_RESV))
2194 btrfs_free_reserved_data_space_noquota(fs_info, len);
2196 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2197 fs_info->delalloc_batch);
2198 spin_lock(&inode->lock);
2199 inode->delalloc_bytes -= len;
2200 if (do_list && inode->delalloc_bytes == 0 &&
2201 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2202 &inode->runtime_flags))
2203 btrfs_del_delalloc_inode(root, inode);
2204 spin_unlock(&inode->lock);
2207 if ((state->state & EXTENT_DELALLOC_NEW) &&
2208 (*bits & EXTENT_DELALLOC_NEW)) {
2209 spin_lock(&inode->lock);
2210 ASSERT(inode->new_delalloc_bytes >= len);
2211 inode->new_delalloc_bytes -= len;
2212 if (*bits & EXTENT_ADD_INODE_BYTES)
2213 inode_add_bytes(&inode->vfs_inode, len);
2214 spin_unlock(&inode->lock);
2219 * btrfs_bio_fits_in_stripe - Checks whether the size of the given bio will fit
2220 * in a chunk's stripe. This function ensures that bios do not span a
2223 * @page - The page we are about to add to the bio
2224 * @size - size we want to add to the bio
2225 * @bio - bio we want to ensure is smaller than a stripe
2226 * @bio_flags - flags of the bio
2228 * return 1 if page cannot be added to the bio
2229 * return 0 if page can be added to the bio
2230 * return error otherwise
2232 int btrfs_bio_fits_in_stripe(struct page *page, size_t size, struct bio *bio,
2233 unsigned long bio_flags)
2235 struct inode *inode = page->mapping->host;
2236 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2237 u64 logical = bio->bi_iter.bi_sector << 9;
2238 u32 bio_len = bio->bi_iter.bi_size;
2239 struct extent_map *em;
2241 struct btrfs_io_geometry geom;
2243 if (bio_flags & EXTENT_BIO_COMPRESSED)
2246 em = btrfs_get_chunk_map(fs_info, logical, fs_info->sectorsize);
2249 ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(bio), logical, &geom);
2253 if (geom.len < bio_len + size)
2256 free_extent_map(em);
2261 * in order to insert checksums into the metadata in large chunks,
2262 * we wait until bio submission time. All the pages in the bio are
2263 * checksummed and sums are attached onto the ordered extent record.
2265 * At IO completion time the cums attached on the ordered extent record
2266 * are inserted into the btree
2268 static blk_status_t btrfs_submit_bio_start(struct inode *inode, struct bio *bio,
2269 u64 dio_file_offset)
2271 return btrfs_csum_one_bio(BTRFS_I(inode), bio, 0, 0);
2274 static blk_status_t extract_ordered_extent(struct btrfs_inode *inode,
2275 struct bio *bio, loff_t file_offset)
2277 struct btrfs_ordered_extent *ordered;
2278 struct extent_map *em = NULL, *em_new = NULL;
2279 struct extent_map_tree *em_tree = &inode->extent_tree;
2280 u64 start = (u64)bio->bi_iter.bi_sector << SECTOR_SHIFT;
2281 u64 len = bio->bi_iter.bi_size;
2282 u64 end = start + len;
2287 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
2288 if (WARN_ON_ONCE(!ordered))
2289 return BLK_STS_IOERR;
2291 /* No need to split */
2292 if (ordered->disk_num_bytes == len)
2295 /* We cannot split once end_bio'd ordered extent */
2296 if (WARN_ON_ONCE(ordered->bytes_left != ordered->disk_num_bytes)) {
2301 /* We cannot split a compressed ordered extent */
2302 if (WARN_ON_ONCE(ordered->disk_num_bytes != ordered->num_bytes)) {
2307 ordered_end = ordered->disk_bytenr + ordered->disk_num_bytes;
2308 /* bio must be in one ordered extent */
2309 if (WARN_ON_ONCE(start < ordered->disk_bytenr || end > ordered_end)) {
2314 /* Checksum list should be empty */
2315 if (WARN_ON_ONCE(!list_empty(&ordered->list))) {
2320 pre = start - ordered->disk_bytenr;
2321 post = ordered_end - end;
2323 ret = btrfs_split_ordered_extent(ordered, pre, post);
2327 read_lock(&em_tree->lock);
2328 em = lookup_extent_mapping(em_tree, ordered->file_offset, len);
2330 read_unlock(&em_tree->lock);
2334 read_unlock(&em_tree->lock);
2336 ASSERT(!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags));
2338 * We cannot reuse em_new here but have to create a new one, as
2339 * unpin_extent_cache() expects the start of the extent map to be the
2340 * logical offset of the file, which does not hold true anymore after
2343 em_new = create_io_em(inode, em->start + pre, len,
2344 em->start + pre, em->block_start + pre, len,
2345 len, len, BTRFS_COMPRESS_NONE,
2346 BTRFS_ORDERED_REGULAR);
2347 if (IS_ERR(em_new)) {
2348 ret = PTR_ERR(em_new);
2351 free_extent_map(em_new);
2354 free_extent_map(em);
2355 btrfs_put_ordered_extent(ordered);
2357 return errno_to_blk_status(ret);
2361 * extent_io.c submission hook. This does the right thing for csum calculation
2362 * on write, or reading the csums from the tree before a read.
2364 * Rules about async/sync submit,
2365 * a) read: sync submit
2367 * b) write without checksum: sync submit
2369 * c) write with checksum:
2370 * c-1) if bio is issued by fsync: sync submit
2371 * (sync_writers != 0)
2373 * c-2) if root is reloc root: sync submit
2374 * (only in case of buffered IO)
2376 * c-3) otherwise: async submit
2378 blk_status_t btrfs_submit_data_bio(struct inode *inode, struct bio *bio,
2379 int mirror_num, unsigned long bio_flags)
2382 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2383 struct btrfs_root *root = BTRFS_I(inode)->root;
2384 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
2385 blk_status_t ret = 0;
2387 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
2389 skip_sum = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) ||
2390 !fs_info->csum_root;
2392 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
2393 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
2395 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
2396 struct page *page = bio_first_bvec_all(bio)->bv_page;
2397 loff_t file_offset = page_offset(page);
2399 ret = extract_ordered_extent(BTRFS_I(inode), bio, file_offset);
2404 if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
2405 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
2409 if (bio_flags & EXTENT_BIO_COMPRESSED) {
2410 ret = btrfs_submit_compressed_read(inode, bio,
2416 * Lookup bio sums does extra checks around whether we
2417 * need to csum or not, which is why we ignore skip_sum
2420 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
2425 } else if (async && !skip_sum) {
2426 /* csum items have already been cloned */
2427 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2429 /* we're doing a write, do the async checksumming */
2430 ret = btrfs_wq_submit_bio(inode, bio, mirror_num, bio_flags,
2431 0, btrfs_submit_bio_start);
2433 } else if (!skip_sum) {
2434 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, 0, 0);
2440 ret = btrfs_map_bio(fs_info, bio, mirror_num);
2444 bio->bi_status = ret;
2451 * given a list of ordered sums record them in the inode. This happens
2452 * at IO completion time based on sums calculated at bio submission time.
2454 static int add_pending_csums(struct btrfs_trans_handle *trans,
2455 struct list_head *list)
2457 struct btrfs_ordered_sum *sum;
2460 list_for_each_entry(sum, list, list) {
2461 trans->adding_csums = true;
2462 ret = btrfs_csum_file_blocks(trans, trans->fs_info->csum_root, sum);
2463 trans->adding_csums = false;
2470 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2473 struct extent_state **cached_state)
2475 u64 search_start = start;
2476 const u64 end = start + len - 1;
2478 while (search_start < end) {
2479 const u64 search_len = end - search_start + 1;
2480 struct extent_map *em;
2484 em = btrfs_get_extent(inode, NULL, 0, search_start, search_len);
2488 if (em->block_start != EXTENT_MAP_HOLE)
2492 if (em->start < search_start)
2493 em_len -= search_start - em->start;
2494 if (em_len > search_len)
2495 em_len = search_len;
2497 ret = set_extent_bit(&inode->io_tree, search_start,
2498 search_start + em_len - 1,
2499 EXTENT_DELALLOC_NEW, 0, NULL, cached_state,
2502 search_start = extent_map_end(em);
2503 free_extent_map(em);
2510 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2511 unsigned int extra_bits,
2512 struct extent_state **cached_state)
2514 WARN_ON(PAGE_ALIGNED(end));
2516 if (start >= i_size_read(&inode->vfs_inode) &&
2517 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2519 * There can't be any extents following eof in this case so just
2520 * set the delalloc new bit for the range directly.
2522 extra_bits |= EXTENT_DELALLOC_NEW;
2526 ret = btrfs_find_new_delalloc_bytes(inode, start,
2533 return set_extent_delalloc(&inode->io_tree, start, end, extra_bits,
2537 /* see btrfs_writepage_start_hook for details on why this is required */
2538 struct btrfs_writepage_fixup {
2540 struct inode *inode;
2541 struct btrfs_work work;
2544 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2546 struct btrfs_writepage_fixup *fixup;
2547 struct btrfs_ordered_extent *ordered;
2548 struct extent_state *cached_state = NULL;
2549 struct extent_changeset *data_reserved = NULL;
2551 struct btrfs_inode *inode;
2555 bool free_delalloc_space = true;
2557 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2559 inode = BTRFS_I(fixup->inode);
2560 page_start = page_offset(page);
2561 page_end = page_offset(page) + PAGE_SIZE - 1;
2564 * This is similar to page_mkwrite, we need to reserve the space before
2565 * we take the page lock.
2567 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2573 * Before we queued this fixup, we took a reference on the page.
2574 * page->mapping may go NULL, but it shouldn't be moved to a different
2577 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2579 * Unfortunately this is a little tricky, either
2581 * 1) We got here and our page had already been dealt with and
2582 * we reserved our space, thus ret == 0, so we need to just
2583 * drop our space reservation and bail. This can happen the
2584 * first time we come into the fixup worker, or could happen
2585 * while waiting for the ordered extent.
2586 * 2) Our page was already dealt with, but we happened to get an
2587 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2588 * this case we obviously don't have anything to release, but
2589 * because the page was already dealt with we don't want to
2590 * mark the page with an error, so make sure we're resetting
2591 * ret to 0. This is why we have this check _before_ the ret
2592 * check, because we do not want to have a surprise ENOSPC
2593 * when the page was already properly dealt with.
2596 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2597 btrfs_delalloc_release_space(inode, data_reserved,
2598 page_start, PAGE_SIZE,
2606 * We can't mess with the page state unless it is locked, so now that
2607 * it is locked bail if we failed to make our space reservation.
2612 lock_extent_bits(&inode->io_tree, page_start, page_end, &cached_state);
2614 /* already ordered? We're done */
2615 if (PageOrdered(page))
2618 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2620 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2623 btrfs_start_ordered_extent(ordered, 1);
2624 btrfs_put_ordered_extent(ordered);
2628 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2634 * Everything went as planned, we're now the owner of a dirty page with
2635 * delayed allocation bits set and space reserved for our COW
2638 * The page was dirty when we started, nothing should have cleaned it.
2640 BUG_ON(!PageDirty(page));
2641 free_delalloc_space = false;
2643 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2644 if (free_delalloc_space)
2645 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2647 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2652 * We hit ENOSPC or other errors. Update the mapping and page
2653 * to reflect the errors and clean the page.
2655 mapping_set_error(page->mapping, ret);
2656 end_extent_writepage(page, ret, page_start, page_end);
2657 clear_page_dirty_for_io(page);
2660 ClearPageChecked(page);
2664 extent_changeset_free(data_reserved);
2666 * As a precaution, do a delayed iput in case it would be the last iput
2667 * that could need flushing space. Recursing back to fixup worker would
2670 btrfs_add_delayed_iput(&inode->vfs_inode);
2674 * There are a few paths in the higher layers of the kernel that directly
2675 * set the page dirty bit without asking the filesystem if it is a
2676 * good idea. This causes problems because we want to make sure COW
2677 * properly happens and the data=ordered rules are followed.
2679 * In our case any range that doesn't have the ORDERED bit set
2680 * hasn't been properly setup for IO. We kick off an async process
2681 * to fix it up. The async helper will wait for ordered extents, set
2682 * the delalloc bit and make it safe to write the page.
2684 int btrfs_writepage_cow_fixup(struct page *page, u64 start, u64 end)
2686 struct inode *inode = page->mapping->host;
2687 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2688 struct btrfs_writepage_fixup *fixup;
2690 /* This page has ordered extent covering it already */
2691 if (PageOrdered(page))
2695 * PageChecked is set below when we create a fixup worker for this page,
2696 * don't try to create another one if we're already PageChecked()
2698 * The extent_io writepage code will redirty the page if we send back
2701 if (PageChecked(page))
2704 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2709 * We are already holding a reference to this inode from
2710 * write_cache_pages. We need to hold it because the space reservation
2711 * takes place outside of the page lock, and we can't trust
2712 * page->mapping outside of the page lock.
2715 SetPageChecked(page);
2717 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
2719 fixup->inode = inode;
2720 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2725 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2726 struct btrfs_inode *inode, u64 file_pos,
2727 struct btrfs_file_extent_item *stack_fi,
2728 const bool update_inode_bytes,
2729 u64 qgroup_reserved)
2731 struct btrfs_root *root = inode->root;
2732 const u64 sectorsize = root->fs_info->sectorsize;
2733 struct btrfs_path *path;
2734 struct extent_buffer *leaf;
2735 struct btrfs_key ins;
2736 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
2737 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
2738 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
2739 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
2740 struct btrfs_drop_extents_args drop_args = { 0 };
2743 path = btrfs_alloc_path();
2748 * we may be replacing one extent in the tree with another.
2749 * The new extent is pinned in the extent map, and we don't want
2750 * to drop it from the cache until it is completely in the btree.
2752 * So, tell btrfs_drop_extents to leave this extent in the cache.
2753 * the caller is expected to unpin it and allow it to be merged
2756 drop_args.path = path;
2757 drop_args.start = file_pos;
2758 drop_args.end = file_pos + num_bytes;
2759 drop_args.replace_extent = true;
2760 drop_args.extent_item_size = sizeof(*stack_fi);
2761 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2765 if (!drop_args.extent_inserted) {
2766 ins.objectid = btrfs_ino(inode);
2767 ins.offset = file_pos;
2768 ins.type = BTRFS_EXTENT_DATA_KEY;
2770 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2775 leaf = path->nodes[0];
2776 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
2777 write_extent_buffer(leaf, stack_fi,
2778 btrfs_item_ptr_offset(leaf, path->slots[0]),
2779 sizeof(struct btrfs_file_extent_item));
2781 btrfs_mark_buffer_dirty(leaf);
2782 btrfs_release_path(path);
2785 * If we dropped an inline extent here, we know the range where it is
2786 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
2787 * number of bytes only for that range containing the inline extent.
2788 * The remaining of the range will be processed when clearning the
2789 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
2791 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
2792 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
2794 inline_size = drop_args.bytes_found - inline_size;
2795 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
2796 drop_args.bytes_found -= inline_size;
2797 num_bytes -= sectorsize;
2800 if (update_inode_bytes)
2801 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
2803 ins.objectid = disk_bytenr;
2804 ins.offset = disk_num_bytes;
2805 ins.type = BTRFS_EXTENT_ITEM_KEY;
2807 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
2811 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
2812 file_pos, qgroup_reserved, &ins);
2814 btrfs_free_path(path);
2819 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2822 struct btrfs_block_group *cache;
2824 cache = btrfs_lookup_block_group(fs_info, start);
2827 spin_lock(&cache->lock);
2828 cache->delalloc_bytes -= len;
2829 spin_unlock(&cache->lock);
2831 btrfs_put_block_group(cache);
2834 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
2835 struct btrfs_ordered_extent *oe)
2837 struct btrfs_file_extent_item stack_fi;
2839 bool update_inode_bytes;
2841 memset(&stack_fi, 0, sizeof(stack_fi));
2842 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
2843 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
2844 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
2845 oe->disk_num_bytes);
2846 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags))
2847 logical_len = oe->truncated_len;
2849 logical_len = oe->num_bytes;
2850 btrfs_set_stack_file_extent_num_bytes(&stack_fi, logical_len);
2851 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, logical_len);
2852 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
2853 /* Encryption and other encoding is reserved and all 0 */
2856 * For delalloc, when completing an ordered extent we update the inode's
2857 * bytes when clearing the range in the inode's io tree, so pass false
2858 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
2859 * except if the ordered extent was truncated.
2861 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
2862 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
2864 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
2865 oe->file_offset, &stack_fi,
2866 update_inode_bytes, oe->qgroup_rsv);
2870 * As ordered data IO finishes, this gets called so we can finish
2871 * an ordered extent if the range of bytes in the file it covers are
2874 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2876 struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
2877 struct btrfs_root *root = inode->root;
2878 struct btrfs_fs_info *fs_info = root->fs_info;
2879 struct btrfs_trans_handle *trans = NULL;
2880 struct extent_io_tree *io_tree = &inode->io_tree;
2881 struct extent_state *cached_state = NULL;
2883 int compress_type = 0;
2885 u64 logical_len = ordered_extent->num_bytes;
2886 bool freespace_inode;
2887 bool truncated = false;
2888 bool clear_reserved_extent = true;
2889 unsigned int clear_bits = EXTENT_DEFRAG;
2891 start = ordered_extent->file_offset;
2892 end = start + ordered_extent->num_bytes - 1;
2894 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2895 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2896 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2897 clear_bits |= EXTENT_DELALLOC_NEW;
2899 freespace_inode = btrfs_is_free_space_inode(inode);
2901 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2906 if (ordered_extent->disk)
2907 btrfs_rewrite_logical_zoned(ordered_extent);
2909 btrfs_free_io_failure_record(inode, start, end);
2911 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2913 logical_len = ordered_extent->truncated_len;
2914 /* Truncated the entire extent, don't bother adding */
2919 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2920 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2922 btrfs_inode_safe_disk_i_size_write(inode, 0);
2923 if (freespace_inode)
2924 trans = btrfs_join_transaction_spacecache(root);
2926 trans = btrfs_join_transaction(root);
2927 if (IS_ERR(trans)) {
2928 ret = PTR_ERR(trans);
2932 trans->block_rsv = &inode->block_rsv;
2933 ret = btrfs_update_inode_fallback(trans, root, inode);
2934 if (ret) /* -ENOMEM or corruption */
2935 btrfs_abort_transaction(trans, ret);
2939 clear_bits |= EXTENT_LOCKED;
2940 lock_extent_bits(io_tree, start, end, &cached_state);
2942 if (freespace_inode)
2943 trans = btrfs_join_transaction_spacecache(root);
2945 trans = btrfs_join_transaction(root);
2946 if (IS_ERR(trans)) {
2947 ret = PTR_ERR(trans);
2952 trans->block_rsv = &inode->block_rsv;
2954 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2955 compress_type = ordered_extent->compress_type;
2956 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2957 BUG_ON(compress_type);
2958 ret = btrfs_mark_extent_written(trans, inode,
2959 ordered_extent->file_offset,
2960 ordered_extent->file_offset +
2963 BUG_ON(root == fs_info->tree_root);
2964 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
2966 clear_reserved_extent = false;
2967 btrfs_release_delalloc_bytes(fs_info,
2968 ordered_extent->disk_bytenr,
2969 ordered_extent->disk_num_bytes);
2972 unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset,
2973 ordered_extent->num_bytes, trans->transid);
2975 btrfs_abort_transaction(trans, ret);
2979 ret = add_pending_csums(trans, &ordered_extent->list);
2981 btrfs_abort_transaction(trans, ret);
2986 * If this is a new delalloc range, clear its new delalloc flag to
2987 * update the inode's number of bytes. This needs to be done first
2988 * before updating the inode item.
2990 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
2991 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
2992 clear_extent_bit(&inode->io_tree, start, end,
2993 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
2994 0, 0, &cached_state);
2996 btrfs_inode_safe_disk_i_size_write(inode, 0);
2997 ret = btrfs_update_inode_fallback(trans, root, inode);
2998 if (ret) { /* -ENOMEM or corruption */
2999 btrfs_abort_transaction(trans, ret);
3004 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
3005 (clear_bits & EXTENT_LOCKED) ? 1 : 0, 0,
3009 btrfs_end_transaction(trans);
3011 if (ret || truncated) {
3012 u64 unwritten_start = start;
3015 * If we failed to finish this ordered extent for any reason we
3016 * need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3017 * extent, and mark the inode with the error if it wasn't
3018 * already set. Any error during writeback would have already
3019 * set the mapping error, so we need to set it if we're the ones
3020 * marking this ordered extent as failed.
3022 if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR,
3023 &ordered_extent->flags))
3024 mapping_set_error(ordered_extent->inode->i_mapping, -EIO);
3027 unwritten_start += logical_len;
3028 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
3030 /* Drop the cache for the part of the extent we didn't write. */
3031 btrfs_drop_extent_cache(inode, unwritten_start, end, 0);
3034 * If the ordered extent had an IOERR or something else went
3035 * wrong we need to return the space for this ordered extent
3036 * back to the allocator. We only free the extent in the
3037 * truncated case if we didn't write out the extent at all.
3039 * If we made it past insert_reserved_file_extent before we
3040 * errored out then we don't need to do this as the accounting
3041 * has already been done.
3043 if ((ret || !logical_len) &&
3044 clear_reserved_extent &&
3045 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3046 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3048 * Discard the range before returning it back to the
3051 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3052 btrfs_discard_extent(fs_info,
3053 ordered_extent->disk_bytenr,
3054 ordered_extent->disk_num_bytes,
3056 btrfs_free_reserved_extent(fs_info,
3057 ordered_extent->disk_bytenr,
3058 ordered_extent->disk_num_bytes, 1);
3063 * This needs to be done to make sure anybody waiting knows we are done
3064 * updating everything for this ordered extent.
3066 btrfs_remove_ordered_extent(inode, ordered_extent);
3069 btrfs_put_ordered_extent(ordered_extent);
3070 /* once for the tree */
3071 btrfs_put_ordered_extent(ordered_extent);
3076 static void finish_ordered_fn(struct btrfs_work *work)
3078 struct btrfs_ordered_extent *ordered_extent;
3079 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3080 btrfs_finish_ordered_io(ordered_extent);
3083 void btrfs_writepage_endio_finish_ordered(struct btrfs_inode *inode,
3084 struct page *page, u64 start,
3085 u64 end, int uptodate)
3087 trace_btrfs_writepage_end_io_hook(inode, start, end, uptodate);
3089 btrfs_mark_ordered_io_finished(inode, page, start, end + 1 - start,
3090 finish_ordered_fn, uptodate);
3094 * check_data_csum - verify checksum of one sector of uncompressed data
3096 * @io_bio: btrfs_io_bio which contains the csum
3097 * @bio_offset: offset to the beginning of the bio (in bytes)
3098 * @page: page where is the data to be verified
3099 * @pgoff: offset inside the page
3100 * @start: logical offset in the file
3102 * The length of such check is always one sector size.
3104 static int check_data_csum(struct inode *inode, struct btrfs_io_bio *io_bio,
3105 u32 bio_offset, struct page *page, u32 pgoff,
3108 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3109 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3111 u32 len = fs_info->sectorsize;
3112 const u32 csum_size = fs_info->csum_size;
3113 unsigned int offset_sectors;
3115 u8 csum[BTRFS_CSUM_SIZE];
3117 ASSERT(pgoff + len <= PAGE_SIZE);
3119 offset_sectors = bio_offset >> fs_info->sectorsize_bits;
3120 csum_expected = ((u8 *)io_bio->csum) + offset_sectors * csum_size;
3122 kaddr = kmap_atomic(page);
3123 shash->tfm = fs_info->csum_shash;
3125 crypto_shash_digest(shash, kaddr + pgoff, len, csum);
3127 if (memcmp(csum, csum_expected, csum_size))
3130 kunmap_atomic(kaddr);
3133 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3134 io_bio->mirror_num);
3136 btrfs_dev_stat_inc_and_print(io_bio->device,
3137 BTRFS_DEV_STAT_CORRUPTION_ERRS);
3138 memset(kaddr + pgoff, 1, len);
3139 flush_dcache_page(page);
3140 kunmap_atomic(kaddr);
3145 * When reads are done, we need to check csums to verify the data is correct.
3146 * if there's a match, we allow the bio to finish. If not, the code in
3147 * extent_io.c will try to find good copies for us.
3149 * @bio_offset: offset to the beginning of the bio (in bytes)
3150 * @start: file offset of the range start
3151 * @end: file offset of the range end (inclusive)
3153 * Return a bitmap where bit set means a csum mismatch, and bit not set means
3156 unsigned int btrfs_verify_data_csum(struct btrfs_io_bio *io_bio, u32 bio_offset,
3157 struct page *page, u64 start, u64 end)
3159 struct inode *inode = page->mapping->host;
3160 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3161 struct btrfs_root *root = BTRFS_I(inode)->root;
3162 const u32 sectorsize = root->fs_info->sectorsize;
3164 unsigned int result = 0;
3166 if (PageChecked(page)) {
3167 ClearPageChecked(page);
3171 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3174 if (!root->fs_info->csum_root)
3177 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3178 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3179 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3183 ASSERT(page_offset(page) <= start &&
3184 end <= page_offset(page) + PAGE_SIZE - 1);
3185 for (pg_off = offset_in_page(start);
3186 pg_off < offset_in_page(end);
3187 pg_off += sectorsize, bio_offset += sectorsize) {
3190 ret = check_data_csum(inode, io_bio, bio_offset, page, pg_off,
3191 page_offset(page) + pg_off);
3193 const int nr_bit = (pg_off - offset_in_page(start)) >>
3194 root->fs_info->sectorsize_bits;
3196 result |= (1U << nr_bit);
3203 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3205 * @inode: The inode we want to perform iput on
3207 * This function uses the generic vfs_inode::i_count to track whether we should
3208 * just decrement it (in case it's > 1) or if this is the last iput then link
3209 * the inode to the delayed iput machinery. Delayed iputs are processed at
3210 * transaction commit time/superblock commit/cleaner kthread.
3212 void btrfs_add_delayed_iput(struct inode *inode)
3214 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3215 struct btrfs_inode *binode = BTRFS_I(inode);
3217 if (atomic_add_unless(&inode->i_count, -1, 1))
3220 atomic_inc(&fs_info->nr_delayed_iputs);
3221 spin_lock(&fs_info->delayed_iput_lock);
3222 ASSERT(list_empty(&binode->delayed_iput));
3223 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3224 spin_unlock(&fs_info->delayed_iput_lock);
3225 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3226 wake_up_process(fs_info->cleaner_kthread);
3229 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3230 struct btrfs_inode *inode)
3232 list_del_init(&inode->delayed_iput);
3233 spin_unlock(&fs_info->delayed_iput_lock);
3234 iput(&inode->vfs_inode);
3235 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3236 wake_up(&fs_info->delayed_iputs_wait);
3237 spin_lock(&fs_info->delayed_iput_lock);
3240 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3241 struct btrfs_inode *inode)
3243 if (!list_empty(&inode->delayed_iput)) {
3244 spin_lock(&fs_info->delayed_iput_lock);
3245 if (!list_empty(&inode->delayed_iput))
3246 run_delayed_iput_locked(fs_info, inode);
3247 spin_unlock(&fs_info->delayed_iput_lock);
3251 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3254 spin_lock(&fs_info->delayed_iput_lock);
3255 while (!list_empty(&fs_info->delayed_iputs)) {
3256 struct btrfs_inode *inode;
3258 inode = list_first_entry(&fs_info->delayed_iputs,
3259 struct btrfs_inode, delayed_iput);
3260 run_delayed_iput_locked(fs_info, inode);
3261 cond_resched_lock(&fs_info->delayed_iput_lock);
3263 spin_unlock(&fs_info->delayed_iput_lock);
3267 * Wait for flushing all delayed iputs
3269 * @fs_info: the filesystem
3271 * This will wait on any delayed iputs that are currently running with KILLABLE
3272 * set. Once they are all done running we will return, unless we are killed in
3273 * which case we return EINTR. This helps in user operations like fallocate etc
3274 * that might get blocked on the iputs.
3276 * Return EINTR if we were killed, 0 if nothing's pending
3278 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3280 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3281 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3288 * This creates an orphan entry for the given inode in case something goes wrong
3289 * in the middle of an unlink.
3291 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3292 struct btrfs_inode *inode)
3296 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3297 if (ret && ret != -EEXIST) {
3298 btrfs_abort_transaction(trans, ret);
3306 * We have done the delete so we can go ahead and remove the orphan item for
3307 * this particular inode.
3309 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3310 struct btrfs_inode *inode)
3312 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3316 * this cleans up any orphans that may be left on the list from the last use
3319 int btrfs_orphan_cleanup(struct btrfs_root *root)
3321 struct btrfs_fs_info *fs_info = root->fs_info;
3322 struct btrfs_path *path;
3323 struct extent_buffer *leaf;
3324 struct btrfs_key key, found_key;
3325 struct btrfs_trans_handle *trans;
3326 struct inode *inode;
3327 u64 last_objectid = 0;
3328 int ret = 0, nr_unlink = 0;
3330 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3333 path = btrfs_alloc_path();
3338 path->reada = READA_BACK;
3340 key.objectid = BTRFS_ORPHAN_OBJECTID;
3341 key.type = BTRFS_ORPHAN_ITEM_KEY;
3342 key.offset = (u64)-1;
3345 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3350 * if ret == 0 means we found what we were searching for, which
3351 * is weird, but possible, so only screw with path if we didn't
3352 * find the key and see if we have stuff that matches
3356 if (path->slots[0] == 0)
3361 /* pull out the item */
3362 leaf = path->nodes[0];
3363 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3365 /* make sure the item matches what we want */
3366 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3368 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3371 /* release the path since we're done with it */
3372 btrfs_release_path(path);
3375 * this is where we are basically btrfs_lookup, without the
3376 * crossing root thing. we store the inode number in the
3377 * offset of the orphan item.
3380 if (found_key.offset == last_objectid) {
3382 "Error removing orphan entry, stopping orphan cleanup");
3387 last_objectid = found_key.offset;
3389 found_key.objectid = found_key.offset;
3390 found_key.type = BTRFS_INODE_ITEM_KEY;
3391 found_key.offset = 0;
3392 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3393 ret = PTR_ERR_OR_ZERO(inode);
3394 if (ret && ret != -ENOENT)
3397 if (ret == -ENOENT && root == fs_info->tree_root) {
3398 struct btrfs_root *dead_root;
3399 int is_dead_root = 0;
3402 * This is an orphan in the tree root. Currently these
3403 * could come from 2 sources:
3404 * a) a root (snapshot/subvolume) deletion in progress
3405 * b) a free space cache inode
3406 * We need to distinguish those two, as the orphan item
3407 * for a root must not get deleted before the deletion
3408 * of the snapshot/subvolume's tree completes.
3410 * btrfs_find_orphan_roots() ran before us, which has
3411 * found all deleted roots and loaded them into
3412 * fs_info->fs_roots_radix. So here we can find if an
3413 * orphan item corresponds to a deleted root by looking
3414 * up the root from that radix tree.
3417 spin_lock(&fs_info->fs_roots_radix_lock);
3418 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3419 (unsigned long)found_key.objectid);
3420 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3422 spin_unlock(&fs_info->fs_roots_radix_lock);
3425 /* prevent this orphan from being found again */
3426 key.offset = found_key.objectid - 1;
3433 * If we have an inode with links, there are a couple of
3434 * possibilities. Old kernels (before v3.12) used to create an
3435 * orphan item for truncate indicating that there were possibly
3436 * extent items past i_size that needed to be deleted. In v3.12,
3437 * truncate was changed to update i_size in sync with the extent
3438 * items, but the (useless) orphan item was still created. Since
3439 * v4.18, we don't create the orphan item for truncate at all.
3441 * So, this item could mean that we need to do a truncate, but
3442 * only if this filesystem was last used on a pre-v3.12 kernel
3443 * and was not cleanly unmounted. The odds of that are quite
3444 * slim, and it's a pain to do the truncate now, so just delete
3447 * It's also possible that this orphan item was supposed to be
3448 * deleted but wasn't. The inode number may have been reused,
3449 * but either way, we can delete the orphan item.
3451 if (ret == -ENOENT || inode->i_nlink) {
3454 trans = btrfs_start_transaction(root, 1);
3455 if (IS_ERR(trans)) {
3456 ret = PTR_ERR(trans);
3459 btrfs_debug(fs_info, "auto deleting %Lu",
3460 found_key.objectid);
3461 ret = btrfs_del_orphan_item(trans, root,
3462 found_key.objectid);
3463 btrfs_end_transaction(trans);
3471 /* this will do delete_inode and everything for us */
3474 /* release the path since we're done with it */
3475 btrfs_release_path(path);
3477 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3479 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3480 trans = btrfs_join_transaction(root);
3482 btrfs_end_transaction(trans);
3486 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3490 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3491 btrfs_free_path(path);
3496 * very simple check to peek ahead in the leaf looking for xattrs. If we
3497 * don't find any xattrs, we know there can't be any acls.
3499 * slot is the slot the inode is in, objectid is the objectid of the inode
3501 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3502 int slot, u64 objectid,
3503 int *first_xattr_slot)
3505 u32 nritems = btrfs_header_nritems(leaf);
3506 struct btrfs_key found_key;
3507 static u64 xattr_access = 0;
3508 static u64 xattr_default = 0;
3511 if (!xattr_access) {
3512 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3513 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3514 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3515 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3519 *first_xattr_slot = -1;
3520 while (slot < nritems) {
3521 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3523 /* we found a different objectid, there must not be acls */
3524 if (found_key.objectid != objectid)
3527 /* we found an xattr, assume we've got an acl */
3528 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3529 if (*first_xattr_slot == -1)
3530 *first_xattr_slot = slot;
3531 if (found_key.offset == xattr_access ||
3532 found_key.offset == xattr_default)
3537 * we found a key greater than an xattr key, there can't
3538 * be any acls later on
3540 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3547 * it goes inode, inode backrefs, xattrs, extents,
3548 * so if there are a ton of hard links to an inode there can
3549 * be a lot of backrefs. Don't waste time searching too hard,
3550 * this is just an optimization
3555 /* we hit the end of the leaf before we found an xattr or
3556 * something larger than an xattr. We have to assume the inode
3559 if (*first_xattr_slot == -1)
3560 *first_xattr_slot = slot;
3565 * read an inode from the btree into the in-memory inode
3567 static int btrfs_read_locked_inode(struct inode *inode,
3568 struct btrfs_path *in_path)
3570 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3571 struct btrfs_path *path = in_path;
3572 struct extent_buffer *leaf;
3573 struct btrfs_inode_item *inode_item;
3574 struct btrfs_root *root = BTRFS_I(inode)->root;
3575 struct btrfs_key location;
3580 bool filled = false;
3581 int first_xattr_slot;
3583 ret = btrfs_fill_inode(inode, &rdev);
3588 path = btrfs_alloc_path();
3593 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3595 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3597 if (path != in_path)
3598 btrfs_free_path(path);
3602 leaf = path->nodes[0];
3607 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3608 struct btrfs_inode_item);
3609 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3610 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3611 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3612 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3613 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3614 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3615 round_up(i_size_read(inode), fs_info->sectorsize));
3617 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3618 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3620 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3621 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3623 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3624 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3626 BTRFS_I(inode)->i_otime.tv_sec =
3627 btrfs_timespec_sec(leaf, &inode_item->otime);
3628 BTRFS_I(inode)->i_otime.tv_nsec =
3629 btrfs_timespec_nsec(leaf, &inode_item->otime);
3631 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3632 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3633 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3635 inode_set_iversion_queried(inode,
3636 btrfs_inode_sequence(leaf, inode_item));
3637 inode->i_generation = BTRFS_I(inode)->generation;
3639 rdev = btrfs_inode_rdev(leaf, inode_item);
3641 BTRFS_I(inode)->index_cnt = (u64)-1;
3642 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3646 * If we were modified in the current generation and evicted from memory
3647 * and then re-read we need to do a full sync since we don't have any
3648 * idea about which extents were modified before we were evicted from
3651 * This is required for both inode re-read from disk and delayed inode
3652 * in delayed_nodes_tree.
3654 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3655 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3656 &BTRFS_I(inode)->runtime_flags);
3659 * We don't persist the id of the transaction where an unlink operation
3660 * against the inode was last made. So here we assume the inode might
3661 * have been evicted, and therefore the exact value of last_unlink_trans
3662 * lost, and set it to last_trans to avoid metadata inconsistencies
3663 * between the inode and its parent if the inode is fsync'ed and the log
3664 * replayed. For example, in the scenario:
3667 * ln mydir/foo mydir/bar
3670 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3671 * xfs_io -c fsync mydir/foo
3673 * mount fs, triggers fsync log replay
3675 * We must make sure that when we fsync our inode foo we also log its
3676 * parent inode, otherwise after log replay the parent still has the
3677 * dentry with the "bar" name but our inode foo has a link count of 1
3678 * and doesn't have an inode ref with the name "bar" anymore.
3680 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3681 * but it guarantees correctness at the expense of occasional full
3682 * transaction commits on fsync if our inode is a directory, or if our
3683 * inode is not a directory, logging its parent unnecessarily.
3685 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3688 * Same logic as for last_unlink_trans. We don't persist the generation
3689 * of the last transaction where this inode was used for a reflink
3690 * operation, so after eviction and reloading the inode we must be
3691 * pessimistic and assume the last transaction that modified the inode.
3693 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3696 if (inode->i_nlink != 1 ||
3697 path->slots[0] >= btrfs_header_nritems(leaf))
3700 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3701 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3704 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3705 if (location.type == BTRFS_INODE_REF_KEY) {
3706 struct btrfs_inode_ref *ref;
3708 ref = (struct btrfs_inode_ref *)ptr;
3709 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3710 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3711 struct btrfs_inode_extref *extref;
3713 extref = (struct btrfs_inode_extref *)ptr;
3714 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3719 * try to precache a NULL acl entry for files that don't have
3720 * any xattrs or acls
3722 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3723 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3724 if (first_xattr_slot != -1) {
3725 path->slots[0] = first_xattr_slot;
3726 ret = btrfs_load_inode_props(inode, path);
3729 "error loading props for ino %llu (root %llu): %d",
3730 btrfs_ino(BTRFS_I(inode)),
3731 root->root_key.objectid, ret);
3733 if (path != in_path)
3734 btrfs_free_path(path);
3737 cache_no_acl(inode);
3739 switch (inode->i_mode & S_IFMT) {
3741 inode->i_mapping->a_ops = &btrfs_aops;
3742 inode->i_fop = &btrfs_file_operations;
3743 inode->i_op = &btrfs_file_inode_operations;
3746 inode->i_fop = &btrfs_dir_file_operations;
3747 inode->i_op = &btrfs_dir_inode_operations;
3750 inode->i_op = &btrfs_symlink_inode_operations;
3751 inode_nohighmem(inode);
3752 inode->i_mapping->a_ops = &btrfs_aops;
3755 inode->i_op = &btrfs_special_inode_operations;
3756 init_special_inode(inode, inode->i_mode, rdev);
3760 btrfs_sync_inode_flags_to_i_flags(inode);
3765 * given a leaf and an inode, copy the inode fields into the leaf
3767 static void fill_inode_item(struct btrfs_trans_handle *trans,
3768 struct extent_buffer *leaf,
3769 struct btrfs_inode_item *item,
3770 struct inode *inode)
3772 struct btrfs_map_token token;
3774 btrfs_init_map_token(&token, leaf);
3776 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3777 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3778 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
3779 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3780 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3782 btrfs_set_token_timespec_sec(&token, &item->atime,
3783 inode->i_atime.tv_sec);
3784 btrfs_set_token_timespec_nsec(&token, &item->atime,
3785 inode->i_atime.tv_nsec);
3787 btrfs_set_token_timespec_sec(&token, &item->mtime,
3788 inode->i_mtime.tv_sec);
3789 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3790 inode->i_mtime.tv_nsec);
3792 btrfs_set_token_timespec_sec(&token, &item->ctime,
3793 inode->i_ctime.tv_sec);
3794 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3795 inode->i_ctime.tv_nsec);
3797 btrfs_set_token_timespec_sec(&token, &item->otime,
3798 BTRFS_I(inode)->i_otime.tv_sec);
3799 btrfs_set_token_timespec_nsec(&token, &item->otime,
3800 BTRFS_I(inode)->i_otime.tv_nsec);
3802 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
3803 btrfs_set_token_inode_generation(&token, item,
3804 BTRFS_I(inode)->generation);
3805 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
3806 btrfs_set_token_inode_transid(&token, item, trans->transid);
3807 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
3808 btrfs_set_token_inode_flags(&token, item, BTRFS_I(inode)->flags);
3809 btrfs_set_token_inode_block_group(&token, item, 0);
3813 * copy everything in the in-memory inode into the btree.
3815 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3816 struct btrfs_root *root,
3817 struct btrfs_inode *inode)
3819 struct btrfs_inode_item *inode_item;
3820 struct btrfs_path *path;
3821 struct extent_buffer *leaf;
3824 path = btrfs_alloc_path();
3828 ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1);
3835 leaf = path->nodes[0];
3836 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3837 struct btrfs_inode_item);
3839 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
3840 btrfs_mark_buffer_dirty(leaf);
3841 btrfs_set_inode_last_trans(trans, inode);
3844 btrfs_free_path(path);
3849 * copy everything in the in-memory inode into the btree.
3851 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3852 struct btrfs_root *root,
3853 struct btrfs_inode *inode)
3855 struct btrfs_fs_info *fs_info = root->fs_info;
3859 * If the inode is a free space inode, we can deadlock during commit
3860 * if we put it into the delayed code.
3862 * The data relocation inode should also be directly updated
3865 if (!btrfs_is_free_space_inode(inode)
3866 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3867 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
3868 btrfs_update_root_times(trans, root);
3870 ret = btrfs_delayed_update_inode(trans, root, inode);
3872 btrfs_set_inode_last_trans(trans, inode);
3876 return btrfs_update_inode_item(trans, root, inode);
3879 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3880 struct btrfs_root *root, struct btrfs_inode *inode)
3884 ret = btrfs_update_inode(trans, root, inode);
3886 return btrfs_update_inode_item(trans, root, inode);
3891 * unlink helper that gets used here in inode.c and in the tree logging
3892 * recovery code. It remove a link in a directory with a given name, and
3893 * also drops the back refs in the inode to the directory
3895 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3896 struct btrfs_root *root,
3897 struct btrfs_inode *dir,
3898 struct btrfs_inode *inode,
3899 const char *name, int name_len)
3901 struct btrfs_fs_info *fs_info = root->fs_info;
3902 struct btrfs_path *path;
3904 struct btrfs_dir_item *di;
3906 u64 ino = btrfs_ino(inode);
3907 u64 dir_ino = btrfs_ino(dir);
3909 path = btrfs_alloc_path();
3915 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3916 name, name_len, -1);
3917 if (IS_ERR_OR_NULL(di)) {
3918 ret = di ? PTR_ERR(di) : -ENOENT;
3921 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3924 btrfs_release_path(path);
3927 * If we don't have dir index, we have to get it by looking up
3928 * the inode ref, since we get the inode ref, remove it directly,
3929 * it is unnecessary to do delayed deletion.
3931 * But if we have dir index, needn't search inode ref to get it.
3932 * Since the inode ref is close to the inode item, it is better
3933 * that we delay to delete it, and just do this deletion when
3934 * we update the inode item.
3936 if (inode->dir_index) {
3937 ret = btrfs_delayed_delete_inode_ref(inode);
3939 index = inode->dir_index;
3944 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3948 "failed to delete reference to %.*s, inode %llu parent %llu",
3949 name_len, name, ino, dir_ino);
3950 btrfs_abort_transaction(trans, ret);
3954 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
3956 btrfs_abort_transaction(trans, ret);
3960 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
3962 if (ret != 0 && ret != -ENOENT) {
3963 btrfs_abort_transaction(trans, ret);
3967 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
3972 btrfs_abort_transaction(trans, ret);
3975 * If we have a pending delayed iput we could end up with the final iput
3976 * being run in btrfs-cleaner context. If we have enough of these built
3977 * up we can end up burning a lot of time in btrfs-cleaner without any
3978 * way to throttle the unlinks. Since we're currently holding a ref on
3979 * the inode we can run the delayed iput here without any issues as the
3980 * final iput won't be done until after we drop the ref we're currently
3983 btrfs_run_delayed_iput(fs_info, inode);
3985 btrfs_free_path(path);
3989 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
3990 inode_inc_iversion(&inode->vfs_inode);
3991 inode_inc_iversion(&dir->vfs_inode);
3992 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
3993 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
3994 ret = btrfs_update_inode(trans, root, dir);
3999 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4000 struct btrfs_root *root,
4001 struct btrfs_inode *dir, struct btrfs_inode *inode,
4002 const char *name, int name_len)
4005 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4007 drop_nlink(&inode->vfs_inode);
4008 ret = btrfs_update_inode(trans, root, inode);
4014 * helper to start transaction for unlink and rmdir.
4016 * unlink and rmdir are special in btrfs, they do not always free space, so
4017 * if we cannot make our reservations the normal way try and see if there is
4018 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4019 * allow the unlink to occur.
4021 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4023 struct btrfs_root *root = BTRFS_I(dir)->root;
4026 * 1 for the possible orphan item
4027 * 1 for the dir item
4028 * 1 for the dir index
4029 * 1 for the inode ref
4032 return btrfs_start_transaction_fallback_global_rsv(root, 5);
4035 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4037 struct btrfs_root *root = BTRFS_I(dir)->root;
4038 struct btrfs_trans_handle *trans;
4039 struct inode *inode = d_inode(dentry);
4042 trans = __unlink_start_trans(dir);
4044 return PTR_ERR(trans);
4046 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4049 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4050 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4051 dentry->d_name.len);
4055 if (inode->i_nlink == 0) {
4056 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4062 btrfs_end_transaction(trans);
4063 btrfs_btree_balance_dirty(root->fs_info);
4067 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4068 struct inode *dir, struct dentry *dentry)
4070 struct btrfs_root *root = BTRFS_I(dir)->root;
4071 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4072 struct btrfs_path *path;
4073 struct extent_buffer *leaf;
4074 struct btrfs_dir_item *di;
4075 struct btrfs_key key;
4076 const char *name = dentry->d_name.name;
4077 int name_len = dentry->d_name.len;
4081 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4083 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4084 objectid = inode->root->root_key.objectid;
4085 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4086 objectid = inode->location.objectid;
4092 path = btrfs_alloc_path();
4096 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4097 name, name_len, -1);
4098 if (IS_ERR_OR_NULL(di)) {
4099 ret = di ? PTR_ERR(di) : -ENOENT;
4103 leaf = path->nodes[0];
4104 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4105 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4106 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4108 btrfs_abort_transaction(trans, ret);
4111 btrfs_release_path(path);
4114 * This is a placeholder inode for a subvolume we didn't have a
4115 * reference to at the time of the snapshot creation. In the meantime
4116 * we could have renamed the real subvol link into our snapshot, so
4117 * depending on btrfs_del_root_ref to return -ENOENT here is incorrect.
4118 * Instead simply lookup the dir_index_item for this entry so we can
4119 * remove it. Otherwise we know we have a ref to the root and we can
4120 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4122 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4123 di = btrfs_search_dir_index_item(root, path, dir_ino,
4125 if (IS_ERR_OR_NULL(di)) {
4130 btrfs_abort_transaction(trans, ret);
4134 leaf = path->nodes[0];
4135 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4137 btrfs_release_path(path);
4139 ret = btrfs_del_root_ref(trans, objectid,
4140 root->root_key.objectid, dir_ino,
4141 &index, name, name_len);
4143 btrfs_abort_transaction(trans, ret);
4148 ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index);
4150 btrfs_abort_transaction(trans, ret);
4154 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4155 inode_inc_iversion(dir);
4156 dir->i_mtime = dir->i_ctime = current_time(dir);
4157 ret = btrfs_update_inode_fallback(trans, root, BTRFS_I(dir));
4159 btrfs_abort_transaction(trans, ret);
4161 btrfs_free_path(path);
4166 * Helper to check if the subvolume references other subvolumes or if it's
4169 static noinline int may_destroy_subvol(struct btrfs_root *root)
4171 struct btrfs_fs_info *fs_info = root->fs_info;
4172 struct btrfs_path *path;
4173 struct btrfs_dir_item *di;
4174 struct btrfs_key key;
4178 path = btrfs_alloc_path();
4182 /* Make sure this root isn't set as the default subvol */
4183 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4184 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4185 dir_id, "default", 7, 0);
4186 if (di && !IS_ERR(di)) {
4187 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4188 if (key.objectid == root->root_key.objectid) {
4191 "deleting default subvolume %llu is not allowed",
4195 btrfs_release_path(path);
4198 key.objectid = root->root_key.objectid;
4199 key.type = BTRFS_ROOT_REF_KEY;
4200 key.offset = (u64)-1;
4202 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4208 if (path->slots[0] > 0) {
4210 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4211 if (key.objectid == root->root_key.objectid &&
4212 key.type == BTRFS_ROOT_REF_KEY)
4216 btrfs_free_path(path);
4220 /* Delete all dentries for inodes belonging to the root */
4221 static void btrfs_prune_dentries(struct btrfs_root *root)
4223 struct btrfs_fs_info *fs_info = root->fs_info;
4224 struct rb_node *node;
4225 struct rb_node *prev;
4226 struct btrfs_inode *entry;
4227 struct inode *inode;
4230 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
4231 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4233 spin_lock(&root->inode_lock);
4235 node = root->inode_tree.rb_node;
4239 entry = rb_entry(node, struct btrfs_inode, rb_node);
4241 if (objectid < btrfs_ino(entry))
4242 node = node->rb_left;
4243 else if (objectid > btrfs_ino(entry))
4244 node = node->rb_right;
4250 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4251 if (objectid <= btrfs_ino(entry)) {
4255 prev = rb_next(prev);
4259 entry = rb_entry(node, struct btrfs_inode, rb_node);
4260 objectid = btrfs_ino(entry) + 1;
4261 inode = igrab(&entry->vfs_inode);
4263 spin_unlock(&root->inode_lock);
4264 if (atomic_read(&inode->i_count) > 1)
4265 d_prune_aliases(inode);
4267 * btrfs_drop_inode will have it removed from the inode
4268 * cache when its usage count hits zero.
4272 spin_lock(&root->inode_lock);
4276 if (cond_resched_lock(&root->inode_lock))
4279 node = rb_next(node);
4281 spin_unlock(&root->inode_lock);
4284 int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
4286 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4287 struct btrfs_root *root = BTRFS_I(dir)->root;
4288 struct inode *inode = d_inode(dentry);
4289 struct btrfs_root *dest = BTRFS_I(inode)->root;
4290 struct btrfs_trans_handle *trans;
4291 struct btrfs_block_rsv block_rsv;
4296 * Don't allow to delete a subvolume with send in progress. This is
4297 * inside the inode lock so the error handling that has to drop the bit
4298 * again is not run concurrently.
4300 spin_lock(&dest->root_item_lock);
4301 if (dest->send_in_progress) {
4302 spin_unlock(&dest->root_item_lock);
4304 "attempt to delete subvolume %llu during send",
4305 dest->root_key.objectid);
4308 root_flags = btrfs_root_flags(&dest->root_item);
4309 btrfs_set_root_flags(&dest->root_item,
4310 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4311 spin_unlock(&dest->root_item_lock);
4313 down_write(&fs_info->subvol_sem);
4315 ret = may_destroy_subvol(dest);
4319 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4321 * One for dir inode,
4322 * two for dir entries,
4323 * two for root ref/backref.
4325 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4329 trans = btrfs_start_transaction(root, 0);
4330 if (IS_ERR(trans)) {
4331 ret = PTR_ERR(trans);
4334 trans->block_rsv = &block_rsv;
4335 trans->bytes_reserved = block_rsv.size;
4337 btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
4339 ret = btrfs_unlink_subvol(trans, dir, dentry);
4341 btrfs_abort_transaction(trans, ret);
4345 ret = btrfs_record_root_in_trans(trans, dest);
4347 btrfs_abort_transaction(trans, ret);
4351 memset(&dest->root_item.drop_progress, 0,
4352 sizeof(dest->root_item.drop_progress));
4353 btrfs_set_root_drop_level(&dest->root_item, 0);
4354 btrfs_set_root_refs(&dest->root_item, 0);
4356 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4357 ret = btrfs_insert_orphan_item(trans,
4359 dest->root_key.objectid);
4361 btrfs_abort_transaction(trans, ret);
4366 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4367 BTRFS_UUID_KEY_SUBVOL,
4368 dest->root_key.objectid);
4369 if (ret && ret != -ENOENT) {
4370 btrfs_abort_transaction(trans, ret);
4373 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4374 ret = btrfs_uuid_tree_remove(trans,
4375 dest->root_item.received_uuid,
4376 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4377 dest->root_key.objectid);
4378 if (ret && ret != -ENOENT) {
4379 btrfs_abort_transaction(trans, ret);
4384 free_anon_bdev(dest->anon_dev);
4387 trans->block_rsv = NULL;
4388 trans->bytes_reserved = 0;
4389 ret = btrfs_end_transaction(trans);
4390 inode->i_flags |= S_DEAD;
4392 btrfs_subvolume_release_metadata(root, &block_rsv);
4394 up_write(&fs_info->subvol_sem);
4396 spin_lock(&dest->root_item_lock);
4397 root_flags = btrfs_root_flags(&dest->root_item);
4398 btrfs_set_root_flags(&dest->root_item,
4399 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4400 spin_unlock(&dest->root_item_lock);
4402 d_invalidate(dentry);
4403 btrfs_prune_dentries(dest);
4404 ASSERT(dest->send_in_progress == 0);
4410 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4412 struct inode *inode = d_inode(dentry);
4414 struct btrfs_root *root = BTRFS_I(dir)->root;
4415 struct btrfs_trans_handle *trans;
4416 u64 last_unlink_trans;
4418 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4420 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4421 return btrfs_delete_subvolume(dir, dentry);
4423 trans = __unlink_start_trans(dir);
4425 return PTR_ERR(trans);
4427 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4428 err = btrfs_unlink_subvol(trans, dir, dentry);
4432 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4436 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4438 /* now the directory is empty */
4439 err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4440 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4441 dentry->d_name.len);
4443 btrfs_i_size_write(BTRFS_I(inode), 0);
4445 * Propagate the last_unlink_trans value of the deleted dir to
4446 * its parent directory. This is to prevent an unrecoverable
4447 * log tree in the case we do something like this:
4449 * 2) create snapshot under dir foo
4450 * 3) delete the snapshot
4453 * 6) fsync foo or some file inside foo
4455 if (last_unlink_trans >= trans->transid)
4456 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4459 btrfs_end_transaction(trans);
4460 btrfs_btree_balance_dirty(root->fs_info);
4466 * Return this if we need to call truncate_block for the last bit of the
4469 #define NEED_TRUNCATE_BLOCK 1
4472 * Remove inode items from a given root.
4474 * @trans: A transaction handle.
4475 * @root: The root from which to remove items.
4476 * @inode: The inode whose items we want to remove.
4477 * @new_size: The new i_size for the inode. This is only applicable when
4478 * @min_type is BTRFS_EXTENT_DATA_KEY, must be 0 otherwise.
4479 * @min_type: The minimum key type to remove. All keys with a type
4480 * greater than this value are removed and all keys with
4481 * this type are removed only if their offset is >= @new_size.
4482 * @extents_found: Output parameter that will contain the number of file
4483 * extent items that were removed or adjusted to the new
4484 * inode i_size. The caller is responsible for initializing
4485 * the counter. Also, it can be NULL if the caller does not
4486 * need this counter.
4488 * Remove all keys associated with the inode from the given root that have a key
4489 * with a type greater than or equals to @min_type. When @min_type has a value of
4490 * BTRFS_EXTENT_DATA_KEY, only remove file extent items that have an offset value
4491 * greater than or equals to @new_size. If a file extent item that starts before
4492 * @new_size and ends after it is found, its length is adjusted.
4494 * Returns: 0 on success, < 0 on error and NEED_TRUNCATE_BLOCK when @min_type is
4495 * BTRFS_EXTENT_DATA_KEY and the caller must truncate the last block.
4497 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4498 struct btrfs_root *root,
4499 struct btrfs_inode *inode,
4500 u64 new_size, u32 min_type,
4503 struct btrfs_fs_info *fs_info = root->fs_info;
4504 struct btrfs_path *path;
4505 struct extent_buffer *leaf;
4506 struct btrfs_file_extent_item *fi;
4507 struct btrfs_key key;
4508 struct btrfs_key found_key;
4509 u64 extent_start = 0;
4510 u64 extent_num_bytes = 0;
4511 u64 extent_offset = 0;
4513 u64 last_size = new_size;
4514 u32 found_type = (u8)-1;
4517 int pending_del_nr = 0;
4518 int pending_del_slot = 0;
4519 int extent_type = -1;
4521 u64 ino = btrfs_ino(inode);
4522 u64 bytes_deleted = 0;
4523 bool be_nice = false;
4524 bool should_throttle = false;
4525 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
4526 struct extent_state *cached_state = NULL;
4528 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4531 * For non-free space inodes and non-shareable roots, we want to back
4532 * off from time to time. This means all inodes in subvolume roots,
4533 * reloc roots, and data reloc roots.
4535 if (!btrfs_is_free_space_inode(inode) &&
4536 test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
4539 path = btrfs_alloc_path();
4542 path->reada = READA_BACK;
4544 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4545 lock_extent_bits(&inode->io_tree, lock_start, (u64)-1,
4549 * We want to drop from the next block forward in case this
4550 * new size is not block aligned since we will be keeping the
4551 * last block of the extent just the way it is.
4553 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4554 fs_info->sectorsize),
4559 * This function is also used to drop the items in the log tree before
4560 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4561 * it is used to drop the logged items. So we shouldn't kill the delayed
4564 if (min_type == 0 && root == inode->root)
4565 btrfs_kill_delayed_inode_items(inode);
4568 key.offset = (u64)-1;
4573 * with a 16K leaf size and 128MB extents, you can actually queue
4574 * up a huge file in a single leaf. Most of the time that
4575 * bytes_deleted is > 0, it will be huge by the time we get here
4577 if (be_nice && bytes_deleted > SZ_32M &&
4578 btrfs_should_end_transaction(trans)) {
4583 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4589 /* there are no items in the tree for us to truncate, we're
4592 if (path->slots[0] == 0)
4598 u64 clear_start = 0, clear_len = 0;
4601 leaf = path->nodes[0];
4602 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4603 found_type = found_key.type;
4605 if (found_key.objectid != ino)
4608 if (found_type < min_type)
4611 item_end = found_key.offset;
4612 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4613 fi = btrfs_item_ptr(leaf, path->slots[0],
4614 struct btrfs_file_extent_item);
4615 extent_type = btrfs_file_extent_type(leaf, fi);
4616 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4618 btrfs_file_extent_num_bytes(leaf, fi);
4620 trace_btrfs_truncate_show_fi_regular(
4621 inode, leaf, fi, found_key.offset);
4622 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4623 item_end += btrfs_file_extent_ram_bytes(leaf,
4626 trace_btrfs_truncate_show_fi_inline(
4627 inode, leaf, fi, path->slots[0],
4632 if (found_type > min_type) {
4635 if (item_end < new_size)
4637 if (found_key.offset >= new_size)
4643 /* FIXME, shrink the extent if the ref count is only 1 */
4644 if (found_type != BTRFS_EXTENT_DATA_KEY)
4647 if (extents_found != NULL)
4650 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4653 clear_start = found_key.offset;
4654 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4656 u64 orig_num_bytes =
4657 btrfs_file_extent_num_bytes(leaf, fi);
4658 extent_num_bytes = ALIGN(new_size -
4660 fs_info->sectorsize);
4661 clear_start = ALIGN(new_size, fs_info->sectorsize);
4662 btrfs_set_file_extent_num_bytes(leaf, fi,
4664 num_dec = (orig_num_bytes -
4666 if (test_bit(BTRFS_ROOT_SHAREABLE,
4669 inode_sub_bytes(&inode->vfs_inode,
4671 btrfs_mark_buffer_dirty(leaf);
4674 btrfs_file_extent_disk_num_bytes(leaf,
4676 extent_offset = found_key.offset -
4677 btrfs_file_extent_offset(leaf, fi);
4679 /* FIXME blocksize != 4096 */
4680 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4681 if (extent_start != 0) {
4683 if (test_bit(BTRFS_ROOT_SHAREABLE,
4685 inode_sub_bytes(&inode->vfs_inode,
4689 clear_len = num_dec;
4690 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4692 * we can't truncate inline items that have had
4696 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4697 btrfs_file_extent_other_encoding(leaf, fi) == 0 &&
4698 btrfs_file_extent_compression(leaf, fi) == 0) {
4699 u32 size = (u32)(new_size - found_key.offset);
4701 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4702 size = btrfs_file_extent_calc_inline_size(size);
4703 btrfs_truncate_item(path, size, 1);
4704 } else if (!del_item) {
4706 * We have to bail so the last_size is set to
4707 * just before this extent.
4709 ret = NEED_TRUNCATE_BLOCK;
4713 * Inline extents are special, we just treat
4714 * them as a full sector worth in the file
4715 * extent tree just for simplicity sake.
4717 clear_len = fs_info->sectorsize;
4720 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
4721 inode_sub_bytes(&inode->vfs_inode,
4722 item_end + 1 - new_size);
4726 * We use btrfs_truncate_inode_items() to clean up log trees for
4727 * multiple fsyncs, and in this case we don't want to clear the
4728 * file extent range because it's just the log.
4730 if (root == inode->root) {
4731 ret = btrfs_inode_clear_file_extent_range(inode,
4732 clear_start, clear_len);
4734 btrfs_abort_transaction(trans, ret);
4740 last_size = found_key.offset;
4742 last_size = new_size;
4744 if (!pending_del_nr) {
4745 /* no pending yet, add ourselves */
4746 pending_del_slot = path->slots[0];
4748 } else if (pending_del_nr &&
4749 path->slots[0] + 1 == pending_del_slot) {
4750 /* hop on the pending chunk */
4752 pending_del_slot = path->slots[0];
4759 should_throttle = false;
4762 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4763 struct btrfs_ref ref = { 0 };
4765 bytes_deleted += extent_num_bytes;
4767 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF,
4768 extent_start, extent_num_bytes, 0);
4769 ref.real_root = root->root_key.objectid;
4770 btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
4771 ino, extent_offset);
4772 ret = btrfs_free_extent(trans, &ref);
4774 btrfs_abort_transaction(trans, ret);
4778 if (btrfs_should_throttle_delayed_refs(trans))
4779 should_throttle = true;
4783 if (found_type == BTRFS_INODE_ITEM_KEY)
4786 if (path->slots[0] == 0 ||
4787 path->slots[0] != pending_del_slot ||
4789 if (pending_del_nr) {
4790 ret = btrfs_del_items(trans, root, path,
4794 btrfs_abort_transaction(trans, ret);
4799 btrfs_release_path(path);
4802 * We can generate a lot of delayed refs, so we need to
4803 * throttle every once and a while and make sure we're
4804 * adding enough space to keep up with the work we are
4805 * generating. Since we hold a transaction here we
4806 * can't flush, and we don't want to FLUSH_LIMIT because
4807 * we could have generated too many delayed refs to
4808 * actually allocate, so just bail if we're short and
4809 * let the normal reservation dance happen higher up.
4811 if (should_throttle) {
4812 ret = btrfs_delayed_refs_rsv_refill(fs_info,
4813 BTRFS_RESERVE_NO_FLUSH);
4825 if (ret >= 0 && pending_del_nr) {
4828 err = btrfs_del_items(trans, root, path, pending_del_slot,
4831 btrfs_abort_transaction(trans, err);
4835 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4836 ASSERT(last_size >= new_size);
4837 if (!ret && last_size > new_size)
4838 last_size = new_size;
4839 btrfs_inode_safe_disk_i_size_write(inode, last_size);
4840 unlock_extent_cached(&inode->io_tree, lock_start, (u64)-1,
4844 btrfs_free_path(path);
4849 * btrfs_truncate_block - read, zero a chunk and write a block
4850 * @inode - inode that we're zeroing
4851 * @from - the offset to start zeroing
4852 * @len - the length to zero, 0 to zero the entire range respective to the
4854 * @front - zero up to the offset instead of from the offset on
4856 * This will find the block for the "from" offset and cow the block and zero the
4857 * part we want to zero. This is used with truncate and hole punching.
4859 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4862 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4863 struct address_space *mapping = inode->vfs_inode.i_mapping;
4864 struct extent_io_tree *io_tree = &inode->io_tree;
4865 struct btrfs_ordered_extent *ordered;
4866 struct extent_state *cached_state = NULL;
4867 struct extent_changeset *data_reserved = NULL;
4868 bool only_release_metadata = false;
4869 u32 blocksize = fs_info->sectorsize;
4870 pgoff_t index = from >> PAGE_SHIFT;
4871 unsigned offset = from & (blocksize - 1);
4873 gfp_t mask = btrfs_alloc_write_mask(mapping);
4874 size_t write_bytes = blocksize;
4879 if (IS_ALIGNED(offset, blocksize) &&
4880 (!len || IS_ALIGNED(len, blocksize)))
4883 block_start = round_down(from, blocksize);
4884 block_end = block_start + blocksize - 1;
4886 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4889 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes) > 0) {
4890 /* For nocow case, no need to reserve data space */
4891 only_release_metadata = true;
4896 ret = btrfs_delalloc_reserve_metadata(inode, blocksize);
4898 if (!only_release_metadata)
4899 btrfs_free_reserved_data_space(inode, data_reserved,
4900 block_start, blocksize);
4904 page = find_or_create_page(mapping, index, mask);
4906 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4908 btrfs_delalloc_release_extents(inode, blocksize);
4912 ret = set_page_extent_mapped(page);
4916 if (!PageUptodate(page)) {
4917 ret = btrfs_readpage(NULL, page);
4919 if (page->mapping != mapping) {
4924 if (!PageUptodate(page)) {
4929 wait_on_page_writeback(page);
4931 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4933 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4935 unlock_extent_cached(io_tree, block_start, block_end,
4939 btrfs_start_ordered_extent(ordered, 1);
4940 btrfs_put_ordered_extent(ordered);
4944 clear_extent_bit(&inode->io_tree, block_start, block_end,
4945 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4946 0, 0, &cached_state);
4948 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4951 unlock_extent_cached(io_tree, block_start, block_end,
4956 if (offset != blocksize) {
4958 len = blocksize - offset;
4960 memzero_page(page, (block_start - page_offset(page)),
4963 memzero_page(page, (block_start - page_offset(page)) + offset,
4965 flush_dcache_page(page);
4967 ClearPageChecked(page);
4968 btrfs_page_set_dirty(fs_info, page, block_start, block_end + 1 - block_start);
4969 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4971 if (only_release_metadata)
4972 set_extent_bit(&inode->io_tree, block_start, block_end,
4973 EXTENT_NORESERVE, 0, NULL, NULL, GFP_NOFS, NULL);
4977 if (only_release_metadata)
4978 btrfs_delalloc_release_metadata(inode, blocksize, true);
4980 btrfs_delalloc_release_space(inode, data_reserved,
4981 block_start, blocksize, true);
4983 btrfs_delalloc_release_extents(inode, blocksize);
4987 if (only_release_metadata)
4988 btrfs_check_nocow_unlock(inode);
4989 extent_changeset_free(data_reserved);
4993 static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode,
4994 u64 offset, u64 len)
4996 struct btrfs_fs_info *fs_info = root->fs_info;
4997 struct btrfs_trans_handle *trans;
4998 struct btrfs_drop_extents_args drop_args = { 0 };
5002 * Still need to make sure the inode looks like it's been updated so
5003 * that any holes get logged if we fsync.
5005 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
5006 inode->last_trans = fs_info->generation;
5007 inode->last_sub_trans = root->log_transid;
5008 inode->last_log_commit = root->last_log_commit;
5013 * 1 - for the one we're dropping
5014 * 1 - for the one we're adding
5015 * 1 - for updating the inode.
5017 trans = btrfs_start_transaction(root, 3);
5019 return PTR_ERR(trans);
5021 drop_args.start = offset;
5022 drop_args.end = offset + len;
5023 drop_args.drop_cache = true;
5025 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
5027 btrfs_abort_transaction(trans, ret);
5028 btrfs_end_transaction(trans);
5032 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode),
5033 offset, 0, 0, len, 0, len, 0, 0, 0);
5035 btrfs_abort_transaction(trans, ret);
5037 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
5038 btrfs_update_inode(trans, root, inode);
5040 btrfs_end_transaction(trans);
5045 * This function puts in dummy file extents for the area we're creating a hole
5046 * for. So if we are truncating this file to a larger size we need to insert
5047 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
5048 * the range between oldsize and size
5050 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
5052 struct btrfs_root *root = inode->root;
5053 struct btrfs_fs_info *fs_info = root->fs_info;
5054 struct extent_io_tree *io_tree = &inode->io_tree;
5055 struct extent_map *em = NULL;
5056 struct extent_state *cached_state = NULL;
5057 struct extent_map_tree *em_tree = &inode->extent_tree;
5058 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
5059 u64 block_end = ALIGN(size, fs_info->sectorsize);
5066 * If our size started in the middle of a block we need to zero out the
5067 * rest of the block before we expand the i_size, otherwise we could
5068 * expose stale data.
5070 err = btrfs_truncate_block(inode, oldsize, 0, 0);
5074 if (size <= hole_start)
5077 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
5079 cur_offset = hole_start;
5081 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
5082 block_end - cur_offset);
5088 last_byte = min(extent_map_end(em), block_end);
5089 last_byte = ALIGN(last_byte, fs_info->sectorsize);
5090 hole_size = last_byte - cur_offset;
5092 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5093 struct extent_map *hole_em;
5095 err = maybe_insert_hole(root, inode, cur_offset,
5100 err = btrfs_inode_set_file_extent_range(inode,
5101 cur_offset, hole_size);
5105 btrfs_drop_extent_cache(inode, cur_offset,
5106 cur_offset + hole_size - 1, 0);
5107 hole_em = alloc_extent_map();
5109 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5110 &inode->runtime_flags);
5113 hole_em->start = cur_offset;
5114 hole_em->len = hole_size;
5115 hole_em->orig_start = cur_offset;
5117 hole_em->block_start = EXTENT_MAP_HOLE;
5118 hole_em->block_len = 0;
5119 hole_em->orig_block_len = 0;
5120 hole_em->ram_bytes = hole_size;
5121 hole_em->compress_type = BTRFS_COMPRESS_NONE;
5122 hole_em->generation = fs_info->generation;
5125 write_lock(&em_tree->lock);
5126 err = add_extent_mapping(em_tree, hole_em, 1);
5127 write_unlock(&em_tree->lock);
5130 btrfs_drop_extent_cache(inode, cur_offset,
5134 free_extent_map(hole_em);
5136 err = btrfs_inode_set_file_extent_range(inode,
5137 cur_offset, hole_size);
5142 free_extent_map(em);
5144 cur_offset = last_byte;
5145 if (cur_offset >= block_end)
5148 free_extent_map(em);
5149 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
5153 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5155 struct btrfs_root *root = BTRFS_I(inode)->root;
5156 struct btrfs_trans_handle *trans;
5157 loff_t oldsize = i_size_read(inode);
5158 loff_t newsize = attr->ia_size;
5159 int mask = attr->ia_valid;
5163 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5164 * special case where we need to update the times despite not having
5165 * these flags set. For all other operations the VFS set these flags
5166 * explicitly if it wants a timestamp update.
5168 if (newsize != oldsize) {
5169 inode_inc_iversion(inode);
5170 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5171 inode->i_ctime = inode->i_mtime =
5172 current_time(inode);
5175 if (newsize > oldsize) {
5177 * Don't do an expanding truncate while snapshotting is ongoing.
5178 * This is to ensure the snapshot captures a fully consistent
5179 * state of this file - if the snapshot captures this expanding
5180 * truncation, it must capture all writes that happened before
5183 btrfs_drew_write_lock(&root->snapshot_lock);
5184 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
5186 btrfs_drew_write_unlock(&root->snapshot_lock);
5190 trans = btrfs_start_transaction(root, 1);
5191 if (IS_ERR(trans)) {
5192 btrfs_drew_write_unlock(&root->snapshot_lock);
5193 return PTR_ERR(trans);
5196 i_size_write(inode, newsize);
5197 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
5198 pagecache_isize_extended(inode, oldsize, newsize);
5199 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5200 btrfs_drew_write_unlock(&root->snapshot_lock);
5201 btrfs_end_transaction(trans);
5203 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5205 if (btrfs_is_zoned(fs_info)) {
5206 ret = btrfs_wait_ordered_range(inode,
5207 ALIGN(newsize, fs_info->sectorsize),
5214 * We're truncating a file that used to have good data down to
5215 * zero. Make sure any new writes to the file get on disk
5219 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5220 &BTRFS_I(inode)->runtime_flags);
5222 truncate_setsize(inode, newsize);
5224 inode_dio_wait(inode);
5226 ret = btrfs_truncate(inode, newsize == oldsize);
5227 if (ret && inode->i_nlink) {
5231 * Truncate failed, so fix up the in-memory size. We
5232 * adjusted disk_i_size down as we removed extents, so
5233 * wait for disk_i_size to be stable and then update the
5234 * in-memory size to match.
5236 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5239 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5246 static int btrfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
5249 struct inode *inode = d_inode(dentry);
5250 struct btrfs_root *root = BTRFS_I(inode)->root;
5253 if (btrfs_root_readonly(root))
5256 err = setattr_prepare(&init_user_ns, dentry, attr);
5260 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5261 err = btrfs_setsize(inode, attr);
5266 if (attr->ia_valid) {
5267 setattr_copy(&init_user_ns, inode, attr);
5268 inode_inc_iversion(inode);
5269 err = btrfs_dirty_inode(inode);
5271 if (!err && attr->ia_valid & ATTR_MODE)
5272 err = posix_acl_chmod(&init_user_ns, inode,
5280 * While truncating the inode pages during eviction, we get the VFS calling
5281 * btrfs_invalidatepage() against each page of the inode. This is slow because
5282 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5283 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5284 * extent_state structures over and over, wasting lots of time.
5286 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5287 * those expensive operations on a per page basis and do only the ordered io
5288 * finishing, while we release here the extent_map and extent_state structures,
5289 * without the excessive merging and splitting.
5291 static void evict_inode_truncate_pages(struct inode *inode)
5293 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5294 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5295 struct rb_node *node;
5297 ASSERT(inode->i_state & I_FREEING);
5298 truncate_inode_pages_final(&inode->i_data);
5300 write_lock(&map_tree->lock);
5301 while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) {
5302 struct extent_map *em;
5304 node = rb_first_cached(&map_tree->map);
5305 em = rb_entry(node, struct extent_map, rb_node);
5306 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5307 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5308 remove_extent_mapping(map_tree, em);
5309 free_extent_map(em);
5310 if (need_resched()) {
5311 write_unlock(&map_tree->lock);
5313 write_lock(&map_tree->lock);
5316 write_unlock(&map_tree->lock);
5319 * Keep looping until we have no more ranges in the io tree.
5320 * We can have ongoing bios started by readahead that have
5321 * their endio callback (extent_io.c:end_bio_extent_readpage)
5322 * still in progress (unlocked the pages in the bio but did not yet
5323 * unlocked the ranges in the io tree). Therefore this means some
5324 * ranges can still be locked and eviction started because before
5325 * submitting those bios, which are executed by a separate task (work
5326 * queue kthread), inode references (inode->i_count) were not taken
5327 * (which would be dropped in the end io callback of each bio).
5328 * Therefore here we effectively end up waiting for those bios and
5329 * anyone else holding locked ranges without having bumped the inode's
5330 * reference count - if we don't do it, when they access the inode's
5331 * io_tree to unlock a range it may be too late, leading to an
5332 * use-after-free issue.
5334 spin_lock(&io_tree->lock);
5335 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5336 struct extent_state *state;
5337 struct extent_state *cached_state = NULL;
5340 unsigned state_flags;
5342 node = rb_first(&io_tree->state);
5343 state = rb_entry(node, struct extent_state, rb_node);
5344 start = state->start;
5346 state_flags = state->state;
5347 spin_unlock(&io_tree->lock);
5349 lock_extent_bits(io_tree, start, end, &cached_state);
5352 * If still has DELALLOC flag, the extent didn't reach disk,
5353 * and its reserved space won't be freed by delayed_ref.
5354 * So we need to free its reserved space here.
5355 * (Refer to comment in btrfs_invalidatepage, case 2)
5357 * Note, end is the bytenr of last byte, so we need + 1 here.
5359 if (state_flags & EXTENT_DELALLOC)
5360 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5363 clear_extent_bit(io_tree, start, end,
5364 EXTENT_LOCKED | EXTENT_DELALLOC |
5365 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
5369 spin_lock(&io_tree->lock);
5371 spin_unlock(&io_tree->lock);
5374 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5375 struct btrfs_block_rsv *rsv)
5377 struct btrfs_fs_info *fs_info = root->fs_info;
5378 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5379 struct btrfs_trans_handle *trans;
5380 u64 delayed_refs_extra = btrfs_calc_insert_metadata_size(fs_info, 1);
5384 * Eviction should be taking place at some place safe because of our
5385 * delayed iputs. However the normal flushing code will run delayed
5386 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5388 * We reserve the delayed_refs_extra here again because we can't use
5389 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5390 * above. We reserve our extra bit here because we generate a ton of
5391 * delayed refs activity by truncating.
5393 * If we cannot make our reservation we'll attempt to steal from the
5394 * global reserve, because we really want to be able to free up space.
5396 ret = btrfs_block_rsv_refill(root, rsv, rsv->size + delayed_refs_extra,
5397 BTRFS_RESERVE_FLUSH_EVICT);
5400 * Try to steal from the global reserve if there is space for
5403 if (btrfs_check_space_for_delayed_refs(fs_info) ||
5404 btrfs_block_rsv_migrate(global_rsv, rsv, rsv->size, 0)) {
5406 "could not allocate space for delete; will truncate on mount");
5407 return ERR_PTR(-ENOSPC);
5409 delayed_refs_extra = 0;
5412 trans = btrfs_join_transaction(root);
5416 if (delayed_refs_extra) {
5417 trans->block_rsv = &fs_info->trans_block_rsv;
5418 trans->bytes_reserved = delayed_refs_extra;
5419 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5420 delayed_refs_extra, 1);
5425 void btrfs_evict_inode(struct inode *inode)
5427 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5428 struct btrfs_trans_handle *trans;
5429 struct btrfs_root *root = BTRFS_I(inode)->root;
5430 struct btrfs_block_rsv *rsv;
5433 trace_btrfs_inode_evict(inode);
5440 evict_inode_truncate_pages(inode);
5442 if (inode->i_nlink &&
5443 ((btrfs_root_refs(&root->root_item) != 0 &&
5444 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5445 btrfs_is_free_space_inode(BTRFS_I(inode))))
5448 if (is_bad_inode(inode))
5451 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5453 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5456 if (inode->i_nlink > 0) {
5457 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5458 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5462 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5466 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5469 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5472 btrfs_i_size_write(BTRFS_I(inode), 0);
5475 trans = evict_refill_and_join(root, rsv);
5479 trans->block_rsv = rsv;
5481 ret = btrfs_truncate_inode_items(trans, root, BTRFS_I(inode),
5483 trans->block_rsv = &fs_info->trans_block_rsv;
5484 btrfs_end_transaction(trans);
5485 btrfs_btree_balance_dirty(fs_info);
5486 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5493 * Errors here aren't a big deal, it just means we leave orphan items in
5494 * the tree. They will be cleaned up on the next mount. If the inode
5495 * number gets reused, cleanup deletes the orphan item without doing
5496 * anything, and unlink reuses the existing orphan item.
5498 * If it turns out that we are dropping too many of these, we might want
5499 * to add a mechanism for retrying these after a commit.
5501 trans = evict_refill_and_join(root, rsv);
5502 if (!IS_ERR(trans)) {
5503 trans->block_rsv = rsv;
5504 btrfs_orphan_del(trans, BTRFS_I(inode));
5505 trans->block_rsv = &fs_info->trans_block_rsv;
5506 btrfs_end_transaction(trans);
5510 btrfs_free_block_rsv(fs_info, rsv);
5513 * If we didn't successfully delete, the orphan item will still be in
5514 * the tree and we'll retry on the next mount. Again, we might also want
5515 * to retry these periodically in the future.
5517 btrfs_remove_delayed_node(BTRFS_I(inode));
5522 * Return the key found in the dir entry in the location pointer, fill @type
5523 * with BTRFS_FT_*, and return 0.
5525 * If no dir entries were found, returns -ENOENT.
5526 * If found a corrupted location in dir entry, returns -EUCLEAN.
5528 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5529 struct btrfs_key *location, u8 *type)
5531 const char *name = dentry->d_name.name;
5532 int namelen = dentry->d_name.len;
5533 struct btrfs_dir_item *di;
5534 struct btrfs_path *path;
5535 struct btrfs_root *root = BTRFS_I(dir)->root;
5538 path = btrfs_alloc_path();
5542 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5544 if (IS_ERR_OR_NULL(di)) {
5545 ret = di ? PTR_ERR(di) : -ENOENT;
5549 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5550 if (location->type != BTRFS_INODE_ITEM_KEY &&
5551 location->type != BTRFS_ROOT_ITEM_KEY) {
5553 btrfs_warn(root->fs_info,
5554 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5555 __func__, name, btrfs_ino(BTRFS_I(dir)),
5556 location->objectid, location->type, location->offset);
5559 *type = btrfs_dir_type(path->nodes[0], di);
5561 btrfs_free_path(path);
5566 * when we hit a tree root in a directory, the btrfs part of the inode
5567 * needs to be changed to reflect the root directory of the tree root. This
5568 * is kind of like crossing a mount point.
5570 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5572 struct dentry *dentry,
5573 struct btrfs_key *location,
5574 struct btrfs_root **sub_root)
5576 struct btrfs_path *path;
5577 struct btrfs_root *new_root;
5578 struct btrfs_root_ref *ref;
5579 struct extent_buffer *leaf;
5580 struct btrfs_key key;
5584 path = btrfs_alloc_path();
5591 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5592 key.type = BTRFS_ROOT_REF_KEY;
5593 key.offset = location->objectid;
5595 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5602 leaf = path->nodes[0];
5603 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5604 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5605 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5608 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5609 (unsigned long)(ref + 1),
5610 dentry->d_name.len);
5614 btrfs_release_path(path);
5616 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5617 if (IS_ERR(new_root)) {
5618 err = PTR_ERR(new_root);
5622 *sub_root = new_root;
5623 location->objectid = btrfs_root_dirid(&new_root->root_item);
5624 location->type = BTRFS_INODE_ITEM_KEY;
5625 location->offset = 0;
5628 btrfs_free_path(path);
5632 static void inode_tree_add(struct inode *inode)
5634 struct btrfs_root *root = BTRFS_I(inode)->root;
5635 struct btrfs_inode *entry;
5637 struct rb_node *parent;
5638 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5639 u64 ino = btrfs_ino(BTRFS_I(inode));
5641 if (inode_unhashed(inode))
5644 spin_lock(&root->inode_lock);
5645 p = &root->inode_tree.rb_node;
5648 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5650 if (ino < btrfs_ino(entry))
5651 p = &parent->rb_left;
5652 else if (ino > btrfs_ino(entry))
5653 p = &parent->rb_right;
5655 WARN_ON(!(entry->vfs_inode.i_state &
5656 (I_WILL_FREE | I_FREEING)));
5657 rb_replace_node(parent, new, &root->inode_tree);
5658 RB_CLEAR_NODE(parent);
5659 spin_unlock(&root->inode_lock);
5663 rb_link_node(new, parent, p);
5664 rb_insert_color(new, &root->inode_tree);
5665 spin_unlock(&root->inode_lock);
5668 static void inode_tree_del(struct btrfs_inode *inode)
5670 struct btrfs_root *root = inode->root;
5673 spin_lock(&root->inode_lock);
5674 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5675 rb_erase(&inode->rb_node, &root->inode_tree);
5676 RB_CLEAR_NODE(&inode->rb_node);
5677 empty = RB_EMPTY_ROOT(&root->inode_tree);
5679 spin_unlock(&root->inode_lock);
5681 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5682 spin_lock(&root->inode_lock);
5683 empty = RB_EMPTY_ROOT(&root->inode_tree);
5684 spin_unlock(&root->inode_lock);
5686 btrfs_add_dead_root(root);
5691 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5693 struct btrfs_iget_args *args = p;
5695 inode->i_ino = args->ino;
5696 BTRFS_I(inode)->location.objectid = args->ino;
5697 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5698 BTRFS_I(inode)->location.offset = 0;
5699 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5700 BUG_ON(args->root && !BTRFS_I(inode)->root);
5704 static int btrfs_find_actor(struct inode *inode, void *opaque)
5706 struct btrfs_iget_args *args = opaque;
5708 return args->ino == BTRFS_I(inode)->location.objectid &&
5709 args->root == BTRFS_I(inode)->root;
5712 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5713 struct btrfs_root *root)
5715 struct inode *inode;
5716 struct btrfs_iget_args args;
5717 unsigned long hashval = btrfs_inode_hash(ino, root);
5722 inode = iget5_locked(s, hashval, btrfs_find_actor,
5723 btrfs_init_locked_inode,
5729 * Get an inode object given its inode number and corresponding root.
5730 * Path can be preallocated to prevent recursing back to iget through
5731 * allocator. NULL is also valid but may require an additional allocation
5734 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5735 struct btrfs_root *root, struct btrfs_path *path)
5737 struct inode *inode;
5739 inode = btrfs_iget_locked(s, ino, root);
5741 return ERR_PTR(-ENOMEM);
5743 if (inode->i_state & I_NEW) {
5746 ret = btrfs_read_locked_inode(inode, path);
5748 inode_tree_add(inode);
5749 unlock_new_inode(inode);
5753 * ret > 0 can come from btrfs_search_slot called by
5754 * btrfs_read_locked_inode, this means the inode item
5759 inode = ERR_PTR(ret);
5766 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5768 return btrfs_iget_path(s, ino, root, NULL);
5771 static struct inode *new_simple_dir(struct super_block *s,
5772 struct btrfs_key *key,
5773 struct btrfs_root *root)
5775 struct inode *inode = new_inode(s);
5778 return ERR_PTR(-ENOMEM);
5780 BTRFS_I(inode)->root = btrfs_grab_root(root);
5781 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5782 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5784 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5786 * We only need lookup, the rest is read-only and there's no inode
5787 * associated with the dentry
5789 inode->i_op = &simple_dir_inode_operations;
5790 inode->i_opflags &= ~IOP_XATTR;
5791 inode->i_fop = &simple_dir_operations;
5792 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5793 inode->i_mtime = current_time(inode);
5794 inode->i_atime = inode->i_mtime;
5795 inode->i_ctime = inode->i_mtime;
5796 BTRFS_I(inode)->i_otime = inode->i_mtime;
5801 static inline u8 btrfs_inode_type(struct inode *inode)
5804 * Compile-time asserts that generic FT_* types still match
5807 BUILD_BUG_ON(BTRFS_FT_UNKNOWN != FT_UNKNOWN);
5808 BUILD_BUG_ON(BTRFS_FT_REG_FILE != FT_REG_FILE);
5809 BUILD_BUG_ON(BTRFS_FT_DIR != FT_DIR);
5810 BUILD_BUG_ON(BTRFS_FT_CHRDEV != FT_CHRDEV);
5811 BUILD_BUG_ON(BTRFS_FT_BLKDEV != FT_BLKDEV);
5812 BUILD_BUG_ON(BTRFS_FT_FIFO != FT_FIFO);
5813 BUILD_BUG_ON(BTRFS_FT_SOCK != FT_SOCK);
5814 BUILD_BUG_ON(BTRFS_FT_SYMLINK != FT_SYMLINK);
5816 return fs_umode_to_ftype(inode->i_mode);
5819 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5821 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5822 struct inode *inode;
5823 struct btrfs_root *root = BTRFS_I(dir)->root;
5824 struct btrfs_root *sub_root = root;
5825 struct btrfs_key location;
5829 if (dentry->d_name.len > BTRFS_NAME_LEN)
5830 return ERR_PTR(-ENAMETOOLONG);
5832 ret = btrfs_inode_by_name(dir, dentry, &location, &di_type);
5834 return ERR_PTR(ret);
5836 if (location.type == BTRFS_INODE_ITEM_KEY) {
5837 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5841 /* Do extra check against inode mode with di_type */
5842 if (btrfs_inode_type(inode) != di_type) {
5844 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5845 inode->i_mode, btrfs_inode_type(inode),
5848 return ERR_PTR(-EUCLEAN);
5853 ret = fixup_tree_root_location(fs_info, dir, dentry,
5854 &location, &sub_root);
5857 inode = ERR_PTR(ret);
5859 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5861 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5863 if (root != sub_root)
5864 btrfs_put_root(sub_root);
5866 if (!IS_ERR(inode) && root != sub_root) {
5867 down_read(&fs_info->cleanup_work_sem);
5868 if (!sb_rdonly(inode->i_sb))
5869 ret = btrfs_orphan_cleanup(sub_root);
5870 up_read(&fs_info->cleanup_work_sem);
5873 inode = ERR_PTR(ret);
5880 static int btrfs_dentry_delete(const struct dentry *dentry)
5882 struct btrfs_root *root;
5883 struct inode *inode = d_inode(dentry);
5885 if (!inode && !IS_ROOT(dentry))
5886 inode = d_inode(dentry->d_parent);
5889 root = BTRFS_I(inode)->root;
5890 if (btrfs_root_refs(&root->root_item) == 0)
5893 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5899 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5902 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5904 if (inode == ERR_PTR(-ENOENT))
5906 return d_splice_alias(inode, dentry);
5910 * All this infrastructure exists because dir_emit can fault, and we are holding
5911 * the tree lock when doing readdir. For now just allocate a buffer and copy
5912 * our information into that, and then dir_emit from the buffer. This is
5913 * similar to what NFS does, only we don't keep the buffer around in pagecache
5914 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5915 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5918 static int btrfs_opendir(struct inode *inode, struct file *file)
5920 struct btrfs_file_private *private;
5922 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5925 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5926 if (!private->filldir_buf) {
5930 file->private_data = private;
5941 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5944 struct dir_entry *entry = addr;
5945 char *name = (char *)(entry + 1);
5947 ctx->pos = get_unaligned(&entry->offset);
5948 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5949 get_unaligned(&entry->ino),
5950 get_unaligned(&entry->type)))
5952 addr += sizeof(struct dir_entry) +
5953 get_unaligned(&entry->name_len);
5959 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5961 struct inode *inode = file_inode(file);
5962 struct btrfs_root *root = BTRFS_I(inode)->root;
5963 struct btrfs_file_private *private = file->private_data;
5964 struct btrfs_dir_item *di;
5965 struct btrfs_key key;
5966 struct btrfs_key found_key;
5967 struct btrfs_path *path;
5969 struct list_head ins_list;
5970 struct list_head del_list;
5972 struct extent_buffer *leaf;
5979 struct btrfs_key location;
5981 if (!dir_emit_dots(file, ctx))
5984 path = btrfs_alloc_path();
5988 addr = private->filldir_buf;
5989 path->reada = READA_FORWARD;
5991 INIT_LIST_HEAD(&ins_list);
5992 INIT_LIST_HEAD(&del_list);
5993 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5996 key.type = BTRFS_DIR_INDEX_KEY;
5997 key.offset = ctx->pos;
5998 key.objectid = btrfs_ino(BTRFS_I(inode));
6000 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6005 struct dir_entry *entry;
6007 leaf = path->nodes[0];
6008 slot = path->slots[0];
6009 if (slot >= btrfs_header_nritems(leaf)) {
6010 ret = btrfs_next_leaf(root, path);
6018 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6020 if (found_key.objectid != key.objectid)
6022 if (found_key.type != BTRFS_DIR_INDEX_KEY)
6024 if (found_key.offset < ctx->pos)
6026 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
6028 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
6029 name_len = btrfs_dir_name_len(leaf, di);
6030 if ((total_len + sizeof(struct dir_entry) + name_len) >=
6032 btrfs_release_path(path);
6033 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6036 addr = private->filldir_buf;
6043 put_unaligned(name_len, &entry->name_len);
6044 name_ptr = (char *)(entry + 1);
6045 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
6047 put_unaligned(fs_ftype_to_dtype(btrfs_dir_type(leaf, di)),
6049 btrfs_dir_item_key_to_cpu(leaf, di, &location);
6050 put_unaligned(location.objectid, &entry->ino);
6051 put_unaligned(found_key.offset, &entry->offset);
6053 addr += sizeof(struct dir_entry) + name_len;
6054 total_len += sizeof(struct dir_entry) + name_len;
6058 btrfs_release_path(path);
6060 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6064 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
6069 * Stop new entries from being returned after we return the last
6072 * New directory entries are assigned a strictly increasing
6073 * offset. This means that new entries created during readdir
6074 * are *guaranteed* to be seen in the future by that readdir.
6075 * This has broken buggy programs which operate on names as
6076 * they're returned by readdir. Until we re-use freed offsets
6077 * we have this hack to stop new entries from being returned
6078 * under the assumption that they'll never reach this huge
6081 * This is being careful not to overflow 32bit loff_t unless the
6082 * last entry requires it because doing so has broken 32bit apps
6085 if (ctx->pos >= INT_MAX)
6086 ctx->pos = LLONG_MAX;
6093 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6094 btrfs_free_path(path);
6099 * This is somewhat expensive, updating the tree every time the
6100 * inode changes. But, it is most likely to find the inode in cache.
6101 * FIXME, needs more benchmarking...there are no reasons other than performance
6102 * to keep or drop this code.
6104 static int btrfs_dirty_inode(struct inode *inode)
6106 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6107 struct btrfs_root *root = BTRFS_I(inode)->root;
6108 struct btrfs_trans_handle *trans;
6111 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6114 trans = btrfs_join_transaction(root);
6116 return PTR_ERR(trans);
6118 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
6119 if (ret && (ret == -ENOSPC || ret == -EDQUOT)) {
6120 /* whoops, lets try again with the full transaction */
6121 btrfs_end_transaction(trans);
6122 trans = btrfs_start_transaction(root, 1);
6124 return PTR_ERR(trans);
6126 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
6128 btrfs_end_transaction(trans);
6129 if (BTRFS_I(inode)->delayed_node)
6130 btrfs_balance_delayed_items(fs_info);
6136 * This is a copy of file_update_time. We need this so we can return error on
6137 * ENOSPC for updating the inode in the case of file write and mmap writes.
6139 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
6142 struct btrfs_root *root = BTRFS_I(inode)->root;
6143 bool dirty = flags & ~S_VERSION;
6145 if (btrfs_root_readonly(root))
6148 if (flags & S_VERSION)
6149 dirty |= inode_maybe_inc_iversion(inode, dirty);
6150 if (flags & S_CTIME)
6151 inode->i_ctime = *now;
6152 if (flags & S_MTIME)
6153 inode->i_mtime = *now;
6154 if (flags & S_ATIME)
6155 inode->i_atime = *now;
6156 return dirty ? btrfs_dirty_inode(inode) : 0;
6160 * find the highest existing sequence number in a directory
6161 * and then set the in-memory index_cnt variable to reflect
6162 * free sequence numbers
6164 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6166 struct btrfs_root *root = inode->root;
6167 struct btrfs_key key, found_key;
6168 struct btrfs_path *path;
6169 struct extent_buffer *leaf;
6172 key.objectid = btrfs_ino(inode);
6173 key.type = BTRFS_DIR_INDEX_KEY;
6174 key.offset = (u64)-1;
6176 path = btrfs_alloc_path();
6180 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6183 /* FIXME: we should be able to handle this */
6189 * MAGIC NUMBER EXPLANATION:
6190 * since we search a directory based on f_pos we have to start at 2
6191 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6192 * else has to start at 2
6194 if (path->slots[0] == 0) {
6195 inode->index_cnt = 2;
6201 leaf = path->nodes[0];
6202 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6204 if (found_key.objectid != btrfs_ino(inode) ||
6205 found_key.type != BTRFS_DIR_INDEX_KEY) {
6206 inode->index_cnt = 2;
6210 inode->index_cnt = found_key.offset + 1;
6212 btrfs_free_path(path);
6217 * helper to find a free sequence number in a given directory. This current
6218 * code is very simple, later versions will do smarter things in the btree
6220 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6224 if (dir->index_cnt == (u64)-1) {
6225 ret = btrfs_inode_delayed_dir_index_count(dir);
6227 ret = btrfs_set_inode_index_count(dir);
6233 *index = dir->index_cnt;
6239 static int btrfs_insert_inode_locked(struct inode *inode)
6241 struct btrfs_iget_args args;
6243 args.ino = BTRFS_I(inode)->location.objectid;
6244 args.root = BTRFS_I(inode)->root;
6246 return insert_inode_locked4(inode,
6247 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6248 btrfs_find_actor, &args);
6252 * Inherit flags from the parent inode.
6254 * Currently only the compression flags and the cow flags are inherited.
6256 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6263 flags = BTRFS_I(dir)->flags;
6265 if (flags & BTRFS_INODE_NOCOMPRESS) {
6266 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6267 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6268 } else if (flags & BTRFS_INODE_COMPRESS) {
6269 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6270 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6273 if (flags & BTRFS_INODE_NODATACOW) {
6274 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6275 if (S_ISREG(inode->i_mode))
6276 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6279 btrfs_sync_inode_flags_to_i_flags(inode);
6282 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6283 struct btrfs_root *root,
6285 const char *name, int name_len,
6286 u64 ref_objectid, u64 objectid,
6287 umode_t mode, u64 *index)
6289 struct btrfs_fs_info *fs_info = root->fs_info;
6290 struct inode *inode;
6291 struct btrfs_inode_item *inode_item;
6292 struct btrfs_key *location;
6293 struct btrfs_path *path;
6294 struct btrfs_inode_ref *ref;
6295 struct btrfs_key key[2];
6297 int nitems = name ? 2 : 1;
6299 unsigned int nofs_flag;
6302 path = btrfs_alloc_path();
6304 return ERR_PTR(-ENOMEM);
6306 nofs_flag = memalloc_nofs_save();
6307 inode = new_inode(fs_info->sb);
6308 memalloc_nofs_restore(nofs_flag);
6310 btrfs_free_path(path);
6311 return ERR_PTR(-ENOMEM);
6315 * O_TMPFILE, set link count to 0, so that after this point,
6316 * we fill in an inode item with the correct link count.
6319 set_nlink(inode, 0);
6322 * we have to initialize this early, so we can reclaim the inode
6323 * number if we fail afterwards in this function.
6325 inode->i_ino = objectid;
6328 trace_btrfs_inode_request(dir);
6330 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6332 btrfs_free_path(path);
6334 return ERR_PTR(ret);
6340 * index_cnt is ignored for everything but a dir,
6341 * btrfs_set_inode_index_count has an explanation for the magic
6344 BTRFS_I(inode)->index_cnt = 2;
6345 BTRFS_I(inode)->dir_index = *index;
6346 BTRFS_I(inode)->root = btrfs_grab_root(root);
6347 BTRFS_I(inode)->generation = trans->transid;
6348 inode->i_generation = BTRFS_I(inode)->generation;
6351 * We could have gotten an inode number from somebody who was fsynced
6352 * and then removed in this same transaction, so let's just set full
6353 * sync since it will be a full sync anyway and this will blow away the
6354 * old info in the log.
6356 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6358 key[0].objectid = objectid;
6359 key[0].type = BTRFS_INODE_ITEM_KEY;
6362 sizes[0] = sizeof(struct btrfs_inode_item);
6366 * Start new inodes with an inode_ref. This is slightly more
6367 * efficient for small numbers of hard links since they will
6368 * be packed into one item. Extended refs will kick in if we
6369 * add more hard links than can fit in the ref item.
6371 key[1].objectid = objectid;
6372 key[1].type = BTRFS_INODE_REF_KEY;
6373 key[1].offset = ref_objectid;
6375 sizes[1] = name_len + sizeof(*ref);
6378 location = &BTRFS_I(inode)->location;
6379 location->objectid = objectid;
6380 location->offset = 0;
6381 location->type = BTRFS_INODE_ITEM_KEY;
6383 ret = btrfs_insert_inode_locked(inode);
6389 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6393 inode_init_owner(&init_user_ns, inode, dir, mode);
6394 inode_set_bytes(inode, 0);
6396 inode->i_mtime = current_time(inode);
6397 inode->i_atime = inode->i_mtime;
6398 inode->i_ctime = inode->i_mtime;
6399 BTRFS_I(inode)->i_otime = inode->i_mtime;
6401 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6402 struct btrfs_inode_item);
6403 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6404 sizeof(*inode_item));
6405 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6408 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6409 struct btrfs_inode_ref);
6410 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6411 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6412 ptr = (unsigned long)(ref + 1);
6413 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6416 btrfs_mark_buffer_dirty(path->nodes[0]);
6417 btrfs_free_path(path);
6419 btrfs_inherit_iflags(inode, dir);
6421 if (S_ISREG(mode)) {
6422 if (btrfs_test_opt(fs_info, NODATASUM))
6423 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6424 if (btrfs_test_opt(fs_info, NODATACOW))
6425 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6426 BTRFS_INODE_NODATASUM;
6429 inode_tree_add(inode);
6431 trace_btrfs_inode_new(inode);
6432 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6434 btrfs_update_root_times(trans, root);
6436 ret = btrfs_inode_inherit_props(trans, inode, dir);
6439 "error inheriting props for ino %llu (root %llu): %d",
6440 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6445 discard_new_inode(inode);
6448 BTRFS_I(dir)->index_cnt--;
6449 btrfs_free_path(path);
6450 return ERR_PTR(ret);
6454 * utility function to add 'inode' into 'parent_inode' with
6455 * a give name and a given sequence number.
6456 * if 'add_backref' is true, also insert a backref from the
6457 * inode to the parent directory.
6459 int btrfs_add_link(struct btrfs_trans_handle *trans,
6460 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6461 const char *name, int name_len, int add_backref, u64 index)
6464 struct btrfs_key key;
6465 struct btrfs_root *root = parent_inode->root;
6466 u64 ino = btrfs_ino(inode);
6467 u64 parent_ino = btrfs_ino(parent_inode);
6469 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6470 memcpy(&key, &inode->root->root_key, sizeof(key));
6473 key.type = BTRFS_INODE_ITEM_KEY;
6477 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6478 ret = btrfs_add_root_ref(trans, key.objectid,
6479 root->root_key.objectid, parent_ino,
6480 index, name, name_len);
6481 } else if (add_backref) {
6482 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6486 /* Nothing to clean up yet */
6490 ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key,
6491 btrfs_inode_type(&inode->vfs_inode), index);
6492 if (ret == -EEXIST || ret == -EOVERFLOW)
6495 btrfs_abort_transaction(trans, ret);
6499 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6501 inode_inc_iversion(&parent_inode->vfs_inode);
6503 * If we are replaying a log tree, we do not want to update the mtime
6504 * and ctime of the parent directory with the current time, since the
6505 * log replay procedure is responsible for setting them to their correct
6506 * values (the ones it had when the fsync was done).
6508 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
6509 struct timespec64 now = current_time(&parent_inode->vfs_inode);
6511 parent_inode->vfs_inode.i_mtime = now;
6512 parent_inode->vfs_inode.i_ctime = now;
6514 ret = btrfs_update_inode(trans, root, parent_inode);
6516 btrfs_abort_transaction(trans, ret);
6520 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6523 err = btrfs_del_root_ref(trans, key.objectid,
6524 root->root_key.objectid, parent_ino,
6525 &local_index, name, name_len);
6527 btrfs_abort_transaction(trans, err);
6528 } else if (add_backref) {
6532 err = btrfs_del_inode_ref(trans, root, name, name_len,
6533 ino, parent_ino, &local_index);
6535 btrfs_abort_transaction(trans, err);
6538 /* Return the original error code */
6542 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6543 struct btrfs_inode *dir, struct dentry *dentry,
6544 struct btrfs_inode *inode, int backref, u64 index)
6546 int err = btrfs_add_link(trans, dir, inode,
6547 dentry->d_name.name, dentry->d_name.len,
6554 static int btrfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
6555 struct dentry *dentry, umode_t mode, dev_t rdev)
6557 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6558 struct btrfs_trans_handle *trans;
6559 struct btrfs_root *root = BTRFS_I(dir)->root;
6560 struct inode *inode = NULL;
6566 * 2 for inode item and ref
6568 * 1 for xattr if selinux is on
6570 trans = btrfs_start_transaction(root, 5);
6572 return PTR_ERR(trans);
6574 err = btrfs_get_free_objectid(root, &objectid);
6578 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6579 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6581 if (IS_ERR(inode)) {
6582 err = PTR_ERR(inode);
6588 * If the active LSM wants to access the inode during
6589 * d_instantiate it needs these. Smack checks to see
6590 * if the filesystem supports xattrs by looking at the
6593 inode->i_op = &btrfs_special_inode_operations;
6594 init_special_inode(inode, inode->i_mode, rdev);
6596 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6600 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6605 btrfs_update_inode(trans, root, BTRFS_I(inode));
6606 d_instantiate_new(dentry, inode);
6609 btrfs_end_transaction(trans);
6610 btrfs_btree_balance_dirty(fs_info);
6612 inode_dec_link_count(inode);
6613 discard_new_inode(inode);
6618 static int btrfs_create(struct user_namespace *mnt_userns, struct inode *dir,
6619 struct dentry *dentry, umode_t mode, bool excl)
6621 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6622 struct btrfs_trans_handle *trans;
6623 struct btrfs_root *root = BTRFS_I(dir)->root;
6624 struct inode *inode = NULL;
6630 * 2 for inode item and ref
6632 * 1 for xattr if selinux is on
6634 trans = btrfs_start_transaction(root, 5);
6636 return PTR_ERR(trans);
6638 err = btrfs_get_free_objectid(root, &objectid);
6642 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6643 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6645 if (IS_ERR(inode)) {
6646 err = PTR_ERR(inode);
6651 * If the active LSM wants to access the inode during
6652 * d_instantiate it needs these. Smack checks to see
6653 * if the filesystem supports xattrs by looking at the
6656 inode->i_fop = &btrfs_file_operations;
6657 inode->i_op = &btrfs_file_inode_operations;
6658 inode->i_mapping->a_ops = &btrfs_aops;
6660 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6664 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6668 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6673 d_instantiate_new(dentry, inode);
6676 btrfs_end_transaction(trans);
6678 inode_dec_link_count(inode);
6679 discard_new_inode(inode);
6681 btrfs_btree_balance_dirty(fs_info);
6685 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6686 struct dentry *dentry)
6688 struct btrfs_trans_handle *trans = NULL;
6689 struct btrfs_root *root = BTRFS_I(dir)->root;
6690 struct inode *inode = d_inode(old_dentry);
6691 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6696 /* do not allow sys_link's with other subvols of the same device */
6697 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6700 if (inode->i_nlink >= BTRFS_LINK_MAX)
6703 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6708 * 2 items for inode and inode ref
6709 * 2 items for dir items
6710 * 1 item for parent inode
6711 * 1 item for orphan item deletion if O_TMPFILE
6713 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6714 if (IS_ERR(trans)) {
6715 err = PTR_ERR(trans);
6720 /* There are several dir indexes for this inode, clear the cache. */
6721 BTRFS_I(inode)->dir_index = 0ULL;
6723 inode_inc_iversion(inode);
6724 inode->i_ctime = current_time(inode);
6726 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6728 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6734 struct dentry *parent = dentry->d_parent;
6736 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6739 if (inode->i_nlink == 1) {
6741 * If new hard link count is 1, it's a file created
6742 * with open(2) O_TMPFILE flag.
6744 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6748 d_instantiate(dentry, inode);
6749 btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent);
6754 btrfs_end_transaction(trans);
6756 inode_dec_link_count(inode);
6759 btrfs_btree_balance_dirty(fs_info);
6763 static int btrfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
6764 struct dentry *dentry, umode_t mode)
6766 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6767 struct inode *inode = NULL;
6768 struct btrfs_trans_handle *trans;
6769 struct btrfs_root *root = BTRFS_I(dir)->root;
6775 * 2 items for inode and ref
6776 * 2 items for dir items
6777 * 1 for xattr if selinux is on
6779 trans = btrfs_start_transaction(root, 5);
6781 return PTR_ERR(trans);
6783 err = btrfs_get_free_objectid(root, &objectid);
6787 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6788 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6789 S_IFDIR | mode, &index);
6790 if (IS_ERR(inode)) {
6791 err = PTR_ERR(inode);
6796 /* these must be set before we unlock the inode */
6797 inode->i_op = &btrfs_dir_inode_operations;
6798 inode->i_fop = &btrfs_dir_file_operations;
6800 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6804 btrfs_i_size_write(BTRFS_I(inode), 0);
6805 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6809 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6810 dentry->d_name.name,
6811 dentry->d_name.len, 0, index);
6815 d_instantiate_new(dentry, inode);
6818 btrfs_end_transaction(trans);
6820 inode_dec_link_count(inode);
6821 discard_new_inode(inode);
6823 btrfs_btree_balance_dirty(fs_info);
6827 static noinline int uncompress_inline(struct btrfs_path *path,
6829 size_t pg_offset, u64 extent_offset,
6830 struct btrfs_file_extent_item *item)
6833 struct extent_buffer *leaf = path->nodes[0];
6836 unsigned long inline_size;
6840 WARN_ON(pg_offset != 0);
6841 compress_type = btrfs_file_extent_compression(leaf, item);
6842 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6843 inline_size = btrfs_file_extent_inline_item_len(leaf,
6844 btrfs_item_nr(path->slots[0]));
6845 tmp = kmalloc(inline_size, GFP_NOFS);
6848 ptr = btrfs_file_extent_inline_start(item);
6850 read_extent_buffer(leaf, tmp, ptr, inline_size);
6852 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6853 ret = btrfs_decompress(compress_type, tmp, page,
6854 extent_offset, inline_size, max_size);
6857 * decompression code contains a memset to fill in any space between the end
6858 * of the uncompressed data and the end of max_size in case the decompressed
6859 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6860 * the end of an inline extent and the beginning of the next block, so we
6861 * cover that region here.
6864 if (max_size + pg_offset < PAGE_SIZE)
6865 memzero_page(page, pg_offset + max_size,
6866 PAGE_SIZE - max_size - pg_offset);
6872 * btrfs_get_extent - Lookup the first extent overlapping a range in a file.
6873 * @inode: file to search in
6874 * @page: page to read extent data into if the extent is inline
6875 * @pg_offset: offset into @page to copy to
6876 * @start: file offset
6877 * @len: length of range starting at @start
6879 * This returns the first &struct extent_map which overlaps with the given
6880 * range, reading it from the B-tree and caching it if necessary. Note that
6881 * there may be more extents which overlap the given range after the returned
6884 * If @page is not NULL and the extent is inline, this also reads the extent
6885 * data directly into the page and marks the extent up to date in the io_tree.
6887 * Return: ERR_PTR on error, non-NULL extent_map on success.
6889 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6890 struct page *page, size_t pg_offset,
6893 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6895 u64 extent_start = 0;
6897 u64 objectid = btrfs_ino(inode);
6898 int extent_type = -1;
6899 struct btrfs_path *path = NULL;
6900 struct btrfs_root *root = inode->root;
6901 struct btrfs_file_extent_item *item;
6902 struct extent_buffer *leaf;
6903 struct btrfs_key found_key;
6904 struct extent_map *em = NULL;
6905 struct extent_map_tree *em_tree = &inode->extent_tree;
6906 struct extent_io_tree *io_tree = &inode->io_tree;
6908 read_lock(&em_tree->lock);
6909 em = lookup_extent_mapping(em_tree, start, len);
6910 read_unlock(&em_tree->lock);
6913 if (em->start > start || em->start + em->len <= start)
6914 free_extent_map(em);
6915 else if (em->block_start == EXTENT_MAP_INLINE && page)
6916 free_extent_map(em);
6920 em = alloc_extent_map();
6925 em->start = EXTENT_MAP_HOLE;
6926 em->orig_start = EXTENT_MAP_HOLE;
6928 em->block_len = (u64)-1;
6930 path = btrfs_alloc_path();
6936 /* Chances are we'll be called again, so go ahead and do readahead */
6937 path->reada = READA_FORWARD;
6940 * The same explanation in load_free_space_cache applies here as well,
6941 * we only read when we're loading the free space cache, and at that
6942 * point the commit_root has everything we need.
6944 if (btrfs_is_free_space_inode(inode)) {
6945 path->search_commit_root = 1;
6946 path->skip_locking = 1;
6949 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6952 } else if (ret > 0) {
6953 if (path->slots[0] == 0)
6959 leaf = path->nodes[0];
6960 item = btrfs_item_ptr(leaf, path->slots[0],
6961 struct btrfs_file_extent_item);
6962 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6963 if (found_key.objectid != objectid ||
6964 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6966 * If we backup past the first extent we want to move forward
6967 * and see if there is an extent in front of us, otherwise we'll
6968 * say there is a hole for our whole search range which can
6975 extent_type = btrfs_file_extent_type(leaf, item);
6976 extent_start = found_key.offset;
6977 extent_end = btrfs_file_extent_end(path);
6978 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6979 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6980 /* Only regular file could have regular/prealloc extent */
6981 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6984 "regular/prealloc extent found for non-regular inode %llu",
6988 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6990 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6991 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6996 if (start >= extent_end) {
6998 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6999 ret = btrfs_next_leaf(root, path);
7005 leaf = path->nodes[0];
7007 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7008 if (found_key.objectid != objectid ||
7009 found_key.type != BTRFS_EXTENT_DATA_KEY)
7011 if (start + len <= found_key.offset)
7013 if (start > found_key.offset)
7016 /* New extent overlaps with existing one */
7018 em->orig_start = start;
7019 em->len = found_key.offset - start;
7020 em->block_start = EXTENT_MAP_HOLE;
7024 btrfs_extent_item_to_extent_map(inode, path, item, !page, em);
7026 if (extent_type == BTRFS_FILE_EXTENT_REG ||
7027 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
7029 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
7033 size_t extent_offset;
7039 size = btrfs_file_extent_ram_bytes(leaf, item);
7040 extent_offset = page_offset(page) + pg_offset - extent_start;
7041 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
7042 size - extent_offset);
7043 em->start = extent_start + extent_offset;
7044 em->len = ALIGN(copy_size, fs_info->sectorsize);
7045 em->orig_block_len = em->len;
7046 em->orig_start = em->start;
7047 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
7049 if (!PageUptodate(page)) {
7050 if (btrfs_file_extent_compression(leaf, item) !=
7051 BTRFS_COMPRESS_NONE) {
7052 ret = uncompress_inline(path, page, pg_offset,
7053 extent_offset, item);
7057 map = kmap_local_page(page);
7058 read_extent_buffer(leaf, map + pg_offset, ptr,
7060 if (pg_offset + copy_size < PAGE_SIZE) {
7061 memset(map + pg_offset + copy_size, 0,
7062 PAGE_SIZE - pg_offset -
7067 flush_dcache_page(page);
7069 set_extent_uptodate(io_tree, em->start,
7070 extent_map_end(em) - 1, NULL, GFP_NOFS);
7075 em->orig_start = start;
7077 em->block_start = EXTENT_MAP_HOLE;
7080 btrfs_release_path(path);
7081 if (em->start > start || extent_map_end(em) <= start) {
7083 "bad extent! em: [%llu %llu] passed [%llu %llu]",
7084 em->start, em->len, start, len);
7089 write_lock(&em_tree->lock);
7090 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
7091 write_unlock(&em_tree->lock);
7093 btrfs_free_path(path);
7095 trace_btrfs_get_extent(root, inode, em);
7098 free_extent_map(em);
7099 return ERR_PTR(ret);
7104 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
7107 struct extent_map *em;
7108 struct extent_map *hole_em = NULL;
7109 u64 delalloc_start = start;
7115 em = btrfs_get_extent(inode, NULL, 0, start, len);
7119 * If our em maps to:
7121 * - a pre-alloc extent,
7122 * there might actually be delalloc bytes behind it.
7124 if (em->block_start != EXTENT_MAP_HOLE &&
7125 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7130 /* check to see if we've wrapped (len == -1 or similar) */
7139 /* ok, we didn't find anything, lets look for delalloc */
7140 delalloc_len = count_range_bits(&inode->io_tree, &delalloc_start,
7141 end, len, EXTENT_DELALLOC, 1);
7142 delalloc_end = delalloc_start + delalloc_len;
7143 if (delalloc_end < delalloc_start)
7144 delalloc_end = (u64)-1;
7147 * We didn't find anything useful, return the original results from
7150 if (delalloc_start > end || delalloc_end <= start) {
7157 * Adjust the delalloc_start to make sure it doesn't go backwards from
7158 * the start they passed in
7160 delalloc_start = max(start, delalloc_start);
7161 delalloc_len = delalloc_end - delalloc_start;
7163 if (delalloc_len > 0) {
7166 const u64 hole_end = extent_map_end(hole_em);
7168 em = alloc_extent_map();
7176 * When btrfs_get_extent can't find anything it returns one
7179 * Make sure what it found really fits our range, and adjust to
7180 * make sure it is based on the start from the caller
7182 if (hole_end <= start || hole_em->start > end) {
7183 free_extent_map(hole_em);
7186 hole_start = max(hole_em->start, start);
7187 hole_len = hole_end - hole_start;
7190 if (hole_em && delalloc_start > hole_start) {
7192 * Our hole starts before our delalloc, so we have to
7193 * return just the parts of the hole that go until the
7196 em->len = min(hole_len, delalloc_start - hole_start);
7197 em->start = hole_start;
7198 em->orig_start = hole_start;
7200 * Don't adjust block start at all, it is fixed at
7203 em->block_start = hole_em->block_start;
7204 em->block_len = hole_len;
7205 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7206 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7209 * Hole is out of passed range or it starts after
7212 em->start = delalloc_start;
7213 em->len = delalloc_len;
7214 em->orig_start = delalloc_start;
7215 em->block_start = EXTENT_MAP_DELALLOC;
7216 em->block_len = delalloc_len;
7223 free_extent_map(hole_em);
7225 free_extent_map(em);
7226 return ERR_PTR(err);
7231 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
7234 const u64 orig_start,
7235 const u64 block_start,
7236 const u64 block_len,
7237 const u64 orig_block_len,
7238 const u64 ram_bytes,
7241 struct extent_map *em = NULL;
7244 if (type != BTRFS_ORDERED_NOCOW) {
7245 em = create_io_em(inode, start, len, orig_start, block_start,
7246 block_len, orig_block_len, ram_bytes,
7247 BTRFS_COMPRESS_NONE, /* compress_type */
7252 ret = btrfs_add_ordered_extent_dio(inode, start, block_start, len,
7256 free_extent_map(em);
7257 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
7266 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
7269 struct btrfs_root *root = inode->root;
7270 struct btrfs_fs_info *fs_info = root->fs_info;
7271 struct extent_map *em;
7272 struct btrfs_key ins;
7276 alloc_hint = get_extent_allocation_hint(inode, start, len);
7277 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7278 0, alloc_hint, &ins, 1, 1);
7280 return ERR_PTR(ret);
7282 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7283 ins.objectid, ins.offset, ins.offset,
7284 ins.offset, BTRFS_ORDERED_REGULAR);
7285 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7287 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7293 static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
7295 struct btrfs_block_group *block_group;
7296 bool readonly = false;
7298 block_group = btrfs_lookup_block_group(fs_info, bytenr);
7299 if (!block_group || block_group->ro)
7302 btrfs_put_block_group(block_group);
7307 * Check if we can do nocow write into the range [@offset, @offset + @len)
7309 * @offset: File offset
7310 * @len: The length to write, will be updated to the nocow writeable
7312 * @orig_start: (optional) Return the original file offset of the file extent
7313 * @orig_len: (optional) Return the original on-disk length of the file extent
7314 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7315 * @strict: if true, omit optimizations that might force us into unnecessary
7316 * cow. e.g., don't trust generation number.
7319 * >0 and update @len if we can do nocow write
7320 * 0 if we can't do nocow write
7321 * <0 if error happened
7323 * NOTE: This only checks the file extents, caller is responsible to wait for
7324 * any ordered extents.
7326 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7327 u64 *orig_start, u64 *orig_block_len,
7328 u64 *ram_bytes, bool strict)
7330 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7331 struct btrfs_path *path;
7333 struct extent_buffer *leaf;
7334 struct btrfs_root *root = BTRFS_I(inode)->root;
7335 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7336 struct btrfs_file_extent_item *fi;
7337 struct btrfs_key key;
7344 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7346 path = btrfs_alloc_path();
7350 ret = btrfs_lookup_file_extent(NULL, root, path,
7351 btrfs_ino(BTRFS_I(inode)), offset, 0);
7355 slot = path->slots[0];
7358 /* can't find the item, must cow */
7365 leaf = path->nodes[0];
7366 btrfs_item_key_to_cpu(leaf, &key, slot);
7367 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7368 key.type != BTRFS_EXTENT_DATA_KEY) {
7369 /* not our file or wrong item type, must cow */
7373 if (key.offset > offset) {
7374 /* Wrong offset, must cow */
7378 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7379 found_type = btrfs_file_extent_type(leaf, fi);
7380 if (found_type != BTRFS_FILE_EXTENT_REG &&
7381 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7382 /* not a regular extent, must cow */
7386 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7389 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7390 if (extent_end <= offset)
7393 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7394 if (disk_bytenr == 0)
7397 if (btrfs_file_extent_compression(leaf, fi) ||
7398 btrfs_file_extent_encryption(leaf, fi) ||
7399 btrfs_file_extent_other_encoding(leaf, fi))
7403 * Do the same check as in btrfs_cross_ref_exist but without the
7404 * unnecessary search.
7407 (btrfs_file_extent_generation(leaf, fi) <=
7408 btrfs_root_last_snapshot(&root->root_item)))
7411 backref_offset = btrfs_file_extent_offset(leaf, fi);
7414 *orig_start = key.offset - backref_offset;
7415 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7416 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7419 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7422 num_bytes = min(offset + *len, extent_end) - offset;
7423 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7426 range_end = round_up(offset + num_bytes,
7427 root->fs_info->sectorsize) - 1;
7428 ret = test_range_bit(io_tree, offset, range_end,
7429 EXTENT_DELALLOC, 0, NULL);
7436 btrfs_release_path(path);
7439 * look for other files referencing this extent, if we
7440 * find any we must cow
7443 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7444 key.offset - backref_offset, disk_bytenr,
7452 * adjust disk_bytenr and num_bytes to cover just the bytes
7453 * in this extent we are about to write. If there
7454 * are any csums in that range we have to cow in order
7455 * to keep the csums correct
7457 disk_bytenr += backref_offset;
7458 disk_bytenr += offset - key.offset;
7459 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7462 * all of the above have passed, it is safe to overwrite this extent
7468 btrfs_free_path(path);
7472 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7473 struct extent_state **cached_state, bool writing)
7475 struct btrfs_ordered_extent *ordered;
7479 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7482 * We're concerned with the entire range that we're going to be
7483 * doing DIO to, so we need to make sure there's no ordered
7484 * extents in this range.
7486 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7487 lockend - lockstart + 1);
7490 * We need to make sure there are no buffered pages in this
7491 * range either, we could have raced between the invalidate in
7492 * generic_file_direct_write and locking the extent. The
7493 * invalidate needs to happen so that reads after a write do not
7497 (!writing || !filemap_range_has_page(inode->i_mapping,
7498 lockstart, lockend)))
7501 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7506 * If we are doing a DIO read and the ordered extent we
7507 * found is for a buffered write, we can not wait for it
7508 * to complete and retry, because if we do so we can
7509 * deadlock with concurrent buffered writes on page
7510 * locks. This happens only if our DIO read covers more
7511 * than one extent map, if at this point has already
7512 * created an ordered extent for a previous extent map
7513 * and locked its range in the inode's io tree, and a
7514 * concurrent write against that previous extent map's
7515 * range and this range started (we unlock the ranges
7516 * in the io tree only when the bios complete and
7517 * buffered writes always lock pages before attempting
7518 * to lock range in the io tree).
7521 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7522 btrfs_start_ordered_extent(ordered, 1);
7525 btrfs_put_ordered_extent(ordered);
7528 * We could trigger writeback for this range (and wait
7529 * for it to complete) and then invalidate the pages for
7530 * this range (through invalidate_inode_pages2_range()),
7531 * but that can lead us to a deadlock with a concurrent
7532 * call to readahead (a buffered read or a defrag call
7533 * triggered a readahead) on a page lock due to an
7534 * ordered dio extent we created before but did not have
7535 * yet a corresponding bio submitted (whence it can not
7536 * complete), which makes readahead wait for that
7537 * ordered extent to complete while holding a lock on
7552 /* The callers of this must take lock_extent() */
7553 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7554 u64 len, u64 orig_start, u64 block_start,
7555 u64 block_len, u64 orig_block_len,
7556 u64 ram_bytes, int compress_type,
7559 struct extent_map_tree *em_tree;
7560 struct extent_map *em;
7563 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7564 type == BTRFS_ORDERED_COMPRESSED ||
7565 type == BTRFS_ORDERED_NOCOW ||
7566 type == BTRFS_ORDERED_REGULAR);
7568 em_tree = &inode->extent_tree;
7569 em = alloc_extent_map();
7571 return ERR_PTR(-ENOMEM);
7574 em->orig_start = orig_start;
7576 em->block_len = block_len;
7577 em->block_start = block_start;
7578 em->orig_block_len = orig_block_len;
7579 em->ram_bytes = ram_bytes;
7580 em->generation = -1;
7581 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7582 if (type == BTRFS_ORDERED_PREALLOC) {
7583 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7584 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7585 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7586 em->compress_type = compress_type;
7590 btrfs_drop_extent_cache(inode, em->start,
7591 em->start + em->len - 1, 0);
7592 write_lock(&em_tree->lock);
7593 ret = add_extent_mapping(em_tree, em, 1);
7594 write_unlock(&em_tree->lock);
7596 * The caller has taken lock_extent(), who could race with us
7599 } while (ret == -EEXIST);
7602 free_extent_map(em);
7603 return ERR_PTR(ret);
7606 /* em got 2 refs now, callers needs to do free_extent_map once. */
7611 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7612 struct inode *inode,
7613 struct btrfs_dio_data *dio_data,
7616 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7617 struct extent_map *em = *map;
7621 * We don't allocate a new extent in the following cases
7623 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7625 * 2) The extent is marked as PREALLOC. We're good to go here and can
7626 * just use the extent.
7629 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7630 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7631 em->block_start != EXTENT_MAP_HOLE)) {
7633 u64 block_start, orig_start, orig_block_len, ram_bytes;
7635 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7636 type = BTRFS_ORDERED_PREALLOC;
7638 type = BTRFS_ORDERED_NOCOW;
7639 len = min(len, em->len - (start - em->start));
7640 block_start = em->block_start + (start - em->start);
7642 if (can_nocow_extent(inode, start, &len, &orig_start,
7643 &orig_block_len, &ram_bytes, false) == 1 &&
7644 btrfs_inc_nocow_writers(fs_info, block_start)) {
7645 struct extent_map *em2;
7647 em2 = btrfs_create_dio_extent(BTRFS_I(inode), start, len,
7648 orig_start, block_start,
7649 len, orig_block_len,
7651 btrfs_dec_nocow_writers(fs_info, block_start);
7652 if (type == BTRFS_ORDERED_PREALLOC) {
7653 free_extent_map(em);
7657 if (em2 && IS_ERR(em2)) {
7662 * For inode marked NODATACOW or extent marked PREALLOC,
7663 * use the existing or preallocated extent, so does not
7664 * need to adjust btrfs_space_info's bytes_may_use.
7666 btrfs_free_reserved_data_space_noquota(fs_info, len);
7671 /* this will cow the extent */
7672 free_extent_map(em);
7673 *map = em = btrfs_new_extent_direct(BTRFS_I(inode), start, len);
7679 len = min(len, em->len - (start - em->start));
7683 * Need to update the i_size under the extent lock so buffered
7684 * readers will get the updated i_size when we unlock.
7686 if (start + len > i_size_read(inode))
7687 i_size_write(inode, start + len);
7689 dio_data->reserve -= len;
7694 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7695 loff_t length, unsigned int flags, struct iomap *iomap,
7696 struct iomap *srcmap)
7698 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7699 struct extent_map *em;
7700 struct extent_state *cached_state = NULL;
7701 struct btrfs_dio_data *dio_data = NULL;
7702 u64 lockstart, lockend;
7703 const bool write = !!(flags & IOMAP_WRITE);
7706 bool unlock_extents = false;
7709 len = min_t(u64, len, fs_info->sectorsize);
7712 lockend = start + len - 1;
7715 * The generic stuff only does filemap_write_and_wait_range, which
7716 * isn't enough if we've written compressed pages to this area, so we
7717 * need to flush the dirty pages again to make absolutely sure that any
7718 * outstanding dirty pages are on disk.
7720 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7721 &BTRFS_I(inode)->runtime_flags)) {
7722 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7723 start + length - 1);
7728 dio_data = kzalloc(sizeof(*dio_data), GFP_NOFS);
7732 dio_data->length = length;
7734 dio_data->reserve = round_up(length, fs_info->sectorsize);
7735 ret = btrfs_delalloc_reserve_space(BTRFS_I(inode),
7736 &dio_data->data_reserved,
7737 start, dio_data->reserve);
7739 extent_changeset_free(dio_data->data_reserved);
7744 iomap->private = dio_data;
7748 * If this errors out it's because we couldn't invalidate pagecache for
7749 * this range and we need to fallback to buffered.
7751 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, write)) {
7756 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7763 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7764 * io. INLINE is special, and we could probably kludge it in here, but
7765 * it's still buffered so for safety lets just fall back to the generic
7768 * For COMPRESSED we _have_ to read the entire extent in so we can
7769 * decompress it, so there will be buffering required no matter what we
7770 * do, so go ahead and fallback to buffered.
7772 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7773 * to buffered IO. Don't blame me, this is the price we pay for using
7776 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7777 em->block_start == EXTENT_MAP_INLINE) {
7778 free_extent_map(em);
7783 len = min(len, em->len - (start - em->start));
7785 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7789 unlock_extents = true;
7790 /* Recalc len in case the new em is smaller than requested */
7791 len = min(len, em->len - (start - em->start));
7794 * We need to unlock only the end area that we aren't using.
7795 * The rest is going to be unlocked by the endio routine.
7797 lockstart = start + len;
7798 if (lockstart < lockend)
7799 unlock_extents = true;
7803 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
7804 lockstart, lockend, &cached_state);
7806 free_extent_state(cached_state);
7809 * Translate extent map information to iomap.
7810 * We trim the extents (and move the addr) even though iomap code does
7811 * that, since we have locked only the parts we are performing I/O in.
7813 if ((em->block_start == EXTENT_MAP_HOLE) ||
7814 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7815 iomap->addr = IOMAP_NULL_ADDR;
7816 iomap->type = IOMAP_HOLE;
7818 iomap->addr = em->block_start + (start - em->start);
7819 iomap->type = IOMAP_MAPPED;
7821 iomap->offset = start;
7822 iomap->bdev = fs_info->fs_devices->latest_bdev;
7823 iomap->length = len;
7825 if (write && btrfs_use_zone_append(BTRFS_I(inode), em->block_start))
7826 iomap->flags |= IOMAP_F_ZONE_APPEND;
7828 free_extent_map(em);
7833 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7837 btrfs_delalloc_release_space(BTRFS_I(inode),
7838 dio_data->data_reserved, start,
7839 dio_data->reserve, true);
7840 btrfs_delalloc_release_extents(BTRFS_I(inode), dio_data->reserve);
7841 extent_changeset_free(dio_data->data_reserved);
7847 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7848 ssize_t written, unsigned int flags, struct iomap *iomap)
7851 struct btrfs_dio_data *dio_data = iomap->private;
7852 size_t submitted = dio_data->submitted;
7853 const bool write = !!(flags & IOMAP_WRITE);
7855 if (!write && (iomap->type == IOMAP_HOLE)) {
7856 /* If reading from a hole, unlock and return */
7857 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1);
7861 if (submitted < length) {
7863 length -= submitted;
7865 __endio_write_update_ordered(BTRFS_I(inode), pos,
7868 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7874 if (dio_data->reserve)
7875 btrfs_delalloc_release_space(BTRFS_I(inode),
7876 dio_data->data_reserved, pos,
7877 dio_data->reserve, true);
7878 btrfs_delalloc_release_extents(BTRFS_I(inode), dio_data->length);
7879 extent_changeset_free(dio_data->data_reserved);
7883 iomap->private = NULL;
7888 static void btrfs_dio_private_put(struct btrfs_dio_private *dip)
7891 * This implies a barrier so that stores to dio_bio->bi_status before
7892 * this and loads of dio_bio->bi_status after this are fully ordered.
7894 if (!refcount_dec_and_test(&dip->refs))
7897 if (btrfs_op(dip->dio_bio) == BTRFS_MAP_WRITE) {
7898 __endio_write_update_ordered(BTRFS_I(dip->inode),
7899 dip->logical_offset,
7901 !dip->dio_bio->bi_status);
7903 unlock_extent(&BTRFS_I(dip->inode)->io_tree,
7904 dip->logical_offset,
7905 dip->logical_offset + dip->bytes - 1);
7908 bio_endio(dip->dio_bio);
7912 static blk_status_t submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7914 unsigned long bio_flags)
7916 struct btrfs_dio_private *dip = bio->bi_private;
7917 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7920 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7922 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
7926 refcount_inc(&dip->refs);
7927 ret = btrfs_map_bio(fs_info, bio, mirror_num);
7929 refcount_dec(&dip->refs);
7933 static blk_status_t btrfs_check_read_dio_bio(struct inode *inode,
7934 struct btrfs_io_bio *io_bio,
7935 const bool uptodate)
7937 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
7938 const u32 sectorsize = fs_info->sectorsize;
7939 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7940 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7941 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7942 struct bio_vec bvec;
7943 struct bvec_iter iter;
7944 u64 start = io_bio->logical;
7946 blk_status_t err = BLK_STS_OK;
7948 __bio_for_each_segment(bvec, &io_bio->bio, iter, io_bio->iter) {
7949 unsigned int i, nr_sectors, pgoff;
7951 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7952 pgoff = bvec.bv_offset;
7953 for (i = 0; i < nr_sectors; i++) {
7954 ASSERT(pgoff < PAGE_SIZE);
7956 (!csum || !check_data_csum(inode, io_bio,
7957 bio_offset, bvec.bv_page,
7959 clean_io_failure(fs_info, failure_tree, io_tree,
7960 start, bvec.bv_page,
7961 btrfs_ino(BTRFS_I(inode)),
7966 ASSERT((start - io_bio->logical) < UINT_MAX);
7967 ret = btrfs_repair_one_sector(inode,
7969 start - io_bio->logical,
7970 bvec.bv_page, pgoff,
7971 start, io_bio->mirror_num,
7972 submit_dio_repair_bio);
7974 err = errno_to_blk_status(ret);
7976 start += sectorsize;
7977 ASSERT(bio_offset + sectorsize > bio_offset);
7978 bio_offset += sectorsize;
7979 pgoff += sectorsize;
7985 static void __endio_write_update_ordered(struct btrfs_inode *inode,
7986 const u64 offset, const u64 bytes,
7987 const bool uptodate)
7989 btrfs_mark_ordered_io_finished(inode, NULL, offset, bytes,
7990 finish_ordered_fn, uptodate);
7993 static blk_status_t btrfs_submit_bio_start_direct_io(struct inode *inode,
7995 u64 dio_file_offset)
7997 return btrfs_csum_one_bio(BTRFS_I(inode), bio, dio_file_offset, 1);
8000 static void btrfs_end_dio_bio(struct bio *bio)
8002 struct btrfs_dio_private *dip = bio->bi_private;
8003 blk_status_t err = bio->bi_status;
8006 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8007 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8008 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8009 bio->bi_opf, bio->bi_iter.bi_sector,
8010 bio->bi_iter.bi_size, err);
8012 if (bio_op(bio) == REQ_OP_READ) {
8013 err = btrfs_check_read_dio_bio(dip->inode, btrfs_io_bio(bio),
8018 dip->dio_bio->bi_status = err;
8020 btrfs_record_physical_zoned(dip->inode, dip->logical_offset, bio);
8023 btrfs_dio_private_put(dip);
8026 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
8027 struct inode *inode, u64 file_offset, int async_submit)
8029 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8030 struct btrfs_dio_private *dip = bio->bi_private;
8031 bool write = btrfs_op(bio) == BTRFS_MAP_WRITE;
8034 /* Check btrfs_submit_bio_hook() for rules about async submit. */
8036 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8039 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8044 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
8047 if (write && async_submit) {
8048 ret = btrfs_wq_submit_bio(inode, bio, 0, 0, file_offset,
8049 btrfs_submit_bio_start_direct_io);
8053 * If we aren't doing async submit, calculate the csum of the
8056 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, file_offset, 1);
8062 csum_offset = file_offset - dip->logical_offset;
8063 csum_offset >>= fs_info->sectorsize_bits;
8064 csum_offset *= fs_info->csum_size;
8065 btrfs_io_bio(bio)->csum = dip->csums + csum_offset;
8068 ret = btrfs_map_bio(fs_info, bio, 0);
8074 * If this succeeds, the btrfs_dio_private is responsible for cleaning up locked
8075 * or ordered extents whether or not we submit any bios.
8077 static struct btrfs_dio_private *btrfs_create_dio_private(struct bio *dio_bio,
8078 struct inode *inode,
8081 const bool write = (btrfs_op(dio_bio) == BTRFS_MAP_WRITE);
8082 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
8084 struct btrfs_dio_private *dip;
8086 dip_size = sizeof(*dip);
8087 if (!write && csum) {
8088 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8091 nblocks = dio_bio->bi_iter.bi_size >> fs_info->sectorsize_bits;
8092 dip_size += fs_info->csum_size * nblocks;
8095 dip = kzalloc(dip_size, GFP_NOFS);
8100 dip->logical_offset = file_offset;
8101 dip->bytes = dio_bio->bi_iter.bi_size;
8102 dip->disk_bytenr = dio_bio->bi_iter.bi_sector << 9;
8103 dip->dio_bio = dio_bio;
8104 refcount_set(&dip->refs, 1);
8108 static blk_qc_t btrfs_submit_direct(struct inode *inode, struct iomap *iomap,
8109 struct bio *dio_bio, loff_t file_offset)
8111 const bool write = (btrfs_op(dio_bio) == BTRFS_MAP_WRITE);
8112 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8113 const bool raid56 = (btrfs_data_alloc_profile(fs_info) &
8114 BTRFS_BLOCK_GROUP_RAID56_MASK);
8115 struct btrfs_dio_private *dip;
8118 int async_submit = 0;
8120 int clone_offset = 0;
8124 blk_status_t status;
8125 struct btrfs_io_geometry geom;
8126 struct btrfs_dio_data *dio_data = iomap->private;
8127 struct extent_map *em = NULL;
8129 dip = btrfs_create_dio_private(dio_bio, inode, file_offset);
8132 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8133 file_offset + dio_bio->bi_iter.bi_size - 1);
8135 dio_bio->bi_status = BLK_STS_RESOURCE;
8137 return BLK_QC_T_NONE;
8142 * Load the csums up front to reduce csum tree searches and
8143 * contention when submitting bios.
8145 * If we have csums disabled this will do nothing.
8147 status = btrfs_lookup_bio_sums(inode, dio_bio, dip->csums);
8148 if (status != BLK_STS_OK)
8152 start_sector = dio_bio->bi_iter.bi_sector;
8153 submit_len = dio_bio->bi_iter.bi_size;
8156 logical = start_sector << 9;
8157 em = btrfs_get_chunk_map(fs_info, logical, submit_len);
8159 status = errno_to_blk_status(PTR_ERR(em));
8163 ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(dio_bio),
8166 status = errno_to_blk_status(ret);
8169 ASSERT(geom.len <= INT_MAX);
8171 clone_len = min_t(int, submit_len, geom.len);
8174 * This will never fail as it's passing GPF_NOFS and
8175 * the allocation is backed by btrfs_bioset.
8177 bio = btrfs_bio_clone_partial(dio_bio, clone_offset, clone_len);
8178 bio->bi_private = dip;
8179 bio->bi_end_io = btrfs_end_dio_bio;
8180 btrfs_io_bio(bio)->logical = file_offset;
8182 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
8183 status = extract_ordered_extent(BTRFS_I(inode), bio,
8191 ASSERT(submit_len >= clone_len);
8192 submit_len -= clone_len;
8195 * Increase the count before we submit the bio so we know
8196 * the end IO handler won't happen before we increase the
8197 * count. Otherwise, the dip might get freed before we're
8198 * done setting it up.
8200 * We transfer the initial reference to the last bio, so we
8201 * don't need to increment the reference count for the last one.
8203 if (submit_len > 0) {
8204 refcount_inc(&dip->refs);
8206 * If we are submitting more than one bio, submit them
8207 * all asynchronously. The exception is RAID 5 or 6, as
8208 * asynchronous checksums make it difficult to collect
8209 * full stripe writes.
8215 status = btrfs_submit_dio_bio(bio, inode, file_offset,
8220 refcount_dec(&dip->refs);
8224 dio_data->submitted += clone_len;
8225 clone_offset += clone_len;
8226 start_sector += clone_len >> 9;
8227 file_offset += clone_len;
8229 free_extent_map(em);
8230 } while (submit_len > 0);
8231 return BLK_QC_T_NONE;
8234 free_extent_map(em);
8236 dip->dio_bio->bi_status = status;
8237 btrfs_dio_private_put(dip);
8239 return BLK_QC_T_NONE;
8242 const struct iomap_ops btrfs_dio_iomap_ops = {
8243 .iomap_begin = btrfs_dio_iomap_begin,
8244 .iomap_end = btrfs_dio_iomap_end,
8247 const struct iomap_dio_ops btrfs_dio_ops = {
8248 .submit_io = btrfs_submit_direct,
8251 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8256 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
8260 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
8263 int btrfs_readpage(struct file *file, struct page *page)
8265 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
8266 u64 start = page_offset(page);
8267 u64 end = start + PAGE_SIZE - 1;
8268 struct btrfs_bio_ctrl bio_ctrl = { 0 };
8271 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
8273 ret = btrfs_do_readpage(page, NULL, &bio_ctrl, 0, NULL);
8275 ret = submit_one_bio(bio_ctrl.bio, 0, bio_ctrl.bio_flags);
8279 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8281 struct inode *inode = page->mapping->host;
8284 if (current->flags & PF_MEMALLOC) {
8285 redirty_page_for_writepage(wbc, page);
8291 * If we are under memory pressure we will call this directly from the
8292 * VM, we need to make sure we have the inode referenced for the ordered
8293 * extent. If not just return like we didn't do anything.
8295 if (!igrab(inode)) {
8296 redirty_page_for_writepage(wbc, page);
8297 return AOP_WRITEPAGE_ACTIVATE;
8299 ret = extent_write_full_page(page, wbc);
8300 btrfs_add_delayed_iput(inode);
8304 static int btrfs_writepages(struct address_space *mapping,
8305 struct writeback_control *wbc)
8307 return extent_writepages(mapping, wbc);
8310 static void btrfs_readahead(struct readahead_control *rac)
8312 extent_readahead(rac);
8315 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8317 int ret = try_release_extent_mapping(page, gfp_flags);
8319 clear_page_extent_mapped(page);
8323 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8325 if (PageWriteback(page) || PageDirty(page))
8327 return __btrfs_releasepage(page, gfp_flags);
8330 #ifdef CONFIG_MIGRATION
8331 static int btrfs_migratepage(struct address_space *mapping,
8332 struct page *newpage, struct page *page,
8333 enum migrate_mode mode)
8337 ret = migrate_page_move_mapping(mapping, newpage, page, 0);
8338 if (ret != MIGRATEPAGE_SUCCESS)
8341 if (page_has_private(page))
8342 attach_page_private(newpage, detach_page_private(page));
8344 if (PageOrdered(page)) {
8345 ClearPageOrdered(page);
8346 SetPageOrdered(newpage);
8349 if (mode != MIGRATE_SYNC_NO_COPY)
8350 migrate_page_copy(newpage, page);
8352 migrate_page_states(newpage, page);
8353 return MIGRATEPAGE_SUCCESS;
8357 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8358 unsigned int length)
8360 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
8361 struct btrfs_fs_info *fs_info = inode->root->fs_info;
8362 struct extent_io_tree *tree = &inode->io_tree;
8363 struct extent_state *cached_state = NULL;
8364 u64 page_start = page_offset(page);
8365 u64 page_end = page_start + PAGE_SIZE - 1;
8367 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
8370 * We have page locked so no new ordered extent can be created on this
8371 * page, nor bio can be submitted for this page.
8373 * But already submitted bio can still be finished on this page.
8374 * Furthermore, endio function won't skip page which has Ordered
8375 * (Private2) already cleared, so it's possible for endio and
8376 * invalidatepage to do the same ordered extent accounting twice
8379 * So here we wait for any submitted bios to finish, so that we won't
8380 * do double ordered extent accounting on the same page.
8382 wait_on_page_writeback(page);
8385 * For subpage case, we have call sites like
8386 * btrfs_punch_hole_lock_range() which passes range not aligned to
8388 * If the range doesn't cover the full page, we don't need to and
8389 * shouldn't clear page extent mapped, as page->private can still
8390 * record subpage dirty bits for other part of the range.
8392 * For cases that can invalidate the full even the range doesn't
8393 * cover the full page, like invalidating the last page, we're
8394 * still safe to wait for ordered extent to finish.
8396 if (!(offset == 0 && length == PAGE_SIZE)) {
8397 btrfs_releasepage(page, GFP_NOFS);
8401 if (!inode_evicting)
8402 lock_extent_bits(tree, page_start, page_end, &cached_state);
8405 while (cur < page_end) {
8406 struct btrfs_ordered_extent *ordered;
8411 ordered = btrfs_lookup_first_ordered_range(inode, cur,
8412 page_end + 1 - cur);
8414 range_end = page_end;
8416 * No ordered extent covering this range, we are safe
8417 * to delete all extent states in the range.
8419 delete_states = true;
8422 if (ordered->file_offset > cur) {
8424 * There is a range between [cur, oe->file_offset) not
8425 * covered by any ordered extent.
8426 * We are safe to delete all extent states, and handle
8427 * the ordered extent in the next iteration.
8429 range_end = ordered->file_offset - 1;
8430 delete_states = true;
8434 range_end = min(ordered->file_offset + ordered->num_bytes - 1,
8436 ASSERT(range_end + 1 - cur < U32_MAX);
8437 range_len = range_end + 1 - cur;
8438 if (!btrfs_page_test_ordered(fs_info, page, cur, range_len)) {
8440 * If Ordered (Private2) is cleared, it means endio has
8441 * already been executed for the range.
8442 * We can't delete the extent states as
8443 * btrfs_finish_ordered_io() may still use some of them.
8445 delete_states = false;
8448 btrfs_page_clear_ordered(fs_info, page, cur, range_len);
8451 * IO on this page will never be started, so we need to account
8452 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8453 * here, must leave that up for the ordered extent completion.
8455 * This will also unlock the range for incoming
8456 * btrfs_finish_ordered_io().
8458 if (!inode_evicting)
8459 clear_extent_bit(tree, cur, range_end,
8461 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8462 EXTENT_DEFRAG, 1, 0, &cached_state);
8464 spin_lock_irq(&inode->ordered_tree.lock);
8465 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8466 ordered->truncated_len = min(ordered->truncated_len,
8467 cur - ordered->file_offset);
8468 spin_unlock_irq(&inode->ordered_tree.lock);
8470 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8471 cur, range_end + 1 - cur, 1)) {
8472 btrfs_finish_ordered_io(ordered);
8474 * The ordered extent has finished, now we're again
8475 * safe to delete all extent states of the range.
8477 delete_states = true;
8480 * btrfs_finish_ordered_io() will get executed by endio
8481 * of other pages, thus we can't delete extent states
8484 delete_states = false;
8488 btrfs_put_ordered_extent(ordered);
8490 * Qgroup reserved space handler
8491 * Sector(s) here will be either:
8493 * 1) Already written to disk or bio already finished
8494 * Then its QGROUP_RESERVED bit in io_tree is already cleared.
8495 * Qgroup will be handled by its qgroup_record then.
8496 * btrfs_qgroup_free_data() call will do nothing here.
8498 * 2) Not written to disk yet
8499 * Then btrfs_qgroup_free_data() call will clear the
8500 * QGROUP_RESERVED bit of its io_tree, and free the qgroup
8501 * reserved data space.
8502 * Since the IO will never happen for this page.
8504 btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur);
8505 if (!inode_evicting) {
8506 clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED |
8507 EXTENT_DELALLOC | EXTENT_UPTODATE |
8508 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1,
8509 delete_states, &cached_state);
8511 cur = range_end + 1;
8514 * We have iterated through all ordered extents of the page, the page
8515 * should not have Ordered (Private2) anymore, or the above iteration
8516 * did something wrong.
8518 ASSERT(!PageOrdered(page));
8519 if (!inode_evicting)
8520 __btrfs_releasepage(page, GFP_NOFS);
8521 ClearPageChecked(page);
8522 clear_page_extent_mapped(page);
8526 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8527 * called from a page fault handler when a page is first dirtied. Hence we must
8528 * be careful to check for EOF conditions here. We set the page up correctly
8529 * for a written page which means we get ENOSPC checking when writing into
8530 * holes and correct delalloc and unwritten extent mapping on filesystems that
8531 * support these features.
8533 * We are not allowed to take the i_mutex here so we have to play games to
8534 * protect against truncate races as the page could now be beyond EOF. Because
8535 * truncate_setsize() writes the inode size before removing pages, once we have
8536 * the page lock we can determine safely if the page is beyond EOF. If it is not
8537 * beyond EOF, then the page is guaranteed safe against truncation until we
8540 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8542 struct page *page = vmf->page;
8543 struct inode *inode = file_inode(vmf->vma->vm_file);
8544 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8545 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8546 struct btrfs_ordered_extent *ordered;
8547 struct extent_state *cached_state = NULL;
8548 struct extent_changeset *data_reserved = NULL;
8549 unsigned long zero_start;
8559 reserved_space = PAGE_SIZE;
8561 sb_start_pagefault(inode->i_sb);
8562 page_start = page_offset(page);
8563 page_end = page_start + PAGE_SIZE - 1;
8567 * Reserving delalloc space after obtaining the page lock can lead to
8568 * deadlock. For example, if a dirty page is locked by this function
8569 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8570 * dirty page write out, then the btrfs_writepage() function could
8571 * end up waiting indefinitely to get a lock on the page currently
8572 * being processed by btrfs_page_mkwrite() function.
8574 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8575 page_start, reserved_space);
8577 ret2 = file_update_time(vmf->vma->vm_file);
8581 ret = vmf_error(ret2);
8587 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8589 down_read(&BTRFS_I(inode)->i_mmap_lock);
8591 size = i_size_read(inode);
8593 if ((page->mapping != inode->i_mapping) ||
8594 (page_start >= size)) {
8595 /* page got truncated out from underneath us */
8598 wait_on_page_writeback(page);
8600 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8601 ret2 = set_page_extent_mapped(page);
8603 ret = vmf_error(ret2);
8604 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8609 * we can't set the delalloc bits if there are pending ordered
8610 * extents. Drop our locks and wait for them to finish
8612 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8615 unlock_extent_cached(io_tree, page_start, page_end,
8618 up_read(&BTRFS_I(inode)->i_mmap_lock);
8619 btrfs_start_ordered_extent(ordered, 1);
8620 btrfs_put_ordered_extent(ordered);
8624 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8625 reserved_space = round_up(size - page_start,
8626 fs_info->sectorsize);
8627 if (reserved_space < PAGE_SIZE) {
8628 end = page_start + reserved_space - 1;
8629 btrfs_delalloc_release_space(BTRFS_I(inode),
8630 data_reserved, page_start,
8631 PAGE_SIZE - reserved_space, true);
8636 * page_mkwrite gets called when the page is firstly dirtied after it's
8637 * faulted in, but write(2) could also dirty a page and set delalloc
8638 * bits, thus in this case for space account reason, we still need to
8639 * clear any delalloc bits within this page range since we have to
8640 * reserve data&meta space before lock_page() (see above comments).
8642 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8643 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8644 EXTENT_DEFRAG, 0, 0, &cached_state);
8646 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8649 unlock_extent_cached(io_tree, page_start, page_end,
8651 ret = VM_FAULT_SIGBUS;
8655 /* page is wholly or partially inside EOF */
8656 if (page_start + PAGE_SIZE > size)
8657 zero_start = offset_in_page(size);
8659 zero_start = PAGE_SIZE;
8661 if (zero_start != PAGE_SIZE) {
8662 memzero_page(page, zero_start, PAGE_SIZE - zero_start);
8663 flush_dcache_page(page);
8665 ClearPageChecked(page);
8666 btrfs_page_set_dirty(fs_info, page, page_start, end + 1 - page_start);
8667 btrfs_page_set_uptodate(fs_info, page, page_start, end + 1 - page_start);
8669 btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
8671 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8672 up_read(&BTRFS_I(inode)->i_mmap_lock);
8674 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8675 sb_end_pagefault(inode->i_sb);
8676 extent_changeset_free(data_reserved);
8677 return VM_FAULT_LOCKED;
8681 up_read(&BTRFS_I(inode)->i_mmap_lock);
8683 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8684 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8685 reserved_space, (ret != 0));
8687 sb_end_pagefault(inode->i_sb);
8688 extent_changeset_free(data_reserved);
8692 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
8694 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8695 struct btrfs_root *root = BTRFS_I(inode)->root;
8696 struct btrfs_block_rsv *rsv;
8698 struct btrfs_trans_handle *trans;
8699 u64 mask = fs_info->sectorsize - 1;
8700 u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8701 u64 extents_found = 0;
8703 if (!skip_writeback) {
8704 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8711 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8712 * things going on here:
8714 * 1) We need to reserve space to update our inode.
8716 * 2) We need to have something to cache all the space that is going to
8717 * be free'd up by the truncate operation, but also have some slack
8718 * space reserved in case it uses space during the truncate (thank you
8719 * very much snapshotting).
8721 * And we need these to be separate. The fact is we can use a lot of
8722 * space doing the truncate, and we have no earthly idea how much space
8723 * we will use, so we need the truncate reservation to be separate so it
8724 * doesn't end up using space reserved for updating the inode. We also
8725 * need to be able to stop the transaction and start a new one, which
8726 * means we need to be able to update the inode several times, and we
8727 * have no idea of knowing how many times that will be, so we can't just
8728 * reserve 1 item for the entirety of the operation, so that has to be
8729 * done separately as well.
8731 * So that leaves us with
8733 * 1) rsv - for the truncate reservation, which we will steal from the
8734 * transaction reservation.
8735 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8736 * updating the inode.
8738 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8741 rsv->size = min_size;
8745 * 1 for the truncate slack space
8746 * 1 for updating the inode.
8748 trans = btrfs_start_transaction(root, 2);
8749 if (IS_ERR(trans)) {
8750 ret = PTR_ERR(trans);
8754 /* Migrate the slack space for the truncate to our reserve */
8755 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8759 trans->block_rsv = rsv;
8762 ret = btrfs_truncate_inode_items(trans, root, BTRFS_I(inode),
8764 BTRFS_EXTENT_DATA_KEY,
8766 trans->block_rsv = &fs_info->trans_block_rsv;
8767 if (ret != -ENOSPC && ret != -EAGAIN)
8770 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
8774 btrfs_end_transaction(trans);
8775 btrfs_btree_balance_dirty(fs_info);
8777 trans = btrfs_start_transaction(root, 2);
8778 if (IS_ERR(trans)) {
8779 ret = PTR_ERR(trans);
8784 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8785 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8786 rsv, min_size, false);
8787 BUG_ON(ret); /* shouldn't happen */
8788 trans->block_rsv = rsv;
8792 * We can't call btrfs_truncate_block inside a trans handle as we could
8793 * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
8794 * we've truncated everything except the last little bit, and can do
8795 * btrfs_truncate_block and then update the disk_i_size.
8797 if (ret == NEED_TRUNCATE_BLOCK) {
8798 btrfs_end_transaction(trans);
8799 btrfs_btree_balance_dirty(fs_info);
8801 ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0);
8804 trans = btrfs_start_transaction(root, 1);
8805 if (IS_ERR(trans)) {
8806 ret = PTR_ERR(trans);
8809 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
8815 trans->block_rsv = &fs_info->trans_block_rsv;
8816 ret2 = btrfs_update_inode(trans, root, BTRFS_I(inode));
8820 ret2 = btrfs_end_transaction(trans);
8823 btrfs_btree_balance_dirty(fs_info);
8826 btrfs_free_block_rsv(fs_info, rsv);
8828 * So if we truncate and then write and fsync we normally would just
8829 * write the extents that changed, which is a problem if we need to
8830 * first truncate that entire inode. So set this flag so we write out
8831 * all of the extents in the inode to the sync log so we're completely
8834 * If no extents were dropped or trimmed we don't need to force the next
8835 * fsync to truncate all the inode's items from the log and re-log them
8836 * all. This means the truncate operation did not change the file size,
8837 * or changed it to a smaller size but there was only an implicit hole
8838 * between the old i_size and the new i_size, and there were no prealloc
8839 * extents beyond i_size to drop.
8841 if (extents_found > 0)
8842 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
8848 * create a new subvolume directory/inode (helper for the ioctl).
8850 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8851 struct btrfs_root *new_root,
8852 struct btrfs_root *parent_root)
8854 struct inode *inode;
8859 err = btrfs_get_free_objectid(new_root, &ino);
8863 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, ino, ino,
8864 S_IFDIR | (~current_umask() & S_IRWXUGO),
8867 return PTR_ERR(inode);
8868 inode->i_op = &btrfs_dir_inode_operations;
8869 inode->i_fop = &btrfs_dir_file_operations;
8871 set_nlink(inode, 1);
8872 btrfs_i_size_write(BTRFS_I(inode), 0);
8873 unlock_new_inode(inode);
8875 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8877 btrfs_err(new_root->fs_info,
8878 "error inheriting subvolume %llu properties: %d",
8879 new_root->root_key.objectid, err);
8881 err = btrfs_update_inode(trans, new_root, BTRFS_I(inode));
8887 struct inode *btrfs_alloc_inode(struct super_block *sb)
8889 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8890 struct btrfs_inode *ei;
8891 struct inode *inode;
8893 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL);
8900 ei->last_sub_trans = 0;
8901 ei->logged_trans = 0;
8902 ei->delalloc_bytes = 0;
8903 ei->new_delalloc_bytes = 0;
8904 ei->defrag_bytes = 0;
8905 ei->disk_i_size = 0;
8908 ei->index_cnt = (u64)-1;
8910 ei->last_unlink_trans = 0;
8911 ei->last_reflink_trans = 0;
8912 ei->last_log_commit = 0;
8914 spin_lock_init(&ei->lock);
8915 ei->outstanding_extents = 0;
8916 if (sb->s_magic != BTRFS_TEST_MAGIC)
8917 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8918 BTRFS_BLOCK_RSV_DELALLOC);
8919 ei->runtime_flags = 0;
8920 ei->prop_compress = BTRFS_COMPRESS_NONE;
8921 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8923 ei->delayed_node = NULL;
8925 ei->i_otime.tv_sec = 0;
8926 ei->i_otime.tv_nsec = 0;
8928 inode = &ei->vfs_inode;
8929 extent_map_tree_init(&ei->extent_tree);
8930 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO, inode);
8931 extent_io_tree_init(fs_info, &ei->io_failure_tree,
8932 IO_TREE_INODE_IO_FAILURE, inode);
8933 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8934 IO_TREE_INODE_FILE_EXTENT, inode);
8935 ei->io_tree.track_uptodate = true;
8936 ei->io_failure_tree.track_uptodate = true;
8937 atomic_set(&ei->sync_writers, 0);
8938 mutex_init(&ei->log_mutex);
8939 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8940 INIT_LIST_HEAD(&ei->delalloc_inodes);
8941 INIT_LIST_HEAD(&ei->delayed_iput);
8942 RB_CLEAR_NODE(&ei->rb_node);
8943 init_rwsem(&ei->i_mmap_lock);
8948 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8949 void btrfs_test_destroy_inode(struct inode *inode)
8951 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
8952 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8956 void btrfs_free_inode(struct inode *inode)
8958 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8961 void btrfs_destroy_inode(struct inode *vfs_inode)
8963 struct btrfs_ordered_extent *ordered;
8964 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8965 struct btrfs_root *root = inode->root;
8967 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8968 WARN_ON(vfs_inode->i_data.nrpages);
8969 WARN_ON(inode->block_rsv.reserved);
8970 WARN_ON(inode->block_rsv.size);
8971 WARN_ON(inode->outstanding_extents);
8972 WARN_ON(inode->delalloc_bytes);
8973 WARN_ON(inode->new_delalloc_bytes);
8974 WARN_ON(inode->csum_bytes);
8975 WARN_ON(inode->defrag_bytes);
8978 * This can happen where we create an inode, but somebody else also
8979 * created the same inode and we need to destroy the one we already
8986 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8990 btrfs_err(root->fs_info,
8991 "found ordered extent %llu %llu on inode cleanup",
8992 ordered->file_offset, ordered->num_bytes);
8993 btrfs_remove_ordered_extent(inode, ordered);
8994 btrfs_put_ordered_extent(ordered);
8995 btrfs_put_ordered_extent(ordered);
8998 btrfs_qgroup_check_reserved_leak(inode);
8999 inode_tree_del(inode);
9000 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9001 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
9002 btrfs_put_root(inode->root);
9005 int btrfs_drop_inode(struct inode *inode)
9007 struct btrfs_root *root = BTRFS_I(inode)->root;
9012 /* the snap/subvol tree is on deleting */
9013 if (btrfs_root_refs(&root->root_item) == 0)
9016 return generic_drop_inode(inode);
9019 static void init_once(void *foo)
9021 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9023 inode_init_once(&ei->vfs_inode);
9026 void __cold btrfs_destroy_cachep(void)
9029 * Make sure all delayed rcu free inodes are flushed before we
9033 kmem_cache_destroy(btrfs_inode_cachep);
9034 kmem_cache_destroy(btrfs_trans_handle_cachep);
9035 kmem_cache_destroy(btrfs_path_cachep);
9036 kmem_cache_destroy(btrfs_free_space_cachep);
9037 kmem_cache_destroy(btrfs_free_space_bitmap_cachep);
9040 int __init btrfs_init_cachep(void)
9042 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9043 sizeof(struct btrfs_inode), 0,
9044 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9046 if (!btrfs_inode_cachep)
9049 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9050 sizeof(struct btrfs_trans_handle), 0,
9051 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9052 if (!btrfs_trans_handle_cachep)
9055 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9056 sizeof(struct btrfs_path), 0,
9057 SLAB_MEM_SPREAD, NULL);
9058 if (!btrfs_path_cachep)
9061 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9062 sizeof(struct btrfs_free_space), 0,
9063 SLAB_MEM_SPREAD, NULL);
9064 if (!btrfs_free_space_cachep)
9067 btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap",
9068 PAGE_SIZE, PAGE_SIZE,
9069 SLAB_MEM_SPREAD, NULL);
9070 if (!btrfs_free_space_bitmap_cachep)
9075 btrfs_destroy_cachep();
9079 static int btrfs_getattr(struct user_namespace *mnt_userns,
9080 const struct path *path, struct kstat *stat,
9081 u32 request_mask, unsigned int flags)
9085 struct inode *inode = d_inode(path->dentry);
9086 u32 blocksize = inode->i_sb->s_blocksize;
9087 u32 bi_flags = BTRFS_I(inode)->flags;
9089 stat->result_mask |= STATX_BTIME;
9090 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9091 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9092 if (bi_flags & BTRFS_INODE_APPEND)
9093 stat->attributes |= STATX_ATTR_APPEND;
9094 if (bi_flags & BTRFS_INODE_COMPRESS)
9095 stat->attributes |= STATX_ATTR_COMPRESSED;
9096 if (bi_flags & BTRFS_INODE_IMMUTABLE)
9097 stat->attributes |= STATX_ATTR_IMMUTABLE;
9098 if (bi_flags & BTRFS_INODE_NODUMP)
9099 stat->attributes |= STATX_ATTR_NODUMP;
9101 stat->attributes_mask |= (STATX_ATTR_APPEND |
9102 STATX_ATTR_COMPRESSED |
9103 STATX_ATTR_IMMUTABLE |
9106 generic_fillattr(&init_user_ns, inode, stat);
9107 stat->dev = BTRFS_I(inode)->root->anon_dev;
9109 spin_lock(&BTRFS_I(inode)->lock);
9110 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9111 inode_bytes = inode_get_bytes(inode);
9112 spin_unlock(&BTRFS_I(inode)->lock);
9113 stat->blocks = (ALIGN(inode_bytes, blocksize) +
9114 ALIGN(delalloc_bytes, blocksize)) >> 9;
9118 static int btrfs_rename_exchange(struct inode *old_dir,
9119 struct dentry *old_dentry,
9120 struct inode *new_dir,
9121 struct dentry *new_dentry)
9123 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9124 struct btrfs_trans_handle *trans;
9125 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9126 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9127 struct inode *new_inode = new_dentry->d_inode;
9128 struct inode *old_inode = old_dentry->d_inode;
9129 struct timespec64 ctime = current_time(old_inode);
9130 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9131 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9136 bool root_log_pinned = false;
9137 bool dest_log_pinned = false;
9138 bool need_abort = false;
9140 /* we only allow rename subvolume link between subvolumes */
9141 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9144 /* close the race window with snapshot create/destroy ioctl */
9145 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
9146 new_ino == BTRFS_FIRST_FREE_OBJECTID)
9147 down_read(&fs_info->subvol_sem);
9150 * We want to reserve the absolute worst case amount of items. So if
9151 * both inodes are subvols and we need to unlink them then that would
9152 * require 4 item modifications, but if they are both normal inodes it
9153 * would require 5 item modifications, so we'll assume their normal
9154 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9155 * should cover the worst case number of items we'll modify.
9157 trans = btrfs_start_transaction(root, 12);
9158 if (IS_ERR(trans)) {
9159 ret = PTR_ERR(trans);
9164 ret = btrfs_record_root_in_trans(trans, dest);
9170 * We need to find a free sequence number both in the source and
9171 * in the destination directory for the exchange.
9173 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9176 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9180 BTRFS_I(old_inode)->dir_index = 0ULL;
9181 BTRFS_I(new_inode)->dir_index = 0ULL;
9183 /* Reference for the source. */
9184 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9185 /* force full log commit if subvolume involved. */
9186 btrfs_set_log_full_commit(trans);
9188 btrfs_pin_log_trans(root);
9189 root_log_pinned = true;
9190 ret = btrfs_insert_inode_ref(trans, dest,
9191 new_dentry->d_name.name,
9192 new_dentry->d_name.len,
9194 btrfs_ino(BTRFS_I(new_dir)),
9201 /* And now for the dest. */
9202 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9203 /* force full log commit if subvolume involved. */
9204 btrfs_set_log_full_commit(trans);
9206 btrfs_pin_log_trans(dest);
9207 dest_log_pinned = true;
9208 ret = btrfs_insert_inode_ref(trans, root,
9209 old_dentry->d_name.name,
9210 old_dentry->d_name.len,
9212 btrfs_ino(BTRFS_I(old_dir)),
9216 btrfs_abort_transaction(trans, ret);
9221 /* Update inode version and ctime/mtime. */
9222 inode_inc_iversion(old_dir);
9223 inode_inc_iversion(new_dir);
9224 inode_inc_iversion(old_inode);
9225 inode_inc_iversion(new_inode);
9226 old_dir->i_ctime = old_dir->i_mtime = ctime;
9227 new_dir->i_ctime = new_dir->i_mtime = ctime;
9228 old_inode->i_ctime = ctime;
9229 new_inode->i_ctime = ctime;
9231 if (old_dentry->d_parent != new_dentry->d_parent) {
9232 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9233 BTRFS_I(old_inode), 1);
9234 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9235 BTRFS_I(new_inode), 1);
9238 /* src is a subvolume */
9239 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9240 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9241 } else { /* src is an inode */
9242 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9243 BTRFS_I(old_dentry->d_inode),
9244 old_dentry->d_name.name,
9245 old_dentry->d_name.len);
9247 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9250 btrfs_abort_transaction(trans, ret);
9254 /* dest is a subvolume */
9255 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9256 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9257 } else { /* dest is an inode */
9258 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9259 BTRFS_I(new_dentry->d_inode),
9260 new_dentry->d_name.name,
9261 new_dentry->d_name.len);
9263 ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode));
9266 btrfs_abort_transaction(trans, ret);
9270 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9271 new_dentry->d_name.name,
9272 new_dentry->d_name.len, 0, old_idx);
9274 btrfs_abort_transaction(trans, ret);
9278 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9279 old_dentry->d_name.name,
9280 old_dentry->d_name.len, 0, new_idx);
9282 btrfs_abort_transaction(trans, ret);
9286 if (old_inode->i_nlink == 1)
9287 BTRFS_I(old_inode)->dir_index = old_idx;
9288 if (new_inode->i_nlink == 1)
9289 BTRFS_I(new_inode)->dir_index = new_idx;
9291 if (root_log_pinned) {
9292 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9293 new_dentry->d_parent);
9294 btrfs_end_log_trans(root);
9295 root_log_pinned = false;
9297 if (dest_log_pinned) {
9298 btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir),
9299 old_dentry->d_parent);
9300 btrfs_end_log_trans(dest);
9301 dest_log_pinned = false;
9305 * If we have pinned a log and an error happened, we unpin tasks
9306 * trying to sync the log and force them to fallback to a transaction
9307 * commit if the log currently contains any of the inodes involved in
9308 * this rename operation (to ensure we do not persist a log with an
9309 * inconsistent state for any of these inodes or leading to any
9310 * inconsistencies when replayed). If the transaction was aborted, the
9311 * abortion reason is propagated to userspace when attempting to commit
9312 * the transaction. If the log does not contain any of these inodes, we
9313 * allow the tasks to sync it.
9315 if (ret && (root_log_pinned || dest_log_pinned)) {
9316 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9317 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9318 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9320 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9321 btrfs_set_log_full_commit(trans);
9323 if (root_log_pinned) {
9324 btrfs_end_log_trans(root);
9325 root_log_pinned = false;
9327 if (dest_log_pinned) {
9328 btrfs_end_log_trans(dest);
9329 dest_log_pinned = false;
9332 ret2 = btrfs_end_transaction(trans);
9333 ret = ret ? ret : ret2;
9335 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
9336 old_ino == BTRFS_FIRST_FREE_OBJECTID)
9337 up_read(&fs_info->subvol_sem);
9342 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9343 struct btrfs_root *root,
9345 struct dentry *dentry)
9348 struct inode *inode;
9352 ret = btrfs_get_free_objectid(root, &objectid);
9356 inode = btrfs_new_inode(trans, root, dir,
9357 dentry->d_name.name,
9359 btrfs_ino(BTRFS_I(dir)),
9361 S_IFCHR | WHITEOUT_MODE,
9364 if (IS_ERR(inode)) {
9365 ret = PTR_ERR(inode);
9369 inode->i_op = &btrfs_special_inode_operations;
9370 init_special_inode(inode, inode->i_mode,
9373 ret = btrfs_init_inode_security(trans, inode, dir,
9378 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9379 BTRFS_I(inode), 0, index);
9383 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
9385 unlock_new_inode(inode);
9387 inode_dec_link_count(inode);
9393 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9394 struct inode *new_dir, struct dentry *new_dentry,
9397 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9398 struct btrfs_trans_handle *trans;
9399 unsigned int trans_num_items;
9400 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9401 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9402 struct inode *new_inode = d_inode(new_dentry);
9403 struct inode *old_inode = d_inode(old_dentry);
9407 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9408 bool log_pinned = false;
9410 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9413 /* we only allow rename subvolume link between subvolumes */
9414 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9417 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9418 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9421 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9422 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9426 /* check for collisions, even if the name isn't there */
9427 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9428 new_dentry->d_name.name,
9429 new_dentry->d_name.len);
9432 if (ret == -EEXIST) {
9434 * eexist without a new_inode */
9435 if (WARN_ON(!new_inode)) {
9439 /* maybe -EOVERFLOW */
9446 * we're using rename to replace one file with another. Start IO on it
9447 * now so we don't add too much work to the end of the transaction
9449 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9450 filemap_flush(old_inode->i_mapping);
9452 /* close the racy window with snapshot create/destroy ioctl */
9453 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9454 down_read(&fs_info->subvol_sem);
9456 * We want to reserve the absolute worst case amount of items. So if
9457 * both inodes are subvols and we need to unlink them then that would
9458 * require 4 item modifications, but if they are both normal inodes it
9459 * would require 5 item modifications, so we'll assume they are normal
9460 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9461 * should cover the worst case number of items we'll modify.
9462 * If our rename has the whiteout flag, we need more 5 units for the
9463 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9464 * when selinux is enabled).
9466 trans_num_items = 11;
9467 if (flags & RENAME_WHITEOUT)
9468 trans_num_items += 5;
9469 trans = btrfs_start_transaction(root, trans_num_items);
9470 if (IS_ERR(trans)) {
9471 ret = PTR_ERR(trans);
9476 ret = btrfs_record_root_in_trans(trans, dest);
9481 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9485 BTRFS_I(old_inode)->dir_index = 0ULL;
9486 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9487 /* force full log commit if subvolume involved. */
9488 btrfs_set_log_full_commit(trans);
9490 btrfs_pin_log_trans(root);
9492 ret = btrfs_insert_inode_ref(trans, dest,
9493 new_dentry->d_name.name,
9494 new_dentry->d_name.len,
9496 btrfs_ino(BTRFS_I(new_dir)), index);
9501 inode_inc_iversion(old_dir);
9502 inode_inc_iversion(new_dir);
9503 inode_inc_iversion(old_inode);
9504 old_dir->i_ctime = old_dir->i_mtime =
9505 new_dir->i_ctime = new_dir->i_mtime =
9506 old_inode->i_ctime = current_time(old_dir);
9508 if (old_dentry->d_parent != new_dentry->d_parent)
9509 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9510 BTRFS_I(old_inode), 1);
9512 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9513 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9515 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9516 BTRFS_I(d_inode(old_dentry)),
9517 old_dentry->d_name.name,
9518 old_dentry->d_name.len);
9520 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9523 btrfs_abort_transaction(trans, ret);
9528 inode_inc_iversion(new_inode);
9529 new_inode->i_ctime = current_time(new_inode);
9530 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9531 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9532 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9533 BUG_ON(new_inode->i_nlink == 0);
9535 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9536 BTRFS_I(d_inode(new_dentry)),
9537 new_dentry->d_name.name,
9538 new_dentry->d_name.len);
9540 if (!ret && new_inode->i_nlink == 0)
9541 ret = btrfs_orphan_add(trans,
9542 BTRFS_I(d_inode(new_dentry)));
9544 btrfs_abort_transaction(trans, ret);
9549 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9550 new_dentry->d_name.name,
9551 new_dentry->d_name.len, 0, index);
9553 btrfs_abort_transaction(trans, ret);
9557 if (old_inode->i_nlink == 1)
9558 BTRFS_I(old_inode)->dir_index = index;
9561 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9562 new_dentry->d_parent);
9563 btrfs_end_log_trans(root);
9567 if (flags & RENAME_WHITEOUT) {
9568 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9572 btrfs_abort_transaction(trans, ret);
9578 * If we have pinned the log and an error happened, we unpin tasks
9579 * trying to sync the log and force them to fallback to a transaction
9580 * commit if the log currently contains any of the inodes involved in
9581 * this rename operation (to ensure we do not persist a log with an
9582 * inconsistent state for any of these inodes or leading to any
9583 * inconsistencies when replayed). If the transaction was aborted, the
9584 * abortion reason is propagated to userspace when attempting to commit
9585 * the transaction. If the log does not contain any of these inodes, we
9586 * allow the tasks to sync it.
9588 if (ret && log_pinned) {
9589 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9590 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9591 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9593 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9594 btrfs_set_log_full_commit(trans);
9596 btrfs_end_log_trans(root);
9599 ret2 = btrfs_end_transaction(trans);
9600 ret = ret ? ret : ret2;
9602 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9603 up_read(&fs_info->subvol_sem);
9608 static int btrfs_rename2(struct user_namespace *mnt_userns, struct inode *old_dir,
9609 struct dentry *old_dentry, struct inode *new_dir,
9610 struct dentry *new_dentry, unsigned int flags)
9612 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9615 if (flags & RENAME_EXCHANGE)
9616 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9619 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9622 struct btrfs_delalloc_work {
9623 struct inode *inode;
9624 struct completion completion;
9625 struct list_head list;
9626 struct btrfs_work work;
9629 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9631 struct btrfs_delalloc_work *delalloc_work;
9632 struct inode *inode;
9634 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9636 inode = delalloc_work->inode;
9637 filemap_flush(inode->i_mapping);
9638 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9639 &BTRFS_I(inode)->runtime_flags))
9640 filemap_flush(inode->i_mapping);
9643 complete(&delalloc_work->completion);
9646 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9648 struct btrfs_delalloc_work *work;
9650 work = kmalloc(sizeof(*work), GFP_NOFS);
9654 init_completion(&work->completion);
9655 INIT_LIST_HEAD(&work->list);
9656 work->inode = inode;
9657 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9663 * some fairly slow code that needs optimization. This walks the list
9664 * of all the inodes with pending delalloc and forces them to disk.
9666 static int start_delalloc_inodes(struct btrfs_root *root,
9667 struct writeback_control *wbc, bool snapshot,
9668 bool in_reclaim_context)
9670 struct btrfs_inode *binode;
9671 struct inode *inode;
9672 struct btrfs_delalloc_work *work, *next;
9673 struct list_head works;
9674 struct list_head splice;
9676 bool full_flush = wbc->nr_to_write == LONG_MAX;
9678 INIT_LIST_HEAD(&works);
9679 INIT_LIST_HEAD(&splice);
9681 mutex_lock(&root->delalloc_mutex);
9682 spin_lock(&root->delalloc_lock);
9683 list_splice_init(&root->delalloc_inodes, &splice);
9684 while (!list_empty(&splice)) {
9685 binode = list_entry(splice.next, struct btrfs_inode,
9688 list_move_tail(&binode->delalloc_inodes,
9689 &root->delalloc_inodes);
9691 if (in_reclaim_context &&
9692 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9695 inode = igrab(&binode->vfs_inode);
9697 cond_resched_lock(&root->delalloc_lock);
9700 spin_unlock(&root->delalloc_lock);
9703 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9704 &binode->runtime_flags);
9706 work = btrfs_alloc_delalloc_work(inode);
9712 list_add_tail(&work->list, &works);
9713 btrfs_queue_work(root->fs_info->flush_workers,
9716 ret = sync_inode(inode, wbc);
9718 test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9719 &BTRFS_I(inode)->runtime_flags))
9720 ret = sync_inode(inode, wbc);
9721 btrfs_add_delayed_iput(inode);
9722 if (ret || wbc->nr_to_write <= 0)
9726 spin_lock(&root->delalloc_lock);
9728 spin_unlock(&root->delalloc_lock);
9731 list_for_each_entry_safe(work, next, &works, list) {
9732 list_del_init(&work->list);
9733 wait_for_completion(&work->completion);
9737 if (!list_empty(&splice)) {
9738 spin_lock(&root->delalloc_lock);
9739 list_splice_tail(&splice, &root->delalloc_inodes);
9740 spin_unlock(&root->delalloc_lock);
9742 mutex_unlock(&root->delalloc_mutex);
9746 int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9748 struct writeback_control wbc = {
9749 .nr_to_write = LONG_MAX,
9750 .sync_mode = WB_SYNC_NONE,
9752 .range_end = LLONG_MAX,
9754 struct btrfs_fs_info *fs_info = root->fs_info;
9756 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9759 return start_delalloc_inodes(root, &wbc, true, in_reclaim_context);
9762 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9763 bool in_reclaim_context)
9765 struct writeback_control wbc = {
9767 .sync_mode = WB_SYNC_NONE,
9769 .range_end = LLONG_MAX,
9771 struct btrfs_root *root;
9772 struct list_head splice;
9775 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9778 INIT_LIST_HEAD(&splice);
9780 mutex_lock(&fs_info->delalloc_root_mutex);
9781 spin_lock(&fs_info->delalloc_root_lock);
9782 list_splice_init(&fs_info->delalloc_roots, &splice);
9783 while (!list_empty(&splice)) {
9785 * Reset nr_to_write here so we know that we're doing a full
9789 wbc.nr_to_write = LONG_MAX;
9791 root = list_first_entry(&splice, struct btrfs_root,
9793 root = btrfs_grab_root(root);
9795 list_move_tail(&root->delalloc_root,
9796 &fs_info->delalloc_roots);
9797 spin_unlock(&fs_info->delalloc_root_lock);
9799 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9800 btrfs_put_root(root);
9801 if (ret < 0 || wbc.nr_to_write <= 0)
9803 spin_lock(&fs_info->delalloc_root_lock);
9805 spin_unlock(&fs_info->delalloc_root_lock);
9809 if (!list_empty(&splice)) {
9810 spin_lock(&fs_info->delalloc_root_lock);
9811 list_splice_tail(&splice, &fs_info->delalloc_roots);
9812 spin_unlock(&fs_info->delalloc_root_lock);
9814 mutex_unlock(&fs_info->delalloc_root_mutex);
9818 static int btrfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
9819 struct dentry *dentry, const char *symname)
9821 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9822 struct btrfs_trans_handle *trans;
9823 struct btrfs_root *root = BTRFS_I(dir)->root;
9824 struct btrfs_path *path;
9825 struct btrfs_key key;
9826 struct inode *inode = NULL;
9833 struct btrfs_file_extent_item *ei;
9834 struct extent_buffer *leaf;
9836 name_len = strlen(symname);
9837 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9838 return -ENAMETOOLONG;
9841 * 2 items for inode item and ref
9842 * 2 items for dir items
9843 * 1 item for updating parent inode item
9844 * 1 item for the inline extent item
9845 * 1 item for xattr if selinux is on
9847 trans = btrfs_start_transaction(root, 7);
9849 return PTR_ERR(trans);
9851 err = btrfs_get_free_objectid(root, &objectid);
9855 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9856 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
9857 objectid, S_IFLNK|S_IRWXUGO, &index);
9858 if (IS_ERR(inode)) {
9859 err = PTR_ERR(inode);
9865 * If the active LSM wants to access the inode during
9866 * d_instantiate it needs these. Smack checks to see
9867 * if the filesystem supports xattrs by looking at the
9870 inode->i_fop = &btrfs_file_operations;
9871 inode->i_op = &btrfs_file_inode_operations;
9872 inode->i_mapping->a_ops = &btrfs_aops;
9874 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9878 path = btrfs_alloc_path();
9883 key.objectid = btrfs_ino(BTRFS_I(inode));
9885 key.type = BTRFS_EXTENT_DATA_KEY;
9886 datasize = btrfs_file_extent_calc_inline_size(name_len);
9887 err = btrfs_insert_empty_item(trans, root, path, &key,
9890 btrfs_free_path(path);
9893 leaf = path->nodes[0];
9894 ei = btrfs_item_ptr(leaf, path->slots[0],
9895 struct btrfs_file_extent_item);
9896 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9897 btrfs_set_file_extent_type(leaf, ei,
9898 BTRFS_FILE_EXTENT_INLINE);
9899 btrfs_set_file_extent_encryption(leaf, ei, 0);
9900 btrfs_set_file_extent_compression(leaf, ei, 0);
9901 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9902 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9904 ptr = btrfs_file_extent_inline_start(ei);
9905 write_extent_buffer(leaf, symname, ptr, name_len);
9906 btrfs_mark_buffer_dirty(leaf);
9907 btrfs_free_path(path);
9909 inode->i_op = &btrfs_symlink_inode_operations;
9910 inode_nohighmem(inode);
9911 inode_set_bytes(inode, name_len);
9912 btrfs_i_size_write(BTRFS_I(inode), name_len);
9913 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
9915 * Last step, add directory indexes for our symlink inode. This is the
9916 * last step to avoid extra cleanup of these indexes if an error happens
9920 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9921 BTRFS_I(inode), 0, index);
9925 d_instantiate_new(dentry, inode);
9928 btrfs_end_transaction(trans);
9930 inode_dec_link_count(inode);
9931 discard_new_inode(inode);
9933 btrfs_btree_balance_dirty(fs_info);
9937 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9938 struct btrfs_trans_handle *trans_in,
9939 struct btrfs_inode *inode,
9940 struct btrfs_key *ins,
9943 struct btrfs_file_extent_item stack_fi;
9944 struct btrfs_replace_extent_info extent_info;
9945 struct btrfs_trans_handle *trans = trans_in;
9946 struct btrfs_path *path;
9947 u64 start = ins->objectid;
9948 u64 len = ins->offset;
9949 int qgroup_released;
9952 memset(&stack_fi, 0, sizeof(stack_fi));
9954 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9955 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9956 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9957 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9958 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9959 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9960 /* Encryption and other encoding is reserved and all 0 */
9962 qgroup_released = btrfs_qgroup_release_data(inode, file_offset, len);
9963 if (qgroup_released < 0)
9964 return ERR_PTR(qgroup_released);
9967 ret = insert_reserved_file_extent(trans, inode,
9968 file_offset, &stack_fi,
9969 true, qgroup_released);
9975 extent_info.disk_offset = start;
9976 extent_info.disk_len = len;
9977 extent_info.data_offset = 0;
9978 extent_info.data_len = len;
9979 extent_info.file_offset = file_offset;
9980 extent_info.extent_buf = (char *)&stack_fi;
9981 extent_info.is_new_extent = true;
9982 extent_info.qgroup_reserved = qgroup_released;
9983 extent_info.insertions = 0;
9985 path = btrfs_alloc_path();
9991 ret = btrfs_replace_file_extents(inode, path, file_offset,
9992 file_offset + len - 1, &extent_info,
9994 btrfs_free_path(path);
10001 * We have released qgroup data range at the beginning of the function,
10002 * and normally qgroup_released bytes will be freed when committing
10004 * But if we error out early, we have to free what we have released
10005 * or we leak qgroup data reservation.
10007 btrfs_qgroup_free_refroot(inode->root->fs_info,
10008 inode->root->root_key.objectid, qgroup_released,
10009 BTRFS_QGROUP_RSV_DATA);
10010 return ERR_PTR(ret);
10013 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10014 u64 start, u64 num_bytes, u64 min_size,
10015 loff_t actual_len, u64 *alloc_hint,
10016 struct btrfs_trans_handle *trans)
10018 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
10019 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10020 struct extent_map *em;
10021 struct btrfs_root *root = BTRFS_I(inode)->root;
10022 struct btrfs_key ins;
10023 u64 cur_offset = start;
10024 u64 clear_offset = start;
10027 u64 last_alloc = (u64)-1;
10029 bool own_trans = true;
10030 u64 end = start + num_bytes - 1;
10034 while (num_bytes > 0) {
10035 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10036 cur_bytes = max(cur_bytes, min_size);
10038 * If we are severely fragmented we could end up with really
10039 * small allocations, so if the allocator is returning small
10040 * chunks lets make its job easier by only searching for those
10043 cur_bytes = min(cur_bytes, last_alloc);
10044 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10045 min_size, 0, *alloc_hint, &ins, 1, 0);
10050 * We've reserved this space, and thus converted it from
10051 * ->bytes_may_use to ->bytes_reserved. Any error that happens
10052 * from here on out we will only need to clear our reservation
10053 * for the remaining unreserved area, so advance our
10054 * clear_offset by our extent size.
10056 clear_offset += ins.offset;
10058 last_alloc = ins.offset;
10059 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
10062 * Now that we inserted the prealloc extent we can finally
10063 * decrement the number of reservations in the block group.
10064 * If we did it before, we could race with relocation and have
10065 * relocation miss the reserved extent, making it fail later.
10067 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10068 if (IS_ERR(trans)) {
10069 ret = PTR_ERR(trans);
10070 btrfs_free_reserved_extent(fs_info, ins.objectid,
10075 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10076 cur_offset + ins.offset -1, 0);
10078 em = alloc_extent_map();
10080 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10081 &BTRFS_I(inode)->runtime_flags);
10085 em->start = cur_offset;
10086 em->orig_start = cur_offset;
10087 em->len = ins.offset;
10088 em->block_start = ins.objectid;
10089 em->block_len = ins.offset;
10090 em->orig_block_len = ins.offset;
10091 em->ram_bytes = ins.offset;
10092 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10093 em->generation = trans->transid;
10096 write_lock(&em_tree->lock);
10097 ret = add_extent_mapping(em_tree, em, 1);
10098 write_unlock(&em_tree->lock);
10099 if (ret != -EEXIST)
10101 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10102 cur_offset + ins.offset - 1,
10105 free_extent_map(em);
10107 num_bytes -= ins.offset;
10108 cur_offset += ins.offset;
10109 *alloc_hint = ins.objectid + ins.offset;
10111 inode_inc_iversion(inode);
10112 inode->i_ctime = current_time(inode);
10113 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10114 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10115 (actual_len > inode->i_size) &&
10116 (cur_offset > inode->i_size)) {
10117 if (cur_offset > actual_len)
10118 i_size = actual_len;
10120 i_size = cur_offset;
10121 i_size_write(inode, i_size);
10122 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
10125 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
10128 btrfs_abort_transaction(trans, ret);
10130 btrfs_end_transaction(trans);
10135 btrfs_end_transaction(trans);
10139 if (clear_offset < end)
10140 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
10141 end - clear_offset + 1);
10145 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10146 u64 start, u64 num_bytes, u64 min_size,
10147 loff_t actual_len, u64 *alloc_hint)
10149 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10150 min_size, actual_len, alloc_hint,
10154 int btrfs_prealloc_file_range_trans(struct inode *inode,
10155 struct btrfs_trans_handle *trans, int mode,
10156 u64 start, u64 num_bytes, u64 min_size,
10157 loff_t actual_len, u64 *alloc_hint)
10159 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10160 min_size, actual_len, alloc_hint, trans);
10163 static int btrfs_set_page_dirty(struct page *page)
10165 return __set_page_dirty_nobuffers(page);
10168 static int btrfs_permission(struct user_namespace *mnt_userns,
10169 struct inode *inode, int mask)
10171 struct btrfs_root *root = BTRFS_I(inode)->root;
10172 umode_t mode = inode->i_mode;
10174 if (mask & MAY_WRITE &&
10175 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10176 if (btrfs_root_readonly(root))
10178 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10181 return generic_permission(&init_user_ns, inode, mask);
10184 static int btrfs_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
10185 struct dentry *dentry, umode_t mode)
10187 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10188 struct btrfs_trans_handle *trans;
10189 struct btrfs_root *root = BTRFS_I(dir)->root;
10190 struct inode *inode = NULL;
10196 * 5 units required for adding orphan entry
10198 trans = btrfs_start_transaction(root, 5);
10200 return PTR_ERR(trans);
10202 ret = btrfs_get_free_objectid(root, &objectid);
10206 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10207 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10208 if (IS_ERR(inode)) {
10209 ret = PTR_ERR(inode);
10214 inode->i_fop = &btrfs_file_operations;
10215 inode->i_op = &btrfs_file_inode_operations;
10217 inode->i_mapping->a_ops = &btrfs_aops;
10219 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10223 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
10226 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10231 * We set number of links to 0 in btrfs_new_inode(), and here we set
10232 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10235 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10237 set_nlink(inode, 1);
10238 d_tmpfile(dentry, inode);
10239 unlock_new_inode(inode);
10240 mark_inode_dirty(inode);
10242 btrfs_end_transaction(trans);
10244 discard_new_inode(inode);
10245 btrfs_btree_balance_dirty(fs_info);
10249 void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end)
10251 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10252 unsigned long index = start >> PAGE_SHIFT;
10253 unsigned long end_index = end >> PAGE_SHIFT;
10257 ASSERT(end + 1 - start <= U32_MAX);
10258 len = end + 1 - start;
10259 while (index <= end_index) {
10260 page = find_get_page(inode->vfs_inode.i_mapping, index);
10261 ASSERT(page); /* Pages should be in the extent_io_tree */
10263 btrfs_page_set_writeback(fs_info, page, start, len);
10271 * Add an entry indicating a block group or device which is pinned by a
10272 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10273 * negative errno on failure.
10275 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10276 bool is_block_group)
10278 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10279 struct btrfs_swapfile_pin *sp, *entry;
10280 struct rb_node **p;
10281 struct rb_node *parent = NULL;
10283 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10288 sp->is_block_group = is_block_group;
10289 sp->bg_extent_count = 1;
10291 spin_lock(&fs_info->swapfile_pins_lock);
10292 p = &fs_info->swapfile_pins.rb_node;
10295 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10296 if (sp->ptr < entry->ptr ||
10297 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10298 p = &(*p)->rb_left;
10299 } else if (sp->ptr > entry->ptr ||
10300 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10301 p = &(*p)->rb_right;
10303 if (is_block_group)
10304 entry->bg_extent_count++;
10305 spin_unlock(&fs_info->swapfile_pins_lock);
10310 rb_link_node(&sp->node, parent, p);
10311 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10312 spin_unlock(&fs_info->swapfile_pins_lock);
10316 /* Free all of the entries pinned by this swapfile. */
10317 static void btrfs_free_swapfile_pins(struct inode *inode)
10319 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10320 struct btrfs_swapfile_pin *sp;
10321 struct rb_node *node, *next;
10323 spin_lock(&fs_info->swapfile_pins_lock);
10324 node = rb_first(&fs_info->swapfile_pins);
10326 next = rb_next(node);
10327 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10328 if (sp->inode == inode) {
10329 rb_erase(&sp->node, &fs_info->swapfile_pins);
10330 if (sp->is_block_group) {
10331 btrfs_dec_block_group_swap_extents(sp->ptr,
10332 sp->bg_extent_count);
10333 btrfs_put_block_group(sp->ptr);
10339 spin_unlock(&fs_info->swapfile_pins_lock);
10342 struct btrfs_swap_info {
10348 unsigned long nr_pages;
10352 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10353 struct btrfs_swap_info *bsi)
10355 unsigned long nr_pages;
10356 u64 first_ppage, first_ppage_reported, next_ppage;
10359 first_ppage = ALIGN(bsi->block_start, PAGE_SIZE) >> PAGE_SHIFT;
10360 next_ppage = ALIGN_DOWN(bsi->block_start + bsi->block_len,
10361 PAGE_SIZE) >> PAGE_SHIFT;
10363 if (first_ppage >= next_ppage)
10365 nr_pages = next_ppage - first_ppage;
10367 first_ppage_reported = first_ppage;
10368 if (bsi->start == 0)
10369 first_ppage_reported++;
10370 if (bsi->lowest_ppage > first_ppage_reported)
10371 bsi->lowest_ppage = first_ppage_reported;
10372 if (bsi->highest_ppage < (next_ppage - 1))
10373 bsi->highest_ppage = next_ppage - 1;
10375 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10378 bsi->nr_extents += ret;
10379 bsi->nr_pages += nr_pages;
10383 static void btrfs_swap_deactivate(struct file *file)
10385 struct inode *inode = file_inode(file);
10387 btrfs_free_swapfile_pins(inode);
10388 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10391 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10394 struct inode *inode = file_inode(file);
10395 struct btrfs_root *root = BTRFS_I(inode)->root;
10396 struct btrfs_fs_info *fs_info = root->fs_info;
10397 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10398 struct extent_state *cached_state = NULL;
10399 struct extent_map *em = NULL;
10400 struct btrfs_device *device = NULL;
10401 struct btrfs_swap_info bsi = {
10402 .lowest_ppage = (sector_t)-1ULL,
10409 * If the swap file was just created, make sure delalloc is done. If the
10410 * file changes again after this, the user is doing something stupid and
10411 * we don't really care.
10413 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10418 * The inode is locked, so these flags won't change after we check them.
10420 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10421 btrfs_warn(fs_info, "swapfile must not be compressed");
10424 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10425 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10428 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10429 btrfs_warn(fs_info, "swapfile must not be checksummed");
10434 * Balance or device remove/replace/resize can move stuff around from
10435 * under us. The exclop protection makes sure they aren't running/won't
10436 * run concurrently while we are mapping the swap extents, and
10437 * fs_info->swapfile_pins prevents them from running while the swap
10438 * file is active and moving the extents. Note that this also prevents
10439 * a concurrent device add which isn't actually necessary, but it's not
10440 * really worth the trouble to allow it.
10442 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10443 btrfs_warn(fs_info,
10444 "cannot activate swapfile while exclusive operation is running");
10449 * Prevent snapshot creation while we are activating the swap file.
10450 * We do not want to race with snapshot creation. If snapshot creation
10451 * already started before we bumped nr_swapfiles from 0 to 1 and
10452 * completes before the first write into the swap file after it is
10453 * activated, than that write would fallback to COW.
10455 if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) {
10456 btrfs_exclop_finish(fs_info);
10457 btrfs_warn(fs_info,
10458 "cannot activate swapfile because snapshot creation is in progress");
10462 * Snapshots can create extents which require COW even if NODATACOW is
10463 * set. We use this counter to prevent snapshots. We must increment it
10464 * before walking the extents because we don't want a concurrent
10465 * snapshot to run after we've already checked the extents.
10467 atomic_inc(&root->nr_swapfiles);
10469 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10471 lock_extent_bits(io_tree, 0, isize - 1, &cached_state);
10473 while (start < isize) {
10474 u64 logical_block_start, physical_block_start;
10475 struct btrfs_block_group *bg;
10476 u64 len = isize - start;
10478 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
10484 if (em->block_start == EXTENT_MAP_HOLE) {
10485 btrfs_warn(fs_info, "swapfile must not have holes");
10489 if (em->block_start == EXTENT_MAP_INLINE) {
10491 * It's unlikely we'll ever actually find ourselves
10492 * here, as a file small enough to fit inline won't be
10493 * big enough to store more than the swap header, but in
10494 * case something changes in the future, let's catch it
10495 * here rather than later.
10497 btrfs_warn(fs_info, "swapfile must not be inline");
10501 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10502 btrfs_warn(fs_info, "swapfile must not be compressed");
10507 logical_block_start = em->block_start + (start - em->start);
10508 len = min(len, em->len - (start - em->start));
10509 free_extent_map(em);
10512 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, true);
10518 btrfs_warn(fs_info,
10519 "swapfile must not be copy-on-write");
10524 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10530 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10531 btrfs_warn(fs_info,
10532 "swapfile must have single data profile");
10537 if (device == NULL) {
10538 device = em->map_lookup->stripes[0].dev;
10539 ret = btrfs_add_swapfile_pin(inode, device, false);
10544 } else if (device != em->map_lookup->stripes[0].dev) {
10545 btrfs_warn(fs_info, "swapfile must be on one device");
10550 physical_block_start = (em->map_lookup->stripes[0].physical +
10551 (logical_block_start - em->start));
10552 len = min(len, em->len - (logical_block_start - em->start));
10553 free_extent_map(em);
10556 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10558 btrfs_warn(fs_info,
10559 "could not find block group containing swapfile");
10564 if (!btrfs_inc_block_group_swap_extents(bg)) {
10565 btrfs_warn(fs_info,
10566 "block group for swapfile at %llu is read-only%s",
10568 atomic_read(&fs_info->scrubs_running) ?
10569 " (scrub running)" : "");
10570 btrfs_put_block_group(bg);
10575 ret = btrfs_add_swapfile_pin(inode, bg, true);
10577 btrfs_put_block_group(bg);
10584 if (bsi.block_len &&
10585 bsi.block_start + bsi.block_len == physical_block_start) {
10586 bsi.block_len += len;
10588 if (bsi.block_len) {
10589 ret = btrfs_add_swap_extent(sis, &bsi);
10594 bsi.block_start = physical_block_start;
10595 bsi.block_len = len;
10602 ret = btrfs_add_swap_extent(sis, &bsi);
10605 if (!IS_ERR_OR_NULL(em))
10606 free_extent_map(em);
10608 unlock_extent_cached(io_tree, 0, isize - 1, &cached_state);
10611 btrfs_swap_deactivate(file);
10613 btrfs_drew_write_unlock(&root->snapshot_lock);
10615 btrfs_exclop_finish(fs_info);
10621 sis->bdev = device->bdev;
10622 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10623 sis->max = bsi.nr_pages;
10624 sis->pages = bsi.nr_pages - 1;
10625 sis->highest_bit = bsi.nr_pages - 1;
10626 return bsi.nr_extents;
10629 static void btrfs_swap_deactivate(struct file *file)
10633 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10636 return -EOPNOTSUPP;
10641 * Update the number of bytes used in the VFS' inode. When we replace extents in
10642 * a range (clone, dedupe, fallocate's zero range), we must update the number of
10643 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
10644 * always get a correct value.
10646 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
10647 const u64 add_bytes,
10648 const u64 del_bytes)
10650 if (add_bytes == del_bytes)
10653 spin_lock(&inode->lock);
10655 inode_sub_bytes(&inode->vfs_inode, del_bytes);
10657 inode_add_bytes(&inode->vfs_inode, add_bytes);
10658 spin_unlock(&inode->lock);
10661 static const struct inode_operations btrfs_dir_inode_operations = {
10662 .getattr = btrfs_getattr,
10663 .lookup = btrfs_lookup,
10664 .create = btrfs_create,
10665 .unlink = btrfs_unlink,
10666 .link = btrfs_link,
10667 .mkdir = btrfs_mkdir,
10668 .rmdir = btrfs_rmdir,
10669 .rename = btrfs_rename2,
10670 .symlink = btrfs_symlink,
10671 .setattr = btrfs_setattr,
10672 .mknod = btrfs_mknod,
10673 .listxattr = btrfs_listxattr,
10674 .permission = btrfs_permission,
10675 .get_acl = btrfs_get_acl,
10676 .set_acl = btrfs_set_acl,
10677 .update_time = btrfs_update_time,
10678 .tmpfile = btrfs_tmpfile,
10679 .fileattr_get = btrfs_fileattr_get,
10680 .fileattr_set = btrfs_fileattr_set,
10683 static const struct file_operations btrfs_dir_file_operations = {
10684 .llseek = generic_file_llseek,
10685 .read = generic_read_dir,
10686 .iterate_shared = btrfs_real_readdir,
10687 .open = btrfs_opendir,
10688 .unlocked_ioctl = btrfs_ioctl,
10689 #ifdef CONFIG_COMPAT
10690 .compat_ioctl = btrfs_compat_ioctl,
10692 .release = btrfs_release_file,
10693 .fsync = btrfs_sync_file,
10697 * btrfs doesn't support the bmap operation because swapfiles
10698 * use bmap to make a mapping of extents in the file. They assume
10699 * these extents won't change over the life of the file and they
10700 * use the bmap result to do IO directly to the drive.
10702 * the btrfs bmap call would return logical addresses that aren't
10703 * suitable for IO and they also will change frequently as COW
10704 * operations happen. So, swapfile + btrfs == corruption.
10706 * For now we're avoiding this by dropping bmap.
10708 static const struct address_space_operations btrfs_aops = {
10709 .readpage = btrfs_readpage,
10710 .writepage = btrfs_writepage,
10711 .writepages = btrfs_writepages,
10712 .readahead = btrfs_readahead,
10713 .direct_IO = noop_direct_IO,
10714 .invalidatepage = btrfs_invalidatepage,
10715 .releasepage = btrfs_releasepage,
10716 #ifdef CONFIG_MIGRATION
10717 .migratepage = btrfs_migratepage,
10719 .set_page_dirty = btrfs_set_page_dirty,
10720 .error_remove_page = generic_error_remove_page,
10721 .swap_activate = btrfs_swap_activate,
10722 .swap_deactivate = btrfs_swap_deactivate,
10725 static const struct inode_operations btrfs_file_inode_operations = {
10726 .getattr = btrfs_getattr,
10727 .setattr = btrfs_setattr,
10728 .listxattr = btrfs_listxattr,
10729 .permission = btrfs_permission,
10730 .fiemap = btrfs_fiemap,
10731 .get_acl = btrfs_get_acl,
10732 .set_acl = btrfs_set_acl,
10733 .update_time = btrfs_update_time,
10734 .fileattr_get = btrfs_fileattr_get,
10735 .fileattr_set = btrfs_fileattr_set,
10737 static const struct inode_operations btrfs_special_inode_operations = {
10738 .getattr = btrfs_getattr,
10739 .setattr = btrfs_setattr,
10740 .permission = btrfs_permission,
10741 .listxattr = btrfs_listxattr,
10742 .get_acl = btrfs_get_acl,
10743 .set_acl = btrfs_set_acl,
10744 .update_time = btrfs_update_time,
10746 static const struct inode_operations btrfs_symlink_inode_operations = {
10747 .get_link = page_get_link,
10748 .getattr = btrfs_getattr,
10749 .setattr = btrfs_setattr,
10750 .permission = btrfs_permission,
10751 .listxattr = btrfs_listxattr,
10752 .update_time = btrfs_update_time,
10755 const struct dentry_operations btrfs_dentry_operations = {
10756 .d_delete = btrfs_dentry_delete,