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/blk-cgroup.h>
10 #include <linux/file.h>
12 #include <linux/pagemap.h>
13 #include <linux/highmem.h>
14 #include <linux/time.h>
15 #include <linux/init.h>
16 #include <linux/string.h>
17 #include <linux/backing-dev.h>
18 #include <linux/writeback.h>
19 #include <linux/compat.h>
20 #include <linux/xattr.h>
21 #include <linux/posix_acl.h>
22 #include <linux/falloc.h>
23 #include <linux/slab.h>
24 #include <linux/ratelimit.h>
25 #include <linux/btrfs.h>
26 #include <linux/blkdev.h>
27 #include <linux/posix_acl_xattr.h>
28 #include <linux/uio.h>
29 #include <linux/magic.h>
30 #include <linux/iversion.h>
31 #include <linux/swap.h>
32 #include <linux/migrate.h>
33 #include <linux/sched/mm.h>
34 #include <linux/iomap.h>
35 #include <asm/unaligned.h>
36 #include <linux/fsverity.h>
40 #include "transaction.h"
41 #include "btrfs_inode.h"
42 #include "print-tree.h"
43 #include "ordered-data.h"
47 #include "compression.h"
49 #include "free-space-cache.h"
52 #include "delalloc-space.h"
53 #include "block-group.h"
54 #include "space-info.h"
57 #include "inode-item.h"
59 struct btrfs_iget_args {
61 struct btrfs_root *root;
64 struct btrfs_dio_data {
66 struct extent_changeset *data_reserved;
69 struct btrfs_rename_ctx {
70 /* Output field. Stores the index number of the old directory entry. */
74 static const struct inode_operations btrfs_dir_inode_operations;
75 static const struct inode_operations btrfs_symlink_inode_operations;
76 static const struct inode_operations btrfs_special_inode_operations;
77 static const struct inode_operations btrfs_file_inode_operations;
78 static const struct address_space_operations btrfs_aops;
79 static const struct file_operations btrfs_dir_file_operations;
81 static struct kmem_cache *btrfs_inode_cachep;
82 struct kmem_cache *btrfs_trans_handle_cachep;
83 struct kmem_cache *btrfs_path_cachep;
84 struct kmem_cache *btrfs_free_space_cachep;
85 struct kmem_cache *btrfs_free_space_bitmap_cachep;
87 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
88 static int btrfs_truncate(struct inode *inode, bool skip_writeback);
89 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
90 static noinline int cow_file_range(struct btrfs_inode *inode,
91 struct page *locked_page,
92 u64 start, u64 end, int *page_started,
93 unsigned long *nr_written, int unlock);
94 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
95 u64 len, u64 orig_start, u64 block_start,
96 u64 block_len, u64 orig_block_len,
97 u64 ram_bytes, int compress_type,
100 static void __endio_write_update_ordered(struct btrfs_inode *inode,
101 const u64 offset, const u64 bytes,
102 const bool uptodate);
105 * btrfs_inode_lock - lock inode i_rwsem based on arguments passed
107 * ilock_flags can have the following bit set:
109 * BTRFS_ILOCK_SHARED - acquire a shared lock on the inode
110 * BTRFS_ILOCK_TRY - try to acquire the lock, if fails on first attempt
112 * BTRFS_ILOCK_MMAP - acquire a write lock on the i_mmap_lock
114 int btrfs_inode_lock(struct inode *inode, unsigned int ilock_flags)
116 if (ilock_flags & BTRFS_ILOCK_SHARED) {
117 if (ilock_flags & BTRFS_ILOCK_TRY) {
118 if (!inode_trylock_shared(inode))
123 inode_lock_shared(inode);
125 if (ilock_flags & BTRFS_ILOCK_TRY) {
126 if (!inode_trylock(inode))
133 if (ilock_flags & BTRFS_ILOCK_MMAP)
134 down_write(&BTRFS_I(inode)->i_mmap_lock);
139 * btrfs_inode_unlock - unock inode i_rwsem
141 * ilock_flags should contain the same bits set as passed to btrfs_inode_lock()
142 * to decide whether the lock acquired is shared or exclusive.
144 void btrfs_inode_unlock(struct inode *inode, unsigned int ilock_flags)
146 if (ilock_flags & BTRFS_ILOCK_MMAP)
147 up_write(&BTRFS_I(inode)->i_mmap_lock);
148 if (ilock_flags & BTRFS_ILOCK_SHARED)
149 inode_unlock_shared(inode);
155 * Cleanup all submitted ordered extents in specified range to handle errors
156 * from the btrfs_run_delalloc_range() callback.
158 * NOTE: caller must ensure that when an error happens, it can not call
159 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
160 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
161 * to be released, which we want to happen only when finishing the ordered
162 * extent (btrfs_finish_ordered_io()).
164 static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode,
165 struct page *locked_page,
166 u64 offset, u64 bytes)
168 unsigned long index = offset >> PAGE_SHIFT;
169 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
170 u64 page_start = page_offset(locked_page);
171 u64 page_end = page_start + PAGE_SIZE - 1;
175 while (index <= end_index) {
177 * For locked page, we will call end_extent_writepage() on it
178 * in run_delalloc_range() for the error handling. That
179 * end_extent_writepage() function will call
180 * btrfs_mark_ordered_io_finished() to clear page Ordered and
181 * run the ordered extent accounting.
183 * Here we can't just clear the Ordered bit, or
184 * btrfs_mark_ordered_io_finished() would skip the accounting
185 * for the page range, and the ordered extent will never finish.
187 if (index == (page_offset(locked_page) >> PAGE_SHIFT)) {
191 page = find_get_page(inode->vfs_inode.i_mapping, index);
197 * Here we just clear all Ordered bits for every page in the
198 * range, then __endio_write_update_ordered() will handle
199 * the ordered extent accounting for the range.
201 btrfs_page_clamp_clear_ordered(inode->root->fs_info, page,
206 /* The locked page covers the full range, nothing needs to be done */
207 if (bytes + offset <= page_offset(locked_page) + PAGE_SIZE)
210 * In case this page belongs to the delalloc range being instantiated
211 * then skip it, since the first page of a range is going to be
212 * properly cleaned up by the caller of run_delalloc_range
214 if (page_start >= offset && page_end <= (offset + bytes - 1)) {
215 bytes = offset + bytes - page_offset(locked_page) - PAGE_SIZE;
216 offset = page_offset(locked_page) + PAGE_SIZE;
219 return __endio_write_update_ordered(inode, offset, bytes, false);
222 static int btrfs_dirty_inode(struct inode *inode);
224 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
225 struct inode *inode, struct inode *dir,
226 const struct qstr *qstr)
230 err = btrfs_init_acl(trans, inode, dir);
232 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
237 * this does all the hard work for inserting an inline extent into
238 * the btree. The caller should have done a btrfs_drop_extents so that
239 * no overlapping inline items exist in the btree
241 static int insert_inline_extent(struct btrfs_trans_handle *trans,
242 struct btrfs_path *path,
243 struct btrfs_inode *inode, bool extent_inserted,
244 size_t size, size_t compressed_size,
246 struct page **compressed_pages)
248 struct btrfs_root *root = inode->root;
249 struct extent_buffer *leaf;
250 struct page *page = NULL;
253 struct btrfs_file_extent_item *ei;
255 size_t cur_size = size;
257 ASSERT((compressed_size > 0 && compressed_pages) ||
258 (compressed_size == 0 && !compressed_pages));
260 if (compressed_size && compressed_pages)
261 cur_size = compressed_size;
263 if (!extent_inserted) {
264 struct btrfs_key key;
267 key.objectid = btrfs_ino(inode);
269 key.type = BTRFS_EXTENT_DATA_KEY;
271 datasize = btrfs_file_extent_calc_inline_size(cur_size);
272 ret = btrfs_insert_empty_item(trans, root, path, &key,
277 leaf = path->nodes[0];
278 ei = btrfs_item_ptr(leaf, path->slots[0],
279 struct btrfs_file_extent_item);
280 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
281 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
282 btrfs_set_file_extent_encryption(leaf, ei, 0);
283 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
284 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
285 ptr = btrfs_file_extent_inline_start(ei);
287 if (compress_type != BTRFS_COMPRESS_NONE) {
290 while (compressed_size > 0) {
291 cpage = compressed_pages[i];
292 cur_size = min_t(unsigned long, compressed_size,
295 kaddr = kmap_atomic(cpage);
296 write_extent_buffer(leaf, kaddr, ptr, cur_size);
297 kunmap_atomic(kaddr);
301 compressed_size -= cur_size;
303 btrfs_set_file_extent_compression(leaf, ei,
306 page = find_get_page(inode->vfs_inode.i_mapping, 0);
307 btrfs_set_file_extent_compression(leaf, ei, 0);
308 kaddr = kmap_atomic(page);
309 write_extent_buffer(leaf, kaddr, ptr, size);
310 kunmap_atomic(kaddr);
313 btrfs_mark_buffer_dirty(leaf);
314 btrfs_release_path(path);
317 * We align size to sectorsize for inline extents just for simplicity
320 ret = btrfs_inode_set_file_extent_range(inode, 0,
321 ALIGN(size, root->fs_info->sectorsize));
326 * we're an inline extent, so nobody can
327 * extend the file past i_size without locking
328 * a page we already have locked.
330 * We must do any isize and inode updates
331 * before we unlock the pages. Otherwise we
332 * could end up racing with unlink.
334 inode->disk_i_size = i_size_read(&inode->vfs_inode);
342 * conditionally insert an inline extent into the file. This
343 * does the checks required to make sure the data is small enough
344 * to fit as an inline extent.
346 static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 size,
347 size_t compressed_size,
349 struct page **compressed_pages)
351 struct btrfs_drop_extents_args drop_args = { 0 };
352 struct btrfs_root *root = inode->root;
353 struct btrfs_fs_info *fs_info = root->fs_info;
354 struct btrfs_trans_handle *trans;
355 u64 data_len = (compressed_size ?: size);
357 struct btrfs_path *path;
360 * We can create an inline extent if it ends at or beyond the current
361 * i_size, is no larger than a sector (decompressed), and the (possibly
362 * compressed) data fits in a leaf and the configured maximum inline
365 if (size < i_size_read(&inode->vfs_inode) ||
366 size > fs_info->sectorsize ||
367 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
368 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;
384 drop_args.end = fs_info->sectorsize;
385 drop_args.drop_cache = true;
386 drop_args.replace_extent = true;
387 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(data_len);
388 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
390 btrfs_abort_transaction(trans, ret);
394 ret = insert_inline_extent(trans, path, inode, drop_args.extent_inserted,
395 size, compressed_size, compress_type,
397 if (ret && ret != -ENOSPC) {
398 btrfs_abort_transaction(trans, ret);
400 } else if (ret == -ENOSPC) {
405 btrfs_update_inode_bytes(inode, size, drop_args.bytes_found);
406 ret = btrfs_update_inode(trans, root, inode);
407 if (ret && ret != -ENOSPC) {
408 btrfs_abort_transaction(trans, ret);
410 } else if (ret == -ENOSPC) {
415 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
418 * Don't forget to free the reserved space, as for inlined extent
419 * it won't count as data extent, free them directly here.
420 * And at reserve time, it's always aligned to page size, so
421 * just free one page here.
423 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
424 btrfs_free_path(path);
425 btrfs_end_transaction(trans);
429 struct async_extent {
434 unsigned long nr_pages;
436 struct list_head list;
441 struct page *locked_page;
444 unsigned int write_flags;
445 struct list_head extents;
446 struct cgroup_subsys_state *blkcg_css;
447 struct btrfs_work work;
448 struct async_cow *async_cow;
453 struct async_chunk chunks[];
456 static noinline int add_async_extent(struct async_chunk *cow,
457 u64 start, u64 ram_size,
460 unsigned long nr_pages,
463 struct async_extent *async_extent;
465 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
466 BUG_ON(!async_extent); /* -ENOMEM */
467 async_extent->start = start;
468 async_extent->ram_size = ram_size;
469 async_extent->compressed_size = compressed_size;
470 async_extent->pages = pages;
471 async_extent->nr_pages = nr_pages;
472 async_extent->compress_type = compress_type;
473 list_add_tail(&async_extent->list, &cow->extents);
478 * Check if the inode has flags compatible with compression
480 static inline bool inode_can_compress(struct btrfs_inode *inode)
482 if (inode->flags & BTRFS_INODE_NODATACOW ||
483 inode->flags & BTRFS_INODE_NODATASUM)
489 * Check if the inode needs to be submitted to compression, based on mount
490 * options, defragmentation, properties or heuristics.
492 static inline int inode_need_compress(struct btrfs_inode *inode, u64 start,
495 struct btrfs_fs_info *fs_info = inode->root->fs_info;
497 if (!inode_can_compress(inode)) {
498 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
499 KERN_ERR "BTRFS: unexpected compression for ino %llu\n",
504 * Special check for subpage.
506 * We lock the full page then run each delalloc range in the page, thus
507 * for the following case, we will hit some subpage specific corner case:
510 * | |///////| |///////|
513 * In above case, both range A and range B will try to unlock the full
514 * page [0, 64K), causing the one finished later will have page
515 * unlocked already, triggering various page lock requirement BUG_ON()s.
517 * So here we add an artificial limit that subpage compression can only
518 * if the range is fully page aligned.
520 * In theory we only need to ensure the first page is fully covered, but
521 * the tailing partial page will be locked until the full compression
522 * finishes, delaying the write of other range.
524 * TODO: Make btrfs_run_delalloc_range() to lock all delalloc range
525 * first to prevent any submitted async extent to unlock the full page.
526 * By this, we can ensure for subpage case that only the last async_cow
527 * will unlock the full page.
529 if (fs_info->sectorsize < PAGE_SIZE) {
530 if (!IS_ALIGNED(start, PAGE_SIZE) ||
531 !IS_ALIGNED(end + 1, PAGE_SIZE))
536 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
539 if (inode->defrag_compress)
541 /* bad compression ratios */
542 if (inode->flags & BTRFS_INODE_NOCOMPRESS)
544 if (btrfs_test_opt(fs_info, COMPRESS) ||
545 inode->flags & BTRFS_INODE_COMPRESS ||
546 inode->prop_compress)
547 return btrfs_compress_heuristic(&inode->vfs_inode, start, end);
551 static inline void inode_should_defrag(struct btrfs_inode *inode,
552 u64 start, u64 end, u64 num_bytes, u32 small_write)
554 /* If this is a small write inside eof, kick off a defrag */
555 if (num_bytes < small_write &&
556 (start > 0 || end + 1 < inode->disk_i_size))
557 btrfs_add_inode_defrag(NULL, inode, small_write);
561 * we create compressed extents in two phases. The first
562 * phase compresses a range of pages that have already been
563 * locked (both pages and state bits are locked).
565 * This is done inside an ordered work queue, and the compression
566 * is spread across many cpus. The actual IO submission is step
567 * two, and the ordered work queue takes care of making sure that
568 * happens in the same order things were put onto the queue by
569 * writepages and friends.
571 * If this code finds it can't get good compression, it puts an
572 * entry onto the work queue to write the uncompressed bytes. This
573 * makes sure that both compressed inodes and uncompressed inodes
574 * are written in the same order that the flusher thread sent them
577 static noinline int compress_file_range(struct async_chunk *async_chunk)
579 struct inode *inode = async_chunk->inode;
580 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
581 u64 blocksize = fs_info->sectorsize;
582 u64 start = async_chunk->start;
583 u64 end = async_chunk->end;
587 struct page **pages = NULL;
588 unsigned long nr_pages;
589 unsigned long total_compressed = 0;
590 unsigned long total_in = 0;
593 int compress_type = fs_info->compress_type;
594 int compressed_extents = 0;
597 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
601 * We need to save i_size before now because it could change in between
602 * us evaluating the size and assigning it. This is because we lock and
603 * unlock the page in truncate and fallocate, and then modify the i_size
606 * The barriers are to emulate READ_ONCE, remove that once i_size_read
610 i_size = i_size_read(inode);
612 actual_end = min_t(u64, i_size, end + 1);
615 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
616 nr_pages = min_t(unsigned long, nr_pages,
617 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
620 * we don't want to send crud past the end of i_size through
621 * compression, that's just a waste of CPU time. So, if the
622 * end of the file is before the start of our current
623 * requested range of bytes, we bail out to the uncompressed
624 * cleanup code that can deal with all of this.
626 * It isn't really the fastest way to fix things, but this is a
627 * very uncommon corner.
629 if (actual_end <= start)
630 goto cleanup_and_bail_uncompressed;
632 total_compressed = actual_end - start;
635 * Skip compression for a small file range(<=blocksize) that
636 * isn't an inline extent, since it doesn't save disk space at all.
638 if (total_compressed <= blocksize &&
639 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
640 goto cleanup_and_bail_uncompressed;
643 * For subpage case, we require full page alignment for the sector
645 * Thus we must also check against @actual_end, not just @end.
647 if (blocksize < PAGE_SIZE) {
648 if (!IS_ALIGNED(start, PAGE_SIZE) ||
649 !IS_ALIGNED(round_up(actual_end, blocksize), PAGE_SIZE))
650 goto cleanup_and_bail_uncompressed;
653 total_compressed = min_t(unsigned long, total_compressed,
654 BTRFS_MAX_UNCOMPRESSED);
659 * we do compression for mount -o compress and when the
660 * inode has not been flagged as nocompress. This flag can
661 * change at any time if we discover bad compression ratios.
663 if (inode_need_compress(BTRFS_I(inode), start, end)) {
665 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
667 /* just bail out to the uncompressed code */
672 if (BTRFS_I(inode)->defrag_compress)
673 compress_type = BTRFS_I(inode)->defrag_compress;
674 else if (BTRFS_I(inode)->prop_compress)
675 compress_type = BTRFS_I(inode)->prop_compress;
678 * we need to call clear_page_dirty_for_io on each
679 * page in the range. Otherwise applications with the file
680 * mmap'd can wander in and change the page contents while
681 * we are compressing them.
683 * If the compression fails for any reason, we set the pages
684 * dirty again later on.
686 * Note that the remaining part is redirtied, the start pointer
687 * has moved, the end is the original one.
690 extent_range_clear_dirty_for_io(inode, start, end);
694 /* Compression level is applied here and only here */
695 ret = btrfs_compress_pages(
696 compress_type | (fs_info->compress_level << 4),
697 inode->i_mapping, start,
704 unsigned long offset = offset_in_page(total_compressed);
705 struct page *page = pages[nr_pages - 1];
707 /* zero the tail end of the last page, we might be
708 * sending it down to disk
711 memzero_page(page, offset, PAGE_SIZE - offset);
717 * Check cow_file_range() for why we don't even try to create inline
718 * extent for subpage case.
720 if (start == 0 && fs_info->sectorsize == PAGE_SIZE) {
721 /* lets try to make an inline extent */
722 if (ret || total_in < actual_end) {
723 /* we didn't compress the entire range, try
724 * to make an uncompressed inline extent.
726 ret = cow_file_range_inline(BTRFS_I(inode), actual_end,
727 0, BTRFS_COMPRESS_NONE,
730 /* try making a compressed inline extent */
731 ret = cow_file_range_inline(BTRFS_I(inode), actual_end,
733 compress_type, pages);
736 unsigned long clear_flags = EXTENT_DELALLOC |
737 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
738 EXTENT_DO_ACCOUNTING;
739 unsigned long page_error_op;
741 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
744 * inline extent creation worked or returned error,
745 * we don't need to create any more async work items.
746 * Unlock and free up our temp pages.
748 * We use DO_ACCOUNTING here because we need the
749 * delalloc_release_metadata to be done _after_ we drop
750 * our outstanding extent for clearing delalloc for this
753 extent_clear_unlock_delalloc(BTRFS_I(inode), start, end,
757 PAGE_START_WRITEBACK |
762 * Ensure we only free the compressed pages if we have
763 * them allocated, as we can still reach here with
764 * inode_need_compress() == false.
767 for (i = 0; i < nr_pages; i++) {
768 WARN_ON(pages[i]->mapping);
779 * we aren't doing an inline extent round the compressed size
780 * up to a block size boundary so the allocator does sane
783 total_compressed = ALIGN(total_compressed, blocksize);
786 * one last check to make sure the compression is really a
787 * win, compare the page count read with the blocks on disk,
788 * compression must free at least one sector size
790 total_in = round_up(total_in, fs_info->sectorsize);
791 if (total_compressed + blocksize <= total_in) {
792 compressed_extents++;
795 * The async work queues will take care of doing actual
796 * allocation on disk for these compressed pages, and
797 * will submit them to the elevator.
799 add_async_extent(async_chunk, start, total_in,
800 total_compressed, pages, nr_pages,
803 if (start + total_in < end) {
809 return compressed_extents;
814 * the compression code ran but failed to make things smaller,
815 * free any pages it allocated and our page pointer array
817 for (i = 0; i < nr_pages; i++) {
818 WARN_ON(pages[i]->mapping);
823 total_compressed = 0;
826 /* flag the file so we don't compress in the future */
827 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
828 !(BTRFS_I(inode)->prop_compress)) {
829 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
832 cleanup_and_bail_uncompressed:
834 * No compression, but we still need to write the pages in the file
835 * we've been given so far. redirty the locked page if it corresponds
836 * to our extent and set things up for the async work queue to run
837 * cow_file_range to do the normal delalloc dance.
839 if (async_chunk->locked_page &&
840 (page_offset(async_chunk->locked_page) >= start &&
841 page_offset(async_chunk->locked_page)) <= end) {
842 __set_page_dirty_nobuffers(async_chunk->locked_page);
843 /* unlocked later on in the async handlers */
847 extent_range_redirty_for_io(inode, start, end);
848 add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
849 BTRFS_COMPRESS_NONE);
850 compressed_extents++;
852 return compressed_extents;
855 static void free_async_extent_pages(struct async_extent *async_extent)
859 if (!async_extent->pages)
862 for (i = 0; i < async_extent->nr_pages; i++) {
863 WARN_ON(async_extent->pages[i]->mapping);
864 put_page(async_extent->pages[i]);
866 kfree(async_extent->pages);
867 async_extent->nr_pages = 0;
868 async_extent->pages = NULL;
871 static int submit_uncompressed_range(struct btrfs_inode *inode,
872 struct async_extent *async_extent,
873 struct page *locked_page)
875 u64 start = async_extent->start;
876 u64 end = async_extent->start + async_extent->ram_size - 1;
877 unsigned long nr_written = 0;
878 int page_started = 0;
882 * Call cow_file_range() to run the delalloc range directly, since we
883 * won't go to NOCOW or async path again.
885 * Also we call cow_file_range() with @unlock_page == 0, so that we
886 * can directly submit them without interruption.
888 ret = cow_file_range(inode, locked_page, start, end, &page_started,
890 /* Inline extent inserted, page gets unlocked and everything is done */
897 unlock_page(locked_page);
901 ret = extent_write_locked_range(&inode->vfs_inode, start, end);
902 /* All pages will be unlocked, including @locked_page */
908 static int submit_one_async_extent(struct btrfs_inode *inode,
909 struct async_chunk *async_chunk,
910 struct async_extent *async_extent,
913 struct extent_io_tree *io_tree = &inode->io_tree;
914 struct btrfs_root *root = inode->root;
915 struct btrfs_fs_info *fs_info = root->fs_info;
916 struct btrfs_key ins;
917 struct page *locked_page = NULL;
918 struct extent_map *em;
920 u64 start = async_extent->start;
921 u64 end = async_extent->start + async_extent->ram_size - 1;
924 * If async_chunk->locked_page is in the async_extent range, we need to
927 if (async_chunk->locked_page) {
928 u64 locked_page_start = page_offset(async_chunk->locked_page);
929 u64 locked_page_end = locked_page_start + PAGE_SIZE - 1;
931 if (!(start >= locked_page_end || end <= locked_page_start))
932 locked_page = async_chunk->locked_page;
934 lock_extent(io_tree, start, end);
936 /* We have fall back to uncompressed write */
937 if (!async_extent->pages)
938 return submit_uncompressed_range(inode, async_extent, locked_page);
940 ret = btrfs_reserve_extent(root, async_extent->ram_size,
941 async_extent->compressed_size,
942 async_extent->compressed_size,
943 0, *alloc_hint, &ins, 1, 1);
945 free_async_extent_pages(async_extent);
947 * Here we used to try again by going back to non-compressed
948 * path for ENOSPC. But we can't reserve space even for
949 * compressed size, how could it work for uncompressed size
950 * which requires larger size? So here we directly go error
956 /* Here we're doing allocation and writeback of the compressed pages */
957 em = create_io_em(inode, start,
958 async_extent->ram_size, /* len */
959 start, /* orig_start */
960 ins.objectid, /* block_start */
961 ins.offset, /* block_len */
962 ins.offset, /* orig_block_len */
963 async_extent->ram_size, /* ram_bytes */
964 async_extent->compress_type,
965 BTRFS_ORDERED_COMPRESSED);
968 goto out_free_reserve;
972 ret = btrfs_add_ordered_extent(inode, start, /* file_offset */
973 async_extent->ram_size, /* num_bytes */
974 async_extent->ram_size, /* ram_bytes */
975 ins.objectid, /* disk_bytenr */
976 ins.offset, /* disk_num_bytes */
978 1 << BTRFS_ORDERED_COMPRESSED,
979 async_extent->compress_type);
981 btrfs_drop_extent_cache(inode, start, end, 0);
982 goto out_free_reserve;
984 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
986 /* Clear dirty, set writeback and unlock the pages. */
987 extent_clear_unlock_delalloc(inode, start, end,
988 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
989 PAGE_UNLOCK | PAGE_START_WRITEBACK);
990 if (btrfs_submit_compressed_write(inode, start, /* file_offset */
991 async_extent->ram_size, /* num_bytes */
992 ins.objectid, /* disk_bytenr */
993 ins.offset, /* compressed_len */
994 async_extent->pages, /* compressed_pages */
995 async_extent->nr_pages,
996 async_chunk->write_flags,
997 async_chunk->blkcg_css)) {
998 const u64 start = async_extent->start;
999 const u64 end = start + async_extent->ram_size - 1;
1001 btrfs_writepage_endio_finish_ordered(inode, NULL, start, end, 0);
1003 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
1004 PAGE_END_WRITEBACK | PAGE_SET_ERROR);
1005 free_async_extent_pages(async_extent);
1007 *alloc_hint = ins.objectid + ins.offset;
1008 kfree(async_extent);
1012 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1013 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1015 extent_clear_unlock_delalloc(inode, start, end,
1016 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
1017 EXTENT_DELALLOC_NEW |
1018 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
1019 PAGE_UNLOCK | PAGE_START_WRITEBACK |
1020 PAGE_END_WRITEBACK | PAGE_SET_ERROR);
1021 free_async_extent_pages(async_extent);
1022 kfree(async_extent);
1027 * Phase two of compressed writeback. This is the ordered portion of the code,
1028 * which only gets called in the order the work was queued. We walk all the
1029 * async extents created by compress_file_range and send them down to the disk.
1031 static noinline void submit_compressed_extents(struct async_chunk *async_chunk)
1033 struct btrfs_inode *inode = BTRFS_I(async_chunk->inode);
1034 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1035 struct async_extent *async_extent;
1039 while (!list_empty(&async_chunk->extents)) {
1043 async_extent = list_entry(async_chunk->extents.next,
1044 struct async_extent, list);
1045 list_del(&async_extent->list);
1046 extent_start = async_extent->start;
1047 ram_size = async_extent->ram_size;
1049 ret = submit_one_async_extent(inode, async_chunk, async_extent,
1051 btrfs_debug(fs_info,
1052 "async extent submission failed root=%lld inode=%llu start=%llu len=%llu ret=%d",
1053 inode->root->root_key.objectid,
1054 btrfs_ino(inode), extent_start, ram_size, ret);
1058 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
1061 struct extent_map_tree *em_tree = &inode->extent_tree;
1062 struct extent_map *em;
1065 read_lock(&em_tree->lock);
1066 em = search_extent_mapping(em_tree, start, num_bytes);
1069 * if block start isn't an actual block number then find the
1070 * first block in this inode and use that as a hint. If that
1071 * block is also bogus then just don't worry about it.
1073 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1074 free_extent_map(em);
1075 em = search_extent_mapping(em_tree, 0, 0);
1076 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
1077 alloc_hint = em->block_start;
1079 free_extent_map(em);
1081 alloc_hint = em->block_start;
1082 free_extent_map(em);
1085 read_unlock(&em_tree->lock);
1091 * when extent_io.c finds a delayed allocation range in the file,
1092 * the call backs end up in this code. The basic idea is to
1093 * allocate extents on disk for the range, and create ordered data structs
1094 * in ram to track those extents.
1096 * locked_page is the page that writepage had locked already. We use
1097 * it to make sure we don't do extra locks or unlocks.
1099 * *page_started is set to one if we unlock locked_page and do everything
1100 * required to start IO on it. It may be clean and already done with
1101 * IO when we return.
1103 static noinline int cow_file_range(struct btrfs_inode *inode,
1104 struct page *locked_page,
1105 u64 start, u64 end, int *page_started,
1106 unsigned long *nr_written, int unlock)
1108 struct btrfs_root *root = inode->root;
1109 struct btrfs_fs_info *fs_info = root->fs_info;
1112 unsigned long ram_size;
1113 u64 cur_alloc_size = 0;
1115 u64 blocksize = fs_info->sectorsize;
1116 struct btrfs_key ins;
1117 struct extent_map *em;
1118 unsigned clear_bits;
1119 unsigned long page_ops;
1120 bool extent_reserved = false;
1123 if (btrfs_is_free_space_inode(inode)) {
1129 num_bytes = ALIGN(end - start + 1, blocksize);
1130 num_bytes = max(blocksize, num_bytes);
1131 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1133 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1136 * Due to the page size limit, for subpage we can only trigger the
1137 * writeback for the dirty sectors of page, that means data writeback
1138 * is doing more writeback than what we want.
1140 * This is especially unexpected for some call sites like fallocate,
1141 * where we only increase i_size after everything is done.
1142 * This means we can trigger inline extent even if we didn't want to.
1143 * So here we skip inline extent creation completely.
1145 if (start == 0 && fs_info->sectorsize == PAGE_SIZE) {
1146 u64 actual_end = min_t(u64, i_size_read(&inode->vfs_inode),
1149 /* lets try to make an inline extent */
1150 ret = cow_file_range_inline(inode, actual_end, 0,
1151 BTRFS_COMPRESS_NONE, NULL);
1154 * We use DO_ACCOUNTING here because we need the
1155 * delalloc_release_metadata to be run _after_ we drop
1156 * our outstanding extent for clearing delalloc for this
1159 extent_clear_unlock_delalloc(inode, start, end,
1161 EXTENT_LOCKED | EXTENT_DELALLOC |
1162 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1163 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1164 PAGE_START_WRITEBACK | PAGE_END_WRITEBACK);
1165 *nr_written = *nr_written +
1166 (end - start + PAGE_SIZE) / PAGE_SIZE;
1169 * locked_page is locked by the caller of
1170 * writepage_delalloc(), not locked by
1171 * __process_pages_contig().
1173 * We can't let __process_pages_contig() to unlock it,
1174 * as it doesn't have any subpage::writers recorded.
1176 * Here we manually unlock the page, since the caller
1177 * can't use page_started to determine if it's an
1178 * inline extent or a compressed extent.
1180 unlock_page(locked_page);
1182 } else if (ret < 0) {
1187 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1188 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
1191 * Relocation relies on the relocated extents to have exactly the same
1192 * size as the original extents. Normally writeback for relocation data
1193 * extents follows a NOCOW path because relocation preallocates the
1194 * extents. However, due to an operation such as scrub turning a block
1195 * group to RO mode, it may fallback to COW mode, so we must make sure
1196 * an extent allocated during COW has exactly the requested size and can
1197 * not be split into smaller extents, otherwise relocation breaks and
1198 * fails during the stage where it updates the bytenr of file extent
1201 if (btrfs_is_data_reloc_root(root))
1202 min_alloc_size = num_bytes;
1204 min_alloc_size = fs_info->sectorsize;
1206 while (num_bytes > 0) {
1207 cur_alloc_size = num_bytes;
1208 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1209 min_alloc_size, 0, alloc_hint,
1213 cur_alloc_size = ins.offset;
1214 extent_reserved = true;
1216 ram_size = ins.offset;
1217 em = create_io_em(inode, start, ins.offset, /* len */
1218 start, /* orig_start */
1219 ins.objectid, /* block_start */
1220 ins.offset, /* block_len */
1221 ins.offset, /* orig_block_len */
1222 ram_size, /* ram_bytes */
1223 BTRFS_COMPRESS_NONE, /* compress_type */
1224 BTRFS_ORDERED_REGULAR /* type */);
1229 free_extent_map(em);
1231 ret = btrfs_add_ordered_extent(inode, start, ram_size, ram_size,
1232 ins.objectid, cur_alloc_size, 0,
1233 1 << BTRFS_ORDERED_REGULAR,
1234 BTRFS_COMPRESS_NONE);
1236 goto out_drop_extent_cache;
1238 if (btrfs_is_data_reloc_root(root)) {
1239 ret = btrfs_reloc_clone_csums(inode, start,
1242 * Only drop cache here, and process as normal.
1244 * We must not allow extent_clear_unlock_delalloc()
1245 * at out_unlock label to free meta of this ordered
1246 * extent, as its meta should be freed by
1247 * btrfs_finish_ordered_io().
1249 * So we must continue until @start is increased to
1250 * skip current ordered extent.
1253 btrfs_drop_extent_cache(inode, start,
1254 start + ram_size - 1, 0);
1257 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1260 * We're not doing compressed IO, don't unlock the first page
1261 * (which the caller expects to stay locked), don't clear any
1262 * dirty bits and don't set any writeback bits
1264 * Do set the Ordered (Private2) bit so we know this page was
1265 * properly setup for writepage.
1267 page_ops = unlock ? PAGE_UNLOCK : 0;
1268 page_ops |= PAGE_SET_ORDERED;
1270 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1272 EXTENT_LOCKED | EXTENT_DELALLOC,
1274 if (num_bytes < cur_alloc_size)
1277 num_bytes -= cur_alloc_size;
1278 alloc_hint = ins.objectid + ins.offset;
1279 start += cur_alloc_size;
1280 extent_reserved = false;
1283 * btrfs_reloc_clone_csums() error, since start is increased
1284 * extent_clear_unlock_delalloc() at out_unlock label won't
1285 * free metadata of current ordered extent, we're OK to exit.
1293 out_drop_extent_cache:
1294 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1296 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1297 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1299 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1300 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1301 page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK;
1303 * If we reserved an extent for our delalloc range (or a subrange) and
1304 * failed to create the respective ordered extent, then it means that
1305 * when we reserved the extent we decremented the extent's size from
1306 * the data space_info's bytes_may_use counter and incremented the
1307 * space_info's bytes_reserved counter by the same amount. We must make
1308 * sure extent_clear_unlock_delalloc() does not try to decrement again
1309 * the data space_info's bytes_may_use counter, therefore we do not pass
1310 * it the flag EXTENT_CLEAR_DATA_RESV.
1312 if (extent_reserved) {
1313 extent_clear_unlock_delalloc(inode, start,
1314 start + cur_alloc_size - 1,
1318 start += cur_alloc_size;
1322 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1323 clear_bits | EXTENT_CLEAR_DATA_RESV,
1329 * work queue call back to started compression on a file and pages
1331 static noinline void async_cow_start(struct btrfs_work *work)
1333 struct async_chunk *async_chunk;
1334 int compressed_extents;
1336 async_chunk = container_of(work, struct async_chunk, work);
1338 compressed_extents = compress_file_range(async_chunk);
1339 if (compressed_extents == 0) {
1340 btrfs_add_delayed_iput(async_chunk->inode);
1341 async_chunk->inode = NULL;
1346 * work queue call back to submit previously compressed pages
1348 static noinline void async_cow_submit(struct btrfs_work *work)
1350 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1352 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1353 unsigned long nr_pages;
1355 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1359 * ->inode could be NULL if async_chunk_start has failed to compress,
1360 * in which case we don't have anything to submit, yet we need to
1361 * always adjust ->async_delalloc_pages as its paired with the init
1362 * happening in cow_file_range_async
1364 if (async_chunk->inode)
1365 submit_compressed_extents(async_chunk);
1367 /* atomic_sub_return implies a barrier */
1368 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1370 cond_wake_up_nomb(&fs_info->async_submit_wait);
1373 static noinline void async_cow_free(struct btrfs_work *work)
1375 struct async_chunk *async_chunk;
1376 struct async_cow *async_cow;
1378 async_chunk = container_of(work, struct async_chunk, work);
1379 if (async_chunk->inode)
1380 btrfs_add_delayed_iput(async_chunk->inode);
1381 if (async_chunk->blkcg_css)
1382 css_put(async_chunk->blkcg_css);
1384 async_cow = async_chunk->async_cow;
1385 if (atomic_dec_and_test(&async_cow->num_chunks))
1389 static int cow_file_range_async(struct btrfs_inode *inode,
1390 struct writeback_control *wbc,
1391 struct page *locked_page,
1392 u64 start, u64 end, int *page_started,
1393 unsigned long *nr_written)
1395 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1396 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1397 struct async_cow *ctx;
1398 struct async_chunk *async_chunk;
1399 unsigned long nr_pages;
1401 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1403 bool should_compress;
1405 const unsigned int write_flags = wbc_to_write_flags(wbc);
1407 unlock_extent(&inode->io_tree, start, end);
1409 if (inode->flags & BTRFS_INODE_NOCOMPRESS &&
1410 !btrfs_test_opt(fs_info, FORCE_COMPRESS)) {
1412 should_compress = false;
1414 should_compress = true;
1417 nofs_flag = memalloc_nofs_save();
1418 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1419 memalloc_nofs_restore(nofs_flag);
1422 unsigned clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC |
1423 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1424 EXTENT_DO_ACCOUNTING;
1425 unsigned long page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK |
1426 PAGE_END_WRITEBACK | PAGE_SET_ERROR;
1428 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1429 clear_bits, page_ops);
1433 async_chunk = ctx->chunks;
1434 atomic_set(&ctx->num_chunks, num_chunks);
1436 for (i = 0; i < num_chunks; i++) {
1437 if (should_compress)
1438 cur_end = min(end, start + SZ_512K - 1);
1443 * igrab is called higher up in the call chain, take only the
1444 * lightweight reference for the callback lifetime
1446 ihold(&inode->vfs_inode);
1447 async_chunk[i].async_cow = ctx;
1448 async_chunk[i].inode = &inode->vfs_inode;
1449 async_chunk[i].start = start;
1450 async_chunk[i].end = cur_end;
1451 async_chunk[i].write_flags = write_flags;
1452 INIT_LIST_HEAD(&async_chunk[i].extents);
1455 * The locked_page comes all the way from writepage and its
1456 * the original page we were actually given. As we spread
1457 * this large delalloc region across multiple async_chunk
1458 * structs, only the first struct needs a pointer to locked_page
1460 * This way we don't need racey decisions about who is supposed
1465 * Depending on the compressibility, the pages might or
1466 * might not go through async. We want all of them to
1467 * be accounted against wbc once. Let's do it here
1468 * before the paths diverge. wbc accounting is used
1469 * only for foreign writeback detection and doesn't
1470 * need full accuracy. Just account the whole thing
1471 * against the first page.
1473 wbc_account_cgroup_owner(wbc, locked_page,
1475 async_chunk[i].locked_page = locked_page;
1478 async_chunk[i].locked_page = NULL;
1481 if (blkcg_css != blkcg_root_css) {
1483 async_chunk[i].blkcg_css = blkcg_css;
1485 async_chunk[i].blkcg_css = NULL;
1488 btrfs_init_work(&async_chunk[i].work, async_cow_start,
1489 async_cow_submit, async_cow_free);
1491 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1492 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1494 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1496 *nr_written += nr_pages;
1497 start = cur_end + 1;
1503 static noinline int run_delalloc_zoned(struct btrfs_inode *inode,
1504 struct page *locked_page, u64 start,
1505 u64 end, int *page_started,
1506 unsigned long *nr_written)
1510 ret = cow_file_range(inode, locked_page, start, end, page_started,
1518 __set_page_dirty_nobuffers(locked_page);
1519 account_page_redirty(locked_page);
1520 extent_write_locked_range(&inode->vfs_inode, start, end);
1526 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1527 u64 bytenr, u64 num_bytes)
1529 struct btrfs_root *csum_root = btrfs_csum_root(fs_info, bytenr);
1530 struct btrfs_ordered_sum *sums;
1534 ret = btrfs_lookup_csums_range(csum_root, bytenr,
1535 bytenr + num_bytes - 1, &list, 0);
1536 if (ret == 0 && list_empty(&list))
1539 while (!list_empty(&list)) {
1540 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1541 list_del(&sums->list);
1549 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1550 const u64 start, const u64 end,
1551 int *page_started, unsigned long *nr_written)
1553 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1554 const bool is_reloc_ino = btrfs_is_data_reloc_root(inode->root);
1555 const u64 range_bytes = end + 1 - start;
1556 struct extent_io_tree *io_tree = &inode->io_tree;
1557 u64 range_start = start;
1561 * If EXTENT_NORESERVE is set it means that when the buffered write was
1562 * made we had not enough available data space and therefore we did not
1563 * reserve data space for it, since we though we could do NOCOW for the
1564 * respective file range (either there is prealloc extent or the inode
1565 * has the NOCOW bit set).
1567 * However when we need to fallback to COW mode (because for example the
1568 * block group for the corresponding extent was turned to RO mode by a
1569 * scrub or relocation) we need to do the following:
1571 * 1) We increment the bytes_may_use counter of the data space info.
1572 * If COW succeeds, it allocates a new data extent and after doing
1573 * that it decrements the space info's bytes_may_use counter and
1574 * increments its bytes_reserved counter by the same amount (we do
1575 * this at btrfs_add_reserved_bytes()). So we need to increment the
1576 * bytes_may_use counter to compensate (when space is reserved at
1577 * buffered write time, the bytes_may_use counter is incremented);
1579 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1580 * that if the COW path fails for any reason, it decrements (through
1581 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1582 * data space info, which we incremented in the step above.
1584 * If we need to fallback to cow and the inode corresponds to a free
1585 * space cache inode or an inode of the data relocation tree, we must
1586 * also increment bytes_may_use of the data space_info for the same
1587 * reason. Space caches and relocated data extents always get a prealloc
1588 * extent for them, however scrub or balance may have set the block
1589 * group that contains that extent to RO mode and therefore force COW
1590 * when starting writeback.
1592 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1593 EXTENT_NORESERVE, 0);
1594 if (count > 0 || is_space_ino || is_reloc_ino) {
1596 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1597 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1599 if (is_space_ino || is_reloc_ino)
1600 bytes = range_bytes;
1602 spin_lock(&sinfo->lock);
1603 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1604 spin_unlock(&sinfo->lock);
1607 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1611 return cow_file_range(inode, locked_page, start, end, page_started,
1616 * when nowcow writeback call back. This checks for snapshots or COW copies
1617 * of the extents that exist in the file, and COWs the file as required.
1619 * If no cow copies or snapshots exist, we write directly to the existing
1622 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
1623 struct page *locked_page,
1624 const u64 start, const u64 end,
1626 unsigned long *nr_written)
1628 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1629 struct btrfs_root *root = inode->root;
1630 struct btrfs_path *path;
1631 u64 cow_start = (u64)-1;
1632 u64 cur_offset = start;
1634 bool check_prev = true;
1635 const bool freespace_inode = btrfs_is_free_space_inode(inode);
1636 u64 ino = btrfs_ino(inode);
1638 u64 disk_bytenr = 0;
1639 const bool force = inode->flags & BTRFS_INODE_NODATACOW;
1641 path = btrfs_alloc_path();
1643 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1644 EXTENT_LOCKED | EXTENT_DELALLOC |
1645 EXTENT_DO_ACCOUNTING |
1646 EXTENT_DEFRAG, PAGE_UNLOCK |
1647 PAGE_START_WRITEBACK |
1648 PAGE_END_WRITEBACK);
1653 struct btrfs_key found_key;
1654 struct btrfs_file_extent_item *fi;
1655 struct extent_buffer *leaf;
1665 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1671 * If there is no extent for our range when doing the initial
1672 * search, then go back to the previous slot as it will be the
1673 * one containing the search offset
1675 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1676 leaf = path->nodes[0];
1677 btrfs_item_key_to_cpu(leaf, &found_key,
1678 path->slots[0] - 1);
1679 if (found_key.objectid == ino &&
1680 found_key.type == BTRFS_EXTENT_DATA_KEY)
1685 /* Go to next leaf if we have exhausted the current one */
1686 leaf = path->nodes[0];
1687 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1688 ret = btrfs_next_leaf(root, path);
1690 if (cow_start != (u64)-1)
1691 cur_offset = cow_start;
1696 leaf = path->nodes[0];
1699 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1701 /* Didn't find anything for our INO */
1702 if (found_key.objectid > ino)
1705 * Keep searching until we find an EXTENT_ITEM or there are no
1706 * more extents for this inode
1708 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1709 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1714 /* Found key is not EXTENT_DATA_KEY or starts after req range */
1715 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1716 found_key.offset > end)
1720 * If the found extent starts after requested offset, then
1721 * adjust extent_end to be right before this extent begins
1723 if (found_key.offset > cur_offset) {
1724 extent_end = found_key.offset;
1730 * Found extent which begins before our range and potentially
1733 fi = btrfs_item_ptr(leaf, path->slots[0],
1734 struct btrfs_file_extent_item);
1735 extent_type = btrfs_file_extent_type(leaf, fi);
1737 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1738 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1739 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1740 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1741 extent_offset = btrfs_file_extent_offset(leaf, fi);
1742 extent_end = found_key.offset +
1743 btrfs_file_extent_num_bytes(leaf, fi);
1745 btrfs_file_extent_disk_num_bytes(leaf, fi);
1747 * If the extent we got ends before our current offset,
1748 * skip to the next extent.
1750 if (extent_end <= cur_offset) {
1755 if (disk_bytenr == 0)
1757 /* Skip compressed/encrypted/encoded extents */
1758 if (btrfs_file_extent_compression(leaf, fi) ||
1759 btrfs_file_extent_encryption(leaf, fi) ||
1760 btrfs_file_extent_other_encoding(leaf, fi))
1763 * If extent is created before the last volume's snapshot
1764 * this implies the extent is shared, hence we can't do
1765 * nocow. This is the same check as in
1766 * btrfs_cross_ref_exist but without calling
1767 * btrfs_search_slot.
1769 if (!freespace_inode &&
1770 btrfs_file_extent_generation(leaf, fi) <=
1771 btrfs_root_last_snapshot(&root->root_item))
1773 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1777 * The following checks can be expensive, as they need to
1778 * take other locks and do btree or rbtree searches, so
1779 * release the path to avoid blocking other tasks for too
1782 btrfs_release_path(path);
1784 ret = btrfs_cross_ref_exist(root, ino,
1786 extent_offset, disk_bytenr, false);
1789 * ret could be -EIO if the above fails to read
1793 if (cow_start != (u64)-1)
1794 cur_offset = cow_start;
1798 WARN_ON_ONCE(freespace_inode);
1801 disk_bytenr += extent_offset;
1802 disk_bytenr += cur_offset - found_key.offset;
1803 num_bytes = min(end + 1, extent_end) - cur_offset;
1805 * If there are pending snapshots for this root, we
1806 * fall into common COW way
1808 if (!freespace_inode && atomic_read(&root->snapshot_force_cow))
1811 * force cow if csum exists in the range.
1812 * this ensure that csum for a given extent are
1813 * either valid or do not exist.
1815 ret = csum_exist_in_range(fs_info, disk_bytenr,
1819 * ret could be -EIO if the above fails to read
1823 if (cow_start != (u64)-1)
1824 cur_offset = cow_start;
1827 WARN_ON_ONCE(freespace_inode);
1830 /* If the extent's block group is RO, we must COW */
1831 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr))
1834 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1835 extent_end = found_key.offset + ram_bytes;
1836 extent_end = ALIGN(extent_end, fs_info->sectorsize);
1837 /* Skip extents outside of our requested range */
1838 if (extent_end <= start) {
1843 /* If this triggers then we have a memory corruption */
1848 * If nocow is false then record the beginning of the range
1849 * that needs to be COWed
1852 if (cow_start == (u64)-1)
1853 cow_start = cur_offset;
1854 cur_offset = extent_end;
1855 if (cur_offset > end)
1857 if (!path->nodes[0])
1864 * COW range from cow_start to found_key.offset - 1. As the key
1865 * will contain the beginning of the first extent that can be
1866 * NOCOW, following one which needs to be COW'ed
1868 if (cow_start != (u64)-1) {
1869 ret = fallback_to_cow(inode, locked_page,
1870 cow_start, found_key.offset - 1,
1871 page_started, nr_written);
1874 cow_start = (u64)-1;
1877 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1878 u64 orig_start = found_key.offset - extent_offset;
1879 struct extent_map *em;
1881 em = create_io_em(inode, cur_offset, num_bytes,
1883 disk_bytenr, /* block_start */
1884 num_bytes, /* block_len */
1885 disk_num_bytes, /* orig_block_len */
1886 ram_bytes, BTRFS_COMPRESS_NONE,
1887 BTRFS_ORDERED_PREALLOC);
1892 free_extent_map(em);
1893 ret = btrfs_add_ordered_extent(inode,
1894 cur_offset, num_bytes, num_bytes,
1895 disk_bytenr, num_bytes, 0,
1896 1 << BTRFS_ORDERED_PREALLOC,
1897 BTRFS_COMPRESS_NONE);
1899 btrfs_drop_extent_cache(inode, cur_offset,
1900 cur_offset + num_bytes - 1,
1905 ret = btrfs_add_ordered_extent(inode, cur_offset,
1906 num_bytes, num_bytes,
1907 disk_bytenr, num_bytes,
1909 1 << BTRFS_ORDERED_NOCOW,
1910 BTRFS_COMPRESS_NONE);
1916 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1919 if (btrfs_is_data_reloc_root(root))
1921 * Error handled later, as we must prevent
1922 * extent_clear_unlock_delalloc() in error handler
1923 * from freeing metadata of created ordered extent.
1925 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1928 extent_clear_unlock_delalloc(inode, cur_offset,
1929 cur_offset + num_bytes - 1,
1930 locked_page, EXTENT_LOCKED |
1932 EXTENT_CLEAR_DATA_RESV,
1933 PAGE_UNLOCK | PAGE_SET_ORDERED);
1935 cur_offset = extent_end;
1938 * btrfs_reloc_clone_csums() error, now we're OK to call error
1939 * handler, as metadata for created ordered extent will only
1940 * be freed by btrfs_finish_ordered_io().
1944 if (cur_offset > end)
1947 btrfs_release_path(path);
1949 if (cur_offset <= end && cow_start == (u64)-1)
1950 cow_start = cur_offset;
1952 if (cow_start != (u64)-1) {
1954 ret = fallback_to_cow(inode, locked_page, cow_start, end,
1955 page_started, nr_written);
1962 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1964 if (ret && cur_offset < end)
1965 extent_clear_unlock_delalloc(inode, cur_offset, end,
1966 locked_page, EXTENT_LOCKED |
1967 EXTENT_DELALLOC | EXTENT_DEFRAG |
1968 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1969 PAGE_START_WRITEBACK |
1970 PAGE_END_WRITEBACK);
1971 btrfs_free_path(path);
1975 static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end)
1977 if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) {
1978 if (inode->defrag_bytes &&
1979 test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG,
1988 * Function to process delayed allocation (create CoW) for ranges which are
1989 * being touched for the first time.
1991 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
1992 u64 start, u64 end, int *page_started, unsigned long *nr_written,
1993 struct writeback_control *wbc)
1996 const bool zoned = btrfs_is_zoned(inode->root->fs_info);
1999 * The range must cover part of the @locked_page, or the returned
2000 * @page_started can confuse the caller.
2002 ASSERT(!(end <= page_offset(locked_page) ||
2003 start >= page_offset(locked_page) + PAGE_SIZE));
2005 if (should_nocow(inode, start, end)) {
2007 * Normally on a zoned device we're only doing COW writes, but
2008 * in case of relocation on a zoned filesystem we have taken
2009 * precaution, that we're only writing sequentially. It's safe
2010 * to use run_delalloc_nocow() here, like for regular
2011 * preallocated inodes.
2014 (zoned && btrfs_is_data_reloc_root(inode->root)));
2015 ret = run_delalloc_nocow(inode, locked_page, start, end,
2016 page_started, nr_written);
2017 } else if (!inode_can_compress(inode) ||
2018 !inode_need_compress(inode, start, end)) {
2020 ret = run_delalloc_zoned(inode, locked_page, start, end,
2021 page_started, nr_written);
2023 ret = cow_file_range(inode, locked_page, start, end,
2024 page_started, nr_written, 1);
2026 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
2027 ret = cow_file_range_async(inode, wbc, locked_page, start, end,
2028 page_started, nr_written);
2032 btrfs_cleanup_ordered_extents(inode, locked_page, start,
2037 void btrfs_split_delalloc_extent(struct inode *inode,
2038 struct extent_state *orig, u64 split)
2042 /* not delalloc, ignore it */
2043 if (!(orig->state & EXTENT_DELALLOC))
2046 size = orig->end - orig->start + 1;
2047 if (size > BTRFS_MAX_EXTENT_SIZE) {
2052 * See the explanation in btrfs_merge_delalloc_extent, the same
2053 * applies here, just in reverse.
2055 new_size = orig->end - split + 1;
2056 num_extents = count_max_extents(new_size);
2057 new_size = split - orig->start;
2058 num_extents += count_max_extents(new_size);
2059 if (count_max_extents(size) >= num_extents)
2063 spin_lock(&BTRFS_I(inode)->lock);
2064 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
2065 spin_unlock(&BTRFS_I(inode)->lock);
2069 * Handle merged delayed allocation extents so we can keep track of new extents
2070 * that are just merged onto old extents, such as when we are doing sequential
2071 * writes, so we can properly account for the metadata space we'll need.
2073 void btrfs_merge_delalloc_extent(struct inode *inode, struct extent_state *new,
2074 struct extent_state *other)
2076 u64 new_size, old_size;
2079 /* not delalloc, ignore it */
2080 if (!(other->state & EXTENT_DELALLOC))
2083 if (new->start > other->start)
2084 new_size = new->end - other->start + 1;
2086 new_size = other->end - new->start + 1;
2088 /* we're not bigger than the max, unreserve the space and go */
2089 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
2090 spin_lock(&BTRFS_I(inode)->lock);
2091 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
2092 spin_unlock(&BTRFS_I(inode)->lock);
2097 * We have to add up either side to figure out how many extents were
2098 * accounted for before we merged into one big extent. If the number of
2099 * extents we accounted for is <= the amount we need for the new range
2100 * then we can return, otherwise drop. Think of it like this
2104 * So we've grown the extent by a MAX_SIZE extent, this would mean we
2105 * need 2 outstanding extents, on one side we have 1 and the other side
2106 * we have 1 so they are == and we can return. But in this case
2108 * [MAX_SIZE+4k][MAX_SIZE+4k]
2110 * Each range on their own accounts for 2 extents, but merged together
2111 * they are only 3 extents worth of accounting, so we need to drop in
2114 old_size = other->end - other->start + 1;
2115 num_extents = count_max_extents(old_size);
2116 old_size = new->end - new->start + 1;
2117 num_extents += count_max_extents(old_size);
2118 if (count_max_extents(new_size) >= num_extents)
2121 spin_lock(&BTRFS_I(inode)->lock);
2122 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
2123 spin_unlock(&BTRFS_I(inode)->lock);
2126 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
2127 struct inode *inode)
2129 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2131 spin_lock(&root->delalloc_lock);
2132 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
2133 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
2134 &root->delalloc_inodes);
2135 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2136 &BTRFS_I(inode)->runtime_flags);
2137 root->nr_delalloc_inodes++;
2138 if (root->nr_delalloc_inodes == 1) {
2139 spin_lock(&fs_info->delalloc_root_lock);
2140 BUG_ON(!list_empty(&root->delalloc_root));
2141 list_add_tail(&root->delalloc_root,
2142 &fs_info->delalloc_roots);
2143 spin_unlock(&fs_info->delalloc_root_lock);
2146 spin_unlock(&root->delalloc_lock);
2150 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
2151 struct btrfs_inode *inode)
2153 struct btrfs_fs_info *fs_info = root->fs_info;
2155 if (!list_empty(&inode->delalloc_inodes)) {
2156 list_del_init(&inode->delalloc_inodes);
2157 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2158 &inode->runtime_flags);
2159 root->nr_delalloc_inodes--;
2160 if (!root->nr_delalloc_inodes) {
2161 ASSERT(list_empty(&root->delalloc_inodes));
2162 spin_lock(&fs_info->delalloc_root_lock);
2163 BUG_ON(list_empty(&root->delalloc_root));
2164 list_del_init(&root->delalloc_root);
2165 spin_unlock(&fs_info->delalloc_root_lock);
2170 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
2171 struct btrfs_inode *inode)
2173 spin_lock(&root->delalloc_lock);
2174 __btrfs_del_delalloc_inode(root, inode);
2175 spin_unlock(&root->delalloc_lock);
2179 * Properly track delayed allocation bytes in the inode and to maintain the
2180 * list of inodes that have pending delalloc work to be done.
2182 void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state,
2185 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2187 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
2190 * set_bit and clear bit hooks normally require _irqsave/restore
2191 * but in this case, we are only testing for the DELALLOC
2192 * bit, which is only set or cleared with irqs on
2194 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2195 struct btrfs_root *root = BTRFS_I(inode)->root;
2196 u64 len = state->end + 1 - state->start;
2197 u32 num_extents = count_max_extents(len);
2198 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
2200 spin_lock(&BTRFS_I(inode)->lock);
2201 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
2202 spin_unlock(&BTRFS_I(inode)->lock);
2204 /* For sanity tests */
2205 if (btrfs_is_testing(fs_info))
2208 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2209 fs_info->delalloc_batch);
2210 spin_lock(&BTRFS_I(inode)->lock);
2211 BTRFS_I(inode)->delalloc_bytes += len;
2212 if (*bits & EXTENT_DEFRAG)
2213 BTRFS_I(inode)->defrag_bytes += len;
2214 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2215 &BTRFS_I(inode)->runtime_flags))
2216 btrfs_add_delalloc_inodes(root, inode);
2217 spin_unlock(&BTRFS_I(inode)->lock);
2220 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2221 (*bits & EXTENT_DELALLOC_NEW)) {
2222 spin_lock(&BTRFS_I(inode)->lock);
2223 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
2225 spin_unlock(&BTRFS_I(inode)->lock);
2230 * Once a range is no longer delalloc this function ensures that proper
2231 * accounting happens.
2233 void btrfs_clear_delalloc_extent(struct inode *vfs_inode,
2234 struct extent_state *state, unsigned *bits)
2236 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
2237 struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb);
2238 u64 len = state->end + 1 - state->start;
2239 u32 num_extents = count_max_extents(len);
2241 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
2242 spin_lock(&inode->lock);
2243 inode->defrag_bytes -= len;
2244 spin_unlock(&inode->lock);
2248 * set_bit and clear bit hooks normally require _irqsave/restore
2249 * but in this case, we are only testing for the DELALLOC
2250 * bit, which is only set or cleared with irqs on
2252 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2253 struct btrfs_root *root = inode->root;
2254 bool do_list = !btrfs_is_free_space_inode(inode);
2256 spin_lock(&inode->lock);
2257 btrfs_mod_outstanding_extents(inode, -num_extents);
2258 spin_unlock(&inode->lock);
2261 * We don't reserve metadata space for space cache inodes so we
2262 * don't need to call delalloc_release_metadata if there is an
2265 if (*bits & EXTENT_CLEAR_META_RESV &&
2266 root != fs_info->tree_root)
2267 btrfs_delalloc_release_metadata(inode, len, false);
2269 /* For sanity tests. */
2270 if (btrfs_is_testing(fs_info))
2273 if (!btrfs_is_data_reloc_root(root) &&
2274 do_list && !(state->state & EXTENT_NORESERVE) &&
2275 (*bits & EXTENT_CLEAR_DATA_RESV))
2276 btrfs_free_reserved_data_space_noquota(fs_info, len);
2278 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2279 fs_info->delalloc_batch);
2280 spin_lock(&inode->lock);
2281 inode->delalloc_bytes -= len;
2282 if (do_list && inode->delalloc_bytes == 0 &&
2283 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2284 &inode->runtime_flags))
2285 btrfs_del_delalloc_inode(root, inode);
2286 spin_unlock(&inode->lock);
2289 if ((state->state & EXTENT_DELALLOC_NEW) &&
2290 (*bits & EXTENT_DELALLOC_NEW)) {
2291 spin_lock(&inode->lock);
2292 ASSERT(inode->new_delalloc_bytes >= len);
2293 inode->new_delalloc_bytes -= len;
2294 if (*bits & EXTENT_ADD_INODE_BYTES)
2295 inode_add_bytes(&inode->vfs_inode, len);
2296 spin_unlock(&inode->lock);
2301 * in order to insert checksums into the metadata in large chunks,
2302 * we wait until bio submission time. All the pages in the bio are
2303 * checksummed and sums are attached onto the ordered extent record.
2305 * At IO completion time the cums attached on the ordered extent record
2306 * are inserted into the btree
2308 static blk_status_t btrfs_submit_bio_start(struct inode *inode, struct bio *bio,
2309 u64 dio_file_offset)
2311 return btrfs_csum_one_bio(BTRFS_I(inode), bio, (u64)-1, false);
2315 * Split an extent_map at [start, start + len]
2317 * This function is intended to be used only for extract_ordered_extent().
2319 static int split_zoned_em(struct btrfs_inode *inode, u64 start, u64 len,
2322 struct extent_map_tree *em_tree = &inode->extent_tree;
2323 struct extent_map *em;
2324 struct extent_map *split_pre = NULL;
2325 struct extent_map *split_mid = NULL;
2326 struct extent_map *split_post = NULL;
2328 unsigned long flags;
2331 if (pre == 0 && post == 0)
2334 split_pre = alloc_extent_map();
2336 split_mid = alloc_extent_map();
2338 split_post = alloc_extent_map();
2339 if (!split_pre || (pre && !split_mid) || (post && !split_post)) {
2344 ASSERT(pre + post < len);
2346 lock_extent(&inode->io_tree, start, start + len - 1);
2347 write_lock(&em_tree->lock);
2348 em = lookup_extent_mapping(em_tree, start, len);
2354 ASSERT(em->len == len);
2355 ASSERT(!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags));
2356 ASSERT(em->block_start < EXTENT_MAP_LAST_BYTE);
2357 ASSERT(test_bit(EXTENT_FLAG_PINNED, &em->flags));
2358 ASSERT(!test_bit(EXTENT_FLAG_LOGGING, &em->flags));
2359 ASSERT(!list_empty(&em->list));
2362 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
2364 /* First, replace the em with a new extent_map starting from * em->start */
2365 split_pre->start = em->start;
2366 split_pre->len = (pre ? pre : em->len - post);
2367 split_pre->orig_start = split_pre->start;
2368 split_pre->block_start = em->block_start;
2369 split_pre->block_len = split_pre->len;
2370 split_pre->orig_block_len = split_pre->block_len;
2371 split_pre->ram_bytes = split_pre->len;
2372 split_pre->flags = flags;
2373 split_pre->compress_type = em->compress_type;
2374 split_pre->generation = em->generation;
2376 replace_extent_mapping(em_tree, em, split_pre, 1);
2379 * Now we only have an extent_map at:
2380 * [em->start, em->start + pre] if pre != 0
2381 * [em->start, em->start + em->len - post] if pre == 0
2385 /* Insert the middle extent_map */
2386 split_mid->start = em->start + pre;
2387 split_mid->len = em->len - pre - post;
2388 split_mid->orig_start = split_mid->start;
2389 split_mid->block_start = em->block_start + pre;
2390 split_mid->block_len = split_mid->len;
2391 split_mid->orig_block_len = split_mid->block_len;
2392 split_mid->ram_bytes = split_mid->len;
2393 split_mid->flags = flags;
2394 split_mid->compress_type = em->compress_type;
2395 split_mid->generation = em->generation;
2396 add_extent_mapping(em_tree, split_mid, 1);
2400 split_post->start = em->start + em->len - post;
2401 split_post->len = post;
2402 split_post->orig_start = split_post->start;
2403 split_post->block_start = em->block_start + em->len - post;
2404 split_post->block_len = split_post->len;
2405 split_post->orig_block_len = split_post->block_len;
2406 split_post->ram_bytes = split_post->len;
2407 split_post->flags = flags;
2408 split_post->compress_type = em->compress_type;
2409 split_post->generation = em->generation;
2410 add_extent_mapping(em_tree, split_post, 1);
2414 free_extent_map(em);
2415 /* Once for the tree */
2416 free_extent_map(em);
2419 write_unlock(&em_tree->lock);
2420 unlock_extent(&inode->io_tree, start, start + len - 1);
2422 free_extent_map(split_pre);
2423 free_extent_map(split_mid);
2424 free_extent_map(split_post);
2429 static blk_status_t extract_ordered_extent(struct btrfs_inode *inode,
2430 struct bio *bio, loff_t file_offset)
2432 struct btrfs_ordered_extent *ordered;
2433 u64 start = (u64)bio->bi_iter.bi_sector << SECTOR_SHIFT;
2435 u64 len = bio->bi_iter.bi_size;
2436 u64 end = start + len;
2441 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
2442 if (WARN_ON_ONCE(!ordered))
2443 return BLK_STS_IOERR;
2445 /* No need to split */
2446 if (ordered->disk_num_bytes == len)
2449 /* We cannot split once end_bio'd ordered extent */
2450 if (WARN_ON_ONCE(ordered->bytes_left != ordered->disk_num_bytes)) {
2455 /* We cannot split a compressed ordered extent */
2456 if (WARN_ON_ONCE(ordered->disk_num_bytes != ordered->num_bytes)) {
2461 ordered_end = ordered->disk_bytenr + ordered->disk_num_bytes;
2462 /* bio must be in one ordered extent */
2463 if (WARN_ON_ONCE(start < ordered->disk_bytenr || end > ordered_end)) {
2468 /* Checksum list should be empty */
2469 if (WARN_ON_ONCE(!list_empty(&ordered->list))) {
2474 file_len = ordered->num_bytes;
2475 pre = start - ordered->disk_bytenr;
2476 post = ordered_end - end;
2478 ret = btrfs_split_ordered_extent(ordered, pre, post);
2481 ret = split_zoned_em(inode, file_offset, file_len, pre, post);
2484 btrfs_put_ordered_extent(ordered);
2486 return errno_to_blk_status(ret);
2490 * extent_io.c submission hook. This does the right thing for csum calculation
2491 * on write, or reading the csums from the tree before a read.
2493 * Rules about async/sync submit,
2494 * a) read: sync submit
2496 * b) write without checksum: sync submit
2498 * c) write with checksum:
2499 * c-1) if bio is issued by fsync: sync submit
2500 * (sync_writers != 0)
2502 * c-2) if root is reloc root: sync submit
2503 * (only in case of buffered IO)
2505 * c-3) otherwise: async submit
2507 blk_status_t btrfs_submit_data_bio(struct inode *inode, struct bio *bio,
2508 int mirror_num, unsigned long bio_flags)
2511 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2512 struct btrfs_root *root = BTRFS_I(inode)->root;
2513 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
2514 blk_status_t ret = 0;
2516 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
2518 skip_sum = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) ||
2519 test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2521 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
2522 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
2524 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
2525 struct page *page = bio_first_bvec_all(bio)->bv_page;
2526 loff_t file_offset = page_offset(page);
2528 ret = extract_ordered_extent(BTRFS_I(inode), bio, file_offset);
2533 if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
2534 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
2538 if (bio_flags & EXTENT_BIO_COMPRESSED) {
2539 ret = btrfs_submit_compressed_read(inode, bio,
2545 * Lookup bio sums does extra checks around whether we
2546 * need to csum or not, which is why we ignore skip_sum
2549 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
2554 } else if (async && !skip_sum) {
2555 /* csum items have already been cloned */
2556 if (btrfs_is_data_reloc_root(root))
2558 /* we're doing a write, do the async checksumming */
2559 ret = btrfs_wq_submit_bio(inode, bio, mirror_num, bio_flags,
2560 0, btrfs_submit_bio_start);
2562 } else if (!skip_sum) {
2563 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, (u64)-1, false);
2569 ret = btrfs_map_bio(fs_info, bio, mirror_num);
2573 bio->bi_status = ret;
2580 * given a list of ordered sums record them in the inode. This happens
2581 * at IO completion time based on sums calculated at bio submission time.
2583 static int add_pending_csums(struct btrfs_trans_handle *trans,
2584 struct list_head *list)
2586 struct btrfs_ordered_sum *sum;
2587 struct btrfs_root *csum_root = NULL;
2590 list_for_each_entry(sum, list, list) {
2591 trans->adding_csums = true;
2593 csum_root = btrfs_csum_root(trans->fs_info,
2595 ret = btrfs_csum_file_blocks(trans, csum_root, sum);
2596 trans->adding_csums = false;
2603 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2606 struct extent_state **cached_state)
2608 u64 search_start = start;
2609 const u64 end = start + len - 1;
2611 while (search_start < end) {
2612 const u64 search_len = end - search_start + 1;
2613 struct extent_map *em;
2617 em = btrfs_get_extent(inode, NULL, 0, search_start, search_len);
2621 if (em->block_start != EXTENT_MAP_HOLE)
2625 if (em->start < search_start)
2626 em_len -= search_start - em->start;
2627 if (em_len > search_len)
2628 em_len = search_len;
2630 ret = set_extent_bit(&inode->io_tree, search_start,
2631 search_start + em_len - 1,
2632 EXTENT_DELALLOC_NEW, 0, NULL, cached_state,
2635 search_start = extent_map_end(em);
2636 free_extent_map(em);
2643 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2644 unsigned int extra_bits,
2645 struct extent_state **cached_state)
2647 WARN_ON(PAGE_ALIGNED(end));
2649 if (start >= i_size_read(&inode->vfs_inode) &&
2650 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2652 * There can't be any extents following eof in this case so just
2653 * set the delalloc new bit for the range directly.
2655 extra_bits |= EXTENT_DELALLOC_NEW;
2659 ret = btrfs_find_new_delalloc_bytes(inode, start,
2666 return set_extent_delalloc(&inode->io_tree, start, end, extra_bits,
2670 /* see btrfs_writepage_start_hook for details on why this is required */
2671 struct btrfs_writepage_fixup {
2673 struct inode *inode;
2674 struct btrfs_work work;
2677 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2679 struct btrfs_writepage_fixup *fixup;
2680 struct btrfs_ordered_extent *ordered;
2681 struct extent_state *cached_state = NULL;
2682 struct extent_changeset *data_reserved = NULL;
2684 struct btrfs_inode *inode;
2688 bool free_delalloc_space = true;
2690 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2692 inode = BTRFS_I(fixup->inode);
2693 page_start = page_offset(page);
2694 page_end = page_offset(page) + PAGE_SIZE - 1;
2697 * This is similar to page_mkwrite, we need to reserve the space before
2698 * we take the page lock.
2700 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2706 * Before we queued this fixup, we took a reference on the page.
2707 * page->mapping may go NULL, but it shouldn't be moved to a different
2710 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2712 * Unfortunately this is a little tricky, either
2714 * 1) We got here and our page had already been dealt with and
2715 * we reserved our space, thus ret == 0, so we need to just
2716 * drop our space reservation and bail. This can happen the
2717 * first time we come into the fixup worker, or could happen
2718 * while waiting for the ordered extent.
2719 * 2) Our page was already dealt with, but we happened to get an
2720 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2721 * this case we obviously don't have anything to release, but
2722 * because the page was already dealt with we don't want to
2723 * mark the page with an error, so make sure we're resetting
2724 * ret to 0. This is why we have this check _before_ the ret
2725 * check, because we do not want to have a surprise ENOSPC
2726 * when the page was already properly dealt with.
2729 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2730 btrfs_delalloc_release_space(inode, data_reserved,
2731 page_start, PAGE_SIZE,
2739 * We can't mess with the page state unless it is locked, so now that
2740 * it is locked bail if we failed to make our space reservation.
2745 lock_extent_bits(&inode->io_tree, page_start, page_end, &cached_state);
2747 /* already ordered? We're done */
2748 if (PageOrdered(page))
2751 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2753 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2756 btrfs_start_ordered_extent(ordered, 1);
2757 btrfs_put_ordered_extent(ordered);
2761 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2767 * Everything went as planned, we're now the owner of a dirty page with
2768 * delayed allocation bits set and space reserved for our COW
2771 * The page was dirty when we started, nothing should have cleaned it.
2773 BUG_ON(!PageDirty(page));
2774 free_delalloc_space = false;
2776 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2777 if (free_delalloc_space)
2778 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2780 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2785 * We hit ENOSPC or other errors. Update the mapping and page
2786 * to reflect the errors and clean the page.
2788 mapping_set_error(page->mapping, ret);
2789 end_extent_writepage(page, ret, page_start, page_end);
2790 clear_page_dirty_for_io(page);
2793 btrfs_page_clear_checked(inode->root->fs_info, page, page_start, PAGE_SIZE);
2797 extent_changeset_free(data_reserved);
2799 * As a precaution, do a delayed iput in case it would be the last iput
2800 * that could need flushing space. Recursing back to fixup worker would
2803 btrfs_add_delayed_iput(&inode->vfs_inode);
2807 * There are a few paths in the higher layers of the kernel that directly
2808 * set the page dirty bit without asking the filesystem if it is a
2809 * good idea. This causes problems because we want to make sure COW
2810 * properly happens and the data=ordered rules are followed.
2812 * In our case any range that doesn't have the ORDERED bit set
2813 * hasn't been properly setup for IO. We kick off an async process
2814 * to fix it up. The async helper will wait for ordered extents, set
2815 * the delalloc bit and make it safe to write the page.
2817 int btrfs_writepage_cow_fixup(struct page *page)
2819 struct inode *inode = page->mapping->host;
2820 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2821 struct btrfs_writepage_fixup *fixup;
2823 /* This page has ordered extent covering it already */
2824 if (PageOrdered(page))
2828 * PageChecked is set below when we create a fixup worker for this page,
2829 * don't try to create another one if we're already PageChecked()
2831 * The extent_io writepage code will redirty the page if we send back
2834 if (PageChecked(page))
2837 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2842 * We are already holding a reference to this inode from
2843 * write_cache_pages. We need to hold it because the space reservation
2844 * takes place outside of the page lock, and we can't trust
2845 * page->mapping outside of the page lock.
2848 btrfs_page_set_checked(fs_info, page, page_offset(page), PAGE_SIZE);
2850 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
2852 fixup->inode = inode;
2853 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2858 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2859 struct btrfs_inode *inode, u64 file_pos,
2860 struct btrfs_file_extent_item *stack_fi,
2861 const bool update_inode_bytes,
2862 u64 qgroup_reserved)
2864 struct btrfs_root *root = inode->root;
2865 const u64 sectorsize = root->fs_info->sectorsize;
2866 struct btrfs_path *path;
2867 struct extent_buffer *leaf;
2868 struct btrfs_key ins;
2869 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
2870 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
2871 u64 offset = btrfs_stack_file_extent_offset(stack_fi);
2872 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
2873 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
2874 struct btrfs_drop_extents_args drop_args = { 0 };
2877 path = btrfs_alloc_path();
2882 * we may be replacing one extent in the tree with another.
2883 * The new extent is pinned in the extent map, and we don't want
2884 * to drop it from the cache until it is completely in the btree.
2886 * So, tell btrfs_drop_extents to leave this extent in the cache.
2887 * the caller is expected to unpin it and allow it to be merged
2890 drop_args.path = path;
2891 drop_args.start = file_pos;
2892 drop_args.end = file_pos + num_bytes;
2893 drop_args.replace_extent = true;
2894 drop_args.extent_item_size = sizeof(*stack_fi);
2895 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2899 if (!drop_args.extent_inserted) {
2900 ins.objectid = btrfs_ino(inode);
2901 ins.offset = file_pos;
2902 ins.type = BTRFS_EXTENT_DATA_KEY;
2904 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2909 leaf = path->nodes[0];
2910 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
2911 write_extent_buffer(leaf, stack_fi,
2912 btrfs_item_ptr_offset(leaf, path->slots[0]),
2913 sizeof(struct btrfs_file_extent_item));
2915 btrfs_mark_buffer_dirty(leaf);
2916 btrfs_release_path(path);
2919 * If we dropped an inline extent here, we know the range where it is
2920 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
2921 * number of bytes only for that range containing the inline extent.
2922 * The remaining of the range will be processed when clearning the
2923 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
2925 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
2926 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
2928 inline_size = drop_args.bytes_found - inline_size;
2929 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
2930 drop_args.bytes_found -= inline_size;
2931 num_bytes -= sectorsize;
2934 if (update_inode_bytes)
2935 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
2937 ins.objectid = disk_bytenr;
2938 ins.offset = disk_num_bytes;
2939 ins.type = BTRFS_EXTENT_ITEM_KEY;
2941 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
2945 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
2947 qgroup_reserved, &ins);
2949 btrfs_free_path(path);
2954 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2957 struct btrfs_block_group *cache;
2959 cache = btrfs_lookup_block_group(fs_info, start);
2962 spin_lock(&cache->lock);
2963 cache->delalloc_bytes -= len;
2964 spin_unlock(&cache->lock);
2966 btrfs_put_block_group(cache);
2969 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
2970 struct btrfs_ordered_extent *oe)
2972 struct btrfs_file_extent_item stack_fi;
2973 bool update_inode_bytes;
2974 u64 num_bytes = oe->num_bytes;
2975 u64 ram_bytes = oe->ram_bytes;
2977 memset(&stack_fi, 0, sizeof(stack_fi));
2978 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
2979 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
2980 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
2981 oe->disk_num_bytes);
2982 btrfs_set_stack_file_extent_offset(&stack_fi, oe->offset);
2983 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags))
2984 num_bytes = ram_bytes = oe->truncated_len;
2985 btrfs_set_stack_file_extent_num_bytes(&stack_fi, num_bytes);
2986 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, ram_bytes);
2987 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
2988 /* Encryption and other encoding is reserved and all 0 */
2991 * For delalloc, when completing an ordered extent we update the inode's
2992 * bytes when clearing the range in the inode's io tree, so pass false
2993 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
2994 * except if the ordered extent was truncated.
2996 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
2997 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
2999 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
3000 oe->file_offset, &stack_fi,
3001 update_inode_bytes, oe->qgroup_rsv);
3005 * As ordered data IO finishes, this gets called so we can finish
3006 * an ordered extent if the range of bytes in the file it covers are
3009 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
3011 struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
3012 struct btrfs_root *root = inode->root;
3013 struct btrfs_fs_info *fs_info = root->fs_info;
3014 struct btrfs_trans_handle *trans = NULL;
3015 struct extent_io_tree *io_tree = &inode->io_tree;
3016 struct extent_state *cached_state = NULL;
3018 int compress_type = 0;
3020 u64 logical_len = ordered_extent->num_bytes;
3021 bool freespace_inode;
3022 bool truncated = false;
3023 bool clear_reserved_extent = true;
3024 unsigned int clear_bits = EXTENT_DEFRAG;
3026 start = ordered_extent->file_offset;
3027 end = start + ordered_extent->num_bytes - 1;
3029 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3030 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
3031 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
3032 clear_bits |= EXTENT_DELALLOC_NEW;
3034 freespace_inode = btrfs_is_free_space_inode(inode);
3036 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
3041 /* A valid bdev implies a write on a sequential zone */
3042 if (ordered_extent->bdev) {
3043 btrfs_rewrite_logical_zoned(ordered_extent);
3044 btrfs_zone_finish_endio(fs_info, ordered_extent->disk_bytenr,
3045 ordered_extent->disk_num_bytes);
3048 btrfs_free_io_failure_record(inode, start, end);
3050 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
3052 logical_len = ordered_extent->truncated_len;
3053 /* Truncated the entire extent, don't bother adding */
3058 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
3059 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
3061 btrfs_inode_safe_disk_i_size_write(inode, 0);
3062 if (freespace_inode)
3063 trans = btrfs_join_transaction_spacecache(root);
3065 trans = btrfs_join_transaction(root);
3066 if (IS_ERR(trans)) {
3067 ret = PTR_ERR(trans);
3071 trans->block_rsv = &inode->block_rsv;
3072 ret = btrfs_update_inode_fallback(trans, root, inode);
3073 if (ret) /* -ENOMEM or corruption */
3074 btrfs_abort_transaction(trans, ret);
3078 clear_bits |= EXTENT_LOCKED;
3079 lock_extent_bits(io_tree, start, end, &cached_state);
3081 if (freespace_inode)
3082 trans = btrfs_join_transaction_spacecache(root);
3084 trans = btrfs_join_transaction(root);
3085 if (IS_ERR(trans)) {
3086 ret = PTR_ERR(trans);
3091 trans->block_rsv = &inode->block_rsv;
3093 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3094 compress_type = ordered_extent->compress_type;
3095 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3096 BUG_ON(compress_type);
3097 ret = btrfs_mark_extent_written(trans, inode,
3098 ordered_extent->file_offset,
3099 ordered_extent->file_offset +
3102 BUG_ON(root == fs_info->tree_root);
3103 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
3105 clear_reserved_extent = false;
3106 btrfs_release_delalloc_bytes(fs_info,
3107 ordered_extent->disk_bytenr,
3108 ordered_extent->disk_num_bytes);
3111 unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset,
3112 ordered_extent->num_bytes, trans->transid);
3114 btrfs_abort_transaction(trans, ret);
3118 ret = add_pending_csums(trans, &ordered_extent->list);
3120 btrfs_abort_transaction(trans, ret);
3125 * If this is a new delalloc range, clear its new delalloc flag to
3126 * update the inode's number of bytes. This needs to be done first
3127 * before updating the inode item.
3129 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
3130 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
3131 clear_extent_bit(&inode->io_tree, start, end,
3132 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
3133 0, 0, &cached_state);
3135 btrfs_inode_safe_disk_i_size_write(inode, 0);
3136 ret = btrfs_update_inode_fallback(trans, root, inode);
3137 if (ret) { /* -ENOMEM or corruption */
3138 btrfs_abort_transaction(trans, ret);
3143 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
3144 (clear_bits & EXTENT_LOCKED) ? 1 : 0, 0,
3148 btrfs_end_transaction(trans);
3150 if (ret || truncated) {
3151 u64 unwritten_start = start;
3154 * If we failed to finish this ordered extent for any reason we
3155 * need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3156 * extent, and mark the inode with the error if it wasn't
3157 * already set. Any error during writeback would have already
3158 * set the mapping error, so we need to set it if we're the ones
3159 * marking this ordered extent as failed.
3161 if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR,
3162 &ordered_extent->flags))
3163 mapping_set_error(ordered_extent->inode->i_mapping, -EIO);
3166 unwritten_start += logical_len;
3167 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
3169 /* Drop the cache for the part of the extent we didn't write. */
3170 btrfs_drop_extent_cache(inode, unwritten_start, end, 0);
3173 * If the ordered extent had an IOERR or something else went
3174 * wrong we need to return the space for this ordered extent
3175 * back to the allocator. We only free the extent in the
3176 * truncated case if we didn't write out the extent at all.
3178 * If we made it past insert_reserved_file_extent before we
3179 * errored out then we don't need to do this as the accounting
3180 * has already been done.
3182 if ((ret || !logical_len) &&
3183 clear_reserved_extent &&
3184 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3185 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3187 * Discard the range before returning it back to the
3190 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3191 btrfs_discard_extent(fs_info,
3192 ordered_extent->disk_bytenr,
3193 ordered_extent->disk_num_bytes,
3195 btrfs_free_reserved_extent(fs_info,
3196 ordered_extent->disk_bytenr,
3197 ordered_extent->disk_num_bytes, 1);
3202 * This needs to be done to make sure anybody waiting knows we are done
3203 * updating everything for this ordered extent.
3205 btrfs_remove_ordered_extent(inode, ordered_extent);
3208 btrfs_put_ordered_extent(ordered_extent);
3209 /* once for the tree */
3210 btrfs_put_ordered_extent(ordered_extent);
3215 static void finish_ordered_fn(struct btrfs_work *work)
3217 struct btrfs_ordered_extent *ordered_extent;
3218 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3219 btrfs_finish_ordered_io(ordered_extent);
3222 void btrfs_writepage_endio_finish_ordered(struct btrfs_inode *inode,
3223 struct page *page, u64 start,
3224 u64 end, bool uptodate)
3226 trace_btrfs_writepage_end_io_hook(inode, start, end, uptodate);
3228 btrfs_mark_ordered_io_finished(inode, page, start, end + 1 - start,
3229 finish_ordered_fn, uptodate);
3233 * check_data_csum - verify checksum of one sector of uncompressed data
3235 * @io_bio: btrfs_io_bio which contains the csum
3236 * @bio_offset: offset to the beginning of the bio (in bytes)
3237 * @page: page where is the data to be verified
3238 * @pgoff: offset inside the page
3239 * @start: logical offset in the file
3241 * The length of such check is always one sector size.
3243 static int check_data_csum(struct inode *inode, struct btrfs_bio *bbio,
3244 u32 bio_offset, struct page *page, u32 pgoff,
3247 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3248 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3250 u32 len = fs_info->sectorsize;
3251 const u32 csum_size = fs_info->csum_size;
3252 unsigned int offset_sectors;
3254 u8 csum[BTRFS_CSUM_SIZE];
3256 ASSERT(pgoff + len <= PAGE_SIZE);
3258 offset_sectors = bio_offset >> fs_info->sectorsize_bits;
3259 csum_expected = ((u8 *)bbio->csum) + offset_sectors * csum_size;
3261 kaddr = kmap_atomic(page);
3262 shash->tfm = fs_info->csum_shash;
3264 crypto_shash_digest(shash, kaddr + pgoff, len, csum);
3266 if (memcmp(csum, csum_expected, csum_size))
3269 kunmap_atomic(kaddr);
3272 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3275 btrfs_dev_stat_inc_and_print(bbio->device,
3276 BTRFS_DEV_STAT_CORRUPTION_ERRS);
3277 memset(kaddr + pgoff, 1, len);
3278 flush_dcache_page(page);
3279 kunmap_atomic(kaddr);
3284 * When reads are done, we need to check csums to verify the data is correct.
3285 * if there's a match, we allow the bio to finish. If not, the code in
3286 * extent_io.c will try to find good copies for us.
3288 * @bio_offset: offset to the beginning of the bio (in bytes)
3289 * @start: file offset of the range start
3290 * @end: file offset of the range end (inclusive)
3292 * Return a bitmap where bit set means a csum mismatch, and bit not set means
3295 unsigned int btrfs_verify_data_csum(struct btrfs_bio *bbio,
3296 u32 bio_offset, struct page *page,
3299 struct inode *inode = page->mapping->host;
3300 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3301 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3302 struct btrfs_root *root = BTRFS_I(inode)->root;
3303 const u32 sectorsize = root->fs_info->sectorsize;
3305 unsigned int result = 0;
3307 if (btrfs_page_test_checked(fs_info, page, start, end + 1 - start)) {
3308 btrfs_page_clear_checked(fs_info, page, start, end + 1 - start);
3313 * This only happens for NODATASUM or compressed read.
3314 * Normally this should be covered by above check for compressed read
3315 * or the next check for NODATASUM. Just do a quicker exit here.
3317 if (bbio->csum == NULL)
3320 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3323 if (unlikely(test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state)))
3326 ASSERT(page_offset(page) <= start &&
3327 end <= page_offset(page) + PAGE_SIZE - 1);
3328 for (pg_off = offset_in_page(start);
3329 pg_off < offset_in_page(end);
3330 pg_off += sectorsize, bio_offset += sectorsize) {
3331 u64 file_offset = pg_off + page_offset(page);
3334 if (btrfs_is_data_reloc_root(root) &&
3335 test_range_bit(io_tree, file_offset,
3336 file_offset + sectorsize - 1,
3337 EXTENT_NODATASUM, 1, NULL)) {
3338 /* Skip the range without csum for data reloc inode */
3339 clear_extent_bits(io_tree, file_offset,
3340 file_offset + sectorsize - 1,
3344 ret = check_data_csum(inode, bbio, bio_offset, page, pg_off,
3345 page_offset(page) + pg_off);
3347 const int nr_bit = (pg_off - offset_in_page(start)) >>
3348 root->fs_info->sectorsize_bits;
3350 result |= (1U << nr_bit);
3357 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3359 * @inode: The inode we want to perform iput on
3361 * This function uses the generic vfs_inode::i_count to track whether we should
3362 * just decrement it (in case it's > 1) or if this is the last iput then link
3363 * the inode to the delayed iput machinery. Delayed iputs are processed at
3364 * transaction commit time/superblock commit/cleaner kthread.
3366 void btrfs_add_delayed_iput(struct inode *inode)
3368 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3369 struct btrfs_inode *binode = BTRFS_I(inode);
3371 if (atomic_add_unless(&inode->i_count, -1, 1))
3374 atomic_inc(&fs_info->nr_delayed_iputs);
3375 spin_lock(&fs_info->delayed_iput_lock);
3376 ASSERT(list_empty(&binode->delayed_iput));
3377 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3378 spin_unlock(&fs_info->delayed_iput_lock);
3379 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3380 wake_up_process(fs_info->cleaner_kthread);
3383 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3384 struct btrfs_inode *inode)
3386 list_del_init(&inode->delayed_iput);
3387 spin_unlock(&fs_info->delayed_iput_lock);
3388 iput(&inode->vfs_inode);
3389 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3390 wake_up(&fs_info->delayed_iputs_wait);
3391 spin_lock(&fs_info->delayed_iput_lock);
3394 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3395 struct btrfs_inode *inode)
3397 if (!list_empty(&inode->delayed_iput)) {
3398 spin_lock(&fs_info->delayed_iput_lock);
3399 if (!list_empty(&inode->delayed_iput))
3400 run_delayed_iput_locked(fs_info, inode);
3401 spin_unlock(&fs_info->delayed_iput_lock);
3405 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3408 spin_lock(&fs_info->delayed_iput_lock);
3409 while (!list_empty(&fs_info->delayed_iputs)) {
3410 struct btrfs_inode *inode;
3412 inode = list_first_entry(&fs_info->delayed_iputs,
3413 struct btrfs_inode, delayed_iput);
3414 run_delayed_iput_locked(fs_info, inode);
3415 cond_resched_lock(&fs_info->delayed_iput_lock);
3417 spin_unlock(&fs_info->delayed_iput_lock);
3421 * Wait for flushing all delayed iputs
3423 * @fs_info: the filesystem
3425 * This will wait on any delayed iputs that are currently running with KILLABLE
3426 * set. Once they are all done running we will return, unless we are killed in
3427 * which case we return EINTR. This helps in user operations like fallocate etc
3428 * that might get blocked on the iputs.
3430 * Return EINTR if we were killed, 0 if nothing's pending
3432 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3434 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3435 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3442 * This creates an orphan entry for the given inode in case something goes wrong
3443 * in the middle of an unlink.
3445 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3446 struct btrfs_inode *inode)
3450 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3451 if (ret && ret != -EEXIST) {
3452 btrfs_abort_transaction(trans, ret);
3460 * We have done the delete so we can go ahead and remove the orphan item for
3461 * this particular inode.
3463 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3464 struct btrfs_inode *inode)
3466 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3470 * this cleans up any orphans that may be left on the list from the last use
3473 int btrfs_orphan_cleanup(struct btrfs_root *root)
3475 struct btrfs_fs_info *fs_info = root->fs_info;
3476 struct btrfs_path *path;
3477 struct extent_buffer *leaf;
3478 struct btrfs_key key, found_key;
3479 struct btrfs_trans_handle *trans;
3480 struct inode *inode;
3481 u64 last_objectid = 0;
3482 int ret = 0, nr_unlink = 0;
3484 if (test_and_set_bit(BTRFS_ROOT_ORPHAN_CLEANUP, &root->state))
3487 path = btrfs_alloc_path();
3492 path->reada = READA_BACK;
3494 key.objectid = BTRFS_ORPHAN_OBJECTID;
3495 key.type = BTRFS_ORPHAN_ITEM_KEY;
3496 key.offset = (u64)-1;
3499 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3504 * if ret == 0 means we found what we were searching for, which
3505 * is weird, but possible, so only screw with path if we didn't
3506 * find the key and see if we have stuff that matches
3510 if (path->slots[0] == 0)
3515 /* pull out the item */
3516 leaf = path->nodes[0];
3517 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3519 /* make sure the item matches what we want */
3520 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3522 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3525 /* release the path since we're done with it */
3526 btrfs_release_path(path);
3529 * this is where we are basically btrfs_lookup, without the
3530 * crossing root thing. we store the inode number in the
3531 * offset of the orphan item.
3534 if (found_key.offset == last_objectid) {
3536 "Error removing orphan entry, stopping orphan cleanup");
3541 last_objectid = found_key.offset;
3543 found_key.objectid = found_key.offset;
3544 found_key.type = BTRFS_INODE_ITEM_KEY;
3545 found_key.offset = 0;
3546 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3547 ret = PTR_ERR_OR_ZERO(inode);
3548 if (ret && ret != -ENOENT)
3551 if (ret == -ENOENT && root == fs_info->tree_root) {
3552 struct btrfs_root *dead_root;
3553 int is_dead_root = 0;
3556 * This is an orphan in the tree root. Currently these
3557 * could come from 2 sources:
3558 * a) a root (snapshot/subvolume) deletion in progress
3559 * b) a free space cache inode
3560 * We need to distinguish those two, as the orphan item
3561 * for a root must not get deleted before the deletion
3562 * of the snapshot/subvolume's tree completes.
3564 * btrfs_find_orphan_roots() ran before us, which has
3565 * found all deleted roots and loaded them into
3566 * fs_info->fs_roots_radix. So here we can find if an
3567 * orphan item corresponds to a deleted root by looking
3568 * up the root from that radix tree.
3571 spin_lock(&fs_info->fs_roots_radix_lock);
3572 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3573 (unsigned long)found_key.objectid);
3574 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3576 spin_unlock(&fs_info->fs_roots_radix_lock);
3579 /* prevent this orphan from being found again */
3580 key.offset = found_key.objectid - 1;
3587 * If we have an inode with links, there are a couple of
3590 * 1. We were halfway through creating fsverity metadata for the
3591 * file. In that case, the orphan item represents incomplete
3592 * fsverity metadata which must be cleaned up with
3593 * btrfs_drop_verity_items and deleting the orphan item.
3595 * 2. Old kernels (before v3.12) used to create an
3596 * orphan item for truncate indicating that there were possibly
3597 * extent items past i_size that needed to be deleted. In v3.12,
3598 * truncate was changed to update i_size in sync with the extent
3599 * items, but the (useless) orphan item was still created. Since
3600 * v4.18, we don't create the orphan item for truncate at all.
3602 * So, this item could mean that we need to do a truncate, but
3603 * only if this filesystem was last used on a pre-v3.12 kernel
3604 * and was not cleanly unmounted. The odds of that are quite
3605 * slim, and it's a pain to do the truncate now, so just delete
3608 * It's also possible that this orphan item was supposed to be
3609 * deleted but wasn't. The inode number may have been reused,
3610 * but either way, we can delete the orphan item.
3612 if (ret == -ENOENT || inode->i_nlink) {
3614 ret = btrfs_drop_verity_items(BTRFS_I(inode));
3619 trans = btrfs_start_transaction(root, 1);
3620 if (IS_ERR(trans)) {
3621 ret = PTR_ERR(trans);
3624 btrfs_debug(fs_info, "auto deleting %Lu",
3625 found_key.objectid);
3626 ret = btrfs_del_orphan_item(trans, root,
3627 found_key.objectid);
3628 btrfs_end_transaction(trans);
3636 /* this will do delete_inode and everything for us */
3639 /* release the path since we're done with it */
3640 btrfs_release_path(path);
3642 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3643 trans = btrfs_join_transaction(root);
3645 btrfs_end_transaction(trans);
3649 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3653 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3654 btrfs_free_path(path);
3659 * very simple check to peek ahead in the leaf looking for xattrs. If we
3660 * don't find any xattrs, we know there can't be any acls.
3662 * slot is the slot the inode is in, objectid is the objectid of the inode
3664 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3665 int slot, u64 objectid,
3666 int *first_xattr_slot)
3668 u32 nritems = btrfs_header_nritems(leaf);
3669 struct btrfs_key found_key;
3670 static u64 xattr_access = 0;
3671 static u64 xattr_default = 0;
3674 if (!xattr_access) {
3675 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3676 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3677 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3678 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3682 *first_xattr_slot = -1;
3683 while (slot < nritems) {
3684 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3686 /* we found a different objectid, there must not be acls */
3687 if (found_key.objectid != objectid)
3690 /* we found an xattr, assume we've got an acl */
3691 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3692 if (*first_xattr_slot == -1)
3693 *first_xattr_slot = slot;
3694 if (found_key.offset == xattr_access ||
3695 found_key.offset == xattr_default)
3700 * we found a key greater than an xattr key, there can't
3701 * be any acls later on
3703 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3710 * it goes inode, inode backrefs, xattrs, extents,
3711 * so if there are a ton of hard links to an inode there can
3712 * be a lot of backrefs. Don't waste time searching too hard,
3713 * this is just an optimization
3718 /* we hit the end of the leaf before we found an xattr or
3719 * something larger than an xattr. We have to assume the inode
3722 if (*first_xattr_slot == -1)
3723 *first_xattr_slot = slot;
3728 * read an inode from the btree into the in-memory inode
3730 static int btrfs_read_locked_inode(struct inode *inode,
3731 struct btrfs_path *in_path)
3733 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3734 struct btrfs_path *path = in_path;
3735 struct extent_buffer *leaf;
3736 struct btrfs_inode_item *inode_item;
3737 struct btrfs_root *root = BTRFS_I(inode)->root;
3738 struct btrfs_key location;
3743 bool filled = false;
3744 int first_xattr_slot;
3746 ret = btrfs_fill_inode(inode, &rdev);
3751 path = btrfs_alloc_path();
3756 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3758 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3760 if (path != in_path)
3761 btrfs_free_path(path);
3765 leaf = path->nodes[0];
3770 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3771 struct btrfs_inode_item);
3772 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3773 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3774 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3775 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3776 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3777 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3778 round_up(i_size_read(inode), fs_info->sectorsize));
3780 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3781 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3783 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3784 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3786 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3787 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3789 BTRFS_I(inode)->i_otime.tv_sec =
3790 btrfs_timespec_sec(leaf, &inode_item->otime);
3791 BTRFS_I(inode)->i_otime.tv_nsec =
3792 btrfs_timespec_nsec(leaf, &inode_item->otime);
3794 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3795 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3796 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3798 inode_set_iversion_queried(inode,
3799 btrfs_inode_sequence(leaf, inode_item));
3800 inode->i_generation = BTRFS_I(inode)->generation;
3802 rdev = btrfs_inode_rdev(leaf, inode_item);
3804 BTRFS_I(inode)->index_cnt = (u64)-1;
3805 btrfs_inode_split_flags(btrfs_inode_flags(leaf, inode_item),
3806 &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags);
3810 * If we were modified in the current generation and evicted from memory
3811 * and then re-read we need to do a full sync since we don't have any
3812 * idea about which extents were modified before we were evicted from
3815 * This is required for both inode re-read from disk and delayed inode
3816 * in delayed_nodes_tree.
3818 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3819 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3820 &BTRFS_I(inode)->runtime_flags);
3823 * We don't persist the id of the transaction where an unlink operation
3824 * against the inode was last made. So here we assume the inode might
3825 * have been evicted, and therefore the exact value of last_unlink_trans
3826 * lost, and set it to last_trans to avoid metadata inconsistencies
3827 * between the inode and its parent if the inode is fsync'ed and the log
3828 * replayed. For example, in the scenario:
3831 * ln mydir/foo mydir/bar
3834 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3835 * xfs_io -c fsync mydir/foo
3837 * mount fs, triggers fsync log replay
3839 * We must make sure that when we fsync our inode foo we also log its
3840 * parent inode, otherwise after log replay the parent still has the
3841 * dentry with the "bar" name but our inode foo has a link count of 1
3842 * and doesn't have an inode ref with the name "bar" anymore.
3844 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3845 * but it guarantees correctness at the expense of occasional full
3846 * transaction commits on fsync if our inode is a directory, or if our
3847 * inode is not a directory, logging its parent unnecessarily.
3849 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3852 * Same logic as for last_unlink_trans. We don't persist the generation
3853 * of the last transaction where this inode was used for a reflink
3854 * operation, so after eviction and reloading the inode we must be
3855 * pessimistic and assume the last transaction that modified the inode.
3857 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3860 if (inode->i_nlink != 1 ||
3861 path->slots[0] >= btrfs_header_nritems(leaf))
3864 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3865 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3868 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3869 if (location.type == BTRFS_INODE_REF_KEY) {
3870 struct btrfs_inode_ref *ref;
3872 ref = (struct btrfs_inode_ref *)ptr;
3873 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3874 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3875 struct btrfs_inode_extref *extref;
3877 extref = (struct btrfs_inode_extref *)ptr;
3878 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3883 * try to precache a NULL acl entry for files that don't have
3884 * any xattrs or acls
3886 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3887 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3888 if (first_xattr_slot != -1) {
3889 path->slots[0] = first_xattr_slot;
3890 ret = btrfs_load_inode_props(inode, path);
3893 "error loading props for ino %llu (root %llu): %d",
3894 btrfs_ino(BTRFS_I(inode)),
3895 root->root_key.objectid, ret);
3897 if (path != in_path)
3898 btrfs_free_path(path);
3901 cache_no_acl(inode);
3903 switch (inode->i_mode & S_IFMT) {
3905 inode->i_mapping->a_ops = &btrfs_aops;
3906 inode->i_fop = &btrfs_file_operations;
3907 inode->i_op = &btrfs_file_inode_operations;
3910 inode->i_fop = &btrfs_dir_file_operations;
3911 inode->i_op = &btrfs_dir_inode_operations;
3914 inode->i_op = &btrfs_symlink_inode_operations;
3915 inode_nohighmem(inode);
3916 inode->i_mapping->a_ops = &btrfs_aops;
3919 inode->i_op = &btrfs_special_inode_operations;
3920 init_special_inode(inode, inode->i_mode, rdev);
3924 btrfs_sync_inode_flags_to_i_flags(inode);
3929 * given a leaf and an inode, copy the inode fields into the leaf
3931 static void fill_inode_item(struct btrfs_trans_handle *trans,
3932 struct extent_buffer *leaf,
3933 struct btrfs_inode_item *item,
3934 struct inode *inode)
3936 struct btrfs_map_token token;
3939 btrfs_init_map_token(&token, leaf);
3941 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3942 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3943 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
3944 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3945 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3947 btrfs_set_token_timespec_sec(&token, &item->atime,
3948 inode->i_atime.tv_sec);
3949 btrfs_set_token_timespec_nsec(&token, &item->atime,
3950 inode->i_atime.tv_nsec);
3952 btrfs_set_token_timespec_sec(&token, &item->mtime,
3953 inode->i_mtime.tv_sec);
3954 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3955 inode->i_mtime.tv_nsec);
3957 btrfs_set_token_timespec_sec(&token, &item->ctime,
3958 inode->i_ctime.tv_sec);
3959 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3960 inode->i_ctime.tv_nsec);
3962 btrfs_set_token_timespec_sec(&token, &item->otime,
3963 BTRFS_I(inode)->i_otime.tv_sec);
3964 btrfs_set_token_timespec_nsec(&token, &item->otime,
3965 BTRFS_I(inode)->i_otime.tv_nsec);
3967 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
3968 btrfs_set_token_inode_generation(&token, item,
3969 BTRFS_I(inode)->generation);
3970 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
3971 btrfs_set_token_inode_transid(&token, item, trans->transid);
3972 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
3973 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
3974 BTRFS_I(inode)->ro_flags);
3975 btrfs_set_token_inode_flags(&token, item, flags);
3976 btrfs_set_token_inode_block_group(&token, item, 0);
3980 * copy everything in the in-memory inode into the btree.
3982 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3983 struct btrfs_root *root,
3984 struct btrfs_inode *inode)
3986 struct btrfs_inode_item *inode_item;
3987 struct btrfs_path *path;
3988 struct extent_buffer *leaf;
3991 path = btrfs_alloc_path();
3995 ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1);
4002 leaf = path->nodes[0];
4003 inode_item = btrfs_item_ptr(leaf, path->slots[0],
4004 struct btrfs_inode_item);
4006 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
4007 btrfs_mark_buffer_dirty(leaf);
4008 btrfs_set_inode_last_trans(trans, inode);
4011 btrfs_free_path(path);
4016 * copy everything in the in-memory inode into the btree.
4018 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
4019 struct btrfs_root *root,
4020 struct btrfs_inode *inode)
4022 struct btrfs_fs_info *fs_info = root->fs_info;
4026 * If the inode is a free space inode, we can deadlock during commit
4027 * if we put it into the delayed code.
4029 * The data relocation inode should also be directly updated
4032 if (!btrfs_is_free_space_inode(inode)
4033 && !btrfs_is_data_reloc_root(root)
4034 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4035 btrfs_update_root_times(trans, root);
4037 ret = btrfs_delayed_update_inode(trans, root, inode);
4039 btrfs_set_inode_last_trans(trans, inode);
4043 return btrfs_update_inode_item(trans, root, inode);
4046 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4047 struct btrfs_root *root, struct btrfs_inode *inode)
4051 ret = btrfs_update_inode(trans, root, inode);
4053 return btrfs_update_inode_item(trans, root, inode);
4058 * unlink helper that gets used here in inode.c and in the tree logging
4059 * recovery code. It remove a link in a directory with a given name, and
4060 * also drops the back refs in the inode to the directory
4062 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4063 struct btrfs_inode *dir,
4064 struct btrfs_inode *inode,
4065 const char *name, int name_len,
4066 struct btrfs_rename_ctx *rename_ctx)
4068 struct btrfs_root *root = dir->root;
4069 struct btrfs_fs_info *fs_info = root->fs_info;
4070 struct btrfs_path *path;
4072 struct btrfs_dir_item *di;
4074 u64 ino = btrfs_ino(inode);
4075 u64 dir_ino = btrfs_ino(dir);
4077 path = btrfs_alloc_path();
4083 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4084 name, name_len, -1);
4085 if (IS_ERR_OR_NULL(di)) {
4086 ret = di ? PTR_ERR(di) : -ENOENT;
4089 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4092 btrfs_release_path(path);
4095 * If we don't have dir index, we have to get it by looking up
4096 * the inode ref, since we get the inode ref, remove it directly,
4097 * it is unnecessary to do delayed deletion.
4099 * But if we have dir index, needn't search inode ref to get it.
4100 * Since the inode ref is close to the inode item, it is better
4101 * that we delay to delete it, and just do this deletion when
4102 * we update the inode item.
4104 if (inode->dir_index) {
4105 ret = btrfs_delayed_delete_inode_ref(inode);
4107 index = inode->dir_index;
4112 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4116 "failed to delete reference to %.*s, inode %llu parent %llu",
4117 name_len, name, ino, dir_ino);
4118 btrfs_abort_transaction(trans, ret);
4123 rename_ctx->index = index;
4125 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4127 btrfs_abort_transaction(trans, ret);
4132 * If we are in a rename context, we don't need to update anything in the
4133 * log. That will be done later during the rename by btrfs_log_new_name().
4134 * Besides that, doing it here would only cause extra unncessary btree
4135 * operations on the log tree, increasing latency for applications.
4138 btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
4140 btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
4145 * If we have a pending delayed iput we could end up with the final iput
4146 * being run in btrfs-cleaner context. If we have enough of these built
4147 * up we can end up burning a lot of time in btrfs-cleaner without any
4148 * way to throttle the unlinks. Since we're currently holding a ref on
4149 * the inode we can run the delayed iput here without any issues as the
4150 * final iput won't be done until after we drop the ref we're currently
4153 btrfs_run_delayed_iput(fs_info, inode);
4155 btrfs_free_path(path);
4159 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
4160 inode_inc_iversion(&inode->vfs_inode);
4161 inode_inc_iversion(&dir->vfs_inode);
4162 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
4163 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4164 ret = btrfs_update_inode(trans, root, dir);
4169 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4170 struct btrfs_inode *dir, struct btrfs_inode *inode,
4171 const char *name, int name_len)
4174 ret = __btrfs_unlink_inode(trans, dir, inode, name, name_len, NULL);
4176 drop_nlink(&inode->vfs_inode);
4177 ret = btrfs_update_inode(trans, inode->root, inode);
4183 * helper to start transaction for unlink and rmdir.
4185 * unlink and rmdir are special in btrfs, they do not always free space, so
4186 * if we cannot make our reservations the normal way try and see if there is
4187 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4188 * allow the unlink to occur.
4190 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4192 struct btrfs_root *root = BTRFS_I(dir)->root;
4195 * 1 for the possible orphan item
4196 * 1 for the dir item
4197 * 1 for the dir index
4198 * 1 for the inode ref
4201 return btrfs_start_transaction_fallback_global_rsv(root, 5);
4204 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4206 struct btrfs_trans_handle *trans;
4207 struct inode *inode = d_inode(dentry);
4210 trans = __unlink_start_trans(dir);
4212 return PTR_ERR(trans);
4214 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4217 ret = btrfs_unlink_inode(trans, BTRFS_I(dir),
4218 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4219 dentry->d_name.len);
4223 if (inode->i_nlink == 0) {
4224 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4230 btrfs_end_transaction(trans);
4231 btrfs_btree_balance_dirty(BTRFS_I(dir)->root->fs_info);
4235 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4236 struct inode *dir, struct dentry *dentry)
4238 struct btrfs_root *root = BTRFS_I(dir)->root;
4239 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4240 struct btrfs_path *path;
4241 struct extent_buffer *leaf;
4242 struct btrfs_dir_item *di;
4243 struct btrfs_key key;
4244 const char *name = dentry->d_name.name;
4245 int name_len = dentry->d_name.len;
4249 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4251 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4252 objectid = inode->root->root_key.objectid;
4253 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4254 objectid = inode->location.objectid;
4260 path = btrfs_alloc_path();
4264 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4265 name, name_len, -1);
4266 if (IS_ERR_OR_NULL(di)) {
4267 ret = di ? PTR_ERR(di) : -ENOENT;
4271 leaf = path->nodes[0];
4272 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4273 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4274 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4276 btrfs_abort_transaction(trans, ret);
4279 btrfs_release_path(path);
4282 * This is a placeholder inode for a subvolume we didn't have a
4283 * reference to at the time of the snapshot creation. In the meantime
4284 * we could have renamed the real subvol link into our snapshot, so
4285 * depending on btrfs_del_root_ref to return -ENOENT here is incorrect.
4286 * Instead simply lookup the dir_index_item for this entry so we can
4287 * remove it. Otherwise we know we have a ref to the root and we can
4288 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4290 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4291 di = btrfs_search_dir_index_item(root, path, dir_ino,
4293 if (IS_ERR_OR_NULL(di)) {
4298 btrfs_abort_transaction(trans, ret);
4302 leaf = path->nodes[0];
4303 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4305 btrfs_release_path(path);
4307 ret = btrfs_del_root_ref(trans, objectid,
4308 root->root_key.objectid, dir_ino,
4309 &index, name, name_len);
4311 btrfs_abort_transaction(trans, ret);
4316 ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index);
4318 btrfs_abort_transaction(trans, ret);
4322 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4323 inode_inc_iversion(dir);
4324 dir->i_mtime = dir->i_ctime = current_time(dir);
4325 ret = btrfs_update_inode_fallback(trans, root, BTRFS_I(dir));
4327 btrfs_abort_transaction(trans, ret);
4329 btrfs_free_path(path);
4334 * Helper to check if the subvolume references other subvolumes or if it's
4337 static noinline int may_destroy_subvol(struct btrfs_root *root)
4339 struct btrfs_fs_info *fs_info = root->fs_info;
4340 struct btrfs_path *path;
4341 struct btrfs_dir_item *di;
4342 struct btrfs_key key;
4346 path = btrfs_alloc_path();
4350 /* Make sure this root isn't set as the default subvol */
4351 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4352 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4353 dir_id, "default", 7, 0);
4354 if (di && !IS_ERR(di)) {
4355 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4356 if (key.objectid == root->root_key.objectid) {
4359 "deleting default subvolume %llu is not allowed",
4363 btrfs_release_path(path);
4366 key.objectid = root->root_key.objectid;
4367 key.type = BTRFS_ROOT_REF_KEY;
4368 key.offset = (u64)-1;
4370 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4376 if (path->slots[0] > 0) {
4378 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4379 if (key.objectid == root->root_key.objectid &&
4380 key.type == BTRFS_ROOT_REF_KEY)
4384 btrfs_free_path(path);
4388 /* Delete all dentries for inodes belonging to the root */
4389 static void btrfs_prune_dentries(struct btrfs_root *root)
4391 struct btrfs_fs_info *fs_info = root->fs_info;
4392 struct rb_node *node;
4393 struct rb_node *prev;
4394 struct btrfs_inode *entry;
4395 struct inode *inode;
4398 if (!BTRFS_FS_ERROR(fs_info))
4399 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4401 spin_lock(&root->inode_lock);
4403 node = root->inode_tree.rb_node;
4407 entry = rb_entry(node, struct btrfs_inode, rb_node);
4409 if (objectid < btrfs_ino(entry))
4410 node = node->rb_left;
4411 else if (objectid > btrfs_ino(entry))
4412 node = node->rb_right;
4418 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4419 if (objectid <= btrfs_ino(entry)) {
4423 prev = rb_next(prev);
4427 entry = rb_entry(node, struct btrfs_inode, rb_node);
4428 objectid = btrfs_ino(entry) + 1;
4429 inode = igrab(&entry->vfs_inode);
4431 spin_unlock(&root->inode_lock);
4432 if (atomic_read(&inode->i_count) > 1)
4433 d_prune_aliases(inode);
4435 * btrfs_drop_inode will have it removed from the inode
4436 * cache when its usage count hits zero.
4440 spin_lock(&root->inode_lock);
4444 if (cond_resched_lock(&root->inode_lock))
4447 node = rb_next(node);
4449 spin_unlock(&root->inode_lock);
4452 int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
4454 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4455 struct btrfs_root *root = BTRFS_I(dir)->root;
4456 struct inode *inode = d_inode(dentry);
4457 struct btrfs_root *dest = BTRFS_I(inode)->root;
4458 struct btrfs_trans_handle *trans;
4459 struct btrfs_block_rsv block_rsv;
4464 * Don't allow to delete a subvolume with send in progress. This is
4465 * inside the inode lock so the error handling that has to drop the bit
4466 * again is not run concurrently.
4468 spin_lock(&dest->root_item_lock);
4469 if (dest->send_in_progress) {
4470 spin_unlock(&dest->root_item_lock);
4472 "attempt to delete subvolume %llu during send",
4473 dest->root_key.objectid);
4476 root_flags = btrfs_root_flags(&dest->root_item);
4477 btrfs_set_root_flags(&dest->root_item,
4478 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4479 spin_unlock(&dest->root_item_lock);
4481 down_write(&fs_info->subvol_sem);
4483 ret = may_destroy_subvol(dest);
4487 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4489 * One for dir inode,
4490 * two for dir entries,
4491 * two for root ref/backref.
4493 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4497 trans = btrfs_start_transaction(root, 0);
4498 if (IS_ERR(trans)) {
4499 ret = PTR_ERR(trans);
4502 trans->block_rsv = &block_rsv;
4503 trans->bytes_reserved = block_rsv.size;
4505 btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
4507 ret = btrfs_unlink_subvol(trans, dir, dentry);
4509 btrfs_abort_transaction(trans, ret);
4513 ret = btrfs_record_root_in_trans(trans, dest);
4515 btrfs_abort_transaction(trans, ret);
4519 memset(&dest->root_item.drop_progress, 0,
4520 sizeof(dest->root_item.drop_progress));
4521 btrfs_set_root_drop_level(&dest->root_item, 0);
4522 btrfs_set_root_refs(&dest->root_item, 0);
4524 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4525 ret = btrfs_insert_orphan_item(trans,
4527 dest->root_key.objectid);
4529 btrfs_abort_transaction(trans, ret);
4534 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4535 BTRFS_UUID_KEY_SUBVOL,
4536 dest->root_key.objectid);
4537 if (ret && ret != -ENOENT) {
4538 btrfs_abort_transaction(trans, ret);
4541 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4542 ret = btrfs_uuid_tree_remove(trans,
4543 dest->root_item.received_uuid,
4544 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4545 dest->root_key.objectid);
4546 if (ret && ret != -ENOENT) {
4547 btrfs_abort_transaction(trans, ret);
4552 free_anon_bdev(dest->anon_dev);
4555 trans->block_rsv = NULL;
4556 trans->bytes_reserved = 0;
4557 ret = btrfs_end_transaction(trans);
4558 inode->i_flags |= S_DEAD;
4560 btrfs_subvolume_release_metadata(root, &block_rsv);
4562 up_write(&fs_info->subvol_sem);
4564 spin_lock(&dest->root_item_lock);
4565 root_flags = btrfs_root_flags(&dest->root_item);
4566 btrfs_set_root_flags(&dest->root_item,
4567 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4568 spin_unlock(&dest->root_item_lock);
4570 d_invalidate(dentry);
4571 btrfs_prune_dentries(dest);
4572 ASSERT(dest->send_in_progress == 0);
4578 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4580 struct inode *inode = d_inode(dentry);
4581 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
4583 struct btrfs_trans_handle *trans;
4584 u64 last_unlink_trans;
4586 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4588 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) {
4589 if (unlikely(btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))) {
4591 "extent tree v2 doesn't support snapshot deletion yet");
4594 return btrfs_delete_subvolume(dir, dentry);
4597 trans = __unlink_start_trans(dir);
4599 return PTR_ERR(trans);
4601 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4602 err = btrfs_unlink_subvol(trans, dir, dentry);
4606 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4610 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4612 /* now the directory is empty */
4613 err = btrfs_unlink_inode(trans, BTRFS_I(dir),
4614 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4615 dentry->d_name.len);
4617 btrfs_i_size_write(BTRFS_I(inode), 0);
4619 * Propagate the last_unlink_trans value of the deleted dir to
4620 * its parent directory. This is to prevent an unrecoverable
4621 * log tree in the case we do something like this:
4623 * 2) create snapshot under dir foo
4624 * 3) delete the snapshot
4627 * 6) fsync foo or some file inside foo
4629 if (last_unlink_trans >= trans->transid)
4630 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4633 btrfs_end_transaction(trans);
4634 btrfs_btree_balance_dirty(fs_info);
4640 * btrfs_truncate_block - read, zero a chunk and write a block
4641 * @inode - inode that we're zeroing
4642 * @from - the offset to start zeroing
4643 * @len - the length to zero, 0 to zero the entire range respective to the
4645 * @front - zero up to the offset instead of from the offset on
4647 * This will find the block for the "from" offset and cow the block and zero the
4648 * part we want to zero. This is used with truncate and hole punching.
4650 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4653 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4654 struct address_space *mapping = inode->vfs_inode.i_mapping;
4655 struct extent_io_tree *io_tree = &inode->io_tree;
4656 struct btrfs_ordered_extent *ordered;
4657 struct extent_state *cached_state = NULL;
4658 struct extent_changeset *data_reserved = NULL;
4659 bool only_release_metadata = false;
4660 u32 blocksize = fs_info->sectorsize;
4661 pgoff_t index = from >> PAGE_SHIFT;
4662 unsigned offset = from & (blocksize - 1);
4664 gfp_t mask = btrfs_alloc_write_mask(mapping);
4665 size_t write_bytes = blocksize;
4670 if (IS_ALIGNED(offset, blocksize) &&
4671 (!len || IS_ALIGNED(len, blocksize)))
4674 block_start = round_down(from, blocksize);
4675 block_end = block_start + blocksize - 1;
4677 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4680 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes) > 0) {
4681 /* For nocow case, no need to reserve data space */
4682 only_release_metadata = true;
4687 ret = btrfs_delalloc_reserve_metadata(inode, blocksize, blocksize);
4689 if (!only_release_metadata)
4690 btrfs_free_reserved_data_space(inode, data_reserved,
4691 block_start, blocksize);
4695 page = find_or_create_page(mapping, index, mask);
4697 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4699 btrfs_delalloc_release_extents(inode, blocksize);
4703 ret = set_page_extent_mapped(page);
4707 if (!PageUptodate(page)) {
4708 ret = btrfs_readpage(NULL, page);
4710 if (page->mapping != mapping) {
4715 if (!PageUptodate(page)) {
4720 wait_on_page_writeback(page);
4722 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4724 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4726 unlock_extent_cached(io_tree, block_start, block_end,
4730 btrfs_start_ordered_extent(ordered, 1);
4731 btrfs_put_ordered_extent(ordered);
4735 clear_extent_bit(&inode->io_tree, block_start, block_end,
4736 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4737 0, 0, &cached_state);
4739 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4742 unlock_extent_cached(io_tree, block_start, block_end,
4747 if (offset != blocksize) {
4749 len = blocksize - offset;
4751 memzero_page(page, (block_start - page_offset(page)),
4754 memzero_page(page, (block_start - page_offset(page)) + offset,
4756 flush_dcache_page(page);
4758 btrfs_page_clear_checked(fs_info, page, block_start,
4759 block_end + 1 - block_start);
4760 btrfs_page_set_dirty(fs_info, page, block_start, block_end + 1 - block_start);
4761 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4763 if (only_release_metadata)
4764 set_extent_bit(&inode->io_tree, block_start, block_end,
4765 EXTENT_NORESERVE, 0, NULL, NULL, GFP_NOFS, NULL);
4769 if (only_release_metadata)
4770 btrfs_delalloc_release_metadata(inode, blocksize, true);
4772 btrfs_delalloc_release_space(inode, data_reserved,
4773 block_start, blocksize, true);
4775 btrfs_delalloc_release_extents(inode, blocksize);
4779 if (only_release_metadata)
4780 btrfs_check_nocow_unlock(inode);
4781 extent_changeset_free(data_reserved);
4785 static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode,
4786 u64 offset, u64 len)
4788 struct btrfs_fs_info *fs_info = root->fs_info;
4789 struct btrfs_trans_handle *trans;
4790 struct btrfs_drop_extents_args drop_args = { 0 };
4794 * If NO_HOLES is enabled, we don't need to do anything.
4795 * Later, up in the call chain, either btrfs_set_inode_last_sub_trans()
4796 * or btrfs_update_inode() will be called, which guarantee that the next
4797 * fsync will know this inode was changed and needs to be logged.
4799 if (btrfs_fs_incompat(fs_info, NO_HOLES))
4803 * 1 - for the one we're dropping
4804 * 1 - for the one we're adding
4805 * 1 - for updating the inode.
4807 trans = btrfs_start_transaction(root, 3);
4809 return PTR_ERR(trans);
4811 drop_args.start = offset;
4812 drop_args.end = offset + len;
4813 drop_args.drop_cache = true;
4815 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
4817 btrfs_abort_transaction(trans, ret);
4818 btrfs_end_transaction(trans);
4822 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode),
4823 offset, 0, 0, len, 0, len, 0, 0, 0);
4825 btrfs_abort_transaction(trans, ret);
4827 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
4828 btrfs_update_inode(trans, root, inode);
4830 btrfs_end_transaction(trans);
4835 * This function puts in dummy file extents for the area we're creating a hole
4836 * for. So if we are truncating this file to a larger size we need to insert
4837 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4838 * the range between oldsize and size
4840 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
4842 struct btrfs_root *root = inode->root;
4843 struct btrfs_fs_info *fs_info = root->fs_info;
4844 struct extent_io_tree *io_tree = &inode->io_tree;
4845 struct extent_map *em = NULL;
4846 struct extent_state *cached_state = NULL;
4847 struct extent_map_tree *em_tree = &inode->extent_tree;
4848 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4849 u64 block_end = ALIGN(size, fs_info->sectorsize);
4856 * If our size started in the middle of a block we need to zero out the
4857 * rest of the block before we expand the i_size, otherwise we could
4858 * expose stale data.
4860 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4864 if (size <= hole_start)
4867 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
4869 cur_offset = hole_start;
4871 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4872 block_end - cur_offset);
4878 last_byte = min(extent_map_end(em), block_end);
4879 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4880 hole_size = last_byte - cur_offset;
4882 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4883 struct extent_map *hole_em;
4885 err = maybe_insert_hole(root, inode, cur_offset,
4890 err = btrfs_inode_set_file_extent_range(inode,
4891 cur_offset, hole_size);
4895 btrfs_drop_extent_cache(inode, cur_offset,
4896 cur_offset + hole_size - 1, 0);
4897 hole_em = alloc_extent_map();
4899 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4900 &inode->runtime_flags);
4903 hole_em->start = cur_offset;
4904 hole_em->len = hole_size;
4905 hole_em->orig_start = cur_offset;
4907 hole_em->block_start = EXTENT_MAP_HOLE;
4908 hole_em->block_len = 0;
4909 hole_em->orig_block_len = 0;
4910 hole_em->ram_bytes = hole_size;
4911 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4912 hole_em->generation = fs_info->generation;
4915 write_lock(&em_tree->lock);
4916 err = add_extent_mapping(em_tree, hole_em, 1);
4917 write_unlock(&em_tree->lock);
4920 btrfs_drop_extent_cache(inode, cur_offset,
4924 free_extent_map(hole_em);
4926 err = btrfs_inode_set_file_extent_range(inode,
4927 cur_offset, hole_size);
4932 free_extent_map(em);
4934 cur_offset = last_byte;
4935 if (cur_offset >= block_end)
4938 free_extent_map(em);
4939 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
4943 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4945 struct btrfs_root *root = BTRFS_I(inode)->root;
4946 struct btrfs_trans_handle *trans;
4947 loff_t oldsize = i_size_read(inode);
4948 loff_t newsize = attr->ia_size;
4949 int mask = attr->ia_valid;
4953 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4954 * special case where we need to update the times despite not having
4955 * these flags set. For all other operations the VFS set these flags
4956 * explicitly if it wants a timestamp update.
4958 if (newsize != oldsize) {
4959 inode_inc_iversion(inode);
4960 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4961 inode->i_ctime = inode->i_mtime =
4962 current_time(inode);
4965 if (newsize > oldsize) {
4967 * Don't do an expanding truncate while snapshotting is ongoing.
4968 * This is to ensure the snapshot captures a fully consistent
4969 * state of this file - if the snapshot captures this expanding
4970 * truncation, it must capture all writes that happened before
4973 btrfs_drew_write_lock(&root->snapshot_lock);
4974 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
4976 btrfs_drew_write_unlock(&root->snapshot_lock);
4980 trans = btrfs_start_transaction(root, 1);
4981 if (IS_ERR(trans)) {
4982 btrfs_drew_write_unlock(&root->snapshot_lock);
4983 return PTR_ERR(trans);
4986 i_size_write(inode, newsize);
4987 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
4988 pagecache_isize_extended(inode, oldsize, newsize);
4989 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
4990 btrfs_drew_write_unlock(&root->snapshot_lock);
4991 btrfs_end_transaction(trans);
4993 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4995 if (btrfs_is_zoned(fs_info)) {
4996 ret = btrfs_wait_ordered_range(inode,
4997 ALIGN(newsize, fs_info->sectorsize),
5004 * We're truncating a file that used to have good data down to
5005 * zero. Make sure any new writes to the file get on disk
5009 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5010 &BTRFS_I(inode)->runtime_flags);
5012 truncate_setsize(inode, newsize);
5014 inode_dio_wait(inode);
5016 ret = btrfs_truncate(inode, newsize == oldsize);
5017 if (ret && inode->i_nlink) {
5021 * Truncate failed, so fix up the in-memory size. We
5022 * adjusted disk_i_size down as we removed extents, so
5023 * wait for disk_i_size to be stable and then update the
5024 * in-memory size to match.
5026 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5029 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5036 static int btrfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
5039 struct inode *inode = d_inode(dentry);
5040 struct btrfs_root *root = BTRFS_I(inode)->root;
5043 if (btrfs_root_readonly(root))
5046 err = setattr_prepare(mnt_userns, dentry, attr);
5050 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5051 err = btrfs_setsize(inode, attr);
5056 if (attr->ia_valid) {
5057 setattr_copy(mnt_userns, inode, attr);
5058 inode_inc_iversion(inode);
5059 err = btrfs_dirty_inode(inode);
5061 if (!err && attr->ia_valid & ATTR_MODE)
5062 err = posix_acl_chmod(mnt_userns, inode, inode->i_mode);
5069 * While truncating the inode pages during eviction, we get the VFS calling
5070 * btrfs_invalidatepage() against each page of the inode. This is slow because
5071 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5072 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5073 * extent_state structures over and over, wasting lots of time.
5075 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5076 * those expensive operations on a per page basis and do only the ordered io
5077 * finishing, while we release here the extent_map and extent_state structures,
5078 * without the excessive merging and splitting.
5080 static void evict_inode_truncate_pages(struct inode *inode)
5082 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5083 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5084 struct rb_node *node;
5086 ASSERT(inode->i_state & I_FREEING);
5087 truncate_inode_pages_final(&inode->i_data);
5089 write_lock(&map_tree->lock);
5090 while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) {
5091 struct extent_map *em;
5093 node = rb_first_cached(&map_tree->map);
5094 em = rb_entry(node, struct extent_map, rb_node);
5095 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5096 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5097 remove_extent_mapping(map_tree, em);
5098 free_extent_map(em);
5099 if (need_resched()) {
5100 write_unlock(&map_tree->lock);
5102 write_lock(&map_tree->lock);
5105 write_unlock(&map_tree->lock);
5108 * Keep looping until we have no more ranges in the io tree.
5109 * We can have ongoing bios started by readahead that have
5110 * their endio callback (extent_io.c:end_bio_extent_readpage)
5111 * still in progress (unlocked the pages in the bio but did not yet
5112 * unlocked the ranges in the io tree). Therefore this means some
5113 * ranges can still be locked and eviction started because before
5114 * submitting those bios, which are executed by a separate task (work
5115 * queue kthread), inode references (inode->i_count) were not taken
5116 * (which would be dropped in the end io callback of each bio).
5117 * Therefore here we effectively end up waiting for those bios and
5118 * anyone else holding locked ranges without having bumped the inode's
5119 * reference count - if we don't do it, when they access the inode's
5120 * io_tree to unlock a range it may be too late, leading to an
5121 * use-after-free issue.
5123 spin_lock(&io_tree->lock);
5124 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5125 struct extent_state *state;
5126 struct extent_state *cached_state = NULL;
5129 unsigned state_flags;
5131 node = rb_first(&io_tree->state);
5132 state = rb_entry(node, struct extent_state, rb_node);
5133 start = state->start;
5135 state_flags = state->state;
5136 spin_unlock(&io_tree->lock);
5138 lock_extent_bits(io_tree, start, end, &cached_state);
5141 * If still has DELALLOC flag, the extent didn't reach disk,
5142 * and its reserved space won't be freed by delayed_ref.
5143 * So we need to free its reserved space here.
5144 * (Refer to comment in btrfs_invalidatepage, case 2)
5146 * Note, end is the bytenr of last byte, so we need + 1 here.
5148 if (state_flags & EXTENT_DELALLOC)
5149 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5152 clear_extent_bit(io_tree, start, end,
5153 EXTENT_LOCKED | EXTENT_DELALLOC |
5154 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
5158 spin_lock(&io_tree->lock);
5160 spin_unlock(&io_tree->lock);
5163 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5164 struct btrfs_block_rsv *rsv)
5166 struct btrfs_fs_info *fs_info = root->fs_info;
5167 struct btrfs_trans_handle *trans;
5168 u64 delayed_refs_extra = btrfs_calc_insert_metadata_size(fs_info, 1);
5172 * Eviction should be taking place at some place safe because of our
5173 * delayed iputs. However the normal flushing code will run delayed
5174 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5176 * We reserve the delayed_refs_extra here again because we can't use
5177 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5178 * above. We reserve our extra bit here because we generate a ton of
5179 * delayed refs activity by truncating.
5181 * BTRFS_RESERVE_FLUSH_EVICT will steal from the global_rsv if it can,
5182 * if we fail to make this reservation we can re-try without the
5183 * delayed_refs_extra so we can make some forward progress.
5185 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size + delayed_refs_extra,
5186 BTRFS_RESERVE_FLUSH_EVICT);
5188 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size,
5189 BTRFS_RESERVE_FLUSH_EVICT);
5192 "could not allocate space for delete; will truncate on mount");
5193 return ERR_PTR(-ENOSPC);
5195 delayed_refs_extra = 0;
5198 trans = btrfs_join_transaction(root);
5202 if (delayed_refs_extra) {
5203 trans->block_rsv = &fs_info->trans_block_rsv;
5204 trans->bytes_reserved = delayed_refs_extra;
5205 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5206 delayed_refs_extra, 1);
5211 void btrfs_evict_inode(struct inode *inode)
5213 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5214 struct btrfs_trans_handle *trans;
5215 struct btrfs_root *root = BTRFS_I(inode)->root;
5216 struct btrfs_block_rsv *rsv;
5219 trace_btrfs_inode_evict(inode);
5222 fsverity_cleanup_inode(inode);
5227 evict_inode_truncate_pages(inode);
5229 if (inode->i_nlink &&
5230 ((btrfs_root_refs(&root->root_item) != 0 &&
5231 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5232 btrfs_is_free_space_inode(BTRFS_I(inode))))
5235 if (is_bad_inode(inode))
5238 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5240 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5243 if (inode->i_nlink > 0) {
5244 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5245 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5250 * This makes sure the inode item in tree is uptodate and the space for
5251 * the inode update is released.
5253 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5258 * This drops any pending insert or delete operations we have for this
5259 * inode. We could have a delayed dir index deletion queued up, but
5260 * we're removing the inode completely so that'll be taken care of in
5263 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
5265 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5268 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5271 btrfs_i_size_write(BTRFS_I(inode), 0);
5274 struct btrfs_truncate_control control = {
5275 .inode = BTRFS_I(inode),
5276 .ino = btrfs_ino(BTRFS_I(inode)),
5281 trans = evict_refill_and_join(root, rsv);
5285 trans->block_rsv = rsv;
5287 ret = btrfs_truncate_inode_items(trans, root, &control);
5288 trans->block_rsv = &fs_info->trans_block_rsv;
5289 btrfs_end_transaction(trans);
5290 btrfs_btree_balance_dirty(fs_info);
5291 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5298 * Errors here aren't a big deal, it just means we leave orphan items in
5299 * the tree. They will be cleaned up on the next mount. If the inode
5300 * number gets reused, cleanup deletes the orphan item without doing
5301 * anything, and unlink reuses the existing orphan item.
5303 * If it turns out that we are dropping too many of these, we might want
5304 * to add a mechanism for retrying these after a commit.
5306 trans = evict_refill_and_join(root, rsv);
5307 if (!IS_ERR(trans)) {
5308 trans->block_rsv = rsv;
5309 btrfs_orphan_del(trans, BTRFS_I(inode));
5310 trans->block_rsv = &fs_info->trans_block_rsv;
5311 btrfs_end_transaction(trans);
5315 btrfs_free_block_rsv(fs_info, rsv);
5318 * If we didn't successfully delete, the orphan item will still be in
5319 * the tree and we'll retry on the next mount. Again, we might also want
5320 * to retry these periodically in the future.
5322 btrfs_remove_delayed_node(BTRFS_I(inode));
5323 fsverity_cleanup_inode(inode);
5328 * Return the key found in the dir entry in the location pointer, fill @type
5329 * with BTRFS_FT_*, and return 0.
5331 * If no dir entries were found, returns -ENOENT.
5332 * If found a corrupted location in dir entry, returns -EUCLEAN.
5334 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5335 struct btrfs_key *location, u8 *type)
5337 const char *name = dentry->d_name.name;
5338 int namelen = dentry->d_name.len;
5339 struct btrfs_dir_item *di;
5340 struct btrfs_path *path;
5341 struct btrfs_root *root = BTRFS_I(dir)->root;
5344 path = btrfs_alloc_path();
5348 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5350 if (IS_ERR_OR_NULL(di)) {
5351 ret = di ? PTR_ERR(di) : -ENOENT;
5355 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5356 if (location->type != BTRFS_INODE_ITEM_KEY &&
5357 location->type != BTRFS_ROOT_ITEM_KEY) {
5359 btrfs_warn(root->fs_info,
5360 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5361 __func__, name, btrfs_ino(BTRFS_I(dir)),
5362 location->objectid, location->type, location->offset);
5365 *type = btrfs_dir_type(path->nodes[0], di);
5367 btrfs_free_path(path);
5372 * when we hit a tree root in a directory, the btrfs part of the inode
5373 * needs to be changed to reflect the root directory of the tree root. This
5374 * is kind of like crossing a mount point.
5376 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5378 struct dentry *dentry,
5379 struct btrfs_key *location,
5380 struct btrfs_root **sub_root)
5382 struct btrfs_path *path;
5383 struct btrfs_root *new_root;
5384 struct btrfs_root_ref *ref;
5385 struct extent_buffer *leaf;
5386 struct btrfs_key key;
5390 path = btrfs_alloc_path();
5397 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5398 key.type = BTRFS_ROOT_REF_KEY;
5399 key.offset = location->objectid;
5401 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5408 leaf = path->nodes[0];
5409 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5410 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5411 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5414 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5415 (unsigned long)(ref + 1),
5416 dentry->d_name.len);
5420 btrfs_release_path(path);
5422 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5423 if (IS_ERR(new_root)) {
5424 err = PTR_ERR(new_root);
5428 *sub_root = new_root;
5429 location->objectid = btrfs_root_dirid(&new_root->root_item);
5430 location->type = BTRFS_INODE_ITEM_KEY;
5431 location->offset = 0;
5434 btrfs_free_path(path);
5438 static void inode_tree_add(struct inode *inode)
5440 struct btrfs_root *root = BTRFS_I(inode)->root;
5441 struct btrfs_inode *entry;
5443 struct rb_node *parent;
5444 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5445 u64 ino = btrfs_ino(BTRFS_I(inode));
5447 if (inode_unhashed(inode))
5450 spin_lock(&root->inode_lock);
5451 p = &root->inode_tree.rb_node;
5454 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5456 if (ino < btrfs_ino(entry))
5457 p = &parent->rb_left;
5458 else if (ino > btrfs_ino(entry))
5459 p = &parent->rb_right;
5461 WARN_ON(!(entry->vfs_inode.i_state &
5462 (I_WILL_FREE | I_FREEING)));
5463 rb_replace_node(parent, new, &root->inode_tree);
5464 RB_CLEAR_NODE(parent);
5465 spin_unlock(&root->inode_lock);
5469 rb_link_node(new, parent, p);
5470 rb_insert_color(new, &root->inode_tree);
5471 spin_unlock(&root->inode_lock);
5474 static void inode_tree_del(struct btrfs_inode *inode)
5476 struct btrfs_root *root = inode->root;
5479 spin_lock(&root->inode_lock);
5480 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5481 rb_erase(&inode->rb_node, &root->inode_tree);
5482 RB_CLEAR_NODE(&inode->rb_node);
5483 empty = RB_EMPTY_ROOT(&root->inode_tree);
5485 spin_unlock(&root->inode_lock);
5487 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5488 spin_lock(&root->inode_lock);
5489 empty = RB_EMPTY_ROOT(&root->inode_tree);
5490 spin_unlock(&root->inode_lock);
5492 btrfs_add_dead_root(root);
5497 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5499 struct btrfs_iget_args *args = p;
5501 inode->i_ino = args->ino;
5502 BTRFS_I(inode)->location.objectid = args->ino;
5503 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5504 BTRFS_I(inode)->location.offset = 0;
5505 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5506 BUG_ON(args->root && !BTRFS_I(inode)->root);
5510 static int btrfs_find_actor(struct inode *inode, void *opaque)
5512 struct btrfs_iget_args *args = opaque;
5514 return args->ino == BTRFS_I(inode)->location.objectid &&
5515 args->root == BTRFS_I(inode)->root;
5518 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5519 struct btrfs_root *root)
5521 struct inode *inode;
5522 struct btrfs_iget_args args;
5523 unsigned long hashval = btrfs_inode_hash(ino, root);
5528 inode = iget5_locked(s, hashval, btrfs_find_actor,
5529 btrfs_init_locked_inode,
5535 * Get an inode object given its inode number and corresponding root.
5536 * Path can be preallocated to prevent recursing back to iget through
5537 * allocator. NULL is also valid but may require an additional allocation
5540 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5541 struct btrfs_root *root, struct btrfs_path *path)
5543 struct inode *inode;
5545 inode = btrfs_iget_locked(s, ino, root);
5547 return ERR_PTR(-ENOMEM);
5549 if (inode->i_state & I_NEW) {
5552 ret = btrfs_read_locked_inode(inode, path);
5554 inode_tree_add(inode);
5555 unlock_new_inode(inode);
5559 * ret > 0 can come from btrfs_search_slot called by
5560 * btrfs_read_locked_inode, this means the inode item
5565 inode = ERR_PTR(ret);
5572 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5574 return btrfs_iget_path(s, ino, root, NULL);
5577 static struct inode *new_simple_dir(struct super_block *s,
5578 struct btrfs_key *key,
5579 struct btrfs_root *root)
5581 struct inode *inode = new_inode(s);
5584 return ERR_PTR(-ENOMEM);
5586 BTRFS_I(inode)->root = btrfs_grab_root(root);
5587 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5588 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5590 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5592 * We only need lookup, the rest is read-only and there's no inode
5593 * associated with the dentry
5595 inode->i_op = &simple_dir_inode_operations;
5596 inode->i_opflags &= ~IOP_XATTR;
5597 inode->i_fop = &simple_dir_operations;
5598 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5599 inode->i_mtime = current_time(inode);
5600 inode->i_atime = inode->i_mtime;
5601 inode->i_ctime = inode->i_mtime;
5602 BTRFS_I(inode)->i_otime = inode->i_mtime;
5607 static_assert(BTRFS_FT_UNKNOWN == FT_UNKNOWN);
5608 static_assert(BTRFS_FT_REG_FILE == FT_REG_FILE);
5609 static_assert(BTRFS_FT_DIR == FT_DIR);
5610 static_assert(BTRFS_FT_CHRDEV == FT_CHRDEV);
5611 static_assert(BTRFS_FT_BLKDEV == FT_BLKDEV);
5612 static_assert(BTRFS_FT_FIFO == FT_FIFO);
5613 static_assert(BTRFS_FT_SOCK == FT_SOCK);
5614 static_assert(BTRFS_FT_SYMLINK == FT_SYMLINK);
5616 static inline u8 btrfs_inode_type(struct inode *inode)
5618 return fs_umode_to_ftype(inode->i_mode);
5621 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5623 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5624 struct inode *inode;
5625 struct btrfs_root *root = BTRFS_I(dir)->root;
5626 struct btrfs_root *sub_root = root;
5627 struct btrfs_key location;
5631 if (dentry->d_name.len > BTRFS_NAME_LEN)
5632 return ERR_PTR(-ENAMETOOLONG);
5634 ret = btrfs_inode_by_name(dir, dentry, &location, &di_type);
5636 return ERR_PTR(ret);
5638 if (location.type == BTRFS_INODE_ITEM_KEY) {
5639 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5643 /* Do extra check against inode mode with di_type */
5644 if (btrfs_inode_type(inode) != di_type) {
5646 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5647 inode->i_mode, btrfs_inode_type(inode),
5650 return ERR_PTR(-EUCLEAN);
5655 ret = fixup_tree_root_location(fs_info, dir, dentry,
5656 &location, &sub_root);
5659 inode = ERR_PTR(ret);
5661 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5663 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5665 if (root != sub_root)
5666 btrfs_put_root(sub_root);
5668 if (!IS_ERR(inode) && root != sub_root) {
5669 down_read(&fs_info->cleanup_work_sem);
5670 if (!sb_rdonly(inode->i_sb))
5671 ret = btrfs_orphan_cleanup(sub_root);
5672 up_read(&fs_info->cleanup_work_sem);
5675 inode = ERR_PTR(ret);
5682 static int btrfs_dentry_delete(const struct dentry *dentry)
5684 struct btrfs_root *root;
5685 struct inode *inode = d_inode(dentry);
5687 if (!inode && !IS_ROOT(dentry))
5688 inode = d_inode(dentry->d_parent);
5691 root = BTRFS_I(inode)->root;
5692 if (btrfs_root_refs(&root->root_item) == 0)
5695 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5701 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5704 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5706 if (inode == ERR_PTR(-ENOENT))
5708 return d_splice_alias(inode, dentry);
5712 * All this infrastructure exists because dir_emit can fault, and we are holding
5713 * the tree lock when doing readdir. For now just allocate a buffer and copy
5714 * our information into that, and then dir_emit from the buffer. This is
5715 * similar to what NFS does, only we don't keep the buffer around in pagecache
5716 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5717 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5720 static int btrfs_opendir(struct inode *inode, struct file *file)
5722 struct btrfs_file_private *private;
5724 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5727 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5728 if (!private->filldir_buf) {
5732 file->private_data = private;
5743 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5746 struct dir_entry *entry = addr;
5747 char *name = (char *)(entry + 1);
5749 ctx->pos = get_unaligned(&entry->offset);
5750 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5751 get_unaligned(&entry->ino),
5752 get_unaligned(&entry->type)))
5754 addr += sizeof(struct dir_entry) +
5755 get_unaligned(&entry->name_len);
5761 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5763 struct inode *inode = file_inode(file);
5764 struct btrfs_root *root = BTRFS_I(inode)->root;
5765 struct btrfs_file_private *private = file->private_data;
5766 struct btrfs_dir_item *di;
5767 struct btrfs_key key;
5768 struct btrfs_key found_key;
5769 struct btrfs_path *path;
5771 struct list_head ins_list;
5772 struct list_head del_list;
5774 struct extent_buffer *leaf;
5781 struct btrfs_key location;
5783 if (!dir_emit_dots(file, ctx))
5786 path = btrfs_alloc_path();
5790 addr = private->filldir_buf;
5791 path->reada = READA_FORWARD;
5793 INIT_LIST_HEAD(&ins_list);
5794 INIT_LIST_HEAD(&del_list);
5795 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5798 key.type = BTRFS_DIR_INDEX_KEY;
5799 key.offset = ctx->pos;
5800 key.objectid = btrfs_ino(BTRFS_I(inode));
5802 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5807 struct dir_entry *entry;
5809 leaf = path->nodes[0];
5810 slot = path->slots[0];
5811 if (slot >= btrfs_header_nritems(leaf)) {
5812 ret = btrfs_next_leaf(root, path);
5820 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5822 if (found_key.objectid != key.objectid)
5824 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5826 if (found_key.offset < ctx->pos)
5828 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5830 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5831 name_len = btrfs_dir_name_len(leaf, di);
5832 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5834 btrfs_release_path(path);
5835 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5838 addr = private->filldir_buf;
5845 put_unaligned(name_len, &entry->name_len);
5846 name_ptr = (char *)(entry + 1);
5847 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5849 put_unaligned(fs_ftype_to_dtype(btrfs_dir_type(leaf, di)),
5851 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5852 put_unaligned(location.objectid, &entry->ino);
5853 put_unaligned(found_key.offset, &entry->offset);
5855 addr += sizeof(struct dir_entry) + name_len;
5856 total_len += sizeof(struct dir_entry) + name_len;
5860 btrfs_release_path(path);
5862 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5866 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5871 * Stop new entries from being returned after we return the last
5874 * New directory entries are assigned a strictly increasing
5875 * offset. This means that new entries created during readdir
5876 * are *guaranteed* to be seen in the future by that readdir.
5877 * This has broken buggy programs which operate on names as
5878 * they're returned by readdir. Until we re-use freed offsets
5879 * we have this hack to stop new entries from being returned
5880 * under the assumption that they'll never reach this huge
5883 * This is being careful not to overflow 32bit loff_t unless the
5884 * last entry requires it because doing so has broken 32bit apps
5887 if (ctx->pos >= INT_MAX)
5888 ctx->pos = LLONG_MAX;
5895 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5896 btrfs_free_path(path);
5901 * This is somewhat expensive, updating the tree every time the
5902 * inode changes. But, it is most likely to find the inode in cache.
5903 * FIXME, needs more benchmarking...there are no reasons other than performance
5904 * to keep or drop this code.
5906 static int btrfs_dirty_inode(struct inode *inode)
5908 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5909 struct btrfs_root *root = BTRFS_I(inode)->root;
5910 struct btrfs_trans_handle *trans;
5913 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5916 trans = btrfs_join_transaction(root);
5918 return PTR_ERR(trans);
5920 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5921 if (ret && (ret == -ENOSPC || ret == -EDQUOT)) {
5922 /* whoops, lets try again with the full transaction */
5923 btrfs_end_transaction(trans);
5924 trans = btrfs_start_transaction(root, 1);
5926 return PTR_ERR(trans);
5928 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5930 btrfs_end_transaction(trans);
5931 if (BTRFS_I(inode)->delayed_node)
5932 btrfs_balance_delayed_items(fs_info);
5938 * This is a copy of file_update_time. We need this so we can return error on
5939 * ENOSPC for updating the inode in the case of file write and mmap writes.
5941 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
5944 struct btrfs_root *root = BTRFS_I(inode)->root;
5945 bool dirty = flags & ~S_VERSION;
5947 if (btrfs_root_readonly(root))
5950 if (flags & S_VERSION)
5951 dirty |= inode_maybe_inc_iversion(inode, dirty);
5952 if (flags & S_CTIME)
5953 inode->i_ctime = *now;
5954 if (flags & S_MTIME)
5955 inode->i_mtime = *now;
5956 if (flags & S_ATIME)
5957 inode->i_atime = *now;
5958 return dirty ? btrfs_dirty_inode(inode) : 0;
5962 * find the highest existing sequence number in a directory
5963 * and then set the in-memory index_cnt variable to reflect
5964 * free sequence numbers
5966 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
5968 struct btrfs_root *root = inode->root;
5969 struct btrfs_key key, found_key;
5970 struct btrfs_path *path;
5971 struct extent_buffer *leaf;
5974 key.objectid = btrfs_ino(inode);
5975 key.type = BTRFS_DIR_INDEX_KEY;
5976 key.offset = (u64)-1;
5978 path = btrfs_alloc_path();
5982 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5985 /* FIXME: we should be able to handle this */
5990 if (path->slots[0] == 0) {
5991 inode->index_cnt = BTRFS_DIR_START_INDEX;
5997 leaf = path->nodes[0];
5998 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6000 if (found_key.objectid != btrfs_ino(inode) ||
6001 found_key.type != BTRFS_DIR_INDEX_KEY) {
6002 inode->index_cnt = BTRFS_DIR_START_INDEX;
6006 inode->index_cnt = found_key.offset + 1;
6008 btrfs_free_path(path);
6013 * helper to find a free sequence number in a given directory. This current
6014 * code is very simple, later versions will do smarter things in the btree
6016 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6020 if (dir->index_cnt == (u64)-1) {
6021 ret = btrfs_inode_delayed_dir_index_count(dir);
6023 ret = btrfs_set_inode_index_count(dir);
6029 *index = dir->index_cnt;
6035 static int btrfs_insert_inode_locked(struct inode *inode)
6037 struct btrfs_iget_args args;
6039 args.ino = BTRFS_I(inode)->location.objectid;
6040 args.root = BTRFS_I(inode)->root;
6042 return insert_inode_locked4(inode,
6043 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6044 btrfs_find_actor, &args);
6048 * Inherit flags from the parent inode.
6050 * Currently only the compression flags and the cow flags are inherited.
6052 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6059 flags = BTRFS_I(dir)->flags;
6061 if (flags & BTRFS_INODE_NOCOMPRESS) {
6062 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6063 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6064 } else if (flags & BTRFS_INODE_COMPRESS) {
6065 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6066 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6069 if (flags & BTRFS_INODE_NODATACOW) {
6070 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6071 if (S_ISREG(inode->i_mode))
6072 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6075 btrfs_sync_inode_flags_to_i_flags(inode);
6078 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6079 struct btrfs_root *root,
6080 struct user_namespace *mnt_userns,
6082 const char *name, int name_len,
6083 u64 ref_objectid, u64 objectid,
6084 umode_t mode, u64 *index)
6086 struct btrfs_fs_info *fs_info = root->fs_info;
6087 struct inode *inode;
6088 struct btrfs_inode_item *inode_item;
6089 struct btrfs_key *location;
6090 struct btrfs_path *path;
6091 struct btrfs_inode_ref *ref;
6092 struct btrfs_key key[2];
6094 struct btrfs_item_batch batch;
6096 unsigned int nofs_flag;
6099 path = btrfs_alloc_path();
6101 return ERR_PTR(-ENOMEM);
6103 nofs_flag = memalloc_nofs_save();
6104 inode = new_inode(fs_info->sb);
6105 memalloc_nofs_restore(nofs_flag);
6107 btrfs_free_path(path);
6108 return ERR_PTR(-ENOMEM);
6112 * O_TMPFILE, set link count to 0, so that after this point,
6113 * we fill in an inode item with the correct link count.
6116 set_nlink(inode, 0);
6119 * we have to initialize this early, so we can reclaim the inode
6120 * number if we fail afterwards in this function.
6122 inode->i_ino = objectid;
6125 trace_btrfs_inode_request(dir);
6127 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6129 btrfs_free_path(path);
6131 return ERR_PTR(ret);
6137 * index_cnt is ignored for everything but a dir,
6138 * btrfs_set_inode_index_count has an explanation for the magic
6141 BTRFS_I(inode)->index_cnt = 2;
6142 BTRFS_I(inode)->dir_index = *index;
6143 BTRFS_I(inode)->root = btrfs_grab_root(root);
6144 BTRFS_I(inode)->generation = trans->transid;
6145 inode->i_generation = BTRFS_I(inode)->generation;
6148 * We could have gotten an inode number from somebody who was fsynced
6149 * and then removed in this same transaction, so let's just set full
6150 * sync since it will be a full sync anyway and this will blow away the
6151 * old info in the log.
6153 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6155 key[0].objectid = objectid;
6156 key[0].type = BTRFS_INODE_ITEM_KEY;
6159 sizes[0] = sizeof(struct btrfs_inode_item);
6163 * Start new inodes with an inode_ref. This is slightly more
6164 * efficient for small numbers of hard links since they will
6165 * be packed into one item. Extended refs will kick in if we
6166 * add more hard links than can fit in the ref item.
6168 key[1].objectid = objectid;
6169 key[1].type = BTRFS_INODE_REF_KEY;
6170 key[1].offset = ref_objectid;
6172 sizes[1] = name_len + sizeof(*ref);
6175 location = &BTRFS_I(inode)->location;
6176 location->objectid = objectid;
6177 location->offset = 0;
6178 location->type = BTRFS_INODE_ITEM_KEY;
6180 ret = btrfs_insert_inode_locked(inode);
6186 batch.keys = &key[0];
6187 batch.data_sizes = &sizes[0];
6188 batch.total_data_size = sizes[0] + (name ? sizes[1] : 0);
6189 batch.nr = name ? 2 : 1;
6190 ret = btrfs_insert_empty_items(trans, root, path, &batch);
6194 inode_init_owner(mnt_userns, inode, dir, mode);
6195 inode_set_bytes(inode, 0);
6197 inode->i_mtime = current_time(inode);
6198 inode->i_atime = inode->i_mtime;
6199 inode->i_ctime = inode->i_mtime;
6200 BTRFS_I(inode)->i_otime = inode->i_mtime;
6202 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6203 struct btrfs_inode_item);
6204 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6205 sizeof(*inode_item));
6206 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6209 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6210 struct btrfs_inode_ref);
6211 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6212 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6213 ptr = (unsigned long)(ref + 1);
6214 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6217 btrfs_mark_buffer_dirty(path->nodes[0]);
6218 btrfs_free_path(path);
6220 btrfs_inherit_iflags(inode, dir);
6222 if (S_ISREG(mode)) {
6223 if (btrfs_test_opt(fs_info, NODATASUM))
6224 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6225 if (btrfs_test_opt(fs_info, NODATACOW))
6226 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6227 BTRFS_INODE_NODATASUM;
6230 inode_tree_add(inode);
6232 trace_btrfs_inode_new(inode);
6233 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6235 btrfs_update_root_times(trans, root);
6237 ret = btrfs_inode_inherit_props(trans, inode, dir);
6240 "error inheriting props for ino %llu (root %llu): %d",
6241 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6246 discard_new_inode(inode);
6249 BTRFS_I(dir)->index_cnt--;
6250 btrfs_free_path(path);
6251 return ERR_PTR(ret);
6255 * utility function to add 'inode' into 'parent_inode' with
6256 * a give name and a given sequence number.
6257 * if 'add_backref' is true, also insert a backref from the
6258 * inode to the parent directory.
6260 int btrfs_add_link(struct btrfs_trans_handle *trans,
6261 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6262 const char *name, int name_len, int add_backref, u64 index)
6265 struct btrfs_key key;
6266 struct btrfs_root *root = parent_inode->root;
6267 u64 ino = btrfs_ino(inode);
6268 u64 parent_ino = btrfs_ino(parent_inode);
6270 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6271 memcpy(&key, &inode->root->root_key, sizeof(key));
6274 key.type = BTRFS_INODE_ITEM_KEY;
6278 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6279 ret = btrfs_add_root_ref(trans, key.objectid,
6280 root->root_key.objectid, parent_ino,
6281 index, name, name_len);
6282 } else if (add_backref) {
6283 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6287 /* Nothing to clean up yet */
6291 ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key,
6292 btrfs_inode_type(&inode->vfs_inode), index);
6293 if (ret == -EEXIST || ret == -EOVERFLOW)
6296 btrfs_abort_transaction(trans, ret);
6300 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6302 inode_inc_iversion(&parent_inode->vfs_inode);
6304 * If we are replaying a log tree, we do not want to update the mtime
6305 * and ctime of the parent directory with the current time, since the
6306 * log replay procedure is responsible for setting them to their correct
6307 * values (the ones it had when the fsync was done).
6309 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
6310 struct timespec64 now = current_time(&parent_inode->vfs_inode);
6312 parent_inode->vfs_inode.i_mtime = now;
6313 parent_inode->vfs_inode.i_ctime = now;
6315 ret = btrfs_update_inode(trans, root, parent_inode);
6317 btrfs_abort_transaction(trans, ret);
6321 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6324 err = btrfs_del_root_ref(trans, key.objectid,
6325 root->root_key.objectid, parent_ino,
6326 &local_index, name, name_len);
6328 btrfs_abort_transaction(trans, err);
6329 } else if (add_backref) {
6333 err = btrfs_del_inode_ref(trans, root, name, name_len,
6334 ino, parent_ino, &local_index);
6336 btrfs_abort_transaction(trans, err);
6339 /* Return the original error code */
6343 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6344 struct btrfs_inode *dir, struct dentry *dentry,
6345 struct btrfs_inode *inode, int backref, u64 index)
6347 int err = btrfs_add_link(trans, dir, inode,
6348 dentry->d_name.name, dentry->d_name.len,
6355 static int btrfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
6356 struct dentry *dentry, umode_t mode, dev_t rdev)
6358 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6359 struct btrfs_trans_handle *trans;
6360 struct btrfs_root *root = BTRFS_I(dir)->root;
6361 struct inode *inode = NULL;
6367 * 2 for inode item and ref
6369 * 1 for xattr if selinux is on
6371 trans = btrfs_start_transaction(root, 5);
6373 return PTR_ERR(trans);
6375 err = btrfs_get_free_objectid(root, &objectid);
6379 inode = btrfs_new_inode(trans, root, mnt_userns, dir,
6380 dentry->d_name.name, dentry->d_name.len,
6381 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
6382 if (IS_ERR(inode)) {
6383 err = PTR_ERR(inode);
6389 * If the active LSM wants to access the inode during
6390 * d_instantiate it needs these. Smack checks to see
6391 * if the filesystem supports xattrs by looking at the
6394 inode->i_op = &btrfs_special_inode_operations;
6395 init_special_inode(inode, inode->i_mode, rdev);
6397 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6401 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6406 btrfs_update_inode(trans, root, BTRFS_I(inode));
6407 d_instantiate_new(dentry, inode);
6410 btrfs_end_transaction(trans);
6411 btrfs_btree_balance_dirty(fs_info);
6413 inode_dec_link_count(inode);
6414 discard_new_inode(inode);
6419 static int btrfs_create(struct user_namespace *mnt_userns, struct inode *dir,
6420 struct dentry *dentry, umode_t mode, bool excl)
6422 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6423 struct btrfs_trans_handle *trans;
6424 struct btrfs_root *root = BTRFS_I(dir)->root;
6425 struct inode *inode = NULL;
6431 * 2 for inode item and ref
6433 * 1 for xattr if selinux is on
6435 trans = btrfs_start_transaction(root, 5);
6437 return PTR_ERR(trans);
6439 err = btrfs_get_free_objectid(root, &objectid);
6443 inode = btrfs_new_inode(trans, root, mnt_userns, dir,
6444 dentry->d_name.name, dentry->d_name.len,
6445 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
6446 if (IS_ERR(inode)) {
6447 err = PTR_ERR(inode);
6452 * If the active LSM wants to access the inode during
6453 * d_instantiate it needs these. Smack checks to see
6454 * if the filesystem supports xattrs by looking at the
6457 inode->i_fop = &btrfs_file_operations;
6458 inode->i_op = &btrfs_file_inode_operations;
6459 inode->i_mapping->a_ops = &btrfs_aops;
6461 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6465 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6469 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6474 d_instantiate_new(dentry, inode);
6477 btrfs_end_transaction(trans);
6479 inode_dec_link_count(inode);
6480 discard_new_inode(inode);
6482 btrfs_btree_balance_dirty(fs_info);
6486 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6487 struct dentry *dentry)
6489 struct btrfs_trans_handle *trans = NULL;
6490 struct btrfs_root *root = BTRFS_I(dir)->root;
6491 struct inode *inode = d_inode(old_dentry);
6492 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6497 /* do not allow sys_link's with other subvols of the same device */
6498 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6501 if (inode->i_nlink >= BTRFS_LINK_MAX)
6504 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6509 * 2 items for inode and inode ref
6510 * 2 items for dir items
6511 * 1 item for parent inode
6512 * 1 item for orphan item deletion if O_TMPFILE
6514 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6515 if (IS_ERR(trans)) {
6516 err = PTR_ERR(trans);
6521 /* There are several dir indexes for this inode, clear the cache. */
6522 BTRFS_I(inode)->dir_index = 0ULL;
6524 inode_inc_iversion(inode);
6525 inode->i_ctime = current_time(inode);
6527 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6529 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6535 struct dentry *parent = dentry->d_parent;
6537 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6540 if (inode->i_nlink == 1) {
6542 * If new hard link count is 1, it's a file created
6543 * with open(2) O_TMPFILE flag.
6545 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6549 d_instantiate(dentry, inode);
6550 btrfs_log_new_name(trans, old_dentry, NULL, 0, parent);
6555 btrfs_end_transaction(trans);
6557 inode_dec_link_count(inode);
6560 btrfs_btree_balance_dirty(fs_info);
6564 static int btrfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
6565 struct dentry *dentry, umode_t mode)
6567 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6568 struct inode *inode = NULL;
6569 struct btrfs_trans_handle *trans;
6570 struct btrfs_root *root = BTRFS_I(dir)->root;
6576 * 2 items for inode and ref
6577 * 2 items for dir items
6578 * 1 for xattr if selinux is on
6580 trans = btrfs_start_transaction(root, 5);
6582 return PTR_ERR(trans);
6584 err = btrfs_get_free_objectid(root, &objectid);
6588 inode = btrfs_new_inode(trans, root, mnt_userns, dir,
6589 dentry->d_name.name, dentry->d_name.len,
6590 btrfs_ino(BTRFS_I(dir)), objectid,
6591 S_IFDIR | mode, &index);
6592 if (IS_ERR(inode)) {
6593 err = PTR_ERR(inode);
6598 /* these must be set before we unlock the inode */
6599 inode->i_op = &btrfs_dir_inode_operations;
6600 inode->i_fop = &btrfs_dir_file_operations;
6602 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6606 btrfs_i_size_write(BTRFS_I(inode), 0);
6607 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6611 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6612 dentry->d_name.name,
6613 dentry->d_name.len, 0, index);
6617 d_instantiate_new(dentry, inode);
6620 btrfs_end_transaction(trans);
6622 inode_dec_link_count(inode);
6623 discard_new_inode(inode);
6625 btrfs_btree_balance_dirty(fs_info);
6629 static noinline int uncompress_inline(struct btrfs_path *path,
6631 size_t pg_offset, u64 extent_offset,
6632 struct btrfs_file_extent_item *item)
6635 struct extent_buffer *leaf = path->nodes[0];
6638 unsigned long inline_size;
6642 WARN_ON(pg_offset != 0);
6643 compress_type = btrfs_file_extent_compression(leaf, item);
6644 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6645 inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]);
6646 tmp = kmalloc(inline_size, GFP_NOFS);
6649 ptr = btrfs_file_extent_inline_start(item);
6651 read_extent_buffer(leaf, tmp, ptr, inline_size);
6653 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6654 ret = btrfs_decompress(compress_type, tmp, page,
6655 extent_offset, inline_size, max_size);
6658 * decompression code contains a memset to fill in any space between the end
6659 * of the uncompressed data and the end of max_size in case the decompressed
6660 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6661 * the end of an inline extent and the beginning of the next block, so we
6662 * cover that region here.
6665 if (max_size + pg_offset < PAGE_SIZE)
6666 memzero_page(page, pg_offset + max_size,
6667 PAGE_SIZE - max_size - pg_offset);
6673 * btrfs_get_extent - Lookup the first extent overlapping a range in a file.
6674 * @inode: file to search in
6675 * @page: page to read extent data into if the extent is inline
6676 * @pg_offset: offset into @page to copy to
6677 * @start: file offset
6678 * @len: length of range starting at @start
6680 * This returns the first &struct extent_map which overlaps with the given
6681 * range, reading it from the B-tree and caching it if necessary. Note that
6682 * there may be more extents which overlap the given range after the returned
6685 * If @page is not NULL and the extent is inline, this also reads the extent
6686 * data directly into the page and marks the extent up to date in the io_tree.
6688 * Return: ERR_PTR on error, non-NULL extent_map on success.
6690 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6691 struct page *page, size_t pg_offset,
6694 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6696 u64 extent_start = 0;
6698 u64 objectid = btrfs_ino(inode);
6699 int extent_type = -1;
6700 struct btrfs_path *path = NULL;
6701 struct btrfs_root *root = inode->root;
6702 struct btrfs_file_extent_item *item;
6703 struct extent_buffer *leaf;
6704 struct btrfs_key found_key;
6705 struct extent_map *em = NULL;
6706 struct extent_map_tree *em_tree = &inode->extent_tree;
6707 struct extent_io_tree *io_tree = &inode->io_tree;
6709 read_lock(&em_tree->lock);
6710 em = lookup_extent_mapping(em_tree, start, len);
6711 read_unlock(&em_tree->lock);
6714 if (em->start > start || em->start + em->len <= start)
6715 free_extent_map(em);
6716 else if (em->block_start == EXTENT_MAP_INLINE && page)
6717 free_extent_map(em);
6721 em = alloc_extent_map();
6726 em->start = EXTENT_MAP_HOLE;
6727 em->orig_start = EXTENT_MAP_HOLE;
6729 em->block_len = (u64)-1;
6731 path = btrfs_alloc_path();
6737 /* Chances are we'll be called again, so go ahead and do readahead */
6738 path->reada = READA_FORWARD;
6741 * The same explanation in load_free_space_cache applies here as well,
6742 * we only read when we're loading the free space cache, and at that
6743 * point the commit_root has everything we need.
6745 if (btrfs_is_free_space_inode(inode)) {
6746 path->search_commit_root = 1;
6747 path->skip_locking = 1;
6750 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6753 } else if (ret > 0) {
6754 if (path->slots[0] == 0)
6760 leaf = path->nodes[0];
6761 item = btrfs_item_ptr(leaf, path->slots[0],
6762 struct btrfs_file_extent_item);
6763 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6764 if (found_key.objectid != objectid ||
6765 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6767 * If we backup past the first extent we want to move forward
6768 * and see if there is an extent in front of us, otherwise we'll
6769 * say there is a hole for our whole search range which can
6776 extent_type = btrfs_file_extent_type(leaf, item);
6777 extent_start = found_key.offset;
6778 extent_end = btrfs_file_extent_end(path);
6779 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6780 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6781 /* Only regular file could have regular/prealloc extent */
6782 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6785 "regular/prealloc extent found for non-regular inode %llu",
6789 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6791 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6792 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6797 if (start >= extent_end) {
6799 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6800 ret = btrfs_next_leaf(root, path);
6806 leaf = path->nodes[0];
6808 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6809 if (found_key.objectid != objectid ||
6810 found_key.type != BTRFS_EXTENT_DATA_KEY)
6812 if (start + len <= found_key.offset)
6814 if (start > found_key.offset)
6817 /* New extent overlaps with existing one */
6819 em->orig_start = start;
6820 em->len = found_key.offset - start;
6821 em->block_start = EXTENT_MAP_HOLE;
6825 btrfs_extent_item_to_extent_map(inode, path, item, !page, em);
6827 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6828 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6830 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6834 size_t extent_offset;
6840 size = btrfs_file_extent_ram_bytes(leaf, item);
6841 extent_offset = page_offset(page) + pg_offset - extent_start;
6842 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6843 size - extent_offset);
6844 em->start = extent_start + extent_offset;
6845 em->len = ALIGN(copy_size, fs_info->sectorsize);
6846 em->orig_block_len = em->len;
6847 em->orig_start = em->start;
6848 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6850 if (!PageUptodate(page)) {
6851 if (btrfs_file_extent_compression(leaf, item) !=
6852 BTRFS_COMPRESS_NONE) {
6853 ret = uncompress_inline(path, page, pg_offset,
6854 extent_offset, item);
6858 map = kmap_local_page(page);
6859 read_extent_buffer(leaf, map + pg_offset, ptr,
6861 if (pg_offset + copy_size < PAGE_SIZE) {
6862 memset(map + pg_offset + copy_size, 0,
6863 PAGE_SIZE - pg_offset -
6868 flush_dcache_page(page);
6870 set_extent_uptodate(io_tree, em->start,
6871 extent_map_end(em) - 1, NULL, GFP_NOFS);
6876 em->orig_start = start;
6878 em->block_start = EXTENT_MAP_HOLE;
6881 btrfs_release_path(path);
6882 if (em->start > start || extent_map_end(em) <= start) {
6884 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6885 em->start, em->len, start, len);
6890 write_lock(&em_tree->lock);
6891 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6892 write_unlock(&em_tree->lock);
6894 btrfs_free_path(path);
6896 trace_btrfs_get_extent(root, inode, em);
6899 free_extent_map(em);
6900 return ERR_PTR(ret);
6905 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
6908 struct extent_map *em;
6909 struct extent_map *hole_em = NULL;
6910 u64 delalloc_start = start;
6916 em = btrfs_get_extent(inode, NULL, 0, start, len);
6920 * If our em maps to:
6922 * - a pre-alloc extent,
6923 * there might actually be delalloc bytes behind it.
6925 if (em->block_start != EXTENT_MAP_HOLE &&
6926 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6931 /* check to see if we've wrapped (len == -1 or similar) */
6940 /* ok, we didn't find anything, lets look for delalloc */
6941 delalloc_len = count_range_bits(&inode->io_tree, &delalloc_start,
6942 end, len, EXTENT_DELALLOC, 1);
6943 delalloc_end = delalloc_start + delalloc_len;
6944 if (delalloc_end < delalloc_start)
6945 delalloc_end = (u64)-1;
6948 * We didn't find anything useful, return the original results from
6951 if (delalloc_start > end || delalloc_end <= start) {
6958 * Adjust the delalloc_start to make sure it doesn't go backwards from
6959 * the start they passed in
6961 delalloc_start = max(start, delalloc_start);
6962 delalloc_len = delalloc_end - delalloc_start;
6964 if (delalloc_len > 0) {
6967 const u64 hole_end = extent_map_end(hole_em);
6969 em = alloc_extent_map();
6977 * When btrfs_get_extent can't find anything it returns one
6980 * Make sure what it found really fits our range, and adjust to
6981 * make sure it is based on the start from the caller
6983 if (hole_end <= start || hole_em->start > end) {
6984 free_extent_map(hole_em);
6987 hole_start = max(hole_em->start, start);
6988 hole_len = hole_end - hole_start;
6991 if (hole_em && delalloc_start > hole_start) {
6993 * Our hole starts before our delalloc, so we have to
6994 * return just the parts of the hole that go until the
6997 em->len = min(hole_len, delalloc_start - hole_start);
6998 em->start = hole_start;
6999 em->orig_start = hole_start;
7001 * Don't adjust block start at all, it is fixed at
7004 em->block_start = hole_em->block_start;
7005 em->block_len = hole_len;
7006 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7007 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7010 * Hole is out of passed range or it starts after
7013 em->start = delalloc_start;
7014 em->len = delalloc_len;
7015 em->orig_start = delalloc_start;
7016 em->block_start = EXTENT_MAP_DELALLOC;
7017 em->block_len = delalloc_len;
7024 free_extent_map(hole_em);
7026 free_extent_map(em);
7027 return ERR_PTR(err);
7032 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
7035 const u64 orig_start,
7036 const u64 block_start,
7037 const u64 block_len,
7038 const u64 orig_block_len,
7039 const u64 ram_bytes,
7042 struct extent_map *em = NULL;
7045 if (type != BTRFS_ORDERED_NOCOW) {
7046 em = create_io_em(inode, start, len, orig_start, block_start,
7047 block_len, orig_block_len, ram_bytes,
7048 BTRFS_COMPRESS_NONE, /* compress_type */
7053 ret = btrfs_add_ordered_extent(inode, start, len, len, block_start,
7056 (1 << BTRFS_ORDERED_DIRECT),
7057 BTRFS_COMPRESS_NONE);
7060 free_extent_map(em);
7061 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
7070 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
7073 struct btrfs_root *root = inode->root;
7074 struct btrfs_fs_info *fs_info = root->fs_info;
7075 struct extent_map *em;
7076 struct btrfs_key ins;
7080 alloc_hint = get_extent_allocation_hint(inode, start, len);
7081 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7082 0, alloc_hint, &ins, 1, 1);
7084 return ERR_PTR(ret);
7086 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7087 ins.objectid, ins.offset, ins.offset,
7088 ins.offset, BTRFS_ORDERED_REGULAR);
7089 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7091 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7097 static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
7099 struct btrfs_block_group *block_group;
7100 bool readonly = false;
7102 block_group = btrfs_lookup_block_group(fs_info, bytenr);
7103 if (!block_group || block_group->ro)
7106 btrfs_put_block_group(block_group);
7111 * Check if we can do nocow write into the range [@offset, @offset + @len)
7113 * @offset: File offset
7114 * @len: The length to write, will be updated to the nocow writeable
7116 * @orig_start: (optional) Return the original file offset of the file extent
7117 * @orig_len: (optional) Return the original on-disk length of the file extent
7118 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7119 * @strict: if true, omit optimizations that might force us into unnecessary
7120 * cow. e.g., don't trust generation number.
7123 * >0 and update @len if we can do nocow write
7124 * 0 if we can't do nocow write
7125 * <0 if error happened
7127 * NOTE: This only checks the file extents, caller is responsible to wait for
7128 * any ordered extents.
7130 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7131 u64 *orig_start, u64 *orig_block_len,
7132 u64 *ram_bytes, bool strict)
7134 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7135 struct btrfs_path *path;
7137 struct extent_buffer *leaf;
7138 struct btrfs_root *root = BTRFS_I(inode)->root;
7139 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7140 struct btrfs_file_extent_item *fi;
7141 struct btrfs_key key;
7148 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7150 path = btrfs_alloc_path();
7154 ret = btrfs_lookup_file_extent(NULL, root, path,
7155 btrfs_ino(BTRFS_I(inode)), offset, 0);
7159 slot = path->slots[0];
7162 /* can't find the item, must cow */
7169 leaf = path->nodes[0];
7170 btrfs_item_key_to_cpu(leaf, &key, slot);
7171 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7172 key.type != BTRFS_EXTENT_DATA_KEY) {
7173 /* not our file or wrong item type, must cow */
7177 if (key.offset > offset) {
7178 /* Wrong offset, must cow */
7182 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7183 found_type = btrfs_file_extent_type(leaf, fi);
7184 if (found_type != BTRFS_FILE_EXTENT_REG &&
7185 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7186 /* not a regular extent, must cow */
7190 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7193 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7194 if (extent_end <= offset)
7197 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7198 if (disk_bytenr == 0)
7201 if (btrfs_file_extent_compression(leaf, fi) ||
7202 btrfs_file_extent_encryption(leaf, fi) ||
7203 btrfs_file_extent_other_encoding(leaf, fi))
7207 * Do the same check as in btrfs_cross_ref_exist but without the
7208 * unnecessary search.
7211 (btrfs_file_extent_generation(leaf, fi) <=
7212 btrfs_root_last_snapshot(&root->root_item)))
7215 backref_offset = btrfs_file_extent_offset(leaf, fi);
7218 *orig_start = key.offset - backref_offset;
7219 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7220 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7223 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7226 num_bytes = min(offset + *len, extent_end) - offset;
7227 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7230 range_end = round_up(offset + num_bytes,
7231 root->fs_info->sectorsize) - 1;
7232 ret = test_range_bit(io_tree, offset, range_end,
7233 EXTENT_DELALLOC, 0, NULL);
7240 btrfs_release_path(path);
7243 * look for other files referencing this extent, if we
7244 * find any we must cow
7247 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7248 key.offset - backref_offset, disk_bytenr,
7256 * adjust disk_bytenr and num_bytes to cover just the bytes
7257 * in this extent we are about to write. If there
7258 * are any csums in that range we have to cow in order
7259 * to keep the csums correct
7261 disk_bytenr += backref_offset;
7262 disk_bytenr += offset - key.offset;
7263 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7266 * all of the above have passed, it is safe to overwrite this extent
7272 btrfs_free_path(path);
7276 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7277 struct extent_state **cached_state, bool writing)
7279 struct btrfs_ordered_extent *ordered;
7283 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7286 * We're concerned with the entire range that we're going to be
7287 * doing DIO to, so we need to make sure there's no ordered
7288 * extents in this range.
7290 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7291 lockend - lockstart + 1);
7294 * We need to make sure there are no buffered pages in this
7295 * range either, we could have raced between the invalidate in
7296 * generic_file_direct_write and locking the extent. The
7297 * invalidate needs to happen so that reads after a write do not
7301 (!writing || !filemap_range_has_page(inode->i_mapping,
7302 lockstart, lockend)))
7305 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7310 * If we are doing a DIO read and the ordered extent we
7311 * found is for a buffered write, we can not wait for it
7312 * to complete and retry, because if we do so we can
7313 * deadlock with concurrent buffered writes on page
7314 * locks. This happens only if our DIO read covers more
7315 * than one extent map, if at this point has already
7316 * created an ordered extent for a previous extent map
7317 * and locked its range in the inode's io tree, and a
7318 * concurrent write against that previous extent map's
7319 * range and this range started (we unlock the ranges
7320 * in the io tree only when the bios complete and
7321 * buffered writes always lock pages before attempting
7322 * to lock range in the io tree).
7325 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7326 btrfs_start_ordered_extent(ordered, 1);
7329 btrfs_put_ordered_extent(ordered);
7332 * We could trigger writeback for this range (and wait
7333 * for it to complete) and then invalidate the pages for
7334 * this range (through invalidate_inode_pages2_range()),
7335 * but that can lead us to a deadlock with a concurrent
7336 * call to readahead (a buffered read or a defrag call
7337 * triggered a readahead) on a page lock due to an
7338 * ordered dio extent we created before but did not have
7339 * yet a corresponding bio submitted (whence it can not
7340 * complete), which makes readahead wait for that
7341 * ordered extent to complete while holding a lock on
7356 /* The callers of this must take lock_extent() */
7357 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7358 u64 len, u64 orig_start, u64 block_start,
7359 u64 block_len, u64 orig_block_len,
7360 u64 ram_bytes, int compress_type,
7363 struct extent_map_tree *em_tree;
7364 struct extent_map *em;
7367 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7368 type == BTRFS_ORDERED_COMPRESSED ||
7369 type == BTRFS_ORDERED_NOCOW ||
7370 type == BTRFS_ORDERED_REGULAR);
7372 em_tree = &inode->extent_tree;
7373 em = alloc_extent_map();
7375 return ERR_PTR(-ENOMEM);
7378 em->orig_start = orig_start;
7380 em->block_len = block_len;
7381 em->block_start = block_start;
7382 em->orig_block_len = orig_block_len;
7383 em->ram_bytes = ram_bytes;
7384 em->generation = -1;
7385 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7386 if (type == BTRFS_ORDERED_PREALLOC) {
7387 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7388 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7389 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7390 em->compress_type = compress_type;
7394 btrfs_drop_extent_cache(inode, em->start,
7395 em->start + em->len - 1, 0);
7396 write_lock(&em_tree->lock);
7397 ret = add_extent_mapping(em_tree, em, 1);
7398 write_unlock(&em_tree->lock);
7400 * The caller has taken lock_extent(), who could race with us
7403 } while (ret == -EEXIST);
7406 free_extent_map(em);
7407 return ERR_PTR(ret);
7410 /* em got 2 refs now, callers needs to do free_extent_map once. */
7415 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7416 struct inode *inode,
7417 struct btrfs_dio_data *dio_data,
7420 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7421 struct extent_map *em = *map;
7423 u64 block_start, orig_start, orig_block_len, ram_bytes;
7424 bool can_nocow = false;
7425 bool space_reserved = false;
7429 * We don't allocate a new extent in the following cases
7431 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7433 * 2) The extent is marked as PREALLOC. We're good to go here and can
7434 * just use the extent.
7437 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7438 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7439 em->block_start != EXTENT_MAP_HOLE)) {
7440 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7441 type = BTRFS_ORDERED_PREALLOC;
7443 type = BTRFS_ORDERED_NOCOW;
7444 len = min(len, em->len - (start - em->start));
7445 block_start = em->block_start + (start - em->start);
7447 if (can_nocow_extent(inode, start, &len, &orig_start,
7448 &orig_block_len, &ram_bytes, false) == 1 &&
7449 btrfs_inc_nocow_writers(fs_info, block_start))
7454 struct extent_map *em2;
7456 /* We can NOCOW, so only need to reserve metadata space. */
7457 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len);
7459 /* Our caller expects us to free the input extent map. */
7460 free_extent_map(em);
7462 btrfs_dec_nocow_writers(fs_info, block_start);
7465 space_reserved = true;
7467 em2 = btrfs_create_dio_extent(BTRFS_I(inode), start, len,
7468 orig_start, block_start,
7469 len, orig_block_len,
7471 btrfs_dec_nocow_writers(fs_info, block_start);
7472 if (type == BTRFS_ORDERED_PREALLOC) {
7473 free_extent_map(em);
7482 const u64 prev_len = len;
7484 /* Our caller expects us to free the input extent map. */
7485 free_extent_map(em);
7488 /* We have to COW, so need to reserve metadata and data space. */
7489 ret = btrfs_delalloc_reserve_space(BTRFS_I(inode),
7490 &dio_data->data_reserved,
7494 space_reserved = true;
7496 em = btrfs_new_extent_direct(BTRFS_I(inode), start, len);
7502 len = min(len, em->len - (start - em->start));
7504 btrfs_delalloc_release_space(BTRFS_I(inode),
7505 dio_data->data_reserved,
7506 start + len, prev_len - len,
7511 * We have created our ordered extent, so we can now release our reservation
7512 * for an outstanding extent.
7514 btrfs_delalloc_release_extents(BTRFS_I(inode), len);
7517 * Need to update the i_size under the extent lock so buffered
7518 * readers will get the updated i_size when we unlock.
7520 if (start + len > i_size_read(inode))
7521 i_size_write(inode, start + len);
7523 if (ret && space_reserved) {
7524 btrfs_delalloc_release_extents(BTRFS_I(inode), len);
7526 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, true);
7528 btrfs_delalloc_release_space(BTRFS_I(inode),
7529 dio_data->data_reserved,
7531 extent_changeset_free(dio_data->data_reserved);
7532 dio_data->data_reserved = NULL;
7538 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7539 loff_t length, unsigned int flags, struct iomap *iomap,
7540 struct iomap *srcmap)
7542 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7543 struct extent_map *em;
7544 struct extent_state *cached_state = NULL;
7545 struct btrfs_dio_data *dio_data = NULL;
7546 u64 lockstart, lockend;
7547 const bool write = !!(flags & IOMAP_WRITE);
7550 bool unlock_extents = false;
7553 len = min_t(u64, len, fs_info->sectorsize);
7556 lockend = start + len - 1;
7559 * The generic stuff only does filemap_write_and_wait_range, which
7560 * isn't enough if we've written compressed pages to this area, so we
7561 * need to flush the dirty pages again to make absolutely sure that any
7562 * outstanding dirty pages are on disk.
7564 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7565 &BTRFS_I(inode)->runtime_flags)) {
7566 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7567 start + length - 1);
7572 dio_data = kzalloc(sizeof(*dio_data), GFP_NOFS);
7576 iomap->private = dio_data;
7580 * If this errors out it's because we couldn't invalidate pagecache for
7581 * this range and we need to fallback to buffered.
7583 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, write)) {
7588 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7595 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7596 * io. INLINE is special, and we could probably kludge it in here, but
7597 * it's still buffered so for safety lets just fall back to the generic
7600 * For COMPRESSED we _have_ to read the entire extent in so we can
7601 * decompress it, so there will be buffering required no matter what we
7602 * do, so go ahead and fallback to buffered.
7604 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7605 * to buffered IO. Don't blame me, this is the price we pay for using
7608 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7609 em->block_start == EXTENT_MAP_INLINE) {
7610 free_extent_map(em);
7615 len = min(len, em->len - (start - em->start));
7618 * If we have a NOWAIT request and the range contains multiple extents
7619 * (or a mix of extents and holes), then we return -EAGAIN to make the
7620 * caller fallback to a context where it can do a blocking (without
7621 * NOWAIT) request. This way we avoid doing partial IO and returning
7622 * success to the caller, which is not optimal for writes and for reads
7623 * it can result in unexpected behaviour for an application.
7625 * When doing a read, because we use IOMAP_DIO_PARTIAL when calling
7626 * iomap_dio_rw(), we can end up returning less data then what the caller
7627 * asked for, resulting in an unexpected, and incorrect, short read.
7628 * That is, the caller asked to read N bytes and we return less than that,
7629 * which is wrong unless we are crossing EOF. This happens if we get a
7630 * page fault error when trying to fault in pages for the buffer that is
7631 * associated to the struct iov_iter passed to iomap_dio_rw(), and we
7632 * have previously submitted bios for other extents in the range, in
7633 * which case iomap_dio_rw() may return us EIOCBQUEUED if not all of
7634 * those bios have completed by the time we get the page fault error,
7635 * which we return back to our caller - we should only return EIOCBQUEUED
7636 * after we have submitted bios for all the extents in the range.
7638 if ((flags & IOMAP_NOWAIT) && len < length) {
7639 free_extent_map(em);
7645 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7649 unlock_extents = true;
7650 /* Recalc len in case the new em is smaller than requested */
7651 len = min(len, em->len - (start - em->start));
7654 * We need to unlock only the end area that we aren't using.
7655 * The rest is going to be unlocked by the endio routine.
7657 lockstart = start + len;
7658 if (lockstart < lockend)
7659 unlock_extents = true;
7663 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
7664 lockstart, lockend, &cached_state);
7666 free_extent_state(cached_state);
7669 * Translate extent map information to iomap.
7670 * We trim the extents (and move the addr) even though iomap code does
7671 * that, since we have locked only the parts we are performing I/O in.
7673 if ((em->block_start == EXTENT_MAP_HOLE) ||
7674 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7675 iomap->addr = IOMAP_NULL_ADDR;
7676 iomap->type = IOMAP_HOLE;
7678 iomap->addr = em->block_start + (start - em->start);
7679 iomap->type = IOMAP_MAPPED;
7681 iomap->offset = start;
7682 iomap->bdev = fs_info->fs_devices->latest_dev->bdev;
7683 iomap->length = len;
7685 if (write && btrfs_use_zone_append(BTRFS_I(inode), em->block_start))
7686 iomap->flags |= IOMAP_F_ZONE_APPEND;
7688 free_extent_map(em);
7693 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7701 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7702 ssize_t written, unsigned int flags, struct iomap *iomap)
7705 struct btrfs_dio_data *dio_data = iomap->private;
7706 size_t submitted = dio_data->submitted;
7707 const bool write = !!(flags & IOMAP_WRITE);
7709 if (!write && (iomap->type == IOMAP_HOLE)) {
7710 /* If reading from a hole, unlock and return */
7711 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1);
7715 if (submitted < length) {
7717 length -= submitted;
7719 __endio_write_update_ordered(BTRFS_I(inode), pos,
7722 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7728 extent_changeset_free(dio_data->data_reserved);
7731 iomap->private = NULL;
7736 static void btrfs_dio_private_put(struct btrfs_dio_private *dip)
7739 * This implies a barrier so that stores to dio_bio->bi_status before
7740 * this and loads of dio_bio->bi_status after this are fully ordered.
7742 if (!refcount_dec_and_test(&dip->refs))
7745 if (btrfs_op(dip->dio_bio) == BTRFS_MAP_WRITE) {
7746 __endio_write_update_ordered(BTRFS_I(dip->inode),
7749 !dip->dio_bio->bi_status);
7751 unlock_extent(&BTRFS_I(dip->inode)->io_tree,
7753 dip->file_offset + dip->bytes - 1);
7756 bio_endio(dip->dio_bio);
7760 static blk_status_t submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7762 unsigned long bio_flags)
7764 struct btrfs_dio_private *dip = bio->bi_private;
7765 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7768 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7770 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
7774 refcount_inc(&dip->refs);
7775 ret = btrfs_map_bio(fs_info, bio, mirror_num);
7777 refcount_dec(&dip->refs);
7781 static blk_status_t btrfs_check_read_dio_bio(struct btrfs_dio_private *dip,
7782 struct btrfs_bio *bbio,
7783 const bool uptodate)
7785 struct inode *inode = dip->inode;
7786 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
7787 const u32 sectorsize = fs_info->sectorsize;
7788 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7789 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7790 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7791 struct bio_vec bvec;
7792 struct bvec_iter iter;
7793 const u64 orig_file_offset = dip->file_offset;
7794 u64 start = orig_file_offset;
7796 blk_status_t err = BLK_STS_OK;
7798 __bio_for_each_segment(bvec, &bbio->bio, iter, bbio->iter) {
7799 unsigned int i, nr_sectors, pgoff;
7801 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7802 pgoff = bvec.bv_offset;
7803 for (i = 0; i < nr_sectors; i++) {
7804 ASSERT(pgoff < PAGE_SIZE);
7806 (!csum || !check_data_csum(inode, bbio,
7807 bio_offset, bvec.bv_page,
7809 clean_io_failure(fs_info, failure_tree, io_tree,
7810 start, bvec.bv_page,
7811 btrfs_ino(BTRFS_I(inode)),
7816 ASSERT((start - orig_file_offset) < UINT_MAX);
7817 ret = btrfs_repair_one_sector(inode,
7819 start - orig_file_offset,
7820 bvec.bv_page, pgoff,
7821 start, bbio->mirror_num,
7822 submit_dio_repair_bio);
7824 err = errno_to_blk_status(ret);
7826 start += sectorsize;
7827 ASSERT(bio_offset + sectorsize > bio_offset);
7828 bio_offset += sectorsize;
7829 pgoff += sectorsize;
7835 static void __endio_write_update_ordered(struct btrfs_inode *inode,
7836 const u64 offset, const u64 bytes,
7837 const bool uptodate)
7839 btrfs_mark_ordered_io_finished(inode, NULL, offset, bytes,
7840 finish_ordered_fn, uptodate);
7843 static blk_status_t btrfs_submit_bio_start_direct_io(struct inode *inode,
7845 u64 dio_file_offset)
7847 return btrfs_csum_one_bio(BTRFS_I(inode), bio, dio_file_offset, false);
7850 static void btrfs_end_dio_bio(struct bio *bio)
7852 struct btrfs_dio_private *dip = bio->bi_private;
7853 blk_status_t err = bio->bi_status;
7856 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
7857 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
7858 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
7859 bio->bi_opf, bio->bi_iter.bi_sector,
7860 bio->bi_iter.bi_size, err);
7862 if (bio_op(bio) == REQ_OP_READ)
7863 err = btrfs_check_read_dio_bio(dip, btrfs_bio(bio), !err);
7866 dip->dio_bio->bi_status = err;
7868 btrfs_record_physical_zoned(dip->inode, dip->file_offset, bio);
7871 btrfs_dio_private_put(dip);
7874 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
7875 struct inode *inode, u64 file_offset, int async_submit)
7877 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7878 struct btrfs_dio_private *dip = bio->bi_private;
7879 bool write = btrfs_op(bio) == BTRFS_MAP_WRITE;
7882 /* Check btrfs_submit_bio_hook() for rules about async submit. */
7884 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7887 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
7892 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
7895 if (write && async_submit) {
7896 ret = btrfs_wq_submit_bio(inode, bio, 0, 0, file_offset,
7897 btrfs_submit_bio_start_direct_io);
7901 * If we aren't doing async submit, calculate the csum of the
7904 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, file_offset, false);
7910 csum_offset = file_offset - dip->file_offset;
7911 csum_offset >>= fs_info->sectorsize_bits;
7912 csum_offset *= fs_info->csum_size;
7913 btrfs_bio(bio)->csum = dip->csums + csum_offset;
7916 ret = btrfs_map_bio(fs_info, bio, 0);
7922 * If this succeeds, the btrfs_dio_private is responsible for cleaning up locked
7923 * or ordered extents whether or not we submit any bios.
7925 static struct btrfs_dio_private *btrfs_create_dio_private(struct bio *dio_bio,
7926 struct inode *inode,
7929 const bool write = (btrfs_op(dio_bio) == BTRFS_MAP_WRITE);
7930 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7932 struct btrfs_dio_private *dip;
7934 dip_size = sizeof(*dip);
7935 if (!write && csum) {
7936 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7939 nblocks = dio_bio->bi_iter.bi_size >> fs_info->sectorsize_bits;
7940 dip_size += fs_info->csum_size * nblocks;
7943 dip = kzalloc(dip_size, GFP_NOFS);
7948 dip->file_offset = file_offset;
7949 dip->bytes = dio_bio->bi_iter.bi_size;
7950 dip->disk_bytenr = dio_bio->bi_iter.bi_sector << 9;
7951 dip->dio_bio = dio_bio;
7952 refcount_set(&dip->refs, 1);
7956 static void btrfs_submit_direct(const struct iomap_iter *iter,
7957 struct bio *dio_bio, loff_t file_offset)
7959 struct inode *inode = iter->inode;
7960 const bool write = (btrfs_op(dio_bio) == BTRFS_MAP_WRITE);
7961 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7962 const bool raid56 = (btrfs_data_alloc_profile(fs_info) &
7963 BTRFS_BLOCK_GROUP_RAID56_MASK);
7964 struct btrfs_dio_private *dip;
7967 int async_submit = 0;
7969 u64 clone_offset = 0;
7973 blk_status_t status;
7974 struct btrfs_io_geometry geom;
7975 struct btrfs_dio_data *dio_data = iter->iomap.private;
7976 struct extent_map *em = NULL;
7978 dip = btrfs_create_dio_private(dio_bio, inode, file_offset);
7981 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
7982 file_offset + dio_bio->bi_iter.bi_size - 1);
7984 dio_bio->bi_status = BLK_STS_RESOURCE;
7991 * Load the csums up front to reduce csum tree searches and
7992 * contention when submitting bios.
7994 * If we have csums disabled this will do nothing.
7996 status = btrfs_lookup_bio_sums(inode, dio_bio, dip->csums);
7997 if (status != BLK_STS_OK)
8001 start_sector = dio_bio->bi_iter.bi_sector;
8002 submit_len = dio_bio->bi_iter.bi_size;
8005 logical = start_sector << 9;
8006 em = btrfs_get_chunk_map(fs_info, logical, submit_len);
8008 status = errno_to_blk_status(PTR_ERR(em));
8012 ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(dio_bio),
8015 status = errno_to_blk_status(ret);
8019 clone_len = min(submit_len, geom.len);
8020 ASSERT(clone_len <= UINT_MAX);
8023 * This will never fail as it's passing GPF_NOFS and
8024 * the allocation is backed by btrfs_bioset.
8026 bio = btrfs_bio_clone_partial(dio_bio, clone_offset, clone_len);
8027 bio->bi_private = dip;
8028 bio->bi_end_io = btrfs_end_dio_bio;
8030 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
8031 status = extract_ordered_extent(BTRFS_I(inode), bio,
8039 ASSERT(submit_len >= clone_len);
8040 submit_len -= clone_len;
8043 * Increase the count before we submit the bio so we know
8044 * the end IO handler won't happen before we increase the
8045 * count. Otherwise, the dip might get freed before we're
8046 * done setting it up.
8048 * We transfer the initial reference to the last bio, so we
8049 * don't need to increment the reference count for the last one.
8051 if (submit_len > 0) {
8052 refcount_inc(&dip->refs);
8054 * If we are submitting more than one bio, submit them
8055 * all asynchronously. The exception is RAID 5 or 6, as
8056 * asynchronous checksums make it difficult to collect
8057 * full stripe writes.
8063 status = btrfs_submit_dio_bio(bio, inode, file_offset,
8068 refcount_dec(&dip->refs);
8072 dio_data->submitted += clone_len;
8073 clone_offset += clone_len;
8074 start_sector += clone_len >> 9;
8075 file_offset += clone_len;
8077 free_extent_map(em);
8078 } while (submit_len > 0);
8082 free_extent_map(em);
8084 dip->dio_bio->bi_status = status;
8085 btrfs_dio_private_put(dip);
8088 const struct iomap_ops btrfs_dio_iomap_ops = {
8089 .iomap_begin = btrfs_dio_iomap_begin,
8090 .iomap_end = btrfs_dio_iomap_end,
8093 const struct iomap_dio_ops btrfs_dio_ops = {
8094 .submit_io = btrfs_submit_direct,
8097 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8102 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
8106 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
8109 int btrfs_readpage(struct file *file, struct page *page)
8111 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
8112 u64 start = page_offset(page);
8113 u64 end = start + PAGE_SIZE - 1;
8114 struct btrfs_bio_ctrl bio_ctrl = { 0 };
8117 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
8119 ret = btrfs_do_readpage(page, NULL, &bio_ctrl, 0, NULL);
8123 ret2 = submit_one_bio(bio_ctrl.bio, 0, bio_ctrl.bio_flags);
8130 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8132 struct inode *inode = page->mapping->host;
8135 if (current->flags & PF_MEMALLOC) {
8136 redirty_page_for_writepage(wbc, page);
8142 * If we are under memory pressure we will call this directly from the
8143 * VM, we need to make sure we have the inode referenced for the ordered
8144 * extent. If not just return like we didn't do anything.
8146 if (!igrab(inode)) {
8147 redirty_page_for_writepage(wbc, page);
8148 return AOP_WRITEPAGE_ACTIVATE;
8150 ret = extent_write_full_page(page, wbc);
8151 btrfs_add_delayed_iput(inode);
8155 static int btrfs_writepages(struct address_space *mapping,
8156 struct writeback_control *wbc)
8158 return extent_writepages(mapping, wbc);
8161 static void btrfs_readahead(struct readahead_control *rac)
8163 extent_readahead(rac);
8167 * For releasepage() and invalidatepage() we have a race window where
8168 * end_page_writeback() is called but the subpage spinlock is not yet released.
8169 * If we continue to release/invalidate the page, we could cause use-after-free
8170 * for subpage spinlock. So this function is to spin and wait for subpage
8173 static void wait_subpage_spinlock(struct page *page)
8175 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
8176 struct btrfs_subpage *subpage;
8178 if (fs_info->sectorsize == PAGE_SIZE)
8181 ASSERT(PagePrivate(page) && page->private);
8182 subpage = (struct btrfs_subpage *)page->private;
8185 * This may look insane as we just acquire the spinlock and release it,
8186 * without doing anything. But we just want to make sure no one is
8187 * still holding the subpage spinlock.
8188 * And since the page is not dirty nor writeback, and we have page
8189 * locked, the only possible way to hold a spinlock is from the endio
8190 * function to clear page writeback.
8192 * Here we just acquire the spinlock so that all existing callers
8193 * should exit and we're safe to release/invalidate the page.
8195 spin_lock_irq(&subpage->lock);
8196 spin_unlock_irq(&subpage->lock);
8199 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8201 int ret = try_release_extent_mapping(page, gfp_flags);
8204 wait_subpage_spinlock(page);
8205 clear_page_extent_mapped(page);
8210 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8212 if (PageWriteback(page) || PageDirty(page))
8214 return __btrfs_releasepage(page, gfp_flags);
8217 #ifdef CONFIG_MIGRATION
8218 static int btrfs_migratepage(struct address_space *mapping,
8219 struct page *newpage, struct page *page,
8220 enum migrate_mode mode)
8224 ret = migrate_page_move_mapping(mapping, newpage, page, 0);
8225 if (ret != MIGRATEPAGE_SUCCESS)
8228 if (page_has_private(page))
8229 attach_page_private(newpage, detach_page_private(page));
8231 if (PageOrdered(page)) {
8232 ClearPageOrdered(page);
8233 SetPageOrdered(newpage);
8236 if (mode != MIGRATE_SYNC_NO_COPY)
8237 migrate_page_copy(newpage, page);
8239 migrate_page_states(newpage, page);
8240 return MIGRATEPAGE_SUCCESS;
8244 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8245 unsigned int length)
8247 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
8248 struct btrfs_fs_info *fs_info = inode->root->fs_info;
8249 struct extent_io_tree *tree = &inode->io_tree;
8250 struct extent_state *cached_state = NULL;
8251 u64 page_start = page_offset(page);
8252 u64 page_end = page_start + PAGE_SIZE - 1;
8254 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
8257 * We have page locked so no new ordered extent can be created on this
8258 * page, nor bio can be submitted for this page.
8260 * But already submitted bio can still be finished on this page.
8261 * Furthermore, endio function won't skip page which has Ordered
8262 * (Private2) already cleared, so it's possible for endio and
8263 * invalidatepage to do the same ordered extent accounting twice
8266 * So here we wait for any submitted bios to finish, so that we won't
8267 * do double ordered extent accounting on the same page.
8269 wait_on_page_writeback(page);
8270 wait_subpage_spinlock(page);
8273 * For subpage case, we have call sites like
8274 * btrfs_punch_hole_lock_range() which passes range not aligned to
8276 * If the range doesn't cover the full page, we don't need to and
8277 * shouldn't clear page extent mapped, as page->private can still
8278 * record subpage dirty bits for other part of the range.
8280 * For cases that can invalidate the full even the range doesn't
8281 * cover the full page, like invalidating the last page, we're
8282 * still safe to wait for ordered extent to finish.
8284 if (!(offset == 0 && length == PAGE_SIZE)) {
8285 btrfs_releasepage(page, GFP_NOFS);
8289 if (!inode_evicting)
8290 lock_extent_bits(tree, page_start, page_end, &cached_state);
8293 while (cur < page_end) {
8294 struct btrfs_ordered_extent *ordered;
8299 ordered = btrfs_lookup_first_ordered_range(inode, cur,
8300 page_end + 1 - cur);
8302 range_end = page_end;
8304 * No ordered extent covering this range, we are safe
8305 * to delete all extent states in the range.
8307 delete_states = true;
8310 if (ordered->file_offset > cur) {
8312 * There is a range between [cur, oe->file_offset) not
8313 * covered by any ordered extent.
8314 * We are safe to delete all extent states, and handle
8315 * the ordered extent in the next iteration.
8317 range_end = ordered->file_offset - 1;
8318 delete_states = true;
8322 range_end = min(ordered->file_offset + ordered->num_bytes - 1,
8324 ASSERT(range_end + 1 - cur < U32_MAX);
8325 range_len = range_end + 1 - cur;
8326 if (!btrfs_page_test_ordered(fs_info, page, cur, range_len)) {
8328 * If Ordered (Private2) is cleared, it means endio has
8329 * already been executed for the range.
8330 * We can't delete the extent states as
8331 * btrfs_finish_ordered_io() may still use some of them.
8333 delete_states = false;
8336 btrfs_page_clear_ordered(fs_info, page, cur, range_len);
8339 * IO on this page will never be started, so we need to account
8340 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8341 * here, must leave that up for the ordered extent completion.
8343 * This will also unlock the range for incoming
8344 * btrfs_finish_ordered_io().
8346 if (!inode_evicting)
8347 clear_extent_bit(tree, cur, range_end,
8349 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8350 EXTENT_DEFRAG, 1, 0, &cached_state);
8352 spin_lock_irq(&inode->ordered_tree.lock);
8353 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8354 ordered->truncated_len = min(ordered->truncated_len,
8355 cur - ordered->file_offset);
8356 spin_unlock_irq(&inode->ordered_tree.lock);
8358 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8359 cur, range_end + 1 - cur)) {
8360 btrfs_finish_ordered_io(ordered);
8362 * The ordered extent has finished, now we're again
8363 * safe to delete all extent states of the range.
8365 delete_states = true;
8368 * btrfs_finish_ordered_io() will get executed by endio
8369 * of other pages, thus we can't delete extent states
8372 delete_states = false;
8376 btrfs_put_ordered_extent(ordered);
8378 * Qgroup reserved space handler
8379 * Sector(s) here will be either:
8381 * 1) Already written to disk or bio already finished
8382 * Then its QGROUP_RESERVED bit in io_tree is already cleared.
8383 * Qgroup will be handled by its qgroup_record then.
8384 * btrfs_qgroup_free_data() call will do nothing here.
8386 * 2) Not written to disk yet
8387 * Then btrfs_qgroup_free_data() call will clear the
8388 * QGROUP_RESERVED bit of its io_tree, and free the qgroup
8389 * reserved data space.
8390 * Since the IO will never happen for this page.
8392 btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur);
8393 if (!inode_evicting) {
8394 clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED |
8395 EXTENT_DELALLOC | EXTENT_UPTODATE |
8396 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1,
8397 delete_states, &cached_state);
8399 cur = range_end + 1;
8402 * We have iterated through all ordered extents of the page, the page
8403 * should not have Ordered (Private2) anymore, or the above iteration
8404 * did something wrong.
8406 ASSERT(!PageOrdered(page));
8407 btrfs_page_clear_checked(fs_info, page, page_offset(page), PAGE_SIZE);
8408 if (!inode_evicting)
8409 __btrfs_releasepage(page, GFP_NOFS);
8410 clear_page_extent_mapped(page);
8414 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8415 * called from a page fault handler when a page is first dirtied. Hence we must
8416 * be careful to check for EOF conditions here. We set the page up correctly
8417 * for a written page which means we get ENOSPC checking when writing into
8418 * holes and correct delalloc and unwritten extent mapping on filesystems that
8419 * support these features.
8421 * We are not allowed to take the i_mutex here so we have to play games to
8422 * protect against truncate races as the page could now be beyond EOF. Because
8423 * truncate_setsize() writes the inode size before removing pages, once we have
8424 * the page lock we can determine safely if the page is beyond EOF. If it is not
8425 * beyond EOF, then the page is guaranteed safe against truncation until we
8428 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8430 struct page *page = vmf->page;
8431 struct inode *inode = file_inode(vmf->vma->vm_file);
8432 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8433 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8434 struct btrfs_ordered_extent *ordered;
8435 struct extent_state *cached_state = NULL;
8436 struct extent_changeset *data_reserved = NULL;
8437 unsigned long zero_start;
8447 reserved_space = PAGE_SIZE;
8449 sb_start_pagefault(inode->i_sb);
8450 page_start = page_offset(page);
8451 page_end = page_start + PAGE_SIZE - 1;
8455 * Reserving delalloc space after obtaining the page lock can lead to
8456 * deadlock. For example, if a dirty page is locked by this function
8457 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8458 * dirty page write out, then the btrfs_writepage() function could
8459 * end up waiting indefinitely to get a lock on the page currently
8460 * being processed by btrfs_page_mkwrite() function.
8462 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8463 page_start, reserved_space);
8465 ret2 = file_update_time(vmf->vma->vm_file);
8469 ret = vmf_error(ret2);
8475 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8477 down_read(&BTRFS_I(inode)->i_mmap_lock);
8479 size = i_size_read(inode);
8481 if ((page->mapping != inode->i_mapping) ||
8482 (page_start >= size)) {
8483 /* page got truncated out from underneath us */
8486 wait_on_page_writeback(page);
8488 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8489 ret2 = set_page_extent_mapped(page);
8491 ret = vmf_error(ret2);
8492 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8497 * we can't set the delalloc bits if there are pending ordered
8498 * extents. Drop our locks and wait for them to finish
8500 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8503 unlock_extent_cached(io_tree, page_start, page_end,
8506 up_read(&BTRFS_I(inode)->i_mmap_lock);
8507 btrfs_start_ordered_extent(ordered, 1);
8508 btrfs_put_ordered_extent(ordered);
8512 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8513 reserved_space = round_up(size - page_start,
8514 fs_info->sectorsize);
8515 if (reserved_space < PAGE_SIZE) {
8516 end = page_start + reserved_space - 1;
8517 btrfs_delalloc_release_space(BTRFS_I(inode),
8518 data_reserved, page_start,
8519 PAGE_SIZE - reserved_space, true);
8524 * page_mkwrite gets called when the page is firstly dirtied after it's
8525 * faulted in, but write(2) could also dirty a page and set delalloc
8526 * bits, thus in this case for space account reason, we still need to
8527 * clear any delalloc bits within this page range since we have to
8528 * reserve data&meta space before lock_page() (see above comments).
8530 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8531 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8532 EXTENT_DEFRAG, 0, 0, &cached_state);
8534 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8537 unlock_extent_cached(io_tree, page_start, page_end,
8539 ret = VM_FAULT_SIGBUS;
8543 /* page is wholly or partially inside EOF */
8544 if (page_start + PAGE_SIZE > size)
8545 zero_start = offset_in_page(size);
8547 zero_start = PAGE_SIZE;
8549 if (zero_start != PAGE_SIZE) {
8550 memzero_page(page, zero_start, PAGE_SIZE - zero_start);
8551 flush_dcache_page(page);
8553 btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE);
8554 btrfs_page_set_dirty(fs_info, page, page_start, end + 1 - page_start);
8555 btrfs_page_set_uptodate(fs_info, page, page_start, end + 1 - page_start);
8557 btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
8559 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8560 up_read(&BTRFS_I(inode)->i_mmap_lock);
8562 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8563 sb_end_pagefault(inode->i_sb);
8564 extent_changeset_free(data_reserved);
8565 return VM_FAULT_LOCKED;
8569 up_read(&BTRFS_I(inode)->i_mmap_lock);
8571 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8572 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8573 reserved_space, (ret != 0));
8575 sb_end_pagefault(inode->i_sb);
8576 extent_changeset_free(data_reserved);
8580 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
8582 struct btrfs_truncate_control control = {
8583 .inode = BTRFS_I(inode),
8584 .ino = btrfs_ino(BTRFS_I(inode)),
8585 .min_type = BTRFS_EXTENT_DATA_KEY,
8586 .clear_extent_range = true,
8588 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8589 struct btrfs_root *root = BTRFS_I(inode)->root;
8590 struct btrfs_block_rsv *rsv;
8592 struct btrfs_trans_handle *trans;
8593 u64 mask = fs_info->sectorsize - 1;
8594 u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8596 if (!skip_writeback) {
8597 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8604 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8605 * things going on here:
8607 * 1) We need to reserve space to update our inode.
8609 * 2) We need to have something to cache all the space that is going to
8610 * be free'd up by the truncate operation, but also have some slack
8611 * space reserved in case it uses space during the truncate (thank you
8612 * very much snapshotting).
8614 * And we need these to be separate. The fact is we can use a lot of
8615 * space doing the truncate, and we have no earthly idea how much space
8616 * we will use, so we need the truncate reservation to be separate so it
8617 * doesn't end up using space reserved for updating the inode. We also
8618 * need to be able to stop the transaction and start a new one, which
8619 * means we need to be able to update the inode several times, and we
8620 * have no idea of knowing how many times that will be, so we can't just
8621 * reserve 1 item for the entirety of the operation, so that has to be
8622 * done separately as well.
8624 * So that leaves us with
8626 * 1) rsv - for the truncate reservation, which we will steal from the
8627 * transaction reservation.
8628 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8629 * updating the inode.
8631 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8634 rsv->size = min_size;
8638 * 1 for the truncate slack space
8639 * 1 for updating the inode.
8641 trans = btrfs_start_transaction(root, 2);
8642 if (IS_ERR(trans)) {
8643 ret = PTR_ERR(trans);
8647 /* Migrate the slack space for the truncate to our reserve */
8648 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8652 trans->block_rsv = rsv;
8655 struct extent_state *cached_state = NULL;
8656 const u64 new_size = inode->i_size;
8657 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
8659 control.new_size = new_size;
8660 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, (u64)-1,
8663 * We want to drop from the next block forward in case this new
8664 * size is not block aligned since we will be keeping the last
8665 * block of the extent just the way it is.
8667 btrfs_drop_extent_cache(BTRFS_I(inode),
8668 ALIGN(new_size, fs_info->sectorsize),
8671 ret = btrfs_truncate_inode_items(trans, root, &control);
8673 inode_sub_bytes(inode, control.sub_bytes);
8674 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), control.last_size);
8676 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start,
8677 (u64)-1, &cached_state);
8679 trans->block_rsv = &fs_info->trans_block_rsv;
8680 if (ret != -ENOSPC && ret != -EAGAIN)
8683 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
8687 btrfs_end_transaction(trans);
8688 btrfs_btree_balance_dirty(fs_info);
8690 trans = btrfs_start_transaction(root, 2);
8691 if (IS_ERR(trans)) {
8692 ret = PTR_ERR(trans);
8697 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8698 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8699 rsv, min_size, false);
8700 BUG_ON(ret); /* shouldn't happen */
8701 trans->block_rsv = rsv;
8705 * We can't call btrfs_truncate_block inside a trans handle as we could
8706 * deadlock with freeze, if we got BTRFS_NEED_TRUNCATE_BLOCK then we
8707 * know we've truncated everything except the last little bit, and can
8708 * do btrfs_truncate_block and then update the disk_i_size.
8710 if (ret == BTRFS_NEED_TRUNCATE_BLOCK) {
8711 btrfs_end_transaction(trans);
8712 btrfs_btree_balance_dirty(fs_info);
8714 ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0);
8717 trans = btrfs_start_transaction(root, 1);
8718 if (IS_ERR(trans)) {
8719 ret = PTR_ERR(trans);
8722 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
8728 trans->block_rsv = &fs_info->trans_block_rsv;
8729 ret2 = btrfs_update_inode(trans, root, BTRFS_I(inode));
8733 ret2 = btrfs_end_transaction(trans);
8736 btrfs_btree_balance_dirty(fs_info);
8739 btrfs_free_block_rsv(fs_info, rsv);
8741 * So if we truncate and then write and fsync we normally would just
8742 * write the extents that changed, which is a problem if we need to
8743 * first truncate that entire inode. So set this flag so we write out
8744 * all of the extents in the inode to the sync log so we're completely
8747 * If no extents were dropped or trimmed we don't need to force the next
8748 * fsync to truncate all the inode's items from the log and re-log them
8749 * all. This means the truncate operation did not change the file size,
8750 * or changed it to a smaller size but there was only an implicit hole
8751 * between the old i_size and the new i_size, and there were no prealloc
8752 * extents beyond i_size to drop.
8754 if (control.extents_found > 0)
8755 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
8761 * create a new subvolume directory/inode (helper for the ioctl).
8763 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8764 struct btrfs_root *new_root,
8765 struct btrfs_root *parent_root,
8766 struct user_namespace *mnt_userns)
8768 struct inode *inode;
8773 err = btrfs_get_free_objectid(new_root, &ino);
8777 inode = btrfs_new_inode(trans, new_root, mnt_userns, NULL, "..", 2,
8779 S_IFDIR | (~current_umask() & S_IRWXUGO),
8782 return PTR_ERR(inode);
8783 inode->i_op = &btrfs_dir_inode_operations;
8784 inode->i_fop = &btrfs_dir_file_operations;
8786 set_nlink(inode, 1);
8787 btrfs_i_size_write(BTRFS_I(inode), 0);
8788 unlock_new_inode(inode);
8790 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8792 btrfs_err(new_root->fs_info,
8793 "error inheriting subvolume %llu properties: %d",
8794 new_root->root_key.objectid, err);
8796 err = btrfs_update_inode(trans, new_root, BTRFS_I(inode));
8802 struct inode *btrfs_alloc_inode(struct super_block *sb)
8804 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8805 struct btrfs_inode *ei;
8806 struct inode *inode;
8808 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL);
8815 ei->last_sub_trans = 0;
8816 ei->logged_trans = 0;
8817 ei->delalloc_bytes = 0;
8818 ei->new_delalloc_bytes = 0;
8819 ei->defrag_bytes = 0;
8820 ei->disk_i_size = 0;
8824 ei->index_cnt = (u64)-1;
8826 ei->last_unlink_trans = 0;
8827 ei->last_reflink_trans = 0;
8828 ei->last_log_commit = 0;
8830 spin_lock_init(&ei->lock);
8831 ei->outstanding_extents = 0;
8832 if (sb->s_magic != BTRFS_TEST_MAGIC)
8833 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8834 BTRFS_BLOCK_RSV_DELALLOC);
8835 ei->runtime_flags = 0;
8836 ei->prop_compress = BTRFS_COMPRESS_NONE;
8837 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8839 ei->delayed_node = NULL;
8841 ei->i_otime.tv_sec = 0;
8842 ei->i_otime.tv_nsec = 0;
8844 inode = &ei->vfs_inode;
8845 extent_map_tree_init(&ei->extent_tree);
8846 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO, inode);
8847 extent_io_tree_init(fs_info, &ei->io_failure_tree,
8848 IO_TREE_INODE_IO_FAILURE, inode);
8849 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8850 IO_TREE_INODE_FILE_EXTENT, inode);
8851 ei->io_tree.track_uptodate = true;
8852 ei->io_failure_tree.track_uptodate = true;
8853 atomic_set(&ei->sync_writers, 0);
8854 mutex_init(&ei->log_mutex);
8855 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8856 INIT_LIST_HEAD(&ei->delalloc_inodes);
8857 INIT_LIST_HEAD(&ei->delayed_iput);
8858 RB_CLEAR_NODE(&ei->rb_node);
8859 init_rwsem(&ei->i_mmap_lock);
8864 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8865 void btrfs_test_destroy_inode(struct inode *inode)
8867 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
8868 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8872 void btrfs_free_inode(struct inode *inode)
8874 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8877 void btrfs_destroy_inode(struct inode *vfs_inode)
8879 struct btrfs_ordered_extent *ordered;
8880 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8881 struct btrfs_root *root = inode->root;
8883 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8884 WARN_ON(vfs_inode->i_data.nrpages);
8885 WARN_ON(inode->block_rsv.reserved);
8886 WARN_ON(inode->block_rsv.size);
8887 WARN_ON(inode->outstanding_extents);
8888 if (!S_ISDIR(vfs_inode->i_mode)) {
8889 WARN_ON(inode->delalloc_bytes);
8890 WARN_ON(inode->new_delalloc_bytes);
8892 WARN_ON(inode->csum_bytes);
8893 WARN_ON(inode->defrag_bytes);
8896 * This can happen where we create an inode, but somebody else also
8897 * created the same inode and we need to destroy the one we already
8904 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8908 btrfs_err(root->fs_info,
8909 "found ordered extent %llu %llu on inode cleanup",
8910 ordered->file_offset, ordered->num_bytes);
8911 btrfs_remove_ordered_extent(inode, ordered);
8912 btrfs_put_ordered_extent(ordered);
8913 btrfs_put_ordered_extent(ordered);
8916 btrfs_qgroup_check_reserved_leak(inode);
8917 inode_tree_del(inode);
8918 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8919 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
8920 btrfs_put_root(inode->root);
8923 int btrfs_drop_inode(struct inode *inode)
8925 struct btrfs_root *root = BTRFS_I(inode)->root;
8930 /* the snap/subvol tree is on deleting */
8931 if (btrfs_root_refs(&root->root_item) == 0)
8934 return generic_drop_inode(inode);
8937 static void init_once(void *foo)
8939 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8941 inode_init_once(&ei->vfs_inode);
8944 void __cold btrfs_destroy_cachep(void)
8947 * Make sure all delayed rcu free inodes are flushed before we
8951 kmem_cache_destroy(btrfs_inode_cachep);
8952 kmem_cache_destroy(btrfs_trans_handle_cachep);
8953 kmem_cache_destroy(btrfs_path_cachep);
8954 kmem_cache_destroy(btrfs_free_space_cachep);
8955 kmem_cache_destroy(btrfs_free_space_bitmap_cachep);
8958 int __init btrfs_init_cachep(void)
8960 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8961 sizeof(struct btrfs_inode), 0,
8962 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
8964 if (!btrfs_inode_cachep)
8967 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8968 sizeof(struct btrfs_trans_handle), 0,
8969 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
8970 if (!btrfs_trans_handle_cachep)
8973 btrfs_path_cachep = kmem_cache_create("btrfs_path",
8974 sizeof(struct btrfs_path), 0,
8975 SLAB_MEM_SPREAD, NULL);
8976 if (!btrfs_path_cachep)
8979 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8980 sizeof(struct btrfs_free_space), 0,
8981 SLAB_MEM_SPREAD, NULL);
8982 if (!btrfs_free_space_cachep)
8985 btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap",
8986 PAGE_SIZE, PAGE_SIZE,
8987 SLAB_MEM_SPREAD, NULL);
8988 if (!btrfs_free_space_bitmap_cachep)
8993 btrfs_destroy_cachep();
8997 static int btrfs_getattr(struct user_namespace *mnt_userns,
8998 const struct path *path, struct kstat *stat,
8999 u32 request_mask, unsigned int flags)
9003 struct inode *inode = d_inode(path->dentry);
9004 u32 blocksize = inode->i_sb->s_blocksize;
9005 u32 bi_flags = BTRFS_I(inode)->flags;
9006 u32 bi_ro_flags = BTRFS_I(inode)->ro_flags;
9008 stat->result_mask |= STATX_BTIME;
9009 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9010 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9011 if (bi_flags & BTRFS_INODE_APPEND)
9012 stat->attributes |= STATX_ATTR_APPEND;
9013 if (bi_flags & BTRFS_INODE_COMPRESS)
9014 stat->attributes |= STATX_ATTR_COMPRESSED;
9015 if (bi_flags & BTRFS_INODE_IMMUTABLE)
9016 stat->attributes |= STATX_ATTR_IMMUTABLE;
9017 if (bi_flags & BTRFS_INODE_NODUMP)
9018 stat->attributes |= STATX_ATTR_NODUMP;
9019 if (bi_ro_flags & BTRFS_INODE_RO_VERITY)
9020 stat->attributes |= STATX_ATTR_VERITY;
9022 stat->attributes_mask |= (STATX_ATTR_APPEND |
9023 STATX_ATTR_COMPRESSED |
9024 STATX_ATTR_IMMUTABLE |
9027 generic_fillattr(mnt_userns, inode, stat);
9028 stat->dev = BTRFS_I(inode)->root->anon_dev;
9030 spin_lock(&BTRFS_I(inode)->lock);
9031 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9032 inode_bytes = inode_get_bytes(inode);
9033 spin_unlock(&BTRFS_I(inode)->lock);
9034 stat->blocks = (ALIGN(inode_bytes, blocksize) +
9035 ALIGN(delalloc_bytes, blocksize)) >> 9;
9039 static int btrfs_rename_exchange(struct inode *old_dir,
9040 struct dentry *old_dentry,
9041 struct inode *new_dir,
9042 struct dentry *new_dentry)
9044 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9045 struct btrfs_trans_handle *trans;
9046 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9047 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9048 struct inode *new_inode = new_dentry->d_inode;
9049 struct inode *old_inode = old_dentry->d_inode;
9050 struct timespec64 ctime = current_time(old_inode);
9051 struct btrfs_rename_ctx old_rename_ctx;
9052 struct btrfs_rename_ctx new_rename_ctx;
9053 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9054 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9059 bool need_abort = false;
9062 * For non-subvolumes allow exchange only within one subvolume, in the
9063 * same inode namespace. Two subvolumes (represented as directory) can
9064 * be exchanged as they're a logical link and have a fixed inode number.
9067 (old_ino != BTRFS_FIRST_FREE_OBJECTID ||
9068 new_ino != BTRFS_FIRST_FREE_OBJECTID))
9071 /* close the race window with snapshot create/destroy ioctl */
9072 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
9073 new_ino == BTRFS_FIRST_FREE_OBJECTID)
9074 down_read(&fs_info->subvol_sem);
9077 * We want to reserve the absolute worst case amount of items. So if
9078 * both inodes are subvols and we need to unlink them then that would
9079 * require 4 item modifications, but if they are both normal inodes it
9080 * would require 5 item modifications, so we'll assume their normal
9081 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9082 * should cover the worst case number of items we'll modify.
9084 trans = btrfs_start_transaction(root, 12);
9085 if (IS_ERR(trans)) {
9086 ret = PTR_ERR(trans);
9091 ret = btrfs_record_root_in_trans(trans, dest);
9097 * We need to find a free sequence number both in the source and
9098 * in the destination directory for the exchange.
9100 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9103 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9107 BTRFS_I(old_inode)->dir_index = 0ULL;
9108 BTRFS_I(new_inode)->dir_index = 0ULL;
9110 /* Reference for the source. */
9111 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9112 /* force full log commit if subvolume involved. */
9113 btrfs_set_log_full_commit(trans);
9115 ret = btrfs_insert_inode_ref(trans, dest,
9116 new_dentry->d_name.name,
9117 new_dentry->d_name.len,
9119 btrfs_ino(BTRFS_I(new_dir)),
9126 /* And now for the dest. */
9127 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9128 /* force full log commit if subvolume involved. */
9129 btrfs_set_log_full_commit(trans);
9131 ret = btrfs_insert_inode_ref(trans, root,
9132 old_dentry->d_name.name,
9133 old_dentry->d_name.len,
9135 btrfs_ino(BTRFS_I(old_dir)),
9139 btrfs_abort_transaction(trans, ret);
9144 /* Update inode version and ctime/mtime. */
9145 inode_inc_iversion(old_dir);
9146 inode_inc_iversion(new_dir);
9147 inode_inc_iversion(old_inode);
9148 inode_inc_iversion(new_inode);
9149 old_dir->i_ctime = old_dir->i_mtime = ctime;
9150 new_dir->i_ctime = new_dir->i_mtime = ctime;
9151 old_inode->i_ctime = ctime;
9152 new_inode->i_ctime = ctime;
9154 if (old_dentry->d_parent != new_dentry->d_parent) {
9155 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9156 BTRFS_I(old_inode), 1);
9157 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9158 BTRFS_I(new_inode), 1);
9161 /* src is a subvolume */
9162 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9163 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9164 } else { /* src is an inode */
9165 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9166 BTRFS_I(old_dentry->d_inode),
9167 old_dentry->d_name.name,
9168 old_dentry->d_name.len,
9171 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9174 btrfs_abort_transaction(trans, ret);
9178 /* dest is a subvolume */
9179 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9180 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9181 } else { /* dest is an inode */
9182 ret = __btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9183 BTRFS_I(new_dentry->d_inode),
9184 new_dentry->d_name.name,
9185 new_dentry->d_name.len,
9188 ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode));
9191 btrfs_abort_transaction(trans, ret);
9195 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9196 new_dentry->d_name.name,
9197 new_dentry->d_name.len, 0, old_idx);
9199 btrfs_abort_transaction(trans, ret);
9203 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9204 old_dentry->d_name.name,
9205 old_dentry->d_name.len, 0, new_idx);
9207 btrfs_abort_transaction(trans, ret);
9211 if (old_inode->i_nlink == 1)
9212 BTRFS_I(old_inode)->dir_index = old_idx;
9213 if (new_inode->i_nlink == 1)
9214 BTRFS_I(new_inode)->dir_index = new_idx;
9217 * Now pin the logs of the roots. We do it to ensure that no other task
9218 * can sync the logs while we are in progress with the rename, because
9219 * that could result in an inconsistency in case any of the inodes that
9220 * are part of this rename operation were logged before.
9222 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9223 btrfs_pin_log_trans(root);
9224 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9225 btrfs_pin_log_trans(dest);
9227 /* Do the log updates for all inodes. */
9228 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9229 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9230 old_rename_ctx.index, new_dentry->d_parent);
9231 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9232 btrfs_log_new_name(trans, new_dentry, BTRFS_I(new_dir),
9233 new_rename_ctx.index, old_dentry->d_parent);
9235 /* Now unpin the logs. */
9236 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9237 btrfs_end_log_trans(root);
9238 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9239 btrfs_end_log_trans(dest);
9241 ret2 = btrfs_end_transaction(trans);
9242 ret = ret ? ret : ret2;
9244 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
9245 old_ino == BTRFS_FIRST_FREE_OBJECTID)
9246 up_read(&fs_info->subvol_sem);
9251 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9252 struct btrfs_root *root,
9253 struct user_namespace *mnt_userns,
9255 struct dentry *dentry)
9258 struct inode *inode;
9262 ret = btrfs_get_free_objectid(root, &objectid);
9266 inode = btrfs_new_inode(trans, root, mnt_userns, dir,
9267 dentry->d_name.name,
9269 btrfs_ino(BTRFS_I(dir)),
9271 S_IFCHR | WHITEOUT_MODE,
9274 if (IS_ERR(inode)) {
9275 ret = PTR_ERR(inode);
9279 inode->i_op = &btrfs_special_inode_operations;
9280 init_special_inode(inode, inode->i_mode,
9283 ret = btrfs_init_inode_security(trans, inode, dir,
9288 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9289 BTRFS_I(inode), 0, index);
9293 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
9295 unlock_new_inode(inode);
9297 inode_dec_link_count(inode);
9303 static int btrfs_rename(struct user_namespace *mnt_userns,
9304 struct inode *old_dir, struct dentry *old_dentry,
9305 struct inode *new_dir, struct dentry *new_dentry,
9308 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9309 struct btrfs_trans_handle *trans;
9310 unsigned int trans_num_items;
9311 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9312 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9313 struct inode *new_inode = d_inode(new_dentry);
9314 struct inode *old_inode = d_inode(old_dentry);
9315 struct btrfs_rename_ctx rename_ctx;
9319 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9321 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9324 /* we only allow rename subvolume link between subvolumes */
9325 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9328 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9329 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9332 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9333 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9337 /* check for collisions, even if the name isn't there */
9338 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9339 new_dentry->d_name.name,
9340 new_dentry->d_name.len);
9343 if (ret == -EEXIST) {
9345 * eexist without a new_inode */
9346 if (WARN_ON(!new_inode)) {
9350 /* maybe -EOVERFLOW */
9357 * we're using rename to replace one file with another. Start IO on it
9358 * now so we don't add too much work to the end of the transaction
9360 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9361 filemap_flush(old_inode->i_mapping);
9363 /* close the racy window with snapshot create/destroy ioctl */
9364 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9365 down_read(&fs_info->subvol_sem);
9367 * We want to reserve the absolute worst case amount of items. So if
9368 * both inodes are subvols and we need to unlink them then that would
9369 * require 4 item modifications, but if they are both normal inodes it
9370 * would require 5 item modifications, so we'll assume they are normal
9371 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9372 * should cover the worst case number of items we'll modify.
9373 * If our rename has the whiteout flag, we need more 5 units for the
9374 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9375 * when selinux is enabled).
9377 trans_num_items = 11;
9378 if (flags & RENAME_WHITEOUT)
9379 trans_num_items += 5;
9380 trans = btrfs_start_transaction(root, trans_num_items);
9381 if (IS_ERR(trans)) {
9382 ret = PTR_ERR(trans);
9387 ret = btrfs_record_root_in_trans(trans, dest);
9392 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9396 BTRFS_I(old_inode)->dir_index = 0ULL;
9397 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9398 /* force full log commit if subvolume involved. */
9399 btrfs_set_log_full_commit(trans);
9401 ret = btrfs_insert_inode_ref(trans, dest,
9402 new_dentry->d_name.name,
9403 new_dentry->d_name.len,
9405 btrfs_ino(BTRFS_I(new_dir)), index);
9410 inode_inc_iversion(old_dir);
9411 inode_inc_iversion(new_dir);
9412 inode_inc_iversion(old_inode);
9413 old_dir->i_ctime = old_dir->i_mtime =
9414 new_dir->i_ctime = new_dir->i_mtime =
9415 old_inode->i_ctime = current_time(old_dir);
9417 if (old_dentry->d_parent != new_dentry->d_parent)
9418 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9419 BTRFS_I(old_inode), 1);
9421 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9422 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9424 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9425 BTRFS_I(d_inode(old_dentry)),
9426 old_dentry->d_name.name,
9427 old_dentry->d_name.len,
9430 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9433 btrfs_abort_transaction(trans, ret);
9438 inode_inc_iversion(new_inode);
9439 new_inode->i_ctime = current_time(new_inode);
9440 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9441 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9442 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9443 BUG_ON(new_inode->i_nlink == 0);
9445 ret = btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9446 BTRFS_I(d_inode(new_dentry)),
9447 new_dentry->d_name.name,
9448 new_dentry->d_name.len);
9450 if (!ret && new_inode->i_nlink == 0)
9451 ret = btrfs_orphan_add(trans,
9452 BTRFS_I(d_inode(new_dentry)));
9454 btrfs_abort_transaction(trans, ret);
9459 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9460 new_dentry->d_name.name,
9461 new_dentry->d_name.len, 0, index);
9463 btrfs_abort_transaction(trans, ret);
9467 if (old_inode->i_nlink == 1)
9468 BTRFS_I(old_inode)->dir_index = index;
9470 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9471 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9472 rename_ctx.index, new_dentry->d_parent);
9474 if (flags & RENAME_WHITEOUT) {
9475 ret = btrfs_whiteout_for_rename(trans, root, mnt_userns,
9476 old_dir, old_dentry);
9479 btrfs_abort_transaction(trans, ret);
9484 ret2 = btrfs_end_transaction(trans);
9485 ret = ret ? ret : ret2;
9487 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9488 up_read(&fs_info->subvol_sem);
9493 static int btrfs_rename2(struct user_namespace *mnt_userns, struct inode *old_dir,
9494 struct dentry *old_dentry, struct inode *new_dir,
9495 struct dentry *new_dentry, unsigned int flags)
9497 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9500 if (flags & RENAME_EXCHANGE)
9501 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9504 return btrfs_rename(mnt_userns, old_dir, old_dentry, new_dir,
9508 struct btrfs_delalloc_work {
9509 struct inode *inode;
9510 struct completion completion;
9511 struct list_head list;
9512 struct btrfs_work work;
9515 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9517 struct btrfs_delalloc_work *delalloc_work;
9518 struct inode *inode;
9520 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9522 inode = delalloc_work->inode;
9523 filemap_flush(inode->i_mapping);
9524 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9525 &BTRFS_I(inode)->runtime_flags))
9526 filemap_flush(inode->i_mapping);
9529 complete(&delalloc_work->completion);
9532 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9534 struct btrfs_delalloc_work *work;
9536 work = kmalloc(sizeof(*work), GFP_NOFS);
9540 init_completion(&work->completion);
9541 INIT_LIST_HEAD(&work->list);
9542 work->inode = inode;
9543 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9549 * some fairly slow code that needs optimization. This walks the list
9550 * of all the inodes with pending delalloc and forces them to disk.
9552 static int start_delalloc_inodes(struct btrfs_root *root,
9553 struct writeback_control *wbc, bool snapshot,
9554 bool in_reclaim_context)
9556 struct btrfs_inode *binode;
9557 struct inode *inode;
9558 struct btrfs_delalloc_work *work, *next;
9559 struct list_head works;
9560 struct list_head splice;
9562 bool full_flush = wbc->nr_to_write == LONG_MAX;
9564 INIT_LIST_HEAD(&works);
9565 INIT_LIST_HEAD(&splice);
9567 mutex_lock(&root->delalloc_mutex);
9568 spin_lock(&root->delalloc_lock);
9569 list_splice_init(&root->delalloc_inodes, &splice);
9570 while (!list_empty(&splice)) {
9571 binode = list_entry(splice.next, struct btrfs_inode,
9574 list_move_tail(&binode->delalloc_inodes,
9575 &root->delalloc_inodes);
9577 if (in_reclaim_context &&
9578 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9581 inode = igrab(&binode->vfs_inode);
9583 cond_resched_lock(&root->delalloc_lock);
9586 spin_unlock(&root->delalloc_lock);
9589 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9590 &binode->runtime_flags);
9592 work = btrfs_alloc_delalloc_work(inode);
9598 list_add_tail(&work->list, &works);
9599 btrfs_queue_work(root->fs_info->flush_workers,
9602 ret = filemap_fdatawrite_wbc(inode->i_mapping, wbc);
9603 btrfs_add_delayed_iput(inode);
9604 if (ret || wbc->nr_to_write <= 0)
9608 spin_lock(&root->delalloc_lock);
9610 spin_unlock(&root->delalloc_lock);
9613 list_for_each_entry_safe(work, next, &works, list) {
9614 list_del_init(&work->list);
9615 wait_for_completion(&work->completion);
9619 if (!list_empty(&splice)) {
9620 spin_lock(&root->delalloc_lock);
9621 list_splice_tail(&splice, &root->delalloc_inodes);
9622 spin_unlock(&root->delalloc_lock);
9624 mutex_unlock(&root->delalloc_mutex);
9628 int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9630 struct writeback_control wbc = {
9631 .nr_to_write = LONG_MAX,
9632 .sync_mode = WB_SYNC_NONE,
9634 .range_end = LLONG_MAX,
9636 struct btrfs_fs_info *fs_info = root->fs_info;
9638 if (BTRFS_FS_ERROR(fs_info))
9641 return start_delalloc_inodes(root, &wbc, true, in_reclaim_context);
9644 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9645 bool in_reclaim_context)
9647 struct writeback_control wbc = {
9649 .sync_mode = WB_SYNC_NONE,
9651 .range_end = LLONG_MAX,
9653 struct btrfs_root *root;
9654 struct list_head splice;
9657 if (BTRFS_FS_ERROR(fs_info))
9660 INIT_LIST_HEAD(&splice);
9662 mutex_lock(&fs_info->delalloc_root_mutex);
9663 spin_lock(&fs_info->delalloc_root_lock);
9664 list_splice_init(&fs_info->delalloc_roots, &splice);
9665 while (!list_empty(&splice)) {
9667 * Reset nr_to_write here so we know that we're doing a full
9671 wbc.nr_to_write = LONG_MAX;
9673 root = list_first_entry(&splice, struct btrfs_root,
9675 root = btrfs_grab_root(root);
9677 list_move_tail(&root->delalloc_root,
9678 &fs_info->delalloc_roots);
9679 spin_unlock(&fs_info->delalloc_root_lock);
9681 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9682 btrfs_put_root(root);
9683 if (ret < 0 || wbc.nr_to_write <= 0)
9685 spin_lock(&fs_info->delalloc_root_lock);
9687 spin_unlock(&fs_info->delalloc_root_lock);
9691 if (!list_empty(&splice)) {
9692 spin_lock(&fs_info->delalloc_root_lock);
9693 list_splice_tail(&splice, &fs_info->delalloc_roots);
9694 spin_unlock(&fs_info->delalloc_root_lock);
9696 mutex_unlock(&fs_info->delalloc_root_mutex);
9700 static int btrfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
9701 struct dentry *dentry, const char *symname)
9703 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9704 struct btrfs_trans_handle *trans;
9705 struct btrfs_root *root = BTRFS_I(dir)->root;
9706 struct btrfs_path *path;
9707 struct btrfs_key key;
9708 struct inode *inode = NULL;
9715 struct btrfs_file_extent_item *ei;
9716 struct extent_buffer *leaf;
9718 name_len = strlen(symname);
9719 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9720 return -ENAMETOOLONG;
9723 * 2 items for inode item and ref
9724 * 2 items for dir items
9725 * 1 item for updating parent inode item
9726 * 1 item for the inline extent item
9727 * 1 item for xattr if selinux is on
9729 trans = btrfs_start_transaction(root, 7);
9731 return PTR_ERR(trans);
9733 err = btrfs_get_free_objectid(root, &objectid);
9737 inode = btrfs_new_inode(trans, root, mnt_userns, dir,
9738 dentry->d_name.name, dentry->d_name.len,
9739 btrfs_ino(BTRFS_I(dir)), objectid,
9740 S_IFLNK | S_IRWXUGO, &index);
9741 if (IS_ERR(inode)) {
9742 err = PTR_ERR(inode);
9748 * If the active LSM wants to access the inode during
9749 * d_instantiate it needs these. Smack checks to see
9750 * if the filesystem supports xattrs by looking at the
9753 inode->i_fop = &btrfs_file_operations;
9754 inode->i_op = &btrfs_file_inode_operations;
9755 inode->i_mapping->a_ops = &btrfs_aops;
9757 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9761 path = btrfs_alloc_path();
9766 key.objectid = btrfs_ino(BTRFS_I(inode));
9768 key.type = BTRFS_EXTENT_DATA_KEY;
9769 datasize = btrfs_file_extent_calc_inline_size(name_len);
9770 err = btrfs_insert_empty_item(trans, root, path, &key,
9773 btrfs_free_path(path);
9776 leaf = path->nodes[0];
9777 ei = btrfs_item_ptr(leaf, path->slots[0],
9778 struct btrfs_file_extent_item);
9779 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9780 btrfs_set_file_extent_type(leaf, ei,
9781 BTRFS_FILE_EXTENT_INLINE);
9782 btrfs_set_file_extent_encryption(leaf, ei, 0);
9783 btrfs_set_file_extent_compression(leaf, ei, 0);
9784 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9785 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9787 ptr = btrfs_file_extent_inline_start(ei);
9788 write_extent_buffer(leaf, symname, ptr, name_len);
9789 btrfs_mark_buffer_dirty(leaf);
9790 btrfs_free_path(path);
9792 inode->i_op = &btrfs_symlink_inode_operations;
9793 inode_nohighmem(inode);
9794 inode_set_bytes(inode, name_len);
9795 btrfs_i_size_write(BTRFS_I(inode), name_len);
9796 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
9798 * Last step, add directory indexes for our symlink inode. This is the
9799 * last step to avoid extra cleanup of these indexes if an error happens
9803 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9804 BTRFS_I(inode), 0, index);
9808 d_instantiate_new(dentry, inode);
9811 btrfs_end_transaction(trans);
9813 inode_dec_link_count(inode);
9814 discard_new_inode(inode);
9816 btrfs_btree_balance_dirty(fs_info);
9820 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9821 struct btrfs_trans_handle *trans_in,
9822 struct btrfs_inode *inode,
9823 struct btrfs_key *ins,
9826 struct btrfs_file_extent_item stack_fi;
9827 struct btrfs_replace_extent_info extent_info;
9828 struct btrfs_trans_handle *trans = trans_in;
9829 struct btrfs_path *path;
9830 u64 start = ins->objectid;
9831 u64 len = ins->offset;
9832 int qgroup_released;
9835 memset(&stack_fi, 0, sizeof(stack_fi));
9837 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9838 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9839 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9840 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9841 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9842 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9843 /* Encryption and other encoding is reserved and all 0 */
9845 qgroup_released = btrfs_qgroup_release_data(inode, file_offset, len);
9846 if (qgroup_released < 0)
9847 return ERR_PTR(qgroup_released);
9850 ret = insert_reserved_file_extent(trans, inode,
9851 file_offset, &stack_fi,
9852 true, qgroup_released);
9858 extent_info.disk_offset = start;
9859 extent_info.disk_len = len;
9860 extent_info.data_offset = 0;
9861 extent_info.data_len = len;
9862 extent_info.file_offset = file_offset;
9863 extent_info.extent_buf = (char *)&stack_fi;
9864 extent_info.is_new_extent = true;
9865 extent_info.qgroup_reserved = qgroup_released;
9866 extent_info.insertions = 0;
9868 path = btrfs_alloc_path();
9874 ret = btrfs_replace_file_extents(inode, path, file_offset,
9875 file_offset + len - 1, &extent_info,
9877 btrfs_free_path(path);
9884 * We have released qgroup data range at the beginning of the function,
9885 * and normally qgroup_released bytes will be freed when committing
9887 * But if we error out early, we have to free what we have released
9888 * or we leak qgroup data reservation.
9890 btrfs_qgroup_free_refroot(inode->root->fs_info,
9891 inode->root->root_key.objectid, qgroup_released,
9892 BTRFS_QGROUP_RSV_DATA);
9893 return ERR_PTR(ret);
9896 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9897 u64 start, u64 num_bytes, u64 min_size,
9898 loff_t actual_len, u64 *alloc_hint,
9899 struct btrfs_trans_handle *trans)
9901 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9902 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9903 struct extent_map *em;
9904 struct btrfs_root *root = BTRFS_I(inode)->root;
9905 struct btrfs_key ins;
9906 u64 cur_offset = start;
9907 u64 clear_offset = start;
9910 u64 last_alloc = (u64)-1;
9912 bool own_trans = true;
9913 u64 end = start + num_bytes - 1;
9917 while (num_bytes > 0) {
9918 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9919 cur_bytes = max(cur_bytes, min_size);
9921 * If we are severely fragmented we could end up with really
9922 * small allocations, so if the allocator is returning small
9923 * chunks lets make its job easier by only searching for those
9926 cur_bytes = min(cur_bytes, last_alloc);
9927 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9928 min_size, 0, *alloc_hint, &ins, 1, 0);
9933 * We've reserved this space, and thus converted it from
9934 * ->bytes_may_use to ->bytes_reserved. Any error that happens
9935 * from here on out we will only need to clear our reservation
9936 * for the remaining unreserved area, so advance our
9937 * clear_offset by our extent size.
9939 clear_offset += ins.offset;
9941 last_alloc = ins.offset;
9942 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
9945 * Now that we inserted the prealloc extent we can finally
9946 * decrement the number of reservations in the block group.
9947 * If we did it before, we could race with relocation and have
9948 * relocation miss the reserved extent, making it fail later.
9950 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
9951 if (IS_ERR(trans)) {
9952 ret = PTR_ERR(trans);
9953 btrfs_free_reserved_extent(fs_info, ins.objectid,
9958 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
9959 cur_offset + ins.offset -1, 0);
9961 em = alloc_extent_map();
9963 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9964 &BTRFS_I(inode)->runtime_flags);
9968 em->start = cur_offset;
9969 em->orig_start = cur_offset;
9970 em->len = ins.offset;
9971 em->block_start = ins.objectid;
9972 em->block_len = ins.offset;
9973 em->orig_block_len = ins.offset;
9974 em->ram_bytes = ins.offset;
9975 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9976 em->generation = trans->transid;
9979 write_lock(&em_tree->lock);
9980 ret = add_extent_mapping(em_tree, em, 1);
9981 write_unlock(&em_tree->lock);
9984 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
9985 cur_offset + ins.offset - 1,
9988 free_extent_map(em);
9990 num_bytes -= ins.offset;
9991 cur_offset += ins.offset;
9992 *alloc_hint = ins.objectid + ins.offset;
9994 inode_inc_iversion(inode);
9995 inode->i_ctime = current_time(inode);
9996 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9997 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9998 (actual_len > inode->i_size) &&
9999 (cur_offset > inode->i_size)) {
10000 if (cur_offset > actual_len)
10001 i_size = actual_len;
10003 i_size = cur_offset;
10004 i_size_write(inode, i_size);
10005 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
10008 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
10011 btrfs_abort_transaction(trans, ret);
10013 btrfs_end_transaction(trans);
10018 btrfs_end_transaction(trans);
10022 if (clear_offset < end)
10023 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
10024 end - clear_offset + 1);
10028 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10029 u64 start, u64 num_bytes, u64 min_size,
10030 loff_t actual_len, u64 *alloc_hint)
10032 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10033 min_size, actual_len, alloc_hint,
10037 int btrfs_prealloc_file_range_trans(struct inode *inode,
10038 struct btrfs_trans_handle *trans, int mode,
10039 u64 start, u64 num_bytes, u64 min_size,
10040 loff_t actual_len, u64 *alloc_hint)
10042 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10043 min_size, actual_len, alloc_hint, trans);
10046 static int btrfs_set_page_dirty(struct page *page)
10048 return __set_page_dirty_nobuffers(page);
10051 static int btrfs_permission(struct user_namespace *mnt_userns,
10052 struct inode *inode, int mask)
10054 struct btrfs_root *root = BTRFS_I(inode)->root;
10055 umode_t mode = inode->i_mode;
10057 if (mask & MAY_WRITE &&
10058 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10059 if (btrfs_root_readonly(root))
10061 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10064 return generic_permission(mnt_userns, inode, mask);
10067 static int btrfs_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
10068 struct dentry *dentry, umode_t mode)
10070 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10071 struct btrfs_trans_handle *trans;
10072 struct btrfs_root *root = BTRFS_I(dir)->root;
10073 struct inode *inode = NULL;
10079 * 5 units required for adding orphan entry
10081 trans = btrfs_start_transaction(root, 5);
10083 return PTR_ERR(trans);
10085 ret = btrfs_get_free_objectid(root, &objectid);
10089 inode = btrfs_new_inode(trans, root, mnt_userns, dir, NULL, 0,
10090 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10091 if (IS_ERR(inode)) {
10092 ret = PTR_ERR(inode);
10097 inode->i_fop = &btrfs_file_operations;
10098 inode->i_op = &btrfs_file_inode_operations;
10100 inode->i_mapping->a_ops = &btrfs_aops;
10102 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10106 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
10109 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10114 * We set number of links to 0 in btrfs_new_inode(), and here we set
10115 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10118 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10120 set_nlink(inode, 1);
10121 d_tmpfile(dentry, inode);
10122 unlock_new_inode(inode);
10123 mark_inode_dirty(inode);
10125 btrfs_end_transaction(trans);
10127 discard_new_inode(inode);
10128 btrfs_btree_balance_dirty(fs_info);
10132 void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end)
10134 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10135 unsigned long index = start >> PAGE_SHIFT;
10136 unsigned long end_index = end >> PAGE_SHIFT;
10140 ASSERT(end + 1 - start <= U32_MAX);
10141 len = end + 1 - start;
10142 while (index <= end_index) {
10143 page = find_get_page(inode->vfs_inode.i_mapping, index);
10144 ASSERT(page); /* Pages should be in the extent_io_tree */
10146 btrfs_page_set_writeback(fs_info, page, start, len);
10154 * Add an entry indicating a block group or device which is pinned by a
10155 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10156 * negative errno on failure.
10158 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10159 bool is_block_group)
10161 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10162 struct btrfs_swapfile_pin *sp, *entry;
10163 struct rb_node **p;
10164 struct rb_node *parent = NULL;
10166 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10171 sp->is_block_group = is_block_group;
10172 sp->bg_extent_count = 1;
10174 spin_lock(&fs_info->swapfile_pins_lock);
10175 p = &fs_info->swapfile_pins.rb_node;
10178 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10179 if (sp->ptr < entry->ptr ||
10180 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10181 p = &(*p)->rb_left;
10182 } else if (sp->ptr > entry->ptr ||
10183 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10184 p = &(*p)->rb_right;
10186 if (is_block_group)
10187 entry->bg_extent_count++;
10188 spin_unlock(&fs_info->swapfile_pins_lock);
10193 rb_link_node(&sp->node, parent, p);
10194 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10195 spin_unlock(&fs_info->swapfile_pins_lock);
10199 /* Free all of the entries pinned by this swapfile. */
10200 static void btrfs_free_swapfile_pins(struct inode *inode)
10202 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10203 struct btrfs_swapfile_pin *sp;
10204 struct rb_node *node, *next;
10206 spin_lock(&fs_info->swapfile_pins_lock);
10207 node = rb_first(&fs_info->swapfile_pins);
10209 next = rb_next(node);
10210 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10211 if (sp->inode == inode) {
10212 rb_erase(&sp->node, &fs_info->swapfile_pins);
10213 if (sp->is_block_group) {
10214 btrfs_dec_block_group_swap_extents(sp->ptr,
10215 sp->bg_extent_count);
10216 btrfs_put_block_group(sp->ptr);
10222 spin_unlock(&fs_info->swapfile_pins_lock);
10225 struct btrfs_swap_info {
10231 unsigned long nr_pages;
10235 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10236 struct btrfs_swap_info *bsi)
10238 unsigned long nr_pages;
10239 unsigned long max_pages;
10240 u64 first_ppage, first_ppage_reported, next_ppage;
10244 * Our swapfile may have had its size extended after the swap header was
10245 * written. In that case activating the swapfile should not go beyond
10246 * the max size set in the swap header.
10248 if (bsi->nr_pages >= sis->max)
10251 max_pages = sis->max - bsi->nr_pages;
10252 first_ppage = ALIGN(bsi->block_start, PAGE_SIZE) >> PAGE_SHIFT;
10253 next_ppage = ALIGN_DOWN(bsi->block_start + bsi->block_len,
10254 PAGE_SIZE) >> PAGE_SHIFT;
10256 if (first_ppage >= next_ppage)
10258 nr_pages = next_ppage - first_ppage;
10259 nr_pages = min(nr_pages, max_pages);
10261 first_ppage_reported = first_ppage;
10262 if (bsi->start == 0)
10263 first_ppage_reported++;
10264 if (bsi->lowest_ppage > first_ppage_reported)
10265 bsi->lowest_ppage = first_ppage_reported;
10266 if (bsi->highest_ppage < (next_ppage - 1))
10267 bsi->highest_ppage = next_ppage - 1;
10269 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10272 bsi->nr_extents += ret;
10273 bsi->nr_pages += nr_pages;
10277 static void btrfs_swap_deactivate(struct file *file)
10279 struct inode *inode = file_inode(file);
10281 btrfs_free_swapfile_pins(inode);
10282 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10285 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10288 struct inode *inode = file_inode(file);
10289 struct btrfs_root *root = BTRFS_I(inode)->root;
10290 struct btrfs_fs_info *fs_info = root->fs_info;
10291 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10292 struct extent_state *cached_state = NULL;
10293 struct extent_map *em = NULL;
10294 struct btrfs_device *device = NULL;
10295 struct btrfs_swap_info bsi = {
10296 .lowest_ppage = (sector_t)-1ULL,
10303 * If the swap file was just created, make sure delalloc is done. If the
10304 * file changes again after this, the user is doing something stupid and
10305 * we don't really care.
10307 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10312 * The inode is locked, so these flags won't change after we check them.
10314 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10315 btrfs_warn(fs_info, "swapfile must not be compressed");
10318 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10319 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10322 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10323 btrfs_warn(fs_info, "swapfile must not be checksummed");
10328 * Balance or device remove/replace/resize can move stuff around from
10329 * under us. The exclop protection makes sure they aren't running/won't
10330 * run concurrently while we are mapping the swap extents, and
10331 * fs_info->swapfile_pins prevents them from running while the swap
10332 * file is active and moving the extents. Note that this also prevents
10333 * a concurrent device add which isn't actually necessary, but it's not
10334 * really worth the trouble to allow it.
10336 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10337 btrfs_warn(fs_info,
10338 "cannot activate swapfile while exclusive operation is running");
10343 * Prevent snapshot creation while we are activating the swap file.
10344 * We do not want to race with snapshot creation. If snapshot creation
10345 * already started before we bumped nr_swapfiles from 0 to 1 and
10346 * completes before the first write into the swap file after it is
10347 * activated, than that write would fallback to COW.
10349 if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) {
10350 btrfs_exclop_finish(fs_info);
10351 btrfs_warn(fs_info,
10352 "cannot activate swapfile because snapshot creation is in progress");
10356 * Snapshots can create extents which require COW even if NODATACOW is
10357 * set. We use this counter to prevent snapshots. We must increment it
10358 * before walking the extents because we don't want a concurrent
10359 * snapshot to run after we've already checked the extents.
10361 atomic_inc(&root->nr_swapfiles);
10363 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10365 lock_extent_bits(io_tree, 0, isize - 1, &cached_state);
10367 while (start < isize) {
10368 u64 logical_block_start, physical_block_start;
10369 struct btrfs_block_group *bg;
10370 u64 len = isize - start;
10372 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
10378 if (em->block_start == EXTENT_MAP_HOLE) {
10379 btrfs_warn(fs_info, "swapfile must not have holes");
10383 if (em->block_start == EXTENT_MAP_INLINE) {
10385 * It's unlikely we'll ever actually find ourselves
10386 * here, as a file small enough to fit inline won't be
10387 * big enough to store more than the swap header, but in
10388 * case something changes in the future, let's catch it
10389 * here rather than later.
10391 btrfs_warn(fs_info, "swapfile must not be inline");
10395 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10396 btrfs_warn(fs_info, "swapfile must not be compressed");
10401 logical_block_start = em->block_start + (start - em->start);
10402 len = min(len, em->len - (start - em->start));
10403 free_extent_map(em);
10406 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, true);
10412 btrfs_warn(fs_info,
10413 "swapfile must not be copy-on-write");
10418 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10424 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10425 btrfs_warn(fs_info,
10426 "swapfile must have single data profile");
10431 if (device == NULL) {
10432 device = em->map_lookup->stripes[0].dev;
10433 ret = btrfs_add_swapfile_pin(inode, device, false);
10438 } else if (device != em->map_lookup->stripes[0].dev) {
10439 btrfs_warn(fs_info, "swapfile must be on one device");
10444 physical_block_start = (em->map_lookup->stripes[0].physical +
10445 (logical_block_start - em->start));
10446 len = min(len, em->len - (logical_block_start - em->start));
10447 free_extent_map(em);
10450 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10452 btrfs_warn(fs_info,
10453 "could not find block group containing swapfile");
10458 if (!btrfs_inc_block_group_swap_extents(bg)) {
10459 btrfs_warn(fs_info,
10460 "block group for swapfile at %llu is read-only%s",
10462 atomic_read(&fs_info->scrubs_running) ?
10463 " (scrub running)" : "");
10464 btrfs_put_block_group(bg);
10469 ret = btrfs_add_swapfile_pin(inode, bg, true);
10471 btrfs_put_block_group(bg);
10478 if (bsi.block_len &&
10479 bsi.block_start + bsi.block_len == physical_block_start) {
10480 bsi.block_len += len;
10482 if (bsi.block_len) {
10483 ret = btrfs_add_swap_extent(sis, &bsi);
10488 bsi.block_start = physical_block_start;
10489 bsi.block_len = len;
10496 ret = btrfs_add_swap_extent(sis, &bsi);
10499 if (!IS_ERR_OR_NULL(em))
10500 free_extent_map(em);
10502 unlock_extent_cached(io_tree, 0, isize - 1, &cached_state);
10505 btrfs_swap_deactivate(file);
10507 btrfs_drew_write_unlock(&root->snapshot_lock);
10509 btrfs_exclop_finish(fs_info);
10515 sis->bdev = device->bdev;
10516 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10517 sis->max = bsi.nr_pages;
10518 sis->pages = bsi.nr_pages - 1;
10519 sis->highest_bit = bsi.nr_pages - 1;
10520 return bsi.nr_extents;
10523 static void btrfs_swap_deactivate(struct file *file)
10527 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10530 return -EOPNOTSUPP;
10535 * Update the number of bytes used in the VFS' inode. When we replace extents in
10536 * a range (clone, dedupe, fallocate's zero range), we must update the number of
10537 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
10538 * always get a correct value.
10540 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
10541 const u64 add_bytes,
10542 const u64 del_bytes)
10544 if (add_bytes == del_bytes)
10547 spin_lock(&inode->lock);
10549 inode_sub_bytes(&inode->vfs_inode, del_bytes);
10551 inode_add_bytes(&inode->vfs_inode, add_bytes);
10552 spin_unlock(&inode->lock);
10555 static const struct inode_operations btrfs_dir_inode_operations = {
10556 .getattr = btrfs_getattr,
10557 .lookup = btrfs_lookup,
10558 .create = btrfs_create,
10559 .unlink = btrfs_unlink,
10560 .link = btrfs_link,
10561 .mkdir = btrfs_mkdir,
10562 .rmdir = btrfs_rmdir,
10563 .rename = btrfs_rename2,
10564 .symlink = btrfs_symlink,
10565 .setattr = btrfs_setattr,
10566 .mknod = btrfs_mknod,
10567 .listxattr = btrfs_listxattr,
10568 .permission = btrfs_permission,
10569 .get_acl = btrfs_get_acl,
10570 .set_acl = btrfs_set_acl,
10571 .update_time = btrfs_update_time,
10572 .tmpfile = btrfs_tmpfile,
10573 .fileattr_get = btrfs_fileattr_get,
10574 .fileattr_set = btrfs_fileattr_set,
10577 static const struct file_operations btrfs_dir_file_operations = {
10578 .llseek = generic_file_llseek,
10579 .read = generic_read_dir,
10580 .iterate_shared = btrfs_real_readdir,
10581 .open = btrfs_opendir,
10582 .unlocked_ioctl = btrfs_ioctl,
10583 #ifdef CONFIG_COMPAT
10584 .compat_ioctl = btrfs_compat_ioctl,
10586 .release = btrfs_release_file,
10587 .fsync = btrfs_sync_file,
10591 * btrfs doesn't support the bmap operation because swapfiles
10592 * use bmap to make a mapping of extents in the file. They assume
10593 * these extents won't change over the life of the file and they
10594 * use the bmap result to do IO directly to the drive.
10596 * the btrfs bmap call would return logical addresses that aren't
10597 * suitable for IO and they also will change frequently as COW
10598 * operations happen. So, swapfile + btrfs == corruption.
10600 * For now we're avoiding this by dropping bmap.
10602 static const struct address_space_operations btrfs_aops = {
10603 .readpage = btrfs_readpage,
10604 .writepage = btrfs_writepage,
10605 .writepages = btrfs_writepages,
10606 .readahead = btrfs_readahead,
10607 .direct_IO = noop_direct_IO,
10608 .invalidatepage = btrfs_invalidatepage,
10609 .releasepage = btrfs_releasepage,
10610 #ifdef CONFIG_MIGRATION
10611 .migratepage = btrfs_migratepage,
10613 .set_page_dirty = btrfs_set_page_dirty,
10614 .error_remove_page = generic_error_remove_page,
10615 .swap_activate = btrfs_swap_activate,
10616 .swap_deactivate = btrfs_swap_deactivate,
10619 static const struct inode_operations btrfs_file_inode_operations = {
10620 .getattr = btrfs_getattr,
10621 .setattr = btrfs_setattr,
10622 .listxattr = btrfs_listxattr,
10623 .permission = btrfs_permission,
10624 .fiemap = btrfs_fiemap,
10625 .get_acl = btrfs_get_acl,
10626 .set_acl = btrfs_set_acl,
10627 .update_time = btrfs_update_time,
10628 .fileattr_get = btrfs_fileattr_get,
10629 .fileattr_set = btrfs_fileattr_set,
10631 static const struct inode_operations btrfs_special_inode_operations = {
10632 .getattr = btrfs_getattr,
10633 .setattr = btrfs_setattr,
10634 .permission = btrfs_permission,
10635 .listxattr = btrfs_listxattr,
10636 .get_acl = btrfs_get_acl,
10637 .set_acl = btrfs_set_acl,
10638 .update_time = btrfs_update_time,
10640 static const struct inode_operations btrfs_symlink_inode_operations = {
10641 .get_link = page_get_link,
10642 .getattr = btrfs_getattr,
10643 .setattr = btrfs_setattr,
10644 .permission = btrfs_permission,
10645 .listxattr = btrfs_listxattr,
10646 .update_time = btrfs_update_time,
10649 const struct dentry_operations btrfs_dentry_operations = {
10650 .d_delete = btrfs_dentry_delete,