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,
249 struct btrfs_root *root = inode->root;
250 struct extent_buffer *leaf;
251 struct page *page = NULL;
254 struct btrfs_file_extent_item *ei;
256 size_t cur_size = size;
259 ASSERT((compressed_size > 0 && compressed_pages) ||
260 (compressed_size == 0 && !compressed_pages));
262 if (compressed_size && compressed_pages)
263 cur_size = compressed_size;
265 if (!extent_inserted) {
266 struct btrfs_key key;
269 key.objectid = btrfs_ino(inode);
271 key.type = BTRFS_EXTENT_DATA_KEY;
273 datasize = btrfs_file_extent_calc_inline_size(cur_size);
274 ret = btrfs_insert_empty_item(trans, root, path, &key,
279 leaf = path->nodes[0];
280 ei = btrfs_item_ptr(leaf, path->slots[0],
281 struct btrfs_file_extent_item);
282 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
283 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
284 btrfs_set_file_extent_encryption(leaf, ei, 0);
285 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
286 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
287 ptr = btrfs_file_extent_inline_start(ei);
289 if (compress_type != BTRFS_COMPRESS_NONE) {
292 while (compressed_size > 0) {
293 cpage = compressed_pages[i];
294 cur_size = min_t(unsigned long, compressed_size,
297 kaddr = kmap_atomic(cpage);
298 write_extent_buffer(leaf, kaddr, ptr, cur_size);
299 kunmap_atomic(kaddr);
303 compressed_size -= cur_size;
305 btrfs_set_file_extent_compression(leaf, ei,
308 page = find_get_page(inode->vfs_inode.i_mapping, 0);
309 btrfs_set_file_extent_compression(leaf, ei, 0);
310 kaddr = kmap_atomic(page);
311 write_extent_buffer(leaf, kaddr, ptr, size);
312 kunmap_atomic(kaddr);
315 btrfs_mark_buffer_dirty(leaf);
316 btrfs_release_path(path);
319 * We align size to sectorsize for inline extents just for simplicity
322 ret = btrfs_inode_set_file_extent_range(inode, 0,
323 ALIGN(size, root->fs_info->sectorsize));
328 * We're an inline extent, so nobody can extend the file past i_size
329 * without locking a page we already have locked.
331 * We must do any i_size and inode updates before we unlock the pages.
332 * Otherwise we could end up racing with unlink.
334 i_size = i_size_read(&inode->vfs_inode);
335 if (update_i_size && size > i_size) {
336 i_size_write(&inode->vfs_inode, size);
339 inode->disk_i_size = i_size;
347 * conditionally insert an inline extent into the file. This
348 * does the checks required to make sure the data is small enough
349 * to fit as an inline extent.
351 static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 size,
352 size_t compressed_size,
354 struct page **compressed_pages,
357 struct btrfs_drop_extents_args drop_args = { 0 };
358 struct btrfs_root *root = inode->root;
359 struct btrfs_fs_info *fs_info = root->fs_info;
360 struct btrfs_trans_handle *trans;
361 u64 data_len = (compressed_size ?: size);
363 struct btrfs_path *path;
366 * We can create an inline extent if it ends at or beyond the current
367 * i_size, is no larger than a sector (decompressed), and the (possibly
368 * compressed) data fits in a leaf and the configured maximum inline
371 if (size < i_size_read(&inode->vfs_inode) ||
372 size > fs_info->sectorsize ||
373 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
374 data_len > fs_info->max_inline)
377 path = btrfs_alloc_path();
381 trans = btrfs_join_transaction(root);
383 btrfs_free_path(path);
384 return PTR_ERR(trans);
386 trans->block_rsv = &inode->block_rsv;
388 drop_args.path = path;
390 drop_args.end = fs_info->sectorsize;
391 drop_args.drop_cache = true;
392 drop_args.replace_extent = true;
393 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(data_len);
394 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
396 btrfs_abort_transaction(trans, ret);
400 ret = insert_inline_extent(trans, path, inode, drop_args.extent_inserted,
401 size, compressed_size, compress_type,
402 compressed_pages, update_i_size);
403 if (ret && ret != -ENOSPC) {
404 btrfs_abort_transaction(trans, ret);
406 } else if (ret == -ENOSPC) {
411 btrfs_update_inode_bytes(inode, size, drop_args.bytes_found);
412 ret = btrfs_update_inode(trans, root, inode);
413 if (ret && ret != -ENOSPC) {
414 btrfs_abort_transaction(trans, ret);
416 } else if (ret == -ENOSPC) {
421 btrfs_set_inode_full_sync(inode);
424 * Don't forget to free the reserved space, as for inlined extent
425 * it won't count as data extent, free them directly here.
426 * And at reserve time, it's always aligned to page size, so
427 * just free one page here.
429 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
430 btrfs_free_path(path);
431 btrfs_end_transaction(trans);
435 struct async_extent {
440 unsigned long nr_pages;
442 struct list_head list;
447 struct page *locked_page;
450 unsigned int write_flags;
451 struct list_head extents;
452 struct cgroup_subsys_state *blkcg_css;
453 struct btrfs_work work;
454 struct async_cow *async_cow;
459 struct async_chunk chunks[];
462 static noinline int add_async_extent(struct async_chunk *cow,
463 u64 start, u64 ram_size,
466 unsigned long nr_pages,
469 struct async_extent *async_extent;
471 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
472 BUG_ON(!async_extent); /* -ENOMEM */
473 async_extent->start = start;
474 async_extent->ram_size = ram_size;
475 async_extent->compressed_size = compressed_size;
476 async_extent->pages = pages;
477 async_extent->nr_pages = nr_pages;
478 async_extent->compress_type = compress_type;
479 list_add_tail(&async_extent->list, &cow->extents);
484 * Check if the inode has flags compatible with compression
486 static inline bool inode_can_compress(struct btrfs_inode *inode)
488 if (inode->flags & BTRFS_INODE_NODATACOW ||
489 inode->flags & BTRFS_INODE_NODATASUM)
495 * Check if the inode needs to be submitted to compression, based on mount
496 * options, defragmentation, properties or heuristics.
498 static inline int inode_need_compress(struct btrfs_inode *inode, u64 start,
501 struct btrfs_fs_info *fs_info = inode->root->fs_info;
503 if (!inode_can_compress(inode)) {
504 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
505 KERN_ERR "BTRFS: unexpected compression for ino %llu\n",
510 * Special check for subpage.
512 * We lock the full page then run each delalloc range in the page, thus
513 * for the following case, we will hit some subpage specific corner case:
516 * | |///////| |///////|
519 * In above case, both range A and range B will try to unlock the full
520 * page [0, 64K), causing the one finished later will have page
521 * unlocked already, triggering various page lock requirement BUG_ON()s.
523 * So here we add an artificial limit that subpage compression can only
524 * if the range is fully page aligned.
526 * In theory we only need to ensure the first page is fully covered, but
527 * the tailing partial page will be locked until the full compression
528 * finishes, delaying the write of other range.
530 * TODO: Make btrfs_run_delalloc_range() to lock all delalloc range
531 * first to prevent any submitted async extent to unlock the full page.
532 * By this, we can ensure for subpage case that only the last async_cow
533 * will unlock the full page.
535 if (fs_info->sectorsize < PAGE_SIZE) {
536 if (!IS_ALIGNED(start, PAGE_SIZE) ||
537 !IS_ALIGNED(end + 1, PAGE_SIZE))
542 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
545 if (inode->defrag_compress)
547 /* bad compression ratios */
548 if (inode->flags & BTRFS_INODE_NOCOMPRESS)
550 if (btrfs_test_opt(fs_info, COMPRESS) ||
551 inode->flags & BTRFS_INODE_COMPRESS ||
552 inode->prop_compress)
553 return btrfs_compress_heuristic(&inode->vfs_inode, start, end);
557 static inline void inode_should_defrag(struct btrfs_inode *inode,
558 u64 start, u64 end, u64 num_bytes, u32 small_write)
560 /* If this is a small write inside eof, kick off a defrag */
561 if (num_bytes < small_write &&
562 (start > 0 || end + 1 < inode->disk_i_size))
563 btrfs_add_inode_defrag(NULL, inode, small_write);
567 * we create compressed extents in two phases. The first
568 * phase compresses a range of pages that have already been
569 * locked (both pages and state bits are locked).
571 * This is done inside an ordered work queue, and the compression
572 * is spread across many cpus. The actual IO submission is step
573 * two, and the ordered work queue takes care of making sure that
574 * happens in the same order things were put onto the queue by
575 * writepages and friends.
577 * If this code finds it can't get good compression, it puts an
578 * entry onto the work queue to write the uncompressed bytes. This
579 * makes sure that both compressed inodes and uncompressed inodes
580 * are written in the same order that the flusher thread sent them
583 static noinline int compress_file_range(struct async_chunk *async_chunk)
585 struct inode *inode = async_chunk->inode;
586 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
587 u64 blocksize = fs_info->sectorsize;
588 u64 start = async_chunk->start;
589 u64 end = async_chunk->end;
593 struct page **pages = NULL;
594 unsigned long nr_pages;
595 unsigned long total_compressed = 0;
596 unsigned long total_in = 0;
599 int compress_type = fs_info->compress_type;
600 int compressed_extents = 0;
603 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
607 * We need to save i_size before now because it could change in between
608 * us evaluating the size and assigning it. This is because we lock and
609 * unlock the page in truncate and fallocate, and then modify the i_size
612 * The barriers are to emulate READ_ONCE, remove that once i_size_read
616 i_size = i_size_read(inode);
618 actual_end = min_t(u64, i_size, end + 1);
621 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
622 nr_pages = min_t(unsigned long, nr_pages,
623 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
626 * we don't want to send crud past the end of i_size through
627 * compression, that's just a waste of CPU time. So, if the
628 * end of the file is before the start of our current
629 * requested range of bytes, we bail out to the uncompressed
630 * cleanup code that can deal with all of this.
632 * It isn't really the fastest way to fix things, but this is a
633 * very uncommon corner.
635 if (actual_end <= start)
636 goto cleanup_and_bail_uncompressed;
638 total_compressed = actual_end - start;
641 * Skip compression for a small file range(<=blocksize) that
642 * isn't an inline extent, since it doesn't save disk space at all.
644 if (total_compressed <= blocksize &&
645 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
646 goto cleanup_and_bail_uncompressed;
649 * For subpage case, we require full page alignment for the sector
651 * Thus we must also check against @actual_end, not just @end.
653 if (blocksize < PAGE_SIZE) {
654 if (!IS_ALIGNED(start, PAGE_SIZE) ||
655 !IS_ALIGNED(round_up(actual_end, blocksize), PAGE_SIZE))
656 goto cleanup_and_bail_uncompressed;
659 total_compressed = min_t(unsigned long, total_compressed,
660 BTRFS_MAX_UNCOMPRESSED);
665 * we do compression for mount -o compress and when the
666 * inode has not been flagged as nocompress. This flag can
667 * change at any time if we discover bad compression ratios.
669 if (inode_need_compress(BTRFS_I(inode), start, end)) {
671 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
673 /* just bail out to the uncompressed code */
678 if (BTRFS_I(inode)->defrag_compress)
679 compress_type = BTRFS_I(inode)->defrag_compress;
680 else if (BTRFS_I(inode)->prop_compress)
681 compress_type = BTRFS_I(inode)->prop_compress;
684 * we need to call clear_page_dirty_for_io on each
685 * page in the range. Otherwise applications with the file
686 * mmap'd can wander in and change the page contents while
687 * we are compressing them.
689 * If the compression fails for any reason, we set the pages
690 * dirty again later on.
692 * Note that the remaining part is redirtied, the start pointer
693 * has moved, the end is the original one.
696 extent_range_clear_dirty_for_io(inode, start, end);
700 /* Compression level is applied here and only here */
701 ret = btrfs_compress_pages(
702 compress_type | (fs_info->compress_level << 4),
703 inode->i_mapping, start,
710 unsigned long offset = offset_in_page(total_compressed);
711 struct page *page = pages[nr_pages - 1];
713 /* zero the tail end of the last page, we might be
714 * sending it down to disk
717 memzero_page(page, offset, PAGE_SIZE - offset);
723 * Check cow_file_range() for why we don't even try to create inline
724 * extent for subpage case.
726 if (start == 0 && fs_info->sectorsize == PAGE_SIZE) {
727 /* lets try to make an inline extent */
728 if (ret || total_in < actual_end) {
729 /* we didn't compress the entire range, try
730 * to make an uncompressed inline extent.
732 ret = cow_file_range_inline(BTRFS_I(inode), actual_end,
733 0, BTRFS_COMPRESS_NONE,
736 /* try making a compressed inline extent */
737 ret = cow_file_range_inline(BTRFS_I(inode), actual_end,
739 compress_type, pages,
743 unsigned long clear_flags = EXTENT_DELALLOC |
744 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
745 EXTENT_DO_ACCOUNTING;
746 unsigned long page_error_op;
748 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
751 * inline extent creation worked or returned error,
752 * we don't need to create any more async work items.
753 * Unlock and free up our temp pages.
755 * We use DO_ACCOUNTING here because we need the
756 * delalloc_release_metadata to be done _after_ we drop
757 * our outstanding extent for clearing delalloc for this
760 extent_clear_unlock_delalloc(BTRFS_I(inode), start, end,
764 PAGE_START_WRITEBACK |
769 * Ensure we only free the compressed pages if we have
770 * them allocated, as we can still reach here with
771 * inode_need_compress() == false.
774 for (i = 0; i < nr_pages; i++) {
775 WARN_ON(pages[i]->mapping);
786 * we aren't doing an inline extent round the compressed size
787 * up to a block size boundary so the allocator does sane
790 total_compressed = ALIGN(total_compressed, blocksize);
793 * one last check to make sure the compression is really a
794 * win, compare the page count read with the blocks on disk,
795 * compression must free at least one sector size
797 total_in = round_up(total_in, fs_info->sectorsize);
798 if (total_compressed + blocksize <= total_in) {
799 compressed_extents++;
802 * The async work queues will take care of doing actual
803 * allocation on disk for these compressed pages, and
804 * will submit them to the elevator.
806 add_async_extent(async_chunk, start, total_in,
807 total_compressed, pages, nr_pages,
810 if (start + total_in < end) {
816 return compressed_extents;
821 * the compression code ran but failed to make things smaller,
822 * free any pages it allocated and our page pointer array
824 for (i = 0; i < nr_pages; i++) {
825 WARN_ON(pages[i]->mapping);
830 total_compressed = 0;
833 /* flag the file so we don't compress in the future */
834 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
835 !(BTRFS_I(inode)->prop_compress)) {
836 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
839 cleanup_and_bail_uncompressed:
841 * No compression, but we still need to write the pages in the file
842 * we've been given so far. redirty the locked page if it corresponds
843 * to our extent and set things up for the async work queue to run
844 * cow_file_range to do the normal delalloc dance.
846 if (async_chunk->locked_page &&
847 (page_offset(async_chunk->locked_page) >= start &&
848 page_offset(async_chunk->locked_page)) <= end) {
849 __set_page_dirty_nobuffers(async_chunk->locked_page);
850 /* unlocked later on in the async handlers */
854 extent_range_redirty_for_io(inode, start, end);
855 add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
856 BTRFS_COMPRESS_NONE);
857 compressed_extents++;
859 return compressed_extents;
862 static void free_async_extent_pages(struct async_extent *async_extent)
866 if (!async_extent->pages)
869 for (i = 0; i < async_extent->nr_pages; i++) {
870 WARN_ON(async_extent->pages[i]->mapping);
871 put_page(async_extent->pages[i]);
873 kfree(async_extent->pages);
874 async_extent->nr_pages = 0;
875 async_extent->pages = NULL;
878 static int submit_uncompressed_range(struct btrfs_inode *inode,
879 struct async_extent *async_extent,
880 struct page *locked_page)
882 u64 start = async_extent->start;
883 u64 end = async_extent->start + async_extent->ram_size - 1;
884 unsigned long nr_written = 0;
885 int page_started = 0;
889 * Call cow_file_range() to run the delalloc range directly, since we
890 * won't go to NOCOW or async path again.
892 * Also we call cow_file_range() with @unlock_page == 0, so that we
893 * can directly submit them without interruption.
895 ret = cow_file_range(inode, locked_page, start, end, &page_started,
897 /* Inline extent inserted, page gets unlocked and everything is done */
904 unlock_page(locked_page);
908 ret = extent_write_locked_range(&inode->vfs_inode, start, end);
909 /* All pages will be unlocked, including @locked_page */
915 static int submit_one_async_extent(struct btrfs_inode *inode,
916 struct async_chunk *async_chunk,
917 struct async_extent *async_extent,
920 struct extent_io_tree *io_tree = &inode->io_tree;
921 struct btrfs_root *root = inode->root;
922 struct btrfs_fs_info *fs_info = root->fs_info;
923 struct btrfs_key ins;
924 struct page *locked_page = NULL;
925 struct extent_map *em;
927 u64 start = async_extent->start;
928 u64 end = async_extent->start + async_extent->ram_size - 1;
931 * If async_chunk->locked_page is in the async_extent range, we need to
934 if (async_chunk->locked_page) {
935 u64 locked_page_start = page_offset(async_chunk->locked_page);
936 u64 locked_page_end = locked_page_start + PAGE_SIZE - 1;
938 if (!(start >= locked_page_end || end <= locked_page_start))
939 locked_page = async_chunk->locked_page;
941 lock_extent(io_tree, start, end);
943 /* We have fall back to uncompressed write */
944 if (!async_extent->pages)
945 return submit_uncompressed_range(inode, async_extent, locked_page);
947 ret = btrfs_reserve_extent(root, async_extent->ram_size,
948 async_extent->compressed_size,
949 async_extent->compressed_size,
950 0, *alloc_hint, &ins, 1, 1);
952 free_async_extent_pages(async_extent);
954 * Here we used to try again by going back to non-compressed
955 * path for ENOSPC. But we can't reserve space even for
956 * compressed size, how could it work for uncompressed size
957 * which requires larger size? So here we directly go error
963 /* Here we're doing allocation and writeback of the compressed pages */
964 em = create_io_em(inode, start,
965 async_extent->ram_size, /* len */
966 start, /* orig_start */
967 ins.objectid, /* block_start */
968 ins.offset, /* block_len */
969 ins.offset, /* orig_block_len */
970 async_extent->ram_size, /* ram_bytes */
971 async_extent->compress_type,
972 BTRFS_ORDERED_COMPRESSED);
975 goto out_free_reserve;
979 ret = btrfs_add_ordered_extent(inode, start, /* file_offset */
980 async_extent->ram_size, /* num_bytes */
981 async_extent->ram_size, /* ram_bytes */
982 ins.objectid, /* disk_bytenr */
983 ins.offset, /* disk_num_bytes */
985 1 << BTRFS_ORDERED_COMPRESSED,
986 async_extent->compress_type);
988 btrfs_drop_extent_cache(inode, start, end, 0);
989 goto out_free_reserve;
991 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
993 /* Clear dirty, set writeback and unlock the pages. */
994 extent_clear_unlock_delalloc(inode, start, end,
995 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
996 PAGE_UNLOCK | PAGE_START_WRITEBACK);
997 if (btrfs_submit_compressed_write(inode, start, /* file_offset */
998 async_extent->ram_size, /* num_bytes */
999 ins.objectid, /* disk_bytenr */
1000 ins.offset, /* compressed_len */
1001 async_extent->pages, /* compressed_pages */
1002 async_extent->nr_pages,
1003 async_chunk->write_flags,
1004 async_chunk->blkcg_css, true)) {
1005 const u64 start = async_extent->start;
1006 const u64 end = start + async_extent->ram_size - 1;
1008 btrfs_writepage_endio_finish_ordered(inode, NULL, start, end, 0);
1010 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
1011 PAGE_END_WRITEBACK | PAGE_SET_ERROR);
1012 free_async_extent_pages(async_extent);
1014 *alloc_hint = ins.objectid + ins.offset;
1015 kfree(async_extent);
1019 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1020 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1022 extent_clear_unlock_delalloc(inode, start, end,
1023 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
1024 EXTENT_DELALLOC_NEW |
1025 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
1026 PAGE_UNLOCK | PAGE_START_WRITEBACK |
1027 PAGE_END_WRITEBACK | PAGE_SET_ERROR);
1028 free_async_extent_pages(async_extent);
1029 kfree(async_extent);
1034 * Phase two of compressed writeback. This is the ordered portion of the code,
1035 * which only gets called in the order the work was queued. We walk all the
1036 * async extents created by compress_file_range and send them down to the disk.
1038 static noinline void submit_compressed_extents(struct async_chunk *async_chunk)
1040 struct btrfs_inode *inode = BTRFS_I(async_chunk->inode);
1041 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1042 struct async_extent *async_extent;
1046 while (!list_empty(&async_chunk->extents)) {
1050 async_extent = list_entry(async_chunk->extents.next,
1051 struct async_extent, list);
1052 list_del(&async_extent->list);
1053 extent_start = async_extent->start;
1054 ram_size = async_extent->ram_size;
1056 ret = submit_one_async_extent(inode, async_chunk, async_extent,
1058 btrfs_debug(fs_info,
1059 "async extent submission failed root=%lld inode=%llu start=%llu len=%llu ret=%d",
1060 inode->root->root_key.objectid,
1061 btrfs_ino(inode), extent_start, ram_size, ret);
1065 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
1068 struct extent_map_tree *em_tree = &inode->extent_tree;
1069 struct extent_map *em;
1072 read_lock(&em_tree->lock);
1073 em = search_extent_mapping(em_tree, start, num_bytes);
1076 * if block start isn't an actual block number then find the
1077 * first block in this inode and use that as a hint. If that
1078 * block is also bogus then just don't worry about it.
1080 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1081 free_extent_map(em);
1082 em = search_extent_mapping(em_tree, 0, 0);
1083 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
1084 alloc_hint = em->block_start;
1086 free_extent_map(em);
1088 alloc_hint = em->block_start;
1089 free_extent_map(em);
1092 read_unlock(&em_tree->lock);
1098 * when extent_io.c finds a delayed allocation range in the file,
1099 * the call backs end up in this code. The basic idea is to
1100 * allocate extents on disk for the range, and create ordered data structs
1101 * in ram to track those extents.
1103 * locked_page is the page that writepage had locked already. We use
1104 * it to make sure we don't do extra locks or unlocks.
1106 * *page_started is set to one if we unlock locked_page and do everything
1107 * required to start IO on it. It may be clean and already done with
1108 * IO when we return.
1110 static noinline int cow_file_range(struct btrfs_inode *inode,
1111 struct page *locked_page,
1112 u64 start, u64 end, int *page_started,
1113 unsigned long *nr_written, int unlock)
1115 struct btrfs_root *root = inode->root;
1116 struct btrfs_fs_info *fs_info = root->fs_info;
1119 unsigned long ram_size;
1120 u64 cur_alloc_size = 0;
1122 u64 blocksize = fs_info->sectorsize;
1123 struct btrfs_key ins;
1124 struct extent_map *em;
1125 unsigned clear_bits;
1126 unsigned long page_ops;
1127 bool extent_reserved = false;
1130 if (btrfs_is_free_space_inode(inode)) {
1135 num_bytes = ALIGN(end - start + 1, blocksize);
1136 num_bytes = max(blocksize, num_bytes);
1137 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1139 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1142 * Due to the page size limit, for subpage we can only trigger the
1143 * writeback for the dirty sectors of page, that means data writeback
1144 * is doing more writeback than what we want.
1146 * This is especially unexpected for some call sites like fallocate,
1147 * where we only increase i_size after everything is done.
1148 * This means we can trigger inline extent even if we didn't want to.
1149 * So here we skip inline extent creation completely.
1151 if (start == 0 && fs_info->sectorsize == PAGE_SIZE) {
1152 u64 actual_end = min_t(u64, i_size_read(&inode->vfs_inode),
1155 /* lets try to make an inline extent */
1156 ret = cow_file_range_inline(inode, actual_end, 0,
1157 BTRFS_COMPRESS_NONE, NULL, false);
1160 * We use DO_ACCOUNTING here because we need the
1161 * delalloc_release_metadata to be run _after_ we drop
1162 * our outstanding extent for clearing delalloc for this
1165 extent_clear_unlock_delalloc(inode, start, end,
1167 EXTENT_LOCKED | EXTENT_DELALLOC |
1168 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1169 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1170 PAGE_START_WRITEBACK | PAGE_END_WRITEBACK);
1171 *nr_written = *nr_written +
1172 (end - start + PAGE_SIZE) / PAGE_SIZE;
1175 * locked_page is locked by the caller of
1176 * writepage_delalloc(), not locked by
1177 * __process_pages_contig().
1179 * We can't let __process_pages_contig() to unlock it,
1180 * as it doesn't have any subpage::writers recorded.
1182 * Here we manually unlock the page, since the caller
1183 * can't use page_started to determine if it's an
1184 * inline extent or a compressed extent.
1186 unlock_page(locked_page);
1188 } else if (ret < 0) {
1193 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1194 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
1197 * Relocation relies on the relocated extents to have exactly the same
1198 * size as the original extents. Normally writeback for relocation data
1199 * extents follows a NOCOW path because relocation preallocates the
1200 * extents. However, due to an operation such as scrub turning a block
1201 * group to RO mode, it may fallback to COW mode, so we must make sure
1202 * an extent allocated during COW has exactly the requested size and can
1203 * not be split into smaller extents, otherwise relocation breaks and
1204 * fails during the stage where it updates the bytenr of file extent
1207 if (btrfs_is_data_reloc_root(root))
1208 min_alloc_size = num_bytes;
1210 min_alloc_size = fs_info->sectorsize;
1212 while (num_bytes > 0) {
1213 cur_alloc_size = num_bytes;
1214 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1215 min_alloc_size, 0, alloc_hint,
1219 cur_alloc_size = ins.offset;
1220 extent_reserved = true;
1222 ram_size = ins.offset;
1223 em = create_io_em(inode, start, ins.offset, /* len */
1224 start, /* orig_start */
1225 ins.objectid, /* block_start */
1226 ins.offset, /* block_len */
1227 ins.offset, /* orig_block_len */
1228 ram_size, /* ram_bytes */
1229 BTRFS_COMPRESS_NONE, /* compress_type */
1230 BTRFS_ORDERED_REGULAR /* type */);
1235 free_extent_map(em);
1237 ret = btrfs_add_ordered_extent(inode, start, ram_size, ram_size,
1238 ins.objectid, cur_alloc_size, 0,
1239 1 << BTRFS_ORDERED_REGULAR,
1240 BTRFS_COMPRESS_NONE);
1242 goto out_drop_extent_cache;
1244 if (btrfs_is_data_reloc_root(root)) {
1245 ret = btrfs_reloc_clone_csums(inode, start,
1248 * Only drop cache here, and process as normal.
1250 * We must not allow extent_clear_unlock_delalloc()
1251 * at out_unlock label to free meta of this ordered
1252 * extent, as its meta should be freed by
1253 * btrfs_finish_ordered_io().
1255 * So we must continue until @start is increased to
1256 * skip current ordered extent.
1259 btrfs_drop_extent_cache(inode, start,
1260 start + ram_size - 1, 0);
1263 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1266 * We're not doing compressed IO, don't unlock the first page
1267 * (which the caller expects to stay locked), don't clear any
1268 * dirty bits and don't set any writeback bits
1270 * Do set the Ordered (Private2) bit so we know this page was
1271 * properly setup for writepage.
1273 page_ops = unlock ? PAGE_UNLOCK : 0;
1274 page_ops |= PAGE_SET_ORDERED;
1276 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1278 EXTENT_LOCKED | EXTENT_DELALLOC,
1280 if (num_bytes < cur_alloc_size)
1283 num_bytes -= cur_alloc_size;
1284 alloc_hint = ins.objectid + ins.offset;
1285 start += cur_alloc_size;
1286 extent_reserved = false;
1289 * btrfs_reloc_clone_csums() error, since start is increased
1290 * extent_clear_unlock_delalloc() at out_unlock label won't
1291 * free metadata of current ordered extent, we're OK to exit.
1299 out_drop_extent_cache:
1300 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1302 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1303 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1305 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1306 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1307 page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK;
1309 * If we reserved an extent for our delalloc range (or a subrange) and
1310 * failed to create the respective ordered extent, then it means that
1311 * when we reserved the extent we decremented the extent's size from
1312 * the data space_info's bytes_may_use counter and incremented the
1313 * space_info's bytes_reserved counter by the same amount. We must make
1314 * sure extent_clear_unlock_delalloc() does not try to decrement again
1315 * the data space_info's bytes_may_use counter, therefore we do not pass
1316 * it the flag EXTENT_CLEAR_DATA_RESV.
1318 if (extent_reserved) {
1319 extent_clear_unlock_delalloc(inode, start,
1320 start + cur_alloc_size - 1,
1324 start += cur_alloc_size;
1328 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1329 clear_bits | EXTENT_CLEAR_DATA_RESV,
1335 * work queue call back to started compression on a file and pages
1337 static noinline void async_cow_start(struct btrfs_work *work)
1339 struct async_chunk *async_chunk;
1340 int compressed_extents;
1342 async_chunk = container_of(work, struct async_chunk, work);
1344 compressed_extents = compress_file_range(async_chunk);
1345 if (compressed_extents == 0) {
1346 btrfs_add_delayed_iput(async_chunk->inode);
1347 async_chunk->inode = NULL;
1352 * work queue call back to submit previously compressed pages
1354 static noinline void async_cow_submit(struct btrfs_work *work)
1356 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1358 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1359 unsigned long nr_pages;
1361 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1365 * ->inode could be NULL if async_chunk_start has failed to compress,
1366 * in which case we don't have anything to submit, yet we need to
1367 * always adjust ->async_delalloc_pages as its paired with the init
1368 * happening in cow_file_range_async
1370 if (async_chunk->inode)
1371 submit_compressed_extents(async_chunk);
1373 /* atomic_sub_return implies a barrier */
1374 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1376 cond_wake_up_nomb(&fs_info->async_submit_wait);
1379 static noinline void async_cow_free(struct btrfs_work *work)
1381 struct async_chunk *async_chunk;
1382 struct async_cow *async_cow;
1384 async_chunk = container_of(work, struct async_chunk, work);
1385 if (async_chunk->inode)
1386 btrfs_add_delayed_iput(async_chunk->inode);
1387 if (async_chunk->blkcg_css)
1388 css_put(async_chunk->blkcg_css);
1390 async_cow = async_chunk->async_cow;
1391 if (atomic_dec_and_test(&async_cow->num_chunks))
1395 static int cow_file_range_async(struct btrfs_inode *inode,
1396 struct writeback_control *wbc,
1397 struct page *locked_page,
1398 u64 start, u64 end, int *page_started,
1399 unsigned long *nr_written)
1401 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1402 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1403 struct async_cow *ctx;
1404 struct async_chunk *async_chunk;
1405 unsigned long nr_pages;
1407 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1409 bool should_compress;
1411 const unsigned int write_flags = wbc_to_write_flags(wbc);
1413 unlock_extent(&inode->io_tree, start, end);
1415 if (inode->flags & BTRFS_INODE_NOCOMPRESS &&
1416 !btrfs_test_opt(fs_info, FORCE_COMPRESS)) {
1418 should_compress = false;
1420 should_compress = true;
1423 nofs_flag = memalloc_nofs_save();
1424 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1425 memalloc_nofs_restore(nofs_flag);
1428 unsigned clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC |
1429 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1430 EXTENT_DO_ACCOUNTING;
1431 unsigned long page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK |
1432 PAGE_END_WRITEBACK | PAGE_SET_ERROR;
1434 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1435 clear_bits, page_ops);
1439 async_chunk = ctx->chunks;
1440 atomic_set(&ctx->num_chunks, num_chunks);
1442 for (i = 0; i < num_chunks; i++) {
1443 if (should_compress)
1444 cur_end = min(end, start + SZ_512K - 1);
1449 * igrab is called higher up in the call chain, take only the
1450 * lightweight reference for the callback lifetime
1452 ihold(&inode->vfs_inode);
1453 async_chunk[i].async_cow = ctx;
1454 async_chunk[i].inode = &inode->vfs_inode;
1455 async_chunk[i].start = start;
1456 async_chunk[i].end = cur_end;
1457 async_chunk[i].write_flags = write_flags;
1458 INIT_LIST_HEAD(&async_chunk[i].extents);
1461 * The locked_page comes all the way from writepage and its
1462 * the original page we were actually given. As we spread
1463 * this large delalloc region across multiple async_chunk
1464 * structs, only the first struct needs a pointer to locked_page
1466 * This way we don't need racey decisions about who is supposed
1471 * Depending on the compressibility, the pages might or
1472 * might not go through async. We want all of them to
1473 * be accounted against wbc once. Let's do it here
1474 * before the paths diverge. wbc accounting is used
1475 * only for foreign writeback detection and doesn't
1476 * need full accuracy. Just account the whole thing
1477 * against the first page.
1479 wbc_account_cgroup_owner(wbc, locked_page,
1481 async_chunk[i].locked_page = locked_page;
1484 async_chunk[i].locked_page = NULL;
1487 if (blkcg_css != blkcg_root_css) {
1489 async_chunk[i].blkcg_css = blkcg_css;
1491 async_chunk[i].blkcg_css = NULL;
1494 btrfs_init_work(&async_chunk[i].work, async_cow_start,
1495 async_cow_submit, async_cow_free);
1497 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1498 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1500 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1502 *nr_written += nr_pages;
1503 start = cur_end + 1;
1509 static noinline int run_delalloc_zoned(struct btrfs_inode *inode,
1510 struct page *locked_page, u64 start,
1511 u64 end, int *page_started,
1512 unsigned long *nr_written)
1516 ret = cow_file_range(inode, locked_page, start, end, page_started,
1524 __set_page_dirty_nobuffers(locked_page);
1525 account_page_redirty(locked_page);
1526 extent_write_locked_range(&inode->vfs_inode, start, end);
1532 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1533 u64 bytenr, u64 num_bytes)
1535 struct btrfs_root *csum_root = btrfs_csum_root(fs_info, bytenr);
1536 struct btrfs_ordered_sum *sums;
1540 ret = btrfs_lookup_csums_range(csum_root, bytenr,
1541 bytenr + num_bytes - 1, &list, 0);
1542 if (ret == 0 && list_empty(&list))
1545 while (!list_empty(&list)) {
1546 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1547 list_del(&sums->list);
1555 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1556 const u64 start, const u64 end,
1557 int *page_started, unsigned long *nr_written)
1559 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1560 const bool is_reloc_ino = btrfs_is_data_reloc_root(inode->root);
1561 const u64 range_bytes = end + 1 - start;
1562 struct extent_io_tree *io_tree = &inode->io_tree;
1563 u64 range_start = start;
1567 * If EXTENT_NORESERVE is set it means that when the buffered write was
1568 * made we had not enough available data space and therefore we did not
1569 * reserve data space for it, since we though we could do NOCOW for the
1570 * respective file range (either there is prealloc extent or the inode
1571 * has the NOCOW bit set).
1573 * However when we need to fallback to COW mode (because for example the
1574 * block group for the corresponding extent was turned to RO mode by a
1575 * scrub or relocation) we need to do the following:
1577 * 1) We increment the bytes_may_use counter of the data space info.
1578 * If COW succeeds, it allocates a new data extent and after doing
1579 * that it decrements the space info's bytes_may_use counter and
1580 * increments its bytes_reserved counter by the same amount (we do
1581 * this at btrfs_add_reserved_bytes()). So we need to increment the
1582 * bytes_may_use counter to compensate (when space is reserved at
1583 * buffered write time, the bytes_may_use counter is incremented);
1585 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1586 * that if the COW path fails for any reason, it decrements (through
1587 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1588 * data space info, which we incremented in the step above.
1590 * If we need to fallback to cow and the inode corresponds to a free
1591 * space cache inode or an inode of the data relocation tree, we must
1592 * also increment bytes_may_use of the data space_info for the same
1593 * reason. Space caches and relocated data extents always get a prealloc
1594 * extent for them, however scrub or balance may have set the block
1595 * group that contains that extent to RO mode and therefore force COW
1596 * when starting writeback.
1598 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1599 EXTENT_NORESERVE, 0);
1600 if (count > 0 || is_space_ino || is_reloc_ino) {
1602 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1603 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1605 if (is_space_ino || is_reloc_ino)
1606 bytes = range_bytes;
1608 spin_lock(&sinfo->lock);
1609 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1610 spin_unlock(&sinfo->lock);
1613 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1617 return cow_file_range(inode, locked_page, start, end, page_started,
1622 * when nowcow writeback call back. This checks for snapshots or COW copies
1623 * of the extents that exist in the file, and COWs the file as required.
1625 * If no cow copies or snapshots exist, we write directly to the existing
1628 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
1629 struct page *locked_page,
1630 const u64 start, const u64 end,
1632 unsigned long *nr_written)
1634 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1635 struct btrfs_root *root = inode->root;
1636 struct btrfs_path *path;
1637 u64 cow_start = (u64)-1;
1638 u64 cur_offset = start;
1640 bool check_prev = true;
1641 const bool freespace_inode = btrfs_is_free_space_inode(inode);
1642 u64 ino = btrfs_ino(inode);
1644 u64 disk_bytenr = 0;
1645 const bool force = inode->flags & BTRFS_INODE_NODATACOW;
1647 path = btrfs_alloc_path();
1649 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1650 EXTENT_LOCKED | EXTENT_DELALLOC |
1651 EXTENT_DO_ACCOUNTING |
1652 EXTENT_DEFRAG, PAGE_UNLOCK |
1653 PAGE_START_WRITEBACK |
1654 PAGE_END_WRITEBACK);
1659 struct btrfs_key found_key;
1660 struct btrfs_file_extent_item *fi;
1661 struct extent_buffer *leaf;
1671 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1677 * If there is no extent for our range when doing the initial
1678 * search, then go back to the previous slot as it will be the
1679 * one containing the search offset
1681 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1682 leaf = path->nodes[0];
1683 btrfs_item_key_to_cpu(leaf, &found_key,
1684 path->slots[0] - 1);
1685 if (found_key.objectid == ino &&
1686 found_key.type == BTRFS_EXTENT_DATA_KEY)
1691 /* Go to next leaf if we have exhausted the current one */
1692 leaf = path->nodes[0];
1693 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1694 ret = btrfs_next_leaf(root, path);
1696 if (cow_start != (u64)-1)
1697 cur_offset = cow_start;
1702 leaf = path->nodes[0];
1705 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1707 /* Didn't find anything for our INO */
1708 if (found_key.objectid > ino)
1711 * Keep searching until we find an EXTENT_ITEM or there are no
1712 * more extents for this inode
1714 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1715 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1720 /* Found key is not EXTENT_DATA_KEY or starts after req range */
1721 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1722 found_key.offset > end)
1726 * If the found extent starts after requested offset, then
1727 * adjust extent_end to be right before this extent begins
1729 if (found_key.offset > cur_offset) {
1730 extent_end = found_key.offset;
1736 * Found extent which begins before our range and potentially
1739 fi = btrfs_item_ptr(leaf, path->slots[0],
1740 struct btrfs_file_extent_item);
1741 extent_type = btrfs_file_extent_type(leaf, fi);
1743 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1744 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1745 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1746 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1747 extent_offset = btrfs_file_extent_offset(leaf, fi);
1748 extent_end = found_key.offset +
1749 btrfs_file_extent_num_bytes(leaf, fi);
1751 btrfs_file_extent_disk_num_bytes(leaf, fi);
1753 * If the extent we got ends before our current offset,
1754 * skip to the next extent.
1756 if (extent_end <= cur_offset) {
1761 if (disk_bytenr == 0)
1763 /* Skip compressed/encrypted/encoded extents */
1764 if (btrfs_file_extent_compression(leaf, fi) ||
1765 btrfs_file_extent_encryption(leaf, fi) ||
1766 btrfs_file_extent_other_encoding(leaf, fi))
1769 * If extent is created before the last volume's snapshot
1770 * this implies the extent is shared, hence we can't do
1771 * nocow. This is the same check as in
1772 * btrfs_cross_ref_exist but without calling
1773 * btrfs_search_slot.
1775 if (!freespace_inode &&
1776 btrfs_file_extent_generation(leaf, fi) <=
1777 btrfs_root_last_snapshot(&root->root_item))
1779 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1783 * The following checks can be expensive, as they need to
1784 * take other locks and do btree or rbtree searches, so
1785 * release the path to avoid blocking other tasks for too
1788 btrfs_release_path(path);
1790 ret = btrfs_cross_ref_exist(root, ino,
1792 extent_offset, disk_bytenr, false);
1795 * ret could be -EIO if the above fails to read
1799 if (cow_start != (u64)-1)
1800 cur_offset = cow_start;
1804 WARN_ON_ONCE(freespace_inode);
1807 disk_bytenr += extent_offset;
1808 disk_bytenr += cur_offset - found_key.offset;
1809 num_bytes = min(end + 1, extent_end) - cur_offset;
1811 * If there are pending snapshots for this root, we
1812 * fall into common COW way
1814 if (!freespace_inode && atomic_read(&root->snapshot_force_cow))
1817 * force cow if csum exists in the range.
1818 * this ensure that csum for a given extent are
1819 * either valid or do not exist.
1821 ret = csum_exist_in_range(fs_info, disk_bytenr,
1825 * ret could be -EIO if the above fails to read
1829 if (cow_start != (u64)-1)
1830 cur_offset = cow_start;
1833 WARN_ON_ONCE(freespace_inode);
1836 /* If the extent's block group is RO, we must COW */
1837 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr))
1840 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1841 extent_end = found_key.offset + ram_bytes;
1842 extent_end = ALIGN(extent_end, fs_info->sectorsize);
1843 /* Skip extents outside of our requested range */
1844 if (extent_end <= start) {
1849 /* If this triggers then we have a memory corruption */
1854 * If nocow is false then record the beginning of the range
1855 * that needs to be COWed
1858 if (cow_start == (u64)-1)
1859 cow_start = cur_offset;
1860 cur_offset = extent_end;
1861 if (cur_offset > end)
1863 if (!path->nodes[0])
1870 * COW range from cow_start to found_key.offset - 1. As the key
1871 * will contain the beginning of the first extent that can be
1872 * NOCOW, following one which needs to be COW'ed
1874 if (cow_start != (u64)-1) {
1875 ret = fallback_to_cow(inode, locked_page,
1876 cow_start, found_key.offset - 1,
1877 page_started, nr_written);
1880 cow_start = (u64)-1;
1883 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1884 u64 orig_start = found_key.offset - extent_offset;
1885 struct extent_map *em;
1887 em = create_io_em(inode, cur_offset, num_bytes,
1889 disk_bytenr, /* block_start */
1890 num_bytes, /* block_len */
1891 disk_num_bytes, /* orig_block_len */
1892 ram_bytes, BTRFS_COMPRESS_NONE,
1893 BTRFS_ORDERED_PREALLOC);
1898 free_extent_map(em);
1899 ret = btrfs_add_ordered_extent(inode,
1900 cur_offset, num_bytes, num_bytes,
1901 disk_bytenr, num_bytes, 0,
1902 1 << BTRFS_ORDERED_PREALLOC,
1903 BTRFS_COMPRESS_NONE);
1905 btrfs_drop_extent_cache(inode, cur_offset,
1906 cur_offset + num_bytes - 1,
1911 ret = btrfs_add_ordered_extent(inode, cur_offset,
1912 num_bytes, num_bytes,
1913 disk_bytenr, num_bytes,
1915 1 << BTRFS_ORDERED_NOCOW,
1916 BTRFS_COMPRESS_NONE);
1922 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1925 if (btrfs_is_data_reloc_root(root))
1927 * Error handled later, as we must prevent
1928 * extent_clear_unlock_delalloc() in error handler
1929 * from freeing metadata of created ordered extent.
1931 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1934 extent_clear_unlock_delalloc(inode, cur_offset,
1935 cur_offset + num_bytes - 1,
1936 locked_page, EXTENT_LOCKED |
1938 EXTENT_CLEAR_DATA_RESV,
1939 PAGE_UNLOCK | PAGE_SET_ORDERED);
1941 cur_offset = extent_end;
1944 * btrfs_reloc_clone_csums() error, now we're OK to call error
1945 * handler, as metadata for created ordered extent will only
1946 * be freed by btrfs_finish_ordered_io().
1950 if (cur_offset > end)
1953 btrfs_release_path(path);
1955 if (cur_offset <= end && cow_start == (u64)-1)
1956 cow_start = cur_offset;
1958 if (cow_start != (u64)-1) {
1960 ret = fallback_to_cow(inode, locked_page, cow_start, end,
1961 page_started, nr_written);
1968 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1970 if (ret && cur_offset < end)
1971 extent_clear_unlock_delalloc(inode, cur_offset, end,
1972 locked_page, EXTENT_LOCKED |
1973 EXTENT_DELALLOC | EXTENT_DEFRAG |
1974 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1975 PAGE_START_WRITEBACK |
1976 PAGE_END_WRITEBACK);
1977 btrfs_free_path(path);
1981 static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end)
1983 if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) {
1984 if (inode->defrag_bytes &&
1985 test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG,
1994 * Function to process delayed allocation (create CoW) for ranges which are
1995 * being touched for the first time.
1997 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
1998 u64 start, u64 end, int *page_started, unsigned long *nr_written,
1999 struct writeback_control *wbc)
2002 const bool zoned = btrfs_is_zoned(inode->root->fs_info);
2005 * The range must cover part of the @locked_page, or the returned
2006 * @page_started can confuse the caller.
2008 ASSERT(!(end <= page_offset(locked_page) ||
2009 start >= page_offset(locked_page) + PAGE_SIZE));
2011 if (should_nocow(inode, start, end)) {
2013 * Normally on a zoned device we're only doing COW writes, but
2014 * in case of relocation on a zoned filesystem we have taken
2015 * precaution, that we're only writing sequentially. It's safe
2016 * to use run_delalloc_nocow() here, like for regular
2017 * preallocated inodes.
2019 ASSERT(!zoned || btrfs_is_data_reloc_root(inode->root));
2020 ret = run_delalloc_nocow(inode, locked_page, start, end,
2021 page_started, nr_written);
2022 } else if (!inode_can_compress(inode) ||
2023 !inode_need_compress(inode, start, end)) {
2025 ret = run_delalloc_zoned(inode, locked_page, start, end,
2026 page_started, nr_written);
2028 ret = cow_file_range(inode, locked_page, start, end,
2029 page_started, nr_written, 1);
2031 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
2032 ret = cow_file_range_async(inode, wbc, locked_page, start, end,
2033 page_started, nr_written);
2037 btrfs_cleanup_ordered_extents(inode, locked_page, start,
2042 void btrfs_split_delalloc_extent(struct inode *inode,
2043 struct extent_state *orig, u64 split)
2047 /* not delalloc, ignore it */
2048 if (!(orig->state & EXTENT_DELALLOC))
2051 size = orig->end - orig->start + 1;
2052 if (size > BTRFS_MAX_EXTENT_SIZE) {
2057 * See the explanation in btrfs_merge_delalloc_extent, the same
2058 * applies here, just in reverse.
2060 new_size = orig->end - split + 1;
2061 num_extents = count_max_extents(new_size);
2062 new_size = split - orig->start;
2063 num_extents += count_max_extents(new_size);
2064 if (count_max_extents(size) >= num_extents)
2068 spin_lock(&BTRFS_I(inode)->lock);
2069 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
2070 spin_unlock(&BTRFS_I(inode)->lock);
2074 * Handle merged delayed allocation extents so we can keep track of new extents
2075 * that are just merged onto old extents, such as when we are doing sequential
2076 * writes, so we can properly account for the metadata space we'll need.
2078 void btrfs_merge_delalloc_extent(struct inode *inode, struct extent_state *new,
2079 struct extent_state *other)
2081 u64 new_size, old_size;
2084 /* not delalloc, ignore it */
2085 if (!(other->state & EXTENT_DELALLOC))
2088 if (new->start > other->start)
2089 new_size = new->end - other->start + 1;
2091 new_size = other->end - new->start + 1;
2093 /* we're not bigger than the max, unreserve the space and go */
2094 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
2095 spin_lock(&BTRFS_I(inode)->lock);
2096 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
2097 spin_unlock(&BTRFS_I(inode)->lock);
2102 * We have to add up either side to figure out how many extents were
2103 * accounted for before we merged into one big extent. If the number of
2104 * extents we accounted for is <= the amount we need for the new range
2105 * then we can return, otherwise drop. Think of it like this
2109 * So we've grown the extent by a MAX_SIZE extent, this would mean we
2110 * need 2 outstanding extents, on one side we have 1 and the other side
2111 * we have 1 so they are == and we can return. But in this case
2113 * [MAX_SIZE+4k][MAX_SIZE+4k]
2115 * Each range on their own accounts for 2 extents, but merged together
2116 * they are only 3 extents worth of accounting, so we need to drop in
2119 old_size = other->end - other->start + 1;
2120 num_extents = count_max_extents(old_size);
2121 old_size = new->end - new->start + 1;
2122 num_extents += count_max_extents(old_size);
2123 if (count_max_extents(new_size) >= num_extents)
2126 spin_lock(&BTRFS_I(inode)->lock);
2127 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
2128 spin_unlock(&BTRFS_I(inode)->lock);
2131 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
2132 struct inode *inode)
2134 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2136 spin_lock(&root->delalloc_lock);
2137 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
2138 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
2139 &root->delalloc_inodes);
2140 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2141 &BTRFS_I(inode)->runtime_flags);
2142 root->nr_delalloc_inodes++;
2143 if (root->nr_delalloc_inodes == 1) {
2144 spin_lock(&fs_info->delalloc_root_lock);
2145 BUG_ON(!list_empty(&root->delalloc_root));
2146 list_add_tail(&root->delalloc_root,
2147 &fs_info->delalloc_roots);
2148 spin_unlock(&fs_info->delalloc_root_lock);
2151 spin_unlock(&root->delalloc_lock);
2155 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
2156 struct btrfs_inode *inode)
2158 struct btrfs_fs_info *fs_info = root->fs_info;
2160 if (!list_empty(&inode->delalloc_inodes)) {
2161 list_del_init(&inode->delalloc_inodes);
2162 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2163 &inode->runtime_flags);
2164 root->nr_delalloc_inodes--;
2165 if (!root->nr_delalloc_inodes) {
2166 ASSERT(list_empty(&root->delalloc_inodes));
2167 spin_lock(&fs_info->delalloc_root_lock);
2168 BUG_ON(list_empty(&root->delalloc_root));
2169 list_del_init(&root->delalloc_root);
2170 spin_unlock(&fs_info->delalloc_root_lock);
2175 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
2176 struct btrfs_inode *inode)
2178 spin_lock(&root->delalloc_lock);
2179 __btrfs_del_delalloc_inode(root, inode);
2180 spin_unlock(&root->delalloc_lock);
2184 * Properly track delayed allocation bytes in the inode and to maintain the
2185 * list of inodes that have pending delalloc work to be done.
2187 void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state,
2190 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2192 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
2195 * set_bit and clear bit hooks normally require _irqsave/restore
2196 * but in this case, we are only testing for the DELALLOC
2197 * bit, which is only set or cleared with irqs on
2199 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2200 struct btrfs_root *root = BTRFS_I(inode)->root;
2201 u64 len = state->end + 1 - state->start;
2202 u32 num_extents = count_max_extents(len);
2203 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
2205 spin_lock(&BTRFS_I(inode)->lock);
2206 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
2207 spin_unlock(&BTRFS_I(inode)->lock);
2209 /* For sanity tests */
2210 if (btrfs_is_testing(fs_info))
2213 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2214 fs_info->delalloc_batch);
2215 spin_lock(&BTRFS_I(inode)->lock);
2216 BTRFS_I(inode)->delalloc_bytes += len;
2217 if (*bits & EXTENT_DEFRAG)
2218 BTRFS_I(inode)->defrag_bytes += len;
2219 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2220 &BTRFS_I(inode)->runtime_flags))
2221 btrfs_add_delalloc_inodes(root, inode);
2222 spin_unlock(&BTRFS_I(inode)->lock);
2225 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2226 (*bits & EXTENT_DELALLOC_NEW)) {
2227 spin_lock(&BTRFS_I(inode)->lock);
2228 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
2230 spin_unlock(&BTRFS_I(inode)->lock);
2235 * Once a range is no longer delalloc this function ensures that proper
2236 * accounting happens.
2238 void btrfs_clear_delalloc_extent(struct inode *vfs_inode,
2239 struct extent_state *state, unsigned *bits)
2241 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
2242 struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb);
2243 u64 len = state->end + 1 - state->start;
2244 u32 num_extents = count_max_extents(len);
2246 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
2247 spin_lock(&inode->lock);
2248 inode->defrag_bytes -= len;
2249 spin_unlock(&inode->lock);
2253 * set_bit and clear bit hooks normally require _irqsave/restore
2254 * but in this case, we are only testing for the DELALLOC
2255 * bit, which is only set or cleared with irqs on
2257 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2258 struct btrfs_root *root = inode->root;
2259 bool do_list = !btrfs_is_free_space_inode(inode);
2261 spin_lock(&inode->lock);
2262 btrfs_mod_outstanding_extents(inode, -num_extents);
2263 spin_unlock(&inode->lock);
2266 * We don't reserve metadata space for space cache inodes so we
2267 * don't need to call delalloc_release_metadata if there is an
2270 if (*bits & EXTENT_CLEAR_META_RESV &&
2271 root != fs_info->tree_root)
2272 btrfs_delalloc_release_metadata(inode, len, false);
2274 /* For sanity tests. */
2275 if (btrfs_is_testing(fs_info))
2278 if (!btrfs_is_data_reloc_root(root) &&
2279 do_list && !(state->state & EXTENT_NORESERVE) &&
2280 (*bits & EXTENT_CLEAR_DATA_RESV))
2281 btrfs_free_reserved_data_space_noquota(fs_info, len);
2283 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2284 fs_info->delalloc_batch);
2285 spin_lock(&inode->lock);
2286 inode->delalloc_bytes -= len;
2287 if (do_list && inode->delalloc_bytes == 0 &&
2288 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2289 &inode->runtime_flags))
2290 btrfs_del_delalloc_inode(root, inode);
2291 spin_unlock(&inode->lock);
2294 if ((state->state & EXTENT_DELALLOC_NEW) &&
2295 (*bits & EXTENT_DELALLOC_NEW)) {
2296 spin_lock(&inode->lock);
2297 ASSERT(inode->new_delalloc_bytes >= len);
2298 inode->new_delalloc_bytes -= len;
2299 if (*bits & EXTENT_ADD_INODE_BYTES)
2300 inode_add_bytes(&inode->vfs_inode, len);
2301 spin_unlock(&inode->lock);
2306 * in order to insert checksums into the metadata in large chunks,
2307 * we wait until bio submission time. All the pages in the bio are
2308 * checksummed and sums are attached onto the ordered extent record.
2310 * At IO completion time the cums attached on the ordered extent record
2311 * are inserted into the btree
2313 static blk_status_t btrfs_submit_bio_start(struct inode *inode, struct bio *bio,
2314 u64 dio_file_offset)
2316 return btrfs_csum_one_bio(BTRFS_I(inode), bio, (u64)-1, false);
2320 * Split an extent_map at [start, start + len]
2322 * This function is intended to be used only for extract_ordered_extent().
2324 static int split_zoned_em(struct btrfs_inode *inode, u64 start, u64 len,
2327 struct extent_map_tree *em_tree = &inode->extent_tree;
2328 struct extent_map *em;
2329 struct extent_map *split_pre = NULL;
2330 struct extent_map *split_mid = NULL;
2331 struct extent_map *split_post = NULL;
2333 unsigned long flags;
2336 if (pre == 0 && post == 0)
2339 split_pre = alloc_extent_map();
2341 split_mid = alloc_extent_map();
2343 split_post = alloc_extent_map();
2344 if (!split_pre || (pre && !split_mid) || (post && !split_post)) {
2349 ASSERT(pre + post < len);
2351 lock_extent(&inode->io_tree, start, start + len - 1);
2352 write_lock(&em_tree->lock);
2353 em = lookup_extent_mapping(em_tree, start, len);
2359 ASSERT(em->len == len);
2360 ASSERT(!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags));
2361 ASSERT(em->block_start < EXTENT_MAP_LAST_BYTE);
2362 ASSERT(test_bit(EXTENT_FLAG_PINNED, &em->flags));
2363 ASSERT(!test_bit(EXTENT_FLAG_LOGGING, &em->flags));
2364 ASSERT(!list_empty(&em->list));
2367 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
2369 /* First, replace the em with a new extent_map starting from * em->start */
2370 split_pre->start = em->start;
2371 split_pre->len = (pre ? pre : em->len - post);
2372 split_pre->orig_start = split_pre->start;
2373 split_pre->block_start = em->block_start;
2374 split_pre->block_len = split_pre->len;
2375 split_pre->orig_block_len = split_pre->block_len;
2376 split_pre->ram_bytes = split_pre->len;
2377 split_pre->flags = flags;
2378 split_pre->compress_type = em->compress_type;
2379 split_pre->generation = em->generation;
2381 replace_extent_mapping(em_tree, em, split_pre, 1);
2384 * Now we only have an extent_map at:
2385 * [em->start, em->start + pre] if pre != 0
2386 * [em->start, em->start + em->len - post] if pre == 0
2390 /* Insert the middle extent_map */
2391 split_mid->start = em->start + pre;
2392 split_mid->len = em->len - pre - post;
2393 split_mid->orig_start = split_mid->start;
2394 split_mid->block_start = em->block_start + pre;
2395 split_mid->block_len = split_mid->len;
2396 split_mid->orig_block_len = split_mid->block_len;
2397 split_mid->ram_bytes = split_mid->len;
2398 split_mid->flags = flags;
2399 split_mid->compress_type = em->compress_type;
2400 split_mid->generation = em->generation;
2401 add_extent_mapping(em_tree, split_mid, 1);
2405 split_post->start = em->start + em->len - post;
2406 split_post->len = post;
2407 split_post->orig_start = split_post->start;
2408 split_post->block_start = em->block_start + em->len - post;
2409 split_post->block_len = split_post->len;
2410 split_post->orig_block_len = split_post->block_len;
2411 split_post->ram_bytes = split_post->len;
2412 split_post->flags = flags;
2413 split_post->compress_type = em->compress_type;
2414 split_post->generation = em->generation;
2415 add_extent_mapping(em_tree, split_post, 1);
2419 free_extent_map(em);
2420 /* Once for the tree */
2421 free_extent_map(em);
2424 write_unlock(&em_tree->lock);
2425 unlock_extent(&inode->io_tree, start, start + len - 1);
2427 free_extent_map(split_pre);
2428 free_extent_map(split_mid);
2429 free_extent_map(split_post);
2434 static blk_status_t extract_ordered_extent(struct btrfs_inode *inode,
2435 struct bio *bio, loff_t file_offset)
2437 struct btrfs_ordered_extent *ordered;
2438 u64 start = (u64)bio->bi_iter.bi_sector << SECTOR_SHIFT;
2440 u64 len = bio->bi_iter.bi_size;
2441 u64 end = start + len;
2446 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
2447 if (WARN_ON_ONCE(!ordered))
2448 return BLK_STS_IOERR;
2450 /* No need to split */
2451 if (ordered->disk_num_bytes == len)
2454 /* We cannot split once end_bio'd ordered extent */
2455 if (WARN_ON_ONCE(ordered->bytes_left != ordered->disk_num_bytes)) {
2460 /* We cannot split a compressed ordered extent */
2461 if (WARN_ON_ONCE(ordered->disk_num_bytes != ordered->num_bytes)) {
2466 ordered_end = ordered->disk_bytenr + ordered->disk_num_bytes;
2467 /* bio must be in one ordered extent */
2468 if (WARN_ON_ONCE(start < ordered->disk_bytenr || end > ordered_end)) {
2473 /* Checksum list should be empty */
2474 if (WARN_ON_ONCE(!list_empty(&ordered->list))) {
2479 file_len = ordered->num_bytes;
2480 pre = start - ordered->disk_bytenr;
2481 post = ordered_end - end;
2483 ret = btrfs_split_ordered_extent(ordered, pre, post);
2486 ret = split_zoned_em(inode, file_offset, file_len, pre, post);
2489 btrfs_put_ordered_extent(ordered);
2491 return errno_to_blk_status(ret);
2495 * extent_io.c submission hook. This does the right thing for csum calculation
2496 * on write, or reading the csums from the tree before a read.
2498 * Rules about async/sync submit,
2499 * a) read: sync submit
2501 * b) write without checksum: sync submit
2503 * c) write with checksum:
2504 * c-1) if bio is issued by fsync: sync submit
2505 * (sync_writers != 0)
2507 * c-2) if root is reloc root: sync submit
2508 * (only in case of buffered IO)
2510 * c-3) otherwise: async submit
2512 blk_status_t btrfs_submit_data_bio(struct inode *inode, struct bio *bio,
2513 int mirror_num, unsigned long bio_flags)
2516 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2517 struct btrfs_root *root = BTRFS_I(inode)->root;
2518 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
2519 blk_status_t ret = 0;
2521 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
2523 skip_sum = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) ||
2524 test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2526 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
2527 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
2529 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
2530 struct page *page = bio_first_bvec_all(bio)->bv_page;
2531 loff_t file_offset = page_offset(page);
2533 ret = extract_ordered_extent(BTRFS_I(inode), bio, file_offset);
2538 if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
2539 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
2543 if (bio_flags & EXTENT_BIO_COMPRESSED) {
2545 * btrfs_submit_compressed_read will handle completing
2546 * the bio if there were any errors, so just return
2549 ret = btrfs_submit_compressed_read(inode, bio,
2555 * Lookup bio sums does extra checks around whether we
2556 * need to csum or not, which is why we ignore skip_sum
2559 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
2564 } else if (async && !skip_sum) {
2565 /* csum items have already been cloned */
2566 if (btrfs_is_data_reloc_root(root))
2568 /* we're doing a write, do the async checksumming */
2569 ret = btrfs_wq_submit_bio(inode, bio, mirror_num, bio_flags,
2570 0, btrfs_submit_bio_start);
2572 } else if (!skip_sum) {
2573 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, (u64)-1, false);
2579 ret = btrfs_map_bio(fs_info, bio, mirror_num);
2583 bio->bi_status = ret;
2591 * given a list of ordered sums record them in the inode. This happens
2592 * at IO completion time based on sums calculated at bio submission time.
2594 static int add_pending_csums(struct btrfs_trans_handle *trans,
2595 struct list_head *list)
2597 struct btrfs_ordered_sum *sum;
2598 struct btrfs_root *csum_root = NULL;
2601 list_for_each_entry(sum, list, list) {
2602 trans->adding_csums = true;
2604 csum_root = btrfs_csum_root(trans->fs_info,
2606 ret = btrfs_csum_file_blocks(trans, csum_root, sum);
2607 trans->adding_csums = false;
2614 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2617 struct extent_state **cached_state)
2619 u64 search_start = start;
2620 const u64 end = start + len - 1;
2622 while (search_start < end) {
2623 const u64 search_len = end - search_start + 1;
2624 struct extent_map *em;
2628 em = btrfs_get_extent(inode, NULL, 0, search_start, search_len);
2632 if (em->block_start != EXTENT_MAP_HOLE)
2636 if (em->start < search_start)
2637 em_len -= search_start - em->start;
2638 if (em_len > search_len)
2639 em_len = search_len;
2641 ret = set_extent_bit(&inode->io_tree, search_start,
2642 search_start + em_len - 1,
2643 EXTENT_DELALLOC_NEW, 0, NULL, cached_state,
2646 search_start = extent_map_end(em);
2647 free_extent_map(em);
2654 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2655 unsigned int extra_bits,
2656 struct extent_state **cached_state)
2658 WARN_ON(PAGE_ALIGNED(end));
2660 if (start >= i_size_read(&inode->vfs_inode) &&
2661 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2663 * There can't be any extents following eof in this case so just
2664 * set the delalloc new bit for the range directly.
2666 extra_bits |= EXTENT_DELALLOC_NEW;
2670 ret = btrfs_find_new_delalloc_bytes(inode, start,
2677 return set_extent_delalloc(&inode->io_tree, start, end, extra_bits,
2681 /* see btrfs_writepage_start_hook for details on why this is required */
2682 struct btrfs_writepage_fixup {
2684 struct inode *inode;
2685 struct btrfs_work work;
2688 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2690 struct btrfs_writepage_fixup *fixup;
2691 struct btrfs_ordered_extent *ordered;
2692 struct extent_state *cached_state = NULL;
2693 struct extent_changeset *data_reserved = NULL;
2695 struct btrfs_inode *inode;
2699 bool free_delalloc_space = true;
2701 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2703 inode = BTRFS_I(fixup->inode);
2704 page_start = page_offset(page);
2705 page_end = page_offset(page) + PAGE_SIZE - 1;
2708 * This is similar to page_mkwrite, we need to reserve the space before
2709 * we take the page lock.
2711 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2717 * Before we queued this fixup, we took a reference on the page.
2718 * page->mapping may go NULL, but it shouldn't be moved to a different
2721 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2723 * Unfortunately this is a little tricky, either
2725 * 1) We got here and our page had already been dealt with and
2726 * we reserved our space, thus ret == 0, so we need to just
2727 * drop our space reservation and bail. This can happen the
2728 * first time we come into the fixup worker, or could happen
2729 * while waiting for the ordered extent.
2730 * 2) Our page was already dealt with, but we happened to get an
2731 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2732 * this case we obviously don't have anything to release, but
2733 * because the page was already dealt with we don't want to
2734 * mark the page with an error, so make sure we're resetting
2735 * ret to 0. This is why we have this check _before_ the ret
2736 * check, because we do not want to have a surprise ENOSPC
2737 * when the page was already properly dealt with.
2740 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2741 btrfs_delalloc_release_space(inode, data_reserved,
2742 page_start, PAGE_SIZE,
2750 * We can't mess with the page state unless it is locked, so now that
2751 * it is locked bail if we failed to make our space reservation.
2756 lock_extent_bits(&inode->io_tree, page_start, page_end, &cached_state);
2758 /* already ordered? We're done */
2759 if (PageOrdered(page))
2762 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2764 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2767 btrfs_start_ordered_extent(ordered, 1);
2768 btrfs_put_ordered_extent(ordered);
2772 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2778 * Everything went as planned, we're now the owner of a dirty page with
2779 * delayed allocation bits set and space reserved for our COW
2782 * The page was dirty when we started, nothing should have cleaned it.
2784 BUG_ON(!PageDirty(page));
2785 free_delalloc_space = false;
2787 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2788 if (free_delalloc_space)
2789 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2791 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2796 * We hit ENOSPC or other errors. Update the mapping and page
2797 * to reflect the errors and clean the page.
2799 mapping_set_error(page->mapping, ret);
2800 end_extent_writepage(page, ret, page_start, page_end);
2801 clear_page_dirty_for_io(page);
2804 btrfs_page_clear_checked(inode->root->fs_info, page, page_start, PAGE_SIZE);
2808 extent_changeset_free(data_reserved);
2810 * As a precaution, do a delayed iput in case it would be the last iput
2811 * that could need flushing space. Recursing back to fixup worker would
2814 btrfs_add_delayed_iput(&inode->vfs_inode);
2818 * There are a few paths in the higher layers of the kernel that directly
2819 * set the page dirty bit without asking the filesystem if it is a
2820 * good idea. This causes problems because we want to make sure COW
2821 * properly happens and the data=ordered rules are followed.
2823 * In our case any range that doesn't have the ORDERED bit set
2824 * hasn't been properly setup for IO. We kick off an async process
2825 * to fix it up. The async helper will wait for ordered extents, set
2826 * the delalloc bit and make it safe to write the page.
2828 int btrfs_writepage_cow_fixup(struct page *page)
2830 struct inode *inode = page->mapping->host;
2831 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2832 struct btrfs_writepage_fixup *fixup;
2834 /* This page has ordered extent covering it already */
2835 if (PageOrdered(page))
2839 * PageChecked is set below when we create a fixup worker for this page,
2840 * don't try to create another one if we're already PageChecked()
2842 * The extent_io writepage code will redirty the page if we send back
2845 if (PageChecked(page))
2848 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2853 * We are already holding a reference to this inode from
2854 * write_cache_pages. We need to hold it because the space reservation
2855 * takes place outside of the page lock, and we can't trust
2856 * page->mapping outside of the page lock.
2859 btrfs_page_set_checked(fs_info, page, page_offset(page), PAGE_SIZE);
2861 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
2863 fixup->inode = inode;
2864 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2869 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2870 struct btrfs_inode *inode, u64 file_pos,
2871 struct btrfs_file_extent_item *stack_fi,
2872 const bool update_inode_bytes,
2873 u64 qgroup_reserved)
2875 struct btrfs_root *root = inode->root;
2876 const u64 sectorsize = root->fs_info->sectorsize;
2877 struct btrfs_path *path;
2878 struct extent_buffer *leaf;
2879 struct btrfs_key ins;
2880 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
2881 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
2882 u64 offset = btrfs_stack_file_extent_offset(stack_fi);
2883 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
2884 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
2885 struct btrfs_drop_extents_args drop_args = { 0 };
2888 path = btrfs_alloc_path();
2893 * we may be replacing one extent in the tree with another.
2894 * The new extent is pinned in the extent map, and we don't want
2895 * to drop it from the cache until it is completely in the btree.
2897 * So, tell btrfs_drop_extents to leave this extent in the cache.
2898 * the caller is expected to unpin it and allow it to be merged
2901 drop_args.path = path;
2902 drop_args.start = file_pos;
2903 drop_args.end = file_pos + num_bytes;
2904 drop_args.replace_extent = true;
2905 drop_args.extent_item_size = sizeof(*stack_fi);
2906 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2910 if (!drop_args.extent_inserted) {
2911 ins.objectid = btrfs_ino(inode);
2912 ins.offset = file_pos;
2913 ins.type = BTRFS_EXTENT_DATA_KEY;
2915 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2920 leaf = path->nodes[0];
2921 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
2922 write_extent_buffer(leaf, stack_fi,
2923 btrfs_item_ptr_offset(leaf, path->slots[0]),
2924 sizeof(struct btrfs_file_extent_item));
2926 btrfs_mark_buffer_dirty(leaf);
2927 btrfs_release_path(path);
2930 * If we dropped an inline extent here, we know the range where it is
2931 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
2932 * number of bytes only for that range containing the inline extent.
2933 * The remaining of the range will be processed when clearning the
2934 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
2936 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
2937 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
2939 inline_size = drop_args.bytes_found - inline_size;
2940 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
2941 drop_args.bytes_found -= inline_size;
2942 num_bytes -= sectorsize;
2945 if (update_inode_bytes)
2946 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
2948 ins.objectid = disk_bytenr;
2949 ins.offset = disk_num_bytes;
2950 ins.type = BTRFS_EXTENT_ITEM_KEY;
2952 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
2956 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
2958 qgroup_reserved, &ins);
2960 btrfs_free_path(path);
2965 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2968 struct btrfs_block_group *cache;
2970 cache = btrfs_lookup_block_group(fs_info, start);
2973 spin_lock(&cache->lock);
2974 cache->delalloc_bytes -= len;
2975 spin_unlock(&cache->lock);
2977 btrfs_put_block_group(cache);
2980 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
2981 struct btrfs_ordered_extent *oe)
2983 struct btrfs_file_extent_item stack_fi;
2984 bool update_inode_bytes;
2985 u64 num_bytes = oe->num_bytes;
2986 u64 ram_bytes = oe->ram_bytes;
2988 memset(&stack_fi, 0, sizeof(stack_fi));
2989 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
2990 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
2991 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
2992 oe->disk_num_bytes);
2993 btrfs_set_stack_file_extent_offset(&stack_fi, oe->offset);
2994 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags))
2995 num_bytes = ram_bytes = oe->truncated_len;
2996 btrfs_set_stack_file_extent_num_bytes(&stack_fi, num_bytes);
2997 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, ram_bytes);
2998 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
2999 /* Encryption and other encoding is reserved and all 0 */
3002 * For delalloc, when completing an ordered extent we update the inode's
3003 * bytes when clearing the range in the inode's io tree, so pass false
3004 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
3005 * except if the ordered extent was truncated.
3007 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
3008 test_bit(BTRFS_ORDERED_ENCODED, &oe->flags) ||
3009 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
3011 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
3012 oe->file_offset, &stack_fi,
3013 update_inode_bytes, oe->qgroup_rsv);
3017 * As ordered data IO finishes, this gets called so we can finish
3018 * an ordered extent if the range of bytes in the file it covers are
3021 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
3023 struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
3024 struct btrfs_root *root = inode->root;
3025 struct btrfs_fs_info *fs_info = root->fs_info;
3026 struct btrfs_trans_handle *trans = NULL;
3027 struct extent_io_tree *io_tree = &inode->io_tree;
3028 struct extent_state *cached_state = NULL;
3030 int compress_type = 0;
3032 u64 logical_len = ordered_extent->num_bytes;
3033 bool freespace_inode;
3034 bool truncated = false;
3035 bool clear_reserved_extent = true;
3036 unsigned int clear_bits = EXTENT_DEFRAG;
3038 start = ordered_extent->file_offset;
3039 end = start + ordered_extent->num_bytes - 1;
3041 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3042 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
3043 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags) &&
3044 !test_bit(BTRFS_ORDERED_ENCODED, &ordered_extent->flags))
3045 clear_bits |= EXTENT_DELALLOC_NEW;
3047 freespace_inode = btrfs_is_free_space_inode(inode);
3049 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
3054 /* A valid bdev implies a write on a sequential zone */
3055 if (ordered_extent->bdev) {
3056 btrfs_rewrite_logical_zoned(ordered_extent);
3057 btrfs_zone_finish_endio(fs_info, ordered_extent->disk_bytenr,
3058 ordered_extent->disk_num_bytes);
3061 btrfs_free_io_failure_record(inode, start, end);
3063 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
3065 logical_len = ordered_extent->truncated_len;
3066 /* Truncated the entire extent, don't bother adding */
3071 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
3072 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
3074 btrfs_inode_safe_disk_i_size_write(inode, 0);
3075 if (freespace_inode)
3076 trans = btrfs_join_transaction_spacecache(root);
3078 trans = btrfs_join_transaction(root);
3079 if (IS_ERR(trans)) {
3080 ret = PTR_ERR(trans);
3084 trans->block_rsv = &inode->block_rsv;
3085 ret = btrfs_update_inode_fallback(trans, root, inode);
3086 if (ret) /* -ENOMEM or corruption */
3087 btrfs_abort_transaction(trans, ret);
3091 clear_bits |= EXTENT_LOCKED;
3092 lock_extent_bits(io_tree, start, end, &cached_state);
3094 if (freespace_inode)
3095 trans = btrfs_join_transaction_spacecache(root);
3097 trans = btrfs_join_transaction(root);
3098 if (IS_ERR(trans)) {
3099 ret = PTR_ERR(trans);
3104 trans->block_rsv = &inode->block_rsv;
3106 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3107 compress_type = ordered_extent->compress_type;
3108 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3109 BUG_ON(compress_type);
3110 ret = btrfs_mark_extent_written(trans, inode,
3111 ordered_extent->file_offset,
3112 ordered_extent->file_offset +
3115 BUG_ON(root == fs_info->tree_root);
3116 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
3118 clear_reserved_extent = false;
3119 btrfs_release_delalloc_bytes(fs_info,
3120 ordered_extent->disk_bytenr,
3121 ordered_extent->disk_num_bytes);
3124 unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset,
3125 ordered_extent->num_bytes, trans->transid);
3127 btrfs_abort_transaction(trans, ret);
3131 ret = add_pending_csums(trans, &ordered_extent->list);
3133 btrfs_abort_transaction(trans, ret);
3138 * If this is a new delalloc range, clear its new delalloc flag to
3139 * update the inode's number of bytes. This needs to be done first
3140 * before updating the inode item.
3142 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
3143 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
3144 clear_extent_bit(&inode->io_tree, start, end,
3145 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
3146 0, 0, &cached_state);
3148 btrfs_inode_safe_disk_i_size_write(inode, 0);
3149 ret = btrfs_update_inode_fallback(trans, root, inode);
3150 if (ret) { /* -ENOMEM or corruption */
3151 btrfs_abort_transaction(trans, ret);
3156 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
3157 (clear_bits & EXTENT_LOCKED) ? 1 : 0, 0,
3161 btrfs_end_transaction(trans);
3163 if (ret || truncated) {
3164 u64 unwritten_start = start;
3167 * If we failed to finish this ordered extent for any reason we
3168 * need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3169 * extent, and mark the inode with the error if it wasn't
3170 * already set. Any error during writeback would have already
3171 * set the mapping error, so we need to set it if we're the ones
3172 * marking this ordered extent as failed.
3174 if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR,
3175 &ordered_extent->flags))
3176 mapping_set_error(ordered_extent->inode->i_mapping, -EIO);
3179 unwritten_start += logical_len;
3180 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
3182 /* Drop the cache for the part of the extent we didn't write. */
3183 btrfs_drop_extent_cache(inode, unwritten_start, end, 0);
3186 * If the ordered extent had an IOERR or something else went
3187 * wrong we need to return the space for this ordered extent
3188 * back to the allocator. We only free the extent in the
3189 * truncated case if we didn't write out the extent at all.
3191 * If we made it past insert_reserved_file_extent before we
3192 * errored out then we don't need to do this as the accounting
3193 * has already been done.
3195 if ((ret || !logical_len) &&
3196 clear_reserved_extent &&
3197 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3198 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3200 * Discard the range before returning it back to the
3203 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3204 btrfs_discard_extent(fs_info,
3205 ordered_extent->disk_bytenr,
3206 ordered_extent->disk_num_bytes,
3208 btrfs_free_reserved_extent(fs_info,
3209 ordered_extent->disk_bytenr,
3210 ordered_extent->disk_num_bytes, 1);
3215 * This needs to be done to make sure anybody waiting knows we are done
3216 * updating everything for this ordered extent.
3218 btrfs_remove_ordered_extent(inode, ordered_extent);
3221 btrfs_put_ordered_extent(ordered_extent);
3222 /* once for the tree */
3223 btrfs_put_ordered_extent(ordered_extent);
3228 static void finish_ordered_fn(struct btrfs_work *work)
3230 struct btrfs_ordered_extent *ordered_extent;
3231 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3232 btrfs_finish_ordered_io(ordered_extent);
3235 void btrfs_writepage_endio_finish_ordered(struct btrfs_inode *inode,
3236 struct page *page, u64 start,
3237 u64 end, bool uptodate)
3239 trace_btrfs_writepage_end_io_hook(inode, start, end, uptodate);
3241 btrfs_mark_ordered_io_finished(inode, page, start, end + 1 - start,
3242 finish_ordered_fn, uptodate);
3246 * check_data_csum - verify checksum of one sector of uncompressed data
3248 * @io_bio: btrfs_io_bio which contains the csum
3249 * @bio_offset: offset to the beginning of the bio (in bytes)
3250 * @page: page where is the data to be verified
3251 * @pgoff: offset inside the page
3252 * @start: logical offset in the file
3254 * The length of such check is always one sector size.
3256 static int check_data_csum(struct inode *inode, struct btrfs_bio *bbio,
3257 u32 bio_offset, struct page *page, u32 pgoff,
3260 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3261 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3263 u32 len = fs_info->sectorsize;
3264 const u32 csum_size = fs_info->csum_size;
3265 unsigned int offset_sectors;
3267 u8 csum[BTRFS_CSUM_SIZE];
3269 ASSERT(pgoff + len <= PAGE_SIZE);
3271 offset_sectors = bio_offset >> fs_info->sectorsize_bits;
3272 csum_expected = ((u8 *)bbio->csum) + offset_sectors * csum_size;
3274 kaddr = kmap_atomic(page);
3275 shash->tfm = fs_info->csum_shash;
3277 crypto_shash_digest(shash, kaddr + pgoff, len, csum);
3279 if (memcmp(csum, csum_expected, csum_size))
3282 kunmap_atomic(kaddr);
3285 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3288 btrfs_dev_stat_inc_and_print(bbio->device,
3289 BTRFS_DEV_STAT_CORRUPTION_ERRS);
3290 memset(kaddr + pgoff, 1, len);
3291 flush_dcache_page(page);
3292 kunmap_atomic(kaddr);
3297 * When reads are done, we need to check csums to verify the data is correct.
3298 * if there's a match, we allow the bio to finish. If not, the code in
3299 * extent_io.c will try to find good copies for us.
3301 * @bio_offset: offset to the beginning of the bio (in bytes)
3302 * @start: file offset of the range start
3303 * @end: file offset of the range end (inclusive)
3305 * Return a bitmap where bit set means a csum mismatch, and bit not set means
3308 unsigned int btrfs_verify_data_csum(struct btrfs_bio *bbio,
3309 u32 bio_offset, struct page *page,
3312 struct inode *inode = page->mapping->host;
3313 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3314 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3315 struct btrfs_root *root = BTRFS_I(inode)->root;
3316 const u32 sectorsize = root->fs_info->sectorsize;
3318 unsigned int result = 0;
3320 if (btrfs_page_test_checked(fs_info, page, start, end + 1 - start)) {
3321 btrfs_page_clear_checked(fs_info, page, start, end + 1 - start);
3326 * This only happens for NODATASUM or compressed read.
3327 * Normally this should be covered by above check for compressed read
3328 * or the next check for NODATASUM. Just do a quicker exit here.
3330 if (bbio->csum == NULL)
3333 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3336 if (unlikely(test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state)))
3339 ASSERT(page_offset(page) <= start &&
3340 end <= page_offset(page) + PAGE_SIZE - 1);
3341 for (pg_off = offset_in_page(start);
3342 pg_off < offset_in_page(end);
3343 pg_off += sectorsize, bio_offset += sectorsize) {
3344 u64 file_offset = pg_off + page_offset(page);
3347 if (btrfs_is_data_reloc_root(root) &&
3348 test_range_bit(io_tree, file_offset,
3349 file_offset + sectorsize - 1,
3350 EXTENT_NODATASUM, 1, NULL)) {
3351 /* Skip the range without csum for data reloc inode */
3352 clear_extent_bits(io_tree, file_offset,
3353 file_offset + sectorsize - 1,
3357 ret = check_data_csum(inode, bbio, bio_offset, page, pg_off,
3358 page_offset(page) + pg_off);
3360 const int nr_bit = (pg_off - offset_in_page(start)) >>
3361 root->fs_info->sectorsize_bits;
3363 result |= (1U << nr_bit);
3370 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3372 * @inode: The inode we want to perform iput on
3374 * This function uses the generic vfs_inode::i_count to track whether we should
3375 * just decrement it (in case it's > 1) or if this is the last iput then link
3376 * the inode to the delayed iput machinery. Delayed iputs are processed at
3377 * transaction commit time/superblock commit/cleaner kthread.
3379 void btrfs_add_delayed_iput(struct inode *inode)
3381 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3382 struct btrfs_inode *binode = BTRFS_I(inode);
3384 if (atomic_add_unless(&inode->i_count, -1, 1))
3387 atomic_inc(&fs_info->nr_delayed_iputs);
3388 spin_lock(&fs_info->delayed_iput_lock);
3389 ASSERT(list_empty(&binode->delayed_iput));
3390 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3391 spin_unlock(&fs_info->delayed_iput_lock);
3392 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3393 wake_up_process(fs_info->cleaner_kthread);
3396 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3397 struct btrfs_inode *inode)
3399 list_del_init(&inode->delayed_iput);
3400 spin_unlock(&fs_info->delayed_iput_lock);
3401 iput(&inode->vfs_inode);
3402 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3403 wake_up(&fs_info->delayed_iputs_wait);
3404 spin_lock(&fs_info->delayed_iput_lock);
3407 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3408 struct btrfs_inode *inode)
3410 if (!list_empty(&inode->delayed_iput)) {
3411 spin_lock(&fs_info->delayed_iput_lock);
3412 if (!list_empty(&inode->delayed_iput))
3413 run_delayed_iput_locked(fs_info, inode);
3414 spin_unlock(&fs_info->delayed_iput_lock);
3418 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3421 spin_lock(&fs_info->delayed_iput_lock);
3422 while (!list_empty(&fs_info->delayed_iputs)) {
3423 struct btrfs_inode *inode;
3425 inode = list_first_entry(&fs_info->delayed_iputs,
3426 struct btrfs_inode, delayed_iput);
3427 run_delayed_iput_locked(fs_info, inode);
3428 cond_resched_lock(&fs_info->delayed_iput_lock);
3430 spin_unlock(&fs_info->delayed_iput_lock);
3434 * Wait for flushing all delayed iputs
3436 * @fs_info: the filesystem
3438 * This will wait on any delayed iputs that are currently running with KILLABLE
3439 * set. Once they are all done running we will return, unless we are killed in
3440 * which case we return EINTR. This helps in user operations like fallocate etc
3441 * that might get blocked on the iputs.
3443 * Return EINTR if we were killed, 0 if nothing's pending
3445 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3447 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3448 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3455 * This creates an orphan entry for the given inode in case something goes wrong
3456 * in the middle of an unlink.
3458 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3459 struct btrfs_inode *inode)
3463 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3464 if (ret && ret != -EEXIST) {
3465 btrfs_abort_transaction(trans, ret);
3473 * We have done the delete so we can go ahead and remove the orphan item for
3474 * this particular inode.
3476 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3477 struct btrfs_inode *inode)
3479 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3483 * this cleans up any orphans that may be left on the list from the last use
3486 int btrfs_orphan_cleanup(struct btrfs_root *root)
3488 struct btrfs_fs_info *fs_info = root->fs_info;
3489 struct btrfs_path *path;
3490 struct extent_buffer *leaf;
3491 struct btrfs_key key, found_key;
3492 struct btrfs_trans_handle *trans;
3493 struct inode *inode;
3494 u64 last_objectid = 0;
3495 int ret = 0, nr_unlink = 0;
3497 if (test_and_set_bit(BTRFS_ROOT_ORPHAN_CLEANUP, &root->state))
3500 path = btrfs_alloc_path();
3505 path->reada = READA_BACK;
3507 key.objectid = BTRFS_ORPHAN_OBJECTID;
3508 key.type = BTRFS_ORPHAN_ITEM_KEY;
3509 key.offset = (u64)-1;
3512 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3517 * if ret == 0 means we found what we were searching for, which
3518 * is weird, but possible, so only screw with path if we didn't
3519 * find the key and see if we have stuff that matches
3523 if (path->slots[0] == 0)
3528 /* pull out the item */
3529 leaf = path->nodes[0];
3530 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3532 /* make sure the item matches what we want */
3533 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3535 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3538 /* release the path since we're done with it */
3539 btrfs_release_path(path);
3542 * this is where we are basically btrfs_lookup, without the
3543 * crossing root thing. we store the inode number in the
3544 * offset of the orphan item.
3547 if (found_key.offset == last_objectid) {
3549 "Error removing orphan entry, stopping orphan cleanup");
3554 last_objectid = found_key.offset;
3556 found_key.objectid = found_key.offset;
3557 found_key.type = BTRFS_INODE_ITEM_KEY;
3558 found_key.offset = 0;
3559 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3560 ret = PTR_ERR_OR_ZERO(inode);
3561 if (ret && ret != -ENOENT)
3564 if (ret == -ENOENT && root == fs_info->tree_root) {
3565 struct btrfs_root *dead_root;
3566 int is_dead_root = 0;
3569 * This is an orphan in the tree root. Currently these
3570 * could come from 2 sources:
3571 * a) a root (snapshot/subvolume) deletion in progress
3572 * b) a free space cache inode
3573 * We need to distinguish those two, as the orphan item
3574 * for a root must not get deleted before the deletion
3575 * of the snapshot/subvolume's tree completes.
3577 * btrfs_find_orphan_roots() ran before us, which has
3578 * found all deleted roots and loaded them into
3579 * fs_info->fs_roots_radix. So here we can find if an
3580 * orphan item corresponds to a deleted root by looking
3581 * up the root from that radix tree.
3584 spin_lock(&fs_info->fs_roots_radix_lock);
3585 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3586 (unsigned long)found_key.objectid);
3587 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3589 spin_unlock(&fs_info->fs_roots_radix_lock);
3592 /* prevent this orphan from being found again */
3593 key.offset = found_key.objectid - 1;
3600 * If we have an inode with links, there are a couple of
3603 * 1. We were halfway through creating fsverity metadata for the
3604 * file. In that case, the orphan item represents incomplete
3605 * fsverity metadata which must be cleaned up with
3606 * btrfs_drop_verity_items and deleting the orphan item.
3608 * 2. Old kernels (before v3.12) used to create an
3609 * orphan item for truncate indicating that there were possibly
3610 * extent items past i_size that needed to be deleted. In v3.12,
3611 * truncate was changed to update i_size in sync with the extent
3612 * items, but the (useless) orphan item was still created. Since
3613 * v4.18, we don't create the orphan item for truncate at all.
3615 * So, this item could mean that we need to do a truncate, but
3616 * only if this filesystem was last used on a pre-v3.12 kernel
3617 * and was not cleanly unmounted. The odds of that are quite
3618 * slim, and it's a pain to do the truncate now, so just delete
3621 * It's also possible that this orphan item was supposed to be
3622 * deleted but wasn't. The inode number may have been reused,
3623 * but either way, we can delete the orphan item.
3625 if (ret == -ENOENT || inode->i_nlink) {
3627 ret = btrfs_drop_verity_items(BTRFS_I(inode));
3632 trans = btrfs_start_transaction(root, 1);
3633 if (IS_ERR(trans)) {
3634 ret = PTR_ERR(trans);
3637 btrfs_debug(fs_info, "auto deleting %Lu",
3638 found_key.objectid);
3639 ret = btrfs_del_orphan_item(trans, root,
3640 found_key.objectid);
3641 btrfs_end_transaction(trans);
3649 /* this will do delete_inode and everything for us */
3652 /* release the path since we're done with it */
3653 btrfs_release_path(path);
3655 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3656 trans = btrfs_join_transaction(root);
3658 btrfs_end_transaction(trans);
3662 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3666 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3667 btrfs_free_path(path);
3672 * very simple check to peek ahead in the leaf looking for xattrs. If we
3673 * don't find any xattrs, we know there can't be any acls.
3675 * slot is the slot the inode is in, objectid is the objectid of the inode
3677 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3678 int slot, u64 objectid,
3679 int *first_xattr_slot)
3681 u32 nritems = btrfs_header_nritems(leaf);
3682 struct btrfs_key found_key;
3683 static u64 xattr_access = 0;
3684 static u64 xattr_default = 0;
3687 if (!xattr_access) {
3688 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3689 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3690 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3691 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3695 *first_xattr_slot = -1;
3696 while (slot < nritems) {
3697 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3699 /* we found a different objectid, there must not be acls */
3700 if (found_key.objectid != objectid)
3703 /* we found an xattr, assume we've got an acl */
3704 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3705 if (*first_xattr_slot == -1)
3706 *first_xattr_slot = slot;
3707 if (found_key.offset == xattr_access ||
3708 found_key.offset == xattr_default)
3713 * we found a key greater than an xattr key, there can't
3714 * be any acls later on
3716 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3723 * it goes inode, inode backrefs, xattrs, extents,
3724 * so if there are a ton of hard links to an inode there can
3725 * be a lot of backrefs. Don't waste time searching too hard,
3726 * this is just an optimization
3731 /* we hit the end of the leaf before we found an xattr or
3732 * something larger than an xattr. We have to assume the inode
3735 if (*first_xattr_slot == -1)
3736 *first_xattr_slot = slot;
3741 * read an inode from the btree into the in-memory inode
3743 static int btrfs_read_locked_inode(struct inode *inode,
3744 struct btrfs_path *in_path)
3746 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3747 struct btrfs_path *path = in_path;
3748 struct extent_buffer *leaf;
3749 struct btrfs_inode_item *inode_item;
3750 struct btrfs_root *root = BTRFS_I(inode)->root;
3751 struct btrfs_key location;
3756 bool filled = false;
3757 int first_xattr_slot;
3759 ret = btrfs_fill_inode(inode, &rdev);
3764 path = btrfs_alloc_path();
3769 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3771 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3773 if (path != in_path)
3774 btrfs_free_path(path);
3778 leaf = path->nodes[0];
3783 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3784 struct btrfs_inode_item);
3785 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3786 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3787 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3788 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3789 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3790 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3791 round_up(i_size_read(inode), fs_info->sectorsize));
3793 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3794 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3796 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3797 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3799 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3800 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3802 BTRFS_I(inode)->i_otime.tv_sec =
3803 btrfs_timespec_sec(leaf, &inode_item->otime);
3804 BTRFS_I(inode)->i_otime.tv_nsec =
3805 btrfs_timespec_nsec(leaf, &inode_item->otime);
3807 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3808 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3809 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3811 inode_set_iversion_queried(inode,
3812 btrfs_inode_sequence(leaf, inode_item));
3813 inode->i_generation = BTRFS_I(inode)->generation;
3815 rdev = btrfs_inode_rdev(leaf, inode_item);
3817 BTRFS_I(inode)->index_cnt = (u64)-1;
3818 btrfs_inode_split_flags(btrfs_inode_flags(leaf, inode_item),
3819 &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags);
3823 * If we were modified in the current generation and evicted from memory
3824 * and then re-read we need to do a full sync since we don't have any
3825 * idea about which extents were modified before we were evicted from
3828 * This is required for both inode re-read from disk and delayed inode
3829 * in delayed_nodes_tree.
3831 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3832 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3833 &BTRFS_I(inode)->runtime_flags);
3836 * We don't persist the id of the transaction where an unlink operation
3837 * against the inode was last made. So here we assume the inode might
3838 * have been evicted, and therefore the exact value of last_unlink_trans
3839 * lost, and set it to last_trans to avoid metadata inconsistencies
3840 * between the inode and its parent if the inode is fsync'ed and the log
3841 * replayed. For example, in the scenario:
3844 * ln mydir/foo mydir/bar
3847 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3848 * xfs_io -c fsync mydir/foo
3850 * mount fs, triggers fsync log replay
3852 * We must make sure that when we fsync our inode foo we also log its
3853 * parent inode, otherwise after log replay the parent still has the
3854 * dentry with the "bar" name but our inode foo has a link count of 1
3855 * and doesn't have an inode ref with the name "bar" anymore.
3857 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3858 * but it guarantees correctness at the expense of occasional full
3859 * transaction commits on fsync if our inode is a directory, or if our
3860 * inode is not a directory, logging its parent unnecessarily.
3862 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3865 * Same logic as for last_unlink_trans. We don't persist the generation
3866 * of the last transaction where this inode was used for a reflink
3867 * operation, so after eviction and reloading the inode we must be
3868 * pessimistic and assume the last transaction that modified the inode.
3870 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3873 if (inode->i_nlink != 1 ||
3874 path->slots[0] >= btrfs_header_nritems(leaf))
3877 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3878 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3881 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3882 if (location.type == BTRFS_INODE_REF_KEY) {
3883 struct btrfs_inode_ref *ref;
3885 ref = (struct btrfs_inode_ref *)ptr;
3886 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3887 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3888 struct btrfs_inode_extref *extref;
3890 extref = (struct btrfs_inode_extref *)ptr;
3891 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3896 * try to precache a NULL acl entry for files that don't have
3897 * any xattrs or acls
3899 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3900 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3901 if (first_xattr_slot != -1) {
3902 path->slots[0] = first_xattr_slot;
3903 ret = btrfs_load_inode_props(inode, path);
3906 "error loading props for ino %llu (root %llu): %d",
3907 btrfs_ino(BTRFS_I(inode)),
3908 root->root_key.objectid, ret);
3910 if (path != in_path)
3911 btrfs_free_path(path);
3914 cache_no_acl(inode);
3916 switch (inode->i_mode & S_IFMT) {
3918 inode->i_mapping->a_ops = &btrfs_aops;
3919 inode->i_fop = &btrfs_file_operations;
3920 inode->i_op = &btrfs_file_inode_operations;
3923 inode->i_fop = &btrfs_dir_file_operations;
3924 inode->i_op = &btrfs_dir_inode_operations;
3927 inode->i_op = &btrfs_symlink_inode_operations;
3928 inode_nohighmem(inode);
3929 inode->i_mapping->a_ops = &btrfs_aops;
3932 inode->i_op = &btrfs_special_inode_operations;
3933 init_special_inode(inode, inode->i_mode, rdev);
3937 btrfs_sync_inode_flags_to_i_flags(inode);
3942 * given a leaf and an inode, copy the inode fields into the leaf
3944 static void fill_inode_item(struct btrfs_trans_handle *trans,
3945 struct extent_buffer *leaf,
3946 struct btrfs_inode_item *item,
3947 struct inode *inode)
3949 struct btrfs_map_token token;
3952 btrfs_init_map_token(&token, leaf);
3954 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3955 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3956 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
3957 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3958 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3960 btrfs_set_token_timespec_sec(&token, &item->atime,
3961 inode->i_atime.tv_sec);
3962 btrfs_set_token_timespec_nsec(&token, &item->atime,
3963 inode->i_atime.tv_nsec);
3965 btrfs_set_token_timespec_sec(&token, &item->mtime,
3966 inode->i_mtime.tv_sec);
3967 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3968 inode->i_mtime.tv_nsec);
3970 btrfs_set_token_timespec_sec(&token, &item->ctime,
3971 inode->i_ctime.tv_sec);
3972 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3973 inode->i_ctime.tv_nsec);
3975 btrfs_set_token_timespec_sec(&token, &item->otime,
3976 BTRFS_I(inode)->i_otime.tv_sec);
3977 btrfs_set_token_timespec_nsec(&token, &item->otime,
3978 BTRFS_I(inode)->i_otime.tv_nsec);
3980 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
3981 btrfs_set_token_inode_generation(&token, item,
3982 BTRFS_I(inode)->generation);
3983 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
3984 btrfs_set_token_inode_transid(&token, item, trans->transid);
3985 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
3986 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
3987 BTRFS_I(inode)->ro_flags);
3988 btrfs_set_token_inode_flags(&token, item, flags);
3989 btrfs_set_token_inode_block_group(&token, item, 0);
3993 * copy everything in the in-memory inode into the btree.
3995 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3996 struct btrfs_root *root,
3997 struct btrfs_inode *inode)
3999 struct btrfs_inode_item *inode_item;
4000 struct btrfs_path *path;
4001 struct extent_buffer *leaf;
4004 path = btrfs_alloc_path();
4008 ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1);
4015 leaf = path->nodes[0];
4016 inode_item = btrfs_item_ptr(leaf, path->slots[0],
4017 struct btrfs_inode_item);
4019 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
4020 btrfs_mark_buffer_dirty(leaf);
4021 btrfs_set_inode_last_trans(trans, inode);
4024 btrfs_free_path(path);
4029 * copy everything in the in-memory inode into the btree.
4031 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
4032 struct btrfs_root *root,
4033 struct btrfs_inode *inode)
4035 struct btrfs_fs_info *fs_info = root->fs_info;
4039 * If the inode is a free space inode, we can deadlock during commit
4040 * if we put it into the delayed code.
4042 * The data relocation inode should also be directly updated
4045 if (!btrfs_is_free_space_inode(inode)
4046 && !btrfs_is_data_reloc_root(root)
4047 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4048 btrfs_update_root_times(trans, root);
4050 ret = btrfs_delayed_update_inode(trans, root, inode);
4052 btrfs_set_inode_last_trans(trans, inode);
4056 return btrfs_update_inode_item(trans, root, inode);
4059 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4060 struct btrfs_root *root, struct btrfs_inode *inode)
4064 ret = btrfs_update_inode(trans, root, inode);
4066 return btrfs_update_inode_item(trans, root, inode);
4071 * unlink helper that gets used here in inode.c and in the tree logging
4072 * recovery code. It remove a link in a directory with a given name, and
4073 * also drops the back refs in the inode to the directory
4075 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4076 struct btrfs_inode *dir,
4077 struct btrfs_inode *inode,
4078 const char *name, int name_len,
4079 struct btrfs_rename_ctx *rename_ctx)
4081 struct btrfs_root *root = dir->root;
4082 struct btrfs_fs_info *fs_info = root->fs_info;
4083 struct btrfs_path *path;
4085 struct btrfs_dir_item *di;
4087 u64 ino = btrfs_ino(inode);
4088 u64 dir_ino = btrfs_ino(dir);
4090 path = btrfs_alloc_path();
4096 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4097 name, name_len, -1);
4098 if (IS_ERR_OR_NULL(di)) {
4099 ret = di ? PTR_ERR(di) : -ENOENT;
4102 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4105 btrfs_release_path(path);
4108 * If we don't have dir index, we have to get it by looking up
4109 * the inode ref, since we get the inode ref, remove it directly,
4110 * it is unnecessary to do delayed deletion.
4112 * But if we have dir index, needn't search inode ref to get it.
4113 * Since the inode ref is close to the inode item, it is better
4114 * that we delay to delete it, and just do this deletion when
4115 * we update the inode item.
4117 if (inode->dir_index) {
4118 ret = btrfs_delayed_delete_inode_ref(inode);
4120 index = inode->dir_index;
4125 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4129 "failed to delete reference to %.*s, inode %llu parent %llu",
4130 name_len, name, ino, dir_ino);
4131 btrfs_abort_transaction(trans, ret);
4136 rename_ctx->index = index;
4138 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4140 btrfs_abort_transaction(trans, ret);
4145 * If we are in a rename context, we don't need to update anything in the
4146 * log. That will be done later during the rename by btrfs_log_new_name().
4147 * Besides that, doing it here would only cause extra unncessary btree
4148 * operations on the log tree, increasing latency for applications.
4151 btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
4153 btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
4158 * If we have a pending delayed iput we could end up with the final iput
4159 * being run in btrfs-cleaner context. If we have enough of these built
4160 * up we can end up burning a lot of time in btrfs-cleaner without any
4161 * way to throttle the unlinks. Since we're currently holding a ref on
4162 * the inode we can run the delayed iput here without any issues as the
4163 * final iput won't be done until after we drop the ref we're currently
4166 btrfs_run_delayed_iput(fs_info, inode);
4168 btrfs_free_path(path);
4172 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
4173 inode_inc_iversion(&inode->vfs_inode);
4174 inode_inc_iversion(&dir->vfs_inode);
4175 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
4176 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4177 ret = btrfs_update_inode(trans, root, dir);
4182 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4183 struct btrfs_inode *dir, struct btrfs_inode *inode,
4184 const char *name, int name_len)
4187 ret = __btrfs_unlink_inode(trans, dir, inode, name, name_len, NULL);
4189 drop_nlink(&inode->vfs_inode);
4190 ret = btrfs_update_inode(trans, inode->root, inode);
4196 * helper to start transaction for unlink and rmdir.
4198 * unlink and rmdir are special in btrfs, they do not always free space, so
4199 * if we cannot make our reservations the normal way try and see if there is
4200 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4201 * allow the unlink to occur.
4203 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4205 struct btrfs_root *root = BTRFS_I(dir)->root;
4208 * 1 for the possible orphan item
4209 * 1 for the dir item
4210 * 1 for the dir index
4211 * 1 for the inode ref
4214 return btrfs_start_transaction_fallback_global_rsv(root, 5);
4217 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4219 struct btrfs_trans_handle *trans;
4220 struct inode *inode = d_inode(dentry);
4223 trans = __unlink_start_trans(dir);
4225 return PTR_ERR(trans);
4227 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4230 ret = btrfs_unlink_inode(trans, BTRFS_I(dir),
4231 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4232 dentry->d_name.len);
4236 if (inode->i_nlink == 0) {
4237 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4243 btrfs_end_transaction(trans);
4244 btrfs_btree_balance_dirty(BTRFS_I(dir)->root->fs_info);
4248 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4249 struct inode *dir, struct dentry *dentry)
4251 struct btrfs_root *root = BTRFS_I(dir)->root;
4252 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4253 struct btrfs_path *path;
4254 struct extent_buffer *leaf;
4255 struct btrfs_dir_item *di;
4256 struct btrfs_key key;
4257 const char *name = dentry->d_name.name;
4258 int name_len = dentry->d_name.len;
4262 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4264 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4265 objectid = inode->root->root_key.objectid;
4266 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4267 objectid = inode->location.objectid;
4273 path = btrfs_alloc_path();
4277 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4278 name, name_len, -1);
4279 if (IS_ERR_OR_NULL(di)) {
4280 ret = di ? PTR_ERR(di) : -ENOENT;
4284 leaf = path->nodes[0];
4285 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4286 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4287 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4289 btrfs_abort_transaction(trans, ret);
4292 btrfs_release_path(path);
4295 * This is a placeholder inode for a subvolume we didn't have a
4296 * reference to at the time of the snapshot creation. In the meantime
4297 * we could have renamed the real subvol link into our snapshot, so
4298 * depending on btrfs_del_root_ref to return -ENOENT here is incorrect.
4299 * Instead simply lookup the dir_index_item for this entry so we can
4300 * remove it. Otherwise we know we have a ref to the root and we can
4301 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4303 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4304 di = btrfs_search_dir_index_item(root, path, dir_ino,
4306 if (IS_ERR_OR_NULL(di)) {
4311 btrfs_abort_transaction(trans, ret);
4315 leaf = path->nodes[0];
4316 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4318 btrfs_release_path(path);
4320 ret = btrfs_del_root_ref(trans, objectid,
4321 root->root_key.objectid, dir_ino,
4322 &index, name, name_len);
4324 btrfs_abort_transaction(trans, ret);
4329 ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index);
4331 btrfs_abort_transaction(trans, ret);
4335 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4336 inode_inc_iversion(dir);
4337 dir->i_mtime = dir->i_ctime = current_time(dir);
4338 ret = btrfs_update_inode_fallback(trans, root, BTRFS_I(dir));
4340 btrfs_abort_transaction(trans, ret);
4342 btrfs_free_path(path);
4347 * Helper to check if the subvolume references other subvolumes or if it's
4350 static noinline int may_destroy_subvol(struct btrfs_root *root)
4352 struct btrfs_fs_info *fs_info = root->fs_info;
4353 struct btrfs_path *path;
4354 struct btrfs_dir_item *di;
4355 struct btrfs_key key;
4359 path = btrfs_alloc_path();
4363 /* Make sure this root isn't set as the default subvol */
4364 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4365 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4366 dir_id, "default", 7, 0);
4367 if (di && !IS_ERR(di)) {
4368 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4369 if (key.objectid == root->root_key.objectid) {
4372 "deleting default subvolume %llu is not allowed",
4376 btrfs_release_path(path);
4379 key.objectid = root->root_key.objectid;
4380 key.type = BTRFS_ROOT_REF_KEY;
4381 key.offset = (u64)-1;
4383 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4389 if (path->slots[0] > 0) {
4391 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4392 if (key.objectid == root->root_key.objectid &&
4393 key.type == BTRFS_ROOT_REF_KEY)
4397 btrfs_free_path(path);
4401 /* Delete all dentries for inodes belonging to the root */
4402 static void btrfs_prune_dentries(struct btrfs_root *root)
4404 struct btrfs_fs_info *fs_info = root->fs_info;
4405 struct rb_node *node;
4406 struct rb_node *prev;
4407 struct btrfs_inode *entry;
4408 struct inode *inode;
4411 if (!BTRFS_FS_ERROR(fs_info))
4412 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4414 spin_lock(&root->inode_lock);
4416 node = root->inode_tree.rb_node;
4420 entry = rb_entry(node, struct btrfs_inode, rb_node);
4422 if (objectid < btrfs_ino(entry))
4423 node = node->rb_left;
4424 else if (objectid > btrfs_ino(entry))
4425 node = node->rb_right;
4431 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4432 if (objectid <= btrfs_ino(entry)) {
4436 prev = rb_next(prev);
4440 entry = rb_entry(node, struct btrfs_inode, rb_node);
4441 objectid = btrfs_ino(entry) + 1;
4442 inode = igrab(&entry->vfs_inode);
4444 spin_unlock(&root->inode_lock);
4445 if (atomic_read(&inode->i_count) > 1)
4446 d_prune_aliases(inode);
4448 * btrfs_drop_inode will have it removed from the inode
4449 * cache when its usage count hits zero.
4453 spin_lock(&root->inode_lock);
4457 if (cond_resched_lock(&root->inode_lock))
4460 node = rb_next(node);
4462 spin_unlock(&root->inode_lock);
4465 int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
4467 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4468 struct btrfs_root *root = BTRFS_I(dir)->root;
4469 struct inode *inode = d_inode(dentry);
4470 struct btrfs_root *dest = BTRFS_I(inode)->root;
4471 struct btrfs_trans_handle *trans;
4472 struct btrfs_block_rsv block_rsv;
4477 * Don't allow to delete a subvolume with send in progress. This is
4478 * inside the inode lock so the error handling that has to drop the bit
4479 * again is not run concurrently.
4481 spin_lock(&dest->root_item_lock);
4482 if (dest->send_in_progress) {
4483 spin_unlock(&dest->root_item_lock);
4485 "attempt to delete subvolume %llu during send",
4486 dest->root_key.objectid);
4489 if (atomic_read(&dest->nr_swapfiles)) {
4490 spin_unlock(&dest->root_item_lock);
4492 "attempt to delete subvolume %llu with active swapfile",
4493 root->root_key.objectid);
4496 root_flags = btrfs_root_flags(&dest->root_item);
4497 btrfs_set_root_flags(&dest->root_item,
4498 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4499 spin_unlock(&dest->root_item_lock);
4501 down_write(&fs_info->subvol_sem);
4503 ret = may_destroy_subvol(dest);
4507 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4509 * One for dir inode,
4510 * two for dir entries,
4511 * two for root ref/backref.
4513 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4517 trans = btrfs_start_transaction(root, 0);
4518 if (IS_ERR(trans)) {
4519 ret = PTR_ERR(trans);
4522 trans->block_rsv = &block_rsv;
4523 trans->bytes_reserved = block_rsv.size;
4525 btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
4527 ret = btrfs_unlink_subvol(trans, dir, dentry);
4529 btrfs_abort_transaction(trans, ret);
4533 ret = btrfs_record_root_in_trans(trans, dest);
4535 btrfs_abort_transaction(trans, ret);
4539 memset(&dest->root_item.drop_progress, 0,
4540 sizeof(dest->root_item.drop_progress));
4541 btrfs_set_root_drop_level(&dest->root_item, 0);
4542 btrfs_set_root_refs(&dest->root_item, 0);
4544 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4545 ret = btrfs_insert_orphan_item(trans,
4547 dest->root_key.objectid);
4549 btrfs_abort_transaction(trans, ret);
4554 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4555 BTRFS_UUID_KEY_SUBVOL,
4556 dest->root_key.objectid);
4557 if (ret && ret != -ENOENT) {
4558 btrfs_abort_transaction(trans, ret);
4561 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4562 ret = btrfs_uuid_tree_remove(trans,
4563 dest->root_item.received_uuid,
4564 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4565 dest->root_key.objectid);
4566 if (ret && ret != -ENOENT) {
4567 btrfs_abort_transaction(trans, ret);
4572 free_anon_bdev(dest->anon_dev);
4575 trans->block_rsv = NULL;
4576 trans->bytes_reserved = 0;
4577 ret = btrfs_end_transaction(trans);
4578 inode->i_flags |= S_DEAD;
4580 btrfs_subvolume_release_metadata(root, &block_rsv);
4582 up_write(&fs_info->subvol_sem);
4584 spin_lock(&dest->root_item_lock);
4585 root_flags = btrfs_root_flags(&dest->root_item);
4586 btrfs_set_root_flags(&dest->root_item,
4587 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4588 spin_unlock(&dest->root_item_lock);
4590 d_invalidate(dentry);
4591 btrfs_prune_dentries(dest);
4592 ASSERT(dest->send_in_progress == 0);
4598 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4600 struct inode *inode = d_inode(dentry);
4601 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
4603 struct btrfs_trans_handle *trans;
4604 u64 last_unlink_trans;
4606 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4608 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) {
4609 if (unlikely(btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))) {
4611 "extent tree v2 doesn't support snapshot deletion yet");
4614 return btrfs_delete_subvolume(dir, dentry);
4617 trans = __unlink_start_trans(dir);
4619 return PTR_ERR(trans);
4621 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4622 err = btrfs_unlink_subvol(trans, dir, dentry);
4626 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4630 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4632 /* now the directory is empty */
4633 err = btrfs_unlink_inode(trans, BTRFS_I(dir),
4634 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4635 dentry->d_name.len);
4637 btrfs_i_size_write(BTRFS_I(inode), 0);
4639 * Propagate the last_unlink_trans value of the deleted dir to
4640 * its parent directory. This is to prevent an unrecoverable
4641 * log tree in the case we do something like this:
4643 * 2) create snapshot under dir foo
4644 * 3) delete the snapshot
4647 * 6) fsync foo or some file inside foo
4649 if (last_unlink_trans >= trans->transid)
4650 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4653 btrfs_end_transaction(trans);
4654 btrfs_btree_balance_dirty(fs_info);
4660 * btrfs_truncate_block - read, zero a chunk and write a block
4661 * @inode - inode that we're zeroing
4662 * @from - the offset to start zeroing
4663 * @len - the length to zero, 0 to zero the entire range respective to the
4665 * @front - zero up to the offset instead of from the offset on
4667 * This will find the block for the "from" offset and cow the block and zero the
4668 * part we want to zero. This is used with truncate and hole punching.
4670 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4673 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4674 struct address_space *mapping = inode->vfs_inode.i_mapping;
4675 struct extent_io_tree *io_tree = &inode->io_tree;
4676 struct btrfs_ordered_extent *ordered;
4677 struct extent_state *cached_state = NULL;
4678 struct extent_changeset *data_reserved = NULL;
4679 bool only_release_metadata = false;
4680 u32 blocksize = fs_info->sectorsize;
4681 pgoff_t index = from >> PAGE_SHIFT;
4682 unsigned offset = from & (blocksize - 1);
4684 gfp_t mask = btrfs_alloc_write_mask(mapping);
4685 size_t write_bytes = blocksize;
4690 if (IS_ALIGNED(offset, blocksize) &&
4691 (!len || IS_ALIGNED(len, blocksize)))
4694 block_start = round_down(from, blocksize);
4695 block_end = block_start + blocksize - 1;
4697 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4700 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes) > 0) {
4701 /* For nocow case, no need to reserve data space */
4702 only_release_metadata = true;
4707 ret = btrfs_delalloc_reserve_metadata(inode, blocksize, blocksize);
4709 if (!only_release_metadata)
4710 btrfs_free_reserved_data_space(inode, data_reserved,
4711 block_start, blocksize);
4715 page = find_or_create_page(mapping, index, mask);
4717 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4719 btrfs_delalloc_release_extents(inode, blocksize);
4723 ret = set_page_extent_mapped(page);
4727 if (!PageUptodate(page)) {
4728 ret = btrfs_readpage(NULL, page);
4730 if (page->mapping != mapping) {
4735 if (!PageUptodate(page)) {
4740 wait_on_page_writeback(page);
4742 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4744 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4746 unlock_extent_cached(io_tree, block_start, block_end,
4750 btrfs_start_ordered_extent(ordered, 1);
4751 btrfs_put_ordered_extent(ordered);
4755 clear_extent_bit(&inode->io_tree, block_start, block_end,
4756 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4757 0, 0, &cached_state);
4759 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4762 unlock_extent_cached(io_tree, block_start, block_end,
4767 if (offset != blocksize) {
4769 len = blocksize - offset;
4771 memzero_page(page, (block_start - page_offset(page)),
4774 memzero_page(page, (block_start - page_offset(page)) + offset,
4776 flush_dcache_page(page);
4778 btrfs_page_clear_checked(fs_info, page, block_start,
4779 block_end + 1 - block_start);
4780 btrfs_page_set_dirty(fs_info, page, block_start, block_end + 1 - block_start);
4781 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4783 if (only_release_metadata)
4784 set_extent_bit(&inode->io_tree, block_start, block_end,
4785 EXTENT_NORESERVE, 0, NULL, NULL, GFP_NOFS, NULL);
4789 if (only_release_metadata)
4790 btrfs_delalloc_release_metadata(inode, blocksize, true);
4792 btrfs_delalloc_release_space(inode, data_reserved,
4793 block_start, blocksize, true);
4795 btrfs_delalloc_release_extents(inode, blocksize);
4799 if (only_release_metadata)
4800 btrfs_check_nocow_unlock(inode);
4801 extent_changeset_free(data_reserved);
4805 static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode,
4806 u64 offset, u64 len)
4808 struct btrfs_fs_info *fs_info = root->fs_info;
4809 struct btrfs_trans_handle *trans;
4810 struct btrfs_drop_extents_args drop_args = { 0 };
4814 * If NO_HOLES is enabled, we don't need to do anything.
4815 * Later, up in the call chain, either btrfs_set_inode_last_sub_trans()
4816 * or btrfs_update_inode() will be called, which guarantee that the next
4817 * fsync will know this inode was changed and needs to be logged.
4819 if (btrfs_fs_incompat(fs_info, NO_HOLES))
4823 * 1 - for the one we're dropping
4824 * 1 - for the one we're adding
4825 * 1 - for updating the inode.
4827 trans = btrfs_start_transaction(root, 3);
4829 return PTR_ERR(trans);
4831 drop_args.start = offset;
4832 drop_args.end = offset + len;
4833 drop_args.drop_cache = true;
4835 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
4837 btrfs_abort_transaction(trans, ret);
4838 btrfs_end_transaction(trans);
4842 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode),
4843 offset, 0, 0, len, 0, len, 0, 0, 0);
4845 btrfs_abort_transaction(trans, ret);
4847 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
4848 btrfs_update_inode(trans, root, inode);
4850 btrfs_end_transaction(trans);
4855 * This function puts in dummy file extents for the area we're creating a hole
4856 * for. So if we are truncating this file to a larger size we need to insert
4857 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4858 * the range between oldsize and size
4860 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
4862 struct btrfs_root *root = inode->root;
4863 struct btrfs_fs_info *fs_info = root->fs_info;
4864 struct extent_io_tree *io_tree = &inode->io_tree;
4865 struct extent_map *em = NULL;
4866 struct extent_state *cached_state = NULL;
4867 struct extent_map_tree *em_tree = &inode->extent_tree;
4868 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4869 u64 block_end = ALIGN(size, fs_info->sectorsize);
4876 * If our size started in the middle of a block we need to zero out the
4877 * rest of the block before we expand the i_size, otherwise we could
4878 * expose stale data.
4880 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4884 if (size <= hole_start)
4887 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
4889 cur_offset = hole_start;
4891 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4892 block_end - cur_offset);
4898 last_byte = min(extent_map_end(em), block_end);
4899 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4900 hole_size = last_byte - cur_offset;
4902 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4903 struct extent_map *hole_em;
4905 err = maybe_insert_hole(root, inode, cur_offset,
4910 err = btrfs_inode_set_file_extent_range(inode,
4911 cur_offset, hole_size);
4915 btrfs_drop_extent_cache(inode, cur_offset,
4916 cur_offset + hole_size - 1, 0);
4917 hole_em = alloc_extent_map();
4919 btrfs_set_inode_full_sync(inode);
4922 hole_em->start = cur_offset;
4923 hole_em->len = hole_size;
4924 hole_em->orig_start = cur_offset;
4926 hole_em->block_start = EXTENT_MAP_HOLE;
4927 hole_em->block_len = 0;
4928 hole_em->orig_block_len = 0;
4929 hole_em->ram_bytes = hole_size;
4930 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4931 hole_em->generation = fs_info->generation;
4934 write_lock(&em_tree->lock);
4935 err = add_extent_mapping(em_tree, hole_em, 1);
4936 write_unlock(&em_tree->lock);
4939 btrfs_drop_extent_cache(inode, cur_offset,
4943 free_extent_map(hole_em);
4945 err = btrfs_inode_set_file_extent_range(inode,
4946 cur_offset, hole_size);
4951 free_extent_map(em);
4953 cur_offset = last_byte;
4954 if (cur_offset >= block_end)
4957 free_extent_map(em);
4958 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
4962 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4964 struct btrfs_root *root = BTRFS_I(inode)->root;
4965 struct btrfs_trans_handle *trans;
4966 loff_t oldsize = i_size_read(inode);
4967 loff_t newsize = attr->ia_size;
4968 int mask = attr->ia_valid;
4972 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4973 * special case where we need to update the times despite not having
4974 * these flags set. For all other operations the VFS set these flags
4975 * explicitly if it wants a timestamp update.
4977 if (newsize != oldsize) {
4978 inode_inc_iversion(inode);
4979 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4980 inode->i_ctime = inode->i_mtime =
4981 current_time(inode);
4984 if (newsize > oldsize) {
4986 * Don't do an expanding truncate while snapshotting is ongoing.
4987 * This is to ensure the snapshot captures a fully consistent
4988 * state of this file - if the snapshot captures this expanding
4989 * truncation, it must capture all writes that happened before
4992 btrfs_drew_write_lock(&root->snapshot_lock);
4993 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
4995 btrfs_drew_write_unlock(&root->snapshot_lock);
4999 trans = btrfs_start_transaction(root, 1);
5000 if (IS_ERR(trans)) {
5001 btrfs_drew_write_unlock(&root->snapshot_lock);
5002 return PTR_ERR(trans);
5005 i_size_write(inode, newsize);
5006 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
5007 pagecache_isize_extended(inode, oldsize, newsize);
5008 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5009 btrfs_drew_write_unlock(&root->snapshot_lock);
5010 btrfs_end_transaction(trans);
5012 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5014 if (btrfs_is_zoned(fs_info)) {
5015 ret = btrfs_wait_ordered_range(inode,
5016 ALIGN(newsize, fs_info->sectorsize),
5023 * We're truncating a file that used to have good data down to
5024 * zero. Make sure any new writes to the file get on disk
5028 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5029 &BTRFS_I(inode)->runtime_flags);
5031 truncate_setsize(inode, newsize);
5033 inode_dio_wait(inode);
5035 ret = btrfs_truncate(inode, newsize == oldsize);
5036 if (ret && inode->i_nlink) {
5040 * Truncate failed, so fix up the in-memory size. We
5041 * adjusted disk_i_size down as we removed extents, so
5042 * wait for disk_i_size to be stable and then update the
5043 * in-memory size to match.
5045 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5048 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5055 static int btrfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
5058 struct inode *inode = d_inode(dentry);
5059 struct btrfs_root *root = BTRFS_I(inode)->root;
5062 if (btrfs_root_readonly(root))
5065 err = setattr_prepare(mnt_userns, dentry, attr);
5069 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5070 err = btrfs_setsize(inode, attr);
5075 if (attr->ia_valid) {
5076 setattr_copy(mnt_userns, inode, attr);
5077 inode_inc_iversion(inode);
5078 err = btrfs_dirty_inode(inode);
5080 if (!err && attr->ia_valid & ATTR_MODE)
5081 err = posix_acl_chmod(mnt_userns, inode, inode->i_mode);
5088 * While truncating the inode pages during eviction, we get the VFS
5089 * calling btrfs_invalidate_folio() against each folio of the inode. This
5090 * is slow because the calls to btrfs_invalidate_folio() result in a
5091 * huge amount of calls to lock_extent_bits() and clear_extent_bit(),
5092 * which keep merging and splitting extent_state structures over and over,
5093 * wasting lots of time.
5095 * Therefore if the inode is being evicted, let btrfs_invalidate_folio()
5096 * skip all those expensive operations on a per folio basis and do only
5097 * the ordered io finishing, while we release here the extent_map and
5098 * extent_state structures, without the excessive merging and splitting.
5100 static void evict_inode_truncate_pages(struct inode *inode)
5102 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5103 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5104 struct rb_node *node;
5106 ASSERT(inode->i_state & I_FREEING);
5107 truncate_inode_pages_final(&inode->i_data);
5109 write_lock(&map_tree->lock);
5110 while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) {
5111 struct extent_map *em;
5113 node = rb_first_cached(&map_tree->map);
5114 em = rb_entry(node, struct extent_map, rb_node);
5115 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5116 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5117 remove_extent_mapping(map_tree, em);
5118 free_extent_map(em);
5119 if (need_resched()) {
5120 write_unlock(&map_tree->lock);
5122 write_lock(&map_tree->lock);
5125 write_unlock(&map_tree->lock);
5128 * Keep looping until we have no more ranges in the io tree.
5129 * We can have ongoing bios started by readahead that have
5130 * their endio callback (extent_io.c:end_bio_extent_readpage)
5131 * still in progress (unlocked the pages in the bio but did not yet
5132 * unlocked the ranges in the io tree). Therefore this means some
5133 * ranges can still be locked and eviction started because before
5134 * submitting those bios, which are executed by a separate task (work
5135 * queue kthread), inode references (inode->i_count) were not taken
5136 * (which would be dropped in the end io callback of each bio).
5137 * Therefore here we effectively end up waiting for those bios and
5138 * anyone else holding locked ranges without having bumped the inode's
5139 * reference count - if we don't do it, when they access the inode's
5140 * io_tree to unlock a range it may be too late, leading to an
5141 * use-after-free issue.
5143 spin_lock(&io_tree->lock);
5144 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5145 struct extent_state *state;
5146 struct extent_state *cached_state = NULL;
5149 unsigned state_flags;
5151 node = rb_first(&io_tree->state);
5152 state = rb_entry(node, struct extent_state, rb_node);
5153 start = state->start;
5155 state_flags = state->state;
5156 spin_unlock(&io_tree->lock);
5158 lock_extent_bits(io_tree, start, end, &cached_state);
5161 * If still has DELALLOC flag, the extent didn't reach disk,
5162 * and its reserved space won't be freed by delayed_ref.
5163 * So we need to free its reserved space here.
5164 * (Refer to comment in btrfs_invalidate_folio, case 2)
5166 * Note, end is the bytenr of last byte, so we need + 1 here.
5168 if (state_flags & EXTENT_DELALLOC)
5169 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5172 clear_extent_bit(io_tree, start, end,
5173 EXTENT_LOCKED | EXTENT_DELALLOC |
5174 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
5178 spin_lock(&io_tree->lock);
5180 spin_unlock(&io_tree->lock);
5183 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5184 struct btrfs_block_rsv *rsv)
5186 struct btrfs_fs_info *fs_info = root->fs_info;
5187 struct btrfs_trans_handle *trans;
5188 u64 delayed_refs_extra = btrfs_calc_insert_metadata_size(fs_info, 1);
5192 * Eviction should be taking place at some place safe because of our
5193 * delayed iputs. However the normal flushing code will run delayed
5194 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5196 * We reserve the delayed_refs_extra here again because we can't use
5197 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5198 * above. We reserve our extra bit here because we generate a ton of
5199 * delayed refs activity by truncating.
5201 * BTRFS_RESERVE_FLUSH_EVICT will steal from the global_rsv if it can,
5202 * if we fail to make this reservation we can re-try without the
5203 * delayed_refs_extra so we can make some forward progress.
5205 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size + delayed_refs_extra,
5206 BTRFS_RESERVE_FLUSH_EVICT);
5208 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size,
5209 BTRFS_RESERVE_FLUSH_EVICT);
5212 "could not allocate space for delete; will truncate on mount");
5213 return ERR_PTR(-ENOSPC);
5215 delayed_refs_extra = 0;
5218 trans = btrfs_join_transaction(root);
5222 if (delayed_refs_extra) {
5223 trans->block_rsv = &fs_info->trans_block_rsv;
5224 trans->bytes_reserved = delayed_refs_extra;
5225 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5226 delayed_refs_extra, 1);
5231 void btrfs_evict_inode(struct inode *inode)
5233 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5234 struct btrfs_trans_handle *trans;
5235 struct btrfs_root *root = BTRFS_I(inode)->root;
5236 struct btrfs_block_rsv *rsv;
5239 trace_btrfs_inode_evict(inode);
5242 fsverity_cleanup_inode(inode);
5247 evict_inode_truncate_pages(inode);
5249 if (inode->i_nlink &&
5250 ((btrfs_root_refs(&root->root_item) != 0 &&
5251 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5252 btrfs_is_free_space_inode(BTRFS_I(inode))))
5255 if (is_bad_inode(inode))
5258 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5260 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5263 if (inode->i_nlink > 0) {
5264 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5265 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5270 * This makes sure the inode item in tree is uptodate and the space for
5271 * the inode update is released.
5273 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5278 * This drops any pending insert or delete operations we have for this
5279 * inode. We could have a delayed dir index deletion queued up, but
5280 * we're removing the inode completely so that'll be taken care of in
5283 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
5285 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5288 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5291 btrfs_i_size_write(BTRFS_I(inode), 0);
5294 struct btrfs_truncate_control control = {
5295 .inode = BTRFS_I(inode),
5296 .ino = btrfs_ino(BTRFS_I(inode)),
5301 trans = evict_refill_and_join(root, rsv);
5305 trans->block_rsv = rsv;
5307 ret = btrfs_truncate_inode_items(trans, root, &control);
5308 trans->block_rsv = &fs_info->trans_block_rsv;
5309 btrfs_end_transaction(trans);
5310 btrfs_btree_balance_dirty(fs_info);
5311 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5318 * Errors here aren't a big deal, it just means we leave orphan items in
5319 * the tree. They will be cleaned up on the next mount. If the inode
5320 * number gets reused, cleanup deletes the orphan item without doing
5321 * anything, and unlink reuses the existing orphan item.
5323 * If it turns out that we are dropping too many of these, we might want
5324 * to add a mechanism for retrying these after a commit.
5326 trans = evict_refill_and_join(root, rsv);
5327 if (!IS_ERR(trans)) {
5328 trans->block_rsv = rsv;
5329 btrfs_orphan_del(trans, BTRFS_I(inode));
5330 trans->block_rsv = &fs_info->trans_block_rsv;
5331 btrfs_end_transaction(trans);
5335 btrfs_free_block_rsv(fs_info, rsv);
5338 * If we didn't successfully delete, the orphan item will still be in
5339 * the tree and we'll retry on the next mount. Again, we might also want
5340 * to retry these periodically in the future.
5342 btrfs_remove_delayed_node(BTRFS_I(inode));
5343 fsverity_cleanup_inode(inode);
5348 * Return the key found in the dir entry in the location pointer, fill @type
5349 * with BTRFS_FT_*, and return 0.
5351 * If no dir entries were found, returns -ENOENT.
5352 * If found a corrupted location in dir entry, returns -EUCLEAN.
5354 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5355 struct btrfs_key *location, u8 *type)
5357 const char *name = dentry->d_name.name;
5358 int namelen = dentry->d_name.len;
5359 struct btrfs_dir_item *di;
5360 struct btrfs_path *path;
5361 struct btrfs_root *root = BTRFS_I(dir)->root;
5364 path = btrfs_alloc_path();
5368 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5370 if (IS_ERR_OR_NULL(di)) {
5371 ret = di ? PTR_ERR(di) : -ENOENT;
5375 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5376 if (location->type != BTRFS_INODE_ITEM_KEY &&
5377 location->type != BTRFS_ROOT_ITEM_KEY) {
5379 btrfs_warn(root->fs_info,
5380 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5381 __func__, name, btrfs_ino(BTRFS_I(dir)),
5382 location->objectid, location->type, location->offset);
5385 *type = btrfs_dir_type(path->nodes[0], di);
5387 btrfs_free_path(path);
5392 * when we hit a tree root in a directory, the btrfs part of the inode
5393 * needs to be changed to reflect the root directory of the tree root. This
5394 * is kind of like crossing a mount point.
5396 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5398 struct dentry *dentry,
5399 struct btrfs_key *location,
5400 struct btrfs_root **sub_root)
5402 struct btrfs_path *path;
5403 struct btrfs_root *new_root;
5404 struct btrfs_root_ref *ref;
5405 struct extent_buffer *leaf;
5406 struct btrfs_key key;
5410 path = btrfs_alloc_path();
5417 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5418 key.type = BTRFS_ROOT_REF_KEY;
5419 key.offset = location->objectid;
5421 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5428 leaf = path->nodes[0];
5429 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5430 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5431 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5434 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5435 (unsigned long)(ref + 1),
5436 dentry->d_name.len);
5440 btrfs_release_path(path);
5442 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5443 if (IS_ERR(new_root)) {
5444 err = PTR_ERR(new_root);
5448 *sub_root = new_root;
5449 location->objectid = btrfs_root_dirid(&new_root->root_item);
5450 location->type = BTRFS_INODE_ITEM_KEY;
5451 location->offset = 0;
5454 btrfs_free_path(path);
5458 static void inode_tree_add(struct inode *inode)
5460 struct btrfs_root *root = BTRFS_I(inode)->root;
5461 struct btrfs_inode *entry;
5463 struct rb_node *parent;
5464 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5465 u64 ino = btrfs_ino(BTRFS_I(inode));
5467 if (inode_unhashed(inode))
5470 spin_lock(&root->inode_lock);
5471 p = &root->inode_tree.rb_node;
5474 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5476 if (ino < btrfs_ino(entry))
5477 p = &parent->rb_left;
5478 else if (ino > btrfs_ino(entry))
5479 p = &parent->rb_right;
5481 WARN_ON(!(entry->vfs_inode.i_state &
5482 (I_WILL_FREE | I_FREEING)));
5483 rb_replace_node(parent, new, &root->inode_tree);
5484 RB_CLEAR_NODE(parent);
5485 spin_unlock(&root->inode_lock);
5489 rb_link_node(new, parent, p);
5490 rb_insert_color(new, &root->inode_tree);
5491 spin_unlock(&root->inode_lock);
5494 static void inode_tree_del(struct btrfs_inode *inode)
5496 struct btrfs_root *root = inode->root;
5499 spin_lock(&root->inode_lock);
5500 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5501 rb_erase(&inode->rb_node, &root->inode_tree);
5502 RB_CLEAR_NODE(&inode->rb_node);
5503 empty = RB_EMPTY_ROOT(&root->inode_tree);
5505 spin_unlock(&root->inode_lock);
5507 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5508 spin_lock(&root->inode_lock);
5509 empty = RB_EMPTY_ROOT(&root->inode_tree);
5510 spin_unlock(&root->inode_lock);
5512 btrfs_add_dead_root(root);
5517 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5519 struct btrfs_iget_args *args = p;
5521 inode->i_ino = args->ino;
5522 BTRFS_I(inode)->location.objectid = args->ino;
5523 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5524 BTRFS_I(inode)->location.offset = 0;
5525 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5526 BUG_ON(args->root && !BTRFS_I(inode)->root);
5530 static int btrfs_find_actor(struct inode *inode, void *opaque)
5532 struct btrfs_iget_args *args = opaque;
5534 return args->ino == BTRFS_I(inode)->location.objectid &&
5535 args->root == BTRFS_I(inode)->root;
5538 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5539 struct btrfs_root *root)
5541 struct inode *inode;
5542 struct btrfs_iget_args args;
5543 unsigned long hashval = btrfs_inode_hash(ino, root);
5548 inode = iget5_locked(s, hashval, btrfs_find_actor,
5549 btrfs_init_locked_inode,
5555 * Get an inode object given its inode number and corresponding root.
5556 * Path can be preallocated to prevent recursing back to iget through
5557 * allocator. NULL is also valid but may require an additional allocation
5560 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5561 struct btrfs_root *root, struct btrfs_path *path)
5563 struct inode *inode;
5565 inode = btrfs_iget_locked(s, ino, root);
5567 return ERR_PTR(-ENOMEM);
5569 if (inode->i_state & I_NEW) {
5572 ret = btrfs_read_locked_inode(inode, path);
5574 inode_tree_add(inode);
5575 unlock_new_inode(inode);
5579 * ret > 0 can come from btrfs_search_slot called by
5580 * btrfs_read_locked_inode, this means the inode item
5585 inode = ERR_PTR(ret);
5592 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5594 return btrfs_iget_path(s, ino, root, NULL);
5597 static struct inode *new_simple_dir(struct super_block *s,
5598 struct btrfs_key *key,
5599 struct btrfs_root *root)
5601 struct inode *inode = new_inode(s);
5604 return ERR_PTR(-ENOMEM);
5606 BTRFS_I(inode)->root = btrfs_grab_root(root);
5607 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5608 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5610 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5612 * We only need lookup, the rest is read-only and there's no inode
5613 * associated with the dentry
5615 inode->i_op = &simple_dir_inode_operations;
5616 inode->i_opflags &= ~IOP_XATTR;
5617 inode->i_fop = &simple_dir_operations;
5618 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5619 inode->i_mtime = current_time(inode);
5620 inode->i_atime = inode->i_mtime;
5621 inode->i_ctime = inode->i_mtime;
5622 BTRFS_I(inode)->i_otime = inode->i_mtime;
5627 static_assert(BTRFS_FT_UNKNOWN == FT_UNKNOWN);
5628 static_assert(BTRFS_FT_REG_FILE == FT_REG_FILE);
5629 static_assert(BTRFS_FT_DIR == FT_DIR);
5630 static_assert(BTRFS_FT_CHRDEV == FT_CHRDEV);
5631 static_assert(BTRFS_FT_BLKDEV == FT_BLKDEV);
5632 static_assert(BTRFS_FT_FIFO == FT_FIFO);
5633 static_assert(BTRFS_FT_SOCK == FT_SOCK);
5634 static_assert(BTRFS_FT_SYMLINK == FT_SYMLINK);
5636 static inline u8 btrfs_inode_type(struct inode *inode)
5638 return fs_umode_to_ftype(inode->i_mode);
5641 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5643 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5644 struct inode *inode;
5645 struct btrfs_root *root = BTRFS_I(dir)->root;
5646 struct btrfs_root *sub_root = root;
5647 struct btrfs_key location;
5651 if (dentry->d_name.len > BTRFS_NAME_LEN)
5652 return ERR_PTR(-ENAMETOOLONG);
5654 ret = btrfs_inode_by_name(dir, dentry, &location, &di_type);
5656 return ERR_PTR(ret);
5658 if (location.type == BTRFS_INODE_ITEM_KEY) {
5659 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5663 /* Do extra check against inode mode with di_type */
5664 if (btrfs_inode_type(inode) != di_type) {
5666 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5667 inode->i_mode, btrfs_inode_type(inode),
5670 return ERR_PTR(-EUCLEAN);
5675 ret = fixup_tree_root_location(fs_info, dir, dentry,
5676 &location, &sub_root);
5679 inode = ERR_PTR(ret);
5681 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5683 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5685 if (root != sub_root)
5686 btrfs_put_root(sub_root);
5688 if (!IS_ERR(inode) && root != sub_root) {
5689 down_read(&fs_info->cleanup_work_sem);
5690 if (!sb_rdonly(inode->i_sb))
5691 ret = btrfs_orphan_cleanup(sub_root);
5692 up_read(&fs_info->cleanup_work_sem);
5695 inode = ERR_PTR(ret);
5702 static int btrfs_dentry_delete(const struct dentry *dentry)
5704 struct btrfs_root *root;
5705 struct inode *inode = d_inode(dentry);
5707 if (!inode && !IS_ROOT(dentry))
5708 inode = d_inode(dentry->d_parent);
5711 root = BTRFS_I(inode)->root;
5712 if (btrfs_root_refs(&root->root_item) == 0)
5715 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5721 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5724 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5726 if (inode == ERR_PTR(-ENOENT))
5728 return d_splice_alias(inode, dentry);
5732 * All this infrastructure exists because dir_emit can fault, and we are holding
5733 * the tree lock when doing readdir. For now just allocate a buffer and copy
5734 * our information into that, and then dir_emit from the buffer. This is
5735 * similar to what NFS does, only we don't keep the buffer around in pagecache
5736 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5737 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5740 static int btrfs_opendir(struct inode *inode, struct file *file)
5742 struct btrfs_file_private *private;
5744 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5747 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5748 if (!private->filldir_buf) {
5752 file->private_data = private;
5763 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5766 struct dir_entry *entry = addr;
5767 char *name = (char *)(entry + 1);
5769 ctx->pos = get_unaligned(&entry->offset);
5770 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5771 get_unaligned(&entry->ino),
5772 get_unaligned(&entry->type)))
5774 addr += sizeof(struct dir_entry) +
5775 get_unaligned(&entry->name_len);
5781 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5783 struct inode *inode = file_inode(file);
5784 struct btrfs_root *root = BTRFS_I(inode)->root;
5785 struct btrfs_file_private *private = file->private_data;
5786 struct btrfs_dir_item *di;
5787 struct btrfs_key key;
5788 struct btrfs_key found_key;
5789 struct btrfs_path *path;
5791 struct list_head ins_list;
5792 struct list_head del_list;
5794 struct extent_buffer *leaf;
5801 struct btrfs_key location;
5803 if (!dir_emit_dots(file, ctx))
5806 path = btrfs_alloc_path();
5810 addr = private->filldir_buf;
5811 path->reada = READA_FORWARD;
5813 INIT_LIST_HEAD(&ins_list);
5814 INIT_LIST_HEAD(&del_list);
5815 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5818 key.type = BTRFS_DIR_INDEX_KEY;
5819 key.offset = ctx->pos;
5820 key.objectid = btrfs_ino(BTRFS_I(inode));
5822 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5827 struct dir_entry *entry;
5829 leaf = path->nodes[0];
5830 slot = path->slots[0];
5831 if (slot >= btrfs_header_nritems(leaf)) {
5832 ret = btrfs_next_leaf(root, path);
5840 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5842 if (found_key.objectid != key.objectid)
5844 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5846 if (found_key.offset < ctx->pos)
5848 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5850 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5851 name_len = btrfs_dir_name_len(leaf, di);
5852 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5854 btrfs_release_path(path);
5855 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5858 addr = private->filldir_buf;
5865 put_unaligned(name_len, &entry->name_len);
5866 name_ptr = (char *)(entry + 1);
5867 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5869 put_unaligned(fs_ftype_to_dtype(btrfs_dir_type(leaf, di)),
5871 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5872 put_unaligned(location.objectid, &entry->ino);
5873 put_unaligned(found_key.offset, &entry->offset);
5875 addr += sizeof(struct dir_entry) + name_len;
5876 total_len += sizeof(struct dir_entry) + name_len;
5880 btrfs_release_path(path);
5882 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5886 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5891 * Stop new entries from being returned after we return the last
5894 * New directory entries are assigned a strictly increasing
5895 * offset. This means that new entries created during readdir
5896 * are *guaranteed* to be seen in the future by that readdir.
5897 * This has broken buggy programs which operate on names as
5898 * they're returned by readdir. Until we re-use freed offsets
5899 * we have this hack to stop new entries from being returned
5900 * under the assumption that they'll never reach this huge
5903 * This is being careful not to overflow 32bit loff_t unless the
5904 * last entry requires it because doing so has broken 32bit apps
5907 if (ctx->pos >= INT_MAX)
5908 ctx->pos = LLONG_MAX;
5915 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5916 btrfs_free_path(path);
5921 * This is somewhat expensive, updating the tree every time the
5922 * inode changes. But, it is most likely to find the inode in cache.
5923 * FIXME, needs more benchmarking...there are no reasons other than performance
5924 * to keep or drop this code.
5926 static int btrfs_dirty_inode(struct inode *inode)
5928 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5929 struct btrfs_root *root = BTRFS_I(inode)->root;
5930 struct btrfs_trans_handle *trans;
5933 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5936 trans = btrfs_join_transaction(root);
5938 return PTR_ERR(trans);
5940 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5941 if (ret && (ret == -ENOSPC || ret == -EDQUOT)) {
5942 /* whoops, lets try again with the full transaction */
5943 btrfs_end_transaction(trans);
5944 trans = btrfs_start_transaction(root, 1);
5946 return PTR_ERR(trans);
5948 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5950 btrfs_end_transaction(trans);
5951 if (BTRFS_I(inode)->delayed_node)
5952 btrfs_balance_delayed_items(fs_info);
5958 * This is a copy of file_update_time. We need this so we can return error on
5959 * ENOSPC for updating the inode in the case of file write and mmap writes.
5961 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
5964 struct btrfs_root *root = BTRFS_I(inode)->root;
5965 bool dirty = flags & ~S_VERSION;
5967 if (btrfs_root_readonly(root))
5970 if (flags & S_VERSION)
5971 dirty |= inode_maybe_inc_iversion(inode, dirty);
5972 if (flags & S_CTIME)
5973 inode->i_ctime = *now;
5974 if (flags & S_MTIME)
5975 inode->i_mtime = *now;
5976 if (flags & S_ATIME)
5977 inode->i_atime = *now;
5978 return dirty ? btrfs_dirty_inode(inode) : 0;
5982 * find the highest existing sequence number in a directory
5983 * and then set the in-memory index_cnt variable to reflect
5984 * free sequence numbers
5986 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
5988 struct btrfs_root *root = inode->root;
5989 struct btrfs_key key, found_key;
5990 struct btrfs_path *path;
5991 struct extent_buffer *leaf;
5994 key.objectid = btrfs_ino(inode);
5995 key.type = BTRFS_DIR_INDEX_KEY;
5996 key.offset = (u64)-1;
5998 path = btrfs_alloc_path();
6002 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6005 /* FIXME: we should be able to handle this */
6010 if (path->slots[0] == 0) {
6011 inode->index_cnt = BTRFS_DIR_START_INDEX;
6017 leaf = path->nodes[0];
6018 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6020 if (found_key.objectid != btrfs_ino(inode) ||
6021 found_key.type != BTRFS_DIR_INDEX_KEY) {
6022 inode->index_cnt = BTRFS_DIR_START_INDEX;
6026 inode->index_cnt = found_key.offset + 1;
6028 btrfs_free_path(path);
6033 * helper to find a free sequence number in a given directory. This current
6034 * code is very simple, later versions will do smarter things in the btree
6036 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6040 if (dir->index_cnt == (u64)-1) {
6041 ret = btrfs_inode_delayed_dir_index_count(dir);
6043 ret = btrfs_set_inode_index_count(dir);
6049 *index = dir->index_cnt;
6055 static int btrfs_insert_inode_locked(struct inode *inode)
6057 struct btrfs_iget_args args;
6059 args.ino = BTRFS_I(inode)->location.objectid;
6060 args.root = BTRFS_I(inode)->root;
6062 return insert_inode_locked4(inode,
6063 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6064 btrfs_find_actor, &args);
6068 * Inherit flags from the parent inode.
6070 * Currently only the compression flags and the cow flags are inherited.
6072 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6079 flags = BTRFS_I(dir)->flags;
6081 if (flags & BTRFS_INODE_NOCOMPRESS) {
6082 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6083 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6084 } else if (flags & BTRFS_INODE_COMPRESS) {
6085 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6086 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6089 if (flags & BTRFS_INODE_NODATACOW) {
6090 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6091 if (S_ISREG(inode->i_mode))
6092 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6095 btrfs_sync_inode_flags_to_i_flags(inode);
6098 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6099 struct btrfs_root *root,
6100 struct user_namespace *mnt_userns,
6102 const char *name, int name_len,
6103 u64 ref_objectid, u64 objectid,
6104 umode_t mode, u64 *index)
6106 struct btrfs_fs_info *fs_info = root->fs_info;
6107 struct inode *inode;
6108 struct btrfs_inode_item *inode_item;
6109 struct btrfs_key *location;
6110 struct btrfs_path *path;
6111 struct btrfs_inode_ref *ref;
6112 struct btrfs_key key[2];
6114 struct btrfs_item_batch batch;
6116 unsigned int nofs_flag;
6119 path = btrfs_alloc_path();
6121 return ERR_PTR(-ENOMEM);
6123 nofs_flag = memalloc_nofs_save();
6124 inode = new_inode(fs_info->sb);
6125 memalloc_nofs_restore(nofs_flag);
6127 btrfs_free_path(path);
6128 return ERR_PTR(-ENOMEM);
6132 * O_TMPFILE, set link count to 0, so that after this point,
6133 * we fill in an inode item with the correct link count.
6136 set_nlink(inode, 0);
6139 * we have to initialize this early, so we can reclaim the inode
6140 * number if we fail afterwards in this function.
6142 inode->i_ino = objectid;
6145 trace_btrfs_inode_request(dir);
6147 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6149 btrfs_free_path(path);
6151 return ERR_PTR(ret);
6157 * index_cnt is ignored for everything but a dir,
6158 * btrfs_set_inode_index_count has an explanation for the magic
6161 BTRFS_I(inode)->index_cnt = 2;
6162 BTRFS_I(inode)->dir_index = *index;
6163 BTRFS_I(inode)->root = btrfs_grab_root(root);
6164 BTRFS_I(inode)->generation = trans->transid;
6165 inode->i_generation = BTRFS_I(inode)->generation;
6168 * We could have gotten an inode number from somebody who was fsynced
6169 * and then removed in this same transaction, so let's just set full
6170 * sync since it will be a full sync anyway and this will blow away the
6171 * old info in the log.
6173 btrfs_set_inode_full_sync(BTRFS_I(inode));
6175 key[0].objectid = objectid;
6176 key[0].type = BTRFS_INODE_ITEM_KEY;
6179 sizes[0] = sizeof(struct btrfs_inode_item);
6183 * Start new inodes with an inode_ref. This is slightly more
6184 * efficient for small numbers of hard links since they will
6185 * be packed into one item. Extended refs will kick in if we
6186 * add more hard links than can fit in the ref item.
6188 key[1].objectid = objectid;
6189 key[1].type = BTRFS_INODE_REF_KEY;
6190 key[1].offset = ref_objectid;
6192 sizes[1] = name_len + sizeof(*ref);
6195 location = &BTRFS_I(inode)->location;
6196 location->objectid = objectid;
6197 location->offset = 0;
6198 location->type = BTRFS_INODE_ITEM_KEY;
6200 ret = btrfs_insert_inode_locked(inode);
6206 batch.keys = &key[0];
6207 batch.data_sizes = &sizes[0];
6208 batch.total_data_size = sizes[0] + (name ? sizes[1] : 0);
6209 batch.nr = name ? 2 : 1;
6210 ret = btrfs_insert_empty_items(trans, root, path, &batch);
6214 inode_init_owner(mnt_userns, inode, dir, mode);
6215 inode_set_bytes(inode, 0);
6217 inode->i_mtime = current_time(inode);
6218 inode->i_atime = inode->i_mtime;
6219 inode->i_ctime = inode->i_mtime;
6220 BTRFS_I(inode)->i_otime = inode->i_mtime;
6222 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6223 struct btrfs_inode_item);
6224 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6225 sizeof(*inode_item));
6226 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6229 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6230 struct btrfs_inode_ref);
6231 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6232 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6233 ptr = (unsigned long)(ref + 1);
6234 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6237 btrfs_mark_buffer_dirty(path->nodes[0]);
6238 btrfs_free_path(path);
6240 btrfs_inherit_iflags(inode, dir);
6242 if (S_ISREG(mode)) {
6243 if (btrfs_test_opt(fs_info, NODATASUM))
6244 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6245 if (btrfs_test_opt(fs_info, NODATACOW))
6246 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6247 BTRFS_INODE_NODATASUM;
6250 inode_tree_add(inode);
6252 trace_btrfs_inode_new(inode);
6253 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6255 btrfs_update_root_times(trans, root);
6257 ret = btrfs_inode_inherit_props(trans, inode, dir);
6260 "error inheriting props for ino %llu (root %llu): %d",
6261 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6266 discard_new_inode(inode);
6269 BTRFS_I(dir)->index_cnt--;
6270 btrfs_free_path(path);
6271 return ERR_PTR(ret);
6275 * utility function to add 'inode' into 'parent_inode' with
6276 * a give name and a given sequence number.
6277 * if 'add_backref' is true, also insert a backref from the
6278 * inode to the parent directory.
6280 int btrfs_add_link(struct btrfs_trans_handle *trans,
6281 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6282 const char *name, int name_len, int add_backref, u64 index)
6285 struct btrfs_key key;
6286 struct btrfs_root *root = parent_inode->root;
6287 u64 ino = btrfs_ino(inode);
6288 u64 parent_ino = btrfs_ino(parent_inode);
6290 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6291 memcpy(&key, &inode->root->root_key, sizeof(key));
6294 key.type = BTRFS_INODE_ITEM_KEY;
6298 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6299 ret = btrfs_add_root_ref(trans, key.objectid,
6300 root->root_key.objectid, parent_ino,
6301 index, name, name_len);
6302 } else if (add_backref) {
6303 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6307 /* Nothing to clean up yet */
6311 ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key,
6312 btrfs_inode_type(&inode->vfs_inode), index);
6313 if (ret == -EEXIST || ret == -EOVERFLOW)
6316 btrfs_abort_transaction(trans, ret);
6320 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6322 inode_inc_iversion(&parent_inode->vfs_inode);
6324 * If we are replaying a log tree, we do not want to update the mtime
6325 * and ctime of the parent directory with the current time, since the
6326 * log replay procedure is responsible for setting them to their correct
6327 * values (the ones it had when the fsync was done).
6329 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
6330 struct timespec64 now = current_time(&parent_inode->vfs_inode);
6332 parent_inode->vfs_inode.i_mtime = now;
6333 parent_inode->vfs_inode.i_ctime = now;
6335 ret = btrfs_update_inode(trans, root, parent_inode);
6337 btrfs_abort_transaction(trans, ret);
6341 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6344 err = btrfs_del_root_ref(trans, key.objectid,
6345 root->root_key.objectid, parent_ino,
6346 &local_index, name, name_len);
6348 btrfs_abort_transaction(trans, err);
6349 } else if (add_backref) {
6353 err = btrfs_del_inode_ref(trans, root, name, name_len,
6354 ino, parent_ino, &local_index);
6356 btrfs_abort_transaction(trans, err);
6359 /* Return the original error code */
6363 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6364 struct btrfs_inode *dir, struct dentry *dentry,
6365 struct btrfs_inode *inode, int backref, u64 index)
6367 int err = btrfs_add_link(trans, dir, inode,
6368 dentry->d_name.name, dentry->d_name.len,
6375 static int btrfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
6376 struct dentry *dentry, umode_t mode, dev_t rdev)
6378 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6379 struct btrfs_trans_handle *trans;
6380 struct btrfs_root *root = BTRFS_I(dir)->root;
6381 struct inode *inode = NULL;
6387 * 2 for inode item and ref
6389 * 1 for xattr if selinux is on
6391 trans = btrfs_start_transaction(root, 5);
6393 return PTR_ERR(trans);
6395 err = btrfs_get_free_objectid(root, &objectid);
6399 inode = btrfs_new_inode(trans, root, mnt_userns, dir,
6400 dentry->d_name.name, dentry->d_name.len,
6401 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
6402 if (IS_ERR(inode)) {
6403 err = PTR_ERR(inode);
6409 * If the active LSM wants to access the inode during
6410 * d_instantiate it needs these. Smack checks to see
6411 * if the filesystem supports xattrs by looking at the
6414 inode->i_op = &btrfs_special_inode_operations;
6415 init_special_inode(inode, inode->i_mode, rdev);
6417 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6421 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6426 btrfs_update_inode(trans, root, BTRFS_I(inode));
6427 d_instantiate_new(dentry, inode);
6430 btrfs_end_transaction(trans);
6431 btrfs_btree_balance_dirty(fs_info);
6433 inode_dec_link_count(inode);
6434 discard_new_inode(inode);
6439 static int btrfs_create(struct user_namespace *mnt_userns, struct inode *dir,
6440 struct dentry *dentry, umode_t mode, bool excl)
6442 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6443 struct btrfs_trans_handle *trans;
6444 struct btrfs_root *root = BTRFS_I(dir)->root;
6445 struct inode *inode = NULL;
6451 * 2 for inode item and ref
6453 * 1 for xattr if selinux is on
6455 trans = btrfs_start_transaction(root, 5);
6457 return PTR_ERR(trans);
6459 err = btrfs_get_free_objectid(root, &objectid);
6463 inode = btrfs_new_inode(trans, root, mnt_userns, dir,
6464 dentry->d_name.name, dentry->d_name.len,
6465 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
6466 if (IS_ERR(inode)) {
6467 err = PTR_ERR(inode);
6472 * If the active LSM wants to access the inode during
6473 * d_instantiate it needs these. Smack checks to see
6474 * if the filesystem supports xattrs by looking at the
6477 inode->i_fop = &btrfs_file_operations;
6478 inode->i_op = &btrfs_file_inode_operations;
6479 inode->i_mapping->a_ops = &btrfs_aops;
6481 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6485 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6489 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6494 d_instantiate_new(dentry, inode);
6497 btrfs_end_transaction(trans);
6499 inode_dec_link_count(inode);
6500 discard_new_inode(inode);
6502 btrfs_btree_balance_dirty(fs_info);
6506 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6507 struct dentry *dentry)
6509 struct btrfs_trans_handle *trans = NULL;
6510 struct btrfs_root *root = BTRFS_I(dir)->root;
6511 struct inode *inode = d_inode(old_dentry);
6512 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6517 /* do not allow sys_link's with other subvols of the same device */
6518 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6521 if (inode->i_nlink >= BTRFS_LINK_MAX)
6524 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6529 * 2 items for inode and inode ref
6530 * 2 items for dir items
6531 * 1 item for parent inode
6532 * 1 item for orphan item deletion if O_TMPFILE
6534 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6535 if (IS_ERR(trans)) {
6536 err = PTR_ERR(trans);
6541 /* There are several dir indexes for this inode, clear the cache. */
6542 BTRFS_I(inode)->dir_index = 0ULL;
6544 inode_inc_iversion(inode);
6545 inode->i_ctime = current_time(inode);
6547 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6549 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6555 struct dentry *parent = dentry->d_parent;
6557 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6560 if (inode->i_nlink == 1) {
6562 * If new hard link count is 1, it's a file created
6563 * with open(2) O_TMPFILE flag.
6565 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6569 d_instantiate(dentry, inode);
6570 btrfs_log_new_name(trans, old_dentry, NULL, 0, parent);
6575 btrfs_end_transaction(trans);
6577 inode_dec_link_count(inode);
6580 btrfs_btree_balance_dirty(fs_info);
6584 static int btrfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
6585 struct dentry *dentry, umode_t mode)
6587 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6588 struct inode *inode = NULL;
6589 struct btrfs_trans_handle *trans;
6590 struct btrfs_root *root = BTRFS_I(dir)->root;
6596 * 2 items for inode and ref
6597 * 2 items for dir items
6598 * 1 for xattr if selinux is on
6600 trans = btrfs_start_transaction(root, 5);
6602 return PTR_ERR(trans);
6604 err = btrfs_get_free_objectid(root, &objectid);
6608 inode = btrfs_new_inode(trans, root, mnt_userns, dir,
6609 dentry->d_name.name, dentry->d_name.len,
6610 btrfs_ino(BTRFS_I(dir)), objectid,
6611 S_IFDIR | mode, &index);
6612 if (IS_ERR(inode)) {
6613 err = PTR_ERR(inode);
6618 /* these must be set before we unlock the inode */
6619 inode->i_op = &btrfs_dir_inode_operations;
6620 inode->i_fop = &btrfs_dir_file_operations;
6622 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6626 btrfs_i_size_write(BTRFS_I(inode), 0);
6627 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6631 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6632 dentry->d_name.name,
6633 dentry->d_name.len, 0, index);
6637 d_instantiate_new(dentry, inode);
6640 btrfs_end_transaction(trans);
6642 inode_dec_link_count(inode);
6643 discard_new_inode(inode);
6645 btrfs_btree_balance_dirty(fs_info);
6649 static noinline int uncompress_inline(struct btrfs_path *path,
6651 size_t pg_offset, u64 extent_offset,
6652 struct btrfs_file_extent_item *item)
6655 struct extent_buffer *leaf = path->nodes[0];
6658 unsigned long inline_size;
6662 WARN_ON(pg_offset != 0);
6663 compress_type = btrfs_file_extent_compression(leaf, item);
6664 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6665 inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]);
6666 tmp = kmalloc(inline_size, GFP_NOFS);
6669 ptr = btrfs_file_extent_inline_start(item);
6671 read_extent_buffer(leaf, tmp, ptr, inline_size);
6673 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6674 ret = btrfs_decompress(compress_type, tmp, page,
6675 extent_offset, inline_size, max_size);
6678 * decompression code contains a memset to fill in any space between the end
6679 * of the uncompressed data and the end of max_size in case the decompressed
6680 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6681 * the end of an inline extent and the beginning of the next block, so we
6682 * cover that region here.
6685 if (max_size + pg_offset < PAGE_SIZE)
6686 memzero_page(page, pg_offset + max_size,
6687 PAGE_SIZE - max_size - pg_offset);
6693 * btrfs_get_extent - Lookup the first extent overlapping a range in a file.
6694 * @inode: file to search in
6695 * @page: page to read extent data into if the extent is inline
6696 * @pg_offset: offset into @page to copy to
6697 * @start: file offset
6698 * @len: length of range starting at @start
6700 * This returns the first &struct extent_map which overlaps with the given
6701 * range, reading it from the B-tree and caching it if necessary. Note that
6702 * there may be more extents which overlap the given range after the returned
6705 * If @page is not NULL and the extent is inline, this also reads the extent
6706 * data directly into the page and marks the extent up to date in the io_tree.
6708 * Return: ERR_PTR on error, non-NULL extent_map on success.
6710 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6711 struct page *page, size_t pg_offset,
6714 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6716 u64 extent_start = 0;
6718 u64 objectid = btrfs_ino(inode);
6719 int extent_type = -1;
6720 struct btrfs_path *path = NULL;
6721 struct btrfs_root *root = inode->root;
6722 struct btrfs_file_extent_item *item;
6723 struct extent_buffer *leaf;
6724 struct btrfs_key found_key;
6725 struct extent_map *em = NULL;
6726 struct extent_map_tree *em_tree = &inode->extent_tree;
6727 struct extent_io_tree *io_tree = &inode->io_tree;
6729 read_lock(&em_tree->lock);
6730 em = lookup_extent_mapping(em_tree, start, len);
6731 read_unlock(&em_tree->lock);
6734 if (em->start > start || em->start + em->len <= start)
6735 free_extent_map(em);
6736 else if (em->block_start == EXTENT_MAP_INLINE && page)
6737 free_extent_map(em);
6741 em = alloc_extent_map();
6746 em->start = EXTENT_MAP_HOLE;
6747 em->orig_start = EXTENT_MAP_HOLE;
6749 em->block_len = (u64)-1;
6751 path = btrfs_alloc_path();
6757 /* Chances are we'll be called again, so go ahead and do readahead */
6758 path->reada = READA_FORWARD;
6761 * The same explanation in load_free_space_cache applies here as well,
6762 * we only read when we're loading the free space cache, and at that
6763 * point the commit_root has everything we need.
6765 if (btrfs_is_free_space_inode(inode)) {
6766 path->search_commit_root = 1;
6767 path->skip_locking = 1;
6770 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6773 } else if (ret > 0) {
6774 if (path->slots[0] == 0)
6780 leaf = path->nodes[0];
6781 item = btrfs_item_ptr(leaf, path->slots[0],
6782 struct btrfs_file_extent_item);
6783 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6784 if (found_key.objectid != objectid ||
6785 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6787 * If we backup past the first extent we want to move forward
6788 * and see if there is an extent in front of us, otherwise we'll
6789 * say there is a hole for our whole search range which can
6796 extent_type = btrfs_file_extent_type(leaf, item);
6797 extent_start = found_key.offset;
6798 extent_end = btrfs_file_extent_end(path);
6799 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6800 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6801 /* Only regular file could have regular/prealloc extent */
6802 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6805 "regular/prealloc extent found for non-regular inode %llu",
6809 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6811 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6812 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6817 if (start >= extent_end) {
6819 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6820 ret = btrfs_next_leaf(root, path);
6826 leaf = path->nodes[0];
6828 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6829 if (found_key.objectid != objectid ||
6830 found_key.type != BTRFS_EXTENT_DATA_KEY)
6832 if (start + len <= found_key.offset)
6834 if (start > found_key.offset)
6837 /* New extent overlaps with existing one */
6839 em->orig_start = start;
6840 em->len = found_key.offset - start;
6841 em->block_start = EXTENT_MAP_HOLE;
6845 btrfs_extent_item_to_extent_map(inode, path, item, !page, em);
6847 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6848 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6850 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6854 size_t extent_offset;
6860 size = btrfs_file_extent_ram_bytes(leaf, item);
6861 extent_offset = page_offset(page) + pg_offset - extent_start;
6862 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6863 size - extent_offset);
6864 em->start = extent_start + extent_offset;
6865 em->len = ALIGN(copy_size, fs_info->sectorsize);
6866 em->orig_block_len = em->len;
6867 em->orig_start = em->start;
6868 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6870 if (!PageUptodate(page)) {
6871 if (btrfs_file_extent_compression(leaf, item) !=
6872 BTRFS_COMPRESS_NONE) {
6873 ret = uncompress_inline(path, page, pg_offset,
6874 extent_offset, item);
6878 map = kmap_local_page(page);
6879 read_extent_buffer(leaf, map + pg_offset, ptr,
6881 if (pg_offset + copy_size < PAGE_SIZE) {
6882 memset(map + pg_offset + copy_size, 0,
6883 PAGE_SIZE - pg_offset -
6888 flush_dcache_page(page);
6890 set_extent_uptodate(io_tree, em->start,
6891 extent_map_end(em) - 1, NULL, GFP_NOFS);
6896 em->orig_start = start;
6898 em->block_start = EXTENT_MAP_HOLE;
6901 btrfs_release_path(path);
6902 if (em->start > start || extent_map_end(em) <= start) {
6904 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6905 em->start, em->len, start, len);
6910 write_lock(&em_tree->lock);
6911 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6912 write_unlock(&em_tree->lock);
6914 btrfs_free_path(path);
6916 trace_btrfs_get_extent(root, inode, em);
6919 free_extent_map(em);
6920 return ERR_PTR(ret);
6925 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
6928 struct extent_map *em;
6929 struct extent_map *hole_em = NULL;
6930 u64 delalloc_start = start;
6936 em = btrfs_get_extent(inode, NULL, 0, start, len);
6940 * If our em maps to:
6942 * - a pre-alloc extent,
6943 * there might actually be delalloc bytes behind it.
6945 if (em->block_start != EXTENT_MAP_HOLE &&
6946 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6951 /* check to see if we've wrapped (len == -1 or similar) */
6960 /* ok, we didn't find anything, lets look for delalloc */
6961 delalloc_len = count_range_bits(&inode->io_tree, &delalloc_start,
6962 end, len, EXTENT_DELALLOC, 1);
6963 delalloc_end = delalloc_start + delalloc_len;
6964 if (delalloc_end < delalloc_start)
6965 delalloc_end = (u64)-1;
6968 * We didn't find anything useful, return the original results from
6971 if (delalloc_start > end || delalloc_end <= start) {
6978 * Adjust the delalloc_start to make sure it doesn't go backwards from
6979 * the start they passed in
6981 delalloc_start = max(start, delalloc_start);
6982 delalloc_len = delalloc_end - delalloc_start;
6984 if (delalloc_len > 0) {
6987 const u64 hole_end = extent_map_end(hole_em);
6989 em = alloc_extent_map();
6997 * When btrfs_get_extent can't find anything it returns one
7000 * Make sure what it found really fits our range, and adjust to
7001 * make sure it is based on the start from the caller
7003 if (hole_end <= start || hole_em->start > end) {
7004 free_extent_map(hole_em);
7007 hole_start = max(hole_em->start, start);
7008 hole_len = hole_end - hole_start;
7011 if (hole_em && delalloc_start > hole_start) {
7013 * Our hole starts before our delalloc, so we have to
7014 * return just the parts of the hole that go until the
7017 em->len = min(hole_len, delalloc_start - hole_start);
7018 em->start = hole_start;
7019 em->orig_start = hole_start;
7021 * Don't adjust block start at all, it is fixed at
7024 em->block_start = hole_em->block_start;
7025 em->block_len = hole_len;
7026 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7027 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7030 * Hole is out of passed range or it starts after
7033 em->start = delalloc_start;
7034 em->len = delalloc_len;
7035 em->orig_start = delalloc_start;
7036 em->block_start = EXTENT_MAP_DELALLOC;
7037 em->block_len = delalloc_len;
7044 free_extent_map(hole_em);
7046 free_extent_map(em);
7047 return ERR_PTR(err);
7052 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
7055 const u64 orig_start,
7056 const u64 block_start,
7057 const u64 block_len,
7058 const u64 orig_block_len,
7059 const u64 ram_bytes,
7062 struct extent_map *em = NULL;
7065 if (type != BTRFS_ORDERED_NOCOW) {
7066 em = create_io_em(inode, start, len, orig_start, block_start,
7067 block_len, orig_block_len, ram_bytes,
7068 BTRFS_COMPRESS_NONE, /* compress_type */
7073 ret = btrfs_add_ordered_extent(inode, start, len, len, block_start,
7076 (1 << BTRFS_ORDERED_DIRECT),
7077 BTRFS_COMPRESS_NONE);
7080 free_extent_map(em);
7081 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
7090 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
7093 struct btrfs_root *root = inode->root;
7094 struct btrfs_fs_info *fs_info = root->fs_info;
7095 struct extent_map *em;
7096 struct btrfs_key ins;
7100 alloc_hint = get_extent_allocation_hint(inode, start, len);
7101 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7102 0, alloc_hint, &ins, 1, 1);
7104 return ERR_PTR(ret);
7106 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7107 ins.objectid, ins.offset, ins.offset,
7108 ins.offset, BTRFS_ORDERED_REGULAR);
7109 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7111 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7117 static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
7119 struct btrfs_block_group *block_group;
7120 bool readonly = false;
7122 block_group = btrfs_lookup_block_group(fs_info, bytenr);
7123 if (!block_group || block_group->ro)
7126 btrfs_put_block_group(block_group);
7131 * Check if we can do nocow write into the range [@offset, @offset + @len)
7133 * @offset: File offset
7134 * @len: The length to write, will be updated to the nocow writeable
7136 * @orig_start: (optional) Return the original file offset of the file extent
7137 * @orig_len: (optional) Return the original on-disk length of the file extent
7138 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7139 * @strict: if true, omit optimizations that might force us into unnecessary
7140 * cow. e.g., don't trust generation number.
7143 * >0 and update @len if we can do nocow write
7144 * 0 if we can't do nocow write
7145 * <0 if error happened
7147 * NOTE: This only checks the file extents, caller is responsible to wait for
7148 * any ordered extents.
7150 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7151 u64 *orig_start, u64 *orig_block_len,
7152 u64 *ram_bytes, bool strict)
7154 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7155 struct btrfs_path *path;
7157 struct extent_buffer *leaf;
7158 struct btrfs_root *root = BTRFS_I(inode)->root;
7159 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7160 struct btrfs_file_extent_item *fi;
7161 struct btrfs_key key;
7168 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7170 path = btrfs_alloc_path();
7174 ret = btrfs_lookup_file_extent(NULL, root, path,
7175 btrfs_ino(BTRFS_I(inode)), offset, 0);
7179 slot = path->slots[0];
7182 /* can't find the item, must cow */
7189 leaf = path->nodes[0];
7190 btrfs_item_key_to_cpu(leaf, &key, slot);
7191 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7192 key.type != BTRFS_EXTENT_DATA_KEY) {
7193 /* not our file or wrong item type, must cow */
7197 if (key.offset > offset) {
7198 /* Wrong offset, must cow */
7202 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7203 found_type = btrfs_file_extent_type(leaf, fi);
7204 if (found_type != BTRFS_FILE_EXTENT_REG &&
7205 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7206 /* not a regular extent, must cow */
7210 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7213 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7214 if (extent_end <= offset)
7217 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7218 if (disk_bytenr == 0)
7221 if (btrfs_file_extent_compression(leaf, fi) ||
7222 btrfs_file_extent_encryption(leaf, fi) ||
7223 btrfs_file_extent_other_encoding(leaf, fi))
7227 * Do the same check as in btrfs_cross_ref_exist but without the
7228 * unnecessary search.
7231 (btrfs_file_extent_generation(leaf, fi) <=
7232 btrfs_root_last_snapshot(&root->root_item)))
7235 backref_offset = btrfs_file_extent_offset(leaf, fi);
7238 *orig_start = key.offset - backref_offset;
7239 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7240 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7243 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7246 num_bytes = min(offset + *len, extent_end) - offset;
7247 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7250 range_end = round_up(offset + num_bytes,
7251 root->fs_info->sectorsize) - 1;
7252 ret = test_range_bit(io_tree, offset, range_end,
7253 EXTENT_DELALLOC, 0, NULL);
7260 btrfs_release_path(path);
7263 * look for other files referencing this extent, if we
7264 * find any we must cow
7267 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7268 key.offset - backref_offset, disk_bytenr,
7276 * adjust disk_bytenr and num_bytes to cover just the bytes
7277 * in this extent we are about to write. If there
7278 * are any csums in that range we have to cow in order
7279 * to keep the csums correct
7281 disk_bytenr += backref_offset;
7282 disk_bytenr += offset - key.offset;
7283 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7286 * all of the above have passed, it is safe to overwrite this extent
7292 btrfs_free_path(path);
7296 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7297 struct extent_state **cached_state, bool writing)
7299 struct btrfs_ordered_extent *ordered;
7303 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7306 * We're concerned with the entire range that we're going to be
7307 * doing DIO to, so we need to make sure there's no ordered
7308 * extents in this range.
7310 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7311 lockend - lockstart + 1);
7314 * We need to make sure there are no buffered pages in this
7315 * range either, we could have raced between the invalidate in
7316 * generic_file_direct_write and locking the extent. The
7317 * invalidate needs to happen so that reads after a write do not
7321 (!writing || !filemap_range_has_page(inode->i_mapping,
7322 lockstart, lockend)))
7325 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7330 * If we are doing a DIO read and the ordered extent we
7331 * found is for a buffered write, we can not wait for it
7332 * to complete and retry, because if we do so we can
7333 * deadlock with concurrent buffered writes on page
7334 * locks. This happens only if our DIO read covers more
7335 * than one extent map, if at this point has already
7336 * created an ordered extent for a previous extent map
7337 * and locked its range in the inode's io tree, and a
7338 * concurrent write against that previous extent map's
7339 * range and this range started (we unlock the ranges
7340 * in the io tree only when the bios complete and
7341 * buffered writes always lock pages before attempting
7342 * to lock range in the io tree).
7345 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7346 btrfs_start_ordered_extent(ordered, 1);
7349 btrfs_put_ordered_extent(ordered);
7352 * We could trigger writeback for this range (and wait
7353 * for it to complete) and then invalidate the pages for
7354 * this range (through invalidate_inode_pages2_range()),
7355 * but that can lead us to a deadlock with a concurrent
7356 * call to readahead (a buffered read or a defrag call
7357 * triggered a readahead) on a page lock due to an
7358 * ordered dio extent we created before but did not have
7359 * yet a corresponding bio submitted (whence it can not
7360 * complete), which makes readahead wait for that
7361 * ordered extent to complete while holding a lock on
7376 /* The callers of this must take lock_extent() */
7377 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7378 u64 len, u64 orig_start, u64 block_start,
7379 u64 block_len, u64 orig_block_len,
7380 u64 ram_bytes, int compress_type,
7383 struct extent_map_tree *em_tree;
7384 struct extent_map *em;
7387 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7388 type == BTRFS_ORDERED_COMPRESSED ||
7389 type == BTRFS_ORDERED_NOCOW ||
7390 type == BTRFS_ORDERED_REGULAR);
7392 em_tree = &inode->extent_tree;
7393 em = alloc_extent_map();
7395 return ERR_PTR(-ENOMEM);
7398 em->orig_start = orig_start;
7400 em->block_len = block_len;
7401 em->block_start = block_start;
7402 em->orig_block_len = orig_block_len;
7403 em->ram_bytes = ram_bytes;
7404 em->generation = -1;
7405 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7406 if (type == BTRFS_ORDERED_PREALLOC) {
7407 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7408 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7409 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7410 em->compress_type = compress_type;
7414 btrfs_drop_extent_cache(inode, em->start,
7415 em->start + em->len - 1, 0);
7416 write_lock(&em_tree->lock);
7417 ret = add_extent_mapping(em_tree, em, 1);
7418 write_unlock(&em_tree->lock);
7420 * The caller has taken lock_extent(), who could race with us
7423 } while (ret == -EEXIST);
7426 free_extent_map(em);
7427 return ERR_PTR(ret);
7430 /* em got 2 refs now, callers needs to do free_extent_map once. */
7435 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7436 struct inode *inode,
7437 struct btrfs_dio_data *dio_data,
7440 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7441 struct extent_map *em = *map;
7443 u64 block_start, orig_start, orig_block_len, ram_bytes;
7444 bool can_nocow = false;
7445 bool space_reserved = false;
7450 * We don't allocate a new extent in the following cases
7452 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7454 * 2) The extent is marked as PREALLOC. We're good to go here and can
7455 * just use the extent.
7458 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7459 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7460 em->block_start != EXTENT_MAP_HOLE)) {
7461 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7462 type = BTRFS_ORDERED_PREALLOC;
7464 type = BTRFS_ORDERED_NOCOW;
7465 len = min(len, em->len - (start - em->start));
7466 block_start = em->block_start + (start - em->start);
7468 if (can_nocow_extent(inode, start, &len, &orig_start,
7469 &orig_block_len, &ram_bytes, false) == 1 &&
7470 btrfs_inc_nocow_writers(fs_info, block_start))
7476 struct extent_map *em2;
7478 /* We can NOCOW, so only need to reserve metadata space. */
7479 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len);
7481 /* Our caller expects us to free the input extent map. */
7482 free_extent_map(em);
7484 btrfs_dec_nocow_writers(fs_info, block_start);
7487 space_reserved = true;
7489 em2 = btrfs_create_dio_extent(BTRFS_I(inode), start, len,
7490 orig_start, block_start,
7491 len, orig_block_len,
7493 btrfs_dec_nocow_writers(fs_info, block_start);
7494 if (type == BTRFS_ORDERED_PREALLOC) {
7495 free_extent_map(em);
7504 /* Our caller expects us to free the input extent map. */
7505 free_extent_map(em);
7508 /* We have to COW, so need to reserve metadata and data space. */
7509 ret = btrfs_delalloc_reserve_space(BTRFS_I(inode),
7510 &dio_data->data_reserved,
7514 space_reserved = true;
7516 em = btrfs_new_extent_direct(BTRFS_I(inode), start, len);
7522 len = min(len, em->len - (start - em->start));
7524 btrfs_delalloc_release_space(BTRFS_I(inode),
7525 dio_data->data_reserved,
7526 start + len, prev_len - len,
7531 * We have created our ordered extent, so we can now release our reservation
7532 * for an outstanding extent.
7534 btrfs_delalloc_release_extents(BTRFS_I(inode), prev_len);
7537 * Need to update the i_size under the extent lock so buffered
7538 * readers will get the updated i_size when we unlock.
7540 if (start + len > i_size_read(inode))
7541 i_size_write(inode, start + len);
7543 if (ret && space_reserved) {
7544 btrfs_delalloc_release_extents(BTRFS_I(inode), len);
7546 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, true);
7548 btrfs_delalloc_release_space(BTRFS_I(inode),
7549 dio_data->data_reserved,
7551 extent_changeset_free(dio_data->data_reserved);
7552 dio_data->data_reserved = NULL;
7558 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7559 loff_t length, unsigned int flags, struct iomap *iomap,
7560 struct iomap *srcmap)
7562 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7563 struct extent_map *em;
7564 struct extent_state *cached_state = NULL;
7565 struct btrfs_dio_data *dio_data = NULL;
7566 u64 lockstart, lockend;
7567 const bool write = !!(flags & IOMAP_WRITE);
7570 bool unlock_extents = false;
7573 len = min_t(u64, len, fs_info->sectorsize);
7576 lockend = start + len - 1;
7579 * The generic stuff only does filemap_write_and_wait_range, which
7580 * isn't enough if we've written compressed pages to this area, so we
7581 * need to flush the dirty pages again to make absolutely sure that any
7582 * outstanding dirty pages are on disk.
7584 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7585 &BTRFS_I(inode)->runtime_flags)) {
7586 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7587 start + length - 1);
7592 dio_data = kzalloc(sizeof(*dio_data), GFP_NOFS);
7596 iomap->private = dio_data;
7600 * If this errors out it's because we couldn't invalidate pagecache for
7601 * this range and we need to fallback to buffered.
7603 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, write)) {
7608 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7615 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7616 * io. INLINE is special, and we could probably kludge it in here, but
7617 * it's still buffered so for safety lets just fall back to the generic
7620 * For COMPRESSED we _have_ to read the entire extent in so we can
7621 * decompress it, so there will be buffering required no matter what we
7622 * do, so go ahead and fallback to buffered.
7624 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7625 * to buffered IO. Don't blame me, this is the price we pay for using
7628 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7629 em->block_start == EXTENT_MAP_INLINE) {
7630 free_extent_map(em);
7635 len = min(len, em->len - (start - em->start));
7638 * If we have a NOWAIT request and the range contains multiple extents
7639 * (or a mix of extents and holes), then we return -EAGAIN to make the
7640 * caller fallback to a context where it can do a blocking (without
7641 * NOWAIT) request. This way we avoid doing partial IO and returning
7642 * success to the caller, which is not optimal for writes and for reads
7643 * it can result in unexpected behaviour for an application.
7645 * When doing a read, because we use IOMAP_DIO_PARTIAL when calling
7646 * iomap_dio_rw(), we can end up returning less data then what the caller
7647 * asked for, resulting in an unexpected, and incorrect, short read.
7648 * That is, the caller asked to read N bytes and we return less than that,
7649 * which is wrong unless we are crossing EOF. This happens if we get a
7650 * page fault error when trying to fault in pages for the buffer that is
7651 * associated to the struct iov_iter passed to iomap_dio_rw(), and we
7652 * have previously submitted bios for other extents in the range, in
7653 * which case iomap_dio_rw() may return us EIOCBQUEUED if not all of
7654 * those bios have completed by the time we get the page fault error,
7655 * which we return back to our caller - we should only return EIOCBQUEUED
7656 * after we have submitted bios for all the extents in the range.
7658 if ((flags & IOMAP_NOWAIT) && len < length) {
7659 free_extent_map(em);
7665 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7669 unlock_extents = true;
7670 /* Recalc len in case the new em is smaller than requested */
7671 len = min(len, em->len - (start - em->start));
7674 * We need to unlock only the end area that we aren't using.
7675 * The rest is going to be unlocked by the endio routine.
7677 lockstart = start + len;
7678 if (lockstart < lockend)
7679 unlock_extents = true;
7683 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
7684 lockstart, lockend, &cached_state);
7686 free_extent_state(cached_state);
7689 * Translate extent map information to iomap.
7690 * We trim the extents (and move the addr) even though iomap code does
7691 * that, since we have locked only the parts we are performing I/O in.
7693 if ((em->block_start == EXTENT_MAP_HOLE) ||
7694 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7695 iomap->addr = IOMAP_NULL_ADDR;
7696 iomap->type = IOMAP_HOLE;
7698 iomap->addr = em->block_start + (start - em->start);
7699 iomap->type = IOMAP_MAPPED;
7701 iomap->offset = start;
7702 iomap->bdev = fs_info->fs_devices->latest_dev->bdev;
7703 iomap->length = len;
7705 if (write && btrfs_use_zone_append(BTRFS_I(inode), em->block_start))
7706 iomap->flags |= IOMAP_F_ZONE_APPEND;
7708 free_extent_map(em);
7713 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7721 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7722 ssize_t written, unsigned int flags, struct iomap *iomap)
7725 struct btrfs_dio_data *dio_data = iomap->private;
7726 size_t submitted = dio_data->submitted;
7727 const bool write = !!(flags & IOMAP_WRITE);
7729 if (!write && (iomap->type == IOMAP_HOLE)) {
7730 /* If reading from a hole, unlock and return */
7731 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1);
7735 if (submitted < length) {
7737 length -= submitted;
7739 __endio_write_update_ordered(BTRFS_I(inode), pos,
7742 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7748 extent_changeset_free(dio_data->data_reserved);
7751 iomap->private = NULL;
7756 static void btrfs_dio_private_put(struct btrfs_dio_private *dip)
7759 * This implies a barrier so that stores to dio_bio->bi_status before
7760 * this and loads of dio_bio->bi_status after this are fully ordered.
7762 if (!refcount_dec_and_test(&dip->refs))
7765 if (btrfs_op(dip->dio_bio) == BTRFS_MAP_WRITE) {
7766 __endio_write_update_ordered(BTRFS_I(dip->inode),
7769 !dip->dio_bio->bi_status);
7771 unlock_extent(&BTRFS_I(dip->inode)->io_tree,
7773 dip->file_offset + dip->bytes - 1);
7776 bio_endio(dip->dio_bio);
7780 static blk_status_t submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7782 unsigned long bio_flags)
7784 struct btrfs_dio_private *dip = bio->bi_private;
7785 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7788 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7790 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
7794 refcount_inc(&dip->refs);
7795 ret = btrfs_map_bio(fs_info, bio, mirror_num);
7797 refcount_dec(&dip->refs);
7801 static blk_status_t btrfs_check_read_dio_bio(struct btrfs_dio_private *dip,
7802 struct btrfs_bio *bbio,
7803 const bool uptodate)
7805 struct inode *inode = dip->inode;
7806 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
7807 const u32 sectorsize = fs_info->sectorsize;
7808 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7809 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7810 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7811 struct bio_vec bvec;
7812 struct bvec_iter iter;
7814 blk_status_t err = BLK_STS_OK;
7816 __bio_for_each_segment(bvec, &bbio->bio, iter, bbio->iter) {
7817 unsigned int i, nr_sectors, pgoff;
7819 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7820 pgoff = bvec.bv_offset;
7821 for (i = 0; i < nr_sectors; i++) {
7822 u64 start = bbio->file_offset + bio_offset;
7824 ASSERT(pgoff < PAGE_SIZE);
7826 (!csum || !check_data_csum(inode, bbio,
7827 bio_offset, bvec.bv_page,
7829 clean_io_failure(fs_info, failure_tree, io_tree,
7830 start, bvec.bv_page,
7831 btrfs_ino(BTRFS_I(inode)),
7836 ret = btrfs_repair_one_sector(inode, &bbio->bio,
7837 bio_offset, bvec.bv_page, pgoff,
7838 start, bbio->mirror_num,
7839 submit_dio_repair_bio);
7841 err = errno_to_blk_status(ret);
7843 ASSERT(bio_offset + sectorsize > bio_offset);
7844 bio_offset += sectorsize;
7845 pgoff += sectorsize;
7851 static void __endio_write_update_ordered(struct btrfs_inode *inode,
7852 const u64 offset, const u64 bytes,
7853 const bool uptodate)
7855 btrfs_mark_ordered_io_finished(inode, NULL, offset, bytes,
7856 finish_ordered_fn, uptodate);
7859 static blk_status_t btrfs_submit_bio_start_direct_io(struct inode *inode,
7861 u64 dio_file_offset)
7863 return btrfs_csum_one_bio(BTRFS_I(inode), bio, dio_file_offset, false);
7866 static void btrfs_end_dio_bio(struct bio *bio)
7868 struct btrfs_dio_private *dip = bio->bi_private;
7869 struct btrfs_bio *bbio = btrfs_bio(bio);
7870 blk_status_t err = bio->bi_status;
7873 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
7874 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
7875 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
7876 bio->bi_opf, bio->bi_iter.bi_sector,
7877 bio->bi_iter.bi_size, err);
7879 if (bio_op(bio) == REQ_OP_READ)
7880 err = btrfs_check_read_dio_bio(dip, bbio, !err);
7883 dip->dio_bio->bi_status = err;
7885 btrfs_record_physical_zoned(dip->inode, bbio->file_offset, bio);
7888 btrfs_dio_private_put(dip);
7891 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
7892 struct inode *inode, u64 file_offset, int async_submit)
7894 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7895 struct btrfs_dio_private *dip = bio->bi_private;
7896 bool write = btrfs_op(bio) == BTRFS_MAP_WRITE;
7899 /* Check btrfs_submit_bio_hook() for rules about async submit. */
7901 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7904 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
7909 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
7912 if (write && async_submit) {
7913 ret = btrfs_wq_submit_bio(inode, bio, 0, 0, file_offset,
7914 btrfs_submit_bio_start_direct_io);
7918 * If we aren't doing async submit, calculate the csum of the
7921 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, file_offset, false);
7927 csum_offset = file_offset - dip->file_offset;
7928 csum_offset >>= fs_info->sectorsize_bits;
7929 csum_offset *= fs_info->csum_size;
7930 btrfs_bio(bio)->csum = dip->csums + csum_offset;
7933 ret = btrfs_map_bio(fs_info, bio, 0);
7939 * If this succeeds, the btrfs_dio_private is responsible for cleaning up locked
7940 * or ordered extents whether or not we submit any bios.
7942 static struct btrfs_dio_private *btrfs_create_dio_private(struct bio *dio_bio,
7943 struct inode *inode,
7946 const bool write = (btrfs_op(dio_bio) == BTRFS_MAP_WRITE);
7947 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7949 struct btrfs_dio_private *dip;
7951 dip_size = sizeof(*dip);
7952 if (!write && csum) {
7953 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7956 nblocks = dio_bio->bi_iter.bi_size >> fs_info->sectorsize_bits;
7957 dip_size += fs_info->csum_size * nblocks;
7960 dip = kzalloc(dip_size, GFP_NOFS);
7965 dip->file_offset = file_offset;
7966 dip->bytes = dio_bio->bi_iter.bi_size;
7967 dip->disk_bytenr = dio_bio->bi_iter.bi_sector << 9;
7968 dip->dio_bio = dio_bio;
7969 refcount_set(&dip->refs, 1);
7973 static void btrfs_submit_direct(const struct iomap_iter *iter,
7974 struct bio *dio_bio, loff_t file_offset)
7976 struct inode *inode = iter->inode;
7977 const bool write = (btrfs_op(dio_bio) == BTRFS_MAP_WRITE);
7978 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7979 const bool raid56 = (btrfs_data_alloc_profile(fs_info) &
7980 BTRFS_BLOCK_GROUP_RAID56_MASK);
7981 struct btrfs_dio_private *dip;
7984 int async_submit = 0;
7986 u64 clone_offset = 0;
7990 blk_status_t status;
7991 struct btrfs_io_geometry geom;
7992 struct btrfs_dio_data *dio_data = iter->iomap.private;
7993 struct extent_map *em = NULL;
7995 dip = btrfs_create_dio_private(dio_bio, inode, file_offset);
7998 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
7999 file_offset + dio_bio->bi_iter.bi_size - 1);
8001 dio_bio->bi_status = BLK_STS_RESOURCE;
8008 * Load the csums up front to reduce csum tree searches and
8009 * contention when submitting bios.
8011 * If we have csums disabled this will do nothing.
8013 status = btrfs_lookup_bio_sums(inode, dio_bio, dip->csums);
8014 if (status != BLK_STS_OK)
8018 start_sector = dio_bio->bi_iter.bi_sector;
8019 submit_len = dio_bio->bi_iter.bi_size;
8022 logical = start_sector << 9;
8023 em = btrfs_get_chunk_map(fs_info, logical, submit_len);
8025 status = errno_to_blk_status(PTR_ERR(em));
8029 ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(dio_bio),
8032 status = errno_to_blk_status(ret);
8036 clone_len = min(submit_len, geom.len);
8037 ASSERT(clone_len <= UINT_MAX);
8040 * This will never fail as it's passing GPF_NOFS and
8041 * the allocation is backed by btrfs_bioset.
8043 bio = btrfs_bio_clone_partial(dio_bio, clone_offset, clone_len);
8044 bio->bi_private = dip;
8045 bio->bi_end_io = btrfs_end_dio_bio;
8046 btrfs_bio(bio)->file_offset = file_offset;
8048 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
8049 status = extract_ordered_extent(BTRFS_I(inode), bio,
8057 ASSERT(submit_len >= clone_len);
8058 submit_len -= clone_len;
8061 * Increase the count before we submit the bio so we know
8062 * the end IO handler won't happen before we increase the
8063 * count. Otherwise, the dip might get freed before we're
8064 * done setting it up.
8066 * We transfer the initial reference to the last bio, so we
8067 * don't need to increment the reference count for the last one.
8069 if (submit_len > 0) {
8070 refcount_inc(&dip->refs);
8072 * If we are submitting more than one bio, submit them
8073 * all asynchronously. The exception is RAID 5 or 6, as
8074 * asynchronous checksums make it difficult to collect
8075 * full stripe writes.
8081 status = btrfs_submit_dio_bio(bio, inode, file_offset,
8086 refcount_dec(&dip->refs);
8090 dio_data->submitted += clone_len;
8091 clone_offset += clone_len;
8092 start_sector += clone_len >> 9;
8093 file_offset += clone_len;
8095 free_extent_map(em);
8096 } while (submit_len > 0);
8100 free_extent_map(em);
8102 dip->dio_bio->bi_status = status;
8103 btrfs_dio_private_put(dip);
8106 const struct iomap_ops btrfs_dio_iomap_ops = {
8107 .iomap_begin = btrfs_dio_iomap_begin,
8108 .iomap_end = btrfs_dio_iomap_end,
8111 const struct iomap_dio_ops btrfs_dio_ops = {
8112 .submit_io = btrfs_submit_direct,
8115 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8120 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
8124 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
8127 int btrfs_readpage(struct file *file, struct page *page)
8129 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
8130 u64 start = page_offset(page);
8131 u64 end = start + PAGE_SIZE - 1;
8132 struct btrfs_bio_ctrl bio_ctrl = { 0 };
8135 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
8137 ret = btrfs_do_readpage(page, NULL, &bio_ctrl, 0, NULL);
8141 ret2 = submit_one_bio(bio_ctrl.bio, 0, bio_ctrl.bio_flags);
8148 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8150 struct inode *inode = page->mapping->host;
8153 if (current->flags & PF_MEMALLOC) {
8154 redirty_page_for_writepage(wbc, page);
8160 * If we are under memory pressure we will call this directly from the
8161 * VM, we need to make sure we have the inode referenced for the ordered
8162 * extent. If not just return like we didn't do anything.
8164 if (!igrab(inode)) {
8165 redirty_page_for_writepage(wbc, page);
8166 return AOP_WRITEPAGE_ACTIVATE;
8168 ret = extent_write_full_page(page, wbc);
8169 btrfs_add_delayed_iput(inode);
8173 static int btrfs_writepages(struct address_space *mapping,
8174 struct writeback_control *wbc)
8176 return extent_writepages(mapping, wbc);
8179 static void btrfs_readahead(struct readahead_control *rac)
8181 extent_readahead(rac);
8185 * For releasepage() and invalidate_folio() we have a race window where
8186 * folio_end_writeback() is called but the subpage spinlock is not yet released.
8187 * If we continue to release/invalidate the page, we could cause use-after-free
8188 * for subpage spinlock. So this function is to spin and wait for subpage
8191 static void wait_subpage_spinlock(struct page *page)
8193 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
8194 struct btrfs_subpage *subpage;
8196 if (fs_info->sectorsize == PAGE_SIZE)
8199 ASSERT(PagePrivate(page) && page->private);
8200 subpage = (struct btrfs_subpage *)page->private;
8203 * This may look insane as we just acquire the spinlock and release it,
8204 * without doing anything. But we just want to make sure no one is
8205 * still holding the subpage spinlock.
8206 * And since the page is not dirty nor writeback, and we have page
8207 * locked, the only possible way to hold a spinlock is from the endio
8208 * function to clear page writeback.
8210 * Here we just acquire the spinlock so that all existing callers
8211 * should exit and we're safe to release/invalidate the page.
8213 spin_lock_irq(&subpage->lock);
8214 spin_unlock_irq(&subpage->lock);
8217 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8219 int ret = try_release_extent_mapping(page, gfp_flags);
8222 wait_subpage_spinlock(page);
8223 clear_page_extent_mapped(page);
8228 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8230 if (PageWriteback(page) || PageDirty(page))
8232 return __btrfs_releasepage(page, gfp_flags);
8235 #ifdef CONFIG_MIGRATION
8236 static int btrfs_migratepage(struct address_space *mapping,
8237 struct page *newpage, struct page *page,
8238 enum migrate_mode mode)
8242 ret = migrate_page_move_mapping(mapping, newpage, page, 0);
8243 if (ret != MIGRATEPAGE_SUCCESS)
8246 if (page_has_private(page))
8247 attach_page_private(newpage, detach_page_private(page));
8249 if (PageOrdered(page)) {
8250 ClearPageOrdered(page);
8251 SetPageOrdered(newpage);
8254 if (mode != MIGRATE_SYNC_NO_COPY)
8255 migrate_page_copy(newpage, page);
8257 migrate_page_states(newpage, page);
8258 return MIGRATEPAGE_SUCCESS;
8262 static void btrfs_invalidate_folio(struct folio *folio, size_t offset,
8265 struct btrfs_inode *inode = BTRFS_I(folio->mapping->host);
8266 struct btrfs_fs_info *fs_info = inode->root->fs_info;
8267 struct extent_io_tree *tree = &inode->io_tree;
8268 struct extent_state *cached_state = NULL;
8269 u64 page_start = folio_pos(folio);
8270 u64 page_end = page_start + folio_size(folio) - 1;
8272 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
8275 * We have folio locked so no new ordered extent can be created on this
8276 * page, nor bio can be submitted for this folio.
8278 * But already submitted bio can still be finished on this folio.
8279 * Furthermore, endio function won't skip folio which has Ordered
8280 * (Private2) already cleared, so it's possible for endio and
8281 * invalidate_folio to do the same ordered extent accounting twice
8284 * So here we wait for any submitted bios to finish, so that we won't
8285 * do double ordered extent accounting on the same folio.
8287 folio_wait_writeback(folio);
8288 wait_subpage_spinlock(&folio->page);
8291 * For subpage case, we have call sites like
8292 * btrfs_punch_hole_lock_range() which passes range not aligned to
8294 * If the range doesn't cover the full folio, we don't need to and
8295 * shouldn't clear page extent mapped, as folio->private can still
8296 * record subpage dirty bits for other part of the range.
8298 * For cases that invalidate the full folio even the range doesn't
8299 * cover the full folio, like invalidating the last folio, we're
8300 * still safe to wait for ordered extent to finish.
8302 if (!(offset == 0 && length == folio_size(folio))) {
8303 btrfs_releasepage(&folio->page, GFP_NOFS);
8307 if (!inode_evicting)
8308 lock_extent_bits(tree, page_start, page_end, &cached_state);
8311 while (cur < page_end) {
8312 struct btrfs_ordered_extent *ordered;
8317 ordered = btrfs_lookup_first_ordered_range(inode, cur,
8318 page_end + 1 - cur);
8320 range_end = page_end;
8322 * No ordered extent covering this range, we are safe
8323 * to delete all extent states in the range.
8325 delete_states = true;
8328 if (ordered->file_offset > cur) {
8330 * There is a range between [cur, oe->file_offset) not
8331 * covered by any ordered extent.
8332 * We are safe to delete all extent states, and handle
8333 * the ordered extent in the next iteration.
8335 range_end = ordered->file_offset - 1;
8336 delete_states = true;
8340 range_end = min(ordered->file_offset + ordered->num_bytes - 1,
8342 ASSERT(range_end + 1 - cur < U32_MAX);
8343 range_len = range_end + 1 - cur;
8344 if (!btrfs_page_test_ordered(fs_info, &folio->page, cur, range_len)) {
8346 * If Ordered (Private2) is cleared, it means endio has
8347 * already been executed for the range.
8348 * We can't delete the extent states as
8349 * btrfs_finish_ordered_io() may still use some of them.
8351 delete_states = false;
8354 btrfs_page_clear_ordered(fs_info, &folio->page, cur, range_len);
8357 * IO on this page will never be started, so we need to account
8358 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8359 * here, must leave that up for the ordered extent completion.
8361 * This will also unlock the range for incoming
8362 * btrfs_finish_ordered_io().
8364 if (!inode_evicting)
8365 clear_extent_bit(tree, cur, range_end,
8367 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8368 EXTENT_DEFRAG, 1, 0, &cached_state);
8370 spin_lock_irq(&inode->ordered_tree.lock);
8371 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8372 ordered->truncated_len = min(ordered->truncated_len,
8373 cur - ordered->file_offset);
8374 spin_unlock_irq(&inode->ordered_tree.lock);
8376 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8377 cur, range_end + 1 - cur)) {
8378 btrfs_finish_ordered_io(ordered);
8380 * The ordered extent has finished, now we're again
8381 * safe to delete all extent states of the range.
8383 delete_states = true;
8386 * btrfs_finish_ordered_io() will get executed by endio
8387 * of other pages, thus we can't delete extent states
8390 delete_states = false;
8394 btrfs_put_ordered_extent(ordered);
8396 * Qgroup reserved space handler
8397 * Sector(s) here will be either:
8399 * 1) Already written to disk or bio already finished
8400 * Then its QGROUP_RESERVED bit in io_tree is already cleared.
8401 * Qgroup will be handled by its qgroup_record then.
8402 * btrfs_qgroup_free_data() call will do nothing here.
8404 * 2) Not written to disk yet
8405 * Then btrfs_qgroup_free_data() call will clear the
8406 * QGROUP_RESERVED bit of its io_tree, and free the qgroup
8407 * reserved data space.
8408 * Since the IO will never happen for this page.
8410 btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur);
8411 if (!inode_evicting) {
8412 clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED |
8413 EXTENT_DELALLOC | EXTENT_UPTODATE |
8414 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1,
8415 delete_states, &cached_state);
8417 cur = range_end + 1;
8420 * We have iterated through all ordered extents of the page, the page
8421 * should not have Ordered (Private2) anymore, or the above iteration
8422 * did something wrong.
8424 ASSERT(!folio_test_ordered(folio));
8425 btrfs_page_clear_checked(fs_info, &folio->page, folio_pos(folio), folio_size(folio));
8426 if (!inode_evicting)
8427 __btrfs_releasepage(&folio->page, GFP_NOFS);
8428 clear_page_extent_mapped(&folio->page);
8432 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8433 * called from a page fault handler when a page is first dirtied. Hence we must
8434 * be careful to check for EOF conditions here. We set the page up correctly
8435 * for a written page which means we get ENOSPC checking when writing into
8436 * holes and correct delalloc and unwritten extent mapping on filesystems that
8437 * support these features.
8439 * We are not allowed to take the i_mutex here so we have to play games to
8440 * protect against truncate races as the page could now be beyond EOF. Because
8441 * truncate_setsize() writes the inode size before removing pages, once we have
8442 * the page lock we can determine safely if the page is beyond EOF. If it is not
8443 * beyond EOF, then the page is guaranteed safe against truncation until we
8446 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8448 struct page *page = vmf->page;
8449 struct inode *inode = file_inode(vmf->vma->vm_file);
8450 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8451 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8452 struct btrfs_ordered_extent *ordered;
8453 struct extent_state *cached_state = NULL;
8454 struct extent_changeset *data_reserved = NULL;
8455 unsigned long zero_start;
8465 reserved_space = PAGE_SIZE;
8467 sb_start_pagefault(inode->i_sb);
8468 page_start = page_offset(page);
8469 page_end = page_start + PAGE_SIZE - 1;
8473 * Reserving delalloc space after obtaining the page lock can lead to
8474 * deadlock. For example, if a dirty page is locked by this function
8475 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8476 * dirty page write out, then the btrfs_writepage() function could
8477 * end up waiting indefinitely to get a lock on the page currently
8478 * being processed by btrfs_page_mkwrite() function.
8480 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8481 page_start, reserved_space);
8483 ret2 = file_update_time(vmf->vma->vm_file);
8487 ret = vmf_error(ret2);
8493 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8495 down_read(&BTRFS_I(inode)->i_mmap_lock);
8497 size = i_size_read(inode);
8499 if ((page->mapping != inode->i_mapping) ||
8500 (page_start >= size)) {
8501 /* page got truncated out from underneath us */
8504 wait_on_page_writeback(page);
8506 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8507 ret2 = set_page_extent_mapped(page);
8509 ret = vmf_error(ret2);
8510 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8515 * we can't set the delalloc bits if there are pending ordered
8516 * extents. Drop our locks and wait for them to finish
8518 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8521 unlock_extent_cached(io_tree, page_start, page_end,
8524 up_read(&BTRFS_I(inode)->i_mmap_lock);
8525 btrfs_start_ordered_extent(ordered, 1);
8526 btrfs_put_ordered_extent(ordered);
8530 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8531 reserved_space = round_up(size - page_start,
8532 fs_info->sectorsize);
8533 if (reserved_space < PAGE_SIZE) {
8534 end = page_start + reserved_space - 1;
8535 btrfs_delalloc_release_space(BTRFS_I(inode),
8536 data_reserved, page_start,
8537 PAGE_SIZE - reserved_space, true);
8542 * page_mkwrite gets called when the page is firstly dirtied after it's
8543 * faulted in, but write(2) could also dirty a page and set delalloc
8544 * bits, thus in this case for space account reason, we still need to
8545 * clear any delalloc bits within this page range since we have to
8546 * reserve data&meta space before lock_page() (see above comments).
8548 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8549 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8550 EXTENT_DEFRAG, 0, 0, &cached_state);
8552 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8555 unlock_extent_cached(io_tree, page_start, page_end,
8557 ret = VM_FAULT_SIGBUS;
8561 /* page is wholly or partially inside EOF */
8562 if (page_start + PAGE_SIZE > size)
8563 zero_start = offset_in_page(size);
8565 zero_start = PAGE_SIZE;
8567 if (zero_start != PAGE_SIZE) {
8568 memzero_page(page, zero_start, PAGE_SIZE - zero_start);
8569 flush_dcache_page(page);
8571 btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE);
8572 btrfs_page_set_dirty(fs_info, page, page_start, end + 1 - page_start);
8573 btrfs_page_set_uptodate(fs_info, page, page_start, end + 1 - page_start);
8575 btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
8577 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8578 up_read(&BTRFS_I(inode)->i_mmap_lock);
8580 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8581 sb_end_pagefault(inode->i_sb);
8582 extent_changeset_free(data_reserved);
8583 return VM_FAULT_LOCKED;
8587 up_read(&BTRFS_I(inode)->i_mmap_lock);
8589 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8590 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8591 reserved_space, (ret != 0));
8593 sb_end_pagefault(inode->i_sb);
8594 extent_changeset_free(data_reserved);
8598 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
8600 struct btrfs_truncate_control control = {
8601 .inode = BTRFS_I(inode),
8602 .ino = btrfs_ino(BTRFS_I(inode)),
8603 .min_type = BTRFS_EXTENT_DATA_KEY,
8604 .clear_extent_range = true,
8606 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8607 struct btrfs_root *root = BTRFS_I(inode)->root;
8608 struct btrfs_block_rsv *rsv;
8610 struct btrfs_trans_handle *trans;
8611 u64 mask = fs_info->sectorsize - 1;
8612 u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8614 if (!skip_writeback) {
8615 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8622 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8623 * things going on here:
8625 * 1) We need to reserve space to update our inode.
8627 * 2) We need to have something to cache all the space that is going to
8628 * be free'd up by the truncate operation, but also have some slack
8629 * space reserved in case it uses space during the truncate (thank you
8630 * very much snapshotting).
8632 * And we need these to be separate. The fact is we can use a lot of
8633 * space doing the truncate, and we have no earthly idea how much space
8634 * we will use, so we need the truncate reservation to be separate so it
8635 * doesn't end up using space reserved for updating the inode. We also
8636 * need to be able to stop the transaction and start a new one, which
8637 * means we need to be able to update the inode several times, and we
8638 * have no idea of knowing how many times that will be, so we can't just
8639 * reserve 1 item for the entirety of the operation, so that has to be
8640 * done separately as well.
8642 * So that leaves us with
8644 * 1) rsv - for the truncate reservation, which we will steal from the
8645 * transaction reservation.
8646 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8647 * updating the inode.
8649 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8652 rsv->size = min_size;
8656 * 1 for the truncate slack space
8657 * 1 for updating the inode.
8659 trans = btrfs_start_transaction(root, 2);
8660 if (IS_ERR(trans)) {
8661 ret = PTR_ERR(trans);
8665 /* Migrate the slack space for the truncate to our reserve */
8666 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8670 trans->block_rsv = rsv;
8673 struct extent_state *cached_state = NULL;
8674 const u64 new_size = inode->i_size;
8675 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
8677 control.new_size = new_size;
8678 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, (u64)-1,
8681 * We want to drop from the next block forward in case this new
8682 * size is not block aligned since we will be keeping the last
8683 * block of the extent just the way it is.
8685 btrfs_drop_extent_cache(BTRFS_I(inode),
8686 ALIGN(new_size, fs_info->sectorsize),
8689 ret = btrfs_truncate_inode_items(trans, root, &control);
8691 inode_sub_bytes(inode, control.sub_bytes);
8692 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), control.last_size);
8694 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start,
8695 (u64)-1, &cached_state);
8697 trans->block_rsv = &fs_info->trans_block_rsv;
8698 if (ret != -ENOSPC && ret != -EAGAIN)
8701 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
8705 btrfs_end_transaction(trans);
8706 btrfs_btree_balance_dirty(fs_info);
8708 trans = btrfs_start_transaction(root, 2);
8709 if (IS_ERR(trans)) {
8710 ret = PTR_ERR(trans);
8715 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8716 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8717 rsv, min_size, false);
8718 BUG_ON(ret); /* shouldn't happen */
8719 trans->block_rsv = rsv;
8723 * We can't call btrfs_truncate_block inside a trans handle as we could
8724 * deadlock with freeze, if we got BTRFS_NEED_TRUNCATE_BLOCK then we
8725 * know we've truncated everything except the last little bit, and can
8726 * do btrfs_truncate_block and then update the disk_i_size.
8728 if (ret == BTRFS_NEED_TRUNCATE_BLOCK) {
8729 btrfs_end_transaction(trans);
8730 btrfs_btree_balance_dirty(fs_info);
8732 ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0);
8735 trans = btrfs_start_transaction(root, 1);
8736 if (IS_ERR(trans)) {
8737 ret = PTR_ERR(trans);
8740 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
8746 trans->block_rsv = &fs_info->trans_block_rsv;
8747 ret2 = btrfs_update_inode(trans, root, BTRFS_I(inode));
8751 ret2 = btrfs_end_transaction(trans);
8754 btrfs_btree_balance_dirty(fs_info);
8757 btrfs_free_block_rsv(fs_info, rsv);
8759 * So if we truncate and then write and fsync we normally would just
8760 * write the extents that changed, which is a problem if we need to
8761 * first truncate that entire inode. So set this flag so we write out
8762 * all of the extents in the inode to the sync log so we're completely
8765 * If no extents were dropped or trimmed we don't need to force the next
8766 * fsync to truncate all the inode's items from the log and re-log them
8767 * all. This means the truncate operation did not change the file size,
8768 * or changed it to a smaller size but there was only an implicit hole
8769 * between the old i_size and the new i_size, and there were no prealloc
8770 * extents beyond i_size to drop.
8772 if (control.extents_found > 0)
8773 btrfs_set_inode_full_sync(BTRFS_I(inode));
8779 * create a new subvolume directory/inode (helper for the ioctl).
8781 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8782 struct btrfs_root *new_root,
8783 struct btrfs_root *parent_root,
8784 struct user_namespace *mnt_userns)
8786 struct inode *inode;
8791 err = btrfs_get_free_objectid(new_root, &ino);
8795 inode = btrfs_new_inode(trans, new_root, mnt_userns, NULL, "..", 2,
8797 S_IFDIR | (~current_umask() & S_IRWXUGO),
8800 return PTR_ERR(inode);
8801 inode->i_op = &btrfs_dir_inode_operations;
8802 inode->i_fop = &btrfs_dir_file_operations;
8804 set_nlink(inode, 1);
8805 btrfs_i_size_write(BTRFS_I(inode), 0);
8806 unlock_new_inode(inode);
8808 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8810 btrfs_err(new_root->fs_info,
8811 "error inheriting subvolume %llu properties: %d",
8812 new_root->root_key.objectid, err);
8814 err = btrfs_update_inode(trans, new_root, BTRFS_I(inode));
8820 struct inode *btrfs_alloc_inode(struct super_block *sb)
8822 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8823 struct btrfs_inode *ei;
8824 struct inode *inode;
8826 ei = alloc_inode_sb(sb, btrfs_inode_cachep, GFP_KERNEL);
8833 ei->last_sub_trans = 0;
8834 ei->logged_trans = 0;
8835 ei->delalloc_bytes = 0;
8836 ei->new_delalloc_bytes = 0;
8837 ei->defrag_bytes = 0;
8838 ei->disk_i_size = 0;
8842 ei->index_cnt = (u64)-1;
8844 ei->last_unlink_trans = 0;
8845 ei->last_reflink_trans = 0;
8846 ei->last_log_commit = 0;
8848 spin_lock_init(&ei->lock);
8849 ei->outstanding_extents = 0;
8850 if (sb->s_magic != BTRFS_TEST_MAGIC)
8851 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8852 BTRFS_BLOCK_RSV_DELALLOC);
8853 ei->runtime_flags = 0;
8854 ei->prop_compress = BTRFS_COMPRESS_NONE;
8855 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8857 ei->delayed_node = NULL;
8859 ei->i_otime.tv_sec = 0;
8860 ei->i_otime.tv_nsec = 0;
8862 inode = &ei->vfs_inode;
8863 extent_map_tree_init(&ei->extent_tree);
8864 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO, inode);
8865 extent_io_tree_init(fs_info, &ei->io_failure_tree,
8866 IO_TREE_INODE_IO_FAILURE, inode);
8867 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8868 IO_TREE_INODE_FILE_EXTENT, inode);
8869 ei->io_tree.track_uptodate = true;
8870 ei->io_failure_tree.track_uptodate = true;
8871 atomic_set(&ei->sync_writers, 0);
8872 mutex_init(&ei->log_mutex);
8873 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8874 INIT_LIST_HEAD(&ei->delalloc_inodes);
8875 INIT_LIST_HEAD(&ei->delayed_iput);
8876 RB_CLEAR_NODE(&ei->rb_node);
8877 init_rwsem(&ei->i_mmap_lock);
8882 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8883 void btrfs_test_destroy_inode(struct inode *inode)
8885 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
8886 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8890 void btrfs_free_inode(struct inode *inode)
8892 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8895 void btrfs_destroy_inode(struct inode *vfs_inode)
8897 struct btrfs_ordered_extent *ordered;
8898 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8899 struct btrfs_root *root = inode->root;
8901 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8902 WARN_ON(vfs_inode->i_data.nrpages);
8903 WARN_ON(inode->block_rsv.reserved);
8904 WARN_ON(inode->block_rsv.size);
8905 WARN_ON(inode->outstanding_extents);
8906 if (!S_ISDIR(vfs_inode->i_mode)) {
8907 WARN_ON(inode->delalloc_bytes);
8908 WARN_ON(inode->new_delalloc_bytes);
8910 WARN_ON(inode->csum_bytes);
8911 WARN_ON(inode->defrag_bytes);
8914 * This can happen where we create an inode, but somebody else also
8915 * created the same inode and we need to destroy the one we already
8922 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8926 btrfs_err(root->fs_info,
8927 "found ordered extent %llu %llu on inode cleanup",
8928 ordered->file_offset, ordered->num_bytes);
8929 btrfs_remove_ordered_extent(inode, ordered);
8930 btrfs_put_ordered_extent(ordered);
8931 btrfs_put_ordered_extent(ordered);
8934 btrfs_qgroup_check_reserved_leak(inode);
8935 inode_tree_del(inode);
8936 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8937 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
8938 btrfs_put_root(inode->root);
8941 int btrfs_drop_inode(struct inode *inode)
8943 struct btrfs_root *root = BTRFS_I(inode)->root;
8948 /* the snap/subvol tree is on deleting */
8949 if (btrfs_root_refs(&root->root_item) == 0)
8952 return generic_drop_inode(inode);
8955 static void init_once(void *foo)
8957 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8959 inode_init_once(&ei->vfs_inode);
8962 void __cold btrfs_destroy_cachep(void)
8965 * Make sure all delayed rcu free inodes are flushed before we
8969 kmem_cache_destroy(btrfs_inode_cachep);
8970 kmem_cache_destroy(btrfs_trans_handle_cachep);
8971 kmem_cache_destroy(btrfs_path_cachep);
8972 kmem_cache_destroy(btrfs_free_space_cachep);
8973 kmem_cache_destroy(btrfs_free_space_bitmap_cachep);
8976 int __init btrfs_init_cachep(void)
8978 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8979 sizeof(struct btrfs_inode), 0,
8980 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
8982 if (!btrfs_inode_cachep)
8985 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8986 sizeof(struct btrfs_trans_handle), 0,
8987 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
8988 if (!btrfs_trans_handle_cachep)
8991 btrfs_path_cachep = kmem_cache_create("btrfs_path",
8992 sizeof(struct btrfs_path), 0,
8993 SLAB_MEM_SPREAD, NULL);
8994 if (!btrfs_path_cachep)
8997 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8998 sizeof(struct btrfs_free_space), 0,
8999 SLAB_MEM_SPREAD, NULL);
9000 if (!btrfs_free_space_cachep)
9003 btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap",
9004 PAGE_SIZE, PAGE_SIZE,
9005 SLAB_MEM_SPREAD, NULL);
9006 if (!btrfs_free_space_bitmap_cachep)
9011 btrfs_destroy_cachep();
9015 static int btrfs_getattr(struct user_namespace *mnt_userns,
9016 const struct path *path, struct kstat *stat,
9017 u32 request_mask, unsigned int flags)
9021 struct inode *inode = d_inode(path->dentry);
9022 u32 blocksize = inode->i_sb->s_blocksize;
9023 u32 bi_flags = BTRFS_I(inode)->flags;
9024 u32 bi_ro_flags = BTRFS_I(inode)->ro_flags;
9026 stat->result_mask |= STATX_BTIME;
9027 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9028 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9029 if (bi_flags & BTRFS_INODE_APPEND)
9030 stat->attributes |= STATX_ATTR_APPEND;
9031 if (bi_flags & BTRFS_INODE_COMPRESS)
9032 stat->attributes |= STATX_ATTR_COMPRESSED;
9033 if (bi_flags & BTRFS_INODE_IMMUTABLE)
9034 stat->attributes |= STATX_ATTR_IMMUTABLE;
9035 if (bi_flags & BTRFS_INODE_NODUMP)
9036 stat->attributes |= STATX_ATTR_NODUMP;
9037 if (bi_ro_flags & BTRFS_INODE_RO_VERITY)
9038 stat->attributes |= STATX_ATTR_VERITY;
9040 stat->attributes_mask |= (STATX_ATTR_APPEND |
9041 STATX_ATTR_COMPRESSED |
9042 STATX_ATTR_IMMUTABLE |
9045 generic_fillattr(mnt_userns, inode, stat);
9046 stat->dev = BTRFS_I(inode)->root->anon_dev;
9048 spin_lock(&BTRFS_I(inode)->lock);
9049 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9050 inode_bytes = inode_get_bytes(inode);
9051 spin_unlock(&BTRFS_I(inode)->lock);
9052 stat->blocks = (ALIGN(inode_bytes, blocksize) +
9053 ALIGN(delalloc_bytes, blocksize)) >> 9;
9057 static int btrfs_rename_exchange(struct inode *old_dir,
9058 struct dentry *old_dentry,
9059 struct inode *new_dir,
9060 struct dentry *new_dentry)
9062 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9063 struct btrfs_trans_handle *trans;
9064 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9065 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9066 struct inode *new_inode = new_dentry->d_inode;
9067 struct inode *old_inode = old_dentry->d_inode;
9068 struct timespec64 ctime = current_time(old_inode);
9069 struct btrfs_rename_ctx old_rename_ctx;
9070 struct btrfs_rename_ctx new_rename_ctx;
9071 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9072 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9077 bool need_abort = false;
9080 * For non-subvolumes allow exchange only within one subvolume, in the
9081 * same inode namespace. Two subvolumes (represented as directory) can
9082 * be exchanged as they're a logical link and have a fixed inode number.
9085 (old_ino != BTRFS_FIRST_FREE_OBJECTID ||
9086 new_ino != BTRFS_FIRST_FREE_OBJECTID))
9089 /* close the race window with snapshot create/destroy ioctl */
9090 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
9091 new_ino == BTRFS_FIRST_FREE_OBJECTID)
9092 down_read(&fs_info->subvol_sem);
9095 * We want to reserve the absolute worst case amount of items. So if
9096 * both inodes are subvols and we need to unlink them then that would
9097 * require 4 item modifications, but if they are both normal inodes it
9098 * would require 5 item modifications, so we'll assume their normal
9099 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9100 * should cover the worst case number of items we'll modify.
9102 trans = btrfs_start_transaction(root, 12);
9103 if (IS_ERR(trans)) {
9104 ret = PTR_ERR(trans);
9109 ret = btrfs_record_root_in_trans(trans, dest);
9115 * We need to find a free sequence number both in the source and
9116 * in the destination directory for the exchange.
9118 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9121 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9125 BTRFS_I(old_inode)->dir_index = 0ULL;
9126 BTRFS_I(new_inode)->dir_index = 0ULL;
9128 /* Reference for the source. */
9129 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9130 /* force full log commit if subvolume involved. */
9131 btrfs_set_log_full_commit(trans);
9133 ret = btrfs_insert_inode_ref(trans, dest,
9134 new_dentry->d_name.name,
9135 new_dentry->d_name.len,
9137 btrfs_ino(BTRFS_I(new_dir)),
9144 /* And now for the dest. */
9145 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9146 /* force full log commit if subvolume involved. */
9147 btrfs_set_log_full_commit(trans);
9149 ret = btrfs_insert_inode_ref(trans, root,
9150 old_dentry->d_name.name,
9151 old_dentry->d_name.len,
9153 btrfs_ino(BTRFS_I(old_dir)),
9157 btrfs_abort_transaction(trans, ret);
9162 /* Update inode version and ctime/mtime. */
9163 inode_inc_iversion(old_dir);
9164 inode_inc_iversion(new_dir);
9165 inode_inc_iversion(old_inode);
9166 inode_inc_iversion(new_inode);
9167 old_dir->i_ctime = old_dir->i_mtime = ctime;
9168 new_dir->i_ctime = new_dir->i_mtime = ctime;
9169 old_inode->i_ctime = ctime;
9170 new_inode->i_ctime = ctime;
9172 if (old_dentry->d_parent != new_dentry->d_parent) {
9173 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9174 BTRFS_I(old_inode), 1);
9175 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9176 BTRFS_I(new_inode), 1);
9179 /* src is a subvolume */
9180 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9181 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9182 } else { /* src is an inode */
9183 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9184 BTRFS_I(old_dentry->d_inode),
9185 old_dentry->d_name.name,
9186 old_dentry->d_name.len,
9189 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9192 btrfs_abort_transaction(trans, ret);
9196 /* dest is a subvolume */
9197 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9198 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9199 } else { /* dest is an inode */
9200 ret = __btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9201 BTRFS_I(new_dentry->d_inode),
9202 new_dentry->d_name.name,
9203 new_dentry->d_name.len,
9206 ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode));
9209 btrfs_abort_transaction(trans, ret);
9213 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9214 new_dentry->d_name.name,
9215 new_dentry->d_name.len, 0, old_idx);
9217 btrfs_abort_transaction(trans, ret);
9221 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9222 old_dentry->d_name.name,
9223 old_dentry->d_name.len, 0, new_idx);
9225 btrfs_abort_transaction(trans, ret);
9229 if (old_inode->i_nlink == 1)
9230 BTRFS_I(old_inode)->dir_index = old_idx;
9231 if (new_inode->i_nlink == 1)
9232 BTRFS_I(new_inode)->dir_index = new_idx;
9235 * Now pin the logs of the roots. We do it to ensure that no other task
9236 * can sync the logs while we are in progress with the rename, because
9237 * that could result in an inconsistency in case any of the inodes that
9238 * are part of this rename operation were logged before.
9240 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9241 btrfs_pin_log_trans(root);
9242 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9243 btrfs_pin_log_trans(dest);
9245 /* Do the log updates for all inodes. */
9246 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9247 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9248 old_rename_ctx.index, new_dentry->d_parent);
9249 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9250 btrfs_log_new_name(trans, new_dentry, BTRFS_I(new_dir),
9251 new_rename_ctx.index, old_dentry->d_parent);
9253 /* Now unpin the logs. */
9254 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9255 btrfs_end_log_trans(root);
9256 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9257 btrfs_end_log_trans(dest);
9259 ret2 = btrfs_end_transaction(trans);
9260 ret = ret ? ret : ret2;
9262 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
9263 old_ino == BTRFS_FIRST_FREE_OBJECTID)
9264 up_read(&fs_info->subvol_sem);
9269 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9270 struct btrfs_root *root,
9271 struct user_namespace *mnt_userns,
9273 struct dentry *dentry)
9276 struct inode *inode;
9280 ret = btrfs_get_free_objectid(root, &objectid);
9284 inode = btrfs_new_inode(trans, root, mnt_userns, dir,
9285 dentry->d_name.name,
9287 btrfs_ino(BTRFS_I(dir)),
9289 S_IFCHR | WHITEOUT_MODE,
9292 if (IS_ERR(inode)) {
9293 ret = PTR_ERR(inode);
9297 inode->i_op = &btrfs_special_inode_operations;
9298 init_special_inode(inode, inode->i_mode,
9301 ret = btrfs_init_inode_security(trans, inode, dir,
9306 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9307 BTRFS_I(inode), 0, index);
9311 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
9313 unlock_new_inode(inode);
9315 inode_dec_link_count(inode);
9321 static int btrfs_rename(struct user_namespace *mnt_userns,
9322 struct inode *old_dir, struct dentry *old_dentry,
9323 struct inode *new_dir, struct dentry *new_dentry,
9326 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9327 struct btrfs_trans_handle *trans;
9328 unsigned int trans_num_items;
9329 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9330 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9331 struct inode *new_inode = d_inode(new_dentry);
9332 struct inode *old_inode = d_inode(old_dentry);
9333 struct btrfs_rename_ctx rename_ctx;
9337 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9339 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9342 /* we only allow rename subvolume link between subvolumes */
9343 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9346 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9347 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9350 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9351 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9355 /* check for collisions, even if the name isn't there */
9356 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9357 new_dentry->d_name.name,
9358 new_dentry->d_name.len);
9361 if (ret == -EEXIST) {
9363 * eexist without a new_inode */
9364 if (WARN_ON(!new_inode)) {
9368 /* maybe -EOVERFLOW */
9375 * we're using rename to replace one file with another. Start IO on it
9376 * now so we don't add too much work to the end of the transaction
9378 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9379 filemap_flush(old_inode->i_mapping);
9381 /* close the racy window with snapshot create/destroy ioctl */
9382 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9383 down_read(&fs_info->subvol_sem);
9385 * We want to reserve the absolute worst case amount of items. So if
9386 * both inodes are subvols and we need to unlink them then that would
9387 * require 4 item modifications, but if they are both normal inodes it
9388 * would require 5 item modifications, so we'll assume they are normal
9389 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9390 * should cover the worst case number of items we'll modify.
9391 * If our rename has the whiteout flag, we need more 5 units for the
9392 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9393 * when selinux is enabled).
9395 trans_num_items = 11;
9396 if (flags & RENAME_WHITEOUT)
9397 trans_num_items += 5;
9398 trans = btrfs_start_transaction(root, trans_num_items);
9399 if (IS_ERR(trans)) {
9400 ret = PTR_ERR(trans);
9405 ret = btrfs_record_root_in_trans(trans, dest);
9410 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9414 BTRFS_I(old_inode)->dir_index = 0ULL;
9415 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9416 /* force full log commit if subvolume involved. */
9417 btrfs_set_log_full_commit(trans);
9419 ret = btrfs_insert_inode_ref(trans, dest,
9420 new_dentry->d_name.name,
9421 new_dentry->d_name.len,
9423 btrfs_ino(BTRFS_I(new_dir)), index);
9428 inode_inc_iversion(old_dir);
9429 inode_inc_iversion(new_dir);
9430 inode_inc_iversion(old_inode);
9431 old_dir->i_ctime = old_dir->i_mtime =
9432 new_dir->i_ctime = new_dir->i_mtime =
9433 old_inode->i_ctime = current_time(old_dir);
9435 if (old_dentry->d_parent != new_dentry->d_parent)
9436 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9437 BTRFS_I(old_inode), 1);
9439 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9440 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9442 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9443 BTRFS_I(d_inode(old_dentry)),
9444 old_dentry->d_name.name,
9445 old_dentry->d_name.len,
9448 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9451 btrfs_abort_transaction(trans, ret);
9456 inode_inc_iversion(new_inode);
9457 new_inode->i_ctime = current_time(new_inode);
9458 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9459 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9460 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9461 BUG_ON(new_inode->i_nlink == 0);
9463 ret = btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9464 BTRFS_I(d_inode(new_dentry)),
9465 new_dentry->d_name.name,
9466 new_dentry->d_name.len);
9468 if (!ret && new_inode->i_nlink == 0)
9469 ret = btrfs_orphan_add(trans,
9470 BTRFS_I(d_inode(new_dentry)));
9472 btrfs_abort_transaction(trans, ret);
9477 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9478 new_dentry->d_name.name,
9479 new_dentry->d_name.len, 0, index);
9481 btrfs_abort_transaction(trans, ret);
9485 if (old_inode->i_nlink == 1)
9486 BTRFS_I(old_inode)->dir_index = index;
9488 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9489 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9490 rename_ctx.index, new_dentry->d_parent);
9492 if (flags & RENAME_WHITEOUT) {
9493 ret = btrfs_whiteout_for_rename(trans, root, mnt_userns,
9494 old_dir, old_dentry);
9497 btrfs_abort_transaction(trans, ret);
9502 ret2 = btrfs_end_transaction(trans);
9503 ret = ret ? ret : ret2;
9505 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9506 up_read(&fs_info->subvol_sem);
9511 static int btrfs_rename2(struct user_namespace *mnt_userns, struct inode *old_dir,
9512 struct dentry *old_dentry, struct inode *new_dir,
9513 struct dentry *new_dentry, unsigned int flags)
9515 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9518 if (flags & RENAME_EXCHANGE)
9519 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9522 return btrfs_rename(mnt_userns, old_dir, old_dentry, new_dir,
9526 struct btrfs_delalloc_work {
9527 struct inode *inode;
9528 struct completion completion;
9529 struct list_head list;
9530 struct btrfs_work work;
9533 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9535 struct btrfs_delalloc_work *delalloc_work;
9536 struct inode *inode;
9538 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9540 inode = delalloc_work->inode;
9541 filemap_flush(inode->i_mapping);
9542 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9543 &BTRFS_I(inode)->runtime_flags))
9544 filemap_flush(inode->i_mapping);
9547 complete(&delalloc_work->completion);
9550 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9552 struct btrfs_delalloc_work *work;
9554 work = kmalloc(sizeof(*work), GFP_NOFS);
9558 init_completion(&work->completion);
9559 INIT_LIST_HEAD(&work->list);
9560 work->inode = inode;
9561 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9567 * some fairly slow code that needs optimization. This walks the list
9568 * of all the inodes with pending delalloc and forces them to disk.
9570 static int start_delalloc_inodes(struct btrfs_root *root,
9571 struct writeback_control *wbc, bool snapshot,
9572 bool in_reclaim_context)
9574 struct btrfs_inode *binode;
9575 struct inode *inode;
9576 struct btrfs_delalloc_work *work, *next;
9577 struct list_head works;
9578 struct list_head splice;
9580 bool full_flush = wbc->nr_to_write == LONG_MAX;
9582 INIT_LIST_HEAD(&works);
9583 INIT_LIST_HEAD(&splice);
9585 mutex_lock(&root->delalloc_mutex);
9586 spin_lock(&root->delalloc_lock);
9587 list_splice_init(&root->delalloc_inodes, &splice);
9588 while (!list_empty(&splice)) {
9589 binode = list_entry(splice.next, struct btrfs_inode,
9592 list_move_tail(&binode->delalloc_inodes,
9593 &root->delalloc_inodes);
9595 if (in_reclaim_context &&
9596 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9599 inode = igrab(&binode->vfs_inode);
9601 cond_resched_lock(&root->delalloc_lock);
9604 spin_unlock(&root->delalloc_lock);
9607 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9608 &binode->runtime_flags);
9610 work = btrfs_alloc_delalloc_work(inode);
9616 list_add_tail(&work->list, &works);
9617 btrfs_queue_work(root->fs_info->flush_workers,
9620 ret = filemap_fdatawrite_wbc(inode->i_mapping, wbc);
9621 btrfs_add_delayed_iput(inode);
9622 if (ret || wbc->nr_to_write <= 0)
9626 spin_lock(&root->delalloc_lock);
9628 spin_unlock(&root->delalloc_lock);
9631 list_for_each_entry_safe(work, next, &works, list) {
9632 list_del_init(&work->list);
9633 wait_for_completion(&work->completion);
9637 if (!list_empty(&splice)) {
9638 spin_lock(&root->delalloc_lock);
9639 list_splice_tail(&splice, &root->delalloc_inodes);
9640 spin_unlock(&root->delalloc_lock);
9642 mutex_unlock(&root->delalloc_mutex);
9646 int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9648 struct writeback_control wbc = {
9649 .nr_to_write = LONG_MAX,
9650 .sync_mode = WB_SYNC_NONE,
9652 .range_end = LLONG_MAX,
9654 struct btrfs_fs_info *fs_info = root->fs_info;
9656 if (BTRFS_FS_ERROR(fs_info))
9659 return start_delalloc_inodes(root, &wbc, true, in_reclaim_context);
9662 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9663 bool in_reclaim_context)
9665 struct writeback_control wbc = {
9667 .sync_mode = WB_SYNC_NONE,
9669 .range_end = LLONG_MAX,
9671 struct btrfs_root *root;
9672 struct list_head splice;
9675 if (BTRFS_FS_ERROR(fs_info))
9678 INIT_LIST_HEAD(&splice);
9680 mutex_lock(&fs_info->delalloc_root_mutex);
9681 spin_lock(&fs_info->delalloc_root_lock);
9682 list_splice_init(&fs_info->delalloc_roots, &splice);
9683 while (!list_empty(&splice)) {
9685 * Reset nr_to_write here so we know that we're doing a full
9689 wbc.nr_to_write = LONG_MAX;
9691 root = list_first_entry(&splice, struct btrfs_root,
9693 root = btrfs_grab_root(root);
9695 list_move_tail(&root->delalloc_root,
9696 &fs_info->delalloc_roots);
9697 spin_unlock(&fs_info->delalloc_root_lock);
9699 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9700 btrfs_put_root(root);
9701 if (ret < 0 || wbc.nr_to_write <= 0)
9703 spin_lock(&fs_info->delalloc_root_lock);
9705 spin_unlock(&fs_info->delalloc_root_lock);
9709 if (!list_empty(&splice)) {
9710 spin_lock(&fs_info->delalloc_root_lock);
9711 list_splice_tail(&splice, &fs_info->delalloc_roots);
9712 spin_unlock(&fs_info->delalloc_root_lock);
9714 mutex_unlock(&fs_info->delalloc_root_mutex);
9718 static int btrfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
9719 struct dentry *dentry, const char *symname)
9721 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9722 struct btrfs_trans_handle *trans;
9723 struct btrfs_root *root = BTRFS_I(dir)->root;
9724 struct btrfs_path *path;
9725 struct btrfs_key key;
9726 struct inode *inode = NULL;
9733 struct btrfs_file_extent_item *ei;
9734 struct extent_buffer *leaf;
9736 name_len = strlen(symname);
9737 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9738 return -ENAMETOOLONG;
9741 * 2 items for inode item and ref
9742 * 2 items for dir items
9743 * 1 item for updating parent inode item
9744 * 1 item for the inline extent item
9745 * 1 item for xattr if selinux is on
9747 trans = btrfs_start_transaction(root, 7);
9749 return PTR_ERR(trans);
9751 err = btrfs_get_free_objectid(root, &objectid);
9755 inode = btrfs_new_inode(trans, root, mnt_userns, dir,
9756 dentry->d_name.name, dentry->d_name.len,
9757 btrfs_ino(BTRFS_I(dir)), objectid,
9758 S_IFLNK | S_IRWXUGO, &index);
9759 if (IS_ERR(inode)) {
9760 err = PTR_ERR(inode);
9766 * If the active LSM wants to access the inode during
9767 * d_instantiate it needs these. Smack checks to see
9768 * if the filesystem supports xattrs by looking at the
9771 inode->i_fop = &btrfs_file_operations;
9772 inode->i_op = &btrfs_file_inode_operations;
9773 inode->i_mapping->a_ops = &btrfs_aops;
9775 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9779 path = btrfs_alloc_path();
9784 key.objectid = btrfs_ino(BTRFS_I(inode));
9786 key.type = BTRFS_EXTENT_DATA_KEY;
9787 datasize = btrfs_file_extent_calc_inline_size(name_len);
9788 err = btrfs_insert_empty_item(trans, root, path, &key,
9791 btrfs_free_path(path);
9794 leaf = path->nodes[0];
9795 ei = btrfs_item_ptr(leaf, path->slots[0],
9796 struct btrfs_file_extent_item);
9797 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9798 btrfs_set_file_extent_type(leaf, ei,
9799 BTRFS_FILE_EXTENT_INLINE);
9800 btrfs_set_file_extent_encryption(leaf, ei, 0);
9801 btrfs_set_file_extent_compression(leaf, ei, 0);
9802 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9803 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9805 ptr = btrfs_file_extent_inline_start(ei);
9806 write_extent_buffer(leaf, symname, ptr, name_len);
9807 btrfs_mark_buffer_dirty(leaf);
9808 btrfs_free_path(path);
9810 inode->i_op = &btrfs_symlink_inode_operations;
9811 inode_nohighmem(inode);
9812 inode_set_bytes(inode, name_len);
9813 btrfs_i_size_write(BTRFS_I(inode), name_len);
9814 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
9816 * Last step, add directory indexes for our symlink inode. This is the
9817 * last step to avoid extra cleanup of these indexes if an error happens
9821 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9822 BTRFS_I(inode), 0, index);
9826 d_instantiate_new(dentry, inode);
9829 btrfs_end_transaction(trans);
9831 inode_dec_link_count(inode);
9832 discard_new_inode(inode);
9834 btrfs_btree_balance_dirty(fs_info);
9838 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9839 struct btrfs_trans_handle *trans_in,
9840 struct btrfs_inode *inode,
9841 struct btrfs_key *ins,
9844 struct btrfs_file_extent_item stack_fi;
9845 struct btrfs_replace_extent_info extent_info;
9846 struct btrfs_trans_handle *trans = trans_in;
9847 struct btrfs_path *path;
9848 u64 start = ins->objectid;
9849 u64 len = ins->offset;
9850 int qgroup_released;
9853 memset(&stack_fi, 0, sizeof(stack_fi));
9855 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9856 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9857 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9858 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9859 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9860 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9861 /* Encryption and other encoding is reserved and all 0 */
9863 qgroup_released = btrfs_qgroup_release_data(inode, file_offset, len);
9864 if (qgroup_released < 0)
9865 return ERR_PTR(qgroup_released);
9868 ret = insert_reserved_file_extent(trans, inode,
9869 file_offset, &stack_fi,
9870 true, qgroup_released);
9876 extent_info.disk_offset = start;
9877 extent_info.disk_len = len;
9878 extent_info.data_offset = 0;
9879 extent_info.data_len = len;
9880 extent_info.file_offset = file_offset;
9881 extent_info.extent_buf = (char *)&stack_fi;
9882 extent_info.is_new_extent = true;
9883 extent_info.qgroup_reserved = qgroup_released;
9884 extent_info.insertions = 0;
9886 path = btrfs_alloc_path();
9892 ret = btrfs_replace_file_extents(inode, path, file_offset,
9893 file_offset + len - 1, &extent_info,
9895 btrfs_free_path(path);
9902 * We have released qgroup data range at the beginning of the function,
9903 * and normally qgroup_released bytes will be freed when committing
9905 * But if we error out early, we have to free what we have released
9906 * or we leak qgroup data reservation.
9908 btrfs_qgroup_free_refroot(inode->root->fs_info,
9909 inode->root->root_key.objectid, qgroup_released,
9910 BTRFS_QGROUP_RSV_DATA);
9911 return ERR_PTR(ret);
9914 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9915 u64 start, u64 num_bytes, u64 min_size,
9916 loff_t actual_len, u64 *alloc_hint,
9917 struct btrfs_trans_handle *trans)
9919 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9920 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9921 struct extent_map *em;
9922 struct btrfs_root *root = BTRFS_I(inode)->root;
9923 struct btrfs_key ins;
9924 u64 cur_offset = start;
9925 u64 clear_offset = start;
9928 u64 last_alloc = (u64)-1;
9930 bool own_trans = true;
9931 u64 end = start + num_bytes - 1;
9935 while (num_bytes > 0) {
9936 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9937 cur_bytes = max(cur_bytes, min_size);
9939 * If we are severely fragmented we could end up with really
9940 * small allocations, so if the allocator is returning small
9941 * chunks lets make its job easier by only searching for those
9944 cur_bytes = min(cur_bytes, last_alloc);
9945 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9946 min_size, 0, *alloc_hint, &ins, 1, 0);
9951 * We've reserved this space, and thus converted it from
9952 * ->bytes_may_use to ->bytes_reserved. Any error that happens
9953 * from here on out we will only need to clear our reservation
9954 * for the remaining unreserved area, so advance our
9955 * clear_offset by our extent size.
9957 clear_offset += ins.offset;
9959 last_alloc = ins.offset;
9960 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
9963 * Now that we inserted the prealloc extent we can finally
9964 * decrement the number of reservations in the block group.
9965 * If we did it before, we could race with relocation and have
9966 * relocation miss the reserved extent, making it fail later.
9968 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
9969 if (IS_ERR(trans)) {
9970 ret = PTR_ERR(trans);
9971 btrfs_free_reserved_extent(fs_info, ins.objectid,
9976 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
9977 cur_offset + ins.offset -1, 0);
9979 em = alloc_extent_map();
9981 btrfs_set_inode_full_sync(BTRFS_I(inode));
9985 em->start = cur_offset;
9986 em->orig_start = cur_offset;
9987 em->len = ins.offset;
9988 em->block_start = ins.objectid;
9989 em->block_len = ins.offset;
9990 em->orig_block_len = ins.offset;
9991 em->ram_bytes = ins.offset;
9992 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9993 em->generation = trans->transid;
9996 write_lock(&em_tree->lock);
9997 ret = add_extent_mapping(em_tree, em, 1);
9998 write_unlock(&em_tree->lock);
10001 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10002 cur_offset + ins.offset - 1,
10005 free_extent_map(em);
10007 num_bytes -= ins.offset;
10008 cur_offset += ins.offset;
10009 *alloc_hint = ins.objectid + ins.offset;
10011 inode_inc_iversion(inode);
10012 inode->i_ctime = current_time(inode);
10013 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10014 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10015 (actual_len > inode->i_size) &&
10016 (cur_offset > inode->i_size)) {
10017 if (cur_offset > actual_len)
10018 i_size = actual_len;
10020 i_size = cur_offset;
10021 i_size_write(inode, i_size);
10022 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
10025 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
10028 btrfs_abort_transaction(trans, ret);
10030 btrfs_end_transaction(trans);
10035 btrfs_end_transaction(trans);
10039 if (clear_offset < end)
10040 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
10041 end - clear_offset + 1);
10045 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10046 u64 start, u64 num_bytes, u64 min_size,
10047 loff_t actual_len, u64 *alloc_hint)
10049 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10050 min_size, actual_len, alloc_hint,
10054 int btrfs_prealloc_file_range_trans(struct inode *inode,
10055 struct btrfs_trans_handle *trans, int mode,
10056 u64 start, u64 num_bytes, u64 min_size,
10057 loff_t actual_len, u64 *alloc_hint)
10059 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10060 min_size, actual_len, alloc_hint, trans);
10063 static int btrfs_permission(struct user_namespace *mnt_userns,
10064 struct inode *inode, int mask)
10066 struct btrfs_root *root = BTRFS_I(inode)->root;
10067 umode_t mode = inode->i_mode;
10069 if (mask & MAY_WRITE &&
10070 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10071 if (btrfs_root_readonly(root))
10073 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10076 return generic_permission(mnt_userns, inode, mask);
10079 static int btrfs_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
10080 struct dentry *dentry, umode_t mode)
10082 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10083 struct btrfs_trans_handle *trans;
10084 struct btrfs_root *root = BTRFS_I(dir)->root;
10085 struct inode *inode = NULL;
10091 * 5 units required for adding orphan entry
10093 trans = btrfs_start_transaction(root, 5);
10095 return PTR_ERR(trans);
10097 ret = btrfs_get_free_objectid(root, &objectid);
10101 inode = btrfs_new_inode(trans, root, mnt_userns, dir, NULL, 0,
10102 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10103 if (IS_ERR(inode)) {
10104 ret = PTR_ERR(inode);
10109 inode->i_fop = &btrfs_file_operations;
10110 inode->i_op = &btrfs_file_inode_operations;
10112 inode->i_mapping->a_ops = &btrfs_aops;
10114 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10118 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
10121 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10126 * We set number of links to 0 in btrfs_new_inode(), and here we set
10127 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10130 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10132 set_nlink(inode, 1);
10133 d_tmpfile(dentry, inode);
10134 unlock_new_inode(inode);
10135 mark_inode_dirty(inode);
10137 btrfs_end_transaction(trans);
10139 discard_new_inode(inode);
10140 btrfs_btree_balance_dirty(fs_info);
10144 void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end)
10146 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10147 unsigned long index = start >> PAGE_SHIFT;
10148 unsigned long end_index = end >> PAGE_SHIFT;
10152 ASSERT(end + 1 - start <= U32_MAX);
10153 len = end + 1 - start;
10154 while (index <= end_index) {
10155 page = find_get_page(inode->vfs_inode.i_mapping, index);
10156 ASSERT(page); /* Pages should be in the extent_io_tree */
10158 btrfs_page_set_writeback(fs_info, page, start, len);
10164 static int btrfs_encoded_io_compression_from_extent(
10165 struct btrfs_fs_info *fs_info,
10168 switch (compress_type) {
10169 case BTRFS_COMPRESS_NONE:
10170 return BTRFS_ENCODED_IO_COMPRESSION_NONE;
10171 case BTRFS_COMPRESS_ZLIB:
10172 return BTRFS_ENCODED_IO_COMPRESSION_ZLIB;
10173 case BTRFS_COMPRESS_LZO:
10175 * The LZO format depends on the sector size. 64K is the maximum
10176 * sector size that we support.
10178 if (fs_info->sectorsize < SZ_4K || fs_info->sectorsize > SZ_64K)
10180 return BTRFS_ENCODED_IO_COMPRESSION_LZO_4K +
10181 (fs_info->sectorsize_bits - 12);
10182 case BTRFS_COMPRESS_ZSTD:
10183 return BTRFS_ENCODED_IO_COMPRESSION_ZSTD;
10189 static ssize_t btrfs_encoded_read_inline(
10190 struct kiocb *iocb,
10191 struct iov_iter *iter, u64 start,
10193 struct extent_state **cached_state,
10194 u64 extent_start, size_t count,
10195 struct btrfs_ioctl_encoded_io_args *encoded,
10198 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10199 struct btrfs_root *root = inode->root;
10200 struct btrfs_fs_info *fs_info = root->fs_info;
10201 struct extent_io_tree *io_tree = &inode->io_tree;
10202 struct btrfs_path *path;
10203 struct extent_buffer *leaf;
10204 struct btrfs_file_extent_item *item;
10210 path = btrfs_alloc_path();
10215 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
10219 /* The extent item disappeared? */
10224 leaf = path->nodes[0];
10225 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
10227 ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
10228 ptr = btrfs_file_extent_inline_start(item);
10230 encoded->len = min_t(u64, extent_start + ram_bytes,
10231 inode->vfs_inode.i_size) - iocb->ki_pos;
10232 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10233 btrfs_file_extent_compression(leaf, item));
10236 encoded->compression = ret;
10237 if (encoded->compression) {
10238 size_t inline_size;
10240 inline_size = btrfs_file_extent_inline_item_len(leaf,
10242 if (inline_size > count) {
10246 count = inline_size;
10247 encoded->unencoded_len = ram_bytes;
10248 encoded->unencoded_offset = iocb->ki_pos - extent_start;
10250 count = min_t(u64, count, encoded->len);
10251 encoded->len = count;
10252 encoded->unencoded_len = count;
10253 ptr += iocb->ki_pos - extent_start;
10256 tmp = kmalloc(count, GFP_NOFS);
10261 read_extent_buffer(leaf, tmp, ptr, count);
10262 btrfs_release_path(path);
10263 unlock_extent_cached(io_tree, start, lockend, cached_state);
10264 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10267 ret = copy_to_iter(tmp, count, iter);
10272 btrfs_free_path(path);
10276 struct btrfs_encoded_read_private {
10277 struct btrfs_inode *inode;
10279 wait_queue_head_t wait;
10281 blk_status_t status;
10285 static blk_status_t submit_encoded_read_bio(struct btrfs_inode *inode,
10286 struct bio *bio, int mirror_num)
10288 struct btrfs_encoded_read_private *priv = bio->bi_private;
10289 struct btrfs_bio *bbio = btrfs_bio(bio);
10290 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10293 if (!priv->skip_csum) {
10294 ret = btrfs_lookup_bio_sums(&inode->vfs_inode, bio, NULL);
10299 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
10301 btrfs_bio_free_csum(bbio);
10305 atomic_inc(&priv->pending);
10306 ret = btrfs_map_bio(fs_info, bio, mirror_num);
10308 atomic_dec(&priv->pending);
10309 btrfs_bio_free_csum(bbio);
10314 static blk_status_t btrfs_encoded_read_verify_csum(struct btrfs_bio *bbio)
10316 const bool uptodate = (bbio->bio.bi_status == BLK_STS_OK);
10317 struct btrfs_encoded_read_private *priv = bbio->bio.bi_private;
10318 struct btrfs_inode *inode = priv->inode;
10319 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10320 u32 sectorsize = fs_info->sectorsize;
10321 struct bio_vec *bvec;
10322 struct bvec_iter_all iter_all;
10323 u64 start = priv->file_offset;
10324 u32 bio_offset = 0;
10326 if (priv->skip_csum || !uptodate)
10327 return bbio->bio.bi_status;
10329 bio_for_each_segment_all(bvec, &bbio->bio, iter_all) {
10330 unsigned int i, nr_sectors, pgoff;
10332 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
10333 pgoff = bvec->bv_offset;
10334 for (i = 0; i < nr_sectors; i++) {
10335 ASSERT(pgoff < PAGE_SIZE);
10336 if (check_data_csum(&inode->vfs_inode, bbio, bio_offset,
10337 bvec->bv_page, pgoff, start))
10338 return BLK_STS_IOERR;
10339 start += sectorsize;
10340 bio_offset += sectorsize;
10341 pgoff += sectorsize;
10347 static void btrfs_encoded_read_endio(struct bio *bio)
10349 struct btrfs_encoded_read_private *priv = bio->bi_private;
10350 struct btrfs_bio *bbio = btrfs_bio(bio);
10351 blk_status_t status;
10353 status = btrfs_encoded_read_verify_csum(bbio);
10356 * The memory barrier implied by the atomic_dec_return() here
10357 * pairs with the memory barrier implied by the
10358 * atomic_dec_return() or io_wait_event() in
10359 * btrfs_encoded_read_regular_fill_pages() to ensure that this
10360 * write is observed before the load of status in
10361 * btrfs_encoded_read_regular_fill_pages().
10363 WRITE_ONCE(priv->status, status);
10365 if (!atomic_dec_return(&priv->pending))
10366 wake_up(&priv->wait);
10367 btrfs_bio_free_csum(bbio);
10371 static int btrfs_encoded_read_regular_fill_pages(struct btrfs_inode *inode,
10375 struct page **pages)
10377 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10378 struct btrfs_encoded_read_private priv = {
10380 .file_offset = file_offset,
10381 .pending = ATOMIC_INIT(1),
10382 .skip_csum = (inode->flags & BTRFS_INODE_NODATASUM),
10384 unsigned long i = 0;
10388 init_waitqueue_head(&priv.wait);
10390 * Submit bios for the extent, splitting due to bio or stripe limits as
10393 while (cur < disk_io_size) {
10394 struct extent_map *em;
10395 struct btrfs_io_geometry geom;
10396 struct bio *bio = NULL;
10399 em = btrfs_get_chunk_map(fs_info, disk_bytenr + cur,
10400 disk_io_size - cur);
10404 ret = btrfs_get_io_geometry(fs_info, em, BTRFS_MAP_READ,
10405 disk_bytenr + cur, &geom);
10406 free_extent_map(em);
10409 WRITE_ONCE(priv.status, errno_to_blk_status(ret));
10412 remaining = min(geom.len, disk_io_size - cur);
10413 while (bio || remaining) {
10414 size_t bytes = min_t(u64, remaining, PAGE_SIZE);
10417 bio = btrfs_bio_alloc(BIO_MAX_VECS);
10418 bio->bi_iter.bi_sector =
10419 (disk_bytenr + cur) >> SECTOR_SHIFT;
10420 bio->bi_end_io = btrfs_encoded_read_endio;
10421 bio->bi_private = &priv;
10422 bio->bi_opf = REQ_OP_READ;
10426 bio_add_page(bio, pages[i], bytes, 0) < bytes) {
10427 blk_status_t status;
10429 status = submit_encoded_read_bio(inode, bio, 0);
10431 WRITE_ONCE(priv.status, status);
10441 remaining -= bytes;
10446 if (atomic_dec_return(&priv.pending))
10447 io_wait_event(priv.wait, !atomic_read(&priv.pending));
10448 /* See btrfs_encoded_read_endio() for ordering. */
10449 return blk_status_to_errno(READ_ONCE(priv.status));
10452 static ssize_t btrfs_encoded_read_regular(struct kiocb *iocb,
10453 struct iov_iter *iter,
10454 u64 start, u64 lockend,
10455 struct extent_state **cached_state,
10456 u64 disk_bytenr, u64 disk_io_size,
10457 size_t count, bool compressed,
10460 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10461 struct extent_io_tree *io_tree = &inode->io_tree;
10462 struct page **pages;
10463 unsigned long nr_pages, i;
10465 size_t page_offset;
10468 nr_pages = DIV_ROUND_UP(disk_io_size, PAGE_SIZE);
10469 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
10472 for (i = 0; i < nr_pages; i++) {
10473 pages[i] = alloc_page(GFP_NOFS);
10480 ret = btrfs_encoded_read_regular_fill_pages(inode, start, disk_bytenr,
10481 disk_io_size, pages);
10485 unlock_extent_cached(io_tree, start, lockend, cached_state);
10486 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10493 i = (iocb->ki_pos - start) >> PAGE_SHIFT;
10494 page_offset = (iocb->ki_pos - start) & (PAGE_SIZE - 1);
10497 while (cur < count) {
10498 size_t bytes = min_t(size_t, count - cur,
10499 PAGE_SIZE - page_offset);
10501 if (copy_page_to_iter(pages[i], page_offset, bytes,
10512 for (i = 0; i < nr_pages; i++) {
10514 __free_page(pages[i]);
10520 ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter,
10521 struct btrfs_ioctl_encoded_io_args *encoded)
10523 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10524 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10525 struct extent_io_tree *io_tree = &inode->io_tree;
10527 size_t count = iov_iter_count(iter);
10528 u64 start, lockend, disk_bytenr, disk_io_size;
10529 struct extent_state *cached_state = NULL;
10530 struct extent_map *em;
10531 bool unlocked = false;
10533 file_accessed(iocb->ki_filp);
10535 btrfs_inode_lock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10537 if (iocb->ki_pos >= inode->vfs_inode.i_size) {
10538 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10541 start = ALIGN_DOWN(iocb->ki_pos, fs_info->sectorsize);
10543 * We don't know how long the extent containing iocb->ki_pos is, but if
10544 * it's compressed we know that it won't be longer than this.
10546 lockend = start + BTRFS_MAX_UNCOMPRESSED - 1;
10549 struct btrfs_ordered_extent *ordered;
10551 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start,
10552 lockend - start + 1);
10554 goto out_unlock_inode;
10555 lock_extent_bits(io_tree, start, lockend, &cached_state);
10556 ordered = btrfs_lookup_ordered_range(inode, start,
10557 lockend - start + 1);
10560 btrfs_put_ordered_extent(ordered);
10561 unlock_extent_cached(io_tree, start, lockend, &cached_state);
10565 em = btrfs_get_extent(inode, NULL, 0, start, lockend - start + 1);
10568 goto out_unlock_extent;
10571 if (em->block_start == EXTENT_MAP_INLINE) {
10572 u64 extent_start = em->start;
10575 * For inline extents we get everything we need out of the
10578 free_extent_map(em);
10580 ret = btrfs_encoded_read_inline(iocb, iter, start, lockend,
10581 &cached_state, extent_start,
10582 count, encoded, &unlocked);
10587 * We only want to return up to EOF even if the extent extends beyond
10590 encoded->len = min_t(u64, extent_map_end(em),
10591 inode->vfs_inode.i_size) - iocb->ki_pos;
10592 if (em->block_start == EXTENT_MAP_HOLE ||
10593 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
10594 disk_bytenr = EXTENT_MAP_HOLE;
10595 count = min_t(u64, count, encoded->len);
10596 encoded->len = count;
10597 encoded->unencoded_len = count;
10598 } else if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10599 disk_bytenr = em->block_start;
10601 * Bail if the buffer isn't large enough to return the whole
10602 * compressed extent.
10604 if (em->block_len > count) {
10608 disk_io_size = count = em->block_len;
10609 encoded->unencoded_len = em->ram_bytes;
10610 encoded->unencoded_offset = iocb->ki_pos - em->orig_start;
10611 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10612 em->compress_type);
10615 encoded->compression = ret;
10617 disk_bytenr = em->block_start + (start - em->start);
10618 if (encoded->len > count)
10619 encoded->len = count;
10621 * Don't read beyond what we locked. This also limits the page
10622 * allocations that we'll do.
10624 disk_io_size = min(lockend + 1, iocb->ki_pos + encoded->len) - start;
10625 count = start + disk_io_size - iocb->ki_pos;
10626 encoded->len = count;
10627 encoded->unencoded_len = count;
10628 disk_io_size = ALIGN(disk_io_size, fs_info->sectorsize);
10630 free_extent_map(em);
10633 if (disk_bytenr == EXTENT_MAP_HOLE) {
10634 unlock_extent_cached(io_tree, start, lockend, &cached_state);
10635 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10637 ret = iov_iter_zero(count, iter);
10641 ret = btrfs_encoded_read_regular(iocb, iter, start, lockend,
10642 &cached_state, disk_bytenr,
10643 disk_io_size, count,
10644 encoded->compression,
10650 iocb->ki_pos += encoded->len;
10652 free_extent_map(em);
10655 unlock_extent_cached(io_tree, start, lockend, &cached_state);
10658 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10662 ssize_t btrfs_do_encoded_write(struct kiocb *iocb, struct iov_iter *from,
10663 const struct btrfs_ioctl_encoded_io_args *encoded)
10665 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10666 struct btrfs_root *root = inode->root;
10667 struct btrfs_fs_info *fs_info = root->fs_info;
10668 struct extent_io_tree *io_tree = &inode->io_tree;
10669 struct extent_changeset *data_reserved = NULL;
10670 struct extent_state *cached_state = NULL;
10674 u64 num_bytes, ram_bytes, disk_num_bytes;
10675 unsigned long nr_pages, i;
10676 struct page **pages;
10677 struct btrfs_key ins;
10678 bool extent_reserved = false;
10679 struct extent_map *em;
10682 switch (encoded->compression) {
10683 case BTRFS_ENCODED_IO_COMPRESSION_ZLIB:
10684 compression = BTRFS_COMPRESS_ZLIB;
10686 case BTRFS_ENCODED_IO_COMPRESSION_ZSTD:
10687 compression = BTRFS_COMPRESS_ZSTD;
10689 case BTRFS_ENCODED_IO_COMPRESSION_LZO_4K:
10690 case BTRFS_ENCODED_IO_COMPRESSION_LZO_8K:
10691 case BTRFS_ENCODED_IO_COMPRESSION_LZO_16K:
10692 case BTRFS_ENCODED_IO_COMPRESSION_LZO_32K:
10693 case BTRFS_ENCODED_IO_COMPRESSION_LZO_64K:
10694 /* The sector size must match for LZO. */
10695 if (encoded->compression -
10696 BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + 12 !=
10697 fs_info->sectorsize_bits)
10699 compression = BTRFS_COMPRESS_LZO;
10704 if (encoded->encryption != BTRFS_ENCODED_IO_ENCRYPTION_NONE)
10707 orig_count = iov_iter_count(from);
10709 /* The extent size must be sane. */
10710 if (encoded->unencoded_len > BTRFS_MAX_UNCOMPRESSED ||
10711 orig_count > BTRFS_MAX_COMPRESSED || orig_count == 0)
10715 * The compressed data must be smaller than the decompressed data.
10717 * It's of course possible for data to compress to larger or the same
10718 * size, but the buffered I/O path falls back to no compression for such
10719 * data, and we don't want to break any assumptions by creating these
10722 * Note that this is less strict than the current check we have that the
10723 * compressed data must be at least one sector smaller than the
10724 * decompressed data. We only want to enforce the weaker requirement
10725 * from old kernels that it is at least one byte smaller.
10727 if (orig_count >= encoded->unencoded_len)
10730 /* The extent must start on a sector boundary. */
10731 start = iocb->ki_pos;
10732 if (!IS_ALIGNED(start, fs_info->sectorsize))
10736 * The extent must end on a sector boundary. However, we allow a write
10737 * which ends at or extends i_size to have an unaligned length; we round
10738 * up the extent size and set i_size to the unaligned end.
10740 if (start + encoded->len < inode->vfs_inode.i_size &&
10741 !IS_ALIGNED(start + encoded->len, fs_info->sectorsize))
10744 /* Finally, the offset in the unencoded data must be sector-aligned. */
10745 if (!IS_ALIGNED(encoded->unencoded_offset, fs_info->sectorsize))
10748 num_bytes = ALIGN(encoded->len, fs_info->sectorsize);
10749 ram_bytes = ALIGN(encoded->unencoded_len, fs_info->sectorsize);
10750 end = start + num_bytes - 1;
10753 * If the extent cannot be inline, the compressed data on disk must be
10754 * sector-aligned. For convenience, we extend it with zeroes if it
10757 disk_num_bytes = ALIGN(orig_count, fs_info->sectorsize);
10758 nr_pages = DIV_ROUND_UP(disk_num_bytes, PAGE_SIZE);
10759 pages = kvcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL_ACCOUNT);
10762 for (i = 0; i < nr_pages; i++) {
10763 size_t bytes = min_t(size_t, PAGE_SIZE, iov_iter_count(from));
10766 pages[i] = alloc_page(GFP_KERNEL_ACCOUNT);
10771 kaddr = kmap(pages[i]);
10772 if (copy_from_iter(kaddr, bytes, from) != bytes) {
10777 if (bytes < PAGE_SIZE)
10778 memset(kaddr + bytes, 0, PAGE_SIZE - bytes);
10783 struct btrfs_ordered_extent *ordered;
10785 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start, num_bytes);
10788 ret = invalidate_inode_pages2_range(inode->vfs_inode.i_mapping,
10789 start >> PAGE_SHIFT,
10790 end >> PAGE_SHIFT);
10793 lock_extent_bits(io_tree, start, end, &cached_state);
10794 ordered = btrfs_lookup_ordered_range(inode, start, num_bytes);
10796 !filemap_range_has_page(inode->vfs_inode.i_mapping, start, end))
10799 btrfs_put_ordered_extent(ordered);
10800 unlock_extent_cached(io_tree, start, end, &cached_state);
10805 * We don't use the higher-level delalloc space functions because our
10806 * num_bytes and disk_num_bytes are different.
10808 ret = btrfs_alloc_data_chunk_ondemand(inode, disk_num_bytes);
10811 ret = btrfs_qgroup_reserve_data(inode, &data_reserved, start, num_bytes);
10813 goto out_free_data_space;
10814 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes, disk_num_bytes);
10816 goto out_qgroup_free_data;
10818 /* Try an inline extent first. */
10819 if (start == 0 && encoded->unencoded_len == encoded->len &&
10820 encoded->unencoded_offset == 0) {
10821 ret = cow_file_range_inline(inode, encoded->len, orig_count,
10822 compression, pages, true);
10826 goto out_delalloc_release;
10830 ret = btrfs_reserve_extent(root, disk_num_bytes, disk_num_bytes,
10831 disk_num_bytes, 0, 0, &ins, 1, 1);
10833 goto out_delalloc_release;
10834 extent_reserved = true;
10836 em = create_io_em(inode, start, num_bytes,
10837 start - encoded->unencoded_offset, ins.objectid,
10838 ins.offset, ins.offset, ram_bytes, compression,
10839 BTRFS_ORDERED_COMPRESSED);
10842 goto out_free_reserved;
10844 free_extent_map(em);
10846 ret = btrfs_add_ordered_extent(inode, start, num_bytes, ram_bytes,
10847 ins.objectid, ins.offset,
10848 encoded->unencoded_offset,
10849 (1 << BTRFS_ORDERED_ENCODED) |
10850 (1 << BTRFS_ORDERED_COMPRESSED),
10853 btrfs_drop_extent_cache(inode, start, end, 0);
10854 goto out_free_reserved;
10856 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10858 if (start + encoded->len > inode->vfs_inode.i_size)
10859 i_size_write(&inode->vfs_inode, start + encoded->len);
10861 unlock_extent_cached(io_tree, start, end, &cached_state);
10863 btrfs_delalloc_release_extents(inode, num_bytes);
10865 if (btrfs_submit_compressed_write(inode, start, num_bytes, ins.objectid,
10866 ins.offset, pages, nr_pages, 0, NULL,
10868 btrfs_writepage_endio_finish_ordered(inode, pages[0], start, end, 0);
10876 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10877 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
10878 out_delalloc_release:
10879 btrfs_delalloc_release_extents(inode, num_bytes);
10880 btrfs_delalloc_release_metadata(inode, disk_num_bytes, ret < 0);
10881 out_qgroup_free_data:
10883 btrfs_qgroup_free_data(inode, data_reserved, start, num_bytes);
10884 out_free_data_space:
10886 * If btrfs_reserve_extent() succeeded, then we already decremented
10889 if (!extent_reserved)
10890 btrfs_free_reserved_data_space_noquota(fs_info, disk_num_bytes);
10892 unlock_extent_cached(io_tree, start, end, &cached_state);
10894 for (i = 0; i < nr_pages; i++) {
10896 __free_page(pages[i]);
10901 iocb->ki_pos += encoded->len;
10907 * Add an entry indicating a block group or device which is pinned by a
10908 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10909 * negative errno on failure.
10911 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10912 bool is_block_group)
10914 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10915 struct btrfs_swapfile_pin *sp, *entry;
10916 struct rb_node **p;
10917 struct rb_node *parent = NULL;
10919 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10924 sp->is_block_group = is_block_group;
10925 sp->bg_extent_count = 1;
10927 spin_lock(&fs_info->swapfile_pins_lock);
10928 p = &fs_info->swapfile_pins.rb_node;
10931 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10932 if (sp->ptr < entry->ptr ||
10933 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10934 p = &(*p)->rb_left;
10935 } else if (sp->ptr > entry->ptr ||
10936 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10937 p = &(*p)->rb_right;
10939 if (is_block_group)
10940 entry->bg_extent_count++;
10941 spin_unlock(&fs_info->swapfile_pins_lock);
10946 rb_link_node(&sp->node, parent, p);
10947 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10948 spin_unlock(&fs_info->swapfile_pins_lock);
10952 /* Free all of the entries pinned by this swapfile. */
10953 static void btrfs_free_swapfile_pins(struct inode *inode)
10955 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10956 struct btrfs_swapfile_pin *sp;
10957 struct rb_node *node, *next;
10959 spin_lock(&fs_info->swapfile_pins_lock);
10960 node = rb_first(&fs_info->swapfile_pins);
10962 next = rb_next(node);
10963 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10964 if (sp->inode == inode) {
10965 rb_erase(&sp->node, &fs_info->swapfile_pins);
10966 if (sp->is_block_group) {
10967 btrfs_dec_block_group_swap_extents(sp->ptr,
10968 sp->bg_extent_count);
10969 btrfs_put_block_group(sp->ptr);
10975 spin_unlock(&fs_info->swapfile_pins_lock);
10978 struct btrfs_swap_info {
10984 unsigned long nr_pages;
10988 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10989 struct btrfs_swap_info *bsi)
10991 unsigned long nr_pages;
10992 unsigned long max_pages;
10993 u64 first_ppage, first_ppage_reported, next_ppage;
10997 * Our swapfile may have had its size extended after the swap header was
10998 * written. In that case activating the swapfile should not go beyond
10999 * the max size set in the swap header.
11001 if (bsi->nr_pages >= sis->max)
11004 max_pages = sis->max - bsi->nr_pages;
11005 first_ppage = ALIGN(bsi->block_start, PAGE_SIZE) >> PAGE_SHIFT;
11006 next_ppage = ALIGN_DOWN(bsi->block_start + bsi->block_len,
11007 PAGE_SIZE) >> PAGE_SHIFT;
11009 if (first_ppage >= next_ppage)
11011 nr_pages = next_ppage - first_ppage;
11012 nr_pages = min(nr_pages, max_pages);
11014 first_ppage_reported = first_ppage;
11015 if (bsi->start == 0)
11016 first_ppage_reported++;
11017 if (bsi->lowest_ppage > first_ppage_reported)
11018 bsi->lowest_ppage = first_ppage_reported;
11019 if (bsi->highest_ppage < (next_ppage - 1))
11020 bsi->highest_ppage = next_ppage - 1;
11022 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
11025 bsi->nr_extents += ret;
11026 bsi->nr_pages += nr_pages;
11030 static void btrfs_swap_deactivate(struct file *file)
11032 struct inode *inode = file_inode(file);
11034 btrfs_free_swapfile_pins(inode);
11035 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
11038 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
11041 struct inode *inode = file_inode(file);
11042 struct btrfs_root *root = BTRFS_I(inode)->root;
11043 struct btrfs_fs_info *fs_info = root->fs_info;
11044 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
11045 struct extent_state *cached_state = NULL;
11046 struct extent_map *em = NULL;
11047 struct btrfs_device *device = NULL;
11048 struct btrfs_swap_info bsi = {
11049 .lowest_ppage = (sector_t)-1ULL,
11056 * If the swap file was just created, make sure delalloc is done. If the
11057 * file changes again after this, the user is doing something stupid and
11058 * we don't really care.
11060 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
11065 * The inode is locked, so these flags won't change after we check them.
11067 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
11068 btrfs_warn(fs_info, "swapfile must not be compressed");
11071 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
11072 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
11075 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
11076 btrfs_warn(fs_info, "swapfile must not be checksummed");
11081 * Balance or device remove/replace/resize can move stuff around from
11082 * under us. The exclop protection makes sure they aren't running/won't
11083 * run concurrently while we are mapping the swap extents, and
11084 * fs_info->swapfile_pins prevents them from running while the swap
11085 * file is active and moving the extents. Note that this also prevents
11086 * a concurrent device add which isn't actually necessary, but it's not
11087 * really worth the trouble to allow it.
11089 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
11090 btrfs_warn(fs_info,
11091 "cannot activate swapfile while exclusive operation is running");
11096 * Prevent snapshot creation while we are activating the swap file.
11097 * We do not want to race with snapshot creation. If snapshot creation
11098 * already started before we bumped nr_swapfiles from 0 to 1 and
11099 * completes before the first write into the swap file after it is
11100 * activated, than that write would fallback to COW.
11102 if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) {
11103 btrfs_exclop_finish(fs_info);
11104 btrfs_warn(fs_info,
11105 "cannot activate swapfile because snapshot creation is in progress");
11109 * Snapshots can create extents which require COW even if NODATACOW is
11110 * set. We use this counter to prevent snapshots. We must increment it
11111 * before walking the extents because we don't want a concurrent
11112 * snapshot to run after we've already checked the extents.
11114 * It is possible that subvolume is marked for deletion but still not
11115 * removed yet. To prevent this race, we check the root status before
11116 * activating the swapfile.
11118 spin_lock(&root->root_item_lock);
11119 if (btrfs_root_dead(root)) {
11120 spin_unlock(&root->root_item_lock);
11122 btrfs_exclop_finish(fs_info);
11123 btrfs_warn(fs_info,
11124 "cannot activate swapfile because subvolume %llu is being deleted",
11125 root->root_key.objectid);
11128 atomic_inc(&root->nr_swapfiles);
11129 spin_unlock(&root->root_item_lock);
11131 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
11133 lock_extent_bits(io_tree, 0, isize - 1, &cached_state);
11135 while (start < isize) {
11136 u64 logical_block_start, physical_block_start;
11137 struct btrfs_block_group *bg;
11138 u64 len = isize - start;
11140 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
11146 if (em->block_start == EXTENT_MAP_HOLE) {
11147 btrfs_warn(fs_info, "swapfile must not have holes");
11151 if (em->block_start == EXTENT_MAP_INLINE) {
11153 * It's unlikely we'll ever actually find ourselves
11154 * here, as a file small enough to fit inline won't be
11155 * big enough to store more than the swap header, but in
11156 * case something changes in the future, let's catch it
11157 * here rather than later.
11159 btrfs_warn(fs_info, "swapfile must not be inline");
11163 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
11164 btrfs_warn(fs_info, "swapfile must not be compressed");
11169 logical_block_start = em->block_start + (start - em->start);
11170 len = min(len, em->len - (start - em->start));
11171 free_extent_map(em);
11174 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, true);
11180 btrfs_warn(fs_info,
11181 "swapfile must not be copy-on-write");
11186 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
11192 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
11193 btrfs_warn(fs_info,
11194 "swapfile must have single data profile");
11199 if (device == NULL) {
11200 device = em->map_lookup->stripes[0].dev;
11201 ret = btrfs_add_swapfile_pin(inode, device, false);
11206 } else if (device != em->map_lookup->stripes[0].dev) {
11207 btrfs_warn(fs_info, "swapfile must be on one device");
11212 physical_block_start = (em->map_lookup->stripes[0].physical +
11213 (logical_block_start - em->start));
11214 len = min(len, em->len - (logical_block_start - em->start));
11215 free_extent_map(em);
11218 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
11220 btrfs_warn(fs_info,
11221 "could not find block group containing swapfile");
11226 if (!btrfs_inc_block_group_swap_extents(bg)) {
11227 btrfs_warn(fs_info,
11228 "block group for swapfile at %llu is read-only%s",
11230 atomic_read(&fs_info->scrubs_running) ?
11231 " (scrub running)" : "");
11232 btrfs_put_block_group(bg);
11237 ret = btrfs_add_swapfile_pin(inode, bg, true);
11239 btrfs_put_block_group(bg);
11246 if (bsi.block_len &&
11247 bsi.block_start + bsi.block_len == physical_block_start) {
11248 bsi.block_len += len;
11250 if (bsi.block_len) {
11251 ret = btrfs_add_swap_extent(sis, &bsi);
11256 bsi.block_start = physical_block_start;
11257 bsi.block_len = len;
11264 ret = btrfs_add_swap_extent(sis, &bsi);
11267 if (!IS_ERR_OR_NULL(em))
11268 free_extent_map(em);
11270 unlock_extent_cached(io_tree, 0, isize - 1, &cached_state);
11273 btrfs_swap_deactivate(file);
11275 btrfs_drew_write_unlock(&root->snapshot_lock);
11277 btrfs_exclop_finish(fs_info);
11283 sis->bdev = device->bdev;
11284 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
11285 sis->max = bsi.nr_pages;
11286 sis->pages = bsi.nr_pages - 1;
11287 sis->highest_bit = bsi.nr_pages - 1;
11288 return bsi.nr_extents;
11291 static void btrfs_swap_deactivate(struct file *file)
11295 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
11298 return -EOPNOTSUPP;
11303 * Update the number of bytes used in the VFS' inode. When we replace extents in
11304 * a range (clone, dedupe, fallocate's zero range), we must update the number of
11305 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
11306 * always get a correct value.
11308 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
11309 const u64 add_bytes,
11310 const u64 del_bytes)
11312 if (add_bytes == del_bytes)
11315 spin_lock(&inode->lock);
11317 inode_sub_bytes(&inode->vfs_inode, del_bytes);
11319 inode_add_bytes(&inode->vfs_inode, add_bytes);
11320 spin_unlock(&inode->lock);
11323 static const struct inode_operations btrfs_dir_inode_operations = {
11324 .getattr = btrfs_getattr,
11325 .lookup = btrfs_lookup,
11326 .create = btrfs_create,
11327 .unlink = btrfs_unlink,
11328 .link = btrfs_link,
11329 .mkdir = btrfs_mkdir,
11330 .rmdir = btrfs_rmdir,
11331 .rename = btrfs_rename2,
11332 .symlink = btrfs_symlink,
11333 .setattr = btrfs_setattr,
11334 .mknod = btrfs_mknod,
11335 .listxattr = btrfs_listxattr,
11336 .permission = btrfs_permission,
11337 .get_acl = btrfs_get_acl,
11338 .set_acl = btrfs_set_acl,
11339 .update_time = btrfs_update_time,
11340 .tmpfile = btrfs_tmpfile,
11341 .fileattr_get = btrfs_fileattr_get,
11342 .fileattr_set = btrfs_fileattr_set,
11345 static const struct file_operations btrfs_dir_file_operations = {
11346 .llseek = generic_file_llseek,
11347 .read = generic_read_dir,
11348 .iterate_shared = btrfs_real_readdir,
11349 .open = btrfs_opendir,
11350 .unlocked_ioctl = btrfs_ioctl,
11351 #ifdef CONFIG_COMPAT
11352 .compat_ioctl = btrfs_compat_ioctl,
11354 .release = btrfs_release_file,
11355 .fsync = btrfs_sync_file,
11359 * btrfs doesn't support the bmap operation because swapfiles
11360 * use bmap to make a mapping of extents in the file. They assume
11361 * these extents won't change over the life of the file and they
11362 * use the bmap result to do IO directly to the drive.
11364 * the btrfs bmap call would return logical addresses that aren't
11365 * suitable for IO and they also will change frequently as COW
11366 * operations happen. So, swapfile + btrfs == corruption.
11368 * For now we're avoiding this by dropping bmap.
11370 static const struct address_space_operations btrfs_aops = {
11371 .readpage = btrfs_readpage,
11372 .writepage = btrfs_writepage,
11373 .writepages = btrfs_writepages,
11374 .readahead = btrfs_readahead,
11375 .direct_IO = noop_direct_IO,
11376 .invalidate_folio = btrfs_invalidate_folio,
11377 .releasepage = btrfs_releasepage,
11378 #ifdef CONFIG_MIGRATION
11379 .migratepage = btrfs_migratepage,
11381 .dirty_folio = filemap_dirty_folio,
11382 .error_remove_page = generic_error_remove_page,
11383 .swap_activate = btrfs_swap_activate,
11384 .swap_deactivate = btrfs_swap_deactivate,
11387 static const struct inode_operations btrfs_file_inode_operations = {
11388 .getattr = btrfs_getattr,
11389 .setattr = btrfs_setattr,
11390 .listxattr = btrfs_listxattr,
11391 .permission = btrfs_permission,
11392 .fiemap = btrfs_fiemap,
11393 .get_acl = btrfs_get_acl,
11394 .set_acl = btrfs_set_acl,
11395 .update_time = btrfs_update_time,
11396 .fileattr_get = btrfs_fileattr_get,
11397 .fileattr_set = btrfs_fileattr_set,
11399 static const struct inode_operations btrfs_special_inode_operations = {
11400 .getattr = btrfs_getattr,
11401 .setattr = btrfs_setattr,
11402 .permission = btrfs_permission,
11403 .listxattr = btrfs_listxattr,
11404 .get_acl = btrfs_get_acl,
11405 .set_acl = btrfs_set_acl,
11406 .update_time = btrfs_update_time,
11408 static const struct inode_operations btrfs_symlink_inode_operations = {
11409 .get_link = page_get_link,
11410 .getattr = btrfs_getattr,
11411 .setattr = btrfs_setattr,
11412 .permission = btrfs_permission,
11413 .listxattr = btrfs_listxattr,
11414 .update_time = btrfs_update_time,
11417 const struct dentry_operations btrfs_dentry_operations = {
11418 .d_delete = btrfs_dentry_delete,