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 static const struct inode_operations btrfs_dir_inode_operations;
70 static const struct inode_operations btrfs_symlink_inode_operations;
71 static const struct inode_operations btrfs_special_inode_operations;
72 static const struct inode_operations btrfs_file_inode_operations;
73 static const struct address_space_operations btrfs_aops;
74 static const struct file_operations btrfs_dir_file_operations;
76 static struct kmem_cache *btrfs_inode_cachep;
77 struct kmem_cache *btrfs_trans_handle_cachep;
78 struct kmem_cache *btrfs_path_cachep;
79 struct kmem_cache *btrfs_free_space_cachep;
80 struct kmem_cache *btrfs_free_space_bitmap_cachep;
82 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
83 static int btrfs_truncate(struct inode *inode, bool skip_writeback);
84 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
85 static noinline int cow_file_range(struct btrfs_inode *inode,
86 struct page *locked_page,
87 u64 start, u64 end, int *page_started,
88 unsigned long *nr_written, int unlock);
89 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
90 u64 len, u64 orig_start, u64 block_start,
91 u64 block_len, u64 orig_block_len,
92 u64 ram_bytes, int compress_type,
95 static void __endio_write_update_ordered(struct btrfs_inode *inode,
96 const u64 offset, const u64 bytes,
100 * btrfs_inode_lock - lock inode i_rwsem based on arguments passed
102 * ilock_flags can have the following bit set:
104 * BTRFS_ILOCK_SHARED - acquire a shared lock on the inode
105 * BTRFS_ILOCK_TRY - try to acquire the lock, if fails on first attempt
107 * BTRFS_ILOCK_MMAP - acquire a write lock on the i_mmap_lock
109 int btrfs_inode_lock(struct inode *inode, unsigned int ilock_flags)
111 if (ilock_flags & BTRFS_ILOCK_SHARED) {
112 if (ilock_flags & BTRFS_ILOCK_TRY) {
113 if (!inode_trylock_shared(inode))
118 inode_lock_shared(inode);
120 if (ilock_flags & BTRFS_ILOCK_TRY) {
121 if (!inode_trylock(inode))
128 if (ilock_flags & BTRFS_ILOCK_MMAP)
129 down_write(&BTRFS_I(inode)->i_mmap_lock);
134 * btrfs_inode_unlock - unock inode i_rwsem
136 * ilock_flags should contain the same bits set as passed to btrfs_inode_lock()
137 * to decide whether the lock acquired is shared or exclusive.
139 void btrfs_inode_unlock(struct inode *inode, unsigned int ilock_flags)
141 if (ilock_flags & BTRFS_ILOCK_MMAP)
142 up_write(&BTRFS_I(inode)->i_mmap_lock);
143 if (ilock_flags & BTRFS_ILOCK_SHARED)
144 inode_unlock_shared(inode);
150 * Cleanup all submitted ordered extents in specified range to handle errors
151 * from the btrfs_run_delalloc_range() callback.
153 * NOTE: caller must ensure that when an error happens, it can not call
154 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
155 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
156 * to be released, which we want to happen only when finishing the ordered
157 * extent (btrfs_finish_ordered_io()).
159 static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode,
160 struct page *locked_page,
161 u64 offset, u64 bytes)
163 unsigned long index = offset >> PAGE_SHIFT;
164 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
165 u64 page_start = page_offset(locked_page);
166 u64 page_end = page_start + PAGE_SIZE - 1;
170 while (index <= end_index) {
172 * For locked page, we will call end_extent_writepage() on it
173 * in run_delalloc_range() for the error handling. That
174 * end_extent_writepage() function will call
175 * btrfs_mark_ordered_io_finished() to clear page Ordered and
176 * run the ordered extent accounting.
178 * Here we can't just clear the Ordered bit, or
179 * btrfs_mark_ordered_io_finished() would skip the accounting
180 * for the page range, and the ordered extent will never finish.
182 if (index == (page_offset(locked_page) >> PAGE_SHIFT)) {
186 page = find_get_page(inode->vfs_inode.i_mapping, index);
192 * Here we just clear all Ordered bits for every page in the
193 * range, then __endio_write_update_ordered() will handle
194 * the ordered extent accounting for the range.
196 btrfs_page_clamp_clear_ordered(inode->root->fs_info, page,
201 /* The locked page covers the full range, nothing needs to be done */
202 if (bytes + offset <= page_offset(locked_page) + PAGE_SIZE)
205 * In case this page belongs to the delalloc range being instantiated
206 * then skip it, since the first page of a range is going to be
207 * properly cleaned up by the caller of run_delalloc_range
209 if (page_start >= offset && page_end <= (offset + bytes - 1)) {
210 bytes = offset + bytes - page_offset(locked_page) - PAGE_SIZE;
211 offset = page_offset(locked_page) + PAGE_SIZE;
214 return __endio_write_update_ordered(inode, offset, bytes, false);
217 static int btrfs_dirty_inode(struct inode *inode);
219 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
220 struct inode *inode, struct inode *dir,
221 const struct qstr *qstr)
225 err = btrfs_init_acl(trans, inode, dir);
227 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
232 * this does all the hard work for inserting an inline extent into
233 * the btree. The caller should have done a btrfs_drop_extents so that
234 * no overlapping inline items exist in the btree
236 static int insert_inline_extent(struct btrfs_trans_handle *trans,
237 struct btrfs_path *path, bool extent_inserted,
238 struct btrfs_root *root, struct inode *inode,
239 u64 start, size_t size, size_t compressed_size,
241 struct page **compressed_pages)
243 struct extent_buffer *leaf;
244 struct page *page = NULL;
247 struct btrfs_file_extent_item *ei;
249 size_t cur_size = size;
250 unsigned long offset;
252 ASSERT((compressed_size > 0 && compressed_pages) ||
253 (compressed_size == 0 && !compressed_pages));
255 if (compressed_size && compressed_pages)
256 cur_size = compressed_size;
258 if (!extent_inserted) {
259 struct btrfs_key key;
262 key.objectid = btrfs_ino(BTRFS_I(inode));
264 key.type = BTRFS_EXTENT_DATA_KEY;
266 datasize = btrfs_file_extent_calc_inline_size(cur_size);
267 ret = btrfs_insert_empty_item(trans, root, path, &key,
272 leaf = path->nodes[0];
273 ei = btrfs_item_ptr(leaf, path->slots[0],
274 struct btrfs_file_extent_item);
275 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
276 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
277 btrfs_set_file_extent_encryption(leaf, ei, 0);
278 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
279 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
280 ptr = btrfs_file_extent_inline_start(ei);
282 if (compress_type != BTRFS_COMPRESS_NONE) {
285 while (compressed_size > 0) {
286 cpage = compressed_pages[i];
287 cur_size = min_t(unsigned long, compressed_size,
290 kaddr = kmap_atomic(cpage);
291 write_extent_buffer(leaf, kaddr, ptr, cur_size);
292 kunmap_atomic(kaddr);
296 compressed_size -= cur_size;
298 btrfs_set_file_extent_compression(leaf, ei,
301 page = find_get_page(inode->i_mapping,
302 start >> PAGE_SHIFT);
303 btrfs_set_file_extent_compression(leaf, ei, 0);
304 kaddr = kmap_atomic(page);
305 offset = offset_in_page(start);
306 write_extent_buffer(leaf, kaddr + offset, ptr, size);
307 kunmap_atomic(kaddr);
310 btrfs_mark_buffer_dirty(leaf);
311 btrfs_release_path(path);
314 * We align size to sectorsize for inline extents just for simplicity
317 size = ALIGN(size, root->fs_info->sectorsize);
318 ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start, size);
323 * we're an inline extent, so nobody can
324 * extend the file past i_size without locking
325 * a page we already have locked.
327 * We must do any isize and inode updates
328 * before we unlock the pages. Otherwise we
329 * could end up racing with unlink.
331 BTRFS_I(inode)->disk_i_size = inode->i_size;
338 * conditionally insert an inline extent into the file. This
339 * does the checks required to make sure the data is small enough
340 * to fit as an inline extent.
342 static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 start,
343 u64 end, size_t compressed_size,
345 struct page **compressed_pages)
347 struct btrfs_drop_extents_args drop_args = { 0 };
348 struct btrfs_root *root = inode->root;
349 struct btrfs_fs_info *fs_info = root->fs_info;
350 struct btrfs_trans_handle *trans;
351 u64 isize = i_size_read(&inode->vfs_inode);
352 u64 actual_end = min(end + 1, isize);
353 u64 inline_len = actual_end - start;
354 u64 aligned_end = ALIGN(end, fs_info->sectorsize);
355 u64 data_len = inline_len;
357 struct btrfs_path *path;
360 data_len = compressed_size;
363 actual_end > fs_info->sectorsize ||
364 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
366 (actual_end & (fs_info->sectorsize - 1)) == 0) ||
368 data_len > fs_info->max_inline) {
372 path = btrfs_alloc_path();
376 trans = btrfs_join_transaction(root);
378 btrfs_free_path(path);
379 return PTR_ERR(trans);
381 trans->block_rsv = &inode->block_rsv;
383 drop_args.path = path;
384 drop_args.start = start;
385 drop_args.end = aligned_end;
386 drop_args.drop_cache = true;
387 drop_args.replace_extent = true;
389 if (compressed_size && compressed_pages)
390 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(
393 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(
396 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
398 btrfs_abort_transaction(trans, ret);
402 if (isize > actual_end)
403 inline_len = min_t(u64, isize, actual_end);
404 ret = insert_inline_extent(trans, path, drop_args.extent_inserted,
405 root, &inode->vfs_inode, start,
406 inline_len, compressed_size,
407 compress_type, compressed_pages);
408 if (ret && ret != -ENOSPC) {
409 btrfs_abort_transaction(trans, ret);
411 } else if (ret == -ENOSPC) {
416 btrfs_update_inode_bytes(inode, inline_len, drop_args.bytes_found);
417 ret = btrfs_update_inode(trans, root, inode);
418 if (ret && ret != -ENOSPC) {
419 btrfs_abort_transaction(trans, ret);
421 } else if (ret == -ENOSPC) {
426 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
429 * Don't forget to free the reserved space, as for inlined extent
430 * it won't count as data extent, free them directly here.
431 * And at reserve time, it's always aligned to page size, so
432 * just free one page here.
434 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
435 btrfs_free_path(path);
436 btrfs_end_transaction(trans);
440 struct async_extent {
445 unsigned long nr_pages;
447 struct list_head list;
452 struct page *locked_page;
455 unsigned int write_flags;
456 struct list_head extents;
457 struct cgroup_subsys_state *blkcg_css;
458 struct btrfs_work work;
459 struct async_cow *async_cow;
464 struct async_chunk chunks[];
467 static noinline int add_async_extent(struct async_chunk *cow,
468 u64 start, u64 ram_size,
471 unsigned long nr_pages,
474 struct async_extent *async_extent;
476 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
477 BUG_ON(!async_extent); /* -ENOMEM */
478 async_extent->start = start;
479 async_extent->ram_size = ram_size;
480 async_extent->compressed_size = compressed_size;
481 async_extent->pages = pages;
482 async_extent->nr_pages = nr_pages;
483 async_extent->compress_type = compress_type;
484 list_add_tail(&async_extent->list, &cow->extents);
489 * Check if the inode has flags compatible with compression
491 static inline bool inode_can_compress(struct btrfs_inode *inode)
493 if (inode->flags & BTRFS_INODE_NODATACOW ||
494 inode->flags & BTRFS_INODE_NODATASUM)
500 * Check if the inode needs to be submitted to compression, based on mount
501 * options, defragmentation, properties or heuristics.
503 static inline int inode_need_compress(struct btrfs_inode *inode, u64 start,
506 struct btrfs_fs_info *fs_info = inode->root->fs_info;
508 if (!inode_can_compress(inode)) {
509 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
510 KERN_ERR "BTRFS: unexpected compression for ino %llu\n",
515 * Special check for subpage.
517 * We lock the full page then run each delalloc range in the page, thus
518 * for the following case, we will hit some subpage specific corner case:
521 * | |///////| |///////|
524 * In above case, both range A and range B will try to unlock the full
525 * page [0, 64K), causing the one finished later will have page
526 * unlocked already, triggering various page lock requirement BUG_ON()s.
528 * So here we add an artificial limit that subpage compression can only
529 * if the range is fully page aligned.
531 * In theory we only need to ensure the first page is fully covered, but
532 * the tailing partial page will be locked until the full compression
533 * finishes, delaying the write of other range.
535 * TODO: Make btrfs_run_delalloc_range() to lock all delalloc range
536 * first to prevent any submitted async extent to unlock the full page.
537 * By this, we can ensure for subpage case that only the last async_cow
538 * will unlock the full page.
540 if (fs_info->sectorsize < PAGE_SIZE) {
541 if (!IS_ALIGNED(start, PAGE_SIZE) ||
542 !IS_ALIGNED(end + 1, PAGE_SIZE))
547 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
550 if (inode->defrag_compress)
552 /* bad compression ratios */
553 if (inode->flags & BTRFS_INODE_NOCOMPRESS)
555 if (btrfs_test_opt(fs_info, COMPRESS) ||
556 inode->flags & BTRFS_INODE_COMPRESS ||
557 inode->prop_compress)
558 return btrfs_compress_heuristic(&inode->vfs_inode, start, end);
562 static inline void inode_should_defrag(struct btrfs_inode *inode,
563 u64 start, u64 end, u64 num_bytes, u32 small_write)
565 /* If this is a small write inside eof, kick off a defrag */
566 if (num_bytes < small_write &&
567 (start > 0 || end + 1 < inode->disk_i_size))
568 btrfs_add_inode_defrag(NULL, inode, small_write);
572 * we create compressed extents in two phases. The first
573 * phase compresses a range of pages that have already been
574 * locked (both pages and state bits are locked).
576 * This is done inside an ordered work queue, and the compression
577 * is spread across many cpus. The actual IO submission is step
578 * two, and the ordered work queue takes care of making sure that
579 * happens in the same order things were put onto the queue by
580 * writepages and friends.
582 * If this code finds it can't get good compression, it puts an
583 * entry onto the work queue to write the uncompressed bytes. This
584 * makes sure that both compressed inodes and uncompressed inodes
585 * are written in the same order that the flusher thread sent them
588 static noinline int compress_file_range(struct async_chunk *async_chunk)
590 struct inode *inode = async_chunk->inode;
591 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
592 u64 blocksize = fs_info->sectorsize;
593 u64 start = async_chunk->start;
594 u64 end = async_chunk->end;
598 struct page **pages = NULL;
599 unsigned long nr_pages;
600 unsigned long total_compressed = 0;
601 unsigned long total_in = 0;
604 int compress_type = fs_info->compress_type;
605 int compressed_extents = 0;
608 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
612 * We need to save i_size before now because it could change in between
613 * us evaluating the size and assigning it. This is because we lock and
614 * unlock the page in truncate and fallocate, and then modify the i_size
617 * The barriers are to emulate READ_ONCE, remove that once i_size_read
621 i_size = i_size_read(inode);
623 actual_end = min_t(u64, i_size, end + 1);
626 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
627 BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
628 nr_pages = min_t(unsigned long, nr_pages,
629 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
632 * we don't want to send crud past the end of i_size through
633 * compression, that's just a waste of CPU time. So, if the
634 * end of the file is before the start of our current
635 * requested range of bytes, we bail out to the uncompressed
636 * cleanup code that can deal with all of this.
638 * It isn't really the fastest way to fix things, but this is a
639 * very uncommon corner.
641 if (actual_end <= start)
642 goto cleanup_and_bail_uncompressed;
644 total_compressed = actual_end - start;
647 * Skip compression for a small file range(<=blocksize) that
648 * isn't an inline extent, since it doesn't save disk space at all.
650 if (total_compressed <= blocksize &&
651 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
652 goto cleanup_and_bail_uncompressed;
655 * For subpage case, we require full page alignment for the sector
657 * Thus we must also check against @actual_end, not just @end.
659 if (blocksize < PAGE_SIZE) {
660 if (!IS_ALIGNED(start, PAGE_SIZE) ||
661 !IS_ALIGNED(round_up(actual_end, blocksize), PAGE_SIZE))
662 goto cleanup_and_bail_uncompressed;
665 total_compressed = min_t(unsigned long, total_compressed,
666 BTRFS_MAX_UNCOMPRESSED);
671 * we do compression for mount -o compress and when the
672 * inode has not been flagged as nocompress. This flag can
673 * change at any time if we discover bad compression ratios.
675 if (inode_need_compress(BTRFS_I(inode), start, end)) {
677 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
679 /* just bail out to the uncompressed code */
684 if (BTRFS_I(inode)->defrag_compress)
685 compress_type = BTRFS_I(inode)->defrag_compress;
686 else if (BTRFS_I(inode)->prop_compress)
687 compress_type = BTRFS_I(inode)->prop_compress;
690 * we need to call clear_page_dirty_for_io on each
691 * page in the range. Otherwise applications with the file
692 * mmap'd can wander in and change the page contents while
693 * we are compressing them.
695 * If the compression fails for any reason, we set the pages
696 * dirty again later on.
698 * Note that the remaining part is redirtied, the start pointer
699 * has moved, the end is the original one.
702 extent_range_clear_dirty_for_io(inode, start, end);
706 /* Compression level is applied here and only here */
707 ret = btrfs_compress_pages(
708 compress_type | (fs_info->compress_level << 4),
709 inode->i_mapping, start,
716 unsigned long offset = offset_in_page(total_compressed);
717 struct page *page = pages[nr_pages - 1];
719 /* zero the tail end of the last page, we might be
720 * sending it down to disk
723 memzero_page(page, offset, PAGE_SIZE - offset);
729 * Check cow_file_range() for why we don't even try to create inline
730 * extent for subpage case.
732 if (start == 0 && fs_info->sectorsize == PAGE_SIZE) {
733 /* lets try to make an inline extent */
734 if (ret || total_in < actual_end) {
735 /* we didn't compress the entire range, try
736 * to make an uncompressed inline extent.
738 ret = cow_file_range_inline(BTRFS_I(inode), start, end,
739 0, BTRFS_COMPRESS_NONE,
742 /* try making a compressed inline extent */
743 ret = cow_file_range_inline(BTRFS_I(inode), start, end,
745 compress_type, pages);
748 unsigned long clear_flags = EXTENT_DELALLOC |
749 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
750 EXTENT_DO_ACCOUNTING;
751 unsigned long page_error_op;
753 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
756 * inline extent creation worked or returned error,
757 * we don't need to create any more async work items.
758 * Unlock and free up our temp pages.
760 * We use DO_ACCOUNTING here because we need the
761 * delalloc_release_metadata to be done _after_ we drop
762 * our outstanding extent for clearing delalloc for this
765 extent_clear_unlock_delalloc(BTRFS_I(inode), start, end,
769 PAGE_START_WRITEBACK |
774 * Ensure we only free the compressed pages if we have
775 * them allocated, as we can still reach here with
776 * inode_need_compress() == false.
779 for (i = 0; i < nr_pages; i++) {
780 WARN_ON(pages[i]->mapping);
791 * we aren't doing an inline extent round the compressed size
792 * up to a block size boundary so the allocator does sane
795 total_compressed = ALIGN(total_compressed, blocksize);
798 * one last check to make sure the compression is really a
799 * win, compare the page count read with the blocks on disk,
800 * compression must free at least one sector size
802 total_in = round_up(total_in, fs_info->sectorsize);
803 if (total_compressed + blocksize <= total_in) {
804 compressed_extents++;
807 * The async work queues will take care of doing actual
808 * allocation on disk for these compressed pages, and
809 * will submit them to the elevator.
811 add_async_extent(async_chunk, start, total_in,
812 total_compressed, pages, nr_pages,
815 if (start + total_in < end) {
821 return compressed_extents;
826 * the compression code ran but failed to make things smaller,
827 * free any pages it allocated and our page pointer array
829 for (i = 0; i < nr_pages; i++) {
830 WARN_ON(pages[i]->mapping);
835 total_compressed = 0;
838 /* flag the file so we don't compress in the future */
839 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
840 !(BTRFS_I(inode)->prop_compress)) {
841 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
844 cleanup_and_bail_uncompressed:
846 * No compression, but we still need to write the pages in the file
847 * we've been given so far. redirty the locked page if it corresponds
848 * to our extent and set things up for the async work queue to run
849 * cow_file_range to do the normal delalloc dance.
851 if (async_chunk->locked_page &&
852 (page_offset(async_chunk->locked_page) >= start &&
853 page_offset(async_chunk->locked_page)) <= end) {
854 __set_page_dirty_nobuffers(async_chunk->locked_page);
855 /* unlocked later on in the async handlers */
859 extent_range_redirty_for_io(inode, start, end);
860 add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
861 BTRFS_COMPRESS_NONE);
862 compressed_extents++;
864 return compressed_extents;
867 static void free_async_extent_pages(struct async_extent *async_extent)
871 if (!async_extent->pages)
874 for (i = 0; i < async_extent->nr_pages; i++) {
875 WARN_ON(async_extent->pages[i]->mapping);
876 put_page(async_extent->pages[i]);
878 kfree(async_extent->pages);
879 async_extent->nr_pages = 0;
880 async_extent->pages = NULL;
883 static int submit_uncompressed_range(struct btrfs_inode *inode,
884 struct async_extent *async_extent,
885 struct page *locked_page)
887 u64 start = async_extent->start;
888 u64 end = async_extent->start + async_extent->ram_size - 1;
889 unsigned long nr_written = 0;
890 int page_started = 0;
894 * Call cow_file_range() to run the delalloc range directly, since we
895 * won't go to NOCOW or async path again.
897 * Also we call cow_file_range() with @unlock_page == 0, so that we
898 * can directly submit them without interruption.
900 ret = cow_file_range(inode, locked_page, start, end, &page_started,
902 /* Inline extent inserted, page gets unlocked and everything is done */
909 unlock_page(locked_page);
913 ret = extent_write_locked_range(&inode->vfs_inode, start, end);
914 /* All pages will be unlocked, including @locked_page */
920 static int submit_one_async_extent(struct btrfs_inode *inode,
921 struct async_chunk *async_chunk,
922 struct async_extent *async_extent,
925 struct extent_io_tree *io_tree = &inode->io_tree;
926 struct btrfs_root *root = inode->root;
927 struct btrfs_fs_info *fs_info = root->fs_info;
928 struct btrfs_key ins;
929 struct page *locked_page = NULL;
930 struct extent_map *em;
932 u64 start = async_extent->start;
933 u64 end = async_extent->start + async_extent->ram_size - 1;
936 * If async_chunk->locked_page is in the async_extent range, we need to
939 if (async_chunk->locked_page) {
940 u64 locked_page_start = page_offset(async_chunk->locked_page);
941 u64 locked_page_end = locked_page_start + PAGE_SIZE - 1;
943 if (!(start >= locked_page_end || end <= locked_page_start))
944 locked_page = async_chunk->locked_page;
946 lock_extent(io_tree, start, end);
948 /* We have fall back to uncompressed write */
949 if (!async_extent->pages)
950 return submit_uncompressed_range(inode, async_extent, locked_page);
952 ret = btrfs_reserve_extent(root, async_extent->ram_size,
953 async_extent->compressed_size,
954 async_extent->compressed_size,
955 0, *alloc_hint, &ins, 1, 1);
957 free_async_extent_pages(async_extent);
959 * Here we used to try again by going back to non-compressed
960 * path for ENOSPC. But we can't reserve space even for
961 * compressed size, how could it work for uncompressed size
962 * which requires larger size? So here we directly go error
968 /* Here we're doing allocation and writeback of the compressed pages */
969 em = create_io_em(inode, start,
970 async_extent->ram_size, /* len */
971 start, /* orig_start */
972 ins.objectid, /* block_start */
973 ins.offset, /* block_len */
974 ins.offset, /* orig_block_len */
975 async_extent->ram_size, /* ram_bytes */
976 async_extent->compress_type,
977 BTRFS_ORDERED_COMPRESSED);
980 goto out_free_reserve;
984 ret = btrfs_add_ordered_extent_compress(inode, start, /* file_offset */
985 ins.objectid, /* disk_bytenr */
986 async_extent->ram_size, /* num_bytes */
987 ins.offset, /* disk_num_bytes */
988 async_extent->compress_type);
990 btrfs_drop_extent_cache(inode, start, end, 0);
991 goto out_free_reserve;
993 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
995 /* Clear dirty, set writeback and unlock the pages. */
996 extent_clear_unlock_delalloc(inode, start, end,
997 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
998 PAGE_UNLOCK | PAGE_START_WRITEBACK);
999 if (btrfs_submit_compressed_write(inode, start, /* file_offset */
1000 async_extent->ram_size, /* num_bytes */
1001 ins.objectid, /* disk_bytenr */
1002 ins.offset, /* compressed_len */
1003 async_extent->pages, /* compressed_pages */
1004 async_extent->nr_pages,
1005 async_chunk->write_flags,
1006 async_chunk->blkcg_css)) {
1007 const u64 start = async_extent->start;
1008 const u64 end = start + async_extent->ram_size - 1;
1010 btrfs_writepage_endio_finish_ordered(inode, NULL, start, end, 0);
1012 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
1013 PAGE_END_WRITEBACK | PAGE_SET_ERROR);
1014 free_async_extent_pages(async_extent);
1016 *alloc_hint = ins.objectid + ins.offset;
1017 kfree(async_extent);
1021 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1022 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1024 extent_clear_unlock_delalloc(inode, start, end,
1025 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
1026 EXTENT_DELALLOC_NEW |
1027 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
1028 PAGE_UNLOCK | PAGE_START_WRITEBACK |
1029 PAGE_END_WRITEBACK | PAGE_SET_ERROR);
1030 free_async_extent_pages(async_extent);
1031 kfree(async_extent);
1036 * Phase two of compressed writeback. This is the ordered portion of the code,
1037 * which only gets called in the order the work was queued. We walk all the
1038 * async extents created by compress_file_range and send them down to the disk.
1040 static noinline void submit_compressed_extents(struct async_chunk *async_chunk)
1042 struct btrfs_inode *inode = BTRFS_I(async_chunk->inode);
1043 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1044 struct async_extent *async_extent;
1048 while (!list_empty(&async_chunk->extents)) {
1052 async_extent = list_entry(async_chunk->extents.next,
1053 struct async_extent, list);
1054 list_del(&async_extent->list);
1055 extent_start = async_extent->start;
1056 ram_size = async_extent->ram_size;
1058 ret = submit_one_async_extent(inode, async_chunk, async_extent,
1060 btrfs_debug(fs_info,
1061 "async extent submission failed root=%lld inode=%llu start=%llu len=%llu ret=%d",
1062 inode->root->root_key.objectid,
1063 btrfs_ino(inode), extent_start, ram_size, ret);
1067 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
1070 struct extent_map_tree *em_tree = &inode->extent_tree;
1071 struct extent_map *em;
1074 read_lock(&em_tree->lock);
1075 em = search_extent_mapping(em_tree, start, num_bytes);
1078 * if block start isn't an actual block number then find the
1079 * first block in this inode and use that as a hint. If that
1080 * block is also bogus then just don't worry about it.
1082 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1083 free_extent_map(em);
1084 em = search_extent_mapping(em_tree, 0, 0);
1085 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
1086 alloc_hint = em->block_start;
1088 free_extent_map(em);
1090 alloc_hint = em->block_start;
1091 free_extent_map(em);
1094 read_unlock(&em_tree->lock);
1100 * when extent_io.c finds a delayed allocation range in the file,
1101 * the call backs end up in this code. The basic idea is to
1102 * allocate extents on disk for the range, and create ordered data structs
1103 * in ram to track those extents.
1105 * locked_page is the page that writepage had locked already. We use
1106 * it to make sure we don't do extra locks or unlocks.
1108 * *page_started is set to one if we unlock locked_page and do everything
1109 * required to start IO on it. It may be clean and already done with
1110 * IO when we return.
1112 static noinline int cow_file_range(struct btrfs_inode *inode,
1113 struct page *locked_page,
1114 u64 start, u64 end, int *page_started,
1115 unsigned long *nr_written, int unlock)
1117 struct btrfs_root *root = inode->root;
1118 struct btrfs_fs_info *fs_info = root->fs_info;
1121 unsigned long ram_size;
1122 u64 cur_alloc_size = 0;
1124 u64 blocksize = fs_info->sectorsize;
1125 struct btrfs_key ins;
1126 struct extent_map *em;
1127 unsigned clear_bits;
1128 unsigned long page_ops;
1129 bool extent_reserved = false;
1132 if (btrfs_is_free_space_inode(inode)) {
1138 num_bytes = ALIGN(end - start + 1, blocksize);
1139 num_bytes = max(blocksize, num_bytes);
1140 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1142 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1145 * Due to the page size limit, for subpage we can only trigger the
1146 * writeback for the dirty sectors of page, that means data writeback
1147 * is doing more writeback than what we want.
1149 * This is especially unexpected for some call sites like fallocate,
1150 * where we only increase i_size after everything is done.
1151 * This means we can trigger inline extent even if we didn't want to.
1152 * So here we skip inline extent creation completely.
1154 if (start == 0 && fs_info->sectorsize == PAGE_SIZE) {
1155 /* lets try to make an inline extent */
1156 ret = cow_file_range_inline(inode, start, end, 0,
1157 BTRFS_COMPRESS_NONE, NULL);
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, ins.objectid,
1238 ram_size, cur_alloc_size,
1239 BTRFS_ORDERED_REGULAR);
1241 goto out_drop_extent_cache;
1243 if (btrfs_is_data_reloc_root(root)) {
1244 ret = btrfs_reloc_clone_csums(inode, start,
1247 * Only drop cache here, and process as normal.
1249 * We must not allow extent_clear_unlock_delalloc()
1250 * at out_unlock label to free meta of this ordered
1251 * extent, as its meta should be freed by
1252 * btrfs_finish_ordered_io().
1254 * So we must continue until @start is increased to
1255 * skip current ordered extent.
1258 btrfs_drop_extent_cache(inode, start,
1259 start + ram_size - 1, 0);
1262 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1265 * We're not doing compressed IO, don't unlock the first page
1266 * (which the caller expects to stay locked), don't clear any
1267 * dirty bits and don't set any writeback bits
1269 * Do set the Ordered (Private2) bit so we know this page was
1270 * properly setup for writepage.
1272 page_ops = unlock ? PAGE_UNLOCK : 0;
1273 page_ops |= PAGE_SET_ORDERED;
1275 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1277 EXTENT_LOCKED | EXTENT_DELALLOC,
1279 if (num_bytes < cur_alloc_size)
1282 num_bytes -= cur_alloc_size;
1283 alloc_hint = ins.objectid + ins.offset;
1284 start += cur_alloc_size;
1285 extent_reserved = false;
1288 * btrfs_reloc_clone_csums() error, since start is increased
1289 * extent_clear_unlock_delalloc() at out_unlock label won't
1290 * free metadata of current ordered extent, we're OK to exit.
1298 out_drop_extent_cache:
1299 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1301 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1302 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1304 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1305 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1306 page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK;
1308 * If we reserved an extent for our delalloc range (or a subrange) and
1309 * failed to create the respective ordered extent, then it means that
1310 * when we reserved the extent we decremented the extent's size from
1311 * the data space_info's bytes_may_use counter and incremented the
1312 * space_info's bytes_reserved counter by the same amount. We must make
1313 * sure extent_clear_unlock_delalloc() does not try to decrement again
1314 * the data space_info's bytes_may_use counter, therefore we do not pass
1315 * it the flag EXTENT_CLEAR_DATA_RESV.
1317 if (extent_reserved) {
1318 extent_clear_unlock_delalloc(inode, start,
1319 start + cur_alloc_size - 1,
1323 start += cur_alloc_size;
1327 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1328 clear_bits | EXTENT_CLEAR_DATA_RESV,
1334 * work queue call back to started compression on a file and pages
1336 static noinline void async_cow_start(struct btrfs_work *work)
1338 struct async_chunk *async_chunk;
1339 int compressed_extents;
1341 async_chunk = container_of(work, struct async_chunk, work);
1343 compressed_extents = compress_file_range(async_chunk);
1344 if (compressed_extents == 0) {
1345 btrfs_add_delayed_iput(async_chunk->inode);
1346 async_chunk->inode = NULL;
1351 * work queue call back to submit previously compressed pages
1353 static noinline void async_cow_submit(struct btrfs_work *work)
1355 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1357 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1358 unsigned long nr_pages;
1360 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1364 * ->inode could be NULL if async_chunk_start has failed to compress,
1365 * in which case we don't have anything to submit, yet we need to
1366 * always adjust ->async_delalloc_pages as its paired with the init
1367 * happening in cow_file_range_async
1369 if (async_chunk->inode)
1370 submit_compressed_extents(async_chunk);
1372 /* atomic_sub_return implies a barrier */
1373 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1375 cond_wake_up_nomb(&fs_info->async_submit_wait);
1378 static noinline void async_cow_free(struct btrfs_work *work)
1380 struct async_chunk *async_chunk;
1381 struct async_cow *async_cow;
1383 async_chunk = container_of(work, struct async_chunk, work);
1384 if (async_chunk->inode)
1385 btrfs_add_delayed_iput(async_chunk->inode);
1386 if (async_chunk->blkcg_css)
1387 css_put(async_chunk->blkcg_css);
1389 async_cow = async_chunk->async_cow;
1390 if (atomic_dec_and_test(&async_cow->num_chunks))
1394 static int cow_file_range_async(struct btrfs_inode *inode,
1395 struct writeback_control *wbc,
1396 struct page *locked_page,
1397 u64 start, u64 end, int *page_started,
1398 unsigned long *nr_written)
1400 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1401 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1402 struct async_cow *ctx;
1403 struct async_chunk *async_chunk;
1404 unsigned long nr_pages;
1406 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1408 bool should_compress;
1410 const unsigned int write_flags = wbc_to_write_flags(wbc);
1412 unlock_extent(&inode->io_tree, start, end);
1414 if (inode->flags & BTRFS_INODE_NOCOMPRESS &&
1415 !btrfs_test_opt(fs_info, FORCE_COMPRESS)) {
1417 should_compress = false;
1419 should_compress = true;
1422 nofs_flag = memalloc_nofs_save();
1423 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1424 memalloc_nofs_restore(nofs_flag);
1427 unsigned clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC |
1428 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1429 EXTENT_DO_ACCOUNTING;
1430 unsigned long page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK |
1431 PAGE_END_WRITEBACK | PAGE_SET_ERROR;
1433 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1434 clear_bits, page_ops);
1438 async_chunk = ctx->chunks;
1439 atomic_set(&ctx->num_chunks, num_chunks);
1441 for (i = 0; i < num_chunks; i++) {
1442 if (should_compress)
1443 cur_end = min(end, start + SZ_512K - 1);
1448 * igrab is called higher up in the call chain, take only the
1449 * lightweight reference for the callback lifetime
1451 ihold(&inode->vfs_inode);
1452 async_chunk[i].async_cow = ctx;
1453 async_chunk[i].inode = &inode->vfs_inode;
1454 async_chunk[i].start = start;
1455 async_chunk[i].end = cur_end;
1456 async_chunk[i].write_flags = write_flags;
1457 INIT_LIST_HEAD(&async_chunk[i].extents);
1460 * The locked_page comes all the way from writepage and its
1461 * the original page we were actually given. As we spread
1462 * this large delalloc region across multiple async_chunk
1463 * structs, only the first struct needs a pointer to locked_page
1465 * This way we don't need racey decisions about who is supposed
1470 * Depending on the compressibility, the pages might or
1471 * might not go through async. We want all of them to
1472 * be accounted against wbc once. Let's do it here
1473 * before the paths diverge. wbc accounting is used
1474 * only for foreign writeback detection and doesn't
1475 * need full accuracy. Just account the whole thing
1476 * against the first page.
1478 wbc_account_cgroup_owner(wbc, locked_page,
1480 async_chunk[i].locked_page = locked_page;
1483 async_chunk[i].locked_page = NULL;
1486 if (blkcg_css != blkcg_root_css) {
1488 async_chunk[i].blkcg_css = blkcg_css;
1490 async_chunk[i].blkcg_css = NULL;
1493 btrfs_init_work(&async_chunk[i].work, async_cow_start,
1494 async_cow_submit, async_cow_free);
1496 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1497 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1499 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1501 *nr_written += nr_pages;
1502 start = cur_end + 1;
1508 static noinline int run_delalloc_zoned(struct btrfs_inode *inode,
1509 struct page *locked_page, u64 start,
1510 u64 end, int *page_started,
1511 unsigned long *nr_written)
1515 ret = cow_file_range(inode, locked_page, start, end, page_started,
1523 __set_page_dirty_nobuffers(locked_page);
1524 account_page_redirty(locked_page);
1525 extent_write_locked_range(&inode->vfs_inode, start, end);
1531 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1532 u64 bytenr, u64 num_bytes)
1534 struct btrfs_root *csum_root = btrfs_csum_root(fs_info, bytenr);
1535 struct btrfs_ordered_sum *sums;
1539 ret = btrfs_lookup_csums_range(csum_root, bytenr,
1540 bytenr + num_bytes - 1, &list, 0);
1541 if (ret == 0 && list_empty(&list))
1544 while (!list_empty(&list)) {
1545 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1546 list_del(&sums->list);
1554 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1555 const u64 start, const u64 end,
1556 int *page_started, unsigned long *nr_written)
1558 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1559 const bool is_reloc_ino = btrfs_is_data_reloc_root(inode->root);
1560 const u64 range_bytes = end + 1 - start;
1561 struct extent_io_tree *io_tree = &inode->io_tree;
1562 u64 range_start = start;
1566 * If EXTENT_NORESERVE is set it means that when the buffered write was
1567 * made we had not enough available data space and therefore we did not
1568 * reserve data space for it, since we though we could do NOCOW for the
1569 * respective file range (either there is prealloc extent or the inode
1570 * has the NOCOW bit set).
1572 * However when we need to fallback to COW mode (because for example the
1573 * block group for the corresponding extent was turned to RO mode by a
1574 * scrub or relocation) we need to do the following:
1576 * 1) We increment the bytes_may_use counter of the data space info.
1577 * If COW succeeds, it allocates a new data extent and after doing
1578 * that it decrements the space info's bytes_may_use counter and
1579 * increments its bytes_reserved counter by the same amount (we do
1580 * this at btrfs_add_reserved_bytes()). So we need to increment the
1581 * bytes_may_use counter to compensate (when space is reserved at
1582 * buffered write time, the bytes_may_use counter is incremented);
1584 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1585 * that if the COW path fails for any reason, it decrements (through
1586 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1587 * data space info, which we incremented in the step above.
1589 * If we need to fallback to cow and the inode corresponds to a free
1590 * space cache inode or an inode of the data relocation tree, we must
1591 * also increment bytes_may_use of the data space_info for the same
1592 * reason. Space caches and relocated data extents always get a prealloc
1593 * extent for them, however scrub or balance may have set the block
1594 * group that contains that extent to RO mode and therefore force COW
1595 * when starting writeback.
1597 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1598 EXTENT_NORESERVE, 0);
1599 if (count > 0 || is_space_ino || is_reloc_ino) {
1601 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1602 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1604 if (is_space_ino || is_reloc_ino)
1605 bytes = range_bytes;
1607 spin_lock(&sinfo->lock);
1608 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1609 spin_unlock(&sinfo->lock);
1612 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1616 return cow_file_range(inode, locked_page, start, end, page_started,
1621 * when nowcow writeback call back. This checks for snapshots or COW copies
1622 * of the extents that exist in the file, and COWs the file as required.
1624 * If no cow copies or snapshots exist, we write directly to the existing
1627 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
1628 struct page *locked_page,
1629 const u64 start, const u64 end,
1631 unsigned long *nr_written)
1633 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1634 struct btrfs_root *root = inode->root;
1635 struct btrfs_path *path;
1636 u64 cow_start = (u64)-1;
1637 u64 cur_offset = start;
1639 bool check_prev = true;
1640 const bool freespace_inode = btrfs_is_free_space_inode(inode);
1641 u64 ino = btrfs_ino(inode);
1643 u64 disk_bytenr = 0;
1644 const bool force = inode->flags & BTRFS_INODE_NODATACOW;
1646 path = btrfs_alloc_path();
1648 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1649 EXTENT_LOCKED | EXTENT_DELALLOC |
1650 EXTENT_DO_ACCOUNTING |
1651 EXTENT_DEFRAG, PAGE_UNLOCK |
1652 PAGE_START_WRITEBACK |
1653 PAGE_END_WRITEBACK);
1658 struct btrfs_key found_key;
1659 struct btrfs_file_extent_item *fi;
1660 struct extent_buffer *leaf;
1670 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1676 * If there is no extent for our range when doing the initial
1677 * search, then go back to the previous slot as it will be the
1678 * one containing the search offset
1680 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1681 leaf = path->nodes[0];
1682 btrfs_item_key_to_cpu(leaf, &found_key,
1683 path->slots[0] - 1);
1684 if (found_key.objectid == ino &&
1685 found_key.type == BTRFS_EXTENT_DATA_KEY)
1690 /* Go to next leaf if we have exhausted the current one */
1691 leaf = path->nodes[0];
1692 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1693 ret = btrfs_next_leaf(root, path);
1695 if (cow_start != (u64)-1)
1696 cur_offset = cow_start;
1701 leaf = path->nodes[0];
1704 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1706 /* Didn't find anything for our INO */
1707 if (found_key.objectid > ino)
1710 * Keep searching until we find an EXTENT_ITEM or there are no
1711 * more extents for this inode
1713 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1714 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1719 /* Found key is not EXTENT_DATA_KEY or starts after req range */
1720 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1721 found_key.offset > end)
1725 * If the found extent starts after requested offset, then
1726 * adjust extent_end to be right before this extent begins
1728 if (found_key.offset > cur_offset) {
1729 extent_end = found_key.offset;
1735 * Found extent which begins before our range and potentially
1738 fi = btrfs_item_ptr(leaf, path->slots[0],
1739 struct btrfs_file_extent_item);
1740 extent_type = btrfs_file_extent_type(leaf, fi);
1742 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1743 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1744 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1745 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1746 extent_offset = btrfs_file_extent_offset(leaf, fi);
1747 extent_end = found_key.offset +
1748 btrfs_file_extent_num_bytes(leaf, fi);
1750 btrfs_file_extent_disk_num_bytes(leaf, fi);
1752 * If the extent we got ends before our current offset,
1753 * skip to the next extent.
1755 if (extent_end <= cur_offset) {
1760 if (disk_bytenr == 0)
1762 /* Skip compressed/encrypted/encoded extents */
1763 if (btrfs_file_extent_compression(leaf, fi) ||
1764 btrfs_file_extent_encryption(leaf, fi) ||
1765 btrfs_file_extent_other_encoding(leaf, fi))
1768 * If extent is created before the last volume's snapshot
1769 * this implies the extent is shared, hence we can't do
1770 * nocow. This is the same check as in
1771 * btrfs_cross_ref_exist but without calling
1772 * btrfs_search_slot.
1774 if (!freespace_inode &&
1775 btrfs_file_extent_generation(leaf, fi) <=
1776 btrfs_root_last_snapshot(&root->root_item))
1778 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1782 * The following checks can be expensive, as they need to
1783 * take other locks and do btree or rbtree searches, so
1784 * release the path to avoid blocking other tasks for too
1787 btrfs_release_path(path);
1789 ret = btrfs_cross_ref_exist(root, ino,
1791 extent_offset, disk_bytenr, false);
1794 * ret could be -EIO if the above fails to read
1798 if (cow_start != (u64)-1)
1799 cur_offset = cow_start;
1803 WARN_ON_ONCE(freespace_inode);
1806 disk_bytenr += extent_offset;
1807 disk_bytenr += cur_offset - found_key.offset;
1808 num_bytes = min(end + 1, extent_end) - cur_offset;
1810 * If there are pending snapshots for this root, we
1811 * fall into common COW way
1813 if (!freespace_inode && atomic_read(&root->snapshot_force_cow))
1816 * force cow if csum exists in the range.
1817 * this ensure that csum for a given extent are
1818 * either valid or do not exist.
1820 ret = csum_exist_in_range(fs_info, disk_bytenr,
1824 * ret could be -EIO if the above fails to read
1828 if (cow_start != (u64)-1)
1829 cur_offset = cow_start;
1832 WARN_ON_ONCE(freespace_inode);
1835 /* If the extent's block group is RO, we must COW */
1836 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr))
1839 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1840 extent_end = found_key.offset + ram_bytes;
1841 extent_end = ALIGN(extent_end, fs_info->sectorsize);
1842 /* Skip extents outside of our requested range */
1843 if (extent_end <= start) {
1848 /* If this triggers then we have a memory corruption */
1853 * If nocow is false then record the beginning of the range
1854 * that needs to be COWed
1857 if (cow_start == (u64)-1)
1858 cow_start = cur_offset;
1859 cur_offset = extent_end;
1860 if (cur_offset > end)
1862 if (!path->nodes[0])
1869 * COW range from cow_start to found_key.offset - 1. As the key
1870 * will contain the beginning of the first extent that can be
1871 * NOCOW, following one which needs to be COW'ed
1873 if (cow_start != (u64)-1) {
1874 ret = fallback_to_cow(inode, locked_page,
1875 cow_start, found_key.offset - 1,
1876 page_started, nr_written);
1879 cow_start = (u64)-1;
1882 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1883 u64 orig_start = found_key.offset - extent_offset;
1884 struct extent_map *em;
1886 em = create_io_em(inode, cur_offset, num_bytes,
1888 disk_bytenr, /* block_start */
1889 num_bytes, /* block_len */
1890 disk_num_bytes, /* orig_block_len */
1891 ram_bytes, BTRFS_COMPRESS_NONE,
1892 BTRFS_ORDERED_PREALLOC);
1897 free_extent_map(em);
1898 ret = btrfs_add_ordered_extent(inode, cur_offset,
1899 disk_bytenr, num_bytes,
1901 BTRFS_ORDERED_PREALLOC);
1903 btrfs_drop_extent_cache(inode, cur_offset,
1904 cur_offset + num_bytes - 1,
1909 ret = btrfs_add_ordered_extent(inode, cur_offset,
1910 disk_bytenr, num_bytes,
1912 BTRFS_ORDERED_NOCOW);
1918 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1921 if (btrfs_is_data_reloc_root(root))
1923 * Error handled later, as we must prevent
1924 * extent_clear_unlock_delalloc() in error handler
1925 * from freeing metadata of created ordered extent.
1927 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1930 extent_clear_unlock_delalloc(inode, cur_offset,
1931 cur_offset + num_bytes - 1,
1932 locked_page, EXTENT_LOCKED |
1934 EXTENT_CLEAR_DATA_RESV,
1935 PAGE_UNLOCK | PAGE_SET_ORDERED);
1937 cur_offset = extent_end;
1940 * btrfs_reloc_clone_csums() error, now we're OK to call error
1941 * handler, as metadata for created ordered extent will only
1942 * be freed by btrfs_finish_ordered_io().
1946 if (cur_offset > end)
1949 btrfs_release_path(path);
1951 if (cur_offset <= end && cow_start == (u64)-1)
1952 cow_start = cur_offset;
1954 if (cow_start != (u64)-1) {
1956 ret = fallback_to_cow(inode, locked_page, cow_start, end,
1957 page_started, nr_written);
1964 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1966 if (ret && cur_offset < end)
1967 extent_clear_unlock_delalloc(inode, cur_offset, end,
1968 locked_page, EXTENT_LOCKED |
1969 EXTENT_DELALLOC | EXTENT_DEFRAG |
1970 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1971 PAGE_START_WRITEBACK |
1972 PAGE_END_WRITEBACK);
1973 btrfs_free_path(path);
1977 static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end)
1979 if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) {
1980 if (inode->defrag_bytes &&
1981 test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG,
1990 * Function to process delayed allocation (create CoW) for ranges which are
1991 * being touched for the first time.
1993 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
1994 u64 start, u64 end, int *page_started, unsigned long *nr_written,
1995 struct writeback_control *wbc)
1998 const bool zoned = btrfs_is_zoned(inode->root->fs_info);
2001 * The range must cover part of the @locked_page, or the returned
2002 * @page_started can confuse the caller.
2004 ASSERT(!(end <= page_offset(locked_page) ||
2005 start >= page_offset(locked_page) + PAGE_SIZE));
2007 if (should_nocow(inode, start, end)) {
2009 * Normally on a zoned device we're only doing COW writes, but
2010 * in case of relocation on a zoned filesystem we have taken
2011 * precaution, that we're only writing sequentially. It's safe
2012 * to use run_delalloc_nocow() here, like for regular
2013 * preallocated inodes.
2016 (zoned && btrfs_is_data_reloc_root(inode->root)));
2017 ret = run_delalloc_nocow(inode, locked_page, start, end,
2018 page_started, nr_written);
2019 } else if (!inode_can_compress(inode) ||
2020 !inode_need_compress(inode, start, end)) {
2022 ret = run_delalloc_zoned(inode, locked_page, start, end,
2023 page_started, nr_written);
2025 ret = cow_file_range(inode, locked_page, start, end,
2026 page_started, nr_written, 1);
2028 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
2029 ret = cow_file_range_async(inode, wbc, locked_page, start, end,
2030 page_started, nr_written);
2034 btrfs_cleanup_ordered_extents(inode, locked_page, start,
2039 void btrfs_split_delalloc_extent(struct inode *inode,
2040 struct extent_state *orig, u64 split)
2044 /* not delalloc, ignore it */
2045 if (!(orig->state & EXTENT_DELALLOC))
2048 size = orig->end - orig->start + 1;
2049 if (size > BTRFS_MAX_EXTENT_SIZE) {
2054 * See the explanation in btrfs_merge_delalloc_extent, the same
2055 * applies here, just in reverse.
2057 new_size = orig->end - split + 1;
2058 num_extents = count_max_extents(new_size);
2059 new_size = split - orig->start;
2060 num_extents += count_max_extents(new_size);
2061 if (count_max_extents(size) >= num_extents)
2065 spin_lock(&BTRFS_I(inode)->lock);
2066 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
2067 spin_unlock(&BTRFS_I(inode)->lock);
2071 * Handle merged delayed allocation extents so we can keep track of new extents
2072 * that are just merged onto old extents, such as when we are doing sequential
2073 * writes, so we can properly account for the metadata space we'll need.
2075 void btrfs_merge_delalloc_extent(struct inode *inode, struct extent_state *new,
2076 struct extent_state *other)
2078 u64 new_size, old_size;
2081 /* not delalloc, ignore it */
2082 if (!(other->state & EXTENT_DELALLOC))
2085 if (new->start > other->start)
2086 new_size = new->end - other->start + 1;
2088 new_size = other->end - new->start + 1;
2090 /* we're not bigger than the max, unreserve the space and go */
2091 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
2092 spin_lock(&BTRFS_I(inode)->lock);
2093 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
2094 spin_unlock(&BTRFS_I(inode)->lock);
2099 * We have to add up either side to figure out how many extents were
2100 * accounted for before we merged into one big extent. If the number of
2101 * extents we accounted for is <= the amount we need for the new range
2102 * then we can return, otherwise drop. Think of it like this
2106 * So we've grown the extent by a MAX_SIZE extent, this would mean we
2107 * need 2 outstanding extents, on one side we have 1 and the other side
2108 * we have 1 so they are == and we can return. But in this case
2110 * [MAX_SIZE+4k][MAX_SIZE+4k]
2112 * Each range on their own accounts for 2 extents, but merged together
2113 * they are only 3 extents worth of accounting, so we need to drop in
2116 old_size = other->end - other->start + 1;
2117 num_extents = count_max_extents(old_size);
2118 old_size = new->end - new->start + 1;
2119 num_extents += count_max_extents(old_size);
2120 if (count_max_extents(new_size) >= num_extents)
2123 spin_lock(&BTRFS_I(inode)->lock);
2124 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
2125 spin_unlock(&BTRFS_I(inode)->lock);
2128 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
2129 struct inode *inode)
2131 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2133 spin_lock(&root->delalloc_lock);
2134 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
2135 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
2136 &root->delalloc_inodes);
2137 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2138 &BTRFS_I(inode)->runtime_flags);
2139 root->nr_delalloc_inodes++;
2140 if (root->nr_delalloc_inodes == 1) {
2141 spin_lock(&fs_info->delalloc_root_lock);
2142 BUG_ON(!list_empty(&root->delalloc_root));
2143 list_add_tail(&root->delalloc_root,
2144 &fs_info->delalloc_roots);
2145 spin_unlock(&fs_info->delalloc_root_lock);
2148 spin_unlock(&root->delalloc_lock);
2152 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
2153 struct btrfs_inode *inode)
2155 struct btrfs_fs_info *fs_info = root->fs_info;
2157 if (!list_empty(&inode->delalloc_inodes)) {
2158 list_del_init(&inode->delalloc_inodes);
2159 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2160 &inode->runtime_flags);
2161 root->nr_delalloc_inodes--;
2162 if (!root->nr_delalloc_inodes) {
2163 ASSERT(list_empty(&root->delalloc_inodes));
2164 spin_lock(&fs_info->delalloc_root_lock);
2165 BUG_ON(list_empty(&root->delalloc_root));
2166 list_del_init(&root->delalloc_root);
2167 spin_unlock(&fs_info->delalloc_root_lock);
2172 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
2173 struct btrfs_inode *inode)
2175 spin_lock(&root->delalloc_lock);
2176 __btrfs_del_delalloc_inode(root, inode);
2177 spin_unlock(&root->delalloc_lock);
2181 * Properly track delayed allocation bytes in the inode and to maintain the
2182 * list of inodes that have pending delalloc work to be done.
2184 void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state,
2187 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2189 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
2192 * set_bit and clear bit hooks normally require _irqsave/restore
2193 * but in this case, we are only testing for the DELALLOC
2194 * bit, which is only set or cleared with irqs on
2196 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2197 struct btrfs_root *root = BTRFS_I(inode)->root;
2198 u64 len = state->end + 1 - state->start;
2199 u32 num_extents = count_max_extents(len);
2200 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
2202 spin_lock(&BTRFS_I(inode)->lock);
2203 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
2204 spin_unlock(&BTRFS_I(inode)->lock);
2206 /* For sanity tests */
2207 if (btrfs_is_testing(fs_info))
2210 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2211 fs_info->delalloc_batch);
2212 spin_lock(&BTRFS_I(inode)->lock);
2213 BTRFS_I(inode)->delalloc_bytes += len;
2214 if (*bits & EXTENT_DEFRAG)
2215 BTRFS_I(inode)->defrag_bytes += len;
2216 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2217 &BTRFS_I(inode)->runtime_flags))
2218 btrfs_add_delalloc_inodes(root, inode);
2219 spin_unlock(&BTRFS_I(inode)->lock);
2222 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2223 (*bits & EXTENT_DELALLOC_NEW)) {
2224 spin_lock(&BTRFS_I(inode)->lock);
2225 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
2227 spin_unlock(&BTRFS_I(inode)->lock);
2232 * Once a range is no longer delalloc this function ensures that proper
2233 * accounting happens.
2235 void btrfs_clear_delalloc_extent(struct inode *vfs_inode,
2236 struct extent_state *state, unsigned *bits)
2238 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
2239 struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb);
2240 u64 len = state->end + 1 - state->start;
2241 u32 num_extents = count_max_extents(len);
2243 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
2244 spin_lock(&inode->lock);
2245 inode->defrag_bytes -= len;
2246 spin_unlock(&inode->lock);
2250 * set_bit and clear bit hooks normally require _irqsave/restore
2251 * but in this case, we are only testing for the DELALLOC
2252 * bit, which is only set or cleared with irqs on
2254 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2255 struct btrfs_root *root = inode->root;
2256 bool do_list = !btrfs_is_free_space_inode(inode);
2258 spin_lock(&inode->lock);
2259 btrfs_mod_outstanding_extents(inode, -num_extents);
2260 spin_unlock(&inode->lock);
2263 * We don't reserve metadata space for space cache inodes so we
2264 * don't need to call delalloc_release_metadata if there is an
2267 if (*bits & EXTENT_CLEAR_META_RESV &&
2268 root != fs_info->tree_root)
2269 btrfs_delalloc_release_metadata(inode, len, false);
2271 /* For sanity tests. */
2272 if (btrfs_is_testing(fs_info))
2275 if (!btrfs_is_data_reloc_root(root) &&
2276 do_list && !(state->state & EXTENT_NORESERVE) &&
2277 (*bits & EXTENT_CLEAR_DATA_RESV))
2278 btrfs_free_reserved_data_space_noquota(fs_info, len);
2280 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2281 fs_info->delalloc_batch);
2282 spin_lock(&inode->lock);
2283 inode->delalloc_bytes -= len;
2284 if (do_list && inode->delalloc_bytes == 0 &&
2285 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2286 &inode->runtime_flags))
2287 btrfs_del_delalloc_inode(root, inode);
2288 spin_unlock(&inode->lock);
2291 if ((state->state & EXTENT_DELALLOC_NEW) &&
2292 (*bits & EXTENT_DELALLOC_NEW)) {
2293 spin_lock(&inode->lock);
2294 ASSERT(inode->new_delalloc_bytes >= len);
2295 inode->new_delalloc_bytes -= len;
2296 if (*bits & EXTENT_ADD_INODE_BYTES)
2297 inode_add_bytes(&inode->vfs_inode, len);
2298 spin_unlock(&inode->lock);
2303 * in order to insert checksums into the metadata in large chunks,
2304 * we wait until bio submission time. All the pages in the bio are
2305 * checksummed and sums are attached onto the ordered extent record.
2307 * At IO completion time the cums attached on the ordered extent record
2308 * are inserted into the btree
2310 static blk_status_t btrfs_submit_bio_start(struct inode *inode, struct bio *bio,
2311 u64 dio_file_offset)
2313 return btrfs_csum_one_bio(BTRFS_I(inode), bio, 0, 0);
2317 * Split an extent_map at [start, start + len]
2319 * This function is intended to be used only for extract_ordered_extent().
2321 static int split_zoned_em(struct btrfs_inode *inode, u64 start, u64 len,
2324 struct extent_map_tree *em_tree = &inode->extent_tree;
2325 struct extent_map *em;
2326 struct extent_map *split_pre = NULL;
2327 struct extent_map *split_mid = NULL;
2328 struct extent_map *split_post = NULL;
2330 unsigned long flags;
2333 if (pre == 0 && post == 0)
2336 split_pre = alloc_extent_map();
2338 split_mid = alloc_extent_map();
2340 split_post = alloc_extent_map();
2341 if (!split_pre || (pre && !split_mid) || (post && !split_post)) {
2346 ASSERT(pre + post < len);
2348 lock_extent(&inode->io_tree, start, start + len - 1);
2349 write_lock(&em_tree->lock);
2350 em = lookup_extent_mapping(em_tree, start, len);
2356 ASSERT(em->len == len);
2357 ASSERT(!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags));
2358 ASSERT(em->block_start < EXTENT_MAP_LAST_BYTE);
2359 ASSERT(test_bit(EXTENT_FLAG_PINNED, &em->flags));
2360 ASSERT(!test_bit(EXTENT_FLAG_LOGGING, &em->flags));
2361 ASSERT(!list_empty(&em->list));
2364 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
2366 /* First, replace the em with a new extent_map starting from * em->start */
2367 split_pre->start = em->start;
2368 split_pre->len = (pre ? pre : em->len - post);
2369 split_pre->orig_start = split_pre->start;
2370 split_pre->block_start = em->block_start;
2371 split_pre->block_len = split_pre->len;
2372 split_pre->orig_block_len = split_pre->block_len;
2373 split_pre->ram_bytes = split_pre->len;
2374 split_pre->flags = flags;
2375 split_pre->compress_type = em->compress_type;
2376 split_pre->generation = em->generation;
2378 replace_extent_mapping(em_tree, em, split_pre, 1);
2381 * Now we only have an extent_map at:
2382 * [em->start, em->start + pre] if pre != 0
2383 * [em->start, em->start + em->len - post] if pre == 0
2387 /* Insert the middle extent_map */
2388 split_mid->start = em->start + pre;
2389 split_mid->len = em->len - pre - post;
2390 split_mid->orig_start = split_mid->start;
2391 split_mid->block_start = em->block_start + pre;
2392 split_mid->block_len = split_mid->len;
2393 split_mid->orig_block_len = split_mid->block_len;
2394 split_mid->ram_bytes = split_mid->len;
2395 split_mid->flags = flags;
2396 split_mid->compress_type = em->compress_type;
2397 split_mid->generation = em->generation;
2398 add_extent_mapping(em_tree, split_mid, 1);
2402 split_post->start = em->start + em->len - post;
2403 split_post->len = post;
2404 split_post->orig_start = split_post->start;
2405 split_post->block_start = em->block_start + em->len - post;
2406 split_post->block_len = split_post->len;
2407 split_post->orig_block_len = split_post->block_len;
2408 split_post->ram_bytes = split_post->len;
2409 split_post->flags = flags;
2410 split_post->compress_type = em->compress_type;
2411 split_post->generation = em->generation;
2412 add_extent_mapping(em_tree, split_post, 1);
2416 free_extent_map(em);
2417 /* Once for the tree */
2418 free_extent_map(em);
2421 write_unlock(&em_tree->lock);
2422 unlock_extent(&inode->io_tree, start, start + len - 1);
2424 free_extent_map(split_pre);
2425 free_extent_map(split_mid);
2426 free_extent_map(split_post);
2431 static blk_status_t extract_ordered_extent(struct btrfs_inode *inode,
2432 struct bio *bio, loff_t file_offset)
2434 struct btrfs_ordered_extent *ordered;
2435 u64 start = (u64)bio->bi_iter.bi_sector << SECTOR_SHIFT;
2437 u64 len = bio->bi_iter.bi_size;
2438 u64 end = start + len;
2443 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
2444 if (WARN_ON_ONCE(!ordered))
2445 return BLK_STS_IOERR;
2447 /* No need to split */
2448 if (ordered->disk_num_bytes == len)
2451 /* We cannot split once end_bio'd ordered extent */
2452 if (WARN_ON_ONCE(ordered->bytes_left != ordered->disk_num_bytes)) {
2457 /* We cannot split a compressed ordered extent */
2458 if (WARN_ON_ONCE(ordered->disk_num_bytes != ordered->num_bytes)) {
2463 ordered_end = ordered->disk_bytenr + ordered->disk_num_bytes;
2464 /* bio must be in one ordered extent */
2465 if (WARN_ON_ONCE(start < ordered->disk_bytenr || end > ordered_end)) {
2470 /* Checksum list should be empty */
2471 if (WARN_ON_ONCE(!list_empty(&ordered->list))) {
2476 file_len = ordered->num_bytes;
2477 pre = start - ordered->disk_bytenr;
2478 post = ordered_end - end;
2480 ret = btrfs_split_ordered_extent(ordered, pre, post);
2483 ret = split_zoned_em(inode, file_offset, file_len, pre, post);
2486 btrfs_put_ordered_extent(ordered);
2488 return errno_to_blk_status(ret);
2492 * extent_io.c submission hook. This does the right thing for csum calculation
2493 * on write, or reading the csums from the tree before a read.
2495 * Rules about async/sync submit,
2496 * a) read: sync submit
2498 * b) write without checksum: sync submit
2500 * c) write with checksum:
2501 * c-1) if bio is issued by fsync: sync submit
2502 * (sync_writers != 0)
2504 * c-2) if root is reloc root: sync submit
2505 * (only in case of buffered IO)
2507 * c-3) otherwise: async submit
2509 blk_status_t btrfs_submit_data_bio(struct inode *inode, struct bio *bio,
2510 int mirror_num, unsigned long bio_flags)
2513 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2514 struct btrfs_root *root = BTRFS_I(inode)->root;
2515 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
2516 blk_status_t ret = 0;
2518 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
2520 skip_sum = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) ||
2521 test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2523 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
2524 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
2526 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
2527 struct page *page = bio_first_bvec_all(bio)->bv_page;
2528 loff_t file_offset = page_offset(page);
2530 ret = extract_ordered_extent(BTRFS_I(inode), bio, file_offset);
2535 if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
2536 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
2540 if (bio_flags & EXTENT_BIO_COMPRESSED) {
2541 ret = btrfs_submit_compressed_read(inode, bio,
2547 * Lookup bio sums does extra checks around whether we
2548 * need to csum or not, which is why we ignore skip_sum
2551 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
2556 } else if (async && !skip_sum) {
2557 /* csum items have already been cloned */
2558 if (btrfs_is_data_reloc_root(root))
2560 /* we're doing a write, do the async checksumming */
2561 ret = btrfs_wq_submit_bio(inode, bio, mirror_num, bio_flags,
2562 0, btrfs_submit_bio_start);
2564 } else if (!skip_sum) {
2565 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, 0, 0);
2571 ret = btrfs_map_bio(fs_info, bio, mirror_num);
2575 bio->bi_status = ret;
2582 * given a list of ordered sums record them in the inode. This happens
2583 * at IO completion time based on sums calculated at bio submission time.
2585 static int add_pending_csums(struct btrfs_trans_handle *trans,
2586 struct list_head *list)
2588 struct btrfs_ordered_sum *sum;
2589 struct btrfs_root *csum_root = NULL;
2592 list_for_each_entry(sum, list, list) {
2593 trans->adding_csums = true;
2595 csum_root = btrfs_csum_root(trans->fs_info,
2597 ret = btrfs_csum_file_blocks(trans, csum_root, sum);
2598 trans->adding_csums = false;
2605 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2608 struct extent_state **cached_state)
2610 u64 search_start = start;
2611 const u64 end = start + len - 1;
2613 while (search_start < end) {
2614 const u64 search_len = end - search_start + 1;
2615 struct extent_map *em;
2619 em = btrfs_get_extent(inode, NULL, 0, search_start, search_len);
2623 if (em->block_start != EXTENT_MAP_HOLE)
2627 if (em->start < search_start)
2628 em_len -= search_start - em->start;
2629 if (em_len > search_len)
2630 em_len = search_len;
2632 ret = set_extent_bit(&inode->io_tree, search_start,
2633 search_start + em_len - 1,
2634 EXTENT_DELALLOC_NEW, 0, NULL, cached_state,
2637 search_start = extent_map_end(em);
2638 free_extent_map(em);
2645 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2646 unsigned int extra_bits,
2647 struct extent_state **cached_state)
2649 WARN_ON(PAGE_ALIGNED(end));
2651 if (start >= i_size_read(&inode->vfs_inode) &&
2652 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2654 * There can't be any extents following eof in this case so just
2655 * set the delalloc new bit for the range directly.
2657 extra_bits |= EXTENT_DELALLOC_NEW;
2661 ret = btrfs_find_new_delalloc_bytes(inode, start,
2668 return set_extent_delalloc(&inode->io_tree, start, end, extra_bits,
2672 /* see btrfs_writepage_start_hook for details on why this is required */
2673 struct btrfs_writepage_fixup {
2675 struct inode *inode;
2676 struct btrfs_work work;
2679 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2681 struct btrfs_writepage_fixup *fixup;
2682 struct btrfs_ordered_extent *ordered;
2683 struct extent_state *cached_state = NULL;
2684 struct extent_changeset *data_reserved = NULL;
2686 struct btrfs_inode *inode;
2690 bool free_delalloc_space = true;
2692 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2694 inode = BTRFS_I(fixup->inode);
2695 page_start = page_offset(page);
2696 page_end = page_offset(page) + PAGE_SIZE - 1;
2699 * This is similar to page_mkwrite, we need to reserve the space before
2700 * we take the page lock.
2702 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2708 * Before we queued this fixup, we took a reference on the page.
2709 * page->mapping may go NULL, but it shouldn't be moved to a different
2712 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2714 * Unfortunately this is a little tricky, either
2716 * 1) We got here and our page had already been dealt with and
2717 * we reserved our space, thus ret == 0, so we need to just
2718 * drop our space reservation and bail. This can happen the
2719 * first time we come into the fixup worker, or could happen
2720 * while waiting for the ordered extent.
2721 * 2) Our page was already dealt with, but we happened to get an
2722 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2723 * this case we obviously don't have anything to release, but
2724 * because the page was already dealt with we don't want to
2725 * mark the page with an error, so make sure we're resetting
2726 * ret to 0. This is why we have this check _before_ the ret
2727 * check, because we do not want to have a surprise ENOSPC
2728 * when the page was already properly dealt with.
2731 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2732 btrfs_delalloc_release_space(inode, data_reserved,
2733 page_start, PAGE_SIZE,
2741 * We can't mess with the page state unless it is locked, so now that
2742 * it is locked bail if we failed to make our space reservation.
2747 lock_extent_bits(&inode->io_tree, page_start, page_end, &cached_state);
2749 /* already ordered? We're done */
2750 if (PageOrdered(page))
2753 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2755 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2758 btrfs_start_ordered_extent(ordered, 1);
2759 btrfs_put_ordered_extent(ordered);
2763 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2769 * Everything went as planned, we're now the owner of a dirty page with
2770 * delayed allocation bits set and space reserved for our COW
2773 * The page was dirty when we started, nothing should have cleaned it.
2775 BUG_ON(!PageDirty(page));
2776 free_delalloc_space = false;
2778 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2779 if (free_delalloc_space)
2780 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2782 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2787 * We hit ENOSPC or other errors. Update the mapping and page
2788 * to reflect the errors and clean the page.
2790 mapping_set_error(page->mapping, ret);
2791 end_extent_writepage(page, ret, page_start, page_end);
2792 clear_page_dirty_for_io(page);
2795 btrfs_page_clear_checked(inode->root->fs_info, page, page_start, PAGE_SIZE);
2799 extent_changeset_free(data_reserved);
2801 * As a precaution, do a delayed iput in case it would be the last iput
2802 * that could need flushing space. Recursing back to fixup worker would
2805 btrfs_add_delayed_iput(&inode->vfs_inode);
2809 * There are a few paths in the higher layers of the kernel that directly
2810 * set the page dirty bit without asking the filesystem if it is a
2811 * good idea. This causes problems because we want to make sure COW
2812 * properly happens and the data=ordered rules are followed.
2814 * In our case any range that doesn't have the ORDERED bit set
2815 * hasn't been properly setup for IO. We kick off an async process
2816 * to fix it up. The async helper will wait for ordered extents, set
2817 * the delalloc bit and make it safe to write the page.
2819 int btrfs_writepage_cow_fixup(struct page *page)
2821 struct inode *inode = page->mapping->host;
2822 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2823 struct btrfs_writepage_fixup *fixup;
2825 /* This page has ordered extent covering it already */
2826 if (PageOrdered(page))
2830 * PageChecked is set below when we create a fixup worker for this page,
2831 * don't try to create another one if we're already PageChecked()
2833 * The extent_io writepage code will redirty the page if we send back
2836 if (PageChecked(page))
2839 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2844 * We are already holding a reference to this inode from
2845 * write_cache_pages. We need to hold it because the space reservation
2846 * takes place outside of the page lock, and we can't trust
2847 * page->mapping outside of the page lock.
2850 btrfs_page_set_checked(fs_info, page, page_offset(page), PAGE_SIZE);
2852 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
2854 fixup->inode = inode;
2855 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2860 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2861 struct btrfs_inode *inode, u64 file_pos,
2862 struct btrfs_file_extent_item *stack_fi,
2863 const bool update_inode_bytes,
2864 u64 qgroup_reserved)
2866 struct btrfs_root *root = inode->root;
2867 const u64 sectorsize = root->fs_info->sectorsize;
2868 struct btrfs_path *path;
2869 struct extent_buffer *leaf;
2870 struct btrfs_key ins;
2871 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
2872 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
2873 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
2874 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
2875 struct btrfs_drop_extents_args drop_args = { 0 };
2878 path = btrfs_alloc_path();
2883 * we may be replacing one extent in the tree with another.
2884 * The new extent is pinned in the extent map, and we don't want
2885 * to drop it from the cache until it is completely in the btree.
2887 * So, tell btrfs_drop_extents to leave this extent in the cache.
2888 * the caller is expected to unpin it and allow it to be merged
2891 drop_args.path = path;
2892 drop_args.start = file_pos;
2893 drop_args.end = file_pos + num_bytes;
2894 drop_args.replace_extent = true;
2895 drop_args.extent_item_size = sizeof(*stack_fi);
2896 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2900 if (!drop_args.extent_inserted) {
2901 ins.objectid = btrfs_ino(inode);
2902 ins.offset = file_pos;
2903 ins.type = BTRFS_EXTENT_DATA_KEY;
2905 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2910 leaf = path->nodes[0];
2911 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
2912 write_extent_buffer(leaf, stack_fi,
2913 btrfs_item_ptr_offset(leaf, path->slots[0]),
2914 sizeof(struct btrfs_file_extent_item));
2916 btrfs_mark_buffer_dirty(leaf);
2917 btrfs_release_path(path);
2920 * If we dropped an inline extent here, we know the range where it is
2921 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
2922 * number of bytes only for that range containing the inline extent.
2923 * The remaining of the range will be processed when clearning the
2924 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
2926 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
2927 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
2929 inline_size = drop_args.bytes_found - inline_size;
2930 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
2931 drop_args.bytes_found -= inline_size;
2932 num_bytes -= sectorsize;
2935 if (update_inode_bytes)
2936 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
2938 ins.objectid = disk_bytenr;
2939 ins.offset = disk_num_bytes;
2940 ins.type = BTRFS_EXTENT_ITEM_KEY;
2942 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
2946 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
2947 file_pos, qgroup_reserved, &ins);
2949 btrfs_free_path(path);
2954 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2957 struct btrfs_block_group *cache;
2959 cache = btrfs_lookup_block_group(fs_info, start);
2962 spin_lock(&cache->lock);
2963 cache->delalloc_bytes -= len;
2964 spin_unlock(&cache->lock);
2966 btrfs_put_block_group(cache);
2969 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
2970 struct btrfs_ordered_extent *oe)
2972 struct btrfs_file_extent_item stack_fi;
2974 bool update_inode_bytes;
2976 memset(&stack_fi, 0, sizeof(stack_fi));
2977 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
2978 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
2979 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
2980 oe->disk_num_bytes);
2981 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags))
2982 logical_len = oe->truncated_len;
2984 logical_len = oe->num_bytes;
2985 btrfs_set_stack_file_extent_num_bytes(&stack_fi, logical_len);
2986 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, logical_len);
2987 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
2988 /* Encryption and other encoding is reserved and all 0 */
2991 * For delalloc, when completing an ordered extent we update the inode's
2992 * bytes when clearing the range in the inode's io tree, so pass false
2993 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
2994 * except if the ordered extent was truncated.
2996 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
2997 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
2999 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
3000 oe->file_offset, &stack_fi,
3001 update_inode_bytes, oe->qgroup_rsv);
3005 * As ordered data IO finishes, this gets called so we can finish
3006 * an ordered extent if the range of bytes in the file it covers are
3009 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
3011 struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
3012 struct btrfs_root *root = inode->root;
3013 struct btrfs_fs_info *fs_info = root->fs_info;
3014 struct btrfs_trans_handle *trans = NULL;
3015 struct extent_io_tree *io_tree = &inode->io_tree;
3016 struct extent_state *cached_state = NULL;
3018 int compress_type = 0;
3020 u64 logical_len = ordered_extent->num_bytes;
3021 bool freespace_inode;
3022 bool truncated = false;
3023 bool clear_reserved_extent = true;
3024 unsigned int clear_bits = EXTENT_DEFRAG;
3026 start = ordered_extent->file_offset;
3027 end = start + ordered_extent->num_bytes - 1;
3029 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3030 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
3031 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
3032 clear_bits |= EXTENT_DELALLOC_NEW;
3034 freespace_inode = btrfs_is_free_space_inode(inode);
3036 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
3041 /* A valid bdev implies a write on a sequential zone */
3042 if (ordered_extent->bdev) {
3043 btrfs_rewrite_logical_zoned(ordered_extent);
3044 btrfs_zone_finish_endio(fs_info, ordered_extent->disk_bytenr,
3045 ordered_extent->disk_num_bytes);
3048 btrfs_free_io_failure_record(inode, start, end);
3050 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
3052 logical_len = ordered_extent->truncated_len;
3053 /* Truncated the entire extent, don't bother adding */
3058 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
3059 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
3061 btrfs_inode_safe_disk_i_size_write(inode, 0);
3062 if (freespace_inode)
3063 trans = btrfs_join_transaction_spacecache(root);
3065 trans = btrfs_join_transaction(root);
3066 if (IS_ERR(trans)) {
3067 ret = PTR_ERR(trans);
3071 trans->block_rsv = &inode->block_rsv;
3072 ret = btrfs_update_inode_fallback(trans, root, inode);
3073 if (ret) /* -ENOMEM or corruption */
3074 btrfs_abort_transaction(trans, ret);
3078 clear_bits |= EXTENT_LOCKED;
3079 lock_extent_bits(io_tree, start, end, &cached_state);
3081 if (freespace_inode)
3082 trans = btrfs_join_transaction_spacecache(root);
3084 trans = btrfs_join_transaction(root);
3085 if (IS_ERR(trans)) {
3086 ret = PTR_ERR(trans);
3091 trans->block_rsv = &inode->block_rsv;
3093 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3094 compress_type = ordered_extent->compress_type;
3095 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3096 BUG_ON(compress_type);
3097 ret = btrfs_mark_extent_written(trans, inode,
3098 ordered_extent->file_offset,
3099 ordered_extent->file_offset +
3102 BUG_ON(root == fs_info->tree_root);
3103 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
3105 clear_reserved_extent = false;
3106 btrfs_release_delalloc_bytes(fs_info,
3107 ordered_extent->disk_bytenr,
3108 ordered_extent->disk_num_bytes);
3111 unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset,
3112 ordered_extent->num_bytes, trans->transid);
3114 btrfs_abort_transaction(trans, ret);
3118 ret = add_pending_csums(trans, &ordered_extent->list);
3120 btrfs_abort_transaction(trans, ret);
3125 * If this is a new delalloc range, clear its new delalloc flag to
3126 * update the inode's number of bytes. This needs to be done first
3127 * before updating the inode item.
3129 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
3130 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
3131 clear_extent_bit(&inode->io_tree, start, end,
3132 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
3133 0, 0, &cached_state);
3135 btrfs_inode_safe_disk_i_size_write(inode, 0);
3136 ret = btrfs_update_inode_fallback(trans, root, inode);
3137 if (ret) { /* -ENOMEM or corruption */
3138 btrfs_abort_transaction(trans, ret);
3143 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
3144 (clear_bits & EXTENT_LOCKED) ? 1 : 0, 0,
3148 btrfs_end_transaction(trans);
3150 if (ret || truncated) {
3151 u64 unwritten_start = start;
3154 * If we failed to finish this ordered extent for any reason we
3155 * need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3156 * extent, and mark the inode with the error if it wasn't
3157 * already set. Any error during writeback would have already
3158 * set the mapping error, so we need to set it if we're the ones
3159 * marking this ordered extent as failed.
3161 if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR,
3162 &ordered_extent->flags))
3163 mapping_set_error(ordered_extent->inode->i_mapping, -EIO);
3166 unwritten_start += logical_len;
3167 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
3169 /* Drop the cache for the part of the extent we didn't write. */
3170 btrfs_drop_extent_cache(inode, unwritten_start, end, 0);
3173 * If the ordered extent had an IOERR or something else went
3174 * wrong we need to return the space for this ordered extent
3175 * back to the allocator. We only free the extent in the
3176 * truncated case if we didn't write out the extent at all.
3178 * If we made it past insert_reserved_file_extent before we
3179 * errored out then we don't need to do this as the accounting
3180 * has already been done.
3182 if ((ret || !logical_len) &&
3183 clear_reserved_extent &&
3184 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3185 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3187 * Discard the range before returning it back to the
3190 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3191 btrfs_discard_extent(fs_info,
3192 ordered_extent->disk_bytenr,
3193 ordered_extent->disk_num_bytes,
3195 btrfs_free_reserved_extent(fs_info,
3196 ordered_extent->disk_bytenr,
3197 ordered_extent->disk_num_bytes, 1);
3202 * This needs to be done to make sure anybody waiting knows we are done
3203 * updating everything for this ordered extent.
3205 btrfs_remove_ordered_extent(inode, ordered_extent);
3208 btrfs_put_ordered_extent(ordered_extent);
3209 /* once for the tree */
3210 btrfs_put_ordered_extent(ordered_extent);
3215 static void finish_ordered_fn(struct btrfs_work *work)
3217 struct btrfs_ordered_extent *ordered_extent;
3218 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3219 btrfs_finish_ordered_io(ordered_extent);
3222 void btrfs_writepage_endio_finish_ordered(struct btrfs_inode *inode,
3223 struct page *page, u64 start,
3224 u64 end, bool uptodate)
3226 trace_btrfs_writepage_end_io_hook(inode, start, end, uptodate);
3228 btrfs_mark_ordered_io_finished(inode, page, start, end + 1 - start,
3229 finish_ordered_fn, uptodate);
3233 * check_data_csum - verify checksum of one sector of uncompressed data
3235 * @io_bio: btrfs_io_bio which contains the csum
3236 * @bio_offset: offset to the beginning of the bio (in bytes)
3237 * @page: page where is the data to be verified
3238 * @pgoff: offset inside the page
3239 * @start: logical offset in the file
3241 * The length of such check is always one sector size.
3243 static int check_data_csum(struct inode *inode, struct btrfs_bio *bbio,
3244 u32 bio_offset, struct page *page, u32 pgoff,
3247 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3248 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3250 u32 len = fs_info->sectorsize;
3251 const u32 csum_size = fs_info->csum_size;
3252 unsigned int offset_sectors;
3254 u8 csum[BTRFS_CSUM_SIZE];
3256 ASSERT(pgoff + len <= PAGE_SIZE);
3258 offset_sectors = bio_offset >> fs_info->sectorsize_bits;
3259 csum_expected = ((u8 *)bbio->csum) + offset_sectors * csum_size;
3261 kaddr = kmap_atomic(page);
3262 shash->tfm = fs_info->csum_shash;
3264 crypto_shash_digest(shash, kaddr + pgoff, len, csum);
3266 if (memcmp(csum, csum_expected, csum_size))
3269 kunmap_atomic(kaddr);
3272 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3275 btrfs_dev_stat_inc_and_print(bbio->device,
3276 BTRFS_DEV_STAT_CORRUPTION_ERRS);
3277 memset(kaddr + pgoff, 1, len);
3278 flush_dcache_page(page);
3279 kunmap_atomic(kaddr);
3284 * When reads are done, we need to check csums to verify the data is correct.
3285 * if there's a match, we allow the bio to finish. If not, the code in
3286 * extent_io.c will try to find good copies for us.
3288 * @bio_offset: offset to the beginning of the bio (in bytes)
3289 * @start: file offset of the range start
3290 * @end: file offset of the range end (inclusive)
3292 * Return a bitmap where bit set means a csum mismatch, and bit not set means
3295 unsigned int btrfs_verify_data_csum(struct btrfs_bio *bbio,
3296 u32 bio_offset, struct page *page,
3299 struct inode *inode = page->mapping->host;
3300 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3301 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3302 struct btrfs_root *root = BTRFS_I(inode)->root;
3303 const u32 sectorsize = root->fs_info->sectorsize;
3305 unsigned int result = 0;
3307 if (btrfs_page_test_checked(fs_info, page, start, end + 1 - start)) {
3308 btrfs_page_clear_checked(fs_info, page, start, end + 1 - start);
3313 * This only happens for NODATASUM or compressed read.
3314 * Normally this should be covered by above check for compressed read
3315 * or the next check for NODATASUM. Just do a quicker exit here.
3317 if (bbio->csum == NULL)
3320 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3323 if (unlikely(test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state)))
3326 ASSERT(page_offset(page) <= start &&
3327 end <= page_offset(page) + PAGE_SIZE - 1);
3328 for (pg_off = offset_in_page(start);
3329 pg_off < offset_in_page(end);
3330 pg_off += sectorsize, bio_offset += sectorsize) {
3331 u64 file_offset = pg_off + page_offset(page);
3334 if (btrfs_is_data_reloc_root(root) &&
3335 test_range_bit(io_tree, file_offset,
3336 file_offset + sectorsize - 1,
3337 EXTENT_NODATASUM, 1, NULL)) {
3338 /* Skip the range without csum for data reloc inode */
3339 clear_extent_bits(io_tree, file_offset,
3340 file_offset + sectorsize - 1,
3344 ret = check_data_csum(inode, bbio, bio_offset, page, pg_off,
3345 page_offset(page) + pg_off);
3347 const int nr_bit = (pg_off - offset_in_page(start)) >>
3348 root->fs_info->sectorsize_bits;
3350 result |= (1U << nr_bit);
3357 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3359 * @inode: The inode we want to perform iput on
3361 * This function uses the generic vfs_inode::i_count to track whether we should
3362 * just decrement it (in case it's > 1) or if this is the last iput then link
3363 * the inode to the delayed iput machinery. Delayed iputs are processed at
3364 * transaction commit time/superblock commit/cleaner kthread.
3366 void btrfs_add_delayed_iput(struct inode *inode)
3368 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3369 struct btrfs_inode *binode = BTRFS_I(inode);
3371 if (atomic_add_unless(&inode->i_count, -1, 1))
3374 atomic_inc(&fs_info->nr_delayed_iputs);
3375 spin_lock(&fs_info->delayed_iput_lock);
3376 ASSERT(list_empty(&binode->delayed_iput));
3377 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3378 spin_unlock(&fs_info->delayed_iput_lock);
3379 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3380 wake_up_process(fs_info->cleaner_kthread);
3383 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3384 struct btrfs_inode *inode)
3386 list_del_init(&inode->delayed_iput);
3387 spin_unlock(&fs_info->delayed_iput_lock);
3388 iput(&inode->vfs_inode);
3389 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3390 wake_up(&fs_info->delayed_iputs_wait);
3391 spin_lock(&fs_info->delayed_iput_lock);
3394 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3395 struct btrfs_inode *inode)
3397 if (!list_empty(&inode->delayed_iput)) {
3398 spin_lock(&fs_info->delayed_iput_lock);
3399 if (!list_empty(&inode->delayed_iput))
3400 run_delayed_iput_locked(fs_info, inode);
3401 spin_unlock(&fs_info->delayed_iput_lock);
3405 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3408 spin_lock(&fs_info->delayed_iput_lock);
3409 while (!list_empty(&fs_info->delayed_iputs)) {
3410 struct btrfs_inode *inode;
3412 inode = list_first_entry(&fs_info->delayed_iputs,
3413 struct btrfs_inode, delayed_iput);
3414 run_delayed_iput_locked(fs_info, inode);
3415 cond_resched_lock(&fs_info->delayed_iput_lock);
3417 spin_unlock(&fs_info->delayed_iput_lock);
3421 * Wait for flushing all delayed iputs
3423 * @fs_info: the filesystem
3425 * This will wait on any delayed iputs that are currently running with KILLABLE
3426 * set. Once they are all done running we will return, unless we are killed in
3427 * which case we return EINTR. This helps in user operations like fallocate etc
3428 * that might get blocked on the iputs.
3430 * Return EINTR if we were killed, 0 if nothing's pending
3432 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3434 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3435 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3442 * This creates an orphan entry for the given inode in case something goes wrong
3443 * in the middle of an unlink.
3445 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3446 struct btrfs_inode *inode)
3450 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3451 if (ret && ret != -EEXIST) {
3452 btrfs_abort_transaction(trans, ret);
3460 * We have done the delete so we can go ahead and remove the orphan item for
3461 * this particular inode.
3463 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3464 struct btrfs_inode *inode)
3466 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3470 * this cleans up any orphans that may be left on the list from the last use
3473 int btrfs_orphan_cleanup(struct btrfs_root *root)
3475 struct btrfs_fs_info *fs_info = root->fs_info;
3476 struct btrfs_path *path;
3477 struct extent_buffer *leaf;
3478 struct btrfs_key key, found_key;
3479 struct btrfs_trans_handle *trans;
3480 struct inode *inode;
3481 u64 last_objectid = 0;
3482 int ret = 0, nr_unlink = 0;
3484 if (test_and_set_bit(BTRFS_ROOT_ORPHAN_CLEANUP, &root->state))
3487 path = btrfs_alloc_path();
3492 path->reada = READA_BACK;
3494 key.objectid = BTRFS_ORPHAN_OBJECTID;
3495 key.type = BTRFS_ORPHAN_ITEM_KEY;
3496 key.offset = (u64)-1;
3499 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3504 * if ret == 0 means we found what we were searching for, which
3505 * is weird, but possible, so only screw with path if we didn't
3506 * find the key and see if we have stuff that matches
3510 if (path->slots[0] == 0)
3515 /* pull out the item */
3516 leaf = path->nodes[0];
3517 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3519 /* make sure the item matches what we want */
3520 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3522 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3525 /* release the path since we're done with it */
3526 btrfs_release_path(path);
3529 * this is where we are basically btrfs_lookup, without the
3530 * crossing root thing. we store the inode number in the
3531 * offset of the orphan item.
3534 if (found_key.offset == last_objectid) {
3536 "Error removing orphan entry, stopping orphan cleanup");
3541 last_objectid = found_key.offset;
3543 found_key.objectid = found_key.offset;
3544 found_key.type = BTRFS_INODE_ITEM_KEY;
3545 found_key.offset = 0;
3546 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3547 ret = PTR_ERR_OR_ZERO(inode);
3548 if (ret && ret != -ENOENT)
3551 if (ret == -ENOENT && root == fs_info->tree_root) {
3552 struct btrfs_root *dead_root;
3553 int is_dead_root = 0;
3556 * This is an orphan in the tree root. Currently these
3557 * could come from 2 sources:
3558 * a) a root (snapshot/subvolume) deletion in progress
3559 * b) a free space cache inode
3560 * We need to distinguish those two, as the orphan item
3561 * for a root must not get deleted before the deletion
3562 * of the snapshot/subvolume's tree completes.
3564 * btrfs_find_orphan_roots() ran before us, which has
3565 * found all deleted roots and loaded them into
3566 * fs_info->fs_roots_radix. So here we can find if an
3567 * orphan item corresponds to a deleted root by looking
3568 * up the root from that radix tree.
3571 spin_lock(&fs_info->fs_roots_radix_lock);
3572 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3573 (unsigned long)found_key.objectid);
3574 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3576 spin_unlock(&fs_info->fs_roots_radix_lock);
3579 /* prevent this orphan from being found again */
3580 key.offset = found_key.objectid - 1;
3587 * If we have an inode with links, there are a couple of
3590 * 1. We were halfway through creating fsverity metadata for the
3591 * file. In that case, the orphan item represents incomplete
3592 * fsverity metadata which must be cleaned up with
3593 * btrfs_drop_verity_items and deleting the orphan item.
3595 * 2. Old kernels (before v3.12) used to create an
3596 * orphan item for truncate indicating that there were possibly
3597 * extent items past i_size that needed to be deleted. In v3.12,
3598 * truncate was changed to update i_size in sync with the extent
3599 * items, but the (useless) orphan item was still created. Since
3600 * v4.18, we don't create the orphan item for truncate at all.
3602 * So, this item could mean that we need to do a truncate, but
3603 * only if this filesystem was last used on a pre-v3.12 kernel
3604 * and was not cleanly unmounted. The odds of that are quite
3605 * slim, and it's a pain to do the truncate now, so just delete
3608 * It's also possible that this orphan item was supposed to be
3609 * deleted but wasn't. The inode number may have been reused,
3610 * but either way, we can delete the orphan item.
3612 if (ret == -ENOENT || inode->i_nlink) {
3614 ret = btrfs_drop_verity_items(BTRFS_I(inode));
3619 trans = btrfs_start_transaction(root, 1);
3620 if (IS_ERR(trans)) {
3621 ret = PTR_ERR(trans);
3624 btrfs_debug(fs_info, "auto deleting %Lu",
3625 found_key.objectid);
3626 ret = btrfs_del_orphan_item(trans, root,
3627 found_key.objectid);
3628 btrfs_end_transaction(trans);
3636 /* this will do delete_inode and everything for us */
3639 /* release the path since we're done with it */
3640 btrfs_release_path(path);
3642 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3643 trans = btrfs_join_transaction(root);
3645 btrfs_end_transaction(trans);
3649 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3653 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3654 btrfs_free_path(path);
3659 * very simple check to peek ahead in the leaf looking for xattrs. If we
3660 * don't find any xattrs, we know there can't be any acls.
3662 * slot is the slot the inode is in, objectid is the objectid of the inode
3664 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3665 int slot, u64 objectid,
3666 int *first_xattr_slot)
3668 u32 nritems = btrfs_header_nritems(leaf);
3669 struct btrfs_key found_key;
3670 static u64 xattr_access = 0;
3671 static u64 xattr_default = 0;
3674 if (!xattr_access) {
3675 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3676 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3677 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3678 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3682 *first_xattr_slot = -1;
3683 while (slot < nritems) {
3684 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3686 /* we found a different objectid, there must not be acls */
3687 if (found_key.objectid != objectid)
3690 /* we found an xattr, assume we've got an acl */
3691 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3692 if (*first_xattr_slot == -1)
3693 *first_xattr_slot = slot;
3694 if (found_key.offset == xattr_access ||
3695 found_key.offset == xattr_default)
3700 * we found a key greater than an xattr key, there can't
3701 * be any acls later on
3703 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3710 * it goes inode, inode backrefs, xattrs, extents,
3711 * so if there are a ton of hard links to an inode there can
3712 * be a lot of backrefs. Don't waste time searching too hard,
3713 * this is just an optimization
3718 /* we hit the end of the leaf before we found an xattr or
3719 * something larger than an xattr. We have to assume the inode
3722 if (*first_xattr_slot == -1)
3723 *first_xattr_slot = slot;
3728 * read an inode from the btree into the in-memory inode
3730 static int btrfs_read_locked_inode(struct inode *inode,
3731 struct btrfs_path *in_path)
3733 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3734 struct btrfs_path *path = in_path;
3735 struct extent_buffer *leaf;
3736 struct btrfs_inode_item *inode_item;
3737 struct btrfs_root *root = BTRFS_I(inode)->root;
3738 struct btrfs_key location;
3743 bool filled = false;
3744 int first_xattr_slot;
3746 ret = btrfs_fill_inode(inode, &rdev);
3751 path = btrfs_alloc_path();
3756 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3758 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3760 if (path != in_path)
3761 btrfs_free_path(path);
3765 leaf = path->nodes[0];
3770 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3771 struct btrfs_inode_item);
3772 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3773 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3774 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3775 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3776 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3777 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3778 round_up(i_size_read(inode), fs_info->sectorsize));
3780 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3781 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3783 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3784 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3786 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3787 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3789 BTRFS_I(inode)->i_otime.tv_sec =
3790 btrfs_timespec_sec(leaf, &inode_item->otime);
3791 BTRFS_I(inode)->i_otime.tv_nsec =
3792 btrfs_timespec_nsec(leaf, &inode_item->otime);
3794 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3795 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3796 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3798 inode_set_iversion_queried(inode,
3799 btrfs_inode_sequence(leaf, inode_item));
3800 inode->i_generation = BTRFS_I(inode)->generation;
3802 rdev = btrfs_inode_rdev(leaf, inode_item);
3804 BTRFS_I(inode)->index_cnt = (u64)-1;
3805 btrfs_inode_split_flags(btrfs_inode_flags(leaf, inode_item),
3806 &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags);
3810 * If we were modified in the current generation and evicted from memory
3811 * and then re-read we need to do a full sync since we don't have any
3812 * idea about which extents were modified before we were evicted from
3815 * This is required for both inode re-read from disk and delayed inode
3816 * in delayed_nodes_tree.
3818 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3819 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3820 &BTRFS_I(inode)->runtime_flags);
3823 * We don't persist the id of the transaction where an unlink operation
3824 * against the inode was last made. So here we assume the inode might
3825 * have been evicted, and therefore the exact value of last_unlink_trans
3826 * lost, and set it to last_trans to avoid metadata inconsistencies
3827 * between the inode and its parent if the inode is fsync'ed and the log
3828 * replayed. For example, in the scenario:
3831 * ln mydir/foo mydir/bar
3834 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3835 * xfs_io -c fsync mydir/foo
3837 * mount fs, triggers fsync log replay
3839 * We must make sure that when we fsync our inode foo we also log its
3840 * parent inode, otherwise after log replay the parent still has the
3841 * dentry with the "bar" name but our inode foo has a link count of 1
3842 * and doesn't have an inode ref with the name "bar" anymore.
3844 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3845 * but it guarantees correctness at the expense of occasional full
3846 * transaction commits on fsync if our inode is a directory, or if our
3847 * inode is not a directory, logging its parent unnecessarily.
3849 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3852 * Same logic as for last_unlink_trans. We don't persist the generation
3853 * of the last transaction where this inode was used for a reflink
3854 * operation, so after eviction and reloading the inode we must be
3855 * pessimistic and assume the last transaction that modified the inode.
3857 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3860 if (inode->i_nlink != 1 ||
3861 path->slots[0] >= btrfs_header_nritems(leaf))
3864 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3865 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3868 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3869 if (location.type == BTRFS_INODE_REF_KEY) {
3870 struct btrfs_inode_ref *ref;
3872 ref = (struct btrfs_inode_ref *)ptr;
3873 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3874 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3875 struct btrfs_inode_extref *extref;
3877 extref = (struct btrfs_inode_extref *)ptr;
3878 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3883 * try to precache a NULL acl entry for files that don't have
3884 * any xattrs or acls
3886 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3887 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3888 if (first_xattr_slot != -1) {
3889 path->slots[0] = first_xattr_slot;
3890 ret = btrfs_load_inode_props(inode, path);
3893 "error loading props for ino %llu (root %llu): %d",
3894 btrfs_ino(BTRFS_I(inode)),
3895 root->root_key.objectid, ret);
3897 if (path != in_path)
3898 btrfs_free_path(path);
3901 cache_no_acl(inode);
3903 switch (inode->i_mode & S_IFMT) {
3905 inode->i_mapping->a_ops = &btrfs_aops;
3906 inode->i_fop = &btrfs_file_operations;
3907 inode->i_op = &btrfs_file_inode_operations;
3910 inode->i_fop = &btrfs_dir_file_operations;
3911 inode->i_op = &btrfs_dir_inode_operations;
3914 inode->i_op = &btrfs_symlink_inode_operations;
3915 inode_nohighmem(inode);
3916 inode->i_mapping->a_ops = &btrfs_aops;
3919 inode->i_op = &btrfs_special_inode_operations;
3920 init_special_inode(inode, inode->i_mode, rdev);
3924 btrfs_sync_inode_flags_to_i_flags(inode);
3929 * given a leaf and an inode, copy the inode fields into the leaf
3931 static void fill_inode_item(struct btrfs_trans_handle *trans,
3932 struct extent_buffer *leaf,
3933 struct btrfs_inode_item *item,
3934 struct inode *inode)
3936 struct btrfs_map_token token;
3939 btrfs_init_map_token(&token, leaf);
3941 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3942 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3943 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
3944 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3945 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3947 btrfs_set_token_timespec_sec(&token, &item->atime,
3948 inode->i_atime.tv_sec);
3949 btrfs_set_token_timespec_nsec(&token, &item->atime,
3950 inode->i_atime.tv_nsec);
3952 btrfs_set_token_timespec_sec(&token, &item->mtime,
3953 inode->i_mtime.tv_sec);
3954 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3955 inode->i_mtime.tv_nsec);
3957 btrfs_set_token_timespec_sec(&token, &item->ctime,
3958 inode->i_ctime.tv_sec);
3959 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3960 inode->i_ctime.tv_nsec);
3962 btrfs_set_token_timespec_sec(&token, &item->otime,
3963 BTRFS_I(inode)->i_otime.tv_sec);
3964 btrfs_set_token_timespec_nsec(&token, &item->otime,
3965 BTRFS_I(inode)->i_otime.tv_nsec);
3967 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
3968 btrfs_set_token_inode_generation(&token, item,
3969 BTRFS_I(inode)->generation);
3970 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
3971 btrfs_set_token_inode_transid(&token, item, trans->transid);
3972 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
3973 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
3974 BTRFS_I(inode)->ro_flags);
3975 btrfs_set_token_inode_flags(&token, item, flags);
3976 btrfs_set_token_inode_block_group(&token, item, 0);
3980 * copy everything in the in-memory inode into the btree.
3982 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3983 struct btrfs_root *root,
3984 struct btrfs_inode *inode)
3986 struct btrfs_inode_item *inode_item;
3987 struct btrfs_path *path;
3988 struct extent_buffer *leaf;
3991 path = btrfs_alloc_path();
3995 ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1);
4002 leaf = path->nodes[0];
4003 inode_item = btrfs_item_ptr(leaf, path->slots[0],
4004 struct btrfs_inode_item);
4006 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
4007 btrfs_mark_buffer_dirty(leaf);
4008 btrfs_set_inode_last_trans(trans, inode);
4011 btrfs_free_path(path);
4016 * copy everything in the in-memory inode into the btree.
4018 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
4019 struct btrfs_root *root,
4020 struct btrfs_inode *inode)
4022 struct btrfs_fs_info *fs_info = root->fs_info;
4026 * If the inode is a free space inode, we can deadlock during commit
4027 * if we put it into the delayed code.
4029 * The data relocation inode should also be directly updated
4032 if (!btrfs_is_free_space_inode(inode)
4033 && !btrfs_is_data_reloc_root(root)
4034 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4035 btrfs_update_root_times(trans, root);
4037 ret = btrfs_delayed_update_inode(trans, root, inode);
4039 btrfs_set_inode_last_trans(trans, inode);
4043 return btrfs_update_inode_item(trans, root, inode);
4046 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4047 struct btrfs_root *root, struct btrfs_inode *inode)
4051 ret = btrfs_update_inode(trans, root, inode);
4053 return btrfs_update_inode_item(trans, root, inode);
4058 * unlink helper that gets used here in inode.c and in the tree logging
4059 * recovery code. It remove a link in a directory with a given name, and
4060 * also drops the back refs in the inode to the directory
4062 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4063 struct btrfs_inode *dir,
4064 struct btrfs_inode *inode,
4065 const char *name, int name_len)
4067 struct btrfs_root *root = dir->root;
4068 struct btrfs_fs_info *fs_info = root->fs_info;
4069 struct btrfs_path *path;
4071 struct btrfs_dir_item *di;
4073 u64 ino = btrfs_ino(inode);
4074 u64 dir_ino = btrfs_ino(dir);
4076 path = btrfs_alloc_path();
4082 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4083 name, name_len, -1);
4084 if (IS_ERR_OR_NULL(di)) {
4085 ret = di ? PTR_ERR(di) : -ENOENT;
4088 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4091 btrfs_release_path(path);
4094 * If we don't have dir index, we have to get it by looking up
4095 * the inode ref, since we get the inode ref, remove it directly,
4096 * it is unnecessary to do delayed deletion.
4098 * But if we have dir index, needn't search inode ref to get it.
4099 * Since the inode ref is close to the inode item, it is better
4100 * that we delay to delete it, and just do this deletion when
4101 * we update the inode item.
4103 if (inode->dir_index) {
4104 ret = btrfs_delayed_delete_inode_ref(inode);
4106 index = inode->dir_index;
4111 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4115 "failed to delete reference to %.*s, inode %llu parent %llu",
4116 name_len, name, ino, dir_ino);
4117 btrfs_abort_transaction(trans, ret);
4121 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4123 btrfs_abort_transaction(trans, ret);
4127 btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
4129 btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir, index);
4132 * If we have a pending delayed iput we could end up with the final iput
4133 * being run in btrfs-cleaner context. If we have enough of these built
4134 * up we can end up burning a lot of time in btrfs-cleaner without any
4135 * way to throttle the unlinks. Since we're currently holding a ref on
4136 * the inode we can run the delayed iput here without any issues as the
4137 * final iput won't be done until after we drop the ref we're currently
4140 btrfs_run_delayed_iput(fs_info, inode);
4142 btrfs_free_path(path);
4146 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
4147 inode_inc_iversion(&inode->vfs_inode);
4148 inode_inc_iversion(&dir->vfs_inode);
4149 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
4150 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4151 ret = btrfs_update_inode(trans, root, dir);
4156 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4157 struct btrfs_inode *dir, struct btrfs_inode *inode,
4158 const char *name, int name_len)
4161 ret = __btrfs_unlink_inode(trans, dir, inode, name, name_len);
4163 drop_nlink(&inode->vfs_inode);
4164 ret = btrfs_update_inode(trans, inode->root, inode);
4170 * helper to start transaction for unlink and rmdir.
4172 * unlink and rmdir are special in btrfs, they do not always free space, so
4173 * if we cannot make our reservations the normal way try and see if there is
4174 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4175 * allow the unlink to occur.
4177 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4179 struct btrfs_root *root = BTRFS_I(dir)->root;
4182 * 1 for the possible orphan item
4183 * 1 for the dir item
4184 * 1 for the dir index
4185 * 1 for the inode ref
4188 return btrfs_start_transaction_fallback_global_rsv(root, 5);
4191 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4193 struct btrfs_trans_handle *trans;
4194 struct inode *inode = d_inode(dentry);
4197 trans = __unlink_start_trans(dir);
4199 return PTR_ERR(trans);
4201 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4204 ret = btrfs_unlink_inode(trans, BTRFS_I(dir),
4205 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4206 dentry->d_name.len);
4210 if (inode->i_nlink == 0) {
4211 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4217 btrfs_end_transaction(trans);
4218 btrfs_btree_balance_dirty(BTRFS_I(dir)->root->fs_info);
4222 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4223 struct inode *dir, struct dentry *dentry)
4225 struct btrfs_root *root = BTRFS_I(dir)->root;
4226 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4227 struct btrfs_path *path;
4228 struct extent_buffer *leaf;
4229 struct btrfs_dir_item *di;
4230 struct btrfs_key key;
4231 const char *name = dentry->d_name.name;
4232 int name_len = dentry->d_name.len;
4236 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4238 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4239 objectid = inode->root->root_key.objectid;
4240 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4241 objectid = inode->location.objectid;
4247 path = btrfs_alloc_path();
4251 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4252 name, name_len, -1);
4253 if (IS_ERR_OR_NULL(di)) {
4254 ret = di ? PTR_ERR(di) : -ENOENT;
4258 leaf = path->nodes[0];
4259 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4260 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4261 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4263 btrfs_abort_transaction(trans, ret);
4266 btrfs_release_path(path);
4269 * This is a placeholder inode for a subvolume we didn't have a
4270 * reference to at the time of the snapshot creation. In the meantime
4271 * we could have renamed the real subvol link into our snapshot, so
4272 * depending on btrfs_del_root_ref to return -ENOENT here is incorrect.
4273 * Instead simply lookup the dir_index_item for this entry so we can
4274 * remove it. Otherwise we know we have a ref to the root and we can
4275 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4277 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4278 di = btrfs_search_dir_index_item(root, path, dir_ino,
4280 if (IS_ERR_OR_NULL(di)) {
4285 btrfs_abort_transaction(trans, ret);
4289 leaf = path->nodes[0];
4290 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4292 btrfs_release_path(path);
4294 ret = btrfs_del_root_ref(trans, objectid,
4295 root->root_key.objectid, dir_ino,
4296 &index, name, name_len);
4298 btrfs_abort_transaction(trans, ret);
4303 ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index);
4305 btrfs_abort_transaction(trans, ret);
4309 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4310 inode_inc_iversion(dir);
4311 dir->i_mtime = dir->i_ctime = current_time(dir);
4312 ret = btrfs_update_inode_fallback(trans, root, BTRFS_I(dir));
4314 btrfs_abort_transaction(trans, ret);
4316 btrfs_free_path(path);
4321 * Helper to check if the subvolume references other subvolumes or if it's
4324 static noinline int may_destroy_subvol(struct btrfs_root *root)
4326 struct btrfs_fs_info *fs_info = root->fs_info;
4327 struct btrfs_path *path;
4328 struct btrfs_dir_item *di;
4329 struct btrfs_key key;
4333 path = btrfs_alloc_path();
4337 /* Make sure this root isn't set as the default subvol */
4338 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4339 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4340 dir_id, "default", 7, 0);
4341 if (di && !IS_ERR(di)) {
4342 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4343 if (key.objectid == root->root_key.objectid) {
4346 "deleting default subvolume %llu is not allowed",
4350 btrfs_release_path(path);
4353 key.objectid = root->root_key.objectid;
4354 key.type = BTRFS_ROOT_REF_KEY;
4355 key.offset = (u64)-1;
4357 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4363 if (path->slots[0] > 0) {
4365 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4366 if (key.objectid == root->root_key.objectid &&
4367 key.type == BTRFS_ROOT_REF_KEY)
4371 btrfs_free_path(path);
4375 /* Delete all dentries for inodes belonging to the root */
4376 static void btrfs_prune_dentries(struct btrfs_root *root)
4378 struct btrfs_fs_info *fs_info = root->fs_info;
4379 struct rb_node *node;
4380 struct rb_node *prev;
4381 struct btrfs_inode *entry;
4382 struct inode *inode;
4385 if (!BTRFS_FS_ERROR(fs_info))
4386 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4388 spin_lock(&root->inode_lock);
4390 node = root->inode_tree.rb_node;
4394 entry = rb_entry(node, struct btrfs_inode, rb_node);
4396 if (objectid < btrfs_ino(entry))
4397 node = node->rb_left;
4398 else if (objectid > btrfs_ino(entry))
4399 node = node->rb_right;
4405 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4406 if (objectid <= btrfs_ino(entry)) {
4410 prev = rb_next(prev);
4414 entry = rb_entry(node, struct btrfs_inode, rb_node);
4415 objectid = btrfs_ino(entry) + 1;
4416 inode = igrab(&entry->vfs_inode);
4418 spin_unlock(&root->inode_lock);
4419 if (atomic_read(&inode->i_count) > 1)
4420 d_prune_aliases(inode);
4422 * btrfs_drop_inode will have it removed from the inode
4423 * cache when its usage count hits zero.
4427 spin_lock(&root->inode_lock);
4431 if (cond_resched_lock(&root->inode_lock))
4434 node = rb_next(node);
4436 spin_unlock(&root->inode_lock);
4439 int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
4441 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4442 struct btrfs_root *root = BTRFS_I(dir)->root;
4443 struct inode *inode = d_inode(dentry);
4444 struct btrfs_root *dest = BTRFS_I(inode)->root;
4445 struct btrfs_trans_handle *trans;
4446 struct btrfs_block_rsv block_rsv;
4451 * Don't allow to delete a subvolume with send in progress. This is
4452 * inside the inode lock so the error handling that has to drop the bit
4453 * again is not run concurrently.
4455 spin_lock(&dest->root_item_lock);
4456 if (dest->send_in_progress) {
4457 spin_unlock(&dest->root_item_lock);
4459 "attempt to delete subvolume %llu during send",
4460 dest->root_key.objectid);
4463 root_flags = btrfs_root_flags(&dest->root_item);
4464 btrfs_set_root_flags(&dest->root_item,
4465 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4466 spin_unlock(&dest->root_item_lock);
4468 down_write(&fs_info->subvol_sem);
4470 ret = may_destroy_subvol(dest);
4474 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4476 * One for dir inode,
4477 * two for dir entries,
4478 * two for root ref/backref.
4480 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4484 trans = btrfs_start_transaction(root, 0);
4485 if (IS_ERR(trans)) {
4486 ret = PTR_ERR(trans);
4489 trans->block_rsv = &block_rsv;
4490 trans->bytes_reserved = block_rsv.size;
4492 btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
4494 ret = btrfs_unlink_subvol(trans, dir, dentry);
4496 btrfs_abort_transaction(trans, ret);
4500 ret = btrfs_record_root_in_trans(trans, dest);
4502 btrfs_abort_transaction(trans, ret);
4506 memset(&dest->root_item.drop_progress, 0,
4507 sizeof(dest->root_item.drop_progress));
4508 btrfs_set_root_drop_level(&dest->root_item, 0);
4509 btrfs_set_root_refs(&dest->root_item, 0);
4511 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4512 ret = btrfs_insert_orphan_item(trans,
4514 dest->root_key.objectid);
4516 btrfs_abort_transaction(trans, ret);
4521 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4522 BTRFS_UUID_KEY_SUBVOL,
4523 dest->root_key.objectid);
4524 if (ret && ret != -ENOENT) {
4525 btrfs_abort_transaction(trans, ret);
4528 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4529 ret = btrfs_uuid_tree_remove(trans,
4530 dest->root_item.received_uuid,
4531 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4532 dest->root_key.objectid);
4533 if (ret && ret != -ENOENT) {
4534 btrfs_abort_transaction(trans, ret);
4539 free_anon_bdev(dest->anon_dev);
4542 trans->block_rsv = NULL;
4543 trans->bytes_reserved = 0;
4544 ret = btrfs_end_transaction(trans);
4545 inode->i_flags |= S_DEAD;
4547 btrfs_subvolume_release_metadata(root, &block_rsv);
4549 up_write(&fs_info->subvol_sem);
4551 spin_lock(&dest->root_item_lock);
4552 root_flags = btrfs_root_flags(&dest->root_item);
4553 btrfs_set_root_flags(&dest->root_item,
4554 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4555 spin_unlock(&dest->root_item_lock);
4557 d_invalidate(dentry);
4558 btrfs_prune_dentries(dest);
4559 ASSERT(dest->send_in_progress == 0);
4565 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4567 struct inode *inode = d_inode(dentry);
4569 struct btrfs_trans_handle *trans;
4570 u64 last_unlink_trans;
4572 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4574 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4575 return btrfs_delete_subvolume(dir, dentry);
4577 trans = __unlink_start_trans(dir);
4579 return PTR_ERR(trans);
4581 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4582 err = btrfs_unlink_subvol(trans, dir, dentry);
4586 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4590 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4592 /* now the directory is empty */
4593 err = btrfs_unlink_inode(trans, BTRFS_I(dir),
4594 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4595 dentry->d_name.len);
4597 btrfs_i_size_write(BTRFS_I(inode), 0);
4599 * Propagate the last_unlink_trans value of the deleted dir to
4600 * its parent directory. This is to prevent an unrecoverable
4601 * log tree in the case we do something like this:
4603 * 2) create snapshot under dir foo
4604 * 3) delete the snapshot
4607 * 6) fsync foo or some file inside foo
4609 if (last_unlink_trans >= trans->transid)
4610 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4613 btrfs_end_transaction(trans);
4614 btrfs_btree_balance_dirty(BTRFS_I(dir)->root->fs_info);
4620 * btrfs_truncate_block - read, zero a chunk and write a block
4621 * @inode - inode that we're zeroing
4622 * @from - the offset to start zeroing
4623 * @len - the length to zero, 0 to zero the entire range respective to the
4625 * @front - zero up to the offset instead of from the offset on
4627 * This will find the block for the "from" offset and cow the block and zero the
4628 * part we want to zero. This is used with truncate and hole punching.
4630 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4633 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4634 struct address_space *mapping = inode->vfs_inode.i_mapping;
4635 struct extent_io_tree *io_tree = &inode->io_tree;
4636 struct btrfs_ordered_extent *ordered;
4637 struct extent_state *cached_state = NULL;
4638 struct extent_changeset *data_reserved = NULL;
4639 bool only_release_metadata = false;
4640 u32 blocksize = fs_info->sectorsize;
4641 pgoff_t index = from >> PAGE_SHIFT;
4642 unsigned offset = from & (blocksize - 1);
4644 gfp_t mask = btrfs_alloc_write_mask(mapping);
4645 size_t write_bytes = blocksize;
4650 if (IS_ALIGNED(offset, blocksize) &&
4651 (!len || IS_ALIGNED(len, blocksize)))
4654 block_start = round_down(from, blocksize);
4655 block_end = block_start + blocksize - 1;
4657 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4660 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes) > 0) {
4661 /* For nocow case, no need to reserve data space */
4662 only_release_metadata = true;
4667 ret = btrfs_delalloc_reserve_metadata(inode, blocksize);
4669 if (!only_release_metadata)
4670 btrfs_free_reserved_data_space(inode, data_reserved,
4671 block_start, blocksize);
4675 page = find_or_create_page(mapping, index, mask);
4677 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4679 btrfs_delalloc_release_extents(inode, blocksize);
4683 ret = set_page_extent_mapped(page);
4687 if (!PageUptodate(page)) {
4688 ret = btrfs_readpage(NULL, page);
4690 if (page->mapping != mapping) {
4695 if (!PageUptodate(page)) {
4700 wait_on_page_writeback(page);
4702 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4704 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4706 unlock_extent_cached(io_tree, block_start, block_end,
4710 btrfs_start_ordered_extent(ordered, 1);
4711 btrfs_put_ordered_extent(ordered);
4715 clear_extent_bit(&inode->io_tree, block_start, block_end,
4716 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4717 0, 0, &cached_state);
4719 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4722 unlock_extent_cached(io_tree, block_start, block_end,
4727 if (offset != blocksize) {
4729 len = blocksize - offset;
4731 memzero_page(page, (block_start - page_offset(page)),
4734 memzero_page(page, (block_start - page_offset(page)) + offset,
4736 flush_dcache_page(page);
4738 btrfs_page_clear_checked(fs_info, page, block_start,
4739 block_end + 1 - block_start);
4740 btrfs_page_set_dirty(fs_info, page, block_start, block_end + 1 - block_start);
4741 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4743 if (only_release_metadata)
4744 set_extent_bit(&inode->io_tree, block_start, block_end,
4745 EXTENT_NORESERVE, 0, NULL, NULL, GFP_NOFS, NULL);
4749 if (only_release_metadata)
4750 btrfs_delalloc_release_metadata(inode, blocksize, true);
4752 btrfs_delalloc_release_space(inode, data_reserved,
4753 block_start, blocksize, true);
4755 btrfs_delalloc_release_extents(inode, blocksize);
4759 if (only_release_metadata)
4760 btrfs_check_nocow_unlock(inode);
4761 extent_changeset_free(data_reserved);
4765 static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode,
4766 u64 offset, u64 len)
4768 struct btrfs_fs_info *fs_info = root->fs_info;
4769 struct btrfs_trans_handle *trans;
4770 struct btrfs_drop_extents_args drop_args = { 0 };
4774 * If NO_HOLES is enabled, we don't need to do anything.
4775 * Later, up in the call chain, either btrfs_set_inode_last_sub_trans()
4776 * or btrfs_update_inode() will be called, which guarantee that the next
4777 * fsync will know this inode was changed and needs to be logged.
4779 if (btrfs_fs_incompat(fs_info, NO_HOLES))
4783 * 1 - for the one we're dropping
4784 * 1 - for the one we're adding
4785 * 1 - for updating the inode.
4787 trans = btrfs_start_transaction(root, 3);
4789 return PTR_ERR(trans);
4791 drop_args.start = offset;
4792 drop_args.end = offset + len;
4793 drop_args.drop_cache = true;
4795 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
4797 btrfs_abort_transaction(trans, ret);
4798 btrfs_end_transaction(trans);
4802 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode),
4803 offset, 0, 0, len, 0, len, 0, 0, 0);
4805 btrfs_abort_transaction(trans, ret);
4807 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
4808 btrfs_update_inode(trans, root, inode);
4810 btrfs_end_transaction(trans);
4815 * This function puts in dummy file extents for the area we're creating a hole
4816 * for. So if we are truncating this file to a larger size we need to insert
4817 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4818 * the range between oldsize and size
4820 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
4822 struct btrfs_root *root = inode->root;
4823 struct btrfs_fs_info *fs_info = root->fs_info;
4824 struct extent_io_tree *io_tree = &inode->io_tree;
4825 struct extent_map *em = NULL;
4826 struct extent_state *cached_state = NULL;
4827 struct extent_map_tree *em_tree = &inode->extent_tree;
4828 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4829 u64 block_end = ALIGN(size, fs_info->sectorsize);
4836 * If our size started in the middle of a block we need to zero out the
4837 * rest of the block before we expand the i_size, otherwise we could
4838 * expose stale data.
4840 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4844 if (size <= hole_start)
4847 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
4849 cur_offset = hole_start;
4851 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4852 block_end - cur_offset);
4858 last_byte = min(extent_map_end(em), block_end);
4859 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4860 hole_size = last_byte - cur_offset;
4862 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4863 struct extent_map *hole_em;
4865 err = maybe_insert_hole(root, inode, cur_offset,
4870 err = btrfs_inode_set_file_extent_range(inode,
4871 cur_offset, hole_size);
4875 btrfs_drop_extent_cache(inode, cur_offset,
4876 cur_offset + hole_size - 1, 0);
4877 hole_em = alloc_extent_map();
4879 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4880 &inode->runtime_flags);
4883 hole_em->start = cur_offset;
4884 hole_em->len = hole_size;
4885 hole_em->orig_start = cur_offset;
4887 hole_em->block_start = EXTENT_MAP_HOLE;
4888 hole_em->block_len = 0;
4889 hole_em->orig_block_len = 0;
4890 hole_em->ram_bytes = hole_size;
4891 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4892 hole_em->generation = fs_info->generation;
4895 write_lock(&em_tree->lock);
4896 err = add_extent_mapping(em_tree, hole_em, 1);
4897 write_unlock(&em_tree->lock);
4900 btrfs_drop_extent_cache(inode, cur_offset,
4904 free_extent_map(hole_em);
4906 err = btrfs_inode_set_file_extent_range(inode,
4907 cur_offset, hole_size);
4912 free_extent_map(em);
4914 cur_offset = last_byte;
4915 if (cur_offset >= block_end)
4918 free_extent_map(em);
4919 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
4923 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4925 struct btrfs_root *root = BTRFS_I(inode)->root;
4926 struct btrfs_trans_handle *trans;
4927 loff_t oldsize = i_size_read(inode);
4928 loff_t newsize = attr->ia_size;
4929 int mask = attr->ia_valid;
4933 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4934 * special case where we need to update the times despite not having
4935 * these flags set. For all other operations the VFS set these flags
4936 * explicitly if it wants a timestamp update.
4938 if (newsize != oldsize) {
4939 inode_inc_iversion(inode);
4940 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4941 inode->i_ctime = inode->i_mtime =
4942 current_time(inode);
4945 if (newsize > oldsize) {
4947 * Don't do an expanding truncate while snapshotting is ongoing.
4948 * This is to ensure the snapshot captures a fully consistent
4949 * state of this file - if the snapshot captures this expanding
4950 * truncation, it must capture all writes that happened before
4953 btrfs_drew_write_lock(&root->snapshot_lock);
4954 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
4956 btrfs_drew_write_unlock(&root->snapshot_lock);
4960 trans = btrfs_start_transaction(root, 1);
4961 if (IS_ERR(trans)) {
4962 btrfs_drew_write_unlock(&root->snapshot_lock);
4963 return PTR_ERR(trans);
4966 i_size_write(inode, newsize);
4967 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
4968 pagecache_isize_extended(inode, oldsize, newsize);
4969 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
4970 btrfs_drew_write_unlock(&root->snapshot_lock);
4971 btrfs_end_transaction(trans);
4973 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4975 if (btrfs_is_zoned(fs_info)) {
4976 ret = btrfs_wait_ordered_range(inode,
4977 ALIGN(newsize, fs_info->sectorsize),
4984 * We're truncating a file that used to have good data down to
4985 * zero. Make sure any new writes to the file get on disk
4989 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
4990 &BTRFS_I(inode)->runtime_flags);
4992 truncate_setsize(inode, newsize);
4994 inode_dio_wait(inode);
4996 ret = btrfs_truncate(inode, newsize == oldsize);
4997 if (ret && inode->i_nlink) {
5001 * Truncate failed, so fix up the in-memory size. We
5002 * adjusted disk_i_size down as we removed extents, so
5003 * wait for disk_i_size to be stable and then update the
5004 * in-memory size to match.
5006 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5009 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5016 static int btrfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
5019 struct inode *inode = d_inode(dentry);
5020 struct btrfs_root *root = BTRFS_I(inode)->root;
5023 if (btrfs_root_readonly(root))
5026 err = setattr_prepare(mnt_userns, dentry, attr);
5030 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5031 err = btrfs_setsize(inode, attr);
5036 if (attr->ia_valid) {
5037 setattr_copy(mnt_userns, inode, attr);
5038 inode_inc_iversion(inode);
5039 err = btrfs_dirty_inode(inode);
5041 if (!err && attr->ia_valid & ATTR_MODE)
5042 err = posix_acl_chmod(mnt_userns, inode, inode->i_mode);
5049 * While truncating the inode pages during eviction, we get the VFS calling
5050 * btrfs_invalidatepage() against each page of the inode. This is slow because
5051 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5052 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5053 * extent_state structures over and over, wasting lots of time.
5055 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5056 * those expensive operations on a per page basis and do only the ordered io
5057 * finishing, while we release here the extent_map and extent_state structures,
5058 * without the excessive merging and splitting.
5060 static void evict_inode_truncate_pages(struct inode *inode)
5062 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5063 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5064 struct rb_node *node;
5066 ASSERT(inode->i_state & I_FREEING);
5067 truncate_inode_pages_final(&inode->i_data);
5069 write_lock(&map_tree->lock);
5070 while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) {
5071 struct extent_map *em;
5073 node = rb_first_cached(&map_tree->map);
5074 em = rb_entry(node, struct extent_map, rb_node);
5075 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5076 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5077 remove_extent_mapping(map_tree, em);
5078 free_extent_map(em);
5079 if (need_resched()) {
5080 write_unlock(&map_tree->lock);
5082 write_lock(&map_tree->lock);
5085 write_unlock(&map_tree->lock);
5088 * Keep looping until we have no more ranges in the io tree.
5089 * We can have ongoing bios started by readahead that have
5090 * their endio callback (extent_io.c:end_bio_extent_readpage)
5091 * still in progress (unlocked the pages in the bio but did not yet
5092 * unlocked the ranges in the io tree). Therefore this means some
5093 * ranges can still be locked and eviction started because before
5094 * submitting those bios, which are executed by a separate task (work
5095 * queue kthread), inode references (inode->i_count) were not taken
5096 * (which would be dropped in the end io callback of each bio).
5097 * Therefore here we effectively end up waiting for those bios and
5098 * anyone else holding locked ranges without having bumped the inode's
5099 * reference count - if we don't do it, when they access the inode's
5100 * io_tree to unlock a range it may be too late, leading to an
5101 * use-after-free issue.
5103 spin_lock(&io_tree->lock);
5104 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5105 struct extent_state *state;
5106 struct extent_state *cached_state = NULL;
5109 unsigned state_flags;
5111 node = rb_first(&io_tree->state);
5112 state = rb_entry(node, struct extent_state, rb_node);
5113 start = state->start;
5115 state_flags = state->state;
5116 spin_unlock(&io_tree->lock);
5118 lock_extent_bits(io_tree, start, end, &cached_state);
5121 * If still has DELALLOC flag, the extent didn't reach disk,
5122 * and its reserved space won't be freed by delayed_ref.
5123 * So we need to free its reserved space here.
5124 * (Refer to comment in btrfs_invalidatepage, case 2)
5126 * Note, end is the bytenr of last byte, so we need + 1 here.
5128 if (state_flags & EXTENT_DELALLOC)
5129 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5132 clear_extent_bit(io_tree, start, end,
5133 EXTENT_LOCKED | EXTENT_DELALLOC |
5134 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
5138 spin_lock(&io_tree->lock);
5140 spin_unlock(&io_tree->lock);
5143 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5144 struct btrfs_block_rsv *rsv)
5146 struct btrfs_fs_info *fs_info = root->fs_info;
5147 struct btrfs_trans_handle *trans;
5148 u64 delayed_refs_extra = btrfs_calc_insert_metadata_size(fs_info, 1);
5152 * Eviction should be taking place at some place safe because of our
5153 * delayed iputs. However the normal flushing code will run delayed
5154 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5156 * We reserve the delayed_refs_extra here again because we can't use
5157 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5158 * above. We reserve our extra bit here because we generate a ton of
5159 * delayed refs activity by truncating.
5161 * BTRFS_RESERVE_FLUSH_EVICT will steal from the global_rsv if it can,
5162 * if we fail to make this reservation we can re-try without the
5163 * delayed_refs_extra so we can make some forward progress.
5165 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size + delayed_refs_extra,
5166 BTRFS_RESERVE_FLUSH_EVICT);
5168 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size,
5169 BTRFS_RESERVE_FLUSH_EVICT);
5172 "could not allocate space for delete; will truncate on mount");
5173 return ERR_PTR(-ENOSPC);
5175 delayed_refs_extra = 0;
5178 trans = btrfs_join_transaction(root);
5182 if (delayed_refs_extra) {
5183 trans->block_rsv = &fs_info->trans_block_rsv;
5184 trans->bytes_reserved = delayed_refs_extra;
5185 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5186 delayed_refs_extra, 1);
5191 void btrfs_evict_inode(struct inode *inode)
5193 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5194 struct btrfs_trans_handle *trans;
5195 struct btrfs_root *root = BTRFS_I(inode)->root;
5196 struct btrfs_block_rsv *rsv;
5199 trace_btrfs_inode_evict(inode);
5202 fsverity_cleanup_inode(inode);
5207 evict_inode_truncate_pages(inode);
5209 if (inode->i_nlink &&
5210 ((btrfs_root_refs(&root->root_item) != 0 &&
5211 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5212 btrfs_is_free_space_inode(BTRFS_I(inode))))
5215 if (is_bad_inode(inode))
5218 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5220 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5223 if (inode->i_nlink > 0) {
5224 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5225 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5230 * This makes sure the inode item in tree is uptodate and the space for
5231 * the inode update is released.
5233 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5238 * This drops any pending insert or delete operations we have for this
5239 * inode. We could have a delayed dir index deletion queued up, but
5240 * we're removing the inode completely so that'll be taken care of in
5243 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
5245 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5248 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5251 btrfs_i_size_write(BTRFS_I(inode), 0);
5254 struct btrfs_truncate_control control = {
5255 .inode = BTRFS_I(inode),
5256 .ino = btrfs_ino(BTRFS_I(inode)),
5261 trans = evict_refill_and_join(root, rsv);
5265 trans->block_rsv = rsv;
5267 ret = btrfs_truncate_inode_items(trans, root, &control);
5268 trans->block_rsv = &fs_info->trans_block_rsv;
5269 btrfs_end_transaction(trans);
5270 btrfs_btree_balance_dirty(fs_info);
5271 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5278 * Errors here aren't a big deal, it just means we leave orphan items in
5279 * the tree. They will be cleaned up on the next mount. If the inode
5280 * number gets reused, cleanup deletes the orphan item without doing
5281 * anything, and unlink reuses the existing orphan item.
5283 * If it turns out that we are dropping too many of these, we might want
5284 * to add a mechanism for retrying these after a commit.
5286 trans = evict_refill_and_join(root, rsv);
5287 if (!IS_ERR(trans)) {
5288 trans->block_rsv = rsv;
5289 btrfs_orphan_del(trans, BTRFS_I(inode));
5290 trans->block_rsv = &fs_info->trans_block_rsv;
5291 btrfs_end_transaction(trans);
5295 btrfs_free_block_rsv(fs_info, rsv);
5298 * If we didn't successfully delete, the orphan item will still be in
5299 * the tree and we'll retry on the next mount. Again, we might also want
5300 * to retry these periodically in the future.
5302 btrfs_remove_delayed_node(BTRFS_I(inode));
5303 fsverity_cleanup_inode(inode);
5308 * Return the key found in the dir entry in the location pointer, fill @type
5309 * with BTRFS_FT_*, and return 0.
5311 * If no dir entries were found, returns -ENOENT.
5312 * If found a corrupted location in dir entry, returns -EUCLEAN.
5314 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5315 struct btrfs_key *location, u8 *type)
5317 const char *name = dentry->d_name.name;
5318 int namelen = dentry->d_name.len;
5319 struct btrfs_dir_item *di;
5320 struct btrfs_path *path;
5321 struct btrfs_root *root = BTRFS_I(dir)->root;
5324 path = btrfs_alloc_path();
5328 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5330 if (IS_ERR_OR_NULL(di)) {
5331 ret = di ? PTR_ERR(di) : -ENOENT;
5335 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5336 if (location->type != BTRFS_INODE_ITEM_KEY &&
5337 location->type != BTRFS_ROOT_ITEM_KEY) {
5339 btrfs_warn(root->fs_info,
5340 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5341 __func__, name, btrfs_ino(BTRFS_I(dir)),
5342 location->objectid, location->type, location->offset);
5345 *type = btrfs_dir_type(path->nodes[0], di);
5347 btrfs_free_path(path);
5352 * when we hit a tree root in a directory, the btrfs part of the inode
5353 * needs to be changed to reflect the root directory of the tree root. This
5354 * is kind of like crossing a mount point.
5356 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5358 struct dentry *dentry,
5359 struct btrfs_key *location,
5360 struct btrfs_root **sub_root)
5362 struct btrfs_path *path;
5363 struct btrfs_root *new_root;
5364 struct btrfs_root_ref *ref;
5365 struct extent_buffer *leaf;
5366 struct btrfs_key key;
5370 path = btrfs_alloc_path();
5377 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5378 key.type = BTRFS_ROOT_REF_KEY;
5379 key.offset = location->objectid;
5381 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5388 leaf = path->nodes[0];
5389 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5390 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5391 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5394 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5395 (unsigned long)(ref + 1),
5396 dentry->d_name.len);
5400 btrfs_release_path(path);
5402 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5403 if (IS_ERR(new_root)) {
5404 err = PTR_ERR(new_root);
5408 *sub_root = new_root;
5409 location->objectid = btrfs_root_dirid(&new_root->root_item);
5410 location->type = BTRFS_INODE_ITEM_KEY;
5411 location->offset = 0;
5414 btrfs_free_path(path);
5418 static void inode_tree_add(struct inode *inode)
5420 struct btrfs_root *root = BTRFS_I(inode)->root;
5421 struct btrfs_inode *entry;
5423 struct rb_node *parent;
5424 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5425 u64 ino = btrfs_ino(BTRFS_I(inode));
5427 if (inode_unhashed(inode))
5430 spin_lock(&root->inode_lock);
5431 p = &root->inode_tree.rb_node;
5434 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5436 if (ino < btrfs_ino(entry))
5437 p = &parent->rb_left;
5438 else if (ino > btrfs_ino(entry))
5439 p = &parent->rb_right;
5441 WARN_ON(!(entry->vfs_inode.i_state &
5442 (I_WILL_FREE | I_FREEING)));
5443 rb_replace_node(parent, new, &root->inode_tree);
5444 RB_CLEAR_NODE(parent);
5445 spin_unlock(&root->inode_lock);
5449 rb_link_node(new, parent, p);
5450 rb_insert_color(new, &root->inode_tree);
5451 spin_unlock(&root->inode_lock);
5454 static void inode_tree_del(struct btrfs_inode *inode)
5456 struct btrfs_root *root = inode->root;
5459 spin_lock(&root->inode_lock);
5460 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5461 rb_erase(&inode->rb_node, &root->inode_tree);
5462 RB_CLEAR_NODE(&inode->rb_node);
5463 empty = RB_EMPTY_ROOT(&root->inode_tree);
5465 spin_unlock(&root->inode_lock);
5467 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5468 spin_lock(&root->inode_lock);
5469 empty = RB_EMPTY_ROOT(&root->inode_tree);
5470 spin_unlock(&root->inode_lock);
5472 btrfs_add_dead_root(root);
5477 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5479 struct btrfs_iget_args *args = p;
5481 inode->i_ino = args->ino;
5482 BTRFS_I(inode)->location.objectid = args->ino;
5483 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5484 BTRFS_I(inode)->location.offset = 0;
5485 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5486 BUG_ON(args->root && !BTRFS_I(inode)->root);
5490 static int btrfs_find_actor(struct inode *inode, void *opaque)
5492 struct btrfs_iget_args *args = opaque;
5494 return args->ino == BTRFS_I(inode)->location.objectid &&
5495 args->root == BTRFS_I(inode)->root;
5498 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5499 struct btrfs_root *root)
5501 struct inode *inode;
5502 struct btrfs_iget_args args;
5503 unsigned long hashval = btrfs_inode_hash(ino, root);
5508 inode = iget5_locked(s, hashval, btrfs_find_actor,
5509 btrfs_init_locked_inode,
5515 * Get an inode object given its inode number and corresponding root.
5516 * Path can be preallocated to prevent recursing back to iget through
5517 * allocator. NULL is also valid but may require an additional allocation
5520 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5521 struct btrfs_root *root, struct btrfs_path *path)
5523 struct inode *inode;
5525 inode = btrfs_iget_locked(s, ino, root);
5527 return ERR_PTR(-ENOMEM);
5529 if (inode->i_state & I_NEW) {
5532 ret = btrfs_read_locked_inode(inode, path);
5534 inode_tree_add(inode);
5535 unlock_new_inode(inode);
5539 * ret > 0 can come from btrfs_search_slot called by
5540 * btrfs_read_locked_inode, this means the inode item
5545 inode = ERR_PTR(ret);
5552 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5554 return btrfs_iget_path(s, ino, root, NULL);
5557 static struct inode *new_simple_dir(struct super_block *s,
5558 struct btrfs_key *key,
5559 struct btrfs_root *root)
5561 struct inode *inode = new_inode(s);
5564 return ERR_PTR(-ENOMEM);
5566 BTRFS_I(inode)->root = btrfs_grab_root(root);
5567 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5568 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5570 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5572 * We only need lookup, the rest is read-only and there's no inode
5573 * associated with the dentry
5575 inode->i_op = &simple_dir_inode_operations;
5576 inode->i_opflags &= ~IOP_XATTR;
5577 inode->i_fop = &simple_dir_operations;
5578 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5579 inode->i_mtime = current_time(inode);
5580 inode->i_atime = inode->i_mtime;
5581 inode->i_ctime = inode->i_mtime;
5582 BTRFS_I(inode)->i_otime = inode->i_mtime;
5587 static inline u8 btrfs_inode_type(struct inode *inode)
5590 * Compile-time asserts that generic FT_* types still match
5593 BUILD_BUG_ON(BTRFS_FT_UNKNOWN != FT_UNKNOWN);
5594 BUILD_BUG_ON(BTRFS_FT_REG_FILE != FT_REG_FILE);
5595 BUILD_BUG_ON(BTRFS_FT_DIR != FT_DIR);
5596 BUILD_BUG_ON(BTRFS_FT_CHRDEV != FT_CHRDEV);
5597 BUILD_BUG_ON(BTRFS_FT_BLKDEV != FT_BLKDEV);
5598 BUILD_BUG_ON(BTRFS_FT_FIFO != FT_FIFO);
5599 BUILD_BUG_ON(BTRFS_FT_SOCK != FT_SOCK);
5600 BUILD_BUG_ON(BTRFS_FT_SYMLINK != FT_SYMLINK);
5602 return fs_umode_to_ftype(inode->i_mode);
5605 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5607 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5608 struct inode *inode;
5609 struct btrfs_root *root = BTRFS_I(dir)->root;
5610 struct btrfs_root *sub_root = root;
5611 struct btrfs_key location;
5615 if (dentry->d_name.len > BTRFS_NAME_LEN)
5616 return ERR_PTR(-ENAMETOOLONG);
5618 ret = btrfs_inode_by_name(dir, dentry, &location, &di_type);
5620 return ERR_PTR(ret);
5622 if (location.type == BTRFS_INODE_ITEM_KEY) {
5623 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5627 /* Do extra check against inode mode with di_type */
5628 if (btrfs_inode_type(inode) != di_type) {
5630 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5631 inode->i_mode, btrfs_inode_type(inode),
5634 return ERR_PTR(-EUCLEAN);
5639 ret = fixup_tree_root_location(fs_info, dir, dentry,
5640 &location, &sub_root);
5643 inode = ERR_PTR(ret);
5645 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5647 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5649 if (root != sub_root)
5650 btrfs_put_root(sub_root);
5652 if (!IS_ERR(inode) && root != sub_root) {
5653 down_read(&fs_info->cleanup_work_sem);
5654 if (!sb_rdonly(inode->i_sb))
5655 ret = btrfs_orphan_cleanup(sub_root);
5656 up_read(&fs_info->cleanup_work_sem);
5659 inode = ERR_PTR(ret);
5666 static int btrfs_dentry_delete(const struct dentry *dentry)
5668 struct btrfs_root *root;
5669 struct inode *inode = d_inode(dentry);
5671 if (!inode && !IS_ROOT(dentry))
5672 inode = d_inode(dentry->d_parent);
5675 root = BTRFS_I(inode)->root;
5676 if (btrfs_root_refs(&root->root_item) == 0)
5679 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5685 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5688 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5690 if (inode == ERR_PTR(-ENOENT))
5692 return d_splice_alias(inode, dentry);
5696 * All this infrastructure exists because dir_emit can fault, and we are holding
5697 * the tree lock when doing readdir. For now just allocate a buffer and copy
5698 * our information into that, and then dir_emit from the buffer. This is
5699 * similar to what NFS does, only we don't keep the buffer around in pagecache
5700 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5701 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5704 static int btrfs_opendir(struct inode *inode, struct file *file)
5706 struct btrfs_file_private *private;
5708 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5711 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5712 if (!private->filldir_buf) {
5716 file->private_data = private;
5727 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5730 struct dir_entry *entry = addr;
5731 char *name = (char *)(entry + 1);
5733 ctx->pos = get_unaligned(&entry->offset);
5734 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5735 get_unaligned(&entry->ino),
5736 get_unaligned(&entry->type)))
5738 addr += sizeof(struct dir_entry) +
5739 get_unaligned(&entry->name_len);
5745 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5747 struct inode *inode = file_inode(file);
5748 struct btrfs_root *root = BTRFS_I(inode)->root;
5749 struct btrfs_file_private *private = file->private_data;
5750 struct btrfs_dir_item *di;
5751 struct btrfs_key key;
5752 struct btrfs_key found_key;
5753 struct btrfs_path *path;
5755 struct list_head ins_list;
5756 struct list_head del_list;
5758 struct extent_buffer *leaf;
5765 struct btrfs_key location;
5767 if (!dir_emit_dots(file, ctx))
5770 path = btrfs_alloc_path();
5774 addr = private->filldir_buf;
5775 path->reada = READA_FORWARD;
5777 INIT_LIST_HEAD(&ins_list);
5778 INIT_LIST_HEAD(&del_list);
5779 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5782 key.type = BTRFS_DIR_INDEX_KEY;
5783 key.offset = ctx->pos;
5784 key.objectid = btrfs_ino(BTRFS_I(inode));
5786 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5791 struct dir_entry *entry;
5793 leaf = path->nodes[0];
5794 slot = path->slots[0];
5795 if (slot >= btrfs_header_nritems(leaf)) {
5796 ret = btrfs_next_leaf(root, path);
5804 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5806 if (found_key.objectid != key.objectid)
5808 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5810 if (found_key.offset < ctx->pos)
5812 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5814 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5815 name_len = btrfs_dir_name_len(leaf, di);
5816 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5818 btrfs_release_path(path);
5819 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5822 addr = private->filldir_buf;
5829 put_unaligned(name_len, &entry->name_len);
5830 name_ptr = (char *)(entry + 1);
5831 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5833 put_unaligned(fs_ftype_to_dtype(btrfs_dir_type(leaf, di)),
5835 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5836 put_unaligned(location.objectid, &entry->ino);
5837 put_unaligned(found_key.offset, &entry->offset);
5839 addr += sizeof(struct dir_entry) + name_len;
5840 total_len += sizeof(struct dir_entry) + name_len;
5844 btrfs_release_path(path);
5846 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5850 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5855 * Stop new entries from being returned after we return the last
5858 * New directory entries are assigned a strictly increasing
5859 * offset. This means that new entries created during readdir
5860 * are *guaranteed* to be seen in the future by that readdir.
5861 * This has broken buggy programs which operate on names as
5862 * they're returned by readdir. Until we re-use freed offsets
5863 * we have this hack to stop new entries from being returned
5864 * under the assumption that they'll never reach this huge
5867 * This is being careful not to overflow 32bit loff_t unless the
5868 * last entry requires it because doing so has broken 32bit apps
5871 if (ctx->pos >= INT_MAX)
5872 ctx->pos = LLONG_MAX;
5879 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5880 btrfs_free_path(path);
5885 * This is somewhat expensive, updating the tree every time the
5886 * inode changes. But, it is most likely to find the inode in cache.
5887 * FIXME, needs more benchmarking...there are no reasons other than performance
5888 * to keep or drop this code.
5890 static int btrfs_dirty_inode(struct inode *inode)
5892 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5893 struct btrfs_root *root = BTRFS_I(inode)->root;
5894 struct btrfs_trans_handle *trans;
5897 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5900 trans = btrfs_join_transaction(root);
5902 return PTR_ERR(trans);
5904 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5905 if (ret && (ret == -ENOSPC || ret == -EDQUOT)) {
5906 /* whoops, lets try again with the full transaction */
5907 btrfs_end_transaction(trans);
5908 trans = btrfs_start_transaction(root, 1);
5910 return PTR_ERR(trans);
5912 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5914 btrfs_end_transaction(trans);
5915 if (BTRFS_I(inode)->delayed_node)
5916 btrfs_balance_delayed_items(fs_info);
5922 * This is a copy of file_update_time. We need this so we can return error on
5923 * ENOSPC for updating the inode in the case of file write and mmap writes.
5925 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
5928 struct btrfs_root *root = BTRFS_I(inode)->root;
5929 bool dirty = flags & ~S_VERSION;
5931 if (btrfs_root_readonly(root))
5934 if (flags & S_VERSION)
5935 dirty |= inode_maybe_inc_iversion(inode, dirty);
5936 if (flags & S_CTIME)
5937 inode->i_ctime = *now;
5938 if (flags & S_MTIME)
5939 inode->i_mtime = *now;
5940 if (flags & S_ATIME)
5941 inode->i_atime = *now;
5942 return dirty ? btrfs_dirty_inode(inode) : 0;
5946 * find the highest existing sequence number in a directory
5947 * and then set the in-memory index_cnt variable to reflect
5948 * free sequence numbers
5950 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
5952 struct btrfs_root *root = inode->root;
5953 struct btrfs_key key, found_key;
5954 struct btrfs_path *path;
5955 struct extent_buffer *leaf;
5958 key.objectid = btrfs_ino(inode);
5959 key.type = BTRFS_DIR_INDEX_KEY;
5960 key.offset = (u64)-1;
5962 path = btrfs_alloc_path();
5966 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5969 /* FIXME: we should be able to handle this */
5975 * MAGIC NUMBER EXPLANATION:
5976 * since we search a directory based on f_pos we have to start at 2
5977 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5978 * else has to start at 2
5980 if (path->slots[0] == 0) {
5981 inode->index_cnt = 2;
5987 leaf = path->nodes[0];
5988 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5990 if (found_key.objectid != btrfs_ino(inode) ||
5991 found_key.type != BTRFS_DIR_INDEX_KEY) {
5992 inode->index_cnt = 2;
5996 inode->index_cnt = found_key.offset + 1;
5998 btrfs_free_path(path);
6003 * helper to find a free sequence number in a given directory. This current
6004 * code is very simple, later versions will do smarter things in the btree
6006 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6010 if (dir->index_cnt == (u64)-1) {
6011 ret = btrfs_inode_delayed_dir_index_count(dir);
6013 ret = btrfs_set_inode_index_count(dir);
6019 *index = dir->index_cnt;
6025 static int btrfs_insert_inode_locked(struct inode *inode)
6027 struct btrfs_iget_args args;
6029 args.ino = BTRFS_I(inode)->location.objectid;
6030 args.root = BTRFS_I(inode)->root;
6032 return insert_inode_locked4(inode,
6033 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6034 btrfs_find_actor, &args);
6038 * Inherit flags from the parent inode.
6040 * Currently only the compression flags and the cow flags are inherited.
6042 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6049 flags = BTRFS_I(dir)->flags;
6051 if (flags & BTRFS_INODE_NOCOMPRESS) {
6052 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6053 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6054 } else if (flags & BTRFS_INODE_COMPRESS) {
6055 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6056 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6059 if (flags & BTRFS_INODE_NODATACOW) {
6060 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6061 if (S_ISREG(inode->i_mode))
6062 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6065 btrfs_sync_inode_flags_to_i_flags(inode);
6068 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6069 struct btrfs_root *root,
6070 struct user_namespace *mnt_userns,
6072 const char *name, int name_len,
6073 u64 ref_objectid, u64 objectid,
6074 umode_t mode, u64 *index)
6076 struct btrfs_fs_info *fs_info = root->fs_info;
6077 struct inode *inode;
6078 struct btrfs_inode_item *inode_item;
6079 struct btrfs_key *location;
6080 struct btrfs_path *path;
6081 struct btrfs_inode_ref *ref;
6082 struct btrfs_key key[2];
6084 struct btrfs_item_batch batch;
6086 unsigned int nofs_flag;
6089 path = btrfs_alloc_path();
6091 return ERR_PTR(-ENOMEM);
6093 nofs_flag = memalloc_nofs_save();
6094 inode = new_inode(fs_info->sb);
6095 memalloc_nofs_restore(nofs_flag);
6097 btrfs_free_path(path);
6098 return ERR_PTR(-ENOMEM);
6102 * O_TMPFILE, set link count to 0, so that after this point,
6103 * we fill in an inode item with the correct link count.
6106 set_nlink(inode, 0);
6109 * we have to initialize this early, so we can reclaim the inode
6110 * number if we fail afterwards in this function.
6112 inode->i_ino = objectid;
6115 trace_btrfs_inode_request(dir);
6117 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6119 btrfs_free_path(path);
6121 return ERR_PTR(ret);
6127 * index_cnt is ignored for everything but a dir,
6128 * btrfs_set_inode_index_count has an explanation for the magic
6131 BTRFS_I(inode)->index_cnt = 2;
6132 BTRFS_I(inode)->dir_index = *index;
6133 BTRFS_I(inode)->root = btrfs_grab_root(root);
6134 BTRFS_I(inode)->generation = trans->transid;
6135 inode->i_generation = BTRFS_I(inode)->generation;
6138 * We could have gotten an inode number from somebody who was fsynced
6139 * and then removed in this same transaction, so let's just set full
6140 * sync since it will be a full sync anyway and this will blow away the
6141 * old info in the log.
6143 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6145 key[0].objectid = objectid;
6146 key[0].type = BTRFS_INODE_ITEM_KEY;
6149 sizes[0] = sizeof(struct btrfs_inode_item);
6153 * Start new inodes with an inode_ref. This is slightly more
6154 * efficient for small numbers of hard links since they will
6155 * be packed into one item. Extended refs will kick in if we
6156 * add more hard links than can fit in the ref item.
6158 key[1].objectid = objectid;
6159 key[1].type = BTRFS_INODE_REF_KEY;
6160 key[1].offset = ref_objectid;
6162 sizes[1] = name_len + sizeof(*ref);
6165 location = &BTRFS_I(inode)->location;
6166 location->objectid = objectid;
6167 location->offset = 0;
6168 location->type = BTRFS_INODE_ITEM_KEY;
6170 ret = btrfs_insert_inode_locked(inode);
6176 batch.keys = &key[0];
6177 batch.data_sizes = &sizes[0];
6178 batch.total_data_size = sizes[0] + (name ? sizes[1] : 0);
6179 batch.nr = name ? 2 : 1;
6180 ret = btrfs_insert_empty_items(trans, root, path, &batch);
6184 inode_init_owner(mnt_userns, inode, dir, mode);
6185 inode_set_bytes(inode, 0);
6187 inode->i_mtime = current_time(inode);
6188 inode->i_atime = inode->i_mtime;
6189 inode->i_ctime = inode->i_mtime;
6190 BTRFS_I(inode)->i_otime = inode->i_mtime;
6192 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6193 struct btrfs_inode_item);
6194 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6195 sizeof(*inode_item));
6196 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6199 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6200 struct btrfs_inode_ref);
6201 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6202 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6203 ptr = (unsigned long)(ref + 1);
6204 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6207 btrfs_mark_buffer_dirty(path->nodes[0]);
6208 btrfs_free_path(path);
6210 btrfs_inherit_iflags(inode, dir);
6212 if (S_ISREG(mode)) {
6213 if (btrfs_test_opt(fs_info, NODATASUM))
6214 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6215 if (btrfs_test_opt(fs_info, NODATACOW))
6216 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6217 BTRFS_INODE_NODATASUM;
6220 inode_tree_add(inode);
6222 trace_btrfs_inode_new(inode);
6223 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6225 btrfs_update_root_times(trans, root);
6227 ret = btrfs_inode_inherit_props(trans, inode, dir);
6230 "error inheriting props for ino %llu (root %llu): %d",
6231 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6236 discard_new_inode(inode);
6239 BTRFS_I(dir)->index_cnt--;
6240 btrfs_free_path(path);
6241 return ERR_PTR(ret);
6245 * utility function to add 'inode' into 'parent_inode' with
6246 * a give name and a given sequence number.
6247 * if 'add_backref' is true, also insert a backref from the
6248 * inode to the parent directory.
6250 int btrfs_add_link(struct btrfs_trans_handle *trans,
6251 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6252 const char *name, int name_len, int add_backref, u64 index)
6255 struct btrfs_key key;
6256 struct btrfs_root *root = parent_inode->root;
6257 u64 ino = btrfs_ino(inode);
6258 u64 parent_ino = btrfs_ino(parent_inode);
6260 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6261 memcpy(&key, &inode->root->root_key, sizeof(key));
6264 key.type = BTRFS_INODE_ITEM_KEY;
6268 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6269 ret = btrfs_add_root_ref(trans, key.objectid,
6270 root->root_key.objectid, parent_ino,
6271 index, name, name_len);
6272 } else if (add_backref) {
6273 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6277 /* Nothing to clean up yet */
6281 ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key,
6282 btrfs_inode_type(&inode->vfs_inode), index);
6283 if (ret == -EEXIST || ret == -EOVERFLOW)
6286 btrfs_abort_transaction(trans, ret);
6290 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6292 inode_inc_iversion(&parent_inode->vfs_inode);
6294 * If we are replaying a log tree, we do not want to update the mtime
6295 * and ctime of the parent directory with the current time, since the
6296 * log replay procedure is responsible for setting them to their correct
6297 * values (the ones it had when the fsync was done).
6299 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
6300 struct timespec64 now = current_time(&parent_inode->vfs_inode);
6302 parent_inode->vfs_inode.i_mtime = now;
6303 parent_inode->vfs_inode.i_ctime = now;
6305 ret = btrfs_update_inode(trans, root, parent_inode);
6307 btrfs_abort_transaction(trans, ret);
6311 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6314 err = btrfs_del_root_ref(trans, key.objectid,
6315 root->root_key.objectid, parent_ino,
6316 &local_index, name, name_len);
6318 btrfs_abort_transaction(trans, err);
6319 } else if (add_backref) {
6323 err = btrfs_del_inode_ref(trans, root, name, name_len,
6324 ino, parent_ino, &local_index);
6326 btrfs_abort_transaction(trans, err);
6329 /* Return the original error code */
6333 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6334 struct btrfs_inode *dir, struct dentry *dentry,
6335 struct btrfs_inode *inode, int backref, u64 index)
6337 int err = btrfs_add_link(trans, dir, inode,
6338 dentry->d_name.name, dentry->d_name.len,
6345 static int btrfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
6346 struct dentry *dentry, umode_t mode, dev_t rdev)
6348 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6349 struct btrfs_trans_handle *trans;
6350 struct btrfs_root *root = BTRFS_I(dir)->root;
6351 struct inode *inode = NULL;
6357 * 2 for inode item and ref
6359 * 1 for xattr if selinux is on
6361 trans = btrfs_start_transaction(root, 5);
6363 return PTR_ERR(trans);
6365 err = btrfs_get_free_objectid(root, &objectid);
6369 inode = btrfs_new_inode(trans, root, mnt_userns, dir,
6370 dentry->d_name.name, dentry->d_name.len,
6371 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
6372 if (IS_ERR(inode)) {
6373 err = PTR_ERR(inode);
6379 * If the active LSM wants to access the inode during
6380 * d_instantiate it needs these. Smack checks to see
6381 * if the filesystem supports xattrs by looking at the
6384 inode->i_op = &btrfs_special_inode_operations;
6385 init_special_inode(inode, inode->i_mode, rdev);
6387 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6391 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6396 btrfs_update_inode(trans, root, BTRFS_I(inode));
6397 d_instantiate_new(dentry, inode);
6400 btrfs_end_transaction(trans);
6401 btrfs_btree_balance_dirty(fs_info);
6403 inode_dec_link_count(inode);
6404 discard_new_inode(inode);
6409 static int btrfs_create(struct user_namespace *mnt_userns, struct inode *dir,
6410 struct dentry *dentry, umode_t mode, bool excl)
6412 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6413 struct btrfs_trans_handle *trans;
6414 struct btrfs_root *root = BTRFS_I(dir)->root;
6415 struct inode *inode = NULL;
6421 * 2 for inode item and ref
6423 * 1 for xattr if selinux is on
6425 trans = btrfs_start_transaction(root, 5);
6427 return PTR_ERR(trans);
6429 err = btrfs_get_free_objectid(root, &objectid);
6433 inode = btrfs_new_inode(trans, root, mnt_userns, dir,
6434 dentry->d_name.name, dentry->d_name.len,
6435 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
6436 if (IS_ERR(inode)) {
6437 err = PTR_ERR(inode);
6442 * If the active LSM wants to access the inode during
6443 * d_instantiate it needs these. Smack checks to see
6444 * if the filesystem supports xattrs by looking at the
6447 inode->i_fop = &btrfs_file_operations;
6448 inode->i_op = &btrfs_file_inode_operations;
6449 inode->i_mapping->a_ops = &btrfs_aops;
6451 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6455 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6459 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6464 d_instantiate_new(dentry, inode);
6467 btrfs_end_transaction(trans);
6469 inode_dec_link_count(inode);
6470 discard_new_inode(inode);
6472 btrfs_btree_balance_dirty(fs_info);
6476 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6477 struct dentry *dentry)
6479 struct btrfs_trans_handle *trans = NULL;
6480 struct btrfs_root *root = BTRFS_I(dir)->root;
6481 struct inode *inode = d_inode(old_dentry);
6482 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6487 /* do not allow sys_link's with other subvols of the same device */
6488 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6491 if (inode->i_nlink >= BTRFS_LINK_MAX)
6494 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6499 * 2 items for inode and inode ref
6500 * 2 items for dir items
6501 * 1 item for parent inode
6502 * 1 item for orphan item deletion if O_TMPFILE
6504 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6505 if (IS_ERR(trans)) {
6506 err = PTR_ERR(trans);
6511 /* There are several dir indexes for this inode, clear the cache. */
6512 BTRFS_I(inode)->dir_index = 0ULL;
6514 inode_inc_iversion(inode);
6515 inode->i_ctime = current_time(inode);
6517 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6519 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6525 struct dentry *parent = dentry->d_parent;
6527 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6530 if (inode->i_nlink == 1) {
6532 * If new hard link count is 1, it's a file created
6533 * with open(2) O_TMPFILE flag.
6535 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6539 d_instantiate(dentry, inode);
6540 btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent);
6545 btrfs_end_transaction(trans);
6547 inode_dec_link_count(inode);
6550 btrfs_btree_balance_dirty(fs_info);
6554 static int btrfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
6555 struct dentry *dentry, umode_t mode)
6557 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6558 struct inode *inode = NULL;
6559 struct btrfs_trans_handle *trans;
6560 struct btrfs_root *root = BTRFS_I(dir)->root;
6566 * 2 items for inode and ref
6567 * 2 items for dir items
6568 * 1 for xattr if selinux is on
6570 trans = btrfs_start_transaction(root, 5);
6572 return PTR_ERR(trans);
6574 err = btrfs_get_free_objectid(root, &objectid);
6578 inode = btrfs_new_inode(trans, root, mnt_userns, dir,
6579 dentry->d_name.name, dentry->d_name.len,
6580 btrfs_ino(BTRFS_I(dir)), objectid,
6581 S_IFDIR | mode, &index);
6582 if (IS_ERR(inode)) {
6583 err = PTR_ERR(inode);
6588 /* these must be set before we unlock the inode */
6589 inode->i_op = &btrfs_dir_inode_operations;
6590 inode->i_fop = &btrfs_dir_file_operations;
6592 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6596 btrfs_i_size_write(BTRFS_I(inode), 0);
6597 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6601 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6602 dentry->d_name.name,
6603 dentry->d_name.len, 0, index);
6607 d_instantiate_new(dentry, inode);
6610 btrfs_end_transaction(trans);
6612 inode_dec_link_count(inode);
6613 discard_new_inode(inode);
6615 btrfs_btree_balance_dirty(fs_info);
6619 static noinline int uncompress_inline(struct btrfs_path *path,
6621 size_t pg_offset, u64 extent_offset,
6622 struct btrfs_file_extent_item *item)
6625 struct extent_buffer *leaf = path->nodes[0];
6628 unsigned long inline_size;
6632 WARN_ON(pg_offset != 0);
6633 compress_type = btrfs_file_extent_compression(leaf, item);
6634 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6635 inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]);
6636 tmp = kmalloc(inline_size, GFP_NOFS);
6639 ptr = btrfs_file_extent_inline_start(item);
6641 read_extent_buffer(leaf, tmp, ptr, inline_size);
6643 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6644 ret = btrfs_decompress(compress_type, tmp, page,
6645 extent_offset, inline_size, max_size);
6648 * decompression code contains a memset to fill in any space between the end
6649 * of the uncompressed data and the end of max_size in case the decompressed
6650 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6651 * the end of an inline extent and the beginning of the next block, so we
6652 * cover that region here.
6655 if (max_size + pg_offset < PAGE_SIZE)
6656 memzero_page(page, pg_offset + max_size,
6657 PAGE_SIZE - max_size - pg_offset);
6663 * btrfs_get_extent - Lookup the first extent overlapping a range in a file.
6664 * @inode: file to search in
6665 * @page: page to read extent data into if the extent is inline
6666 * @pg_offset: offset into @page to copy to
6667 * @start: file offset
6668 * @len: length of range starting at @start
6670 * This returns the first &struct extent_map which overlaps with the given
6671 * range, reading it from the B-tree and caching it if necessary. Note that
6672 * there may be more extents which overlap the given range after the returned
6675 * If @page is not NULL and the extent is inline, this also reads the extent
6676 * data directly into the page and marks the extent up to date in the io_tree.
6678 * Return: ERR_PTR on error, non-NULL extent_map on success.
6680 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6681 struct page *page, size_t pg_offset,
6684 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6686 u64 extent_start = 0;
6688 u64 objectid = btrfs_ino(inode);
6689 int extent_type = -1;
6690 struct btrfs_path *path = NULL;
6691 struct btrfs_root *root = inode->root;
6692 struct btrfs_file_extent_item *item;
6693 struct extent_buffer *leaf;
6694 struct btrfs_key found_key;
6695 struct extent_map *em = NULL;
6696 struct extent_map_tree *em_tree = &inode->extent_tree;
6697 struct extent_io_tree *io_tree = &inode->io_tree;
6699 read_lock(&em_tree->lock);
6700 em = lookup_extent_mapping(em_tree, start, len);
6701 read_unlock(&em_tree->lock);
6704 if (em->start > start || em->start + em->len <= start)
6705 free_extent_map(em);
6706 else if (em->block_start == EXTENT_MAP_INLINE && page)
6707 free_extent_map(em);
6711 em = alloc_extent_map();
6716 em->start = EXTENT_MAP_HOLE;
6717 em->orig_start = EXTENT_MAP_HOLE;
6719 em->block_len = (u64)-1;
6721 path = btrfs_alloc_path();
6727 /* Chances are we'll be called again, so go ahead and do readahead */
6728 path->reada = READA_FORWARD;
6731 * The same explanation in load_free_space_cache applies here as well,
6732 * we only read when we're loading the free space cache, and at that
6733 * point the commit_root has everything we need.
6735 if (btrfs_is_free_space_inode(inode)) {
6736 path->search_commit_root = 1;
6737 path->skip_locking = 1;
6740 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6743 } else if (ret > 0) {
6744 if (path->slots[0] == 0)
6750 leaf = path->nodes[0];
6751 item = btrfs_item_ptr(leaf, path->slots[0],
6752 struct btrfs_file_extent_item);
6753 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6754 if (found_key.objectid != objectid ||
6755 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6757 * If we backup past the first extent we want to move forward
6758 * and see if there is an extent in front of us, otherwise we'll
6759 * say there is a hole for our whole search range which can
6766 extent_type = btrfs_file_extent_type(leaf, item);
6767 extent_start = found_key.offset;
6768 extent_end = btrfs_file_extent_end(path);
6769 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6770 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6771 /* Only regular file could have regular/prealloc extent */
6772 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6775 "regular/prealloc extent found for non-regular inode %llu",
6779 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6781 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6782 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6787 if (start >= extent_end) {
6789 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6790 ret = btrfs_next_leaf(root, path);
6796 leaf = path->nodes[0];
6798 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6799 if (found_key.objectid != objectid ||
6800 found_key.type != BTRFS_EXTENT_DATA_KEY)
6802 if (start + len <= found_key.offset)
6804 if (start > found_key.offset)
6807 /* New extent overlaps with existing one */
6809 em->orig_start = start;
6810 em->len = found_key.offset - start;
6811 em->block_start = EXTENT_MAP_HOLE;
6815 btrfs_extent_item_to_extent_map(inode, path, item, !page, em);
6817 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6818 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6820 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6824 size_t extent_offset;
6830 size = btrfs_file_extent_ram_bytes(leaf, item);
6831 extent_offset = page_offset(page) + pg_offset - extent_start;
6832 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6833 size - extent_offset);
6834 em->start = extent_start + extent_offset;
6835 em->len = ALIGN(copy_size, fs_info->sectorsize);
6836 em->orig_block_len = em->len;
6837 em->orig_start = em->start;
6838 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6840 if (!PageUptodate(page)) {
6841 if (btrfs_file_extent_compression(leaf, item) !=
6842 BTRFS_COMPRESS_NONE) {
6843 ret = uncompress_inline(path, page, pg_offset,
6844 extent_offset, item);
6848 map = kmap_local_page(page);
6849 read_extent_buffer(leaf, map + pg_offset, ptr,
6851 if (pg_offset + copy_size < PAGE_SIZE) {
6852 memset(map + pg_offset + copy_size, 0,
6853 PAGE_SIZE - pg_offset -
6858 flush_dcache_page(page);
6860 set_extent_uptodate(io_tree, em->start,
6861 extent_map_end(em) - 1, NULL, GFP_NOFS);
6866 em->orig_start = start;
6868 em->block_start = EXTENT_MAP_HOLE;
6871 btrfs_release_path(path);
6872 if (em->start > start || extent_map_end(em) <= start) {
6874 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6875 em->start, em->len, start, len);
6880 write_lock(&em_tree->lock);
6881 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6882 write_unlock(&em_tree->lock);
6884 btrfs_free_path(path);
6886 trace_btrfs_get_extent(root, inode, em);
6889 free_extent_map(em);
6890 return ERR_PTR(ret);
6895 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
6898 struct extent_map *em;
6899 struct extent_map *hole_em = NULL;
6900 u64 delalloc_start = start;
6906 em = btrfs_get_extent(inode, NULL, 0, start, len);
6910 * If our em maps to:
6912 * - a pre-alloc extent,
6913 * there might actually be delalloc bytes behind it.
6915 if (em->block_start != EXTENT_MAP_HOLE &&
6916 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6921 /* check to see if we've wrapped (len == -1 or similar) */
6930 /* ok, we didn't find anything, lets look for delalloc */
6931 delalloc_len = count_range_bits(&inode->io_tree, &delalloc_start,
6932 end, len, EXTENT_DELALLOC, 1);
6933 delalloc_end = delalloc_start + delalloc_len;
6934 if (delalloc_end < delalloc_start)
6935 delalloc_end = (u64)-1;
6938 * We didn't find anything useful, return the original results from
6941 if (delalloc_start > end || delalloc_end <= start) {
6948 * Adjust the delalloc_start to make sure it doesn't go backwards from
6949 * the start they passed in
6951 delalloc_start = max(start, delalloc_start);
6952 delalloc_len = delalloc_end - delalloc_start;
6954 if (delalloc_len > 0) {
6957 const u64 hole_end = extent_map_end(hole_em);
6959 em = alloc_extent_map();
6967 * When btrfs_get_extent can't find anything it returns one
6970 * Make sure what it found really fits our range, and adjust to
6971 * make sure it is based on the start from the caller
6973 if (hole_end <= start || hole_em->start > end) {
6974 free_extent_map(hole_em);
6977 hole_start = max(hole_em->start, start);
6978 hole_len = hole_end - hole_start;
6981 if (hole_em && delalloc_start > hole_start) {
6983 * Our hole starts before our delalloc, so we have to
6984 * return just the parts of the hole that go until the
6987 em->len = min(hole_len, delalloc_start - hole_start);
6988 em->start = hole_start;
6989 em->orig_start = hole_start;
6991 * Don't adjust block start at all, it is fixed at
6994 em->block_start = hole_em->block_start;
6995 em->block_len = hole_len;
6996 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6997 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7000 * Hole is out of passed range or it starts after
7003 em->start = delalloc_start;
7004 em->len = delalloc_len;
7005 em->orig_start = delalloc_start;
7006 em->block_start = EXTENT_MAP_DELALLOC;
7007 em->block_len = delalloc_len;
7014 free_extent_map(hole_em);
7016 free_extent_map(em);
7017 return ERR_PTR(err);
7022 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
7025 const u64 orig_start,
7026 const u64 block_start,
7027 const u64 block_len,
7028 const u64 orig_block_len,
7029 const u64 ram_bytes,
7032 struct extent_map *em = NULL;
7035 if (type != BTRFS_ORDERED_NOCOW) {
7036 em = create_io_em(inode, start, len, orig_start, block_start,
7037 block_len, orig_block_len, ram_bytes,
7038 BTRFS_COMPRESS_NONE, /* compress_type */
7043 ret = btrfs_add_ordered_extent_dio(inode, start, block_start, len,
7047 free_extent_map(em);
7048 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
7057 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
7060 struct btrfs_root *root = inode->root;
7061 struct btrfs_fs_info *fs_info = root->fs_info;
7062 struct extent_map *em;
7063 struct btrfs_key ins;
7067 alloc_hint = get_extent_allocation_hint(inode, start, len);
7068 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7069 0, alloc_hint, &ins, 1, 1);
7071 return ERR_PTR(ret);
7073 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7074 ins.objectid, ins.offset, ins.offset,
7075 ins.offset, BTRFS_ORDERED_REGULAR);
7076 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7078 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7084 static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
7086 struct btrfs_block_group *block_group;
7087 bool readonly = false;
7089 block_group = btrfs_lookup_block_group(fs_info, bytenr);
7090 if (!block_group || block_group->ro)
7093 btrfs_put_block_group(block_group);
7098 * Check if we can do nocow write into the range [@offset, @offset + @len)
7100 * @offset: File offset
7101 * @len: The length to write, will be updated to the nocow writeable
7103 * @orig_start: (optional) Return the original file offset of the file extent
7104 * @orig_len: (optional) Return the original on-disk length of the file extent
7105 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7106 * @strict: if true, omit optimizations that might force us into unnecessary
7107 * cow. e.g., don't trust generation number.
7110 * >0 and update @len if we can do nocow write
7111 * 0 if we can't do nocow write
7112 * <0 if error happened
7114 * NOTE: This only checks the file extents, caller is responsible to wait for
7115 * any ordered extents.
7117 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7118 u64 *orig_start, u64 *orig_block_len,
7119 u64 *ram_bytes, bool strict)
7121 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7122 struct btrfs_path *path;
7124 struct extent_buffer *leaf;
7125 struct btrfs_root *root = BTRFS_I(inode)->root;
7126 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7127 struct btrfs_file_extent_item *fi;
7128 struct btrfs_key key;
7135 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7137 path = btrfs_alloc_path();
7141 ret = btrfs_lookup_file_extent(NULL, root, path,
7142 btrfs_ino(BTRFS_I(inode)), offset, 0);
7146 slot = path->slots[0];
7149 /* can't find the item, must cow */
7156 leaf = path->nodes[0];
7157 btrfs_item_key_to_cpu(leaf, &key, slot);
7158 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7159 key.type != BTRFS_EXTENT_DATA_KEY) {
7160 /* not our file or wrong item type, must cow */
7164 if (key.offset > offset) {
7165 /* Wrong offset, must cow */
7169 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7170 found_type = btrfs_file_extent_type(leaf, fi);
7171 if (found_type != BTRFS_FILE_EXTENT_REG &&
7172 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7173 /* not a regular extent, must cow */
7177 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7180 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7181 if (extent_end <= offset)
7184 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7185 if (disk_bytenr == 0)
7188 if (btrfs_file_extent_compression(leaf, fi) ||
7189 btrfs_file_extent_encryption(leaf, fi) ||
7190 btrfs_file_extent_other_encoding(leaf, fi))
7194 * Do the same check as in btrfs_cross_ref_exist but without the
7195 * unnecessary search.
7198 (btrfs_file_extent_generation(leaf, fi) <=
7199 btrfs_root_last_snapshot(&root->root_item)))
7202 backref_offset = btrfs_file_extent_offset(leaf, fi);
7205 *orig_start = key.offset - backref_offset;
7206 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7207 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7210 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7213 num_bytes = min(offset + *len, extent_end) - offset;
7214 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7217 range_end = round_up(offset + num_bytes,
7218 root->fs_info->sectorsize) - 1;
7219 ret = test_range_bit(io_tree, offset, range_end,
7220 EXTENT_DELALLOC, 0, NULL);
7227 btrfs_release_path(path);
7230 * look for other files referencing this extent, if we
7231 * find any we must cow
7234 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7235 key.offset - backref_offset, disk_bytenr,
7243 * adjust disk_bytenr and num_bytes to cover just the bytes
7244 * in this extent we are about to write. If there
7245 * are any csums in that range we have to cow in order
7246 * to keep the csums correct
7248 disk_bytenr += backref_offset;
7249 disk_bytenr += offset - key.offset;
7250 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7253 * all of the above have passed, it is safe to overwrite this extent
7259 btrfs_free_path(path);
7263 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7264 struct extent_state **cached_state, bool writing)
7266 struct btrfs_ordered_extent *ordered;
7270 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7273 * We're concerned with the entire range that we're going to be
7274 * doing DIO to, so we need to make sure there's no ordered
7275 * extents in this range.
7277 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7278 lockend - lockstart + 1);
7281 * We need to make sure there are no buffered pages in this
7282 * range either, we could have raced between the invalidate in
7283 * generic_file_direct_write and locking the extent. The
7284 * invalidate needs to happen so that reads after a write do not
7288 (!writing || !filemap_range_has_page(inode->i_mapping,
7289 lockstart, lockend)))
7292 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7297 * If we are doing a DIO read and the ordered extent we
7298 * found is for a buffered write, we can not wait for it
7299 * to complete and retry, because if we do so we can
7300 * deadlock with concurrent buffered writes on page
7301 * locks. This happens only if our DIO read covers more
7302 * than one extent map, if at this point has already
7303 * created an ordered extent for a previous extent map
7304 * and locked its range in the inode's io tree, and a
7305 * concurrent write against that previous extent map's
7306 * range and this range started (we unlock the ranges
7307 * in the io tree only when the bios complete and
7308 * buffered writes always lock pages before attempting
7309 * to lock range in the io tree).
7312 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7313 btrfs_start_ordered_extent(ordered, 1);
7316 btrfs_put_ordered_extent(ordered);
7319 * We could trigger writeback for this range (and wait
7320 * for it to complete) and then invalidate the pages for
7321 * this range (through invalidate_inode_pages2_range()),
7322 * but that can lead us to a deadlock with a concurrent
7323 * call to readahead (a buffered read or a defrag call
7324 * triggered a readahead) on a page lock due to an
7325 * ordered dio extent we created before but did not have
7326 * yet a corresponding bio submitted (whence it can not
7327 * complete), which makes readahead wait for that
7328 * ordered extent to complete while holding a lock on
7343 /* The callers of this must take lock_extent() */
7344 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7345 u64 len, u64 orig_start, u64 block_start,
7346 u64 block_len, u64 orig_block_len,
7347 u64 ram_bytes, int compress_type,
7350 struct extent_map_tree *em_tree;
7351 struct extent_map *em;
7354 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7355 type == BTRFS_ORDERED_COMPRESSED ||
7356 type == BTRFS_ORDERED_NOCOW ||
7357 type == BTRFS_ORDERED_REGULAR);
7359 em_tree = &inode->extent_tree;
7360 em = alloc_extent_map();
7362 return ERR_PTR(-ENOMEM);
7365 em->orig_start = orig_start;
7367 em->block_len = block_len;
7368 em->block_start = block_start;
7369 em->orig_block_len = orig_block_len;
7370 em->ram_bytes = ram_bytes;
7371 em->generation = -1;
7372 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7373 if (type == BTRFS_ORDERED_PREALLOC) {
7374 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7375 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7376 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7377 em->compress_type = compress_type;
7381 btrfs_drop_extent_cache(inode, em->start,
7382 em->start + em->len - 1, 0);
7383 write_lock(&em_tree->lock);
7384 ret = add_extent_mapping(em_tree, em, 1);
7385 write_unlock(&em_tree->lock);
7387 * The caller has taken lock_extent(), who could race with us
7390 } while (ret == -EEXIST);
7393 free_extent_map(em);
7394 return ERR_PTR(ret);
7397 /* em got 2 refs now, callers needs to do free_extent_map once. */
7402 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7403 struct inode *inode,
7404 struct btrfs_dio_data *dio_data,
7407 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7408 struct extent_map *em = *map;
7410 u64 block_start, orig_start, orig_block_len, ram_bytes;
7411 bool can_nocow = false;
7412 bool space_reserved = false;
7416 * We don't allocate a new extent in the following cases
7418 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7420 * 2) The extent is marked as PREALLOC. We're good to go here and can
7421 * just use the extent.
7424 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7425 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7426 em->block_start != EXTENT_MAP_HOLE)) {
7427 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7428 type = BTRFS_ORDERED_PREALLOC;
7430 type = BTRFS_ORDERED_NOCOW;
7431 len = min(len, em->len - (start - em->start));
7432 block_start = em->block_start + (start - em->start);
7434 if (can_nocow_extent(inode, start, &len, &orig_start,
7435 &orig_block_len, &ram_bytes, false) == 1 &&
7436 btrfs_inc_nocow_writers(fs_info, block_start))
7441 struct extent_map *em2;
7443 /* We can NOCOW, so only need to reserve metadata space. */
7444 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len);
7446 /* Our caller expects us to free the input extent map. */
7447 free_extent_map(em);
7449 btrfs_dec_nocow_writers(fs_info, block_start);
7452 space_reserved = true;
7454 em2 = btrfs_create_dio_extent(BTRFS_I(inode), start, len,
7455 orig_start, block_start,
7456 len, orig_block_len,
7458 btrfs_dec_nocow_writers(fs_info, block_start);
7459 if (type == BTRFS_ORDERED_PREALLOC) {
7460 free_extent_map(em);
7469 const u64 prev_len = len;
7471 /* Our caller expects us to free the input extent map. */
7472 free_extent_map(em);
7475 /* We have to COW, so need to reserve metadata and data space. */
7476 ret = btrfs_delalloc_reserve_space(BTRFS_I(inode),
7477 &dio_data->data_reserved,
7481 space_reserved = true;
7483 em = btrfs_new_extent_direct(BTRFS_I(inode), start, len);
7489 len = min(len, em->len - (start - em->start));
7491 btrfs_delalloc_release_space(BTRFS_I(inode),
7492 dio_data->data_reserved,
7493 start + len, prev_len - len,
7498 * We have created our ordered extent, so we can now release our reservation
7499 * for an outstanding extent.
7501 btrfs_delalloc_release_extents(BTRFS_I(inode), len);
7504 * Need to update the i_size under the extent lock so buffered
7505 * readers will get the updated i_size when we unlock.
7507 if (start + len > i_size_read(inode))
7508 i_size_write(inode, start + len);
7510 if (ret && space_reserved) {
7511 btrfs_delalloc_release_extents(BTRFS_I(inode), len);
7513 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, true);
7515 btrfs_delalloc_release_space(BTRFS_I(inode),
7516 dio_data->data_reserved,
7518 extent_changeset_free(dio_data->data_reserved);
7519 dio_data->data_reserved = NULL;
7525 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7526 loff_t length, unsigned int flags, struct iomap *iomap,
7527 struct iomap *srcmap)
7529 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7530 struct extent_map *em;
7531 struct extent_state *cached_state = NULL;
7532 struct btrfs_dio_data *dio_data = NULL;
7533 u64 lockstart, lockend;
7534 const bool write = !!(flags & IOMAP_WRITE);
7537 bool unlock_extents = false;
7540 len = min_t(u64, len, fs_info->sectorsize);
7543 lockend = start + len - 1;
7546 * The generic stuff only does filemap_write_and_wait_range, which
7547 * isn't enough if we've written compressed pages to this area, so we
7548 * need to flush the dirty pages again to make absolutely sure that any
7549 * outstanding dirty pages are on disk.
7551 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7552 &BTRFS_I(inode)->runtime_flags)) {
7553 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7554 start + length - 1);
7559 dio_data = kzalloc(sizeof(*dio_data), GFP_NOFS);
7563 iomap->private = dio_data;
7567 * If this errors out it's because we couldn't invalidate pagecache for
7568 * this range and we need to fallback to buffered.
7570 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, write)) {
7575 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7582 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7583 * io. INLINE is special, and we could probably kludge it in here, but
7584 * it's still buffered so for safety lets just fall back to the generic
7587 * For COMPRESSED we _have_ to read the entire extent in so we can
7588 * decompress it, so there will be buffering required no matter what we
7589 * do, so go ahead and fallback to buffered.
7591 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7592 * to buffered IO. Don't blame me, this is the price we pay for using
7595 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7596 em->block_start == EXTENT_MAP_INLINE) {
7597 free_extent_map(em);
7602 len = min(len, em->len - (start - em->start));
7604 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7608 unlock_extents = true;
7609 /* Recalc len in case the new em is smaller than requested */
7610 len = min(len, em->len - (start - em->start));
7613 * We need to unlock only the end area that we aren't using.
7614 * The rest is going to be unlocked by the endio routine.
7616 lockstart = start + len;
7617 if (lockstart < lockend)
7618 unlock_extents = true;
7622 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
7623 lockstart, lockend, &cached_state);
7625 free_extent_state(cached_state);
7628 * Translate extent map information to iomap.
7629 * We trim the extents (and move the addr) even though iomap code does
7630 * that, since we have locked only the parts we are performing I/O in.
7632 if ((em->block_start == EXTENT_MAP_HOLE) ||
7633 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7634 iomap->addr = IOMAP_NULL_ADDR;
7635 iomap->type = IOMAP_HOLE;
7637 iomap->addr = em->block_start + (start - em->start);
7638 iomap->type = IOMAP_MAPPED;
7640 iomap->offset = start;
7641 iomap->bdev = fs_info->fs_devices->latest_dev->bdev;
7642 iomap->length = len;
7644 if (write && btrfs_use_zone_append(BTRFS_I(inode), em->block_start))
7645 iomap->flags |= IOMAP_F_ZONE_APPEND;
7647 free_extent_map(em);
7652 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7660 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7661 ssize_t written, unsigned int flags, struct iomap *iomap)
7664 struct btrfs_dio_data *dio_data = iomap->private;
7665 size_t submitted = dio_data->submitted;
7666 const bool write = !!(flags & IOMAP_WRITE);
7668 if (!write && (iomap->type == IOMAP_HOLE)) {
7669 /* If reading from a hole, unlock and return */
7670 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1);
7674 if (submitted < length) {
7676 length -= submitted;
7678 __endio_write_update_ordered(BTRFS_I(inode), pos,
7681 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7687 extent_changeset_free(dio_data->data_reserved);
7690 iomap->private = NULL;
7695 static void btrfs_dio_private_put(struct btrfs_dio_private *dip)
7698 * This implies a barrier so that stores to dio_bio->bi_status before
7699 * this and loads of dio_bio->bi_status after this are fully ordered.
7701 if (!refcount_dec_and_test(&dip->refs))
7704 if (btrfs_op(dip->dio_bio) == BTRFS_MAP_WRITE) {
7705 __endio_write_update_ordered(BTRFS_I(dip->inode),
7708 !dip->dio_bio->bi_status);
7710 unlock_extent(&BTRFS_I(dip->inode)->io_tree,
7712 dip->file_offset + dip->bytes - 1);
7715 bio_endio(dip->dio_bio);
7719 static blk_status_t submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7721 unsigned long bio_flags)
7723 struct btrfs_dio_private *dip = bio->bi_private;
7724 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7727 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7729 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
7733 refcount_inc(&dip->refs);
7734 ret = btrfs_map_bio(fs_info, bio, mirror_num);
7736 refcount_dec(&dip->refs);
7740 static blk_status_t btrfs_check_read_dio_bio(struct btrfs_dio_private *dip,
7741 struct btrfs_bio *bbio,
7742 const bool uptodate)
7744 struct inode *inode = dip->inode;
7745 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
7746 const u32 sectorsize = fs_info->sectorsize;
7747 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7748 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7749 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7750 struct bio_vec bvec;
7751 struct bvec_iter iter;
7752 const u64 orig_file_offset = dip->file_offset;
7753 u64 start = orig_file_offset;
7755 blk_status_t err = BLK_STS_OK;
7757 __bio_for_each_segment(bvec, &bbio->bio, iter, bbio->iter) {
7758 unsigned int i, nr_sectors, pgoff;
7760 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7761 pgoff = bvec.bv_offset;
7762 for (i = 0; i < nr_sectors; i++) {
7763 ASSERT(pgoff < PAGE_SIZE);
7765 (!csum || !check_data_csum(inode, bbio,
7766 bio_offset, bvec.bv_page,
7768 clean_io_failure(fs_info, failure_tree, io_tree,
7769 start, bvec.bv_page,
7770 btrfs_ino(BTRFS_I(inode)),
7775 ASSERT((start - orig_file_offset) < UINT_MAX);
7776 ret = btrfs_repair_one_sector(inode,
7778 start - orig_file_offset,
7779 bvec.bv_page, pgoff,
7780 start, bbio->mirror_num,
7781 submit_dio_repair_bio);
7783 err = errno_to_blk_status(ret);
7785 start += sectorsize;
7786 ASSERT(bio_offset + sectorsize > bio_offset);
7787 bio_offset += sectorsize;
7788 pgoff += sectorsize;
7794 static void __endio_write_update_ordered(struct btrfs_inode *inode,
7795 const u64 offset, const u64 bytes,
7796 const bool uptodate)
7798 btrfs_mark_ordered_io_finished(inode, NULL, offset, bytes,
7799 finish_ordered_fn, uptodate);
7802 static blk_status_t btrfs_submit_bio_start_direct_io(struct inode *inode,
7804 u64 dio_file_offset)
7806 return btrfs_csum_one_bio(BTRFS_I(inode), bio, dio_file_offset, 1);
7809 static void btrfs_end_dio_bio(struct bio *bio)
7811 struct btrfs_dio_private *dip = bio->bi_private;
7812 blk_status_t err = bio->bi_status;
7815 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
7816 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
7817 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
7818 bio->bi_opf, bio->bi_iter.bi_sector,
7819 bio->bi_iter.bi_size, err);
7821 if (bio_op(bio) == REQ_OP_READ)
7822 err = btrfs_check_read_dio_bio(dip, btrfs_bio(bio), !err);
7825 dip->dio_bio->bi_status = err;
7827 btrfs_record_physical_zoned(dip->inode, dip->file_offset, bio);
7830 btrfs_dio_private_put(dip);
7833 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
7834 struct inode *inode, u64 file_offset, int async_submit)
7836 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7837 struct btrfs_dio_private *dip = bio->bi_private;
7838 bool write = btrfs_op(bio) == BTRFS_MAP_WRITE;
7841 /* Check btrfs_submit_bio_hook() for rules about async submit. */
7843 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7846 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
7851 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
7854 if (write && async_submit) {
7855 ret = btrfs_wq_submit_bio(inode, bio, 0, 0, file_offset,
7856 btrfs_submit_bio_start_direct_io);
7860 * If we aren't doing async submit, calculate the csum of the
7863 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, file_offset, 1);
7869 csum_offset = file_offset - dip->file_offset;
7870 csum_offset >>= fs_info->sectorsize_bits;
7871 csum_offset *= fs_info->csum_size;
7872 btrfs_bio(bio)->csum = dip->csums + csum_offset;
7875 ret = btrfs_map_bio(fs_info, bio, 0);
7881 * If this succeeds, the btrfs_dio_private is responsible for cleaning up locked
7882 * or ordered extents whether or not we submit any bios.
7884 static struct btrfs_dio_private *btrfs_create_dio_private(struct bio *dio_bio,
7885 struct inode *inode,
7888 const bool write = (btrfs_op(dio_bio) == BTRFS_MAP_WRITE);
7889 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7891 struct btrfs_dio_private *dip;
7893 dip_size = sizeof(*dip);
7894 if (!write && csum) {
7895 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7898 nblocks = dio_bio->bi_iter.bi_size >> fs_info->sectorsize_bits;
7899 dip_size += fs_info->csum_size * nblocks;
7902 dip = kzalloc(dip_size, GFP_NOFS);
7907 dip->file_offset = file_offset;
7908 dip->bytes = dio_bio->bi_iter.bi_size;
7909 dip->disk_bytenr = dio_bio->bi_iter.bi_sector << 9;
7910 dip->dio_bio = dio_bio;
7911 refcount_set(&dip->refs, 1);
7915 static void btrfs_submit_direct(const struct iomap_iter *iter,
7916 struct bio *dio_bio, loff_t file_offset)
7918 struct inode *inode = iter->inode;
7919 const bool write = (btrfs_op(dio_bio) == BTRFS_MAP_WRITE);
7920 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7921 const bool raid56 = (btrfs_data_alloc_profile(fs_info) &
7922 BTRFS_BLOCK_GROUP_RAID56_MASK);
7923 struct btrfs_dio_private *dip;
7926 int async_submit = 0;
7928 u64 clone_offset = 0;
7932 blk_status_t status;
7933 struct btrfs_io_geometry geom;
7934 struct btrfs_dio_data *dio_data = iter->iomap.private;
7935 struct extent_map *em = NULL;
7937 dip = btrfs_create_dio_private(dio_bio, inode, file_offset);
7940 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
7941 file_offset + dio_bio->bi_iter.bi_size - 1);
7943 dio_bio->bi_status = BLK_STS_RESOURCE;
7950 * Load the csums up front to reduce csum tree searches and
7951 * contention when submitting bios.
7953 * If we have csums disabled this will do nothing.
7955 status = btrfs_lookup_bio_sums(inode, dio_bio, dip->csums);
7956 if (status != BLK_STS_OK)
7960 start_sector = dio_bio->bi_iter.bi_sector;
7961 submit_len = dio_bio->bi_iter.bi_size;
7964 logical = start_sector << 9;
7965 em = btrfs_get_chunk_map(fs_info, logical, submit_len);
7967 status = errno_to_blk_status(PTR_ERR(em));
7971 ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(dio_bio),
7974 status = errno_to_blk_status(ret);
7978 clone_len = min(submit_len, geom.len);
7979 ASSERT(clone_len <= UINT_MAX);
7982 * This will never fail as it's passing GPF_NOFS and
7983 * the allocation is backed by btrfs_bioset.
7985 bio = btrfs_bio_clone_partial(dio_bio, clone_offset, clone_len);
7986 bio->bi_private = dip;
7987 bio->bi_end_io = btrfs_end_dio_bio;
7989 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
7990 status = extract_ordered_extent(BTRFS_I(inode), bio,
7998 ASSERT(submit_len >= clone_len);
7999 submit_len -= clone_len;
8002 * Increase the count before we submit the bio so we know
8003 * the end IO handler won't happen before we increase the
8004 * count. Otherwise, the dip might get freed before we're
8005 * done setting it up.
8007 * We transfer the initial reference to the last bio, so we
8008 * don't need to increment the reference count for the last one.
8010 if (submit_len > 0) {
8011 refcount_inc(&dip->refs);
8013 * If we are submitting more than one bio, submit them
8014 * all asynchronously. The exception is RAID 5 or 6, as
8015 * asynchronous checksums make it difficult to collect
8016 * full stripe writes.
8022 status = btrfs_submit_dio_bio(bio, inode, file_offset,
8027 refcount_dec(&dip->refs);
8031 dio_data->submitted += clone_len;
8032 clone_offset += clone_len;
8033 start_sector += clone_len >> 9;
8034 file_offset += clone_len;
8036 free_extent_map(em);
8037 } while (submit_len > 0);
8041 free_extent_map(em);
8043 dip->dio_bio->bi_status = status;
8044 btrfs_dio_private_put(dip);
8047 const struct iomap_ops btrfs_dio_iomap_ops = {
8048 .iomap_begin = btrfs_dio_iomap_begin,
8049 .iomap_end = btrfs_dio_iomap_end,
8052 const struct iomap_dio_ops btrfs_dio_ops = {
8053 .submit_io = btrfs_submit_direct,
8056 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8061 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
8065 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
8068 int btrfs_readpage(struct file *file, struct page *page)
8070 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
8071 u64 start = page_offset(page);
8072 u64 end = start + PAGE_SIZE - 1;
8073 struct btrfs_bio_ctrl bio_ctrl = { 0 };
8076 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
8078 ret = btrfs_do_readpage(page, NULL, &bio_ctrl, 0, NULL);
8080 ret = submit_one_bio(bio_ctrl.bio, 0, bio_ctrl.bio_flags);
8084 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8086 struct inode *inode = page->mapping->host;
8089 if (current->flags & PF_MEMALLOC) {
8090 redirty_page_for_writepage(wbc, page);
8096 * If we are under memory pressure we will call this directly from the
8097 * VM, we need to make sure we have the inode referenced for the ordered
8098 * extent. If not just return like we didn't do anything.
8100 if (!igrab(inode)) {
8101 redirty_page_for_writepage(wbc, page);
8102 return AOP_WRITEPAGE_ACTIVATE;
8104 ret = extent_write_full_page(page, wbc);
8105 btrfs_add_delayed_iput(inode);
8109 static int btrfs_writepages(struct address_space *mapping,
8110 struct writeback_control *wbc)
8112 return extent_writepages(mapping, wbc);
8115 static void btrfs_readahead(struct readahead_control *rac)
8117 extent_readahead(rac);
8121 * For releasepage() and invalidatepage() we have a race window where
8122 * end_page_writeback() is called but the subpage spinlock is not yet released.
8123 * If we continue to release/invalidate the page, we could cause use-after-free
8124 * for subpage spinlock. So this function is to spin and wait for subpage
8127 static void wait_subpage_spinlock(struct page *page)
8129 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
8130 struct btrfs_subpage *subpage;
8132 if (fs_info->sectorsize == PAGE_SIZE)
8135 ASSERT(PagePrivate(page) && page->private);
8136 subpage = (struct btrfs_subpage *)page->private;
8139 * This may look insane as we just acquire the spinlock and release it,
8140 * without doing anything. But we just want to make sure no one is
8141 * still holding the subpage spinlock.
8142 * And since the page is not dirty nor writeback, and we have page
8143 * locked, the only possible way to hold a spinlock is from the endio
8144 * function to clear page writeback.
8146 * Here we just acquire the spinlock so that all existing callers
8147 * should exit and we're safe to release/invalidate the page.
8149 spin_lock_irq(&subpage->lock);
8150 spin_unlock_irq(&subpage->lock);
8153 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8155 int ret = try_release_extent_mapping(page, gfp_flags);
8158 wait_subpage_spinlock(page);
8159 clear_page_extent_mapped(page);
8164 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8166 if (PageWriteback(page) || PageDirty(page))
8168 return __btrfs_releasepage(page, gfp_flags);
8171 #ifdef CONFIG_MIGRATION
8172 static int btrfs_migratepage(struct address_space *mapping,
8173 struct page *newpage, struct page *page,
8174 enum migrate_mode mode)
8178 ret = migrate_page_move_mapping(mapping, newpage, page, 0);
8179 if (ret != MIGRATEPAGE_SUCCESS)
8182 if (page_has_private(page))
8183 attach_page_private(newpage, detach_page_private(page));
8185 if (PageOrdered(page)) {
8186 ClearPageOrdered(page);
8187 SetPageOrdered(newpage);
8190 if (mode != MIGRATE_SYNC_NO_COPY)
8191 migrate_page_copy(newpage, page);
8193 migrate_page_states(newpage, page);
8194 return MIGRATEPAGE_SUCCESS;
8198 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8199 unsigned int length)
8201 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
8202 struct btrfs_fs_info *fs_info = inode->root->fs_info;
8203 struct extent_io_tree *tree = &inode->io_tree;
8204 struct extent_state *cached_state = NULL;
8205 u64 page_start = page_offset(page);
8206 u64 page_end = page_start + PAGE_SIZE - 1;
8208 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
8211 * We have page locked so no new ordered extent can be created on this
8212 * page, nor bio can be submitted for this page.
8214 * But already submitted bio can still be finished on this page.
8215 * Furthermore, endio function won't skip page which has Ordered
8216 * (Private2) already cleared, so it's possible for endio and
8217 * invalidatepage to do the same ordered extent accounting twice
8220 * So here we wait for any submitted bios to finish, so that we won't
8221 * do double ordered extent accounting on the same page.
8223 wait_on_page_writeback(page);
8224 wait_subpage_spinlock(page);
8227 * For subpage case, we have call sites like
8228 * btrfs_punch_hole_lock_range() which passes range not aligned to
8230 * If the range doesn't cover the full page, we don't need to and
8231 * shouldn't clear page extent mapped, as page->private can still
8232 * record subpage dirty bits for other part of the range.
8234 * For cases that can invalidate the full even the range doesn't
8235 * cover the full page, like invalidating the last page, we're
8236 * still safe to wait for ordered extent to finish.
8238 if (!(offset == 0 && length == PAGE_SIZE)) {
8239 btrfs_releasepage(page, GFP_NOFS);
8243 if (!inode_evicting)
8244 lock_extent_bits(tree, page_start, page_end, &cached_state);
8247 while (cur < page_end) {
8248 struct btrfs_ordered_extent *ordered;
8253 ordered = btrfs_lookup_first_ordered_range(inode, cur,
8254 page_end + 1 - cur);
8256 range_end = page_end;
8258 * No ordered extent covering this range, we are safe
8259 * to delete all extent states in the range.
8261 delete_states = true;
8264 if (ordered->file_offset > cur) {
8266 * There is a range between [cur, oe->file_offset) not
8267 * covered by any ordered extent.
8268 * We are safe to delete all extent states, and handle
8269 * the ordered extent in the next iteration.
8271 range_end = ordered->file_offset - 1;
8272 delete_states = true;
8276 range_end = min(ordered->file_offset + ordered->num_bytes - 1,
8278 ASSERT(range_end + 1 - cur < U32_MAX);
8279 range_len = range_end + 1 - cur;
8280 if (!btrfs_page_test_ordered(fs_info, page, cur, range_len)) {
8282 * If Ordered (Private2) is cleared, it means endio has
8283 * already been executed for the range.
8284 * We can't delete the extent states as
8285 * btrfs_finish_ordered_io() may still use some of them.
8287 delete_states = false;
8290 btrfs_page_clear_ordered(fs_info, page, cur, range_len);
8293 * IO on this page will never be started, so we need to account
8294 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8295 * here, must leave that up for the ordered extent completion.
8297 * This will also unlock the range for incoming
8298 * btrfs_finish_ordered_io().
8300 if (!inode_evicting)
8301 clear_extent_bit(tree, cur, range_end,
8303 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8304 EXTENT_DEFRAG, 1, 0, &cached_state);
8306 spin_lock_irq(&inode->ordered_tree.lock);
8307 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8308 ordered->truncated_len = min(ordered->truncated_len,
8309 cur - ordered->file_offset);
8310 spin_unlock_irq(&inode->ordered_tree.lock);
8312 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8313 cur, range_end + 1 - cur)) {
8314 btrfs_finish_ordered_io(ordered);
8316 * The ordered extent has finished, now we're again
8317 * safe to delete all extent states of the range.
8319 delete_states = true;
8322 * btrfs_finish_ordered_io() will get executed by endio
8323 * of other pages, thus we can't delete extent states
8326 delete_states = false;
8330 btrfs_put_ordered_extent(ordered);
8332 * Qgroup reserved space handler
8333 * Sector(s) here will be either:
8335 * 1) Already written to disk or bio already finished
8336 * Then its QGROUP_RESERVED bit in io_tree is already cleared.
8337 * Qgroup will be handled by its qgroup_record then.
8338 * btrfs_qgroup_free_data() call will do nothing here.
8340 * 2) Not written to disk yet
8341 * Then btrfs_qgroup_free_data() call will clear the
8342 * QGROUP_RESERVED bit of its io_tree, and free the qgroup
8343 * reserved data space.
8344 * Since the IO will never happen for this page.
8346 btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur);
8347 if (!inode_evicting) {
8348 clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED |
8349 EXTENT_DELALLOC | EXTENT_UPTODATE |
8350 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1,
8351 delete_states, &cached_state);
8353 cur = range_end + 1;
8356 * We have iterated through all ordered extents of the page, the page
8357 * should not have Ordered (Private2) anymore, or the above iteration
8358 * did something wrong.
8360 ASSERT(!PageOrdered(page));
8361 btrfs_page_clear_checked(fs_info, page, page_offset(page), PAGE_SIZE);
8362 if (!inode_evicting)
8363 __btrfs_releasepage(page, GFP_NOFS);
8364 clear_page_extent_mapped(page);
8368 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8369 * called from a page fault handler when a page is first dirtied. Hence we must
8370 * be careful to check for EOF conditions here. We set the page up correctly
8371 * for a written page which means we get ENOSPC checking when writing into
8372 * holes and correct delalloc and unwritten extent mapping on filesystems that
8373 * support these features.
8375 * We are not allowed to take the i_mutex here so we have to play games to
8376 * protect against truncate races as the page could now be beyond EOF. Because
8377 * truncate_setsize() writes the inode size before removing pages, once we have
8378 * the page lock we can determine safely if the page is beyond EOF. If it is not
8379 * beyond EOF, then the page is guaranteed safe against truncation until we
8382 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8384 struct page *page = vmf->page;
8385 struct inode *inode = file_inode(vmf->vma->vm_file);
8386 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8387 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8388 struct btrfs_ordered_extent *ordered;
8389 struct extent_state *cached_state = NULL;
8390 struct extent_changeset *data_reserved = NULL;
8391 unsigned long zero_start;
8401 reserved_space = PAGE_SIZE;
8403 sb_start_pagefault(inode->i_sb);
8404 page_start = page_offset(page);
8405 page_end = page_start + PAGE_SIZE - 1;
8409 * Reserving delalloc space after obtaining the page lock can lead to
8410 * deadlock. For example, if a dirty page is locked by this function
8411 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8412 * dirty page write out, then the btrfs_writepage() function could
8413 * end up waiting indefinitely to get a lock on the page currently
8414 * being processed by btrfs_page_mkwrite() function.
8416 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8417 page_start, reserved_space);
8419 ret2 = file_update_time(vmf->vma->vm_file);
8423 ret = vmf_error(ret2);
8429 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8431 down_read(&BTRFS_I(inode)->i_mmap_lock);
8433 size = i_size_read(inode);
8435 if ((page->mapping != inode->i_mapping) ||
8436 (page_start >= size)) {
8437 /* page got truncated out from underneath us */
8440 wait_on_page_writeback(page);
8442 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8443 ret2 = set_page_extent_mapped(page);
8445 ret = vmf_error(ret2);
8446 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8451 * we can't set the delalloc bits if there are pending ordered
8452 * extents. Drop our locks and wait for them to finish
8454 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8457 unlock_extent_cached(io_tree, page_start, page_end,
8460 up_read(&BTRFS_I(inode)->i_mmap_lock);
8461 btrfs_start_ordered_extent(ordered, 1);
8462 btrfs_put_ordered_extent(ordered);
8466 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8467 reserved_space = round_up(size - page_start,
8468 fs_info->sectorsize);
8469 if (reserved_space < PAGE_SIZE) {
8470 end = page_start + reserved_space - 1;
8471 btrfs_delalloc_release_space(BTRFS_I(inode),
8472 data_reserved, page_start,
8473 PAGE_SIZE - reserved_space, true);
8478 * page_mkwrite gets called when the page is firstly dirtied after it's
8479 * faulted in, but write(2) could also dirty a page and set delalloc
8480 * bits, thus in this case for space account reason, we still need to
8481 * clear any delalloc bits within this page range since we have to
8482 * reserve data&meta space before lock_page() (see above comments).
8484 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8485 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8486 EXTENT_DEFRAG, 0, 0, &cached_state);
8488 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8491 unlock_extent_cached(io_tree, page_start, page_end,
8493 ret = VM_FAULT_SIGBUS;
8497 /* page is wholly or partially inside EOF */
8498 if (page_start + PAGE_SIZE > size)
8499 zero_start = offset_in_page(size);
8501 zero_start = PAGE_SIZE;
8503 if (zero_start != PAGE_SIZE) {
8504 memzero_page(page, zero_start, PAGE_SIZE - zero_start);
8505 flush_dcache_page(page);
8507 btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE);
8508 btrfs_page_set_dirty(fs_info, page, page_start, end + 1 - page_start);
8509 btrfs_page_set_uptodate(fs_info, page, page_start, end + 1 - page_start);
8511 btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
8513 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8514 up_read(&BTRFS_I(inode)->i_mmap_lock);
8516 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8517 sb_end_pagefault(inode->i_sb);
8518 extent_changeset_free(data_reserved);
8519 return VM_FAULT_LOCKED;
8523 up_read(&BTRFS_I(inode)->i_mmap_lock);
8525 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8526 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8527 reserved_space, (ret != 0));
8529 sb_end_pagefault(inode->i_sb);
8530 extent_changeset_free(data_reserved);
8534 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
8536 struct btrfs_truncate_control control = {
8537 .inode = BTRFS_I(inode),
8538 .ino = btrfs_ino(BTRFS_I(inode)),
8539 .min_type = BTRFS_EXTENT_DATA_KEY,
8540 .clear_extent_range = true,
8542 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8543 struct btrfs_root *root = BTRFS_I(inode)->root;
8544 struct btrfs_block_rsv *rsv;
8546 struct btrfs_trans_handle *trans;
8547 u64 mask = fs_info->sectorsize - 1;
8548 u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8550 if (!skip_writeback) {
8551 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8558 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8559 * things going on here:
8561 * 1) We need to reserve space to update our inode.
8563 * 2) We need to have something to cache all the space that is going to
8564 * be free'd up by the truncate operation, but also have some slack
8565 * space reserved in case it uses space during the truncate (thank you
8566 * very much snapshotting).
8568 * And we need these to be separate. The fact is we can use a lot of
8569 * space doing the truncate, and we have no earthly idea how much space
8570 * we will use, so we need the truncate reservation to be separate so it
8571 * doesn't end up using space reserved for updating the inode. We also
8572 * need to be able to stop the transaction and start a new one, which
8573 * means we need to be able to update the inode several times, and we
8574 * have no idea of knowing how many times that will be, so we can't just
8575 * reserve 1 item for the entirety of the operation, so that has to be
8576 * done separately as well.
8578 * So that leaves us with
8580 * 1) rsv - for the truncate reservation, which we will steal from the
8581 * transaction reservation.
8582 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8583 * updating the inode.
8585 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8588 rsv->size = min_size;
8592 * 1 for the truncate slack space
8593 * 1 for updating the inode.
8595 trans = btrfs_start_transaction(root, 2);
8596 if (IS_ERR(trans)) {
8597 ret = PTR_ERR(trans);
8601 /* Migrate the slack space for the truncate to our reserve */
8602 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8606 trans->block_rsv = rsv;
8609 struct extent_state *cached_state = NULL;
8610 const u64 new_size = inode->i_size;
8611 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
8613 control.new_size = new_size;
8614 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, (u64)-1,
8617 * We want to drop from the next block forward in case this new
8618 * size is not block aligned since we will be keeping the last
8619 * block of the extent just the way it is.
8621 btrfs_drop_extent_cache(BTRFS_I(inode),
8622 ALIGN(new_size, fs_info->sectorsize),
8625 ret = btrfs_truncate_inode_items(trans, root, &control);
8627 inode_sub_bytes(inode, control.sub_bytes);
8628 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), control.last_size);
8630 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start,
8631 (u64)-1, &cached_state);
8633 trans->block_rsv = &fs_info->trans_block_rsv;
8634 if (ret != -ENOSPC && ret != -EAGAIN)
8637 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
8641 btrfs_end_transaction(trans);
8642 btrfs_btree_balance_dirty(fs_info);
8644 trans = btrfs_start_transaction(root, 2);
8645 if (IS_ERR(trans)) {
8646 ret = PTR_ERR(trans);
8651 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8652 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8653 rsv, min_size, false);
8654 BUG_ON(ret); /* shouldn't happen */
8655 trans->block_rsv = rsv;
8659 * We can't call btrfs_truncate_block inside a trans handle as we could
8660 * deadlock with freeze, if we got BTRFS_NEED_TRUNCATE_BLOCK then we
8661 * know we've truncated everything except the last little bit, and can
8662 * do btrfs_truncate_block and then update the disk_i_size.
8664 if (ret == BTRFS_NEED_TRUNCATE_BLOCK) {
8665 btrfs_end_transaction(trans);
8666 btrfs_btree_balance_dirty(fs_info);
8668 ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0);
8671 trans = btrfs_start_transaction(root, 1);
8672 if (IS_ERR(trans)) {
8673 ret = PTR_ERR(trans);
8676 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
8682 trans->block_rsv = &fs_info->trans_block_rsv;
8683 ret2 = btrfs_update_inode(trans, root, BTRFS_I(inode));
8687 ret2 = btrfs_end_transaction(trans);
8690 btrfs_btree_balance_dirty(fs_info);
8693 btrfs_free_block_rsv(fs_info, rsv);
8695 * So if we truncate and then write and fsync we normally would just
8696 * write the extents that changed, which is a problem if we need to
8697 * first truncate that entire inode. So set this flag so we write out
8698 * all of the extents in the inode to the sync log so we're completely
8701 * If no extents were dropped or trimmed we don't need to force the next
8702 * fsync to truncate all the inode's items from the log and re-log them
8703 * all. This means the truncate operation did not change the file size,
8704 * or changed it to a smaller size but there was only an implicit hole
8705 * between the old i_size and the new i_size, and there were no prealloc
8706 * extents beyond i_size to drop.
8708 if (control.extents_found > 0)
8709 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
8715 * create a new subvolume directory/inode (helper for the ioctl).
8717 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8718 struct btrfs_root *new_root,
8719 struct btrfs_root *parent_root,
8720 struct user_namespace *mnt_userns)
8722 struct inode *inode;
8727 err = btrfs_get_free_objectid(new_root, &ino);
8731 inode = btrfs_new_inode(trans, new_root, mnt_userns, NULL, "..", 2,
8733 S_IFDIR | (~current_umask() & S_IRWXUGO),
8736 return PTR_ERR(inode);
8737 inode->i_op = &btrfs_dir_inode_operations;
8738 inode->i_fop = &btrfs_dir_file_operations;
8740 set_nlink(inode, 1);
8741 btrfs_i_size_write(BTRFS_I(inode), 0);
8742 unlock_new_inode(inode);
8744 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8746 btrfs_err(new_root->fs_info,
8747 "error inheriting subvolume %llu properties: %d",
8748 new_root->root_key.objectid, err);
8750 err = btrfs_update_inode(trans, new_root, BTRFS_I(inode));
8756 struct inode *btrfs_alloc_inode(struct super_block *sb)
8758 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8759 struct btrfs_inode *ei;
8760 struct inode *inode;
8762 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL);
8769 ei->last_sub_trans = 0;
8770 ei->logged_trans = 0;
8771 ei->delalloc_bytes = 0;
8772 ei->new_delalloc_bytes = 0;
8773 ei->defrag_bytes = 0;
8774 ei->disk_i_size = 0;
8778 ei->index_cnt = (u64)-1;
8780 ei->last_unlink_trans = 0;
8781 ei->last_reflink_trans = 0;
8782 ei->last_log_commit = 0;
8784 spin_lock_init(&ei->lock);
8785 ei->outstanding_extents = 0;
8786 if (sb->s_magic != BTRFS_TEST_MAGIC)
8787 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8788 BTRFS_BLOCK_RSV_DELALLOC);
8789 ei->runtime_flags = 0;
8790 ei->prop_compress = BTRFS_COMPRESS_NONE;
8791 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8793 ei->delayed_node = NULL;
8795 ei->i_otime.tv_sec = 0;
8796 ei->i_otime.tv_nsec = 0;
8798 inode = &ei->vfs_inode;
8799 extent_map_tree_init(&ei->extent_tree);
8800 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO, inode);
8801 extent_io_tree_init(fs_info, &ei->io_failure_tree,
8802 IO_TREE_INODE_IO_FAILURE, inode);
8803 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8804 IO_TREE_INODE_FILE_EXTENT, inode);
8805 ei->io_tree.track_uptodate = true;
8806 ei->io_failure_tree.track_uptodate = true;
8807 atomic_set(&ei->sync_writers, 0);
8808 mutex_init(&ei->log_mutex);
8809 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8810 INIT_LIST_HEAD(&ei->delalloc_inodes);
8811 INIT_LIST_HEAD(&ei->delayed_iput);
8812 RB_CLEAR_NODE(&ei->rb_node);
8813 init_rwsem(&ei->i_mmap_lock);
8818 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8819 void btrfs_test_destroy_inode(struct inode *inode)
8821 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
8822 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8826 void btrfs_free_inode(struct inode *inode)
8828 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8831 void btrfs_destroy_inode(struct inode *vfs_inode)
8833 struct btrfs_ordered_extent *ordered;
8834 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8835 struct btrfs_root *root = inode->root;
8837 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8838 WARN_ON(vfs_inode->i_data.nrpages);
8839 WARN_ON(inode->block_rsv.reserved);
8840 WARN_ON(inode->block_rsv.size);
8841 WARN_ON(inode->outstanding_extents);
8842 if (!S_ISDIR(vfs_inode->i_mode)) {
8843 WARN_ON(inode->delalloc_bytes);
8844 WARN_ON(inode->new_delalloc_bytes);
8846 WARN_ON(inode->csum_bytes);
8847 WARN_ON(inode->defrag_bytes);
8850 * This can happen where we create an inode, but somebody else also
8851 * created the same inode and we need to destroy the one we already
8858 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8862 btrfs_err(root->fs_info,
8863 "found ordered extent %llu %llu on inode cleanup",
8864 ordered->file_offset, ordered->num_bytes);
8865 btrfs_remove_ordered_extent(inode, ordered);
8866 btrfs_put_ordered_extent(ordered);
8867 btrfs_put_ordered_extent(ordered);
8870 btrfs_qgroup_check_reserved_leak(inode);
8871 inode_tree_del(inode);
8872 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8873 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
8874 btrfs_put_root(inode->root);
8877 int btrfs_drop_inode(struct inode *inode)
8879 struct btrfs_root *root = BTRFS_I(inode)->root;
8884 /* the snap/subvol tree is on deleting */
8885 if (btrfs_root_refs(&root->root_item) == 0)
8888 return generic_drop_inode(inode);
8891 static void init_once(void *foo)
8893 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8895 inode_init_once(&ei->vfs_inode);
8898 void __cold btrfs_destroy_cachep(void)
8901 * Make sure all delayed rcu free inodes are flushed before we
8905 kmem_cache_destroy(btrfs_inode_cachep);
8906 kmem_cache_destroy(btrfs_trans_handle_cachep);
8907 kmem_cache_destroy(btrfs_path_cachep);
8908 kmem_cache_destroy(btrfs_free_space_cachep);
8909 kmem_cache_destroy(btrfs_free_space_bitmap_cachep);
8912 int __init btrfs_init_cachep(void)
8914 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8915 sizeof(struct btrfs_inode), 0,
8916 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
8918 if (!btrfs_inode_cachep)
8921 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8922 sizeof(struct btrfs_trans_handle), 0,
8923 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
8924 if (!btrfs_trans_handle_cachep)
8927 btrfs_path_cachep = kmem_cache_create("btrfs_path",
8928 sizeof(struct btrfs_path), 0,
8929 SLAB_MEM_SPREAD, NULL);
8930 if (!btrfs_path_cachep)
8933 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8934 sizeof(struct btrfs_free_space), 0,
8935 SLAB_MEM_SPREAD, NULL);
8936 if (!btrfs_free_space_cachep)
8939 btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap",
8940 PAGE_SIZE, PAGE_SIZE,
8941 SLAB_MEM_SPREAD, NULL);
8942 if (!btrfs_free_space_bitmap_cachep)
8947 btrfs_destroy_cachep();
8951 static int btrfs_getattr(struct user_namespace *mnt_userns,
8952 const struct path *path, struct kstat *stat,
8953 u32 request_mask, unsigned int flags)
8957 struct inode *inode = d_inode(path->dentry);
8958 u32 blocksize = inode->i_sb->s_blocksize;
8959 u32 bi_flags = BTRFS_I(inode)->flags;
8960 u32 bi_ro_flags = BTRFS_I(inode)->ro_flags;
8962 stat->result_mask |= STATX_BTIME;
8963 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
8964 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
8965 if (bi_flags & BTRFS_INODE_APPEND)
8966 stat->attributes |= STATX_ATTR_APPEND;
8967 if (bi_flags & BTRFS_INODE_COMPRESS)
8968 stat->attributes |= STATX_ATTR_COMPRESSED;
8969 if (bi_flags & BTRFS_INODE_IMMUTABLE)
8970 stat->attributes |= STATX_ATTR_IMMUTABLE;
8971 if (bi_flags & BTRFS_INODE_NODUMP)
8972 stat->attributes |= STATX_ATTR_NODUMP;
8973 if (bi_ro_flags & BTRFS_INODE_RO_VERITY)
8974 stat->attributes |= STATX_ATTR_VERITY;
8976 stat->attributes_mask |= (STATX_ATTR_APPEND |
8977 STATX_ATTR_COMPRESSED |
8978 STATX_ATTR_IMMUTABLE |
8981 generic_fillattr(mnt_userns, inode, stat);
8982 stat->dev = BTRFS_I(inode)->root->anon_dev;
8984 spin_lock(&BTRFS_I(inode)->lock);
8985 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
8986 inode_bytes = inode_get_bytes(inode);
8987 spin_unlock(&BTRFS_I(inode)->lock);
8988 stat->blocks = (ALIGN(inode_bytes, blocksize) +
8989 ALIGN(delalloc_bytes, blocksize)) >> 9;
8993 static int btrfs_rename_exchange(struct inode *old_dir,
8994 struct dentry *old_dentry,
8995 struct inode *new_dir,
8996 struct dentry *new_dentry)
8998 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
8999 struct btrfs_trans_handle *trans;
9000 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9001 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9002 struct inode *new_inode = new_dentry->d_inode;
9003 struct inode *old_inode = old_dentry->d_inode;
9004 struct timespec64 ctime = current_time(old_inode);
9005 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9006 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9011 bool root_log_pinned = false;
9012 bool dest_log_pinned = false;
9013 bool need_abort = false;
9016 * For non-subvolumes allow exchange only within one subvolume, in the
9017 * same inode namespace. Two subvolumes (represented as directory) can
9018 * be exchanged as they're a logical link and have a fixed inode number.
9021 (old_ino != BTRFS_FIRST_FREE_OBJECTID ||
9022 new_ino != BTRFS_FIRST_FREE_OBJECTID))
9025 /* close the race window with snapshot create/destroy ioctl */
9026 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
9027 new_ino == BTRFS_FIRST_FREE_OBJECTID)
9028 down_read(&fs_info->subvol_sem);
9031 * We want to reserve the absolute worst case amount of items. So if
9032 * both inodes are subvols and we need to unlink them then that would
9033 * require 4 item modifications, but if they are both normal inodes it
9034 * would require 5 item modifications, so we'll assume their normal
9035 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9036 * should cover the worst case number of items we'll modify.
9038 trans = btrfs_start_transaction(root, 12);
9039 if (IS_ERR(trans)) {
9040 ret = PTR_ERR(trans);
9045 ret = btrfs_record_root_in_trans(trans, dest);
9051 * We need to find a free sequence number both in the source and
9052 * in the destination directory for the exchange.
9054 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9057 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9061 BTRFS_I(old_inode)->dir_index = 0ULL;
9062 BTRFS_I(new_inode)->dir_index = 0ULL;
9064 /* Reference for the source. */
9065 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9066 /* force full log commit if subvolume involved. */
9067 btrfs_set_log_full_commit(trans);
9069 ret = btrfs_insert_inode_ref(trans, dest,
9070 new_dentry->d_name.name,
9071 new_dentry->d_name.len,
9073 btrfs_ino(BTRFS_I(new_dir)),
9080 /* And now for the dest. */
9081 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9082 /* force full log commit if subvolume involved. */
9083 btrfs_set_log_full_commit(trans);
9085 ret = btrfs_insert_inode_ref(trans, root,
9086 old_dentry->d_name.name,
9087 old_dentry->d_name.len,
9089 btrfs_ino(BTRFS_I(old_dir)),
9093 btrfs_abort_transaction(trans, ret);
9098 /* Update inode version and ctime/mtime. */
9099 inode_inc_iversion(old_dir);
9100 inode_inc_iversion(new_dir);
9101 inode_inc_iversion(old_inode);
9102 inode_inc_iversion(new_inode);
9103 old_dir->i_ctime = old_dir->i_mtime = ctime;
9104 new_dir->i_ctime = new_dir->i_mtime = ctime;
9105 old_inode->i_ctime = ctime;
9106 new_inode->i_ctime = ctime;
9108 if (old_dentry->d_parent != new_dentry->d_parent) {
9109 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9110 BTRFS_I(old_inode), 1);
9111 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9112 BTRFS_I(new_inode), 1);
9116 * Now pin the logs of the roots. We do it to ensure that no other task
9117 * can sync the logs while we are in progress with the rename, because
9118 * that could result in an inconsistency in case any of the inodes that
9119 * are part of this rename operation were logged before.
9121 * We pin the logs even if at this precise moment none of the inodes was
9122 * logged before. This is because right after we checked for that, some
9123 * other task fsyncing some other inode not involved with this rename
9124 * operation could log that one of our inodes exists.
9126 * We don't need to pin the logs before the above calls to
9127 * btrfs_insert_inode_ref(), since those don't ever need to change a log.
9129 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
9130 btrfs_pin_log_trans(root);
9131 root_log_pinned = true;
9133 if (new_ino != BTRFS_FIRST_FREE_OBJECTID) {
9134 btrfs_pin_log_trans(dest);
9135 dest_log_pinned = true;
9138 /* src is a subvolume */
9139 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9140 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9141 } else { /* src is an inode */
9142 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9143 BTRFS_I(old_dentry->d_inode),
9144 old_dentry->d_name.name,
9145 old_dentry->d_name.len);
9147 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9150 btrfs_abort_transaction(trans, ret);
9154 /* dest is a subvolume */
9155 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9156 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9157 } else { /* dest is an inode */
9158 ret = __btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9159 BTRFS_I(new_dentry->d_inode),
9160 new_dentry->d_name.name,
9161 new_dentry->d_name.len);
9163 ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode));
9166 btrfs_abort_transaction(trans, ret);
9170 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9171 new_dentry->d_name.name,
9172 new_dentry->d_name.len, 0, old_idx);
9174 btrfs_abort_transaction(trans, ret);
9178 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9179 old_dentry->d_name.name,
9180 old_dentry->d_name.len, 0, new_idx);
9182 btrfs_abort_transaction(trans, ret);
9186 if (old_inode->i_nlink == 1)
9187 BTRFS_I(old_inode)->dir_index = old_idx;
9188 if (new_inode->i_nlink == 1)
9189 BTRFS_I(new_inode)->dir_index = new_idx;
9191 if (root_log_pinned) {
9192 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9193 new_dentry->d_parent);
9194 btrfs_end_log_trans(root);
9195 root_log_pinned = false;
9197 if (dest_log_pinned) {
9198 btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir),
9199 old_dentry->d_parent);
9200 btrfs_end_log_trans(dest);
9201 dest_log_pinned = false;
9205 * If we have pinned a log and an error happened, we unpin tasks
9206 * trying to sync the log and force them to fallback to a transaction
9207 * commit if the log currently contains any of the inodes involved in
9208 * this rename operation (to ensure we do not persist a log with an
9209 * inconsistent state for any of these inodes or leading to any
9210 * inconsistencies when replayed). If the transaction was aborted, the
9211 * abortion reason is propagated to userspace when attempting to commit
9212 * the transaction. If the log does not contain any of these inodes, we
9213 * allow the tasks to sync it.
9215 if (ret && (root_log_pinned || dest_log_pinned)) {
9216 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9217 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9218 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9219 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation))
9220 btrfs_set_log_full_commit(trans);
9222 if (root_log_pinned) {
9223 btrfs_end_log_trans(root);
9224 root_log_pinned = false;
9226 if (dest_log_pinned) {
9227 btrfs_end_log_trans(dest);
9228 dest_log_pinned = false;
9231 ret2 = btrfs_end_transaction(trans);
9232 ret = ret ? ret : ret2;
9234 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
9235 old_ino == BTRFS_FIRST_FREE_OBJECTID)
9236 up_read(&fs_info->subvol_sem);
9241 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9242 struct btrfs_root *root,
9243 struct user_namespace *mnt_userns,
9245 struct dentry *dentry)
9248 struct inode *inode;
9252 ret = btrfs_get_free_objectid(root, &objectid);
9256 inode = btrfs_new_inode(trans, root, mnt_userns, dir,
9257 dentry->d_name.name,
9259 btrfs_ino(BTRFS_I(dir)),
9261 S_IFCHR | WHITEOUT_MODE,
9264 if (IS_ERR(inode)) {
9265 ret = PTR_ERR(inode);
9269 inode->i_op = &btrfs_special_inode_operations;
9270 init_special_inode(inode, inode->i_mode,
9273 ret = btrfs_init_inode_security(trans, inode, dir,
9278 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9279 BTRFS_I(inode), 0, index);
9283 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
9285 unlock_new_inode(inode);
9287 inode_dec_link_count(inode);
9293 static int btrfs_rename(struct user_namespace *mnt_userns,
9294 struct inode *old_dir, struct dentry *old_dentry,
9295 struct inode *new_dir, struct dentry *new_dentry,
9298 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9299 struct btrfs_trans_handle *trans;
9300 unsigned int trans_num_items;
9301 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9302 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9303 struct inode *new_inode = d_inode(new_dentry);
9304 struct inode *old_inode = d_inode(old_dentry);
9308 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9309 bool log_pinned = false;
9311 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9314 /* we only allow rename subvolume link between subvolumes */
9315 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9318 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9319 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9322 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9323 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9327 /* check for collisions, even if the name isn't there */
9328 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9329 new_dentry->d_name.name,
9330 new_dentry->d_name.len);
9333 if (ret == -EEXIST) {
9335 * eexist without a new_inode */
9336 if (WARN_ON(!new_inode)) {
9340 /* maybe -EOVERFLOW */
9347 * we're using rename to replace one file with another. Start IO on it
9348 * now so we don't add too much work to the end of the transaction
9350 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9351 filemap_flush(old_inode->i_mapping);
9353 /* close the racy window with snapshot create/destroy ioctl */
9354 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9355 down_read(&fs_info->subvol_sem);
9357 * We want to reserve the absolute worst case amount of items. So if
9358 * both inodes are subvols and we need to unlink them then that would
9359 * require 4 item modifications, but if they are both normal inodes it
9360 * would require 5 item modifications, so we'll assume they are normal
9361 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9362 * should cover the worst case number of items we'll modify.
9363 * If our rename has the whiteout flag, we need more 5 units for the
9364 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9365 * when selinux is enabled).
9367 trans_num_items = 11;
9368 if (flags & RENAME_WHITEOUT)
9369 trans_num_items += 5;
9370 trans = btrfs_start_transaction(root, trans_num_items);
9371 if (IS_ERR(trans)) {
9372 ret = PTR_ERR(trans);
9377 ret = btrfs_record_root_in_trans(trans, dest);
9382 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9386 BTRFS_I(old_inode)->dir_index = 0ULL;
9387 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9388 /* force full log commit if subvolume involved. */
9389 btrfs_set_log_full_commit(trans);
9391 ret = btrfs_insert_inode_ref(trans, dest,
9392 new_dentry->d_name.name,
9393 new_dentry->d_name.len,
9395 btrfs_ino(BTRFS_I(new_dir)), index);
9400 inode_inc_iversion(old_dir);
9401 inode_inc_iversion(new_dir);
9402 inode_inc_iversion(old_inode);
9403 old_dir->i_ctime = old_dir->i_mtime =
9404 new_dir->i_ctime = new_dir->i_mtime =
9405 old_inode->i_ctime = current_time(old_dir);
9407 if (old_dentry->d_parent != new_dentry->d_parent)
9408 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9409 BTRFS_I(old_inode), 1);
9411 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9412 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9415 * Now pin the log. We do it to ensure that no other task can
9416 * sync the log while we are in progress with the rename, as
9417 * that could result in an inconsistency in case any of the
9418 * inodes that are part of this rename operation were logged
9421 * We pin the log even if at this precise moment none of the
9422 * inodes was logged before. This is because right after we
9423 * checked for that, some other task fsyncing some other inode
9424 * not involved with this rename operation could log that one of
9425 * our inodes exists.
9427 * We don't need to pin the logs before the above call to
9428 * btrfs_insert_inode_ref(), since that does not need to change
9431 btrfs_pin_log_trans(root);
9433 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9434 BTRFS_I(d_inode(old_dentry)),
9435 old_dentry->d_name.name,
9436 old_dentry->d_name.len);
9438 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9441 btrfs_abort_transaction(trans, ret);
9446 inode_inc_iversion(new_inode);
9447 new_inode->i_ctime = current_time(new_inode);
9448 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9449 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9450 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9451 BUG_ON(new_inode->i_nlink == 0);
9453 ret = btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9454 BTRFS_I(d_inode(new_dentry)),
9455 new_dentry->d_name.name,
9456 new_dentry->d_name.len);
9458 if (!ret && new_inode->i_nlink == 0)
9459 ret = btrfs_orphan_add(trans,
9460 BTRFS_I(d_inode(new_dentry)));
9462 btrfs_abort_transaction(trans, ret);
9467 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9468 new_dentry->d_name.name,
9469 new_dentry->d_name.len, 0, index);
9471 btrfs_abort_transaction(trans, ret);
9475 if (old_inode->i_nlink == 1)
9476 BTRFS_I(old_inode)->dir_index = index;
9479 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9480 new_dentry->d_parent);
9481 btrfs_end_log_trans(root);
9485 if (flags & RENAME_WHITEOUT) {
9486 ret = btrfs_whiteout_for_rename(trans, root, mnt_userns,
9487 old_dir, old_dentry);
9490 btrfs_abort_transaction(trans, ret);
9496 * If we have pinned the log and an error happened, we unpin tasks
9497 * trying to sync the log and force them to fallback to a transaction
9498 * commit if the log currently contains any of the inodes involved in
9499 * this rename operation (to ensure we do not persist a log with an
9500 * inconsistent state for any of these inodes or leading to any
9501 * inconsistencies when replayed). If the transaction was aborted, the
9502 * abortion reason is propagated to userspace when attempting to commit
9503 * the transaction. If the log does not contain any of these inodes, we
9504 * allow the tasks to sync it.
9506 if (ret && log_pinned) {
9507 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9508 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9509 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9511 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9512 btrfs_set_log_full_commit(trans);
9514 btrfs_end_log_trans(root);
9517 ret2 = btrfs_end_transaction(trans);
9518 ret = ret ? ret : ret2;
9520 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9521 up_read(&fs_info->subvol_sem);
9526 static int btrfs_rename2(struct user_namespace *mnt_userns, struct inode *old_dir,
9527 struct dentry *old_dentry, struct inode *new_dir,
9528 struct dentry *new_dentry, unsigned int flags)
9530 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9533 if (flags & RENAME_EXCHANGE)
9534 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9537 return btrfs_rename(mnt_userns, old_dir, old_dentry, new_dir,
9541 struct btrfs_delalloc_work {
9542 struct inode *inode;
9543 struct completion completion;
9544 struct list_head list;
9545 struct btrfs_work work;
9548 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9550 struct btrfs_delalloc_work *delalloc_work;
9551 struct inode *inode;
9553 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9555 inode = delalloc_work->inode;
9556 filemap_flush(inode->i_mapping);
9557 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9558 &BTRFS_I(inode)->runtime_flags))
9559 filemap_flush(inode->i_mapping);
9562 complete(&delalloc_work->completion);
9565 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9567 struct btrfs_delalloc_work *work;
9569 work = kmalloc(sizeof(*work), GFP_NOFS);
9573 init_completion(&work->completion);
9574 INIT_LIST_HEAD(&work->list);
9575 work->inode = inode;
9576 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9582 * some fairly slow code that needs optimization. This walks the list
9583 * of all the inodes with pending delalloc and forces them to disk.
9585 static int start_delalloc_inodes(struct btrfs_root *root,
9586 struct writeback_control *wbc, bool snapshot,
9587 bool in_reclaim_context)
9589 struct btrfs_inode *binode;
9590 struct inode *inode;
9591 struct btrfs_delalloc_work *work, *next;
9592 struct list_head works;
9593 struct list_head splice;
9595 bool full_flush = wbc->nr_to_write == LONG_MAX;
9597 INIT_LIST_HEAD(&works);
9598 INIT_LIST_HEAD(&splice);
9600 mutex_lock(&root->delalloc_mutex);
9601 spin_lock(&root->delalloc_lock);
9602 list_splice_init(&root->delalloc_inodes, &splice);
9603 while (!list_empty(&splice)) {
9604 binode = list_entry(splice.next, struct btrfs_inode,
9607 list_move_tail(&binode->delalloc_inodes,
9608 &root->delalloc_inodes);
9610 if (in_reclaim_context &&
9611 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9614 inode = igrab(&binode->vfs_inode);
9616 cond_resched_lock(&root->delalloc_lock);
9619 spin_unlock(&root->delalloc_lock);
9622 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9623 &binode->runtime_flags);
9625 work = btrfs_alloc_delalloc_work(inode);
9631 list_add_tail(&work->list, &works);
9632 btrfs_queue_work(root->fs_info->flush_workers,
9635 ret = filemap_fdatawrite_wbc(inode->i_mapping, wbc);
9636 btrfs_add_delayed_iput(inode);
9637 if (ret || wbc->nr_to_write <= 0)
9641 spin_lock(&root->delalloc_lock);
9643 spin_unlock(&root->delalloc_lock);
9646 list_for_each_entry_safe(work, next, &works, list) {
9647 list_del_init(&work->list);
9648 wait_for_completion(&work->completion);
9652 if (!list_empty(&splice)) {
9653 spin_lock(&root->delalloc_lock);
9654 list_splice_tail(&splice, &root->delalloc_inodes);
9655 spin_unlock(&root->delalloc_lock);
9657 mutex_unlock(&root->delalloc_mutex);
9661 int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9663 struct writeback_control wbc = {
9664 .nr_to_write = LONG_MAX,
9665 .sync_mode = WB_SYNC_NONE,
9667 .range_end = LLONG_MAX,
9669 struct btrfs_fs_info *fs_info = root->fs_info;
9671 if (BTRFS_FS_ERROR(fs_info))
9674 return start_delalloc_inodes(root, &wbc, true, in_reclaim_context);
9677 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9678 bool in_reclaim_context)
9680 struct writeback_control wbc = {
9682 .sync_mode = WB_SYNC_NONE,
9684 .range_end = LLONG_MAX,
9686 struct btrfs_root *root;
9687 struct list_head splice;
9690 if (BTRFS_FS_ERROR(fs_info))
9693 INIT_LIST_HEAD(&splice);
9695 mutex_lock(&fs_info->delalloc_root_mutex);
9696 spin_lock(&fs_info->delalloc_root_lock);
9697 list_splice_init(&fs_info->delalloc_roots, &splice);
9698 while (!list_empty(&splice)) {
9700 * Reset nr_to_write here so we know that we're doing a full
9704 wbc.nr_to_write = LONG_MAX;
9706 root = list_first_entry(&splice, struct btrfs_root,
9708 root = btrfs_grab_root(root);
9710 list_move_tail(&root->delalloc_root,
9711 &fs_info->delalloc_roots);
9712 spin_unlock(&fs_info->delalloc_root_lock);
9714 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9715 btrfs_put_root(root);
9716 if (ret < 0 || wbc.nr_to_write <= 0)
9718 spin_lock(&fs_info->delalloc_root_lock);
9720 spin_unlock(&fs_info->delalloc_root_lock);
9724 if (!list_empty(&splice)) {
9725 spin_lock(&fs_info->delalloc_root_lock);
9726 list_splice_tail(&splice, &fs_info->delalloc_roots);
9727 spin_unlock(&fs_info->delalloc_root_lock);
9729 mutex_unlock(&fs_info->delalloc_root_mutex);
9733 static int btrfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
9734 struct dentry *dentry, const char *symname)
9736 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9737 struct btrfs_trans_handle *trans;
9738 struct btrfs_root *root = BTRFS_I(dir)->root;
9739 struct btrfs_path *path;
9740 struct btrfs_key key;
9741 struct inode *inode = NULL;
9748 struct btrfs_file_extent_item *ei;
9749 struct extent_buffer *leaf;
9751 name_len = strlen(symname);
9752 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9753 return -ENAMETOOLONG;
9756 * 2 items for inode item and ref
9757 * 2 items for dir items
9758 * 1 item for updating parent inode item
9759 * 1 item for the inline extent item
9760 * 1 item for xattr if selinux is on
9762 trans = btrfs_start_transaction(root, 7);
9764 return PTR_ERR(trans);
9766 err = btrfs_get_free_objectid(root, &objectid);
9770 inode = btrfs_new_inode(trans, root, mnt_userns, dir,
9771 dentry->d_name.name, dentry->d_name.len,
9772 btrfs_ino(BTRFS_I(dir)), objectid,
9773 S_IFLNK | S_IRWXUGO, &index);
9774 if (IS_ERR(inode)) {
9775 err = PTR_ERR(inode);
9781 * If the active LSM wants to access the inode during
9782 * d_instantiate it needs these. Smack checks to see
9783 * if the filesystem supports xattrs by looking at the
9786 inode->i_fop = &btrfs_file_operations;
9787 inode->i_op = &btrfs_file_inode_operations;
9788 inode->i_mapping->a_ops = &btrfs_aops;
9790 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9794 path = btrfs_alloc_path();
9799 key.objectid = btrfs_ino(BTRFS_I(inode));
9801 key.type = BTRFS_EXTENT_DATA_KEY;
9802 datasize = btrfs_file_extent_calc_inline_size(name_len);
9803 err = btrfs_insert_empty_item(trans, root, path, &key,
9806 btrfs_free_path(path);
9809 leaf = path->nodes[0];
9810 ei = btrfs_item_ptr(leaf, path->slots[0],
9811 struct btrfs_file_extent_item);
9812 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9813 btrfs_set_file_extent_type(leaf, ei,
9814 BTRFS_FILE_EXTENT_INLINE);
9815 btrfs_set_file_extent_encryption(leaf, ei, 0);
9816 btrfs_set_file_extent_compression(leaf, ei, 0);
9817 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9818 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9820 ptr = btrfs_file_extent_inline_start(ei);
9821 write_extent_buffer(leaf, symname, ptr, name_len);
9822 btrfs_mark_buffer_dirty(leaf);
9823 btrfs_free_path(path);
9825 inode->i_op = &btrfs_symlink_inode_operations;
9826 inode_nohighmem(inode);
9827 inode_set_bytes(inode, name_len);
9828 btrfs_i_size_write(BTRFS_I(inode), name_len);
9829 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
9831 * Last step, add directory indexes for our symlink inode. This is the
9832 * last step to avoid extra cleanup of these indexes if an error happens
9836 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9837 BTRFS_I(inode), 0, index);
9841 d_instantiate_new(dentry, inode);
9844 btrfs_end_transaction(trans);
9846 inode_dec_link_count(inode);
9847 discard_new_inode(inode);
9849 btrfs_btree_balance_dirty(fs_info);
9853 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9854 struct btrfs_trans_handle *trans_in,
9855 struct btrfs_inode *inode,
9856 struct btrfs_key *ins,
9859 struct btrfs_file_extent_item stack_fi;
9860 struct btrfs_replace_extent_info extent_info;
9861 struct btrfs_trans_handle *trans = trans_in;
9862 struct btrfs_path *path;
9863 u64 start = ins->objectid;
9864 u64 len = ins->offset;
9865 int qgroup_released;
9868 memset(&stack_fi, 0, sizeof(stack_fi));
9870 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9871 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9872 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9873 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9874 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9875 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9876 /* Encryption and other encoding is reserved and all 0 */
9878 qgroup_released = btrfs_qgroup_release_data(inode, file_offset, len);
9879 if (qgroup_released < 0)
9880 return ERR_PTR(qgroup_released);
9883 ret = insert_reserved_file_extent(trans, inode,
9884 file_offset, &stack_fi,
9885 true, qgroup_released);
9891 extent_info.disk_offset = start;
9892 extent_info.disk_len = len;
9893 extent_info.data_offset = 0;
9894 extent_info.data_len = len;
9895 extent_info.file_offset = file_offset;
9896 extent_info.extent_buf = (char *)&stack_fi;
9897 extent_info.is_new_extent = true;
9898 extent_info.qgroup_reserved = qgroup_released;
9899 extent_info.insertions = 0;
9901 path = btrfs_alloc_path();
9907 ret = btrfs_replace_file_extents(inode, path, file_offset,
9908 file_offset + len - 1, &extent_info,
9910 btrfs_free_path(path);
9917 * We have released qgroup data range at the beginning of the function,
9918 * and normally qgroup_released bytes will be freed when committing
9920 * But if we error out early, we have to free what we have released
9921 * or we leak qgroup data reservation.
9923 btrfs_qgroup_free_refroot(inode->root->fs_info,
9924 inode->root->root_key.objectid, qgroup_released,
9925 BTRFS_QGROUP_RSV_DATA);
9926 return ERR_PTR(ret);
9929 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9930 u64 start, u64 num_bytes, u64 min_size,
9931 loff_t actual_len, u64 *alloc_hint,
9932 struct btrfs_trans_handle *trans)
9934 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9935 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9936 struct extent_map *em;
9937 struct btrfs_root *root = BTRFS_I(inode)->root;
9938 struct btrfs_key ins;
9939 u64 cur_offset = start;
9940 u64 clear_offset = start;
9943 u64 last_alloc = (u64)-1;
9945 bool own_trans = true;
9946 u64 end = start + num_bytes - 1;
9950 while (num_bytes > 0) {
9951 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9952 cur_bytes = max(cur_bytes, min_size);
9954 * If we are severely fragmented we could end up with really
9955 * small allocations, so if the allocator is returning small
9956 * chunks lets make its job easier by only searching for those
9959 cur_bytes = min(cur_bytes, last_alloc);
9960 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9961 min_size, 0, *alloc_hint, &ins, 1, 0);
9966 * We've reserved this space, and thus converted it from
9967 * ->bytes_may_use to ->bytes_reserved. Any error that happens
9968 * from here on out we will only need to clear our reservation
9969 * for the remaining unreserved area, so advance our
9970 * clear_offset by our extent size.
9972 clear_offset += ins.offset;
9974 last_alloc = ins.offset;
9975 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
9978 * Now that we inserted the prealloc extent we can finally
9979 * decrement the number of reservations in the block group.
9980 * If we did it before, we could race with relocation and have
9981 * relocation miss the reserved extent, making it fail later.
9983 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
9984 if (IS_ERR(trans)) {
9985 ret = PTR_ERR(trans);
9986 btrfs_free_reserved_extent(fs_info, ins.objectid,
9991 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
9992 cur_offset + ins.offset -1, 0);
9994 em = alloc_extent_map();
9996 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9997 &BTRFS_I(inode)->runtime_flags);
10001 em->start = cur_offset;
10002 em->orig_start = cur_offset;
10003 em->len = ins.offset;
10004 em->block_start = ins.objectid;
10005 em->block_len = ins.offset;
10006 em->orig_block_len = ins.offset;
10007 em->ram_bytes = ins.offset;
10008 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10009 em->generation = trans->transid;
10012 write_lock(&em_tree->lock);
10013 ret = add_extent_mapping(em_tree, em, 1);
10014 write_unlock(&em_tree->lock);
10015 if (ret != -EEXIST)
10017 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10018 cur_offset + ins.offset - 1,
10021 free_extent_map(em);
10023 num_bytes -= ins.offset;
10024 cur_offset += ins.offset;
10025 *alloc_hint = ins.objectid + ins.offset;
10027 inode_inc_iversion(inode);
10028 inode->i_ctime = current_time(inode);
10029 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10030 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10031 (actual_len > inode->i_size) &&
10032 (cur_offset > inode->i_size)) {
10033 if (cur_offset > actual_len)
10034 i_size = actual_len;
10036 i_size = cur_offset;
10037 i_size_write(inode, i_size);
10038 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
10041 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
10044 btrfs_abort_transaction(trans, ret);
10046 btrfs_end_transaction(trans);
10051 btrfs_end_transaction(trans);
10055 if (clear_offset < end)
10056 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
10057 end - clear_offset + 1);
10061 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10062 u64 start, u64 num_bytes, u64 min_size,
10063 loff_t actual_len, u64 *alloc_hint)
10065 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10066 min_size, actual_len, alloc_hint,
10070 int btrfs_prealloc_file_range_trans(struct inode *inode,
10071 struct btrfs_trans_handle *trans, int mode,
10072 u64 start, u64 num_bytes, u64 min_size,
10073 loff_t actual_len, u64 *alloc_hint)
10075 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10076 min_size, actual_len, alloc_hint, trans);
10079 static int btrfs_set_page_dirty(struct page *page)
10081 return __set_page_dirty_nobuffers(page);
10084 static int btrfs_permission(struct user_namespace *mnt_userns,
10085 struct inode *inode, int mask)
10087 struct btrfs_root *root = BTRFS_I(inode)->root;
10088 umode_t mode = inode->i_mode;
10090 if (mask & MAY_WRITE &&
10091 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10092 if (btrfs_root_readonly(root))
10094 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10097 return generic_permission(mnt_userns, inode, mask);
10100 static int btrfs_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
10101 struct dentry *dentry, umode_t mode)
10103 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10104 struct btrfs_trans_handle *trans;
10105 struct btrfs_root *root = BTRFS_I(dir)->root;
10106 struct inode *inode = NULL;
10112 * 5 units required for adding orphan entry
10114 trans = btrfs_start_transaction(root, 5);
10116 return PTR_ERR(trans);
10118 ret = btrfs_get_free_objectid(root, &objectid);
10122 inode = btrfs_new_inode(trans, root, mnt_userns, dir, NULL, 0,
10123 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10124 if (IS_ERR(inode)) {
10125 ret = PTR_ERR(inode);
10130 inode->i_fop = &btrfs_file_operations;
10131 inode->i_op = &btrfs_file_inode_operations;
10133 inode->i_mapping->a_ops = &btrfs_aops;
10135 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10139 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
10142 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10147 * We set number of links to 0 in btrfs_new_inode(), and here we set
10148 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10151 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10153 set_nlink(inode, 1);
10154 d_tmpfile(dentry, inode);
10155 unlock_new_inode(inode);
10156 mark_inode_dirty(inode);
10158 btrfs_end_transaction(trans);
10160 discard_new_inode(inode);
10161 btrfs_btree_balance_dirty(fs_info);
10165 void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end)
10167 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10168 unsigned long index = start >> PAGE_SHIFT;
10169 unsigned long end_index = end >> PAGE_SHIFT;
10173 ASSERT(end + 1 - start <= U32_MAX);
10174 len = end + 1 - start;
10175 while (index <= end_index) {
10176 page = find_get_page(inode->vfs_inode.i_mapping, index);
10177 ASSERT(page); /* Pages should be in the extent_io_tree */
10179 btrfs_page_set_writeback(fs_info, page, start, len);
10187 * Add an entry indicating a block group or device which is pinned by a
10188 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10189 * negative errno on failure.
10191 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10192 bool is_block_group)
10194 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10195 struct btrfs_swapfile_pin *sp, *entry;
10196 struct rb_node **p;
10197 struct rb_node *parent = NULL;
10199 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10204 sp->is_block_group = is_block_group;
10205 sp->bg_extent_count = 1;
10207 spin_lock(&fs_info->swapfile_pins_lock);
10208 p = &fs_info->swapfile_pins.rb_node;
10211 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10212 if (sp->ptr < entry->ptr ||
10213 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10214 p = &(*p)->rb_left;
10215 } else if (sp->ptr > entry->ptr ||
10216 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10217 p = &(*p)->rb_right;
10219 if (is_block_group)
10220 entry->bg_extent_count++;
10221 spin_unlock(&fs_info->swapfile_pins_lock);
10226 rb_link_node(&sp->node, parent, p);
10227 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10228 spin_unlock(&fs_info->swapfile_pins_lock);
10232 /* Free all of the entries pinned by this swapfile. */
10233 static void btrfs_free_swapfile_pins(struct inode *inode)
10235 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10236 struct btrfs_swapfile_pin *sp;
10237 struct rb_node *node, *next;
10239 spin_lock(&fs_info->swapfile_pins_lock);
10240 node = rb_first(&fs_info->swapfile_pins);
10242 next = rb_next(node);
10243 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10244 if (sp->inode == inode) {
10245 rb_erase(&sp->node, &fs_info->swapfile_pins);
10246 if (sp->is_block_group) {
10247 btrfs_dec_block_group_swap_extents(sp->ptr,
10248 sp->bg_extent_count);
10249 btrfs_put_block_group(sp->ptr);
10255 spin_unlock(&fs_info->swapfile_pins_lock);
10258 struct btrfs_swap_info {
10264 unsigned long nr_pages;
10268 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10269 struct btrfs_swap_info *bsi)
10271 unsigned long nr_pages;
10272 unsigned long max_pages;
10273 u64 first_ppage, first_ppage_reported, next_ppage;
10277 * Our swapfile may have had its size extended after the swap header was
10278 * written. In that case activating the swapfile should not go beyond
10279 * the max size set in the swap header.
10281 if (bsi->nr_pages >= sis->max)
10284 max_pages = sis->max - bsi->nr_pages;
10285 first_ppage = ALIGN(bsi->block_start, PAGE_SIZE) >> PAGE_SHIFT;
10286 next_ppage = ALIGN_DOWN(bsi->block_start + bsi->block_len,
10287 PAGE_SIZE) >> PAGE_SHIFT;
10289 if (first_ppage >= next_ppage)
10291 nr_pages = next_ppage - first_ppage;
10292 nr_pages = min(nr_pages, max_pages);
10294 first_ppage_reported = first_ppage;
10295 if (bsi->start == 0)
10296 first_ppage_reported++;
10297 if (bsi->lowest_ppage > first_ppage_reported)
10298 bsi->lowest_ppage = first_ppage_reported;
10299 if (bsi->highest_ppage < (next_ppage - 1))
10300 bsi->highest_ppage = next_ppage - 1;
10302 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10305 bsi->nr_extents += ret;
10306 bsi->nr_pages += nr_pages;
10310 static void btrfs_swap_deactivate(struct file *file)
10312 struct inode *inode = file_inode(file);
10314 btrfs_free_swapfile_pins(inode);
10315 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10318 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10321 struct inode *inode = file_inode(file);
10322 struct btrfs_root *root = BTRFS_I(inode)->root;
10323 struct btrfs_fs_info *fs_info = root->fs_info;
10324 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10325 struct extent_state *cached_state = NULL;
10326 struct extent_map *em = NULL;
10327 struct btrfs_device *device = NULL;
10328 struct btrfs_swap_info bsi = {
10329 .lowest_ppage = (sector_t)-1ULL,
10336 * If the swap file was just created, make sure delalloc is done. If the
10337 * file changes again after this, the user is doing something stupid and
10338 * we don't really care.
10340 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10345 * The inode is locked, so these flags won't change after we check them.
10347 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10348 btrfs_warn(fs_info, "swapfile must not be compressed");
10351 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10352 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10355 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10356 btrfs_warn(fs_info, "swapfile must not be checksummed");
10361 * Balance or device remove/replace/resize can move stuff around from
10362 * under us. The exclop protection makes sure they aren't running/won't
10363 * run concurrently while we are mapping the swap extents, and
10364 * fs_info->swapfile_pins prevents them from running while the swap
10365 * file is active and moving the extents. Note that this also prevents
10366 * a concurrent device add which isn't actually necessary, but it's not
10367 * really worth the trouble to allow it.
10369 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10370 btrfs_warn(fs_info,
10371 "cannot activate swapfile while exclusive operation is running");
10376 * Prevent snapshot creation while we are activating the swap file.
10377 * We do not want to race with snapshot creation. If snapshot creation
10378 * already started before we bumped nr_swapfiles from 0 to 1 and
10379 * completes before the first write into the swap file after it is
10380 * activated, than that write would fallback to COW.
10382 if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) {
10383 btrfs_exclop_finish(fs_info);
10384 btrfs_warn(fs_info,
10385 "cannot activate swapfile because snapshot creation is in progress");
10389 * Snapshots can create extents which require COW even if NODATACOW is
10390 * set. We use this counter to prevent snapshots. We must increment it
10391 * before walking the extents because we don't want a concurrent
10392 * snapshot to run after we've already checked the extents.
10394 atomic_inc(&root->nr_swapfiles);
10396 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10398 lock_extent_bits(io_tree, 0, isize - 1, &cached_state);
10400 while (start < isize) {
10401 u64 logical_block_start, physical_block_start;
10402 struct btrfs_block_group *bg;
10403 u64 len = isize - start;
10405 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
10411 if (em->block_start == EXTENT_MAP_HOLE) {
10412 btrfs_warn(fs_info, "swapfile must not have holes");
10416 if (em->block_start == EXTENT_MAP_INLINE) {
10418 * It's unlikely we'll ever actually find ourselves
10419 * here, as a file small enough to fit inline won't be
10420 * big enough to store more than the swap header, but in
10421 * case something changes in the future, let's catch it
10422 * here rather than later.
10424 btrfs_warn(fs_info, "swapfile must not be inline");
10428 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10429 btrfs_warn(fs_info, "swapfile must not be compressed");
10434 logical_block_start = em->block_start + (start - em->start);
10435 len = min(len, em->len - (start - em->start));
10436 free_extent_map(em);
10439 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, true);
10445 btrfs_warn(fs_info,
10446 "swapfile must not be copy-on-write");
10451 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10457 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10458 btrfs_warn(fs_info,
10459 "swapfile must have single data profile");
10464 if (device == NULL) {
10465 device = em->map_lookup->stripes[0].dev;
10466 ret = btrfs_add_swapfile_pin(inode, device, false);
10471 } else if (device != em->map_lookup->stripes[0].dev) {
10472 btrfs_warn(fs_info, "swapfile must be on one device");
10477 physical_block_start = (em->map_lookup->stripes[0].physical +
10478 (logical_block_start - em->start));
10479 len = min(len, em->len - (logical_block_start - em->start));
10480 free_extent_map(em);
10483 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10485 btrfs_warn(fs_info,
10486 "could not find block group containing swapfile");
10491 if (!btrfs_inc_block_group_swap_extents(bg)) {
10492 btrfs_warn(fs_info,
10493 "block group for swapfile at %llu is read-only%s",
10495 atomic_read(&fs_info->scrubs_running) ?
10496 " (scrub running)" : "");
10497 btrfs_put_block_group(bg);
10502 ret = btrfs_add_swapfile_pin(inode, bg, true);
10504 btrfs_put_block_group(bg);
10511 if (bsi.block_len &&
10512 bsi.block_start + bsi.block_len == physical_block_start) {
10513 bsi.block_len += len;
10515 if (bsi.block_len) {
10516 ret = btrfs_add_swap_extent(sis, &bsi);
10521 bsi.block_start = physical_block_start;
10522 bsi.block_len = len;
10529 ret = btrfs_add_swap_extent(sis, &bsi);
10532 if (!IS_ERR_OR_NULL(em))
10533 free_extent_map(em);
10535 unlock_extent_cached(io_tree, 0, isize - 1, &cached_state);
10538 btrfs_swap_deactivate(file);
10540 btrfs_drew_write_unlock(&root->snapshot_lock);
10542 btrfs_exclop_finish(fs_info);
10548 sis->bdev = device->bdev;
10549 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10550 sis->max = bsi.nr_pages;
10551 sis->pages = bsi.nr_pages - 1;
10552 sis->highest_bit = bsi.nr_pages - 1;
10553 return bsi.nr_extents;
10556 static void btrfs_swap_deactivate(struct file *file)
10560 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10563 return -EOPNOTSUPP;
10568 * Update the number of bytes used in the VFS' inode. When we replace extents in
10569 * a range (clone, dedupe, fallocate's zero range), we must update the number of
10570 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
10571 * always get a correct value.
10573 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
10574 const u64 add_bytes,
10575 const u64 del_bytes)
10577 if (add_bytes == del_bytes)
10580 spin_lock(&inode->lock);
10582 inode_sub_bytes(&inode->vfs_inode, del_bytes);
10584 inode_add_bytes(&inode->vfs_inode, add_bytes);
10585 spin_unlock(&inode->lock);
10588 static const struct inode_operations btrfs_dir_inode_operations = {
10589 .getattr = btrfs_getattr,
10590 .lookup = btrfs_lookup,
10591 .create = btrfs_create,
10592 .unlink = btrfs_unlink,
10593 .link = btrfs_link,
10594 .mkdir = btrfs_mkdir,
10595 .rmdir = btrfs_rmdir,
10596 .rename = btrfs_rename2,
10597 .symlink = btrfs_symlink,
10598 .setattr = btrfs_setattr,
10599 .mknod = btrfs_mknod,
10600 .listxattr = btrfs_listxattr,
10601 .permission = btrfs_permission,
10602 .get_acl = btrfs_get_acl,
10603 .set_acl = btrfs_set_acl,
10604 .update_time = btrfs_update_time,
10605 .tmpfile = btrfs_tmpfile,
10606 .fileattr_get = btrfs_fileattr_get,
10607 .fileattr_set = btrfs_fileattr_set,
10610 static const struct file_operations btrfs_dir_file_operations = {
10611 .llseek = generic_file_llseek,
10612 .read = generic_read_dir,
10613 .iterate_shared = btrfs_real_readdir,
10614 .open = btrfs_opendir,
10615 .unlocked_ioctl = btrfs_ioctl,
10616 #ifdef CONFIG_COMPAT
10617 .compat_ioctl = btrfs_compat_ioctl,
10619 .release = btrfs_release_file,
10620 .fsync = btrfs_sync_file,
10624 * btrfs doesn't support the bmap operation because swapfiles
10625 * use bmap to make a mapping of extents in the file. They assume
10626 * these extents won't change over the life of the file and they
10627 * use the bmap result to do IO directly to the drive.
10629 * the btrfs bmap call would return logical addresses that aren't
10630 * suitable for IO and they also will change frequently as COW
10631 * operations happen. So, swapfile + btrfs == corruption.
10633 * For now we're avoiding this by dropping bmap.
10635 static const struct address_space_operations btrfs_aops = {
10636 .readpage = btrfs_readpage,
10637 .writepage = btrfs_writepage,
10638 .writepages = btrfs_writepages,
10639 .readahead = btrfs_readahead,
10640 .direct_IO = noop_direct_IO,
10641 .invalidatepage = btrfs_invalidatepage,
10642 .releasepage = btrfs_releasepage,
10643 #ifdef CONFIG_MIGRATION
10644 .migratepage = btrfs_migratepage,
10646 .set_page_dirty = btrfs_set_page_dirty,
10647 .error_remove_page = generic_error_remove_page,
10648 .swap_activate = btrfs_swap_activate,
10649 .swap_deactivate = btrfs_swap_deactivate,
10652 static const struct inode_operations btrfs_file_inode_operations = {
10653 .getattr = btrfs_getattr,
10654 .setattr = btrfs_setattr,
10655 .listxattr = btrfs_listxattr,
10656 .permission = btrfs_permission,
10657 .fiemap = btrfs_fiemap,
10658 .get_acl = btrfs_get_acl,
10659 .set_acl = btrfs_set_acl,
10660 .update_time = btrfs_update_time,
10661 .fileattr_get = btrfs_fileattr_get,
10662 .fileattr_set = btrfs_fileattr_set,
10664 static const struct inode_operations btrfs_special_inode_operations = {
10665 .getattr = btrfs_getattr,
10666 .setattr = btrfs_setattr,
10667 .permission = btrfs_permission,
10668 .listxattr = btrfs_listxattr,
10669 .get_acl = btrfs_get_acl,
10670 .set_acl = btrfs_set_acl,
10671 .update_time = btrfs_update_time,
10673 static const struct inode_operations btrfs_symlink_inode_operations = {
10674 .get_link = page_get_link,
10675 .getattr = btrfs_getattr,
10676 .setattr = btrfs_setattr,
10677 .permission = btrfs_permission,
10678 .listxattr = btrfs_listxattr,
10679 .update_time = btrfs_update_time,
10682 const struct dentry_operations btrfs_dentry_operations = {
10683 .d_delete = btrfs_dentry_delete,