1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <crypto/hash.h>
7 #include <linux/kernel.h>
9 #include <linux/file.h>
11 #include <linux/pagemap.h>
12 #include <linux/highmem.h>
13 #include <linux/time.h>
14 #include <linux/init.h>
15 #include <linux/string.h>
16 #include <linux/backing-dev.h>
17 #include <linux/writeback.h>
18 #include <linux/compat.h>
19 #include <linux/xattr.h>
20 #include <linux/posix_acl.h>
21 #include <linux/falloc.h>
22 #include <linux/slab.h>
23 #include <linux/ratelimit.h>
24 #include <linux/btrfs.h>
25 #include <linux/blkdev.h>
26 #include <linux/posix_acl_xattr.h>
27 #include <linux/uio.h>
28 #include <linux/magic.h>
29 #include <linux/iversion.h>
30 #include <linux/swap.h>
31 #include <linux/migrate.h>
32 #include <linux/sched/mm.h>
33 #include <linux/iomap.h>
34 #include <asm/unaligned.h>
38 #include "transaction.h"
39 #include "btrfs_inode.h"
40 #include "print-tree.h"
41 #include "ordered-data.h"
45 #include "compression.h"
47 #include "free-space-cache.h"
50 #include "delalloc-space.h"
51 #include "block-group.h"
52 #include "space-info.h"
56 struct btrfs_iget_args {
58 struct btrfs_root *root;
61 struct btrfs_dio_data {
65 struct extent_changeset *data_reserved;
68 static const struct inode_operations btrfs_dir_inode_operations;
69 static const struct inode_operations btrfs_symlink_inode_operations;
70 static const struct inode_operations btrfs_special_inode_operations;
71 static const struct inode_operations btrfs_file_inode_operations;
72 static const struct address_space_operations btrfs_aops;
73 static const struct file_operations btrfs_dir_file_operations;
75 static struct kmem_cache *btrfs_inode_cachep;
76 struct kmem_cache *btrfs_trans_handle_cachep;
77 struct kmem_cache *btrfs_path_cachep;
78 struct kmem_cache *btrfs_free_space_cachep;
79 struct kmem_cache *btrfs_free_space_bitmap_cachep;
81 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
82 static int btrfs_truncate(struct inode *inode, bool skip_writeback);
83 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
84 static noinline int cow_file_range(struct btrfs_inode *inode,
85 struct page *locked_page,
86 u64 start, u64 end, int *page_started,
87 unsigned long *nr_written, int unlock);
88 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
89 u64 len, u64 orig_start, u64 block_start,
90 u64 block_len, u64 orig_block_len,
91 u64 ram_bytes, int compress_type,
94 static void __endio_write_update_ordered(struct btrfs_inode *inode,
95 const u64 offset, const u64 bytes,
99 * btrfs_inode_lock - lock inode i_rwsem based on arguments passed
101 * ilock_flags can have the following bit set:
103 * BTRFS_ILOCK_SHARED - acquire a shared lock on the inode
104 * BTRFS_ILOCK_TRY - try to acquire the lock, if fails on first attempt
106 * BTRFS_ILOCK_MMAP - acquire a write lock on the i_mmap_lock
108 int btrfs_inode_lock(struct inode *inode, unsigned int ilock_flags)
110 if (ilock_flags & BTRFS_ILOCK_SHARED) {
111 if (ilock_flags & BTRFS_ILOCK_TRY) {
112 if (!inode_trylock_shared(inode))
117 inode_lock_shared(inode);
119 if (ilock_flags & BTRFS_ILOCK_TRY) {
120 if (!inode_trylock(inode))
127 if (ilock_flags & BTRFS_ILOCK_MMAP)
128 down_write(&BTRFS_I(inode)->i_mmap_lock);
133 * btrfs_inode_unlock - unock inode i_rwsem
135 * ilock_flags should contain the same bits set as passed to btrfs_inode_lock()
136 * to decide whether the lock acquired is shared or exclusive.
138 void btrfs_inode_unlock(struct inode *inode, unsigned int ilock_flags)
140 if (ilock_flags & BTRFS_ILOCK_MMAP)
141 up_write(&BTRFS_I(inode)->i_mmap_lock);
142 if (ilock_flags & BTRFS_ILOCK_SHARED)
143 inode_unlock_shared(inode);
149 * Cleanup all submitted ordered extents in specified range to handle errors
150 * from the btrfs_run_delalloc_range() callback.
152 * NOTE: caller must ensure that when an error happens, it can not call
153 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
154 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
155 * to be released, which we want to happen only when finishing the ordered
156 * extent (btrfs_finish_ordered_io()).
158 static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode,
159 struct page *locked_page,
160 u64 offset, u64 bytes)
162 unsigned long index = offset >> PAGE_SHIFT;
163 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
164 u64 page_start = page_offset(locked_page);
165 u64 page_end = page_start + PAGE_SIZE - 1;
169 while (index <= end_index) {
171 * For locked page, we will call end_extent_writepage() on it
172 * in run_delalloc_range() for the error handling. That
173 * end_extent_writepage() function will call
174 * btrfs_mark_ordered_io_finished() to clear page Ordered and
175 * run the ordered extent accounting.
177 * Here we can't just clear the Ordered bit, or
178 * btrfs_mark_ordered_io_finished() would skip the accounting
179 * for the page range, and the ordered extent will never finish.
181 if (index == (page_offset(locked_page) >> PAGE_SHIFT)) {
185 page = find_get_page(inode->vfs_inode.i_mapping, index);
191 * Here we just clear all Ordered bits for every page in the
192 * range, then __endio_write_update_ordered() will handle
193 * the ordered extent accounting for the range.
195 btrfs_page_clamp_clear_ordered(inode->root->fs_info, page,
200 /* The locked page covers the full range, nothing needs to be done */
201 if (bytes + offset <= page_offset(locked_page) + PAGE_SIZE)
204 * In case this page belongs to the delalloc range being instantiated
205 * then skip it, since the first page of a range is going to be
206 * properly cleaned up by the caller of run_delalloc_range
208 if (page_start >= offset && page_end <= (offset + bytes - 1)) {
209 bytes = offset + bytes - page_offset(locked_page) - PAGE_SIZE;
210 offset = page_offset(locked_page) + PAGE_SIZE;
213 return __endio_write_update_ordered(inode, offset, bytes, false);
216 static int btrfs_dirty_inode(struct inode *inode);
218 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
219 struct inode *inode, struct inode *dir,
220 const struct qstr *qstr)
224 err = btrfs_init_acl(trans, inode, dir);
226 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
231 * this does all the hard work for inserting an inline extent into
232 * the btree. The caller should have done a btrfs_drop_extents so that
233 * no overlapping inline items exist in the btree
235 static int insert_inline_extent(struct btrfs_trans_handle *trans,
236 struct btrfs_path *path, bool extent_inserted,
237 struct btrfs_root *root, struct inode *inode,
238 u64 start, size_t size, size_t compressed_size,
240 struct page **compressed_pages)
242 struct extent_buffer *leaf;
243 struct page *page = NULL;
246 struct btrfs_file_extent_item *ei;
248 size_t cur_size = size;
249 unsigned long offset;
251 ASSERT((compressed_size > 0 && compressed_pages) ||
252 (compressed_size == 0 && !compressed_pages));
254 if (compressed_size && compressed_pages)
255 cur_size = compressed_size;
257 if (!extent_inserted) {
258 struct btrfs_key key;
261 key.objectid = btrfs_ino(BTRFS_I(inode));
263 key.type = BTRFS_EXTENT_DATA_KEY;
265 datasize = btrfs_file_extent_calc_inline_size(cur_size);
266 ret = btrfs_insert_empty_item(trans, root, path, &key,
271 leaf = path->nodes[0];
272 ei = btrfs_item_ptr(leaf, path->slots[0],
273 struct btrfs_file_extent_item);
274 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
275 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
276 btrfs_set_file_extent_encryption(leaf, ei, 0);
277 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
278 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
279 ptr = btrfs_file_extent_inline_start(ei);
281 if (compress_type != BTRFS_COMPRESS_NONE) {
284 while (compressed_size > 0) {
285 cpage = compressed_pages[i];
286 cur_size = min_t(unsigned long, compressed_size,
289 kaddr = kmap_atomic(cpage);
290 write_extent_buffer(leaf, kaddr, ptr, cur_size);
291 kunmap_atomic(kaddr);
295 compressed_size -= cur_size;
297 btrfs_set_file_extent_compression(leaf, ei,
300 page = find_get_page(inode->i_mapping,
301 start >> PAGE_SHIFT);
302 btrfs_set_file_extent_compression(leaf, ei, 0);
303 kaddr = kmap_atomic(page);
304 offset = offset_in_page(start);
305 write_extent_buffer(leaf, kaddr + offset, ptr, size);
306 kunmap_atomic(kaddr);
309 btrfs_mark_buffer_dirty(leaf);
310 btrfs_release_path(path);
313 * We align size to sectorsize for inline extents just for simplicity
316 size = ALIGN(size, root->fs_info->sectorsize);
317 ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start, size);
322 * we're an inline extent, so nobody can
323 * extend the file past i_size without locking
324 * a page we already have locked.
326 * We must do any isize and inode updates
327 * before we unlock the pages. Otherwise we
328 * could end up racing with unlink.
330 BTRFS_I(inode)->disk_i_size = inode->i_size;
337 * conditionally insert an inline extent into the file. This
338 * does the checks required to make sure the data is small enough
339 * to fit as an inline extent.
341 static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 start,
342 u64 end, size_t compressed_size,
344 struct page **compressed_pages)
346 struct btrfs_drop_extents_args drop_args = { 0 };
347 struct btrfs_root *root = inode->root;
348 struct btrfs_fs_info *fs_info = root->fs_info;
349 struct btrfs_trans_handle *trans;
350 u64 isize = i_size_read(&inode->vfs_inode);
351 u64 actual_end = min(end + 1, isize);
352 u64 inline_len = actual_end - start;
353 u64 aligned_end = ALIGN(end, fs_info->sectorsize);
354 u64 data_len = inline_len;
356 struct btrfs_path *path;
359 data_len = compressed_size;
362 actual_end > fs_info->sectorsize ||
363 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
365 (actual_end & (fs_info->sectorsize - 1)) == 0) ||
367 data_len > fs_info->max_inline) {
371 path = btrfs_alloc_path();
375 trans = btrfs_join_transaction(root);
377 btrfs_free_path(path);
378 return PTR_ERR(trans);
380 trans->block_rsv = &inode->block_rsv;
382 drop_args.path = path;
383 drop_args.start = start;
384 drop_args.end = aligned_end;
385 drop_args.drop_cache = true;
386 drop_args.replace_extent = true;
388 if (compressed_size && compressed_pages)
389 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(
392 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(
395 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
397 btrfs_abort_transaction(trans, ret);
401 if (isize > actual_end)
402 inline_len = min_t(u64, isize, actual_end);
403 ret = insert_inline_extent(trans, path, drop_args.extent_inserted,
404 root, &inode->vfs_inode, start,
405 inline_len, compressed_size,
406 compress_type, compressed_pages);
407 if (ret && ret != -ENOSPC) {
408 btrfs_abort_transaction(trans, ret);
410 } else if (ret == -ENOSPC) {
415 btrfs_update_inode_bytes(inode, inline_len, drop_args.bytes_found);
416 ret = btrfs_update_inode(trans, root, inode);
417 if (ret && ret != -ENOSPC) {
418 btrfs_abort_transaction(trans, ret);
420 } else if (ret == -ENOSPC) {
425 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
428 * Don't forget to free the reserved space, as for inlined extent
429 * it won't count as data extent, free them directly here.
430 * And at reserve time, it's always aligned to page size, so
431 * just free one page here.
433 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
434 btrfs_free_path(path);
435 btrfs_end_transaction(trans);
439 struct async_extent {
444 unsigned long nr_pages;
446 struct list_head list;
451 struct page *locked_page;
454 unsigned int write_flags;
455 struct list_head extents;
456 struct cgroup_subsys_state *blkcg_css;
457 struct btrfs_work work;
462 /* Number of chunks in flight; must be first in the structure */
464 struct async_chunk chunks[];
467 static noinline int add_async_extent(struct async_chunk *cow,
468 u64 start, u64 ram_size,
471 unsigned long nr_pages,
474 struct async_extent *async_extent;
476 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
477 BUG_ON(!async_extent); /* -ENOMEM */
478 async_extent->start = start;
479 async_extent->ram_size = ram_size;
480 async_extent->compressed_size = compressed_size;
481 async_extent->pages = pages;
482 async_extent->nr_pages = nr_pages;
483 async_extent->compress_type = compress_type;
484 list_add_tail(&async_extent->list, &cow->extents);
489 * Check if the inode has flags compatible with compression
491 static inline bool inode_can_compress(struct btrfs_inode *inode)
493 if (inode->flags & BTRFS_INODE_NODATACOW ||
494 inode->flags & BTRFS_INODE_NODATASUM)
500 * Check if the inode needs to be submitted to compression, based on mount
501 * options, defragmentation, properties or heuristics.
503 static inline int inode_need_compress(struct btrfs_inode *inode, u64 start,
506 struct btrfs_fs_info *fs_info = inode->root->fs_info;
508 if (!inode_can_compress(inode)) {
509 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
510 KERN_ERR "BTRFS: unexpected compression for ino %llu\n",
515 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
518 if (inode->defrag_compress)
520 /* bad compression ratios */
521 if (inode->flags & BTRFS_INODE_NOCOMPRESS)
523 if (btrfs_test_opt(fs_info, COMPRESS) ||
524 inode->flags & BTRFS_INODE_COMPRESS ||
525 inode->prop_compress)
526 return btrfs_compress_heuristic(&inode->vfs_inode, start, end);
530 static inline void inode_should_defrag(struct btrfs_inode *inode,
531 u64 start, u64 end, u64 num_bytes, u64 small_write)
533 /* If this is a small write inside eof, kick off a defrag */
534 if (num_bytes < small_write &&
535 (start > 0 || end + 1 < inode->disk_i_size))
536 btrfs_add_inode_defrag(NULL, inode);
540 * we create compressed extents in two phases. The first
541 * phase compresses a range of pages that have already been
542 * locked (both pages and state bits are locked).
544 * This is done inside an ordered work queue, and the compression
545 * is spread across many cpus. The actual IO submission is step
546 * two, and the ordered work queue takes care of making sure that
547 * happens in the same order things were put onto the queue by
548 * writepages and friends.
550 * If this code finds it can't get good compression, it puts an
551 * entry onto the work queue to write the uncompressed bytes. This
552 * makes sure that both compressed inodes and uncompressed inodes
553 * are written in the same order that the flusher thread sent them
556 static noinline int compress_file_range(struct async_chunk *async_chunk)
558 struct inode *inode = async_chunk->inode;
559 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
560 u64 blocksize = fs_info->sectorsize;
561 u64 start = async_chunk->start;
562 u64 end = async_chunk->end;
566 struct page **pages = NULL;
567 unsigned long nr_pages;
568 unsigned long total_compressed = 0;
569 unsigned long total_in = 0;
572 int compress_type = fs_info->compress_type;
573 int compressed_extents = 0;
576 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
580 * We need to save i_size before now because it could change in between
581 * us evaluating the size and assigning it. This is because we lock and
582 * unlock the page in truncate and fallocate, and then modify the i_size
585 * The barriers are to emulate READ_ONCE, remove that once i_size_read
589 i_size = i_size_read(inode);
591 actual_end = min_t(u64, i_size, end + 1);
594 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
595 BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
596 nr_pages = min_t(unsigned long, nr_pages,
597 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
600 * we don't want to send crud past the end of i_size through
601 * compression, that's just a waste of CPU time. So, if the
602 * end of the file is before the start of our current
603 * requested range of bytes, we bail out to the uncompressed
604 * cleanup code that can deal with all of this.
606 * It isn't really the fastest way to fix things, but this is a
607 * very uncommon corner.
609 if (actual_end <= start)
610 goto cleanup_and_bail_uncompressed;
612 total_compressed = actual_end - start;
615 * skip compression for a small file range(<=blocksize) that
616 * isn't an inline extent, since it doesn't save disk space at all.
618 if (total_compressed <= blocksize &&
619 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
620 goto cleanup_and_bail_uncompressed;
622 total_compressed = min_t(unsigned long, total_compressed,
623 BTRFS_MAX_UNCOMPRESSED);
628 * we do compression for mount -o compress and when the
629 * inode has not been flagged as nocompress. This flag can
630 * change at any time if we discover bad compression ratios.
632 if (nr_pages > 1 && inode_need_compress(BTRFS_I(inode), start, end)) {
634 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
636 /* just bail out to the uncompressed code */
641 if (BTRFS_I(inode)->defrag_compress)
642 compress_type = BTRFS_I(inode)->defrag_compress;
643 else if (BTRFS_I(inode)->prop_compress)
644 compress_type = BTRFS_I(inode)->prop_compress;
647 * we need to call clear_page_dirty_for_io on each
648 * page in the range. Otherwise applications with the file
649 * mmap'd can wander in and change the page contents while
650 * we are compressing them.
652 * If the compression fails for any reason, we set the pages
653 * dirty again later on.
655 * Note that the remaining part is redirtied, the start pointer
656 * has moved, the end is the original one.
659 extent_range_clear_dirty_for_io(inode, start, end);
663 /* Compression level is applied here and only here */
664 ret = btrfs_compress_pages(
665 compress_type | (fs_info->compress_level << 4),
666 inode->i_mapping, start,
673 unsigned long offset = offset_in_page(total_compressed);
674 struct page *page = pages[nr_pages - 1];
676 /* zero the tail end of the last page, we might be
677 * sending it down to disk
680 memzero_page(page, offset, PAGE_SIZE - offset);
686 /* lets try to make an inline extent */
687 if (ret || total_in < actual_end) {
688 /* we didn't compress the entire range, try
689 * to make an uncompressed inline extent.
691 ret = cow_file_range_inline(BTRFS_I(inode), start, end,
692 0, BTRFS_COMPRESS_NONE,
695 /* try making a compressed inline extent */
696 ret = cow_file_range_inline(BTRFS_I(inode), start, end,
698 compress_type, pages);
701 unsigned long clear_flags = EXTENT_DELALLOC |
702 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
703 EXTENT_DO_ACCOUNTING;
704 unsigned long page_error_op;
706 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
709 * inline extent creation worked or returned error,
710 * we don't need to create any more async work items.
711 * Unlock and free up our temp pages.
713 * We use DO_ACCOUNTING here because we need the
714 * delalloc_release_metadata to be done _after_ we drop
715 * our outstanding extent for clearing delalloc for this
718 extent_clear_unlock_delalloc(BTRFS_I(inode), start, end,
722 PAGE_START_WRITEBACK |
727 * Ensure we only free the compressed pages if we have
728 * them allocated, as we can still reach here with
729 * inode_need_compress() == false.
732 for (i = 0; i < nr_pages; i++) {
733 WARN_ON(pages[i]->mapping);
744 * we aren't doing an inline extent round the compressed size
745 * up to a block size boundary so the allocator does sane
748 total_compressed = ALIGN(total_compressed, blocksize);
751 * one last check to make sure the compression is really a
752 * win, compare the page count read with the blocks on disk,
753 * compression must free at least one sector size
755 total_in = ALIGN(total_in, PAGE_SIZE);
756 if (total_compressed + blocksize <= total_in) {
757 compressed_extents++;
760 * The async work queues will take care of doing actual
761 * allocation on disk for these compressed pages, and
762 * will submit them to the elevator.
764 add_async_extent(async_chunk, start, total_in,
765 total_compressed, pages, nr_pages,
768 if (start + total_in < end) {
774 return compressed_extents;
779 * the compression code ran but failed to make things smaller,
780 * free any pages it allocated and our page pointer array
782 for (i = 0; i < nr_pages; i++) {
783 WARN_ON(pages[i]->mapping);
788 total_compressed = 0;
791 /* flag the file so we don't compress in the future */
792 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
793 !(BTRFS_I(inode)->prop_compress)) {
794 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
797 cleanup_and_bail_uncompressed:
799 * No compression, but we still need to write the pages in the file
800 * we've been given so far. redirty the locked page if it corresponds
801 * to our extent and set things up for the async work queue to run
802 * cow_file_range to do the normal delalloc dance.
804 if (async_chunk->locked_page &&
805 (page_offset(async_chunk->locked_page) >= start &&
806 page_offset(async_chunk->locked_page)) <= end) {
807 __set_page_dirty_nobuffers(async_chunk->locked_page);
808 /* unlocked later on in the async handlers */
812 extent_range_redirty_for_io(inode, start, end);
813 add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
814 BTRFS_COMPRESS_NONE);
815 compressed_extents++;
817 return compressed_extents;
820 static void free_async_extent_pages(struct async_extent *async_extent)
824 if (!async_extent->pages)
827 for (i = 0; i < async_extent->nr_pages; i++) {
828 WARN_ON(async_extent->pages[i]->mapping);
829 put_page(async_extent->pages[i]);
831 kfree(async_extent->pages);
832 async_extent->nr_pages = 0;
833 async_extent->pages = NULL;
837 * phase two of compressed writeback. This is the ordered portion
838 * of the code, which only gets called in the order the work was
839 * queued. We walk all the async extents created by compress_file_range
840 * and send them down to the disk.
842 static noinline void submit_compressed_extents(struct async_chunk *async_chunk)
844 struct btrfs_inode *inode = BTRFS_I(async_chunk->inode);
845 struct btrfs_fs_info *fs_info = inode->root->fs_info;
846 struct async_extent *async_extent;
848 struct btrfs_key ins;
849 struct extent_map *em;
850 struct btrfs_root *root = inode->root;
851 struct extent_io_tree *io_tree = &inode->io_tree;
855 while (!list_empty(&async_chunk->extents)) {
856 async_extent = list_entry(async_chunk->extents.next,
857 struct async_extent, list);
858 list_del(&async_extent->list);
861 lock_extent(io_tree, async_extent->start,
862 async_extent->start + async_extent->ram_size - 1);
863 /* did the compression code fall back to uncompressed IO? */
864 if (!async_extent->pages) {
865 int page_started = 0;
866 unsigned long nr_written = 0;
868 /* allocate blocks */
869 ret = cow_file_range(inode, async_chunk->locked_page,
871 async_extent->start +
872 async_extent->ram_size - 1,
873 &page_started, &nr_written, 0);
878 * if page_started, cow_file_range inserted an
879 * inline extent and took care of all the unlocking
880 * and IO for us. Otherwise, we need to submit
881 * all those pages down to the drive.
883 if (!page_started && !ret)
884 extent_write_locked_range(&inode->vfs_inode,
886 async_extent->start +
887 async_extent->ram_size - 1,
889 else if (ret && async_chunk->locked_page)
890 unlock_page(async_chunk->locked_page);
896 ret = btrfs_reserve_extent(root, async_extent->ram_size,
897 async_extent->compressed_size,
898 async_extent->compressed_size,
899 0, alloc_hint, &ins, 1, 1);
901 free_async_extent_pages(async_extent);
903 if (ret == -ENOSPC) {
904 unlock_extent(io_tree, async_extent->start,
905 async_extent->start +
906 async_extent->ram_size - 1);
909 * we need to redirty the pages if we decide to
910 * fallback to uncompressed IO, otherwise we
911 * will not submit these pages down to lower
914 extent_range_redirty_for_io(&inode->vfs_inode,
916 async_extent->start +
917 async_extent->ram_size - 1);
924 * here we're doing allocation and writeback of the
927 em = create_io_em(inode, async_extent->start,
928 async_extent->ram_size, /* len */
929 async_extent->start, /* orig_start */
930 ins.objectid, /* block_start */
931 ins.offset, /* block_len */
932 ins.offset, /* orig_block_len */
933 async_extent->ram_size, /* ram_bytes */
934 async_extent->compress_type,
935 BTRFS_ORDERED_COMPRESSED);
937 /* ret value is not necessary due to void function */
938 goto out_free_reserve;
941 ret = btrfs_add_ordered_extent_compress(inode,
944 async_extent->ram_size,
946 async_extent->compress_type);
948 btrfs_drop_extent_cache(inode, async_extent->start,
949 async_extent->start +
950 async_extent->ram_size - 1, 0);
951 goto out_free_reserve;
953 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
956 * clear dirty, set writeback and unlock the pages.
958 extent_clear_unlock_delalloc(inode, async_extent->start,
959 async_extent->start +
960 async_extent->ram_size - 1,
961 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
962 PAGE_UNLOCK | PAGE_START_WRITEBACK);
963 if (btrfs_submit_compressed_write(inode, async_extent->start,
964 async_extent->ram_size,
966 ins.offset, async_extent->pages,
967 async_extent->nr_pages,
968 async_chunk->write_flags,
969 async_chunk->blkcg_css)) {
970 struct page *p = async_extent->pages[0];
971 const u64 start = async_extent->start;
972 const u64 end = start + async_extent->ram_size - 1;
974 p->mapping = inode->vfs_inode.i_mapping;
975 btrfs_writepage_endio_finish_ordered(inode, p, start,
979 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
982 free_async_extent_pages(async_extent);
984 alloc_hint = ins.objectid + ins.offset;
990 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
991 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
993 extent_clear_unlock_delalloc(inode, async_extent->start,
994 async_extent->start +
995 async_extent->ram_size - 1,
996 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
997 EXTENT_DELALLOC_NEW |
998 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
999 PAGE_UNLOCK | PAGE_START_WRITEBACK |
1000 PAGE_END_WRITEBACK | PAGE_SET_ERROR);
1001 free_async_extent_pages(async_extent);
1002 kfree(async_extent);
1006 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
1009 struct extent_map_tree *em_tree = &inode->extent_tree;
1010 struct extent_map *em;
1013 read_lock(&em_tree->lock);
1014 em = search_extent_mapping(em_tree, start, num_bytes);
1017 * if block start isn't an actual block number then find the
1018 * first block in this inode and use that as a hint. If that
1019 * block is also bogus then just don't worry about it.
1021 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1022 free_extent_map(em);
1023 em = search_extent_mapping(em_tree, 0, 0);
1024 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
1025 alloc_hint = em->block_start;
1027 free_extent_map(em);
1029 alloc_hint = em->block_start;
1030 free_extent_map(em);
1033 read_unlock(&em_tree->lock);
1039 * when extent_io.c finds a delayed allocation range in the file,
1040 * the call backs end up in this code. The basic idea is to
1041 * allocate extents on disk for the range, and create ordered data structs
1042 * in ram to track those extents.
1044 * locked_page is the page that writepage had locked already. We use
1045 * it to make sure we don't do extra locks or unlocks.
1047 * *page_started is set to one if we unlock locked_page and do everything
1048 * required to start IO on it. It may be clean and already done with
1049 * IO when we return.
1051 static noinline int cow_file_range(struct btrfs_inode *inode,
1052 struct page *locked_page,
1053 u64 start, u64 end, int *page_started,
1054 unsigned long *nr_written, int unlock)
1056 struct btrfs_root *root = inode->root;
1057 struct btrfs_fs_info *fs_info = root->fs_info;
1060 unsigned long ram_size;
1061 u64 cur_alloc_size = 0;
1063 u64 blocksize = fs_info->sectorsize;
1064 struct btrfs_key ins;
1065 struct extent_map *em;
1066 unsigned clear_bits;
1067 unsigned long page_ops;
1068 bool extent_reserved = false;
1071 if (btrfs_is_free_space_inode(inode)) {
1077 num_bytes = ALIGN(end - start + 1, blocksize);
1078 num_bytes = max(blocksize, num_bytes);
1079 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1081 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1084 /* lets try to make an inline extent */
1085 ret = cow_file_range_inline(inode, start, end, 0,
1086 BTRFS_COMPRESS_NONE, NULL);
1089 * We use DO_ACCOUNTING here because we need the
1090 * delalloc_release_metadata to be run _after_ we drop
1091 * our outstanding extent for clearing delalloc for this
1094 extent_clear_unlock_delalloc(inode, start, end,
1096 EXTENT_LOCKED | EXTENT_DELALLOC |
1097 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1098 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1099 PAGE_START_WRITEBACK | PAGE_END_WRITEBACK);
1100 *nr_written = *nr_written +
1101 (end - start + PAGE_SIZE) / PAGE_SIZE;
1104 * locked_page is locked by the caller of
1105 * writepage_delalloc(), not locked by
1106 * __process_pages_contig().
1108 * We can't let __process_pages_contig() to unlock it,
1109 * as it doesn't have any subpage::writers recorded.
1111 * Here we manually unlock the page, since the caller
1112 * can't use page_started to determine if it's an
1113 * inline extent or a compressed extent.
1115 unlock_page(locked_page);
1117 } else if (ret < 0) {
1122 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1123 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
1126 * Relocation relies on the relocated extents to have exactly the same
1127 * size as the original extents. Normally writeback for relocation data
1128 * extents follows a NOCOW path because relocation preallocates the
1129 * extents. However, due to an operation such as scrub turning a block
1130 * group to RO mode, it may fallback to COW mode, so we must make sure
1131 * an extent allocated during COW has exactly the requested size and can
1132 * not be split into smaller extents, otherwise relocation breaks and
1133 * fails during the stage where it updates the bytenr of file extent
1136 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1137 min_alloc_size = num_bytes;
1139 min_alloc_size = fs_info->sectorsize;
1141 while (num_bytes > 0) {
1142 cur_alloc_size = num_bytes;
1143 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1144 min_alloc_size, 0, alloc_hint,
1148 cur_alloc_size = ins.offset;
1149 extent_reserved = true;
1151 ram_size = ins.offset;
1152 em = create_io_em(inode, start, ins.offset, /* len */
1153 start, /* orig_start */
1154 ins.objectid, /* block_start */
1155 ins.offset, /* block_len */
1156 ins.offset, /* orig_block_len */
1157 ram_size, /* ram_bytes */
1158 BTRFS_COMPRESS_NONE, /* compress_type */
1159 BTRFS_ORDERED_REGULAR /* type */);
1164 free_extent_map(em);
1166 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1167 ram_size, cur_alloc_size,
1168 BTRFS_ORDERED_REGULAR);
1170 goto out_drop_extent_cache;
1172 if (root->root_key.objectid ==
1173 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1174 ret = btrfs_reloc_clone_csums(inode, start,
1177 * Only drop cache here, and process as normal.
1179 * We must not allow extent_clear_unlock_delalloc()
1180 * at out_unlock label to free meta of this ordered
1181 * extent, as its meta should be freed by
1182 * btrfs_finish_ordered_io().
1184 * So we must continue until @start is increased to
1185 * skip current ordered extent.
1188 btrfs_drop_extent_cache(inode, start,
1189 start + ram_size - 1, 0);
1192 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1195 * We're not doing compressed IO, don't unlock the first page
1196 * (which the caller expects to stay locked), don't clear any
1197 * dirty bits and don't set any writeback bits
1199 * Do set the Ordered (Private2) bit so we know this page was
1200 * properly setup for writepage.
1202 page_ops = unlock ? PAGE_UNLOCK : 0;
1203 page_ops |= PAGE_SET_ORDERED;
1205 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1207 EXTENT_LOCKED | EXTENT_DELALLOC,
1209 if (num_bytes < cur_alloc_size)
1212 num_bytes -= cur_alloc_size;
1213 alloc_hint = ins.objectid + ins.offset;
1214 start += cur_alloc_size;
1215 extent_reserved = false;
1218 * btrfs_reloc_clone_csums() error, since start is increased
1219 * extent_clear_unlock_delalloc() at out_unlock label won't
1220 * free metadata of current ordered extent, we're OK to exit.
1228 out_drop_extent_cache:
1229 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1231 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1232 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1234 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1235 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1236 page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK;
1238 * If we reserved an extent for our delalloc range (or a subrange) and
1239 * failed to create the respective ordered extent, then it means that
1240 * when we reserved the extent we decremented the extent's size from
1241 * the data space_info's bytes_may_use counter and incremented the
1242 * space_info's bytes_reserved counter by the same amount. We must make
1243 * sure extent_clear_unlock_delalloc() does not try to decrement again
1244 * the data space_info's bytes_may_use counter, therefore we do not pass
1245 * it the flag EXTENT_CLEAR_DATA_RESV.
1247 if (extent_reserved) {
1248 extent_clear_unlock_delalloc(inode, start,
1249 start + cur_alloc_size - 1,
1253 start += cur_alloc_size;
1257 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1258 clear_bits | EXTENT_CLEAR_DATA_RESV,
1264 * work queue call back to started compression on a file and pages
1266 static noinline void async_cow_start(struct btrfs_work *work)
1268 struct async_chunk *async_chunk;
1269 int compressed_extents;
1271 async_chunk = container_of(work, struct async_chunk, work);
1273 compressed_extents = compress_file_range(async_chunk);
1274 if (compressed_extents == 0) {
1275 btrfs_add_delayed_iput(async_chunk->inode);
1276 async_chunk->inode = NULL;
1281 * work queue call back to submit previously compressed pages
1283 static noinline void async_cow_submit(struct btrfs_work *work)
1285 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1287 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1288 unsigned long nr_pages;
1290 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1293 /* atomic_sub_return implies a barrier */
1294 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1296 cond_wake_up_nomb(&fs_info->async_submit_wait);
1299 * ->inode could be NULL if async_chunk_start has failed to compress,
1300 * in which case we don't have anything to submit, yet we need to
1301 * always adjust ->async_delalloc_pages as its paired with the init
1302 * happening in cow_file_range_async
1304 if (async_chunk->inode)
1305 submit_compressed_extents(async_chunk);
1308 static noinline void async_cow_free(struct btrfs_work *work)
1310 struct async_chunk *async_chunk;
1312 async_chunk = container_of(work, struct async_chunk, work);
1313 if (async_chunk->inode)
1314 btrfs_add_delayed_iput(async_chunk->inode);
1315 if (async_chunk->blkcg_css)
1316 css_put(async_chunk->blkcg_css);
1318 * Since the pointer to 'pending' is at the beginning of the array of
1319 * async_chunk's, freeing it ensures the whole array has been freed.
1321 if (atomic_dec_and_test(async_chunk->pending))
1322 kvfree(async_chunk->pending);
1325 static int cow_file_range_async(struct btrfs_inode *inode,
1326 struct writeback_control *wbc,
1327 struct page *locked_page,
1328 u64 start, u64 end, int *page_started,
1329 unsigned long *nr_written)
1331 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1332 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1333 struct async_cow *ctx;
1334 struct async_chunk *async_chunk;
1335 unsigned long nr_pages;
1337 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1339 bool should_compress;
1341 const unsigned int write_flags = wbc_to_write_flags(wbc);
1343 unlock_extent(&inode->io_tree, start, end);
1345 if (inode->flags & BTRFS_INODE_NOCOMPRESS &&
1346 !btrfs_test_opt(fs_info, FORCE_COMPRESS)) {
1348 should_compress = false;
1350 should_compress = true;
1353 nofs_flag = memalloc_nofs_save();
1354 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1355 memalloc_nofs_restore(nofs_flag);
1358 unsigned clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC |
1359 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1360 EXTENT_DO_ACCOUNTING;
1361 unsigned long page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK |
1362 PAGE_END_WRITEBACK | PAGE_SET_ERROR;
1364 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1365 clear_bits, page_ops);
1369 async_chunk = ctx->chunks;
1370 atomic_set(&ctx->num_chunks, num_chunks);
1372 for (i = 0; i < num_chunks; i++) {
1373 if (should_compress)
1374 cur_end = min(end, start + SZ_512K - 1);
1379 * igrab is called higher up in the call chain, take only the
1380 * lightweight reference for the callback lifetime
1382 ihold(&inode->vfs_inode);
1383 async_chunk[i].pending = &ctx->num_chunks;
1384 async_chunk[i].inode = &inode->vfs_inode;
1385 async_chunk[i].start = start;
1386 async_chunk[i].end = cur_end;
1387 async_chunk[i].write_flags = write_flags;
1388 INIT_LIST_HEAD(&async_chunk[i].extents);
1391 * The locked_page comes all the way from writepage and its
1392 * the original page we were actually given. As we spread
1393 * this large delalloc region across multiple async_chunk
1394 * structs, only the first struct needs a pointer to locked_page
1396 * This way we don't need racey decisions about who is supposed
1401 * Depending on the compressibility, the pages might or
1402 * might not go through async. We want all of them to
1403 * be accounted against wbc once. Let's do it here
1404 * before the paths diverge. wbc accounting is used
1405 * only for foreign writeback detection and doesn't
1406 * need full accuracy. Just account the whole thing
1407 * against the first page.
1409 wbc_account_cgroup_owner(wbc, locked_page,
1411 async_chunk[i].locked_page = locked_page;
1414 async_chunk[i].locked_page = NULL;
1417 if (blkcg_css != blkcg_root_css) {
1419 async_chunk[i].blkcg_css = blkcg_css;
1421 async_chunk[i].blkcg_css = NULL;
1424 btrfs_init_work(&async_chunk[i].work, async_cow_start,
1425 async_cow_submit, async_cow_free);
1427 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1428 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1430 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1432 *nr_written += nr_pages;
1433 start = cur_end + 1;
1439 static noinline int run_delalloc_zoned(struct btrfs_inode *inode,
1440 struct page *locked_page, u64 start,
1441 u64 end, int *page_started,
1442 unsigned long *nr_written)
1446 ret = cow_file_range(inode, locked_page, start, end, page_started,
1454 __set_page_dirty_nobuffers(locked_page);
1455 account_page_redirty(locked_page);
1456 extent_write_locked_range(&inode->vfs_inode, start, end, WB_SYNC_ALL);
1462 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1463 u64 bytenr, u64 num_bytes)
1466 struct btrfs_ordered_sum *sums;
1469 ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1470 bytenr + num_bytes - 1, &list, 0);
1471 if (ret == 0 && list_empty(&list))
1474 while (!list_empty(&list)) {
1475 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1476 list_del(&sums->list);
1484 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1485 const u64 start, const u64 end,
1486 int *page_started, unsigned long *nr_written)
1488 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1489 const bool is_reloc_ino = (inode->root->root_key.objectid ==
1490 BTRFS_DATA_RELOC_TREE_OBJECTID);
1491 const u64 range_bytes = end + 1 - start;
1492 struct extent_io_tree *io_tree = &inode->io_tree;
1493 u64 range_start = start;
1497 * If EXTENT_NORESERVE is set it means that when the buffered write was
1498 * made we had not enough available data space and therefore we did not
1499 * reserve data space for it, since we though we could do NOCOW for the
1500 * respective file range (either there is prealloc extent or the inode
1501 * has the NOCOW bit set).
1503 * However when we need to fallback to COW mode (because for example the
1504 * block group for the corresponding extent was turned to RO mode by a
1505 * scrub or relocation) we need to do the following:
1507 * 1) We increment the bytes_may_use counter of the data space info.
1508 * If COW succeeds, it allocates a new data extent and after doing
1509 * that it decrements the space info's bytes_may_use counter and
1510 * increments its bytes_reserved counter by the same amount (we do
1511 * this at btrfs_add_reserved_bytes()). So we need to increment the
1512 * bytes_may_use counter to compensate (when space is reserved at
1513 * buffered write time, the bytes_may_use counter is incremented);
1515 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1516 * that if the COW path fails for any reason, it decrements (through
1517 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1518 * data space info, which we incremented in the step above.
1520 * If we need to fallback to cow and the inode corresponds to a free
1521 * space cache inode or an inode of the data relocation tree, we must
1522 * also increment bytes_may_use of the data space_info for the same
1523 * reason. Space caches and relocated data extents always get a prealloc
1524 * extent for them, however scrub or balance may have set the block
1525 * group that contains that extent to RO mode and therefore force COW
1526 * when starting writeback.
1528 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1529 EXTENT_NORESERVE, 0);
1530 if (count > 0 || is_space_ino || is_reloc_ino) {
1532 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1533 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1535 if (is_space_ino || is_reloc_ino)
1536 bytes = range_bytes;
1538 spin_lock(&sinfo->lock);
1539 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1540 spin_unlock(&sinfo->lock);
1543 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1547 return cow_file_range(inode, locked_page, start, end, page_started,
1552 * when nowcow writeback call back. This checks for snapshots or COW copies
1553 * of the extents that exist in the file, and COWs the file as required.
1555 * If no cow copies or snapshots exist, we write directly to the existing
1558 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
1559 struct page *locked_page,
1560 const u64 start, const u64 end,
1562 unsigned long *nr_written)
1564 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1565 struct btrfs_root *root = inode->root;
1566 struct btrfs_path *path;
1567 u64 cow_start = (u64)-1;
1568 u64 cur_offset = start;
1570 bool check_prev = true;
1571 const bool freespace_inode = btrfs_is_free_space_inode(inode);
1572 u64 ino = btrfs_ino(inode);
1574 u64 disk_bytenr = 0;
1575 const bool force = inode->flags & BTRFS_INODE_NODATACOW;
1577 path = btrfs_alloc_path();
1579 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1580 EXTENT_LOCKED | EXTENT_DELALLOC |
1581 EXTENT_DO_ACCOUNTING |
1582 EXTENT_DEFRAG, PAGE_UNLOCK |
1583 PAGE_START_WRITEBACK |
1584 PAGE_END_WRITEBACK);
1589 struct btrfs_key found_key;
1590 struct btrfs_file_extent_item *fi;
1591 struct extent_buffer *leaf;
1601 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1607 * If there is no extent for our range when doing the initial
1608 * search, then go back to the previous slot as it will be the
1609 * one containing the search offset
1611 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1612 leaf = path->nodes[0];
1613 btrfs_item_key_to_cpu(leaf, &found_key,
1614 path->slots[0] - 1);
1615 if (found_key.objectid == ino &&
1616 found_key.type == BTRFS_EXTENT_DATA_KEY)
1621 /* Go to next leaf if we have exhausted the current one */
1622 leaf = path->nodes[0];
1623 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1624 ret = btrfs_next_leaf(root, path);
1626 if (cow_start != (u64)-1)
1627 cur_offset = cow_start;
1632 leaf = path->nodes[0];
1635 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1637 /* Didn't find anything for our INO */
1638 if (found_key.objectid > ino)
1641 * Keep searching until we find an EXTENT_ITEM or there are no
1642 * more extents for this inode
1644 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1645 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1650 /* Found key is not EXTENT_DATA_KEY or starts after req range */
1651 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1652 found_key.offset > end)
1656 * If the found extent starts after requested offset, then
1657 * adjust extent_end to be right before this extent begins
1659 if (found_key.offset > cur_offset) {
1660 extent_end = found_key.offset;
1666 * Found extent which begins before our range and potentially
1669 fi = btrfs_item_ptr(leaf, path->slots[0],
1670 struct btrfs_file_extent_item);
1671 extent_type = btrfs_file_extent_type(leaf, fi);
1673 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1674 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1675 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1676 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1677 extent_offset = btrfs_file_extent_offset(leaf, fi);
1678 extent_end = found_key.offset +
1679 btrfs_file_extent_num_bytes(leaf, fi);
1681 btrfs_file_extent_disk_num_bytes(leaf, fi);
1683 * If the extent we got ends before our current offset,
1684 * skip to the next extent.
1686 if (extent_end <= cur_offset) {
1691 if (disk_bytenr == 0)
1693 /* Skip compressed/encrypted/encoded extents */
1694 if (btrfs_file_extent_compression(leaf, fi) ||
1695 btrfs_file_extent_encryption(leaf, fi) ||
1696 btrfs_file_extent_other_encoding(leaf, fi))
1699 * If extent is created before the last volume's snapshot
1700 * this implies the extent is shared, hence we can't do
1701 * nocow. This is the same check as in
1702 * btrfs_cross_ref_exist but without calling
1703 * btrfs_search_slot.
1705 if (!freespace_inode &&
1706 btrfs_file_extent_generation(leaf, fi) <=
1707 btrfs_root_last_snapshot(&root->root_item))
1709 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1713 * The following checks can be expensive, as they need to
1714 * take other locks and do btree or rbtree searches, so
1715 * release the path to avoid blocking other tasks for too
1718 btrfs_release_path(path);
1720 ret = btrfs_cross_ref_exist(root, ino,
1722 extent_offset, disk_bytenr, false);
1725 * ret could be -EIO if the above fails to read
1729 if (cow_start != (u64)-1)
1730 cur_offset = cow_start;
1734 WARN_ON_ONCE(freespace_inode);
1737 disk_bytenr += extent_offset;
1738 disk_bytenr += cur_offset - found_key.offset;
1739 num_bytes = min(end + 1, extent_end) - cur_offset;
1741 * If there are pending snapshots for this root, we
1742 * fall into common COW way
1744 if (!freespace_inode && atomic_read(&root->snapshot_force_cow))
1747 * force cow if csum exists in the range.
1748 * this ensure that csum for a given extent are
1749 * either valid or do not exist.
1751 ret = csum_exist_in_range(fs_info, disk_bytenr,
1755 * ret could be -EIO if the above fails to read
1759 if (cow_start != (u64)-1)
1760 cur_offset = cow_start;
1763 WARN_ON_ONCE(freespace_inode);
1766 /* If the extent's block group is RO, we must COW */
1767 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr))
1770 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1771 extent_end = found_key.offset + ram_bytes;
1772 extent_end = ALIGN(extent_end, fs_info->sectorsize);
1773 /* Skip extents outside of our requested range */
1774 if (extent_end <= start) {
1779 /* If this triggers then we have a memory corruption */
1784 * If nocow is false then record the beginning of the range
1785 * that needs to be COWed
1788 if (cow_start == (u64)-1)
1789 cow_start = cur_offset;
1790 cur_offset = extent_end;
1791 if (cur_offset > end)
1793 if (!path->nodes[0])
1800 * COW range from cow_start to found_key.offset - 1. As the key
1801 * will contain the beginning of the first extent that can be
1802 * NOCOW, following one which needs to be COW'ed
1804 if (cow_start != (u64)-1) {
1805 ret = fallback_to_cow(inode, locked_page,
1806 cow_start, found_key.offset - 1,
1807 page_started, nr_written);
1810 cow_start = (u64)-1;
1813 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1814 u64 orig_start = found_key.offset - extent_offset;
1815 struct extent_map *em;
1817 em = create_io_em(inode, cur_offset, num_bytes,
1819 disk_bytenr, /* block_start */
1820 num_bytes, /* block_len */
1821 disk_num_bytes, /* orig_block_len */
1822 ram_bytes, BTRFS_COMPRESS_NONE,
1823 BTRFS_ORDERED_PREALLOC);
1828 free_extent_map(em);
1829 ret = btrfs_add_ordered_extent(inode, cur_offset,
1830 disk_bytenr, num_bytes,
1832 BTRFS_ORDERED_PREALLOC);
1834 btrfs_drop_extent_cache(inode, cur_offset,
1835 cur_offset + num_bytes - 1,
1840 ret = btrfs_add_ordered_extent(inode, cur_offset,
1841 disk_bytenr, num_bytes,
1843 BTRFS_ORDERED_NOCOW);
1849 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1852 if (root->root_key.objectid ==
1853 BTRFS_DATA_RELOC_TREE_OBJECTID)
1855 * Error handled later, as we must prevent
1856 * extent_clear_unlock_delalloc() in error handler
1857 * from freeing metadata of created ordered extent.
1859 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1862 extent_clear_unlock_delalloc(inode, cur_offset,
1863 cur_offset + num_bytes - 1,
1864 locked_page, EXTENT_LOCKED |
1866 EXTENT_CLEAR_DATA_RESV,
1867 PAGE_UNLOCK | PAGE_SET_ORDERED);
1869 cur_offset = extent_end;
1872 * btrfs_reloc_clone_csums() error, now we're OK to call error
1873 * handler, as metadata for created ordered extent will only
1874 * be freed by btrfs_finish_ordered_io().
1878 if (cur_offset > end)
1881 btrfs_release_path(path);
1883 if (cur_offset <= end && cow_start == (u64)-1)
1884 cow_start = cur_offset;
1886 if (cow_start != (u64)-1) {
1888 ret = fallback_to_cow(inode, locked_page, cow_start, end,
1889 page_started, nr_written);
1896 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1898 if (ret && cur_offset < end)
1899 extent_clear_unlock_delalloc(inode, cur_offset, end,
1900 locked_page, EXTENT_LOCKED |
1901 EXTENT_DELALLOC | EXTENT_DEFRAG |
1902 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1903 PAGE_START_WRITEBACK |
1904 PAGE_END_WRITEBACK);
1905 btrfs_free_path(path);
1909 static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end)
1911 if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) {
1912 if (inode->defrag_bytes &&
1913 test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG,
1922 * Function to process delayed allocation (create CoW) for ranges which are
1923 * being touched for the first time.
1925 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
1926 u64 start, u64 end, int *page_started, unsigned long *nr_written,
1927 struct writeback_control *wbc)
1930 const bool zoned = btrfs_is_zoned(inode->root->fs_info);
1932 if (should_nocow(inode, start, end)) {
1934 ret = run_delalloc_nocow(inode, locked_page, start, end,
1935 page_started, nr_written);
1936 } else if (!inode_can_compress(inode) ||
1937 !inode_need_compress(inode, start, end)) {
1939 ret = run_delalloc_zoned(inode, locked_page, start, end,
1940 page_started, nr_written);
1942 ret = cow_file_range(inode, locked_page, start, end,
1943 page_started, nr_written, 1);
1945 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
1946 ret = cow_file_range_async(inode, wbc, locked_page, start, end,
1947 page_started, nr_written);
1950 btrfs_cleanup_ordered_extents(inode, locked_page, start,
1955 void btrfs_split_delalloc_extent(struct inode *inode,
1956 struct extent_state *orig, u64 split)
1960 /* not delalloc, ignore it */
1961 if (!(orig->state & EXTENT_DELALLOC))
1964 size = orig->end - orig->start + 1;
1965 if (size > BTRFS_MAX_EXTENT_SIZE) {
1970 * See the explanation in btrfs_merge_delalloc_extent, the same
1971 * applies here, just in reverse.
1973 new_size = orig->end - split + 1;
1974 num_extents = count_max_extents(new_size);
1975 new_size = split - orig->start;
1976 num_extents += count_max_extents(new_size);
1977 if (count_max_extents(size) >= num_extents)
1981 spin_lock(&BTRFS_I(inode)->lock);
1982 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1983 spin_unlock(&BTRFS_I(inode)->lock);
1987 * Handle merged delayed allocation extents so we can keep track of new extents
1988 * that are just merged onto old extents, such as when we are doing sequential
1989 * writes, so we can properly account for the metadata space we'll need.
1991 void btrfs_merge_delalloc_extent(struct inode *inode, struct extent_state *new,
1992 struct extent_state *other)
1994 u64 new_size, old_size;
1997 /* not delalloc, ignore it */
1998 if (!(other->state & EXTENT_DELALLOC))
2001 if (new->start > other->start)
2002 new_size = new->end - other->start + 1;
2004 new_size = other->end - new->start + 1;
2006 /* we're not bigger than the max, unreserve the space and go */
2007 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
2008 spin_lock(&BTRFS_I(inode)->lock);
2009 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
2010 spin_unlock(&BTRFS_I(inode)->lock);
2015 * We have to add up either side to figure out how many extents were
2016 * accounted for before we merged into one big extent. If the number of
2017 * extents we accounted for is <= the amount we need for the new range
2018 * then we can return, otherwise drop. Think of it like this
2022 * So we've grown the extent by a MAX_SIZE extent, this would mean we
2023 * need 2 outstanding extents, on one side we have 1 and the other side
2024 * we have 1 so they are == and we can return. But in this case
2026 * [MAX_SIZE+4k][MAX_SIZE+4k]
2028 * Each range on their own accounts for 2 extents, but merged together
2029 * they are only 3 extents worth of accounting, so we need to drop in
2032 old_size = other->end - other->start + 1;
2033 num_extents = count_max_extents(old_size);
2034 old_size = new->end - new->start + 1;
2035 num_extents += count_max_extents(old_size);
2036 if (count_max_extents(new_size) >= num_extents)
2039 spin_lock(&BTRFS_I(inode)->lock);
2040 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
2041 spin_unlock(&BTRFS_I(inode)->lock);
2044 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
2045 struct inode *inode)
2047 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2049 spin_lock(&root->delalloc_lock);
2050 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
2051 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
2052 &root->delalloc_inodes);
2053 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2054 &BTRFS_I(inode)->runtime_flags);
2055 root->nr_delalloc_inodes++;
2056 if (root->nr_delalloc_inodes == 1) {
2057 spin_lock(&fs_info->delalloc_root_lock);
2058 BUG_ON(!list_empty(&root->delalloc_root));
2059 list_add_tail(&root->delalloc_root,
2060 &fs_info->delalloc_roots);
2061 spin_unlock(&fs_info->delalloc_root_lock);
2064 spin_unlock(&root->delalloc_lock);
2068 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
2069 struct btrfs_inode *inode)
2071 struct btrfs_fs_info *fs_info = root->fs_info;
2073 if (!list_empty(&inode->delalloc_inodes)) {
2074 list_del_init(&inode->delalloc_inodes);
2075 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2076 &inode->runtime_flags);
2077 root->nr_delalloc_inodes--;
2078 if (!root->nr_delalloc_inodes) {
2079 ASSERT(list_empty(&root->delalloc_inodes));
2080 spin_lock(&fs_info->delalloc_root_lock);
2081 BUG_ON(list_empty(&root->delalloc_root));
2082 list_del_init(&root->delalloc_root);
2083 spin_unlock(&fs_info->delalloc_root_lock);
2088 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
2089 struct btrfs_inode *inode)
2091 spin_lock(&root->delalloc_lock);
2092 __btrfs_del_delalloc_inode(root, inode);
2093 spin_unlock(&root->delalloc_lock);
2097 * Properly track delayed allocation bytes in the inode and to maintain the
2098 * list of inodes that have pending delalloc work to be done.
2100 void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state,
2103 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2105 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
2108 * set_bit and clear bit hooks normally require _irqsave/restore
2109 * but in this case, we are only testing for the DELALLOC
2110 * bit, which is only set or cleared with irqs on
2112 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2113 struct btrfs_root *root = BTRFS_I(inode)->root;
2114 u64 len = state->end + 1 - state->start;
2115 u32 num_extents = count_max_extents(len);
2116 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
2118 spin_lock(&BTRFS_I(inode)->lock);
2119 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
2120 spin_unlock(&BTRFS_I(inode)->lock);
2122 /* For sanity tests */
2123 if (btrfs_is_testing(fs_info))
2126 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2127 fs_info->delalloc_batch);
2128 spin_lock(&BTRFS_I(inode)->lock);
2129 BTRFS_I(inode)->delalloc_bytes += len;
2130 if (*bits & EXTENT_DEFRAG)
2131 BTRFS_I(inode)->defrag_bytes += len;
2132 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2133 &BTRFS_I(inode)->runtime_flags))
2134 btrfs_add_delalloc_inodes(root, inode);
2135 spin_unlock(&BTRFS_I(inode)->lock);
2138 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2139 (*bits & EXTENT_DELALLOC_NEW)) {
2140 spin_lock(&BTRFS_I(inode)->lock);
2141 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
2143 spin_unlock(&BTRFS_I(inode)->lock);
2148 * Once a range is no longer delalloc this function ensures that proper
2149 * accounting happens.
2151 void btrfs_clear_delalloc_extent(struct inode *vfs_inode,
2152 struct extent_state *state, unsigned *bits)
2154 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
2155 struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb);
2156 u64 len = state->end + 1 - state->start;
2157 u32 num_extents = count_max_extents(len);
2159 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
2160 spin_lock(&inode->lock);
2161 inode->defrag_bytes -= len;
2162 spin_unlock(&inode->lock);
2166 * set_bit and clear bit hooks normally require _irqsave/restore
2167 * but in this case, we are only testing for the DELALLOC
2168 * bit, which is only set or cleared with irqs on
2170 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2171 struct btrfs_root *root = inode->root;
2172 bool do_list = !btrfs_is_free_space_inode(inode);
2174 spin_lock(&inode->lock);
2175 btrfs_mod_outstanding_extents(inode, -num_extents);
2176 spin_unlock(&inode->lock);
2179 * We don't reserve metadata space for space cache inodes so we
2180 * don't need to call delalloc_release_metadata if there is an
2183 if (*bits & EXTENT_CLEAR_META_RESV &&
2184 root != fs_info->tree_root)
2185 btrfs_delalloc_release_metadata(inode, len, false);
2187 /* For sanity tests. */
2188 if (btrfs_is_testing(fs_info))
2191 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
2192 do_list && !(state->state & EXTENT_NORESERVE) &&
2193 (*bits & EXTENT_CLEAR_DATA_RESV))
2194 btrfs_free_reserved_data_space_noquota(fs_info, len);
2196 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2197 fs_info->delalloc_batch);
2198 spin_lock(&inode->lock);
2199 inode->delalloc_bytes -= len;
2200 if (do_list && inode->delalloc_bytes == 0 &&
2201 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2202 &inode->runtime_flags))
2203 btrfs_del_delalloc_inode(root, inode);
2204 spin_unlock(&inode->lock);
2207 if ((state->state & EXTENT_DELALLOC_NEW) &&
2208 (*bits & EXTENT_DELALLOC_NEW)) {
2209 spin_lock(&inode->lock);
2210 ASSERT(inode->new_delalloc_bytes >= len);
2211 inode->new_delalloc_bytes -= len;
2212 if (*bits & EXTENT_ADD_INODE_BYTES)
2213 inode_add_bytes(&inode->vfs_inode, len);
2214 spin_unlock(&inode->lock);
2219 * btrfs_bio_fits_in_stripe - Checks whether the size of the given bio will fit
2220 * in a chunk's stripe. This function ensures that bios do not span a
2223 * @page - The page we are about to add to the bio
2224 * @size - size we want to add to the bio
2225 * @bio - bio we want to ensure is smaller than a stripe
2226 * @bio_flags - flags of the bio
2228 * return 1 if page cannot be added to the bio
2229 * return 0 if page can be added to the bio
2230 * return error otherwise
2232 int btrfs_bio_fits_in_stripe(struct page *page, size_t size, struct bio *bio,
2233 unsigned long bio_flags)
2235 struct inode *inode = page->mapping->host;
2236 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2237 u64 logical = bio->bi_iter.bi_sector << 9;
2238 u32 bio_len = bio->bi_iter.bi_size;
2239 struct extent_map *em;
2241 struct btrfs_io_geometry geom;
2243 if (bio_flags & EXTENT_BIO_COMPRESSED)
2246 em = btrfs_get_chunk_map(fs_info, logical, fs_info->sectorsize);
2249 ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(bio), logical, &geom);
2253 if (geom.len < bio_len + size)
2256 free_extent_map(em);
2261 * in order to insert checksums into the metadata in large chunks,
2262 * we wait until bio submission time. All the pages in the bio are
2263 * checksummed and sums are attached onto the ordered extent record.
2265 * At IO completion time the cums attached on the ordered extent record
2266 * are inserted into the btree
2268 static blk_status_t btrfs_submit_bio_start(struct inode *inode, struct bio *bio,
2269 u64 dio_file_offset)
2271 return btrfs_csum_one_bio(BTRFS_I(inode), bio, 0, 0);
2275 * Split an extent_map at [start, start + len]
2277 * This function is intended to be used only for extract_ordered_extent().
2279 static int split_zoned_em(struct btrfs_inode *inode, u64 start, u64 len,
2282 struct extent_map_tree *em_tree = &inode->extent_tree;
2283 struct extent_map *em;
2284 struct extent_map *split_pre = NULL;
2285 struct extent_map *split_mid = NULL;
2286 struct extent_map *split_post = NULL;
2289 unsigned long flags;
2292 if (pre == 0 && post == 0)
2295 split_pre = alloc_extent_map();
2297 split_mid = alloc_extent_map();
2299 split_post = alloc_extent_map();
2300 if (!split_pre || (pre && !split_mid) || (post && !split_post)) {
2305 ASSERT(pre + post < len);
2307 lock_extent(&inode->io_tree, start, start + len - 1);
2308 write_lock(&em_tree->lock);
2309 em = lookup_extent_mapping(em_tree, start, len);
2315 ASSERT(em->len == len);
2316 ASSERT(!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags));
2317 ASSERT(em->block_start < EXTENT_MAP_LAST_BYTE);
2320 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
2321 clear_bit(EXTENT_FLAG_LOGGING, &flags);
2322 modified = !list_empty(&em->list);
2324 /* First, replace the em with a new extent_map starting from * em->start */
2325 split_pre->start = em->start;
2326 split_pre->len = (pre ? pre : em->len - post);
2327 split_pre->orig_start = split_pre->start;
2328 split_pre->block_start = em->block_start;
2329 split_pre->block_len = split_pre->len;
2330 split_pre->orig_block_len = split_pre->block_len;
2331 split_pre->ram_bytes = split_pre->len;
2332 split_pre->flags = flags;
2333 split_pre->compress_type = em->compress_type;
2334 split_pre->generation = em->generation;
2336 replace_extent_mapping(em_tree, em, split_pre, modified);
2339 * Now we only have an extent_map at:
2340 * [em->start, em->start + pre] if pre != 0
2341 * [em->start, em->start + em->len - post] if pre == 0
2345 /* Insert the middle extent_map */
2346 split_mid->start = em->start + pre;
2347 split_mid->len = em->len - pre - post;
2348 split_mid->orig_start = split_mid->start;
2349 split_mid->block_start = em->block_start + pre;
2350 split_mid->block_len = split_mid->len;
2351 split_mid->orig_block_len = split_mid->block_len;
2352 split_mid->ram_bytes = split_mid->len;
2353 split_mid->flags = flags;
2354 split_mid->compress_type = em->compress_type;
2355 split_mid->generation = em->generation;
2356 add_extent_mapping(em_tree, split_mid, modified);
2360 split_post->start = em->start + em->len - post;
2361 split_post->len = post;
2362 split_post->orig_start = split_post->start;
2363 split_post->block_start = em->block_start + em->len - post;
2364 split_post->block_len = split_post->len;
2365 split_post->orig_block_len = split_post->block_len;
2366 split_post->ram_bytes = split_post->len;
2367 split_post->flags = flags;
2368 split_post->compress_type = em->compress_type;
2369 split_post->generation = em->generation;
2370 add_extent_mapping(em_tree, split_post, modified);
2374 free_extent_map(em);
2375 /* Once for the tree */
2376 free_extent_map(em);
2379 write_unlock(&em_tree->lock);
2380 unlock_extent(&inode->io_tree, start, start + len - 1);
2382 free_extent_map(split_pre);
2383 free_extent_map(split_mid);
2384 free_extent_map(split_post);
2389 static blk_status_t extract_ordered_extent(struct btrfs_inode *inode,
2390 struct bio *bio, loff_t file_offset)
2392 struct btrfs_ordered_extent *ordered;
2393 u64 start = (u64)bio->bi_iter.bi_sector << SECTOR_SHIFT;
2395 u64 len = bio->bi_iter.bi_size;
2396 u64 end = start + len;
2401 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
2402 if (WARN_ON_ONCE(!ordered))
2403 return BLK_STS_IOERR;
2405 /* No need to split */
2406 if (ordered->disk_num_bytes == len)
2409 /* We cannot split once end_bio'd ordered extent */
2410 if (WARN_ON_ONCE(ordered->bytes_left != ordered->disk_num_bytes)) {
2415 /* We cannot split a compressed ordered extent */
2416 if (WARN_ON_ONCE(ordered->disk_num_bytes != ordered->num_bytes)) {
2421 ordered_end = ordered->disk_bytenr + ordered->disk_num_bytes;
2422 /* bio must be in one ordered extent */
2423 if (WARN_ON_ONCE(start < ordered->disk_bytenr || end > ordered_end)) {
2428 /* Checksum list should be empty */
2429 if (WARN_ON_ONCE(!list_empty(&ordered->list))) {
2434 file_len = ordered->num_bytes;
2435 pre = start - ordered->disk_bytenr;
2436 post = ordered_end - end;
2438 ret = btrfs_split_ordered_extent(ordered, pre, post);
2441 ret = split_zoned_em(inode, file_offset, file_len, pre, post);
2444 btrfs_put_ordered_extent(ordered);
2446 return errno_to_blk_status(ret);
2450 * extent_io.c submission hook. This does the right thing for csum calculation
2451 * on write, or reading the csums from the tree before a read.
2453 * Rules about async/sync submit,
2454 * a) read: sync submit
2456 * b) write without checksum: sync submit
2458 * c) write with checksum:
2459 * c-1) if bio is issued by fsync: sync submit
2460 * (sync_writers != 0)
2462 * c-2) if root is reloc root: sync submit
2463 * (only in case of buffered IO)
2465 * c-3) otherwise: async submit
2467 blk_status_t btrfs_submit_data_bio(struct inode *inode, struct bio *bio,
2468 int mirror_num, unsigned long bio_flags)
2471 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2472 struct btrfs_root *root = BTRFS_I(inode)->root;
2473 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
2474 blk_status_t ret = 0;
2476 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
2478 skip_sum = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) ||
2479 !fs_info->csum_root;
2481 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
2482 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
2484 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
2485 struct page *page = bio_first_bvec_all(bio)->bv_page;
2486 loff_t file_offset = page_offset(page);
2488 ret = extract_ordered_extent(BTRFS_I(inode), bio, file_offset);
2493 if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
2494 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
2498 if (bio_flags & EXTENT_BIO_COMPRESSED) {
2499 ret = btrfs_submit_compressed_read(inode, bio,
2505 * Lookup bio sums does extra checks around whether we
2506 * need to csum or not, which is why we ignore skip_sum
2509 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
2514 } else if (async && !skip_sum) {
2515 /* csum items have already been cloned */
2516 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2518 /* we're doing a write, do the async checksumming */
2519 ret = btrfs_wq_submit_bio(inode, bio, mirror_num, bio_flags,
2520 0, btrfs_submit_bio_start);
2522 } else if (!skip_sum) {
2523 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, 0, 0);
2529 ret = btrfs_map_bio(fs_info, bio, mirror_num);
2533 bio->bi_status = ret;
2540 * given a list of ordered sums record them in the inode. This happens
2541 * at IO completion time based on sums calculated at bio submission time.
2543 static int add_pending_csums(struct btrfs_trans_handle *trans,
2544 struct list_head *list)
2546 struct btrfs_ordered_sum *sum;
2549 list_for_each_entry(sum, list, list) {
2550 trans->adding_csums = true;
2551 ret = btrfs_csum_file_blocks(trans, trans->fs_info->csum_root, sum);
2552 trans->adding_csums = false;
2559 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2562 struct extent_state **cached_state)
2564 u64 search_start = start;
2565 const u64 end = start + len - 1;
2567 while (search_start < end) {
2568 const u64 search_len = end - search_start + 1;
2569 struct extent_map *em;
2573 em = btrfs_get_extent(inode, NULL, 0, search_start, search_len);
2577 if (em->block_start != EXTENT_MAP_HOLE)
2581 if (em->start < search_start)
2582 em_len -= search_start - em->start;
2583 if (em_len > search_len)
2584 em_len = search_len;
2586 ret = set_extent_bit(&inode->io_tree, search_start,
2587 search_start + em_len - 1,
2588 EXTENT_DELALLOC_NEW, 0, NULL, cached_state,
2591 search_start = extent_map_end(em);
2592 free_extent_map(em);
2599 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2600 unsigned int extra_bits,
2601 struct extent_state **cached_state)
2603 WARN_ON(PAGE_ALIGNED(end));
2605 if (start >= i_size_read(&inode->vfs_inode) &&
2606 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2608 * There can't be any extents following eof in this case so just
2609 * set the delalloc new bit for the range directly.
2611 extra_bits |= EXTENT_DELALLOC_NEW;
2615 ret = btrfs_find_new_delalloc_bytes(inode, start,
2622 return set_extent_delalloc(&inode->io_tree, start, end, extra_bits,
2626 /* see btrfs_writepage_start_hook for details on why this is required */
2627 struct btrfs_writepage_fixup {
2629 struct inode *inode;
2630 struct btrfs_work work;
2633 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2635 struct btrfs_writepage_fixup *fixup;
2636 struct btrfs_ordered_extent *ordered;
2637 struct extent_state *cached_state = NULL;
2638 struct extent_changeset *data_reserved = NULL;
2640 struct btrfs_inode *inode;
2644 bool free_delalloc_space = true;
2646 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2648 inode = BTRFS_I(fixup->inode);
2649 page_start = page_offset(page);
2650 page_end = page_offset(page) + PAGE_SIZE - 1;
2653 * This is similar to page_mkwrite, we need to reserve the space before
2654 * we take the page lock.
2656 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2662 * Before we queued this fixup, we took a reference on the page.
2663 * page->mapping may go NULL, but it shouldn't be moved to a different
2666 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2668 * Unfortunately this is a little tricky, either
2670 * 1) We got here and our page had already been dealt with and
2671 * we reserved our space, thus ret == 0, so we need to just
2672 * drop our space reservation and bail. This can happen the
2673 * first time we come into the fixup worker, or could happen
2674 * while waiting for the ordered extent.
2675 * 2) Our page was already dealt with, but we happened to get an
2676 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2677 * this case we obviously don't have anything to release, but
2678 * because the page was already dealt with we don't want to
2679 * mark the page with an error, so make sure we're resetting
2680 * ret to 0. This is why we have this check _before_ the ret
2681 * check, because we do not want to have a surprise ENOSPC
2682 * when the page was already properly dealt with.
2685 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2686 btrfs_delalloc_release_space(inode, data_reserved,
2687 page_start, PAGE_SIZE,
2695 * We can't mess with the page state unless it is locked, so now that
2696 * it is locked bail if we failed to make our space reservation.
2701 lock_extent_bits(&inode->io_tree, page_start, page_end, &cached_state);
2703 /* already ordered? We're done */
2704 if (PageOrdered(page))
2707 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2709 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2712 btrfs_start_ordered_extent(ordered, 1);
2713 btrfs_put_ordered_extent(ordered);
2717 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2723 * Everything went as planned, we're now the owner of a dirty page with
2724 * delayed allocation bits set and space reserved for our COW
2727 * The page was dirty when we started, nothing should have cleaned it.
2729 BUG_ON(!PageDirty(page));
2730 free_delalloc_space = false;
2732 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2733 if (free_delalloc_space)
2734 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2736 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2741 * We hit ENOSPC or other errors. Update the mapping and page
2742 * to reflect the errors and clean the page.
2744 mapping_set_error(page->mapping, ret);
2745 end_extent_writepage(page, ret, page_start, page_end);
2746 clear_page_dirty_for_io(page);
2749 ClearPageChecked(page);
2753 extent_changeset_free(data_reserved);
2755 * As a precaution, do a delayed iput in case it would be the last iput
2756 * that could need flushing space. Recursing back to fixup worker would
2759 btrfs_add_delayed_iput(&inode->vfs_inode);
2763 * There are a few paths in the higher layers of the kernel that directly
2764 * set the page dirty bit without asking the filesystem if it is a
2765 * good idea. This causes problems because we want to make sure COW
2766 * properly happens and the data=ordered rules are followed.
2768 * In our case any range that doesn't have the ORDERED bit set
2769 * hasn't been properly setup for IO. We kick off an async process
2770 * to fix it up. The async helper will wait for ordered extents, set
2771 * the delalloc bit and make it safe to write the page.
2773 int btrfs_writepage_cow_fixup(struct page *page, u64 start, u64 end)
2775 struct inode *inode = page->mapping->host;
2776 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2777 struct btrfs_writepage_fixup *fixup;
2779 /* This page has ordered extent covering it already */
2780 if (PageOrdered(page))
2784 * PageChecked is set below when we create a fixup worker for this page,
2785 * don't try to create another one if we're already PageChecked()
2787 * The extent_io writepage code will redirty the page if we send back
2790 if (PageChecked(page))
2793 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2798 * We are already holding a reference to this inode from
2799 * write_cache_pages. We need to hold it because the space reservation
2800 * takes place outside of the page lock, and we can't trust
2801 * page->mapping outside of the page lock.
2804 SetPageChecked(page);
2806 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
2808 fixup->inode = inode;
2809 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2814 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2815 struct btrfs_inode *inode, u64 file_pos,
2816 struct btrfs_file_extent_item *stack_fi,
2817 const bool update_inode_bytes,
2818 u64 qgroup_reserved)
2820 struct btrfs_root *root = inode->root;
2821 const u64 sectorsize = root->fs_info->sectorsize;
2822 struct btrfs_path *path;
2823 struct extent_buffer *leaf;
2824 struct btrfs_key ins;
2825 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
2826 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
2827 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
2828 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
2829 struct btrfs_drop_extents_args drop_args = { 0 };
2832 path = btrfs_alloc_path();
2837 * we may be replacing one extent in the tree with another.
2838 * The new extent is pinned in the extent map, and we don't want
2839 * to drop it from the cache until it is completely in the btree.
2841 * So, tell btrfs_drop_extents to leave this extent in the cache.
2842 * the caller is expected to unpin it and allow it to be merged
2845 drop_args.path = path;
2846 drop_args.start = file_pos;
2847 drop_args.end = file_pos + num_bytes;
2848 drop_args.replace_extent = true;
2849 drop_args.extent_item_size = sizeof(*stack_fi);
2850 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2854 if (!drop_args.extent_inserted) {
2855 ins.objectid = btrfs_ino(inode);
2856 ins.offset = file_pos;
2857 ins.type = BTRFS_EXTENT_DATA_KEY;
2859 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2864 leaf = path->nodes[0];
2865 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
2866 write_extent_buffer(leaf, stack_fi,
2867 btrfs_item_ptr_offset(leaf, path->slots[0]),
2868 sizeof(struct btrfs_file_extent_item));
2870 btrfs_mark_buffer_dirty(leaf);
2871 btrfs_release_path(path);
2874 * If we dropped an inline extent here, we know the range where it is
2875 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
2876 * number of bytes only for that range containing the inline extent.
2877 * The remaining of the range will be processed when clearning the
2878 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
2880 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
2881 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
2883 inline_size = drop_args.bytes_found - inline_size;
2884 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
2885 drop_args.bytes_found -= inline_size;
2886 num_bytes -= sectorsize;
2889 if (update_inode_bytes)
2890 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
2892 ins.objectid = disk_bytenr;
2893 ins.offset = disk_num_bytes;
2894 ins.type = BTRFS_EXTENT_ITEM_KEY;
2896 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
2900 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
2901 file_pos, qgroup_reserved, &ins);
2903 btrfs_free_path(path);
2908 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2911 struct btrfs_block_group *cache;
2913 cache = btrfs_lookup_block_group(fs_info, start);
2916 spin_lock(&cache->lock);
2917 cache->delalloc_bytes -= len;
2918 spin_unlock(&cache->lock);
2920 btrfs_put_block_group(cache);
2923 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
2924 struct btrfs_ordered_extent *oe)
2926 struct btrfs_file_extent_item stack_fi;
2928 bool update_inode_bytes;
2930 memset(&stack_fi, 0, sizeof(stack_fi));
2931 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
2932 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
2933 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
2934 oe->disk_num_bytes);
2935 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags))
2936 logical_len = oe->truncated_len;
2938 logical_len = oe->num_bytes;
2939 btrfs_set_stack_file_extent_num_bytes(&stack_fi, logical_len);
2940 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, logical_len);
2941 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
2942 /* Encryption and other encoding is reserved and all 0 */
2945 * For delalloc, when completing an ordered extent we update the inode's
2946 * bytes when clearing the range in the inode's io tree, so pass false
2947 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
2948 * except if the ordered extent was truncated.
2950 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
2951 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
2953 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
2954 oe->file_offset, &stack_fi,
2955 update_inode_bytes, oe->qgroup_rsv);
2959 * As ordered data IO finishes, this gets called so we can finish
2960 * an ordered extent if the range of bytes in the file it covers are
2963 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2965 struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
2966 struct btrfs_root *root = inode->root;
2967 struct btrfs_fs_info *fs_info = root->fs_info;
2968 struct btrfs_trans_handle *trans = NULL;
2969 struct extent_io_tree *io_tree = &inode->io_tree;
2970 struct extent_state *cached_state = NULL;
2972 int compress_type = 0;
2974 u64 logical_len = ordered_extent->num_bytes;
2975 bool freespace_inode;
2976 bool truncated = false;
2977 bool clear_reserved_extent = true;
2978 unsigned int clear_bits = EXTENT_DEFRAG;
2980 start = ordered_extent->file_offset;
2981 end = start + ordered_extent->num_bytes - 1;
2983 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2984 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2985 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2986 clear_bits |= EXTENT_DELALLOC_NEW;
2988 freespace_inode = btrfs_is_free_space_inode(inode);
2990 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2995 if (ordered_extent->disk)
2996 btrfs_rewrite_logical_zoned(ordered_extent);
2998 btrfs_free_io_failure_record(inode, start, end);
3000 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
3002 logical_len = ordered_extent->truncated_len;
3003 /* Truncated the entire extent, don't bother adding */
3008 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
3009 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
3011 btrfs_inode_safe_disk_i_size_write(inode, 0);
3012 if (freespace_inode)
3013 trans = btrfs_join_transaction_spacecache(root);
3015 trans = btrfs_join_transaction(root);
3016 if (IS_ERR(trans)) {
3017 ret = PTR_ERR(trans);
3021 trans->block_rsv = &inode->block_rsv;
3022 ret = btrfs_update_inode_fallback(trans, root, inode);
3023 if (ret) /* -ENOMEM or corruption */
3024 btrfs_abort_transaction(trans, ret);
3028 clear_bits |= EXTENT_LOCKED;
3029 lock_extent_bits(io_tree, start, end, &cached_state);
3031 if (freespace_inode)
3032 trans = btrfs_join_transaction_spacecache(root);
3034 trans = btrfs_join_transaction(root);
3035 if (IS_ERR(trans)) {
3036 ret = PTR_ERR(trans);
3041 trans->block_rsv = &inode->block_rsv;
3043 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3044 compress_type = ordered_extent->compress_type;
3045 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3046 BUG_ON(compress_type);
3047 ret = btrfs_mark_extent_written(trans, inode,
3048 ordered_extent->file_offset,
3049 ordered_extent->file_offset +
3052 BUG_ON(root == fs_info->tree_root);
3053 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
3055 clear_reserved_extent = false;
3056 btrfs_release_delalloc_bytes(fs_info,
3057 ordered_extent->disk_bytenr,
3058 ordered_extent->disk_num_bytes);
3061 unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset,
3062 ordered_extent->num_bytes, trans->transid);
3064 btrfs_abort_transaction(trans, ret);
3068 ret = add_pending_csums(trans, &ordered_extent->list);
3070 btrfs_abort_transaction(trans, ret);
3075 * If this is a new delalloc range, clear its new delalloc flag to
3076 * update the inode's number of bytes. This needs to be done first
3077 * before updating the inode item.
3079 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
3080 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
3081 clear_extent_bit(&inode->io_tree, start, end,
3082 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
3083 0, 0, &cached_state);
3085 btrfs_inode_safe_disk_i_size_write(inode, 0);
3086 ret = btrfs_update_inode_fallback(trans, root, inode);
3087 if (ret) { /* -ENOMEM or corruption */
3088 btrfs_abort_transaction(trans, ret);
3093 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
3094 (clear_bits & EXTENT_LOCKED) ? 1 : 0, 0,
3098 btrfs_end_transaction(trans);
3100 if (ret || truncated) {
3101 u64 unwritten_start = start;
3104 * If we failed to finish this ordered extent for any reason we
3105 * need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3106 * extent, and mark the inode with the error if it wasn't
3107 * already set. Any error during writeback would have already
3108 * set the mapping error, so we need to set it if we're the ones
3109 * marking this ordered extent as failed.
3111 if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR,
3112 &ordered_extent->flags))
3113 mapping_set_error(ordered_extent->inode->i_mapping, -EIO);
3116 unwritten_start += logical_len;
3117 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
3119 /* Drop the cache for the part of the extent we didn't write. */
3120 btrfs_drop_extent_cache(inode, unwritten_start, end, 0);
3123 * If the ordered extent had an IOERR or something else went
3124 * wrong we need to return the space for this ordered extent
3125 * back to the allocator. We only free the extent in the
3126 * truncated case if we didn't write out the extent at all.
3128 * If we made it past insert_reserved_file_extent before we
3129 * errored out then we don't need to do this as the accounting
3130 * has already been done.
3132 if ((ret || !logical_len) &&
3133 clear_reserved_extent &&
3134 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3135 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3137 * Discard the range before returning it back to the
3140 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3141 btrfs_discard_extent(fs_info,
3142 ordered_extent->disk_bytenr,
3143 ordered_extent->disk_num_bytes,
3145 btrfs_free_reserved_extent(fs_info,
3146 ordered_extent->disk_bytenr,
3147 ordered_extent->disk_num_bytes, 1);
3152 * This needs to be done to make sure anybody waiting knows we are done
3153 * updating everything for this ordered extent.
3155 btrfs_remove_ordered_extent(inode, ordered_extent);
3158 btrfs_put_ordered_extent(ordered_extent);
3159 /* once for the tree */
3160 btrfs_put_ordered_extent(ordered_extent);
3165 static void finish_ordered_fn(struct btrfs_work *work)
3167 struct btrfs_ordered_extent *ordered_extent;
3168 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3169 btrfs_finish_ordered_io(ordered_extent);
3172 void btrfs_writepage_endio_finish_ordered(struct btrfs_inode *inode,
3173 struct page *page, u64 start,
3174 u64 end, int uptodate)
3176 trace_btrfs_writepage_end_io_hook(inode, start, end, uptodate);
3178 btrfs_mark_ordered_io_finished(inode, page, start, end + 1 - start,
3179 finish_ordered_fn, uptodate);
3183 * check_data_csum - verify checksum of one sector of uncompressed data
3185 * @io_bio: btrfs_io_bio which contains the csum
3186 * @bio_offset: offset to the beginning of the bio (in bytes)
3187 * @page: page where is the data to be verified
3188 * @pgoff: offset inside the page
3189 * @start: logical offset in the file
3191 * The length of such check is always one sector size.
3193 static int check_data_csum(struct inode *inode, struct btrfs_io_bio *io_bio,
3194 u32 bio_offset, struct page *page, u32 pgoff,
3197 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3198 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3200 u32 len = fs_info->sectorsize;
3201 const u32 csum_size = fs_info->csum_size;
3202 unsigned int offset_sectors;
3204 u8 csum[BTRFS_CSUM_SIZE];
3206 ASSERT(pgoff + len <= PAGE_SIZE);
3208 offset_sectors = bio_offset >> fs_info->sectorsize_bits;
3209 csum_expected = ((u8 *)io_bio->csum) + offset_sectors * csum_size;
3211 kaddr = kmap_atomic(page);
3212 shash->tfm = fs_info->csum_shash;
3214 crypto_shash_digest(shash, kaddr + pgoff, len, csum);
3216 if (memcmp(csum, csum_expected, csum_size))
3219 kunmap_atomic(kaddr);
3222 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3223 io_bio->mirror_num);
3225 btrfs_dev_stat_inc_and_print(io_bio->device,
3226 BTRFS_DEV_STAT_CORRUPTION_ERRS);
3227 memset(kaddr + pgoff, 1, len);
3228 flush_dcache_page(page);
3229 kunmap_atomic(kaddr);
3234 * When reads are done, we need to check csums to verify the data is correct.
3235 * if there's a match, we allow the bio to finish. If not, the code in
3236 * extent_io.c will try to find good copies for us.
3238 * @bio_offset: offset to the beginning of the bio (in bytes)
3239 * @start: file offset of the range start
3240 * @end: file offset of the range end (inclusive)
3242 * Return a bitmap where bit set means a csum mismatch, and bit not set means
3245 unsigned int btrfs_verify_data_csum(struct btrfs_io_bio *io_bio, u32 bio_offset,
3246 struct page *page, u64 start, u64 end)
3248 struct inode *inode = page->mapping->host;
3249 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3250 struct btrfs_root *root = BTRFS_I(inode)->root;
3251 const u32 sectorsize = root->fs_info->sectorsize;
3253 unsigned int result = 0;
3255 if (PageChecked(page)) {
3256 ClearPageChecked(page);
3260 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3263 if (!root->fs_info->csum_root)
3266 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3267 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3268 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3272 ASSERT(page_offset(page) <= start &&
3273 end <= page_offset(page) + PAGE_SIZE - 1);
3274 for (pg_off = offset_in_page(start);
3275 pg_off < offset_in_page(end);
3276 pg_off += sectorsize, bio_offset += sectorsize) {
3279 ret = check_data_csum(inode, io_bio, bio_offset, page, pg_off,
3280 page_offset(page) + pg_off);
3282 const int nr_bit = (pg_off - offset_in_page(start)) >>
3283 root->fs_info->sectorsize_bits;
3285 result |= (1U << nr_bit);
3292 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3294 * @inode: The inode we want to perform iput on
3296 * This function uses the generic vfs_inode::i_count to track whether we should
3297 * just decrement it (in case it's > 1) or if this is the last iput then link
3298 * the inode to the delayed iput machinery. Delayed iputs are processed at
3299 * transaction commit time/superblock commit/cleaner kthread.
3301 void btrfs_add_delayed_iput(struct inode *inode)
3303 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3304 struct btrfs_inode *binode = BTRFS_I(inode);
3306 if (atomic_add_unless(&inode->i_count, -1, 1))
3309 atomic_inc(&fs_info->nr_delayed_iputs);
3310 spin_lock(&fs_info->delayed_iput_lock);
3311 ASSERT(list_empty(&binode->delayed_iput));
3312 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3313 spin_unlock(&fs_info->delayed_iput_lock);
3314 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3315 wake_up_process(fs_info->cleaner_kthread);
3318 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3319 struct btrfs_inode *inode)
3321 list_del_init(&inode->delayed_iput);
3322 spin_unlock(&fs_info->delayed_iput_lock);
3323 iput(&inode->vfs_inode);
3324 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3325 wake_up(&fs_info->delayed_iputs_wait);
3326 spin_lock(&fs_info->delayed_iput_lock);
3329 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3330 struct btrfs_inode *inode)
3332 if (!list_empty(&inode->delayed_iput)) {
3333 spin_lock(&fs_info->delayed_iput_lock);
3334 if (!list_empty(&inode->delayed_iput))
3335 run_delayed_iput_locked(fs_info, inode);
3336 spin_unlock(&fs_info->delayed_iput_lock);
3340 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3343 spin_lock(&fs_info->delayed_iput_lock);
3344 while (!list_empty(&fs_info->delayed_iputs)) {
3345 struct btrfs_inode *inode;
3347 inode = list_first_entry(&fs_info->delayed_iputs,
3348 struct btrfs_inode, delayed_iput);
3349 run_delayed_iput_locked(fs_info, inode);
3350 cond_resched_lock(&fs_info->delayed_iput_lock);
3352 spin_unlock(&fs_info->delayed_iput_lock);
3356 * Wait for flushing all delayed iputs
3358 * @fs_info: the filesystem
3360 * This will wait on any delayed iputs that are currently running with KILLABLE
3361 * set. Once they are all done running we will return, unless we are killed in
3362 * which case we return EINTR. This helps in user operations like fallocate etc
3363 * that might get blocked on the iputs.
3365 * Return EINTR if we were killed, 0 if nothing's pending
3367 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3369 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3370 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3377 * This creates an orphan entry for the given inode in case something goes wrong
3378 * in the middle of an unlink.
3380 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3381 struct btrfs_inode *inode)
3385 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3386 if (ret && ret != -EEXIST) {
3387 btrfs_abort_transaction(trans, ret);
3395 * We have done the delete so we can go ahead and remove the orphan item for
3396 * this particular inode.
3398 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3399 struct btrfs_inode *inode)
3401 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3405 * this cleans up any orphans that may be left on the list from the last use
3408 int btrfs_orphan_cleanup(struct btrfs_root *root)
3410 struct btrfs_fs_info *fs_info = root->fs_info;
3411 struct btrfs_path *path;
3412 struct extent_buffer *leaf;
3413 struct btrfs_key key, found_key;
3414 struct btrfs_trans_handle *trans;
3415 struct inode *inode;
3416 u64 last_objectid = 0;
3417 int ret = 0, nr_unlink = 0;
3419 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3422 path = btrfs_alloc_path();
3427 path->reada = READA_BACK;
3429 key.objectid = BTRFS_ORPHAN_OBJECTID;
3430 key.type = BTRFS_ORPHAN_ITEM_KEY;
3431 key.offset = (u64)-1;
3434 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3439 * if ret == 0 means we found what we were searching for, which
3440 * is weird, but possible, so only screw with path if we didn't
3441 * find the key and see if we have stuff that matches
3445 if (path->slots[0] == 0)
3450 /* pull out the item */
3451 leaf = path->nodes[0];
3452 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3454 /* make sure the item matches what we want */
3455 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3457 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3460 /* release the path since we're done with it */
3461 btrfs_release_path(path);
3464 * this is where we are basically btrfs_lookup, without the
3465 * crossing root thing. we store the inode number in the
3466 * offset of the orphan item.
3469 if (found_key.offset == last_objectid) {
3471 "Error removing orphan entry, stopping orphan cleanup");
3476 last_objectid = found_key.offset;
3478 found_key.objectid = found_key.offset;
3479 found_key.type = BTRFS_INODE_ITEM_KEY;
3480 found_key.offset = 0;
3481 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3482 ret = PTR_ERR_OR_ZERO(inode);
3483 if (ret && ret != -ENOENT)
3486 if (ret == -ENOENT && root == fs_info->tree_root) {
3487 struct btrfs_root *dead_root;
3488 int is_dead_root = 0;
3491 * This is an orphan in the tree root. Currently these
3492 * could come from 2 sources:
3493 * a) a root (snapshot/subvolume) deletion in progress
3494 * b) a free space cache inode
3495 * We need to distinguish those two, as the orphan item
3496 * for a root must not get deleted before the deletion
3497 * of the snapshot/subvolume's tree completes.
3499 * btrfs_find_orphan_roots() ran before us, which has
3500 * found all deleted roots and loaded them into
3501 * fs_info->fs_roots_radix. So here we can find if an
3502 * orphan item corresponds to a deleted root by looking
3503 * up the root from that radix tree.
3506 spin_lock(&fs_info->fs_roots_radix_lock);
3507 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3508 (unsigned long)found_key.objectid);
3509 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3511 spin_unlock(&fs_info->fs_roots_radix_lock);
3514 /* prevent this orphan from being found again */
3515 key.offset = found_key.objectid - 1;
3522 * If we have an inode with links, there are a couple of
3523 * possibilities. Old kernels (before v3.12) used to create an
3524 * orphan item for truncate indicating that there were possibly
3525 * extent items past i_size that needed to be deleted. In v3.12,
3526 * truncate was changed to update i_size in sync with the extent
3527 * items, but the (useless) orphan item was still created. Since
3528 * v4.18, we don't create the orphan item for truncate at all.
3530 * So, this item could mean that we need to do a truncate, but
3531 * only if this filesystem was last used on a pre-v3.12 kernel
3532 * and was not cleanly unmounted. The odds of that are quite
3533 * slim, and it's a pain to do the truncate now, so just delete
3536 * It's also possible that this orphan item was supposed to be
3537 * deleted but wasn't. The inode number may have been reused,
3538 * but either way, we can delete the orphan item.
3540 if (ret == -ENOENT || inode->i_nlink) {
3543 trans = btrfs_start_transaction(root, 1);
3544 if (IS_ERR(trans)) {
3545 ret = PTR_ERR(trans);
3548 btrfs_debug(fs_info, "auto deleting %Lu",
3549 found_key.objectid);
3550 ret = btrfs_del_orphan_item(trans, root,
3551 found_key.objectid);
3552 btrfs_end_transaction(trans);
3560 /* this will do delete_inode and everything for us */
3563 /* release the path since we're done with it */
3564 btrfs_release_path(path);
3566 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3568 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3569 trans = btrfs_join_transaction(root);
3571 btrfs_end_transaction(trans);
3575 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3579 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3580 btrfs_free_path(path);
3585 * very simple check to peek ahead in the leaf looking for xattrs. If we
3586 * don't find any xattrs, we know there can't be any acls.
3588 * slot is the slot the inode is in, objectid is the objectid of the inode
3590 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3591 int slot, u64 objectid,
3592 int *first_xattr_slot)
3594 u32 nritems = btrfs_header_nritems(leaf);
3595 struct btrfs_key found_key;
3596 static u64 xattr_access = 0;
3597 static u64 xattr_default = 0;
3600 if (!xattr_access) {
3601 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3602 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3603 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3604 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3608 *first_xattr_slot = -1;
3609 while (slot < nritems) {
3610 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3612 /* we found a different objectid, there must not be acls */
3613 if (found_key.objectid != objectid)
3616 /* we found an xattr, assume we've got an acl */
3617 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3618 if (*first_xattr_slot == -1)
3619 *first_xattr_slot = slot;
3620 if (found_key.offset == xattr_access ||
3621 found_key.offset == xattr_default)
3626 * we found a key greater than an xattr key, there can't
3627 * be any acls later on
3629 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3636 * it goes inode, inode backrefs, xattrs, extents,
3637 * so if there are a ton of hard links to an inode there can
3638 * be a lot of backrefs. Don't waste time searching too hard,
3639 * this is just an optimization
3644 /* we hit the end of the leaf before we found an xattr or
3645 * something larger than an xattr. We have to assume the inode
3648 if (*first_xattr_slot == -1)
3649 *first_xattr_slot = slot;
3654 * read an inode from the btree into the in-memory inode
3656 static int btrfs_read_locked_inode(struct inode *inode,
3657 struct btrfs_path *in_path)
3659 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3660 struct btrfs_path *path = in_path;
3661 struct extent_buffer *leaf;
3662 struct btrfs_inode_item *inode_item;
3663 struct btrfs_root *root = BTRFS_I(inode)->root;
3664 struct btrfs_key location;
3669 bool filled = false;
3670 int first_xattr_slot;
3672 ret = btrfs_fill_inode(inode, &rdev);
3677 path = btrfs_alloc_path();
3682 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3684 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3686 if (path != in_path)
3687 btrfs_free_path(path);
3691 leaf = path->nodes[0];
3696 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3697 struct btrfs_inode_item);
3698 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3699 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3700 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3701 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3702 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3703 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3704 round_up(i_size_read(inode), fs_info->sectorsize));
3706 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3707 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3709 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3710 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3712 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3713 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3715 BTRFS_I(inode)->i_otime.tv_sec =
3716 btrfs_timespec_sec(leaf, &inode_item->otime);
3717 BTRFS_I(inode)->i_otime.tv_nsec =
3718 btrfs_timespec_nsec(leaf, &inode_item->otime);
3720 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3721 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3722 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3724 inode_set_iversion_queried(inode,
3725 btrfs_inode_sequence(leaf, inode_item));
3726 inode->i_generation = BTRFS_I(inode)->generation;
3728 rdev = btrfs_inode_rdev(leaf, inode_item);
3730 BTRFS_I(inode)->index_cnt = (u64)-1;
3731 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3735 * If we were modified in the current generation and evicted from memory
3736 * and then re-read we need to do a full sync since we don't have any
3737 * idea about which extents were modified before we were evicted from
3740 * This is required for both inode re-read from disk and delayed inode
3741 * in delayed_nodes_tree.
3743 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3744 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3745 &BTRFS_I(inode)->runtime_flags);
3748 * We don't persist the id of the transaction where an unlink operation
3749 * against the inode was last made. So here we assume the inode might
3750 * have been evicted, and therefore the exact value of last_unlink_trans
3751 * lost, and set it to last_trans to avoid metadata inconsistencies
3752 * between the inode and its parent if the inode is fsync'ed and the log
3753 * replayed. For example, in the scenario:
3756 * ln mydir/foo mydir/bar
3759 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3760 * xfs_io -c fsync mydir/foo
3762 * mount fs, triggers fsync log replay
3764 * We must make sure that when we fsync our inode foo we also log its
3765 * parent inode, otherwise after log replay the parent still has the
3766 * dentry with the "bar" name but our inode foo has a link count of 1
3767 * and doesn't have an inode ref with the name "bar" anymore.
3769 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3770 * but it guarantees correctness at the expense of occasional full
3771 * transaction commits on fsync if our inode is a directory, or if our
3772 * inode is not a directory, logging its parent unnecessarily.
3774 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3777 * Same logic as for last_unlink_trans. We don't persist the generation
3778 * of the last transaction where this inode was used for a reflink
3779 * operation, so after eviction and reloading the inode we must be
3780 * pessimistic and assume the last transaction that modified the inode.
3782 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3785 if (inode->i_nlink != 1 ||
3786 path->slots[0] >= btrfs_header_nritems(leaf))
3789 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3790 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3793 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3794 if (location.type == BTRFS_INODE_REF_KEY) {
3795 struct btrfs_inode_ref *ref;
3797 ref = (struct btrfs_inode_ref *)ptr;
3798 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3799 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3800 struct btrfs_inode_extref *extref;
3802 extref = (struct btrfs_inode_extref *)ptr;
3803 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3808 * try to precache a NULL acl entry for files that don't have
3809 * any xattrs or acls
3811 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3812 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3813 if (first_xattr_slot != -1) {
3814 path->slots[0] = first_xattr_slot;
3815 ret = btrfs_load_inode_props(inode, path);
3818 "error loading props for ino %llu (root %llu): %d",
3819 btrfs_ino(BTRFS_I(inode)),
3820 root->root_key.objectid, ret);
3822 if (path != in_path)
3823 btrfs_free_path(path);
3826 cache_no_acl(inode);
3828 switch (inode->i_mode & S_IFMT) {
3830 inode->i_mapping->a_ops = &btrfs_aops;
3831 inode->i_fop = &btrfs_file_operations;
3832 inode->i_op = &btrfs_file_inode_operations;
3835 inode->i_fop = &btrfs_dir_file_operations;
3836 inode->i_op = &btrfs_dir_inode_operations;
3839 inode->i_op = &btrfs_symlink_inode_operations;
3840 inode_nohighmem(inode);
3841 inode->i_mapping->a_ops = &btrfs_aops;
3844 inode->i_op = &btrfs_special_inode_operations;
3845 init_special_inode(inode, inode->i_mode, rdev);
3849 btrfs_sync_inode_flags_to_i_flags(inode);
3854 * given a leaf and an inode, copy the inode fields into the leaf
3856 static void fill_inode_item(struct btrfs_trans_handle *trans,
3857 struct extent_buffer *leaf,
3858 struct btrfs_inode_item *item,
3859 struct inode *inode)
3861 struct btrfs_map_token token;
3863 btrfs_init_map_token(&token, leaf);
3865 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3866 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3867 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
3868 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3869 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3871 btrfs_set_token_timespec_sec(&token, &item->atime,
3872 inode->i_atime.tv_sec);
3873 btrfs_set_token_timespec_nsec(&token, &item->atime,
3874 inode->i_atime.tv_nsec);
3876 btrfs_set_token_timespec_sec(&token, &item->mtime,
3877 inode->i_mtime.tv_sec);
3878 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3879 inode->i_mtime.tv_nsec);
3881 btrfs_set_token_timespec_sec(&token, &item->ctime,
3882 inode->i_ctime.tv_sec);
3883 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3884 inode->i_ctime.tv_nsec);
3886 btrfs_set_token_timespec_sec(&token, &item->otime,
3887 BTRFS_I(inode)->i_otime.tv_sec);
3888 btrfs_set_token_timespec_nsec(&token, &item->otime,
3889 BTRFS_I(inode)->i_otime.tv_nsec);
3891 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
3892 btrfs_set_token_inode_generation(&token, item,
3893 BTRFS_I(inode)->generation);
3894 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
3895 btrfs_set_token_inode_transid(&token, item, trans->transid);
3896 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
3897 btrfs_set_token_inode_flags(&token, item, BTRFS_I(inode)->flags);
3898 btrfs_set_token_inode_block_group(&token, item, 0);
3902 * copy everything in the in-memory inode into the btree.
3904 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3905 struct btrfs_root *root,
3906 struct btrfs_inode *inode)
3908 struct btrfs_inode_item *inode_item;
3909 struct btrfs_path *path;
3910 struct extent_buffer *leaf;
3913 path = btrfs_alloc_path();
3917 ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1);
3924 leaf = path->nodes[0];
3925 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3926 struct btrfs_inode_item);
3928 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
3929 btrfs_mark_buffer_dirty(leaf);
3930 btrfs_set_inode_last_trans(trans, inode);
3933 btrfs_free_path(path);
3938 * copy everything in the in-memory inode into the btree.
3940 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3941 struct btrfs_root *root,
3942 struct btrfs_inode *inode)
3944 struct btrfs_fs_info *fs_info = root->fs_info;
3948 * If the inode is a free space inode, we can deadlock during commit
3949 * if we put it into the delayed code.
3951 * The data relocation inode should also be directly updated
3954 if (!btrfs_is_free_space_inode(inode)
3955 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3956 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
3957 btrfs_update_root_times(trans, root);
3959 ret = btrfs_delayed_update_inode(trans, root, inode);
3961 btrfs_set_inode_last_trans(trans, inode);
3965 return btrfs_update_inode_item(trans, root, inode);
3968 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3969 struct btrfs_root *root, struct btrfs_inode *inode)
3973 ret = btrfs_update_inode(trans, root, inode);
3975 return btrfs_update_inode_item(trans, root, inode);
3980 * unlink helper that gets used here in inode.c and in the tree logging
3981 * recovery code. It remove a link in a directory with a given name, and
3982 * also drops the back refs in the inode to the directory
3984 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3985 struct btrfs_root *root,
3986 struct btrfs_inode *dir,
3987 struct btrfs_inode *inode,
3988 const char *name, int name_len)
3990 struct btrfs_fs_info *fs_info = root->fs_info;
3991 struct btrfs_path *path;
3993 struct btrfs_dir_item *di;
3995 u64 ino = btrfs_ino(inode);
3996 u64 dir_ino = btrfs_ino(dir);
3998 path = btrfs_alloc_path();
4004 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4005 name, name_len, -1);
4006 if (IS_ERR_OR_NULL(di)) {
4007 ret = di ? PTR_ERR(di) : -ENOENT;
4010 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4013 btrfs_release_path(path);
4016 * If we don't have dir index, we have to get it by looking up
4017 * the inode ref, since we get the inode ref, remove it directly,
4018 * it is unnecessary to do delayed deletion.
4020 * But if we have dir index, needn't search inode ref to get it.
4021 * Since the inode ref is close to the inode item, it is better
4022 * that we delay to delete it, and just do this deletion when
4023 * we update the inode item.
4025 if (inode->dir_index) {
4026 ret = btrfs_delayed_delete_inode_ref(inode);
4028 index = inode->dir_index;
4033 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4037 "failed to delete reference to %.*s, inode %llu parent %llu",
4038 name_len, name, ino, dir_ino);
4039 btrfs_abort_transaction(trans, ret);
4043 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4045 btrfs_abort_transaction(trans, ret);
4049 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
4051 if (ret != 0 && ret != -ENOENT) {
4052 btrfs_abort_transaction(trans, ret);
4056 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
4061 btrfs_abort_transaction(trans, ret);
4064 * If we have a pending delayed iput we could end up with the final iput
4065 * being run in btrfs-cleaner context. If we have enough of these built
4066 * up we can end up burning a lot of time in btrfs-cleaner without any
4067 * way to throttle the unlinks. Since we're currently holding a ref on
4068 * the inode we can run the delayed iput here without any issues as the
4069 * final iput won't be done until after we drop the ref we're currently
4072 btrfs_run_delayed_iput(fs_info, inode);
4074 btrfs_free_path(path);
4078 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
4079 inode_inc_iversion(&inode->vfs_inode);
4080 inode_inc_iversion(&dir->vfs_inode);
4081 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
4082 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4083 ret = btrfs_update_inode(trans, root, dir);
4088 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4089 struct btrfs_root *root,
4090 struct btrfs_inode *dir, struct btrfs_inode *inode,
4091 const char *name, int name_len)
4094 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4096 drop_nlink(&inode->vfs_inode);
4097 ret = btrfs_update_inode(trans, root, inode);
4103 * helper to start transaction for unlink and rmdir.
4105 * unlink and rmdir are special in btrfs, they do not always free space, so
4106 * if we cannot make our reservations the normal way try and see if there is
4107 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4108 * allow the unlink to occur.
4110 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4112 struct btrfs_root *root = BTRFS_I(dir)->root;
4115 * 1 for the possible orphan item
4116 * 1 for the dir item
4117 * 1 for the dir index
4118 * 1 for the inode ref
4121 return btrfs_start_transaction_fallback_global_rsv(root, 5);
4124 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4126 struct btrfs_root *root = BTRFS_I(dir)->root;
4127 struct btrfs_trans_handle *trans;
4128 struct inode *inode = d_inode(dentry);
4131 trans = __unlink_start_trans(dir);
4133 return PTR_ERR(trans);
4135 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4138 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4139 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4140 dentry->d_name.len);
4144 if (inode->i_nlink == 0) {
4145 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4151 btrfs_end_transaction(trans);
4152 btrfs_btree_balance_dirty(root->fs_info);
4156 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4157 struct inode *dir, struct dentry *dentry)
4159 struct btrfs_root *root = BTRFS_I(dir)->root;
4160 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4161 struct btrfs_path *path;
4162 struct extent_buffer *leaf;
4163 struct btrfs_dir_item *di;
4164 struct btrfs_key key;
4165 const char *name = dentry->d_name.name;
4166 int name_len = dentry->d_name.len;
4170 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4172 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4173 objectid = inode->root->root_key.objectid;
4174 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4175 objectid = inode->location.objectid;
4181 path = btrfs_alloc_path();
4185 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4186 name, name_len, -1);
4187 if (IS_ERR_OR_NULL(di)) {
4188 ret = di ? PTR_ERR(di) : -ENOENT;
4192 leaf = path->nodes[0];
4193 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4194 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4195 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4197 btrfs_abort_transaction(trans, ret);
4200 btrfs_release_path(path);
4203 * This is a placeholder inode for a subvolume we didn't have a
4204 * reference to at the time of the snapshot creation. In the meantime
4205 * we could have renamed the real subvol link into our snapshot, so
4206 * depending on btrfs_del_root_ref to return -ENOENT here is incorrect.
4207 * Instead simply lookup the dir_index_item for this entry so we can
4208 * remove it. Otherwise we know we have a ref to the root and we can
4209 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4211 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4212 di = btrfs_search_dir_index_item(root, path, dir_ino,
4214 if (IS_ERR_OR_NULL(di)) {
4219 btrfs_abort_transaction(trans, ret);
4223 leaf = path->nodes[0];
4224 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4226 btrfs_release_path(path);
4228 ret = btrfs_del_root_ref(trans, objectid,
4229 root->root_key.objectid, dir_ino,
4230 &index, name, name_len);
4232 btrfs_abort_transaction(trans, ret);
4237 ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index);
4239 btrfs_abort_transaction(trans, ret);
4243 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4244 inode_inc_iversion(dir);
4245 dir->i_mtime = dir->i_ctime = current_time(dir);
4246 ret = btrfs_update_inode_fallback(trans, root, BTRFS_I(dir));
4248 btrfs_abort_transaction(trans, ret);
4250 btrfs_free_path(path);
4255 * Helper to check if the subvolume references other subvolumes or if it's
4258 static noinline int may_destroy_subvol(struct btrfs_root *root)
4260 struct btrfs_fs_info *fs_info = root->fs_info;
4261 struct btrfs_path *path;
4262 struct btrfs_dir_item *di;
4263 struct btrfs_key key;
4267 path = btrfs_alloc_path();
4271 /* Make sure this root isn't set as the default subvol */
4272 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4273 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4274 dir_id, "default", 7, 0);
4275 if (di && !IS_ERR(di)) {
4276 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4277 if (key.objectid == root->root_key.objectid) {
4280 "deleting default subvolume %llu is not allowed",
4284 btrfs_release_path(path);
4287 key.objectid = root->root_key.objectid;
4288 key.type = BTRFS_ROOT_REF_KEY;
4289 key.offset = (u64)-1;
4291 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4297 if (path->slots[0] > 0) {
4299 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4300 if (key.objectid == root->root_key.objectid &&
4301 key.type == BTRFS_ROOT_REF_KEY)
4305 btrfs_free_path(path);
4309 /* Delete all dentries for inodes belonging to the root */
4310 static void btrfs_prune_dentries(struct btrfs_root *root)
4312 struct btrfs_fs_info *fs_info = root->fs_info;
4313 struct rb_node *node;
4314 struct rb_node *prev;
4315 struct btrfs_inode *entry;
4316 struct inode *inode;
4319 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
4320 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4322 spin_lock(&root->inode_lock);
4324 node = root->inode_tree.rb_node;
4328 entry = rb_entry(node, struct btrfs_inode, rb_node);
4330 if (objectid < btrfs_ino(entry))
4331 node = node->rb_left;
4332 else if (objectid > btrfs_ino(entry))
4333 node = node->rb_right;
4339 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4340 if (objectid <= btrfs_ino(entry)) {
4344 prev = rb_next(prev);
4348 entry = rb_entry(node, struct btrfs_inode, rb_node);
4349 objectid = btrfs_ino(entry) + 1;
4350 inode = igrab(&entry->vfs_inode);
4352 spin_unlock(&root->inode_lock);
4353 if (atomic_read(&inode->i_count) > 1)
4354 d_prune_aliases(inode);
4356 * btrfs_drop_inode will have it removed from the inode
4357 * cache when its usage count hits zero.
4361 spin_lock(&root->inode_lock);
4365 if (cond_resched_lock(&root->inode_lock))
4368 node = rb_next(node);
4370 spin_unlock(&root->inode_lock);
4373 int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
4375 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4376 struct btrfs_root *root = BTRFS_I(dir)->root;
4377 struct inode *inode = d_inode(dentry);
4378 struct btrfs_root *dest = BTRFS_I(inode)->root;
4379 struct btrfs_trans_handle *trans;
4380 struct btrfs_block_rsv block_rsv;
4385 * Don't allow to delete a subvolume with send in progress. This is
4386 * inside the inode lock so the error handling that has to drop the bit
4387 * again is not run concurrently.
4389 spin_lock(&dest->root_item_lock);
4390 if (dest->send_in_progress) {
4391 spin_unlock(&dest->root_item_lock);
4393 "attempt to delete subvolume %llu during send",
4394 dest->root_key.objectid);
4397 root_flags = btrfs_root_flags(&dest->root_item);
4398 btrfs_set_root_flags(&dest->root_item,
4399 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4400 spin_unlock(&dest->root_item_lock);
4402 down_write(&fs_info->subvol_sem);
4404 ret = may_destroy_subvol(dest);
4408 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4410 * One for dir inode,
4411 * two for dir entries,
4412 * two for root ref/backref.
4414 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4418 trans = btrfs_start_transaction(root, 0);
4419 if (IS_ERR(trans)) {
4420 ret = PTR_ERR(trans);
4423 trans->block_rsv = &block_rsv;
4424 trans->bytes_reserved = block_rsv.size;
4426 btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
4428 ret = btrfs_unlink_subvol(trans, dir, dentry);
4430 btrfs_abort_transaction(trans, ret);
4434 ret = btrfs_record_root_in_trans(trans, dest);
4436 btrfs_abort_transaction(trans, ret);
4440 memset(&dest->root_item.drop_progress, 0,
4441 sizeof(dest->root_item.drop_progress));
4442 btrfs_set_root_drop_level(&dest->root_item, 0);
4443 btrfs_set_root_refs(&dest->root_item, 0);
4445 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4446 ret = btrfs_insert_orphan_item(trans,
4448 dest->root_key.objectid);
4450 btrfs_abort_transaction(trans, ret);
4455 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4456 BTRFS_UUID_KEY_SUBVOL,
4457 dest->root_key.objectid);
4458 if (ret && ret != -ENOENT) {
4459 btrfs_abort_transaction(trans, ret);
4462 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4463 ret = btrfs_uuid_tree_remove(trans,
4464 dest->root_item.received_uuid,
4465 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4466 dest->root_key.objectid);
4467 if (ret && ret != -ENOENT) {
4468 btrfs_abort_transaction(trans, ret);
4473 free_anon_bdev(dest->anon_dev);
4476 trans->block_rsv = NULL;
4477 trans->bytes_reserved = 0;
4478 ret = btrfs_end_transaction(trans);
4479 inode->i_flags |= S_DEAD;
4481 btrfs_subvolume_release_metadata(root, &block_rsv);
4483 up_write(&fs_info->subvol_sem);
4485 spin_lock(&dest->root_item_lock);
4486 root_flags = btrfs_root_flags(&dest->root_item);
4487 btrfs_set_root_flags(&dest->root_item,
4488 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4489 spin_unlock(&dest->root_item_lock);
4491 d_invalidate(dentry);
4492 btrfs_prune_dentries(dest);
4493 ASSERT(dest->send_in_progress == 0);
4499 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4501 struct inode *inode = d_inode(dentry);
4503 struct btrfs_root *root = BTRFS_I(dir)->root;
4504 struct btrfs_trans_handle *trans;
4505 u64 last_unlink_trans;
4507 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4509 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4510 return btrfs_delete_subvolume(dir, dentry);
4512 trans = __unlink_start_trans(dir);
4514 return PTR_ERR(trans);
4516 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4517 err = btrfs_unlink_subvol(trans, dir, dentry);
4521 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4525 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4527 /* now the directory is empty */
4528 err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4529 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4530 dentry->d_name.len);
4532 btrfs_i_size_write(BTRFS_I(inode), 0);
4534 * Propagate the last_unlink_trans value of the deleted dir to
4535 * its parent directory. This is to prevent an unrecoverable
4536 * log tree in the case we do something like this:
4538 * 2) create snapshot under dir foo
4539 * 3) delete the snapshot
4542 * 6) fsync foo or some file inside foo
4544 if (last_unlink_trans >= trans->transid)
4545 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4548 btrfs_end_transaction(trans);
4549 btrfs_btree_balance_dirty(root->fs_info);
4555 * Return this if we need to call truncate_block for the last bit of the
4558 #define NEED_TRUNCATE_BLOCK 1
4561 * Remove inode items from a given root.
4563 * @trans: A transaction handle.
4564 * @root: The root from which to remove items.
4565 * @inode: The inode whose items we want to remove.
4566 * @new_size: The new i_size for the inode. This is only applicable when
4567 * @min_type is BTRFS_EXTENT_DATA_KEY, must be 0 otherwise.
4568 * @min_type: The minimum key type to remove. All keys with a type
4569 * greater than this value are removed and all keys with
4570 * this type are removed only if their offset is >= @new_size.
4571 * @extents_found: Output parameter that will contain the number of file
4572 * extent items that were removed or adjusted to the new
4573 * inode i_size. The caller is responsible for initializing
4574 * the counter. Also, it can be NULL if the caller does not
4575 * need this counter.
4577 * Remove all keys associated with the inode from the given root that have a key
4578 * with a type greater than or equals to @min_type. When @min_type has a value of
4579 * BTRFS_EXTENT_DATA_KEY, only remove file extent items that have an offset value
4580 * greater than or equals to @new_size. If a file extent item that starts before
4581 * @new_size and ends after it is found, its length is adjusted.
4583 * Returns: 0 on success, < 0 on error and NEED_TRUNCATE_BLOCK when @min_type is
4584 * BTRFS_EXTENT_DATA_KEY and the caller must truncate the last block.
4586 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4587 struct btrfs_root *root,
4588 struct btrfs_inode *inode,
4589 u64 new_size, u32 min_type,
4592 struct btrfs_fs_info *fs_info = root->fs_info;
4593 struct btrfs_path *path;
4594 struct extent_buffer *leaf;
4595 struct btrfs_file_extent_item *fi;
4596 struct btrfs_key key;
4597 struct btrfs_key found_key;
4598 u64 extent_start = 0;
4599 u64 extent_num_bytes = 0;
4600 u64 extent_offset = 0;
4602 u64 last_size = new_size;
4603 u32 found_type = (u8)-1;
4606 int pending_del_nr = 0;
4607 int pending_del_slot = 0;
4608 int extent_type = -1;
4610 u64 ino = btrfs_ino(inode);
4611 u64 bytes_deleted = 0;
4612 bool be_nice = false;
4613 bool should_throttle = false;
4614 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
4615 struct extent_state *cached_state = NULL;
4617 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4620 * For non-free space inodes and non-shareable roots, we want to back
4621 * off from time to time. This means all inodes in subvolume roots,
4622 * reloc roots, and data reloc roots.
4624 if (!btrfs_is_free_space_inode(inode) &&
4625 test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
4628 path = btrfs_alloc_path();
4631 path->reada = READA_BACK;
4633 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4634 lock_extent_bits(&inode->io_tree, lock_start, (u64)-1,
4638 * We want to drop from the next block forward in case this
4639 * new size is not block aligned since we will be keeping the
4640 * last block of the extent just the way it is.
4642 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4643 fs_info->sectorsize),
4648 * This function is also used to drop the items in the log tree before
4649 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4650 * it is used to drop the logged items. So we shouldn't kill the delayed
4653 if (min_type == 0 && root == inode->root)
4654 btrfs_kill_delayed_inode_items(inode);
4657 key.offset = (u64)-1;
4662 * with a 16K leaf size and 128MB extents, you can actually queue
4663 * up a huge file in a single leaf. Most of the time that
4664 * bytes_deleted is > 0, it will be huge by the time we get here
4666 if (be_nice && bytes_deleted > SZ_32M &&
4667 btrfs_should_end_transaction(trans)) {
4672 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4678 /* there are no items in the tree for us to truncate, we're
4681 if (path->slots[0] == 0)
4687 u64 clear_start = 0, clear_len = 0;
4690 leaf = path->nodes[0];
4691 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4692 found_type = found_key.type;
4694 if (found_key.objectid != ino)
4697 if (found_type < min_type)
4700 item_end = found_key.offset;
4701 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4702 fi = btrfs_item_ptr(leaf, path->slots[0],
4703 struct btrfs_file_extent_item);
4704 extent_type = btrfs_file_extent_type(leaf, fi);
4705 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4707 btrfs_file_extent_num_bytes(leaf, fi);
4709 trace_btrfs_truncate_show_fi_regular(
4710 inode, leaf, fi, found_key.offset);
4711 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4712 item_end += btrfs_file_extent_ram_bytes(leaf,
4715 trace_btrfs_truncate_show_fi_inline(
4716 inode, leaf, fi, path->slots[0],
4721 if (found_type > min_type) {
4724 if (item_end < new_size)
4726 if (found_key.offset >= new_size)
4732 /* FIXME, shrink the extent if the ref count is only 1 */
4733 if (found_type != BTRFS_EXTENT_DATA_KEY)
4736 if (extents_found != NULL)
4739 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4742 clear_start = found_key.offset;
4743 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4745 u64 orig_num_bytes =
4746 btrfs_file_extent_num_bytes(leaf, fi);
4747 extent_num_bytes = ALIGN(new_size -
4749 fs_info->sectorsize);
4750 clear_start = ALIGN(new_size, fs_info->sectorsize);
4751 btrfs_set_file_extent_num_bytes(leaf, fi,
4753 num_dec = (orig_num_bytes -
4755 if (test_bit(BTRFS_ROOT_SHAREABLE,
4758 inode_sub_bytes(&inode->vfs_inode,
4760 btrfs_mark_buffer_dirty(leaf);
4763 btrfs_file_extent_disk_num_bytes(leaf,
4765 extent_offset = found_key.offset -
4766 btrfs_file_extent_offset(leaf, fi);
4768 /* FIXME blocksize != 4096 */
4769 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4770 if (extent_start != 0) {
4772 if (test_bit(BTRFS_ROOT_SHAREABLE,
4774 inode_sub_bytes(&inode->vfs_inode,
4778 clear_len = num_dec;
4779 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4781 * we can't truncate inline items that have had
4785 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4786 btrfs_file_extent_other_encoding(leaf, fi) == 0 &&
4787 btrfs_file_extent_compression(leaf, fi) == 0) {
4788 u32 size = (u32)(new_size - found_key.offset);
4790 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4791 size = btrfs_file_extent_calc_inline_size(size);
4792 btrfs_truncate_item(path, size, 1);
4793 } else if (!del_item) {
4795 * We have to bail so the last_size is set to
4796 * just before this extent.
4798 ret = NEED_TRUNCATE_BLOCK;
4802 * Inline extents are special, we just treat
4803 * them as a full sector worth in the file
4804 * extent tree just for simplicity sake.
4806 clear_len = fs_info->sectorsize;
4809 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
4810 inode_sub_bytes(&inode->vfs_inode,
4811 item_end + 1 - new_size);
4815 * We use btrfs_truncate_inode_items() to clean up log trees for
4816 * multiple fsyncs, and in this case we don't want to clear the
4817 * file extent range because it's just the log.
4819 if (root == inode->root) {
4820 ret = btrfs_inode_clear_file_extent_range(inode,
4821 clear_start, clear_len);
4823 btrfs_abort_transaction(trans, ret);
4829 last_size = found_key.offset;
4831 last_size = new_size;
4833 if (!pending_del_nr) {
4834 /* no pending yet, add ourselves */
4835 pending_del_slot = path->slots[0];
4837 } else if (pending_del_nr &&
4838 path->slots[0] + 1 == pending_del_slot) {
4839 /* hop on the pending chunk */
4841 pending_del_slot = path->slots[0];
4848 should_throttle = false;
4851 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4852 struct btrfs_ref ref = { 0 };
4854 bytes_deleted += extent_num_bytes;
4856 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF,
4857 extent_start, extent_num_bytes, 0);
4858 ref.real_root = root->root_key.objectid;
4859 btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
4860 ino, extent_offset);
4861 ret = btrfs_free_extent(trans, &ref);
4863 btrfs_abort_transaction(trans, ret);
4867 if (btrfs_should_throttle_delayed_refs(trans))
4868 should_throttle = true;
4872 if (found_type == BTRFS_INODE_ITEM_KEY)
4875 if (path->slots[0] == 0 ||
4876 path->slots[0] != pending_del_slot ||
4878 if (pending_del_nr) {
4879 ret = btrfs_del_items(trans, root, path,
4883 btrfs_abort_transaction(trans, ret);
4888 btrfs_release_path(path);
4891 * We can generate a lot of delayed refs, so we need to
4892 * throttle every once and a while and make sure we're
4893 * adding enough space to keep up with the work we are
4894 * generating. Since we hold a transaction here we
4895 * can't flush, and we don't want to FLUSH_LIMIT because
4896 * we could have generated too many delayed refs to
4897 * actually allocate, so just bail if we're short and
4898 * let the normal reservation dance happen higher up.
4900 if (should_throttle) {
4901 ret = btrfs_delayed_refs_rsv_refill(fs_info,
4902 BTRFS_RESERVE_NO_FLUSH);
4914 if (ret >= 0 && pending_del_nr) {
4917 err = btrfs_del_items(trans, root, path, pending_del_slot,
4920 btrfs_abort_transaction(trans, err);
4924 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4925 ASSERT(last_size >= new_size);
4926 if (!ret && last_size > new_size)
4927 last_size = new_size;
4928 btrfs_inode_safe_disk_i_size_write(inode, last_size);
4929 unlock_extent_cached(&inode->io_tree, lock_start, (u64)-1,
4933 btrfs_free_path(path);
4938 * btrfs_truncate_block - read, zero a chunk and write a block
4939 * @inode - inode that we're zeroing
4940 * @from - the offset to start zeroing
4941 * @len - the length to zero, 0 to zero the entire range respective to the
4943 * @front - zero up to the offset instead of from the offset on
4945 * This will find the block for the "from" offset and cow the block and zero the
4946 * part we want to zero. This is used with truncate and hole punching.
4948 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4951 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4952 struct address_space *mapping = inode->vfs_inode.i_mapping;
4953 struct extent_io_tree *io_tree = &inode->io_tree;
4954 struct btrfs_ordered_extent *ordered;
4955 struct extent_state *cached_state = NULL;
4956 struct extent_changeset *data_reserved = NULL;
4957 bool only_release_metadata = false;
4958 u32 blocksize = fs_info->sectorsize;
4959 pgoff_t index = from >> PAGE_SHIFT;
4960 unsigned offset = from & (blocksize - 1);
4962 gfp_t mask = btrfs_alloc_write_mask(mapping);
4963 size_t write_bytes = blocksize;
4968 if (IS_ALIGNED(offset, blocksize) &&
4969 (!len || IS_ALIGNED(len, blocksize)))
4972 block_start = round_down(from, blocksize);
4973 block_end = block_start + blocksize - 1;
4975 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4978 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes) > 0) {
4979 /* For nocow case, no need to reserve data space */
4980 only_release_metadata = true;
4985 ret = btrfs_delalloc_reserve_metadata(inode, blocksize);
4987 if (!only_release_metadata)
4988 btrfs_free_reserved_data_space(inode, data_reserved,
4989 block_start, blocksize);
4993 page = find_or_create_page(mapping, index, mask);
4995 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4997 btrfs_delalloc_release_extents(inode, blocksize);
5001 ret = set_page_extent_mapped(page);
5005 if (!PageUptodate(page)) {
5006 ret = btrfs_readpage(NULL, page);
5008 if (page->mapping != mapping) {
5013 if (!PageUptodate(page)) {
5018 wait_on_page_writeback(page);
5020 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
5022 ordered = btrfs_lookup_ordered_extent(inode, block_start);
5024 unlock_extent_cached(io_tree, block_start, block_end,
5028 btrfs_start_ordered_extent(ordered, 1);
5029 btrfs_put_ordered_extent(ordered);
5033 clear_extent_bit(&inode->io_tree, block_start, block_end,
5034 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
5035 0, 0, &cached_state);
5037 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
5040 unlock_extent_cached(io_tree, block_start, block_end,
5045 if (offset != blocksize) {
5047 len = blocksize - offset;
5049 memzero_page(page, (block_start - page_offset(page)),
5052 memzero_page(page, (block_start - page_offset(page)) + offset,
5054 flush_dcache_page(page);
5056 ClearPageChecked(page);
5057 btrfs_page_set_dirty(fs_info, page, block_start, block_end + 1 - block_start);
5058 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
5060 if (only_release_metadata)
5061 set_extent_bit(&inode->io_tree, block_start, block_end,
5062 EXTENT_NORESERVE, 0, NULL, NULL, GFP_NOFS, NULL);
5066 if (only_release_metadata)
5067 btrfs_delalloc_release_metadata(inode, blocksize, true);
5069 btrfs_delalloc_release_space(inode, data_reserved,
5070 block_start, blocksize, true);
5072 btrfs_delalloc_release_extents(inode, blocksize);
5076 if (only_release_metadata)
5077 btrfs_check_nocow_unlock(inode);
5078 extent_changeset_free(data_reserved);
5082 static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode,
5083 u64 offset, u64 len)
5085 struct btrfs_fs_info *fs_info = root->fs_info;
5086 struct btrfs_trans_handle *trans;
5087 struct btrfs_drop_extents_args drop_args = { 0 };
5091 * Still need to make sure the inode looks like it's been updated so
5092 * that any holes get logged if we fsync.
5094 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
5095 inode->last_trans = fs_info->generation;
5096 inode->last_sub_trans = root->log_transid;
5097 inode->last_log_commit = root->last_log_commit;
5102 * 1 - for the one we're dropping
5103 * 1 - for the one we're adding
5104 * 1 - for updating the inode.
5106 trans = btrfs_start_transaction(root, 3);
5108 return PTR_ERR(trans);
5110 drop_args.start = offset;
5111 drop_args.end = offset + len;
5112 drop_args.drop_cache = true;
5114 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
5116 btrfs_abort_transaction(trans, ret);
5117 btrfs_end_transaction(trans);
5121 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode),
5122 offset, 0, 0, len, 0, len, 0, 0, 0);
5124 btrfs_abort_transaction(trans, ret);
5126 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
5127 btrfs_update_inode(trans, root, inode);
5129 btrfs_end_transaction(trans);
5134 * This function puts in dummy file extents for the area we're creating a hole
5135 * for. So if we are truncating this file to a larger size we need to insert
5136 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
5137 * the range between oldsize and size
5139 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
5141 struct btrfs_root *root = inode->root;
5142 struct btrfs_fs_info *fs_info = root->fs_info;
5143 struct extent_io_tree *io_tree = &inode->io_tree;
5144 struct extent_map *em = NULL;
5145 struct extent_state *cached_state = NULL;
5146 struct extent_map_tree *em_tree = &inode->extent_tree;
5147 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
5148 u64 block_end = ALIGN(size, fs_info->sectorsize);
5155 * If our size started in the middle of a block we need to zero out the
5156 * rest of the block before we expand the i_size, otherwise we could
5157 * expose stale data.
5159 err = btrfs_truncate_block(inode, oldsize, 0, 0);
5163 if (size <= hole_start)
5166 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
5168 cur_offset = hole_start;
5170 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
5171 block_end - cur_offset);
5177 last_byte = min(extent_map_end(em), block_end);
5178 last_byte = ALIGN(last_byte, fs_info->sectorsize);
5179 hole_size = last_byte - cur_offset;
5181 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5182 struct extent_map *hole_em;
5184 err = maybe_insert_hole(root, inode, cur_offset,
5189 err = btrfs_inode_set_file_extent_range(inode,
5190 cur_offset, hole_size);
5194 btrfs_drop_extent_cache(inode, cur_offset,
5195 cur_offset + hole_size - 1, 0);
5196 hole_em = alloc_extent_map();
5198 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5199 &inode->runtime_flags);
5202 hole_em->start = cur_offset;
5203 hole_em->len = hole_size;
5204 hole_em->orig_start = cur_offset;
5206 hole_em->block_start = EXTENT_MAP_HOLE;
5207 hole_em->block_len = 0;
5208 hole_em->orig_block_len = 0;
5209 hole_em->ram_bytes = hole_size;
5210 hole_em->compress_type = BTRFS_COMPRESS_NONE;
5211 hole_em->generation = fs_info->generation;
5214 write_lock(&em_tree->lock);
5215 err = add_extent_mapping(em_tree, hole_em, 1);
5216 write_unlock(&em_tree->lock);
5219 btrfs_drop_extent_cache(inode, cur_offset,
5223 free_extent_map(hole_em);
5225 err = btrfs_inode_set_file_extent_range(inode,
5226 cur_offset, hole_size);
5231 free_extent_map(em);
5233 cur_offset = last_byte;
5234 if (cur_offset >= block_end)
5237 free_extent_map(em);
5238 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
5242 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5244 struct btrfs_root *root = BTRFS_I(inode)->root;
5245 struct btrfs_trans_handle *trans;
5246 loff_t oldsize = i_size_read(inode);
5247 loff_t newsize = attr->ia_size;
5248 int mask = attr->ia_valid;
5252 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5253 * special case where we need to update the times despite not having
5254 * these flags set. For all other operations the VFS set these flags
5255 * explicitly if it wants a timestamp update.
5257 if (newsize != oldsize) {
5258 inode_inc_iversion(inode);
5259 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5260 inode->i_ctime = inode->i_mtime =
5261 current_time(inode);
5264 if (newsize > oldsize) {
5266 * Don't do an expanding truncate while snapshotting is ongoing.
5267 * This is to ensure the snapshot captures a fully consistent
5268 * state of this file - if the snapshot captures this expanding
5269 * truncation, it must capture all writes that happened before
5272 btrfs_drew_write_lock(&root->snapshot_lock);
5273 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
5275 btrfs_drew_write_unlock(&root->snapshot_lock);
5279 trans = btrfs_start_transaction(root, 1);
5280 if (IS_ERR(trans)) {
5281 btrfs_drew_write_unlock(&root->snapshot_lock);
5282 return PTR_ERR(trans);
5285 i_size_write(inode, newsize);
5286 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
5287 pagecache_isize_extended(inode, oldsize, newsize);
5288 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5289 btrfs_drew_write_unlock(&root->snapshot_lock);
5290 btrfs_end_transaction(trans);
5292 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5294 if (btrfs_is_zoned(fs_info)) {
5295 ret = btrfs_wait_ordered_range(inode,
5296 ALIGN(newsize, fs_info->sectorsize),
5303 * We're truncating a file that used to have good data down to
5304 * zero. Make sure any new writes to the file get on disk
5308 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5309 &BTRFS_I(inode)->runtime_flags);
5311 truncate_setsize(inode, newsize);
5313 inode_dio_wait(inode);
5315 ret = btrfs_truncate(inode, newsize == oldsize);
5316 if (ret && inode->i_nlink) {
5320 * Truncate failed, so fix up the in-memory size. We
5321 * adjusted disk_i_size down as we removed extents, so
5322 * wait for disk_i_size to be stable and then update the
5323 * in-memory size to match.
5325 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5328 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5335 static int btrfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
5338 struct inode *inode = d_inode(dentry);
5339 struct btrfs_root *root = BTRFS_I(inode)->root;
5342 if (btrfs_root_readonly(root))
5345 err = setattr_prepare(&init_user_ns, dentry, attr);
5349 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5350 err = btrfs_setsize(inode, attr);
5355 if (attr->ia_valid) {
5356 setattr_copy(&init_user_ns, inode, attr);
5357 inode_inc_iversion(inode);
5358 err = btrfs_dirty_inode(inode);
5360 if (!err && attr->ia_valid & ATTR_MODE)
5361 err = posix_acl_chmod(&init_user_ns, inode,
5369 * While truncating the inode pages during eviction, we get the VFS calling
5370 * btrfs_invalidatepage() against each page of the inode. This is slow because
5371 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5372 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5373 * extent_state structures over and over, wasting lots of time.
5375 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5376 * those expensive operations on a per page basis and do only the ordered io
5377 * finishing, while we release here the extent_map and extent_state structures,
5378 * without the excessive merging and splitting.
5380 static void evict_inode_truncate_pages(struct inode *inode)
5382 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5383 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5384 struct rb_node *node;
5386 ASSERT(inode->i_state & I_FREEING);
5387 truncate_inode_pages_final(&inode->i_data);
5389 write_lock(&map_tree->lock);
5390 while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) {
5391 struct extent_map *em;
5393 node = rb_first_cached(&map_tree->map);
5394 em = rb_entry(node, struct extent_map, rb_node);
5395 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5396 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5397 remove_extent_mapping(map_tree, em);
5398 free_extent_map(em);
5399 if (need_resched()) {
5400 write_unlock(&map_tree->lock);
5402 write_lock(&map_tree->lock);
5405 write_unlock(&map_tree->lock);
5408 * Keep looping until we have no more ranges in the io tree.
5409 * We can have ongoing bios started by readahead that have
5410 * their endio callback (extent_io.c:end_bio_extent_readpage)
5411 * still in progress (unlocked the pages in the bio but did not yet
5412 * unlocked the ranges in the io tree). Therefore this means some
5413 * ranges can still be locked and eviction started because before
5414 * submitting those bios, which are executed by a separate task (work
5415 * queue kthread), inode references (inode->i_count) were not taken
5416 * (which would be dropped in the end io callback of each bio).
5417 * Therefore here we effectively end up waiting for those bios and
5418 * anyone else holding locked ranges without having bumped the inode's
5419 * reference count - if we don't do it, when they access the inode's
5420 * io_tree to unlock a range it may be too late, leading to an
5421 * use-after-free issue.
5423 spin_lock(&io_tree->lock);
5424 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5425 struct extent_state *state;
5426 struct extent_state *cached_state = NULL;
5429 unsigned state_flags;
5431 node = rb_first(&io_tree->state);
5432 state = rb_entry(node, struct extent_state, rb_node);
5433 start = state->start;
5435 state_flags = state->state;
5436 spin_unlock(&io_tree->lock);
5438 lock_extent_bits(io_tree, start, end, &cached_state);
5441 * If still has DELALLOC flag, the extent didn't reach disk,
5442 * and its reserved space won't be freed by delayed_ref.
5443 * So we need to free its reserved space here.
5444 * (Refer to comment in btrfs_invalidatepage, case 2)
5446 * Note, end is the bytenr of last byte, so we need + 1 here.
5448 if (state_flags & EXTENT_DELALLOC)
5449 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5452 clear_extent_bit(io_tree, start, end,
5453 EXTENT_LOCKED | EXTENT_DELALLOC |
5454 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
5458 spin_lock(&io_tree->lock);
5460 spin_unlock(&io_tree->lock);
5463 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5464 struct btrfs_block_rsv *rsv)
5466 struct btrfs_fs_info *fs_info = root->fs_info;
5467 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5468 struct btrfs_trans_handle *trans;
5469 u64 delayed_refs_extra = btrfs_calc_insert_metadata_size(fs_info, 1);
5473 * Eviction should be taking place at some place safe because of our
5474 * delayed iputs. However the normal flushing code will run delayed
5475 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5477 * We reserve the delayed_refs_extra here again because we can't use
5478 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5479 * above. We reserve our extra bit here because we generate a ton of
5480 * delayed refs activity by truncating.
5482 * If we cannot make our reservation we'll attempt to steal from the
5483 * global reserve, because we really want to be able to free up space.
5485 ret = btrfs_block_rsv_refill(root, rsv, rsv->size + delayed_refs_extra,
5486 BTRFS_RESERVE_FLUSH_EVICT);
5489 * Try to steal from the global reserve if there is space for
5492 if (btrfs_check_space_for_delayed_refs(fs_info) ||
5493 btrfs_block_rsv_migrate(global_rsv, rsv, rsv->size, 0)) {
5495 "could not allocate space for delete; will truncate on mount");
5496 return ERR_PTR(-ENOSPC);
5498 delayed_refs_extra = 0;
5501 trans = btrfs_join_transaction(root);
5505 if (delayed_refs_extra) {
5506 trans->block_rsv = &fs_info->trans_block_rsv;
5507 trans->bytes_reserved = delayed_refs_extra;
5508 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5509 delayed_refs_extra, 1);
5514 void btrfs_evict_inode(struct inode *inode)
5516 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5517 struct btrfs_trans_handle *trans;
5518 struct btrfs_root *root = BTRFS_I(inode)->root;
5519 struct btrfs_block_rsv *rsv;
5522 trace_btrfs_inode_evict(inode);
5529 evict_inode_truncate_pages(inode);
5531 if (inode->i_nlink &&
5532 ((btrfs_root_refs(&root->root_item) != 0 &&
5533 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5534 btrfs_is_free_space_inode(BTRFS_I(inode))))
5537 if (is_bad_inode(inode))
5540 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5542 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5545 if (inode->i_nlink > 0) {
5546 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5547 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5551 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5555 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5558 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5561 btrfs_i_size_write(BTRFS_I(inode), 0);
5564 trans = evict_refill_and_join(root, rsv);
5568 trans->block_rsv = rsv;
5570 ret = btrfs_truncate_inode_items(trans, root, BTRFS_I(inode),
5572 trans->block_rsv = &fs_info->trans_block_rsv;
5573 btrfs_end_transaction(trans);
5574 btrfs_btree_balance_dirty(fs_info);
5575 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5582 * Errors here aren't a big deal, it just means we leave orphan items in
5583 * the tree. They will be cleaned up on the next mount. If the inode
5584 * number gets reused, cleanup deletes the orphan item without doing
5585 * anything, and unlink reuses the existing orphan item.
5587 * If it turns out that we are dropping too many of these, we might want
5588 * to add a mechanism for retrying these after a commit.
5590 trans = evict_refill_and_join(root, rsv);
5591 if (!IS_ERR(trans)) {
5592 trans->block_rsv = rsv;
5593 btrfs_orphan_del(trans, BTRFS_I(inode));
5594 trans->block_rsv = &fs_info->trans_block_rsv;
5595 btrfs_end_transaction(trans);
5599 btrfs_free_block_rsv(fs_info, rsv);
5602 * If we didn't successfully delete, the orphan item will still be in
5603 * the tree and we'll retry on the next mount. Again, we might also want
5604 * to retry these periodically in the future.
5606 btrfs_remove_delayed_node(BTRFS_I(inode));
5611 * Return the key found in the dir entry in the location pointer, fill @type
5612 * with BTRFS_FT_*, and return 0.
5614 * If no dir entries were found, returns -ENOENT.
5615 * If found a corrupted location in dir entry, returns -EUCLEAN.
5617 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5618 struct btrfs_key *location, u8 *type)
5620 const char *name = dentry->d_name.name;
5621 int namelen = dentry->d_name.len;
5622 struct btrfs_dir_item *di;
5623 struct btrfs_path *path;
5624 struct btrfs_root *root = BTRFS_I(dir)->root;
5627 path = btrfs_alloc_path();
5631 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5633 if (IS_ERR_OR_NULL(di)) {
5634 ret = di ? PTR_ERR(di) : -ENOENT;
5638 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5639 if (location->type != BTRFS_INODE_ITEM_KEY &&
5640 location->type != BTRFS_ROOT_ITEM_KEY) {
5642 btrfs_warn(root->fs_info,
5643 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5644 __func__, name, btrfs_ino(BTRFS_I(dir)),
5645 location->objectid, location->type, location->offset);
5648 *type = btrfs_dir_type(path->nodes[0], di);
5650 btrfs_free_path(path);
5655 * when we hit a tree root in a directory, the btrfs part of the inode
5656 * needs to be changed to reflect the root directory of the tree root. This
5657 * is kind of like crossing a mount point.
5659 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5661 struct dentry *dentry,
5662 struct btrfs_key *location,
5663 struct btrfs_root **sub_root)
5665 struct btrfs_path *path;
5666 struct btrfs_root *new_root;
5667 struct btrfs_root_ref *ref;
5668 struct extent_buffer *leaf;
5669 struct btrfs_key key;
5673 path = btrfs_alloc_path();
5680 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5681 key.type = BTRFS_ROOT_REF_KEY;
5682 key.offset = location->objectid;
5684 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5691 leaf = path->nodes[0];
5692 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5693 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5694 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5697 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5698 (unsigned long)(ref + 1),
5699 dentry->d_name.len);
5703 btrfs_release_path(path);
5705 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5706 if (IS_ERR(new_root)) {
5707 err = PTR_ERR(new_root);
5711 *sub_root = new_root;
5712 location->objectid = btrfs_root_dirid(&new_root->root_item);
5713 location->type = BTRFS_INODE_ITEM_KEY;
5714 location->offset = 0;
5717 btrfs_free_path(path);
5721 static void inode_tree_add(struct inode *inode)
5723 struct btrfs_root *root = BTRFS_I(inode)->root;
5724 struct btrfs_inode *entry;
5726 struct rb_node *parent;
5727 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5728 u64 ino = btrfs_ino(BTRFS_I(inode));
5730 if (inode_unhashed(inode))
5733 spin_lock(&root->inode_lock);
5734 p = &root->inode_tree.rb_node;
5737 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5739 if (ino < btrfs_ino(entry))
5740 p = &parent->rb_left;
5741 else if (ino > btrfs_ino(entry))
5742 p = &parent->rb_right;
5744 WARN_ON(!(entry->vfs_inode.i_state &
5745 (I_WILL_FREE | I_FREEING)));
5746 rb_replace_node(parent, new, &root->inode_tree);
5747 RB_CLEAR_NODE(parent);
5748 spin_unlock(&root->inode_lock);
5752 rb_link_node(new, parent, p);
5753 rb_insert_color(new, &root->inode_tree);
5754 spin_unlock(&root->inode_lock);
5757 static void inode_tree_del(struct btrfs_inode *inode)
5759 struct btrfs_root *root = inode->root;
5762 spin_lock(&root->inode_lock);
5763 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5764 rb_erase(&inode->rb_node, &root->inode_tree);
5765 RB_CLEAR_NODE(&inode->rb_node);
5766 empty = RB_EMPTY_ROOT(&root->inode_tree);
5768 spin_unlock(&root->inode_lock);
5770 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5771 spin_lock(&root->inode_lock);
5772 empty = RB_EMPTY_ROOT(&root->inode_tree);
5773 spin_unlock(&root->inode_lock);
5775 btrfs_add_dead_root(root);
5780 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5782 struct btrfs_iget_args *args = p;
5784 inode->i_ino = args->ino;
5785 BTRFS_I(inode)->location.objectid = args->ino;
5786 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5787 BTRFS_I(inode)->location.offset = 0;
5788 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5789 BUG_ON(args->root && !BTRFS_I(inode)->root);
5793 static int btrfs_find_actor(struct inode *inode, void *opaque)
5795 struct btrfs_iget_args *args = opaque;
5797 return args->ino == BTRFS_I(inode)->location.objectid &&
5798 args->root == BTRFS_I(inode)->root;
5801 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5802 struct btrfs_root *root)
5804 struct inode *inode;
5805 struct btrfs_iget_args args;
5806 unsigned long hashval = btrfs_inode_hash(ino, root);
5811 inode = iget5_locked(s, hashval, btrfs_find_actor,
5812 btrfs_init_locked_inode,
5818 * Get an inode object given its inode number and corresponding root.
5819 * Path can be preallocated to prevent recursing back to iget through
5820 * allocator. NULL is also valid but may require an additional allocation
5823 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5824 struct btrfs_root *root, struct btrfs_path *path)
5826 struct inode *inode;
5828 inode = btrfs_iget_locked(s, ino, root);
5830 return ERR_PTR(-ENOMEM);
5832 if (inode->i_state & I_NEW) {
5835 ret = btrfs_read_locked_inode(inode, path);
5837 inode_tree_add(inode);
5838 unlock_new_inode(inode);
5842 * ret > 0 can come from btrfs_search_slot called by
5843 * btrfs_read_locked_inode, this means the inode item
5848 inode = ERR_PTR(ret);
5855 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5857 return btrfs_iget_path(s, ino, root, NULL);
5860 static struct inode *new_simple_dir(struct super_block *s,
5861 struct btrfs_key *key,
5862 struct btrfs_root *root)
5864 struct inode *inode = new_inode(s);
5867 return ERR_PTR(-ENOMEM);
5869 BTRFS_I(inode)->root = btrfs_grab_root(root);
5870 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5871 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5873 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5875 * We only need lookup, the rest is read-only and there's no inode
5876 * associated with the dentry
5878 inode->i_op = &simple_dir_inode_operations;
5879 inode->i_opflags &= ~IOP_XATTR;
5880 inode->i_fop = &simple_dir_operations;
5881 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5882 inode->i_mtime = current_time(inode);
5883 inode->i_atime = inode->i_mtime;
5884 inode->i_ctime = inode->i_mtime;
5885 BTRFS_I(inode)->i_otime = inode->i_mtime;
5890 static inline u8 btrfs_inode_type(struct inode *inode)
5893 * Compile-time asserts that generic FT_* types still match
5896 BUILD_BUG_ON(BTRFS_FT_UNKNOWN != FT_UNKNOWN);
5897 BUILD_BUG_ON(BTRFS_FT_REG_FILE != FT_REG_FILE);
5898 BUILD_BUG_ON(BTRFS_FT_DIR != FT_DIR);
5899 BUILD_BUG_ON(BTRFS_FT_CHRDEV != FT_CHRDEV);
5900 BUILD_BUG_ON(BTRFS_FT_BLKDEV != FT_BLKDEV);
5901 BUILD_BUG_ON(BTRFS_FT_FIFO != FT_FIFO);
5902 BUILD_BUG_ON(BTRFS_FT_SOCK != FT_SOCK);
5903 BUILD_BUG_ON(BTRFS_FT_SYMLINK != FT_SYMLINK);
5905 return fs_umode_to_ftype(inode->i_mode);
5908 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5910 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5911 struct inode *inode;
5912 struct btrfs_root *root = BTRFS_I(dir)->root;
5913 struct btrfs_root *sub_root = root;
5914 struct btrfs_key location;
5918 if (dentry->d_name.len > BTRFS_NAME_LEN)
5919 return ERR_PTR(-ENAMETOOLONG);
5921 ret = btrfs_inode_by_name(dir, dentry, &location, &di_type);
5923 return ERR_PTR(ret);
5925 if (location.type == BTRFS_INODE_ITEM_KEY) {
5926 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5930 /* Do extra check against inode mode with di_type */
5931 if (btrfs_inode_type(inode) != di_type) {
5933 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5934 inode->i_mode, btrfs_inode_type(inode),
5937 return ERR_PTR(-EUCLEAN);
5942 ret = fixup_tree_root_location(fs_info, dir, dentry,
5943 &location, &sub_root);
5946 inode = ERR_PTR(ret);
5948 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5950 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5952 if (root != sub_root)
5953 btrfs_put_root(sub_root);
5955 if (!IS_ERR(inode) && root != sub_root) {
5956 down_read(&fs_info->cleanup_work_sem);
5957 if (!sb_rdonly(inode->i_sb))
5958 ret = btrfs_orphan_cleanup(sub_root);
5959 up_read(&fs_info->cleanup_work_sem);
5962 inode = ERR_PTR(ret);
5969 static int btrfs_dentry_delete(const struct dentry *dentry)
5971 struct btrfs_root *root;
5972 struct inode *inode = d_inode(dentry);
5974 if (!inode && !IS_ROOT(dentry))
5975 inode = d_inode(dentry->d_parent);
5978 root = BTRFS_I(inode)->root;
5979 if (btrfs_root_refs(&root->root_item) == 0)
5982 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5988 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5991 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5993 if (inode == ERR_PTR(-ENOENT))
5995 return d_splice_alias(inode, dentry);
5999 * All this infrastructure exists because dir_emit can fault, and we are holding
6000 * the tree lock when doing readdir. For now just allocate a buffer and copy
6001 * our information into that, and then dir_emit from the buffer. This is
6002 * similar to what NFS does, only we don't keep the buffer around in pagecache
6003 * because I'm afraid I'll mess that up. Long term we need to make filldir do
6004 * copy_to_user_inatomic so we don't have to worry about page faulting under the
6007 static int btrfs_opendir(struct inode *inode, struct file *file)
6009 struct btrfs_file_private *private;
6011 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
6014 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
6015 if (!private->filldir_buf) {
6019 file->private_data = private;
6030 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
6033 struct dir_entry *entry = addr;
6034 char *name = (char *)(entry + 1);
6036 ctx->pos = get_unaligned(&entry->offset);
6037 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
6038 get_unaligned(&entry->ino),
6039 get_unaligned(&entry->type)))
6041 addr += sizeof(struct dir_entry) +
6042 get_unaligned(&entry->name_len);
6048 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
6050 struct inode *inode = file_inode(file);
6051 struct btrfs_root *root = BTRFS_I(inode)->root;
6052 struct btrfs_file_private *private = file->private_data;
6053 struct btrfs_dir_item *di;
6054 struct btrfs_key key;
6055 struct btrfs_key found_key;
6056 struct btrfs_path *path;
6058 struct list_head ins_list;
6059 struct list_head del_list;
6061 struct extent_buffer *leaf;
6068 struct btrfs_key location;
6070 if (!dir_emit_dots(file, ctx))
6073 path = btrfs_alloc_path();
6077 addr = private->filldir_buf;
6078 path->reada = READA_FORWARD;
6080 INIT_LIST_HEAD(&ins_list);
6081 INIT_LIST_HEAD(&del_list);
6082 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
6085 key.type = BTRFS_DIR_INDEX_KEY;
6086 key.offset = ctx->pos;
6087 key.objectid = btrfs_ino(BTRFS_I(inode));
6089 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6094 struct dir_entry *entry;
6096 leaf = path->nodes[0];
6097 slot = path->slots[0];
6098 if (slot >= btrfs_header_nritems(leaf)) {
6099 ret = btrfs_next_leaf(root, path);
6107 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6109 if (found_key.objectid != key.objectid)
6111 if (found_key.type != BTRFS_DIR_INDEX_KEY)
6113 if (found_key.offset < ctx->pos)
6115 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
6117 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
6118 name_len = btrfs_dir_name_len(leaf, di);
6119 if ((total_len + sizeof(struct dir_entry) + name_len) >=
6121 btrfs_release_path(path);
6122 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6125 addr = private->filldir_buf;
6132 put_unaligned(name_len, &entry->name_len);
6133 name_ptr = (char *)(entry + 1);
6134 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
6136 put_unaligned(fs_ftype_to_dtype(btrfs_dir_type(leaf, di)),
6138 btrfs_dir_item_key_to_cpu(leaf, di, &location);
6139 put_unaligned(location.objectid, &entry->ino);
6140 put_unaligned(found_key.offset, &entry->offset);
6142 addr += sizeof(struct dir_entry) + name_len;
6143 total_len += sizeof(struct dir_entry) + name_len;
6147 btrfs_release_path(path);
6149 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6153 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
6158 * Stop new entries from being returned after we return the last
6161 * New directory entries are assigned a strictly increasing
6162 * offset. This means that new entries created during readdir
6163 * are *guaranteed* to be seen in the future by that readdir.
6164 * This has broken buggy programs which operate on names as
6165 * they're returned by readdir. Until we re-use freed offsets
6166 * we have this hack to stop new entries from being returned
6167 * under the assumption that they'll never reach this huge
6170 * This is being careful not to overflow 32bit loff_t unless the
6171 * last entry requires it because doing so has broken 32bit apps
6174 if (ctx->pos >= INT_MAX)
6175 ctx->pos = LLONG_MAX;
6182 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6183 btrfs_free_path(path);
6188 * This is somewhat expensive, updating the tree every time the
6189 * inode changes. But, it is most likely to find the inode in cache.
6190 * FIXME, needs more benchmarking...there are no reasons other than performance
6191 * to keep or drop this code.
6193 static int btrfs_dirty_inode(struct inode *inode)
6195 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6196 struct btrfs_root *root = BTRFS_I(inode)->root;
6197 struct btrfs_trans_handle *trans;
6200 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6203 trans = btrfs_join_transaction(root);
6205 return PTR_ERR(trans);
6207 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
6208 if (ret && (ret == -ENOSPC || ret == -EDQUOT)) {
6209 /* whoops, lets try again with the full transaction */
6210 btrfs_end_transaction(trans);
6211 trans = btrfs_start_transaction(root, 1);
6213 return PTR_ERR(trans);
6215 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
6217 btrfs_end_transaction(trans);
6218 if (BTRFS_I(inode)->delayed_node)
6219 btrfs_balance_delayed_items(fs_info);
6225 * This is a copy of file_update_time. We need this so we can return error on
6226 * ENOSPC for updating the inode in the case of file write and mmap writes.
6228 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
6231 struct btrfs_root *root = BTRFS_I(inode)->root;
6232 bool dirty = flags & ~S_VERSION;
6234 if (btrfs_root_readonly(root))
6237 if (flags & S_VERSION)
6238 dirty |= inode_maybe_inc_iversion(inode, dirty);
6239 if (flags & S_CTIME)
6240 inode->i_ctime = *now;
6241 if (flags & S_MTIME)
6242 inode->i_mtime = *now;
6243 if (flags & S_ATIME)
6244 inode->i_atime = *now;
6245 return dirty ? btrfs_dirty_inode(inode) : 0;
6249 * find the highest existing sequence number in a directory
6250 * and then set the in-memory index_cnt variable to reflect
6251 * free sequence numbers
6253 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6255 struct btrfs_root *root = inode->root;
6256 struct btrfs_key key, found_key;
6257 struct btrfs_path *path;
6258 struct extent_buffer *leaf;
6261 key.objectid = btrfs_ino(inode);
6262 key.type = BTRFS_DIR_INDEX_KEY;
6263 key.offset = (u64)-1;
6265 path = btrfs_alloc_path();
6269 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6272 /* FIXME: we should be able to handle this */
6278 * MAGIC NUMBER EXPLANATION:
6279 * since we search a directory based on f_pos we have to start at 2
6280 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6281 * else has to start at 2
6283 if (path->slots[0] == 0) {
6284 inode->index_cnt = 2;
6290 leaf = path->nodes[0];
6291 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6293 if (found_key.objectid != btrfs_ino(inode) ||
6294 found_key.type != BTRFS_DIR_INDEX_KEY) {
6295 inode->index_cnt = 2;
6299 inode->index_cnt = found_key.offset + 1;
6301 btrfs_free_path(path);
6306 * helper to find a free sequence number in a given directory. This current
6307 * code is very simple, later versions will do smarter things in the btree
6309 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6313 if (dir->index_cnt == (u64)-1) {
6314 ret = btrfs_inode_delayed_dir_index_count(dir);
6316 ret = btrfs_set_inode_index_count(dir);
6322 *index = dir->index_cnt;
6328 static int btrfs_insert_inode_locked(struct inode *inode)
6330 struct btrfs_iget_args args;
6332 args.ino = BTRFS_I(inode)->location.objectid;
6333 args.root = BTRFS_I(inode)->root;
6335 return insert_inode_locked4(inode,
6336 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6337 btrfs_find_actor, &args);
6341 * Inherit flags from the parent inode.
6343 * Currently only the compression flags and the cow flags are inherited.
6345 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6352 flags = BTRFS_I(dir)->flags;
6354 if (flags & BTRFS_INODE_NOCOMPRESS) {
6355 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6356 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6357 } else if (flags & BTRFS_INODE_COMPRESS) {
6358 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6359 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6362 if (flags & BTRFS_INODE_NODATACOW) {
6363 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6364 if (S_ISREG(inode->i_mode))
6365 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6368 btrfs_sync_inode_flags_to_i_flags(inode);
6371 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6372 struct btrfs_root *root,
6374 const char *name, int name_len,
6375 u64 ref_objectid, u64 objectid,
6376 umode_t mode, u64 *index)
6378 struct btrfs_fs_info *fs_info = root->fs_info;
6379 struct inode *inode;
6380 struct btrfs_inode_item *inode_item;
6381 struct btrfs_key *location;
6382 struct btrfs_path *path;
6383 struct btrfs_inode_ref *ref;
6384 struct btrfs_key key[2];
6386 int nitems = name ? 2 : 1;
6388 unsigned int nofs_flag;
6391 path = btrfs_alloc_path();
6393 return ERR_PTR(-ENOMEM);
6395 nofs_flag = memalloc_nofs_save();
6396 inode = new_inode(fs_info->sb);
6397 memalloc_nofs_restore(nofs_flag);
6399 btrfs_free_path(path);
6400 return ERR_PTR(-ENOMEM);
6404 * O_TMPFILE, set link count to 0, so that after this point,
6405 * we fill in an inode item with the correct link count.
6408 set_nlink(inode, 0);
6411 * we have to initialize this early, so we can reclaim the inode
6412 * number if we fail afterwards in this function.
6414 inode->i_ino = objectid;
6417 trace_btrfs_inode_request(dir);
6419 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6421 btrfs_free_path(path);
6423 return ERR_PTR(ret);
6429 * index_cnt is ignored for everything but a dir,
6430 * btrfs_set_inode_index_count has an explanation for the magic
6433 BTRFS_I(inode)->index_cnt = 2;
6434 BTRFS_I(inode)->dir_index = *index;
6435 BTRFS_I(inode)->root = btrfs_grab_root(root);
6436 BTRFS_I(inode)->generation = trans->transid;
6437 inode->i_generation = BTRFS_I(inode)->generation;
6440 * We could have gotten an inode number from somebody who was fsynced
6441 * and then removed in this same transaction, so let's just set full
6442 * sync since it will be a full sync anyway and this will blow away the
6443 * old info in the log.
6445 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6447 key[0].objectid = objectid;
6448 key[0].type = BTRFS_INODE_ITEM_KEY;
6451 sizes[0] = sizeof(struct btrfs_inode_item);
6455 * Start new inodes with an inode_ref. This is slightly more
6456 * efficient for small numbers of hard links since they will
6457 * be packed into one item. Extended refs will kick in if we
6458 * add more hard links than can fit in the ref item.
6460 key[1].objectid = objectid;
6461 key[1].type = BTRFS_INODE_REF_KEY;
6462 key[1].offset = ref_objectid;
6464 sizes[1] = name_len + sizeof(*ref);
6467 location = &BTRFS_I(inode)->location;
6468 location->objectid = objectid;
6469 location->offset = 0;
6470 location->type = BTRFS_INODE_ITEM_KEY;
6472 ret = btrfs_insert_inode_locked(inode);
6478 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6482 inode_init_owner(&init_user_ns, inode, dir, mode);
6483 inode_set_bytes(inode, 0);
6485 inode->i_mtime = current_time(inode);
6486 inode->i_atime = inode->i_mtime;
6487 inode->i_ctime = inode->i_mtime;
6488 BTRFS_I(inode)->i_otime = inode->i_mtime;
6490 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6491 struct btrfs_inode_item);
6492 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6493 sizeof(*inode_item));
6494 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6497 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6498 struct btrfs_inode_ref);
6499 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6500 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6501 ptr = (unsigned long)(ref + 1);
6502 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6505 btrfs_mark_buffer_dirty(path->nodes[0]);
6506 btrfs_free_path(path);
6508 btrfs_inherit_iflags(inode, dir);
6510 if (S_ISREG(mode)) {
6511 if (btrfs_test_opt(fs_info, NODATASUM))
6512 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6513 if (btrfs_test_opt(fs_info, NODATACOW))
6514 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6515 BTRFS_INODE_NODATASUM;
6518 inode_tree_add(inode);
6520 trace_btrfs_inode_new(inode);
6521 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6523 btrfs_update_root_times(trans, root);
6525 ret = btrfs_inode_inherit_props(trans, inode, dir);
6528 "error inheriting props for ino %llu (root %llu): %d",
6529 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6534 discard_new_inode(inode);
6537 BTRFS_I(dir)->index_cnt--;
6538 btrfs_free_path(path);
6539 return ERR_PTR(ret);
6543 * utility function to add 'inode' into 'parent_inode' with
6544 * a give name and a given sequence number.
6545 * if 'add_backref' is true, also insert a backref from the
6546 * inode to the parent directory.
6548 int btrfs_add_link(struct btrfs_trans_handle *trans,
6549 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6550 const char *name, int name_len, int add_backref, u64 index)
6553 struct btrfs_key key;
6554 struct btrfs_root *root = parent_inode->root;
6555 u64 ino = btrfs_ino(inode);
6556 u64 parent_ino = btrfs_ino(parent_inode);
6558 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6559 memcpy(&key, &inode->root->root_key, sizeof(key));
6562 key.type = BTRFS_INODE_ITEM_KEY;
6566 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6567 ret = btrfs_add_root_ref(trans, key.objectid,
6568 root->root_key.objectid, parent_ino,
6569 index, name, name_len);
6570 } else if (add_backref) {
6571 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6575 /* Nothing to clean up yet */
6579 ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key,
6580 btrfs_inode_type(&inode->vfs_inode), index);
6581 if (ret == -EEXIST || ret == -EOVERFLOW)
6584 btrfs_abort_transaction(trans, ret);
6588 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6590 inode_inc_iversion(&parent_inode->vfs_inode);
6592 * If we are replaying a log tree, we do not want to update the mtime
6593 * and ctime of the parent directory with the current time, since the
6594 * log replay procedure is responsible for setting them to their correct
6595 * values (the ones it had when the fsync was done).
6597 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
6598 struct timespec64 now = current_time(&parent_inode->vfs_inode);
6600 parent_inode->vfs_inode.i_mtime = now;
6601 parent_inode->vfs_inode.i_ctime = now;
6603 ret = btrfs_update_inode(trans, root, parent_inode);
6605 btrfs_abort_transaction(trans, ret);
6609 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6612 err = btrfs_del_root_ref(trans, key.objectid,
6613 root->root_key.objectid, parent_ino,
6614 &local_index, name, name_len);
6616 btrfs_abort_transaction(trans, err);
6617 } else if (add_backref) {
6621 err = btrfs_del_inode_ref(trans, root, name, name_len,
6622 ino, parent_ino, &local_index);
6624 btrfs_abort_transaction(trans, err);
6627 /* Return the original error code */
6631 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6632 struct btrfs_inode *dir, struct dentry *dentry,
6633 struct btrfs_inode *inode, int backref, u64 index)
6635 int err = btrfs_add_link(trans, dir, inode,
6636 dentry->d_name.name, dentry->d_name.len,
6643 static int btrfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
6644 struct dentry *dentry, umode_t mode, dev_t rdev)
6646 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6647 struct btrfs_trans_handle *trans;
6648 struct btrfs_root *root = BTRFS_I(dir)->root;
6649 struct inode *inode = NULL;
6655 * 2 for inode item and ref
6657 * 1 for xattr if selinux is on
6659 trans = btrfs_start_transaction(root, 5);
6661 return PTR_ERR(trans);
6663 err = btrfs_get_free_objectid(root, &objectid);
6667 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6668 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6670 if (IS_ERR(inode)) {
6671 err = PTR_ERR(inode);
6677 * If the active LSM wants to access the inode during
6678 * d_instantiate it needs these. Smack checks to see
6679 * if the filesystem supports xattrs by looking at the
6682 inode->i_op = &btrfs_special_inode_operations;
6683 init_special_inode(inode, inode->i_mode, rdev);
6685 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6689 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6694 btrfs_update_inode(trans, root, BTRFS_I(inode));
6695 d_instantiate_new(dentry, inode);
6698 btrfs_end_transaction(trans);
6699 btrfs_btree_balance_dirty(fs_info);
6701 inode_dec_link_count(inode);
6702 discard_new_inode(inode);
6707 static int btrfs_create(struct user_namespace *mnt_userns, struct inode *dir,
6708 struct dentry *dentry, umode_t mode, bool excl)
6710 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6711 struct btrfs_trans_handle *trans;
6712 struct btrfs_root *root = BTRFS_I(dir)->root;
6713 struct inode *inode = NULL;
6719 * 2 for inode item and ref
6721 * 1 for xattr if selinux is on
6723 trans = btrfs_start_transaction(root, 5);
6725 return PTR_ERR(trans);
6727 err = btrfs_get_free_objectid(root, &objectid);
6731 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6732 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6734 if (IS_ERR(inode)) {
6735 err = PTR_ERR(inode);
6740 * If the active LSM wants to access the inode during
6741 * d_instantiate it needs these. Smack checks to see
6742 * if the filesystem supports xattrs by looking at the
6745 inode->i_fop = &btrfs_file_operations;
6746 inode->i_op = &btrfs_file_inode_operations;
6747 inode->i_mapping->a_ops = &btrfs_aops;
6749 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6753 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6757 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6762 d_instantiate_new(dentry, inode);
6765 btrfs_end_transaction(trans);
6767 inode_dec_link_count(inode);
6768 discard_new_inode(inode);
6770 btrfs_btree_balance_dirty(fs_info);
6774 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6775 struct dentry *dentry)
6777 struct btrfs_trans_handle *trans = NULL;
6778 struct btrfs_root *root = BTRFS_I(dir)->root;
6779 struct inode *inode = d_inode(old_dentry);
6780 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6785 /* do not allow sys_link's with other subvols of the same device */
6786 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6789 if (inode->i_nlink >= BTRFS_LINK_MAX)
6792 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6797 * 2 items for inode and inode ref
6798 * 2 items for dir items
6799 * 1 item for parent inode
6800 * 1 item for orphan item deletion if O_TMPFILE
6802 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6803 if (IS_ERR(trans)) {
6804 err = PTR_ERR(trans);
6809 /* There are several dir indexes for this inode, clear the cache. */
6810 BTRFS_I(inode)->dir_index = 0ULL;
6812 inode_inc_iversion(inode);
6813 inode->i_ctime = current_time(inode);
6815 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6817 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6823 struct dentry *parent = dentry->d_parent;
6825 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6828 if (inode->i_nlink == 1) {
6830 * If new hard link count is 1, it's a file created
6831 * with open(2) O_TMPFILE flag.
6833 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6837 d_instantiate(dentry, inode);
6838 btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent);
6843 btrfs_end_transaction(trans);
6845 inode_dec_link_count(inode);
6848 btrfs_btree_balance_dirty(fs_info);
6852 static int btrfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
6853 struct dentry *dentry, umode_t mode)
6855 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6856 struct inode *inode = NULL;
6857 struct btrfs_trans_handle *trans;
6858 struct btrfs_root *root = BTRFS_I(dir)->root;
6864 * 2 items for inode and ref
6865 * 2 items for dir items
6866 * 1 for xattr if selinux is on
6868 trans = btrfs_start_transaction(root, 5);
6870 return PTR_ERR(trans);
6872 err = btrfs_get_free_objectid(root, &objectid);
6876 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6877 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6878 S_IFDIR | mode, &index);
6879 if (IS_ERR(inode)) {
6880 err = PTR_ERR(inode);
6885 /* these must be set before we unlock the inode */
6886 inode->i_op = &btrfs_dir_inode_operations;
6887 inode->i_fop = &btrfs_dir_file_operations;
6889 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6893 btrfs_i_size_write(BTRFS_I(inode), 0);
6894 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6898 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6899 dentry->d_name.name,
6900 dentry->d_name.len, 0, index);
6904 d_instantiate_new(dentry, inode);
6907 btrfs_end_transaction(trans);
6909 inode_dec_link_count(inode);
6910 discard_new_inode(inode);
6912 btrfs_btree_balance_dirty(fs_info);
6916 static noinline int uncompress_inline(struct btrfs_path *path,
6918 size_t pg_offset, u64 extent_offset,
6919 struct btrfs_file_extent_item *item)
6922 struct extent_buffer *leaf = path->nodes[0];
6925 unsigned long inline_size;
6929 WARN_ON(pg_offset != 0);
6930 compress_type = btrfs_file_extent_compression(leaf, item);
6931 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6932 inline_size = btrfs_file_extent_inline_item_len(leaf,
6933 btrfs_item_nr(path->slots[0]));
6934 tmp = kmalloc(inline_size, GFP_NOFS);
6937 ptr = btrfs_file_extent_inline_start(item);
6939 read_extent_buffer(leaf, tmp, ptr, inline_size);
6941 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6942 ret = btrfs_decompress(compress_type, tmp, page,
6943 extent_offset, inline_size, max_size);
6946 * decompression code contains a memset to fill in any space between the end
6947 * of the uncompressed data and the end of max_size in case the decompressed
6948 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6949 * the end of an inline extent and the beginning of the next block, so we
6950 * cover that region here.
6953 if (max_size + pg_offset < PAGE_SIZE)
6954 memzero_page(page, pg_offset + max_size,
6955 PAGE_SIZE - max_size - pg_offset);
6961 * btrfs_get_extent - Lookup the first extent overlapping a range in a file.
6962 * @inode: file to search in
6963 * @page: page to read extent data into if the extent is inline
6964 * @pg_offset: offset into @page to copy to
6965 * @start: file offset
6966 * @len: length of range starting at @start
6968 * This returns the first &struct extent_map which overlaps with the given
6969 * range, reading it from the B-tree and caching it if necessary. Note that
6970 * there may be more extents which overlap the given range after the returned
6973 * If @page is not NULL and the extent is inline, this also reads the extent
6974 * data directly into the page and marks the extent up to date in the io_tree.
6976 * Return: ERR_PTR on error, non-NULL extent_map on success.
6978 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6979 struct page *page, size_t pg_offset,
6982 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6984 u64 extent_start = 0;
6986 u64 objectid = btrfs_ino(inode);
6987 int extent_type = -1;
6988 struct btrfs_path *path = NULL;
6989 struct btrfs_root *root = inode->root;
6990 struct btrfs_file_extent_item *item;
6991 struct extent_buffer *leaf;
6992 struct btrfs_key found_key;
6993 struct extent_map *em = NULL;
6994 struct extent_map_tree *em_tree = &inode->extent_tree;
6995 struct extent_io_tree *io_tree = &inode->io_tree;
6997 read_lock(&em_tree->lock);
6998 em = lookup_extent_mapping(em_tree, start, len);
6999 read_unlock(&em_tree->lock);
7002 if (em->start > start || em->start + em->len <= start)
7003 free_extent_map(em);
7004 else if (em->block_start == EXTENT_MAP_INLINE && page)
7005 free_extent_map(em);
7009 em = alloc_extent_map();
7014 em->start = EXTENT_MAP_HOLE;
7015 em->orig_start = EXTENT_MAP_HOLE;
7017 em->block_len = (u64)-1;
7019 path = btrfs_alloc_path();
7025 /* Chances are we'll be called again, so go ahead and do readahead */
7026 path->reada = READA_FORWARD;
7029 * The same explanation in load_free_space_cache applies here as well,
7030 * we only read when we're loading the free space cache, and at that
7031 * point the commit_root has everything we need.
7033 if (btrfs_is_free_space_inode(inode)) {
7034 path->search_commit_root = 1;
7035 path->skip_locking = 1;
7038 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
7041 } else if (ret > 0) {
7042 if (path->slots[0] == 0)
7048 leaf = path->nodes[0];
7049 item = btrfs_item_ptr(leaf, path->slots[0],
7050 struct btrfs_file_extent_item);
7051 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7052 if (found_key.objectid != objectid ||
7053 found_key.type != BTRFS_EXTENT_DATA_KEY) {
7055 * If we backup past the first extent we want to move forward
7056 * and see if there is an extent in front of us, otherwise we'll
7057 * say there is a hole for our whole search range which can
7064 extent_type = btrfs_file_extent_type(leaf, item);
7065 extent_start = found_key.offset;
7066 extent_end = btrfs_file_extent_end(path);
7067 if (extent_type == BTRFS_FILE_EXTENT_REG ||
7068 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
7069 /* Only regular file could have regular/prealloc extent */
7070 if (!S_ISREG(inode->vfs_inode.i_mode)) {
7073 "regular/prealloc extent found for non-regular inode %llu",
7077 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
7079 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
7080 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
7085 if (start >= extent_end) {
7087 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
7088 ret = btrfs_next_leaf(root, path);
7094 leaf = path->nodes[0];
7096 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7097 if (found_key.objectid != objectid ||
7098 found_key.type != BTRFS_EXTENT_DATA_KEY)
7100 if (start + len <= found_key.offset)
7102 if (start > found_key.offset)
7105 /* New extent overlaps with existing one */
7107 em->orig_start = start;
7108 em->len = found_key.offset - start;
7109 em->block_start = EXTENT_MAP_HOLE;
7113 btrfs_extent_item_to_extent_map(inode, path, item, !page, em);
7115 if (extent_type == BTRFS_FILE_EXTENT_REG ||
7116 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
7118 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
7122 size_t extent_offset;
7128 size = btrfs_file_extent_ram_bytes(leaf, item);
7129 extent_offset = page_offset(page) + pg_offset - extent_start;
7130 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
7131 size - extent_offset);
7132 em->start = extent_start + extent_offset;
7133 em->len = ALIGN(copy_size, fs_info->sectorsize);
7134 em->orig_block_len = em->len;
7135 em->orig_start = em->start;
7136 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
7138 if (!PageUptodate(page)) {
7139 if (btrfs_file_extent_compression(leaf, item) !=
7140 BTRFS_COMPRESS_NONE) {
7141 ret = uncompress_inline(path, page, pg_offset,
7142 extent_offset, item);
7146 map = kmap_local_page(page);
7147 read_extent_buffer(leaf, map + pg_offset, ptr,
7149 if (pg_offset + copy_size < PAGE_SIZE) {
7150 memset(map + pg_offset + copy_size, 0,
7151 PAGE_SIZE - pg_offset -
7156 flush_dcache_page(page);
7158 set_extent_uptodate(io_tree, em->start,
7159 extent_map_end(em) - 1, NULL, GFP_NOFS);
7164 em->orig_start = start;
7166 em->block_start = EXTENT_MAP_HOLE;
7169 btrfs_release_path(path);
7170 if (em->start > start || extent_map_end(em) <= start) {
7172 "bad extent! em: [%llu %llu] passed [%llu %llu]",
7173 em->start, em->len, start, len);
7178 write_lock(&em_tree->lock);
7179 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
7180 write_unlock(&em_tree->lock);
7182 btrfs_free_path(path);
7184 trace_btrfs_get_extent(root, inode, em);
7187 free_extent_map(em);
7188 return ERR_PTR(ret);
7193 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
7196 struct extent_map *em;
7197 struct extent_map *hole_em = NULL;
7198 u64 delalloc_start = start;
7204 em = btrfs_get_extent(inode, NULL, 0, start, len);
7208 * If our em maps to:
7210 * - a pre-alloc extent,
7211 * there might actually be delalloc bytes behind it.
7213 if (em->block_start != EXTENT_MAP_HOLE &&
7214 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7219 /* check to see if we've wrapped (len == -1 or similar) */
7228 /* ok, we didn't find anything, lets look for delalloc */
7229 delalloc_len = count_range_bits(&inode->io_tree, &delalloc_start,
7230 end, len, EXTENT_DELALLOC, 1);
7231 delalloc_end = delalloc_start + delalloc_len;
7232 if (delalloc_end < delalloc_start)
7233 delalloc_end = (u64)-1;
7236 * We didn't find anything useful, return the original results from
7239 if (delalloc_start > end || delalloc_end <= start) {
7246 * Adjust the delalloc_start to make sure it doesn't go backwards from
7247 * the start they passed in
7249 delalloc_start = max(start, delalloc_start);
7250 delalloc_len = delalloc_end - delalloc_start;
7252 if (delalloc_len > 0) {
7255 const u64 hole_end = extent_map_end(hole_em);
7257 em = alloc_extent_map();
7265 * When btrfs_get_extent can't find anything it returns one
7268 * Make sure what it found really fits our range, and adjust to
7269 * make sure it is based on the start from the caller
7271 if (hole_end <= start || hole_em->start > end) {
7272 free_extent_map(hole_em);
7275 hole_start = max(hole_em->start, start);
7276 hole_len = hole_end - hole_start;
7279 if (hole_em && delalloc_start > hole_start) {
7281 * Our hole starts before our delalloc, so we have to
7282 * return just the parts of the hole that go until the
7285 em->len = min(hole_len, delalloc_start - hole_start);
7286 em->start = hole_start;
7287 em->orig_start = hole_start;
7289 * Don't adjust block start at all, it is fixed at
7292 em->block_start = hole_em->block_start;
7293 em->block_len = hole_len;
7294 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7295 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7298 * Hole is out of passed range or it starts after
7301 em->start = delalloc_start;
7302 em->len = delalloc_len;
7303 em->orig_start = delalloc_start;
7304 em->block_start = EXTENT_MAP_DELALLOC;
7305 em->block_len = delalloc_len;
7312 free_extent_map(hole_em);
7314 free_extent_map(em);
7315 return ERR_PTR(err);
7320 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
7323 const u64 orig_start,
7324 const u64 block_start,
7325 const u64 block_len,
7326 const u64 orig_block_len,
7327 const u64 ram_bytes,
7330 struct extent_map *em = NULL;
7333 if (type != BTRFS_ORDERED_NOCOW) {
7334 em = create_io_em(inode, start, len, orig_start, block_start,
7335 block_len, orig_block_len, ram_bytes,
7336 BTRFS_COMPRESS_NONE, /* compress_type */
7341 ret = btrfs_add_ordered_extent_dio(inode, start, block_start, len,
7345 free_extent_map(em);
7346 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
7355 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
7358 struct btrfs_root *root = inode->root;
7359 struct btrfs_fs_info *fs_info = root->fs_info;
7360 struct extent_map *em;
7361 struct btrfs_key ins;
7365 alloc_hint = get_extent_allocation_hint(inode, start, len);
7366 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7367 0, alloc_hint, &ins, 1, 1);
7369 return ERR_PTR(ret);
7371 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7372 ins.objectid, ins.offset, ins.offset,
7373 ins.offset, BTRFS_ORDERED_REGULAR);
7374 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7376 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7382 static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
7384 struct btrfs_block_group *block_group;
7385 bool readonly = false;
7387 block_group = btrfs_lookup_block_group(fs_info, bytenr);
7388 if (!block_group || block_group->ro)
7391 btrfs_put_block_group(block_group);
7396 * Check if we can do nocow write into the range [@offset, @offset + @len)
7398 * @offset: File offset
7399 * @len: The length to write, will be updated to the nocow writeable
7401 * @orig_start: (optional) Return the original file offset of the file extent
7402 * @orig_len: (optional) Return the original on-disk length of the file extent
7403 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7404 * @strict: if true, omit optimizations that might force us into unnecessary
7405 * cow. e.g., don't trust generation number.
7408 * >0 and update @len if we can do nocow write
7409 * 0 if we can't do nocow write
7410 * <0 if error happened
7412 * NOTE: This only checks the file extents, caller is responsible to wait for
7413 * any ordered extents.
7415 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7416 u64 *orig_start, u64 *orig_block_len,
7417 u64 *ram_bytes, bool strict)
7419 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7420 struct btrfs_path *path;
7422 struct extent_buffer *leaf;
7423 struct btrfs_root *root = BTRFS_I(inode)->root;
7424 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7425 struct btrfs_file_extent_item *fi;
7426 struct btrfs_key key;
7433 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7435 path = btrfs_alloc_path();
7439 ret = btrfs_lookup_file_extent(NULL, root, path,
7440 btrfs_ino(BTRFS_I(inode)), offset, 0);
7444 slot = path->slots[0];
7447 /* can't find the item, must cow */
7454 leaf = path->nodes[0];
7455 btrfs_item_key_to_cpu(leaf, &key, slot);
7456 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7457 key.type != BTRFS_EXTENT_DATA_KEY) {
7458 /* not our file or wrong item type, must cow */
7462 if (key.offset > offset) {
7463 /* Wrong offset, must cow */
7467 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7468 found_type = btrfs_file_extent_type(leaf, fi);
7469 if (found_type != BTRFS_FILE_EXTENT_REG &&
7470 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7471 /* not a regular extent, must cow */
7475 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7478 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7479 if (extent_end <= offset)
7482 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7483 if (disk_bytenr == 0)
7486 if (btrfs_file_extent_compression(leaf, fi) ||
7487 btrfs_file_extent_encryption(leaf, fi) ||
7488 btrfs_file_extent_other_encoding(leaf, fi))
7492 * Do the same check as in btrfs_cross_ref_exist but without the
7493 * unnecessary search.
7496 (btrfs_file_extent_generation(leaf, fi) <=
7497 btrfs_root_last_snapshot(&root->root_item)))
7500 backref_offset = btrfs_file_extent_offset(leaf, fi);
7503 *orig_start = key.offset - backref_offset;
7504 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7505 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7508 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7511 num_bytes = min(offset + *len, extent_end) - offset;
7512 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7515 range_end = round_up(offset + num_bytes,
7516 root->fs_info->sectorsize) - 1;
7517 ret = test_range_bit(io_tree, offset, range_end,
7518 EXTENT_DELALLOC, 0, NULL);
7525 btrfs_release_path(path);
7528 * look for other files referencing this extent, if we
7529 * find any we must cow
7532 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7533 key.offset - backref_offset, disk_bytenr,
7541 * adjust disk_bytenr and num_bytes to cover just the bytes
7542 * in this extent we are about to write. If there
7543 * are any csums in that range we have to cow in order
7544 * to keep the csums correct
7546 disk_bytenr += backref_offset;
7547 disk_bytenr += offset - key.offset;
7548 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7551 * all of the above have passed, it is safe to overwrite this extent
7557 btrfs_free_path(path);
7561 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7562 struct extent_state **cached_state, bool writing)
7564 struct btrfs_ordered_extent *ordered;
7568 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7571 * We're concerned with the entire range that we're going to be
7572 * doing DIO to, so we need to make sure there's no ordered
7573 * extents in this range.
7575 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7576 lockend - lockstart + 1);
7579 * We need to make sure there are no buffered pages in this
7580 * range either, we could have raced between the invalidate in
7581 * generic_file_direct_write and locking the extent. The
7582 * invalidate needs to happen so that reads after a write do not
7586 (!writing || !filemap_range_has_page(inode->i_mapping,
7587 lockstart, lockend)))
7590 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7595 * If we are doing a DIO read and the ordered extent we
7596 * found is for a buffered write, we can not wait for it
7597 * to complete and retry, because if we do so we can
7598 * deadlock with concurrent buffered writes on page
7599 * locks. This happens only if our DIO read covers more
7600 * than one extent map, if at this point has already
7601 * created an ordered extent for a previous extent map
7602 * and locked its range in the inode's io tree, and a
7603 * concurrent write against that previous extent map's
7604 * range and this range started (we unlock the ranges
7605 * in the io tree only when the bios complete and
7606 * buffered writes always lock pages before attempting
7607 * to lock range in the io tree).
7610 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7611 btrfs_start_ordered_extent(ordered, 1);
7614 btrfs_put_ordered_extent(ordered);
7617 * We could trigger writeback for this range (and wait
7618 * for it to complete) and then invalidate the pages for
7619 * this range (through invalidate_inode_pages2_range()),
7620 * but that can lead us to a deadlock with a concurrent
7621 * call to readahead (a buffered read or a defrag call
7622 * triggered a readahead) on a page lock due to an
7623 * ordered dio extent we created before but did not have
7624 * yet a corresponding bio submitted (whence it can not
7625 * complete), which makes readahead wait for that
7626 * ordered extent to complete while holding a lock on
7641 /* The callers of this must take lock_extent() */
7642 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7643 u64 len, u64 orig_start, u64 block_start,
7644 u64 block_len, u64 orig_block_len,
7645 u64 ram_bytes, int compress_type,
7648 struct extent_map_tree *em_tree;
7649 struct extent_map *em;
7652 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7653 type == BTRFS_ORDERED_COMPRESSED ||
7654 type == BTRFS_ORDERED_NOCOW ||
7655 type == BTRFS_ORDERED_REGULAR);
7657 em_tree = &inode->extent_tree;
7658 em = alloc_extent_map();
7660 return ERR_PTR(-ENOMEM);
7663 em->orig_start = orig_start;
7665 em->block_len = block_len;
7666 em->block_start = block_start;
7667 em->orig_block_len = orig_block_len;
7668 em->ram_bytes = ram_bytes;
7669 em->generation = -1;
7670 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7671 if (type == BTRFS_ORDERED_PREALLOC) {
7672 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7673 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7674 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7675 em->compress_type = compress_type;
7679 btrfs_drop_extent_cache(inode, em->start,
7680 em->start + em->len - 1, 0);
7681 write_lock(&em_tree->lock);
7682 ret = add_extent_mapping(em_tree, em, 1);
7683 write_unlock(&em_tree->lock);
7685 * The caller has taken lock_extent(), who could race with us
7688 } while (ret == -EEXIST);
7691 free_extent_map(em);
7692 return ERR_PTR(ret);
7695 /* em got 2 refs now, callers needs to do free_extent_map once. */
7700 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7701 struct inode *inode,
7702 struct btrfs_dio_data *dio_data,
7705 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7706 struct extent_map *em = *map;
7710 * We don't allocate a new extent in the following cases
7712 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7714 * 2) The extent is marked as PREALLOC. We're good to go here and can
7715 * just use the extent.
7718 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7719 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7720 em->block_start != EXTENT_MAP_HOLE)) {
7722 u64 block_start, orig_start, orig_block_len, ram_bytes;
7724 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7725 type = BTRFS_ORDERED_PREALLOC;
7727 type = BTRFS_ORDERED_NOCOW;
7728 len = min(len, em->len - (start - em->start));
7729 block_start = em->block_start + (start - em->start);
7731 if (can_nocow_extent(inode, start, &len, &orig_start,
7732 &orig_block_len, &ram_bytes, false) == 1 &&
7733 btrfs_inc_nocow_writers(fs_info, block_start)) {
7734 struct extent_map *em2;
7736 em2 = btrfs_create_dio_extent(BTRFS_I(inode), start, len,
7737 orig_start, block_start,
7738 len, orig_block_len,
7740 btrfs_dec_nocow_writers(fs_info, block_start);
7741 if (type == BTRFS_ORDERED_PREALLOC) {
7742 free_extent_map(em);
7746 if (em2 && IS_ERR(em2)) {
7751 * For inode marked NODATACOW or extent marked PREALLOC,
7752 * use the existing or preallocated extent, so does not
7753 * need to adjust btrfs_space_info's bytes_may_use.
7755 btrfs_free_reserved_data_space_noquota(fs_info, len);
7760 /* this will cow the extent */
7761 free_extent_map(em);
7762 *map = em = btrfs_new_extent_direct(BTRFS_I(inode), start, len);
7768 len = min(len, em->len - (start - em->start));
7772 * Need to update the i_size under the extent lock so buffered
7773 * readers will get the updated i_size when we unlock.
7775 if (start + len > i_size_read(inode))
7776 i_size_write(inode, start + len);
7778 dio_data->reserve -= len;
7783 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7784 loff_t length, unsigned int flags, struct iomap *iomap,
7785 struct iomap *srcmap)
7787 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7788 struct extent_map *em;
7789 struct extent_state *cached_state = NULL;
7790 struct btrfs_dio_data *dio_data = NULL;
7791 u64 lockstart, lockend;
7792 const bool write = !!(flags & IOMAP_WRITE);
7795 bool unlock_extents = false;
7798 len = min_t(u64, len, fs_info->sectorsize);
7801 lockend = start + len - 1;
7804 * The generic stuff only does filemap_write_and_wait_range, which
7805 * isn't enough if we've written compressed pages to this area, so we
7806 * need to flush the dirty pages again to make absolutely sure that any
7807 * outstanding dirty pages are on disk.
7809 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7810 &BTRFS_I(inode)->runtime_flags)) {
7811 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7812 start + length - 1);
7817 dio_data = kzalloc(sizeof(*dio_data), GFP_NOFS);
7821 dio_data->length = length;
7823 dio_data->reserve = round_up(length, fs_info->sectorsize);
7824 ret = btrfs_delalloc_reserve_space(BTRFS_I(inode),
7825 &dio_data->data_reserved,
7826 start, dio_data->reserve);
7828 extent_changeset_free(dio_data->data_reserved);
7833 iomap->private = dio_data;
7837 * If this errors out it's because we couldn't invalidate pagecache for
7838 * this range and we need to fallback to buffered.
7840 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, write)) {
7845 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7852 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7853 * io. INLINE is special, and we could probably kludge it in here, but
7854 * it's still buffered so for safety lets just fall back to the generic
7857 * For COMPRESSED we _have_ to read the entire extent in so we can
7858 * decompress it, so there will be buffering required no matter what we
7859 * do, so go ahead and fallback to buffered.
7861 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7862 * to buffered IO. Don't blame me, this is the price we pay for using
7865 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7866 em->block_start == EXTENT_MAP_INLINE) {
7867 free_extent_map(em);
7872 len = min(len, em->len - (start - em->start));
7874 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7878 unlock_extents = true;
7879 /* Recalc len in case the new em is smaller than requested */
7880 len = min(len, em->len - (start - em->start));
7883 * We need to unlock only the end area that we aren't using.
7884 * The rest is going to be unlocked by the endio routine.
7886 lockstart = start + len;
7887 if (lockstart < lockend)
7888 unlock_extents = true;
7892 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
7893 lockstart, lockend, &cached_state);
7895 free_extent_state(cached_state);
7898 * Translate extent map information to iomap.
7899 * We trim the extents (and move the addr) even though iomap code does
7900 * that, since we have locked only the parts we are performing I/O in.
7902 if ((em->block_start == EXTENT_MAP_HOLE) ||
7903 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7904 iomap->addr = IOMAP_NULL_ADDR;
7905 iomap->type = IOMAP_HOLE;
7907 iomap->addr = em->block_start + (start - em->start);
7908 iomap->type = IOMAP_MAPPED;
7910 iomap->offset = start;
7911 iomap->bdev = fs_info->fs_devices->latest_bdev;
7912 iomap->length = len;
7914 if (write && btrfs_use_zone_append(BTRFS_I(inode), em->block_start))
7915 iomap->flags |= IOMAP_F_ZONE_APPEND;
7917 free_extent_map(em);
7922 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7926 btrfs_delalloc_release_space(BTRFS_I(inode),
7927 dio_data->data_reserved, start,
7928 dio_data->reserve, true);
7929 btrfs_delalloc_release_extents(BTRFS_I(inode), dio_data->reserve);
7930 extent_changeset_free(dio_data->data_reserved);
7936 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7937 ssize_t written, unsigned int flags, struct iomap *iomap)
7940 struct btrfs_dio_data *dio_data = iomap->private;
7941 size_t submitted = dio_data->submitted;
7942 const bool write = !!(flags & IOMAP_WRITE);
7944 if (!write && (iomap->type == IOMAP_HOLE)) {
7945 /* If reading from a hole, unlock and return */
7946 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1);
7950 if (submitted < length) {
7952 length -= submitted;
7954 __endio_write_update_ordered(BTRFS_I(inode), pos,
7957 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7963 if (dio_data->reserve)
7964 btrfs_delalloc_release_space(BTRFS_I(inode),
7965 dio_data->data_reserved, pos,
7966 dio_data->reserve, true);
7967 btrfs_delalloc_release_extents(BTRFS_I(inode), dio_data->length);
7968 extent_changeset_free(dio_data->data_reserved);
7972 iomap->private = NULL;
7977 static void btrfs_dio_private_put(struct btrfs_dio_private *dip)
7980 * This implies a barrier so that stores to dio_bio->bi_status before
7981 * this and loads of dio_bio->bi_status after this are fully ordered.
7983 if (!refcount_dec_and_test(&dip->refs))
7986 if (btrfs_op(dip->dio_bio) == BTRFS_MAP_WRITE) {
7987 __endio_write_update_ordered(BTRFS_I(dip->inode),
7988 dip->logical_offset,
7990 !dip->dio_bio->bi_status);
7992 unlock_extent(&BTRFS_I(dip->inode)->io_tree,
7993 dip->logical_offset,
7994 dip->logical_offset + dip->bytes - 1);
7997 bio_endio(dip->dio_bio);
8001 static blk_status_t submit_dio_repair_bio(struct inode *inode, struct bio *bio,
8003 unsigned long bio_flags)
8005 struct btrfs_dio_private *dip = bio->bi_private;
8006 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8009 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
8011 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8015 refcount_inc(&dip->refs);
8016 ret = btrfs_map_bio(fs_info, bio, mirror_num);
8018 refcount_dec(&dip->refs);
8022 static blk_status_t btrfs_check_read_dio_bio(struct inode *inode,
8023 struct btrfs_io_bio *io_bio,
8024 const bool uptodate)
8026 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
8027 const u32 sectorsize = fs_info->sectorsize;
8028 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
8029 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8030 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
8031 struct bio_vec bvec;
8032 struct bvec_iter iter;
8033 u64 start = io_bio->logical;
8035 blk_status_t err = BLK_STS_OK;
8037 __bio_for_each_segment(bvec, &io_bio->bio, iter, io_bio->iter) {
8038 unsigned int i, nr_sectors, pgoff;
8040 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
8041 pgoff = bvec.bv_offset;
8042 for (i = 0; i < nr_sectors; i++) {
8043 ASSERT(pgoff < PAGE_SIZE);
8045 (!csum || !check_data_csum(inode, io_bio,
8046 bio_offset, bvec.bv_page,
8048 clean_io_failure(fs_info, failure_tree, io_tree,
8049 start, bvec.bv_page,
8050 btrfs_ino(BTRFS_I(inode)),
8055 ASSERT((start - io_bio->logical) < UINT_MAX);
8056 ret = btrfs_repair_one_sector(inode,
8058 start - io_bio->logical,
8059 bvec.bv_page, pgoff,
8060 start, io_bio->mirror_num,
8061 submit_dio_repair_bio);
8063 err = errno_to_blk_status(ret);
8065 start += sectorsize;
8066 ASSERT(bio_offset + sectorsize > bio_offset);
8067 bio_offset += sectorsize;
8068 pgoff += sectorsize;
8074 static void __endio_write_update_ordered(struct btrfs_inode *inode,
8075 const u64 offset, const u64 bytes,
8076 const bool uptodate)
8078 btrfs_mark_ordered_io_finished(inode, NULL, offset, bytes,
8079 finish_ordered_fn, uptodate);
8082 static blk_status_t btrfs_submit_bio_start_direct_io(struct inode *inode,
8084 u64 dio_file_offset)
8086 return btrfs_csum_one_bio(BTRFS_I(inode), bio, dio_file_offset, 1);
8089 static void btrfs_end_dio_bio(struct bio *bio)
8091 struct btrfs_dio_private *dip = bio->bi_private;
8092 blk_status_t err = bio->bi_status;
8095 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8096 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8097 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8098 bio->bi_opf, bio->bi_iter.bi_sector,
8099 bio->bi_iter.bi_size, err);
8101 if (bio_op(bio) == REQ_OP_READ) {
8102 err = btrfs_check_read_dio_bio(dip->inode, btrfs_io_bio(bio),
8107 dip->dio_bio->bi_status = err;
8109 btrfs_record_physical_zoned(dip->inode, dip->logical_offset, bio);
8112 btrfs_dio_private_put(dip);
8115 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
8116 struct inode *inode, u64 file_offset, int async_submit)
8118 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8119 struct btrfs_dio_private *dip = bio->bi_private;
8120 bool write = btrfs_op(bio) == BTRFS_MAP_WRITE;
8123 /* Check btrfs_submit_bio_hook() for rules about async submit. */
8125 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8128 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8133 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
8136 if (write && async_submit) {
8137 ret = btrfs_wq_submit_bio(inode, bio, 0, 0, file_offset,
8138 btrfs_submit_bio_start_direct_io);
8142 * If we aren't doing async submit, calculate the csum of the
8145 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, file_offset, 1);
8151 csum_offset = file_offset - dip->logical_offset;
8152 csum_offset >>= fs_info->sectorsize_bits;
8153 csum_offset *= fs_info->csum_size;
8154 btrfs_io_bio(bio)->csum = dip->csums + csum_offset;
8157 ret = btrfs_map_bio(fs_info, bio, 0);
8163 * If this succeeds, the btrfs_dio_private is responsible for cleaning up locked
8164 * or ordered extents whether or not we submit any bios.
8166 static struct btrfs_dio_private *btrfs_create_dio_private(struct bio *dio_bio,
8167 struct inode *inode,
8170 const bool write = (btrfs_op(dio_bio) == BTRFS_MAP_WRITE);
8171 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
8173 struct btrfs_dio_private *dip;
8175 dip_size = sizeof(*dip);
8176 if (!write && csum) {
8177 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8180 nblocks = dio_bio->bi_iter.bi_size >> fs_info->sectorsize_bits;
8181 dip_size += fs_info->csum_size * nblocks;
8184 dip = kzalloc(dip_size, GFP_NOFS);
8189 dip->logical_offset = file_offset;
8190 dip->bytes = dio_bio->bi_iter.bi_size;
8191 dip->disk_bytenr = dio_bio->bi_iter.bi_sector << 9;
8192 dip->dio_bio = dio_bio;
8193 refcount_set(&dip->refs, 1);
8197 static blk_qc_t btrfs_submit_direct(struct inode *inode, struct iomap *iomap,
8198 struct bio *dio_bio, loff_t file_offset)
8200 const bool write = (btrfs_op(dio_bio) == BTRFS_MAP_WRITE);
8201 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8202 const bool raid56 = (btrfs_data_alloc_profile(fs_info) &
8203 BTRFS_BLOCK_GROUP_RAID56_MASK);
8204 struct btrfs_dio_private *dip;
8207 int async_submit = 0;
8209 int clone_offset = 0;
8213 blk_status_t status;
8214 struct btrfs_io_geometry geom;
8215 struct btrfs_dio_data *dio_data = iomap->private;
8216 struct extent_map *em = NULL;
8218 dip = btrfs_create_dio_private(dio_bio, inode, file_offset);
8221 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8222 file_offset + dio_bio->bi_iter.bi_size - 1);
8224 dio_bio->bi_status = BLK_STS_RESOURCE;
8226 return BLK_QC_T_NONE;
8231 * Load the csums up front to reduce csum tree searches and
8232 * contention when submitting bios.
8234 * If we have csums disabled this will do nothing.
8236 status = btrfs_lookup_bio_sums(inode, dio_bio, dip->csums);
8237 if (status != BLK_STS_OK)
8241 start_sector = dio_bio->bi_iter.bi_sector;
8242 submit_len = dio_bio->bi_iter.bi_size;
8245 logical = start_sector << 9;
8246 em = btrfs_get_chunk_map(fs_info, logical, submit_len);
8248 status = errno_to_blk_status(PTR_ERR(em));
8252 ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(dio_bio),
8255 status = errno_to_blk_status(ret);
8258 ASSERT(geom.len <= INT_MAX);
8260 clone_len = min_t(int, submit_len, geom.len);
8263 * This will never fail as it's passing GPF_NOFS and
8264 * the allocation is backed by btrfs_bioset.
8266 bio = btrfs_bio_clone_partial(dio_bio, clone_offset, clone_len);
8267 bio->bi_private = dip;
8268 bio->bi_end_io = btrfs_end_dio_bio;
8269 btrfs_io_bio(bio)->logical = file_offset;
8271 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
8272 status = extract_ordered_extent(BTRFS_I(inode), bio,
8280 ASSERT(submit_len >= clone_len);
8281 submit_len -= clone_len;
8284 * Increase the count before we submit the bio so we know
8285 * the end IO handler won't happen before we increase the
8286 * count. Otherwise, the dip might get freed before we're
8287 * done setting it up.
8289 * We transfer the initial reference to the last bio, so we
8290 * don't need to increment the reference count for the last one.
8292 if (submit_len > 0) {
8293 refcount_inc(&dip->refs);
8295 * If we are submitting more than one bio, submit them
8296 * all asynchronously. The exception is RAID 5 or 6, as
8297 * asynchronous checksums make it difficult to collect
8298 * full stripe writes.
8304 status = btrfs_submit_dio_bio(bio, inode, file_offset,
8309 refcount_dec(&dip->refs);
8313 dio_data->submitted += clone_len;
8314 clone_offset += clone_len;
8315 start_sector += clone_len >> 9;
8316 file_offset += clone_len;
8318 free_extent_map(em);
8319 } while (submit_len > 0);
8320 return BLK_QC_T_NONE;
8323 free_extent_map(em);
8325 dip->dio_bio->bi_status = status;
8326 btrfs_dio_private_put(dip);
8328 return BLK_QC_T_NONE;
8331 const struct iomap_ops btrfs_dio_iomap_ops = {
8332 .iomap_begin = btrfs_dio_iomap_begin,
8333 .iomap_end = btrfs_dio_iomap_end,
8336 const struct iomap_dio_ops btrfs_dio_ops = {
8337 .submit_io = btrfs_submit_direct,
8340 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8345 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
8349 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
8352 int btrfs_readpage(struct file *file, struct page *page)
8354 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
8355 u64 start = page_offset(page);
8356 u64 end = start + PAGE_SIZE - 1;
8357 struct btrfs_bio_ctrl bio_ctrl = { 0 };
8360 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
8362 ret = btrfs_do_readpage(page, NULL, &bio_ctrl, 0, NULL);
8364 ret = submit_one_bio(bio_ctrl.bio, 0, bio_ctrl.bio_flags);
8368 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8370 struct inode *inode = page->mapping->host;
8373 if (current->flags & PF_MEMALLOC) {
8374 redirty_page_for_writepage(wbc, page);
8380 * If we are under memory pressure we will call this directly from the
8381 * VM, we need to make sure we have the inode referenced for the ordered
8382 * extent. If not just return like we didn't do anything.
8384 if (!igrab(inode)) {
8385 redirty_page_for_writepage(wbc, page);
8386 return AOP_WRITEPAGE_ACTIVATE;
8388 ret = extent_write_full_page(page, wbc);
8389 btrfs_add_delayed_iput(inode);
8393 static int btrfs_writepages(struct address_space *mapping,
8394 struct writeback_control *wbc)
8396 return extent_writepages(mapping, wbc);
8399 static void btrfs_readahead(struct readahead_control *rac)
8401 extent_readahead(rac);
8404 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8406 int ret = try_release_extent_mapping(page, gfp_flags);
8408 clear_page_extent_mapped(page);
8412 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8414 if (PageWriteback(page) || PageDirty(page))
8416 return __btrfs_releasepage(page, gfp_flags);
8419 #ifdef CONFIG_MIGRATION
8420 static int btrfs_migratepage(struct address_space *mapping,
8421 struct page *newpage, struct page *page,
8422 enum migrate_mode mode)
8426 ret = migrate_page_move_mapping(mapping, newpage, page, 0);
8427 if (ret != MIGRATEPAGE_SUCCESS)
8430 if (page_has_private(page))
8431 attach_page_private(newpage, detach_page_private(page));
8433 if (PageOrdered(page)) {
8434 ClearPageOrdered(page);
8435 SetPageOrdered(newpage);
8438 if (mode != MIGRATE_SYNC_NO_COPY)
8439 migrate_page_copy(newpage, page);
8441 migrate_page_states(newpage, page);
8442 return MIGRATEPAGE_SUCCESS;
8446 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8447 unsigned int length)
8449 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
8450 struct btrfs_fs_info *fs_info = inode->root->fs_info;
8451 struct extent_io_tree *tree = &inode->io_tree;
8452 struct extent_state *cached_state = NULL;
8453 u64 page_start = page_offset(page);
8454 u64 page_end = page_start + PAGE_SIZE - 1;
8456 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
8459 * We have page locked so no new ordered extent can be created on this
8460 * page, nor bio can be submitted for this page.
8462 * But already submitted bio can still be finished on this page.
8463 * Furthermore, endio function won't skip page which has Ordered
8464 * (Private2) already cleared, so it's possible for endio and
8465 * invalidatepage to do the same ordered extent accounting twice
8468 * So here we wait for any submitted bios to finish, so that we won't
8469 * do double ordered extent accounting on the same page.
8471 wait_on_page_writeback(page);
8474 * For subpage case, we have call sites like
8475 * btrfs_punch_hole_lock_range() which passes range not aligned to
8477 * If the range doesn't cover the full page, we don't need to and
8478 * shouldn't clear page extent mapped, as page->private can still
8479 * record subpage dirty bits for other part of the range.
8481 * For cases that can invalidate the full even the range doesn't
8482 * cover the full page, like invalidating the last page, we're
8483 * still safe to wait for ordered extent to finish.
8485 if (!(offset == 0 && length == PAGE_SIZE)) {
8486 btrfs_releasepage(page, GFP_NOFS);
8490 if (!inode_evicting)
8491 lock_extent_bits(tree, page_start, page_end, &cached_state);
8494 while (cur < page_end) {
8495 struct btrfs_ordered_extent *ordered;
8500 ordered = btrfs_lookup_first_ordered_range(inode, cur,
8501 page_end + 1 - cur);
8503 range_end = page_end;
8505 * No ordered extent covering this range, we are safe
8506 * to delete all extent states in the range.
8508 delete_states = true;
8511 if (ordered->file_offset > cur) {
8513 * There is a range between [cur, oe->file_offset) not
8514 * covered by any ordered extent.
8515 * We are safe to delete all extent states, and handle
8516 * the ordered extent in the next iteration.
8518 range_end = ordered->file_offset - 1;
8519 delete_states = true;
8523 range_end = min(ordered->file_offset + ordered->num_bytes - 1,
8525 ASSERT(range_end + 1 - cur < U32_MAX);
8526 range_len = range_end + 1 - cur;
8527 if (!btrfs_page_test_ordered(fs_info, page, cur, range_len)) {
8529 * If Ordered (Private2) is cleared, it means endio has
8530 * already been executed for the range.
8531 * We can't delete the extent states as
8532 * btrfs_finish_ordered_io() may still use some of them.
8534 delete_states = false;
8537 btrfs_page_clear_ordered(fs_info, page, cur, range_len);
8540 * IO on this page will never be started, so we need to account
8541 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8542 * here, must leave that up for the ordered extent completion.
8544 * This will also unlock the range for incoming
8545 * btrfs_finish_ordered_io().
8547 if (!inode_evicting)
8548 clear_extent_bit(tree, cur, range_end,
8550 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8551 EXTENT_DEFRAG, 1, 0, &cached_state);
8553 spin_lock_irq(&inode->ordered_tree.lock);
8554 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8555 ordered->truncated_len = min(ordered->truncated_len,
8556 cur - ordered->file_offset);
8557 spin_unlock_irq(&inode->ordered_tree.lock);
8559 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8560 cur, range_end + 1 - cur, 1)) {
8561 btrfs_finish_ordered_io(ordered);
8563 * The ordered extent has finished, now we're again
8564 * safe to delete all extent states of the range.
8566 delete_states = true;
8569 * btrfs_finish_ordered_io() will get executed by endio
8570 * of other pages, thus we can't delete extent states
8573 delete_states = false;
8577 btrfs_put_ordered_extent(ordered);
8579 * Qgroup reserved space handler
8580 * Sector(s) here will be either:
8582 * 1) Already written to disk or bio already finished
8583 * Then its QGROUP_RESERVED bit in io_tree is already cleared.
8584 * Qgroup will be handled by its qgroup_record then.
8585 * btrfs_qgroup_free_data() call will do nothing here.
8587 * 2) Not written to disk yet
8588 * Then btrfs_qgroup_free_data() call will clear the
8589 * QGROUP_RESERVED bit of its io_tree, and free the qgroup
8590 * reserved data space.
8591 * Since the IO will never happen for this page.
8593 btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur);
8594 if (!inode_evicting) {
8595 clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED |
8596 EXTENT_DELALLOC | EXTENT_UPTODATE |
8597 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1,
8598 delete_states, &cached_state);
8600 cur = range_end + 1;
8603 * We have iterated through all ordered extents of the page, the page
8604 * should not have Ordered (Private2) anymore, or the above iteration
8605 * did something wrong.
8607 ASSERT(!PageOrdered(page));
8608 if (!inode_evicting)
8609 __btrfs_releasepage(page, GFP_NOFS);
8610 ClearPageChecked(page);
8611 clear_page_extent_mapped(page);
8615 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8616 * called from a page fault handler when a page is first dirtied. Hence we must
8617 * be careful to check for EOF conditions here. We set the page up correctly
8618 * for a written page which means we get ENOSPC checking when writing into
8619 * holes and correct delalloc and unwritten extent mapping on filesystems that
8620 * support these features.
8622 * We are not allowed to take the i_mutex here so we have to play games to
8623 * protect against truncate races as the page could now be beyond EOF. Because
8624 * truncate_setsize() writes the inode size before removing pages, once we have
8625 * the page lock we can determine safely if the page is beyond EOF. If it is not
8626 * beyond EOF, then the page is guaranteed safe against truncation until we
8629 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8631 struct page *page = vmf->page;
8632 struct inode *inode = file_inode(vmf->vma->vm_file);
8633 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8634 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8635 struct btrfs_ordered_extent *ordered;
8636 struct extent_state *cached_state = NULL;
8637 struct extent_changeset *data_reserved = NULL;
8638 unsigned long zero_start;
8648 reserved_space = PAGE_SIZE;
8650 sb_start_pagefault(inode->i_sb);
8651 page_start = page_offset(page);
8652 page_end = page_start + PAGE_SIZE - 1;
8656 * Reserving delalloc space after obtaining the page lock can lead to
8657 * deadlock. For example, if a dirty page is locked by this function
8658 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8659 * dirty page write out, then the btrfs_writepage() function could
8660 * end up waiting indefinitely to get a lock on the page currently
8661 * being processed by btrfs_page_mkwrite() function.
8663 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8664 page_start, reserved_space);
8666 ret2 = file_update_time(vmf->vma->vm_file);
8670 ret = vmf_error(ret2);
8676 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8678 down_read(&BTRFS_I(inode)->i_mmap_lock);
8680 size = i_size_read(inode);
8682 if ((page->mapping != inode->i_mapping) ||
8683 (page_start >= size)) {
8684 /* page got truncated out from underneath us */
8687 wait_on_page_writeback(page);
8689 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8690 ret2 = set_page_extent_mapped(page);
8692 ret = vmf_error(ret2);
8693 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8698 * we can't set the delalloc bits if there are pending ordered
8699 * extents. Drop our locks and wait for them to finish
8701 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8704 unlock_extent_cached(io_tree, page_start, page_end,
8707 up_read(&BTRFS_I(inode)->i_mmap_lock);
8708 btrfs_start_ordered_extent(ordered, 1);
8709 btrfs_put_ordered_extent(ordered);
8713 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8714 reserved_space = round_up(size - page_start,
8715 fs_info->sectorsize);
8716 if (reserved_space < PAGE_SIZE) {
8717 end = page_start + reserved_space - 1;
8718 btrfs_delalloc_release_space(BTRFS_I(inode),
8719 data_reserved, page_start,
8720 PAGE_SIZE - reserved_space, true);
8725 * page_mkwrite gets called when the page is firstly dirtied after it's
8726 * faulted in, but write(2) could also dirty a page and set delalloc
8727 * bits, thus in this case for space account reason, we still need to
8728 * clear any delalloc bits within this page range since we have to
8729 * reserve data&meta space before lock_page() (see above comments).
8731 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8732 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8733 EXTENT_DEFRAG, 0, 0, &cached_state);
8735 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8738 unlock_extent_cached(io_tree, page_start, page_end,
8740 ret = VM_FAULT_SIGBUS;
8744 /* page is wholly or partially inside EOF */
8745 if (page_start + PAGE_SIZE > size)
8746 zero_start = offset_in_page(size);
8748 zero_start = PAGE_SIZE;
8750 if (zero_start != PAGE_SIZE) {
8751 memzero_page(page, zero_start, PAGE_SIZE - zero_start);
8752 flush_dcache_page(page);
8754 ClearPageChecked(page);
8755 btrfs_page_set_dirty(fs_info, page, page_start, end + 1 - page_start);
8756 btrfs_page_set_uptodate(fs_info, page, page_start, end + 1 - page_start);
8758 btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
8760 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8761 up_read(&BTRFS_I(inode)->i_mmap_lock);
8763 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8764 sb_end_pagefault(inode->i_sb);
8765 extent_changeset_free(data_reserved);
8766 return VM_FAULT_LOCKED;
8770 up_read(&BTRFS_I(inode)->i_mmap_lock);
8772 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8773 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8774 reserved_space, (ret != 0));
8776 sb_end_pagefault(inode->i_sb);
8777 extent_changeset_free(data_reserved);
8781 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
8783 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8784 struct btrfs_root *root = BTRFS_I(inode)->root;
8785 struct btrfs_block_rsv *rsv;
8787 struct btrfs_trans_handle *trans;
8788 u64 mask = fs_info->sectorsize - 1;
8789 u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8790 u64 extents_found = 0;
8792 if (!skip_writeback) {
8793 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8800 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8801 * things going on here:
8803 * 1) We need to reserve space to update our inode.
8805 * 2) We need to have something to cache all the space that is going to
8806 * be free'd up by the truncate operation, but also have some slack
8807 * space reserved in case it uses space during the truncate (thank you
8808 * very much snapshotting).
8810 * And we need these to be separate. The fact is we can use a lot of
8811 * space doing the truncate, and we have no earthly idea how much space
8812 * we will use, so we need the truncate reservation to be separate so it
8813 * doesn't end up using space reserved for updating the inode. We also
8814 * need to be able to stop the transaction and start a new one, which
8815 * means we need to be able to update the inode several times, and we
8816 * have no idea of knowing how many times that will be, so we can't just
8817 * reserve 1 item for the entirety of the operation, so that has to be
8818 * done separately as well.
8820 * So that leaves us with
8822 * 1) rsv - for the truncate reservation, which we will steal from the
8823 * transaction reservation.
8824 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8825 * updating the inode.
8827 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8830 rsv->size = min_size;
8834 * 1 for the truncate slack space
8835 * 1 for updating the inode.
8837 trans = btrfs_start_transaction(root, 2);
8838 if (IS_ERR(trans)) {
8839 ret = PTR_ERR(trans);
8843 /* Migrate the slack space for the truncate to our reserve */
8844 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8848 trans->block_rsv = rsv;
8851 ret = btrfs_truncate_inode_items(trans, root, BTRFS_I(inode),
8853 BTRFS_EXTENT_DATA_KEY,
8855 trans->block_rsv = &fs_info->trans_block_rsv;
8856 if (ret != -ENOSPC && ret != -EAGAIN)
8859 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
8863 btrfs_end_transaction(trans);
8864 btrfs_btree_balance_dirty(fs_info);
8866 trans = btrfs_start_transaction(root, 2);
8867 if (IS_ERR(trans)) {
8868 ret = PTR_ERR(trans);
8873 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8874 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8875 rsv, min_size, false);
8876 BUG_ON(ret); /* shouldn't happen */
8877 trans->block_rsv = rsv;
8881 * We can't call btrfs_truncate_block inside a trans handle as we could
8882 * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
8883 * we've truncated everything except the last little bit, and can do
8884 * btrfs_truncate_block and then update the disk_i_size.
8886 if (ret == NEED_TRUNCATE_BLOCK) {
8887 btrfs_end_transaction(trans);
8888 btrfs_btree_balance_dirty(fs_info);
8890 ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0);
8893 trans = btrfs_start_transaction(root, 1);
8894 if (IS_ERR(trans)) {
8895 ret = PTR_ERR(trans);
8898 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
8904 trans->block_rsv = &fs_info->trans_block_rsv;
8905 ret2 = btrfs_update_inode(trans, root, BTRFS_I(inode));
8909 ret2 = btrfs_end_transaction(trans);
8912 btrfs_btree_balance_dirty(fs_info);
8915 btrfs_free_block_rsv(fs_info, rsv);
8917 * So if we truncate and then write and fsync we normally would just
8918 * write the extents that changed, which is a problem if we need to
8919 * first truncate that entire inode. So set this flag so we write out
8920 * all of the extents in the inode to the sync log so we're completely
8923 * If no extents were dropped or trimmed we don't need to force the next
8924 * fsync to truncate all the inode's items from the log and re-log them
8925 * all. This means the truncate operation did not change the file size,
8926 * or changed it to a smaller size but there was only an implicit hole
8927 * between the old i_size and the new i_size, and there were no prealloc
8928 * extents beyond i_size to drop.
8930 if (extents_found > 0)
8931 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
8937 * create a new subvolume directory/inode (helper for the ioctl).
8939 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8940 struct btrfs_root *new_root,
8941 struct btrfs_root *parent_root)
8943 struct inode *inode;
8948 err = btrfs_get_free_objectid(new_root, &ino);
8952 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, ino, ino,
8953 S_IFDIR | (~current_umask() & S_IRWXUGO),
8956 return PTR_ERR(inode);
8957 inode->i_op = &btrfs_dir_inode_operations;
8958 inode->i_fop = &btrfs_dir_file_operations;
8960 set_nlink(inode, 1);
8961 btrfs_i_size_write(BTRFS_I(inode), 0);
8962 unlock_new_inode(inode);
8964 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8966 btrfs_err(new_root->fs_info,
8967 "error inheriting subvolume %llu properties: %d",
8968 new_root->root_key.objectid, err);
8970 err = btrfs_update_inode(trans, new_root, BTRFS_I(inode));
8976 struct inode *btrfs_alloc_inode(struct super_block *sb)
8978 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8979 struct btrfs_inode *ei;
8980 struct inode *inode;
8982 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL);
8989 ei->last_sub_trans = 0;
8990 ei->logged_trans = 0;
8991 ei->delalloc_bytes = 0;
8992 ei->new_delalloc_bytes = 0;
8993 ei->defrag_bytes = 0;
8994 ei->disk_i_size = 0;
8997 ei->index_cnt = (u64)-1;
8999 ei->last_unlink_trans = 0;
9000 ei->last_reflink_trans = 0;
9001 ei->last_log_commit = 0;
9003 spin_lock_init(&ei->lock);
9004 ei->outstanding_extents = 0;
9005 if (sb->s_magic != BTRFS_TEST_MAGIC)
9006 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
9007 BTRFS_BLOCK_RSV_DELALLOC);
9008 ei->runtime_flags = 0;
9009 ei->prop_compress = BTRFS_COMPRESS_NONE;
9010 ei->defrag_compress = BTRFS_COMPRESS_NONE;
9012 ei->delayed_node = NULL;
9014 ei->i_otime.tv_sec = 0;
9015 ei->i_otime.tv_nsec = 0;
9017 inode = &ei->vfs_inode;
9018 extent_map_tree_init(&ei->extent_tree);
9019 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO, inode);
9020 extent_io_tree_init(fs_info, &ei->io_failure_tree,
9021 IO_TREE_INODE_IO_FAILURE, inode);
9022 extent_io_tree_init(fs_info, &ei->file_extent_tree,
9023 IO_TREE_INODE_FILE_EXTENT, inode);
9024 ei->io_tree.track_uptodate = true;
9025 ei->io_failure_tree.track_uptodate = true;
9026 atomic_set(&ei->sync_writers, 0);
9027 mutex_init(&ei->log_mutex);
9028 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9029 INIT_LIST_HEAD(&ei->delalloc_inodes);
9030 INIT_LIST_HEAD(&ei->delayed_iput);
9031 RB_CLEAR_NODE(&ei->rb_node);
9032 init_rwsem(&ei->i_mmap_lock);
9037 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9038 void btrfs_test_destroy_inode(struct inode *inode)
9040 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9041 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9045 void btrfs_free_inode(struct inode *inode)
9047 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9050 void btrfs_destroy_inode(struct inode *vfs_inode)
9052 struct btrfs_ordered_extent *ordered;
9053 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
9054 struct btrfs_root *root = inode->root;
9056 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
9057 WARN_ON(vfs_inode->i_data.nrpages);
9058 WARN_ON(inode->block_rsv.reserved);
9059 WARN_ON(inode->block_rsv.size);
9060 WARN_ON(inode->outstanding_extents);
9061 WARN_ON(inode->delalloc_bytes);
9062 WARN_ON(inode->new_delalloc_bytes);
9063 WARN_ON(inode->csum_bytes);
9064 WARN_ON(inode->defrag_bytes);
9067 * This can happen where we create an inode, but somebody else also
9068 * created the same inode and we need to destroy the one we already
9075 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9079 btrfs_err(root->fs_info,
9080 "found ordered extent %llu %llu on inode cleanup",
9081 ordered->file_offset, ordered->num_bytes);
9082 btrfs_remove_ordered_extent(inode, ordered);
9083 btrfs_put_ordered_extent(ordered);
9084 btrfs_put_ordered_extent(ordered);
9087 btrfs_qgroup_check_reserved_leak(inode);
9088 inode_tree_del(inode);
9089 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9090 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
9091 btrfs_put_root(inode->root);
9094 int btrfs_drop_inode(struct inode *inode)
9096 struct btrfs_root *root = BTRFS_I(inode)->root;
9101 /* the snap/subvol tree is on deleting */
9102 if (btrfs_root_refs(&root->root_item) == 0)
9105 return generic_drop_inode(inode);
9108 static void init_once(void *foo)
9110 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9112 inode_init_once(&ei->vfs_inode);
9115 void __cold btrfs_destroy_cachep(void)
9118 * Make sure all delayed rcu free inodes are flushed before we
9122 kmem_cache_destroy(btrfs_inode_cachep);
9123 kmem_cache_destroy(btrfs_trans_handle_cachep);
9124 kmem_cache_destroy(btrfs_path_cachep);
9125 kmem_cache_destroy(btrfs_free_space_cachep);
9126 kmem_cache_destroy(btrfs_free_space_bitmap_cachep);
9129 int __init btrfs_init_cachep(void)
9131 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9132 sizeof(struct btrfs_inode), 0,
9133 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9135 if (!btrfs_inode_cachep)
9138 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9139 sizeof(struct btrfs_trans_handle), 0,
9140 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9141 if (!btrfs_trans_handle_cachep)
9144 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9145 sizeof(struct btrfs_path), 0,
9146 SLAB_MEM_SPREAD, NULL);
9147 if (!btrfs_path_cachep)
9150 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9151 sizeof(struct btrfs_free_space), 0,
9152 SLAB_MEM_SPREAD, NULL);
9153 if (!btrfs_free_space_cachep)
9156 btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap",
9157 PAGE_SIZE, PAGE_SIZE,
9158 SLAB_MEM_SPREAD, NULL);
9159 if (!btrfs_free_space_bitmap_cachep)
9164 btrfs_destroy_cachep();
9168 static int btrfs_getattr(struct user_namespace *mnt_userns,
9169 const struct path *path, struct kstat *stat,
9170 u32 request_mask, unsigned int flags)
9174 struct inode *inode = d_inode(path->dentry);
9175 u32 blocksize = inode->i_sb->s_blocksize;
9176 u32 bi_flags = BTRFS_I(inode)->flags;
9178 stat->result_mask |= STATX_BTIME;
9179 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9180 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9181 if (bi_flags & BTRFS_INODE_APPEND)
9182 stat->attributes |= STATX_ATTR_APPEND;
9183 if (bi_flags & BTRFS_INODE_COMPRESS)
9184 stat->attributes |= STATX_ATTR_COMPRESSED;
9185 if (bi_flags & BTRFS_INODE_IMMUTABLE)
9186 stat->attributes |= STATX_ATTR_IMMUTABLE;
9187 if (bi_flags & BTRFS_INODE_NODUMP)
9188 stat->attributes |= STATX_ATTR_NODUMP;
9190 stat->attributes_mask |= (STATX_ATTR_APPEND |
9191 STATX_ATTR_COMPRESSED |
9192 STATX_ATTR_IMMUTABLE |
9195 generic_fillattr(&init_user_ns, inode, stat);
9196 stat->dev = BTRFS_I(inode)->root->anon_dev;
9198 spin_lock(&BTRFS_I(inode)->lock);
9199 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9200 inode_bytes = inode_get_bytes(inode);
9201 spin_unlock(&BTRFS_I(inode)->lock);
9202 stat->blocks = (ALIGN(inode_bytes, blocksize) +
9203 ALIGN(delalloc_bytes, blocksize)) >> 9;
9207 static int btrfs_rename_exchange(struct inode *old_dir,
9208 struct dentry *old_dentry,
9209 struct inode *new_dir,
9210 struct dentry *new_dentry)
9212 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9213 struct btrfs_trans_handle *trans;
9214 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9215 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9216 struct inode *new_inode = new_dentry->d_inode;
9217 struct inode *old_inode = old_dentry->d_inode;
9218 struct timespec64 ctime = current_time(old_inode);
9219 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9220 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9225 bool root_log_pinned = false;
9226 bool dest_log_pinned = false;
9227 bool need_abort = false;
9229 /* we only allow rename subvolume link between subvolumes */
9230 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9233 /* close the race window with snapshot create/destroy ioctl */
9234 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
9235 new_ino == BTRFS_FIRST_FREE_OBJECTID)
9236 down_read(&fs_info->subvol_sem);
9239 * We want to reserve the absolute worst case amount of items. So if
9240 * both inodes are subvols and we need to unlink them then that would
9241 * require 4 item modifications, but if they are both normal inodes it
9242 * would require 5 item modifications, so we'll assume their normal
9243 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9244 * should cover the worst case number of items we'll modify.
9246 trans = btrfs_start_transaction(root, 12);
9247 if (IS_ERR(trans)) {
9248 ret = PTR_ERR(trans);
9253 ret = btrfs_record_root_in_trans(trans, dest);
9259 * We need to find a free sequence number both in the source and
9260 * in the destination directory for the exchange.
9262 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9265 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9269 BTRFS_I(old_inode)->dir_index = 0ULL;
9270 BTRFS_I(new_inode)->dir_index = 0ULL;
9272 /* Reference for the source. */
9273 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9274 /* force full log commit if subvolume involved. */
9275 btrfs_set_log_full_commit(trans);
9277 btrfs_pin_log_trans(root);
9278 root_log_pinned = true;
9279 ret = btrfs_insert_inode_ref(trans, dest,
9280 new_dentry->d_name.name,
9281 new_dentry->d_name.len,
9283 btrfs_ino(BTRFS_I(new_dir)),
9290 /* And now for the dest. */
9291 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9292 /* force full log commit if subvolume involved. */
9293 btrfs_set_log_full_commit(trans);
9295 btrfs_pin_log_trans(dest);
9296 dest_log_pinned = true;
9297 ret = btrfs_insert_inode_ref(trans, root,
9298 old_dentry->d_name.name,
9299 old_dentry->d_name.len,
9301 btrfs_ino(BTRFS_I(old_dir)),
9305 btrfs_abort_transaction(trans, ret);
9310 /* Update inode version and ctime/mtime. */
9311 inode_inc_iversion(old_dir);
9312 inode_inc_iversion(new_dir);
9313 inode_inc_iversion(old_inode);
9314 inode_inc_iversion(new_inode);
9315 old_dir->i_ctime = old_dir->i_mtime = ctime;
9316 new_dir->i_ctime = new_dir->i_mtime = ctime;
9317 old_inode->i_ctime = ctime;
9318 new_inode->i_ctime = ctime;
9320 if (old_dentry->d_parent != new_dentry->d_parent) {
9321 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9322 BTRFS_I(old_inode), 1);
9323 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9324 BTRFS_I(new_inode), 1);
9327 /* src is a subvolume */
9328 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9329 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9330 } else { /* src is an inode */
9331 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9332 BTRFS_I(old_dentry->d_inode),
9333 old_dentry->d_name.name,
9334 old_dentry->d_name.len);
9336 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9339 btrfs_abort_transaction(trans, ret);
9343 /* dest is a subvolume */
9344 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9345 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9346 } else { /* dest is an inode */
9347 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9348 BTRFS_I(new_dentry->d_inode),
9349 new_dentry->d_name.name,
9350 new_dentry->d_name.len);
9352 ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode));
9355 btrfs_abort_transaction(trans, ret);
9359 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9360 new_dentry->d_name.name,
9361 new_dentry->d_name.len, 0, old_idx);
9363 btrfs_abort_transaction(trans, ret);
9367 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9368 old_dentry->d_name.name,
9369 old_dentry->d_name.len, 0, new_idx);
9371 btrfs_abort_transaction(trans, ret);
9375 if (old_inode->i_nlink == 1)
9376 BTRFS_I(old_inode)->dir_index = old_idx;
9377 if (new_inode->i_nlink == 1)
9378 BTRFS_I(new_inode)->dir_index = new_idx;
9380 if (root_log_pinned) {
9381 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9382 new_dentry->d_parent);
9383 btrfs_end_log_trans(root);
9384 root_log_pinned = false;
9386 if (dest_log_pinned) {
9387 btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir),
9388 old_dentry->d_parent);
9389 btrfs_end_log_trans(dest);
9390 dest_log_pinned = false;
9394 * If we have pinned a log and an error happened, we unpin tasks
9395 * trying to sync the log and force them to fallback to a transaction
9396 * commit if the log currently contains any of the inodes involved in
9397 * this rename operation (to ensure we do not persist a log with an
9398 * inconsistent state for any of these inodes or leading to any
9399 * inconsistencies when replayed). If the transaction was aborted, the
9400 * abortion reason is propagated to userspace when attempting to commit
9401 * the transaction. If the log does not contain any of these inodes, we
9402 * allow the tasks to sync it.
9404 if (ret && (root_log_pinned || dest_log_pinned)) {
9405 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9406 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9407 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9409 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9410 btrfs_set_log_full_commit(trans);
9412 if (root_log_pinned) {
9413 btrfs_end_log_trans(root);
9414 root_log_pinned = false;
9416 if (dest_log_pinned) {
9417 btrfs_end_log_trans(dest);
9418 dest_log_pinned = false;
9421 ret2 = btrfs_end_transaction(trans);
9422 ret = ret ? ret : ret2;
9424 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
9425 old_ino == BTRFS_FIRST_FREE_OBJECTID)
9426 up_read(&fs_info->subvol_sem);
9431 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9432 struct btrfs_root *root,
9434 struct dentry *dentry)
9437 struct inode *inode;
9441 ret = btrfs_get_free_objectid(root, &objectid);
9445 inode = btrfs_new_inode(trans, root, dir,
9446 dentry->d_name.name,
9448 btrfs_ino(BTRFS_I(dir)),
9450 S_IFCHR | WHITEOUT_MODE,
9453 if (IS_ERR(inode)) {
9454 ret = PTR_ERR(inode);
9458 inode->i_op = &btrfs_special_inode_operations;
9459 init_special_inode(inode, inode->i_mode,
9462 ret = btrfs_init_inode_security(trans, inode, dir,
9467 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9468 BTRFS_I(inode), 0, index);
9472 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
9474 unlock_new_inode(inode);
9476 inode_dec_link_count(inode);
9482 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9483 struct inode *new_dir, struct dentry *new_dentry,
9486 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9487 struct btrfs_trans_handle *trans;
9488 unsigned int trans_num_items;
9489 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9490 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9491 struct inode *new_inode = d_inode(new_dentry);
9492 struct inode *old_inode = d_inode(old_dentry);
9496 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9497 bool log_pinned = false;
9499 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9502 /* we only allow rename subvolume link between subvolumes */
9503 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9506 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9507 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9510 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9511 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9515 /* check for collisions, even if the name isn't there */
9516 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9517 new_dentry->d_name.name,
9518 new_dentry->d_name.len);
9521 if (ret == -EEXIST) {
9523 * eexist without a new_inode */
9524 if (WARN_ON(!new_inode)) {
9528 /* maybe -EOVERFLOW */
9535 * we're using rename to replace one file with another. Start IO on it
9536 * now so we don't add too much work to the end of the transaction
9538 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9539 filemap_flush(old_inode->i_mapping);
9541 /* close the racy window with snapshot create/destroy ioctl */
9542 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9543 down_read(&fs_info->subvol_sem);
9545 * We want to reserve the absolute worst case amount of items. So if
9546 * both inodes are subvols and we need to unlink them then that would
9547 * require 4 item modifications, but if they are both normal inodes it
9548 * would require 5 item modifications, so we'll assume they are normal
9549 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9550 * should cover the worst case number of items we'll modify.
9551 * If our rename has the whiteout flag, we need more 5 units for the
9552 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9553 * when selinux is enabled).
9555 trans_num_items = 11;
9556 if (flags & RENAME_WHITEOUT)
9557 trans_num_items += 5;
9558 trans = btrfs_start_transaction(root, trans_num_items);
9559 if (IS_ERR(trans)) {
9560 ret = PTR_ERR(trans);
9565 ret = btrfs_record_root_in_trans(trans, dest);
9570 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9574 BTRFS_I(old_inode)->dir_index = 0ULL;
9575 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9576 /* force full log commit if subvolume involved. */
9577 btrfs_set_log_full_commit(trans);
9579 btrfs_pin_log_trans(root);
9581 ret = btrfs_insert_inode_ref(trans, dest,
9582 new_dentry->d_name.name,
9583 new_dentry->d_name.len,
9585 btrfs_ino(BTRFS_I(new_dir)), index);
9590 inode_inc_iversion(old_dir);
9591 inode_inc_iversion(new_dir);
9592 inode_inc_iversion(old_inode);
9593 old_dir->i_ctime = old_dir->i_mtime =
9594 new_dir->i_ctime = new_dir->i_mtime =
9595 old_inode->i_ctime = current_time(old_dir);
9597 if (old_dentry->d_parent != new_dentry->d_parent)
9598 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9599 BTRFS_I(old_inode), 1);
9601 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9602 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9604 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9605 BTRFS_I(d_inode(old_dentry)),
9606 old_dentry->d_name.name,
9607 old_dentry->d_name.len);
9609 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9612 btrfs_abort_transaction(trans, ret);
9617 inode_inc_iversion(new_inode);
9618 new_inode->i_ctime = current_time(new_inode);
9619 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9620 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9621 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9622 BUG_ON(new_inode->i_nlink == 0);
9624 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9625 BTRFS_I(d_inode(new_dentry)),
9626 new_dentry->d_name.name,
9627 new_dentry->d_name.len);
9629 if (!ret && new_inode->i_nlink == 0)
9630 ret = btrfs_orphan_add(trans,
9631 BTRFS_I(d_inode(new_dentry)));
9633 btrfs_abort_transaction(trans, ret);
9638 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9639 new_dentry->d_name.name,
9640 new_dentry->d_name.len, 0, index);
9642 btrfs_abort_transaction(trans, ret);
9646 if (old_inode->i_nlink == 1)
9647 BTRFS_I(old_inode)->dir_index = index;
9650 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9651 new_dentry->d_parent);
9652 btrfs_end_log_trans(root);
9656 if (flags & RENAME_WHITEOUT) {
9657 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9661 btrfs_abort_transaction(trans, ret);
9667 * If we have pinned the log and an error happened, we unpin tasks
9668 * trying to sync the log and force them to fallback to a transaction
9669 * commit if the log currently contains any of the inodes involved in
9670 * this rename operation (to ensure we do not persist a log with an
9671 * inconsistent state for any of these inodes or leading to any
9672 * inconsistencies when replayed). If the transaction was aborted, the
9673 * abortion reason is propagated to userspace when attempting to commit
9674 * the transaction. If the log does not contain any of these inodes, we
9675 * allow the tasks to sync it.
9677 if (ret && log_pinned) {
9678 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9679 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9680 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9682 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9683 btrfs_set_log_full_commit(trans);
9685 btrfs_end_log_trans(root);
9688 ret2 = btrfs_end_transaction(trans);
9689 ret = ret ? ret : ret2;
9691 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9692 up_read(&fs_info->subvol_sem);
9697 static int btrfs_rename2(struct user_namespace *mnt_userns, struct inode *old_dir,
9698 struct dentry *old_dentry, struct inode *new_dir,
9699 struct dentry *new_dentry, unsigned int flags)
9701 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9704 if (flags & RENAME_EXCHANGE)
9705 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9708 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9711 struct btrfs_delalloc_work {
9712 struct inode *inode;
9713 struct completion completion;
9714 struct list_head list;
9715 struct btrfs_work work;
9718 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9720 struct btrfs_delalloc_work *delalloc_work;
9721 struct inode *inode;
9723 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9725 inode = delalloc_work->inode;
9726 filemap_flush(inode->i_mapping);
9727 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9728 &BTRFS_I(inode)->runtime_flags))
9729 filemap_flush(inode->i_mapping);
9732 complete(&delalloc_work->completion);
9735 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9737 struct btrfs_delalloc_work *work;
9739 work = kmalloc(sizeof(*work), GFP_NOFS);
9743 init_completion(&work->completion);
9744 INIT_LIST_HEAD(&work->list);
9745 work->inode = inode;
9746 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9752 * some fairly slow code that needs optimization. This walks the list
9753 * of all the inodes with pending delalloc and forces them to disk.
9755 static int start_delalloc_inodes(struct btrfs_root *root,
9756 struct writeback_control *wbc, bool snapshot,
9757 bool in_reclaim_context)
9759 struct btrfs_inode *binode;
9760 struct inode *inode;
9761 struct btrfs_delalloc_work *work, *next;
9762 struct list_head works;
9763 struct list_head splice;
9765 bool full_flush = wbc->nr_to_write == LONG_MAX;
9767 INIT_LIST_HEAD(&works);
9768 INIT_LIST_HEAD(&splice);
9770 mutex_lock(&root->delalloc_mutex);
9771 spin_lock(&root->delalloc_lock);
9772 list_splice_init(&root->delalloc_inodes, &splice);
9773 while (!list_empty(&splice)) {
9774 binode = list_entry(splice.next, struct btrfs_inode,
9777 list_move_tail(&binode->delalloc_inodes,
9778 &root->delalloc_inodes);
9780 if (in_reclaim_context &&
9781 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9784 inode = igrab(&binode->vfs_inode);
9786 cond_resched_lock(&root->delalloc_lock);
9789 spin_unlock(&root->delalloc_lock);
9792 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9793 &binode->runtime_flags);
9795 work = btrfs_alloc_delalloc_work(inode);
9801 list_add_tail(&work->list, &works);
9802 btrfs_queue_work(root->fs_info->flush_workers,
9805 ret = sync_inode(inode, wbc);
9807 test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9808 &BTRFS_I(inode)->runtime_flags))
9809 ret = sync_inode(inode, wbc);
9810 btrfs_add_delayed_iput(inode);
9811 if (ret || wbc->nr_to_write <= 0)
9815 spin_lock(&root->delalloc_lock);
9817 spin_unlock(&root->delalloc_lock);
9820 list_for_each_entry_safe(work, next, &works, list) {
9821 list_del_init(&work->list);
9822 wait_for_completion(&work->completion);
9826 if (!list_empty(&splice)) {
9827 spin_lock(&root->delalloc_lock);
9828 list_splice_tail(&splice, &root->delalloc_inodes);
9829 spin_unlock(&root->delalloc_lock);
9831 mutex_unlock(&root->delalloc_mutex);
9835 int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9837 struct writeback_control wbc = {
9838 .nr_to_write = LONG_MAX,
9839 .sync_mode = WB_SYNC_NONE,
9841 .range_end = LLONG_MAX,
9843 struct btrfs_fs_info *fs_info = root->fs_info;
9845 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9848 return start_delalloc_inodes(root, &wbc, true, in_reclaim_context);
9851 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9852 bool in_reclaim_context)
9854 struct writeback_control wbc = {
9856 .sync_mode = WB_SYNC_NONE,
9858 .range_end = LLONG_MAX,
9860 struct btrfs_root *root;
9861 struct list_head splice;
9864 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9867 INIT_LIST_HEAD(&splice);
9869 mutex_lock(&fs_info->delalloc_root_mutex);
9870 spin_lock(&fs_info->delalloc_root_lock);
9871 list_splice_init(&fs_info->delalloc_roots, &splice);
9872 while (!list_empty(&splice)) {
9874 * Reset nr_to_write here so we know that we're doing a full
9878 wbc.nr_to_write = LONG_MAX;
9880 root = list_first_entry(&splice, struct btrfs_root,
9882 root = btrfs_grab_root(root);
9884 list_move_tail(&root->delalloc_root,
9885 &fs_info->delalloc_roots);
9886 spin_unlock(&fs_info->delalloc_root_lock);
9888 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9889 btrfs_put_root(root);
9890 if (ret < 0 || wbc.nr_to_write <= 0)
9892 spin_lock(&fs_info->delalloc_root_lock);
9894 spin_unlock(&fs_info->delalloc_root_lock);
9898 if (!list_empty(&splice)) {
9899 spin_lock(&fs_info->delalloc_root_lock);
9900 list_splice_tail(&splice, &fs_info->delalloc_roots);
9901 spin_unlock(&fs_info->delalloc_root_lock);
9903 mutex_unlock(&fs_info->delalloc_root_mutex);
9907 static int btrfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
9908 struct dentry *dentry, const char *symname)
9910 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9911 struct btrfs_trans_handle *trans;
9912 struct btrfs_root *root = BTRFS_I(dir)->root;
9913 struct btrfs_path *path;
9914 struct btrfs_key key;
9915 struct inode *inode = NULL;
9922 struct btrfs_file_extent_item *ei;
9923 struct extent_buffer *leaf;
9925 name_len = strlen(symname);
9926 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9927 return -ENAMETOOLONG;
9930 * 2 items for inode item and ref
9931 * 2 items for dir items
9932 * 1 item for updating parent inode item
9933 * 1 item for the inline extent item
9934 * 1 item for xattr if selinux is on
9936 trans = btrfs_start_transaction(root, 7);
9938 return PTR_ERR(trans);
9940 err = btrfs_get_free_objectid(root, &objectid);
9944 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9945 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
9946 objectid, S_IFLNK|S_IRWXUGO, &index);
9947 if (IS_ERR(inode)) {
9948 err = PTR_ERR(inode);
9954 * If the active LSM wants to access the inode during
9955 * d_instantiate it needs these. Smack checks to see
9956 * if the filesystem supports xattrs by looking at the
9959 inode->i_fop = &btrfs_file_operations;
9960 inode->i_op = &btrfs_file_inode_operations;
9961 inode->i_mapping->a_ops = &btrfs_aops;
9963 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9967 path = btrfs_alloc_path();
9972 key.objectid = btrfs_ino(BTRFS_I(inode));
9974 key.type = BTRFS_EXTENT_DATA_KEY;
9975 datasize = btrfs_file_extent_calc_inline_size(name_len);
9976 err = btrfs_insert_empty_item(trans, root, path, &key,
9979 btrfs_free_path(path);
9982 leaf = path->nodes[0];
9983 ei = btrfs_item_ptr(leaf, path->slots[0],
9984 struct btrfs_file_extent_item);
9985 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9986 btrfs_set_file_extent_type(leaf, ei,
9987 BTRFS_FILE_EXTENT_INLINE);
9988 btrfs_set_file_extent_encryption(leaf, ei, 0);
9989 btrfs_set_file_extent_compression(leaf, ei, 0);
9990 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9991 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9993 ptr = btrfs_file_extent_inline_start(ei);
9994 write_extent_buffer(leaf, symname, ptr, name_len);
9995 btrfs_mark_buffer_dirty(leaf);
9996 btrfs_free_path(path);
9998 inode->i_op = &btrfs_symlink_inode_operations;
9999 inode_nohighmem(inode);
10000 inode_set_bytes(inode, name_len);
10001 btrfs_i_size_write(BTRFS_I(inode), name_len);
10002 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
10004 * Last step, add directory indexes for our symlink inode. This is the
10005 * last step to avoid extra cleanup of these indexes if an error happens
10009 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
10010 BTRFS_I(inode), 0, index);
10014 d_instantiate_new(dentry, inode);
10017 btrfs_end_transaction(trans);
10018 if (err && inode) {
10019 inode_dec_link_count(inode);
10020 discard_new_inode(inode);
10022 btrfs_btree_balance_dirty(fs_info);
10026 static struct btrfs_trans_handle *insert_prealloc_file_extent(
10027 struct btrfs_trans_handle *trans_in,
10028 struct btrfs_inode *inode,
10029 struct btrfs_key *ins,
10032 struct btrfs_file_extent_item stack_fi;
10033 struct btrfs_replace_extent_info extent_info;
10034 struct btrfs_trans_handle *trans = trans_in;
10035 struct btrfs_path *path;
10036 u64 start = ins->objectid;
10037 u64 len = ins->offset;
10038 int qgroup_released;
10041 memset(&stack_fi, 0, sizeof(stack_fi));
10043 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
10044 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
10045 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
10046 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
10047 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
10048 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
10049 /* Encryption and other encoding is reserved and all 0 */
10051 qgroup_released = btrfs_qgroup_release_data(inode, file_offset, len);
10052 if (qgroup_released < 0)
10053 return ERR_PTR(qgroup_released);
10056 ret = insert_reserved_file_extent(trans, inode,
10057 file_offset, &stack_fi,
10058 true, qgroup_released);
10064 extent_info.disk_offset = start;
10065 extent_info.disk_len = len;
10066 extent_info.data_offset = 0;
10067 extent_info.data_len = len;
10068 extent_info.file_offset = file_offset;
10069 extent_info.extent_buf = (char *)&stack_fi;
10070 extent_info.is_new_extent = true;
10071 extent_info.qgroup_reserved = qgroup_released;
10072 extent_info.insertions = 0;
10074 path = btrfs_alloc_path();
10080 ret = btrfs_replace_file_extents(inode, path, file_offset,
10081 file_offset + len - 1, &extent_info,
10083 btrfs_free_path(path);
10090 * We have released qgroup data range at the beginning of the function,
10091 * and normally qgroup_released bytes will be freed when committing
10093 * But if we error out early, we have to free what we have released
10094 * or we leak qgroup data reservation.
10096 btrfs_qgroup_free_refroot(inode->root->fs_info,
10097 inode->root->root_key.objectid, qgroup_released,
10098 BTRFS_QGROUP_RSV_DATA);
10099 return ERR_PTR(ret);
10102 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10103 u64 start, u64 num_bytes, u64 min_size,
10104 loff_t actual_len, u64 *alloc_hint,
10105 struct btrfs_trans_handle *trans)
10107 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
10108 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10109 struct extent_map *em;
10110 struct btrfs_root *root = BTRFS_I(inode)->root;
10111 struct btrfs_key ins;
10112 u64 cur_offset = start;
10113 u64 clear_offset = start;
10116 u64 last_alloc = (u64)-1;
10118 bool own_trans = true;
10119 u64 end = start + num_bytes - 1;
10123 while (num_bytes > 0) {
10124 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10125 cur_bytes = max(cur_bytes, min_size);
10127 * If we are severely fragmented we could end up with really
10128 * small allocations, so if the allocator is returning small
10129 * chunks lets make its job easier by only searching for those
10132 cur_bytes = min(cur_bytes, last_alloc);
10133 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10134 min_size, 0, *alloc_hint, &ins, 1, 0);
10139 * We've reserved this space, and thus converted it from
10140 * ->bytes_may_use to ->bytes_reserved. Any error that happens
10141 * from here on out we will only need to clear our reservation
10142 * for the remaining unreserved area, so advance our
10143 * clear_offset by our extent size.
10145 clear_offset += ins.offset;
10147 last_alloc = ins.offset;
10148 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
10151 * Now that we inserted the prealloc extent we can finally
10152 * decrement the number of reservations in the block group.
10153 * If we did it before, we could race with relocation and have
10154 * relocation miss the reserved extent, making it fail later.
10156 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10157 if (IS_ERR(trans)) {
10158 ret = PTR_ERR(trans);
10159 btrfs_free_reserved_extent(fs_info, ins.objectid,
10164 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10165 cur_offset + ins.offset -1, 0);
10167 em = alloc_extent_map();
10169 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10170 &BTRFS_I(inode)->runtime_flags);
10174 em->start = cur_offset;
10175 em->orig_start = cur_offset;
10176 em->len = ins.offset;
10177 em->block_start = ins.objectid;
10178 em->block_len = ins.offset;
10179 em->orig_block_len = ins.offset;
10180 em->ram_bytes = ins.offset;
10181 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10182 em->generation = trans->transid;
10185 write_lock(&em_tree->lock);
10186 ret = add_extent_mapping(em_tree, em, 1);
10187 write_unlock(&em_tree->lock);
10188 if (ret != -EEXIST)
10190 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10191 cur_offset + ins.offset - 1,
10194 free_extent_map(em);
10196 num_bytes -= ins.offset;
10197 cur_offset += ins.offset;
10198 *alloc_hint = ins.objectid + ins.offset;
10200 inode_inc_iversion(inode);
10201 inode->i_ctime = current_time(inode);
10202 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10203 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10204 (actual_len > inode->i_size) &&
10205 (cur_offset > inode->i_size)) {
10206 if (cur_offset > actual_len)
10207 i_size = actual_len;
10209 i_size = cur_offset;
10210 i_size_write(inode, i_size);
10211 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
10214 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
10217 btrfs_abort_transaction(trans, ret);
10219 btrfs_end_transaction(trans);
10224 btrfs_end_transaction(trans);
10228 if (clear_offset < end)
10229 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
10230 end - clear_offset + 1);
10234 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10235 u64 start, u64 num_bytes, u64 min_size,
10236 loff_t actual_len, u64 *alloc_hint)
10238 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10239 min_size, actual_len, alloc_hint,
10243 int btrfs_prealloc_file_range_trans(struct inode *inode,
10244 struct btrfs_trans_handle *trans, int mode,
10245 u64 start, u64 num_bytes, u64 min_size,
10246 loff_t actual_len, u64 *alloc_hint)
10248 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10249 min_size, actual_len, alloc_hint, trans);
10252 static int btrfs_set_page_dirty(struct page *page)
10254 return __set_page_dirty_nobuffers(page);
10257 static int btrfs_permission(struct user_namespace *mnt_userns,
10258 struct inode *inode, int mask)
10260 struct btrfs_root *root = BTRFS_I(inode)->root;
10261 umode_t mode = inode->i_mode;
10263 if (mask & MAY_WRITE &&
10264 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10265 if (btrfs_root_readonly(root))
10267 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10270 return generic_permission(&init_user_ns, inode, mask);
10273 static int btrfs_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
10274 struct dentry *dentry, umode_t mode)
10276 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10277 struct btrfs_trans_handle *trans;
10278 struct btrfs_root *root = BTRFS_I(dir)->root;
10279 struct inode *inode = NULL;
10285 * 5 units required for adding orphan entry
10287 trans = btrfs_start_transaction(root, 5);
10289 return PTR_ERR(trans);
10291 ret = btrfs_get_free_objectid(root, &objectid);
10295 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10296 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10297 if (IS_ERR(inode)) {
10298 ret = PTR_ERR(inode);
10303 inode->i_fop = &btrfs_file_operations;
10304 inode->i_op = &btrfs_file_inode_operations;
10306 inode->i_mapping->a_ops = &btrfs_aops;
10308 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10312 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
10315 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10320 * We set number of links to 0 in btrfs_new_inode(), and here we set
10321 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10324 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10326 set_nlink(inode, 1);
10327 d_tmpfile(dentry, inode);
10328 unlock_new_inode(inode);
10329 mark_inode_dirty(inode);
10331 btrfs_end_transaction(trans);
10333 discard_new_inode(inode);
10334 btrfs_btree_balance_dirty(fs_info);
10338 void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end)
10340 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10341 unsigned long index = start >> PAGE_SHIFT;
10342 unsigned long end_index = end >> PAGE_SHIFT;
10346 ASSERT(end + 1 - start <= U32_MAX);
10347 len = end + 1 - start;
10348 while (index <= end_index) {
10349 page = find_get_page(inode->vfs_inode.i_mapping, index);
10350 ASSERT(page); /* Pages should be in the extent_io_tree */
10352 btrfs_page_set_writeback(fs_info, page, start, len);
10360 * Add an entry indicating a block group or device which is pinned by a
10361 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10362 * negative errno on failure.
10364 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10365 bool is_block_group)
10367 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10368 struct btrfs_swapfile_pin *sp, *entry;
10369 struct rb_node **p;
10370 struct rb_node *parent = NULL;
10372 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10377 sp->is_block_group = is_block_group;
10378 sp->bg_extent_count = 1;
10380 spin_lock(&fs_info->swapfile_pins_lock);
10381 p = &fs_info->swapfile_pins.rb_node;
10384 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10385 if (sp->ptr < entry->ptr ||
10386 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10387 p = &(*p)->rb_left;
10388 } else if (sp->ptr > entry->ptr ||
10389 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10390 p = &(*p)->rb_right;
10392 if (is_block_group)
10393 entry->bg_extent_count++;
10394 spin_unlock(&fs_info->swapfile_pins_lock);
10399 rb_link_node(&sp->node, parent, p);
10400 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10401 spin_unlock(&fs_info->swapfile_pins_lock);
10405 /* Free all of the entries pinned by this swapfile. */
10406 static void btrfs_free_swapfile_pins(struct inode *inode)
10408 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10409 struct btrfs_swapfile_pin *sp;
10410 struct rb_node *node, *next;
10412 spin_lock(&fs_info->swapfile_pins_lock);
10413 node = rb_first(&fs_info->swapfile_pins);
10415 next = rb_next(node);
10416 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10417 if (sp->inode == inode) {
10418 rb_erase(&sp->node, &fs_info->swapfile_pins);
10419 if (sp->is_block_group) {
10420 btrfs_dec_block_group_swap_extents(sp->ptr,
10421 sp->bg_extent_count);
10422 btrfs_put_block_group(sp->ptr);
10428 spin_unlock(&fs_info->swapfile_pins_lock);
10431 struct btrfs_swap_info {
10437 unsigned long nr_pages;
10441 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10442 struct btrfs_swap_info *bsi)
10444 unsigned long nr_pages;
10445 u64 first_ppage, first_ppage_reported, next_ppage;
10448 first_ppage = ALIGN(bsi->block_start, PAGE_SIZE) >> PAGE_SHIFT;
10449 next_ppage = ALIGN_DOWN(bsi->block_start + bsi->block_len,
10450 PAGE_SIZE) >> PAGE_SHIFT;
10452 if (first_ppage >= next_ppage)
10454 nr_pages = next_ppage - first_ppage;
10456 first_ppage_reported = first_ppage;
10457 if (bsi->start == 0)
10458 first_ppage_reported++;
10459 if (bsi->lowest_ppage > first_ppage_reported)
10460 bsi->lowest_ppage = first_ppage_reported;
10461 if (bsi->highest_ppage < (next_ppage - 1))
10462 bsi->highest_ppage = next_ppage - 1;
10464 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10467 bsi->nr_extents += ret;
10468 bsi->nr_pages += nr_pages;
10472 static void btrfs_swap_deactivate(struct file *file)
10474 struct inode *inode = file_inode(file);
10476 btrfs_free_swapfile_pins(inode);
10477 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10480 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10483 struct inode *inode = file_inode(file);
10484 struct btrfs_root *root = BTRFS_I(inode)->root;
10485 struct btrfs_fs_info *fs_info = root->fs_info;
10486 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10487 struct extent_state *cached_state = NULL;
10488 struct extent_map *em = NULL;
10489 struct btrfs_device *device = NULL;
10490 struct btrfs_swap_info bsi = {
10491 .lowest_ppage = (sector_t)-1ULL,
10498 * If the swap file was just created, make sure delalloc is done. If the
10499 * file changes again after this, the user is doing something stupid and
10500 * we don't really care.
10502 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10507 * The inode is locked, so these flags won't change after we check them.
10509 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10510 btrfs_warn(fs_info, "swapfile must not be compressed");
10513 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10514 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10517 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10518 btrfs_warn(fs_info, "swapfile must not be checksummed");
10523 * Balance or device remove/replace/resize can move stuff around from
10524 * under us. The exclop protection makes sure they aren't running/won't
10525 * run concurrently while we are mapping the swap extents, and
10526 * fs_info->swapfile_pins prevents them from running while the swap
10527 * file is active and moving the extents. Note that this also prevents
10528 * a concurrent device add which isn't actually necessary, but it's not
10529 * really worth the trouble to allow it.
10531 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10532 btrfs_warn(fs_info,
10533 "cannot activate swapfile while exclusive operation is running");
10538 * Prevent snapshot creation while we are activating the swap file.
10539 * We do not want to race with snapshot creation. If snapshot creation
10540 * already started before we bumped nr_swapfiles from 0 to 1 and
10541 * completes before the first write into the swap file after it is
10542 * activated, than that write would fallback to COW.
10544 if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) {
10545 btrfs_exclop_finish(fs_info);
10546 btrfs_warn(fs_info,
10547 "cannot activate swapfile because snapshot creation is in progress");
10551 * Snapshots can create extents which require COW even if NODATACOW is
10552 * set. We use this counter to prevent snapshots. We must increment it
10553 * before walking the extents because we don't want a concurrent
10554 * snapshot to run after we've already checked the extents.
10556 atomic_inc(&root->nr_swapfiles);
10558 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10560 lock_extent_bits(io_tree, 0, isize - 1, &cached_state);
10562 while (start < isize) {
10563 u64 logical_block_start, physical_block_start;
10564 struct btrfs_block_group *bg;
10565 u64 len = isize - start;
10567 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
10573 if (em->block_start == EXTENT_MAP_HOLE) {
10574 btrfs_warn(fs_info, "swapfile must not have holes");
10578 if (em->block_start == EXTENT_MAP_INLINE) {
10580 * It's unlikely we'll ever actually find ourselves
10581 * here, as a file small enough to fit inline won't be
10582 * big enough to store more than the swap header, but in
10583 * case something changes in the future, let's catch it
10584 * here rather than later.
10586 btrfs_warn(fs_info, "swapfile must not be inline");
10590 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10591 btrfs_warn(fs_info, "swapfile must not be compressed");
10596 logical_block_start = em->block_start + (start - em->start);
10597 len = min(len, em->len - (start - em->start));
10598 free_extent_map(em);
10601 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, true);
10607 btrfs_warn(fs_info,
10608 "swapfile must not be copy-on-write");
10613 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10619 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10620 btrfs_warn(fs_info,
10621 "swapfile must have single data profile");
10626 if (device == NULL) {
10627 device = em->map_lookup->stripes[0].dev;
10628 ret = btrfs_add_swapfile_pin(inode, device, false);
10633 } else if (device != em->map_lookup->stripes[0].dev) {
10634 btrfs_warn(fs_info, "swapfile must be on one device");
10639 physical_block_start = (em->map_lookup->stripes[0].physical +
10640 (logical_block_start - em->start));
10641 len = min(len, em->len - (logical_block_start - em->start));
10642 free_extent_map(em);
10645 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10647 btrfs_warn(fs_info,
10648 "could not find block group containing swapfile");
10653 if (!btrfs_inc_block_group_swap_extents(bg)) {
10654 btrfs_warn(fs_info,
10655 "block group for swapfile at %llu is read-only%s",
10657 atomic_read(&fs_info->scrubs_running) ?
10658 " (scrub running)" : "");
10659 btrfs_put_block_group(bg);
10664 ret = btrfs_add_swapfile_pin(inode, bg, true);
10666 btrfs_put_block_group(bg);
10673 if (bsi.block_len &&
10674 bsi.block_start + bsi.block_len == physical_block_start) {
10675 bsi.block_len += len;
10677 if (bsi.block_len) {
10678 ret = btrfs_add_swap_extent(sis, &bsi);
10683 bsi.block_start = physical_block_start;
10684 bsi.block_len = len;
10691 ret = btrfs_add_swap_extent(sis, &bsi);
10694 if (!IS_ERR_OR_NULL(em))
10695 free_extent_map(em);
10697 unlock_extent_cached(io_tree, 0, isize - 1, &cached_state);
10700 btrfs_swap_deactivate(file);
10702 btrfs_drew_write_unlock(&root->snapshot_lock);
10704 btrfs_exclop_finish(fs_info);
10710 sis->bdev = device->bdev;
10711 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10712 sis->max = bsi.nr_pages;
10713 sis->pages = bsi.nr_pages - 1;
10714 sis->highest_bit = bsi.nr_pages - 1;
10715 return bsi.nr_extents;
10718 static void btrfs_swap_deactivate(struct file *file)
10722 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10725 return -EOPNOTSUPP;
10730 * Update the number of bytes used in the VFS' inode. When we replace extents in
10731 * a range (clone, dedupe, fallocate's zero range), we must update the number of
10732 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
10733 * always get a correct value.
10735 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
10736 const u64 add_bytes,
10737 const u64 del_bytes)
10739 if (add_bytes == del_bytes)
10742 spin_lock(&inode->lock);
10744 inode_sub_bytes(&inode->vfs_inode, del_bytes);
10746 inode_add_bytes(&inode->vfs_inode, add_bytes);
10747 spin_unlock(&inode->lock);
10750 static const struct inode_operations btrfs_dir_inode_operations = {
10751 .getattr = btrfs_getattr,
10752 .lookup = btrfs_lookup,
10753 .create = btrfs_create,
10754 .unlink = btrfs_unlink,
10755 .link = btrfs_link,
10756 .mkdir = btrfs_mkdir,
10757 .rmdir = btrfs_rmdir,
10758 .rename = btrfs_rename2,
10759 .symlink = btrfs_symlink,
10760 .setattr = btrfs_setattr,
10761 .mknod = btrfs_mknod,
10762 .listxattr = btrfs_listxattr,
10763 .permission = btrfs_permission,
10764 .get_acl = btrfs_get_acl,
10765 .set_acl = btrfs_set_acl,
10766 .update_time = btrfs_update_time,
10767 .tmpfile = btrfs_tmpfile,
10768 .fileattr_get = btrfs_fileattr_get,
10769 .fileattr_set = btrfs_fileattr_set,
10772 static const struct file_operations btrfs_dir_file_operations = {
10773 .llseek = generic_file_llseek,
10774 .read = generic_read_dir,
10775 .iterate_shared = btrfs_real_readdir,
10776 .open = btrfs_opendir,
10777 .unlocked_ioctl = btrfs_ioctl,
10778 #ifdef CONFIG_COMPAT
10779 .compat_ioctl = btrfs_compat_ioctl,
10781 .release = btrfs_release_file,
10782 .fsync = btrfs_sync_file,
10786 * btrfs doesn't support the bmap operation because swapfiles
10787 * use bmap to make a mapping of extents in the file. They assume
10788 * these extents won't change over the life of the file and they
10789 * use the bmap result to do IO directly to the drive.
10791 * the btrfs bmap call would return logical addresses that aren't
10792 * suitable for IO and they also will change frequently as COW
10793 * operations happen. So, swapfile + btrfs == corruption.
10795 * For now we're avoiding this by dropping bmap.
10797 static const struct address_space_operations btrfs_aops = {
10798 .readpage = btrfs_readpage,
10799 .writepage = btrfs_writepage,
10800 .writepages = btrfs_writepages,
10801 .readahead = btrfs_readahead,
10802 .direct_IO = noop_direct_IO,
10803 .invalidatepage = btrfs_invalidatepage,
10804 .releasepage = btrfs_releasepage,
10805 #ifdef CONFIG_MIGRATION
10806 .migratepage = btrfs_migratepage,
10808 .set_page_dirty = btrfs_set_page_dirty,
10809 .error_remove_page = generic_error_remove_page,
10810 .swap_activate = btrfs_swap_activate,
10811 .swap_deactivate = btrfs_swap_deactivate,
10814 static const struct inode_operations btrfs_file_inode_operations = {
10815 .getattr = btrfs_getattr,
10816 .setattr = btrfs_setattr,
10817 .listxattr = btrfs_listxattr,
10818 .permission = btrfs_permission,
10819 .fiemap = btrfs_fiemap,
10820 .get_acl = btrfs_get_acl,
10821 .set_acl = btrfs_set_acl,
10822 .update_time = btrfs_update_time,
10823 .fileattr_get = btrfs_fileattr_get,
10824 .fileattr_set = btrfs_fileattr_set,
10826 static const struct inode_operations btrfs_special_inode_operations = {
10827 .getattr = btrfs_getattr,
10828 .setattr = btrfs_setattr,
10829 .permission = btrfs_permission,
10830 .listxattr = btrfs_listxattr,
10831 .get_acl = btrfs_get_acl,
10832 .set_acl = btrfs_set_acl,
10833 .update_time = btrfs_update_time,
10835 static const struct inode_operations btrfs_symlink_inode_operations = {
10836 .get_link = page_get_link,
10837 .getattr = btrfs_getattr,
10838 .setattr = btrfs_setattr,
10839 .permission = btrfs_permission,
10840 .listxattr = btrfs_listxattr,
10841 .update_time = btrfs_update_time,
10844 const struct dentry_operations btrfs_dentry_operations = {
10845 .d_delete = btrfs_dentry_delete,