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"
55 struct btrfs_iget_args {
57 struct btrfs_root *root;
60 struct btrfs_dio_data {
64 struct extent_changeset *data_reserved;
67 static const struct inode_operations btrfs_dir_inode_operations;
68 static const struct inode_operations btrfs_symlink_inode_operations;
69 static const struct inode_operations btrfs_special_inode_operations;
70 static const struct inode_operations btrfs_file_inode_operations;
71 static const struct address_space_operations btrfs_aops;
72 static const struct file_operations btrfs_dir_file_operations;
74 static struct kmem_cache *btrfs_inode_cachep;
75 struct kmem_cache *btrfs_trans_handle_cachep;
76 struct kmem_cache *btrfs_path_cachep;
77 struct kmem_cache *btrfs_free_space_cachep;
78 struct kmem_cache *btrfs_free_space_bitmap_cachep;
80 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
81 static int btrfs_truncate(struct inode *inode, bool skip_writeback);
82 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
83 static noinline int cow_file_range(struct btrfs_inode *inode,
84 struct page *locked_page,
85 u64 start, u64 end, int *page_started,
86 unsigned long *nr_written, int unlock);
87 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
88 u64 len, u64 orig_start, u64 block_start,
89 u64 block_len, u64 orig_block_len,
90 u64 ram_bytes, int compress_type,
93 static void __endio_write_update_ordered(struct btrfs_inode *inode,
94 const u64 offset, const u64 bytes,
98 * btrfs_inode_lock - lock inode i_rwsem based on arguments passed
100 * ilock_flags can have the following bit set:
102 * BTRFS_ILOCK_SHARED - acquire a shared lock on the inode
103 * BTRFS_ILOCK_TRY - try to acquire the lock, if fails on first attempt
106 int btrfs_inode_lock(struct inode *inode, unsigned int ilock_flags)
108 if (ilock_flags & BTRFS_ILOCK_SHARED) {
109 if (ilock_flags & BTRFS_ILOCK_TRY) {
110 if (!inode_trylock_shared(inode))
115 inode_lock_shared(inode);
117 if (ilock_flags & BTRFS_ILOCK_TRY) {
118 if (!inode_trylock(inode))
129 * btrfs_inode_unlock - unock inode i_rwsem
131 * ilock_flags should contain the same bits set as passed to btrfs_inode_lock()
132 * to decide whether the lock acquired is shared or exclusive.
134 void btrfs_inode_unlock(struct inode *inode, unsigned int ilock_flags)
136 if (ilock_flags & BTRFS_ILOCK_SHARED)
137 inode_unlock_shared(inode);
143 * Cleanup all submitted ordered extents in specified range to handle errors
144 * from the btrfs_run_delalloc_range() callback.
146 * NOTE: caller must ensure that when an error happens, it can not call
147 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
148 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
149 * to be released, which we want to happen only when finishing the ordered
150 * extent (btrfs_finish_ordered_io()).
152 static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode,
153 struct page *locked_page,
154 u64 offset, u64 bytes)
156 unsigned long index = offset >> PAGE_SHIFT;
157 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
158 u64 page_start = page_offset(locked_page);
159 u64 page_end = page_start + PAGE_SIZE - 1;
163 while (index <= end_index) {
164 page = find_get_page(inode->vfs_inode.i_mapping, index);
168 ClearPagePrivate2(page);
173 * In case this page belongs to the delalloc range being instantiated
174 * then skip it, since the first page of a range is going to be
175 * properly cleaned up by the caller of run_delalloc_range
177 if (page_start >= offset && page_end <= (offset + bytes - 1)) {
182 return __endio_write_update_ordered(inode, offset, bytes, false);
185 static int btrfs_dirty_inode(struct inode *inode);
187 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
188 struct inode *inode, struct inode *dir,
189 const struct qstr *qstr)
193 err = btrfs_init_acl(trans, inode, dir);
195 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
200 * this does all the hard work for inserting an inline extent into
201 * the btree. The caller should have done a btrfs_drop_extents so that
202 * no overlapping inline items exist in the btree
204 static int insert_inline_extent(struct btrfs_trans_handle *trans,
205 struct btrfs_path *path, bool extent_inserted,
206 struct btrfs_root *root, struct inode *inode,
207 u64 start, size_t size, size_t compressed_size,
209 struct page **compressed_pages)
211 struct extent_buffer *leaf;
212 struct page *page = NULL;
215 struct btrfs_file_extent_item *ei;
217 size_t cur_size = size;
218 unsigned long offset;
220 ASSERT((compressed_size > 0 && compressed_pages) ||
221 (compressed_size == 0 && !compressed_pages));
223 if (compressed_size && compressed_pages)
224 cur_size = compressed_size;
226 if (!extent_inserted) {
227 struct btrfs_key key;
230 key.objectid = btrfs_ino(BTRFS_I(inode));
232 key.type = BTRFS_EXTENT_DATA_KEY;
234 datasize = btrfs_file_extent_calc_inline_size(cur_size);
235 ret = btrfs_insert_empty_item(trans, root, path, &key,
240 leaf = path->nodes[0];
241 ei = btrfs_item_ptr(leaf, path->slots[0],
242 struct btrfs_file_extent_item);
243 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
244 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
245 btrfs_set_file_extent_encryption(leaf, ei, 0);
246 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
247 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
248 ptr = btrfs_file_extent_inline_start(ei);
250 if (compress_type != BTRFS_COMPRESS_NONE) {
253 while (compressed_size > 0) {
254 cpage = compressed_pages[i];
255 cur_size = min_t(unsigned long, compressed_size,
258 kaddr = kmap_atomic(cpage);
259 write_extent_buffer(leaf, kaddr, ptr, cur_size);
260 kunmap_atomic(kaddr);
264 compressed_size -= cur_size;
266 btrfs_set_file_extent_compression(leaf, ei,
269 page = find_get_page(inode->i_mapping,
270 start >> PAGE_SHIFT);
271 btrfs_set_file_extent_compression(leaf, ei, 0);
272 kaddr = kmap_atomic(page);
273 offset = offset_in_page(start);
274 write_extent_buffer(leaf, kaddr + offset, ptr, size);
275 kunmap_atomic(kaddr);
278 btrfs_mark_buffer_dirty(leaf);
279 btrfs_release_path(path);
282 * We align size to sectorsize for inline extents just for simplicity
285 size = ALIGN(size, root->fs_info->sectorsize);
286 ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start, size);
291 * we're an inline extent, so nobody can
292 * extend the file past i_size without locking
293 * a page we already have locked.
295 * We must do any isize and inode updates
296 * before we unlock the pages. Otherwise we
297 * could end up racing with unlink.
299 BTRFS_I(inode)->disk_i_size = inode->i_size;
306 * conditionally insert an inline extent into the file. This
307 * does the checks required to make sure the data is small enough
308 * to fit as an inline extent.
310 static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 start,
311 u64 end, size_t compressed_size,
313 struct page **compressed_pages)
315 struct btrfs_drop_extents_args drop_args = { 0 };
316 struct btrfs_root *root = inode->root;
317 struct btrfs_fs_info *fs_info = root->fs_info;
318 struct btrfs_trans_handle *trans;
319 u64 isize = i_size_read(&inode->vfs_inode);
320 u64 actual_end = min(end + 1, isize);
321 u64 inline_len = actual_end - start;
322 u64 aligned_end = ALIGN(end, fs_info->sectorsize);
323 u64 data_len = inline_len;
325 struct btrfs_path *path;
328 data_len = compressed_size;
331 actual_end > fs_info->sectorsize ||
332 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
334 (actual_end & (fs_info->sectorsize - 1)) == 0) ||
336 data_len > fs_info->max_inline) {
340 path = btrfs_alloc_path();
344 trans = btrfs_join_transaction(root);
346 btrfs_free_path(path);
347 return PTR_ERR(trans);
349 trans->block_rsv = &inode->block_rsv;
351 drop_args.path = path;
352 drop_args.start = start;
353 drop_args.end = aligned_end;
354 drop_args.drop_cache = true;
355 drop_args.replace_extent = true;
357 if (compressed_size && compressed_pages)
358 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(
361 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(
364 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
366 btrfs_abort_transaction(trans, ret);
370 if (isize > actual_end)
371 inline_len = min_t(u64, isize, actual_end);
372 ret = insert_inline_extent(trans, path, drop_args.extent_inserted,
373 root, &inode->vfs_inode, start,
374 inline_len, compressed_size,
375 compress_type, compressed_pages);
376 if (ret && ret != -ENOSPC) {
377 btrfs_abort_transaction(trans, ret);
379 } else if (ret == -ENOSPC) {
384 btrfs_update_inode_bytes(inode, inline_len, drop_args.bytes_found);
385 ret = btrfs_update_inode(trans, root, inode);
386 if (ret && ret != -ENOSPC) {
387 btrfs_abort_transaction(trans, ret);
389 } else if (ret == -ENOSPC) {
394 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
397 * Don't forget to free the reserved space, as for inlined extent
398 * it won't count as data extent, free them directly here.
399 * And at reserve time, it's always aligned to page size, so
400 * just free one page here.
402 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
403 btrfs_free_path(path);
404 btrfs_end_transaction(trans);
408 struct async_extent {
413 unsigned long nr_pages;
415 struct list_head list;
420 struct page *locked_page;
423 unsigned int write_flags;
424 struct list_head extents;
425 struct cgroup_subsys_state *blkcg_css;
426 struct btrfs_work work;
431 /* Number of chunks in flight; must be first in the structure */
433 struct async_chunk chunks[];
436 static noinline int add_async_extent(struct async_chunk *cow,
437 u64 start, u64 ram_size,
440 unsigned long nr_pages,
443 struct async_extent *async_extent;
445 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
446 BUG_ON(!async_extent); /* -ENOMEM */
447 async_extent->start = start;
448 async_extent->ram_size = ram_size;
449 async_extent->compressed_size = compressed_size;
450 async_extent->pages = pages;
451 async_extent->nr_pages = nr_pages;
452 async_extent->compress_type = compress_type;
453 list_add_tail(&async_extent->list, &cow->extents);
458 * Check if the inode has flags compatible with compression
460 static inline bool inode_can_compress(struct btrfs_inode *inode)
462 if (inode->flags & BTRFS_INODE_NODATACOW ||
463 inode->flags & BTRFS_INODE_NODATASUM)
469 * Check if the inode needs to be submitted to compression, based on mount
470 * options, defragmentation, properties or heuristics.
472 static inline int inode_need_compress(struct btrfs_inode *inode, u64 start,
475 struct btrfs_fs_info *fs_info = inode->root->fs_info;
477 if (!inode_can_compress(inode)) {
478 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
479 KERN_ERR "BTRFS: unexpected compression for ino %llu\n",
484 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
487 if (inode->defrag_compress)
489 /* bad compression ratios */
490 if (inode->flags & BTRFS_INODE_NOCOMPRESS)
492 if (btrfs_test_opt(fs_info, COMPRESS) ||
493 inode->flags & BTRFS_INODE_COMPRESS ||
494 inode->prop_compress)
495 return btrfs_compress_heuristic(&inode->vfs_inode, start, end);
499 static inline void inode_should_defrag(struct btrfs_inode *inode,
500 u64 start, u64 end, u64 num_bytes, u64 small_write)
502 /* If this is a small write inside eof, kick off a defrag */
503 if (num_bytes < small_write &&
504 (start > 0 || end + 1 < inode->disk_i_size))
505 btrfs_add_inode_defrag(NULL, inode);
509 * we create compressed extents in two phases. The first
510 * phase compresses a range of pages that have already been
511 * locked (both pages and state bits are locked).
513 * This is done inside an ordered work queue, and the compression
514 * is spread across many cpus. The actual IO submission is step
515 * two, and the ordered work queue takes care of making sure that
516 * happens in the same order things were put onto the queue by
517 * writepages and friends.
519 * If this code finds it can't get good compression, it puts an
520 * entry onto the work queue to write the uncompressed bytes. This
521 * makes sure that both compressed inodes and uncompressed inodes
522 * are written in the same order that the flusher thread sent them
525 static noinline int compress_file_range(struct async_chunk *async_chunk)
527 struct inode *inode = async_chunk->inode;
528 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
529 u64 blocksize = fs_info->sectorsize;
530 u64 start = async_chunk->start;
531 u64 end = async_chunk->end;
535 struct page **pages = NULL;
536 unsigned long nr_pages;
537 unsigned long total_compressed = 0;
538 unsigned long total_in = 0;
541 int compress_type = fs_info->compress_type;
542 int compressed_extents = 0;
545 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
549 * We need to save i_size before now because it could change in between
550 * us evaluating the size and assigning it. This is because we lock and
551 * unlock the page in truncate and fallocate, and then modify the i_size
554 * The barriers are to emulate READ_ONCE, remove that once i_size_read
558 i_size = i_size_read(inode);
560 actual_end = min_t(u64, i_size, end + 1);
563 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
564 BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
565 nr_pages = min_t(unsigned long, nr_pages,
566 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
569 * we don't want to send crud past the end of i_size through
570 * compression, that's just a waste of CPU time. So, if the
571 * end of the file is before the start of our current
572 * requested range of bytes, we bail out to the uncompressed
573 * cleanup code that can deal with all of this.
575 * It isn't really the fastest way to fix things, but this is a
576 * very uncommon corner.
578 if (actual_end <= start)
579 goto cleanup_and_bail_uncompressed;
581 total_compressed = actual_end - start;
584 * skip compression for a small file range(<=blocksize) that
585 * isn't an inline extent, since it doesn't save disk space at all.
587 if (total_compressed <= blocksize &&
588 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
589 goto cleanup_and_bail_uncompressed;
591 total_compressed = min_t(unsigned long, total_compressed,
592 BTRFS_MAX_UNCOMPRESSED);
597 * we do compression for mount -o compress and when the
598 * inode has not been flagged as nocompress. This flag can
599 * change at any time if we discover bad compression ratios.
601 if (inode_need_compress(BTRFS_I(inode), start, end)) {
603 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
605 /* just bail out to the uncompressed code */
610 if (BTRFS_I(inode)->defrag_compress)
611 compress_type = BTRFS_I(inode)->defrag_compress;
612 else if (BTRFS_I(inode)->prop_compress)
613 compress_type = BTRFS_I(inode)->prop_compress;
616 * we need to call clear_page_dirty_for_io on each
617 * page in the range. Otherwise applications with the file
618 * mmap'd can wander in and change the page contents while
619 * we are compressing them.
621 * If the compression fails for any reason, we set the pages
622 * dirty again later on.
624 * Note that the remaining part is redirtied, the start pointer
625 * has moved, the end is the original one.
628 extent_range_clear_dirty_for_io(inode, start, end);
632 /* Compression level is applied here and only here */
633 ret = btrfs_compress_pages(
634 compress_type | (fs_info->compress_level << 4),
635 inode->i_mapping, start,
642 unsigned long offset = offset_in_page(total_compressed);
643 struct page *page = pages[nr_pages - 1];
646 /* zero the tail end of the last page, we might be
647 * sending it down to disk
650 kaddr = kmap_atomic(page);
651 memset(kaddr + offset, 0,
653 kunmap_atomic(kaddr);
660 /* lets try to make an inline extent */
661 if (ret || total_in < actual_end) {
662 /* we didn't compress the entire range, try
663 * to make an uncompressed inline extent.
665 ret = cow_file_range_inline(BTRFS_I(inode), start, end,
666 0, BTRFS_COMPRESS_NONE,
669 /* try making a compressed inline extent */
670 ret = cow_file_range_inline(BTRFS_I(inode), start, end,
672 compress_type, pages);
675 unsigned long clear_flags = EXTENT_DELALLOC |
676 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
677 EXTENT_DO_ACCOUNTING;
678 unsigned long page_error_op;
680 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
683 * inline extent creation worked or returned error,
684 * we don't need to create any more async work items.
685 * Unlock and free up our temp pages.
687 * We use DO_ACCOUNTING here because we need the
688 * delalloc_release_metadata to be done _after_ we drop
689 * our outstanding extent for clearing delalloc for this
692 extent_clear_unlock_delalloc(BTRFS_I(inode), start, end,
696 PAGE_START_WRITEBACK |
701 * Ensure we only free the compressed pages if we have
702 * them allocated, as we can still reach here with
703 * inode_need_compress() == false.
706 for (i = 0; i < nr_pages; i++) {
707 WARN_ON(pages[i]->mapping);
718 * we aren't doing an inline extent round the compressed size
719 * up to a block size boundary so the allocator does sane
722 total_compressed = ALIGN(total_compressed, blocksize);
725 * one last check to make sure the compression is really a
726 * win, compare the page count read with the blocks on disk,
727 * compression must free at least one sector size
729 total_in = ALIGN(total_in, PAGE_SIZE);
730 if (total_compressed + blocksize <= total_in) {
731 compressed_extents++;
734 * The async work queues will take care of doing actual
735 * allocation on disk for these compressed pages, and
736 * will submit them to the elevator.
738 add_async_extent(async_chunk, start, total_in,
739 total_compressed, pages, nr_pages,
742 if (start + total_in < end) {
748 return compressed_extents;
753 * the compression code ran but failed to make things smaller,
754 * free any pages it allocated and our page pointer array
756 for (i = 0; i < nr_pages; i++) {
757 WARN_ON(pages[i]->mapping);
762 total_compressed = 0;
765 /* flag the file so we don't compress in the future */
766 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
767 !(BTRFS_I(inode)->prop_compress)) {
768 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
771 cleanup_and_bail_uncompressed:
773 * No compression, but we still need to write the pages in the file
774 * we've been given so far. redirty the locked page if it corresponds
775 * to our extent and set things up for the async work queue to run
776 * cow_file_range to do the normal delalloc dance.
778 if (async_chunk->locked_page &&
779 (page_offset(async_chunk->locked_page) >= start &&
780 page_offset(async_chunk->locked_page)) <= end) {
781 __set_page_dirty_nobuffers(async_chunk->locked_page);
782 /* unlocked later on in the async handlers */
786 extent_range_redirty_for_io(inode, start, end);
787 add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
788 BTRFS_COMPRESS_NONE);
789 compressed_extents++;
791 return compressed_extents;
794 static void free_async_extent_pages(struct async_extent *async_extent)
798 if (!async_extent->pages)
801 for (i = 0; i < async_extent->nr_pages; i++) {
802 WARN_ON(async_extent->pages[i]->mapping);
803 put_page(async_extent->pages[i]);
805 kfree(async_extent->pages);
806 async_extent->nr_pages = 0;
807 async_extent->pages = NULL;
811 * phase two of compressed writeback. This is the ordered portion
812 * of the code, which only gets called in the order the work was
813 * queued. We walk all the async extents created by compress_file_range
814 * and send them down to the disk.
816 static noinline void submit_compressed_extents(struct async_chunk *async_chunk)
818 struct btrfs_inode *inode = BTRFS_I(async_chunk->inode);
819 struct btrfs_fs_info *fs_info = inode->root->fs_info;
820 struct async_extent *async_extent;
822 struct btrfs_key ins;
823 struct extent_map *em;
824 struct btrfs_root *root = inode->root;
825 struct extent_io_tree *io_tree = &inode->io_tree;
829 while (!list_empty(&async_chunk->extents)) {
830 async_extent = list_entry(async_chunk->extents.next,
831 struct async_extent, list);
832 list_del(&async_extent->list);
835 lock_extent(io_tree, async_extent->start,
836 async_extent->start + async_extent->ram_size - 1);
837 /* did the compression code fall back to uncompressed IO? */
838 if (!async_extent->pages) {
839 int page_started = 0;
840 unsigned long nr_written = 0;
842 /* allocate blocks */
843 ret = cow_file_range(inode, async_chunk->locked_page,
845 async_extent->start +
846 async_extent->ram_size - 1,
847 &page_started, &nr_written, 0);
852 * if page_started, cow_file_range inserted an
853 * inline extent and took care of all the unlocking
854 * and IO for us. Otherwise, we need to submit
855 * all those pages down to the drive.
857 if (!page_started && !ret)
858 extent_write_locked_range(&inode->vfs_inode,
860 async_extent->start +
861 async_extent->ram_size - 1,
863 else if (ret && async_chunk->locked_page)
864 unlock_page(async_chunk->locked_page);
870 ret = btrfs_reserve_extent(root, async_extent->ram_size,
871 async_extent->compressed_size,
872 async_extent->compressed_size,
873 0, alloc_hint, &ins, 1, 1);
875 free_async_extent_pages(async_extent);
877 if (ret == -ENOSPC) {
878 unlock_extent(io_tree, async_extent->start,
879 async_extent->start +
880 async_extent->ram_size - 1);
883 * we need to redirty the pages if we decide to
884 * fallback to uncompressed IO, otherwise we
885 * will not submit these pages down to lower
888 extent_range_redirty_for_io(&inode->vfs_inode,
890 async_extent->start +
891 async_extent->ram_size - 1);
898 * here we're doing allocation and writeback of the
901 em = create_io_em(inode, async_extent->start,
902 async_extent->ram_size, /* len */
903 async_extent->start, /* orig_start */
904 ins.objectid, /* block_start */
905 ins.offset, /* block_len */
906 ins.offset, /* orig_block_len */
907 async_extent->ram_size, /* ram_bytes */
908 async_extent->compress_type,
909 BTRFS_ORDERED_COMPRESSED);
911 /* ret value is not necessary due to void function */
912 goto out_free_reserve;
915 ret = btrfs_add_ordered_extent_compress(inode,
918 async_extent->ram_size,
920 async_extent->compress_type);
922 btrfs_drop_extent_cache(inode, async_extent->start,
923 async_extent->start +
924 async_extent->ram_size - 1, 0);
925 goto out_free_reserve;
927 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
930 * clear dirty, set writeback and unlock the pages.
932 extent_clear_unlock_delalloc(inode, async_extent->start,
933 async_extent->start +
934 async_extent->ram_size - 1,
935 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
936 PAGE_UNLOCK | PAGE_START_WRITEBACK);
937 if (btrfs_submit_compressed_write(inode, async_extent->start,
938 async_extent->ram_size,
940 ins.offset, async_extent->pages,
941 async_extent->nr_pages,
942 async_chunk->write_flags,
943 async_chunk->blkcg_css)) {
944 struct page *p = async_extent->pages[0];
945 const u64 start = async_extent->start;
946 const u64 end = start + async_extent->ram_size - 1;
948 p->mapping = inode->vfs_inode.i_mapping;
949 btrfs_writepage_endio_finish_ordered(p, start, end, 0);
952 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
955 free_async_extent_pages(async_extent);
957 alloc_hint = ins.objectid + ins.offset;
963 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
964 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
966 extent_clear_unlock_delalloc(inode, async_extent->start,
967 async_extent->start +
968 async_extent->ram_size - 1,
969 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
970 EXTENT_DELALLOC_NEW |
971 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
972 PAGE_UNLOCK | PAGE_START_WRITEBACK |
973 PAGE_END_WRITEBACK | PAGE_SET_ERROR);
974 free_async_extent_pages(async_extent);
979 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
982 struct extent_map_tree *em_tree = &inode->extent_tree;
983 struct extent_map *em;
986 read_lock(&em_tree->lock);
987 em = search_extent_mapping(em_tree, start, num_bytes);
990 * if block start isn't an actual block number then find the
991 * first block in this inode and use that as a hint. If that
992 * block is also bogus then just don't worry about it.
994 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
996 em = search_extent_mapping(em_tree, 0, 0);
997 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
998 alloc_hint = em->block_start;
1000 free_extent_map(em);
1002 alloc_hint = em->block_start;
1003 free_extent_map(em);
1006 read_unlock(&em_tree->lock);
1012 * when extent_io.c finds a delayed allocation range in the file,
1013 * the call backs end up in this code. The basic idea is to
1014 * allocate extents on disk for the range, and create ordered data structs
1015 * in ram to track those extents.
1017 * locked_page is the page that writepage had locked already. We use
1018 * it to make sure we don't do extra locks or unlocks.
1020 * *page_started is set to one if we unlock locked_page and do everything
1021 * required to start IO on it. It may be clean and already done with
1022 * IO when we return.
1024 static noinline int cow_file_range(struct btrfs_inode *inode,
1025 struct page *locked_page,
1026 u64 start, u64 end, int *page_started,
1027 unsigned long *nr_written, int unlock)
1029 struct btrfs_root *root = inode->root;
1030 struct btrfs_fs_info *fs_info = root->fs_info;
1033 unsigned long ram_size;
1034 u64 cur_alloc_size = 0;
1036 u64 blocksize = fs_info->sectorsize;
1037 struct btrfs_key ins;
1038 struct extent_map *em;
1039 unsigned clear_bits;
1040 unsigned long page_ops;
1041 bool extent_reserved = false;
1044 if (btrfs_is_free_space_inode(inode)) {
1050 num_bytes = ALIGN(end - start + 1, blocksize);
1051 num_bytes = max(blocksize, num_bytes);
1052 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1054 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1057 /* lets try to make an inline extent */
1058 ret = cow_file_range_inline(inode, start, end, 0,
1059 BTRFS_COMPRESS_NONE, NULL);
1062 * We use DO_ACCOUNTING here because we need the
1063 * delalloc_release_metadata to be run _after_ we drop
1064 * our outstanding extent for clearing delalloc for this
1067 extent_clear_unlock_delalloc(inode, start, end, NULL,
1068 EXTENT_LOCKED | EXTENT_DELALLOC |
1069 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1070 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1071 PAGE_START_WRITEBACK | PAGE_END_WRITEBACK);
1072 *nr_written = *nr_written +
1073 (end - start + PAGE_SIZE) / PAGE_SIZE;
1076 } else if (ret < 0) {
1081 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1082 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
1085 * Relocation relies on the relocated extents to have exactly the same
1086 * size as the original extents. Normally writeback for relocation data
1087 * extents follows a NOCOW path because relocation preallocates the
1088 * extents. However, due to an operation such as scrub turning a block
1089 * group to RO mode, it may fallback to COW mode, so we must make sure
1090 * an extent allocated during COW has exactly the requested size and can
1091 * not be split into smaller extents, otherwise relocation breaks and
1092 * fails during the stage where it updates the bytenr of file extent
1095 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1096 min_alloc_size = num_bytes;
1098 min_alloc_size = fs_info->sectorsize;
1100 while (num_bytes > 0) {
1101 cur_alloc_size = num_bytes;
1102 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1103 min_alloc_size, 0, alloc_hint,
1107 cur_alloc_size = ins.offset;
1108 extent_reserved = true;
1110 ram_size = ins.offset;
1111 em = create_io_em(inode, start, ins.offset, /* len */
1112 start, /* orig_start */
1113 ins.objectid, /* block_start */
1114 ins.offset, /* block_len */
1115 ins.offset, /* orig_block_len */
1116 ram_size, /* ram_bytes */
1117 BTRFS_COMPRESS_NONE, /* compress_type */
1118 BTRFS_ORDERED_REGULAR /* type */);
1123 free_extent_map(em);
1125 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1126 ram_size, cur_alloc_size,
1127 BTRFS_ORDERED_REGULAR);
1129 goto out_drop_extent_cache;
1131 if (root->root_key.objectid ==
1132 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1133 ret = btrfs_reloc_clone_csums(inode, start,
1136 * Only drop cache here, and process as normal.
1138 * We must not allow extent_clear_unlock_delalloc()
1139 * at out_unlock label to free meta of this ordered
1140 * extent, as its meta should be freed by
1141 * btrfs_finish_ordered_io().
1143 * So we must continue until @start is increased to
1144 * skip current ordered extent.
1147 btrfs_drop_extent_cache(inode, start,
1148 start + ram_size - 1, 0);
1151 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1153 /* we're not doing compressed IO, don't unlock the first
1154 * page (which the caller expects to stay locked), don't
1155 * clear any dirty bits and don't set any writeback bits
1157 * Do set the Private2 bit so we know this page was properly
1158 * setup for writepage
1160 page_ops = unlock ? PAGE_UNLOCK : 0;
1161 page_ops |= PAGE_SET_PRIVATE2;
1163 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1165 EXTENT_LOCKED | EXTENT_DELALLOC,
1167 if (num_bytes < cur_alloc_size)
1170 num_bytes -= cur_alloc_size;
1171 alloc_hint = ins.objectid + ins.offset;
1172 start += cur_alloc_size;
1173 extent_reserved = false;
1176 * btrfs_reloc_clone_csums() error, since start is increased
1177 * extent_clear_unlock_delalloc() at out_unlock label won't
1178 * free metadata of current ordered extent, we're OK to exit.
1186 out_drop_extent_cache:
1187 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1189 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1190 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1192 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1193 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1194 page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK;
1196 * If we reserved an extent for our delalloc range (or a subrange) and
1197 * failed to create the respective ordered extent, then it means that
1198 * when we reserved the extent we decremented the extent's size from
1199 * the data space_info's bytes_may_use counter and incremented the
1200 * space_info's bytes_reserved counter by the same amount. We must make
1201 * sure extent_clear_unlock_delalloc() does not try to decrement again
1202 * the data space_info's bytes_may_use counter, therefore we do not pass
1203 * it the flag EXTENT_CLEAR_DATA_RESV.
1205 if (extent_reserved) {
1206 extent_clear_unlock_delalloc(inode, start,
1207 start + cur_alloc_size - 1,
1211 start += cur_alloc_size;
1215 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1216 clear_bits | EXTENT_CLEAR_DATA_RESV,
1222 * work queue call back to started compression on a file and pages
1224 static noinline void async_cow_start(struct btrfs_work *work)
1226 struct async_chunk *async_chunk;
1227 int compressed_extents;
1229 async_chunk = container_of(work, struct async_chunk, work);
1231 compressed_extents = compress_file_range(async_chunk);
1232 if (compressed_extents == 0) {
1233 btrfs_add_delayed_iput(async_chunk->inode);
1234 async_chunk->inode = NULL;
1239 * work queue call back to submit previously compressed pages
1241 static noinline void async_cow_submit(struct btrfs_work *work)
1243 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1245 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1246 unsigned long nr_pages;
1248 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1251 /* atomic_sub_return implies a barrier */
1252 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1254 cond_wake_up_nomb(&fs_info->async_submit_wait);
1257 * ->inode could be NULL if async_chunk_start has failed to compress,
1258 * in which case we don't have anything to submit, yet we need to
1259 * always adjust ->async_delalloc_pages as its paired with the init
1260 * happening in cow_file_range_async
1262 if (async_chunk->inode)
1263 submit_compressed_extents(async_chunk);
1266 static noinline void async_cow_free(struct btrfs_work *work)
1268 struct async_chunk *async_chunk;
1270 async_chunk = container_of(work, struct async_chunk, work);
1271 if (async_chunk->inode)
1272 btrfs_add_delayed_iput(async_chunk->inode);
1273 if (async_chunk->blkcg_css)
1274 css_put(async_chunk->blkcg_css);
1276 * Since the pointer to 'pending' is at the beginning of the array of
1277 * async_chunk's, freeing it ensures the whole array has been freed.
1279 if (atomic_dec_and_test(async_chunk->pending))
1280 kvfree(async_chunk->pending);
1283 static int cow_file_range_async(struct btrfs_inode *inode,
1284 struct writeback_control *wbc,
1285 struct page *locked_page,
1286 u64 start, u64 end, int *page_started,
1287 unsigned long *nr_written)
1289 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1290 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1291 struct async_cow *ctx;
1292 struct async_chunk *async_chunk;
1293 unsigned long nr_pages;
1295 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1297 bool should_compress;
1299 const unsigned int write_flags = wbc_to_write_flags(wbc);
1301 unlock_extent(&inode->io_tree, start, end);
1303 if (inode->flags & BTRFS_INODE_NOCOMPRESS &&
1304 !btrfs_test_opt(fs_info, FORCE_COMPRESS)) {
1306 should_compress = false;
1308 should_compress = true;
1311 nofs_flag = memalloc_nofs_save();
1312 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1313 memalloc_nofs_restore(nofs_flag);
1316 unsigned clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC |
1317 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1318 EXTENT_DO_ACCOUNTING;
1319 unsigned long page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK |
1320 PAGE_END_WRITEBACK | PAGE_SET_ERROR;
1322 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1323 clear_bits, page_ops);
1327 async_chunk = ctx->chunks;
1328 atomic_set(&ctx->num_chunks, num_chunks);
1330 for (i = 0; i < num_chunks; i++) {
1331 if (should_compress)
1332 cur_end = min(end, start + SZ_512K - 1);
1337 * igrab is called higher up in the call chain, take only the
1338 * lightweight reference for the callback lifetime
1340 ihold(&inode->vfs_inode);
1341 async_chunk[i].pending = &ctx->num_chunks;
1342 async_chunk[i].inode = &inode->vfs_inode;
1343 async_chunk[i].start = start;
1344 async_chunk[i].end = cur_end;
1345 async_chunk[i].write_flags = write_flags;
1346 INIT_LIST_HEAD(&async_chunk[i].extents);
1349 * The locked_page comes all the way from writepage and its
1350 * the original page we were actually given. As we spread
1351 * this large delalloc region across multiple async_chunk
1352 * structs, only the first struct needs a pointer to locked_page
1354 * This way we don't need racey decisions about who is supposed
1359 * Depending on the compressibility, the pages might or
1360 * might not go through async. We want all of them to
1361 * be accounted against wbc once. Let's do it here
1362 * before the paths diverge. wbc accounting is used
1363 * only for foreign writeback detection and doesn't
1364 * need full accuracy. Just account the whole thing
1365 * against the first page.
1367 wbc_account_cgroup_owner(wbc, locked_page,
1369 async_chunk[i].locked_page = locked_page;
1372 async_chunk[i].locked_page = NULL;
1375 if (blkcg_css != blkcg_root_css) {
1377 async_chunk[i].blkcg_css = blkcg_css;
1379 async_chunk[i].blkcg_css = NULL;
1382 btrfs_init_work(&async_chunk[i].work, async_cow_start,
1383 async_cow_submit, async_cow_free);
1385 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1386 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1388 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1390 *nr_written += nr_pages;
1391 start = cur_end + 1;
1397 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1398 u64 bytenr, u64 num_bytes)
1401 struct btrfs_ordered_sum *sums;
1404 ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1405 bytenr + num_bytes - 1, &list, 0);
1406 if (ret == 0 && list_empty(&list))
1409 while (!list_empty(&list)) {
1410 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1411 list_del(&sums->list);
1419 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1420 const u64 start, const u64 end,
1421 int *page_started, unsigned long *nr_written)
1423 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1424 const bool is_reloc_ino = (inode->root->root_key.objectid ==
1425 BTRFS_DATA_RELOC_TREE_OBJECTID);
1426 const u64 range_bytes = end + 1 - start;
1427 struct extent_io_tree *io_tree = &inode->io_tree;
1428 u64 range_start = start;
1432 * If EXTENT_NORESERVE is set it means that when the buffered write was
1433 * made we had not enough available data space and therefore we did not
1434 * reserve data space for it, since we though we could do NOCOW for the
1435 * respective file range (either there is prealloc extent or the inode
1436 * has the NOCOW bit set).
1438 * However when we need to fallback to COW mode (because for example the
1439 * block group for the corresponding extent was turned to RO mode by a
1440 * scrub or relocation) we need to do the following:
1442 * 1) We increment the bytes_may_use counter of the data space info.
1443 * If COW succeeds, it allocates a new data extent and after doing
1444 * that it decrements the space info's bytes_may_use counter and
1445 * increments its bytes_reserved counter by the same amount (we do
1446 * this at btrfs_add_reserved_bytes()). So we need to increment the
1447 * bytes_may_use counter to compensate (when space is reserved at
1448 * buffered write time, the bytes_may_use counter is incremented);
1450 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1451 * that if the COW path fails for any reason, it decrements (through
1452 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1453 * data space info, which we incremented in the step above.
1455 * If we need to fallback to cow and the inode corresponds to a free
1456 * space cache inode or an inode of the data relocation tree, we must
1457 * also increment bytes_may_use of the data space_info for the same
1458 * reason. Space caches and relocated data extents always get a prealloc
1459 * extent for them, however scrub or balance may have set the block
1460 * group that contains that extent to RO mode and therefore force COW
1461 * when starting writeback.
1463 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1464 EXTENT_NORESERVE, 0);
1465 if (count > 0 || is_space_ino || is_reloc_ino) {
1467 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1468 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1470 if (is_space_ino || is_reloc_ino)
1471 bytes = range_bytes;
1473 spin_lock(&sinfo->lock);
1474 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1475 spin_unlock(&sinfo->lock);
1478 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1482 return cow_file_range(inode, locked_page, start, end, page_started,
1487 * when nowcow writeback call back. This checks for snapshots or COW copies
1488 * of the extents that exist in the file, and COWs the file as required.
1490 * If no cow copies or snapshots exist, we write directly to the existing
1493 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
1494 struct page *locked_page,
1495 const u64 start, const u64 end,
1496 int *page_started, int force,
1497 unsigned long *nr_written)
1499 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1500 struct btrfs_root *root = inode->root;
1501 struct btrfs_path *path;
1502 u64 cow_start = (u64)-1;
1503 u64 cur_offset = start;
1505 bool check_prev = true;
1506 const bool freespace_inode = btrfs_is_free_space_inode(inode);
1507 u64 ino = btrfs_ino(inode);
1509 u64 disk_bytenr = 0;
1511 path = btrfs_alloc_path();
1513 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1514 EXTENT_LOCKED | EXTENT_DELALLOC |
1515 EXTENT_DO_ACCOUNTING |
1516 EXTENT_DEFRAG, PAGE_UNLOCK |
1517 PAGE_START_WRITEBACK |
1518 PAGE_END_WRITEBACK);
1523 struct btrfs_key found_key;
1524 struct btrfs_file_extent_item *fi;
1525 struct extent_buffer *leaf;
1535 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1541 * If there is no extent for our range when doing the initial
1542 * search, then go back to the previous slot as it will be the
1543 * one containing the search offset
1545 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1546 leaf = path->nodes[0];
1547 btrfs_item_key_to_cpu(leaf, &found_key,
1548 path->slots[0] - 1);
1549 if (found_key.objectid == ino &&
1550 found_key.type == BTRFS_EXTENT_DATA_KEY)
1555 /* Go to next leaf if we have exhausted the current one */
1556 leaf = path->nodes[0];
1557 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1558 ret = btrfs_next_leaf(root, path);
1560 if (cow_start != (u64)-1)
1561 cur_offset = cow_start;
1566 leaf = path->nodes[0];
1569 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1571 /* Didn't find anything for our INO */
1572 if (found_key.objectid > ino)
1575 * Keep searching until we find an EXTENT_ITEM or there are no
1576 * more extents for this inode
1578 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1579 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1584 /* Found key is not EXTENT_DATA_KEY or starts after req range */
1585 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1586 found_key.offset > end)
1590 * If the found extent starts after requested offset, then
1591 * adjust extent_end to be right before this extent begins
1593 if (found_key.offset > cur_offset) {
1594 extent_end = found_key.offset;
1600 * Found extent which begins before our range and potentially
1603 fi = btrfs_item_ptr(leaf, path->slots[0],
1604 struct btrfs_file_extent_item);
1605 extent_type = btrfs_file_extent_type(leaf, fi);
1607 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1608 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1609 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1610 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1611 extent_offset = btrfs_file_extent_offset(leaf, fi);
1612 extent_end = found_key.offset +
1613 btrfs_file_extent_num_bytes(leaf, fi);
1615 btrfs_file_extent_disk_num_bytes(leaf, fi);
1617 * If the extent we got ends before our current offset,
1618 * skip to the next extent.
1620 if (extent_end <= cur_offset) {
1625 if (disk_bytenr == 0)
1627 /* Skip compressed/encrypted/encoded extents */
1628 if (btrfs_file_extent_compression(leaf, fi) ||
1629 btrfs_file_extent_encryption(leaf, fi) ||
1630 btrfs_file_extent_other_encoding(leaf, fi))
1633 * If extent is created before the last volume's snapshot
1634 * this implies the extent is shared, hence we can't do
1635 * nocow. This is the same check as in
1636 * btrfs_cross_ref_exist but without calling
1637 * btrfs_search_slot.
1639 if (!freespace_inode &&
1640 btrfs_file_extent_generation(leaf, fi) <=
1641 btrfs_root_last_snapshot(&root->root_item))
1643 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1647 * The following checks can be expensive, as they need to
1648 * take other locks and do btree or rbtree searches, so
1649 * release the path to avoid blocking other tasks for too
1652 btrfs_release_path(path);
1654 /* If extent is RO, we must COW it */
1655 if (btrfs_extent_readonly(fs_info, disk_bytenr))
1657 ret = btrfs_cross_ref_exist(root, ino,
1659 extent_offset, disk_bytenr, false);
1662 * ret could be -EIO if the above fails to read
1666 if (cow_start != (u64)-1)
1667 cur_offset = cow_start;
1671 WARN_ON_ONCE(freespace_inode);
1674 disk_bytenr += extent_offset;
1675 disk_bytenr += cur_offset - found_key.offset;
1676 num_bytes = min(end + 1, extent_end) - cur_offset;
1678 * If there are pending snapshots for this root, we
1679 * fall into common COW way
1681 if (!freespace_inode && atomic_read(&root->snapshot_force_cow))
1684 * force cow if csum exists in the range.
1685 * this ensure that csum for a given extent are
1686 * either valid or do not exist.
1688 ret = csum_exist_in_range(fs_info, disk_bytenr,
1692 * ret could be -EIO if the above fails to read
1696 if (cow_start != (u64)-1)
1697 cur_offset = cow_start;
1700 WARN_ON_ONCE(freespace_inode);
1703 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr))
1706 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1707 extent_end = found_key.offset + ram_bytes;
1708 extent_end = ALIGN(extent_end, fs_info->sectorsize);
1709 /* Skip extents outside of our requested range */
1710 if (extent_end <= start) {
1715 /* If this triggers then we have a memory corruption */
1720 * If nocow is false then record the beginning of the range
1721 * that needs to be COWed
1724 if (cow_start == (u64)-1)
1725 cow_start = cur_offset;
1726 cur_offset = extent_end;
1727 if (cur_offset > end)
1729 if (!path->nodes[0])
1736 * COW range from cow_start to found_key.offset - 1. As the key
1737 * will contain the beginning of the first extent that can be
1738 * NOCOW, following one which needs to be COW'ed
1740 if (cow_start != (u64)-1) {
1741 ret = fallback_to_cow(inode, locked_page,
1742 cow_start, found_key.offset - 1,
1743 page_started, nr_written);
1746 cow_start = (u64)-1;
1749 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1750 u64 orig_start = found_key.offset - extent_offset;
1751 struct extent_map *em;
1753 em = create_io_em(inode, cur_offset, num_bytes,
1755 disk_bytenr, /* block_start */
1756 num_bytes, /* block_len */
1757 disk_num_bytes, /* orig_block_len */
1758 ram_bytes, BTRFS_COMPRESS_NONE,
1759 BTRFS_ORDERED_PREALLOC);
1764 free_extent_map(em);
1765 ret = btrfs_add_ordered_extent(inode, cur_offset,
1766 disk_bytenr, num_bytes,
1768 BTRFS_ORDERED_PREALLOC);
1770 btrfs_drop_extent_cache(inode, cur_offset,
1771 cur_offset + num_bytes - 1,
1776 ret = btrfs_add_ordered_extent(inode, cur_offset,
1777 disk_bytenr, num_bytes,
1779 BTRFS_ORDERED_NOCOW);
1785 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1788 if (root->root_key.objectid ==
1789 BTRFS_DATA_RELOC_TREE_OBJECTID)
1791 * Error handled later, as we must prevent
1792 * extent_clear_unlock_delalloc() in error handler
1793 * from freeing metadata of created ordered extent.
1795 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1798 extent_clear_unlock_delalloc(inode, cur_offset,
1799 cur_offset + num_bytes - 1,
1800 locked_page, EXTENT_LOCKED |
1802 EXTENT_CLEAR_DATA_RESV,
1803 PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1805 cur_offset = extent_end;
1808 * btrfs_reloc_clone_csums() error, now we're OK to call error
1809 * handler, as metadata for created ordered extent will only
1810 * be freed by btrfs_finish_ordered_io().
1814 if (cur_offset > end)
1817 btrfs_release_path(path);
1819 if (cur_offset <= end && cow_start == (u64)-1)
1820 cow_start = cur_offset;
1822 if (cow_start != (u64)-1) {
1824 ret = fallback_to_cow(inode, locked_page, cow_start, end,
1825 page_started, nr_written);
1832 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1834 if (ret && cur_offset < end)
1835 extent_clear_unlock_delalloc(inode, cur_offset, end,
1836 locked_page, EXTENT_LOCKED |
1837 EXTENT_DELALLOC | EXTENT_DEFRAG |
1838 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1839 PAGE_START_WRITEBACK |
1840 PAGE_END_WRITEBACK);
1841 btrfs_free_path(path);
1845 static inline int need_force_cow(struct btrfs_inode *inode, u64 start, u64 end)
1848 if (!(inode->flags & BTRFS_INODE_NODATACOW) &&
1849 !(inode->flags & BTRFS_INODE_PREALLOC))
1853 * @defrag_bytes is a hint value, no spinlock held here,
1854 * if is not zero, it means the file is defragging.
1855 * Force cow if given extent needs to be defragged.
1857 if (inode->defrag_bytes &&
1858 test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG, 0, NULL))
1865 * Function to process delayed allocation (create CoW) for ranges which are
1866 * being touched for the first time.
1868 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
1869 u64 start, u64 end, int *page_started, unsigned long *nr_written,
1870 struct writeback_control *wbc)
1873 int force_cow = need_force_cow(inode, start, end);
1875 if (inode->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1876 ret = run_delalloc_nocow(inode, locked_page, start, end,
1877 page_started, 1, nr_written);
1878 } else if (inode->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1879 ret = run_delalloc_nocow(inode, locked_page, start, end,
1880 page_started, 0, nr_written);
1881 } else if (!inode_can_compress(inode) ||
1882 !inode_need_compress(inode, start, end)) {
1883 ret = cow_file_range(inode, locked_page, start, end,
1884 page_started, nr_written, 1);
1886 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
1887 ret = cow_file_range_async(inode, wbc, locked_page, start, end,
1888 page_started, nr_written);
1891 btrfs_cleanup_ordered_extents(inode, locked_page, start,
1896 void btrfs_split_delalloc_extent(struct inode *inode,
1897 struct extent_state *orig, u64 split)
1901 /* not delalloc, ignore it */
1902 if (!(orig->state & EXTENT_DELALLOC))
1905 size = orig->end - orig->start + 1;
1906 if (size > BTRFS_MAX_EXTENT_SIZE) {
1911 * See the explanation in btrfs_merge_delalloc_extent, the same
1912 * applies here, just in reverse.
1914 new_size = orig->end - split + 1;
1915 num_extents = count_max_extents(new_size);
1916 new_size = split - orig->start;
1917 num_extents += count_max_extents(new_size);
1918 if (count_max_extents(size) >= num_extents)
1922 spin_lock(&BTRFS_I(inode)->lock);
1923 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1924 spin_unlock(&BTRFS_I(inode)->lock);
1928 * Handle merged delayed allocation extents so we can keep track of new extents
1929 * that are just merged onto old extents, such as when we are doing sequential
1930 * writes, so we can properly account for the metadata space we'll need.
1932 void btrfs_merge_delalloc_extent(struct inode *inode, struct extent_state *new,
1933 struct extent_state *other)
1935 u64 new_size, old_size;
1938 /* not delalloc, ignore it */
1939 if (!(other->state & EXTENT_DELALLOC))
1942 if (new->start > other->start)
1943 new_size = new->end - other->start + 1;
1945 new_size = other->end - new->start + 1;
1947 /* we're not bigger than the max, unreserve the space and go */
1948 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1949 spin_lock(&BTRFS_I(inode)->lock);
1950 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1951 spin_unlock(&BTRFS_I(inode)->lock);
1956 * We have to add up either side to figure out how many extents were
1957 * accounted for before we merged into one big extent. If the number of
1958 * extents we accounted for is <= the amount we need for the new range
1959 * then we can return, otherwise drop. Think of it like this
1963 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1964 * need 2 outstanding extents, on one side we have 1 and the other side
1965 * we have 1 so they are == and we can return. But in this case
1967 * [MAX_SIZE+4k][MAX_SIZE+4k]
1969 * Each range on their own accounts for 2 extents, but merged together
1970 * they are only 3 extents worth of accounting, so we need to drop in
1973 old_size = other->end - other->start + 1;
1974 num_extents = count_max_extents(old_size);
1975 old_size = new->end - new->start + 1;
1976 num_extents += count_max_extents(old_size);
1977 if (count_max_extents(new_size) >= num_extents)
1980 spin_lock(&BTRFS_I(inode)->lock);
1981 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1982 spin_unlock(&BTRFS_I(inode)->lock);
1985 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1986 struct inode *inode)
1988 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1990 spin_lock(&root->delalloc_lock);
1991 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1992 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1993 &root->delalloc_inodes);
1994 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1995 &BTRFS_I(inode)->runtime_flags);
1996 root->nr_delalloc_inodes++;
1997 if (root->nr_delalloc_inodes == 1) {
1998 spin_lock(&fs_info->delalloc_root_lock);
1999 BUG_ON(!list_empty(&root->delalloc_root));
2000 list_add_tail(&root->delalloc_root,
2001 &fs_info->delalloc_roots);
2002 spin_unlock(&fs_info->delalloc_root_lock);
2005 spin_unlock(&root->delalloc_lock);
2009 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
2010 struct btrfs_inode *inode)
2012 struct btrfs_fs_info *fs_info = root->fs_info;
2014 if (!list_empty(&inode->delalloc_inodes)) {
2015 list_del_init(&inode->delalloc_inodes);
2016 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2017 &inode->runtime_flags);
2018 root->nr_delalloc_inodes--;
2019 if (!root->nr_delalloc_inodes) {
2020 ASSERT(list_empty(&root->delalloc_inodes));
2021 spin_lock(&fs_info->delalloc_root_lock);
2022 BUG_ON(list_empty(&root->delalloc_root));
2023 list_del_init(&root->delalloc_root);
2024 spin_unlock(&fs_info->delalloc_root_lock);
2029 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
2030 struct btrfs_inode *inode)
2032 spin_lock(&root->delalloc_lock);
2033 __btrfs_del_delalloc_inode(root, inode);
2034 spin_unlock(&root->delalloc_lock);
2038 * Properly track delayed allocation bytes in the inode and to maintain the
2039 * list of inodes that have pending delalloc work to be done.
2041 void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state,
2044 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2046 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
2049 * set_bit and clear bit hooks normally require _irqsave/restore
2050 * but in this case, we are only testing for the DELALLOC
2051 * bit, which is only set or cleared with irqs on
2053 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2054 struct btrfs_root *root = BTRFS_I(inode)->root;
2055 u64 len = state->end + 1 - state->start;
2056 u32 num_extents = count_max_extents(len);
2057 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
2059 spin_lock(&BTRFS_I(inode)->lock);
2060 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
2061 spin_unlock(&BTRFS_I(inode)->lock);
2063 /* For sanity tests */
2064 if (btrfs_is_testing(fs_info))
2067 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2068 fs_info->delalloc_batch);
2069 spin_lock(&BTRFS_I(inode)->lock);
2070 BTRFS_I(inode)->delalloc_bytes += len;
2071 if (*bits & EXTENT_DEFRAG)
2072 BTRFS_I(inode)->defrag_bytes += len;
2073 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2074 &BTRFS_I(inode)->runtime_flags))
2075 btrfs_add_delalloc_inodes(root, inode);
2076 spin_unlock(&BTRFS_I(inode)->lock);
2079 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2080 (*bits & EXTENT_DELALLOC_NEW)) {
2081 spin_lock(&BTRFS_I(inode)->lock);
2082 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
2084 spin_unlock(&BTRFS_I(inode)->lock);
2089 * Once a range is no longer delalloc this function ensures that proper
2090 * accounting happens.
2092 void btrfs_clear_delalloc_extent(struct inode *vfs_inode,
2093 struct extent_state *state, unsigned *bits)
2095 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
2096 struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb);
2097 u64 len = state->end + 1 - state->start;
2098 u32 num_extents = count_max_extents(len);
2100 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
2101 spin_lock(&inode->lock);
2102 inode->defrag_bytes -= len;
2103 spin_unlock(&inode->lock);
2107 * set_bit and clear bit hooks normally require _irqsave/restore
2108 * but in this case, we are only testing for the DELALLOC
2109 * bit, which is only set or cleared with irqs on
2111 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2112 struct btrfs_root *root = inode->root;
2113 bool do_list = !btrfs_is_free_space_inode(inode);
2115 spin_lock(&inode->lock);
2116 btrfs_mod_outstanding_extents(inode, -num_extents);
2117 spin_unlock(&inode->lock);
2120 * We don't reserve metadata space for space cache inodes so we
2121 * don't need to call delalloc_release_metadata if there is an
2124 if (*bits & EXTENT_CLEAR_META_RESV &&
2125 root != fs_info->tree_root)
2126 btrfs_delalloc_release_metadata(inode, len, false);
2128 /* For sanity tests. */
2129 if (btrfs_is_testing(fs_info))
2132 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
2133 do_list && !(state->state & EXTENT_NORESERVE) &&
2134 (*bits & EXTENT_CLEAR_DATA_RESV))
2135 btrfs_free_reserved_data_space_noquota(fs_info, len);
2137 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2138 fs_info->delalloc_batch);
2139 spin_lock(&inode->lock);
2140 inode->delalloc_bytes -= len;
2141 if (do_list && inode->delalloc_bytes == 0 &&
2142 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2143 &inode->runtime_flags))
2144 btrfs_del_delalloc_inode(root, inode);
2145 spin_unlock(&inode->lock);
2148 if ((state->state & EXTENT_DELALLOC_NEW) &&
2149 (*bits & EXTENT_DELALLOC_NEW)) {
2150 spin_lock(&inode->lock);
2151 ASSERT(inode->new_delalloc_bytes >= len);
2152 inode->new_delalloc_bytes -= len;
2153 if (*bits & EXTENT_ADD_INODE_BYTES)
2154 inode_add_bytes(&inode->vfs_inode, len);
2155 spin_unlock(&inode->lock);
2160 * btrfs_bio_fits_in_stripe - Checks whether the size of the given bio will fit
2161 * in a chunk's stripe. This function ensures that bios do not span a
2164 * @page - The page we are about to add to the bio
2165 * @size - size we want to add to the bio
2166 * @bio - bio we want to ensure is smaller than a stripe
2167 * @bio_flags - flags of the bio
2169 * return 1 if page cannot be added to the bio
2170 * return 0 if page can be added to the bio
2171 * return error otherwise
2173 int btrfs_bio_fits_in_stripe(struct page *page, size_t size, struct bio *bio,
2174 unsigned long bio_flags)
2176 struct inode *inode = page->mapping->host;
2177 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2178 u64 logical = bio->bi_iter.bi_sector << 9;
2179 struct extent_map *em;
2183 struct btrfs_io_geometry geom;
2185 if (bio_flags & EXTENT_BIO_COMPRESSED)
2188 length = bio->bi_iter.bi_size;
2189 map_length = length;
2190 em = btrfs_get_chunk_map(fs_info, logical, map_length);
2193 ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(bio), logical,
2198 if (geom.len < length + size)
2201 free_extent_map(em);
2206 * in order to insert checksums into the metadata in large chunks,
2207 * we wait until bio submission time. All the pages in the bio are
2208 * checksummed and sums are attached onto the ordered extent record.
2210 * At IO completion time the cums attached on the ordered extent record
2211 * are inserted into the btree
2213 static blk_status_t btrfs_submit_bio_start(struct inode *inode, struct bio *bio,
2214 u64 dio_file_offset)
2216 return btrfs_csum_one_bio(BTRFS_I(inode), bio, 0, 0);
2219 bool btrfs_bio_fits_in_ordered_extent(struct page *page, struct bio *bio,
2222 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
2223 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2224 struct btrfs_ordered_extent *ordered;
2225 u64 len = bio->bi_iter.bi_size + size;
2228 ASSERT(btrfs_is_zoned(fs_info));
2229 ASSERT(fs_info->max_zone_append_size > 0);
2230 ASSERT(bio_op(bio) == REQ_OP_ZONE_APPEND);
2232 /* Ordered extent not yet created, so we're good */
2233 ordered = btrfs_lookup_ordered_extent(inode, page_offset(page));
2237 if ((bio->bi_iter.bi_sector << SECTOR_SHIFT) + len >
2238 ordered->disk_bytenr + ordered->disk_num_bytes)
2241 btrfs_put_ordered_extent(ordered);
2246 static blk_status_t extract_ordered_extent(struct btrfs_inode *inode,
2247 struct bio *bio, loff_t file_offset)
2249 struct btrfs_ordered_extent *ordered;
2250 struct extent_map *em = NULL, *em_new = NULL;
2251 struct extent_map_tree *em_tree = &inode->extent_tree;
2252 u64 start = (u64)bio->bi_iter.bi_sector << SECTOR_SHIFT;
2253 u64 len = bio->bi_iter.bi_size;
2254 u64 end = start + len;
2259 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
2260 if (WARN_ON_ONCE(!ordered))
2261 return BLK_STS_IOERR;
2263 /* No need to split */
2264 if (ordered->disk_num_bytes == len)
2267 /* We cannot split once end_bio'd ordered extent */
2268 if (WARN_ON_ONCE(ordered->bytes_left != ordered->disk_num_bytes)) {
2273 /* We cannot split a compressed ordered extent */
2274 if (WARN_ON_ONCE(ordered->disk_num_bytes != ordered->num_bytes)) {
2279 ordered_end = ordered->disk_bytenr + ordered->disk_num_bytes;
2280 /* bio must be in one ordered extent */
2281 if (WARN_ON_ONCE(start < ordered->disk_bytenr || end > ordered_end)) {
2286 /* Checksum list should be empty */
2287 if (WARN_ON_ONCE(!list_empty(&ordered->list))) {
2292 pre = start - ordered->disk_bytenr;
2293 post = ordered_end - end;
2295 ret = btrfs_split_ordered_extent(ordered, pre, post);
2299 read_lock(&em_tree->lock);
2300 em = lookup_extent_mapping(em_tree, ordered->file_offset, len);
2302 read_unlock(&em_tree->lock);
2306 read_unlock(&em_tree->lock);
2308 ASSERT(!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags));
2310 * We cannot reuse em_new here but have to create a new one, as
2311 * unpin_extent_cache() expects the start of the extent map to be the
2312 * logical offset of the file, which does not hold true anymore after
2315 em_new = create_io_em(inode, em->start + pre, len,
2316 em->start + pre, em->block_start + pre, len,
2317 len, len, BTRFS_COMPRESS_NONE,
2318 BTRFS_ORDERED_REGULAR);
2319 if (IS_ERR(em_new)) {
2320 ret = PTR_ERR(em_new);
2323 free_extent_map(em_new);
2326 free_extent_map(em);
2327 btrfs_put_ordered_extent(ordered);
2329 return errno_to_blk_status(ret);
2333 * extent_io.c submission hook. This does the right thing for csum calculation
2334 * on write, or reading the csums from the tree before a read.
2336 * Rules about async/sync submit,
2337 * a) read: sync submit
2339 * b) write without checksum: sync submit
2341 * c) write with checksum:
2342 * c-1) if bio is issued by fsync: sync submit
2343 * (sync_writers != 0)
2345 * c-2) if root is reloc root: sync submit
2346 * (only in case of buffered IO)
2348 * c-3) otherwise: async submit
2350 blk_status_t btrfs_submit_data_bio(struct inode *inode, struct bio *bio,
2351 int mirror_num, unsigned long bio_flags)
2354 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2355 struct btrfs_root *root = BTRFS_I(inode)->root;
2356 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
2357 blk_status_t ret = 0;
2359 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
2361 skip_sum = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) ||
2362 !fs_info->csum_root;
2364 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
2365 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
2367 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
2368 struct page *page = bio_first_bvec_all(bio)->bv_page;
2369 loff_t file_offset = page_offset(page);
2371 ret = extract_ordered_extent(BTRFS_I(inode), bio, file_offset);
2376 if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
2377 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
2381 if (bio_flags & EXTENT_BIO_COMPRESSED) {
2382 ret = btrfs_submit_compressed_read(inode, bio,
2388 * Lookup bio sums does extra checks around whether we
2389 * need to csum or not, which is why we ignore skip_sum
2392 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
2397 } else if (async && !skip_sum) {
2398 /* csum items have already been cloned */
2399 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2401 /* we're doing a write, do the async checksumming */
2402 ret = btrfs_wq_submit_bio(inode, bio, mirror_num, bio_flags,
2403 0, btrfs_submit_bio_start);
2405 } else if (!skip_sum) {
2406 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, 0, 0);
2412 ret = btrfs_map_bio(fs_info, bio, mirror_num);
2416 bio->bi_status = ret;
2423 * given a list of ordered sums record them in the inode. This happens
2424 * at IO completion time based on sums calculated at bio submission time.
2426 static int add_pending_csums(struct btrfs_trans_handle *trans,
2427 struct list_head *list)
2429 struct btrfs_ordered_sum *sum;
2432 list_for_each_entry(sum, list, list) {
2433 trans->adding_csums = true;
2434 ret = btrfs_csum_file_blocks(trans, trans->fs_info->csum_root, sum);
2435 trans->adding_csums = false;
2442 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2445 struct extent_state **cached_state)
2447 u64 search_start = start;
2448 const u64 end = start + len - 1;
2450 while (search_start < end) {
2451 const u64 search_len = end - search_start + 1;
2452 struct extent_map *em;
2456 em = btrfs_get_extent(inode, NULL, 0, search_start, search_len);
2460 if (em->block_start != EXTENT_MAP_HOLE)
2464 if (em->start < search_start)
2465 em_len -= search_start - em->start;
2466 if (em_len > search_len)
2467 em_len = search_len;
2469 ret = set_extent_bit(&inode->io_tree, search_start,
2470 search_start + em_len - 1,
2471 EXTENT_DELALLOC_NEW, 0, NULL, cached_state,
2474 search_start = extent_map_end(em);
2475 free_extent_map(em);
2482 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2483 unsigned int extra_bits,
2484 struct extent_state **cached_state)
2486 WARN_ON(PAGE_ALIGNED(end));
2488 if (start >= i_size_read(&inode->vfs_inode) &&
2489 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2491 * There can't be any extents following eof in this case so just
2492 * set the delalloc new bit for the range directly.
2494 extra_bits |= EXTENT_DELALLOC_NEW;
2498 ret = btrfs_find_new_delalloc_bytes(inode, start,
2505 return set_extent_delalloc(&inode->io_tree, start, end, extra_bits,
2509 /* see btrfs_writepage_start_hook for details on why this is required */
2510 struct btrfs_writepage_fixup {
2512 struct inode *inode;
2513 struct btrfs_work work;
2516 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2518 struct btrfs_writepage_fixup *fixup;
2519 struct btrfs_ordered_extent *ordered;
2520 struct extent_state *cached_state = NULL;
2521 struct extent_changeset *data_reserved = NULL;
2523 struct btrfs_inode *inode;
2527 bool free_delalloc_space = true;
2529 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2531 inode = BTRFS_I(fixup->inode);
2532 page_start = page_offset(page);
2533 page_end = page_offset(page) + PAGE_SIZE - 1;
2536 * This is similar to page_mkwrite, we need to reserve the space before
2537 * we take the page lock.
2539 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2545 * Before we queued this fixup, we took a reference on the page.
2546 * page->mapping may go NULL, but it shouldn't be moved to a different
2549 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2551 * Unfortunately this is a little tricky, either
2553 * 1) We got here and our page had already been dealt with and
2554 * we reserved our space, thus ret == 0, so we need to just
2555 * drop our space reservation and bail. This can happen the
2556 * first time we come into the fixup worker, or could happen
2557 * while waiting for the ordered extent.
2558 * 2) Our page was already dealt with, but we happened to get an
2559 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2560 * this case we obviously don't have anything to release, but
2561 * because the page was already dealt with we don't want to
2562 * mark the page with an error, so make sure we're resetting
2563 * ret to 0. This is why we have this check _before_ the ret
2564 * check, because we do not want to have a surprise ENOSPC
2565 * when the page was already properly dealt with.
2568 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2569 btrfs_delalloc_release_space(inode, data_reserved,
2570 page_start, PAGE_SIZE,
2578 * We can't mess with the page state unless it is locked, so now that
2579 * it is locked bail if we failed to make our space reservation.
2584 lock_extent_bits(&inode->io_tree, page_start, page_end, &cached_state);
2586 /* already ordered? We're done */
2587 if (PagePrivate2(page))
2590 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2592 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2595 btrfs_start_ordered_extent(ordered, 1);
2596 btrfs_put_ordered_extent(ordered);
2600 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2606 * Everything went as planned, we're now the owner of a dirty page with
2607 * delayed allocation bits set and space reserved for our COW
2610 * The page was dirty when we started, nothing should have cleaned it.
2612 BUG_ON(!PageDirty(page));
2613 free_delalloc_space = false;
2615 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2616 if (free_delalloc_space)
2617 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2619 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2624 * We hit ENOSPC or other errors. Update the mapping and page
2625 * to reflect the errors and clean the page.
2627 mapping_set_error(page->mapping, ret);
2628 end_extent_writepage(page, ret, page_start, page_end);
2629 clear_page_dirty_for_io(page);
2632 ClearPageChecked(page);
2636 extent_changeset_free(data_reserved);
2638 * As a precaution, do a delayed iput in case it would be the last iput
2639 * that could need flushing space. Recursing back to fixup worker would
2642 btrfs_add_delayed_iput(&inode->vfs_inode);
2646 * There are a few paths in the higher layers of the kernel that directly
2647 * set the page dirty bit without asking the filesystem if it is a
2648 * good idea. This causes problems because we want to make sure COW
2649 * properly happens and the data=ordered rules are followed.
2651 * In our case any range that doesn't have the ORDERED bit set
2652 * hasn't been properly setup for IO. We kick off an async process
2653 * to fix it up. The async helper will wait for ordered extents, set
2654 * the delalloc bit and make it safe to write the page.
2656 int btrfs_writepage_cow_fixup(struct page *page, u64 start, u64 end)
2658 struct inode *inode = page->mapping->host;
2659 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2660 struct btrfs_writepage_fixup *fixup;
2662 /* this page is properly in the ordered list */
2663 if (TestClearPagePrivate2(page))
2667 * PageChecked is set below when we create a fixup worker for this page,
2668 * don't try to create another one if we're already PageChecked()
2670 * The extent_io writepage code will redirty the page if we send back
2673 if (PageChecked(page))
2676 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2681 * We are already holding a reference to this inode from
2682 * write_cache_pages. We need to hold it because the space reservation
2683 * takes place outside of the page lock, and we can't trust
2684 * page->mapping outside of the page lock.
2687 SetPageChecked(page);
2689 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
2691 fixup->inode = inode;
2692 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2697 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2698 struct btrfs_inode *inode, u64 file_pos,
2699 struct btrfs_file_extent_item *stack_fi,
2700 const bool update_inode_bytes,
2701 u64 qgroup_reserved)
2703 struct btrfs_root *root = inode->root;
2704 const u64 sectorsize = root->fs_info->sectorsize;
2705 struct btrfs_path *path;
2706 struct extent_buffer *leaf;
2707 struct btrfs_key ins;
2708 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
2709 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
2710 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
2711 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
2712 struct btrfs_drop_extents_args drop_args = { 0 };
2715 path = btrfs_alloc_path();
2720 * we may be replacing one extent in the tree with another.
2721 * The new extent is pinned in the extent map, and we don't want
2722 * to drop it from the cache until it is completely in the btree.
2724 * So, tell btrfs_drop_extents to leave this extent in the cache.
2725 * the caller is expected to unpin it and allow it to be merged
2728 drop_args.path = path;
2729 drop_args.start = file_pos;
2730 drop_args.end = file_pos + num_bytes;
2731 drop_args.replace_extent = true;
2732 drop_args.extent_item_size = sizeof(*stack_fi);
2733 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2737 if (!drop_args.extent_inserted) {
2738 ins.objectid = btrfs_ino(inode);
2739 ins.offset = file_pos;
2740 ins.type = BTRFS_EXTENT_DATA_KEY;
2742 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2747 leaf = path->nodes[0];
2748 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
2749 write_extent_buffer(leaf, stack_fi,
2750 btrfs_item_ptr_offset(leaf, path->slots[0]),
2751 sizeof(struct btrfs_file_extent_item));
2753 btrfs_mark_buffer_dirty(leaf);
2754 btrfs_release_path(path);
2757 * If we dropped an inline extent here, we know the range where it is
2758 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
2759 * number of bytes only for that range contaning the inline extent.
2760 * The remaining of the range will be processed when clearning the
2761 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
2763 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
2764 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
2766 inline_size = drop_args.bytes_found - inline_size;
2767 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
2768 drop_args.bytes_found -= inline_size;
2769 num_bytes -= sectorsize;
2772 if (update_inode_bytes)
2773 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
2775 ins.objectid = disk_bytenr;
2776 ins.offset = disk_num_bytes;
2777 ins.type = BTRFS_EXTENT_ITEM_KEY;
2779 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
2783 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
2784 file_pos, qgroup_reserved, &ins);
2786 btrfs_free_path(path);
2791 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2794 struct btrfs_block_group *cache;
2796 cache = btrfs_lookup_block_group(fs_info, start);
2799 spin_lock(&cache->lock);
2800 cache->delalloc_bytes -= len;
2801 spin_unlock(&cache->lock);
2803 btrfs_put_block_group(cache);
2806 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
2807 struct btrfs_ordered_extent *oe)
2809 struct btrfs_file_extent_item stack_fi;
2811 bool update_inode_bytes;
2813 memset(&stack_fi, 0, sizeof(stack_fi));
2814 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
2815 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
2816 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
2817 oe->disk_num_bytes);
2818 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags))
2819 logical_len = oe->truncated_len;
2821 logical_len = oe->num_bytes;
2822 btrfs_set_stack_file_extent_num_bytes(&stack_fi, logical_len);
2823 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, logical_len);
2824 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
2825 /* Encryption and other encoding is reserved and all 0 */
2828 * For delalloc, when completing an ordered extent we update the inode's
2829 * bytes when clearing the range in the inode's io tree, so pass false
2830 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
2831 * except if the ordered extent was truncated.
2833 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
2834 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
2836 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
2837 oe->file_offset, &stack_fi,
2838 update_inode_bytes, oe->qgroup_rsv);
2842 * As ordered data IO finishes, this gets called so we can finish
2843 * an ordered extent if the range of bytes in the file it covers are
2846 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2848 struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
2849 struct btrfs_root *root = inode->root;
2850 struct btrfs_fs_info *fs_info = root->fs_info;
2851 struct btrfs_trans_handle *trans = NULL;
2852 struct extent_io_tree *io_tree = &inode->io_tree;
2853 struct extent_state *cached_state = NULL;
2855 int compress_type = 0;
2857 u64 logical_len = ordered_extent->num_bytes;
2858 bool freespace_inode;
2859 bool truncated = false;
2860 bool clear_reserved_extent = true;
2861 unsigned int clear_bits = EXTENT_DEFRAG;
2863 start = ordered_extent->file_offset;
2864 end = start + ordered_extent->num_bytes - 1;
2866 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2867 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2868 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2869 clear_bits |= EXTENT_DELALLOC_NEW;
2871 freespace_inode = btrfs_is_free_space_inode(inode);
2873 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2878 if (ordered_extent->disk)
2879 btrfs_rewrite_logical_zoned(ordered_extent);
2881 btrfs_free_io_failure_record(inode, start, end);
2883 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2885 logical_len = ordered_extent->truncated_len;
2886 /* Truncated the entire extent, don't bother adding */
2891 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2892 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2894 btrfs_inode_safe_disk_i_size_write(inode, 0);
2895 if (freespace_inode)
2896 trans = btrfs_join_transaction_spacecache(root);
2898 trans = btrfs_join_transaction(root);
2899 if (IS_ERR(trans)) {
2900 ret = PTR_ERR(trans);
2904 trans->block_rsv = &inode->block_rsv;
2905 ret = btrfs_update_inode_fallback(trans, root, inode);
2906 if (ret) /* -ENOMEM or corruption */
2907 btrfs_abort_transaction(trans, ret);
2911 clear_bits |= EXTENT_LOCKED;
2912 lock_extent_bits(io_tree, start, end, &cached_state);
2914 if (freespace_inode)
2915 trans = btrfs_join_transaction_spacecache(root);
2917 trans = btrfs_join_transaction(root);
2918 if (IS_ERR(trans)) {
2919 ret = PTR_ERR(trans);
2924 trans->block_rsv = &inode->block_rsv;
2926 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2927 compress_type = ordered_extent->compress_type;
2928 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2929 BUG_ON(compress_type);
2930 ret = btrfs_mark_extent_written(trans, inode,
2931 ordered_extent->file_offset,
2932 ordered_extent->file_offset +
2935 BUG_ON(root == fs_info->tree_root);
2936 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
2938 clear_reserved_extent = false;
2939 btrfs_release_delalloc_bytes(fs_info,
2940 ordered_extent->disk_bytenr,
2941 ordered_extent->disk_num_bytes);
2944 unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset,
2945 ordered_extent->num_bytes, trans->transid);
2947 btrfs_abort_transaction(trans, ret);
2951 ret = add_pending_csums(trans, &ordered_extent->list);
2953 btrfs_abort_transaction(trans, ret);
2958 * If this is a new delalloc range, clear its new delalloc flag to
2959 * update the inode's number of bytes. This needs to be done first
2960 * before updating the inode item.
2962 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
2963 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
2964 clear_extent_bit(&inode->io_tree, start, end,
2965 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
2966 0, 0, &cached_state);
2968 btrfs_inode_safe_disk_i_size_write(inode, 0);
2969 ret = btrfs_update_inode_fallback(trans, root, inode);
2970 if (ret) { /* -ENOMEM or corruption */
2971 btrfs_abort_transaction(trans, ret);
2976 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
2977 (clear_bits & EXTENT_LOCKED) ? 1 : 0, 0,
2981 btrfs_end_transaction(trans);
2983 if (ret || truncated) {
2984 u64 unwritten_start = start;
2987 unwritten_start += logical_len;
2988 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
2990 /* Drop the cache for the part of the extent we didn't write. */
2991 btrfs_drop_extent_cache(inode, unwritten_start, end, 0);
2994 * If the ordered extent had an IOERR or something else went
2995 * wrong we need to return the space for this ordered extent
2996 * back to the allocator. We only free the extent in the
2997 * truncated case if we didn't write out the extent at all.
2999 * If we made it past insert_reserved_file_extent before we
3000 * errored out then we don't need to do this as the accounting
3001 * has already been done.
3003 if ((ret || !logical_len) &&
3004 clear_reserved_extent &&
3005 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3006 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3008 * Discard the range before returning it back to the
3011 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3012 btrfs_discard_extent(fs_info,
3013 ordered_extent->disk_bytenr,
3014 ordered_extent->disk_num_bytes,
3016 btrfs_free_reserved_extent(fs_info,
3017 ordered_extent->disk_bytenr,
3018 ordered_extent->disk_num_bytes, 1);
3023 * This needs to be done to make sure anybody waiting knows we are done
3024 * updating everything for this ordered extent.
3026 btrfs_remove_ordered_extent(inode, ordered_extent);
3029 btrfs_put_ordered_extent(ordered_extent);
3030 /* once for the tree */
3031 btrfs_put_ordered_extent(ordered_extent);
3036 static void finish_ordered_fn(struct btrfs_work *work)
3038 struct btrfs_ordered_extent *ordered_extent;
3039 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3040 btrfs_finish_ordered_io(ordered_extent);
3043 void btrfs_writepage_endio_finish_ordered(struct page *page, u64 start,
3044 u64 end, int uptodate)
3046 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
3047 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3048 struct btrfs_ordered_extent *ordered_extent = NULL;
3049 struct btrfs_workqueue *wq;
3051 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3053 ClearPagePrivate2(page);
3054 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3055 end - start + 1, uptodate))
3058 if (btrfs_is_free_space_inode(inode))
3059 wq = fs_info->endio_freespace_worker;
3061 wq = fs_info->endio_write_workers;
3063 btrfs_init_work(&ordered_extent->work, finish_ordered_fn, NULL, NULL);
3064 btrfs_queue_work(wq, &ordered_extent->work);
3068 * check_data_csum - verify checksum of one sector of uncompressed data
3070 * @io_bio: btrfs_io_bio which contains the csum
3071 * @bio_offset: offset to the beginning of the bio (in bytes)
3072 * @page: page where is the data to be verified
3073 * @pgoff: offset inside the page
3075 * The length of such check is always one sector size.
3077 static int check_data_csum(struct inode *inode, struct btrfs_io_bio *io_bio,
3078 u32 bio_offset, struct page *page, u32 pgoff)
3080 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3081 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3083 u32 len = fs_info->sectorsize;
3084 const u32 csum_size = fs_info->csum_size;
3085 unsigned int offset_sectors;
3087 u8 csum[BTRFS_CSUM_SIZE];
3089 ASSERT(pgoff + len <= PAGE_SIZE);
3091 offset_sectors = bio_offset >> fs_info->sectorsize_bits;
3092 csum_expected = ((u8 *)io_bio->csum) + offset_sectors * csum_size;
3094 kaddr = kmap_atomic(page);
3095 shash->tfm = fs_info->csum_shash;
3097 crypto_shash_digest(shash, kaddr + pgoff, len, csum);
3099 if (memcmp(csum, csum_expected, csum_size))
3102 kunmap_atomic(kaddr);
3105 btrfs_print_data_csum_error(BTRFS_I(inode), page_offset(page) + pgoff,
3106 csum, csum_expected, io_bio->mirror_num);
3108 btrfs_dev_stat_inc_and_print(io_bio->device,
3109 BTRFS_DEV_STAT_CORRUPTION_ERRS);
3110 memset(kaddr + pgoff, 1, len);
3111 flush_dcache_page(page);
3112 kunmap_atomic(kaddr);
3117 * When reads are done, we need to check csums to verify the data is correct.
3118 * if there's a match, we allow the bio to finish. If not, the code in
3119 * extent_io.c will try to find good copies for us.
3121 * @bio_offset: offset to the beginning of the bio (in bytes)
3122 * @start: file offset of the range start
3123 * @end: file offset of the range end (inclusive)
3124 * @mirror: mirror number
3126 int btrfs_verify_data_csum(struct btrfs_io_bio *io_bio, u32 bio_offset,
3127 struct page *page, u64 start, u64 end, int mirror)
3129 struct inode *inode = page->mapping->host;
3130 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3131 struct btrfs_root *root = BTRFS_I(inode)->root;
3132 const u32 sectorsize = root->fs_info->sectorsize;
3135 if (PageChecked(page)) {
3136 ClearPageChecked(page);
3140 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3143 if (!root->fs_info->csum_root)
3146 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3147 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3148 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3152 ASSERT(page_offset(page) <= start &&
3153 end <= page_offset(page) + PAGE_SIZE - 1);
3154 for (pg_off = offset_in_page(start);
3155 pg_off < offset_in_page(end);
3156 pg_off += sectorsize, bio_offset += sectorsize) {
3159 ret = check_data_csum(inode, io_bio, bio_offset, page, pg_off);
3167 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3169 * @inode: The inode we want to perform iput on
3171 * This function uses the generic vfs_inode::i_count to track whether we should
3172 * just decrement it (in case it's > 1) or if this is the last iput then link
3173 * the inode to the delayed iput machinery. Delayed iputs are processed at
3174 * transaction commit time/superblock commit/cleaner kthread.
3176 void btrfs_add_delayed_iput(struct inode *inode)
3178 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3179 struct btrfs_inode *binode = BTRFS_I(inode);
3181 if (atomic_add_unless(&inode->i_count, -1, 1))
3184 atomic_inc(&fs_info->nr_delayed_iputs);
3185 spin_lock(&fs_info->delayed_iput_lock);
3186 ASSERT(list_empty(&binode->delayed_iput));
3187 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3188 spin_unlock(&fs_info->delayed_iput_lock);
3189 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3190 wake_up_process(fs_info->cleaner_kthread);
3193 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3194 struct btrfs_inode *inode)
3196 list_del_init(&inode->delayed_iput);
3197 spin_unlock(&fs_info->delayed_iput_lock);
3198 iput(&inode->vfs_inode);
3199 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3200 wake_up(&fs_info->delayed_iputs_wait);
3201 spin_lock(&fs_info->delayed_iput_lock);
3204 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3205 struct btrfs_inode *inode)
3207 if (!list_empty(&inode->delayed_iput)) {
3208 spin_lock(&fs_info->delayed_iput_lock);
3209 if (!list_empty(&inode->delayed_iput))
3210 run_delayed_iput_locked(fs_info, inode);
3211 spin_unlock(&fs_info->delayed_iput_lock);
3215 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3218 spin_lock(&fs_info->delayed_iput_lock);
3219 while (!list_empty(&fs_info->delayed_iputs)) {
3220 struct btrfs_inode *inode;
3222 inode = list_first_entry(&fs_info->delayed_iputs,
3223 struct btrfs_inode, delayed_iput);
3224 run_delayed_iput_locked(fs_info, inode);
3226 spin_unlock(&fs_info->delayed_iput_lock);
3230 * Wait for flushing all delayed iputs
3232 * @fs_info: the filesystem
3234 * This will wait on any delayed iputs that are currently running with KILLABLE
3235 * set. Once they are all done running we will return, unless we are killed in
3236 * which case we return EINTR. This helps in user operations like fallocate etc
3237 * that might get blocked on the iputs.
3239 * Return EINTR if we were killed, 0 if nothing's pending
3241 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3243 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3244 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3251 * This creates an orphan entry for the given inode in case something goes wrong
3252 * in the middle of an unlink.
3254 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3255 struct btrfs_inode *inode)
3259 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3260 if (ret && ret != -EEXIST) {
3261 btrfs_abort_transaction(trans, ret);
3269 * We have done the delete so we can go ahead and remove the orphan item for
3270 * this particular inode.
3272 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3273 struct btrfs_inode *inode)
3275 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3279 * this cleans up any orphans that may be left on the list from the last use
3282 int btrfs_orphan_cleanup(struct btrfs_root *root)
3284 struct btrfs_fs_info *fs_info = root->fs_info;
3285 struct btrfs_path *path;
3286 struct extent_buffer *leaf;
3287 struct btrfs_key key, found_key;
3288 struct btrfs_trans_handle *trans;
3289 struct inode *inode;
3290 u64 last_objectid = 0;
3291 int ret = 0, nr_unlink = 0;
3293 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3296 path = btrfs_alloc_path();
3301 path->reada = READA_BACK;
3303 key.objectid = BTRFS_ORPHAN_OBJECTID;
3304 key.type = BTRFS_ORPHAN_ITEM_KEY;
3305 key.offset = (u64)-1;
3308 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3313 * if ret == 0 means we found what we were searching for, which
3314 * is weird, but possible, so only screw with path if we didn't
3315 * find the key and see if we have stuff that matches
3319 if (path->slots[0] == 0)
3324 /* pull out the item */
3325 leaf = path->nodes[0];
3326 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3328 /* make sure the item matches what we want */
3329 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3331 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3334 /* release the path since we're done with it */
3335 btrfs_release_path(path);
3338 * this is where we are basically btrfs_lookup, without the
3339 * crossing root thing. we store the inode number in the
3340 * offset of the orphan item.
3343 if (found_key.offset == last_objectid) {
3345 "Error removing orphan entry, stopping orphan cleanup");
3350 last_objectid = found_key.offset;
3352 found_key.objectid = found_key.offset;
3353 found_key.type = BTRFS_INODE_ITEM_KEY;
3354 found_key.offset = 0;
3355 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3356 ret = PTR_ERR_OR_ZERO(inode);
3357 if (ret && ret != -ENOENT)
3360 if (ret == -ENOENT && root == fs_info->tree_root) {
3361 struct btrfs_root *dead_root;
3362 int is_dead_root = 0;
3365 * this is an orphan in the tree root. Currently these
3366 * could come from 2 sources:
3367 * a) a snapshot deletion in progress
3368 * b) a free space cache inode
3369 * We need to distinguish those two, as the snapshot
3370 * orphan must not get deleted.
3371 * find_dead_roots already ran before us, so if this
3372 * is a snapshot deletion, we should find the root
3373 * in the fs_roots radix tree.
3376 spin_lock(&fs_info->fs_roots_radix_lock);
3377 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3378 (unsigned long)found_key.objectid);
3379 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3381 spin_unlock(&fs_info->fs_roots_radix_lock);
3384 /* prevent this orphan from being found again */
3385 key.offset = found_key.objectid - 1;
3392 * If we have an inode with links, there are a couple of
3393 * possibilities. Old kernels (before v3.12) used to create an
3394 * orphan item for truncate indicating that there were possibly
3395 * extent items past i_size that needed to be deleted. In v3.12,
3396 * truncate was changed to update i_size in sync with the extent
3397 * items, but the (useless) orphan item was still created. Since
3398 * v4.18, we don't create the orphan item for truncate at all.
3400 * So, this item could mean that we need to do a truncate, but
3401 * only if this filesystem was last used on a pre-v3.12 kernel
3402 * and was not cleanly unmounted. The odds of that are quite
3403 * slim, and it's a pain to do the truncate now, so just delete
3406 * It's also possible that this orphan item was supposed to be
3407 * deleted but wasn't. The inode number may have been reused,
3408 * but either way, we can delete the orphan item.
3410 if (ret == -ENOENT || inode->i_nlink) {
3413 trans = btrfs_start_transaction(root, 1);
3414 if (IS_ERR(trans)) {
3415 ret = PTR_ERR(trans);
3418 btrfs_debug(fs_info, "auto deleting %Lu",
3419 found_key.objectid);
3420 ret = btrfs_del_orphan_item(trans, root,
3421 found_key.objectid);
3422 btrfs_end_transaction(trans);
3430 /* this will do delete_inode and everything for us */
3433 /* release the path since we're done with it */
3434 btrfs_release_path(path);
3436 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3438 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3439 trans = btrfs_join_transaction(root);
3441 btrfs_end_transaction(trans);
3445 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3449 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3450 btrfs_free_path(path);
3455 * very simple check to peek ahead in the leaf looking for xattrs. If we
3456 * don't find any xattrs, we know there can't be any acls.
3458 * slot is the slot the inode is in, objectid is the objectid of the inode
3460 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3461 int slot, u64 objectid,
3462 int *first_xattr_slot)
3464 u32 nritems = btrfs_header_nritems(leaf);
3465 struct btrfs_key found_key;
3466 static u64 xattr_access = 0;
3467 static u64 xattr_default = 0;
3470 if (!xattr_access) {
3471 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3472 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3473 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3474 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3478 *first_xattr_slot = -1;
3479 while (slot < nritems) {
3480 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3482 /* we found a different objectid, there must not be acls */
3483 if (found_key.objectid != objectid)
3486 /* we found an xattr, assume we've got an acl */
3487 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3488 if (*first_xattr_slot == -1)
3489 *first_xattr_slot = slot;
3490 if (found_key.offset == xattr_access ||
3491 found_key.offset == xattr_default)
3496 * we found a key greater than an xattr key, there can't
3497 * be any acls later on
3499 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3506 * it goes inode, inode backrefs, xattrs, extents,
3507 * so if there are a ton of hard links to an inode there can
3508 * be a lot of backrefs. Don't waste time searching too hard,
3509 * this is just an optimization
3514 /* we hit the end of the leaf before we found an xattr or
3515 * something larger than an xattr. We have to assume the inode
3518 if (*first_xattr_slot == -1)
3519 *first_xattr_slot = slot;
3524 * read an inode from the btree into the in-memory inode
3526 static int btrfs_read_locked_inode(struct inode *inode,
3527 struct btrfs_path *in_path)
3529 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3530 struct btrfs_path *path = in_path;
3531 struct extent_buffer *leaf;
3532 struct btrfs_inode_item *inode_item;
3533 struct btrfs_root *root = BTRFS_I(inode)->root;
3534 struct btrfs_key location;
3539 bool filled = false;
3540 int first_xattr_slot;
3542 ret = btrfs_fill_inode(inode, &rdev);
3547 path = btrfs_alloc_path();
3552 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3554 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3556 if (path != in_path)
3557 btrfs_free_path(path);
3561 leaf = path->nodes[0];
3566 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3567 struct btrfs_inode_item);
3568 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3569 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3570 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3571 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3572 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3573 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3574 round_up(i_size_read(inode), fs_info->sectorsize));
3576 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3577 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3579 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3580 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3582 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3583 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3585 BTRFS_I(inode)->i_otime.tv_sec =
3586 btrfs_timespec_sec(leaf, &inode_item->otime);
3587 BTRFS_I(inode)->i_otime.tv_nsec =
3588 btrfs_timespec_nsec(leaf, &inode_item->otime);
3590 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3591 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3592 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3594 inode_set_iversion_queried(inode,
3595 btrfs_inode_sequence(leaf, inode_item));
3596 inode->i_generation = BTRFS_I(inode)->generation;
3598 rdev = btrfs_inode_rdev(leaf, inode_item);
3600 BTRFS_I(inode)->index_cnt = (u64)-1;
3601 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3605 * If we were modified in the current generation and evicted from memory
3606 * and then re-read we need to do a full sync since we don't have any
3607 * idea about which extents were modified before we were evicted from
3610 * This is required for both inode re-read from disk and delayed inode
3611 * in delayed_nodes_tree.
3613 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3614 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3615 &BTRFS_I(inode)->runtime_flags);
3618 * We don't persist the id of the transaction where an unlink operation
3619 * against the inode was last made. So here we assume the inode might
3620 * have been evicted, and therefore the exact value of last_unlink_trans
3621 * lost, and set it to last_trans to avoid metadata inconsistencies
3622 * between the inode and its parent if the inode is fsync'ed and the log
3623 * replayed. For example, in the scenario:
3626 * ln mydir/foo mydir/bar
3629 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3630 * xfs_io -c fsync mydir/foo
3632 * mount fs, triggers fsync log replay
3634 * We must make sure that when we fsync our inode foo we also log its
3635 * parent inode, otherwise after log replay the parent still has the
3636 * dentry with the "bar" name but our inode foo has a link count of 1
3637 * and doesn't have an inode ref with the name "bar" anymore.
3639 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3640 * but it guarantees correctness at the expense of occasional full
3641 * transaction commits on fsync if our inode is a directory, or if our
3642 * inode is not a directory, logging its parent unnecessarily.
3644 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3647 * Same logic as for last_unlink_trans. We don't persist the generation
3648 * of the last transaction where this inode was used for a reflink
3649 * operation, so after eviction and reloading the inode we must be
3650 * pessimistic and assume the last transaction that modified the inode.
3652 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3655 if (inode->i_nlink != 1 ||
3656 path->slots[0] >= btrfs_header_nritems(leaf))
3659 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3660 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3663 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3664 if (location.type == BTRFS_INODE_REF_KEY) {
3665 struct btrfs_inode_ref *ref;
3667 ref = (struct btrfs_inode_ref *)ptr;
3668 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3669 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3670 struct btrfs_inode_extref *extref;
3672 extref = (struct btrfs_inode_extref *)ptr;
3673 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3678 * try to precache a NULL acl entry for files that don't have
3679 * any xattrs or acls
3681 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3682 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3683 if (first_xattr_slot != -1) {
3684 path->slots[0] = first_xattr_slot;
3685 ret = btrfs_load_inode_props(inode, path);
3688 "error loading props for ino %llu (root %llu): %d",
3689 btrfs_ino(BTRFS_I(inode)),
3690 root->root_key.objectid, ret);
3692 if (path != in_path)
3693 btrfs_free_path(path);
3696 cache_no_acl(inode);
3698 switch (inode->i_mode & S_IFMT) {
3700 inode->i_mapping->a_ops = &btrfs_aops;
3701 inode->i_fop = &btrfs_file_operations;
3702 inode->i_op = &btrfs_file_inode_operations;
3705 inode->i_fop = &btrfs_dir_file_operations;
3706 inode->i_op = &btrfs_dir_inode_operations;
3709 inode->i_op = &btrfs_symlink_inode_operations;
3710 inode_nohighmem(inode);
3711 inode->i_mapping->a_ops = &btrfs_aops;
3714 inode->i_op = &btrfs_special_inode_operations;
3715 init_special_inode(inode, inode->i_mode, rdev);
3719 btrfs_sync_inode_flags_to_i_flags(inode);
3724 * given a leaf and an inode, copy the inode fields into the leaf
3726 static void fill_inode_item(struct btrfs_trans_handle *trans,
3727 struct extent_buffer *leaf,
3728 struct btrfs_inode_item *item,
3729 struct inode *inode)
3731 struct btrfs_map_token token;
3733 btrfs_init_map_token(&token, leaf);
3735 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3736 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3737 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
3738 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3739 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3741 btrfs_set_token_timespec_sec(&token, &item->atime,
3742 inode->i_atime.tv_sec);
3743 btrfs_set_token_timespec_nsec(&token, &item->atime,
3744 inode->i_atime.tv_nsec);
3746 btrfs_set_token_timespec_sec(&token, &item->mtime,
3747 inode->i_mtime.tv_sec);
3748 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3749 inode->i_mtime.tv_nsec);
3751 btrfs_set_token_timespec_sec(&token, &item->ctime,
3752 inode->i_ctime.tv_sec);
3753 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3754 inode->i_ctime.tv_nsec);
3756 btrfs_set_token_timespec_sec(&token, &item->otime,
3757 BTRFS_I(inode)->i_otime.tv_sec);
3758 btrfs_set_token_timespec_nsec(&token, &item->otime,
3759 BTRFS_I(inode)->i_otime.tv_nsec);
3761 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
3762 btrfs_set_token_inode_generation(&token, item,
3763 BTRFS_I(inode)->generation);
3764 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
3765 btrfs_set_token_inode_transid(&token, item, trans->transid);
3766 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
3767 btrfs_set_token_inode_flags(&token, item, BTRFS_I(inode)->flags);
3768 btrfs_set_token_inode_block_group(&token, item, 0);
3772 * copy everything in the in-memory inode into the btree.
3774 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3775 struct btrfs_root *root,
3776 struct btrfs_inode *inode)
3778 struct btrfs_inode_item *inode_item;
3779 struct btrfs_path *path;
3780 struct extent_buffer *leaf;
3783 path = btrfs_alloc_path();
3787 ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1);
3794 leaf = path->nodes[0];
3795 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3796 struct btrfs_inode_item);
3798 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
3799 btrfs_mark_buffer_dirty(leaf);
3800 btrfs_set_inode_last_trans(trans, inode);
3803 btrfs_free_path(path);
3808 * copy everything in the in-memory inode into the btree.
3810 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3811 struct btrfs_root *root,
3812 struct btrfs_inode *inode)
3814 struct btrfs_fs_info *fs_info = root->fs_info;
3818 * If the inode is a free space inode, we can deadlock during commit
3819 * if we put it into the delayed code.
3821 * The data relocation inode should also be directly updated
3824 if (!btrfs_is_free_space_inode(inode)
3825 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3826 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
3827 btrfs_update_root_times(trans, root);
3829 ret = btrfs_delayed_update_inode(trans, root, inode);
3831 btrfs_set_inode_last_trans(trans, inode);
3835 return btrfs_update_inode_item(trans, root, inode);
3838 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3839 struct btrfs_root *root, struct btrfs_inode *inode)
3843 ret = btrfs_update_inode(trans, root, inode);
3845 return btrfs_update_inode_item(trans, root, inode);
3850 * unlink helper that gets used here in inode.c and in the tree logging
3851 * recovery code. It remove a link in a directory with a given name, and
3852 * also drops the back refs in the inode to the directory
3854 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3855 struct btrfs_root *root,
3856 struct btrfs_inode *dir,
3857 struct btrfs_inode *inode,
3858 const char *name, int name_len)
3860 struct btrfs_fs_info *fs_info = root->fs_info;
3861 struct btrfs_path *path;
3863 struct btrfs_dir_item *di;
3865 u64 ino = btrfs_ino(inode);
3866 u64 dir_ino = btrfs_ino(dir);
3868 path = btrfs_alloc_path();
3874 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3875 name, name_len, -1);
3876 if (IS_ERR_OR_NULL(di)) {
3877 ret = di ? PTR_ERR(di) : -ENOENT;
3880 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3883 btrfs_release_path(path);
3886 * If we don't have dir index, we have to get it by looking up
3887 * the inode ref, since we get the inode ref, remove it directly,
3888 * it is unnecessary to do delayed deletion.
3890 * But if we have dir index, needn't search inode ref to get it.
3891 * Since the inode ref is close to the inode item, it is better
3892 * that we delay to delete it, and just do this deletion when
3893 * we update the inode item.
3895 if (inode->dir_index) {
3896 ret = btrfs_delayed_delete_inode_ref(inode);
3898 index = inode->dir_index;
3903 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3907 "failed to delete reference to %.*s, inode %llu parent %llu",
3908 name_len, name, ino, dir_ino);
3909 btrfs_abort_transaction(trans, ret);
3913 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
3915 btrfs_abort_transaction(trans, ret);
3919 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
3921 if (ret != 0 && ret != -ENOENT) {
3922 btrfs_abort_transaction(trans, ret);
3926 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
3931 btrfs_abort_transaction(trans, ret);
3934 * If we have a pending delayed iput we could end up with the final iput
3935 * being run in btrfs-cleaner context. If we have enough of these built
3936 * up we can end up burning a lot of time in btrfs-cleaner without any
3937 * way to throttle the unlinks. Since we're currently holding a ref on
3938 * the inode we can run the delayed iput here without any issues as the
3939 * final iput won't be done until after we drop the ref we're currently
3942 btrfs_run_delayed_iput(fs_info, inode);
3944 btrfs_free_path(path);
3948 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
3949 inode_inc_iversion(&inode->vfs_inode);
3950 inode_inc_iversion(&dir->vfs_inode);
3951 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
3952 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
3953 ret = btrfs_update_inode(trans, root, dir);
3958 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3959 struct btrfs_root *root,
3960 struct btrfs_inode *dir, struct btrfs_inode *inode,
3961 const char *name, int name_len)
3964 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3966 drop_nlink(&inode->vfs_inode);
3967 ret = btrfs_update_inode(trans, root, inode);
3973 * helper to start transaction for unlink and rmdir.
3975 * unlink and rmdir are special in btrfs, they do not always free space, so
3976 * if we cannot make our reservations the normal way try and see if there is
3977 * plenty of slack room in the global reserve to migrate, otherwise we cannot
3978 * allow the unlink to occur.
3980 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3982 struct btrfs_root *root = BTRFS_I(dir)->root;
3985 * 1 for the possible orphan item
3986 * 1 for the dir item
3987 * 1 for the dir index
3988 * 1 for the inode ref
3991 return btrfs_start_transaction_fallback_global_rsv(root, 5);
3994 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3996 struct btrfs_root *root = BTRFS_I(dir)->root;
3997 struct btrfs_trans_handle *trans;
3998 struct inode *inode = d_inode(dentry);
4001 trans = __unlink_start_trans(dir);
4003 return PTR_ERR(trans);
4005 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4008 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4009 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4010 dentry->d_name.len);
4014 if (inode->i_nlink == 0) {
4015 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4021 btrfs_end_transaction(trans);
4022 btrfs_btree_balance_dirty(root->fs_info);
4026 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4027 struct inode *dir, struct dentry *dentry)
4029 struct btrfs_root *root = BTRFS_I(dir)->root;
4030 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4031 struct btrfs_path *path;
4032 struct extent_buffer *leaf;
4033 struct btrfs_dir_item *di;
4034 struct btrfs_key key;
4035 const char *name = dentry->d_name.name;
4036 int name_len = dentry->d_name.len;
4040 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4042 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4043 objectid = inode->root->root_key.objectid;
4044 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4045 objectid = inode->location.objectid;
4051 path = btrfs_alloc_path();
4055 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4056 name, name_len, -1);
4057 if (IS_ERR_OR_NULL(di)) {
4058 ret = di ? PTR_ERR(di) : -ENOENT;
4062 leaf = path->nodes[0];
4063 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4064 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4065 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4067 btrfs_abort_transaction(trans, ret);
4070 btrfs_release_path(path);
4073 * This is a placeholder inode for a subvolume we didn't have a
4074 * reference to at the time of the snapshot creation. In the meantime
4075 * we could have renamed the real subvol link into our snapshot, so
4076 * depending on btrfs_del_root_ref to return -ENOENT here is incorret.
4077 * Instead simply lookup the dir_index_item for this entry so we can
4078 * remove it. Otherwise we know we have a ref to the root and we can
4079 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4081 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4082 di = btrfs_search_dir_index_item(root, path, dir_ino,
4084 if (IS_ERR_OR_NULL(di)) {
4089 btrfs_abort_transaction(trans, ret);
4093 leaf = path->nodes[0];
4094 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4096 btrfs_release_path(path);
4098 ret = btrfs_del_root_ref(trans, objectid,
4099 root->root_key.objectid, dir_ino,
4100 &index, name, name_len);
4102 btrfs_abort_transaction(trans, ret);
4107 ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index);
4109 btrfs_abort_transaction(trans, ret);
4113 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4114 inode_inc_iversion(dir);
4115 dir->i_mtime = dir->i_ctime = current_time(dir);
4116 ret = btrfs_update_inode_fallback(trans, root, BTRFS_I(dir));
4118 btrfs_abort_transaction(trans, ret);
4120 btrfs_free_path(path);
4125 * Helper to check if the subvolume references other subvolumes or if it's
4128 static noinline int may_destroy_subvol(struct btrfs_root *root)
4130 struct btrfs_fs_info *fs_info = root->fs_info;
4131 struct btrfs_path *path;
4132 struct btrfs_dir_item *di;
4133 struct btrfs_key key;
4137 path = btrfs_alloc_path();
4141 /* Make sure this root isn't set as the default subvol */
4142 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4143 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4144 dir_id, "default", 7, 0);
4145 if (di && !IS_ERR(di)) {
4146 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4147 if (key.objectid == root->root_key.objectid) {
4150 "deleting default subvolume %llu is not allowed",
4154 btrfs_release_path(path);
4157 key.objectid = root->root_key.objectid;
4158 key.type = BTRFS_ROOT_REF_KEY;
4159 key.offset = (u64)-1;
4161 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4167 if (path->slots[0] > 0) {
4169 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4170 if (key.objectid == root->root_key.objectid &&
4171 key.type == BTRFS_ROOT_REF_KEY)
4175 btrfs_free_path(path);
4179 /* Delete all dentries for inodes belonging to the root */
4180 static void btrfs_prune_dentries(struct btrfs_root *root)
4182 struct btrfs_fs_info *fs_info = root->fs_info;
4183 struct rb_node *node;
4184 struct rb_node *prev;
4185 struct btrfs_inode *entry;
4186 struct inode *inode;
4189 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
4190 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4192 spin_lock(&root->inode_lock);
4194 node = root->inode_tree.rb_node;
4198 entry = rb_entry(node, struct btrfs_inode, rb_node);
4200 if (objectid < btrfs_ino(entry))
4201 node = node->rb_left;
4202 else if (objectid > btrfs_ino(entry))
4203 node = node->rb_right;
4209 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4210 if (objectid <= btrfs_ino(entry)) {
4214 prev = rb_next(prev);
4218 entry = rb_entry(node, struct btrfs_inode, rb_node);
4219 objectid = btrfs_ino(entry) + 1;
4220 inode = igrab(&entry->vfs_inode);
4222 spin_unlock(&root->inode_lock);
4223 if (atomic_read(&inode->i_count) > 1)
4224 d_prune_aliases(inode);
4226 * btrfs_drop_inode will have it removed from the inode
4227 * cache when its usage count hits zero.
4231 spin_lock(&root->inode_lock);
4235 if (cond_resched_lock(&root->inode_lock))
4238 node = rb_next(node);
4240 spin_unlock(&root->inode_lock);
4243 int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
4245 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4246 struct btrfs_root *root = BTRFS_I(dir)->root;
4247 struct inode *inode = d_inode(dentry);
4248 struct btrfs_root *dest = BTRFS_I(inode)->root;
4249 struct btrfs_trans_handle *trans;
4250 struct btrfs_block_rsv block_rsv;
4255 * Don't allow to delete a subvolume with send in progress. This is
4256 * inside the inode lock so the error handling that has to drop the bit
4257 * again is not run concurrently.
4259 spin_lock(&dest->root_item_lock);
4260 if (dest->send_in_progress) {
4261 spin_unlock(&dest->root_item_lock);
4263 "attempt to delete subvolume %llu during send",
4264 dest->root_key.objectid);
4267 root_flags = btrfs_root_flags(&dest->root_item);
4268 btrfs_set_root_flags(&dest->root_item,
4269 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4270 spin_unlock(&dest->root_item_lock);
4272 down_write(&fs_info->subvol_sem);
4274 ret = may_destroy_subvol(dest);
4278 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4280 * One for dir inode,
4281 * two for dir entries,
4282 * two for root ref/backref.
4284 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4288 trans = btrfs_start_transaction(root, 0);
4289 if (IS_ERR(trans)) {
4290 ret = PTR_ERR(trans);
4293 trans->block_rsv = &block_rsv;
4294 trans->bytes_reserved = block_rsv.size;
4296 btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
4298 ret = btrfs_unlink_subvol(trans, dir, dentry);
4300 btrfs_abort_transaction(trans, ret);
4304 btrfs_record_root_in_trans(trans, dest);
4306 memset(&dest->root_item.drop_progress, 0,
4307 sizeof(dest->root_item.drop_progress));
4308 btrfs_set_root_drop_level(&dest->root_item, 0);
4309 btrfs_set_root_refs(&dest->root_item, 0);
4311 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4312 ret = btrfs_insert_orphan_item(trans,
4314 dest->root_key.objectid);
4316 btrfs_abort_transaction(trans, ret);
4321 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4322 BTRFS_UUID_KEY_SUBVOL,
4323 dest->root_key.objectid);
4324 if (ret && ret != -ENOENT) {
4325 btrfs_abort_transaction(trans, ret);
4328 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4329 ret = btrfs_uuid_tree_remove(trans,
4330 dest->root_item.received_uuid,
4331 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4332 dest->root_key.objectid);
4333 if (ret && ret != -ENOENT) {
4334 btrfs_abort_transaction(trans, ret);
4339 free_anon_bdev(dest->anon_dev);
4342 trans->block_rsv = NULL;
4343 trans->bytes_reserved = 0;
4344 ret = btrfs_end_transaction(trans);
4345 inode->i_flags |= S_DEAD;
4347 btrfs_subvolume_release_metadata(root, &block_rsv);
4349 up_write(&fs_info->subvol_sem);
4351 spin_lock(&dest->root_item_lock);
4352 root_flags = btrfs_root_flags(&dest->root_item);
4353 btrfs_set_root_flags(&dest->root_item,
4354 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4355 spin_unlock(&dest->root_item_lock);
4357 d_invalidate(dentry);
4358 btrfs_prune_dentries(dest);
4359 ASSERT(dest->send_in_progress == 0);
4365 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4367 struct inode *inode = d_inode(dentry);
4369 struct btrfs_root *root = BTRFS_I(dir)->root;
4370 struct btrfs_trans_handle *trans;
4371 u64 last_unlink_trans;
4373 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4375 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4376 return btrfs_delete_subvolume(dir, dentry);
4378 trans = __unlink_start_trans(dir);
4380 return PTR_ERR(trans);
4382 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4383 err = btrfs_unlink_subvol(trans, dir, dentry);
4387 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4391 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4393 /* now the directory is empty */
4394 err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4395 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4396 dentry->d_name.len);
4398 btrfs_i_size_write(BTRFS_I(inode), 0);
4400 * Propagate the last_unlink_trans value of the deleted dir to
4401 * its parent directory. This is to prevent an unrecoverable
4402 * log tree in the case we do something like this:
4404 * 2) create snapshot under dir foo
4405 * 3) delete the snapshot
4408 * 6) fsync foo or some file inside foo
4410 if (last_unlink_trans >= trans->transid)
4411 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4414 btrfs_end_transaction(trans);
4415 btrfs_btree_balance_dirty(root->fs_info);
4421 * Return this if we need to call truncate_block for the last bit of the
4424 #define NEED_TRUNCATE_BLOCK 1
4427 * this can truncate away extent items, csum items and directory items.
4428 * It starts at a high offset and removes keys until it can't find
4429 * any higher than new_size
4431 * csum items that cross the new i_size are truncated to the new size
4434 * min_type is the minimum key type to truncate down to. If set to 0, this
4435 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4437 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4438 struct btrfs_root *root,
4439 struct btrfs_inode *inode,
4440 u64 new_size, u32 min_type)
4442 struct btrfs_fs_info *fs_info = root->fs_info;
4443 struct btrfs_path *path;
4444 struct extent_buffer *leaf;
4445 struct btrfs_file_extent_item *fi;
4446 struct btrfs_key key;
4447 struct btrfs_key found_key;
4448 u64 extent_start = 0;
4449 u64 extent_num_bytes = 0;
4450 u64 extent_offset = 0;
4452 u64 last_size = new_size;
4453 u32 found_type = (u8)-1;
4456 int pending_del_nr = 0;
4457 int pending_del_slot = 0;
4458 int extent_type = -1;
4460 u64 ino = btrfs_ino(inode);
4461 u64 bytes_deleted = 0;
4462 bool be_nice = false;
4463 bool should_throttle = false;
4464 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
4465 struct extent_state *cached_state = NULL;
4467 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4470 * For non-free space inodes and non-shareable roots, we want to back
4471 * off from time to time. This means all inodes in subvolume roots,
4472 * reloc roots, and data reloc roots.
4474 if (!btrfs_is_free_space_inode(inode) &&
4475 test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
4478 path = btrfs_alloc_path();
4481 path->reada = READA_BACK;
4483 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4484 lock_extent_bits(&inode->io_tree, lock_start, (u64)-1,
4488 * We want to drop from the next block forward in case this
4489 * new size is not block aligned since we will be keeping the
4490 * last block of the extent just the way it is.
4492 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4493 fs_info->sectorsize),
4498 * This function is also used to drop the items in the log tree before
4499 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4500 * it is used to drop the logged items. So we shouldn't kill the delayed
4503 if (min_type == 0 && root == inode->root)
4504 btrfs_kill_delayed_inode_items(inode);
4507 key.offset = (u64)-1;
4512 * with a 16K leaf size and 128MB extents, you can actually queue
4513 * up a huge file in a single leaf. Most of the time that
4514 * bytes_deleted is > 0, it will be huge by the time we get here
4516 if (be_nice && bytes_deleted > SZ_32M &&
4517 btrfs_should_end_transaction(trans)) {
4522 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4528 /* there are no items in the tree for us to truncate, we're
4531 if (path->slots[0] == 0)
4537 u64 clear_start = 0, clear_len = 0;
4540 leaf = path->nodes[0];
4541 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4542 found_type = found_key.type;
4544 if (found_key.objectid != ino)
4547 if (found_type < min_type)
4550 item_end = found_key.offset;
4551 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4552 fi = btrfs_item_ptr(leaf, path->slots[0],
4553 struct btrfs_file_extent_item);
4554 extent_type = btrfs_file_extent_type(leaf, fi);
4555 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4557 btrfs_file_extent_num_bytes(leaf, fi);
4559 trace_btrfs_truncate_show_fi_regular(
4560 inode, leaf, fi, found_key.offset);
4561 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4562 item_end += btrfs_file_extent_ram_bytes(leaf,
4565 trace_btrfs_truncate_show_fi_inline(
4566 inode, leaf, fi, path->slots[0],
4571 if (found_type > min_type) {
4574 if (item_end < new_size)
4576 if (found_key.offset >= new_size)
4582 /* FIXME, shrink the extent if the ref count is only 1 */
4583 if (found_type != BTRFS_EXTENT_DATA_KEY)
4586 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4589 clear_start = found_key.offset;
4590 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4592 u64 orig_num_bytes =
4593 btrfs_file_extent_num_bytes(leaf, fi);
4594 extent_num_bytes = ALIGN(new_size -
4596 fs_info->sectorsize);
4597 clear_start = ALIGN(new_size, fs_info->sectorsize);
4598 btrfs_set_file_extent_num_bytes(leaf, fi,
4600 num_dec = (orig_num_bytes -
4602 if (test_bit(BTRFS_ROOT_SHAREABLE,
4605 inode_sub_bytes(&inode->vfs_inode,
4607 btrfs_mark_buffer_dirty(leaf);
4610 btrfs_file_extent_disk_num_bytes(leaf,
4612 extent_offset = found_key.offset -
4613 btrfs_file_extent_offset(leaf, fi);
4615 /* FIXME blocksize != 4096 */
4616 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4617 if (extent_start != 0) {
4619 if (test_bit(BTRFS_ROOT_SHAREABLE,
4621 inode_sub_bytes(&inode->vfs_inode,
4625 clear_len = num_dec;
4626 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4628 * we can't truncate inline items that have had
4632 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4633 btrfs_file_extent_other_encoding(leaf, fi) == 0 &&
4634 btrfs_file_extent_compression(leaf, fi) == 0) {
4635 u32 size = (u32)(new_size - found_key.offset);
4637 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4638 size = btrfs_file_extent_calc_inline_size(size);
4639 btrfs_truncate_item(path, size, 1);
4640 } else if (!del_item) {
4642 * We have to bail so the last_size is set to
4643 * just before this extent.
4645 ret = NEED_TRUNCATE_BLOCK;
4649 * Inline extents are special, we just treat
4650 * them as a full sector worth in the file
4651 * extent tree just for simplicity sake.
4653 clear_len = fs_info->sectorsize;
4656 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
4657 inode_sub_bytes(&inode->vfs_inode,
4658 item_end + 1 - new_size);
4662 * We use btrfs_truncate_inode_items() to clean up log trees for
4663 * multiple fsyncs, and in this case we don't want to clear the
4664 * file extent range because it's just the log.
4666 if (root == inode->root) {
4667 ret = btrfs_inode_clear_file_extent_range(inode,
4668 clear_start, clear_len);
4670 btrfs_abort_transaction(trans, ret);
4676 last_size = found_key.offset;
4678 last_size = new_size;
4680 if (!pending_del_nr) {
4681 /* no pending yet, add ourselves */
4682 pending_del_slot = path->slots[0];
4684 } else if (pending_del_nr &&
4685 path->slots[0] + 1 == pending_del_slot) {
4686 /* hop on the pending chunk */
4688 pending_del_slot = path->slots[0];
4695 should_throttle = false;
4698 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4699 struct btrfs_ref ref = { 0 };
4701 bytes_deleted += extent_num_bytes;
4703 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF,
4704 extent_start, extent_num_bytes, 0);
4705 ref.real_root = root->root_key.objectid;
4706 btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
4707 ino, extent_offset);
4708 ret = btrfs_free_extent(trans, &ref);
4710 btrfs_abort_transaction(trans, ret);
4714 if (btrfs_should_throttle_delayed_refs(trans))
4715 should_throttle = true;
4719 if (found_type == BTRFS_INODE_ITEM_KEY)
4722 if (path->slots[0] == 0 ||
4723 path->slots[0] != pending_del_slot ||
4725 if (pending_del_nr) {
4726 ret = btrfs_del_items(trans, root, path,
4730 btrfs_abort_transaction(trans, ret);
4735 btrfs_release_path(path);
4738 * We can generate a lot of delayed refs, so we need to
4739 * throttle every once and a while and make sure we're
4740 * adding enough space to keep up with the work we are
4741 * generating. Since we hold a transaction here we
4742 * can't flush, and we don't want to FLUSH_LIMIT because
4743 * we could have generated too many delayed refs to
4744 * actually allocate, so just bail if we're short and
4745 * let the normal reservation dance happen higher up.
4747 if (should_throttle) {
4748 ret = btrfs_delayed_refs_rsv_refill(fs_info,
4749 BTRFS_RESERVE_NO_FLUSH);
4761 if (ret >= 0 && pending_del_nr) {
4764 err = btrfs_del_items(trans, root, path, pending_del_slot,
4767 btrfs_abort_transaction(trans, err);
4771 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4772 ASSERT(last_size >= new_size);
4773 if (!ret && last_size > new_size)
4774 last_size = new_size;
4775 btrfs_inode_safe_disk_i_size_write(inode, last_size);
4776 unlock_extent_cached(&inode->io_tree, lock_start, (u64)-1,
4780 btrfs_free_path(path);
4785 * btrfs_truncate_block - read, zero a chunk and write a block
4786 * @inode - inode that we're zeroing
4787 * @from - the offset to start zeroing
4788 * @len - the length to zero, 0 to zero the entire range respective to the
4790 * @front - zero up to the offset instead of from the offset on
4792 * This will find the block for the "from" offset and cow the block and zero the
4793 * part we want to zero. This is used with truncate and hole punching.
4795 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4798 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4799 struct address_space *mapping = inode->vfs_inode.i_mapping;
4800 struct extent_io_tree *io_tree = &inode->io_tree;
4801 struct btrfs_ordered_extent *ordered;
4802 struct extent_state *cached_state = NULL;
4803 struct extent_changeset *data_reserved = NULL;
4805 bool only_release_metadata = false;
4806 u32 blocksize = fs_info->sectorsize;
4807 pgoff_t index = from >> PAGE_SHIFT;
4808 unsigned offset = from & (blocksize - 1);
4810 gfp_t mask = btrfs_alloc_write_mask(mapping);
4811 size_t write_bytes = blocksize;
4816 if (IS_ALIGNED(offset, blocksize) &&
4817 (!len || IS_ALIGNED(len, blocksize)))
4820 block_start = round_down(from, blocksize);
4821 block_end = block_start + blocksize - 1;
4823 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4826 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes) > 0) {
4827 /* For nocow case, no need to reserve data space */
4828 only_release_metadata = true;
4833 ret = btrfs_delalloc_reserve_metadata(inode, blocksize);
4835 if (!only_release_metadata)
4836 btrfs_free_reserved_data_space(inode, data_reserved,
4837 block_start, blocksize);
4841 page = find_or_create_page(mapping, index, mask);
4843 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4845 btrfs_delalloc_release_extents(inode, blocksize);
4849 ret = set_page_extent_mapped(page);
4853 if (!PageUptodate(page)) {
4854 ret = btrfs_readpage(NULL, page);
4856 if (page->mapping != mapping) {
4861 if (!PageUptodate(page)) {
4866 wait_on_page_writeback(page);
4868 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4870 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4872 unlock_extent_cached(io_tree, block_start, block_end,
4876 btrfs_start_ordered_extent(ordered, 1);
4877 btrfs_put_ordered_extent(ordered);
4881 clear_extent_bit(&inode->io_tree, block_start, block_end,
4882 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4883 0, 0, &cached_state);
4885 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4888 unlock_extent_cached(io_tree, block_start, block_end,
4893 if (offset != blocksize) {
4895 len = blocksize - offset;
4898 memset(kaddr + (block_start - page_offset(page)),
4901 memset(kaddr + (block_start - page_offset(page)) + offset,
4903 flush_dcache_page(page);
4906 ClearPageChecked(page);
4907 set_page_dirty(page);
4908 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4910 if (only_release_metadata)
4911 set_extent_bit(&inode->io_tree, block_start, block_end,
4912 EXTENT_NORESERVE, 0, NULL, NULL, GFP_NOFS, NULL);
4916 if (only_release_metadata)
4917 btrfs_delalloc_release_metadata(inode, blocksize, true);
4919 btrfs_delalloc_release_space(inode, data_reserved,
4920 block_start, blocksize, true);
4922 btrfs_delalloc_release_extents(inode, blocksize);
4926 if (only_release_metadata)
4927 btrfs_check_nocow_unlock(inode);
4928 extent_changeset_free(data_reserved);
4932 static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode,
4933 u64 offset, u64 len)
4935 struct btrfs_fs_info *fs_info = root->fs_info;
4936 struct btrfs_trans_handle *trans;
4937 struct btrfs_drop_extents_args drop_args = { 0 };
4941 * Still need to make sure the inode looks like it's been updated so
4942 * that any holes get logged if we fsync.
4944 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4945 inode->last_trans = fs_info->generation;
4946 inode->last_sub_trans = root->log_transid;
4947 inode->last_log_commit = root->last_log_commit;
4952 * 1 - for the one we're dropping
4953 * 1 - for the one we're adding
4954 * 1 - for updating the inode.
4956 trans = btrfs_start_transaction(root, 3);
4958 return PTR_ERR(trans);
4960 drop_args.start = offset;
4961 drop_args.end = offset + len;
4962 drop_args.drop_cache = true;
4964 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
4966 btrfs_abort_transaction(trans, ret);
4967 btrfs_end_transaction(trans);
4971 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode),
4972 offset, 0, 0, len, 0, len, 0, 0, 0);
4974 btrfs_abort_transaction(trans, ret);
4976 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
4977 btrfs_update_inode(trans, root, inode);
4979 btrfs_end_transaction(trans);
4984 * This function puts in dummy file extents for the area we're creating a hole
4985 * for. So if we are truncating this file to a larger size we need to insert
4986 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4987 * the range between oldsize and size
4989 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
4991 struct btrfs_root *root = inode->root;
4992 struct btrfs_fs_info *fs_info = root->fs_info;
4993 struct extent_io_tree *io_tree = &inode->io_tree;
4994 struct extent_map *em = NULL;
4995 struct extent_state *cached_state = NULL;
4996 struct extent_map_tree *em_tree = &inode->extent_tree;
4997 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4998 u64 block_end = ALIGN(size, fs_info->sectorsize);
5005 * If our size started in the middle of a block we need to zero out the
5006 * rest of the block before we expand the i_size, otherwise we could
5007 * expose stale data.
5009 err = btrfs_truncate_block(inode, oldsize, 0, 0);
5013 if (size <= hole_start)
5016 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
5018 cur_offset = hole_start;
5020 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
5021 block_end - cur_offset);
5027 last_byte = min(extent_map_end(em), block_end);
5028 last_byte = ALIGN(last_byte, fs_info->sectorsize);
5029 hole_size = last_byte - cur_offset;
5031 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5032 struct extent_map *hole_em;
5034 err = maybe_insert_hole(root, inode, cur_offset,
5039 err = btrfs_inode_set_file_extent_range(inode,
5040 cur_offset, hole_size);
5044 btrfs_drop_extent_cache(inode, cur_offset,
5045 cur_offset + hole_size - 1, 0);
5046 hole_em = alloc_extent_map();
5048 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5049 &inode->runtime_flags);
5052 hole_em->start = cur_offset;
5053 hole_em->len = hole_size;
5054 hole_em->orig_start = cur_offset;
5056 hole_em->block_start = EXTENT_MAP_HOLE;
5057 hole_em->block_len = 0;
5058 hole_em->orig_block_len = 0;
5059 hole_em->ram_bytes = hole_size;
5060 hole_em->compress_type = BTRFS_COMPRESS_NONE;
5061 hole_em->generation = fs_info->generation;
5064 write_lock(&em_tree->lock);
5065 err = add_extent_mapping(em_tree, hole_em, 1);
5066 write_unlock(&em_tree->lock);
5069 btrfs_drop_extent_cache(inode, cur_offset,
5073 free_extent_map(hole_em);
5075 err = btrfs_inode_set_file_extent_range(inode,
5076 cur_offset, hole_size);
5081 free_extent_map(em);
5083 cur_offset = last_byte;
5084 if (cur_offset >= block_end)
5087 free_extent_map(em);
5088 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
5092 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5094 struct btrfs_root *root = BTRFS_I(inode)->root;
5095 struct btrfs_trans_handle *trans;
5096 loff_t oldsize = i_size_read(inode);
5097 loff_t newsize = attr->ia_size;
5098 int mask = attr->ia_valid;
5102 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5103 * special case where we need to update the times despite not having
5104 * these flags set. For all other operations the VFS set these flags
5105 * explicitly if it wants a timestamp update.
5107 if (newsize != oldsize) {
5108 inode_inc_iversion(inode);
5109 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5110 inode->i_ctime = inode->i_mtime =
5111 current_time(inode);
5114 if (newsize > oldsize) {
5116 * Don't do an expanding truncate while snapshotting is ongoing.
5117 * This is to ensure the snapshot captures a fully consistent
5118 * state of this file - if the snapshot captures this expanding
5119 * truncation, it must capture all writes that happened before
5122 btrfs_drew_write_lock(&root->snapshot_lock);
5123 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
5125 btrfs_drew_write_unlock(&root->snapshot_lock);
5129 trans = btrfs_start_transaction(root, 1);
5130 if (IS_ERR(trans)) {
5131 btrfs_drew_write_unlock(&root->snapshot_lock);
5132 return PTR_ERR(trans);
5135 i_size_write(inode, newsize);
5136 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
5137 pagecache_isize_extended(inode, oldsize, newsize);
5138 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5139 btrfs_drew_write_unlock(&root->snapshot_lock);
5140 btrfs_end_transaction(trans);
5144 * We're truncating a file that used to have good data down to
5145 * zero. Make sure any new writes to the file get on disk
5149 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5150 &BTRFS_I(inode)->runtime_flags);
5152 truncate_setsize(inode, newsize);
5154 inode_dio_wait(inode);
5156 ret = btrfs_truncate(inode, newsize == oldsize);
5157 if (ret && inode->i_nlink) {
5161 * Truncate failed, so fix up the in-memory size. We
5162 * adjusted disk_i_size down as we removed extents, so
5163 * wait for disk_i_size to be stable and then update the
5164 * in-memory size to match.
5166 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5169 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5176 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5178 struct inode *inode = d_inode(dentry);
5179 struct btrfs_root *root = BTRFS_I(inode)->root;
5182 if (btrfs_root_readonly(root))
5185 err = setattr_prepare(dentry, attr);
5189 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5190 err = btrfs_setsize(inode, attr);
5195 if (attr->ia_valid) {
5196 setattr_copy(inode, attr);
5197 inode_inc_iversion(inode);
5198 err = btrfs_dirty_inode(inode);
5200 if (!err && attr->ia_valid & ATTR_MODE)
5201 err = posix_acl_chmod(inode, inode->i_mode);
5208 * While truncating the inode pages during eviction, we get the VFS calling
5209 * btrfs_invalidatepage() against each page of the inode. This is slow because
5210 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5211 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5212 * extent_state structures over and over, wasting lots of time.
5214 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5215 * those expensive operations on a per page basis and do only the ordered io
5216 * finishing, while we release here the extent_map and extent_state structures,
5217 * without the excessive merging and splitting.
5219 static void evict_inode_truncate_pages(struct inode *inode)
5221 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5222 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5223 struct rb_node *node;
5225 ASSERT(inode->i_state & I_FREEING);
5226 truncate_inode_pages_final(&inode->i_data);
5228 write_lock(&map_tree->lock);
5229 while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) {
5230 struct extent_map *em;
5232 node = rb_first_cached(&map_tree->map);
5233 em = rb_entry(node, struct extent_map, rb_node);
5234 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5235 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5236 remove_extent_mapping(map_tree, em);
5237 free_extent_map(em);
5238 if (need_resched()) {
5239 write_unlock(&map_tree->lock);
5241 write_lock(&map_tree->lock);
5244 write_unlock(&map_tree->lock);
5247 * Keep looping until we have no more ranges in the io tree.
5248 * We can have ongoing bios started by readahead that have
5249 * their endio callback (extent_io.c:end_bio_extent_readpage)
5250 * still in progress (unlocked the pages in the bio but did not yet
5251 * unlocked the ranges in the io tree). Therefore this means some
5252 * ranges can still be locked and eviction started because before
5253 * submitting those bios, which are executed by a separate task (work
5254 * queue kthread), inode references (inode->i_count) were not taken
5255 * (which would be dropped in the end io callback of each bio).
5256 * Therefore here we effectively end up waiting for those bios and
5257 * anyone else holding locked ranges without having bumped the inode's
5258 * reference count - if we don't do it, when they access the inode's
5259 * io_tree to unlock a range it may be too late, leading to an
5260 * use-after-free issue.
5262 spin_lock(&io_tree->lock);
5263 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5264 struct extent_state *state;
5265 struct extent_state *cached_state = NULL;
5268 unsigned state_flags;
5270 node = rb_first(&io_tree->state);
5271 state = rb_entry(node, struct extent_state, rb_node);
5272 start = state->start;
5274 state_flags = state->state;
5275 spin_unlock(&io_tree->lock);
5277 lock_extent_bits(io_tree, start, end, &cached_state);
5280 * If still has DELALLOC flag, the extent didn't reach disk,
5281 * and its reserved space won't be freed by delayed_ref.
5282 * So we need to free its reserved space here.
5283 * (Refer to comment in btrfs_invalidatepage, case 2)
5285 * Note, end is the bytenr of last byte, so we need + 1 here.
5287 if (state_flags & EXTENT_DELALLOC)
5288 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5291 clear_extent_bit(io_tree, start, end,
5292 EXTENT_LOCKED | EXTENT_DELALLOC |
5293 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
5297 spin_lock(&io_tree->lock);
5299 spin_unlock(&io_tree->lock);
5302 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5303 struct btrfs_block_rsv *rsv)
5305 struct btrfs_fs_info *fs_info = root->fs_info;
5306 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5307 struct btrfs_trans_handle *trans;
5308 u64 delayed_refs_extra = btrfs_calc_insert_metadata_size(fs_info, 1);
5312 * Eviction should be taking place at some place safe because of our
5313 * delayed iputs. However the normal flushing code will run delayed
5314 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5316 * We reserve the delayed_refs_extra here again because we can't use
5317 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5318 * above. We reserve our extra bit here because we generate a ton of
5319 * delayed refs activity by truncating.
5321 * If we cannot make our reservation we'll attempt to steal from the
5322 * global reserve, because we really want to be able to free up space.
5324 ret = btrfs_block_rsv_refill(root, rsv, rsv->size + delayed_refs_extra,
5325 BTRFS_RESERVE_FLUSH_EVICT);
5328 * Try to steal from the global reserve if there is space for
5331 if (btrfs_check_space_for_delayed_refs(fs_info) ||
5332 btrfs_block_rsv_migrate(global_rsv, rsv, rsv->size, 0)) {
5334 "could not allocate space for delete; will truncate on mount");
5335 return ERR_PTR(-ENOSPC);
5337 delayed_refs_extra = 0;
5340 trans = btrfs_join_transaction(root);
5344 if (delayed_refs_extra) {
5345 trans->block_rsv = &fs_info->trans_block_rsv;
5346 trans->bytes_reserved = delayed_refs_extra;
5347 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5348 delayed_refs_extra, 1);
5353 void btrfs_evict_inode(struct inode *inode)
5355 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5356 struct btrfs_trans_handle *trans;
5357 struct btrfs_root *root = BTRFS_I(inode)->root;
5358 struct btrfs_block_rsv *rsv;
5361 trace_btrfs_inode_evict(inode);
5368 evict_inode_truncate_pages(inode);
5370 if (inode->i_nlink &&
5371 ((btrfs_root_refs(&root->root_item) != 0 &&
5372 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5373 btrfs_is_free_space_inode(BTRFS_I(inode))))
5376 if (is_bad_inode(inode))
5379 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5381 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5384 if (inode->i_nlink > 0) {
5385 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5386 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5390 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5394 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5397 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5400 btrfs_i_size_write(BTRFS_I(inode), 0);
5403 trans = evict_refill_and_join(root, rsv);
5407 trans->block_rsv = rsv;
5409 ret = btrfs_truncate_inode_items(trans, root, BTRFS_I(inode),
5411 trans->block_rsv = &fs_info->trans_block_rsv;
5412 btrfs_end_transaction(trans);
5413 btrfs_btree_balance_dirty(fs_info);
5414 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5421 * Errors here aren't a big deal, it just means we leave orphan items in
5422 * the tree. They will be cleaned up on the next mount. If the inode
5423 * number gets reused, cleanup deletes the orphan item without doing
5424 * anything, and unlink reuses the existing orphan item.
5426 * If it turns out that we are dropping too many of these, we might want
5427 * to add a mechanism for retrying these after a commit.
5429 trans = evict_refill_and_join(root, rsv);
5430 if (!IS_ERR(trans)) {
5431 trans->block_rsv = rsv;
5432 btrfs_orphan_del(trans, BTRFS_I(inode));
5433 trans->block_rsv = &fs_info->trans_block_rsv;
5434 btrfs_end_transaction(trans);
5438 btrfs_free_block_rsv(fs_info, rsv);
5441 * If we didn't successfully delete, the orphan item will still be in
5442 * the tree and we'll retry on the next mount. Again, we might also want
5443 * to retry these periodically in the future.
5445 btrfs_remove_delayed_node(BTRFS_I(inode));
5450 * Return the key found in the dir entry in the location pointer, fill @type
5451 * with BTRFS_FT_*, and return 0.
5453 * If no dir entries were found, returns -ENOENT.
5454 * If found a corrupted location in dir entry, returns -EUCLEAN.
5456 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5457 struct btrfs_key *location, u8 *type)
5459 const char *name = dentry->d_name.name;
5460 int namelen = dentry->d_name.len;
5461 struct btrfs_dir_item *di;
5462 struct btrfs_path *path;
5463 struct btrfs_root *root = BTRFS_I(dir)->root;
5466 path = btrfs_alloc_path();
5470 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5472 if (IS_ERR_OR_NULL(di)) {
5473 ret = di ? PTR_ERR(di) : -ENOENT;
5477 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5478 if (location->type != BTRFS_INODE_ITEM_KEY &&
5479 location->type != BTRFS_ROOT_ITEM_KEY) {
5481 btrfs_warn(root->fs_info,
5482 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5483 __func__, name, btrfs_ino(BTRFS_I(dir)),
5484 location->objectid, location->type, location->offset);
5487 *type = btrfs_dir_type(path->nodes[0], di);
5489 btrfs_free_path(path);
5494 * when we hit a tree root in a directory, the btrfs part of the inode
5495 * needs to be changed to reflect the root directory of the tree root. This
5496 * is kind of like crossing a mount point.
5498 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5500 struct dentry *dentry,
5501 struct btrfs_key *location,
5502 struct btrfs_root **sub_root)
5504 struct btrfs_path *path;
5505 struct btrfs_root *new_root;
5506 struct btrfs_root_ref *ref;
5507 struct extent_buffer *leaf;
5508 struct btrfs_key key;
5512 path = btrfs_alloc_path();
5519 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5520 key.type = BTRFS_ROOT_REF_KEY;
5521 key.offset = location->objectid;
5523 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5530 leaf = path->nodes[0];
5531 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5532 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5533 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5536 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5537 (unsigned long)(ref + 1),
5538 dentry->d_name.len);
5542 btrfs_release_path(path);
5544 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5545 if (IS_ERR(new_root)) {
5546 err = PTR_ERR(new_root);
5550 *sub_root = new_root;
5551 location->objectid = btrfs_root_dirid(&new_root->root_item);
5552 location->type = BTRFS_INODE_ITEM_KEY;
5553 location->offset = 0;
5556 btrfs_free_path(path);
5560 static void inode_tree_add(struct inode *inode)
5562 struct btrfs_root *root = BTRFS_I(inode)->root;
5563 struct btrfs_inode *entry;
5565 struct rb_node *parent;
5566 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5567 u64 ino = btrfs_ino(BTRFS_I(inode));
5569 if (inode_unhashed(inode))
5572 spin_lock(&root->inode_lock);
5573 p = &root->inode_tree.rb_node;
5576 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5578 if (ino < btrfs_ino(entry))
5579 p = &parent->rb_left;
5580 else if (ino > btrfs_ino(entry))
5581 p = &parent->rb_right;
5583 WARN_ON(!(entry->vfs_inode.i_state &
5584 (I_WILL_FREE | I_FREEING)));
5585 rb_replace_node(parent, new, &root->inode_tree);
5586 RB_CLEAR_NODE(parent);
5587 spin_unlock(&root->inode_lock);
5591 rb_link_node(new, parent, p);
5592 rb_insert_color(new, &root->inode_tree);
5593 spin_unlock(&root->inode_lock);
5596 static void inode_tree_del(struct btrfs_inode *inode)
5598 struct btrfs_root *root = inode->root;
5601 spin_lock(&root->inode_lock);
5602 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5603 rb_erase(&inode->rb_node, &root->inode_tree);
5604 RB_CLEAR_NODE(&inode->rb_node);
5605 empty = RB_EMPTY_ROOT(&root->inode_tree);
5607 spin_unlock(&root->inode_lock);
5609 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5610 spin_lock(&root->inode_lock);
5611 empty = RB_EMPTY_ROOT(&root->inode_tree);
5612 spin_unlock(&root->inode_lock);
5614 btrfs_add_dead_root(root);
5619 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5621 struct btrfs_iget_args *args = p;
5623 inode->i_ino = args->ino;
5624 BTRFS_I(inode)->location.objectid = args->ino;
5625 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5626 BTRFS_I(inode)->location.offset = 0;
5627 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5628 BUG_ON(args->root && !BTRFS_I(inode)->root);
5632 static int btrfs_find_actor(struct inode *inode, void *opaque)
5634 struct btrfs_iget_args *args = opaque;
5636 return args->ino == BTRFS_I(inode)->location.objectid &&
5637 args->root == BTRFS_I(inode)->root;
5640 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5641 struct btrfs_root *root)
5643 struct inode *inode;
5644 struct btrfs_iget_args args;
5645 unsigned long hashval = btrfs_inode_hash(ino, root);
5650 inode = iget5_locked(s, hashval, btrfs_find_actor,
5651 btrfs_init_locked_inode,
5657 * Get an inode object given its inode number and corresponding root.
5658 * Path can be preallocated to prevent recursing back to iget through
5659 * allocator. NULL is also valid but may require an additional allocation
5662 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5663 struct btrfs_root *root, struct btrfs_path *path)
5665 struct inode *inode;
5667 inode = btrfs_iget_locked(s, ino, root);
5669 return ERR_PTR(-ENOMEM);
5671 if (inode->i_state & I_NEW) {
5674 ret = btrfs_read_locked_inode(inode, path);
5676 inode_tree_add(inode);
5677 unlock_new_inode(inode);
5681 * ret > 0 can come from btrfs_search_slot called by
5682 * btrfs_read_locked_inode, this means the inode item
5687 inode = ERR_PTR(ret);
5694 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5696 return btrfs_iget_path(s, ino, root, NULL);
5699 static struct inode *new_simple_dir(struct super_block *s,
5700 struct btrfs_key *key,
5701 struct btrfs_root *root)
5703 struct inode *inode = new_inode(s);
5706 return ERR_PTR(-ENOMEM);
5708 BTRFS_I(inode)->root = btrfs_grab_root(root);
5709 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5710 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5712 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5714 * We only need lookup, the rest is read-only and there's no inode
5715 * associated with the dentry
5717 inode->i_op = &simple_dir_inode_operations;
5718 inode->i_opflags &= ~IOP_XATTR;
5719 inode->i_fop = &simple_dir_operations;
5720 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5721 inode->i_mtime = current_time(inode);
5722 inode->i_atime = inode->i_mtime;
5723 inode->i_ctime = inode->i_mtime;
5724 BTRFS_I(inode)->i_otime = inode->i_mtime;
5729 static inline u8 btrfs_inode_type(struct inode *inode)
5732 * Compile-time asserts that generic FT_* types still match
5735 BUILD_BUG_ON(BTRFS_FT_UNKNOWN != FT_UNKNOWN);
5736 BUILD_BUG_ON(BTRFS_FT_REG_FILE != FT_REG_FILE);
5737 BUILD_BUG_ON(BTRFS_FT_DIR != FT_DIR);
5738 BUILD_BUG_ON(BTRFS_FT_CHRDEV != FT_CHRDEV);
5739 BUILD_BUG_ON(BTRFS_FT_BLKDEV != FT_BLKDEV);
5740 BUILD_BUG_ON(BTRFS_FT_FIFO != FT_FIFO);
5741 BUILD_BUG_ON(BTRFS_FT_SOCK != FT_SOCK);
5742 BUILD_BUG_ON(BTRFS_FT_SYMLINK != FT_SYMLINK);
5744 return fs_umode_to_ftype(inode->i_mode);
5747 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5749 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5750 struct inode *inode;
5751 struct btrfs_root *root = BTRFS_I(dir)->root;
5752 struct btrfs_root *sub_root = root;
5753 struct btrfs_key location;
5757 if (dentry->d_name.len > BTRFS_NAME_LEN)
5758 return ERR_PTR(-ENAMETOOLONG);
5760 ret = btrfs_inode_by_name(dir, dentry, &location, &di_type);
5762 return ERR_PTR(ret);
5764 if (location.type == BTRFS_INODE_ITEM_KEY) {
5765 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5769 /* Do extra check against inode mode with di_type */
5770 if (btrfs_inode_type(inode) != di_type) {
5772 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5773 inode->i_mode, btrfs_inode_type(inode),
5776 return ERR_PTR(-EUCLEAN);
5781 ret = fixup_tree_root_location(fs_info, dir, dentry,
5782 &location, &sub_root);
5785 inode = ERR_PTR(ret);
5787 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5789 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5791 if (root != sub_root)
5792 btrfs_put_root(sub_root);
5794 if (!IS_ERR(inode) && root != sub_root) {
5795 down_read(&fs_info->cleanup_work_sem);
5796 if (!sb_rdonly(inode->i_sb))
5797 ret = btrfs_orphan_cleanup(sub_root);
5798 up_read(&fs_info->cleanup_work_sem);
5801 inode = ERR_PTR(ret);
5808 static int btrfs_dentry_delete(const struct dentry *dentry)
5810 struct btrfs_root *root;
5811 struct inode *inode = d_inode(dentry);
5813 if (!inode && !IS_ROOT(dentry))
5814 inode = d_inode(dentry->d_parent);
5817 root = BTRFS_I(inode)->root;
5818 if (btrfs_root_refs(&root->root_item) == 0)
5821 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5827 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5830 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5832 if (inode == ERR_PTR(-ENOENT))
5834 return d_splice_alias(inode, dentry);
5838 * All this infrastructure exists because dir_emit can fault, and we are holding
5839 * the tree lock when doing readdir. For now just allocate a buffer and copy
5840 * our information into that, and then dir_emit from the buffer. This is
5841 * similar to what NFS does, only we don't keep the buffer around in pagecache
5842 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5843 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5846 static int btrfs_opendir(struct inode *inode, struct file *file)
5848 struct btrfs_file_private *private;
5850 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5853 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5854 if (!private->filldir_buf) {
5858 file->private_data = private;
5869 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5872 struct dir_entry *entry = addr;
5873 char *name = (char *)(entry + 1);
5875 ctx->pos = get_unaligned(&entry->offset);
5876 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5877 get_unaligned(&entry->ino),
5878 get_unaligned(&entry->type)))
5880 addr += sizeof(struct dir_entry) +
5881 get_unaligned(&entry->name_len);
5887 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5889 struct inode *inode = file_inode(file);
5890 struct btrfs_root *root = BTRFS_I(inode)->root;
5891 struct btrfs_file_private *private = file->private_data;
5892 struct btrfs_dir_item *di;
5893 struct btrfs_key key;
5894 struct btrfs_key found_key;
5895 struct btrfs_path *path;
5897 struct list_head ins_list;
5898 struct list_head del_list;
5900 struct extent_buffer *leaf;
5907 struct btrfs_key location;
5909 if (!dir_emit_dots(file, ctx))
5912 path = btrfs_alloc_path();
5916 addr = private->filldir_buf;
5917 path->reada = READA_FORWARD;
5919 INIT_LIST_HEAD(&ins_list);
5920 INIT_LIST_HEAD(&del_list);
5921 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5924 key.type = BTRFS_DIR_INDEX_KEY;
5925 key.offset = ctx->pos;
5926 key.objectid = btrfs_ino(BTRFS_I(inode));
5928 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5933 struct dir_entry *entry;
5935 leaf = path->nodes[0];
5936 slot = path->slots[0];
5937 if (slot >= btrfs_header_nritems(leaf)) {
5938 ret = btrfs_next_leaf(root, path);
5946 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5948 if (found_key.objectid != key.objectid)
5950 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5952 if (found_key.offset < ctx->pos)
5954 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5956 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5957 name_len = btrfs_dir_name_len(leaf, di);
5958 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5960 btrfs_release_path(path);
5961 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5964 addr = private->filldir_buf;
5971 put_unaligned(name_len, &entry->name_len);
5972 name_ptr = (char *)(entry + 1);
5973 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5975 put_unaligned(fs_ftype_to_dtype(btrfs_dir_type(leaf, di)),
5977 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5978 put_unaligned(location.objectid, &entry->ino);
5979 put_unaligned(found_key.offset, &entry->offset);
5981 addr += sizeof(struct dir_entry) + name_len;
5982 total_len += sizeof(struct dir_entry) + name_len;
5986 btrfs_release_path(path);
5988 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5992 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5997 * Stop new entries from being returned after we return the last
6000 * New directory entries are assigned a strictly increasing
6001 * offset. This means that new entries created during readdir
6002 * are *guaranteed* to be seen in the future by that readdir.
6003 * This has broken buggy programs which operate on names as
6004 * they're returned by readdir. Until we re-use freed offsets
6005 * we have this hack to stop new entries from being returned
6006 * under the assumption that they'll never reach this huge
6009 * This is being careful not to overflow 32bit loff_t unless the
6010 * last entry requires it because doing so has broken 32bit apps
6013 if (ctx->pos >= INT_MAX)
6014 ctx->pos = LLONG_MAX;
6021 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6022 btrfs_free_path(path);
6027 * This is somewhat expensive, updating the tree every time the
6028 * inode changes. But, it is most likely to find the inode in cache.
6029 * FIXME, needs more benchmarking...there are no reasons other than performance
6030 * to keep or drop this code.
6032 static int btrfs_dirty_inode(struct inode *inode)
6034 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6035 struct btrfs_root *root = BTRFS_I(inode)->root;
6036 struct btrfs_trans_handle *trans;
6039 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6042 trans = btrfs_join_transaction(root);
6044 return PTR_ERR(trans);
6046 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
6047 if (ret && ret == -ENOSPC) {
6048 /* whoops, lets try again with the full transaction */
6049 btrfs_end_transaction(trans);
6050 trans = btrfs_start_transaction(root, 1);
6052 return PTR_ERR(trans);
6054 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
6056 btrfs_end_transaction(trans);
6057 if (BTRFS_I(inode)->delayed_node)
6058 btrfs_balance_delayed_items(fs_info);
6064 * This is a copy of file_update_time. We need this so we can return error on
6065 * ENOSPC for updating the inode in the case of file write and mmap writes.
6067 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
6070 struct btrfs_root *root = BTRFS_I(inode)->root;
6071 bool dirty = flags & ~S_VERSION;
6073 if (btrfs_root_readonly(root))
6076 if (flags & S_VERSION)
6077 dirty |= inode_maybe_inc_iversion(inode, dirty);
6078 if (flags & S_CTIME)
6079 inode->i_ctime = *now;
6080 if (flags & S_MTIME)
6081 inode->i_mtime = *now;
6082 if (flags & S_ATIME)
6083 inode->i_atime = *now;
6084 return dirty ? btrfs_dirty_inode(inode) : 0;
6088 * find the highest existing sequence number in a directory
6089 * and then set the in-memory index_cnt variable to reflect
6090 * free sequence numbers
6092 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6094 struct btrfs_root *root = inode->root;
6095 struct btrfs_key key, found_key;
6096 struct btrfs_path *path;
6097 struct extent_buffer *leaf;
6100 key.objectid = btrfs_ino(inode);
6101 key.type = BTRFS_DIR_INDEX_KEY;
6102 key.offset = (u64)-1;
6104 path = btrfs_alloc_path();
6108 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6111 /* FIXME: we should be able to handle this */
6117 * MAGIC NUMBER EXPLANATION:
6118 * since we search a directory based on f_pos we have to start at 2
6119 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6120 * else has to start at 2
6122 if (path->slots[0] == 0) {
6123 inode->index_cnt = 2;
6129 leaf = path->nodes[0];
6130 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6132 if (found_key.objectid != btrfs_ino(inode) ||
6133 found_key.type != BTRFS_DIR_INDEX_KEY) {
6134 inode->index_cnt = 2;
6138 inode->index_cnt = found_key.offset + 1;
6140 btrfs_free_path(path);
6145 * helper to find a free sequence number in a given directory. This current
6146 * code is very simple, later versions will do smarter things in the btree
6148 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6152 if (dir->index_cnt == (u64)-1) {
6153 ret = btrfs_inode_delayed_dir_index_count(dir);
6155 ret = btrfs_set_inode_index_count(dir);
6161 *index = dir->index_cnt;
6167 static int btrfs_insert_inode_locked(struct inode *inode)
6169 struct btrfs_iget_args args;
6171 args.ino = BTRFS_I(inode)->location.objectid;
6172 args.root = BTRFS_I(inode)->root;
6174 return insert_inode_locked4(inode,
6175 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6176 btrfs_find_actor, &args);
6180 * Inherit flags from the parent inode.
6182 * Currently only the compression flags and the cow flags are inherited.
6184 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6191 flags = BTRFS_I(dir)->flags;
6193 if (flags & BTRFS_INODE_NOCOMPRESS) {
6194 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6195 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6196 } else if (flags & BTRFS_INODE_COMPRESS) {
6197 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6198 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6201 if (flags & BTRFS_INODE_NODATACOW) {
6202 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6203 if (S_ISREG(inode->i_mode))
6204 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6207 btrfs_sync_inode_flags_to_i_flags(inode);
6210 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6211 struct btrfs_root *root,
6213 const char *name, int name_len,
6214 u64 ref_objectid, u64 objectid,
6215 umode_t mode, u64 *index)
6217 struct btrfs_fs_info *fs_info = root->fs_info;
6218 struct inode *inode;
6219 struct btrfs_inode_item *inode_item;
6220 struct btrfs_key *location;
6221 struct btrfs_path *path;
6222 struct btrfs_inode_ref *ref;
6223 struct btrfs_key key[2];
6225 int nitems = name ? 2 : 1;
6227 unsigned int nofs_flag;
6230 path = btrfs_alloc_path();
6232 return ERR_PTR(-ENOMEM);
6234 nofs_flag = memalloc_nofs_save();
6235 inode = new_inode(fs_info->sb);
6236 memalloc_nofs_restore(nofs_flag);
6238 btrfs_free_path(path);
6239 return ERR_PTR(-ENOMEM);
6243 * O_TMPFILE, set link count to 0, so that after this point,
6244 * we fill in an inode item with the correct link count.
6247 set_nlink(inode, 0);
6250 * we have to initialize this early, so we can reclaim the inode
6251 * number if we fail afterwards in this function.
6253 inode->i_ino = objectid;
6256 trace_btrfs_inode_request(dir);
6258 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6260 btrfs_free_path(path);
6262 return ERR_PTR(ret);
6268 * index_cnt is ignored for everything but a dir,
6269 * btrfs_set_inode_index_count has an explanation for the magic
6272 BTRFS_I(inode)->index_cnt = 2;
6273 BTRFS_I(inode)->dir_index = *index;
6274 BTRFS_I(inode)->root = btrfs_grab_root(root);
6275 BTRFS_I(inode)->generation = trans->transid;
6276 inode->i_generation = BTRFS_I(inode)->generation;
6279 * We could have gotten an inode number from somebody who was fsynced
6280 * and then removed in this same transaction, so let's just set full
6281 * sync since it will be a full sync anyway and this will blow away the
6282 * old info in the log.
6284 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6286 key[0].objectid = objectid;
6287 key[0].type = BTRFS_INODE_ITEM_KEY;
6290 sizes[0] = sizeof(struct btrfs_inode_item);
6294 * Start new inodes with an inode_ref. This is slightly more
6295 * efficient for small numbers of hard links since they will
6296 * be packed into one item. Extended refs will kick in if we
6297 * add more hard links than can fit in the ref item.
6299 key[1].objectid = objectid;
6300 key[1].type = BTRFS_INODE_REF_KEY;
6301 key[1].offset = ref_objectid;
6303 sizes[1] = name_len + sizeof(*ref);
6306 location = &BTRFS_I(inode)->location;
6307 location->objectid = objectid;
6308 location->offset = 0;
6309 location->type = BTRFS_INODE_ITEM_KEY;
6311 ret = btrfs_insert_inode_locked(inode);
6317 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6321 inode_init_owner(inode, dir, mode);
6322 inode_set_bytes(inode, 0);
6324 inode->i_mtime = current_time(inode);
6325 inode->i_atime = inode->i_mtime;
6326 inode->i_ctime = inode->i_mtime;
6327 BTRFS_I(inode)->i_otime = inode->i_mtime;
6329 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6330 struct btrfs_inode_item);
6331 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6332 sizeof(*inode_item));
6333 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6336 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6337 struct btrfs_inode_ref);
6338 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6339 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6340 ptr = (unsigned long)(ref + 1);
6341 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6344 btrfs_mark_buffer_dirty(path->nodes[0]);
6345 btrfs_free_path(path);
6347 btrfs_inherit_iflags(inode, dir);
6349 if (S_ISREG(mode)) {
6350 if (btrfs_test_opt(fs_info, NODATASUM))
6351 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6352 if (btrfs_test_opt(fs_info, NODATACOW))
6353 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6354 BTRFS_INODE_NODATASUM;
6357 inode_tree_add(inode);
6359 trace_btrfs_inode_new(inode);
6360 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6362 btrfs_update_root_times(trans, root);
6364 ret = btrfs_inode_inherit_props(trans, inode, dir);
6367 "error inheriting props for ino %llu (root %llu): %d",
6368 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6373 discard_new_inode(inode);
6376 BTRFS_I(dir)->index_cnt--;
6377 btrfs_free_path(path);
6378 return ERR_PTR(ret);
6382 * utility function to add 'inode' into 'parent_inode' with
6383 * a give name and a given sequence number.
6384 * if 'add_backref' is true, also insert a backref from the
6385 * inode to the parent directory.
6387 int btrfs_add_link(struct btrfs_trans_handle *trans,
6388 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6389 const char *name, int name_len, int add_backref, u64 index)
6392 struct btrfs_key key;
6393 struct btrfs_root *root = parent_inode->root;
6394 u64 ino = btrfs_ino(inode);
6395 u64 parent_ino = btrfs_ino(parent_inode);
6397 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6398 memcpy(&key, &inode->root->root_key, sizeof(key));
6401 key.type = BTRFS_INODE_ITEM_KEY;
6405 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6406 ret = btrfs_add_root_ref(trans, key.objectid,
6407 root->root_key.objectid, parent_ino,
6408 index, name, name_len);
6409 } else if (add_backref) {
6410 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6414 /* Nothing to clean up yet */
6418 ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key,
6419 btrfs_inode_type(&inode->vfs_inode), index);
6420 if (ret == -EEXIST || ret == -EOVERFLOW)
6423 btrfs_abort_transaction(trans, ret);
6427 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6429 inode_inc_iversion(&parent_inode->vfs_inode);
6431 * If we are replaying a log tree, we do not want to update the mtime
6432 * and ctime of the parent directory with the current time, since the
6433 * log replay procedure is responsible for setting them to their correct
6434 * values (the ones it had when the fsync was done).
6436 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
6437 struct timespec64 now = current_time(&parent_inode->vfs_inode);
6439 parent_inode->vfs_inode.i_mtime = now;
6440 parent_inode->vfs_inode.i_ctime = now;
6442 ret = btrfs_update_inode(trans, root, parent_inode);
6444 btrfs_abort_transaction(trans, ret);
6448 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6451 err = btrfs_del_root_ref(trans, key.objectid,
6452 root->root_key.objectid, parent_ino,
6453 &local_index, name, name_len);
6455 btrfs_abort_transaction(trans, err);
6456 } else if (add_backref) {
6460 err = btrfs_del_inode_ref(trans, root, name, name_len,
6461 ino, parent_ino, &local_index);
6463 btrfs_abort_transaction(trans, err);
6466 /* Return the original error code */
6470 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6471 struct btrfs_inode *dir, struct dentry *dentry,
6472 struct btrfs_inode *inode, int backref, u64 index)
6474 int err = btrfs_add_link(trans, dir, inode,
6475 dentry->d_name.name, dentry->d_name.len,
6482 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6483 umode_t mode, dev_t rdev)
6485 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6486 struct btrfs_trans_handle *trans;
6487 struct btrfs_root *root = BTRFS_I(dir)->root;
6488 struct inode *inode = NULL;
6494 * 2 for inode item and ref
6496 * 1 for xattr if selinux is on
6498 trans = btrfs_start_transaction(root, 5);
6500 return PTR_ERR(trans);
6502 err = btrfs_get_free_objectid(root, &objectid);
6506 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6507 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6509 if (IS_ERR(inode)) {
6510 err = PTR_ERR(inode);
6516 * If the active LSM wants to access the inode during
6517 * d_instantiate it needs these. Smack checks to see
6518 * if the filesystem supports xattrs by looking at the
6521 inode->i_op = &btrfs_special_inode_operations;
6522 init_special_inode(inode, inode->i_mode, rdev);
6524 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6528 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6533 btrfs_update_inode(trans, root, BTRFS_I(inode));
6534 d_instantiate_new(dentry, inode);
6537 btrfs_end_transaction(trans);
6538 btrfs_btree_balance_dirty(fs_info);
6540 inode_dec_link_count(inode);
6541 discard_new_inode(inode);
6546 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6547 umode_t mode, bool excl)
6549 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6550 struct btrfs_trans_handle *trans;
6551 struct btrfs_root *root = BTRFS_I(dir)->root;
6552 struct inode *inode = NULL;
6558 * 2 for inode item and ref
6560 * 1 for xattr if selinux is on
6562 trans = btrfs_start_transaction(root, 5);
6564 return PTR_ERR(trans);
6566 err = btrfs_get_free_objectid(root, &objectid);
6570 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6571 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6573 if (IS_ERR(inode)) {
6574 err = PTR_ERR(inode);
6579 * If the active LSM wants to access the inode during
6580 * d_instantiate it needs these. Smack checks to see
6581 * if the filesystem supports xattrs by looking at the
6584 inode->i_fop = &btrfs_file_operations;
6585 inode->i_op = &btrfs_file_inode_operations;
6586 inode->i_mapping->a_ops = &btrfs_aops;
6588 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6592 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6596 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6601 d_instantiate_new(dentry, inode);
6604 btrfs_end_transaction(trans);
6606 inode_dec_link_count(inode);
6607 discard_new_inode(inode);
6609 btrfs_btree_balance_dirty(fs_info);
6613 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6614 struct dentry *dentry)
6616 struct btrfs_trans_handle *trans = NULL;
6617 struct btrfs_root *root = BTRFS_I(dir)->root;
6618 struct inode *inode = d_inode(old_dentry);
6619 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6624 /* do not allow sys_link's with other subvols of the same device */
6625 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6628 if (inode->i_nlink >= BTRFS_LINK_MAX)
6631 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6636 * 2 items for inode and inode ref
6637 * 2 items for dir items
6638 * 1 item for parent inode
6639 * 1 item for orphan item deletion if O_TMPFILE
6641 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6642 if (IS_ERR(trans)) {
6643 err = PTR_ERR(trans);
6648 /* There are several dir indexes for this inode, clear the cache. */
6649 BTRFS_I(inode)->dir_index = 0ULL;
6651 inode_inc_iversion(inode);
6652 inode->i_ctime = current_time(inode);
6654 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6656 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6662 struct dentry *parent = dentry->d_parent;
6664 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6667 if (inode->i_nlink == 1) {
6669 * If new hard link count is 1, it's a file created
6670 * with open(2) O_TMPFILE flag.
6672 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6676 d_instantiate(dentry, inode);
6677 btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent);
6682 btrfs_end_transaction(trans);
6684 inode_dec_link_count(inode);
6687 btrfs_btree_balance_dirty(fs_info);
6691 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6693 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6694 struct inode *inode = NULL;
6695 struct btrfs_trans_handle *trans;
6696 struct btrfs_root *root = BTRFS_I(dir)->root;
6702 * 2 items for inode and ref
6703 * 2 items for dir items
6704 * 1 for xattr if selinux is on
6706 trans = btrfs_start_transaction(root, 5);
6708 return PTR_ERR(trans);
6710 err = btrfs_get_free_objectid(root, &objectid);
6714 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6715 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6716 S_IFDIR | mode, &index);
6717 if (IS_ERR(inode)) {
6718 err = PTR_ERR(inode);
6723 /* these must be set before we unlock the inode */
6724 inode->i_op = &btrfs_dir_inode_operations;
6725 inode->i_fop = &btrfs_dir_file_operations;
6727 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6731 btrfs_i_size_write(BTRFS_I(inode), 0);
6732 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6736 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6737 dentry->d_name.name,
6738 dentry->d_name.len, 0, index);
6742 d_instantiate_new(dentry, inode);
6745 btrfs_end_transaction(trans);
6747 inode_dec_link_count(inode);
6748 discard_new_inode(inode);
6750 btrfs_btree_balance_dirty(fs_info);
6754 static noinline int uncompress_inline(struct btrfs_path *path,
6756 size_t pg_offset, u64 extent_offset,
6757 struct btrfs_file_extent_item *item)
6760 struct extent_buffer *leaf = path->nodes[0];
6763 unsigned long inline_size;
6767 WARN_ON(pg_offset != 0);
6768 compress_type = btrfs_file_extent_compression(leaf, item);
6769 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6770 inline_size = btrfs_file_extent_inline_item_len(leaf,
6771 btrfs_item_nr(path->slots[0]));
6772 tmp = kmalloc(inline_size, GFP_NOFS);
6775 ptr = btrfs_file_extent_inline_start(item);
6777 read_extent_buffer(leaf, tmp, ptr, inline_size);
6779 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6780 ret = btrfs_decompress(compress_type, tmp, page,
6781 extent_offset, inline_size, max_size);
6784 * decompression code contains a memset to fill in any space between the end
6785 * of the uncompressed data and the end of max_size in case the decompressed
6786 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6787 * the end of an inline extent and the beginning of the next block, so we
6788 * cover that region here.
6791 if (max_size + pg_offset < PAGE_SIZE) {
6792 char *map = kmap(page);
6793 memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6801 * btrfs_get_extent - Lookup the first extent overlapping a range in a file.
6802 * @inode: file to search in
6803 * @page: page to read extent data into if the extent is inline
6804 * @pg_offset: offset into @page to copy to
6805 * @start: file offset
6806 * @len: length of range starting at @start
6808 * This returns the first &struct extent_map which overlaps with the given
6809 * range, reading it from the B-tree and caching it if necessary. Note that
6810 * there may be more extents which overlap the given range after the returned
6813 * If @page is not NULL and the extent is inline, this also reads the extent
6814 * data directly into the page and marks the extent up to date in the io_tree.
6816 * Return: ERR_PTR on error, non-NULL extent_map on success.
6818 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6819 struct page *page, size_t pg_offset,
6822 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6824 u64 extent_start = 0;
6826 u64 objectid = btrfs_ino(inode);
6827 int extent_type = -1;
6828 struct btrfs_path *path = NULL;
6829 struct btrfs_root *root = inode->root;
6830 struct btrfs_file_extent_item *item;
6831 struct extent_buffer *leaf;
6832 struct btrfs_key found_key;
6833 struct extent_map *em = NULL;
6834 struct extent_map_tree *em_tree = &inode->extent_tree;
6835 struct extent_io_tree *io_tree = &inode->io_tree;
6837 read_lock(&em_tree->lock);
6838 em = lookup_extent_mapping(em_tree, start, len);
6839 read_unlock(&em_tree->lock);
6842 if (em->start > start || em->start + em->len <= start)
6843 free_extent_map(em);
6844 else if (em->block_start == EXTENT_MAP_INLINE && page)
6845 free_extent_map(em);
6849 em = alloc_extent_map();
6854 em->start = EXTENT_MAP_HOLE;
6855 em->orig_start = EXTENT_MAP_HOLE;
6857 em->block_len = (u64)-1;
6859 path = btrfs_alloc_path();
6865 /* Chances are we'll be called again, so go ahead and do readahead */
6866 path->reada = READA_FORWARD;
6869 * The same explanation in load_free_space_cache applies here as well,
6870 * we only read when we're loading the free space cache, and at that
6871 * point the commit_root has everything we need.
6873 if (btrfs_is_free_space_inode(inode)) {
6874 path->search_commit_root = 1;
6875 path->skip_locking = 1;
6878 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6881 } else if (ret > 0) {
6882 if (path->slots[0] == 0)
6888 leaf = path->nodes[0];
6889 item = btrfs_item_ptr(leaf, path->slots[0],
6890 struct btrfs_file_extent_item);
6891 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6892 if (found_key.objectid != objectid ||
6893 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6895 * If we backup past the first extent we want to move forward
6896 * and see if there is an extent in front of us, otherwise we'll
6897 * say there is a hole for our whole search range which can
6904 extent_type = btrfs_file_extent_type(leaf, item);
6905 extent_start = found_key.offset;
6906 extent_end = btrfs_file_extent_end(path);
6907 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6908 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6909 /* Only regular file could have regular/prealloc extent */
6910 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6913 "regular/prealloc extent found for non-regular inode %llu",
6917 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6919 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6920 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6925 if (start >= extent_end) {
6927 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6928 ret = btrfs_next_leaf(root, path);
6934 leaf = path->nodes[0];
6936 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6937 if (found_key.objectid != objectid ||
6938 found_key.type != BTRFS_EXTENT_DATA_KEY)
6940 if (start + len <= found_key.offset)
6942 if (start > found_key.offset)
6945 /* New extent overlaps with existing one */
6947 em->orig_start = start;
6948 em->len = found_key.offset - start;
6949 em->block_start = EXTENT_MAP_HOLE;
6953 btrfs_extent_item_to_extent_map(inode, path, item, !page, em);
6955 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6956 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6958 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6962 size_t extent_offset;
6968 size = btrfs_file_extent_ram_bytes(leaf, item);
6969 extent_offset = page_offset(page) + pg_offset - extent_start;
6970 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6971 size - extent_offset);
6972 em->start = extent_start + extent_offset;
6973 em->len = ALIGN(copy_size, fs_info->sectorsize);
6974 em->orig_block_len = em->len;
6975 em->orig_start = em->start;
6976 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6978 if (!PageUptodate(page)) {
6979 if (btrfs_file_extent_compression(leaf, item) !=
6980 BTRFS_COMPRESS_NONE) {
6981 ret = uncompress_inline(path, page, pg_offset,
6982 extent_offset, item);
6987 read_extent_buffer(leaf, map + pg_offset, ptr,
6989 if (pg_offset + copy_size < PAGE_SIZE) {
6990 memset(map + pg_offset + copy_size, 0,
6991 PAGE_SIZE - pg_offset -
6996 flush_dcache_page(page);
6998 set_extent_uptodate(io_tree, em->start,
6999 extent_map_end(em) - 1, NULL, GFP_NOFS);
7004 em->orig_start = start;
7006 em->block_start = EXTENT_MAP_HOLE;
7009 btrfs_release_path(path);
7010 if (em->start > start || extent_map_end(em) <= start) {
7012 "bad extent! em: [%llu %llu] passed [%llu %llu]",
7013 em->start, em->len, start, len);
7018 write_lock(&em_tree->lock);
7019 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
7020 write_unlock(&em_tree->lock);
7022 btrfs_free_path(path);
7024 trace_btrfs_get_extent(root, inode, em);
7027 free_extent_map(em);
7028 return ERR_PTR(ret);
7033 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
7036 struct extent_map *em;
7037 struct extent_map *hole_em = NULL;
7038 u64 delalloc_start = start;
7044 em = btrfs_get_extent(inode, NULL, 0, start, len);
7048 * If our em maps to:
7050 * - a pre-alloc extent,
7051 * there might actually be delalloc bytes behind it.
7053 if (em->block_start != EXTENT_MAP_HOLE &&
7054 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7059 /* check to see if we've wrapped (len == -1 or similar) */
7068 /* ok, we didn't find anything, lets look for delalloc */
7069 delalloc_len = count_range_bits(&inode->io_tree, &delalloc_start,
7070 end, len, EXTENT_DELALLOC, 1);
7071 delalloc_end = delalloc_start + delalloc_len;
7072 if (delalloc_end < delalloc_start)
7073 delalloc_end = (u64)-1;
7076 * We didn't find anything useful, return the original results from
7079 if (delalloc_start > end || delalloc_end <= start) {
7086 * Adjust the delalloc_start to make sure it doesn't go backwards from
7087 * the start they passed in
7089 delalloc_start = max(start, delalloc_start);
7090 delalloc_len = delalloc_end - delalloc_start;
7092 if (delalloc_len > 0) {
7095 const u64 hole_end = extent_map_end(hole_em);
7097 em = alloc_extent_map();
7105 * When btrfs_get_extent can't find anything it returns one
7108 * Make sure what it found really fits our range, and adjust to
7109 * make sure it is based on the start from the caller
7111 if (hole_end <= start || hole_em->start > end) {
7112 free_extent_map(hole_em);
7115 hole_start = max(hole_em->start, start);
7116 hole_len = hole_end - hole_start;
7119 if (hole_em && delalloc_start > hole_start) {
7121 * Our hole starts before our delalloc, so we have to
7122 * return just the parts of the hole that go until the
7125 em->len = min(hole_len, delalloc_start - hole_start);
7126 em->start = hole_start;
7127 em->orig_start = hole_start;
7129 * Don't adjust block start at all, it is fixed at
7132 em->block_start = hole_em->block_start;
7133 em->block_len = hole_len;
7134 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7135 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7138 * Hole is out of passed range or it starts after
7141 em->start = delalloc_start;
7142 em->len = delalloc_len;
7143 em->orig_start = delalloc_start;
7144 em->block_start = EXTENT_MAP_DELALLOC;
7145 em->block_len = delalloc_len;
7152 free_extent_map(hole_em);
7154 free_extent_map(em);
7155 return ERR_PTR(err);
7160 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
7163 const u64 orig_start,
7164 const u64 block_start,
7165 const u64 block_len,
7166 const u64 orig_block_len,
7167 const u64 ram_bytes,
7170 struct extent_map *em = NULL;
7173 if (type != BTRFS_ORDERED_NOCOW) {
7174 em = create_io_em(inode, start, len, orig_start, block_start,
7175 block_len, orig_block_len, ram_bytes,
7176 BTRFS_COMPRESS_NONE, /* compress_type */
7181 ret = btrfs_add_ordered_extent_dio(inode, start, block_start, len,
7185 free_extent_map(em);
7186 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
7195 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
7198 struct btrfs_root *root = inode->root;
7199 struct btrfs_fs_info *fs_info = root->fs_info;
7200 struct extent_map *em;
7201 struct btrfs_key ins;
7205 alloc_hint = get_extent_allocation_hint(inode, start, len);
7206 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7207 0, alloc_hint, &ins, 1, 1);
7209 return ERR_PTR(ret);
7211 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7212 ins.objectid, ins.offset, ins.offset,
7213 ins.offset, BTRFS_ORDERED_REGULAR);
7214 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7216 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7223 * Check if we can do nocow write into the range [@offset, @offset + @len)
7225 * @offset: File offset
7226 * @len: The length to write, will be updated to the nocow writeable
7228 * @orig_start: (optional) Return the original file offset of the file extent
7229 * @orig_len: (optional) Return the original on-disk length of the file extent
7230 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7231 * @strict: if true, omit optimizations that might force us into unnecessary
7232 * cow. e.g., don't trust generation number.
7235 * >0 and update @len if we can do nocow write
7236 * 0 if we can't do nocow write
7237 * <0 if error happened
7239 * NOTE: This only checks the file extents, caller is responsible to wait for
7240 * any ordered extents.
7242 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7243 u64 *orig_start, u64 *orig_block_len,
7244 u64 *ram_bytes, bool strict)
7246 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7247 struct btrfs_path *path;
7249 struct extent_buffer *leaf;
7250 struct btrfs_root *root = BTRFS_I(inode)->root;
7251 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7252 struct btrfs_file_extent_item *fi;
7253 struct btrfs_key key;
7260 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7262 path = btrfs_alloc_path();
7266 ret = btrfs_lookup_file_extent(NULL, root, path,
7267 btrfs_ino(BTRFS_I(inode)), offset, 0);
7271 slot = path->slots[0];
7274 /* can't find the item, must cow */
7281 leaf = path->nodes[0];
7282 btrfs_item_key_to_cpu(leaf, &key, slot);
7283 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7284 key.type != BTRFS_EXTENT_DATA_KEY) {
7285 /* not our file or wrong item type, must cow */
7289 if (key.offset > offset) {
7290 /* Wrong offset, must cow */
7294 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7295 found_type = btrfs_file_extent_type(leaf, fi);
7296 if (found_type != BTRFS_FILE_EXTENT_REG &&
7297 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7298 /* not a regular extent, must cow */
7302 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7305 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7306 if (extent_end <= offset)
7309 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7310 if (disk_bytenr == 0)
7313 if (btrfs_file_extent_compression(leaf, fi) ||
7314 btrfs_file_extent_encryption(leaf, fi) ||
7315 btrfs_file_extent_other_encoding(leaf, fi))
7319 * Do the same check as in btrfs_cross_ref_exist but without the
7320 * unnecessary search.
7323 (btrfs_file_extent_generation(leaf, fi) <=
7324 btrfs_root_last_snapshot(&root->root_item)))
7327 backref_offset = btrfs_file_extent_offset(leaf, fi);
7330 *orig_start = key.offset - backref_offset;
7331 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7332 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7335 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7338 num_bytes = min(offset + *len, extent_end) - offset;
7339 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7342 range_end = round_up(offset + num_bytes,
7343 root->fs_info->sectorsize) - 1;
7344 ret = test_range_bit(io_tree, offset, range_end,
7345 EXTENT_DELALLOC, 0, NULL);
7352 btrfs_release_path(path);
7355 * look for other files referencing this extent, if we
7356 * find any we must cow
7359 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7360 key.offset - backref_offset, disk_bytenr,
7368 * adjust disk_bytenr and num_bytes to cover just the bytes
7369 * in this extent we are about to write. If there
7370 * are any csums in that range we have to cow in order
7371 * to keep the csums correct
7373 disk_bytenr += backref_offset;
7374 disk_bytenr += offset - key.offset;
7375 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7378 * all of the above have passed, it is safe to overwrite this extent
7384 btrfs_free_path(path);
7388 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7389 struct extent_state **cached_state, bool writing)
7391 struct btrfs_ordered_extent *ordered;
7395 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7398 * We're concerned with the entire range that we're going to be
7399 * doing DIO to, so we need to make sure there's no ordered
7400 * extents in this range.
7402 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7403 lockend - lockstart + 1);
7406 * We need to make sure there are no buffered pages in this
7407 * range either, we could have raced between the invalidate in
7408 * generic_file_direct_write and locking the extent. The
7409 * invalidate needs to happen so that reads after a write do not
7413 (!writing || !filemap_range_has_page(inode->i_mapping,
7414 lockstart, lockend)))
7417 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7422 * If we are doing a DIO read and the ordered extent we
7423 * found is for a buffered write, we can not wait for it
7424 * to complete and retry, because if we do so we can
7425 * deadlock with concurrent buffered writes on page
7426 * locks. This happens only if our DIO read covers more
7427 * than one extent map, if at this point has already
7428 * created an ordered extent for a previous extent map
7429 * and locked its range in the inode's io tree, and a
7430 * concurrent write against that previous extent map's
7431 * range and this range started (we unlock the ranges
7432 * in the io tree only when the bios complete and
7433 * buffered writes always lock pages before attempting
7434 * to lock range in the io tree).
7437 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7438 btrfs_start_ordered_extent(ordered, 1);
7441 btrfs_put_ordered_extent(ordered);
7444 * We could trigger writeback for this range (and wait
7445 * for it to complete) and then invalidate the pages for
7446 * this range (through invalidate_inode_pages2_range()),
7447 * but that can lead us to a deadlock with a concurrent
7448 * call to readahead (a buffered read or a defrag call
7449 * triggered a readahead) on a page lock due to an
7450 * ordered dio extent we created before but did not have
7451 * yet a corresponding bio submitted (whence it can not
7452 * complete), which makes readahead wait for that
7453 * ordered extent to complete while holding a lock on
7468 /* The callers of this must take lock_extent() */
7469 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7470 u64 len, u64 orig_start, u64 block_start,
7471 u64 block_len, u64 orig_block_len,
7472 u64 ram_bytes, int compress_type,
7475 struct extent_map_tree *em_tree;
7476 struct extent_map *em;
7479 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7480 type == BTRFS_ORDERED_COMPRESSED ||
7481 type == BTRFS_ORDERED_NOCOW ||
7482 type == BTRFS_ORDERED_REGULAR);
7484 em_tree = &inode->extent_tree;
7485 em = alloc_extent_map();
7487 return ERR_PTR(-ENOMEM);
7490 em->orig_start = orig_start;
7492 em->block_len = block_len;
7493 em->block_start = block_start;
7494 em->orig_block_len = orig_block_len;
7495 em->ram_bytes = ram_bytes;
7496 em->generation = -1;
7497 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7498 if (type == BTRFS_ORDERED_PREALLOC) {
7499 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7500 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7501 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7502 em->compress_type = compress_type;
7506 btrfs_drop_extent_cache(inode, em->start,
7507 em->start + em->len - 1, 0);
7508 write_lock(&em_tree->lock);
7509 ret = add_extent_mapping(em_tree, em, 1);
7510 write_unlock(&em_tree->lock);
7512 * The caller has taken lock_extent(), who could race with us
7515 } while (ret == -EEXIST);
7518 free_extent_map(em);
7519 return ERR_PTR(ret);
7522 /* em got 2 refs now, callers needs to do free_extent_map once. */
7527 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7528 struct inode *inode,
7529 struct btrfs_dio_data *dio_data,
7532 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7533 struct extent_map *em = *map;
7537 * We don't allocate a new extent in the following cases
7539 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7541 * 2) The extent is marked as PREALLOC. We're good to go here and can
7542 * just use the extent.
7545 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7546 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7547 em->block_start != EXTENT_MAP_HOLE)) {
7549 u64 block_start, orig_start, orig_block_len, ram_bytes;
7551 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7552 type = BTRFS_ORDERED_PREALLOC;
7554 type = BTRFS_ORDERED_NOCOW;
7555 len = min(len, em->len - (start - em->start));
7556 block_start = em->block_start + (start - em->start);
7558 if (can_nocow_extent(inode, start, &len, &orig_start,
7559 &orig_block_len, &ram_bytes, false) == 1 &&
7560 btrfs_inc_nocow_writers(fs_info, block_start)) {
7561 struct extent_map *em2;
7563 em2 = btrfs_create_dio_extent(BTRFS_I(inode), start, len,
7564 orig_start, block_start,
7565 len, orig_block_len,
7567 btrfs_dec_nocow_writers(fs_info, block_start);
7568 if (type == BTRFS_ORDERED_PREALLOC) {
7569 free_extent_map(em);
7573 if (em2 && IS_ERR(em2)) {
7578 * For inode marked NODATACOW or extent marked PREALLOC,
7579 * use the existing or preallocated extent, so does not
7580 * need to adjust btrfs_space_info's bytes_may_use.
7582 btrfs_free_reserved_data_space_noquota(fs_info, len);
7587 /* this will cow the extent */
7588 free_extent_map(em);
7589 *map = em = btrfs_new_extent_direct(BTRFS_I(inode), start, len);
7595 len = min(len, em->len - (start - em->start));
7599 * Need to update the i_size under the extent lock so buffered
7600 * readers will get the updated i_size when we unlock.
7602 if (start + len > i_size_read(inode))
7603 i_size_write(inode, start + len);
7605 dio_data->reserve -= len;
7610 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7611 loff_t length, unsigned int flags, struct iomap *iomap,
7612 struct iomap *srcmap)
7614 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7615 struct extent_map *em;
7616 struct extent_state *cached_state = NULL;
7617 struct btrfs_dio_data *dio_data = NULL;
7618 u64 lockstart, lockend;
7619 const bool write = !!(flags & IOMAP_WRITE);
7622 bool unlock_extents = false;
7625 len = min_t(u64, len, fs_info->sectorsize);
7628 lockend = start + len - 1;
7631 * The generic stuff only does filemap_write_and_wait_range, which
7632 * isn't enough if we've written compressed pages to this area, so we
7633 * need to flush the dirty pages again to make absolutely sure that any
7634 * outstanding dirty pages are on disk.
7636 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7637 &BTRFS_I(inode)->runtime_flags)) {
7638 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7639 start + length - 1);
7644 dio_data = kzalloc(sizeof(*dio_data), GFP_NOFS);
7648 dio_data->length = length;
7650 dio_data->reserve = round_up(length, fs_info->sectorsize);
7651 ret = btrfs_delalloc_reserve_space(BTRFS_I(inode),
7652 &dio_data->data_reserved,
7653 start, dio_data->reserve);
7655 extent_changeset_free(dio_data->data_reserved);
7660 iomap->private = dio_data;
7664 * If this errors out it's because we couldn't invalidate pagecache for
7665 * this range and we need to fallback to buffered.
7667 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, write)) {
7672 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7679 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7680 * io. INLINE is special, and we could probably kludge it in here, but
7681 * it's still buffered so for safety lets just fall back to the generic
7684 * For COMPRESSED we _have_ to read the entire extent in so we can
7685 * decompress it, so there will be buffering required no matter what we
7686 * do, so go ahead and fallback to buffered.
7688 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7689 * to buffered IO. Don't blame me, this is the price we pay for using
7692 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7693 em->block_start == EXTENT_MAP_INLINE) {
7694 free_extent_map(em);
7699 len = min(len, em->len - (start - em->start));
7701 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7705 unlock_extents = true;
7706 /* Recalc len in case the new em is smaller than requested */
7707 len = min(len, em->len - (start - em->start));
7710 * We need to unlock only the end area that we aren't using.
7711 * The rest is going to be unlocked by the endio routine.
7713 lockstart = start + len;
7714 if (lockstart < lockend)
7715 unlock_extents = true;
7719 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
7720 lockstart, lockend, &cached_state);
7722 free_extent_state(cached_state);
7725 * Translate extent map information to iomap.
7726 * We trim the extents (and move the addr) even though iomap code does
7727 * that, since we have locked only the parts we are performing I/O in.
7729 if ((em->block_start == EXTENT_MAP_HOLE) ||
7730 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7731 iomap->addr = IOMAP_NULL_ADDR;
7732 iomap->type = IOMAP_HOLE;
7734 iomap->addr = em->block_start + (start - em->start);
7735 iomap->type = IOMAP_MAPPED;
7737 iomap->offset = start;
7738 iomap->bdev = fs_info->fs_devices->latest_bdev;
7739 iomap->length = len;
7741 free_extent_map(em);
7746 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7750 btrfs_delalloc_release_space(BTRFS_I(inode),
7751 dio_data->data_reserved, start,
7752 dio_data->reserve, true);
7753 btrfs_delalloc_release_extents(BTRFS_I(inode), dio_data->reserve);
7754 extent_changeset_free(dio_data->data_reserved);
7760 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7761 ssize_t written, unsigned int flags, struct iomap *iomap)
7764 struct btrfs_dio_data *dio_data = iomap->private;
7765 size_t submitted = dio_data->submitted;
7766 const bool write = !!(flags & IOMAP_WRITE);
7768 if (!write && (iomap->type == IOMAP_HOLE)) {
7769 /* If reading from a hole, unlock and return */
7770 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1);
7774 if (submitted < length) {
7776 length -= submitted;
7778 __endio_write_update_ordered(BTRFS_I(inode), pos,
7781 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7787 if (dio_data->reserve)
7788 btrfs_delalloc_release_space(BTRFS_I(inode),
7789 dio_data->data_reserved, pos,
7790 dio_data->reserve, true);
7791 btrfs_delalloc_release_extents(BTRFS_I(inode), dio_data->length);
7792 extent_changeset_free(dio_data->data_reserved);
7796 iomap->private = NULL;
7801 static void btrfs_dio_private_put(struct btrfs_dio_private *dip)
7804 * This implies a barrier so that stores to dio_bio->bi_status before
7805 * this and loads of dio_bio->bi_status after this are fully ordered.
7807 if (!refcount_dec_and_test(&dip->refs))
7810 if (btrfs_op(dip->dio_bio) == BTRFS_MAP_WRITE) {
7811 __endio_write_update_ordered(BTRFS_I(dip->inode),
7812 dip->logical_offset,
7814 !dip->dio_bio->bi_status);
7816 unlock_extent(&BTRFS_I(dip->inode)->io_tree,
7817 dip->logical_offset,
7818 dip->logical_offset + dip->bytes - 1);
7821 bio_endio(dip->dio_bio);
7825 static blk_status_t submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7827 unsigned long bio_flags)
7829 struct btrfs_dio_private *dip = bio->bi_private;
7830 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7833 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7835 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
7839 refcount_inc(&dip->refs);
7840 ret = btrfs_map_bio(fs_info, bio, mirror_num);
7842 refcount_dec(&dip->refs);
7846 static blk_status_t btrfs_check_read_dio_bio(struct inode *inode,
7847 struct btrfs_io_bio *io_bio,
7848 const bool uptodate)
7850 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
7851 const u32 sectorsize = fs_info->sectorsize;
7852 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7853 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7854 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7855 struct bio_vec bvec;
7856 struct bvec_iter iter;
7857 u64 start = io_bio->logical;
7859 blk_status_t err = BLK_STS_OK;
7861 __bio_for_each_segment(bvec, &io_bio->bio, iter, io_bio->iter) {
7862 unsigned int i, nr_sectors, pgoff;
7864 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7865 pgoff = bvec.bv_offset;
7866 for (i = 0; i < nr_sectors; i++) {
7867 ASSERT(pgoff < PAGE_SIZE);
7869 (!csum || !check_data_csum(inode, io_bio,
7870 bio_offset, bvec.bv_page, pgoff))) {
7871 clean_io_failure(fs_info, failure_tree, io_tree,
7872 start, bvec.bv_page,
7873 btrfs_ino(BTRFS_I(inode)),
7876 blk_status_t status;
7878 ASSERT((start - io_bio->logical) < UINT_MAX);
7879 status = btrfs_submit_read_repair(inode,
7881 start - io_bio->logical,
7882 bvec.bv_page, pgoff,
7884 start + sectorsize - 1,
7886 submit_dio_repair_bio);
7890 start += sectorsize;
7891 ASSERT(bio_offset + sectorsize > bio_offset);
7892 bio_offset += sectorsize;
7893 pgoff += sectorsize;
7899 static void __endio_write_update_ordered(struct btrfs_inode *inode,
7900 const u64 offset, const u64 bytes,
7901 const bool uptodate)
7903 struct btrfs_fs_info *fs_info = inode->root->fs_info;
7904 struct btrfs_ordered_extent *ordered = NULL;
7905 struct btrfs_workqueue *wq;
7906 u64 ordered_offset = offset;
7907 u64 ordered_bytes = bytes;
7910 if (btrfs_is_free_space_inode(inode))
7911 wq = fs_info->endio_freespace_worker;
7913 wq = fs_info->endio_write_workers;
7915 while (ordered_offset < offset + bytes) {
7916 last_offset = ordered_offset;
7917 if (btrfs_dec_test_first_ordered_pending(inode, &ordered,
7921 btrfs_init_work(&ordered->work, finish_ordered_fn, NULL,
7923 btrfs_queue_work(wq, &ordered->work);
7926 /* No ordered extent found in the range, exit */
7927 if (ordered_offset == last_offset)
7930 * Our bio might span multiple ordered extents. In this case
7931 * we keep going until we have accounted the whole dio.
7933 if (ordered_offset < offset + bytes) {
7934 ordered_bytes = offset + bytes - ordered_offset;
7940 static blk_status_t btrfs_submit_bio_start_direct_io(struct inode *inode,
7942 u64 dio_file_offset)
7944 return btrfs_csum_one_bio(BTRFS_I(inode), bio, dio_file_offset, 1);
7947 static void btrfs_end_dio_bio(struct bio *bio)
7949 struct btrfs_dio_private *dip = bio->bi_private;
7950 blk_status_t err = bio->bi_status;
7953 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
7954 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
7955 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
7956 bio->bi_opf, bio->bi_iter.bi_sector,
7957 bio->bi_iter.bi_size, err);
7959 if (bio_op(bio) == REQ_OP_READ) {
7960 err = btrfs_check_read_dio_bio(dip->inode, btrfs_io_bio(bio),
7965 dip->dio_bio->bi_status = err;
7968 btrfs_dio_private_put(dip);
7971 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
7972 struct inode *inode, u64 file_offset, int async_submit)
7974 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7975 struct btrfs_dio_private *dip = bio->bi_private;
7976 bool write = btrfs_op(bio) == BTRFS_MAP_WRITE;
7979 /* Check btrfs_submit_bio_hook() for rules about async submit. */
7981 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7984 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
7989 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
7992 if (write && async_submit) {
7993 ret = btrfs_wq_submit_bio(inode, bio, 0, 0, file_offset,
7994 btrfs_submit_bio_start_direct_io);
7998 * If we aren't doing async submit, calculate the csum of the
8001 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, file_offset, 1);
8007 csum_offset = file_offset - dip->logical_offset;
8008 csum_offset >>= fs_info->sectorsize_bits;
8009 csum_offset *= fs_info->csum_size;
8010 btrfs_io_bio(bio)->csum = dip->csums + csum_offset;
8013 ret = btrfs_map_bio(fs_info, bio, 0);
8019 * If this succeeds, the btrfs_dio_private is responsible for cleaning up locked
8020 * or ordered extents whether or not we submit any bios.
8022 static struct btrfs_dio_private *btrfs_create_dio_private(struct bio *dio_bio,
8023 struct inode *inode,
8026 const bool write = (btrfs_op(dio_bio) == BTRFS_MAP_WRITE);
8027 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
8029 struct btrfs_dio_private *dip;
8031 dip_size = sizeof(*dip);
8032 if (!write && csum) {
8033 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8036 nblocks = dio_bio->bi_iter.bi_size >> fs_info->sectorsize_bits;
8037 dip_size += fs_info->csum_size * nblocks;
8040 dip = kzalloc(dip_size, GFP_NOFS);
8045 dip->logical_offset = file_offset;
8046 dip->bytes = dio_bio->bi_iter.bi_size;
8047 dip->disk_bytenr = dio_bio->bi_iter.bi_sector << 9;
8048 dip->dio_bio = dio_bio;
8049 refcount_set(&dip->refs, 1);
8053 static blk_qc_t btrfs_submit_direct(struct inode *inode, struct iomap *iomap,
8054 struct bio *dio_bio, loff_t file_offset)
8056 const bool write = (btrfs_op(dio_bio) == BTRFS_MAP_WRITE);
8057 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8058 const bool raid56 = (btrfs_data_alloc_profile(fs_info) &
8059 BTRFS_BLOCK_GROUP_RAID56_MASK);
8060 struct btrfs_dio_private *dip;
8063 int async_submit = 0;
8065 int clone_offset = 0;
8069 blk_status_t status;
8070 struct btrfs_io_geometry geom;
8071 struct btrfs_dio_data *dio_data = iomap->private;
8072 struct extent_map *em = NULL;
8074 dip = btrfs_create_dio_private(dio_bio, inode, file_offset);
8077 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8078 file_offset + dio_bio->bi_iter.bi_size - 1);
8080 dio_bio->bi_status = BLK_STS_RESOURCE;
8082 return BLK_QC_T_NONE;
8087 * Load the csums up front to reduce csum tree searches and
8088 * contention when submitting bios.
8090 * If we have csums disabled this will do nothing.
8092 status = btrfs_lookup_bio_sums(inode, dio_bio, dip->csums);
8093 if (status != BLK_STS_OK)
8097 start_sector = dio_bio->bi_iter.bi_sector;
8098 submit_len = dio_bio->bi_iter.bi_size;
8101 logical = start_sector << 9;
8102 em = btrfs_get_chunk_map(fs_info, logical, submit_len);
8104 status = errno_to_blk_status(PTR_ERR(em));
8108 ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(dio_bio),
8109 logical, submit_len, &geom);
8111 status = errno_to_blk_status(ret);
8114 ASSERT(geom.len <= INT_MAX);
8116 clone_len = min_t(int, submit_len, geom.len);
8119 * This will never fail as it's passing GPF_NOFS and
8120 * the allocation is backed by btrfs_bioset.
8122 bio = btrfs_bio_clone_partial(dio_bio, clone_offset, clone_len);
8123 bio->bi_private = dip;
8124 bio->bi_end_io = btrfs_end_dio_bio;
8125 btrfs_io_bio(bio)->logical = file_offset;
8127 ASSERT(submit_len >= clone_len);
8128 submit_len -= clone_len;
8131 * Increase the count before we submit the bio so we know
8132 * the end IO handler won't happen before we increase the
8133 * count. Otherwise, the dip might get freed before we're
8134 * done setting it up.
8136 * We transfer the initial reference to the last bio, so we
8137 * don't need to increment the reference count for the last one.
8139 if (submit_len > 0) {
8140 refcount_inc(&dip->refs);
8142 * If we are submitting more than one bio, submit them
8143 * all asynchronously. The exception is RAID 5 or 6, as
8144 * asynchronous checksums make it difficult to collect
8145 * full stripe writes.
8151 status = btrfs_submit_dio_bio(bio, inode, file_offset,
8156 refcount_dec(&dip->refs);
8160 dio_data->submitted += clone_len;
8161 clone_offset += clone_len;
8162 start_sector += clone_len >> 9;
8163 file_offset += clone_len;
8165 free_extent_map(em);
8166 } while (submit_len > 0);
8167 return BLK_QC_T_NONE;
8170 free_extent_map(em);
8172 dip->dio_bio->bi_status = status;
8173 btrfs_dio_private_put(dip);
8175 return BLK_QC_T_NONE;
8178 const struct iomap_ops btrfs_dio_iomap_ops = {
8179 .iomap_begin = btrfs_dio_iomap_begin,
8180 .iomap_end = btrfs_dio_iomap_end,
8183 const struct iomap_dio_ops btrfs_dio_ops = {
8184 .submit_io = btrfs_submit_direct,
8187 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8192 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
8196 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
8199 int btrfs_readpage(struct file *file, struct page *page)
8201 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
8202 u64 start = page_offset(page);
8203 u64 end = start + PAGE_SIZE - 1;
8204 unsigned long bio_flags = 0;
8205 struct bio *bio = NULL;
8208 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
8210 ret = btrfs_do_readpage(page, NULL, &bio, &bio_flags, 0, NULL);
8212 ret = submit_one_bio(bio, 0, bio_flags);
8216 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8218 struct inode *inode = page->mapping->host;
8221 if (current->flags & PF_MEMALLOC) {
8222 redirty_page_for_writepage(wbc, page);
8228 * If we are under memory pressure we will call this directly from the
8229 * VM, we need to make sure we have the inode referenced for the ordered
8230 * extent. If not just return like we didn't do anything.
8232 if (!igrab(inode)) {
8233 redirty_page_for_writepage(wbc, page);
8234 return AOP_WRITEPAGE_ACTIVATE;
8236 ret = extent_write_full_page(page, wbc);
8237 btrfs_add_delayed_iput(inode);
8241 static int btrfs_writepages(struct address_space *mapping,
8242 struct writeback_control *wbc)
8244 return extent_writepages(mapping, wbc);
8247 static void btrfs_readahead(struct readahead_control *rac)
8249 extent_readahead(rac);
8252 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8254 int ret = try_release_extent_mapping(page, gfp_flags);
8256 clear_page_extent_mapped(page);
8260 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8262 if (PageWriteback(page) || PageDirty(page))
8264 return __btrfs_releasepage(page, gfp_flags);
8267 #ifdef CONFIG_MIGRATION
8268 static int btrfs_migratepage(struct address_space *mapping,
8269 struct page *newpage, struct page *page,
8270 enum migrate_mode mode)
8274 ret = migrate_page_move_mapping(mapping, newpage, page, 0);
8275 if (ret != MIGRATEPAGE_SUCCESS)
8278 if (page_has_private(page))
8279 attach_page_private(newpage, detach_page_private(page));
8281 if (PagePrivate2(page)) {
8282 ClearPagePrivate2(page);
8283 SetPagePrivate2(newpage);
8286 if (mode != MIGRATE_SYNC_NO_COPY)
8287 migrate_page_copy(newpage, page);
8289 migrate_page_states(newpage, page);
8290 return MIGRATEPAGE_SUCCESS;
8294 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8295 unsigned int length)
8297 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
8298 struct extent_io_tree *tree = &inode->io_tree;
8299 struct btrfs_ordered_extent *ordered;
8300 struct extent_state *cached_state = NULL;
8301 u64 page_start = page_offset(page);
8302 u64 page_end = page_start + PAGE_SIZE - 1;
8305 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
8306 bool found_ordered = false;
8307 bool completed_ordered = false;
8310 * we have the page locked, so new writeback can't start,
8311 * and the dirty bit won't be cleared while we are here.
8313 * Wait for IO on this page so that we can safely clear
8314 * the PagePrivate2 bit and do ordered accounting
8316 wait_on_page_writeback(page);
8319 btrfs_releasepage(page, GFP_NOFS);
8323 if (!inode_evicting)
8324 lock_extent_bits(tree, page_start, page_end, &cached_state);
8328 ordered = btrfs_lookup_ordered_range(inode, start, page_end - start + 1);
8330 found_ordered = true;
8332 ordered->file_offset + ordered->num_bytes - 1);
8334 * IO on this page will never be started, so we need to account
8335 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8336 * here, must leave that up for the ordered extent completion.
8338 if (!inode_evicting)
8339 clear_extent_bit(tree, start, end,
8341 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8342 EXTENT_DEFRAG, 1, 0, &cached_state);
8344 * whoever cleared the private bit is responsible
8345 * for the finish_ordered_io
8347 if (TestClearPagePrivate2(page)) {
8348 struct btrfs_ordered_inode_tree *tree;
8351 tree = &inode->ordered_tree;
8353 spin_lock_irq(&tree->lock);
8354 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8355 new_len = start - ordered->file_offset;
8356 if (new_len < ordered->truncated_len)
8357 ordered->truncated_len = new_len;
8358 spin_unlock_irq(&tree->lock);
8360 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8362 end - start + 1, 1)) {
8363 btrfs_finish_ordered_io(ordered);
8364 completed_ordered = true;
8367 btrfs_put_ordered_extent(ordered);
8368 if (!inode_evicting) {
8369 cached_state = NULL;
8370 lock_extent_bits(tree, start, end,
8375 if (start < page_end)
8380 * Qgroup reserved space handler
8381 * Page here will be either
8382 * 1) Already written to disk or ordered extent already submitted
8383 * Then its QGROUP_RESERVED bit in io_tree is already cleaned.
8384 * Qgroup will be handled by its qgroup_record then.
8385 * btrfs_qgroup_free_data() call will do nothing here.
8387 * 2) Not written to disk yet
8388 * Then btrfs_qgroup_free_data() call will clear the QGROUP_RESERVED
8389 * bit of its io_tree, and free the qgroup reserved data space.
8390 * Since the IO will never happen for this page.
8392 btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
8393 if (!inode_evicting) {
8397 * If there's an ordered extent for this range and we have not
8398 * finished it ourselves, we must leave EXTENT_DELALLOC_NEW set
8399 * in the range for the ordered extent completion. We must also
8400 * not delete the range, otherwise we would lose that bit (and
8401 * any other bits set in the range). Make sure EXTENT_UPTODATE
8402 * is cleared if we don't delete, otherwise it can lead to
8403 * corruptions if the i_size is extented later.
8405 if (found_ordered && !completed_ordered)
8407 clear_extent_bit(tree, page_start, page_end, EXTENT_LOCKED |
8408 EXTENT_DELALLOC | EXTENT_UPTODATE |
8409 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1,
8410 delete, &cached_state);
8412 __btrfs_releasepage(page, GFP_NOFS);
8415 ClearPageChecked(page);
8416 clear_page_extent_mapped(page);
8420 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8421 * called from a page fault handler when a page is first dirtied. Hence we must
8422 * be careful to check for EOF conditions here. We set the page up correctly
8423 * for a written page which means we get ENOSPC checking when writing into
8424 * holes and correct delalloc and unwritten extent mapping on filesystems that
8425 * support these features.
8427 * We are not allowed to take the i_mutex here so we have to play games to
8428 * protect against truncate races as the page could now be beyond EOF. Because
8429 * truncate_setsize() writes the inode size before removing pages, once we have
8430 * the page lock we can determine safely if the page is beyond EOF. If it is not
8431 * beyond EOF, then the page is guaranteed safe against truncation until we
8434 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8436 struct page *page = vmf->page;
8437 struct inode *inode = file_inode(vmf->vma->vm_file);
8438 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8439 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8440 struct btrfs_ordered_extent *ordered;
8441 struct extent_state *cached_state = NULL;
8442 struct extent_changeset *data_reserved = NULL;
8444 unsigned long zero_start;
8454 reserved_space = PAGE_SIZE;
8456 sb_start_pagefault(inode->i_sb);
8457 page_start = page_offset(page);
8458 page_end = page_start + PAGE_SIZE - 1;
8462 * Reserving delalloc space after obtaining the page lock can lead to
8463 * deadlock. For example, if a dirty page is locked by this function
8464 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8465 * dirty page write out, then the btrfs_writepage() function could
8466 * end up waiting indefinitely to get a lock on the page currently
8467 * being processed by btrfs_page_mkwrite() function.
8469 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8470 page_start, reserved_space);
8472 ret2 = file_update_time(vmf->vma->vm_file);
8476 ret = vmf_error(ret2);
8482 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8485 size = i_size_read(inode);
8487 if ((page->mapping != inode->i_mapping) ||
8488 (page_start >= size)) {
8489 /* page got truncated out from underneath us */
8492 wait_on_page_writeback(page);
8494 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8495 ret2 = set_page_extent_mapped(page);
8497 ret = vmf_error(ret2);
8498 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8503 * we can't set the delalloc bits if there are pending ordered
8504 * extents. Drop our locks and wait for them to finish
8506 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8509 unlock_extent_cached(io_tree, page_start, page_end,
8512 btrfs_start_ordered_extent(ordered, 1);
8513 btrfs_put_ordered_extent(ordered);
8517 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8518 reserved_space = round_up(size - page_start,
8519 fs_info->sectorsize);
8520 if (reserved_space < PAGE_SIZE) {
8521 end = page_start + reserved_space - 1;
8522 btrfs_delalloc_release_space(BTRFS_I(inode),
8523 data_reserved, page_start,
8524 PAGE_SIZE - reserved_space, true);
8529 * page_mkwrite gets called when the page is firstly dirtied after it's
8530 * faulted in, but write(2) could also dirty a page and set delalloc
8531 * bits, thus in this case for space account reason, we still need to
8532 * clear any delalloc bits within this page range since we have to
8533 * reserve data&meta space before lock_page() (see above comments).
8535 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8536 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8537 EXTENT_DEFRAG, 0, 0, &cached_state);
8539 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8542 unlock_extent_cached(io_tree, page_start, page_end,
8544 ret = VM_FAULT_SIGBUS;
8548 /* page is wholly or partially inside EOF */
8549 if (page_start + PAGE_SIZE > size)
8550 zero_start = offset_in_page(size);
8552 zero_start = PAGE_SIZE;
8554 if (zero_start != PAGE_SIZE) {
8556 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
8557 flush_dcache_page(page);
8560 ClearPageChecked(page);
8561 set_page_dirty(page);
8562 SetPageUptodate(page);
8564 BTRFS_I(inode)->last_trans = fs_info->generation;
8565 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8566 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8568 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8570 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8571 sb_end_pagefault(inode->i_sb);
8572 extent_changeset_free(data_reserved);
8573 return VM_FAULT_LOCKED;
8578 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8579 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8580 reserved_space, (ret != 0));
8582 sb_end_pagefault(inode->i_sb);
8583 extent_changeset_free(data_reserved);
8587 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
8589 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8590 struct btrfs_root *root = BTRFS_I(inode)->root;
8591 struct btrfs_block_rsv *rsv;
8593 struct btrfs_trans_handle *trans;
8594 u64 mask = fs_info->sectorsize - 1;
8595 u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8597 if (!skip_writeback) {
8598 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8605 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8606 * things going on here:
8608 * 1) We need to reserve space to update our inode.
8610 * 2) We need to have something to cache all the space that is going to
8611 * be free'd up by the truncate operation, but also have some slack
8612 * space reserved in case it uses space during the truncate (thank you
8613 * very much snapshotting).
8615 * And we need these to be separate. The fact is we can use a lot of
8616 * space doing the truncate, and we have no earthly idea how much space
8617 * we will use, so we need the truncate reservation to be separate so it
8618 * doesn't end up using space reserved for updating the inode. We also
8619 * need to be able to stop the transaction and start a new one, which
8620 * means we need to be able to update the inode several times, and we
8621 * have no idea of knowing how many times that will be, so we can't just
8622 * reserve 1 item for the entirety of the operation, so that has to be
8623 * done separately as well.
8625 * So that leaves us with
8627 * 1) rsv - for the truncate reservation, which we will steal from the
8628 * transaction reservation.
8629 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8630 * updating the inode.
8632 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8635 rsv->size = min_size;
8639 * 1 for the truncate slack space
8640 * 1 for updating the inode.
8642 trans = btrfs_start_transaction(root, 2);
8643 if (IS_ERR(trans)) {
8644 ret = PTR_ERR(trans);
8648 /* Migrate the slack space for the truncate to our reserve */
8649 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8654 * So if we truncate and then write and fsync we normally would just
8655 * write the extents that changed, which is a problem if we need to
8656 * first truncate that entire inode. So set this flag so we write out
8657 * all of the extents in the inode to the sync log so we're completely
8660 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
8661 trans->block_rsv = rsv;
8664 ret = btrfs_truncate_inode_items(trans, root, BTRFS_I(inode),
8666 BTRFS_EXTENT_DATA_KEY);
8667 trans->block_rsv = &fs_info->trans_block_rsv;
8668 if (ret != -ENOSPC && ret != -EAGAIN)
8671 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
8675 btrfs_end_transaction(trans);
8676 btrfs_btree_balance_dirty(fs_info);
8678 trans = btrfs_start_transaction(root, 2);
8679 if (IS_ERR(trans)) {
8680 ret = PTR_ERR(trans);
8685 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8686 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8687 rsv, min_size, false);
8688 BUG_ON(ret); /* shouldn't happen */
8689 trans->block_rsv = rsv;
8693 * We can't call btrfs_truncate_block inside a trans handle as we could
8694 * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
8695 * we've truncated everything except the last little bit, and can do
8696 * btrfs_truncate_block and then update the disk_i_size.
8698 if (ret == NEED_TRUNCATE_BLOCK) {
8699 btrfs_end_transaction(trans);
8700 btrfs_btree_balance_dirty(fs_info);
8702 ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0);
8705 trans = btrfs_start_transaction(root, 1);
8706 if (IS_ERR(trans)) {
8707 ret = PTR_ERR(trans);
8710 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
8716 trans->block_rsv = &fs_info->trans_block_rsv;
8717 ret2 = btrfs_update_inode(trans, root, BTRFS_I(inode));
8721 ret2 = btrfs_end_transaction(trans);
8724 btrfs_btree_balance_dirty(fs_info);
8727 btrfs_free_block_rsv(fs_info, rsv);
8733 * create a new subvolume directory/inode (helper for the ioctl).
8735 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8736 struct btrfs_root *new_root,
8737 struct btrfs_root *parent_root)
8739 struct inode *inode;
8744 err = btrfs_get_free_objectid(new_root, &ino);
8748 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, ino, ino,
8749 S_IFDIR | (~current_umask() & S_IRWXUGO),
8752 return PTR_ERR(inode);
8753 inode->i_op = &btrfs_dir_inode_operations;
8754 inode->i_fop = &btrfs_dir_file_operations;
8756 set_nlink(inode, 1);
8757 btrfs_i_size_write(BTRFS_I(inode), 0);
8758 unlock_new_inode(inode);
8760 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8762 btrfs_err(new_root->fs_info,
8763 "error inheriting subvolume %llu properties: %d",
8764 new_root->root_key.objectid, err);
8766 err = btrfs_update_inode(trans, new_root, BTRFS_I(inode));
8772 struct inode *btrfs_alloc_inode(struct super_block *sb)
8774 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8775 struct btrfs_inode *ei;
8776 struct inode *inode;
8778 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL);
8785 ei->last_sub_trans = 0;
8786 ei->logged_trans = 0;
8787 ei->delalloc_bytes = 0;
8788 ei->new_delalloc_bytes = 0;
8789 ei->defrag_bytes = 0;
8790 ei->disk_i_size = 0;
8793 ei->index_cnt = (u64)-1;
8795 ei->last_unlink_trans = 0;
8796 ei->last_reflink_trans = 0;
8797 ei->last_log_commit = 0;
8799 spin_lock_init(&ei->lock);
8800 ei->outstanding_extents = 0;
8801 if (sb->s_magic != BTRFS_TEST_MAGIC)
8802 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8803 BTRFS_BLOCK_RSV_DELALLOC);
8804 ei->runtime_flags = 0;
8805 ei->prop_compress = BTRFS_COMPRESS_NONE;
8806 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8808 ei->delayed_node = NULL;
8810 ei->i_otime.tv_sec = 0;
8811 ei->i_otime.tv_nsec = 0;
8813 inode = &ei->vfs_inode;
8814 extent_map_tree_init(&ei->extent_tree);
8815 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO, inode);
8816 extent_io_tree_init(fs_info, &ei->io_failure_tree,
8817 IO_TREE_INODE_IO_FAILURE, inode);
8818 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8819 IO_TREE_INODE_FILE_EXTENT, inode);
8820 ei->io_tree.track_uptodate = true;
8821 ei->io_failure_tree.track_uptodate = true;
8822 atomic_set(&ei->sync_writers, 0);
8823 mutex_init(&ei->log_mutex);
8824 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8825 INIT_LIST_HEAD(&ei->delalloc_inodes);
8826 INIT_LIST_HEAD(&ei->delayed_iput);
8827 RB_CLEAR_NODE(&ei->rb_node);
8832 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8833 void btrfs_test_destroy_inode(struct inode *inode)
8835 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
8836 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8840 void btrfs_free_inode(struct inode *inode)
8842 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8845 void btrfs_destroy_inode(struct inode *vfs_inode)
8847 struct btrfs_ordered_extent *ordered;
8848 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8849 struct btrfs_root *root = inode->root;
8851 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8852 WARN_ON(vfs_inode->i_data.nrpages);
8853 WARN_ON(inode->block_rsv.reserved);
8854 WARN_ON(inode->block_rsv.size);
8855 WARN_ON(inode->outstanding_extents);
8856 WARN_ON(inode->delalloc_bytes);
8857 WARN_ON(inode->new_delalloc_bytes);
8858 WARN_ON(inode->csum_bytes);
8859 WARN_ON(inode->defrag_bytes);
8862 * This can happen where we create an inode, but somebody else also
8863 * created the same inode and we need to destroy the one we already
8870 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8874 btrfs_err(root->fs_info,
8875 "found ordered extent %llu %llu on inode cleanup",
8876 ordered->file_offset, ordered->num_bytes);
8877 btrfs_remove_ordered_extent(inode, ordered);
8878 btrfs_put_ordered_extent(ordered);
8879 btrfs_put_ordered_extent(ordered);
8882 btrfs_qgroup_check_reserved_leak(inode);
8883 inode_tree_del(inode);
8884 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8885 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
8886 btrfs_put_root(inode->root);
8889 int btrfs_drop_inode(struct inode *inode)
8891 struct btrfs_root *root = BTRFS_I(inode)->root;
8896 /* the snap/subvol tree is on deleting */
8897 if (btrfs_root_refs(&root->root_item) == 0)
8900 return generic_drop_inode(inode);
8903 static void init_once(void *foo)
8905 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8907 inode_init_once(&ei->vfs_inode);
8910 void __cold btrfs_destroy_cachep(void)
8913 * Make sure all delayed rcu free inodes are flushed before we
8917 kmem_cache_destroy(btrfs_inode_cachep);
8918 kmem_cache_destroy(btrfs_trans_handle_cachep);
8919 kmem_cache_destroy(btrfs_path_cachep);
8920 kmem_cache_destroy(btrfs_free_space_cachep);
8921 kmem_cache_destroy(btrfs_free_space_bitmap_cachep);
8924 int __init btrfs_init_cachep(void)
8926 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8927 sizeof(struct btrfs_inode), 0,
8928 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
8930 if (!btrfs_inode_cachep)
8933 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8934 sizeof(struct btrfs_trans_handle), 0,
8935 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
8936 if (!btrfs_trans_handle_cachep)
8939 btrfs_path_cachep = kmem_cache_create("btrfs_path",
8940 sizeof(struct btrfs_path), 0,
8941 SLAB_MEM_SPREAD, NULL);
8942 if (!btrfs_path_cachep)
8945 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8946 sizeof(struct btrfs_free_space), 0,
8947 SLAB_MEM_SPREAD, NULL);
8948 if (!btrfs_free_space_cachep)
8951 btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap",
8952 PAGE_SIZE, PAGE_SIZE,
8953 SLAB_RED_ZONE, NULL);
8954 if (!btrfs_free_space_bitmap_cachep)
8959 btrfs_destroy_cachep();
8963 static int btrfs_getattr(const struct path *path, struct kstat *stat,
8964 u32 request_mask, unsigned int flags)
8968 struct inode *inode = d_inode(path->dentry);
8969 u32 blocksize = inode->i_sb->s_blocksize;
8970 u32 bi_flags = BTRFS_I(inode)->flags;
8972 stat->result_mask |= STATX_BTIME;
8973 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
8974 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
8975 if (bi_flags & BTRFS_INODE_APPEND)
8976 stat->attributes |= STATX_ATTR_APPEND;
8977 if (bi_flags & BTRFS_INODE_COMPRESS)
8978 stat->attributes |= STATX_ATTR_COMPRESSED;
8979 if (bi_flags & BTRFS_INODE_IMMUTABLE)
8980 stat->attributes |= STATX_ATTR_IMMUTABLE;
8981 if (bi_flags & BTRFS_INODE_NODUMP)
8982 stat->attributes |= STATX_ATTR_NODUMP;
8984 stat->attributes_mask |= (STATX_ATTR_APPEND |
8985 STATX_ATTR_COMPRESSED |
8986 STATX_ATTR_IMMUTABLE |
8989 generic_fillattr(inode, stat);
8990 stat->dev = BTRFS_I(inode)->root->anon_dev;
8992 spin_lock(&BTRFS_I(inode)->lock);
8993 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
8994 inode_bytes = inode_get_bytes(inode);
8995 spin_unlock(&BTRFS_I(inode)->lock);
8996 stat->blocks = (ALIGN(inode_bytes, blocksize) +
8997 ALIGN(delalloc_bytes, blocksize)) >> 9;
9001 static int btrfs_rename_exchange(struct inode *old_dir,
9002 struct dentry *old_dentry,
9003 struct inode *new_dir,
9004 struct dentry *new_dentry)
9006 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9007 struct btrfs_trans_handle *trans;
9008 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9009 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9010 struct inode *new_inode = new_dentry->d_inode;
9011 struct inode *old_inode = old_dentry->d_inode;
9012 struct timespec64 ctime = current_time(old_inode);
9013 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9014 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9019 bool root_log_pinned = false;
9020 bool dest_log_pinned = false;
9022 /* we only allow rename subvolume link between subvolumes */
9023 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9026 /* close the race window with snapshot create/destroy ioctl */
9027 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
9028 new_ino == BTRFS_FIRST_FREE_OBJECTID)
9029 down_read(&fs_info->subvol_sem);
9032 * We want to reserve the absolute worst case amount of items. So if
9033 * both inodes are subvols and we need to unlink them then that would
9034 * require 4 item modifications, but if they are both normal inodes it
9035 * would require 5 item modifications, so we'll assume their normal
9036 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9037 * should cover the worst case number of items we'll modify.
9039 trans = btrfs_start_transaction(root, 12);
9040 if (IS_ERR(trans)) {
9041 ret = PTR_ERR(trans);
9046 btrfs_record_root_in_trans(trans, dest);
9049 * We need to find a free sequence number both in the source and
9050 * in the destination directory for the exchange.
9052 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9055 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9059 BTRFS_I(old_inode)->dir_index = 0ULL;
9060 BTRFS_I(new_inode)->dir_index = 0ULL;
9062 /* Reference for the source. */
9063 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9064 /* force full log commit if subvolume involved. */
9065 btrfs_set_log_full_commit(trans);
9067 btrfs_pin_log_trans(root);
9068 root_log_pinned = true;
9069 ret = btrfs_insert_inode_ref(trans, dest,
9070 new_dentry->d_name.name,
9071 new_dentry->d_name.len,
9073 btrfs_ino(BTRFS_I(new_dir)),
9079 /* And now for the dest. */
9080 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9081 /* force full log commit if subvolume involved. */
9082 btrfs_set_log_full_commit(trans);
9084 btrfs_pin_log_trans(dest);
9085 dest_log_pinned = true;
9086 ret = btrfs_insert_inode_ref(trans, root,
9087 old_dentry->d_name.name,
9088 old_dentry->d_name.len,
9090 btrfs_ino(BTRFS_I(old_dir)),
9096 /* Update inode version and ctime/mtime. */
9097 inode_inc_iversion(old_dir);
9098 inode_inc_iversion(new_dir);
9099 inode_inc_iversion(old_inode);
9100 inode_inc_iversion(new_inode);
9101 old_dir->i_ctime = old_dir->i_mtime = ctime;
9102 new_dir->i_ctime = new_dir->i_mtime = ctime;
9103 old_inode->i_ctime = ctime;
9104 new_inode->i_ctime = ctime;
9106 if (old_dentry->d_parent != new_dentry->d_parent) {
9107 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9108 BTRFS_I(old_inode), 1);
9109 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9110 BTRFS_I(new_inode), 1);
9113 /* src is a subvolume */
9114 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9115 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9116 } else { /* src is an inode */
9117 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9118 BTRFS_I(old_dentry->d_inode),
9119 old_dentry->d_name.name,
9120 old_dentry->d_name.len);
9122 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9125 btrfs_abort_transaction(trans, ret);
9129 /* dest is a subvolume */
9130 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9131 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9132 } else { /* dest is an inode */
9133 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9134 BTRFS_I(new_dentry->d_inode),
9135 new_dentry->d_name.name,
9136 new_dentry->d_name.len);
9138 ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode));
9141 btrfs_abort_transaction(trans, ret);
9145 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9146 new_dentry->d_name.name,
9147 new_dentry->d_name.len, 0, old_idx);
9149 btrfs_abort_transaction(trans, ret);
9153 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9154 old_dentry->d_name.name,
9155 old_dentry->d_name.len, 0, new_idx);
9157 btrfs_abort_transaction(trans, ret);
9161 if (old_inode->i_nlink == 1)
9162 BTRFS_I(old_inode)->dir_index = old_idx;
9163 if (new_inode->i_nlink == 1)
9164 BTRFS_I(new_inode)->dir_index = new_idx;
9166 if (root_log_pinned) {
9167 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9168 new_dentry->d_parent);
9169 btrfs_end_log_trans(root);
9170 root_log_pinned = false;
9172 if (dest_log_pinned) {
9173 btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir),
9174 old_dentry->d_parent);
9175 btrfs_end_log_trans(dest);
9176 dest_log_pinned = false;
9180 * If we have pinned a log and an error happened, we unpin tasks
9181 * trying to sync the log and force them to fallback to a transaction
9182 * commit if the log currently contains any of the inodes involved in
9183 * this rename operation (to ensure we do not persist a log with an
9184 * inconsistent state for any of these inodes or leading to any
9185 * inconsistencies when replayed). If the transaction was aborted, the
9186 * abortion reason is propagated to userspace when attempting to commit
9187 * the transaction. If the log does not contain any of these inodes, we
9188 * allow the tasks to sync it.
9190 if (ret && (root_log_pinned || dest_log_pinned)) {
9191 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9192 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9193 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9195 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9196 btrfs_set_log_full_commit(trans);
9198 if (root_log_pinned) {
9199 btrfs_end_log_trans(root);
9200 root_log_pinned = false;
9202 if (dest_log_pinned) {
9203 btrfs_end_log_trans(dest);
9204 dest_log_pinned = false;
9207 ret2 = btrfs_end_transaction(trans);
9208 ret = ret ? ret : ret2;
9210 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
9211 old_ino == BTRFS_FIRST_FREE_OBJECTID)
9212 up_read(&fs_info->subvol_sem);
9217 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9218 struct btrfs_root *root,
9220 struct dentry *dentry)
9223 struct inode *inode;
9227 ret = btrfs_get_free_objectid(root, &objectid);
9231 inode = btrfs_new_inode(trans, root, dir,
9232 dentry->d_name.name,
9234 btrfs_ino(BTRFS_I(dir)),
9236 S_IFCHR | WHITEOUT_MODE,
9239 if (IS_ERR(inode)) {
9240 ret = PTR_ERR(inode);
9244 inode->i_op = &btrfs_special_inode_operations;
9245 init_special_inode(inode, inode->i_mode,
9248 ret = btrfs_init_inode_security(trans, inode, dir,
9253 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9254 BTRFS_I(inode), 0, index);
9258 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
9260 unlock_new_inode(inode);
9262 inode_dec_link_count(inode);
9268 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9269 struct inode *new_dir, struct dentry *new_dentry,
9272 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9273 struct btrfs_trans_handle *trans;
9274 unsigned int trans_num_items;
9275 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9276 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9277 struct inode *new_inode = d_inode(new_dentry);
9278 struct inode *old_inode = d_inode(old_dentry);
9282 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9283 bool log_pinned = false;
9285 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9288 /* we only allow rename subvolume link between subvolumes */
9289 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9292 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9293 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9296 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9297 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9301 /* check for collisions, even if the name isn't there */
9302 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9303 new_dentry->d_name.name,
9304 new_dentry->d_name.len);
9307 if (ret == -EEXIST) {
9309 * eexist without a new_inode */
9310 if (WARN_ON(!new_inode)) {
9314 /* maybe -EOVERFLOW */
9321 * we're using rename to replace one file with another. Start IO on it
9322 * now so we don't add too much work to the end of the transaction
9324 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9325 filemap_flush(old_inode->i_mapping);
9327 /* close the racy window with snapshot create/destroy ioctl */
9328 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9329 down_read(&fs_info->subvol_sem);
9331 * We want to reserve the absolute worst case amount of items. So if
9332 * both inodes are subvols and we need to unlink them then that would
9333 * require 4 item modifications, but if they are both normal inodes it
9334 * would require 5 item modifications, so we'll assume they are normal
9335 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9336 * should cover the worst case number of items we'll modify.
9337 * If our rename has the whiteout flag, we need more 5 units for the
9338 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9339 * when selinux is enabled).
9341 trans_num_items = 11;
9342 if (flags & RENAME_WHITEOUT)
9343 trans_num_items += 5;
9344 trans = btrfs_start_transaction(root, trans_num_items);
9345 if (IS_ERR(trans)) {
9346 ret = PTR_ERR(trans);
9351 btrfs_record_root_in_trans(trans, dest);
9353 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9357 BTRFS_I(old_inode)->dir_index = 0ULL;
9358 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9359 /* force full log commit if subvolume involved. */
9360 btrfs_set_log_full_commit(trans);
9362 btrfs_pin_log_trans(root);
9364 ret = btrfs_insert_inode_ref(trans, dest,
9365 new_dentry->d_name.name,
9366 new_dentry->d_name.len,
9368 btrfs_ino(BTRFS_I(new_dir)), index);
9373 inode_inc_iversion(old_dir);
9374 inode_inc_iversion(new_dir);
9375 inode_inc_iversion(old_inode);
9376 old_dir->i_ctime = old_dir->i_mtime =
9377 new_dir->i_ctime = new_dir->i_mtime =
9378 old_inode->i_ctime = current_time(old_dir);
9380 if (old_dentry->d_parent != new_dentry->d_parent)
9381 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9382 BTRFS_I(old_inode), 1);
9384 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9385 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9387 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9388 BTRFS_I(d_inode(old_dentry)),
9389 old_dentry->d_name.name,
9390 old_dentry->d_name.len);
9392 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9395 btrfs_abort_transaction(trans, ret);
9400 inode_inc_iversion(new_inode);
9401 new_inode->i_ctime = current_time(new_inode);
9402 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9403 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9404 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9405 BUG_ON(new_inode->i_nlink == 0);
9407 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9408 BTRFS_I(d_inode(new_dentry)),
9409 new_dentry->d_name.name,
9410 new_dentry->d_name.len);
9412 if (!ret && new_inode->i_nlink == 0)
9413 ret = btrfs_orphan_add(trans,
9414 BTRFS_I(d_inode(new_dentry)));
9416 btrfs_abort_transaction(trans, ret);
9421 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9422 new_dentry->d_name.name,
9423 new_dentry->d_name.len, 0, index);
9425 btrfs_abort_transaction(trans, ret);
9429 if (old_inode->i_nlink == 1)
9430 BTRFS_I(old_inode)->dir_index = index;
9433 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9434 new_dentry->d_parent);
9435 btrfs_end_log_trans(root);
9439 if (flags & RENAME_WHITEOUT) {
9440 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9444 btrfs_abort_transaction(trans, ret);
9450 * If we have pinned the log and an error happened, we unpin tasks
9451 * trying to sync the log and force them to fallback to a transaction
9452 * commit if the log currently contains any of the inodes involved in
9453 * this rename operation (to ensure we do not persist a log with an
9454 * inconsistent state for any of these inodes or leading to any
9455 * inconsistencies when replayed). If the transaction was aborted, the
9456 * abortion reason is propagated to userspace when attempting to commit
9457 * the transaction. If the log does not contain any of these inodes, we
9458 * allow the tasks to sync it.
9460 if (ret && log_pinned) {
9461 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9462 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9463 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9465 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9466 btrfs_set_log_full_commit(trans);
9468 btrfs_end_log_trans(root);
9471 ret2 = btrfs_end_transaction(trans);
9472 ret = ret ? ret : ret2;
9474 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9475 up_read(&fs_info->subvol_sem);
9480 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9481 struct inode *new_dir, struct dentry *new_dentry,
9484 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9487 if (flags & RENAME_EXCHANGE)
9488 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9491 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9494 struct btrfs_delalloc_work {
9495 struct inode *inode;
9496 struct completion completion;
9497 struct list_head list;
9498 struct btrfs_work work;
9501 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9503 struct btrfs_delalloc_work *delalloc_work;
9504 struct inode *inode;
9506 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9508 inode = delalloc_work->inode;
9509 filemap_flush(inode->i_mapping);
9510 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9511 &BTRFS_I(inode)->runtime_flags))
9512 filemap_flush(inode->i_mapping);
9515 complete(&delalloc_work->completion);
9518 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9520 struct btrfs_delalloc_work *work;
9522 work = kmalloc(sizeof(*work), GFP_NOFS);
9526 init_completion(&work->completion);
9527 INIT_LIST_HEAD(&work->list);
9528 work->inode = inode;
9529 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9535 * some fairly slow code that needs optimization. This walks the list
9536 * of all the inodes with pending delalloc and forces them to disk.
9538 static int start_delalloc_inodes(struct btrfs_root *root,
9539 struct writeback_control *wbc, bool snapshot,
9540 bool in_reclaim_context)
9542 struct btrfs_inode *binode;
9543 struct inode *inode;
9544 struct btrfs_delalloc_work *work, *next;
9545 struct list_head works;
9546 struct list_head splice;
9548 bool full_flush = wbc->nr_to_write == LONG_MAX;
9550 INIT_LIST_HEAD(&works);
9551 INIT_LIST_HEAD(&splice);
9553 mutex_lock(&root->delalloc_mutex);
9554 spin_lock(&root->delalloc_lock);
9555 list_splice_init(&root->delalloc_inodes, &splice);
9556 while (!list_empty(&splice)) {
9557 binode = list_entry(splice.next, struct btrfs_inode,
9560 list_move_tail(&binode->delalloc_inodes,
9561 &root->delalloc_inodes);
9563 if (in_reclaim_context &&
9564 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9567 inode = igrab(&binode->vfs_inode);
9569 cond_resched_lock(&root->delalloc_lock);
9572 spin_unlock(&root->delalloc_lock);
9575 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9576 &binode->runtime_flags);
9578 work = btrfs_alloc_delalloc_work(inode);
9584 list_add_tail(&work->list, &works);
9585 btrfs_queue_work(root->fs_info->flush_workers,
9588 ret = sync_inode(inode, wbc);
9590 test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9591 &BTRFS_I(inode)->runtime_flags))
9592 ret = sync_inode(inode, wbc);
9593 btrfs_add_delayed_iput(inode);
9594 if (ret || wbc->nr_to_write <= 0)
9598 spin_lock(&root->delalloc_lock);
9600 spin_unlock(&root->delalloc_lock);
9603 list_for_each_entry_safe(work, next, &works, list) {
9604 list_del_init(&work->list);
9605 wait_for_completion(&work->completion);
9609 if (!list_empty(&splice)) {
9610 spin_lock(&root->delalloc_lock);
9611 list_splice_tail(&splice, &root->delalloc_inodes);
9612 spin_unlock(&root->delalloc_lock);
9614 mutex_unlock(&root->delalloc_mutex);
9618 int btrfs_start_delalloc_snapshot(struct btrfs_root *root)
9620 struct writeback_control wbc = {
9621 .nr_to_write = LONG_MAX,
9622 .sync_mode = WB_SYNC_NONE,
9624 .range_end = LLONG_MAX,
9626 struct btrfs_fs_info *fs_info = root->fs_info;
9628 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9631 return start_delalloc_inodes(root, &wbc, true, false);
9634 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9635 bool in_reclaim_context)
9637 struct writeback_control wbc = {
9639 .sync_mode = WB_SYNC_NONE,
9641 .range_end = LLONG_MAX,
9643 struct btrfs_root *root;
9644 struct list_head splice;
9647 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9650 INIT_LIST_HEAD(&splice);
9652 mutex_lock(&fs_info->delalloc_root_mutex);
9653 spin_lock(&fs_info->delalloc_root_lock);
9654 list_splice_init(&fs_info->delalloc_roots, &splice);
9655 while (!list_empty(&splice)) {
9657 * Reset nr_to_write here so we know that we're doing a full
9661 wbc.nr_to_write = LONG_MAX;
9663 root = list_first_entry(&splice, struct btrfs_root,
9665 root = btrfs_grab_root(root);
9667 list_move_tail(&root->delalloc_root,
9668 &fs_info->delalloc_roots);
9669 spin_unlock(&fs_info->delalloc_root_lock);
9671 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9672 btrfs_put_root(root);
9673 if (ret < 0 || wbc.nr_to_write <= 0)
9675 spin_lock(&fs_info->delalloc_root_lock);
9677 spin_unlock(&fs_info->delalloc_root_lock);
9681 if (!list_empty(&splice)) {
9682 spin_lock(&fs_info->delalloc_root_lock);
9683 list_splice_tail(&splice, &fs_info->delalloc_roots);
9684 spin_unlock(&fs_info->delalloc_root_lock);
9686 mutex_unlock(&fs_info->delalloc_root_mutex);
9690 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9691 const char *symname)
9693 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9694 struct btrfs_trans_handle *trans;
9695 struct btrfs_root *root = BTRFS_I(dir)->root;
9696 struct btrfs_path *path;
9697 struct btrfs_key key;
9698 struct inode *inode = NULL;
9705 struct btrfs_file_extent_item *ei;
9706 struct extent_buffer *leaf;
9708 name_len = strlen(symname);
9709 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9710 return -ENAMETOOLONG;
9713 * 2 items for inode item and ref
9714 * 2 items for dir items
9715 * 1 item for updating parent inode item
9716 * 1 item for the inline extent item
9717 * 1 item for xattr if selinux is on
9719 trans = btrfs_start_transaction(root, 7);
9721 return PTR_ERR(trans);
9723 err = btrfs_get_free_objectid(root, &objectid);
9727 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9728 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
9729 objectid, S_IFLNK|S_IRWXUGO, &index);
9730 if (IS_ERR(inode)) {
9731 err = PTR_ERR(inode);
9737 * If the active LSM wants to access the inode during
9738 * d_instantiate it needs these. Smack checks to see
9739 * if the filesystem supports xattrs by looking at the
9742 inode->i_fop = &btrfs_file_operations;
9743 inode->i_op = &btrfs_file_inode_operations;
9744 inode->i_mapping->a_ops = &btrfs_aops;
9746 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9750 path = btrfs_alloc_path();
9755 key.objectid = btrfs_ino(BTRFS_I(inode));
9757 key.type = BTRFS_EXTENT_DATA_KEY;
9758 datasize = btrfs_file_extent_calc_inline_size(name_len);
9759 err = btrfs_insert_empty_item(trans, root, path, &key,
9762 btrfs_free_path(path);
9765 leaf = path->nodes[0];
9766 ei = btrfs_item_ptr(leaf, path->slots[0],
9767 struct btrfs_file_extent_item);
9768 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9769 btrfs_set_file_extent_type(leaf, ei,
9770 BTRFS_FILE_EXTENT_INLINE);
9771 btrfs_set_file_extent_encryption(leaf, ei, 0);
9772 btrfs_set_file_extent_compression(leaf, ei, 0);
9773 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9774 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9776 ptr = btrfs_file_extent_inline_start(ei);
9777 write_extent_buffer(leaf, symname, ptr, name_len);
9778 btrfs_mark_buffer_dirty(leaf);
9779 btrfs_free_path(path);
9781 inode->i_op = &btrfs_symlink_inode_operations;
9782 inode_nohighmem(inode);
9783 inode_set_bytes(inode, name_len);
9784 btrfs_i_size_write(BTRFS_I(inode), name_len);
9785 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
9787 * Last step, add directory indexes for our symlink inode. This is the
9788 * last step to avoid extra cleanup of these indexes if an error happens
9792 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9793 BTRFS_I(inode), 0, index);
9797 d_instantiate_new(dentry, inode);
9800 btrfs_end_transaction(trans);
9802 inode_dec_link_count(inode);
9803 discard_new_inode(inode);
9805 btrfs_btree_balance_dirty(fs_info);
9809 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9810 struct btrfs_trans_handle *trans_in,
9811 struct btrfs_inode *inode,
9812 struct btrfs_key *ins,
9815 struct btrfs_file_extent_item stack_fi;
9816 struct btrfs_replace_extent_info extent_info;
9817 struct btrfs_trans_handle *trans = trans_in;
9818 struct btrfs_path *path;
9819 u64 start = ins->objectid;
9820 u64 len = ins->offset;
9823 memset(&stack_fi, 0, sizeof(stack_fi));
9825 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9826 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9827 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9828 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9829 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9830 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9831 /* Encryption and other encoding is reserved and all 0 */
9833 ret = btrfs_qgroup_release_data(inode, file_offset, len);
9835 return ERR_PTR(ret);
9838 ret = insert_reserved_file_extent(trans, inode,
9839 file_offset, &stack_fi,
9842 return ERR_PTR(ret);
9846 extent_info.disk_offset = start;
9847 extent_info.disk_len = len;
9848 extent_info.data_offset = 0;
9849 extent_info.data_len = len;
9850 extent_info.file_offset = file_offset;
9851 extent_info.extent_buf = (char *)&stack_fi;
9852 extent_info.is_new_extent = true;
9853 extent_info.qgroup_reserved = ret;
9854 extent_info.insertions = 0;
9856 path = btrfs_alloc_path();
9858 return ERR_PTR(-ENOMEM);
9860 ret = btrfs_replace_file_extents(&inode->vfs_inode, path, file_offset,
9861 file_offset + len - 1, &extent_info,
9863 btrfs_free_path(path);
9865 return ERR_PTR(ret);
9870 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9871 u64 start, u64 num_bytes, u64 min_size,
9872 loff_t actual_len, u64 *alloc_hint,
9873 struct btrfs_trans_handle *trans)
9875 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9876 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9877 struct extent_map *em;
9878 struct btrfs_root *root = BTRFS_I(inode)->root;
9879 struct btrfs_key ins;
9880 u64 cur_offset = start;
9881 u64 clear_offset = start;
9884 u64 last_alloc = (u64)-1;
9886 bool own_trans = true;
9887 u64 end = start + num_bytes - 1;
9891 while (num_bytes > 0) {
9892 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9893 cur_bytes = max(cur_bytes, min_size);
9895 * If we are severely fragmented we could end up with really
9896 * small allocations, so if the allocator is returning small
9897 * chunks lets make its job easier by only searching for those
9900 cur_bytes = min(cur_bytes, last_alloc);
9901 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9902 min_size, 0, *alloc_hint, &ins, 1, 0);
9907 * We've reserved this space, and thus converted it from
9908 * ->bytes_may_use to ->bytes_reserved. Any error that happens
9909 * from here on out we will only need to clear our reservation
9910 * for the remaining unreserved area, so advance our
9911 * clear_offset by our extent size.
9913 clear_offset += ins.offset;
9915 last_alloc = ins.offset;
9916 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
9919 * Now that we inserted the prealloc extent we can finally
9920 * decrement the number of reservations in the block group.
9921 * If we did it before, we could race with relocation and have
9922 * relocation miss the reserved extent, making it fail later.
9924 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
9925 if (IS_ERR(trans)) {
9926 ret = PTR_ERR(trans);
9927 btrfs_free_reserved_extent(fs_info, ins.objectid,
9932 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
9933 cur_offset + ins.offset -1, 0);
9935 em = alloc_extent_map();
9937 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9938 &BTRFS_I(inode)->runtime_flags);
9942 em->start = cur_offset;
9943 em->orig_start = cur_offset;
9944 em->len = ins.offset;
9945 em->block_start = ins.objectid;
9946 em->block_len = ins.offset;
9947 em->orig_block_len = ins.offset;
9948 em->ram_bytes = ins.offset;
9949 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9950 em->generation = trans->transid;
9953 write_lock(&em_tree->lock);
9954 ret = add_extent_mapping(em_tree, em, 1);
9955 write_unlock(&em_tree->lock);
9958 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
9959 cur_offset + ins.offset - 1,
9962 free_extent_map(em);
9964 num_bytes -= ins.offset;
9965 cur_offset += ins.offset;
9966 *alloc_hint = ins.objectid + ins.offset;
9968 inode_inc_iversion(inode);
9969 inode->i_ctime = current_time(inode);
9970 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9971 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9972 (actual_len > inode->i_size) &&
9973 (cur_offset > inode->i_size)) {
9974 if (cur_offset > actual_len)
9975 i_size = actual_len;
9977 i_size = cur_offset;
9978 i_size_write(inode, i_size);
9979 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
9982 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
9985 btrfs_abort_transaction(trans, ret);
9987 btrfs_end_transaction(trans);
9992 btrfs_end_transaction(trans);
9996 if (clear_offset < end)
9997 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
9998 end - clear_offset + 1);
10002 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10003 u64 start, u64 num_bytes, u64 min_size,
10004 loff_t actual_len, u64 *alloc_hint)
10006 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10007 min_size, actual_len, alloc_hint,
10011 int btrfs_prealloc_file_range_trans(struct inode *inode,
10012 struct btrfs_trans_handle *trans, int mode,
10013 u64 start, u64 num_bytes, u64 min_size,
10014 loff_t actual_len, u64 *alloc_hint)
10016 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10017 min_size, actual_len, alloc_hint, trans);
10020 static int btrfs_set_page_dirty(struct page *page)
10022 return __set_page_dirty_nobuffers(page);
10025 static int btrfs_permission(struct inode *inode, int mask)
10027 struct btrfs_root *root = BTRFS_I(inode)->root;
10028 umode_t mode = inode->i_mode;
10030 if (mask & MAY_WRITE &&
10031 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10032 if (btrfs_root_readonly(root))
10034 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10037 return generic_permission(inode, mask);
10040 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10042 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10043 struct btrfs_trans_handle *trans;
10044 struct btrfs_root *root = BTRFS_I(dir)->root;
10045 struct inode *inode = NULL;
10051 * 5 units required for adding orphan entry
10053 trans = btrfs_start_transaction(root, 5);
10055 return PTR_ERR(trans);
10057 ret = btrfs_get_free_objectid(root, &objectid);
10061 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10062 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10063 if (IS_ERR(inode)) {
10064 ret = PTR_ERR(inode);
10069 inode->i_fop = &btrfs_file_operations;
10070 inode->i_op = &btrfs_file_inode_operations;
10072 inode->i_mapping->a_ops = &btrfs_aops;
10074 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10078 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
10081 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10086 * We set number of links to 0 in btrfs_new_inode(), and here we set
10087 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10090 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10092 set_nlink(inode, 1);
10093 d_tmpfile(dentry, inode);
10094 unlock_new_inode(inode);
10095 mark_inode_dirty(inode);
10097 btrfs_end_transaction(trans);
10099 discard_new_inode(inode);
10100 btrfs_btree_balance_dirty(fs_info);
10104 void btrfs_set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
10106 struct inode *inode = tree->private_data;
10107 unsigned long index = start >> PAGE_SHIFT;
10108 unsigned long end_index = end >> PAGE_SHIFT;
10111 while (index <= end_index) {
10112 page = find_get_page(inode->i_mapping, index);
10113 ASSERT(page); /* Pages should be in the extent_io_tree */
10114 set_page_writeback(page);
10122 * Add an entry indicating a block group or device which is pinned by a
10123 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10124 * negative errno on failure.
10126 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10127 bool is_block_group)
10129 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10130 struct btrfs_swapfile_pin *sp, *entry;
10131 struct rb_node **p;
10132 struct rb_node *parent = NULL;
10134 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10139 sp->is_block_group = is_block_group;
10141 spin_lock(&fs_info->swapfile_pins_lock);
10142 p = &fs_info->swapfile_pins.rb_node;
10145 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10146 if (sp->ptr < entry->ptr ||
10147 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10148 p = &(*p)->rb_left;
10149 } else if (sp->ptr > entry->ptr ||
10150 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10151 p = &(*p)->rb_right;
10153 spin_unlock(&fs_info->swapfile_pins_lock);
10158 rb_link_node(&sp->node, parent, p);
10159 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10160 spin_unlock(&fs_info->swapfile_pins_lock);
10164 /* Free all of the entries pinned by this swapfile. */
10165 static void btrfs_free_swapfile_pins(struct inode *inode)
10167 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10168 struct btrfs_swapfile_pin *sp;
10169 struct rb_node *node, *next;
10171 spin_lock(&fs_info->swapfile_pins_lock);
10172 node = rb_first(&fs_info->swapfile_pins);
10174 next = rb_next(node);
10175 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10176 if (sp->inode == inode) {
10177 rb_erase(&sp->node, &fs_info->swapfile_pins);
10178 if (sp->is_block_group)
10179 btrfs_put_block_group(sp->ptr);
10184 spin_unlock(&fs_info->swapfile_pins_lock);
10187 struct btrfs_swap_info {
10193 unsigned long nr_pages;
10197 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10198 struct btrfs_swap_info *bsi)
10200 unsigned long nr_pages;
10201 u64 first_ppage, first_ppage_reported, next_ppage;
10204 first_ppage = ALIGN(bsi->block_start, PAGE_SIZE) >> PAGE_SHIFT;
10205 next_ppage = ALIGN_DOWN(bsi->block_start + bsi->block_len,
10206 PAGE_SIZE) >> PAGE_SHIFT;
10208 if (first_ppage >= next_ppage)
10210 nr_pages = next_ppage - first_ppage;
10212 first_ppage_reported = first_ppage;
10213 if (bsi->start == 0)
10214 first_ppage_reported++;
10215 if (bsi->lowest_ppage > first_ppage_reported)
10216 bsi->lowest_ppage = first_ppage_reported;
10217 if (bsi->highest_ppage < (next_ppage - 1))
10218 bsi->highest_ppage = next_ppage - 1;
10220 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10223 bsi->nr_extents += ret;
10224 bsi->nr_pages += nr_pages;
10228 static void btrfs_swap_deactivate(struct file *file)
10230 struct inode *inode = file_inode(file);
10232 btrfs_free_swapfile_pins(inode);
10233 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10236 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10239 struct inode *inode = file_inode(file);
10240 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10241 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10242 struct extent_state *cached_state = NULL;
10243 struct extent_map *em = NULL;
10244 struct btrfs_device *device = NULL;
10245 struct btrfs_swap_info bsi = {
10246 .lowest_ppage = (sector_t)-1ULL,
10253 * If the swap file was just created, make sure delalloc is done. If the
10254 * file changes again after this, the user is doing something stupid and
10255 * we don't really care.
10257 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10262 * The inode is locked, so these flags won't change after we check them.
10264 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10265 btrfs_warn(fs_info, "swapfile must not be compressed");
10268 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10269 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10272 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10273 btrfs_warn(fs_info, "swapfile must not be checksummed");
10278 * Balance or device remove/replace/resize can move stuff around from
10279 * under us. The exclop protection makes sure they aren't running/won't
10280 * run concurrently while we are mapping the swap extents, and
10281 * fs_info->swapfile_pins prevents them from running while the swap
10282 * file is active and moving the extents. Note that this also prevents
10283 * a concurrent device add which isn't actually necessary, but it's not
10284 * really worth the trouble to allow it.
10286 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10287 btrfs_warn(fs_info,
10288 "cannot activate swapfile while exclusive operation is running");
10292 * Snapshots can create extents which require COW even if NODATACOW is
10293 * set. We use this counter to prevent snapshots. We must increment it
10294 * before walking the extents because we don't want a concurrent
10295 * snapshot to run after we've already checked the extents.
10297 atomic_inc(&BTRFS_I(inode)->root->nr_swapfiles);
10299 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10301 lock_extent_bits(io_tree, 0, isize - 1, &cached_state);
10303 while (start < isize) {
10304 u64 logical_block_start, physical_block_start;
10305 struct btrfs_block_group *bg;
10306 u64 len = isize - start;
10308 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
10314 if (em->block_start == EXTENT_MAP_HOLE) {
10315 btrfs_warn(fs_info, "swapfile must not have holes");
10319 if (em->block_start == EXTENT_MAP_INLINE) {
10321 * It's unlikely we'll ever actually find ourselves
10322 * here, as a file small enough to fit inline won't be
10323 * big enough to store more than the swap header, but in
10324 * case something changes in the future, let's catch it
10325 * here rather than later.
10327 btrfs_warn(fs_info, "swapfile must not be inline");
10331 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10332 btrfs_warn(fs_info, "swapfile must not be compressed");
10337 logical_block_start = em->block_start + (start - em->start);
10338 len = min(len, em->len - (start - em->start));
10339 free_extent_map(em);
10342 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, true);
10348 btrfs_warn(fs_info,
10349 "swapfile must not be copy-on-write");
10354 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10360 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10361 btrfs_warn(fs_info,
10362 "swapfile must have single data profile");
10367 if (device == NULL) {
10368 device = em->map_lookup->stripes[0].dev;
10369 ret = btrfs_add_swapfile_pin(inode, device, false);
10374 } else if (device != em->map_lookup->stripes[0].dev) {
10375 btrfs_warn(fs_info, "swapfile must be on one device");
10380 physical_block_start = (em->map_lookup->stripes[0].physical +
10381 (logical_block_start - em->start));
10382 len = min(len, em->len - (logical_block_start - em->start));
10383 free_extent_map(em);
10386 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10388 btrfs_warn(fs_info,
10389 "could not find block group containing swapfile");
10394 ret = btrfs_add_swapfile_pin(inode, bg, true);
10396 btrfs_put_block_group(bg);
10403 if (bsi.block_len &&
10404 bsi.block_start + bsi.block_len == physical_block_start) {
10405 bsi.block_len += len;
10407 if (bsi.block_len) {
10408 ret = btrfs_add_swap_extent(sis, &bsi);
10413 bsi.block_start = physical_block_start;
10414 bsi.block_len = len;
10421 ret = btrfs_add_swap_extent(sis, &bsi);
10424 if (!IS_ERR_OR_NULL(em))
10425 free_extent_map(em);
10427 unlock_extent_cached(io_tree, 0, isize - 1, &cached_state);
10430 btrfs_swap_deactivate(file);
10432 btrfs_exclop_finish(fs_info);
10438 sis->bdev = device->bdev;
10439 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10440 sis->max = bsi.nr_pages;
10441 sis->pages = bsi.nr_pages - 1;
10442 sis->highest_bit = bsi.nr_pages - 1;
10443 return bsi.nr_extents;
10446 static void btrfs_swap_deactivate(struct file *file)
10450 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10453 return -EOPNOTSUPP;
10458 * Update the number of bytes used in the VFS' inode. When we replace extents in
10459 * a range (clone, dedupe, fallocate's zero range), we must update the number of
10460 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
10461 * always get a correct value.
10463 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
10464 const u64 add_bytes,
10465 const u64 del_bytes)
10467 if (add_bytes == del_bytes)
10470 spin_lock(&inode->lock);
10472 inode_sub_bytes(&inode->vfs_inode, del_bytes);
10474 inode_add_bytes(&inode->vfs_inode, add_bytes);
10475 spin_unlock(&inode->lock);
10478 static const struct inode_operations btrfs_dir_inode_operations = {
10479 .getattr = btrfs_getattr,
10480 .lookup = btrfs_lookup,
10481 .create = btrfs_create,
10482 .unlink = btrfs_unlink,
10483 .link = btrfs_link,
10484 .mkdir = btrfs_mkdir,
10485 .rmdir = btrfs_rmdir,
10486 .rename = btrfs_rename2,
10487 .symlink = btrfs_symlink,
10488 .setattr = btrfs_setattr,
10489 .mknod = btrfs_mknod,
10490 .listxattr = btrfs_listxattr,
10491 .permission = btrfs_permission,
10492 .get_acl = btrfs_get_acl,
10493 .set_acl = btrfs_set_acl,
10494 .update_time = btrfs_update_time,
10495 .tmpfile = btrfs_tmpfile,
10498 static const struct file_operations btrfs_dir_file_operations = {
10499 .llseek = generic_file_llseek,
10500 .read = generic_read_dir,
10501 .iterate_shared = btrfs_real_readdir,
10502 .open = btrfs_opendir,
10503 .unlocked_ioctl = btrfs_ioctl,
10504 #ifdef CONFIG_COMPAT
10505 .compat_ioctl = btrfs_compat_ioctl,
10507 .release = btrfs_release_file,
10508 .fsync = btrfs_sync_file,
10512 * btrfs doesn't support the bmap operation because swapfiles
10513 * use bmap to make a mapping of extents in the file. They assume
10514 * these extents won't change over the life of the file and they
10515 * use the bmap result to do IO directly to the drive.
10517 * the btrfs bmap call would return logical addresses that aren't
10518 * suitable for IO and they also will change frequently as COW
10519 * operations happen. So, swapfile + btrfs == corruption.
10521 * For now we're avoiding this by dropping bmap.
10523 static const struct address_space_operations btrfs_aops = {
10524 .readpage = btrfs_readpage,
10525 .writepage = btrfs_writepage,
10526 .writepages = btrfs_writepages,
10527 .readahead = btrfs_readahead,
10528 .direct_IO = noop_direct_IO,
10529 .invalidatepage = btrfs_invalidatepage,
10530 .releasepage = btrfs_releasepage,
10531 #ifdef CONFIG_MIGRATION
10532 .migratepage = btrfs_migratepage,
10534 .set_page_dirty = btrfs_set_page_dirty,
10535 .error_remove_page = generic_error_remove_page,
10536 .swap_activate = btrfs_swap_activate,
10537 .swap_deactivate = btrfs_swap_deactivate,
10540 static const struct inode_operations btrfs_file_inode_operations = {
10541 .getattr = btrfs_getattr,
10542 .setattr = btrfs_setattr,
10543 .listxattr = btrfs_listxattr,
10544 .permission = btrfs_permission,
10545 .fiemap = btrfs_fiemap,
10546 .get_acl = btrfs_get_acl,
10547 .set_acl = btrfs_set_acl,
10548 .update_time = btrfs_update_time,
10550 static const struct inode_operations btrfs_special_inode_operations = {
10551 .getattr = btrfs_getattr,
10552 .setattr = btrfs_setattr,
10553 .permission = btrfs_permission,
10554 .listxattr = btrfs_listxattr,
10555 .get_acl = btrfs_get_acl,
10556 .set_acl = btrfs_set_acl,
10557 .update_time = btrfs_update_time,
10559 static const struct inode_operations btrfs_symlink_inode_operations = {
10560 .get_link = page_get_link,
10561 .getattr = btrfs_getattr,
10562 .setattr = btrfs_setattr,
10563 .permission = btrfs_permission,
10564 .listxattr = btrfs_listxattr,
10565 .update_time = btrfs_update_time,
10568 const struct dentry_operations btrfs_dentry_operations = {
10569 .d_delete = btrfs_dentry_delete,