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"
48 #include "inode-map.h"
51 #include "delalloc-space.h"
52 #include "block-group.h"
53 #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;
68 static const struct inode_operations btrfs_dir_inode_operations;
69 static const struct inode_operations btrfs_symlink_inode_operations;
70 static const struct inode_operations btrfs_special_inode_operations;
71 static const struct inode_operations btrfs_file_inode_operations;
72 static const struct address_space_operations btrfs_aops;
73 static const struct file_operations btrfs_dir_file_operations;
75 static struct kmem_cache *btrfs_inode_cachep;
76 struct kmem_cache *btrfs_trans_handle_cachep;
77 struct kmem_cache *btrfs_path_cachep;
78 struct kmem_cache *btrfs_free_space_cachep;
79 struct kmem_cache *btrfs_free_space_bitmap_cachep;
81 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
82 static int btrfs_truncate(struct inode *inode, bool skip_writeback);
83 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
84 static noinline int cow_file_range(struct btrfs_inode *inode,
85 struct page *locked_page,
86 u64 start, u64 end, int *page_started,
87 unsigned long *nr_written, int unlock);
88 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
89 u64 len, u64 orig_start, u64 block_start,
90 u64 block_len, u64 orig_block_len,
91 u64 ram_bytes, int compress_type,
94 static void __endio_write_update_ordered(struct btrfs_inode *inode,
95 const u64 offset, const u64 bytes,
99 * Cleanup all submitted ordered extents in specified range to handle errors
100 * from the btrfs_run_delalloc_range() callback.
102 * NOTE: caller must ensure that when an error happens, it can not call
103 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
104 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
105 * to be released, which we want to happen only when finishing the ordered
106 * extent (btrfs_finish_ordered_io()).
108 static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode,
109 struct page *locked_page,
110 u64 offset, u64 bytes)
112 unsigned long index = offset >> PAGE_SHIFT;
113 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
114 u64 page_start = page_offset(locked_page);
115 u64 page_end = page_start + PAGE_SIZE - 1;
119 while (index <= end_index) {
120 page = find_get_page(inode->vfs_inode.i_mapping, index);
124 ClearPagePrivate2(page);
129 * In case this page belongs to the delalloc range being instantiated
130 * then skip it, since the first page of a range is going to be
131 * properly cleaned up by the caller of run_delalloc_range
133 if (page_start >= offset && page_end <= (offset + bytes - 1)) {
138 return __endio_write_update_ordered(inode, offset, bytes, false);
141 static int btrfs_dirty_inode(struct inode *inode);
143 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
144 struct inode *inode, struct inode *dir,
145 const struct qstr *qstr)
149 err = btrfs_init_acl(trans, inode, dir);
151 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
156 * this does all the hard work for inserting an inline extent into
157 * the btree. The caller should have done a btrfs_drop_extents so that
158 * no overlapping inline items exist in the btree
160 static int insert_inline_extent(struct btrfs_trans_handle *trans,
161 struct btrfs_path *path, int extent_inserted,
162 struct btrfs_root *root, struct inode *inode,
163 u64 start, size_t size, size_t compressed_size,
165 struct page **compressed_pages)
167 struct extent_buffer *leaf;
168 struct page *page = NULL;
171 struct btrfs_file_extent_item *ei;
173 size_t cur_size = size;
174 unsigned long offset;
176 ASSERT((compressed_size > 0 && compressed_pages) ||
177 (compressed_size == 0 && !compressed_pages));
179 if (compressed_size && compressed_pages)
180 cur_size = compressed_size;
182 inode_add_bytes(inode, size);
184 if (!extent_inserted) {
185 struct btrfs_key key;
188 key.objectid = btrfs_ino(BTRFS_I(inode));
190 key.type = BTRFS_EXTENT_DATA_KEY;
192 datasize = btrfs_file_extent_calc_inline_size(cur_size);
193 path->leave_spinning = 1;
194 ret = btrfs_insert_empty_item(trans, root, path, &key,
199 leaf = path->nodes[0];
200 ei = btrfs_item_ptr(leaf, path->slots[0],
201 struct btrfs_file_extent_item);
202 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
203 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
204 btrfs_set_file_extent_encryption(leaf, ei, 0);
205 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
206 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
207 ptr = btrfs_file_extent_inline_start(ei);
209 if (compress_type != BTRFS_COMPRESS_NONE) {
212 while (compressed_size > 0) {
213 cpage = compressed_pages[i];
214 cur_size = min_t(unsigned long, compressed_size,
217 kaddr = kmap_atomic(cpage);
218 write_extent_buffer(leaf, kaddr, ptr, cur_size);
219 kunmap_atomic(kaddr);
223 compressed_size -= cur_size;
225 btrfs_set_file_extent_compression(leaf, ei,
228 page = find_get_page(inode->i_mapping,
229 start >> PAGE_SHIFT);
230 btrfs_set_file_extent_compression(leaf, ei, 0);
231 kaddr = kmap_atomic(page);
232 offset = offset_in_page(start);
233 write_extent_buffer(leaf, kaddr + offset, ptr, size);
234 kunmap_atomic(kaddr);
237 btrfs_mark_buffer_dirty(leaf);
238 btrfs_release_path(path);
241 * We align size to sectorsize for inline extents just for simplicity
244 size = ALIGN(size, root->fs_info->sectorsize);
245 ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start, size);
250 * we're an inline extent, so nobody can
251 * extend the file past i_size without locking
252 * a page we already have locked.
254 * We must do any isize and inode updates
255 * before we unlock the pages. Otherwise we
256 * could end up racing with unlink.
258 BTRFS_I(inode)->disk_i_size = inode->i_size;
259 ret = btrfs_update_inode(trans, root, inode);
267 * conditionally insert an inline extent into the file. This
268 * does the checks required to make sure the data is small enough
269 * to fit as an inline extent.
271 static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 start,
272 u64 end, size_t compressed_size,
274 struct page **compressed_pages)
276 struct btrfs_root *root = inode->root;
277 struct btrfs_fs_info *fs_info = root->fs_info;
278 struct btrfs_trans_handle *trans;
279 u64 isize = i_size_read(&inode->vfs_inode);
280 u64 actual_end = min(end + 1, isize);
281 u64 inline_len = actual_end - start;
282 u64 aligned_end = ALIGN(end, fs_info->sectorsize);
283 u64 data_len = inline_len;
285 struct btrfs_path *path;
286 int extent_inserted = 0;
287 u32 extent_item_size;
290 data_len = compressed_size;
293 actual_end > fs_info->sectorsize ||
294 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
296 (actual_end & (fs_info->sectorsize - 1)) == 0) ||
298 data_len > fs_info->max_inline) {
302 path = btrfs_alloc_path();
306 trans = btrfs_join_transaction(root);
308 btrfs_free_path(path);
309 return PTR_ERR(trans);
311 trans->block_rsv = &inode->block_rsv;
313 if (compressed_size && compressed_pages)
314 extent_item_size = btrfs_file_extent_calc_inline_size(
317 extent_item_size = btrfs_file_extent_calc_inline_size(
320 ret = __btrfs_drop_extents(trans, root, inode, path, start, aligned_end,
321 NULL, 1, 1, extent_item_size,
324 btrfs_abort_transaction(trans, ret);
328 if (isize > actual_end)
329 inline_len = min_t(u64, isize, actual_end);
330 ret = insert_inline_extent(trans, path, extent_inserted,
331 root, &inode->vfs_inode, start,
332 inline_len, compressed_size,
333 compress_type, compressed_pages);
334 if (ret && ret != -ENOSPC) {
335 btrfs_abort_transaction(trans, ret);
337 } else if (ret == -ENOSPC) {
342 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
343 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
346 * Don't forget to free the reserved space, as for inlined extent
347 * it won't count as data extent, free them directly here.
348 * And at reserve time, it's always aligned to page size, so
349 * just free one page here.
351 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
352 btrfs_free_path(path);
353 btrfs_end_transaction(trans);
357 struct async_extent {
362 unsigned long nr_pages;
364 struct list_head list;
369 struct page *locked_page;
372 unsigned int write_flags;
373 struct list_head extents;
374 struct cgroup_subsys_state *blkcg_css;
375 struct btrfs_work work;
380 /* Number of chunks in flight; must be first in the structure */
382 struct async_chunk chunks[];
385 static noinline int add_async_extent(struct async_chunk *cow,
386 u64 start, u64 ram_size,
389 unsigned long nr_pages,
392 struct async_extent *async_extent;
394 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
395 BUG_ON(!async_extent); /* -ENOMEM */
396 async_extent->start = start;
397 async_extent->ram_size = ram_size;
398 async_extent->compressed_size = compressed_size;
399 async_extent->pages = pages;
400 async_extent->nr_pages = nr_pages;
401 async_extent->compress_type = compress_type;
402 list_add_tail(&async_extent->list, &cow->extents);
407 * Check if the inode has flags compatible with compression
409 static inline bool inode_can_compress(struct btrfs_inode *inode)
411 if (inode->flags & BTRFS_INODE_NODATACOW ||
412 inode->flags & BTRFS_INODE_NODATASUM)
418 * Check if the inode needs to be submitted to compression, based on mount
419 * options, defragmentation, properties or heuristics.
421 static inline int inode_need_compress(struct btrfs_inode *inode, u64 start,
424 struct btrfs_fs_info *fs_info = inode->root->fs_info;
426 if (!inode_can_compress(inode)) {
427 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
428 KERN_ERR "BTRFS: unexpected compression for ino %llu\n",
433 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
436 if (inode->defrag_compress)
438 /* bad compression ratios */
439 if (inode->flags & BTRFS_INODE_NOCOMPRESS)
441 if (btrfs_test_opt(fs_info, COMPRESS) ||
442 inode->flags & BTRFS_INODE_COMPRESS ||
443 inode->prop_compress)
444 return btrfs_compress_heuristic(&inode->vfs_inode, start, end);
448 static inline void inode_should_defrag(struct btrfs_inode *inode,
449 u64 start, u64 end, u64 num_bytes, u64 small_write)
451 /* If this is a small write inside eof, kick off a defrag */
452 if (num_bytes < small_write &&
453 (start > 0 || end + 1 < inode->disk_i_size))
454 btrfs_add_inode_defrag(NULL, inode);
458 * we create compressed extents in two phases. The first
459 * phase compresses a range of pages that have already been
460 * locked (both pages and state bits are locked).
462 * This is done inside an ordered work queue, and the compression
463 * is spread across many cpus. The actual IO submission is step
464 * two, and the ordered work queue takes care of making sure that
465 * happens in the same order things were put onto the queue by
466 * writepages and friends.
468 * If this code finds it can't get good compression, it puts an
469 * entry onto the work queue to write the uncompressed bytes. This
470 * makes sure that both compressed inodes and uncompressed inodes
471 * are written in the same order that the flusher thread sent them
474 static noinline int compress_file_range(struct async_chunk *async_chunk)
476 struct inode *inode = async_chunk->inode;
477 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
478 u64 blocksize = fs_info->sectorsize;
479 u64 start = async_chunk->start;
480 u64 end = async_chunk->end;
484 struct page **pages = NULL;
485 unsigned long nr_pages;
486 unsigned long total_compressed = 0;
487 unsigned long total_in = 0;
490 int compress_type = fs_info->compress_type;
491 int compressed_extents = 0;
494 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
498 * We need to save i_size before now because it could change in between
499 * us evaluating the size and assigning it. This is because we lock and
500 * unlock the page in truncate and fallocate, and then modify the i_size
503 * The barriers are to emulate READ_ONCE, remove that once i_size_read
507 i_size = i_size_read(inode);
509 actual_end = min_t(u64, i_size, end + 1);
512 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
513 BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
514 nr_pages = min_t(unsigned long, nr_pages,
515 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
518 * we don't want to send crud past the end of i_size through
519 * compression, that's just a waste of CPU time. So, if the
520 * end of the file is before the start of our current
521 * requested range of bytes, we bail out to the uncompressed
522 * cleanup code that can deal with all of this.
524 * It isn't really the fastest way to fix things, but this is a
525 * very uncommon corner.
527 if (actual_end <= start)
528 goto cleanup_and_bail_uncompressed;
530 total_compressed = actual_end - start;
533 * skip compression for a small file range(<=blocksize) that
534 * isn't an inline extent, since it doesn't save disk space at all.
536 if (total_compressed <= blocksize &&
537 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
538 goto cleanup_and_bail_uncompressed;
540 total_compressed = min_t(unsigned long, total_compressed,
541 BTRFS_MAX_UNCOMPRESSED);
546 * we do compression for mount -o compress and when the
547 * inode has not been flagged as nocompress. This flag can
548 * change at any time if we discover bad compression ratios.
550 if (inode_need_compress(BTRFS_I(inode), start, end)) {
552 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
554 /* just bail out to the uncompressed code */
559 if (BTRFS_I(inode)->defrag_compress)
560 compress_type = BTRFS_I(inode)->defrag_compress;
561 else if (BTRFS_I(inode)->prop_compress)
562 compress_type = BTRFS_I(inode)->prop_compress;
565 * we need to call clear_page_dirty_for_io on each
566 * page in the range. Otherwise applications with the file
567 * mmap'd can wander in and change the page contents while
568 * we are compressing them.
570 * If the compression fails for any reason, we set the pages
571 * dirty again later on.
573 * Note that the remaining part is redirtied, the start pointer
574 * has moved, the end is the original one.
577 extent_range_clear_dirty_for_io(inode, start, end);
581 /* Compression level is applied here and only here */
582 ret = btrfs_compress_pages(
583 compress_type | (fs_info->compress_level << 4),
584 inode->i_mapping, start,
591 unsigned long offset = offset_in_page(total_compressed);
592 struct page *page = pages[nr_pages - 1];
595 /* zero the tail end of the last page, we might be
596 * sending it down to disk
599 kaddr = kmap_atomic(page);
600 memset(kaddr + offset, 0,
602 kunmap_atomic(kaddr);
609 /* lets try to make an inline extent */
610 if (ret || total_in < actual_end) {
611 /* we didn't compress the entire range, try
612 * to make an uncompressed inline extent.
614 ret = cow_file_range_inline(BTRFS_I(inode), start, end,
615 0, BTRFS_COMPRESS_NONE,
618 /* try making a compressed inline extent */
619 ret = cow_file_range_inline(BTRFS_I(inode), start, end,
621 compress_type, pages);
624 unsigned long clear_flags = EXTENT_DELALLOC |
625 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
626 EXTENT_DO_ACCOUNTING;
627 unsigned long page_error_op;
629 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
632 * inline extent creation worked or returned error,
633 * we don't need to create any more async work items.
634 * Unlock and free up our temp pages.
636 * We use DO_ACCOUNTING here because we need the
637 * delalloc_release_metadata to be done _after_ we drop
638 * our outstanding extent for clearing delalloc for this
641 extent_clear_unlock_delalloc(BTRFS_I(inode), start, end,
651 * Ensure we only free the compressed pages if we have
652 * them allocated, as we can still reach here with
653 * inode_need_compress() == false.
656 for (i = 0; i < nr_pages; i++) {
657 WARN_ON(pages[i]->mapping);
668 * we aren't doing an inline extent round the compressed size
669 * up to a block size boundary so the allocator does sane
672 total_compressed = ALIGN(total_compressed, blocksize);
675 * one last check to make sure the compression is really a
676 * win, compare the page count read with the blocks on disk,
677 * compression must free at least one sector size
679 total_in = ALIGN(total_in, PAGE_SIZE);
680 if (total_compressed + blocksize <= total_in) {
681 compressed_extents++;
684 * The async work queues will take care of doing actual
685 * allocation on disk for these compressed pages, and
686 * will submit them to the elevator.
688 add_async_extent(async_chunk, start, total_in,
689 total_compressed, pages, nr_pages,
692 if (start + total_in < end) {
698 return compressed_extents;
703 * the compression code ran but failed to make things smaller,
704 * free any pages it allocated and our page pointer array
706 for (i = 0; i < nr_pages; i++) {
707 WARN_ON(pages[i]->mapping);
712 total_compressed = 0;
715 /* flag the file so we don't compress in the future */
716 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
717 !(BTRFS_I(inode)->prop_compress)) {
718 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
721 cleanup_and_bail_uncompressed:
723 * No compression, but we still need to write the pages in the file
724 * we've been given so far. redirty the locked page if it corresponds
725 * to our extent and set things up for the async work queue to run
726 * cow_file_range to do the normal delalloc dance.
728 if (async_chunk->locked_page &&
729 (page_offset(async_chunk->locked_page) >= start &&
730 page_offset(async_chunk->locked_page)) <= end) {
731 __set_page_dirty_nobuffers(async_chunk->locked_page);
732 /* unlocked later on in the async handlers */
736 extent_range_redirty_for_io(inode, start, end);
737 add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
738 BTRFS_COMPRESS_NONE);
739 compressed_extents++;
741 return compressed_extents;
744 static void free_async_extent_pages(struct async_extent *async_extent)
748 if (!async_extent->pages)
751 for (i = 0; i < async_extent->nr_pages; i++) {
752 WARN_ON(async_extent->pages[i]->mapping);
753 put_page(async_extent->pages[i]);
755 kfree(async_extent->pages);
756 async_extent->nr_pages = 0;
757 async_extent->pages = NULL;
761 * phase two of compressed writeback. This is the ordered portion
762 * of the code, which only gets called in the order the work was
763 * queued. We walk all the async extents created by compress_file_range
764 * and send them down to the disk.
766 static noinline void submit_compressed_extents(struct async_chunk *async_chunk)
768 struct btrfs_inode *inode = BTRFS_I(async_chunk->inode);
769 struct btrfs_fs_info *fs_info = inode->root->fs_info;
770 struct async_extent *async_extent;
772 struct btrfs_key ins;
773 struct extent_map *em;
774 struct btrfs_root *root = inode->root;
775 struct extent_io_tree *io_tree = &inode->io_tree;
779 while (!list_empty(&async_chunk->extents)) {
780 async_extent = list_entry(async_chunk->extents.next,
781 struct async_extent, list);
782 list_del(&async_extent->list);
785 lock_extent(io_tree, async_extent->start,
786 async_extent->start + async_extent->ram_size - 1);
787 /* did the compression code fall back to uncompressed IO? */
788 if (!async_extent->pages) {
789 int page_started = 0;
790 unsigned long nr_written = 0;
792 /* allocate blocks */
793 ret = cow_file_range(inode, async_chunk->locked_page,
795 async_extent->start +
796 async_extent->ram_size - 1,
797 &page_started, &nr_written, 0);
802 * if page_started, cow_file_range inserted an
803 * inline extent and took care of all the unlocking
804 * and IO for us. Otherwise, we need to submit
805 * all those pages down to the drive.
807 if (!page_started && !ret)
808 extent_write_locked_range(&inode->vfs_inode,
810 async_extent->start +
811 async_extent->ram_size - 1,
813 else if (ret && async_chunk->locked_page)
814 unlock_page(async_chunk->locked_page);
820 ret = btrfs_reserve_extent(root, async_extent->ram_size,
821 async_extent->compressed_size,
822 async_extent->compressed_size,
823 0, alloc_hint, &ins, 1, 1);
825 free_async_extent_pages(async_extent);
827 if (ret == -ENOSPC) {
828 unlock_extent(io_tree, async_extent->start,
829 async_extent->start +
830 async_extent->ram_size - 1);
833 * we need to redirty the pages if we decide to
834 * fallback to uncompressed IO, otherwise we
835 * will not submit these pages down to lower
838 extent_range_redirty_for_io(&inode->vfs_inode,
840 async_extent->start +
841 async_extent->ram_size - 1);
848 * here we're doing allocation and writeback of the
851 em = create_io_em(inode, async_extent->start,
852 async_extent->ram_size, /* len */
853 async_extent->start, /* orig_start */
854 ins.objectid, /* block_start */
855 ins.offset, /* block_len */
856 ins.offset, /* orig_block_len */
857 async_extent->ram_size, /* ram_bytes */
858 async_extent->compress_type,
859 BTRFS_ORDERED_COMPRESSED);
861 /* ret value is not necessary due to void function */
862 goto out_free_reserve;
865 ret = btrfs_add_ordered_extent_compress(inode,
868 async_extent->ram_size,
870 BTRFS_ORDERED_COMPRESSED,
871 async_extent->compress_type);
873 btrfs_drop_extent_cache(inode, async_extent->start,
874 async_extent->start +
875 async_extent->ram_size - 1, 0);
876 goto out_free_reserve;
878 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
881 * clear dirty, set writeback and unlock the pages.
883 extent_clear_unlock_delalloc(inode, async_extent->start,
884 async_extent->start +
885 async_extent->ram_size - 1,
886 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
887 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
889 if (btrfs_submit_compressed_write(inode, async_extent->start,
890 async_extent->ram_size,
892 ins.offset, async_extent->pages,
893 async_extent->nr_pages,
894 async_chunk->write_flags,
895 async_chunk->blkcg_css)) {
896 struct page *p = async_extent->pages[0];
897 const u64 start = async_extent->start;
898 const u64 end = start + async_extent->ram_size - 1;
900 p->mapping = inode->vfs_inode.i_mapping;
901 btrfs_writepage_endio_finish_ordered(p, start, end, 0);
904 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
907 free_async_extent_pages(async_extent);
909 alloc_hint = ins.objectid + ins.offset;
915 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
916 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
918 extent_clear_unlock_delalloc(inode, async_extent->start,
919 async_extent->start +
920 async_extent->ram_size - 1,
921 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
922 EXTENT_DELALLOC_NEW |
923 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
924 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
925 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
927 free_async_extent_pages(async_extent);
932 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
935 struct extent_map_tree *em_tree = &inode->extent_tree;
936 struct extent_map *em;
939 read_lock(&em_tree->lock);
940 em = search_extent_mapping(em_tree, start, num_bytes);
943 * if block start isn't an actual block number then find the
944 * first block in this inode and use that as a hint. If that
945 * block is also bogus then just don't worry about it.
947 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
949 em = search_extent_mapping(em_tree, 0, 0);
950 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
951 alloc_hint = em->block_start;
955 alloc_hint = em->block_start;
959 read_unlock(&em_tree->lock);
965 * when extent_io.c finds a delayed allocation range in the file,
966 * the call backs end up in this code. The basic idea is to
967 * allocate extents on disk for the range, and create ordered data structs
968 * in ram to track those extents.
970 * locked_page is the page that writepage had locked already. We use
971 * it to make sure we don't do extra locks or unlocks.
973 * *page_started is set to one if we unlock locked_page and do everything
974 * required to start IO on it. It may be clean and already done with
977 static noinline int cow_file_range(struct btrfs_inode *inode,
978 struct page *locked_page,
979 u64 start, u64 end, int *page_started,
980 unsigned long *nr_written, int unlock)
982 struct btrfs_root *root = inode->root;
983 struct btrfs_fs_info *fs_info = root->fs_info;
986 unsigned long ram_size;
987 u64 cur_alloc_size = 0;
989 u64 blocksize = fs_info->sectorsize;
990 struct btrfs_key ins;
991 struct extent_map *em;
993 unsigned long page_ops;
994 bool extent_reserved = false;
997 if (btrfs_is_free_space_inode(inode)) {
1003 num_bytes = ALIGN(end - start + 1, blocksize);
1004 num_bytes = max(blocksize, num_bytes);
1005 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1007 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1010 /* lets try to make an inline extent */
1011 ret = cow_file_range_inline(inode, start, end, 0,
1012 BTRFS_COMPRESS_NONE, NULL);
1015 * We use DO_ACCOUNTING here because we need the
1016 * delalloc_release_metadata to be run _after_ we drop
1017 * our outstanding extent for clearing delalloc for this
1020 extent_clear_unlock_delalloc(inode, start, end, NULL,
1021 EXTENT_LOCKED | EXTENT_DELALLOC |
1022 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1023 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1024 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1025 PAGE_END_WRITEBACK);
1026 *nr_written = *nr_written +
1027 (end - start + PAGE_SIZE) / PAGE_SIZE;
1030 } else if (ret < 0) {
1035 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1036 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
1039 * Relocation relies on the relocated extents to have exactly the same
1040 * size as the original extents. Normally writeback for relocation data
1041 * extents follows a NOCOW path because relocation preallocates the
1042 * extents. However, due to an operation such as scrub turning a block
1043 * group to RO mode, it may fallback to COW mode, so we must make sure
1044 * an extent allocated during COW has exactly the requested size and can
1045 * not be split into smaller extents, otherwise relocation breaks and
1046 * fails during the stage where it updates the bytenr of file extent
1049 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1050 min_alloc_size = num_bytes;
1052 min_alloc_size = fs_info->sectorsize;
1054 while (num_bytes > 0) {
1055 cur_alloc_size = num_bytes;
1056 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1057 min_alloc_size, 0, alloc_hint,
1061 cur_alloc_size = ins.offset;
1062 extent_reserved = true;
1064 ram_size = ins.offset;
1065 em = create_io_em(inode, start, ins.offset, /* len */
1066 start, /* orig_start */
1067 ins.objectid, /* block_start */
1068 ins.offset, /* block_len */
1069 ins.offset, /* orig_block_len */
1070 ram_size, /* ram_bytes */
1071 BTRFS_COMPRESS_NONE, /* compress_type */
1072 BTRFS_ORDERED_REGULAR /* type */);
1077 free_extent_map(em);
1079 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1080 ram_size, cur_alloc_size, 0);
1082 goto out_drop_extent_cache;
1084 if (root->root_key.objectid ==
1085 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1086 ret = btrfs_reloc_clone_csums(inode, start,
1089 * Only drop cache here, and process as normal.
1091 * We must not allow extent_clear_unlock_delalloc()
1092 * at out_unlock label to free meta of this ordered
1093 * extent, as its meta should be freed by
1094 * btrfs_finish_ordered_io().
1096 * So we must continue until @start is increased to
1097 * skip current ordered extent.
1100 btrfs_drop_extent_cache(inode, start,
1101 start + ram_size - 1, 0);
1104 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1106 /* we're not doing compressed IO, don't unlock the first
1107 * page (which the caller expects to stay locked), don't
1108 * clear any dirty bits and don't set any writeback bits
1110 * Do set the Private2 bit so we know this page was properly
1111 * setup for writepage
1113 page_ops = unlock ? PAGE_UNLOCK : 0;
1114 page_ops |= PAGE_SET_PRIVATE2;
1116 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1118 EXTENT_LOCKED | EXTENT_DELALLOC,
1120 if (num_bytes < cur_alloc_size)
1123 num_bytes -= cur_alloc_size;
1124 alloc_hint = ins.objectid + ins.offset;
1125 start += cur_alloc_size;
1126 extent_reserved = false;
1129 * btrfs_reloc_clone_csums() error, since start is increased
1130 * extent_clear_unlock_delalloc() at out_unlock label won't
1131 * free metadata of current ordered extent, we're OK to exit.
1139 out_drop_extent_cache:
1140 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1142 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1143 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1145 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1146 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1147 page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1150 * If we reserved an extent for our delalloc range (or a subrange) and
1151 * failed to create the respective ordered extent, then it means that
1152 * when we reserved the extent we decremented the extent's size from
1153 * the data space_info's bytes_may_use counter and incremented the
1154 * space_info's bytes_reserved counter by the same amount. We must make
1155 * sure extent_clear_unlock_delalloc() does not try to decrement again
1156 * the data space_info's bytes_may_use counter, therefore we do not pass
1157 * it the flag EXTENT_CLEAR_DATA_RESV.
1159 if (extent_reserved) {
1160 extent_clear_unlock_delalloc(inode, start,
1161 start + cur_alloc_size - 1,
1165 start += cur_alloc_size;
1169 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1170 clear_bits | EXTENT_CLEAR_DATA_RESV,
1176 * work queue call back to started compression on a file and pages
1178 static noinline void async_cow_start(struct btrfs_work *work)
1180 struct async_chunk *async_chunk;
1181 int compressed_extents;
1183 async_chunk = container_of(work, struct async_chunk, work);
1185 compressed_extents = compress_file_range(async_chunk);
1186 if (compressed_extents == 0) {
1187 btrfs_add_delayed_iput(async_chunk->inode);
1188 async_chunk->inode = NULL;
1193 * work queue call back to submit previously compressed pages
1195 static noinline void async_cow_submit(struct btrfs_work *work)
1197 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1199 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1200 unsigned long nr_pages;
1202 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1205 /* atomic_sub_return implies a barrier */
1206 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1208 cond_wake_up_nomb(&fs_info->async_submit_wait);
1211 * ->inode could be NULL if async_chunk_start has failed to compress,
1212 * in which case we don't have anything to submit, yet we need to
1213 * always adjust ->async_delalloc_pages as its paired with the init
1214 * happening in cow_file_range_async
1216 if (async_chunk->inode)
1217 submit_compressed_extents(async_chunk);
1220 static noinline void async_cow_free(struct btrfs_work *work)
1222 struct async_chunk *async_chunk;
1224 async_chunk = container_of(work, struct async_chunk, work);
1225 if (async_chunk->inode)
1226 btrfs_add_delayed_iput(async_chunk->inode);
1227 if (async_chunk->blkcg_css)
1228 css_put(async_chunk->blkcg_css);
1230 * Since the pointer to 'pending' is at the beginning of the array of
1231 * async_chunk's, freeing it ensures the whole array has been freed.
1233 if (atomic_dec_and_test(async_chunk->pending))
1234 kvfree(async_chunk->pending);
1237 static int cow_file_range_async(struct btrfs_inode *inode,
1238 struct writeback_control *wbc,
1239 struct page *locked_page,
1240 u64 start, u64 end, int *page_started,
1241 unsigned long *nr_written)
1243 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1244 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1245 struct async_cow *ctx;
1246 struct async_chunk *async_chunk;
1247 unsigned long nr_pages;
1249 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1251 bool should_compress;
1253 const unsigned int write_flags = wbc_to_write_flags(wbc);
1255 unlock_extent(&inode->io_tree, start, end);
1257 if (inode->flags & BTRFS_INODE_NOCOMPRESS &&
1258 !btrfs_test_opt(fs_info, FORCE_COMPRESS)) {
1260 should_compress = false;
1262 should_compress = true;
1265 nofs_flag = memalloc_nofs_save();
1266 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1267 memalloc_nofs_restore(nofs_flag);
1270 unsigned clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC |
1271 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1272 EXTENT_DO_ACCOUNTING;
1273 unsigned long page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1274 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
1277 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1278 clear_bits, page_ops);
1282 async_chunk = ctx->chunks;
1283 atomic_set(&ctx->num_chunks, num_chunks);
1285 for (i = 0; i < num_chunks; i++) {
1286 if (should_compress)
1287 cur_end = min(end, start + SZ_512K - 1);
1292 * igrab is called higher up in the call chain, take only the
1293 * lightweight reference for the callback lifetime
1295 ihold(&inode->vfs_inode);
1296 async_chunk[i].pending = &ctx->num_chunks;
1297 async_chunk[i].inode = &inode->vfs_inode;
1298 async_chunk[i].start = start;
1299 async_chunk[i].end = cur_end;
1300 async_chunk[i].write_flags = write_flags;
1301 INIT_LIST_HEAD(&async_chunk[i].extents);
1304 * The locked_page comes all the way from writepage and its
1305 * the original page we were actually given. As we spread
1306 * this large delalloc region across multiple async_chunk
1307 * structs, only the first struct needs a pointer to locked_page
1309 * This way we don't need racey decisions about who is supposed
1314 * Depending on the compressibility, the pages might or
1315 * might not go through async. We want all of them to
1316 * be accounted against wbc once. Let's do it here
1317 * before the paths diverge. wbc accounting is used
1318 * only for foreign writeback detection and doesn't
1319 * need full accuracy. Just account the whole thing
1320 * against the first page.
1322 wbc_account_cgroup_owner(wbc, locked_page,
1324 async_chunk[i].locked_page = locked_page;
1327 async_chunk[i].locked_page = NULL;
1330 if (blkcg_css != blkcg_root_css) {
1332 async_chunk[i].blkcg_css = blkcg_css;
1334 async_chunk[i].blkcg_css = NULL;
1337 btrfs_init_work(&async_chunk[i].work, async_cow_start,
1338 async_cow_submit, async_cow_free);
1340 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1341 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1343 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1345 *nr_written += nr_pages;
1346 start = cur_end + 1;
1352 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1353 u64 bytenr, u64 num_bytes)
1356 struct btrfs_ordered_sum *sums;
1359 ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1360 bytenr + num_bytes - 1, &list, 0);
1361 if (ret == 0 && list_empty(&list))
1364 while (!list_empty(&list)) {
1365 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1366 list_del(&sums->list);
1374 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1375 const u64 start, const u64 end,
1376 int *page_started, unsigned long *nr_written)
1378 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1379 const bool is_reloc_ino = (inode->root->root_key.objectid ==
1380 BTRFS_DATA_RELOC_TREE_OBJECTID);
1381 const u64 range_bytes = end + 1 - start;
1382 struct extent_io_tree *io_tree = &inode->io_tree;
1383 u64 range_start = start;
1387 * If EXTENT_NORESERVE is set it means that when the buffered write was
1388 * made we had not enough available data space and therefore we did not
1389 * reserve data space for it, since we though we could do NOCOW for the
1390 * respective file range (either there is prealloc extent or the inode
1391 * has the NOCOW bit set).
1393 * However when we need to fallback to COW mode (because for example the
1394 * block group for the corresponding extent was turned to RO mode by a
1395 * scrub or relocation) we need to do the following:
1397 * 1) We increment the bytes_may_use counter of the data space info.
1398 * If COW succeeds, it allocates a new data extent and after doing
1399 * that it decrements the space info's bytes_may_use counter and
1400 * increments its bytes_reserved counter by the same amount (we do
1401 * this at btrfs_add_reserved_bytes()). So we need to increment the
1402 * bytes_may_use counter to compensate (when space is reserved at
1403 * buffered write time, the bytes_may_use counter is incremented);
1405 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1406 * that if the COW path fails for any reason, it decrements (through
1407 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1408 * data space info, which we incremented in the step above.
1410 * If we need to fallback to cow and the inode corresponds to a free
1411 * space cache inode or an inode of the data relocation tree, we must
1412 * also increment bytes_may_use of the data space_info for the same
1413 * reason. Space caches and relocated data extents always get a prealloc
1414 * extent for them, however scrub or balance may have set the block
1415 * group that contains that extent to RO mode and therefore force COW
1416 * when starting writeback.
1418 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1419 EXTENT_NORESERVE, 0);
1420 if (count > 0 || is_space_ino || is_reloc_ino) {
1422 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1423 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1425 if (is_space_ino || is_reloc_ino)
1426 bytes = range_bytes;
1428 spin_lock(&sinfo->lock);
1429 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1430 spin_unlock(&sinfo->lock);
1433 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1437 return cow_file_range(inode, locked_page, start, end, page_started,
1442 * when nowcow writeback call back. This checks for snapshots or COW copies
1443 * of the extents that exist in the file, and COWs the file as required.
1445 * If no cow copies or snapshots exist, we write directly to the existing
1448 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
1449 struct page *locked_page,
1450 const u64 start, const u64 end,
1451 int *page_started, int force,
1452 unsigned long *nr_written)
1454 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1455 struct btrfs_root *root = inode->root;
1456 struct btrfs_path *path;
1457 u64 cow_start = (u64)-1;
1458 u64 cur_offset = start;
1460 bool check_prev = true;
1461 const bool freespace_inode = btrfs_is_free_space_inode(inode);
1462 u64 ino = btrfs_ino(inode);
1464 u64 disk_bytenr = 0;
1466 path = btrfs_alloc_path();
1468 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1469 EXTENT_LOCKED | EXTENT_DELALLOC |
1470 EXTENT_DO_ACCOUNTING |
1471 EXTENT_DEFRAG, PAGE_UNLOCK |
1473 PAGE_SET_WRITEBACK |
1474 PAGE_END_WRITEBACK);
1479 struct btrfs_key found_key;
1480 struct btrfs_file_extent_item *fi;
1481 struct extent_buffer *leaf;
1491 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1497 * If there is no extent for our range when doing the initial
1498 * search, then go back to the previous slot as it will be the
1499 * one containing the search offset
1501 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1502 leaf = path->nodes[0];
1503 btrfs_item_key_to_cpu(leaf, &found_key,
1504 path->slots[0] - 1);
1505 if (found_key.objectid == ino &&
1506 found_key.type == BTRFS_EXTENT_DATA_KEY)
1511 /* Go to next leaf if we have exhausted the current one */
1512 leaf = path->nodes[0];
1513 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1514 ret = btrfs_next_leaf(root, path);
1516 if (cow_start != (u64)-1)
1517 cur_offset = cow_start;
1522 leaf = path->nodes[0];
1525 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1527 /* Didn't find anything for our INO */
1528 if (found_key.objectid > ino)
1531 * Keep searching until we find an EXTENT_ITEM or there are no
1532 * more extents for this inode
1534 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1535 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1540 /* Found key is not EXTENT_DATA_KEY or starts after req range */
1541 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1542 found_key.offset > end)
1546 * If the found extent starts after requested offset, then
1547 * adjust extent_end to be right before this extent begins
1549 if (found_key.offset > cur_offset) {
1550 extent_end = found_key.offset;
1556 * Found extent which begins before our range and potentially
1559 fi = btrfs_item_ptr(leaf, path->slots[0],
1560 struct btrfs_file_extent_item);
1561 extent_type = btrfs_file_extent_type(leaf, fi);
1563 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1564 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1565 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1566 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1567 extent_offset = btrfs_file_extent_offset(leaf, fi);
1568 extent_end = found_key.offset +
1569 btrfs_file_extent_num_bytes(leaf, fi);
1571 btrfs_file_extent_disk_num_bytes(leaf, fi);
1573 * If the extent we got ends before our current offset,
1574 * skip to the next extent.
1576 if (extent_end <= cur_offset) {
1581 if (disk_bytenr == 0)
1583 /* Skip compressed/encrypted/encoded extents */
1584 if (btrfs_file_extent_compression(leaf, fi) ||
1585 btrfs_file_extent_encryption(leaf, fi) ||
1586 btrfs_file_extent_other_encoding(leaf, fi))
1589 * If extent is created before the last volume's snapshot
1590 * this implies the extent is shared, hence we can't do
1591 * nocow. This is the same check as in
1592 * btrfs_cross_ref_exist but without calling
1593 * btrfs_search_slot.
1595 if (!freespace_inode &&
1596 btrfs_file_extent_generation(leaf, fi) <=
1597 btrfs_root_last_snapshot(&root->root_item))
1599 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1601 /* If extent is RO, we must COW it */
1602 if (btrfs_extent_readonly(fs_info, disk_bytenr))
1604 ret = btrfs_cross_ref_exist(root, ino,
1606 extent_offset, disk_bytenr, false);
1609 * ret could be -EIO if the above fails to read
1613 if (cow_start != (u64)-1)
1614 cur_offset = cow_start;
1618 WARN_ON_ONCE(freespace_inode);
1621 disk_bytenr += extent_offset;
1622 disk_bytenr += cur_offset - found_key.offset;
1623 num_bytes = min(end + 1, extent_end) - cur_offset;
1625 * If there are pending snapshots for this root, we
1626 * fall into common COW way
1628 if (!freespace_inode && atomic_read(&root->snapshot_force_cow))
1631 * force cow if csum exists in the range.
1632 * this ensure that csum for a given extent are
1633 * either valid or do not exist.
1635 ret = csum_exist_in_range(fs_info, disk_bytenr,
1639 * ret could be -EIO if the above fails to read
1643 if (cow_start != (u64)-1)
1644 cur_offset = cow_start;
1647 WARN_ON_ONCE(freespace_inode);
1650 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr))
1653 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1654 extent_end = found_key.offset + ram_bytes;
1655 extent_end = ALIGN(extent_end, fs_info->sectorsize);
1656 /* Skip extents outside of our requested range */
1657 if (extent_end <= start) {
1662 /* If this triggers then we have a memory corruption */
1667 * If nocow is false then record the beginning of the range
1668 * that needs to be COWed
1671 if (cow_start == (u64)-1)
1672 cow_start = cur_offset;
1673 cur_offset = extent_end;
1674 if (cur_offset > end)
1680 btrfs_release_path(path);
1683 * COW range from cow_start to found_key.offset - 1. As the key
1684 * will contain the beginning of the first extent that can be
1685 * NOCOW, following one which needs to be COW'ed
1687 if (cow_start != (u64)-1) {
1688 ret = fallback_to_cow(inode, locked_page,
1689 cow_start, found_key.offset - 1,
1690 page_started, nr_written);
1693 cow_start = (u64)-1;
1696 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1697 u64 orig_start = found_key.offset - extent_offset;
1698 struct extent_map *em;
1700 em = create_io_em(inode, cur_offset, num_bytes,
1702 disk_bytenr, /* block_start */
1703 num_bytes, /* block_len */
1704 disk_num_bytes, /* orig_block_len */
1705 ram_bytes, BTRFS_COMPRESS_NONE,
1706 BTRFS_ORDERED_PREALLOC);
1711 free_extent_map(em);
1712 ret = btrfs_add_ordered_extent(inode, cur_offset,
1713 disk_bytenr, num_bytes,
1715 BTRFS_ORDERED_PREALLOC);
1717 btrfs_drop_extent_cache(inode, cur_offset,
1718 cur_offset + num_bytes - 1,
1723 ret = btrfs_add_ordered_extent(inode, cur_offset,
1724 disk_bytenr, num_bytes,
1726 BTRFS_ORDERED_NOCOW);
1732 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1735 if (root->root_key.objectid ==
1736 BTRFS_DATA_RELOC_TREE_OBJECTID)
1738 * Error handled later, as we must prevent
1739 * extent_clear_unlock_delalloc() in error handler
1740 * from freeing metadata of created ordered extent.
1742 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1745 extent_clear_unlock_delalloc(inode, cur_offset,
1746 cur_offset + num_bytes - 1,
1747 locked_page, EXTENT_LOCKED |
1749 EXTENT_CLEAR_DATA_RESV,
1750 PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1752 cur_offset = extent_end;
1755 * btrfs_reloc_clone_csums() error, now we're OK to call error
1756 * handler, as metadata for created ordered extent will only
1757 * be freed by btrfs_finish_ordered_io().
1761 if (cur_offset > end)
1764 btrfs_release_path(path);
1766 if (cur_offset <= end && cow_start == (u64)-1)
1767 cow_start = cur_offset;
1769 if (cow_start != (u64)-1) {
1771 ret = fallback_to_cow(inode, locked_page, cow_start, end,
1772 page_started, nr_written);
1779 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1781 if (ret && cur_offset < end)
1782 extent_clear_unlock_delalloc(inode, cur_offset, end,
1783 locked_page, EXTENT_LOCKED |
1784 EXTENT_DELALLOC | EXTENT_DEFRAG |
1785 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1787 PAGE_SET_WRITEBACK |
1788 PAGE_END_WRITEBACK);
1789 btrfs_free_path(path);
1793 static inline int need_force_cow(struct btrfs_inode *inode, u64 start, u64 end)
1796 if (!(inode->flags & BTRFS_INODE_NODATACOW) &&
1797 !(inode->flags & BTRFS_INODE_PREALLOC))
1801 * @defrag_bytes is a hint value, no spinlock held here,
1802 * if is not zero, it means the file is defragging.
1803 * Force cow if given extent needs to be defragged.
1805 if (inode->defrag_bytes &&
1806 test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG, 0, NULL))
1813 * Function to process delayed allocation (create CoW) for ranges which are
1814 * being touched for the first time.
1816 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
1817 u64 start, u64 end, int *page_started, unsigned long *nr_written,
1818 struct writeback_control *wbc)
1821 int force_cow = need_force_cow(inode, start, end);
1823 if (inode->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1824 ret = run_delalloc_nocow(inode, locked_page, start, end,
1825 page_started, 1, nr_written);
1826 } else if (inode->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1827 ret = run_delalloc_nocow(inode, locked_page, start, end,
1828 page_started, 0, nr_written);
1829 } else if (!inode_can_compress(inode) ||
1830 !inode_need_compress(inode, start, end)) {
1831 ret = cow_file_range(inode, locked_page, start, end,
1832 page_started, nr_written, 1);
1834 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
1835 ret = cow_file_range_async(inode, wbc, locked_page, start, end,
1836 page_started, nr_written);
1839 btrfs_cleanup_ordered_extents(inode, locked_page, start,
1844 void btrfs_split_delalloc_extent(struct inode *inode,
1845 struct extent_state *orig, u64 split)
1849 /* not delalloc, ignore it */
1850 if (!(orig->state & EXTENT_DELALLOC))
1853 size = orig->end - orig->start + 1;
1854 if (size > BTRFS_MAX_EXTENT_SIZE) {
1859 * See the explanation in btrfs_merge_delalloc_extent, the same
1860 * applies here, just in reverse.
1862 new_size = orig->end - split + 1;
1863 num_extents = count_max_extents(new_size);
1864 new_size = split - orig->start;
1865 num_extents += count_max_extents(new_size);
1866 if (count_max_extents(size) >= num_extents)
1870 spin_lock(&BTRFS_I(inode)->lock);
1871 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1872 spin_unlock(&BTRFS_I(inode)->lock);
1876 * Handle merged delayed allocation extents so we can keep track of new extents
1877 * that are just merged onto old extents, such as when we are doing sequential
1878 * writes, so we can properly account for the metadata space we'll need.
1880 void btrfs_merge_delalloc_extent(struct inode *inode, struct extent_state *new,
1881 struct extent_state *other)
1883 u64 new_size, old_size;
1886 /* not delalloc, ignore it */
1887 if (!(other->state & EXTENT_DELALLOC))
1890 if (new->start > other->start)
1891 new_size = new->end - other->start + 1;
1893 new_size = other->end - new->start + 1;
1895 /* we're not bigger than the max, unreserve the space and go */
1896 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1897 spin_lock(&BTRFS_I(inode)->lock);
1898 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1899 spin_unlock(&BTRFS_I(inode)->lock);
1904 * We have to add up either side to figure out how many extents were
1905 * accounted for before we merged into one big extent. If the number of
1906 * extents we accounted for is <= the amount we need for the new range
1907 * then we can return, otherwise drop. Think of it like this
1911 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1912 * need 2 outstanding extents, on one side we have 1 and the other side
1913 * we have 1 so they are == and we can return. But in this case
1915 * [MAX_SIZE+4k][MAX_SIZE+4k]
1917 * Each range on their own accounts for 2 extents, but merged together
1918 * they are only 3 extents worth of accounting, so we need to drop in
1921 old_size = other->end - other->start + 1;
1922 num_extents = count_max_extents(old_size);
1923 old_size = new->end - new->start + 1;
1924 num_extents += count_max_extents(old_size);
1925 if (count_max_extents(new_size) >= num_extents)
1928 spin_lock(&BTRFS_I(inode)->lock);
1929 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1930 spin_unlock(&BTRFS_I(inode)->lock);
1933 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1934 struct inode *inode)
1936 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1938 spin_lock(&root->delalloc_lock);
1939 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1940 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1941 &root->delalloc_inodes);
1942 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1943 &BTRFS_I(inode)->runtime_flags);
1944 root->nr_delalloc_inodes++;
1945 if (root->nr_delalloc_inodes == 1) {
1946 spin_lock(&fs_info->delalloc_root_lock);
1947 BUG_ON(!list_empty(&root->delalloc_root));
1948 list_add_tail(&root->delalloc_root,
1949 &fs_info->delalloc_roots);
1950 spin_unlock(&fs_info->delalloc_root_lock);
1953 spin_unlock(&root->delalloc_lock);
1957 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
1958 struct btrfs_inode *inode)
1960 struct btrfs_fs_info *fs_info = root->fs_info;
1962 if (!list_empty(&inode->delalloc_inodes)) {
1963 list_del_init(&inode->delalloc_inodes);
1964 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1965 &inode->runtime_flags);
1966 root->nr_delalloc_inodes--;
1967 if (!root->nr_delalloc_inodes) {
1968 ASSERT(list_empty(&root->delalloc_inodes));
1969 spin_lock(&fs_info->delalloc_root_lock);
1970 BUG_ON(list_empty(&root->delalloc_root));
1971 list_del_init(&root->delalloc_root);
1972 spin_unlock(&fs_info->delalloc_root_lock);
1977 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1978 struct btrfs_inode *inode)
1980 spin_lock(&root->delalloc_lock);
1981 __btrfs_del_delalloc_inode(root, inode);
1982 spin_unlock(&root->delalloc_lock);
1986 * Properly track delayed allocation bytes in the inode and to maintain the
1987 * list of inodes that have pending delalloc work to be done.
1989 void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state,
1992 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1994 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1997 * set_bit and clear bit hooks normally require _irqsave/restore
1998 * but in this case, we are only testing for the DELALLOC
1999 * bit, which is only set or cleared with irqs on
2001 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2002 struct btrfs_root *root = BTRFS_I(inode)->root;
2003 u64 len = state->end + 1 - state->start;
2004 u32 num_extents = count_max_extents(len);
2005 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
2007 spin_lock(&BTRFS_I(inode)->lock);
2008 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
2009 spin_unlock(&BTRFS_I(inode)->lock);
2011 /* For sanity tests */
2012 if (btrfs_is_testing(fs_info))
2015 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2016 fs_info->delalloc_batch);
2017 spin_lock(&BTRFS_I(inode)->lock);
2018 BTRFS_I(inode)->delalloc_bytes += len;
2019 if (*bits & EXTENT_DEFRAG)
2020 BTRFS_I(inode)->defrag_bytes += len;
2021 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2022 &BTRFS_I(inode)->runtime_flags))
2023 btrfs_add_delalloc_inodes(root, inode);
2024 spin_unlock(&BTRFS_I(inode)->lock);
2027 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2028 (*bits & EXTENT_DELALLOC_NEW)) {
2029 spin_lock(&BTRFS_I(inode)->lock);
2030 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
2032 spin_unlock(&BTRFS_I(inode)->lock);
2037 * Once a range is no longer delalloc this function ensures that proper
2038 * accounting happens.
2040 void btrfs_clear_delalloc_extent(struct inode *vfs_inode,
2041 struct extent_state *state, unsigned *bits)
2043 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
2044 struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb);
2045 u64 len = state->end + 1 - state->start;
2046 u32 num_extents = count_max_extents(len);
2048 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
2049 spin_lock(&inode->lock);
2050 inode->defrag_bytes -= len;
2051 spin_unlock(&inode->lock);
2055 * set_bit and clear bit hooks normally require _irqsave/restore
2056 * but in this case, we are only testing for the DELALLOC
2057 * bit, which is only set or cleared with irqs on
2059 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2060 struct btrfs_root *root = inode->root;
2061 bool do_list = !btrfs_is_free_space_inode(inode);
2063 spin_lock(&inode->lock);
2064 btrfs_mod_outstanding_extents(inode, -num_extents);
2065 spin_unlock(&inode->lock);
2068 * We don't reserve metadata space for space cache inodes so we
2069 * don't need to call delalloc_release_metadata if there is an
2072 if (*bits & EXTENT_CLEAR_META_RESV &&
2073 root != fs_info->tree_root)
2074 btrfs_delalloc_release_metadata(inode, len, false);
2076 /* For sanity tests. */
2077 if (btrfs_is_testing(fs_info))
2080 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
2081 do_list && !(state->state & EXTENT_NORESERVE) &&
2082 (*bits & EXTENT_CLEAR_DATA_RESV))
2083 btrfs_free_reserved_data_space_noquota(fs_info, len);
2085 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2086 fs_info->delalloc_batch);
2087 spin_lock(&inode->lock);
2088 inode->delalloc_bytes -= len;
2089 if (do_list && inode->delalloc_bytes == 0 &&
2090 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2091 &inode->runtime_flags))
2092 btrfs_del_delalloc_inode(root, inode);
2093 spin_unlock(&inode->lock);
2096 if ((state->state & EXTENT_DELALLOC_NEW) &&
2097 (*bits & EXTENT_DELALLOC_NEW)) {
2098 spin_lock(&inode->lock);
2099 ASSERT(inode->new_delalloc_bytes >= len);
2100 inode->new_delalloc_bytes -= len;
2101 spin_unlock(&inode->lock);
2106 * btrfs_bio_fits_in_stripe - Checks whether the size of the given bio will fit
2107 * in a chunk's stripe. This function ensures that bios do not span a
2110 * @page - The page we are about to add to the bio
2111 * @size - size we want to add to the bio
2112 * @bio - bio we want to ensure is smaller than a stripe
2113 * @bio_flags - flags of the bio
2115 * return 1 if page cannot be added to the bio
2116 * return 0 if page can be added to the bio
2117 * return error otherwise
2119 int btrfs_bio_fits_in_stripe(struct page *page, size_t size, struct bio *bio,
2120 unsigned long bio_flags)
2122 struct inode *inode = page->mapping->host;
2123 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2124 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
2128 struct btrfs_io_geometry geom;
2130 if (bio_flags & EXTENT_BIO_COMPRESSED)
2133 length = bio->bi_iter.bi_size;
2134 map_length = length;
2135 ret = btrfs_get_io_geometry(fs_info, btrfs_op(bio), logical, map_length,
2140 if (geom.len < length + size)
2146 * in order to insert checksums into the metadata in large chunks,
2147 * we wait until bio submission time. All the pages in the bio are
2148 * checksummed and sums are attached onto the ordered extent record.
2150 * At IO completion time the cums attached on the ordered extent record
2151 * are inserted into the btree
2153 static blk_status_t btrfs_submit_bio_start(void *private_data, struct bio *bio,
2156 struct inode *inode = private_data;
2158 return btrfs_csum_one_bio(BTRFS_I(inode), bio, 0, 0);
2162 * extent_io.c submission hook. This does the right thing for csum calculation
2163 * on write, or reading the csums from the tree before a read.
2165 * Rules about async/sync submit,
2166 * a) read: sync submit
2168 * b) write without checksum: sync submit
2170 * c) write with checksum:
2171 * c-1) if bio is issued by fsync: sync submit
2172 * (sync_writers != 0)
2174 * c-2) if root is reloc root: sync submit
2175 * (only in case of buffered IO)
2177 * c-3) otherwise: async submit
2179 blk_status_t btrfs_submit_data_bio(struct inode *inode, struct bio *bio,
2180 int mirror_num, unsigned long bio_flags)
2183 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2184 struct btrfs_root *root = BTRFS_I(inode)->root;
2185 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
2186 blk_status_t ret = 0;
2188 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
2190 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
2192 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
2193 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
2195 if (bio_op(bio) != REQ_OP_WRITE) {
2196 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
2200 if (bio_flags & EXTENT_BIO_COMPRESSED) {
2201 ret = btrfs_submit_compressed_read(inode, bio,
2205 } else if (!skip_sum) {
2206 ret = btrfs_lookup_bio_sums(inode, bio, (u64)-1, NULL);
2211 } else if (async && !skip_sum) {
2212 /* csum items have already been cloned */
2213 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2215 /* we're doing a write, do the async checksumming */
2216 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags,
2217 0, inode, btrfs_submit_bio_start);
2219 } else if (!skip_sum) {
2220 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, 0, 0);
2226 ret = btrfs_map_bio(fs_info, bio, mirror_num);
2230 bio->bi_status = ret;
2237 * given a list of ordered sums record them in the inode. This happens
2238 * at IO completion time based on sums calculated at bio submission time.
2240 static int add_pending_csums(struct btrfs_trans_handle *trans,
2241 struct list_head *list)
2243 struct btrfs_ordered_sum *sum;
2246 list_for_each_entry(sum, list, list) {
2247 trans->adding_csums = true;
2248 ret = btrfs_csum_file_blocks(trans, trans->fs_info->csum_root, sum);
2249 trans->adding_csums = false;
2256 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2257 unsigned int extra_bits,
2258 struct extent_state **cached_state)
2260 WARN_ON(PAGE_ALIGNED(end));
2261 return set_extent_delalloc(&inode->io_tree, start, end, extra_bits,
2265 /* see btrfs_writepage_start_hook for details on why this is required */
2266 struct btrfs_writepage_fixup {
2268 struct inode *inode;
2269 struct btrfs_work work;
2272 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2274 struct btrfs_writepage_fixup *fixup;
2275 struct btrfs_ordered_extent *ordered;
2276 struct extent_state *cached_state = NULL;
2277 struct extent_changeset *data_reserved = NULL;
2279 struct btrfs_inode *inode;
2283 bool free_delalloc_space = true;
2285 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2287 inode = BTRFS_I(fixup->inode);
2288 page_start = page_offset(page);
2289 page_end = page_offset(page) + PAGE_SIZE - 1;
2292 * This is similar to page_mkwrite, we need to reserve the space before
2293 * we take the page lock.
2295 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2301 * Before we queued this fixup, we took a reference on the page.
2302 * page->mapping may go NULL, but it shouldn't be moved to a different
2305 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2307 * Unfortunately this is a little tricky, either
2309 * 1) We got here and our page had already been dealt with and
2310 * we reserved our space, thus ret == 0, so we need to just
2311 * drop our space reservation and bail. This can happen the
2312 * first time we come into the fixup worker, or could happen
2313 * while waiting for the ordered extent.
2314 * 2) Our page was already dealt with, but we happened to get an
2315 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2316 * this case we obviously don't have anything to release, but
2317 * because the page was already dealt with we don't want to
2318 * mark the page with an error, so make sure we're resetting
2319 * ret to 0. This is why we have this check _before_ the ret
2320 * check, because we do not want to have a surprise ENOSPC
2321 * when the page was already properly dealt with.
2324 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2325 btrfs_delalloc_release_space(inode, data_reserved,
2326 page_start, PAGE_SIZE,
2334 * We can't mess with the page state unless it is locked, so now that
2335 * it is locked bail if we failed to make our space reservation.
2340 lock_extent_bits(&inode->io_tree, page_start, page_end, &cached_state);
2342 /* already ordered? We're done */
2343 if (PagePrivate2(page))
2346 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2348 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2351 btrfs_start_ordered_extent(ordered, 1);
2352 btrfs_put_ordered_extent(ordered);
2356 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2362 * Everything went as planned, we're now the owner of a dirty page with
2363 * delayed allocation bits set and space reserved for our COW
2366 * The page was dirty when we started, nothing should have cleaned it.
2368 BUG_ON(!PageDirty(page));
2369 free_delalloc_space = false;
2371 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2372 if (free_delalloc_space)
2373 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2375 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2380 * We hit ENOSPC or other errors. Update the mapping and page
2381 * to reflect the errors and clean the page.
2383 mapping_set_error(page->mapping, ret);
2384 end_extent_writepage(page, ret, page_start, page_end);
2385 clear_page_dirty_for_io(page);
2388 ClearPageChecked(page);
2392 extent_changeset_free(data_reserved);
2394 * As a precaution, do a delayed iput in case it would be the last iput
2395 * that could need flushing space. Recursing back to fixup worker would
2398 btrfs_add_delayed_iput(&inode->vfs_inode);
2402 * There are a few paths in the higher layers of the kernel that directly
2403 * set the page dirty bit without asking the filesystem if it is a
2404 * good idea. This causes problems because we want to make sure COW
2405 * properly happens and the data=ordered rules are followed.
2407 * In our case any range that doesn't have the ORDERED bit set
2408 * hasn't been properly setup for IO. We kick off an async process
2409 * to fix it up. The async helper will wait for ordered extents, set
2410 * the delalloc bit and make it safe to write the page.
2412 int btrfs_writepage_cow_fixup(struct page *page, u64 start, u64 end)
2414 struct inode *inode = page->mapping->host;
2415 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2416 struct btrfs_writepage_fixup *fixup;
2418 /* this page is properly in the ordered list */
2419 if (TestClearPagePrivate2(page))
2423 * PageChecked is set below when we create a fixup worker for this page,
2424 * don't try to create another one if we're already PageChecked()
2426 * The extent_io writepage code will redirty the page if we send back
2429 if (PageChecked(page))
2432 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2437 * We are already holding a reference to this inode from
2438 * write_cache_pages. We need to hold it because the space reservation
2439 * takes place outside of the page lock, and we can't trust
2440 * page->mapping outside of the page lock.
2443 SetPageChecked(page);
2445 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
2447 fixup->inode = inode;
2448 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2453 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2454 struct btrfs_inode *inode, u64 file_pos,
2455 struct btrfs_file_extent_item *stack_fi,
2456 u64 qgroup_reserved)
2458 struct btrfs_root *root = inode->root;
2459 struct btrfs_path *path;
2460 struct extent_buffer *leaf;
2461 struct btrfs_key ins;
2462 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
2463 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
2464 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
2465 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
2466 int extent_inserted = 0;
2469 path = btrfs_alloc_path();
2474 * we may be replacing one extent in the tree with another.
2475 * The new extent is pinned in the extent map, and we don't want
2476 * to drop it from the cache until it is completely in the btree.
2478 * So, tell btrfs_drop_extents to leave this extent in the cache.
2479 * the caller is expected to unpin it and allow it to be merged
2482 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2483 file_pos + num_bytes, NULL, 0,
2484 1, sizeof(*stack_fi), &extent_inserted);
2488 if (!extent_inserted) {
2489 ins.objectid = btrfs_ino(inode);
2490 ins.offset = file_pos;
2491 ins.type = BTRFS_EXTENT_DATA_KEY;
2493 path->leave_spinning = 1;
2494 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2499 leaf = path->nodes[0];
2500 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
2501 write_extent_buffer(leaf, stack_fi,
2502 btrfs_item_ptr_offset(leaf, path->slots[0]),
2503 sizeof(struct btrfs_file_extent_item));
2505 btrfs_mark_buffer_dirty(leaf);
2506 btrfs_release_path(path);
2508 inode_add_bytes(&inode->vfs_inode, num_bytes);
2510 ins.objectid = disk_bytenr;
2511 ins.offset = disk_num_bytes;
2512 ins.type = BTRFS_EXTENT_ITEM_KEY;
2514 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
2518 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
2519 file_pos, qgroup_reserved, &ins);
2521 btrfs_free_path(path);
2526 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2529 struct btrfs_block_group *cache;
2531 cache = btrfs_lookup_block_group(fs_info, start);
2534 spin_lock(&cache->lock);
2535 cache->delalloc_bytes -= len;
2536 spin_unlock(&cache->lock);
2538 btrfs_put_block_group(cache);
2541 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
2542 struct btrfs_ordered_extent *oe)
2544 struct btrfs_file_extent_item stack_fi;
2547 memset(&stack_fi, 0, sizeof(stack_fi));
2548 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
2549 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
2550 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
2551 oe->disk_num_bytes);
2552 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags))
2553 logical_len = oe->truncated_len;
2555 logical_len = oe->num_bytes;
2556 btrfs_set_stack_file_extent_num_bytes(&stack_fi, logical_len);
2557 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, logical_len);
2558 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
2559 /* Encryption and other encoding is reserved and all 0 */
2561 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
2562 oe->file_offset, &stack_fi,
2567 * As ordered data IO finishes, this gets called so we can finish
2568 * an ordered extent if the range of bytes in the file it covers are
2571 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2573 struct inode *inode = ordered_extent->inode;
2574 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2575 struct btrfs_root *root = BTRFS_I(inode)->root;
2576 struct btrfs_trans_handle *trans = NULL;
2577 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2578 struct extent_state *cached_state = NULL;
2580 int compress_type = 0;
2582 u64 logical_len = ordered_extent->num_bytes;
2583 bool freespace_inode;
2584 bool truncated = false;
2585 bool range_locked = false;
2586 bool clear_new_delalloc_bytes = false;
2587 bool clear_reserved_extent = true;
2588 unsigned int clear_bits;
2590 start = ordered_extent->file_offset;
2591 end = start + ordered_extent->num_bytes - 1;
2593 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2594 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2595 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2596 clear_new_delalloc_bytes = true;
2598 freespace_inode = btrfs_is_free_space_inode(BTRFS_I(inode));
2600 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2605 btrfs_free_io_failure_record(BTRFS_I(inode), start, end);
2607 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2609 logical_len = ordered_extent->truncated_len;
2610 /* Truncated the entire extent, don't bother adding */
2615 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2616 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2618 btrfs_inode_safe_disk_i_size_write(inode, 0);
2619 if (freespace_inode)
2620 trans = btrfs_join_transaction_spacecache(root);
2622 trans = btrfs_join_transaction(root);
2623 if (IS_ERR(trans)) {
2624 ret = PTR_ERR(trans);
2628 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
2629 ret = btrfs_update_inode_fallback(trans, root, inode);
2630 if (ret) /* -ENOMEM or corruption */
2631 btrfs_abort_transaction(trans, ret);
2635 range_locked = true;
2636 lock_extent_bits(io_tree, start, end, &cached_state);
2638 if (freespace_inode)
2639 trans = btrfs_join_transaction_spacecache(root);
2641 trans = btrfs_join_transaction(root);
2642 if (IS_ERR(trans)) {
2643 ret = PTR_ERR(trans);
2648 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
2650 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2651 compress_type = ordered_extent->compress_type;
2652 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2653 BUG_ON(compress_type);
2654 ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
2655 ordered_extent->file_offset,
2656 ordered_extent->file_offset +
2659 BUG_ON(root == fs_info->tree_root);
2660 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
2662 clear_reserved_extent = false;
2663 btrfs_release_delalloc_bytes(fs_info,
2664 ordered_extent->disk_bytenr,
2665 ordered_extent->disk_num_bytes);
2668 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2669 ordered_extent->file_offset,
2670 ordered_extent->num_bytes, trans->transid);
2672 btrfs_abort_transaction(trans, ret);
2676 ret = add_pending_csums(trans, &ordered_extent->list);
2678 btrfs_abort_transaction(trans, ret);
2682 btrfs_inode_safe_disk_i_size_write(inode, 0);
2683 ret = btrfs_update_inode_fallback(trans, root, inode);
2684 if (ret) { /* -ENOMEM or corruption */
2685 btrfs_abort_transaction(trans, ret);
2690 clear_bits = EXTENT_DEFRAG;
2692 clear_bits |= EXTENT_LOCKED;
2693 if (clear_new_delalloc_bytes)
2694 clear_bits |= EXTENT_DELALLOC_NEW;
2695 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, clear_bits,
2696 (clear_bits & EXTENT_LOCKED) ? 1 : 0, 0,
2700 btrfs_end_transaction(trans);
2702 if (ret || truncated) {
2703 u64 unwritten_start = start;
2706 unwritten_start += logical_len;
2707 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
2709 /* Drop the cache for the part of the extent we didn't write. */
2710 btrfs_drop_extent_cache(BTRFS_I(inode), unwritten_start, end, 0);
2713 * If the ordered extent had an IOERR or something else went
2714 * wrong we need to return the space for this ordered extent
2715 * back to the allocator. We only free the extent in the
2716 * truncated case if we didn't write out the extent at all.
2718 * If we made it past insert_reserved_file_extent before we
2719 * errored out then we don't need to do this as the accounting
2720 * has already been done.
2722 if ((ret || !logical_len) &&
2723 clear_reserved_extent &&
2724 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2725 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2727 * Discard the range before returning it back to the
2730 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
2731 btrfs_discard_extent(fs_info,
2732 ordered_extent->disk_bytenr,
2733 ordered_extent->disk_num_bytes,
2735 btrfs_free_reserved_extent(fs_info,
2736 ordered_extent->disk_bytenr,
2737 ordered_extent->disk_num_bytes, 1);
2742 * This needs to be done to make sure anybody waiting knows we are done
2743 * updating everything for this ordered extent.
2745 btrfs_remove_ordered_extent(BTRFS_I(inode), ordered_extent);
2748 btrfs_put_ordered_extent(ordered_extent);
2749 /* once for the tree */
2750 btrfs_put_ordered_extent(ordered_extent);
2755 static void finish_ordered_fn(struct btrfs_work *work)
2757 struct btrfs_ordered_extent *ordered_extent;
2758 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2759 btrfs_finish_ordered_io(ordered_extent);
2762 void btrfs_writepage_endio_finish_ordered(struct page *page, u64 start,
2763 u64 end, int uptodate)
2765 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
2766 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2767 struct btrfs_ordered_extent *ordered_extent = NULL;
2768 struct btrfs_workqueue *wq;
2770 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2772 ClearPagePrivate2(page);
2773 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2774 end - start + 1, uptodate))
2777 if (btrfs_is_free_space_inode(inode))
2778 wq = fs_info->endio_freespace_worker;
2780 wq = fs_info->endio_write_workers;
2782 btrfs_init_work(&ordered_extent->work, finish_ordered_fn, NULL, NULL);
2783 btrfs_queue_work(wq, &ordered_extent->work);
2786 static int check_data_csum(struct inode *inode, struct btrfs_io_bio *io_bio,
2787 int icsum, struct page *page, int pgoff, u64 start,
2790 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2791 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
2793 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2795 u8 csum[BTRFS_CSUM_SIZE];
2797 csum_expected = ((u8 *)io_bio->csum) + icsum * csum_size;
2799 kaddr = kmap_atomic(page);
2800 shash->tfm = fs_info->csum_shash;
2802 crypto_shash_digest(shash, kaddr + pgoff, len, csum);
2804 if (memcmp(csum, csum_expected, csum_size))
2807 kunmap_atomic(kaddr);
2810 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
2811 io_bio->mirror_num);
2813 btrfs_dev_stat_inc_and_print(io_bio->device,
2814 BTRFS_DEV_STAT_CORRUPTION_ERRS);
2815 memset(kaddr + pgoff, 1, len);
2816 flush_dcache_page(page);
2817 kunmap_atomic(kaddr);
2822 * when reads are done, we need to check csums to verify the data is correct
2823 * if there's a match, we allow the bio to finish. If not, the code in
2824 * extent_io.c will try to find good copies for us.
2826 int btrfs_verify_data_csum(struct btrfs_io_bio *io_bio, u64 phy_offset,
2827 struct page *page, u64 start, u64 end, int mirror)
2829 size_t offset = start - page_offset(page);
2830 struct inode *inode = page->mapping->host;
2831 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2832 struct btrfs_root *root = BTRFS_I(inode)->root;
2834 if (PageChecked(page)) {
2835 ClearPageChecked(page);
2839 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2842 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2843 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2844 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
2848 phy_offset >>= inode->i_sb->s_blocksize_bits;
2849 return check_data_csum(inode, io_bio, phy_offset, page, offset, start,
2850 (size_t)(end - start + 1));
2854 * btrfs_add_delayed_iput - perform a delayed iput on @inode
2856 * @inode: The inode we want to perform iput on
2858 * This function uses the generic vfs_inode::i_count to track whether we should
2859 * just decrement it (in case it's > 1) or if this is the last iput then link
2860 * the inode to the delayed iput machinery. Delayed iputs are processed at
2861 * transaction commit time/superblock commit/cleaner kthread.
2863 void btrfs_add_delayed_iput(struct inode *inode)
2865 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2866 struct btrfs_inode *binode = BTRFS_I(inode);
2868 if (atomic_add_unless(&inode->i_count, -1, 1))
2871 atomic_inc(&fs_info->nr_delayed_iputs);
2872 spin_lock(&fs_info->delayed_iput_lock);
2873 ASSERT(list_empty(&binode->delayed_iput));
2874 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
2875 spin_unlock(&fs_info->delayed_iput_lock);
2876 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
2877 wake_up_process(fs_info->cleaner_kthread);
2880 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
2881 struct btrfs_inode *inode)
2883 list_del_init(&inode->delayed_iput);
2884 spin_unlock(&fs_info->delayed_iput_lock);
2885 iput(&inode->vfs_inode);
2886 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
2887 wake_up(&fs_info->delayed_iputs_wait);
2888 spin_lock(&fs_info->delayed_iput_lock);
2891 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
2892 struct btrfs_inode *inode)
2894 if (!list_empty(&inode->delayed_iput)) {
2895 spin_lock(&fs_info->delayed_iput_lock);
2896 if (!list_empty(&inode->delayed_iput))
2897 run_delayed_iput_locked(fs_info, inode);
2898 spin_unlock(&fs_info->delayed_iput_lock);
2902 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
2905 spin_lock(&fs_info->delayed_iput_lock);
2906 while (!list_empty(&fs_info->delayed_iputs)) {
2907 struct btrfs_inode *inode;
2909 inode = list_first_entry(&fs_info->delayed_iputs,
2910 struct btrfs_inode, delayed_iput);
2911 run_delayed_iput_locked(fs_info, inode);
2913 spin_unlock(&fs_info->delayed_iput_lock);
2917 * btrfs_wait_on_delayed_iputs - wait on the delayed iputs to be done running
2918 * @fs_info - the fs_info for this fs
2919 * @return - EINTR if we were killed, 0 if nothing's pending
2921 * This will wait on any delayed iputs that are currently running with KILLABLE
2922 * set. Once they are all done running we will return, unless we are killed in
2923 * which case we return EINTR. This helps in user operations like fallocate etc
2924 * that might get blocked on the iputs.
2926 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
2928 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
2929 atomic_read(&fs_info->nr_delayed_iputs) == 0);
2936 * This creates an orphan entry for the given inode in case something goes wrong
2937 * in the middle of an unlink.
2939 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
2940 struct btrfs_inode *inode)
2944 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
2945 if (ret && ret != -EEXIST) {
2946 btrfs_abort_transaction(trans, ret);
2954 * We have done the delete so we can go ahead and remove the orphan item for
2955 * this particular inode.
2957 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
2958 struct btrfs_inode *inode)
2960 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
2964 * this cleans up any orphans that may be left on the list from the last use
2967 int btrfs_orphan_cleanup(struct btrfs_root *root)
2969 struct btrfs_fs_info *fs_info = root->fs_info;
2970 struct btrfs_path *path;
2971 struct extent_buffer *leaf;
2972 struct btrfs_key key, found_key;
2973 struct btrfs_trans_handle *trans;
2974 struct inode *inode;
2975 u64 last_objectid = 0;
2976 int ret = 0, nr_unlink = 0;
2978 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2981 path = btrfs_alloc_path();
2986 path->reada = READA_BACK;
2988 key.objectid = BTRFS_ORPHAN_OBJECTID;
2989 key.type = BTRFS_ORPHAN_ITEM_KEY;
2990 key.offset = (u64)-1;
2993 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2998 * if ret == 0 means we found what we were searching for, which
2999 * is weird, but possible, so only screw with path if we didn't
3000 * find the key and see if we have stuff that matches
3004 if (path->slots[0] == 0)
3009 /* pull out the item */
3010 leaf = path->nodes[0];
3011 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3013 /* make sure the item matches what we want */
3014 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3016 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3019 /* release the path since we're done with it */
3020 btrfs_release_path(path);
3023 * this is where we are basically btrfs_lookup, without the
3024 * crossing root thing. we store the inode number in the
3025 * offset of the orphan item.
3028 if (found_key.offset == last_objectid) {
3030 "Error removing orphan entry, stopping orphan cleanup");
3035 last_objectid = found_key.offset;
3037 found_key.objectid = found_key.offset;
3038 found_key.type = BTRFS_INODE_ITEM_KEY;
3039 found_key.offset = 0;
3040 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3041 ret = PTR_ERR_OR_ZERO(inode);
3042 if (ret && ret != -ENOENT)
3045 if (ret == -ENOENT && root == fs_info->tree_root) {
3046 struct btrfs_root *dead_root;
3047 int is_dead_root = 0;
3050 * this is an orphan in the tree root. Currently these
3051 * could come from 2 sources:
3052 * a) a snapshot deletion in progress
3053 * b) a free space cache inode
3054 * We need to distinguish those two, as the snapshot
3055 * orphan must not get deleted.
3056 * find_dead_roots already ran before us, so if this
3057 * is a snapshot deletion, we should find the root
3058 * in the fs_roots radix tree.
3061 spin_lock(&fs_info->fs_roots_radix_lock);
3062 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3063 (unsigned long)found_key.objectid);
3064 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3066 spin_unlock(&fs_info->fs_roots_radix_lock);
3069 /* prevent this orphan from being found again */
3070 key.offset = found_key.objectid - 1;
3077 * If we have an inode with links, there are a couple of
3078 * possibilities. Old kernels (before v3.12) used to create an
3079 * orphan item for truncate indicating that there were possibly
3080 * extent items past i_size that needed to be deleted. In v3.12,
3081 * truncate was changed to update i_size in sync with the extent
3082 * items, but the (useless) orphan item was still created. Since
3083 * v4.18, we don't create the orphan item for truncate at all.
3085 * So, this item could mean that we need to do a truncate, but
3086 * only if this filesystem was last used on a pre-v3.12 kernel
3087 * and was not cleanly unmounted. The odds of that are quite
3088 * slim, and it's a pain to do the truncate now, so just delete
3091 * It's also possible that this orphan item was supposed to be
3092 * deleted but wasn't. The inode number may have been reused,
3093 * but either way, we can delete the orphan item.
3095 if (ret == -ENOENT || inode->i_nlink) {
3098 trans = btrfs_start_transaction(root, 1);
3099 if (IS_ERR(trans)) {
3100 ret = PTR_ERR(trans);
3103 btrfs_debug(fs_info, "auto deleting %Lu",
3104 found_key.objectid);
3105 ret = btrfs_del_orphan_item(trans, root,
3106 found_key.objectid);
3107 btrfs_end_transaction(trans);
3115 /* this will do delete_inode and everything for us */
3118 /* release the path since we're done with it */
3119 btrfs_release_path(path);
3121 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3123 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3124 trans = btrfs_join_transaction(root);
3126 btrfs_end_transaction(trans);
3130 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3134 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3135 btrfs_free_path(path);
3140 * very simple check to peek ahead in the leaf looking for xattrs. If we
3141 * don't find any xattrs, we know there can't be any acls.
3143 * slot is the slot the inode is in, objectid is the objectid of the inode
3145 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3146 int slot, u64 objectid,
3147 int *first_xattr_slot)
3149 u32 nritems = btrfs_header_nritems(leaf);
3150 struct btrfs_key found_key;
3151 static u64 xattr_access = 0;
3152 static u64 xattr_default = 0;
3155 if (!xattr_access) {
3156 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3157 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3158 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3159 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3163 *first_xattr_slot = -1;
3164 while (slot < nritems) {
3165 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3167 /* we found a different objectid, there must not be acls */
3168 if (found_key.objectid != objectid)
3171 /* we found an xattr, assume we've got an acl */
3172 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3173 if (*first_xattr_slot == -1)
3174 *first_xattr_slot = slot;
3175 if (found_key.offset == xattr_access ||
3176 found_key.offset == xattr_default)
3181 * we found a key greater than an xattr key, there can't
3182 * be any acls later on
3184 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3191 * it goes inode, inode backrefs, xattrs, extents,
3192 * so if there are a ton of hard links to an inode there can
3193 * be a lot of backrefs. Don't waste time searching too hard,
3194 * this is just an optimization
3199 /* we hit the end of the leaf before we found an xattr or
3200 * something larger than an xattr. We have to assume the inode
3203 if (*first_xattr_slot == -1)
3204 *first_xattr_slot = slot;
3209 * read an inode from the btree into the in-memory inode
3211 static int btrfs_read_locked_inode(struct inode *inode,
3212 struct btrfs_path *in_path)
3214 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3215 struct btrfs_path *path = in_path;
3216 struct extent_buffer *leaf;
3217 struct btrfs_inode_item *inode_item;
3218 struct btrfs_root *root = BTRFS_I(inode)->root;
3219 struct btrfs_key location;
3224 bool filled = false;
3225 int first_xattr_slot;
3227 ret = btrfs_fill_inode(inode, &rdev);
3232 path = btrfs_alloc_path();
3237 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3239 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3241 if (path != in_path)
3242 btrfs_free_path(path);
3246 leaf = path->nodes[0];
3251 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3252 struct btrfs_inode_item);
3253 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3254 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3255 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3256 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3257 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3258 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3259 round_up(i_size_read(inode), fs_info->sectorsize));
3261 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3262 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3264 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3265 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3267 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3268 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3270 BTRFS_I(inode)->i_otime.tv_sec =
3271 btrfs_timespec_sec(leaf, &inode_item->otime);
3272 BTRFS_I(inode)->i_otime.tv_nsec =
3273 btrfs_timespec_nsec(leaf, &inode_item->otime);
3275 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3276 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3277 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3279 inode_set_iversion_queried(inode,
3280 btrfs_inode_sequence(leaf, inode_item));
3281 inode->i_generation = BTRFS_I(inode)->generation;
3283 rdev = btrfs_inode_rdev(leaf, inode_item);
3285 BTRFS_I(inode)->index_cnt = (u64)-1;
3286 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3290 * If we were modified in the current generation and evicted from memory
3291 * and then re-read we need to do a full sync since we don't have any
3292 * idea about which extents were modified before we were evicted from
3295 * This is required for both inode re-read from disk and delayed inode
3296 * in delayed_nodes_tree.
3298 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3299 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3300 &BTRFS_I(inode)->runtime_flags);
3303 * We don't persist the id of the transaction where an unlink operation
3304 * against the inode was last made. So here we assume the inode might
3305 * have been evicted, and therefore the exact value of last_unlink_trans
3306 * lost, and set it to last_trans to avoid metadata inconsistencies
3307 * between the inode and its parent if the inode is fsync'ed and the log
3308 * replayed. For example, in the scenario:
3311 * ln mydir/foo mydir/bar
3314 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3315 * xfs_io -c fsync mydir/foo
3317 * mount fs, triggers fsync log replay
3319 * We must make sure that when we fsync our inode foo we also log its
3320 * parent inode, otherwise after log replay the parent still has the
3321 * dentry with the "bar" name but our inode foo has a link count of 1
3322 * and doesn't have an inode ref with the name "bar" anymore.
3324 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3325 * but it guarantees correctness at the expense of occasional full
3326 * transaction commits on fsync if our inode is a directory, or if our
3327 * inode is not a directory, logging its parent unnecessarily.
3329 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3332 * Same logic as for last_unlink_trans. We don't persist the generation
3333 * of the last transaction where this inode was used for a reflink
3334 * operation, so after eviction and reloading the inode we must be
3335 * pessimistic and assume the last transaction that modified the inode.
3337 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3340 if (inode->i_nlink != 1 ||
3341 path->slots[0] >= btrfs_header_nritems(leaf))
3344 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3345 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3348 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3349 if (location.type == BTRFS_INODE_REF_KEY) {
3350 struct btrfs_inode_ref *ref;
3352 ref = (struct btrfs_inode_ref *)ptr;
3353 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3354 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3355 struct btrfs_inode_extref *extref;
3357 extref = (struct btrfs_inode_extref *)ptr;
3358 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3363 * try to precache a NULL acl entry for files that don't have
3364 * any xattrs or acls
3366 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3367 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3368 if (first_xattr_slot != -1) {
3369 path->slots[0] = first_xattr_slot;
3370 ret = btrfs_load_inode_props(inode, path);
3373 "error loading props for ino %llu (root %llu): %d",
3374 btrfs_ino(BTRFS_I(inode)),
3375 root->root_key.objectid, ret);
3377 if (path != in_path)
3378 btrfs_free_path(path);
3381 cache_no_acl(inode);
3383 switch (inode->i_mode & S_IFMT) {
3385 inode->i_mapping->a_ops = &btrfs_aops;
3386 inode->i_fop = &btrfs_file_operations;
3387 inode->i_op = &btrfs_file_inode_operations;
3390 inode->i_fop = &btrfs_dir_file_operations;
3391 inode->i_op = &btrfs_dir_inode_operations;
3394 inode->i_op = &btrfs_symlink_inode_operations;
3395 inode_nohighmem(inode);
3396 inode->i_mapping->a_ops = &btrfs_aops;
3399 inode->i_op = &btrfs_special_inode_operations;
3400 init_special_inode(inode, inode->i_mode, rdev);
3404 btrfs_sync_inode_flags_to_i_flags(inode);
3409 * given a leaf and an inode, copy the inode fields into the leaf
3411 static void fill_inode_item(struct btrfs_trans_handle *trans,
3412 struct extent_buffer *leaf,
3413 struct btrfs_inode_item *item,
3414 struct inode *inode)
3416 struct btrfs_map_token token;
3418 btrfs_init_map_token(&token, leaf);
3420 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3421 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3422 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
3423 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3424 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3426 btrfs_set_token_timespec_sec(&token, &item->atime,
3427 inode->i_atime.tv_sec);
3428 btrfs_set_token_timespec_nsec(&token, &item->atime,
3429 inode->i_atime.tv_nsec);
3431 btrfs_set_token_timespec_sec(&token, &item->mtime,
3432 inode->i_mtime.tv_sec);
3433 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3434 inode->i_mtime.tv_nsec);
3436 btrfs_set_token_timespec_sec(&token, &item->ctime,
3437 inode->i_ctime.tv_sec);
3438 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3439 inode->i_ctime.tv_nsec);
3441 btrfs_set_token_timespec_sec(&token, &item->otime,
3442 BTRFS_I(inode)->i_otime.tv_sec);
3443 btrfs_set_token_timespec_nsec(&token, &item->otime,
3444 BTRFS_I(inode)->i_otime.tv_nsec);
3446 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
3447 btrfs_set_token_inode_generation(&token, item,
3448 BTRFS_I(inode)->generation);
3449 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
3450 btrfs_set_token_inode_transid(&token, item, trans->transid);
3451 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
3452 btrfs_set_token_inode_flags(&token, item, BTRFS_I(inode)->flags);
3453 btrfs_set_token_inode_block_group(&token, item, 0);
3457 * copy everything in the in-memory inode into the btree.
3459 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3460 struct btrfs_root *root, struct inode *inode)
3462 struct btrfs_inode_item *inode_item;
3463 struct btrfs_path *path;
3464 struct extent_buffer *leaf;
3467 path = btrfs_alloc_path();
3471 path->leave_spinning = 1;
3472 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3480 leaf = path->nodes[0];
3481 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3482 struct btrfs_inode_item);
3484 fill_inode_item(trans, leaf, inode_item, inode);
3485 btrfs_mark_buffer_dirty(leaf);
3486 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
3489 btrfs_free_path(path);
3494 * copy everything in the in-memory inode into the btree.
3496 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3497 struct btrfs_root *root, struct inode *inode)
3499 struct btrfs_fs_info *fs_info = root->fs_info;
3503 * If the inode is a free space inode, we can deadlock during commit
3504 * if we put it into the delayed code.
3506 * The data relocation inode should also be directly updated
3509 if (!btrfs_is_free_space_inode(BTRFS_I(inode))
3510 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3511 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
3512 btrfs_update_root_times(trans, root);
3514 ret = btrfs_delayed_update_inode(trans, root, inode);
3516 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
3520 return btrfs_update_inode_item(trans, root, inode);
3523 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3524 struct btrfs_root *root,
3525 struct inode *inode)
3529 ret = btrfs_update_inode(trans, root, inode);
3531 return btrfs_update_inode_item(trans, root, inode);
3536 * unlink helper that gets used here in inode.c and in the tree logging
3537 * recovery code. It remove a link in a directory with a given name, and
3538 * also drops the back refs in the inode to the directory
3540 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3541 struct btrfs_root *root,
3542 struct btrfs_inode *dir,
3543 struct btrfs_inode *inode,
3544 const char *name, int name_len)
3546 struct btrfs_fs_info *fs_info = root->fs_info;
3547 struct btrfs_path *path;
3549 struct btrfs_dir_item *di;
3551 u64 ino = btrfs_ino(inode);
3552 u64 dir_ino = btrfs_ino(dir);
3554 path = btrfs_alloc_path();
3560 path->leave_spinning = 1;
3561 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3562 name, name_len, -1);
3563 if (IS_ERR_OR_NULL(di)) {
3564 ret = di ? PTR_ERR(di) : -ENOENT;
3567 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3570 btrfs_release_path(path);
3573 * If we don't have dir index, we have to get it by looking up
3574 * the inode ref, since we get the inode ref, remove it directly,
3575 * it is unnecessary to do delayed deletion.
3577 * But if we have dir index, needn't search inode ref to get it.
3578 * Since the inode ref is close to the inode item, it is better
3579 * that we delay to delete it, and just do this deletion when
3580 * we update the inode item.
3582 if (inode->dir_index) {
3583 ret = btrfs_delayed_delete_inode_ref(inode);
3585 index = inode->dir_index;
3590 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3594 "failed to delete reference to %.*s, inode %llu parent %llu",
3595 name_len, name, ino, dir_ino);
3596 btrfs_abort_transaction(trans, ret);
3600 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
3602 btrfs_abort_transaction(trans, ret);
3606 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
3608 if (ret != 0 && ret != -ENOENT) {
3609 btrfs_abort_transaction(trans, ret);
3613 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
3618 btrfs_abort_transaction(trans, ret);
3621 * If we have a pending delayed iput we could end up with the final iput
3622 * being run in btrfs-cleaner context. If we have enough of these built
3623 * up we can end up burning a lot of time in btrfs-cleaner without any
3624 * way to throttle the unlinks. Since we're currently holding a ref on
3625 * the inode we can run the delayed iput here without any issues as the
3626 * final iput won't be done until after we drop the ref we're currently
3629 btrfs_run_delayed_iput(fs_info, inode);
3631 btrfs_free_path(path);
3635 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
3636 inode_inc_iversion(&inode->vfs_inode);
3637 inode_inc_iversion(&dir->vfs_inode);
3638 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
3639 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
3640 ret = btrfs_update_inode(trans, root, &dir->vfs_inode);
3645 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3646 struct btrfs_root *root,
3647 struct btrfs_inode *dir, struct btrfs_inode *inode,
3648 const char *name, int name_len)
3651 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3653 drop_nlink(&inode->vfs_inode);
3654 ret = btrfs_update_inode(trans, root, &inode->vfs_inode);
3660 * helper to start transaction for unlink and rmdir.
3662 * unlink and rmdir are special in btrfs, they do not always free space, so
3663 * if we cannot make our reservations the normal way try and see if there is
3664 * plenty of slack room in the global reserve to migrate, otherwise we cannot
3665 * allow the unlink to occur.
3667 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3669 struct btrfs_root *root = BTRFS_I(dir)->root;
3672 * 1 for the possible orphan item
3673 * 1 for the dir item
3674 * 1 for the dir index
3675 * 1 for the inode ref
3678 return btrfs_start_transaction_fallback_global_rsv(root, 5);
3681 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3683 struct btrfs_root *root = BTRFS_I(dir)->root;
3684 struct btrfs_trans_handle *trans;
3685 struct inode *inode = d_inode(dentry);
3688 trans = __unlink_start_trans(dir);
3690 return PTR_ERR(trans);
3692 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
3695 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
3696 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
3697 dentry->d_name.len);
3701 if (inode->i_nlink == 0) {
3702 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
3708 btrfs_end_transaction(trans);
3709 btrfs_btree_balance_dirty(root->fs_info);
3713 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3714 struct inode *dir, struct dentry *dentry)
3716 struct btrfs_root *root = BTRFS_I(dir)->root;
3717 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
3718 struct btrfs_path *path;
3719 struct extent_buffer *leaf;
3720 struct btrfs_dir_item *di;
3721 struct btrfs_key key;
3722 const char *name = dentry->d_name.name;
3723 int name_len = dentry->d_name.len;
3727 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
3729 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
3730 objectid = inode->root->root_key.objectid;
3731 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
3732 objectid = inode->location.objectid;
3738 path = btrfs_alloc_path();
3742 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3743 name, name_len, -1);
3744 if (IS_ERR_OR_NULL(di)) {
3745 ret = di ? PTR_ERR(di) : -ENOENT;
3749 leaf = path->nodes[0];
3750 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3751 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3752 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3754 btrfs_abort_transaction(trans, ret);
3757 btrfs_release_path(path);
3760 * This is a placeholder inode for a subvolume we didn't have a
3761 * reference to at the time of the snapshot creation. In the meantime
3762 * we could have renamed the real subvol link into our snapshot, so
3763 * depending on btrfs_del_root_ref to return -ENOENT here is incorret.
3764 * Instead simply lookup the dir_index_item for this entry so we can
3765 * remove it. Otherwise we know we have a ref to the root and we can
3766 * call btrfs_del_root_ref, and it _shouldn't_ fail.
3768 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
3769 di = btrfs_search_dir_index_item(root, path, dir_ino,
3771 if (IS_ERR_OR_NULL(di)) {
3776 btrfs_abort_transaction(trans, ret);
3780 leaf = path->nodes[0];
3781 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3783 btrfs_release_path(path);
3785 ret = btrfs_del_root_ref(trans, objectid,
3786 root->root_key.objectid, dir_ino,
3787 &index, name, name_len);
3789 btrfs_abort_transaction(trans, ret);
3794 ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index);
3796 btrfs_abort_transaction(trans, ret);
3800 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
3801 inode_inc_iversion(dir);
3802 dir->i_mtime = dir->i_ctime = current_time(dir);
3803 ret = btrfs_update_inode_fallback(trans, root, dir);
3805 btrfs_abort_transaction(trans, ret);
3807 btrfs_free_path(path);
3812 * Helper to check if the subvolume references other subvolumes or if it's
3815 static noinline int may_destroy_subvol(struct btrfs_root *root)
3817 struct btrfs_fs_info *fs_info = root->fs_info;
3818 struct btrfs_path *path;
3819 struct btrfs_dir_item *di;
3820 struct btrfs_key key;
3824 path = btrfs_alloc_path();
3828 /* Make sure this root isn't set as the default subvol */
3829 dir_id = btrfs_super_root_dir(fs_info->super_copy);
3830 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
3831 dir_id, "default", 7, 0);
3832 if (di && !IS_ERR(di)) {
3833 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
3834 if (key.objectid == root->root_key.objectid) {
3837 "deleting default subvolume %llu is not allowed",
3841 btrfs_release_path(path);
3844 key.objectid = root->root_key.objectid;
3845 key.type = BTRFS_ROOT_REF_KEY;
3846 key.offset = (u64)-1;
3848 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3854 if (path->slots[0] > 0) {
3856 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3857 if (key.objectid == root->root_key.objectid &&
3858 key.type == BTRFS_ROOT_REF_KEY)
3862 btrfs_free_path(path);
3866 /* Delete all dentries for inodes belonging to the root */
3867 static void btrfs_prune_dentries(struct btrfs_root *root)
3869 struct btrfs_fs_info *fs_info = root->fs_info;
3870 struct rb_node *node;
3871 struct rb_node *prev;
3872 struct btrfs_inode *entry;
3873 struct inode *inode;
3876 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3877 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3879 spin_lock(&root->inode_lock);
3881 node = root->inode_tree.rb_node;
3885 entry = rb_entry(node, struct btrfs_inode, rb_node);
3887 if (objectid < btrfs_ino(entry))
3888 node = node->rb_left;
3889 else if (objectid > btrfs_ino(entry))
3890 node = node->rb_right;
3896 entry = rb_entry(prev, struct btrfs_inode, rb_node);
3897 if (objectid <= btrfs_ino(entry)) {
3901 prev = rb_next(prev);
3905 entry = rb_entry(node, struct btrfs_inode, rb_node);
3906 objectid = btrfs_ino(entry) + 1;
3907 inode = igrab(&entry->vfs_inode);
3909 spin_unlock(&root->inode_lock);
3910 if (atomic_read(&inode->i_count) > 1)
3911 d_prune_aliases(inode);
3913 * btrfs_drop_inode will have it removed from the inode
3914 * cache when its usage count hits zero.
3918 spin_lock(&root->inode_lock);
3922 if (cond_resched_lock(&root->inode_lock))
3925 node = rb_next(node);
3927 spin_unlock(&root->inode_lock);
3930 int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
3932 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
3933 struct btrfs_root *root = BTRFS_I(dir)->root;
3934 struct inode *inode = d_inode(dentry);
3935 struct btrfs_root *dest = BTRFS_I(inode)->root;
3936 struct btrfs_trans_handle *trans;
3937 struct btrfs_block_rsv block_rsv;
3943 * Don't allow to delete a subvolume with send in progress. This is
3944 * inside the inode lock so the error handling that has to drop the bit
3945 * again is not run concurrently.
3947 spin_lock(&dest->root_item_lock);
3948 if (dest->send_in_progress) {
3949 spin_unlock(&dest->root_item_lock);
3951 "attempt to delete subvolume %llu during send",
3952 dest->root_key.objectid);
3955 root_flags = btrfs_root_flags(&dest->root_item);
3956 btrfs_set_root_flags(&dest->root_item,
3957 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
3958 spin_unlock(&dest->root_item_lock);
3960 down_write(&fs_info->subvol_sem);
3962 err = may_destroy_subvol(dest);
3966 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
3968 * One for dir inode,
3969 * two for dir entries,
3970 * two for root ref/backref.
3972 err = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
3976 trans = btrfs_start_transaction(root, 0);
3977 if (IS_ERR(trans)) {
3978 err = PTR_ERR(trans);
3981 trans->block_rsv = &block_rsv;
3982 trans->bytes_reserved = block_rsv.size;
3984 btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
3986 ret = btrfs_unlink_subvol(trans, dir, dentry);
3989 btrfs_abort_transaction(trans, ret);
3993 btrfs_record_root_in_trans(trans, dest);
3995 memset(&dest->root_item.drop_progress, 0,
3996 sizeof(dest->root_item.drop_progress));
3997 dest->root_item.drop_level = 0;
3998 btrfs_set_root_refs(&dest->root_item, 0);
4000 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4001 ret = btrfs_insert_orphan_item(trans,
4003 dest->root_key.objectid);
4005 btrfs_abort_transaction(trans, ret);
4011 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4012 BTRFS_UUID_KEY_SUBVOL,
4013 dest->root_key.objectid);
4014 if (ret && ret != -ENOENT) {
4015 btrfs_abort_transaction(trans, ret);
4019 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4020 ret = btrfs_uuid_tree_remove(trans,
4021 dest->root_item.received_uuid,
4022 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4023 dest->root_key.objectid);
4024 if (ret && ret != -ENOENT) {
4025 btrfs_abort_transaction(trans, ret);
4031 free_anon_bdev(dest->anon_dev);
4034 trans->block_rsv = NULL;
4035 trans->bytes_reserved = 0;
4036 ret = btrfs_end_transaction(trans);
4039 inode->i_flags |= S_DEAD;
4041 btrfs_subvolume_release_metadata(root, &block_rsv);
4043 up_write(&fs_info->subvol_sem);
4045 spin_lock(&dest->root_item_lock);
4046 root_flags = btrfs_root_flags(&dest->root_item);
4047 btrfs_set_root_flags(&dest->root_item,
4048 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4049 spin_unlock(&dest->root_item_lock);
4051 d_invalidate(dentry);
4052 btrfs_prune_dentries(dest);
4053 ASSERT(dest->send_in_progress == 0);
4056 if (dest->ino_cache_inode) {
4057 iput(dest->ino_cache_inode);
4058 dest->ino_cache_inode = NULL;
4065 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4067 struct inode *inode = d_inode(dentry);
4069 struct btrfs_root *root = BTRFS_I(dir)->root;
4070 struct btrfs_trans_handle *trans;
4071 u64 last_unlink_trans;
4073 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4075 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4076 return btrfs_delete_subvolume(dir, dentry);
4078 trans = __unlink_start_trans(dir);
4080 return PTR_ERR(trans);
4082 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4083 err = btrfs_unlink_subvol(trans, dir, dentry);
4087 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4091 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4093 /* now the directory is empty */
4094 err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4095 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4096 dentry->d_name.len);
4098 btrfs_i_size_write(BTRFS_I(inode), 0);
4100 * Propagate the last_unlink_trans value of the deleted dir to
4101 * its parent directory. This is to prevent an unrecoverable
4102 * log tree in the case we do something like this:
4104 * 2) create snapshot under dir foo
4105 * 3) delete the snapshot
4108 * 6) fsync foo or some file inside foo
4110 if (last_unlink_trans >= trans->transid)
4111 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4114 btrfs_end_transaction(trans);
4115 btrfs_btree_balance_dirty(root->fs_info);
4121 * Return this if we need to call truncate_block for the last bit of the
4124 #define NEED_TRUNCATE_BLOCK 1
4127 * this can truncate away extent items, csum items and directory items.
4128 * It starts at a high offset and removes keys until it can't find
4129 * any higher than new_size
4131 * csum items that cross the new i_size are truncated to the new size
4134 * min_type is the minimum key type to truncate down to. If set to 0, this
4135 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4137 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4138 struct btrfs_root *root,
4139 struct inode *inode,
4140 u64 new_size, u32 min_type)
4142 struct btrfs_fs_info *fs_info = root->fs_info;
4143 struct btrfs_path *path;
4144 struct extent_buffer *leaf;
4145 struct btrfs_file_extent_item *fi;
4146 struct btrfs_key key;
4147 struct btrfs_key found_key;
4148 u64 extent_start = 0;
4149 u64 extent_num_bytes = 0;
4150 u64 extent_offset = 0;
4152 u64 last_size = new_size;
4153 u32 found_type = (u8)-1;
4156 int pending_del_nr = 0;
4157 int pending_del_slot = 0;
4158 int extent_type = -1;
4160 u64 ino = btrfs_ino(BTRFS_I(inode));
4161 u64 bytes_deleted = 0;
4162 bool be_nice = false;
4163 bool should_throttle = false;
4164 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
4165 struct extent_state *cached_state = NULL;
4167 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4170 * For non-free space inodes and non-shareable roots, we want to back
4171 * off from time to time. This means all inodes in subvolume roots,
4172 * reloc roots, and data reloc roots.
4174 if (!btrfs_is_free_space_inode(BTRFS_I(inode)) &&
4175 test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
4178 path = btrfs_alloc_path();
4181 path->reada = READA_BACK;
4183 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4184 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, (u64)-1,
4188 * We want to drop from the next block forward in case this
4189 * new size is not block aligned since we will be keeping the
4190 * last block of the extent just the way it is.
4192 btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size,
4193 fs_info->sectorsize),
4198 * This function is also used to drop the items in the log tree before
4199 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4200 * it is used to drop the logged items. So we shouldn't kill the delayed
4203 if (min_type == 0 && root == BTRFS_I(inode)->root)
4204 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
4207 key.offset = (u64)-1;
4212 * with a 16K leaf size and 128MB extents, you can actually queue
4213 * up a huge file in a single leaf. Most of the time that
4214 * bytes_deleted is > 0, it will be huge by the time we get here
4216 if (be_nice && bytes_deleted > SZ_32M &&
4217 btrfs_should_end_transaction(trans)) {
4222 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4228 /* there are no items in the tree for us to truncate, we're
4231 if (path->slots[0] == 0)
4237 u64 clear_start = 0, clear_len = 0;
4240 leaf = path->nodes[0];
4241 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4242 found_type = found_key.type;
4244 if (found_key.objectid != ino)
4247 if (found_type < min_type)
4250 item_end = found_key.offset;
4251 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4252 fi = btrfs_item_ptr(leaf, path->slots[0],
4253 struct btrfs_file_extent_item);
4254 extent_type = btrfs_file_extent_type(leaf, fi);
4255 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4257 btrfs_file_extent_num_bytes(leaf, fi);
4259 trace_btrfs_truncate_show_fi_regular(
4260 BTRFS_I(inode), leaf, fi,
4262 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4263 item_end += btrfs_file_extent_ram_bytes(leaf,
4266 trace_btrfs_truncate_show_fi_inline(
4267 BTRFS_I(inode), leaf, fi, path->slots[0],
4272 if (found_type > min_type) {
4275 if (item_end < new_size)
4277 if (found_key.offset >= new_size)
4283 /* FIXME, shrink the extent if the ref count is only 1 */
4284 if (found_type != BTRFS_EXTENT_DATA_KEY)
4287 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4290 clear_start = found_key.offset;
4291 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4293 u64 orig_num_bytes =
4294 btrfs_file_extent_num_bytes(leaf, fi);
4295 extent_num_bytes = ALIGN(new_size -
4297 fs_info->sectorsize);
4298 clear_start = ALIGN(new_size, fs_info->sectorsize);
4299 btrfs_set_file_extent_num_bytes(leaf, fi,
4301 num_dec = (orig_num_bytes -
4303 if (test_bit(BTRFS_ROOT_SHAREABLE,
4306 inode_sub_bytes(inode, num_dec);
4307 btrfs_mark_buffer_dirty(leaf);
4310 btrfs_file_extent_disk_num_bytes(leaf,
4312 extent_offset = found_key.offset -
4313 btrfs_file_extent_offset(leaf, fi);
4315 /* FIXME blocksize != 4096 */
4316 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4317 if (extent_start != 0) {
4319 if (test_bit(BTRFS_ROOT_SHAREABLE,
4321 inode_sub_bytes(inode, num_dec);
4324 clear_len = num_dec;
4325 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4327 * we can't truncate inline items that have had
4331 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4332 btrfs_file_extent_other_encoding(leaf, fi) == 0 &&
4333 btrfs_file_extent_compression(leaf, fi) == 0) {
4334 u32 size = (u32)(new_size - found_key.offset);
4336 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4337 size = btrfs_file_extent_calc_inline_size(size);
4338 btrfs_truncate_item(path, size, 1);
4339 } else if (!del_item) {
4341 * We have to bail so the last_size is set to
4342 * just before this extent.
4344 ret = NEED_TRUNCATE_BLOCK;
4348 * Inline extents are special, we just treat
4349 * them as a full sector worth in the file
4350 * extent tree just for simplicity sake.
4352 clear_len = fs_info->sectorsize;
4355 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
4356 inode_sub_bytes(inode, item_end + 1 - new_size);
4360 * We use btrfs_truncate_inode_items() to clean up log trees for
4361 * multiple fsyncs, and in this case we don't want to clear the
4362 * file extent range because it's just the log.
4364 if (root == BTRFS_I(inode)->root) {
4365 ret = btrfs_inode_clear_file_extent_range(BTRFS_I(inode),
4366 clear_start, clear_len);
4368 btrfs_abort_transaction(trans, ret);
4374 last_size = found_key.offset;
4376 last_size = new_size;
4378 if (!pending_del_nr) {
4379 /* no pending yet, add ourselves */
4380 pending_del_slot = path->slots[0];
4382 } else if (pending_del_nr &&
4383 path->slots[0] + 1 == pending_del_slot) {
4384 /* hop on the pending chunk */
4386 pending_del_slot = path->slots[0];
4393 should_throttle = false;
4396 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4397 struct btrfs_ref ref = { 0 };
4399 bytes_deleted += extent_num_bytes;
4401 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF,
4402 extent_start, extent_num_bytes, 0);
4403 ref.real_root = root->root_key.objectid;
4404 btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
4405 ino, extent_offset);
4406 ret = btrfs_free_extent(trans, &ref);
4408 btrfs_abort_transaction(trans, ret);
4412 if (btrfs_should_throttle_delayed_refs(trans))
4413 should_throttle = true;
4417 if (found_type == BTRFS_INODE_ITEM_KEY)
4420 if (path->slots[0] == 0 ||
4421 path->slots[0] != pending_del_slot ||
4423 if (pending_del_nr) {
4424 ret = btrfs_del_items(trans, root, path,
4428 btrfs_abort_transaction(trans, ret);
4433 btrfs_release_path(path);
4436 * We can generate a lot of delayed refs, so we need to
4437 * throttle every once and a while and make sure we're
4438 * adding enough space to keep up with the work we are
4439 * generating. Since we hold a transaction here we
4440 * can't flush, and we don't want to FLUSH_LIMIT because
4441 * we could have generated too many delayed refs to
4442 * actually allocate, so just bail if we're short and
4443 * let the normal reservation dance happen higher up.
4445 if (should_throttle) {
4446 ret = btrfs_delayed_refs_rsv_refill(fs_info,
4447 BTRFS_RESERVE_NO_FLUSH);
4459 if (ret >= 0 && pending_del_nr) {
4462 err = btrfs_del_items(trans, root, path, pending_del_slot,
4465 btrfs_abort_transaction(trans, err);
4469 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4470 ASSERT(last_size >= new_size);
4471 if (!ret && last_size > new_size)
4472 last_size = new_size;
4473 btrfs_inode_safe_disk_i_size_write(inode, last_size);
4474 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start,
4475 (u64)-1, &cached_state);
4478 btrfs_free_path(path);
4483 * btrfs_truncate_block - read, zero a chunk and write a block
4484 * @inode - inode that we're zeroing
4485 * @from - the offset to start zeroing
4486 * @len - the length to zero, 0 to zero the entire range respective to the
4488 * @front - zero up to the offset instead of from the offset on
4490 * This will find the block for the "from" offset and cow the block and zero the
4491 * part we want to zero. This is used with truncate and hole punching.
4493 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4496 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4497 struct address_space *mapping = inode->i_mapping;
4498 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4499 struct btrfs_ordered_extent *ordered;
4500 struct extent_state *cached_state = NULL;
4501 struct extent_changeset *data_reserved = NULL;
4503 bool only_release_metadata = false;
4504 u32 blocksize = fs_info->sectorsize;
4505 pgoff_t index = from >> PAGE_SHIFT;
4506 unsigned offset = from & (blocksize - 1);
4508 gfp_t mask = btrfs_alloc_write_mask(mapping);
4509 size_t write_bytes = blocksize;
4514 if (IS_ALIGNED(offset, blocksize) &&
4515 (!len || IS_ALIGNED(len, blocksize)))
4518 block_start = round_down(from, blocksize);
4519 block_end = block_start + blocksize - 1;
4521 ret = btrfs_check_data_free_space(BTRFS_I(inode), &data_reserved,
4522 block_start, blocksize);
4524 if (btrfs_check_nocow_lock(BTRFS_I(inode), block_start,
4525 &write_bytes) > 0) {
4526 /* For nocow case, no need to reserve data space */
4527 only_release_metadata = true;
4532 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), blocksize);
4534 if (!only_release_metadata)
4535 btrfs_free_reserved_data_space(BTRFS_I(inode),
4536 data_reserved, block_start, blocksize);
4540 page = find_or_create_page(mapping, index, mask);
4542 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved,
4543 block_start, blocksize, true);
4544 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize);
4549 if (!PageUptodate(page)) {
4550 ret = btrfs_readpage(NULL, page);
4552 if (page->mapping != mapping) {
4557 if (!PageUptodate(page)) {
4562 wait_on_page_writeback(page);
4564 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4565 set_page_extent_mapped(page);
4567 ordered = btrfs_lookup_ordered_extent(BTRFS_I(inode), block_start);
4569 unlock_extent_cached(io_tree, block_start, block_end,
4573 btrfs_start_ordered_extent(ordered, 1);
4574 btrfs_put_ordered_extent(ordered);
4578 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4579 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4580 0, 0, &cached_state);
4582 ret = btrfs_set_extent_delalloc(BTRFS_I(inode), block_start, block_end, 0,
4585 unlock_extent_cached(io_tree, block_start, block_end,
4590 if (offset != blocksize) {
4592 len = blocksize - offset;
4595 memset(kaddr + (block_start - page_offset(page)),
4598 memset(kaddr + (block_start - page_offset(page)) + offset,
4600 flush_dcache_page(page);
4603 ClearPageChecked(page);
4604 set_page_dirty(page);
4605 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4607 if (only_release_metadata)
4608 set_extent_bit(&BTRFS_I(inode)->io_tree, block_start,
4609 block_end, EXTENT_NORESERVE, NULL, NULL,
4614 if (only_release_metadata)
4615 btrfs_delalloc_release_metadata(BTRFS_I(inode),
4618 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved,
4619 block_start, blocksize, true);
4621 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize);
4625 if (only_release_metadata)
4626 btrfs_check_nocow_unlock(BTRFS_I(inode));
4627 extent_changeset_free(data_reserved);
4631 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4632 u64 offset, u64 len)
4634 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4635 struct btrfs_trans_handle *trans;
4639 * Still need to make sure the inode looks like it's been updated so
4640 * that any holes get logged if we fsync.
4642 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4643 BTRFS_I(inode)->last_trans = fs_info->generation;
4644 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4645 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4650 * 1 - for the one we're dropping
4651 * 1 - for the one we're adding
4652 * 1 - for updating the inode.
4654 trans = btrfs_start_transaction(root, 3);
4656 return PTR_ERR(trans);
4658 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4660 btrfs_abort_transaction(trans, ret);
4661 btrfs_end_transaction(trans);
4665 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)),
4666 offset, 0, 0, len, 0, len, 0, 0, 0);
4668 btrfs_abort_transaction(trans, ret);
4670 btrfs_update_inode(trans, root, inode);
4671 btrfs_end_transaction(trans);
4676 * This function puts in dummy file extents for the area we're creating a hole
4677 * for. So if we are truncating this file to a larger size we need to insert
4678 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4679 * the range between oldsize and size
4681 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4683 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4684 struct btrfs_root *root = BTRFS_I(inode)->root;
4685 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4686 struct extent_map *em = NULL;
4687 struct extent_state *cached_state = NULL;
4688 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4689 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4690 u64 block_end = ALIGN(size, fs_info->sectorsize);
4697 * If our size started in the middle of a block we need to zero out the
4698 * rest of the block before we expand the i_size, otherwise we could
4699 * expose stale data.
4701 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4705 if (size <= hole_start)
4708 btrfs_lock_and_flush_ordered_range(BTRFS_I(inode), hole_start,
4709 block_end - 1, &cached_state);
4710 cur_offset = hole_start;
4712 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
4713 block_end - cur_offset);
4719 last_byte = min(extent_map_end(em), block_end);
4720 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4721 hole_size = last_byte - cur_offset;
4723 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4724 struct extent_map *hole_em;
4726 err = maybe_insert_hole(root, inode, cur_offset,
4731 err = btrfs_inode_set_file_extent_range(BTRFS_I(inode),
4732 cur_offset, hole_size);
4736 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
4737 cur_offset + hole_size - 1, 0);
4738 hole_em = alloc_extent_map();
4740 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4741 &BTRFS_I(inode)->runtime_flags);
4744 hole_em->start = cur_offset;
4745 hole_em->len = hole_size;
4746 hole_em->orig_start = cur_offset;
4748 hole_em->block_start = EXTENT_MAP_HOLE;
4749 hole_em->block_len = 0;
4750 hole_em->orig_block_len = 0;
4751 hole_em->ram_bytes = hole_size;
4752 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4753 hole_em->generation = fs_info->generation;
4756 write_lock(&em_tree->lock);
4757 err = add_extent_mapping(em_tree, hole_em, 1);
4758 write_unlock(&em_tree->lock);
4761 btrfs_drop_extent_cache(BTRFS_I(inode),
4766 free_extent_map(hole_em);
4768 err = btrfs_inode_set_file_extent_range(BTRFS_I(inode),
4769 cur_offset, hole_size);
4774 free_extent_map(em);
4776 cur_offset = last_byte;
4777 if (cur_offset >= block_end)
4780 free_extent_map(em);
4781 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
4785 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4787 struct btrfs_root *root = BTRFS_I(inode)->root;
4788 struct btrfs_trans_handle *trans;
4789 loff_t oldsize = i_size_read(inode);
4790 loff_t newsize = attr->ia_size;
4791 int mask = attr->ia_valid;
4795 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4796 * special case where we need to update the times despite not having
4797 * these flags set. For all other operations the VFS set these flags
4798 * explicitly if it wants a timestamp update.
4800 if (newsize != oldsize) {
4801 inode_inc_iversion(inode);
4802 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4803 inode->i_ctime = inode->i_mtime =
4804 current_time(inode);
4807 if (newsize > oldsize) {
4809 * Don't do an expanding truncate while snapshotting is ongoing.
4810 * This is to ensure the snapshot captures a fully consistent
4811 * state of this file - if the snapshot captures this expanding
4812 * truncation, it must capture all writes that happened before
4815 btrfs_drew_write_lock(&root->snapshot_lock);
4816 ret = btrfs_cont_expand(inode, oldsize, newsize);
4818 btrfs_drew_write_unlock(&root->snapshot_lock);
4822 trans = btrfs_start_transaction(root, 1);
4823 if (IS_ERR(trans)) {
4824 btrfs_drew_write_unlock(&root->snapshot_lock);
4825 return PTR_ERR(trans);
4828 i_size_write(inode, newsize);
4829 btrfs_inode_safe_disk_i_size_write(inode, 0);
4830 pagecache_isize_extended(inode, oldsize, newsize);
4831 ret = btrfs_update_inode(trans, root, inode);
4832 btrfs_drew_write_unlock(&root->snapshot_lock);
4833 btrfs_end_transaction(trans);
4837 * We're truncating a file that used to have good data down to
4838 * zero. Make sure any new writes to the file get on disk
4842 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
4843 &BTRFS_I(inode)->runtime_flags);
4845 truncate_setsize(inode, newsize);
4847 inode_dio_wait(inode);
4849 ret = btrfs_truncate(inode, newsize == oldsize);
4850 if (ret && inode->i_nlink) {
4854 * Truncate failed, so fix up the in-memory size. We
4855 * adjusted disk_i_size down as we removed extents, so
4856 * wait for disk_i_size to be stable and then update the
4857 * in-memory size to match.
4859 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
4862 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4869 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4871 struct inode *inode = d_inode(dentry);
4872 struct btrfs_root *root = BTRFS_I(inode)->root;
4875 if (btrfs_root_readonly(root))
4878 err = setattr_prepare(dentry, attr);
4882 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4883 err = btrfs_setsize(inode, attr);
4888 if (attr->ia_valid) {
4889 setattr_copy(inode, attr);
4890 inode_inc_iversion(inode);
4891 err = btrfs_dirty_inode(inode);
4893 if (!err && attr->ia_valid & ATTR_MODE)
4894 err = posix_acl_chmod(inode, inode->i_mode);
4901 * While truncating the inode pages during eviction, we get the VFS calling
4902 * btrfs_invalidatepage() against each page of the inode. This is slow because
4903 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4904 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4905 * extent_state structures over and over, wasting lots of time.
4907 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4908 * those expensive operations on a per page basis and do only the ordered io
4909 * finishing, while we release here the extent_map and extent_state structures,
4910 * without the excessive merging and splitting.
4912 static void evict_inode_truncate_pages(struct inode *inode)
4914 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4915 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
4916 struct rb_node *node;
4918 ASSERT(inode->i_state & I_FREEING);
4919 truncate_inode_pages_final(&inode->i_data);
4921 write_lock(&map_tree->lock);
4922 while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) {
4923 struct extent_map *em;
4925 node = rb_first_cached(&map_tree->map);
4926 em = rb_entry(node, struct extent_map, rb_node);
4927 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
4928 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
4929 remove_extent_mapping(map_tree, em);
4930 free_extent_map(em);
4931 if (need_resched()) {
4932 write_unlock(&map_tree->lock);
4934 write_lock(&map_tree->lock);
4937 write_unlock(&map_tree->lock);
4940 * Keep looping until we have no more ranges in the io tree.
4941 * We can have ongoing bios started by readahead that have
4942 * their endio callback (extent_io.c:end_bio_extent_readpage)
4943 * still in progress (unlocked the pages in the bio but did not yet
4944 * unlocked the ranges in the io tree). Therefore this means some
4945 * ranges can still be locked and eviction started because before
4946 * submitting those bios, which are executed by a separate task (work
4947 * queue kthread), inode references (inode->i_count) were not taken
4948 * (which would be dropped in the end io callback of each bio).
4949 * Therefore here we effectively end up waiting for those bios and
4950 * anyone else holding locked ranges without having bumped the inode's
4951 * reference count - if we don't do it, when they access the inode's
4952 * io_tree to unlock a range it may be too late, leading to an
4953 * use-after-free issue.
4955 spin_lock(&io_tree->lock);
4956 while (!RB_EMPTY_ROOT(&io_tree->state)) {
4957 struct extent_state *state;
4958 struct extent_state *cached_state = NULL;
4961 unsigned state_flags;
4963 node = rb_first(&io_tree->state);
4964 state = rb_entry(node, struct extent_state, rb_node);
4965 start = state->start;
4967 state_flags = state->state;
4968 spin_unlock(&io_tree->lock);
4970 lock_extent_bits(io_tree, start, end, &cached_state);
4973 * If still has DELALLOC flag, the extent didn't reach disk,
4974 * and its reserved space won't be freed by delayed_ref.
4975 * So we need to free its reserved space here.
4976 * (Refer to comment in btrfs_invalidatepage, case 2)
4978 * Note, end is the bytenr of last byte, so we need + 1 here.
4980 if (state_flags & EXTENT_DELALLOC)
4981 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
4984 clear_extent_bit(io_tree, start, end,
4985 EXTENT_LOCKED | EXTENT_DELALLOC |
4986 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
4990 spin_lock(&io_tree->lock);
4992 spin_unlock(&io_tree->lock);
4995 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
4996 struct btrfs_block_rsv *rsv)
4998 struct btrfs_fs_info *fs_info = root->fs_info;
4999 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5000 struct btrfs_trans_handle *trans;
5001 u64 delayed_refs_extra = btrfs_calc_insert_metadata_size(fs_info, 1);
5005 * Eviction should be taking place at some place safe because of our
5006 * delayed iputs. However the normal flushing code will run delayed
5007 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5009 * We reserve the delayed_refs_extra here again because we can't use
5010 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5011 * above. We reserve our extra bit here because we generate a ton of
5012 * delayed refs activity by truncating.
5014 * If we cannot make our reservation we'll attempt to steal from the
5015 * global reserve, because we really want to be able to free up space.
5017 ret = btrfs_block_rsv_refill(root, rsv, rsv->size + delayed_refs_extra,
5018 BTRFS_RESERVE_FLUSH_EVICT);
5021 * Try to steal from the global reserve if there is space for
5024 if (btrfs_check_space_for_delayed_refs(fs_info) ||
5025 btrfs_block_rsv_migrate(global_rsv, rsv, rsv->size, 0)) {
5027 "could not allocate space for delete; will truncate on mount");
5028 return ERR_PTR(-ENOSPC);
5030 delayed_refs_extra = 0;
5033 trans = btrfs_join_transaction(root);
5037 if (delayed_refs_extra) {
5038 trans->block_rsv = &fs_info->trans_block_rsv;
5039 trans->bytes_reserved = delayed_refs_extra;
5040 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5041 delayed_refs_extra, 1);
5046 void btrfs_evict_inode(struct inode *inode)
5048 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5049 struct btrfs_trans_handle *trans;
5050 struct btrfs_root *root = BTRFS_I(inode)->root;
5051 struct btrfs_block_rsv *rsv;
5054 trace_btrfs_inode_evict(inode);
5061 evict_inode_truncate_pages(inode);
5063 if (inode->i_nlink &&
5064 ((btrfs_root_refs(&root->root_item) != 0 &&
5065 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5066 btrfs_is_free_space_inode(BTRFS_I(inode))))
5069 if (is_bad_inode(inode))
5072 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5074 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5077 if (inode->i_nlink > 0) {
5078 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5079 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5083 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5087 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5090 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5093 btrfs_i_size_write(BTRFS_I(inode), 0);
5096 trans = evict_refill_and_join(root, rsv);
5100 trans->block_rsv = rsv;
5102 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5103 trans->block_rsv = &fs_info->trans_block_rsv;
5104 btrfs_end_transaction(trans);
5105 btrfs_btree_balance_dirty(fs_info);
5106 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5113 * Errors here aren't a big deal, it just means we leave orphan items in
5114 * the tree. They will be cleaned up on the next mount. If the inode
5115 * number gets reused, cleanup deletes the orphan item without doing
5116 * anything, and unlink reuses the existing orphan item.
5118 * If it turns out that we are dropping too many of these, we might want
5119 * to add a mechanism for retrying these after a commit.
5121 trans = evict_refill_and_join(root, rsv);
5122 if (!IS_ERR(trans)) {
5123 trans->block_rsv = rsv;
5124 btrfs_orphan_del(trans, BTRFS_I(inode));
5125 trans->block_rsv = &fs_info->trans_block_rsv;
5126 btrfs_end_transaction(trans);
5129 if (!(root == fs_info->tree_root ||
5130 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5131 btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
5134 btrfs_free_block_rsv(fs_info, rsv);
5137 * If we didn't successfully delete, the orphan item will still be in
5138 * the tree and we'll retry on the next mount. Again, we might also want
5139 * to retry these periodically in the future.
5141 btrfs_remove_delayed_node(BTRFS_I(inode));
5146 * Return the key found in the dir entry in the location pointer, fill @type
5147 * with BTRFS_FT_*, and return 0.
5149 * If no dir entries were found, returns -ENOENT.
5150 * If found a corrupted location in dir entry, returns -EUCLEAN.
5152 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5153 struct btrfs_key *location, u8 *type)
5155 const char *name = dentry->d_name.name;
5156 int namelen = dentry->d_name.len;
5157 struct btrfs_dir_item *di;
5158 struct btrfs_path *path;
5159 struct btrfs_root *root = BTRFS_I(dir)->root;
5162 path = btrfs_alloc_path();
5166 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5168 if (IS_ERR_OR_NULL(di)) {
5169 ret = di ? PTR_ERR(di) : -ENOENT;
5173 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5174 if (location->type != BTRFS_INODE_ITEM_KEY &&
5175 location->type != BTRFS_ROOT_ITEM_KEY) {
5177 btrfs_warn(root->fs_info,
5178 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5179 __func__, name, btrfs_ino(BTRFS_I(dir)),
5180 location->objectid, location->type, location->offset);
5183 *type = btrfs_dir_type(path->nodes[0], di);
5185 btrfs_free_path(path);
5190 * when we hit a tree root in a directory, the btrfs part of the inode
5191 * needs to be changed to reflect the root directory of the tree root. This
5192 * is kind of like crossing a mount point.
5194 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5196 struct dentry *dentry,
5197 struct btrfs_key *location,
5198 struct btrfs_root **sub_root)
5200 struct btrfs_path *path;
5201 struct btrfs_root *new_root;
5202 struct btrfs_root_ref *ref;
5203 struct extent_buffer *leaf;
5204 struct btrfs_key key;
5208 path = btrfs_alloc_path();
5215 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5216 key.type = BTRFS_ROOT_REF_KEY;
5217 key.offset = location->objectid;
5219 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5226 leaf = path->nodes[0];
5227 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5228 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5229 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5232 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5233 (unsigned long)(ref + 1),
5234 dentry->d_name.len);
5238 btrfs_release_path(path);
5240 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5241 if (IS_ERR(new_root)) {
5242 err = PTR_ERR(new_root);
5246 *sub_root = new_root;
5247 location->objectid = btrfs_root_dirid(&new_root->root_item);
5248 location->type = BTRFS_INODE_ITEM_KEY;
5249 location->offset = 0;
5252 btrfs_free_path(path);
5256 static void inode_tree_add(struct inode *inode)
5258 struct btrfs_root *root = BTRFS_I(inode)->root;
5259 struct btrfs_inode *entry;
5261 struct rb_node *parent;
5262 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5263 u64 ino = btrfs_ino(BTRFS_I(inode));
5265 if (inode_unhashed(inode))
5268 spin_lock(&root->inode_lock);
5269 p = &root->inode_tree.rb_node;
5272 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5274 if (ino < btrfs_ino(entry))
5275 p = &parent->rb_left;
5276 else if (ino > btrfs_ino(entry))
5277 p = &parent->rb_right;
5279 WARN_ON(!(entry->vfs_inode.i_state &
5280 (I_WILL_FREE | I_FREEING)));
5281 rb_replace_node(parent, new, &root->inode_tree);
5282 RB_CLEAR_NODE(parent);
5283 spin_unlock(&root->inode_lock);
5287 rb_link_node(new, parent, p);
5288 rb_insert_color(new, &root->inode_tree);
5289 spin_unlock(&root->inode_lock);
5292 static void inode_tree_del(struct btrfs_inode *inode)
5294 struct btrfs_root *root = inode->root;
5297 spin_lock(&root->inode_lock);
5298 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5299 rb_erase(&inode->rb_node, &root->inode_tree);
5300 RB_CLEAR_NODE(&inode->rb_node);
5301 empty = RB_EMPTY_ROOT(&root->inode_tree);
5303 spin_unlock(&root->inode_lock);
5305 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5306 spin_lock(&root->inode_lock);
5307 empty = RB_EMPTY_ROOT(&root->inode_tree);
5308 spin_unlock(&root->inode_lock);
5310 btrfs_add_dead_root(root);
5315 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5317 struct btrfs_iget_args *args = p;
5319 inode->i_ino = args->ino;
5320 BTRFS_I(inode)->location.objectid = args->ino;
5321 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5322 BTRFS_I(inode)->location.offset = 0;
5323 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5324 BUG_ON(args->root && !BTRFS_I(inode)->root);
5328 static int btrfs_find_actor(struct inode *inode, void *opaque)
5330 struct btrfs_iget_args *args = opaque;
5332 return args->ino == BTRFS_I(inode)->location.objectid &&
5333 args->root == BTRFS_I(inode)->root;
5336 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5337 struct btrfs_root *root)
5339 struct inode *inode;
5340 struct btrfs_iget_args args;
5341 unsigned long hashval = btrfs_inode_hash(ino, root);
5346 inode = iget5_locked(s, hashval, btrfs_find_actor,
5347 btrfs_init_locked_inode,
5353 * Get an inode object given its inode number and corresponding root.
5354 * Path can be preallocated to prevent recursing back to iget through
5355 * allocator. NULL is also valid but may require an additional allocation
5358 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5359 struct btrfs_root *root, struct btrfs_path *path)
5361 struct inode *inode;
5363 inode = btrfs_iget_locked(s, ino, root);
5365 return ERR_PTR(-ENOMEM);
5367 if (inode->i_state & I_NEW) {
5370 ret = btrfs_read_locked_inode(inode, path);
5372 inode_tree_add(inode);
5373 unlock_new_inode(inode);
5377 * ret > 0 can come from btrfs_search_slot called by
5378 * btrfs_read_locked_inode, this means the inode item
5383 inode = ERR_PTR(ret);
5390 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5392 return btrfs_iget_path(s, ino, root, NULL);
5395 static struct inode *new_simple_dir(struct super_block *s,
5396 struct btrfs_key *key,
5397 struct btrfs_root *root)
5399 struct inode *inode = new_inode(s);
5402 return ERR_PTR(-ENOMEM);
5404 BTRFS_I(inode)->root = btrfs_grab_root(root);
5405 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5406 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5408 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5410 * We only need lookup, the rest is read-only and there's no inode
5411 * associated with the dentry
5413 inode->i_op = &simple_dir_inode_operations;
5414 inode->i_opflags &= ~IOP_XATTR;
5415 inode->i_fop = &simple_dir_operations;
5416 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5417 inode->i_mtime = current_time(inode);
5418 inode->i_atime = inode->i_mtime;
5419 inode->i_ctime = inode->i_mtime;
5420 BTRFS_I(inode)->i_otime = inode->i_mtime;
5425 static inline u8 btrfs_inode_type(struct inode *inode)
5428 * Compile-time asserts that generic FT_* types still match
5431 BUILD_BUG_ON(BTRFS_FT_UNKNOWN != FT_UNKNOWN);
5432 BUILD_BUG_ON(BTRFS_FT_REG_FILE != FT_REG_FILE);
5433 BUILD_BUG_ON(BTRFS_FT_DIR != FT_DIR);
5434 BUILD_BUG_ON(BTRFS_FT_CHRDEV != FT_CHRDEV);
5435 BUILD_BUG_ON(BTRFS_FT_BLKDEV != FT_BLKDEV);
5436 BUILD_BUG_ON(BTRFS_FT_FIFO != FT_FIFO);
5437 BUILD_BUG_ON(BTRFS_FT_SOCK != FT_SOCK);
5438 BUILD_BUG_ON(BTRFS_FT_SYMLINK != FT_SYMLINK);
5440 return fs_umode_to_ftype(inode->i_mode);
5443 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5445 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5446 struct inode *inode;
5447 struct btrfs_root *root = BTRFS_I(dir)->root;
5448 struct btrfs_root *sub_root = root;
5449 struct btrfs_key location;
5453 if (dentry->d_name.len > BTRFS_NAME_LEN)
5454 return ERR_PTR(-ENAMETOOLONG);
5456 ret = btrfs_inode_by_name(dir, dentry, &location, &di_type);
5458 return ERR_PTR(ret);
5460 if (location.type == BTRFS_INODE_ITEM_KEY) {
5461 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5465 /* Do extra check against inode mode with di_type */
5466 if (btrfs_inode_type(inode) != di_type) {
5468 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5469 inode->i_mode, btrfs_inode_type(inode),
5472 return ERR_PTR(-EUCLEAN);
5477 ret = fixup_tree_root_location(fs_info, dir, dentry,
5478 &location, &sub_root);
5481 inode = ERR_PTR(ret);
5483 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5485 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5487 if (root != sub_root)
5488 btrfs_put_root(sub_root);
5490 if (!IS_ERR(inode) && root != sub_root) {
5491 down_read(&fs_info->cleanup_work_sem);
5492 if (!sb_rdonly(inode->i_sb))
5493 ret = btrfs_orphan_cleanup(sub_root);
5494 up_read(&fs_info->cleanup_work_sem);
5497 inode = ERR_PTR(ret);
5504 static int btrfs_dentry_delete(const struct dentry *dentry)
5506 struct btrfs_root *root;
5507 struct inode *inode = d_inode(dentry);
5509 if (!inode && !IS_ROOT(dentry))
5510 inode = d_inode(dentry->d_parent);
5513 root = BTRFS_I(inode)->root;
5514 if (btrfs_root_refs(&root->root_item) == 0)
5517 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5523 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5526 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5528 if (inode == ERR_PTR(-ENOENT))
5530 return d_splice_alias(inode, dentry);
5534 * All this infrastructure exists because dir_emit can fault, and we are holding
5535 * the tree lock when doing readdir. For now just allocate a buffer and copy
5536 * our information into that, and then dir_emit from the buffer. This is
5537 * similar to what NFS does, only we don't keep the buffer around in pagecache
5538 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5539 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5542 static int btrfs_opendir(struct inode *inode, struct file *file)
5544 struct btrfs_file_private *private;
5546 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5549 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5550 if (!private->filldir_buf) {
5554 file->private_data = private;
5565 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5568 struct dir_entry *entry = addr;
5569 char *name = (char *)(entry + 1);
5571 ctx->pos = get_unaligned(&entry->offset);
5572 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5573 get_unaligned(&entry->ino),
5574 get_unaligned(&entry->type)))
5576 addr += sizeof(struct dir_entry) +
5577 get_unaligned(&entry->name_len);
5583 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5585 struct inode *inode = file_inode(file);
5586 struct btrfs_root *root = BTRFS_I(inode)->root;
5587 struct btrfs_file_private *private = file->private_data;
5588 struct btrfs_dir_item *di;
5589 struct btrfs_key key;
5590 struct btrfs_key found_key;
5591 struct btrfs_path *path;
5593 struct list_head ins_list;
5594 struct list_head del_list;
5596 struct extent_buffer *leaf;
5603 struct btrfs_key location;
5605 if (!dir_emit_dots(file, ctx))
5608 path = btrfs_alloc_path();
5612 addr = private->filldir_buf;
5613 path->reada = READA_FORWARD;
5615 INIT_LIST_HEAD(&ins_list);
5616 INIT_LIST_HEAD(&del_list);
5617 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5620 key.type = BTRFS_DIR_INDEX_KEY;
5621 key.offset = ctx->pos;
5622 key.objectid = btrfs_ino(BTRFS_I(inode));
5624 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5629 struct dir_entry *entry;
5631 leaf = path->nodes[0];
5632 slot = path->slots[0];
5633 if (slot >= btrfs_header_nritems(leaf)) {
5634 ret = btrfs_next_leaf(root, path);
5642 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5644 if (found_key.objectid != key.objectid)
5646 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5648 if (found_key.offset < ctx->pos)
5650 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5652 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5653 name_len = btrfs_dir_name_len(leaf, di);
5654 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5656 btrfs_release_path(path);
5657 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5660 addr = private->filldir_buf;
5667 put_unaligned(name_len, &entry->name_len);
5668 name_ptr = (char *)(entry + 1);
5669 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5671 put_unaligned(fs_ftype_to_dtype(btrfs_dir_type(leaf, di)),
5673 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5674 put_unaligned(location.objectid, &entry->ino);
5675 put_unaligned(found_key.offset, &entry->offset);
5677 addr += sizeof(struct dir_entry) + name_len;
5678 total_len += sizeof(struct dir_entry) + name_len;
5682 btrfs_release_path(path);
5684 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5688 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5693 * Stop new entries from being returned after we return the last
5696 * New directory entries are assigned a strictly increasing
5697 * offset. This means that new entries created during readdir
5698 * are *guaranteed* to be seen in the future by that readdir.
5699 * This has broken buggy programs which operate on names as
5700 * they're returned by readdir. Until we re-use freed offsets
5701 * we have this hack to stop new entries from being returned
5702 * under the assumption that they'll never reach this huge
5705 * This is being careful not to overflow 32bit loff_t unless the
5706 * last entry requires it because doing so has broken 32bit apps
5709 if (ctx->pos >= INT_MAX)
5710 ctx->pos = LLONG_MAX;
5717 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5718 btrfs_free_path(path);
5723 * This is somewhat expensive, updating the tree every time the
5724 * inode changes. But, it is most likely to find the inode in cache.
5725 * FIXME, needs more benchmarking...there are no reasons other than performance
5726 * to keep or drop this code.
5728 static int btrfs_dirty_inode(struct inode *inode)
5730 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5731 struct btrfs_root *root = BTRFS_I(inode)->root;
5732 struct btrfs_trans_handle *trans;
5735 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5738 trans = btrfs_join_transaction(root);
5740 return PTR_ERR(trans);
5742 ret = btrfs_update_inode(trans, root, inode);
5743 if (ret && ret == -ENOSPC) {
5744 /* whoops, lets try again with the full transaction */
5745 btrfs_end_transaction(trans);
5746 trans = btrfs_start_transaction(root, 1);
5748 return PTR_ERR(trans);
5750 ret = btrfs_update_inode(trans, root, inode);
5752 btrfs_end_transaction(trans);
5753 if (BTRFS_I(inode)->delayed_node)
5754 btrfs_balance_delayed_items(fs_info);
5760 * This is a copy of file_update_time. We need this so we can return error on
5761 * ENOSPC for updating the inode in the case of file write and mmap writes.
5763 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
5766 struct btrfs_root *root = BTRFS_I(inode)->root;
5767 bool dirty = flags & ~S_VERSION;
5769 if (btrfs_root_readonly(root))
5772 if (flags & S_VERSION)
5773 dirty |= inode_maybe_inc_iversion(inode, dirty);
5774 if (flags & S_CTIME)
5775 inode->i_ctime = *now;
5776 if (flags & S_MTIME)
5777 inode->i_mtime = *now;
5778 if (flags & S_ATIME)
5779 inode->i_atime = *now;
5780 return dirty ? btrfs_dirty_inode(inode) : 0;
5784 * find the highest existing sequence number in a directory
5785 * and then set the in-memory index_cnt variable to reflect
5786 * free sequence numbers
5788 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
5790 struct btrfs_root *root = inode->root;
5791 struct btrfs_key key, found_key;
5792 struct btrfs_path *path;
5793 struct extent_buffer *leaf;
5796 key.objectid = btrfs_ino(inode);
5797 key.type = BTRFS_DIR_INDEX_KEY;
5798 key.offset = (u64)-1;
5800 path = btrfs_alloc_path();
5804 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5807 /* FIXME: we should be able to handle this */
5813 * MAGIC NUMBER EXPLANATION:
5814 * since we search a directory based on f_pos we have to start at 2
5815 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5816 * else has to start at 2
5818 if (path->slots[0] == 0) {
5819 inode->index_cnt = 2;
5825 leaf = path->nodes[0];
5826 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5828 if (found_key.objectid != btrfs_ino(inode) ||
5829 found_key.type != BTRFS_DIR_INDEX_KEY) {
5830 inode->index_cnt = 2;
5834 inode->index_cnt = found_key.offset + 1;
5836 btrfs_free_path(path);
5841 * helper to find a free sequence number in a given directory. This current
5842 * code is very simple, later versions will do smarter things in the btree
5844 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
5848 if (dir->index_cnt == (u64)-1) {
5849 ret = btrfs_inode_delayed_dir_index_count(dir);
5851 ret = btrfs_set_inode_index_count(dir);
5857 *index = dir->index_cnt;
5863 static int btrfs_insert_inode_locked(struct inode *inode)
5865 struct btrfs_iget_args args;
5867 args.ino = BTRFS_I(inode)->location.objectid;
5868 args.root = BTRFS_I(inode)->root;
5870 return insert_inode_locked4(inode,
5871 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
5872 btrfs_find_actor, &args);
5876 * Inherit flags from the parent inode.
5878 * Currently only the compression flags and the cow flags are inherited.
5880 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
5887 flags = BTRFS_I(dir)->flags;
5889 if (flags & BTRFS_INODE_NOCOMPRESS) {
5890 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
5891 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
5892 } else if (flags & BTRFS_INODE_COMPRESS) {
5893 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
5894 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
5897 if (flags & BTRFS_INODE_NODATACOW) {
5898 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
5899 if (S_ISREG(inode->i_mode))
5900 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5903 btrfs_sync_inode_flags_to_i_flags(inode);
5906 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5907 struct btrfs_root *root,
5909 const char *name, int name_len,
5910 u64 ref_objectid, u64 objectid,
5911 umode_t mode, u64 *index)
5913 struct btrfs_fs_info *fs_info = root->fs_info;
5914 struct inode *inode;
5915 struct btrfs_inode_item *inode_item;
5916 struct btrfs_key *location;
5917 struct btrfs_path *path;
5918 struct btrfs_inode_ref *ref;
5919 struct btrfs_key key[2];
5921 int nitems = name ? 2 : 1;
5923 unsigned int nofs_flag;
5926 path = btrfs_alloc_path();
5928 return ERR_PTR(-ENOMEM);
5930 nofs_flag = memalloc_nofs_save();
5931 inode = new_inode(fs_info->sb);
5932 memalloc_nofs_restore(nofs_flag);
5934 btrfs_free_path(path);
5935 return ERR_PTR(-ENOMEM);
5939 * O_TMPFILE, set link count to 0, so that after this point,
5940 * we fill in an inode item with the correct link count.
5943 set_nlink(inode, 0);
5946 * we have to initialize this early, so we can reclaim the inode
5947 * number if we fail afterwards in this function.
5949 inode->i_ino = objectid;
5952 trace_btrfs_inode_request(dir);
5954 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
5956 btrfs_free_path(path);
5958 return ERR_PTR(ret);
5964 * index_cnt is ignored for everything but a dir,
5965 * btrfs_set_inode_index_count has an explanation for the magic
5968 BTRFS_I(inode)->index_cnt = 2;
5969 BTRFS_I(inode)->dir_index = *index;
5970 BTRFS_I(inode)->root = btrfs_grab_root(root);
5971 BTRFS_I(inode)->generation = trans->transid;
5972 inode->i_generation = BTRFS_I(inode)->generation;
5975 * We could have gotten an inode number from somebody who was fsynced
5976 * and then removed in this same transaction, so let's just set full
5977 * sync since it will be a full sync anyway and this will blow away the
5978 * old info in the log.
5980 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5982 key[0].objectid = objectid;
5983 key[0].type = BTRFS_INODE_ITEM_KEY;
5986 sizes[0] = sizeof(struct btrfs_inode_item);
5990 * Start new inodes with an inode_ref. This is slightly more
5991 * efficient for small numbers of hard links since they will
5992 * be packed into one item. Extended refs will kick in if we
5993 * add more hard links than can fit in the ref item.
5995 key[1].objectid = objectid;
5996 key[1].type = BTRFS_INODE_REF_KEY;
5997 key[1].offset = ref_objectid;
5999 sizes[1] = name_len + sizeof(*ref);
6002 location = &BTRFS_I(inode)->location;
6003 location->objectid = objectid;
6004 location->offset = 0;
6005 location->type = BTRFS_INODE_ITEM_KEY;
6007 ret = btrfs_insert_inode_locked(inode);
6013 path->leave_spinning = 1;
6014 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6018 inode_init_owner(inode, dir, mode);
6019 inode_set_bytes(inode, 0);
6021 inode->i_mtime = current_time(inode);
6022 inode->i_atime = inode->i_mtime;
6023 inode->i_ctime = inode->i_mtime;
6024 BTRFS_I(inode)->i_otime = inode->i_mtime;
6026 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6027 struct btrfs_inode_item);
6028 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6029 sizeof(*inode_item));
6030 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6033 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6034 struct btrfs_inode_ref);
6035 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6036 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6037 ptr = (unsigned long)(ref + 1);
6038 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6041 btrfs_mark_buffer_dirty(path->nodes[0]);
6042 btrfs_free_path(path);
6044 btrfs_inherit_iflags(inode, dir);
6046 if (S_ISREG(mode)) {
6047 if (btrfs_test_opt(fs_info, NODATASUM))
6048 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6049 if (btrfs_test_opt(fs_info, NODATACOW))
6050 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6051 BTRFS_INODE_NODATASUM;
6054 inode_tree_add(inode);
6056 trace_btrfs_inode_new(inode);
6057 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6059 btrfs_update_root_times(trans, root);
6061 ret = btrfs_inode_inherit_props(trans, inode, dir);
6064 "error inheriting props for ino %llu (root %llu): %d",
6065 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6070 discard_new_inode(inode);
6073 BTRFS_I(dir)->index_cnt--;
6074 btrfs_free_path(path);
6075 return ERR_PTR(ret);
6079 * utility function to add 'inode' into 'parent_inode' with
6080 * a give name and a given sequence number.
6081 * if 'add_backref' is true, also insert a backref from the
6082 * inode to the parent directory.
6084 int btrfs_add_link(struct btrfs_trans_handle *trans,
6085 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6086 const char *name, int name_len, int add_backref, u64 index)
6089 struct btrfs_key key;
6090 struct btrfs_root *root = parent_inode->root;
6091 u64 ino = btrfs_ino(inode);
6092 u64 parent_ino = btrfs_ino(parent_inode);
6094 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6095 memcpy(&key, &inode->root->root_key, sizeof(key));
6098 key.type = BTRFS_INODE_ITEM_KEY;
6102 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6103 ret = btrfs_add_root_ref(trans, key.objectid,
6104 root->root_key.objectid, parent_ino,
6105 index, name, name_len);
6106 } else if (add_backref) {
6107 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6111 /* Nothing to clean up yet */
6115 ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key,
6116 btrfs_inode_type(&inode->vfs_inode), index);
6117 if (ret == -EEXIST || ret == -EOVERFLOW)
6120 btrfs_abort_transaction(trans, ret);
6124 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6126 inode_inc_iversion(&parent_inode->vfs_inode);
6128 * If we are replaying a log tree, we do not want to update the mtime
6129 * and ctime of the parent directory with the current time, since the
6130 * log replay procedure is responsible for setting them to their correct
6131 * values (the ones it had when the fsync was done).
6133 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
6134 struct timespec64 now = current_time(&parent_inode->vfs_inode);
6136 parent_inode->vfs_inode.i_mtime = now;
6137 parent_inode->vfs_inode.i_ctime = now;
6139 ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode);
6141 btrfs_abort_transaction(trans, ret);
6145 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6148 err = btrfs_del_root_ref(trans, key.objectid,
6149 root->root_key.objectid, parent_ino,
6150 &local_index, name, name_len);
6152 btrfs_abort_transaction(trans, err);
6153 } else if (add_backref) {
6157 err = btrfs_del_inode_ref(trans, root, name, name_len,
6158 ino, parent_ino, &local_index);
6160 btrfs_abort_transaction(trans, err);
6163 /* Return the original error code */
6167 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6168 struct btrfs_inode *dir, struct dentry *dentry,
6169 struct btrfs_inode *inode, int backref, u64 index)
6171 int err = btrfs_add_link(trans, dir, inode,
6172 dentry->d_name.name, dentry->d_name.len,
6179 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6180 umode_t mode, dev_t rdev)
6182 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6183 struct btrfs_trans_handle *trans;
6184 struct btrfs_root *root = BTRFS_I(dir)->root;
6185 struct inode *inode = NULL;
6191 * 2 for inode item and ref
6193 * 1 for xattr if selinux is on
6195 trans = btrfs_start_transaction(root, 5);
6197 return PTR_ERR(trans);
6199 err = btrfs_find_free_ino(root, &objectid);
6203 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6204 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6206 if (IS_ERR(inode)) {
6207 err = PTR_ERR(inode);
6213 * If the active LSM wants to access the inode during
6214 * d_instantiate it needs these. Smack checks to see
6215 * if the filesystem supports xattrs by looking at the
6218 inode->i_op = &btrfs_special_inode_operations;
6219 init_special_inode(inode, inode->i_mode, rdev);
6221 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6225 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6230 btrfs_update_inode(trans, root, inode);
6231 d_instantiate_new(dentry, inode);
6234 btrfs_end_transaction(trans);
6235 btrfs_btree_balance_dirty(fs_info);
6237 inode_dec_link_count(inode);
6238 discard_new_inode(inode);
6243 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6244 umode_t mode, bool excl)
6246 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6247 struct btrfs_trans_handle *trans;
6248 struct btrfs_root *root = BTRFS_I(dir)->root;
6249 struct inode *inode = NULL;
6255 * 2 for inode item and ref
6257 * 1 for xattr if selinux is on
6259 trans = btrfs_start_transaction(root, 5);
6261 return PTR_ERR(trans);
6263 err = btrfs_find_free_ino(root, &objectid);
6267 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6268 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6270 if (IS_ERR(inode)) {
6271 err = PTR_ERR(inode);
6276 * If the active LSM wants to access the inode during
6277 * d_instantiate it needs these. Smack checks to see
6278 * if the filesystem supports xattrs by looking at the
6281 inode->i_fop = &btrfs_file_operations;
6282 inode->i_op = &btrfs_file_inode_operations;
6283 inode->i_mapping->a_ops = &btrfs_aops;
6285 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6289 err = btrfs_update_inode(trans, root, inode);
6293 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6298 d_instantiate_new(dentry, inode);
6301 btrfs_end_transaction(trans);
6303 inode_dec_link_count(inode);
6304 discard_new_inode(inode);
6306 btrfs_btree_balance_dirty(fs_info);
6310 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6311 struct dentry *dentry)
6313 struct btrfs_trans_handle *trans = NULL;
6314 struct btrfs_root *root = BTRFS_I(dir)->root;
6315 struct inode *inode = d_inode(old_dentry);
6316 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6321 /* do not allow sys_link's with other subvols of the same device */
6322 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6325 if (inode->i_nlink >= BTRFS_LINK_MAX)
6328 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6333 * 2 items for inode and inode ref
6334 * 2 items for dir items
6335 * 1 item for parent inode
6336 * 1 item for orphan item deletion if O_TMPFILE
6338 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6339 if (IS_ERR(trans)) {
6340 err = PTR_ERR(trans);
6345 /* There are several dir indexes for this inode, clear the cache. */
6346 BTRFS_I(inode)->dir_index = 0ULL;
6348 inode_inc_iversion(inode);
6349 inode->i_ctime = current_time(inode);
6351 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6353 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6359 struct dentry *parent = dentry->d_parent;
6361 err = btrfs_update_inode(trans, root, inode);
6364 if (inode->i_nlink == 1) {
6366 * If new hard link count is 1, it's a file created
6367 * with open(2) O_TMPFILE flag.
6369 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6373 d_instantiate(dentry, inode);
6374 btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent);
6379 btrfs_end_transaction(trans);
6381 inode_dec_link_count(inode);
6384 btrfs_btree_balance_dirty(fs_info);
6388 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6390 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6391 struct inode *inode = NULL;
6392 struct btrfs_trans_handle *trans;
6393 struct btrfs_root *root = BTRFS_I(dir)->root;
6399 * 2 items for inode and ref
6400 * 2 items for dir items
6401 * 1 for xattr if selinux is on
6403 trans = btrfs_start_transaction(root, 5);
6405 return PTR_ERR(trans);
6407 err = btrfs_find_free_ino(root, &objectid);
6411 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6412 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6413 S_IFDIR | mode, &index);
6414 if (IS_ERR(inode)) {
6415 err = PTR_ERR(inode);
6420 /* these must be set before we unlock the inode */
6421 inode->i_op = &btrfs_dir_inode_operations;
6422 inode->i_fop = &btrfs_dir_file_operations;
6424 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6428 btrfs_i_size_write(BTRFS_I(inode), 0);
6429 err = btrfs_update_inode(trans, root, inode);
6433 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6434 dentry->d_name.name,
6435 dentry->d_name.len, 0, index);
6439 d_instantiate_new(dentry, inode);
6442 btrfs_end_transaction(trans);
6444 inode_dec_link_count(inode);
6445 discard_new_inode(inode);
6447 btrfs_btree_balance_dirty(fs_info);
6451 static noinline int uncompress_inline(struct btrfs_path *path,
6453 size_t pg_offset, u64 extent_offset,
6454 struct btrfs_file_extent_item *item)
6457 struct extent_buffer *leaf = path->nodes[0];
6460 unsigned long inline_size;
6464 WARN_ON(pg_offset != 0);
6465 compress_type = btrfs_file_extent_compression(leaf, item);
6466 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6467 inline_size = btrfs_file_extent_inline_item_len(leaf,
6468 btrfs_item_nr(path->slots[0]));
6469 tmp = kmalloc(inline_size, GFP_NOFS);
6472 ptr = btrfs_file_extent_inline_start(item);
6474 read_extent_buffer(leaf, tmp, ptr, inline_size);
6476 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6477 ret = btrfs_decompress(compress_type, tmp, page,
6478 extent_offset, inline_size, max_size);
6481 * decompression code contains a memset to fill in any space between the end
6482 * of the uncompressed data and the end of max_size in case the decompressed
6483 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6484 * the end of an inline extent and the beginning of the next block, so we
6485 * cover that region here.
6488 if (max_size + pg_offset < PAGE_SIZE) {
6489 char *map = kmap(page);
6490 memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6498 * btrfs_get_extent - Lookup the first extent overlapping a range in a file.
6499 * @inode: file to search in
6500 * @page: page to read extent data into if the extent is inline
6501 * @pg_offset: offset into @page to copy to
6502 * @start: file offset
6503 * @len: length of range starting at @start
6505 * This returns the first &struct extent_map which overlaps with the given
6506 * range, reading it from the B-tree and caching it if necessary. Note that
6507 * there may be more extents which overlap the given range after the returned
6510 * If @page is not NULL and the extent is inline, this also reads the extent
6511 * data directly into the page and marks the extent up to date in the io_tree.
6513 * Return: ERR_PTR on error, non-NULL extent_map on success.
6515 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6516 struct page *page, size_t pg_offset,
6519 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6521 u64 extent_start = 0;
6523 u64 objectid = btrfs_ino(inode);
6524 int extent_type = -1;
6525 struct btrfs_path *path = NULL;
6526 struct btrfs_root *root = inode->root;
6527 struct btrfs_file_extent_item *item;
6528 struct extent_buffer *leaf;
6529 struct btrfs_key found_key;
6530 struct extent_map *em = NULL;
6531 struct extent_map_tree *em_tree = &inode->extent_tree;
6532 struct extent_io_tree *io_tree = &inode->io_tree;
6534 read_lock(&em_tree->lock);
6535 em = lookup_extent_mapping(em_tree, start, len);
6536 read_unlock(&em_tree->lock);
6539 if (em->start > start || em->start + em->len <= start)
6540 free_extent_map(em);
6541 else if (em->block_start == EXTENT_MAP_INLINE && page)
6542 free_extent_map(em);
6546 em = alloc_extent_map();
6551 em->start = EXTENT_MAP_HOLE;
6552 em->orig_start = EXTENT_MAP_HOLE;
6554 em->block_len = (u64)-1;
6556 path = btrfs_alloc_path();
6562 /* Chances are we'll be called again, so go ahead and do readahead */
6563 path->reada = READA_FORWARD;
6566 * Unless we're going to uncompress the inline extent, no sleep would
6569 path->leave_spinning = 1;
6571 path->recurse = btrfs_is_free_space_inode(inode);
6573 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6576 } else if (ret > 0) {
6577 if (path->slots[0] == 0)
6583 leaf = path->nodes[0];
6584 item = btrfs_item_ptr(leaf, path->slots[0],
6585 struct btrfs_file_extent_item);
6586 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6587 if (found_key.objectid != objectid ||
6588 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6590 * If we backup past the first extent we want to move forward
6591 * and see if there is an extent in front of us, otherwise we'll
6592 * say there is a hole for our whole search range which can
6599 extent_type = btrfs_file_extent_type(leaf, item);
6600 extent_start = found_key.offset;
6601 extent_end = btrfs_file_extent_end(path);
6602 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6603 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6604 /* Only regular file could have regular/prealloc extent */
6605 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6608 "regular/prealloc extent found for non-regular inode %llu",
6612 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6614 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6615 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6620 if (start >= extent_end) {
6622 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6623 ret = btrfs_next_leaf(root, path);
6629 leaf = path->nodes[0];
6631 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6632 if (found_key.objectid != objectid ||
6633 found_key.type != BTRFS_EXTENT_DATA_KEY)
6635 if (start + len <= found_key.offset)
6637 if (start > found_key.offset)
6640 /* New extent overlaps with existing one */
6642 em->orig_start = start;
6643 em->len = found_key.offset - start;
6644 em->block_start = EXTENT_MAP_HOLE;
6648 btrfs_extent_item_to_extent_map(inode, path, item, !page, em);
6650 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6651 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6653 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6657 size_t extent_offset;
6663 size = btrfs_file_extent_ram_bytes(leaf, item);
6664 extent_offset = page_offset(page) + pg_offset - extent_start;
6665 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6666 size - extent_offset);
6667 em->start = extent_start + extent_offset;
6668 em->len = ALIGN(copy_size, fs_info->sectorsize);
6669 em->orig_block_len = em->len;
6670 em->orig_start = em->start;
6671 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6673 btrfs_set_path_blocking(path);
6674 if (!PageUptodate(page)) {
6675 if (btrfs_file_extent_compression(leaf, item) !=
6676 BTRFS_COMPRESS_NONE) {
6677 ret = uncompress_inline(path, page, pg_offset,
6678 extent_offset, item);
6683 read_extent_buffer(leaf, map + pg_offset, ptr,
6685 if (pg_offset + copy_size < PAGE_SIZE) {
6686 memset(map + pg_offset + copy_size, 0,
6687 PAGE_SIZE - pg_offset -
6692 flush_dcache_page(page);
6694 set_extent_uptodate(io_tree, em->start,
6695 extent_map_end(em) - 1, NULL, GFP_NOFS);
6700 em->orig_start = start;
6702 em->block_start = EXTENT_MAP_HOLE;
6705 btrfs_release_path(path);
6706 if (em->start > start || extent_map_end(em) <= start) {
6708 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6709 em->start, em->len, start, len);
6714 write_lock(&em_tree->lock);
6715 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6716 write_unlock(&em_tree->lock);
6718 btrfs_free_path(path);
6720 trace_btrfs_get_extent(root, inode, em);
6723 free_extent_map(em);
6724 return ERR_PTR(ret);
6729 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
6732 struct extent_map *em;
6733 struct extent_map *hole_em = NULL;
6734 u64 delalloc_start = start;
6740 em = btrfs_get_extent(inode, NULL, 0, start, len);
6744 * If our em maps to:
6746 * - a pre-alloc extent,
6747 * there might actually be delalloc bytes behind it.
6749 if (em->block_start != EXTENT_MAP_HOLE &&
6750 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6755 /* check to see if we've wrapped (len == -1 or similar) */
6764 /* ok, we didn't find anything, lets look for delalloc */
6765 delalloc_len = count_range_bits(&inode->io_tree, &delalloc_start,
6766 end, len, EXTENT_DELALLOC, 1);
6767 delalloc_end = delalloc_start + delalloc_len;
6768 if (delalloc_end < delalloc_start)
6769 delalloc_end = (u64)-1;
6772 * We didn't find anything useful, return the original results from
6775 if (delalloc_start > end || delalloc_end <= start) {
6782 * Adjust the delalloc_start to make sure it doesn't go backwards from
6783 * the start they passed in
6785 delalloc_start = max(start, delalloc_start);
6786 delalloc_len = delalloc_end - delalloc_start;
6788 if (delalloc_len > 0) {
6791 const u64 hole_end = extent_map_end(hole_em);
6793 em = alloc_extent_map();
6801 * When btrfs_get_extent can't find anything it returns one
6804 * Make sure what it found really fits our range, and adjust to
6805 * make sure it is based on the start from the caller
6807 if (hole_end <= start || hole_em->start > end) {
6808 free_extent_map(hole_em);
6811 hole_start = max(hole_em->start, start);
6812 hole_len = hole_end - hole_start;
6815 if (hole_em && delalloc_start > hole_start) {
6817 * Our hole starts before our delalloc, so we have to
6818 * return just the parts of the hole that go until the
6821 em->len = min(hole_len, delalloc_start - hole_start);
6822 em->start = hole_start;
6823 em->orig_start = hole_start;
6825 * Don't adjust block start at all, it is fixed at
6828 em->block_start = hole_em->block_start;
6829 em->block_len = hole_len;
6830 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6831 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6834 * Hole is out of passed range or it starts after
6837 em->start = delalloc_start;
6838 em->len = delalloc_len;
6839 em->orig_start = delalloc_start;
6840 em->block_start = EXTENT_MAP_DELALLOC;
6841 em->block_len = delalloc_len;
6848 free_extent_map(hole_em);
6850 free_extent_map(em);
6851 return ERR_PTR(err);
6856 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
6859 const u64 orig_start,
6860 const u64 block_start,
6861 const u64 block_len,
6862 const u64 orig_block_len,
6863 const u64 ram_bytes,
6866 struct extent_map *em = NULL;
6869 if (type != BTRFS_ORDERED_NOCOW) {
6870 em = create_io_em(inode, start, len, orig_start, block_start,
6871 block_len, orig_block_len, ram_bytes,
6872 BTRFS_COMPRESS_NONE, /* compress_type */
6877 ret = btrfs_add_ordered_extent_dio(inode, start, block_start, len,
6881 free_extent_map(em);
6882 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
6891 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
6894 struct btrfs_root *root = inode->root;
6895 struct btrfs_fs_info *fs_info = root->fs_info;
6896 struct extent_map *em;
6897 struct btrfs_key ins;
6901 alloc_hint = get_extent_allocation_hint(inode, start, len);
6902 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
6903 0, alloc_hint, &ins, 1, 1);
6905 return ERR_PTR(ret);
6907 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
6908 ins.objectid, ins.offset, ins.offset,
6909 ins.offset, BTRFS_ORDERED_REGULAR);
6910 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
6912 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
6919 * Check if we can do nocow write into the range [@offset, @offset + @len)
6921 * @offset: File offset
6922 * @len: The length to write, will be updated to the nocow writeable
6924 * @orig_start: (optional) Return the original file offset of the file extent
6925 * @orig_len: (optional) Return the original on-disk length of the file extent
6926 * @ram_bytes: (optional) Return the ram_bytes of the file extent
6927 * @strict: if true, omit optimizations that might force us into unnecessary
6928 * cow. e.g., don't trust generation number.
6930 * This function will flush ordered extents in the range to ensure proper
6931 * nocow checks for (nowait == false) case.
6934 * >0 and update @len if we can do nocow write
6935 * 0 if we can't do nocow write
6936 * <0 if error happened
6938 * NOTE: This only checks the file extents, caller is responsible to wait for
6939 * any ordered extents.
6941 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
6942 u64 *orig_start, u64 *orig_block_len,
6943 u64 *ram_bytes, bool strict)
6945 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6946 struct btrfs_path *path;
6948 struct extent_buffer *leaf;
6949 struct btrfs_root *root = BTRFS_I(inode)->root;
6950 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6951 struct btrfs_file_extent_item *fi;
6952 struct btrfs_key key;
6959 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
6961 path = btrfs_alloc_path();
6965 ret = btrfs_lookup_file_extent(NULL, root, path,
6966 btrfs_ino(BTRFS_I(inode)), offset, 0);
6970 slot = path->slots[0];
6973 /* can't find the item, must cow */
6980 leaf = path->nodes[0];
6981 btrfs_item_key_to_cpu(leaf, &key, slot);
6982 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
6983 key.type != BTRFS_EXTENT_DATA_KEY) {
6984 /* not our file or wrong item type, must cow */
6988 if (key.offset > offset) {
6989 /* Wrong offset, must cow */
6993 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6994 found_type = btrfs_file_extent_type(leaf, fi);
6995 if (found_type != BTRFS_FILE_EXTENT_REG &&
6996 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6997 /* not a regular extent, must cow */
7001 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7004 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7005 if (extent_end <= offset)
7008 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7009 if (disk_bytenr == 0)
7012 if (btrfs_file_extent_compression(leaf, fi) ||
7013 btrfs_file_extent_encryption(leaf, fi) ||
7014 btrfs_file_extent_other_encoding(leaf, fi))
7018 * Do the same check as in btrfs_cross_ref_exist but without the
7019 * unnecessary search.
7022 (btrfs_file_extent_generation(leaf, fi) <=
7023 btrfs_root_last_snapshot(&root->root_item)))
7026 backref_offset = btrfs_file_extent_offset(leaf, fi);
7029 *orig_start = key.offset - backref_offset;
7030 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7031 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7034 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7037 num_bytes = min(offset + *len, extent_end) - offset;
7038 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7041 range_end = round_up(offset + num_bytes,
7042 root->fs_info->sectorsize) - 1;
7043 ret = test_range_bit(io_tree, offset, range_end,
7044 EXTENT_DELALLOC, 0, NULL);
7051 btrfs_release_path(path);
7054 * look for other files referencing this extent, if we
7055 * find any we must cow
7058 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7059 key.offset - backref_offset, disk_bytenr,
7067 * adjust disk_bytenr and num_bytes to cover just the bytes
7068 * in this extent we are about to write. If there
7069 * are any csums in that range we have to cow in order
7070 * to keep the csums correct
7072 disk_bytenr += backref_offset;
7073 disk_bytenr += offset - key.offset;
7074 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7077 * all of the above have passed, it is safe to overwrite this extent
7083 btrfs_free_path(path);
7087 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7088 struct extent_state **cached_state, bool writing)
7090 struct btrfs_ordered_extent *ordered;
7094 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7097 * We're concerned with the entire range that we're going to be
7098 * doing DIO to, so we need to make sure there's no ordered
7099 * extents in this range.
7101 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7102 lockend - lockstart + 1);
7105 * We need to make sure there are no buffered pages in this
7106 * range either, we could have raced between the invalidate in
7107 * generic_file_direct_write and locking the extent. The
7108 * invalidate needs to happen so that reads after a write do not
7112 (!writing || !filemap_range_has_page(inode->i_mapping,
7113 lockstart, lockend)))
7116 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7121 * If we are doing a DIO read and the ordered extent we
7122 * found is for a buffered write, we can not wait for it
7123 * to complete and retry, because if we do so we can
7124 * deadlock with concurrent buffered writes on page
7125 * locks. This happens only if our DIO read covers more
7126 * than one extent map, if at this point has already
7127 * created an ordered extent for a previous extent map
7128 * and locked its range in the inode's io tree, and a
7129 * concurrent write against that previous extent map's
7130 * range and this range started (we unlock the ranges
7131 * in the io tree only when the bios complete and
7132 * buffered writes always lock pages before attempting
7133 * to lock range in the io tree).
7136 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7137 btrfs_start_ordered_extent(ordered, 1);
7140 btrfs_put_ordered_extent(ordered);
7143 * We could trigger writeback for this range (and wait
7144 * for it to complete) and then invalidate the pages for
7145 * this range (through invalidate_inode_pages2_range()),
7146 * but that can lead us to a deadlock with a concurrent
7147 * call to readahead (a buffered read or a defrag call
7148 * triggered a readahead) on a page lock due to an
7149 * ordered dio extent we created before but did not have
7150 * yet a corresponding bio submitted (whence it can not
7151 * complete), which makes readahead wait for that
7152 * ordered extent to complete while holding a lock on
7167 /* The callers of this must take lock_extent() */
7168 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7169 u64 len, u64 orig_start, u64 block_start,
7170 u64 block_len, u64 orig_block_len,
7171 u64 ram_bytes, int compress_type,
7174 struct extent_map_tree *em_tree;
7175 struct extent_map *em;
7178 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7179 type == BTRFS_ORDERED_COMPRESSED ||
7180 type == BTRFS_ORDERED_NOCOW ||
7181 type == BTRFS_ORDERED_REGULAR);
7183 em_tree = &inode->extent_tree;
7184 em = alloc_extent_map();
7186 return ERR_PTR(-ENOMEM);
7189 em->orig_start = orig_start;
7191 em->block_len = block_len;
7192 em->block_start = block_start;
7193 em->orig_block_len = orig_block_len;
7194 em->ram_bytes = ram_bytes;
7195 em->generation = -1;
7196 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7197 if (type == BTRFS_ORDERED_PREALLOC) {
7198 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7199 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7200 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7201 em->compress_type = compress_type;
7205 btrfs_drop_extent_cache(inode, em->start,
7206 em->start + em->len - 1, 0);
7207 write_lock(&em_tree->lock);
7208 ret = add_extent_mapping(em_tree, em, 1);
7209 write_unlock(&em_tree->lock);
7211 * The caller has taken lock_extent(), who could race with us
7214 } while (ret == -EEXIST);
7217 free_extent_map(em);
7218 return ERR_PTR(ret);
7221 /* em got 2 refs now, callers needs to do free_extent_map once. */
7226 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7227 struct inode *inode,
7228 struct btrfs_dio_data *dio_data,
7231 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7232 struct extent_map *em = *map;
7236 * We don't allocate a new extent in the following cases
7238 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7240 * 2) The extent is marked as PREALLOC. We're good to go here and can
7241 * just use the extent.
7244 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7245 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7246 em->block_start != EXTENT_MAP_HOLE)) {
7248 u64 block_start, orig_start, orig_block_len, ram_bytes;
7250 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7251 type = BTRFS_ORDERED_PREALLOC;
7253 type = BTRFS_ORDERED_NOCOW;
7254 len = min(len, em->len - (start - em->start));
7255 block_start = em->block_start + (start - em->start);
7257 if (can_nocow_extent(inode, start, &len, &orig_start,
7258 &orig_block_len, &ram_bytes, false) == 1 &&
7259 btrfs_inc_nocow_writers(fs_info, block_start)) {
7260 struct extent_map *em2;
7262 em2 = btrfs_create_dio_extent(BTRFS_I(inode), start, len,
7263 orig_start, block_start,
7264 len, orig_block_len,
7266 btrfs_dec_nocow_writers(fs_info, block_start);
7267 if (type == BTRFS_ORDERED_PREALLOC) {
7268 free_extent_map(em);
7272 if (em2 && IS_ERR(em2)) {
7277 * For inode marked NODATACOW or extent marked PREALLOC,
7278 * use the existing or preallocated extent, so does not
7279 * need to adjust btrfs_space_info's bytes_may_use.
7281 btrfs_free_reserved_data_space_noquota(fs_info, len);
7286 /* this will cow the extent */
7287 free_extent_map(em);
7288 *map = em = btrfs_new_extent_direct(BTRFS_I(inode), start, len);
7294 len = min(len, em->len - (start - em->start));
7298 * Need to update the i_size under the extent lock so buffered
7299 * readers will get the updated i_size when we unlock.
7301 if (start + len > i_size_read(inode))
7302 i_size_write(inode, start + len);
7304 dio_data->reserve -= len;
7309 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7310 loff_t length, unsigned int flags, struct iomap *iomap,
7311 struct iomap *srcmap)
7313 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7314 struct extent_map *em;
7315 struct extent_state *cached_state = NULL;
7316 struct btrfs_dio_data *dio_data = NULL;
7317 u64 lockstart, lockend;
7318 const bool write = !!(flags & IOMAP_WRITE);
7321 bool unlock_extents = false;
7322 bool sync = (current->journal_info == BTRFS_DIO_SYNC_STUB);
7325 * We used current->journal_info here to see if we were sync, but
7326 * there's a lot of tests in the enospc machinery to not do flushing if
7327 * we have a journal_info set, so we need to clear this out and re-set
7330 ASSERT(current->journal_info == NULL ||
7331 current->journal_info == BTRFS_DIO_SYNC_STUB);
7332 current->journal_info = NULL;
7335 len = min_t(u64, len, fs_info->sectorsize);
7338 lockend = start + len - 1;
7341 * The generic stuff only does filemap_write_and_wait_range, which
7342 * isn't enough if we've written compressed pages to this area, so we
7343 * need to flush the dirty pages again to make absolutely sure that any
7344 * outstanding dirty pages are on disk.
7346 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7347 &BTRFS_I(inode)->runtime_flags)) {
7348 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7349 start + length - 1);
7354 dio_data = kzalloc(sizeof(*dio_data), GFP_NOFS);
7358 dio_data->sync = sync;
7359 dio_data->length = length;
7361 dio_data->reserve = round_up(length, fs_info->sectorsize);
7362 ret = btrfs_delalloc_reserve_space(BTRFS_I(inode),
7363 &dio_data->data_reserved,
7364 start, dio_data->reserve);
7366 extent_changeset_free(dio_data->data_reserved);
7371 iomap->private = dio_data;
7375 * If this errors out it's because we couldn't invalidate pagecache for
7376 * this range and we need to fallback to buffered.
7378 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, write)) {
7383 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7390 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7391 * io. INLINE is special, and we could probably kludge it in here, but
7392 * it's still buffered so for safety lets just fall back to the generic
7395 * For COMPRESSED we _have_ to read the entire extent in so we can
7396 * decompress it, so there will be buffering required no matter what we
7397 * do, so go ahead and fallback to buffered.
7399 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7400 * to buffered IO. Don't blame me, this is the price we pay for using
7403 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7404 em->block_start == EXTENT_MAP_INLINE) {
7405 free_extent_map(em);
7410 len = min(len, em->len - (start - em->start));
7412 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7416 unlock_extents = true;
7417 /* Recalc len in case the new em is smaller than requested */
7418 len = min(len, em->len - (start - em->start));
7421 * We need to unlock only the end area that we aren't using.
7422 * The rest is going to be unlocked by the endio routine.
7424 lockstart = start + len;
7425 if (lockstart < lockend)
7426 unlock_extents = true;
7430 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
7431 lockstart, lockend, &cached_state);
7433 free_extent_state(cached_state);
7436 * Translate extent map information to iomap.
7437 * We trim the extents (and move the addr) even though iomap code does
7438 * that, since we have locked only the parts we are performing I/O in.
7440 if ((em->block_start == EXTENT_MAP_HOLE) ||
7441 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7442 iomap->addr = IOMAP_NULL_ADDR;
7443 iomap->type = IOMAP_HOLE;
7445 iomap->addr = em->block_start + (start - em->start);
7446 iomap->type = IOMAP_MAPPED;
7448 iomap->offset = start;
7449 iomap->bdev = fs_info->fs_devices->latest_bdev;
7450 iomap->length = len;
7452 free_extent_map(em);
7457 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7461 btrfs_delalloc_release_space(BTRFS_I(inode),
7462 dio_data->data_reserved, start,
7463 dio_data->reserve, true);
7464 btrfs_delalloc_release_extents(BTRFS_I(inode), dio_data->reserve);
7465 extent_changeset_free(dio_data->data_reserved);
7471 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7472 ssize_t written, unsigned int flags, struct iomap *iomap)
7475 struct btrfs_dio_data *dio_data = iomap->private;
7476 size_t submitted = dio_data->submitted;
7477 const bool write = !!(flags & IOMAP_WRITE);
7479 if (!write && (iomap->type == IOMAP_HOLE)) {
7480 /* If reading from a hole, unlock and return */
7481 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1);
7485 if (submitted < length) {
7487 length -= submitted;
7489 __endio_write_update_ordered(BTRFS_I(inode), pos,
7492 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7498 if (dio_data->reserve)
7499 btrfs_delalloc_release_space(BTRFS_I(inode),
7500 dio_data->data_reserved, pos,
7501 dio_data->reserve, true);
7502 btrfs_delalloc_release_extents(BTRFS_I(inode), dio_data->length);
7503 extent_changeset_free(dio_data->data_reserved);
7507 * We're all done, we can re-set the current->journal_info now safely
7510 if (dio_data->sync) {
7511 ASSERT(current->journal_info == NULL);
7512 current->journal_info = BTRFS_DIO_SYNC_STUB;
7515 iomap->private = NULL;
7520 static void btrfs_dio_private_put(struct btrfs_dio_private *dip)
7523 * This implies a barrier so that stores to dio_bio->bi_status before
7524 * this and loads of dio_bio->bi_status after this are fully ordered.
7526 if (!refcount_dec_and_test(&dip->refs))
7529 if (bio_op(dip->dio_bio) == REQ_OP_WRITE) {
7530 __endio_write_update_ordered(BTRFS_I(dip->inode),
7531 dip->logical_offset,
7533 !dip->dio_bio->bi_status);
7535 unlock_extent(&BTRFS_I(dip->inode)->io_tree,
7536 dip->logical_offset,
7537 dip->logical_offset + dip->bytes - 1);
7540 bio_endio(dip->dio_bio);
7544 static blk_status_t submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7546 unsigned long bio_flags)
7548 struct btrfs_dio_private *dip = bio->bi_private;
7549 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7552 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7554 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
7558 refcount_inc(&dip->refs);
7559 ret = btrfs_map_bio(fs_info, bio, mirror_num);
7561 refcount_dec(&dip->refs);
7565 static blk_status_t btrfs_check_read_dio_bio(struct inode *inode,
7566 struct btrfs_io_bio *io_bio,
7567 const bool uptodate)
7569 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
7570 const u32 sectorsize = fs_info->sectorsize;
7571 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7572 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7573 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7574 struct bio_vec bvec;
7575 struct bvec_iter iter;
7576 u64 start = io_bio->logical;
7578 blk_status_t err = BLK_STS_OK;
7580 __bio_for_each_segment(bvec, &io_bio->bio, iter, io_bio->iter) {
7581 unsigned int i, nr_sectors, pgoff;
7583 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7584 pgoff = bvec.bv_offset;
7585 for (i = 0; i < nr_sectors; i++) {
7586 ASSERT(pgoff < PAGE_SIZE);
7588 (!csum || !check_data_csum(inode, io_bio, icsum,
7589 bvec.bv_page, pgoff,
7590 start, sectorsize))) {
7591 clean_io_failure(fs_info, failure_tree, io_tree,
7592 start, bvec.bv_page,
7593 btrfs_ino(BTRFS_I(inode)),
7596 blk_status_t status;
7598 status = btrfs_submit_read_repair(inode,
7600 start - io_bio->logical,
7601 bvec.bv_page, pgoff,
7603 start + sectorsize - 1,
7605 submit_dio_repair_bio);
7609 start += sectorsize;
7611 pgoff += sectorsize;
7617 static void __endio_write_update_ordered(struct btrfs_inode *inode,
7618 const u64 offset, const u64 bytes,
7619 const bool uptodate)
7621 struct btrfs_fs_info *fs_info = inode->root->fs_info;
7622 struct btrfs_ordered_extent *ordered = NULL;
7623 struct btrfs_workqueue *wq;
7624 u64 ordered_offset = offset;
7625 u64 ordered_bytes = bytes;
7628 if (btrfs_is_free_space_inode(inode))
7629 wq = fs_info->endio_freespace_worker;
7631 wq = fs_info->endio_write_workers;
7633 while (ordered_offset < offset + bytes) {
7634 last_offset = ordered_offset;
7635 if (btrfs_dec_test_first_ordered_pending(inode, &ordered,
7639 btrfs_init_work(&ordered->work, finish_ordered_fn, NULL,
7641 btrfs_queue_work(wq, &ordered->work);
7644 * If btrfs_dec_test_ordered_pending does not find any ordered
7645 * extent in the range, we can exit.
7647 if (ordered_offset == last_offset)
7650 * Our bio might span multiple ordered extents. In this case
7651 * we keep going until we have accounted the whole dio.
7653 if (ordered_offset < offset + bytes) {
7654 ordered_bytes = offset + bytes - ordered_offset;
7660 static blk_status_t btrfs_submit_bio_start_direct_io(void *private_data,
7661 struct bio *bio, u64 offset)
7663 struct inode *inode = private_data;
7665 return btrfs_csum_one_bio(BTRFS_I(inode), bio, offset, 1);
7668 static void btrfs_end_dio_bio(struct bio *bio)
7670 struct btrfs_dio_private *dip = bio->bi_private;
7671 blk_status_t err = bio->bi_status;
7674 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
7675 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
7676 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
7678 (unsigned long long)bio->bi_iter.bi_sector,
7679 bio->bi_iter.bi_size, err);
7681 if (bio_op(bio) == REQ_OP_READ) {
7682 err = btrfs_check_read_dio_bio(dip->inode, btrfs_io_bio(bio),
7687 dip->dio_bio->bi_status = err;
7690 btrfs_dio_private_put(dip);
7693 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
7694 struct inode *inode, u64 file_offset, int async_submit)
7696 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7697 struct btrfs_dio_private *dip = bio->bi_private;
7698 bool write = bio_op(bio) == REQ_OP_WRITE;
7701 /* Check btrfs_submit_bio_hook() for rules about async submit. */
7703 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7706 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
7711 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
7714 if (write && async_submit) {
7715 ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0,
7717 btrfs_submit_bio_start_direct_io);
7721 * If we aren't doing async submit, calculate the csum of the
7724 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, file_offset, 1);
7730 csum_offset = file_offset - dip->logical_offset;
7731 csum_offset >>= inode->i_sb->s_blocksize_bits;
7732 csum_offset *= btrfs_super_csum_size(fs_info->super_copy);
7733 btrfs_io_bio(bio)->csum = dip->csums + csum_offset;
7736 ret = btrfs_map_bio(fs_info, bio, 0);
7742 * If this succeeds, the btrfs_dio_private is responsible for cleaning up locked
7743 * or ordered extents whether or not we submit any bios.
7745 static struct btrfs_dio_private *btrfs_create_dio_private(struct bio *dio_bio,
7746 struct inode *inode,
7749 const bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
7750 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7752 struct btrfs_dio_private *dip;
7754 dip_size = sizeof(*dip);
7755 if (!write && csum) {
7756 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7757 const u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
7760 nblocks = dio_bio->bi_iter.bi_size >> inode->i_sb->s_blocksize_bits;
7761 dip_size += csum_size * nblocks;
7764 dip = kzalloc(dip_size, GFP_NOFS);
7769 dip->logical_offset = file_offset;
7770 dip->bytes = dio_bio->bi_iter.bi_size;
7771 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
7772 dip->dio_bio = dio_bio;
7773 refcount_set(&dip->refs, 1);
7777 static blk_qc_t btrfs_submit_direct(struct inode *inode, struct iomap *iomap,
7778 struct bio *dio_bio, loff_t file_offset)
7780 const bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
7781 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7782 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7783 const bool raid56 = (btrfs_data_alloc_profile(fs_info) &
7784 BTRFS_BLOCK_GROUP_RAID56_MASK);
7785 struct btrfs_dio_private *dip;
7788 int async_submit = 0;
7790 int clone_offset = 0;
7793 blk_status_t status;
7794 struct btrfs_io_geometry geom;
7795 struct btrfs_dio_data *dio_data = iomap->private;
7797 dip = btrfs_create_dio_private(dio_bio, inode, file_offset);
7800 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
7801 file_offset + dio_bio->bi_iter.bi_size - 1);
7803 dio_bio->bi_status = BLK_STS_RESOURCE;
7805 return BLK_QC_T_NONE;
7808 if (!write && csum) {
7810 * Load the csums up front to reduce csum tree searches and
7811 * contention when submitting bios.
7813 status = btrfs_lookup_bio_sums(inode, dio_bio, file_offset,
7815 if (status != BLK_STS_OK)
7819 start_sector = dio_bio->bi_iter.bi_sector;
7820 submit_len = dio_bio->bi_iter.bi_size;
7823 ret = btrfs_get_io_geometry(fs_info, btrfs_op(dio_bio),
7824 start_sector << 9, submit_len,
7827 status = errno_to_blk_status(ret);
7830 ASSERT(geom.len <= INT_MAX);
7832 clone_len = min_t(int, submit_len, geom.len);
7835 * This will never fail as it's passing GPF_NOFS and
7836 * the allocation is backed by btrfs_bioset.
7838 bio = btrfs_bio_clone_partial(dio_bio, clone_offset, clone_len);
7839 bio->bi_private = dip;
7840 bio->bi_end_io = btrfs_end_dio_bio;
7841 btrfs_io_bio(bio)->logical = file_offset;
7843 ASSERT(submit_len >= clone_len);
7844 submit_len -= clone_len;
7847 * Increase the count before we submit the bio so we know
7848 * the end IO handler won't happen before we increase the
7849 * count. Otherwise, the dip might get freed before we're
7850 * done setting it up.
7852 * We transfer the initial reference to the last bio, so we
7853 * don't need to increment the reference count for the last one.
7855 if (submit_len > 0) {
7856 refcount_inc(&dip->refs);
7858 * If we are submitting more than one bio, submit them
7859 * all asynchronously. The exception is RAID 5 or 6, as
7860 * asynchronous checksums make it difficult to collect
7861 * full stripe writes.
7867 status = btrfs_submit_dio_bio(bio, inode, file_offset,
7872 refcount_dec(&dip->refs);
7876 dio_data->submitted += clone_len;
7877 clone_offset += clone_len;
7878 start_sector += clone_len >> 9;
7879 file_offset += clone_len;
7880 } while (submit_len > 0);
7881 return BLK_QC_T_NONE;
7884 dip->dio_bio->bi_status = status;
7885 btrfs_dio_private_put(dip);
7886 return BLK_QC_T_NONE;
7889 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
7890 const struct iov_iter *iter, loff_t offset)
7894 unsigned int blocksize_mask = fs_info->sectorsize - 1;
7895 ssize_t retval = -EINVAL;
7897 if (offset & blocksize_mask)
7900 if (iov_iter_alignment(iter) & blocksize_mask)
7903 /* If this is a write we don't need to check anymore */
7904 if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter))
7907 * Check to make sure we don't have duplicate iov_base's in this
7908 * iovec, if so return EINVAL, otherwise we'll get csum errors
7909 * when reading back.
7911 for (seg = 0; seg < iter->nr_segs; seg++) {
7912 for (i = seg + 1; i < iter->nr_segs; i++) {
7913 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
7922 static inline int btrfs_maybe_fsync_end_io(struct kiocb *iocb, ssize_t size,
7923 int error, unsigned flags)
7926 * Now if we're still in the context of our submitter we know we can't
7927 * safely run generic_write_sync(), so clear our flag here so that the
7928 * caller knows to follow up with a sync.
7930 if (current->journal_info == BTRFS_DIO_SYNC_STUB) {
7931 current->journal_info = NULL;
7939 iocb->ki_flags |= IOCB_DSYNC;
7940 return generic_write_sync(iocb, size);
7946 static const struct iomap_ops btrfs_dio_iomap_ops = {
7947 .iomap_begin = btrfs_dio_iomap_begin,
7948 .iomap_end = btrfs_dio_iomap_end,
7951 static const struct iomap_dio_ops btrfs_dio_ops = {
7952 .submit_io = btrfs_submit_direct,
7955 static const struct iomap_dio_ops btrfs_sync_dops = {
7956 .submit_io = btrfs_submit_direct,
7957 .end_io = btrfs_maybe_fsync_end_io,
7960 ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
7962 struct file *file = iocb->ki_filp;
7963 struct inode *inode = file->f_mapping->host;
7964 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7965 struct extent_changeset *data_reserved = NULL;
7966 loff_t offset = iocb->ki_pos;
7968 bool relock = false;
7971 if (check_direct_IO(fs_info, iter, offset))
7974 count = iov_iter_count(iter);
7975 if (iov_iter_rw(iter) == WRITE) {
7977 * If the write DIO is beyond the EOF, we need update
7978 * the isize, but it is protected by i_mutex. So we can
7979 * not unlock the i_mutex at this case.
7981 if (offset + count <= inode->i_size) {
7982 inode_unlock(inode);
7985 down_read(&BTRFS_I(inode)->dio_sem);
7989 * We have are actually a sync iocb, so we need our fancy endio to know
7990 * if we need to sync.
7992 if (current->journal_info)
7993 ret = iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops,
7994 &btrfs_sync_dops, is_sync_kiocb(iocb));
7996 ret = iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops,
7997 &btrfs_dio_ops, is_sync_kiocb(iocb));
7999 if (ret == -ENOTBLK)
8002 if (iov_iter_rw(iter) == WRITE)
8003 up_read(&BTRFS_I(inode)->dio_sem);
8008 extent_changeset_free(data_reserved);
8012 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8017 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
8021 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
8024 int btrfs_readpage(struct file *file, struct page *page)
8026 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
8027 u64 start = page_offset(page);
8028 u64 end = start + PAGE_SIZE - 1;
8029 unsigned long bio_flags = 0;
8030 struct bio *bio = NULL;
8033 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
8035 ret = btrfs_do_readpage(page, NULL, &bio, &bio_flags, 0, NULL);
8037 ret = submit_one_bio(bio, 0, bio_flags);
8041 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8043 struct inode *inode = page->mapping->host;
8046 if (current->flags & PF_MEMALLOC) {
8047 redirty_page_for_writepage(wbc, page);
8053 * If we are under memory pressure we will call this directly from the
8054 * VM, we need to make sure we have the inode referenced for the ordered
8055 * extent. If not just return like we didn't do anything.
8057 if (!igrab(inode)) {
8058 redirty_page_for_writepage(wbc, page);
8059 return AOP_WRITEPAGE_ACTIVATE;
8061 ret = extent_write_full_page(page, wbc);
8062 btrfs_add_delayed_iput(inode);
8066 static int btrfs_writepages(struct address_space *mapping,
8067 struct writeback_control *wbc)
8069 return extent_writepages(mapping, wbc);
8072 static void btrfs_readahead(struct readahead_control *rac)
8074 extent_readahead(rac);
8077 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8079 int ret = try_release_extent_mapping(page, gfp_flags);
8081 detach_page_private(page);
8085 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8087 if (PageWriteback(page) || PageDirty(page))
8089 return __btrfs_releasepage(page, gfp_flags);
8092 #ifdef CONFIG_MIGRATION
8093 static int btrfs_migratepage(struct address_space *mapping,
8094 struct page *newpage, struct page *page,
8095 enum migrate_mode mode)
8099 ret = migrate_page_move_mapping(mapping, newpage, page, 0);
8100 if (ret != MIGRATEPAGE_SUCCESS)
8103 if (page_has_private(page))
8104 attach_page_private(newpage, detach_page_private(page));
8106 if (PagePrivate2(page)) {
8107 ClearPagePrivate2(page);
8108 SetPagePrivate2(newpage);
8111 if (mode != MIGRATE_SYNC_NO_COPY)
8112 migrate_page_copy(newpage, page);
8114 migrate_page_states(newpage, page);
8115 return MIGRATEPAGE_SUCCESS;
8119 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8120 unsigned int length)
8122 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
8123 struct extent_io_tree *tree = &inode->io_tree;
8124 struct btrfs_ordered_extent *ordered;
8125 struct extent_state *cached_state = NULL;
8126 u64 page_start = page_offset(page);
8127 u64 page_end = page_start + PAGE_SIZE - 1;
8130 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
8133 * we have the page locked, so new writeback can't start,
8134 * and the dirty bit won't be cleared while we are here.
8136 * Wait for IO on this page so that we can safely clear
8137 * the PagePrivate2 bit and do ordered accounting
8139 wait_on_page_writeback(page);
8142 btrfs_releasepage(page, GFP_NOFS);
8146 if (!inode_evicting)
8147 lock_extent_bits(tree, page_start, page_end, &cached_state);
8150 ordered = btrfs_lookup_ordered_range(inode, start, page_end - start + 1);
8153 ordered->file_offset + ordered->num_bytes - 1);
8155 * IO on this page will never be started, so we need
8156 * to account for any ordered extents now
8158 if (!inode_evicting)
8159 clear_extent_bit(tree, start, end,
8160 EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
8161 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8162 EXTENT_DEFRAG, 1, 0, &cached_state);
8164 * whoever cleared the private bit is responsible
8165 * for the finish_ordered_io
8167 if (TestClearPagePrivate2(page)) {
8168 struct btrfs_ordered_inode_tree *tree;
8171 tree = &inode->ordered_tree;
8173 spin_lock_irq(&tree->lock);
8174 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8175 new_len = start - ordered->file_offset;
8176 if (new_len < ordered->truncated_len)
8177 ordered->truncated_len = new_len;
8178 spin_unlock_irq(&tree->lock);
8180 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8182 end - start + 1, 1))
8183 btrfs_finish_ordered_io(ordered);
8185 btrfs_put_ordered_extent(ordered);
8186 if (!inode_evicting) {
8187 cached_state = NULL;
8188 lock_extent_bits(tree, start, end,
8193 if (start < page_end)
8198 * Qgroup reserved space handler
8199 * Page here will be either
8200 * 1) Already written to disk or ordered extent already submitted
8201 * Then its QGROUP_RESERVED bit in io_tree is already cleaned.
8202 * Qgroup will be handled by its qgroup_record then.
8203 * btrfs_qgroup_free_data() call will do nothing here.
8205 * 2) Not written to disk yet
8206 * Then btrfs_qgroup_free_data() call will clear the QGROUP_RESERVED
8207 * bit of its io_tree, and free the qgroup reserved data space.
8208 * Since the IO will never happen for this page.
8210 btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
8211 if (!inode_evicting) {
8212 clear_extent_bit(tree, page_start, page_end, EXTENT_LOCKED |
8213 EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
8214 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
8217 __btrfs_releasepage(page, GFP_NOFS);
8220 ClearPageChecked(page);
8221 detach_page_private(page);
8225 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8226 * called from a page fault handler when a page is first dirtied. Hence we must
8227 * be careful to check for EOF conditions here. We set the page up correctly
8228 * for a written page which means we get ENOSPC checking when writing into
8229 * holes and correct delalloc and unwritten extent mapping on filesystems that
8230 * support these features.
8232 * We are not allowed to take the i_mutex here so we have to play games to
8233 * protect against truncate races as the page could now be beyond EOF. Because
8234 * truncate_setsize() writes the inode size before removing pages, once we have
8235 * the page lock we can determine safely if the page is beyond EOF. If it is not
8236 * beyond EOF, then the page is guaranteed safe against truncation until we
8239 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8241 struct page *page = vmf->page;
8242 struct inode *inode = file_inode(vmf->vma->vm_file);
8243 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8244 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8245 struct btrfs_ordered_extent *ordered;
8246 struct extent_state *cached_state = NULL;
8247 struct extent_changeset *data_reserved = NULL;
8249 unsigned long zero_start;
8259 reserved_space = PAGE_SIZE;
8261 sb_start_pagefault(inode->i_sb);
8262 page_start = page_offset(page);
8263 page_end = page_start + PAGE_SIZE - 1;
8267 * Reserving delalloc space after obtaining the page lock can lead to
8268 * deadlock. For example, if a dirty page is locked by this function
8269 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8270 * dirty page write out, then the btrfs_writepage() function could
8271 * end up waiting indefinitely to get a lock on the page currently
8272 * being processed by btrfs_page_mkwrite() function.
8274 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8275 page_start, reserved_space);
8277 ret2 = file_update_time(vmf->vma->vm_file);
8281 ret = vmf_error(ret2);
8287 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8290 size = i_size_read(inode);
8292 if ((page->mapping != inode->i_mapping) ||
8293 (page_start >= size)) {
8294 /* page got truncated out from underneath us */
8297 wait_on_page_writeback(page);
8299 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8300 set_page_extent_mapped(page);
8303 * we can't set the delalloc bits if there are pending ordered
8304 * extents. Drop our locks and wait for them to finish
8306 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8309 unlock_extent_cached(io_tree, page_start, page_end,
8312 btrfs_start_ordered_extent(ordered, 1);
8313 btrfs_put_ordered_extent(ordered);
8317 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8318 reserved_space = round_up(size - page_start,
8319 fs_info->sectorsize);
8320 if (reserved_space < PAGE_SIZE) {
8321 end = page_start + reserved_space - 1;
8322 btrfs_delalloc_release_space(BTRFS_I(inode),
8323 data_reserved, page_start,
8324 PAGE_SIZE - reserved_space, true);
8329 * page_mkwrite gets called when the page is firstly dirtied after it's
8330 * faulted in, but write(2) could also dirty a page and set delalloc
8331 * bits, thus in this case for space account reason, we still need to
8332 * clear any delalloc bits within this page range since we have to
8333 * reserve data&meta space before lock_page() (see above comments).
8335 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8336 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8337 EXTENT_DEFRAG, 0, 0, &cached_state);
8339 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8342 unlock_extent_cached(io_tree, page_start, page_end,
8344 ret = VM_FAULT_SIGBUS;
8348 /* page is wholly or partially inside EOF */
8349 if (page_start + PAGE_SIZE > size)
8350 zero_start = offset_in_page(size);
8352 zero_start = PAGE_SIZE;
8354 if (zero_start != PAGE_SIZE) {
8356 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
8357 flush_dcache_page(page);
8360 ClearPageChecked(page);
8361 set_page_dirty(page);
8362 SetPageUptodate(page);
8364 BTRFS_I(inode)->last_trans = fs_info->generation;
8365 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8366 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8368 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8370 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8371 sb_end_pagefault(inode->i_sb);
8372 extent_changeset_free(data_reserved);
8373 return VM_FAULT_LOCKED;
8378 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8379 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8380 reserved_space, (ret != 0));
8382 sb_end_pagefault(inode->i_sb);
8383 extent_changeset_free(data_reserved);
8387 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
8389 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8390 struct btrfs_root *root = BTRFS_I(inode)->root;
8391 struct btrfs_block_rsv *rsv;
8393 struct btrfs_trans_handle *trans;
8394 u64 mask = fs_info->sectorsize - 1;
8395 u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8397 if (!skip_writeback) {
8398 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8405 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8406 * things going on here:
8408 * 1) We need to reserve space to update our inode.
8410 * 2) We need to have something to cache all the space that is going to
8411 * be free'd up by the truncate operation, but also have some slack
8412 * space reserved in case it uses space during the truncate (thank you
8413 * very much snapshotting).
8415 * And we need these to be separate. The fact is we can use a lot of
8416 * space doing the truncate, and we have no earthly idea how much space
8417 * we will use, so we need the truncate reservation to be separate so it
8418 * doesn't end up using space reserved for updating the inode. We also
8419 * need to be able to stop the transaction and start a new one, which
8420 * means we need to be able to update the inode several times, and we
8421 * have no idea of knowing how many times that will be, so we can't just
8422 * reserve 1 item for the entirety of the operation, so that has to be
8423 * done separately as well.
8425 * So that leaves us with
8427 * 1) rsv - for the truncate reservation, which we will steal from the
8428 * transaction reservation.
8429 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8430 * updating the inode.
8432 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8435 rsv->size = min_size;
8439 * 1 for the truncate slack space
8440 * 1 for updating the inode.
8442 trans = btrfs_start_transaction(root, 2);
8443 if (IS_ERR(trans)) {
8444 ret = PTR_ERR(trans);
8448 /* Migrate the slack space for the truncate to our reserve */
8449 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8454 * So if we truncate and then write and fsync we normally would just
8455 * write the extents that changed, which is a problem if we need to
8456 * first truncate that entire inode. So set this flag so we write out
8457 * all of the extents in the inode to the sync log so we're completely
8460 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
8461 trans->block_rsv = rsv;
8464 ret = btrfs_truncate_inode_items(trans, root, inode,
8466 BTRFS_EXTENT_DATA_KEY);
8467 trans->block_rsv = &fs_info->trans_block_rsv;
8468 if (ret != -ENOSPC && ret != -EAGAIN)
8471 ret = btrfs_update_inode(trans, root, inode);
8475 btrfs_end_transaction(trans);
8476 btrfs_btree_balance_dirty(fs_info);
8478 trans = btrfs_start_transaction(root, 2);
8479 if (IS_ERR(trans)) {
8480 ret = PTR_ERR(trans);
8485 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8486 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8487 rsv, min_size, false);
8488 BUG_ON(ret); /* shouldn't happen */
8489 trans->block_rsv = rsv;
8493 * We can't call btrfs_truncate_block inside a trans handle as we could
8494 * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
8495 * we've truncated everything except the last little bit, and can do
8496 * btrfs_truncate_block and then update the disk_i_size.
8498 if (ret == NEED_TRUNCATE_BLOCK) {
8499 btrfs_end_transaction(trans);
8500 btrfs_btree_balance_dirty(fs_info);
8502 ret = btrfs_truncate_block(inode, inode->i_size, 0, 0);
8505 trans = btrfs_start_transaction(root, 1);
8506 if (IS_ERR(trans)) {
8507 ret = PTR_ERR(trans);
8510 btrfs_inode_safe_disk_i_size_write(inode, 0);
8516 trans->block_rsv = &fs_info->trans_block_rsv;
8517 ret2 = btrfs_update_inode(trans, root, inode);
8521 ret2 = btrfs_end_transaction(trans);
8524 btrfs_btree_balance_dirty(fs_info);
8527 btrfs_free_block_rsv(fs_info, rsv);
8533 * create a new subvolume directory/inode (helper for the ioctl).
8535 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8536 struct btrfs_root *new_root,
8537 struct btrfs_root *parent_root,
8540 struct inode *inode;
8544 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
8545 new_dirid, new_dirid,
8546 S_IFDIR | (~current_umask() & S_IRWXUGO),
8549 return PTR_ERR(inode);
8550 inode->i_op = &btrfs_dir_inode_operations;
8551 inode->i_fop = &btrfs_dir_file_operations;
8553 set_nlink(inode, 1);
8554 btrfs_i_size_write(BTRFS_I(inode), 0);
8555 unlock_new_inode(inode);
8557 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8559 btrfs_err(new_root->fs_info,
8560 "error inheriting subvolume %llu properties: %d",
8561 new_root->root_key.objectid, err);
8563 err = btrfs_update_inode(trans, new_root, inode);
8569 struct inode *btrfs_alloc_inode(struct super_block *sb)
8571 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8572 struct btrfs_inode *ei;
8573 struct inode *inode;
8575 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL);
8582 ei->last_sub_trans = 0;
8583 ei->logged_trans = 0;
8584 ei->delalloc_bytes = 0;
8585 ei->new_delalloc_bytes = 0;
8586 ei->defrag_bytes = 0;
8587 ei->disk_i_size = 0;
8590 ei->index_cnt = (u64)-1;
8592 ei->last_unlink_trans = 0;
8593 ei->last_reflink_trans = 0;
8594 ei->last_log_commit = 0;
8596 spin_lock_init(&ei->lock);
8597 ei->outstanding_extents = 0;
8598 if (sb->s_magic != BTRFS_TEST_MAGIC)
8599 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8600 BTRFS_BLOCK_RSV_DELALLOC);
8601 ei->runtime_flags = 0;
8602 ei->prop_compress = BTRFS_COMPRESS_NONE;
8603 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8605 ei->delayed_node = NULL;
8607 ei->i_otime.tv_sec = 0;
8608 ei->i_otime.tv_nsec = 0;
8610 inode = &ei->vfs_inode;
8611 extent_map_tree_init(&ei->extent_tree);
8612 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO, inode);
8613 extent_io_tree_init(fs_info, &ei->io_failure_tree,
8614 IO_TREE_INODE_IO_FAILURE, inode);
8615 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8616 IO_TREE_INODE_FILE_EXTENT, inode);
8617 ei->io_tree.track_uptodate = true;
8618 ei->io_failure_tree.track_uptodate = true;
8619 atomic_set(&ei->sync_writers, 0);
8620 mutex_init(&ei->log_mutex);
8621 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8622 INIT_LIST_HEAD(&ei->delalloc_inodes);
8623 INIT_LIST_HEAD(&ei->delayed_iput);
8624 RB_CLEAR_NODE(&ei->rb_node);
8625 init_rwsem(&ei->dio_sem);
8630 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8631 void btrfs_test_destroy_inode(struct inode *inode)
8633 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
8634 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8638 void btrfs_free_inode(struct inode *inode)
8640 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8643 void btrfs_destroy_inode(struct inode *vfs_inode)
8645 struct btrfs_ordered_extent *ordered;
8646 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8647 struct btrfs_root *root = inode->root;
8649 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8650 WARN_ON(vfs_inode->i_data.nrpages);
8651 WARN_ON(inode->block_rsv.reserved);
8652 WARN_ON(inode->block_rsv.size);
8653 WARN_ON(inode->outstanding_extents);
8654 WARN_ON(inode->delalloc_bytes);
8655 WARN_ON(inode->new_delalloc_bytes);
8656 WARN_ON(inode->csum_bytes);
8657 WARN_ON(inode->defrag_bytes);
8660 * This can happen where we create an inode, but somebody else also
8661 * created the same inode and we need to destroy the one we already
8668 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8672 btrfs_err(root->fs_info,
8673 "found ordered extent %llu %llu on inode cleanup",
8674 ordered->file_offset, ordered->num_bytes);
8675 btrfs_remove_ordered_extent(inode, ordered);
8676 btrfs_put_ordered_extent(ordered);
8677 btrfs_put_ordered_extent(ordered);
8680 btrfs_qgroup_check_reserved_leak(inode);
8681 inode_tree_del(inode);
8682 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8683 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
8684 btrfs_put_root(inode->root);
8687 int btrfs_drop_inode(struct inode *inode)
8689 struct btrfs_root *root = BTRFS_I(inode)->root;
8694 /* the snap/subvol tree is on deleting */
8695 if (btrfs_root_refs(&root->root_item) == 0)
8698 return generic_drop_inode(inode);
8701 static void init_once(void *foo)
8703 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8705 inode_init_once(&ei->vfs_inode);
8708 void __cold btrfs_destroy_cachep(void)
8711 * Make sure all delayed rcu free inodes are flushed before we
8715 kmem_cache_destroy(btrfs_inode_cachep);
8716 kmem_cache_destroy(btrfs_trans_handle_cachep);
8717 kmem_cache_destroy(btrfs_path_cachep);
8718 kmem_cache_destroy(btrfs_free_space_cachep);
8719 kmem_cache_destroy(btrfs_free_space_bitmap_cachep);
8722 int __init btrfs_init_cachep(void)
8724 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8725 sizeof(struct btrfs_inode), 0,
8726 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
8728 if (!btrfs_inode_cachep)
8731 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8732 sizeof(struct btrfs_trans_handle), 0,
8733 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
8734 if (!btrfs_trans_handle_cachep)
8737 btrfs_path_cachep = kmem_cache_create("btrfs_path",
8738 sizeof(struct btrfs_path), 0,
8739 SLAB_MEM_SPREAD, NULL);
8740 if (!btrfs_path_cachep)
8743 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8744 sizeof(struct btrfs_free_space), 0,
8745 SLAB_MEM_SPREAD, NULL);
8746 if (!btrfs_free_space_cachep)
8749 btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap",
8750 PAGE_SIZE, PAGE_SIZE,
8751 SLAB_RED_ZONE, NULL);
8752 if (!btrfs_free_space_bitmap_cachep)
8757 btrfs_destroy_cachep();
8761 static int btrfs_getattr(const struct path *path, struct kstat *stat,
8762 u32 request_mask, unsigned int flags)
8765 struct inode *inode = d_inode(path->dentry);
8766 u32 blocksize = inode->i_sb->s_blocksize;
8767 u32 bi_flags = BTRFS_I(inode)->flags;
8769 stat->result_mask |= STATX_BTIME;
8770 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
8771 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
8772 if (bi_flags & BTRFS_INODE_APPEND)
8773 stat->attributes |= STATX_ATTR_APPEND;
8774 if (bi_flags & BTRFS_INODE_COMPRESS)
8775 stat->attributes |= STATX_ATTR_COMPRESSED;
8776 if (bi_flags & BTRFS_INODE_IMMUTABLE)
8777 stat->attributes |= STATX_ATTR_IMMUTABLE;
8778 if (bi_flags & BTRFS_INODE_NODUMP)
8779 stat->attributes |= STATX_ATTR_NODUMP;
8781 stat->attributes_mask |= (STATX_ATTR_APPEND |
8782 STATX_ATTR_COMPRESSED |
8783 STATX_ATTR_IMMUTABLE |
8786 generic_fillattr(inode, stat);
8787 stat->dev = BTRFS_I(inode)->root->anon_dev;
8789 spin_lock(&BTRFS_I(inode)->lock);
8790 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
8791 spin_unlock(&BTRFS_I(inode)->lock);
8792 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8793 ALIGN(delalloc_bytes, blocksize)) >> 9;
8797 static int btrfs_rename_exchange(struct inode *old_dir,
8798 struct dentry *old_dentry,
8799 struct inode *new_dir,
8800 struct dentry *new_dentry)
8802 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
8803 struct btrfs_trans_handle *trans;
8804 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8805 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8806 struct inode *new_inode = new_dentry->d_inode;
8807 struct inode *old_inode = old_dentry->d_inode;
8808 struct timespec64 ctime = current_time(old_inode);
8809 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
8810 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
8815 bool root_log_pinned = false;
8816 bool dest_log_pinned = false;
8818 /* we only allow rename subvolume link between subvolumes */
8819 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8822 /* close the race window with snapshot create/destroy ioctl */
8823 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
8824 new_ino == BTRFS_FIRST_FREE_OBJECTID)
8825 down_read(&fs_info->subvol_sem);
8828 * We want to reserve the absolute worst case amount of items. So if
8829 * both inodes are subvols and we need to unlink them then that would
8830 * require 4 item modifications, but if they are both normal inodes it
8831 * would require 5 item modifications, so we'll assume their normal
8832 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
8833 * should cover the worst case number of items we'll modify.
8835 trans = btrfs_start_transaction(root, 12);
8836 if (IS_ERR(trans)) {
8837 ret = PTR_ERR(trans);
8842 btrfs_record_root_in_trans(trans, dest);
8845 * We need to find a free sequence number both in the source and
8846 * in the destination directory for the exchange.
8848 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
8851 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
8855 BTRFS_I(old_inode)->dir_index = 0ULL;
8856 BTRFS_I(new_inode)->dir_index = 0ULL;
8858 /* Reference for the source. */
8859 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8860 /* force full log commit if subvolume involved. */
8861 btrfs_set_log_full_commit(trans);
8863 btrfs_pin_log_trans(root);
8864 root_log_pinned = true;
8865 ret = btrfs_insert_inode_ref(trans, dest,
8866 new_dentry->d_name.name,
8867 new_dentry->d_name.len,
8869 btrfs_ino(BTRFS_I(new_dir)),
8875 /* And now for the dest. */
8876 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8877 /* force full log commit if subvolume involved. */
8878 btrfs_set_log_full_commit(trans);
8880 btrfs_pin_log_trans(dest);
8881 dest_log_pinned = true;
8882 ret = btrfs_insert_inode_ref(trans, root,
8883 old_dentry->d_name.name,
8884 old_dentry->d_name.len,
8886 btrfs_ino(BTRFS_I(old_dir)),
8892 /* Update inode version and ctime/mtime. */
8893 inode_inc_iversion(old_dir);
8894 inode_inc_iversion(new_dir);
8895 inode_inc_iversion(old_inode);
8896 inode_inc_iversion(new_inode);
8897 old_dir->i_ctime = old_dir->i_mtime = ctime;
8898 new_dir->i_ctime = new_dir->i_mtime = ctime;
8899 old_inode->i_ctime = ctime;
8900 new_inode->i_ctime = ctime;
8902 if (old_dentry->d_parent != new_dentry->d_parent) {
8903 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
8904 BTRFS_I(old_inode), 1);
8905 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
8906 BTRFS_I(new_inode), 1);
8909 /* src is a subvolume */
8910 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8911 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
8912 } else { /* src is an inode */
8913 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
8914 BTRFS_I(old_dentry->d_inode),
8915 old_dentry->d_name.name,
8916 old_dentry->d_name.len);
8918 ret = btrfs_update_inode(trans, root, old_inode);
8921 btrfs_abort_transaction(trans, ret);
8925 /* dest is a subvolume */
8926 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8927 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
8928 } else { /* dest is an inode */
8929 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
8930 BTRFS_I(new_dentry->d_inode),
8931 new_dentry->d_name.name,
8932 new_dentry->d_name.len);
8934 ret = btrfs_update_inode(trans, dest, new_inode);
8937 btrfs_abort_transaction(trans, ret);
8941 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
8942 new_dentry->d_name.name,
8943 new_dentry->d_name.len, 0, old_idx);
8945 btrfs_abort_transaction(trans, ret);
8949 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
8950 old_dentry->d_name.name,
8951 old_dentry->d_name.len, 0, new_idx);
8953 btrfs_abort_transaction(trans, ret);
8957 if (old_inode->i_nlink == 1)
8958 BTRFS_I(old_inode)->dir_index = old_idx;
8959 if (new_inode->i_nlink == 1)
8960 BTRFS_I(new_inode)->dir_index = new_idx;
8962 if (root_log_pinned) {
8963 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
8964 new_dentry->d_parent);
8965 btrfs_end_log_trans(root);
8966 root_log_pinned = false;
8968 if (dest_log_pinned) {
8969 btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir),
8970 old_dentry->d_parent);
8971 btrfs_end_log_trans(dest);
8972 dest_log_pinned = false;
8976 * If we have pinned a log and an error happened, we unpin tasks
8977 * trying to sync the log and force them to fallback to a transaction
8978 * commit if the log currently contains any of the inodes involved in
8979 * this rename operation (to ensure we do not persist a log with an
8980 * inconsistent state for any of these inodes or leading to any
8981 * inconsistencies when replayed). If the transaction was aborted, the
8982 * abortion reason is propagated to userspace when attempting to commit
8983 * the transaction. If the log does not contain any of these inodes, we
8984 * allow the tasks to sync it.
8986 if (ret && (root_log_pinned || dest_log_pinned)) {
8987 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
8988 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
8989 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
8991 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
8992 btrfs_set_log_full_commit(trans);
8994 if (root_log_pinned) {
8995 btrfs_end_log_trans(root);
8996 root_log_pinned = false;
8998 if (dest_log_pinned) {
8999 btrfs_end_log_trans(dest);
9000 dest_log_pinned = false;
9003 ret2 = btrfs_end_transaction(trans);
9004 ret = ret ? ret : ret2;
9006 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
9007 old_ino == BTRFS_FIRST_FREE_OBJECTID)
9008 up_read(&fs_info->subvol_sem);
9013 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9014 struct btrfs_root *root,
9016 struct dentry *dentry)
9019 struct inode *inode;
9023 ret = btrfs_find_free_ino(root, &objectid);
9027 inode = btrfs_new_inode(trans, root, dir,
9028 dentry->d_name.name,
9030 btrfs_ino(BTRFS_I(dir)),
9032 S_IFCHR | WHITEOUT_MODE,
9035 if (IS_ERR(inode)) {
9036 ret = PTR_ERR(inode);
9040 inode->i_op = &btrfs_special_inode_operations;
9041 init_special_inode(inode, inode->i_mode,
9044 ret = btrfs_init_inode_security(trans, inode, dir,
9049 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9050 BTRFS_I(inode), 0, index);
9054 ret = btrfs_update_inode(trans, root, inode);
9056 unlock_new_inode(inode);
9058 inode_dec_link_count(inode);
9064 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9065 struct inode *new_dir, struct dentry *new_dentry,
9068 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9069 struct btrfs_trans_handle *trans;
9070 unsigned int trans_num_items;
9071 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9072 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9073 struct inode *new_inode = d_inode(new_dentry);
9074 struct inode *old_inode = d_inode(old_dentry);
9078 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9079 bool log_pinned = false;
9081 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9084 /* we only allow rename subvolume link between subvolumes */
9085 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9088 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9089 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9092 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9093 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9097 /* check for collisions, even if the name isn't there */
9098 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9099 new_dentry->d_name.name,
9100 new_dentry->d_name.len);
9103 if (ret == -EEXIST) {
9105 * eexist without a new_inode */
9106 if (WARN_ON(!new_inode)) {
9110 /* maybe -EOVERFLOW */
9117 * we're using rename to replace one file with another. Start IO on it
9118 * now so we don't add too much work to the end of the transaction
9120 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9121 filemap_flush(old_inode->i_mapping);
9123 /* close the racy window with snapshot create/destroy ioctl */
9124 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9125 down_read(&fs_info->subvol_sem);
9127 * We want to reserve the absolute worst case amount of items. So if
9128 * both inodes are subvols and we need to unlink them then that would
9129 * require 4 item modifications, but if they are both normal inodes it
9130 * would require 5 item modifications, so we'll assume they are normal
9131 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9132 * should cover the worst case number of items we'll modify.
9133 * If our rename has the whiteout flag, we need more 5 units for the
9134 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9135 * when selinux is enabled).
9137 trans_num_items = 11;
9138 if (flags & RENAME_WHITEOUT)
9139 trans_num_items += 5;
9140 trans = btrfs_start_transaction(root, trans_num_items);
9141 if (IS_ERR(trans)) {
9142 ret = PTR_ERR(trans);
9147 btrfs_record_root_in_trans(trans, dest);
9149 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9153 BTRFS_I(old_inode)->dir_index = 0ULL;
9154 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9155 /* force full log commit if subvolume involved. */
9156 btrfs_set_log_full_commit(trans);
9158 btrfs_pin_log_trans(root);
9160 ret = btrfs_insert_inode_ref(trans, dest,
9161 new_dentry->d_name.name,
9162 new_dentry->d_name.len,
9164 btrfs_ino(BTRFS_I(new_dir)), index);
9169 inode_inc_iversion(old_dir);
9170 inode_inc_iversion(new_dir);
9171 inode_inc_iversion(old_inode);
9172 old_dir->i_ctime = old_dir->i_mtime =
9173 new_dir->i_ctime = new_dir->i_mtime =
9174 old_inode->i_ctime = current_time(old_dir);
9176 if (old_dentry->d_parent != new_dentry->d_parent)
9177 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9178 BTRFS_I(old_inode), 1);
9180 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9181 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9183 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9184 BTRFS_I(d_inode(old_dentry)),
9185 old_dentry->d_name.name,
9186 old_dentry->d_name.len);
9188 ret = btrfs_update_inode(trans, root, old_inode);
9191 btrfs_abort_transaction(trans, ret);
9196 inode_inc_iversion(new_inode);
9197 new_inode->i_ctime = current_time(new_inode);
9198 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9199 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9200 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9201 BUG_ON(new_inode->i_nlink == 0);
9203 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9204 BTRFS_I(d_inode(new_dentry)),
9205 new_dentry->d_name.name,
9206 new_dentry->d_name.len);
9208 if (!ret && new_inode->i_nlink == 0)
9209 ret = btrfs_orphan_add(trans,
9210 BTRFS_I(d_inode(new_dentry)));
9212 btrfs_abort_transaction(trans, ret);
9217 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9218 new_dentry->d_name.name,
9219 new_dentry->d_name.len, 0, index);
9221 btrfs_abort_transaction(trans, ret);
9225 if (old_inode->i_nlink == 1)
9226 BTRFS_I(old_inode)->dir_index = index;
9229 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9230 new_dentry->d_parent);
9231 btrfs_end_log_trans(root);
9235 if (flags & RENAME_WHITEOUT) {
9236 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9240 btrfs_abort_transaction(trans, ret);
9246 * If we have pinned the log and an error happened, we unpin tasks
9247 * trying to sync the log and force them to fallback to a transaction
9248 * commit if the log currently contains any of the inodes involved in
9249 * this rename operation (to ensure we do not persist a log with an
9250 * inconsistent state for any of these inodes or leading to any
9251 * inconsistencies when replayed). If the transaction was aborted, the
9252 * abortion reason is propagated to userspace when attempting to commit
9253 * the transaction. If the log does not contain any of these inodes, we
9254 * allow the tasks to sync it.
9256 if (ret && log_pinned) {
9257 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9258 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9259 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9261 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9262 btrfs_set_log_full_commit(trans);
9264 btrfs_end_log_trans(root);
9267 ret2 = btrfs_end_transaction(trans);
9268 ret = ret ? ret : ret2;
9270 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9271 up_read(&fs_info->subvol_sem);
9276 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9277 struct inode *new_dir, struct dentry *new_dentry,
9280 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9283 if (flags & RENAME_EXCHANGE)
9284 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9287 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9290 struct btrfs_delalloc_work {
9291 struct inode *inode;
9292 struct completion completion;
9293 struct list_head list;
9294 struct btrfs_work work;
9297 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9299 struct btrfs_delalloc_work *delalloc_work;
9300 struct inode *inode;
9302 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9304 inode = delalloc_work->inode;
9305 filemap_flush(inode->i_mapping);
9306 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9307 &BTRFS_I(inode)->runtime_flags))
9308 filemap_flush(inode->i_mapping);
9311 complete(&delalloc_work->completion);
9314 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9316 struct btrfs_delalloc_work *work;
9318 work = kmalloc(sizeof(*work), GFP_NOFS);
9322 init_completion(&work->completion);
9323 INIT_LIST_HEAD(&work->list);
9324 work->inode = inode;
9325 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9331 * some fairly slow code that needs optimization. This walks the list
9332 * of all the inodes with pending delalloc and forces them to disk.
9334 static int start_delalloc_inodes(struct btrfs_root *root, u64 *nr, bool snapshot)
9336 struct btrfs_inode *binode;
9337 struct inode *inode;
9338 struct btrfs_delalloc_work *work, *next;
9339 struct list_head works;
9340 struct list_head splice;
9343 INIT_LIST_HEAD(&works);
9344 INIT_LIST_HEAD(&splice);
9346 mutex_lock(&root->delalloc_mutex);
9347 spin_lock(&root->delalloc_lock);
9348 list_splice_init(&root->delalloc_inodes, &splice);
9349 while (!list_empty(&splice)) {
9350 binode = list_entry(splice.next, struct btrfs_inode,
9353 list_move_tail(&binode->delalloc_inodes,
9354 &root->delalloc_inodes);
9355 inode = igrab(&binode->vfs_inode);
9357 cond_resched_lock(&root->delalloc_lock);
9360 spin_unlock(&root->delalloc_lock);
9363 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9364 &binode->runtime_flags);
9365 work = btrfs_alloc_delalloc_work(inode);
9371 list_add_tail(&work->list, &works);
9372 btrfs_queue_work(root->fs_info->flush_workers,
9374 if (*nr != U64_MAX) {
9380 spin_lock(&root->delalloc_lock);
9382 spin_unlock(&root->delalloc_lock);
9385 list_for_each_entry_safe(work, next, &works, list) {
9386 list_del_init(&work->list);
9387 wait_for_completion(&work->completion);
9391 if (!list_empty(&splice)) {
9392 spin_lock(&root->delalloc_lock);
9393 list_splice_tail(&splice, &root->delalloc_inodes);
9394 spin_unlock(&root->delalloc_lock);
9396 mutex_unlock(&root->delalloc_mutex);
9400 int btrfs_start_delalloc_snapshot(struct btrfs_root *root)
9402 struct btrfs_fs_info *fs_info = root->fs_info;
9405 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9408 return start_delalloc_inodes(root, &nr, true);
9411 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, u64 nr)
9413 struct btrfs_root *root;
9414 struct list_head splice;
9417 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9420 INIT_LIST_HEAD(&splice);
9422 mutex_lock(&fs_info->delalloc_root_mutex);
9423 spin_lock(&fs_info->delalloc_root_lock);
9424 list_splice_init(&fs_info->delalloc_roots, &splice);
9425 while (!list_empty(&splice) && nr) {
9426 root = list_first_entry(&splice, struct btrfs_root,
9428 root = btrfs_grab_root(root);
9430 list_move_tail(&root->delalloc_root,
9431 &fs_info->delalloc_roots);
9432 spin_unlock(&fs_info->delalloc_root_lock);
9434 ret = start_delalloc_inodes(root, &nr, false);
9435 btrfs_put_root(root);
9438 spin_lock(&fs_info->delalloc_root_lock);
9440 spin_unlock(&fs_info->delalloc_root_lock);
9444 if (!list_empty(&splice)) {
9445 spin_lock(&fs_info->delalloc_root_lock);
9446 list_splice_tail(&splice, &fs_info->delalloc_roots);
9447 spin_unlock(&fs_info->delalloc_root_lock);
9449 mutex_unlock(&fs_info->delalloc_root_mutex);
9453 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9454 const char *symname)
9456 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9457 struct btrfs_trans_handle *trans;
9458 struct btrfs_root *root = BTRFS_I(dir)->root;
9459 struct btrfs_path *path;
9460 struct btrfs_key key;
9461 struct inode *inode = NULL;
9468 struct btrfs_file_extent_item *ei;
9469 struct extent_buffer *leaf;
9471 name_len = strlen(symname);
9472 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9473 return -ENAMETOOLONG;
9476 * 2 items for inode item and ref
9477 * 2 items for dir items
9478 * 1 item for updating parent inode item
9479 * 1 item for the inline extent item
9480 * 1 item for xattr if selinux is on
9482 trans = btrfs_start_transaction(root, 7);
9484 return PTR_ERR(trans);
9486 err = btrfs_find_free_ino(root, &objectid);
9490 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9491 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
9492 objectid, S_IFLNK|S_IRWXUGO, &index);
9493 if (IS_ERR(inode)) {
9494 err = PTR_ERR(inode);
9500 * If the active LSM wants to access the inode during
9501 * d_instantiate it needs these. Smack checks to see
9502 * if the filesystem supports xattrs by looking at the
9505 inode->i_fop = &btrfs_file_operations;
9506 inode->i_op = &btrfs_file_inode_operations;
9507 inode->i_mapping->a_ops = &btrfs_aops;
9509 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9513 path = btrfs_alloc_path();
9518 key.objectid = btrfs_ino(BTRFS_I(inode));
9520 key.type = BTRFS_EXTENT_DATA_KEY;
9521 datasize = btrfs_file_extent_calc_inline_size(name_len);
9522 err = btrfs_insert_empty_item(trans, root, path, &key,
9525 btrfs_free_path(path);
9528 leaf = path->nodes[0];
9529 ei = btrfs_item_ptr(leaf, path->slots[0],
9530 struct btrfs_file_extent_item);
9531 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9532 btrfs_set_file_extent_type(leaf, ei,
9533 BTRFS_FILE_EXTENT_INLINE);
9534 btrfs_set_file_extent_encryption(leaf, ei, 0);
9535 btrfs_set_file_extent_compression(leaf, ei, 0);
9536 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9537 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9539 ptr = btrfs_file_extent_inline_start(ei);
9540 write_extent_buffer(leaf, symname, ptr, name_len);
9541 btrfs_mark_buffer_dirty(leaf);
9542 btrfs_free_path(path);
9544 inode->i_op = &btrfs_symlink_inode_operations;
9545 inode_nohighmem(inode);
9546 inode_set_bytes(inode, name_len);
9547 btrfs_i_size_write(BTRFS_I(inode), name_len);
9548 err = btrfs_update_inode(trans, root, inode);
9550 * Last step, add directory indexes for our symlink inode. This is the
9551 * last step to avoid extra cleanup of these indexes if an error happens
9555 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9556 BTRFS_I(inode), 0, index);
9560 d_instantiate_new(dentry, inode);
9563 btrfs_end_transaction(trans);
9565 inode_dec_link_count(inode);
9566 discard_new_inode(inode);
9568 btrfs_btree_balance_dirty(fs_info);
9572 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9573 struct btrfs_trans_handle *trans_in,
9574 struct inode *inode, struct btrfs_key *ins,
9577 struct btrfs_file_extent_item stack_fi;
9578 struct btrfs_replace_extent_info extent_info;
9579 struct btrfs_trans_handle *trans = trans_in;
9580 struct btrfs_path *path;
9581 u64 start = ins->objectid;
9582 u64 len = ins->offset;
9585 memset(&stack_fi, 0, sizeof(stack_fi));
9587 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9588 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9589 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9590 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9591 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9592 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9593 /* Encryption and other encoding is reserved and all 0 */
9595 ret = btrfs_qgroup_release_data(BTRFS_I(inode), file_offset, len);
9597 return ERR_PTR(ret);
9600 ret = insert_reserved_file_extent(trans, BTRFS_I(inode),
9601 file_offset, &stack_fi, ret);
9603 return ERR_PTR(ret);
9607 extent_info.disk_offset = start;
9608 extent_info.disk_len = len;
9609 extent_info.data_offset = 0;
9610 extent_info.data_len = len;
9611 extent_info.file_offset = file_offset;
9612 extent_info.extent_buf = (char *)&stack_fi;
9613 extent_info.is_new_extent = true;
9614 extent_info.qgroup_reserved = ret;
9615 extent_info.insertions = 0;
9617 path = btrfs_alloc_path();
9619 return ERR_PTR(-ENOMEM);
9621 ret = btrfs_replace_file_extents(inode, path, file_offset,
9622 file_offset + len - 1, &extent_info,
9624 btrfs_free_path(path);
9626 return ERR_PTR(ret);
9631 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9632 u64 start, u64 num_bytes, u64 min_size,
9633 loff_t actual_len, u64 *alloc_hint,
9634 struct btrfs_trans_handle *trans)
9636 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9637 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9638 struct extent_map *em;
9639 struct btrfs_root *root = BTRFS_I(inode)->root;
9640 struct btrfs_key ins;
9641 u64 cur_offset = start;
9642 u64 clear_offset = start;
9645 u64 last_alloc = (u64)-1;
9647 bool own_trans = true;
9648 u64 end = start + num_bytes - 1;
9652 while (num_bytes > 0) {
9653 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9654 cur_bytes = max(cur_bytes, min_size);
9656 * If we are severely fragmented we could end up with really
9657 * small allocations, so if the allocator is returning small
9658 * chunks lets make its job easier by only searching for those
9661 cur_bytes = min(cur_bytes, last_alloc);
9662 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9663 min_size, 0, *alloc_hint, &ins, 1, 0);
9668 * We've reserved this space, and thus converted it from
9669 * ->bytes_may_use to ->bytes_reserved. Any error that happens
9670 * from here on out we will only need to clear our reservation
9671 * for the remaining unreserved area, so advance our
9672 * clear_offset by our extent size.
9674 clear_offset += ins.offset;
9675 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
9677 last_alloc = ins.offset;
9678 trans = insert_prealloc_file_extent(trans, inode, &ins, cur_offset);
9679 if (IS_ERR(trans)) {
9680 ret = PTR_ERR(trans);
9681 btrfs_free_reserved_extent(fs_info, ins.objectid,
9686 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
9687 cur_offset + ins.offset -1, 0);
9689 em = alloc_extent_map();
9691 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9692 &BTRFS_I(inode)->runtime_flags);
9696 em->start = cur_offset;
9697 em->orig_start = cur_offset;
9698 em->len = ins.offset;
9699 em->block_start = ins.objectid;
9700 em->block_len = ins.offset;
9701 em->orig_block_len = ins.offset;
9702 em->ram_bytes = ins.offset;
9703 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9704 em->generation = trans->transid;
9707 write_lock(&em_tree->lock);
9708 ret = add_extent_mapping(em_tree, em, 1);
9709 write_unlock(&em_tree->lock);
9712 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
9713 cur_offset + ins.offset - 1,
9716 free_extent_map(em);
9718 num_bytes -= ins.offset;
9719 cur_offset += ins.offset;
9720 *alloc_hint = ins.objectid + ins.offset;
9722 inode_inc_iversion(inode);
9723 inode->i_ctime = current_time(inode);
9724 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9725 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9726 (actual_len > inode->i_size) &&
9727 (cur_offset > inode->i_size)) {
9728 if (cur_offset > actual_len)
9729 i_size = actual_len;
9731 i_size = cur_offset;
9732 i_size_write(inode, i_size);
9733 btrfs_inode_safe_disk_i_size_write(inode, 0);
9736 ret = btrfs_update_inode(trans, root, inode);
9739 btrfs_abort_transaction(trans, ret);
9741 btrfs_end_transaction(trans);
9746 btrfs_end_transaction(trans);
9750 if (clear_offset < end)
9751 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
9752 end - clear_offset + 1);
9756 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9757 u64 start, u64 num_bytes, u64 min_size,
9758 loff_t actual_len, u64 *alloc_hint)
9760 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9761 min_size, actual_len, alloc_hint,
9765 int btrfs_prealloc_file_range_trans(struct inode *inode,
9766 struct btrfs_trans_handle *trans, int mode,
9767 u64 start, u64 num_bytes, u64 min_size,
9768 loff_t actual_len, u64 *alloc_hint)
9770 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9771 min_size, actual_len, alloc_hint, trans);
9774 static int btrfs_set_page_dirty(struct page *page)
9776 return __set_page_dirty_nobuffers(page);
9779 static int btrfs_permission(struct inode *inode, int mask)
9781 struct btrfs_root *root = BTRFS_I(inode)->root;
9782 umode_t mode = inode->i_mode;
9784 if (mask & MAY_WRITE &&
9785 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9786 if (btrfs_root_readonly(root))
9788 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9791 return generic_permission(inode, mask);
9794 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
9796 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9797 struct btrfs_trans_handle *trans;
9798 struct btrfs_root *root = BTRFS_I(dir)->root;
9799 struct inode *inode = NULL;
9805 * 5 units required for adding orphan entry
9807 trans = btrfs_start_transaction(root, 5);
9809 return PTR_ERR(trans);
9811 ret = btrfs_find_free_ino(root, &objectid);
9815 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
9816 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
9817 if (IS_ERR(inode)) {
9818 ret = PTR_ERR(inode);
9823 inode->i_fop = &btrfs_file_operations;
9824 inode->i_op = &btrfs_file_inode_operations;
9826 inode->i_mapping->a_ops = &btrfs_aops;
9828 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
9832 ret = btrfs_update_inode(trans, root, inode);
9835 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
9840 * We set number of links to 0 in btrfs_new_inode(), and here we set
9841 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9844 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9846 set_nlink(inode, 1);
9847 d_tmpfile(dentry, inode);
9848 unlock_new_inode(inode);
9849 mark_inode_dirty(inode);
9851 btrfs_end_transaction(trans);
9853 discard_new_inode(inode);
9854 btrfs_btree_balance_dirty(fs_info);
9858 void btrfs_set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
9860 struct inode *inode = tree->private_data;
9861 unsigned long index = start >> PAGE_SHIFT;
9862 unsigned long end_index = end >> PAGE_SHIFT;
9865 while (index <= end_index) {
9866 page = find_get_page(inode->i_mapping, index);
9867 ASSERT(page); /* Pages should be in the extent_io_tree */
9868 set_page_writeback(page);
9876 * Add an entry indicating a block group or device which is pinned by a
9877 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
9878 * negative errno on failure.
9880 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
9881 bool is_block_group)
9883 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
9884 struct btrfs_swapfile_pin *sp, *entry;
9886 struct rb_node *parent = NULL;
9888 sp = kmalloc(sizeof(*sp), GFP_NOFS);
9893 sp->is_block_group = is_block_group;
9895 spin_lock(&fs_info->swapfile_pins_lock);
9896 p = &fs_info->swapfile_pins.rb_node;
9899 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
9900 if (sp->ptr < entry->ptr ||
9901 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
9903 } else if (sp->ptr > entry->ptr ||
9904 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
9905 p = &(*p)->rb_right;
9907 spin_unlock(&fs_info->swapfile_pins_lock);
9912 rb_link_node(&sp->node, parent, p);
9913 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
9914 spin_unlock(&fs_info->swapfile_pins_lock);
9918 /* Free all of the entries pinned by this swapfile. */
9919 static void btrfs_free_swapfile_pins(struct inode *inode)
9921 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
9922 struct btrfs_swapfile_pin *sp;
9923 struct rb_node *node, *next;
9925 spin_lock(&fs_info->swapfile_pins_lock);
9926 node = rb_first(&fs_info->swapfile_pins);
9928 next = rb_next(node);
9929 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
9930 if (sp->inode == inode) {
9931 rb_erase(&sp->node, &fs_info->swapfile_pins);
9932 if (sp->is_block_group)
9933 btrfs_put_block_group(sp->ptr);
9938 spin_unlock(&fs_info->swapfile_pins_lock);
9941 struct btrfs_swap_info {
9947 unsigned long nr_pages;
9951 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
9952 struct btrfs_swap_info *bsi)
9954 unsigned long nr_pages;
9955 u64 first_ppage, first_ppage_reported, next_ppage;
9958 first_ppage = ALIGN(bsi->block_start, PAGE_SIZE) >> PAGE_SHIFT;
9959 next_ppage = ALIGN_DOWN(bsi->block_start + bsi->block_len,
9960 PAGE_SIZE) >> PAGE_SHIFT;
9962 if (first_ppage >= next_ppage)
9964 nr_pages = next_ppage - first_ppage;
9966 first_ppage_reported = first_ppage;
9967 if (bsi->start == 0)
9968 first_ppage_reported++;
9969 if (bsi->lowest_ppage > first_ppage_reported)
9970 bsi->lowest_ppage = first_ppage_reported;
9971 if (bsi->highest_ppage < (next_ppage - 1))
9972 bsi->highest_ppage = next_ppage - 1;
9974 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
9977 bsi->nr_extents += ret;
9978 bsi->nr_pages += nr_pages;
9982 static void btrfs_swap_deactivate(struct file *file)
9984 struct inode *inode = file_inode(file);
9986 btrfs_free_swapfile_pins(inode);
9987 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
9990 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
9993 struct inode *inode = file_inode(file);
9994 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
9995 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
9996 struct extent_state *cached_state = NULL;
9997 struct extent_map *em = NULL;
9998 struct btrfs_device *device = NULL;
9999 struct btrfs_swap_info bsi = {
10000 .lowest_ppage = (sector_t)-1ULL,
10007 * If the swap file was just created, make sure delalloc is done. If the
10008 * file changes again after this, the user is doing something stupid and
10009 * we don't really care.
10011 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10016 * The inode is locked, so these flags won't change after we check them.
10018 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10019 btrfs_warn(fs_info, "swapfile must not be compressed");
10022 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10023 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10026 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10027 btrfs_warn(fs_info, "swapfile must not be checksummed");
10032 * Balance or device remove/replace/resize can move stuff around from
10033 * under us. The exclop protection makes sure they aren't running/won't
10034 * run concurrently while we are mapping the swap extents, and
10035 * fs_info->swapfile_pins prevents them from running while the swap
10036 * file is active and moving the extents. Note that this also prevents
10037 * a concurrent device add which isn't actually necessary, but it's not
10038 * really worth the trouble to allow it.
10040 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10041 btrfs_warn(fs_info,
10042 "cannot activate swapfile while exclusive operation is running");
10046 * Snapshots can create extents which require COW even if NODATACOW is
10047 * set. We use this counter to prevent snapshots. We must increment it
10048 * before walking the extents because we don't want a concurrent
10049 * snapshot to run after we've already checked the extents.
10051 atomic_inc(&BTRFS_I(inode)->root->nr_swapfiles);
10053 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10055 lock_extent_bits(io_tree, 0, isize - 1, &cached_state);
10057 while (start < isize) {
10058 u64 logical_block_start, physical_block_start;
10059 struct btrfs_block_group *bg;
10060 u64 len = isize - start;
10062 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
10068 if (em->block_start == EXTENT_MAP_HOLE) {
10069 btrfs_warn(fs_info, "swapfile must not have holes");
10073 if (em->block_start == EXTENT_MAP_INLINE) {
10075 * It's unlikely we'll ever actually find ourselves
10076 * here, as a file small enough to fit inline won't be
10077 * big enough to store more than the swap header, but in
10078 * case something changes in the future, let's catch it
10079 * here rather than later.
10081 btrfs_warn(fs_info, "swapfile must not be inline");
10085 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10086 btrfs_warn(fs_info, "swapfile must not be compressed");
10091 logical_block_start = em->block_start + (start - em->start);
10092 len = min(len, em->len - (start - em->start));
10093 free_extent_map(em);
10096 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, true);
10102 btrfs_warn(fs_info,
10103 "swapfile must not be copy-on-write");
10108 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10114 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10115 btrfs_warn(fs_info,
10116 "swapfile must have single data profile");
10121 if (device == NULL) {
10122 device = em->map_lookup->stripes[0].dev;
10123 ret = btrfs_add_swapfile_pin(inode, device, false);
10128 } else if (device != em->map_lookup->stripes[0].dev) {
10129 btrfs_warn(fs_info, "swapfile must be on one device");
10134 physical_block_start = (em->map_lookup->stripes[0].physical +
10135 (logical_block_start - em->start));
10136 len = min(len, em->len - (logical_block_start - em->start));
10137 free_extent_map(em);
10140 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10142 btrfs_warn(fs_info,
10143 "could not find block group containing swapfile");
10148 ret = btrfs_add_swapfile_pin(inode, bg, true);
10150 btrfs_put_block_group(bg);
10157 if (bsi.block_len &&
10158 bsi.block_start + bsi.block_len == physical_block_start) {
10159 bsi.block_len += len;
10161 if (bsi.block_len) {
10162 ret = btrfs_add_swap_extent(sis, &bsi);
10167 bsi.block_start = physical_block_start;
10168 bsi.block_len = len;
10175 ret = btrfs_add_swap_extent(sis, &bsi);
10178 if (!IS_ERR_OR_NULL(em))
10179 free_extent_map(em);
10181 unlock_extent_cached(io_tree, 0, isize - 1, &cached_state);
10184 btrfs_swap_deactivate(file);
10186 btrfs_exclop_finish(fs_info);
10192 sis->bdev = device->bdev;
10193 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10194 sis->max = bsi.nr_pages;
10195 sis->pages = bsi.nr_pages - 1;
10196 sis->highest_bit = bsi.nr_pages - 1;
10197 return bsi.nr_extents;
10200 static void btrfs_swap_deactivate(struct file *file)
10204 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10207 return -EOPNOTSUPP;
10211 static const struct inode_operations btrfs_dir_inode_operations = {
10212 .getattr = btrfs_getattr,
10213 .lookup = btrfs_lookup,
10214 .create = btrfs_create,
10215 .unlink = btrfs_unlink,
10216 .link = btrfs_link,
10217 .mkdir = btrfs_mkdir,
10218 .rmdir = btrfs_rmdir,
10219 .rename = btrfs_rename2,
10220 .symlink = btrfs_symlink,
10221 .setattr = btrfs_setattr,
10222 .mknod = btrfs_mknod,
10223 .listxattr = btrfs_listxattr,
10224 .permission = btrfs_permission,
10225 .get_acl = btrfs_get_acl,
10226 .set_acl = btrfs_set_acl,
10227 .update_time = btrfs_update_time,
10228 .tmpfile = btrfs_tmpfile,
10231 static const struct file_operations btrfs_dir_file_operations = {
10232 .llseek = generic_file_llseek,
10233 .read = generic_read_dir,
10234 .iterate_shared = btrfs_real_readdir,
10235 .open = btrfs_opendir,
10236 .unlocked_ioctl = btrfs_ioctl,
10237 #ifdef CONFIG_COMPAT
10238 .compat_ioctl = btrfs_compat_ioctl,
10240 .release = btrfs_release_file,
10241 .fsync = btrfs_sync_file,
10245 * btrfs doesn't support the bmap operation because swapfiles
10246 * use bmap to make a mapping of extents in the file. They assume
10247 * these extents won't change over the life of the file and they
10248 * use the bmap result to do IO directly to the drive.
10250 * the btrfs bmap call would return logical addresses that aren't
10251 * suitable for IO and they also will change frequently as COW
10252 * operations happen. So, swapfile + btrfs == corruption.
10254 * For now we're avoiding this by dropping bmap.
10256 static const struct address_space_operations btrfs_aops = {
10257 .readpage = btrfs_readpage,
10258 .writepage = btrfs_writepage,
10259 .writepages = btrfs_writepages,
10260 .readahead = btrfs_readahead,
10261 .direct_IO = noop_direct_IO,
10262 .invalidatepage = btrfs_invalidatepage,
10263 .releasepage = btrfs_releasepage,
10264 #ifdef CONFIG_MIGRATION
10265 .migratepage = btrfs_migratepage,
10267 .set_page_dirty = btrfs_set_page_dirty,
10268 .error_remove_page = generic_error_remove_page,
10269 .swap_activate = btrfs_swap_activate,
10270 .swap_deactivate = btrfs_swap_deactivate,
10273 static const struct inode_operations btrfs_file_inode_operations = {
10274 .getattr = btrfs_getattr,
10275 .setattr = btrfs_setattr,
10276 .listxattr = btrfs_listxattr,
10277 .permission = btrfs_permission,
10278 .fiemap = btrfs_fiemap,
10279 .get_acl = btrfs_get_acl,
10280 .set_acl = btrfs_set_acl,
10281 .update_time = btrfs_update_time,
10283 static const struct inode_operations btrfs_special_inode_operations = {
10284 .getattr = btrfs_getattr,
10285 .setattr = btrfs_setattr,
10286 .permission = btrfs_permission,
10287 .listxattr = btrfs_listxattr,
10288 .get_acl = btrfs_get_acl,
10289 .set_acl = btrfs_set_acl,
10290 .update_time = btrfs_update_time,
10292 static const struct inode_operations btrfs_symlink_inode_operations = {
10293 .get_link = page_get_link,
10294 .getattr = btrfs_getattr,
10295 .setattr = btrfs_setattr,
10296 .permission = btrfs_permission,
10297 .listxattr = btrfs_listxattr,
10298 .update_time = btrfs_update_time,
10301 const struct dentry_operations btrfs_dentry_operations = {
10302 .d_delete = btrfs_dentry_delete,