1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <linux/kernel.h>
8 #include <linux/buffer_head.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/sched/mm.h>
32 #include <asm/unaligned.h>
36 #include "transaction.h"
37 #include "btrfs_inode.h"
38 #include "print-tree.h"
39 #include "ordered-data.h"
43 #include "compression.h"
45 #include "free-space-cache.h"
46 #include "inode-map.h"
49 #include "delalloc-space.h"
50 #include "block-group.h"
52 struct btrfs_iget_args {
53 struct btrfs_key *location;
54 struct btrfs_root *root;
57 struct btrfs_dio_data {
59 u64 unsubmitted_oe_range_start;
60 u64 unsubmitted_oe_range_end;
64 static const struct inode_operations btrfs_dir_inode_operations;
65 static const struct inode_operations btrfs_symlink_inode_operations;
66 static const struct inode_operations btrfs_special_inode_operations;
67 static const struct inode_operations btrfs_file_inode_operations;
68 static const struct address_space_operations btrfs_aops;
69 static const struct file_operations btrfs_dir_file_operations;
70 static const struct extent_io_ops btrfs_extent_io_ops;
72 static struct kmem_cache *btrfs_inode_cachep;
73 struct kmem_cache *btrfs_trans_handle_cachep;
74 struct kmem_cache *btrfs_path_cachep;
75 struct kmem_cache *btrfs_free_space_cachep;
76 struct kmem_cache *btrfs_free_space_bitmap_cachep;
78 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
79 static int btrfs_truncate(struct inode *inode, bool skip_writeback);
80 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
81 static noinline int cow_file_range(struct inode *inode,
82 struct page *locked_page,
83 u64 start, u64 end, int *page_started,
84 unsigned long *nr_written, int unlock);
85 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
86 u64 orig_start, u64 block_start,
87 u64 block_len, u64 orig_block_len,
88 u64 ram_bytes, int compress_type,
91 static void __endio_write_update_ordered(struct inode *inode,
92 const u64 offset, const u64 bytes,
96 * Cleanup all submitted ordered extents in specified range to handle errors
97 * from the btrfs_run_delalloc_range() callback.
99 * NOTE: caller must ensure that when an error happens, it can not call
100 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
101 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
102 * to be released, which we want to happen only when finishing the ordered
103 * extent (btrfs_finish_ordered_io()).
105 static inline void btrfs_cleanup_ordered_extents(struct inode *inode,
106 struct page *locked_page,
107 u64 offset, u64 bytes)
109 unsigned long index = offset >> PAGE_SHIFT;
110 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
111 u64 page_start = page_offset(locked_page);
112 u64 page_end = page_start + PAGE_SIZE - 1;
116 while (index <= end_index) {
117 page = find_get_page(inode->i_mapping, index);
121 ClearPagePrivate2(page);
126 * In case this page belongs to the delalloc range being instantiated
127 * then skip it, since the first page of a range is going to be
128 * properly cleaned up by the caller of run_delalloc_range
130 if (page_start >= offset && page_end <= (offset + bytes - 1)) {
135 return __endio_write_update_ordered(inode, offset, bytes, false);
138 static int btrfs_dirty_inode(struct inode *inode);
140 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
141 void btrfs_test_inode_set_ops(struct inode *inode)
143 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
147 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
148 struct inode *inode, struct inode *dir,
149 const struct qstr *qstr)
153 err = btrfs_init_acl(trans, inode, dir);
155 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
160 * this does all the hard work for inserting an inline extent into
161 * the btree. The caller should have done a btrfs_drop_extents so that
162 * no overlapping inline items exist in the btree
164 static int insert_inline_extent(struct btrfs_trans_handle *trans,
165 struct btrfs_path *path, int extent_inserted,
166 struct btrfs_root *root, struct inode *inode,
167 u64 start, size_t size, size_t compressed_size,
169 struct page **compressed_pages)
171 struct extent_buffer *leaf;
172 struct page *page = NULL;
175 struct btrfs_file_extent_item *ei;
177 size_t cur_size = size;
178 unsigned long offset;
180 ASSERT((compressed_size > 0 && compressed_pages) ||
181 (compressed_size == 0 && !compressed_pages));
183 if (compressed_size && compressed_pages)
184 cur_size = compressed_size;
186 inode_add_bytes(inode, size);
188 if (!extent_inserted) {
189 struct btrfs_key key;
192 key.objectid = btrfs_ino(BTRFS_I(inode));
194 key.type = BTRFS_EXTENT_DATA_KEY;
196 datasize = btrfs_file_extent_calc_inline_size(cur_size);
197 path->leave_spinning = 1;
198 ret = btrfs_insert_empty_item(trans, root, path, &key,
203 leaf = path->nodes[0];
204 ei = btrfs_item_ptr(leaf, path->slots[0],
205 struct btrfs_file_extent_item);
206 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
207 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
208 btrfs_set_file_extent_encryption(leaf, ei, 0);
209 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
210 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
211 ptr = btrfs_file_extent_inline_start(ei);
213 if (compress_type != BTRFS_COMPRESS_NONE) {
216 while (compressed_size > 0) {
217 cpage = compressed_pages[i];
218 cur_size = min_t(unsigned long, compressed_size,
221 kaddr = kmap_atomic(cpage);
222 write_extent_buffer(leaf, kaddr, ptr, cur_size);
223 kunmap_atomic(kaddr);
227 compressed_size -= cur_size;
229 btrfs_set_file_extent_compression(leaf, ei,
232 page = find_get_page(inode->i_mapping,
233 start >> PAGE_SHIFT);
234 btrfs_set_file_extent_compression(leaf, ei, 0);
235 kaddr = kmap_atomic(page);
236 offset = offset_in_page(start);
237 write_extent_buffer(leaf, kaddr + offset, ptr, size);
238 kunmap_atomic(kaddr);
241 btrfs_mark_buffer_dirty(leaf);
242 btrfs_release_path(path);
245 * we're an inline extent, so nobody can
246 * extend the file past i_size without locking
247 * a page we already have locked.
249 * We must do any isize and inode updates
250 * before we unlock the pages. Otherwise we
251 * could end up racing with unlink.
253 BTRFS_I(inode)->disk_i_size = inode->i_size;
254 ret = btrfs_update_inode(trans, root, inode);
262 * conditionally insert an inline extent into the file. This
263 * does the checks required to make sure the data is small enough
264 * to fit as an inline extent.
266 static noinline int cow_file_range_inline(struct inode *inode, u64 start,
267 u64 end, size_t compressed_size,
269 struct page **compressed_pages)
271 struct btrfs_root *root = BTRFS_I(inode)->root;
272 struct btrfs_fs_info *fs_info = root->fs_info;
273 struct btrfs_trans_handle *trans;
274 u64 isize = i_size_read(inode);
275 u64 actual_end = min(end + 1, isize);
276 u64 inline_len = actual_end - start;
277 u64 aligned_end = ALIGN(end, fs_info->sectorsize);
278 u64 data_len = inline_len;
280 struct btrfs_path *path;
281 int extent_inserted = 0;
282 u32 extent_item_size;
285 data_len = compressed_size;
288 actual_end > fs_info->sectorsize ||
289 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
291 (actual_end & (fs_info->sectorsize - 1)) == 0) ||
293 data_len > fs_info->max_inline) {
297 path = btrfs_alloc_path();
301 trans = btrfs_join_transaction(root);
303 btrfs_free_path(path);
304 return PTR_ERR(trans);
306 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
308 if (compressed_size && compressed_pages)
309 extent_item_size = btrfs_file_extent_calc_inline_size(
312 extent_item_size = btrfs_file_extent_calc_inline_size(
315 ret = __btrfs_drop_extents(trans, root, inode, path,
316 start, aligned_end, NULL,
317 1, 1, extent_item_size, &extent_inserted);
319 btrfs_abort_transaction(trans, ret);
323 if (isize > actual_end)
324 inline_len = min_t(u64, isize, actual_end);
325 ret = insert_inline_extent(trans, path, extent_inserted,
327 inline_len, compressed_size,
328 compress_type, compressed_pages);
329 if (ret && ret != -ENOSPC) {
330 btrfs_abort_transaction(trans, ret);
332 } else if (ret == -ENOSPC) {
337 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
338 btrfs_drop_extent_cache(BTRFS_I(inode), start, aligned_end - 1, 0);
341 * Don't forget to free the reserved space, as for inlined extent
342 * it won't count as data extent, free them directly here.
343 * And at reserve time, it's always aligned to page size, so
344 * just free one page here.
346 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
347 btrfs_free_path(path);
348 btrfs_end_transaction(trans);
352 struct async_extent {
357 unsigned long nr_pages;
359 struct list_head list;
364 struct page *locked_page;
367 unsigned int write_flags;
368 struct list_head extents;
369 struct cgroup_subsys_state *blkcg_css;
370 struct btrfs_work work;
375 /* Number of chunks in flight; must be first in the structure */
377 struct async_chunk chunks[];
380 static noinline int add_async_extent(struct async_chunk *cow,
381 u64 start, u64 ram_size,
384 unsigned long nr_pages,
387 struct async_extent *async_extent;
389 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
390 BUG_ON(!async_extent); /* -ENOMEM */
391 async_extent->start = start;
392 async_extent->ram_size = ram_size;
393 async_extent->compressed_size = compressed_size;
394 async_extent->pages = pages;
395 async_extent->nr_pages = nr_pages;
396 async_extent->compress_type = compress_type;
397 list_add_tail(&async_extent->list, &cow->extents);
402 * Check if the inode has flags compatible with compression
404 static inline bool inode_can_compress(struct inode *inode)
406 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW ||
407 BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
413 * Check if the inode needs to be submitted to compression, based on mount
414 * options, defragmentation, properties or heuristics.
416 static inline int inode_need_compress(struct inode *inode, u64 start, u64 end)
418 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
420 if (!inode_can_compress(inode)) {
421 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
422 KERN_ERR "BTRFS: unexpected compression for ino %llu\n",
423 btrfs_ino(BTRFS_I(inode)));
427 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
430 if (BTRFS_I(inode)->defrag_compress)
432 /* bad compression ratios */
433 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
435 if (btrfs_test_opt(fs_info, COMPRESS) ||
436 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
437 BTRFS_I(inode)->prop_compress)
438 return btrfs_compress_heuristic(inode, start, end);
442 static inline void inode_should_defrag(struct btrfs_inode *inode,
443 u64 start, u64 end, u64 num_bytes, u64 small_write)
445 /* If this is a small write inside eof, kick off a defrag */
446 if (num_bytes < small_write &&
447 (start > 0 || end + 1 < inode->disk_i_size))
448 btrfs_add_inode_defrag(NULL, inode);
452 * we create compressed extents in two phases. The first
453 * phase compresses a range of pages that have already been
454 * locked (both pages and state bits are locked).
456 * This is done inside an ordered work queue, and the compression
457 * is spread across many cpus. The actual IO submission is step
458 * two, and the ordered work queue takes care of making sure that
459 * happens in the same order things were put onto the queue by
460 * writepages and friends.
462 * If this code finds it can't get good compression, it puts an
463 * entry onto the work queue to write the uncompressed bytes. This
464 * makes sure that both compressed inodes and uncompressed inodes
465 * are written in the same order that the flusher thread sent them
468 static noinline int compress_file_range(struct async_chunk *async_chunk)
470 struct inode *inode = async_chunk->inode;
471 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
472 u64 blocksize = fs_info->sectorsize;
473 u64 start = async_chunk->start;
474 u64 end = async_chunk->end;
478 struct page **pages = NULL;
479 unsigned long nr_pages;
480 unsigned long total_compressed = 0;
481 unsigned long total_in = 0;
484 int compress_type = fs_info->compress_type;
485 int compressed_extents = 0;
488 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
492 * We need to save i_size before now because it could change in between
493 * us evaluating the size and assigning it. This is because we lock and
494 * unlock the page in truncate and fallocate, and then modify the i_size
497 * The barriers are to emulate READ_ONCE, remove that once i_size_read
501 i_size = i_size_read(inode);
503 actual_end = min_t(u64, i_size, end + 1);
506 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
507 BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
508 nr_pages = min_t(unsigned long, nr_pages,
509 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
512 * we don't want to send crud past the end of i_size through
513 * compression, that's just a waste of CPU time. So, if the
514 * end of the file is before the start of our current
515 * requested range of bytes, we bail out to the uncompressed
516 * cleanup code that can deal with all of this.
518 * It isn't really the fastest way to fix things, but this is a
519 * very uncommon corner.
521 if (actual_end <= start)
522 goto cleanup_and_bail_uncompressed;
524 total_compressed = actual_end - start;
527 * skip compression for a small file range(<=blocksize) that
528 * isn't an inline extent, since it doesn't save disk space at all.
530 if (total_compressed <= blocksize &&
531 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
532 goto cleanup_and_bail_uncompressed;
534 total_compressed = min_t(unsigned long, total_compressed,
535 BTRFS_MAX_UNCOMPRESSED);
540 * we do compression for mount -o compress and when the
541 * inode has not been flagged as nocompress. This flag can
542 * change at any time if we discover bad compression ratios.
544 if (inode_need_compress(inode, start, end)) {
546 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
548 /* just bail out to the uncompressed code */
553 if (BTRFS_I(inode)->defrag_compress)
554 compress_type = BTRFS_I(inode)->defrag_compress;
555 else if (BTRFS_I(inode)->prop_compress)
556 compress_type = BTRFS_I(inode)->prop_compress;
559 * we need to call clear_page_dirty_for_io on each
560 * page in the range. Otherwise applications with the file
561 * mmap'd can wander in and change the page contents while
562 * we are compressing them.
564 * If the compression fails for any reason, we set the pages
565 * dirty again later on.
567 * Note that the remaining part is redirtied, the start pointer
568 * has moved, the end is the original one.
571 extent_range_clear_dirty_for_io(inode, start, end);
575 /* Compression level is applied here and only here */
576 ret = btrfs_compress_pages(
577 compress_type | (fs_info->compress_level << 4),
578 inode->i_mapping, start,
585 unsigned long offset = offset_in_page(total_compressed);
586 struct page *page = pages[nr_pages - 1];
589 /* zero the tail end of the last page, we might be
590 * sending it down to disk
593 kaddr = kmap_atomic(page);
594 memset(kaddr + offset, 0,
596 kunmap_atomic(kaddr);
603 /* lets try to make an inline extent */
604 if (ret || total_in < actual_end) {
605 /* we didn't compress the entire range, try
606 * to make an uncompressed inline extent.
608 ret = cow_file_range_inline(inode, start, end, 0,
609 BTRFS_COMPRESS_NONE, NULL);
611 /* try making a compressed inline extent */
612 ret = cow_file_range_inline(inode, start, end,
614 compress_type, pages);
617 unsigned long clear_flags = EXTENT_DELALLOC |
618 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
619 EXTENT_DO_ACCOUNTING;
620 unsigned long page_error_op;
622 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
625 * inline extent creation worked or returned error,
626 * we don't need to create any more async work items.
627 * Unlock and free up our temp pages.
629 * We use DO_ACCOUNTING here because we need the
630 * delalloc_release_metadata to be done _after_ we drop
631 * our outstanding extent for clearing delalloc for this
634 extent_clear_unlock_delalloc(inode, start, end, NULL,
642 for (i = 0; i < nr_pages; i++) {
643 WARN_ON(pages[i]->mapping);
654 * we aren't doing an inline extent round the compressed size
655 * up to a block size boundary so the allocator does sane
658 total_compressed = ALIGN(total_compressed, blocksize);
661 * one last check to make sure the compression is really a
662 * win, compare the page count read with the blocks on disk,
663 * compression must free at least one sector size
665 total_in = ALIGN(total_in, PAGE_SIZE);
666 if (total_compressed + blocksize <= total_in) {
667 compressed_extents++;
670 * The async work queues will take care of doing actual
671 * allocation on disk for these compressed pages, and
672 * will submit them to the elevator.
674 add_async_extent(async_chunk, start, total_in,
675 total_compressed, pages, nr_pages,
678 if (start + total_in < end) {
684 return compressed_extents;
689 * the compression code ran but failed to make things smaller,
690 * free any pages it allocated and our page pointer array
692 for (i = 0; i < nr_pages; i++) {
693 WARN_ON(pages[i]->mapping);
698 total_compressed = 0;
701 /* flag the file so we don't compress in the future */
702 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
703 !(BTRFS_I(inode)->prop_compress)) {
704 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
707 cleanup_and_bail_uncompressed:
709 * No compression, but we still need to write the pages in the file
710 * we've been given so far. redirty the locked page if it corresponds
711 * to our extent and set things up for the async work queue to run
712 * cow_file_range to do the normal delalloc dance.
714 if (async_chunk->locked_page &&
715 (page_offset(async_chunk->locked_page) >= start &&
716 page_offset(async_chunk->locked_page)) <= end) {
717 __set_page_dirty_nobuffers(async_chunk->locked_page);
718 /* unlocked later on in the async handlers */
722 extent_range_redirty_for_io(inode, start, end);
723 add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
724 BTRFS_COMPRESS_NONE);
725 compressed_extents++;
727 return compressed_extents;
730 static void free_async_extent_pages(struct async_extent *async_extent)
734 if (!async_extent->pages)
737 for (i = 0; i < async_extent->nr_pages; i++) {
738 WARN_ON(async_extent->pages[i]->mapping);
739 put_page(async_extent->pages[i]);
741 kfree(async_extent->pages);
742 async_extent->nr_pages = 0;
743 async_extent->pages = NULL;
747 * phase two of compressed writeback. This is the ordered portion
748 * of the code, which only gets called in the order the work was
749 * queued. We walk all the async extents created by compress_file_range
750 * and send them down to the disk.
752 static noinline void submit_compressed_extents(struct async_chunk *async_chunk)
754 struct inode *inode = async_chunk->inode;
755 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
756 struct async_extent *async_extent;
758 struct btrfs_key ins;
759 struct extent_map *em;
760 struct btrfs_root *root = BTRFS_I(inode)->root;
761 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
765 while (!list_empty(&async_chunk->extents)) {
766 async_extent = list_entry(async_chunk->extents.next,
767 struct async_extent, list);
768 list_del(&async_extent->list);
771 lock_extent(io_tree, async_extent->start,
772 async_extent->start + async_extent->ram_size - 1);
773 /* did the compression code fall back to uncompressed IO? */
774 if (!async_extent->pages) {
775 int page_started = 0;
776 unsigned long nr_written = 0;
778 /* allocate blocks */
779 ret = cow_file_range(inode, async_chunk->locked_page,
781 async_extent->start +
782 async_extent->ram_size - 1,
783 &page_started, &nr_written, 0);
788 * if page_started, cow_file_range inserted an
789 * inline extent and took care of all the unlocking
790 * and IO for us. Otherwise, we need to submit
791 * all those pages down to the drive.
793 if (!page_started && !ret)
794 extent_write_locked_range(inode,
796 async_extent->start +
797 async_extent->ram_size - 1,
799 else if (ret && async_chunk->locked_page)
800 unlock_page(async_chunk->locked_page);
806 ret = btrfs_reserve_extent(root, async_extent->ram_size,
807 async_extent->compressed_size,
808 async_extent->compressed_size,
809 0, alloc_hint, &ins, 1, 1);
811 free_async_extent_pages(async_extent);
813 if (ret == -ENOSPC) {
814 unlock_extent(io_tree, async_extent->start,
815 async_extent->start +
816 async_extent->ram_size - 1);
819 * we need to redirty the pages if we decide to
820 * fallback to uncompressed IO, otherwise we
821 * will not submit these pages down to lower
824 extent_range_redirty_for_io(inode,
826 async_extent->start +
827 async_extent->ram_size - 1);
834 * here we're doing allocation and writeback of the
837 em = create_io_em(inode, async_extent->start,
838 async_extent->ram_size, /* len */
839 async_extent->start, /* orig_start */
840 ins.objectid, /* block_start */
841 ins.offset, /* block_len */
842 ins.offset, /* orig_block_len */
843 async_extent->ram_size, /* ram_bytes */
844 async_extent->compress_type,
845 BTRFS_ORDERED_COMPRESSED);
847 /* ret value is not necessary due to void function */
848 goto out_free_reserve;
851 ret = btrfs_add_ordered_extent_compress(inode,
854 async_extent->ram_size,
856 BTRFS_ORDERED_COMPRESSED,
857 async_extent->compress_type);
859 btrfs_drop_extent_cache(BTRFS_I(inode),
861 async_extent->start +
862 async_extent->ram_size - 1, 0);
863 goto out_free_reserve;
865 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
868 * clear dirty, set writeback and unlock the pages.
870 extent_clear_unlock_delalloc(inode, async_extent->start,
871 async_extent->start +
872 async_extent->ram_size - 1,
873 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
874 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
876 if (btrfs_submit_compressed_write(inode,
878 async_extent->ram_size,
880 ins.offset, async_extent->pages,
881 async_extent->nr_pages,
882 async_chunk->write_flags,
883 async_chunk->blkcg_css)) {
884 struct page *p = async_extent->pages[0];
885 const u64 start = async_extent->start;
886 const u64 end = start + async_extent->ram_size - 1;
888 p->mapping = inode->i_mapping;
889 btrfs_writepage_endio_finish_ordered(p, start, end, 0);
892 extent_clear_unlock_delalloc(inode, start, end,
896 free_async_extent_pages(async_extent);
898 alloc_hint = ins.objectid + ins.offset;
904 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
905 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
907 extent_clear_unlock_delalloc(inode, async_extent->start,
908 async_extent->start +
909 async_extent->ram_size - 1,
910 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
911 EXTENT_DELALLOC_NEW |
912 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
913 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
914 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
916 free_async_extent_pages(async_extent);
921 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
924 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
925 struct extent_map *em;
928 read_lock(&em_tree->lock);
929 em = search_extent_mapping(em_tree, start, num_bytes);
932 * if block start isn't an actual block number then find the
933 * first block in this inode and use that as a hint. If that
934 * block is also bogus then just don't worry about it.
936 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
938 em = search_extent_mapping(em_tree, 0, 0);
939 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
940 alloc_hint = em->block_start;
944 alloc_hint = em->block_start;
948 read_unlock(&em_tree->lock);
954 * when extent_io.c finds a delayed allocation range in the file,
955 * the call backs end up in this code. The basic idea is to
956 * allocate extents on disk for the range, and create ordered data structs
957 * in ram to track those extents.
959 * locked_page is the page that writepage had locked already. We use
960 * it to make sure we don't do extra locks or unlocks.
962 * *page_started is set to one if we unlock locked_page and do everything
963 * required to start IO on it. It may be clean and already done with
966 static noinline int cow_file_range(struct inode *inode,
967 struct page *locked_page,
968 u64 start, u64 end, int *page_started,
969 unsigned long *nr_written, int unlock)
971 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
972 struct btrfs_root *root = BTRFS_I(inode)->root;
975 unsigned long ram_size;
976 u64 cur_alloc_size = 0;
977 u64 blocksize = fs_info->sectorsize;
978 struct btrfs_key ins;
979 struct extent_map *em;
981 unsigned long page_ops;
982 bool extent_reserved = false;
985 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
991 num_bytes = ALIGN(end - start + 1, blocksize);
992 num_bytes = max(blocksize, num_bytes);
993 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
995 inode_should_defrag(BTRFS_I(inode), start, end, num_bytes, SZ_64K);
998 /* lets try to make an inline extent */
999 ret = cow_file_range_inline(inode, start, end, 0,
1000 BTRFS_COMPRESS_NONE, NULL);
1003 * We use DO_ACCOUNTING here because we need the
1004 * delalloc_release_metadata to be run _after_ we drop
1005 * our outstanding extent for clearing delalloc for this
1008 extent_clear_unlock_delalloc(inode, start, end, NULL,
1009 EXTENT_LOCKED | EXTENT_DELALLOC |
1010 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1011 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1012 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1013 PAGE_END_WRITEBACK);
1014 *nr_written = *nr_written +
1015 (end - start + PAGE_SIZE) / PAGE_SIZE;
1018 } else if (ret < 0) {
1023 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1024 btrfs_drop_extent_cache(BTRFS_I(inode), start,
1025 start + num_bytes - 1, 0);
1027 while (num_bytes > 0) {
1028 cur_alloc_size = num_bytes;
1029 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1030 fs_info->sectorsize, 0, alloc_hint,
1034 cur_alloc_size = ins.offset;
1035 extent_reserved = true;
1037 ram_size = ins.offset;
1038 em = create_io_em(inode, start, ins.offset, /* len */
1039 start, /* orig_start */
1040 ins.objectid, /* block_start */
1041 ins.offset, /* block_len */
1042 ins.offset, /* orig_block_len */
1043 ram_size, /* ram_bytes */
1044 BTRFS_COMPRESS_NONE, /* compress_type */
1045 BTRFS_ORDERED_REGULAR /* type */);
1050 free_extent_map(em);
1052 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1053 ram_size, cur_alloc_size, 0);
1055 goto out_drop_extent_cache;
1057 if (root->root_key.objectid ==
1058 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1059 ret = btrfs_reloc_clone_csums(inode, start,
1062 * Only drop cache here, and process as normal.
1064 * We must not allow extent_clear_unlock_delalloc()
1065 * at out_unlock label to free meta of this ordered
1066 * extent, as its meta should be freed by
1067 * btrfs_finish_ordered_io().
1069 * So we must continue until @start is increased to
1070 * skip current ordered extent.
1073 btrfs_drop_extent_cache(BTRFS_I(inode), start,
1074 start + ram_size - 1, 0);
1077 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1079 /* we're not doing compressed IO, don't unlock the first
1080 * page (which the caller expects to stay locked), don't
1081 * clear any dirty bits and don't set any writeback bits
1083 * Do set the Private2 bit so we know this page was properly
1084 * setup for writepage
1086 page_ops = unlock ? PAGE_UNLOCK : 0;
1087 page_ops |= PAGE_SET_PRIVATE2;
1089 extent_clear_unlock_delalloc(inode, start,
1090 start + ram_size - 1,
1092 EXTENT_LOCKED | EXTENT_DELALLOC,
1094 if (num_bytes < cur_alloc_size)
1097 num_bytes -= cur_alloc_size;
1098 alloc_hint = ins.objectid + ins.offset;
1099 start += cur_alloc_size;
1100 extent_reserved = false;
1103 * btrfs_reloc_clone_csums() error, since start is increased
1104 * extent_clear_unlock_delalloc() at out_unlock label won't
1105 * free metadata of current ordered extent, we're OK to exit.
1113 out_drop_extent_cache:
1114 btrfs_drop_extent_cache(BTRFS_I(inode), start, start + ram_size - 1, 0);
1116 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1117 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1119 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1120 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1121 page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1124 * If we reserved an extent for our delalloc range (or a subrange) and
1125 * failed to create the respective ordered extent, then it means that
1126 * when we reserved the extent we decremented the extent's size from
1127 * the data space_info's bytes_may_use counter and incremented the
1128 * space_info's bytes_reserved counter by the same amount. We must make
1129 * sure extent_clear_unlock_delalloc() does not try to decrement again
1130 * the data space_info's bytes_may_use counter, therefore we do not pass
1131 * it the flag EXTENT_CLEAR_DATA_RESV.
1133 if (extent_reserved) {
1134 extent_clear_unlock_delalloc(inode, start,
1135 start + cur_alloc_size,
1139 start += cur_alloc_size;
1143 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1144 clear_bits | EXTENT_CLEAR_DATA_RESV,
1150 * work queue call back to started compression on a file and pages
1152 static noinline void async_cow_start(struct btrfs_work *work)
1154 struct async_chunk *async_chunk;
1155 int compressed_extents;
1157 async_chunk = container_of(work, struct async_chunk, work);
1159 compressed_extents = compress_file_range(async_chunk);
1160 if (compressed_extents == 0) {
1161 btrfs_add_delayed_iput(async_chunk->inode);
1162 async_chunk->inode = NULL;
1167 * work queue call back to submit previously compressed pages
1169 static noinline void async_cow_submit(struct btrfs_work *work)
1171 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1173 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1174 unsigned long nr_pages;
1176 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1179 /* atomic_sub_return implies a barrier */
1180 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1182 cond_wake_up_nomb(&fs_info->async_submit_wait);
1185 * ->inode could be NULL if async_chunk_start has failed to compress,
1186 * in which case we don't have anything to submit, yet we need to
1187 * always adjust ->async_delalloc_pages as its paired with the init
1188 * happening in cow_file_range_async
1190 if (async_chunk->inode)
1191 submit_compressed_extents(async_chunk);
1194 static noinline void async_cow_free(struct btrfs_work *work)
1196 struct async_chunk *async_chunk;
1198 async_chunk = container_of(work, struct async_chunk, work);
1199 if (async_chunk->inode)
1200 btrfs_add_delayed_iput(async_chunk->inode);
1201 if (async_chunk->blkcg_css)
1202 css_put(async_chunk->blkcg_css);
1204 * Since the pointer to 'pending' is at the beginning of the array of
1205 * async_chunk's, freeing it ensures the whole array has been freed.
1207 if (atomic_dec_and_test(async_chunk->pending))
1208 kvfree(async_chunk->pending);
1211 static int cow_file_range_async(struct inode *inode,
1212 struct writeback_control *wbc,
1213 struct page *locked_page,
1214 u64 start, u64 end, int *page_started,
1215 unsigned long *nr_written)
1217 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1218 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1219 struct async_cow *ctx;
1220 struct async_chunk *async_chunk;
1221 unsigned long nr_pages;
1223 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1225 bool should_compress;
1227 const unsigned int write_flags = wbc_to_write_flags(wbc);
1229 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
1231 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1232 !btrfs_test_opt(fs_info, FORCE_COMPRESS)) {
1234 should_compress = false;
1236 should_compress = true;
1239 nofs_flag = memalloc_nofs_save();
1240 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1241 memalloc_nofs_restore(nofs_flag);
1244 unsigned clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC |
1245 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1246 EXTENT_DO_ACCOUNTING;
1247 unsigned long page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1248 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
1251 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1252 clear_bits, page_ops);
1256 async_chunk = ctx->chunks;
1257 atomic_set(&ctx->num_chunks, num_chunks);
1259 for (i = 0; i < num_chunks; i++) {
1260 if (should_compress)
1261 cur_end = min(end, start + SZ_512K - 1);
1266 * igrab is called higher up in the call chain, take only the
1267 * lightweight reference for the callback lifetime
1270 async_chunk[i].pending = &ctx->num_chunks;
1271 async_chunk[i].inode = inode;
1272 async_chunk[i].start = start;
1273 async_chunk[i].end = cur_end;
1274 async_chunk[i].write_flags = write_flags;
1275 INIT_LIST_HEAD(&async_chunk[i].extents);
1278 * The locked_page comes all the way from writepage and its
1279 * the original page we were actually given. As we spread
1280 * this large delalloc region across multiple async_chunk
1281 * structs, only the first struct needs a pointer to locked_page
1283 * This way we don't need racey decisions about who is supposed
1288 * Depending on the compressibility, the pages might or
1289 * might not go through async. We want all of them to
1290 * be accounted against wbc once. Let's do it here
1291 * before the paths diverge. wbc accounting is used
1292 * only for foreign writeback detection and doesn't
1293 * need full accuracy. Just account the whole thing
1294 * against the first page.
1296 wbc_account_cgroup_owner(wbc, locked_page,
1298 async_chunk[i].locked_page = locked_page;
1301 async_chunk[i].locked_page = NULL;
1304 if (blkcg_css != blkcg_root_css) {
1306 async_chunk[i].blkcg_css = blkcg_css;
1308 async_chunk[i].blkcg_css = NULL;
1311 btrfs_init_work(&async_chunk[i].work, async_cow_start,
1312 async_cow_submit, async_cow_free);
1314 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1315 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1317 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1319 *nr_written += nr_pages;
1320 start = cur_end + 1;
1326 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1327 u64 bytenr, u64 num_bytes)
1330 struct btrfs_ordered_sum *sums;
1333 ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1334 bytenr + num_bytes - 1, &list, 0);
1335 if (ret == 0 && list_empty(&list))
1338 while (!list_empty(&list)) {
1339 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1340 list_del(&sums->list);
1349 * when nowcow writeback call back. This checks for snapshots or COW copies
1350 * of the extents that exist in the file, and COWs the file as required.
1352 * If no cow copies or snapshots exist, we write directly to the existing
1355 static noinline int run_delalloc_nocow(struct inode *inode,
1356 struct page *locked_page,
1357 const u64 start, const u64 end,
1358 int *page_started, int force,
1359 unsigned long *nr_written)
1361 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1362 struct btrfs_root *root = BTRFS_I(inode)->root;
1363 struct btrfs_path *path;
1364 u64 cow_start = (u64)-1;
1365 u64 cur_offset = start;
1367 bool check_prev = true;
1368 const bool freespace_inode = btrfs_is_free_space_inode(BTRFS_I(inode));
1369 u64 ino = btrfs_ino(BTRFS_I(inode));
1371 u64 disk_bytenr = 0;
1373 path = btrfs_alloc_path();
1375 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1376 EXTENT_LOCKED | EXTENT_DELALLOC |
1377 EXTENT_DO_ACCOUNTING |
1378 EXTENT_DEFRAG, PAGE_UNLOCK |
1380 PAGE_SET_WRITEBACK |
1381 PAGE_END_WRITEBACK);
1386 struct btrfs_key found_key;
1387 struct btrfs_file_extent_item *fi;
1388 struct extent_buffer *leaf;
1398 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1404 * If there is no extent for our range when doing the initial
1405 * search, then go back to the previous slot as it will be the
1406 * one containing the search offset
1408 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1409 leaf = path->nodes[0];
1410 btrfs_item_key_to_cpu(leaf, &found_key,
1411 path->slots[0] - 1);
1412 if (found_key.objectid == ino &&
1413 found_key.type == BTRFS_EXTENT_DATA_KEY)
1418 /* Go to next leaf if we have exhausted the current one */
1419 leaf = path->nodes[0];
1420 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1421 ret = btrfs_next_leaf(root, path);
1423 if (cow_start != (u64)-1)
1424 cur_offset = cow_start;
1429 leaf = path->nodes[0];
1432 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1434 /* Didn't find anything for our INO */
1435 if (found_key.objectid > ino)
1438 * Keep searching until we find an EXTENT_ITEM or there are no
1439 * more extents for this inode
1441 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1442 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1447 /* Found key is not EXTENT_DATA_KEY or starts after req range */
1448 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1449 found_key.offset > end)
1453 * If the found extent starts after requested offset, then
1454 * adjust extent_end to be right before this extent begins
1456 if (found_key.offset > cur_offset) {
1457 extent_end = found_key.offset;
1463 * Found extent which begins before our range and potentially
1466 fi = btrfs_item_ptr(leaf, path->slots[0],
1467 struct btrfs_file_extent_item);
1468 extent_type = btrfs_file_extent_type(leaf, fi);
1470 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1471 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1472 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1473 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1474 extent_offset = btrfs_file_extent_offset(leaf, fi);
1475 extent_end = found_key.offset +
1476 btrfs_file_extent_num_bytes(leaf, fi);
1478 btrfs_file_extent_disk_num_bytes(leaf, fi);
1480 * If the extent we got ends before our current offset,
1481 * skip to the next extent.
1483 if (extent_end <= cur_offset) {
1488 if (disk_bytenr == 0)
1490 /* Skip compressed/encrypted/encoded extents */
1491 if (btrfs_file_extent_compression(leaf, fi) ||
1492 btrfs_file_extent_encryption(leaf, fi) ||
1493 btrfs_file_extent_other_encoding(leaf, fi))
1496 * If extent is created before the last volume's snapshot
1497 * this implies the extent is shared, hence we can't do
1498 * nocow. This is the same check as in
1499 * btrfs_cross_ref_exist but without calling
1500 * btrfs_search_slot.
1502 if (!freespace_inode &&
1503 btrfs_file_extent_generation(leaf, fi) <=
1504 btrfs_root_last_snapshot(&root->root_item))
1506 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1508 /* If extent is RO, we must COW it */
1509 if (btrfs_extent_readonly(fs_info, disk_bytenr))
1511 ret = btrfs_cross_ref_exist(root, ino,
1513 extent_offset, disk_bytenr);
1516 * ret could be -EIO if the above fails to read
1520 if (cow_start != (u64)-1)
1521 cur_offset = cow_start;
1525 WARN_ON_ONCE(freespace_inode);
1528 disk_bytenr += extent_offset;
1529 disk_bytenr += cur_offset - found_key.offset;
1530 num_bytes = min(end + 1, extent_end) - cur_offset;
1532 * If there are pending snapshots for this root, we
1533 * fall into common COW way
1535 if (!freespace_inode && atomic_read(&root->snapshot_force_cow))
1538 * force cow if csum exists in the range.
1539 * this ensure that csum for a given extent are
1540 * either valid or do not exist.
1542 ret = csum_exist_in_range(fs_info, disk_bytenr,
1546 * ret could be -EIO if the above fails to read
1550 if (cow_start != (u64)-1)
1551 cur_offset = cow_start;
1554 WARN_ON_ONCE(freespace_inode);
1557 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr))
1560 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1561 extent_end = found_key.offset + ram_bytes;
1562 extent_end = ALIGN(extent_end, fs_info->sectorsize);
1563 /* Skip extents outside of our requested range */
1564 if (extent_end <= start) {
1569 /* If this triggers then we have a memory corruption */
1574 * If nocow is false then record the beginning of the range
1575 * that needs to be COWed
1578 if (cow_start == (u64)-1)
1579 cow_start = cur_offset;
1580 cur_offset = extent_end;
1581 if (cur_offset > end)
1587 btrfs_release_path(path);
1590 * COW range from cow_start to found_key.offset - 1. As the key
1591 * will contain the beginning of the first extent that can be
1592 * NOCOW, following one which needs to be COW'ed
1594 if (cow_start != (u64)-1) {
1595 ret = cow_file_range(inode, locked_page,
1596 cow_start, found_key.offset - 1,
1597 page_started, nr_written, 1);
1600 btrfs_dec_nocow_writers(fs_info,
1604 cow_start = (u64)-1;
1607 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1608 u64 orig_start = found_key.offset - extent_offset;
1609 struct extent_map *em;
1611 em = create_io_em(inode, cur_offset, num_bytes,
1613 disk_bytenr, /* block_start */
1614 num_bytes, /* block_len */
1615 disk_num_bytes, /* orig_block_len */
1616 ram_bytes, BTRFS_COMPRESS_NONE,
1617 BTRFS_ORDERED_PREALLOC);
1620 btrfs_dec_nocow_writers(fs_info,
1625 free_extent_map(em);
1626 ret = btrfs_add_ordered_extent(inode, cur_offset,
1627 disk_bytenr, num_bytes,
1629 BTRFS_ORDERED_PREALLOC);
1631 btrfs_drop_extent_cache(BTRFS_I(inode),
1633 cur_offset + num_bytes - 1,
1638 ret = btrfs_add_ordered_extent(inode, cur_offset,
1639 disk_bytenr, num_bytes,
1641 BTRFS_ORDERED_NOCOW);
1647 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1650 if (root->root_key.objectid ==
1651 BTRFS_DATA_RELOC_TREE_OBJECTID)
1653 * Error handled later, as we must prevent
1654 * extent_clear_unlock_delalloc() in error handler
1655 * from freeing metadata of created ordered extent.
1657 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1660 extent_clear_unlock_delalloc(inode, cur_offset,
1661 cur_offset + num_bytes - 1,
1662 locked_page, EXTENT_LOCKED |
1664 EXTENT_CLEAR_DATA_RESV,
1665 PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1667 cur_offset = extent_end;
1670 * btrfs_reloc_clone_csums() error, now we're OK to call error
1671 * handler, as metadata for created ordered extent will only
1672 * be freed by btrfs_finish_ordered_io().
1676 if (cur_offset > end)
1679 btrfs_release_path(path);
1681 if (cur_offset <= end && cow_start == (u64)-1)
1682 cow_start = cur_offset;
1684 if (cow_start != (u64)-1) {
1686 ret = cow_file_range(inode, locked_page, cow_start, end,
1687 page_started, nr_written, 1);
1694 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1696 if (ret && cur_offset < end)
1697 extent_clear_unlock_delalloc(inode, cur_offset, end,
1698 locked_page, EXTENT_LOCKED |
1699 EXTENT_DELALLOC | EXTENT_DEFRAG |
1700 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1702 PAGE_SET_WRITEBACK |
1703 PAGE_END_WRITEBACK);
1704 btrfs_free_path(path);
1708 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1711 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1712 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1716 * @defrag_bytes is a hint value, no spinlock held here,
1717 * if is not zero, it means the file is defragging.
1718 * Force cow if given extent needs to be defragged.
1720 if (BTRFS_I(inode)->defrag_bytes &&
1721 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1722 EXTENT_DEFRAG, 0, NULL))
1729 * Function to process delayed allocation (create CoW) for ranges which are
1730 * being touched for the first time.
1732 int btrfs_run_delalloc_range(struct inode *inode, struct page *locked_page,
1733 u64 start, u64 end, int *page_started, unsigned long *nr_written,
1734 struct writeback_control *wbc)
1737 int force_cow = need_force_cow(inode, start, end);
1739 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1740 ret = run_delalloc_nocow(inode, locked_page, start, end,
1741 page_started, 1, nr_written);
1742 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1743 ret = run_delalloc_nocow(inode, locked_page, start, end,
1744 page_started, 0, nr_written);
1745 } else if (!inode_can_compress(inode) ||
1746 !inode_need_compress(inode, start, end)) {
1747 ret = cow_file_range(inode, locked_page, start, end,
1748 page_started, nr_written, 1);
1750 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1751 &BTRFS_I(inode)->runtime_flags);
1752 ret = cow_file_range_async(inode, wbc, locked_page, start, end,
1753 page_started, nr_written);
1756 btrfs_cleanup_ordered_extents(inode, locked_page, start,
1761 void btrfs_split_delalloc_extent(struct inode *inode,
1762 struct extent_state *orig, u64 split)
1766 /* not delalloc, ignore it */
1767 if (!(orig->state & EXTENT_DELALLOC))
1770 size = orig->end - orig->start + 1;
1771 if (size > BTRFS_MAX_EXTENT_SIZE) {
1776 * See the explanation in btrfs_merge_delalloc_extent, the same
1777 * applies here, just in reverse.
1779 new_size = orig->end - split + 1;
1780 num_extents = count_max_extents(new_size);
1781 new_size = split - orig->start;
1782 num_extents += count_max_extents(new_size);
1783 if (count_max_extents(size) >= num_extents)
1787 spin_lock(&BTRFS_I(inode)->lock);
1788 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1789 spin_unlock(&BTRFS_I(inode)->lock);
1793 * Handle merged delayed allocation extents so we can keep track of new extents
1794 * that are just merged onto old extents, such as when we are doing sequential
1795 * writes, so we can properly account for the metadata space we'll need.
1797 void btrfs_merge_delalloc_extent(struct inode *inode, struct extent_state *new,
1798 struct extent_state *other)
1800 u64 new_size, old_size;
1803 /* not delalloc, ignore it */
1804 if (!(other->state & EXTENT_DELALLOC))
1807 if (new->start > other->start)
1808 new_size = new->end - other->start + 1;
1810 new_size = other->end - new->start + 1;
1812 /* we're not bigger than the max, unreserve the space and go */
1813 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1814 spin_lock(&BTRFS_I(inode)->lock);
1815 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1816 spin_unlock(&BTRFS_I(inode)->lock);
1821 * We have to add up either side to figure out how many extents were
1822 * accounted for before we merged into one big extent. If the number of
1823 * extents we accounted for is <= the amount we need for the new range
1824 * then we can return, otherwise drop. Think of it like this
1828 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1829 * need 2 outstanding extents, on one side we have 1 and the other side
1830 * we have 1 so they are == and we can return. But in this case
1832 * [MAX_SIZE+4k][MAX_SIZE+4k]
1834 * Each range on their own accounts for 2 extents, but merged together
1835 * they are only 3 extents worth of accounting, so we need to drop in
1838 old_size = other->end - other->start + 1;
1839 num_extents = count_max_extents(old_size);
1840 old_size = new->end - new->start + 1;
1841 num_extents += count_max_extents(old_size);
1842 if (count_max_extents(new_size) >= num_extents)
1845 spin_lock(&BTRFS_I(inode)->lock);
1846 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1847 spin_unlock(&BTRFS_I(inode)->lock);
1850 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1851 struct inode *inode)
1853 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1855 spin_lock(&root->delalloc_lock);
1856 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1857 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1858 &root->delalloc_inodes);
1859 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1860 &BTRFS_I(inode)->runtime_flags);
1861 root->nr_delalloc_inodes++;
1862 if (root->nr_delalloc_inodes == 1) {
1863 spin_lock(&fs_info->delalloc_root_lock);
1864 BUG_ON(!list_empty(&root->delalloc_root));
1865 list_add_tail(&root->delalloc_root,
1866 &fs_info->delalloc_roots);
1867 spin_unlock(&fs_info->delalloc_root_lock);
1870 spin_unlock(&root->delalloc_lock);
1874 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
1875 struct btrfs_inode *inode)
1877 struct btrfs_fs_info *fs_info = root->fs_info;
1879 if (!list_empty(&inode->delalloc_inodes)) {
1880 list_del_init(&inode->delalloc_inodes);
1881 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1882 &inode->runtime_flags);
1883 root->nr_delalloc_inodes--;
1884 if (!root->nr_delalloc_inodes) {
1885 ASSERT(list_empty(&root->delalloc_inodes));
1886 spin_lock(&fs_info->delalloc_root_lock);
1887 BUG_ON(list_empty(&root->delalloc_root));
1888 list_del_init(&root->delalloc_root);
1889 spin_unlock(&fs_info->delalloc_root_lock);
1894 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1895 struct btrfs_inode *inode)
1897 spin_lock(&root->delalloc_lock);
1898 __btrfs_del_delalloc_inode(root, inode);
1899 spin_unlock(&root->delalloc_lock);
1903 * Properly track delayed allocation bytes in the inode and to maintain the
1904 * list of inodes that have pending delalloc work to be done.
1906 void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state,
1909 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1911 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1914 * set_bit and clear bit hooks normally require _irqsave/restore
1915 * but in this case, we are only testing for the DELALLOC
1916 * bit, which is only set or cleared with irqs on
1918 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1919 struct btrfs_root *root = BTRFS_I(inode)->root;
1920 u64 len = state->end + 1 - state->start;
1921 u32 num_extents = count_max_extents(len);
1922 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
1924 spin_lock(&BTRFS_I(inode)->lock);
1925 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
1926 spin_unlock(&BTRFS_I(inode)->lock);
1928 /* For sanity tests */
1929 if (btrfs_is_testing(fs_info))
1932 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
1933 fs_info->delalloc_batch);
1934 spin_lock(&BTRFS_I(inode)->lock);
1935 BTRFS_I(inode)->delalloc_bytes += len;
1936 if (*bits & EXTENT_DEFRAG)
1937 BTRFS_I(inode)->defrag_bytes += len;
1938 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1939 &BTRFS_I(inode)->runtime_flags))
1940 btrfs_add_delalloc_inodes(root, inode);
1941 spin_unlock(&BTRFS_I(inode)->lock);
1944 if (!(state->state & EXTENT_DELALLOC_NEW) &&
1945 (*bits & EXTENT_DELALLOC_NEW)) {
1946 spin_lock(&BTRFS_I(inode)->lock);
1947 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
1949 spin_unlock(&BTRFS_I(inode)->lock);
1954 * Once a range is no longer delalloc this function ensures that proper
1955 * accounting happens.
1957 void btrfs_clear_delalloc_extent(struct inode *vfs_inode,
1958 struct extent_state *state, unsigned *bits)
1960 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
1961 struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb);
1962 u64 len = state->end + 1 - state->start;
1963 u32 num_extents = count_max_extents(len);
1965 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
1966 spin_lock(&inode->lock);
1967 inode->defrag_bytes -= len;
1968 spin_unlock(&inode->lock);
1972 * set_bit and clear bit hooks normally require _irqsave/restore
1973 * but in this case, we are only testing for the DELALLOC
1974 * bit, which is only set or cleared with irqs on
1976 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1977 struct btrfs_root *root = inode->root;
1978 bool do_list = !btrfs_is_free_space_inode(inode);
1980 spin_lock(&inode->lock);
1981 btrfs_mod_outstanding_extents(inode, -num_extents);
1982 spin_unlock(&inode->lock);
1985 * We don't reserve metadata space for space cache inodes so we
1986 * don't need to call delalloc_release_metadata if there is an
1989 if (*bits & EXTENT_CLEAR_META_RESV &&
1990 root != fs_info->tree_root)
1991 btrfs_delalloc_release_metadata(inode, len, false);
1993 /* For sanity tests. */
1994 if (btrfs_is_testing(fs_info))
1997 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
1998 do_list && !(state->state & EXTENT_NORESERVE) &&
1999 (*bits & EXTENT_CLEAR_DATA_RESV))
2000 btrfs_free_reserved_data_space_noquota(
2004 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2005 fs_info->delalloc_batch);
2006 spin_lock(&inode->lock);
2007 inode->delalloc_bytes -= len;
2008 if (do_list && inode->delalloc_bytes == 0 &&
2009 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2010 &inode->runtime_flags))
2011 btrfs_del_delalloc_inode(root, inode);
2012 spin_unlock(&inode->lock);
2015 if ((state->state & EXTENT_DELALLOC_NEW) &&
2016 (*bits & EXTENT_DELALLOC_NEW)) {
2017 spin_lock(&inode->lock);
2018 ASSERT(inode->new_delalloc_bytes >= len);
2019 inode->new_delalloc_bytes -= len;
2020 spin_unlock(&inode->lock);
2025 * btrfs_bio_fits_in_stripe - Checks whether the size of the given bio will fit
2026 * in a chunk's stripe. This function ensures that bios do not span a
2029 * @page - The page we are about to add to the bio
2030 * @size - size we want to add to the bio
2031 * @bio - bio we want to ensure is smaller than a stripe
2032 * @bio_flags - flags of the bio
2034 * return 1 if page cannot be added to the bio
2035 * return 0 if page can be added to the bio
2036 * return error otherwise
2038 int btrfs_bio_fits_in_stripe(struct page *page, size_t size, struct bio *bio,
2039 unsigned long bio_flags)
2041 struct inode *inode = page->mapping->host;
2042 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2043 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
2047 struct btrfs_io_geometry geom;
2049 if (bio_flags & EXTENT_BIO_COMPRESSED)
2052 length = bio->bi_iter.bi_size;
2053 map_length = length;
2054 ret = btrfs_get_io_geometry(fs_info, btrfs_op(bio), logical, map_length,
2059 if (geom.len < length + size)
2065 * in order to insert checksums into the metadata in large chunks,
2066 * we wait until bio submission time. All the pages in the bio are
2067 * checksummed and sums are attached onto the ordered extent record.
2069 * At IO completion time the cums attached on the ordered extent record
2070 * are inserted into the btree
2072 static blk_status_t btrfs_submit_bio_start(void *private_data, struct bio *bio,
2075 struct inode *inode = private_data;
2076 blk_status_t ret = 0;
2078 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
2079 BUG_ON(ret); /* -ENOMEM */
2084 * extent_io.c submission hook. This does the right thing for csum calculation
2085 * on write, or reading the csums from the tree before a read.
2087 * Rules about async/sync submit,
2088 * a) read: sync submit
2090 * b) write without checksum: sync submit
2092 * c) write with checksum:
2093 * c-1) if bio is issued by fsync: sync submit
2094 * (sync_writers != 0)
2096 * c-2) if root is reloc root: sync submit
2097 * (only in case of buffered IO)
2099 * c-3) otherwise: async submit
2101 static blk_status_t btrfs_submit_bio_hook(struct inode *inode, struct bio *bio,
2103 unsigned long bio_flags)
2106 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2107 struct btrfs_root *root = BTRFS_I(inode)->root;
2108 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
2109 blk_status_t ret = 0;
2111 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
2113 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
2115 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
2116 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
2118 if (bio_op(bio) != REQ_OP_WRITE) {
2119 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
2123 if (bio_flags & EXTENT_BIO_COMPRESSED) {
2124 ret = btrfs_submit_compressed_read(inode, bio,
2128 } else if (!skip_sum) {
2129 ret = btrfs_lookup_bio_sums(inode, bio, (u64)-1, NULL);
2134 } else if (async && !skip_sum) {
2135 /* csum items have already been cloned */
2136 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2138 /* we're doing a write, do the async checksumming */
2139 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags,
2140 0, inode, btrfs_submit_bio_start);
2142 } else if (!skip_sum) {
2143 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
2149 ret = btrfs_map_bio(fs_info, bio, mirror_num);
2153 bio->bi_status = ret;
2160 * given a list of ordered sums record them in the inode. This happens
2161 * at IO completion time based on sums calculated at bio submission time.
2163 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
2164 struct inode *inode, struct list_head *list)
2166 struct btrfs_ordered_sum *sum;
2169 list_for_each_entry(sum, list, list) {
2170 trans->adding_csums = true;
2171 ret = btrfs_csum_file_blocks(trans,
2172 BTRFS_I(inode)->root->fs_info->csum_root, sum);
2173 trans->adding_csums = false;
2180 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
2181 unsigned int extra_bits,
2182 struct extent_state **cached_state)
2184 WARN_ON(PAGE_ALIGNED(end));
2185 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2186 extra_bits, cached_state);
2189 /* see btrfs_writepage_start_hook for details on why this is required */
2190 struct btrfs_writepage_fixup {
2192 struct btrfs_work work;
2195 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2197 struct btrfs_writepage_fixup *fixup;
2198 struct btrfs_ordered_extent *ordered;
2199 struct extent_state *cached_state = NULL;
2200 struct extent_changeset *data_reserved = NULL;
2202 struct inode *inode;
2207 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2211 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2212 ClearPageChecked(page);
2216 inode = page->mapping->host;
2217 page_start = page_offset(page);
2218 page_end = page_offset(page) + PAGE_SIZE - 1;
2220 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2223 /* already ordered? We're done */
2224 if (PagePrivate2(page))
2227 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
2230 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2231 page_end, &cached_state);
2233 btrfs_start_ordered_extent(inode, ordered, 1);
2234 btrfs_put_ordered_extent(ordered);
2238 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2241 mapping_set_error(page->mapping, ret);
2242 end_extent_writepage(page, ret, page_start, page_end);
2243 ClearPageChecked(page);
2247 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2250 mapping_set_error(page->mapping, ret);
2251 end_extent_writepage(page, ret, page_start, page_end);
2252 ClearPageChecked(page);
2256 ClearPageChecked(page);
2257 set_page_dirty(page);
2259 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
2261 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2264 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2270 extent_changeset_free(data_reserved);
2274 * There are a few paths in the higher layers of the kernel that directly
2275 * set the page dirty bit without asking the filesystem if it is a
2276 * good idea. This causes problems because we want to make sure COW
2277 * properly happens and the data=ordered rules are followed.
2279 * In our case any range that doesn't have the ORDERED bit set
2280 * hasn't been properly setup for IO. We kick off an async process
2281 * to fix it up. The async helper will wait for ordered extents, set
2282 * the delalloc bit and make it safe to write the page.
2284 int btrfs_writepage_cow_fixup(struct page *page, u64 start, u64 end)
2286 struct inode *inode = page->mapping->host;
2287 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2288 struct btrfs_writepage_fixup *fixup;
2290 /* this page is properly in the ordered list */
2291 if (TestClearPagePrivate2(page))
2294 if (PageChecked(page))
2297 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2301 SetPageChecked(page);
2303 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
2305 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2309 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2310 struct inode *inode, u64 file_pos,
2311 u64 disk_bytenr, u64 disk_num_bytes,
2312 u64 num_bytes, u64 ram_bytes,
2313 u8 compression, u8 encryption,
2314 u16 other_encoding, int extent_type)
2316 struct btrfs_root *root = BTRFS_I(inode)->root;
2317 struct btrfs_file_extent_item *fi;
2318 struct btrfs_path *path;
2319 struct extent_buffer *leaf;
2320 struct btrfs_key ins;
2322 int extent_inserted = 0;
2325 path = btrfs_alloc_path();
2330 * we may be replacing one extent in the tree with another.
2331 * The new extent is pinned in the extent map, and we don't want
2332 * to drop it from the cache until it is completely in the btree.
2334 * So, tell btrfs_drop_extents to leave this extent in the cache.
2335 * the caller is expected to unpin it and allow it to be merged
2338 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2339 file_pos + num_bytes, NULL, 0,
2340 1, sizeof(*fi), &extent_inserted);
2344 if (!extent_inserted) {
2345 ins.objectid = btrfs_ino(BTRFS_I(inode));
2346 ins.offset = file_pos;
2347 ins.type = BTRFS_EXTENT_DATA_KEY;
2349 path->leave_spinning = 1;
2350 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2355 leaf = path->nodes[0];
2356 fi = btrfs_item_ptr(leaf, path->slots[0],
2357 struct btrfs_file_extent_item);
2358 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2359 btrfs_set_file_extent_type(leaf, fi, extent_type);
2360 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2361 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2362 btrfs_set_file_extent_offset(leaf, fi, 0);
2363 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2364 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2365 btrfs_set_file_extent_compression(leaf, fi, compression);
2366 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2367 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2369 btrfs_mark_buffer_dirty(leaf);
2370 btrfs_release_path(path);
2372 inode_add_bytes(inode, num_bytes);
2374 ins.objectid = disk_bytenr;
2375 ins.offset = disk_num_bytes;
2376 ins.type = BTRFS_EXTENT_ITEM_KEY;
2379 * Release the reserved range from inode dirty range map, as it is
2380 * already moved into delayed_ref_head
2382 ret = btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2386 ret = btrfs_alloc_reserved_file_extent(trans, root,
2387 btrfs_ino(BTRFS_I(inode)),
2388 file_pos, qg_released, &ins);
2390 btrfs_free_path(path);
2395 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2398 struct btrfs_block_group *cache;
2400 cache = btrfs_lookup_block_group(fs_info, start);
2403 spin_lock(&cache->lock);
2404 cache->delalloc_bytes -= len;
2405 spin_unlock(&cache->lock);
2407 btrfs_put_block_group(cache);
2410 /* as ordered data IO finishes, this gets called so we can finish
2411 * an ordered extent if the range of bytes in the file it covers are
2414 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2416 struct inode *inode = ordered_extent->inode;
2417 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2418 struct btrfs_root *root = BTRFS_I(inode)->root;
2419 struct btrfs_trans_handle *trans = NULL;
2420 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2421 struct extent_state *cached_state = NULL;
2423 int compress_type = 0;
2425 u64 logical_len = ordered_extent->num_bytes;
2426 bool freespace_inode;
2427 bool truncated = false;
2428 bool range_locked = false;
2429 bool clear_new_delalloc_bytes = false;
2430 bool clear_reserved_extent = true;
2431 unsigned int clear_bits;
2433 start = ordered_extent->file_offset;
2434 end = start + ordered_extent->num_bytes - 1;
2436 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2437 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2438 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2439 clear_new_delalloc_bytes = true;
2441 freespace_inode = btrfs_is_free_space_inode(BTRFS_I(inode));
2443 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2448 btrfs_free_io_failure_record(BTRFS_I(inode), start, end);
2450 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2452 logical_len = ordered_extent->truncated_len;
2453 /* Truncated the entire extent, don't bother adding */
2458 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2459 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2462 * For mwrite(mmap + memset to write) case, we still reserve
2463 * space for NOCOW range.
2464 * As NOCOW won't cause a new delayed ref, just free the space
2466 btrfs_qgroup_free_data(inode, NULL, start,
2467 ordered_extent->num_bytes);
2468 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2469 if (freespace_inode)
2470 trans = btrfs_join_transaction_spacecache(root);
2472 trans = btrfs_join_transaction(root);
2473 if (IS_ERR(trans)) {
2474 ret = PTR_ERR(trans);
2478 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
2479 ret = btrfs_update_inode_fallback(trans, root, inode);
2480 if (ret) /* -ENOMEM or corruption */
2481 btrfs_abort_transaction(trans, ret);
2485 range_locked = true;
2486 lock_extent_bits(io_tree, start, end, &cached_state);
2488 if (freespace_inode)
2489 trans = btrfs_join_transaction_spacecache(root);
2491 trans = btrfs_join_transaction(root);
2492 if (IS_ERR(trans)) {
2493 ret = PTR_ERR(trans);
2498 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
2500 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2501 compress_type = ordered_extent->compress_type;
2502 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2503 BUG_ON(compress_type);
2504 btrfs_qgroup_free_data(inode, NULL, start,
2505 ordered_extent->num_bytes);
2506 ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
2507 ordered_extent->file_offset,
2508 ordered_extent->file_offset +
2511 BUG_ON(root == fs_info->tree_root);
2512 ret = insert_reserved_file_extent(trans, inode, start,
2513 ordered_extent->disk_bytenr,
2514 ordered_extent->disk_num_bytes,
2515 logical_len, logical_len,
2516 compress_type, 0, 0,
2517 BTRFS_FILE_EXTENT_REG);
2519 clear_reserved_extent = false;
2520 btrfs_release_delalloc_bytes(fs_info,
2521 ordered_extent->disk_bytenr,
2522 ordered_extent->disk_num_bytes);
2525 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2526 ordered_extent->file_offset,
2527 ordered_extent->num_bytes, trans->transid);
2529 btrfs_abort_transaction(trans, ret);
2533 ret = add_pending_csums(trans, inode, &ordered_extent->list);
2535 btrfs_abort_transaction(trans, ret);
2539 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2540 ret = btrfs_update_inode_fallback(trans, root, inode);
2541 if (ret) { /* -ENOMEM or corruption */
2542 btrfs_abort_transaction(trans, ret);
2547 clear_bits = EXTENT_DEFRAG;
2549 clear_bits |= EXTENT_LOCKED;
2550 if (clear_new_delalloc_bytes)
2551 clear_bits |= EXTENT_DELALLOC_NEW;
2552 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, clear_bits,
2553 (clear_bits & EXTENT_LOCKED) ? 1 : 0, 0,
2557 btrfs_end_transaction(trans);
2559 if (ret || truncated) {
2560 u64 unwritten_start = start;
2563 unwritten_start += logical_len;
2564 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
2566 /* Drop the cache for the part of the extent we didn't write. */
2567 btrfs_drop_extent_cache(BTRFS_I(inode), unwritten_start, end, 0);
2570 * If the ordered extent had an IOERR or something else went
2571 * wrong we need to return the space for this ordered extent
2572 * back to the allocator. We only free the extent in the
2573 * truncated case if we didn't write out the extent at all.
2575 * If we made it past insert_reserved_file_extent before we
2576 * errored out then we don't need to do this as the accounting
2577 * has already been done.
2579 if ((ret || !logical_len) &&
2580 clear_reserved_extent &&
2581 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2582 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2584 * Discard the range before returning it back to the
2587 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
2588 btrfs_discard_extent(fs_info,
2589 ordered_extent->disk_bytenr,
2590 ordered_extent->disk_num_bytes,
2592 btrfs_free_reserved_extent(fs_info,
2593 ordered_extent->disk_bytenr,
2594 ordered_extent->disk_num_bytes, 1);
2599 * This needs to be done to make sure anybody waiting knows we are done
2600 * updating everything for this ordered extent.
2602 btrfs_remove_ordered_extent(inode, ordered_extent);
2605 btrfs_put_ordered_extent(ordered_extent);
2606 /* once for the tree */
2607 btrfs_put_ordered_extent(ordered_extent);
2612 static void finish_ordered_fn(struct btrfs_work *work)
2614 struct btrfs_ordered_extent *ordered_extent;
2615 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2616 btrfs_finish_ordered_io(ordered_extent);
2619 void btrfs_writepage_endio_finish_ordered(struct page *page, u64 start,
2620 u64 end, int uptodate)
2622 struct inode *inode = page->mapping->host;
2623 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2624 struct btrfs_ordered_extent *ordered_extent = NULL;
2625 struct btrfs_workqueue *wq;
2627 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2629 ClearPagePrivate2(page);
2630 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2631 end - start + 1, uptodate))
2634 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
2635 wq = fs_info->endio_freespace_worker;
2637 wq = fs_info->endio_write_workers;
2639 btrfs_init_work(&ordered_extent->work, finish_ordered_fn, NULL, NULL);
2640 btrfs_queue_work(wq, &ordered_extent->work);
2643 static int __readpage_endio_check(struct inode *inode,
2644 struct btrfs_io_bio *io_bio,
2645 int icsum, struct page *page,
2646 int pgoff, u64 start, size_t len)
2648 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2649 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
2651 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2653 u8 csum[BTRFS_CSUM_SIZE];
2655 csum_expected = ((u8 *)io_bio->csum) + icsum * csum_size;
2657 kaddr = kmap_atomic(page);
2658 shash->tfm = fs_info->csum_shash;
2660 crypto_shash_init(shash);
2661 crypto_shash_update(shash, kaddr + pgoff, len);
2662 crypto_shash_final(shash, csum);
2664 if (memcmp(csum, csum_expected, csum_size))
2667 kunmap_atomic(kaddr);
2670 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
2671 io_bio->mirror_num);
2672 memset(kaddr + pgoff, 1, len);
2673 flush_dcache_page(page);
2674 kunmap_atomic(kaddr);
2679 * when reads are done, we need to check csums to verify the data is correct
2680 * if there's a match, we allow the bio to finish. If not, the code in
2681 * extent_io.c will try to find good copies for us.
2683 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
2684 u64 phy_offset, struct page *page,
2685 u64 start, u64 end, int mirror)
2687 size_t offset = start - page_offset(page);
2688 struct inode *inode = page->mapping->host;
2689 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2690 struct btrfs_root *root = BTRFS_I(inode)->root;
2692 if (PageChecked(page)) {
2693 ClearPageChecked(page);
2697 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2700 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2701 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2702 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
2706 phy_offset >>= inode->i_sb->s_blocksize_bits;
2707 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
2708 start, (size_t)(end - start + 1));
2712 * btrfs_add_delayed_iput - perform a delayed iput on @inode
2714 * @inode: The inode we want to perform iput on
2716 * This function uses the generic vfs_inode::i_count to track whether we should
2717 * just decrement it (in case it's > 1) or if this is the last iput then link
2718 * the inode to the delayed iput machinery. Delayed iputs are processed at
2719 * transaction commit time/superblock commit/cleaner kthread.
2721 void btrfs_add_delayed_iput(struct inode *inode)
2723 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2724 struct btrfs_inode *binode = BTRFS_I(inode);
2726 if (atomic_add_unless(&inode->i_count, -1, 1))
2729 atomic_inc(&fs_info->nr_delayed_iputs);
2730 spin_lock(&fs_info->delayed_iput_lock);
2731 ASSERT(list_empty(&binode->delayed_iput));
2732 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
2733 spin_unlock(&fs_info->delayed_iput_lock);
2734 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
2735 wake_up_process(fs_info->cleaner_kthread);
2738 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
2739 struct btrfs_inode *inode)
2741 list_del_init(&inode->delayed_iput);
2742 spin_unlock(&fs_info->delayed_iput_lock);
2743 iput(&inode->vfs_inode);
2744 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
2745 wake_up(&fs_info->delayed_iputs_wait);
2746 spin_lock(&fs_info->delayed_iput_lock);
2749 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
2750 struct btrfs_inode *inode)
2752 if (!list_empty(&inode->delayed_iput)) {
2753 spin_lock(&fs_info->delayed_iput_lock);
2754 if (!list_empty(&inode->delayed_iput))
2755 run_delayed_iput_locked(fs_info, inode);
2756 spin_unlock(&fs_info->delayed_iput_lock);
2760 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
2763 spin_lock(&fs_info->delayed_iput_lock);
2764 while (!list_empty(&fs_info->delayed_iputs)) {
2765 struct btrfs_inode *inode;
2767 inode = list_first_entry(&fs_info->delayed_iputs,
2768 struct btrfs_inode, delayed_iput);
2769 run_delayed_iput_locked(fs_info, inode);
2771 spin_unlock(&fs_info->delayed_iput_lock);
2775 * btrfs_wait_on_delayed_iputs - wait on the delayed iputs to be done running
2776 * @fs_info - the fs_info for this fs
2777 * @return - EINTR if we were killed, 0 if nothing's pending
2779 * This will wait on any delayed iputs that are currently running with KILLABLE
2780 * set. Once they are all done running we will return, unless we are killed in
2781 * which case we return EINTR. This helps in user operations like fallocate etc
2782 * that might get blocked on the iputs.
2784 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
2786 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
2787 atomic_read(&fs_info->nr_delayed_iputs) == 0);
2794 * This creates an orphan entry for the given inode in case something goes wrong
2795 * in the middle of an unlink.
2797 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
2798 struct btrfs_inode *inode)
2802 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
2803 if (ret && ret != -EEXIST) {
2804 btrfs_abort_transaction(trans, ret);
2812 * We have done the delete so we can go ahead and remove the orphan item for
2813 * this particular inode.
2815 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
2816 struct btrfs_inode *inode)
2818 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
2822 * this cleans up any orphans that may be left on the list from the last use
2825 int btrfs_orphan_cleanup(struct btrfs_root *root)
2827 struct btrfs_fs_info *fs_info = root->fs_info;
2828 struct btrfs_path *path;
2829 struct extent_buffer *leaf;
2830 struct btrfs_key key, found_key;
2831 struct btrfs_trans_handle *trans;
2832 struct inode *inode;
2833 u64 last_objectid = 0;
2834 int ret = 0, nr_unlink = 0;
2836 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2839 path = btrfs_alloc_path();
2844 path->reada = READA_BACK;
2846 key.objectid = BTRFS_ORPHAN_OBJECTID;
2847 key.type = BTRFS_ORPHAN_ITEM_KEY;
2848 key.offset = (u64)-1;
2851 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2856 * if ret == 0 means we found what we were searching for, which
2857 * is weird, but possible, so only screw with path if we didn't
2858 * find the key and see if we have stuff that matches
2862 if (path->slots[0] == 0)
2867 /* pull out the item */
2868 leaf = path->nodes[0];
2869 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2871 /* make sure the item matches what we want */
2872 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2874 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
2877 /* release the path since we're done with it */
2878 btrfs_release_path(path);
2881 * this is where we are basically btrfs_lookup, without the
2882 * crossing root thing. we store the inode number in the
2883 * offset of the orphan item.
2886 if (found_key.offset == last_objectid) {
2888 "Error removing orphan entry, stopping orphan cleanup");
2893 last_objectid = found_key.offset;
2895 found_key.objectid = found_key.offset;
2896 found_key.type = BTRFS_INODE_ITEM_KEY;
2897 found_key.offset = 0;
2898 inode = btrfs_iget(fs_info->sb, &found_key, root);
2899 ret = PTR_ERR_OR_ZERO(inode);
2900 if (ret && ret != -ENOENT)
2903 if (ret == -ENOENT && root == fs_info->tree_root) {
2904 struct btrfs_root *dead_root;
2905 struct btrfs_fs_info *fs_info = root->fs_info;
2906 int is_dead_root = 0;
2909 * this is an orphan in the tree root. Currently these
2910 * could come from 2 sources:
2911 * a) a snapshot deletion in progress
2912 * b) a free space cache inode
2913 * We need to distinguish those two, as the snapshot
2914 * orphan must not get deleted.
2915 * find_dead_roots already ran before us, so if this
2916 * is a snapshot deletion, we should find the root
2917 * in the dead_roots list
2919 spin_lock(&fs_info->trans_lock);
2920 list_for_each_entry(dead_root, &fs_info->dead_roots,
2922 if (dead_root->root_key.objectid ==
2923 found_key.objectid) {
2928 spin_unlock(&fs_info->trans_lock);
2930 /* prevent this orphan from being found again */
2931 key.offset = found_key.objectid - 1;
2938 * If we have an inode with links, there are a couple of
2939 * possibilities. Old kernels (before v3.12) used to create an
2940 * orphan item for truncate indicating that there were possibly
2941 * extent items past i_size that needed to be deleted. In v3.12,
2942 * truncate was changed to update i_size in sync with the extent
2943 * items, but the (useless) orphan item was still created. Since
2944 * v4.18, we don't create the orphan item for truncate at all.
2946 * So, this item could mean that we need to do a truncate, but
2947 * only if this filesystem was last used on a pre-v3.12 kernel
2948 * and was not cleanly unmounted. The odds of that are quite
2949 * slim, and it's a pain to do the truncate now, so just delete
2952 * It's also possible that this orphan item was supposed to be
2953 * deleted but wasn't. The inode number may have been reused,
2954 * but either way, we can delete the orphan item.
2956 if (ret == -ENOENT || inode->i_nlink) {
2959 trans = btrfs_start_transaction(root, 1);
2960 if (IS_ERR(trans)) {
2961 ret = PTR_ERR(trans);
2964 btrfs_debug(fs_info, "auto deleting %Lu",
2965 found_key.objectid);
2966 ret = btrfs_del_orphan_item(trans, root,
2967 found_key.objectid);
2968 btrfs_end_transaction(trans);
2976 /* this will do delete_inode and everything for us */
2979 /* release the path since we're done with it */
2980 btrfs_release_path(path);
2982 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2984 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
2985 trans = btrfs_join_transaction(root);
2987 btrfs_end_transaction(trans);
2991 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
2995 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
2996 btrfs_free_path(path);
3001 * very simple check to peek ahead in the leaf looking for xattrs. If we
3002 * don't find any xattrs, we know there can't be any acls.
3004 * slot is the slot the inode is in, objectid is the objectid of the inode
3006 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3007 int slot, u64 objectid,
3008 int *first_xattr_slot)
3010 u32 nritems = btrfs_header_nritems(leaf);
3011 struct btrfs_key found_key;
3012 static u64 xattr_access = 0;
3013 static u64 xattr_default = 0;
3016 if (!xattr_access) {
3017 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3018 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3019 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3020 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3024 *first_xattr_slot = -1;
3025 while (slot < nritems) {
3026 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3028 /* we found a different objectid, there must not be acls */
3029 if (found_key.objectid != objectid)
3032 /* we found an xattr, assume we've got an acl */
3033 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3034 if (*first_xattr_slot == -1)
3035 *first_xattr_slot = slot;
3036 if (found_key.offset == xattr_access ||
3037 found_key.offset == xattr_default)
3042 * we found a key greater than an xattr key, there can't
3043 * be any acls later on
3045 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3052 * it goes inode, inode backrefs, xattrs, extents,
3053 * so if there are a ton of hard links to an inode there can
3054 * be a lot of backrefs. Don't waste time searching too hard,
3055 * this is just an optimization
3060 /* we hit the end of the leaf before we found an xattr or
3061 * something larger than an xattr. We have to assume the inode
3064 if (*first_xattr_slot == -1)
3065 *first_xattr_slot = slot;
3070 * read an inode from the btree into the in-memory inode
3072 static int btrfs_read_locked_inode(struct inode *inode,
3073 struct btrfs_path *in_path)
3075 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3076 struct btrfs_path *path = in_path;
3077 struct extent_buffer *leaf;
3078 struct btrfs_inode_item *inode_item;
3079 struct btrfs_root *root = BTRFS_I(inode)->root;
3080 struct btrfs_key location;
3085 bool filled = false;
3086 int first_xattr_slot;
3088 ret = btrfs_fill_inode(inode, &rdev);
3093 path = btrfs_alloc_path();
3098 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3100 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3102 if (path != in_path)
3103 btrfs_free_path(path);
3107 leaf = path->nodes[0];
3112 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3113 struct btrfs_inode_item);
3114 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3115 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3116 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3117 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3118 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3120 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3121 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3123 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3124 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3126 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3127 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3129 BTRFS_I(inode)->i_otime.tv_sec =
3130 btrfs_timespec_sec(leaf, &inode_item->otime);
3131 BTRFS_I(inode)->i_otime.tv_nsec =
3132 btrfs_timespec_nsec(leaf, &inode_item->otime);
3134 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3135 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3136 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3138 inode_set_iversion_queried(inode,
3139 btrfs_inode_sequence(leaf, inode_item));
3140 inode->i_generation = BTRFS_I(inode)->generation;
3142 rdev = btrfs_inode_rdev(leaf, inode_item);
3144 BTRFS_I(inode)->index_cnt = (u64)-1;
3145 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3149 * If we were modified in the current generation and evicted from memory
3150 * and then re-read we need to do a full sync since we don't have any
3151 * idea about which extents were modified before we were evicted from
3154 * This is required for both inode re-read from disk and delayed inode
3155 * in delayed_nodes_tree.
3157 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3158 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3159 &BTRFS_I(inode)->runtime_flags);
3162 * We don't persist the id of the transaction where an unlink operation
3163 * against the inode was last made. So here we assume the inode might
3164 * have been evicted, and therefore the exact value of last_unlink_trans
3165 * lost, and set it to last_trans to avoid metadata inconsistencies
3166 * between the inode and its parent if the inode is fsync'ed and the log
3167 * replayed. For example, in the scenario:
3170 * ln mydir/foo mydir/bar
3173 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3174 * xfs_io -c fsync mydir/foo
3176 * mount fs, triggers fsync log replay
3178 * We must make sure that when we fsync our inode foo we also log its
3179 * parent inode, otherwise after log replay the parent still has the
3180 * dentry with the "bar" name but our inode foo has a link count of 1
3181 * and doesn't have an inode ref with the name "bar" anymore.
3183 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3184 * but it guarantees correctness at the expense of occasional full
3185 * transaction commits on fsync if our inode is a directory, or if our
3186 * inode is not a directory, logging its parent unnecessarily.
3188 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3191 if (inode->i_nlink != 1 ||
3192 path->slots[0] >= btrfs_header_nritems(leaf))
3195 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3196 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3199 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3200 if (location.type == BTRFS_INODE_REF_KEY) {
3201 struct btrfs_inode_ref *ref;
3203 ref = (struct btrfs_inode_ref *)ptr;
3204 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3205 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3206 struct btrfs_inode_extref *extref;
3208 extref = (struct btrfs_inode_extref *)ptr;
3209 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3214 * try to precache a NULL acl entry for files that don't have
3215 * any xattrs or acls
3217 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3218 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3219 if (first_xattr_slot != -1) {
3220 path->slots[0] = first_xattr_slot;
3221 ret = btrfs_load_inode_props(inode, path);
3224 "error loading props for ino %llu (root %llu): %d",
3225 btrfs_ino(BTRFS_I(inode)),
3226 root->root_key.objectid, ret);
3228 if (path != in_path)
3229 btrfs_free_path(path);
3232 cache_no_acl(inode);
3234 switch (inode->i_mode & S_IFMT) {
3236 inode->i_mapping->a_ops = &btrfs_aops;
3237 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3238 inode->i_fop = &btrfs_file_operations;
3239 inode->i_op = &btrfs_file_inode_operations;
3242 inode->i_fop = &btrfs_dir_file_operations;
3243 inode->i_op = &btrfs_dir_inode_operations;
3246 inode->i_op = &btrfs_symlink_inode_operations;
3247 inode_nohighmem(inode);
3248 inode->i_mapping->a_ops = &btrfs_aops;
3251 inode->i_op = &btrfs_special_inode_operations;
3252 init_special_inode(inode, inode->i_mode, rdev);
3256 btrfs_sync_inode_flags_to_i_flags(inode);
3261 * given a leaf and an inode, copy the inode fields into the leaf
3263 static void fill_inode_item(struct btrfs_trans_handle *trans,
3264 struct extent_buffer *leaf,
3265 struct btrfs_inode_item *item,
3266 struct inode *inode)
3268 struct btrfs_map_token token;
3270 btrfs_init_map_token(&token, leaf);
3272 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3273 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3274 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3276 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3277 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3279 btrfs_set_token_timespec_sec(leaf, &item->atime,
3280 inode->i_atime.tv_sec, &token);
3281 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3282 inode->i_atime.tv_nsec, &token);
3284 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3285 inode->i_mtime.tv_sec, &token);
3286 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3287 inode->i_mtime.tv_nsec, &token);
3289 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3290 inode->i_ctime.tv_sec, &token);
3291 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3292 inode->i_ctime.tv_nsec, &token);
3294 btrfs_set_token_timespec_sec(leaf, &item->otime,
3295 BTRFS_I(inode)->i_otime.tv_sec, &token);
3296 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3297 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3299 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3301 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3303 btrfs_set_token_inode_sequence(leaf, item, inode_peek_iversion(inode),
3305 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3306 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3307 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3308 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3312 * copy everything in the in-memory inode into the btree.
3314 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3315 struct btrfs_root *root, struct inode *inode)
3317 struct btrfs_inode_item *inode_item;
3318 struct btrfs_path *path;
3319 struct extent_buffer *leaf;
3322 path = btrfs_alloc_path();
3326 path->leave_spinning = 1;
3327 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3335 leaf = path->nodes[0];
3336 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3337 struct btrfs_inode_item);
3339 fill_inode_item(trans, leaf, inode_item, inode);
3340 btrfs_mark_buffer_dirty(leaf);
3341 btrfs_set_inode_last_trans(trans, inode);
3344 btrfs_free_path(path);
3349 * copy everything in the in-memory inode into the btree.
3351 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3352 struct btrfs_root *root, struct inode *inode)
3354 struct btrfs_fs_info *fs_info = root->fs_info;
3358 * If the inode is a free space inode, we can deadlock during commit
3359 * if we put it into the delayed code.
3361 * The data relocation inode should also be directly updated
3364 if (!btrfs_is_free_space_inode(BTRFS_I(inode))
3365 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3366 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
3367 btrfs_update_root_times(trans, root);
3369 ret = btrfs_delayed_update_inode(trans, root, inode);
3371 btrfs_set_inode_last_trans(trans, inode);
3375 return btrfs_update_inode_item(trans, root, inode);
3378 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3379 struct btrfs_root *root,
3380 struct inode *inode)
3384 ret = btrfs_update_inode(trans, root, inode);
3386 return btrfs_update_inode_item(trans, root, inode);
3391 * unlink helper that gets used here in inode.c and in the tree logging
3392 * recovery code. It remove a link in a directory with a given name, and
3393 * also drops the back refs in the inode to the directory
3395 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3396 struct btrfs_root *root,
3397 struct btrfs_inode *dir,
3398 struct btrfs_inode *inode,
3399 const char *name, int name_len)
3401 struct btrfs_fs_info *fs_info = root->fs_info;
3402 struct btrfs_path *path;
3404 struct btrfs_dir_item *di;
3406 u64 ino = btrfs_ino(inode);
3407 u64 dir_ino = btrfs_ino(dir);
3409 path = btrfs_alloc_path();
3415 path->leave_spinning = 1;
3416 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3417 name, name_len, -1);
3418 if (IS_ERR_OR_NULL(di)) {
3419 ret = di ? PTR_ERR(di) : -ENOENT;
3422 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3425 btrfs_release_path(path);
3428 * If we don't have dir index, we have to get it by looking up
3429 * the inode ref, since we get the inode ref, remove it directly,
3430 * it is unnecessary to do delayed deletion.
3432 * But if we have dir index, needn't search inode ref to get it.
3433 * Since the inode ref is close to the inode item, it is better
3434 * that we delay to delete it, and just do this deletion when
3435 * we update the inode item.
3437 if (inode->dir_index) {
3438 ret = btrfs_delayed_delete_inode_ref(inode);
3440 index = inode->dir_index;
3445 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3449 "failed to delete reference to %.*s, inode %llu parent %llu",
3450 name_len, name, ino, dir_ino);
3451 btrfs_abort_transaction(trans, ret);
3455 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
3457 btrfs_abort_transaction(trans, ret);
3461 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
3463 if (ret != 0 && ret != -ENOENT) {
3464 btrfs_abort_transaction(trans, ret);
3468 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
3473 btrfs_abort_transaction(trans, ret);
3476 * If we have a pending delayed iput we could end up with the final iput
3477 * being run in btrfs-cleaner context. If we have enough of these built
3478 * up we can end up burning a lot of time in btrfs-cleaner without any
3479 * way to throttle the unlinks. Since we're currently holding a ref on
3480 * the inode we can run the delayed iput here without any issues as the
3481 * final iput won't be done until after we drop the ref we're currently
3484 btrfs_run_delayed_iput(fs_info, inode);
3486 btrfs_free_path(path);
3490 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
3491 inode_inc_iversion(&inode->vfs_inode);
3492 inode_inc_iversion(&dir->vfs_inode);
3493 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
3494 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
3495 ret = btrfs_update_inode(trans, root, &dir->vfs_inode);
3500 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3501 struct btrfs_root *root,
3502 struct btrfs_inode *dir, struct btrfs_inode *inode,
3503 const char *name, int name_len)
3506 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3508 drop_nlink(&inode->vfs_inode);
3509 ret = btrfs_update_inode(trans, root, &inode->vfs_inode);
3515 * helper to start transaction for unlink and rmdir.
3517 * unlink and rmdir are special in btrfs, they do not always free space, so
3518 * if we cannot make our reservations the normal way try and see if there is
3519 * plenty of slack room in the global reserve to migrate, otherwise we cannot
3520 * allow the unlink to occur.
3522 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3524 struct btrfs_root *root = BTRFS_I(dir)->root;
3527 * 1 for the possible orphan item
3528 * 1 for the dir item
3529 * 1 for the dir index
3530 * 1 for the inode ref
3533 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
3536 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3538 struct btrfs_root *root = BTRFS_I(dir)->root;
3539 struct btrfs_trans_handle *trans;
3540 struct inode *inode = d_inode(dentry);
3543 trans = __unlink_start_trans(dir);
3545 return PTR_ERR(trans);
3547 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
3550 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
3551 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
3552 dentry->d_name.len);
3556 if (inode->i_nlink == 0) {
3557 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
3563 btrfs_end_transaction(trans);
3564 btrfs_btree_balance_dirty(root->fs_info);
3568 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3569 struct inode *dir, struct dentry *dentry)
3571 struct btrfs_root *root = BTRFS_I(dir)->root;
3572 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
3573 struct btrfs_path *path;
3574 struct extent_buffer *leaf;
3575 struct btrfs_dir_item *di;
3576 struct btrfs_key key;
3577 const char *name = dentry->d_name.name;
3578 int name_len = dentry->d_name.len;
3582 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
3584 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
3585 objectid = inode->root->root_key.objectid;
3586 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
3587 objectid = inode->location.objectid;
3593 path = btrfs_alloc_path();
3597 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3598 name, name_len, -1);
3599 if (IS_ERR_OR_NULL(di)) {
3600 ret = di ? PTR_ERR(di) : -ENOENT;
3604 leaf = path->nodes[0];
3605 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3606 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3607 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3609 btrfs_abort_transaction(trans, ret);
3612 btrfs_release_path(path);
3615 * This is a placeholder inode for a subvolume we didn't have a
3616 * reference to at the time of the snapshot creation. In the meantime
3617 * we could have renamed the real subvol link into our snapshot, so
3618 * depending on btrfs_del_root_ref to return -ENOENT here is incorret.
3619 * Instead simply lookup the dir_index_item for this entry so we can
3620 * remove it. Otherwise we know we have a ref to the root and we can
3621 * call btrfs_del_root_ref, and it _shouldn't_ fail.
3623 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
3624 di = btrfs_search_dir_index_item(root, path, dir_ino,
3626 if (IS_ERR_OR_NULL(di)) {
3631 btrfs_abort_transaction(trans, ret);
3635 leaf = path->nodes[0];
3636 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3638 btrfs_release_path(path);
3640 ret = btrfs_del_root_ref(trans, objectid,
3641 root->root_key.objectid, dir_ino,
3642 &index, name, name_len);
3644 btrfs_abort_transaction(trans, ret);
3649 ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index);
3651 btrfs_abort_transaction(trans, ret);
3655 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
3656 inode_inc_iversion(dir);
3657 dir->i_mtime = dir->i_ctime = current_time(dir);
3658 ret = btrfs_update_inode_fallback(trans, root, dir);
3660 btrfs_abort_transaction(trans, ret);
3662 btrfs_free_path(path);
3667 * Helper to check if the subvolume references other subvolumes or if it's
3670 static noinline int may_destroy_subvol(struct btrfs_root *root)
3672 struct btrfs_fs_info *fs_info = root->fs_info;
3673 struct btrfs_path *path;
3674 struct btrfs_dir_item *di;
3675 struct btrfs_key key;
3679 path = btrfs_alloc_path();
3683 /* Make sure this root isn't set as the default subvol */
3684 dir_id = btrfs_super_root_dir(fs_info->super_copy);
3685 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
3686 dir_id, "default", 7, 0);
3687 if (di && !IS_ERR(di)) {
3688 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
3689 if (key.objectid == root->root_key.objectid) {
3692 "deleting default subvolume %llu is not allowed",
3696 btrfs_release_path(path);
3699 key.objectid = root->root_key.objectid;
3700 key.type = BTRFS_ROOT_REF_KEY;
3701 key.offset = (u64)-1;
3703 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3709 if (path->slots[0] > 0) {
3711 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3712 if (key.objectid == root->root_key.objectid &&
3713 key.type == BTRFS_ROOT_REF_KEY)
3717 btrfs_free_path(path);
3721 /* Delete all dentries for inodes belonging to the root */
3722 static void btrfs_prune_dentries(struct btrfs_root *root)
3724 struct btrfs_fs_info *fs_info = root->fs_info;
3725 struct rb_node *node;
3726 struct rb_node *prev;
3727 struct btrfs_inode *entry;
3728 struct inode *inode;
3731 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3732 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3734 spin_lock(&root->inode_lock);
3736 node = root->inode_tree.rb_node;
3740 entry = rb_entry(node, struct btrfs_inode, rb_node);
3742 if (objectid < btrfs_ino(entry))
3743 node = node->rb_left;
3744 else if (objectid > btrfs_ino(entry))
3745 node = node->rb_right;
3751 entry = rb_entry(prev, struct btrfs_inode, rb_node);
3752 if (objectid <= btrfs_ino(entry)) {
3756 prev = rb_next(prev);
3760 entry = rb_entry(node, struct btrfs_inode, rb_node);
3761 objectid = btrfs_ino(entry) + 1;
3762 inode = igrab(&entry->vfs_inode);
3764 spin_unlock(&root->inode_lock);
3765 if (atomic_read(&inode->i_count) > 1)
3766 d_prune_aliases(inode);
3768 * btrfs_drop_inode will have it removed from the inode
3769 * cache when its usage count hits zero.
3773 spin_lock(&root->inode_lock);
3777 if (cond_resched_lock(&root->inode_lock))
3780 node = rb_next(node);
3782 spin_unlock(&root->inode_lock);
3785 int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
3787 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
3788 struct btrfs_root *root = BTRFS_I(dir)->root;
3789 struct inode *inode = d_inode(dentry);
3790 struct btrfs_root *dest = BTRFS_I(inode)->root;
3791 struct btrfs_trans_handle *trans;
3792 struct btrfs_block_rsv block_rsv;
3798 * Don't allow to delete a subvolume with send in progress. This is
3799 * inside the inode lock so the error handling that has to drop the bit
3800 * again is not run concurrently.
3802 spin_lock(&dest->root_item_lock);
3803 if (dest->send_in_progress) {
3804 spin_unlock(&dest->root_item_lock);
3806 "attempt to delete subvolume %llu during send",
3807 dest->root_key.objectid);
3810 root_flags = btrfs_root_flags(&dest->root_item);
3811 btrfs_set_root_flags(&dest->root_item,
3812 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
3813 spin_unlock(&dest->root_item_lock);
3815 down_write(&fs_info->subvol_sem);
3817 err = may_destroy_subvol(dest);
3821 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
3823 * One for dir inode,
3824 * two for dir entries,
3825 * two for root ref/backref.
3827 err = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
3831 trans = btrfs_start_transaction(root, 0);
3832 if (IS_ERR(trans)) {
3833 err = PTR_ERR(trans);
3836 trans->block_rsv = &block_rsv;
3837 trans->bytes_reserved = block_rsv.size;
3839 btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
3841 ret = btrfs_unlink_subvol(trans, dir, dentry);
3844 btrfs_abort_transaction(trans, ret);
3848 btrfs_record_root_in_trans(trans, dest);
3850 memset(&dest->root_item.drop_progress, 0,
3851 sizeof(dest->root_item.drop_progress));
3852 dest->root_item.drop_level = 0;
3853 btrfs_set_root_refs(&dest->root_item, 0);
3855 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
3856 ret = btrfs_insert_orphan_item(trans,
3858 dest->root_key.objectid);
3860 btrfs_abort_transaction(trans, ret);
3866 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
3867 BTRFS_UUID_KEY_SUBVOL,
3868 dest->root_key.objectid);
3869 if (ret && ret != -ENOENT) {
3870 btrfs_abort_transaction(trans, ret);
3874 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
3875 ret = btrfs_uuid_tree_remove(trans,
3876 dest->root_item.received_uuid,
3877 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
3878 dest->root_key.objectid);
3879 if (ret && ret != -ENOENT) {
3880 btrfs_abort_transaction(trans, ret);
3887 trans->block_rsv = NULL;
3888 trans->bytes_reserved = 0;
3889 ret = btrfs_end_transaction(trans);
3892 inode->i_flags |= S_DEAD;
3894 btrfs_subvolume_release_metadata(fs_info, &block_rsv);
3896 up_write(&fs_info->subvol_sem);
3898 spin_lock(&dest->root_item_lock);
3899 root_flags = btrfs_root_flags(&dest->root_item);
3900 btrfs_set_root_flags(&dest->root_item,
3901 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
3902 spin_unlock(&dest->root_item_lock);
3904 d_invalidate(dentry);
3905 btrfs_prune_dentries(dest);
3906 ASSERT(dest->send_in_progress == 0);
3909 if (dest->ino_cache_inode) {
3910 iput(dest->ino_cache_inode);
3911 dest->ino_cache_inode = NULL;
3918 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3920 struct inode *inode = d_inode(dentry);
3922 struct btrfs_root *root = BTRFS_I(dir)->root;
3923 struct btrfs_trans_handle *trans;
3924 u64 last_unlink_trans;
3926 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3928 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
3929 return btrfs_delete_subvolume(dir, dentry);
3931 trans = __unlink_start_trans(dir);
3933 return PTR_ERR(trans);
3935 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3936 err = btrfs_unlink_subvol(trans, dir, dentry);
3940 err = btrfs_orphan_add(trans, BTRFS_I(inode));
3944 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
3946 /* now the directory is empty */
3947 err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
3948 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
3949 dentry->d_name.len);
3951 btrfs_i_size_write(BTRFS_I(inode), 0);
3953 * Propagate the last_unlink_trans value of the deleted dir to
3954 * its parent directory. This is to prevent an unrecoverable
3955 * log tree in the case we do something like this:
3957 * 2) create snapshot under dir foo
3958 * 3) delete the snapshot
3961 * 6) fsync foo or some file inside foo
3963 if (last_unlink_trans >= trans->transid)
3964 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
3967 btrfs_end_transaction(trans);
3968 btrfs_btree_balance_dirty(root->fs_info);
3974 * Return this if we need to call truncate_block for the last bit of the
3977 #define NEED_TRUNCATE_BLOCK 1
3980 * this can truncate away extent items, csum items and directory items.
3981 * It starts at a high offset and removes keys until it can't find
3982 * any higher than new_size
3984 * csum items that cross the new i_size are truncated to the new size
3987 * min_type is the minimum key type to truncate down to. If set to 0, this
3988 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3990 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3991 struct btrfs_root *root,
3992 struct inode *inode,
3993 u64 new_size, u32 min_type)
3995 struct btrfs_fs_info *fs_info = root->fs_info;
3996 struct btrfs_path *path;
3997 struct extent_buffer *leaf;
3998 struct btrfs_file_extent_item *fi;
3999 struct btrfs_key key;
4000 struct btrfs_key found_key;
4001 u64 extent_start = 0;
4002 u64 extent_num_bytes = 0;
4003 u64 extent_offset = 0;
4005 u64 last_size = new_size;
4006 u32 found_type = (u8)-1;
4009 int pending_del_nr = 0;
4010 int pending_del_slot = 0;
4011 int extent_type = -1;
4013 u64 ino = btrfs_ino(BTRFS_I(inode));
4014 u64 bytes_deleted = 0;
4015 bool be_nice = false;
4016 bool should_throttle = false;
4018 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4021 * for non-free space inodes and ref cows, we want to back off from
4024 if (!btrfs_is_free_space_inode(BTRFS_I(inode)) &&
4025 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4028 path = btrfs_alloc_path();
4031 path->reada = READA_BACK;
4034 * We want to drop from the next block forward in case this new size is
4035 * not block aligned since we will be keeping the last block of the
4036 * extent just the way it is.
4038 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4039 root == fs_info->tree_root)
4040 btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size,
4041 fs_info->sectorsize),
4045 * This function is also used to drop the items in the log tree before
4046 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4047 * it is used to drop the logged items. So we shouldn't kill the delayed
4050 if (min_type == 0 && root == BTRFS_I(inode)->root)
4051 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
4054 key.offset = (u64)-1;
4059 * with a 16K leaf size and 128MB extents, you can actually queue
4060 * up a huge file in a single leaf. Most of the time that
4061 * bytes_deleted is > 0, it will be huge by the time we get here
4063 if (be_nice && bytes_deleted > SZ_32M &&
4064 btrfs_should_end_transaction(trans)) {
4069 path->leave_spinning = 1;
4070 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4076 /* there are no items in the tree for us to truncate, we're
4079 if (path->slots[0] == 0)
4086 leaf = path->nodes[0];
4087 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4088 found_type = found_key.type;
4090 if (found_key.objectid != ino)
4093 if (found_type < min_type)
4096 item_end = found_key.offset;
4097 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4098 fi = btrfs_item_ptr(leaf, path->slots[0],
4099 struct btrfs_file_extent_item);
4100 extent_type = btrfs_file_extent_type(leaf, fi);
4101 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4103 btrfs_file_extent_num_bytes(leaf, fi);
4105 trace_btrfs_truncate_show_fi_regular(
4106 BTRFS_I(inode), leaf, fi,
4108 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4109 item_end += btrfs_file_extent_ram_bytes(leaf,
4112 trace_btrfs_truncate_show_fi_inline(
4113 BTRFS_I(inode), leaf, fi, path->slots[0],
4118 if (found_type > min_type) {
4121 if (item_end < new_size)
4123 if (found_key.offset >= new_size)
4129 /* FIXME, shrink the extent if the ref count is only 1 */
4130 if (found_type != BTRFS_EXTENT_DATA_KEY)
4133 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4135 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4137 u64 orig_num_bytes =
4138 btrfs_file_extent_num_bytes(leaf, fi);
4139 extent_num_bytes = ALIGN(new_size -
4141 fs_info->sectorsize);
4142 btrfs_set_file_extent_num_bytes(leaf, fi,
4144 num_dec = (orig_num_bytes -
4146 if (test_bit(BTRFS_ROOT_REF_COWS,
4149 inode_sub_bytes(inode, num_dec);
4150 btrfs_mark_buffer_dirty(leaf);
4153 btrfs_file_extent_disk_num_bytes(leaf,
4155 extent_offset = found_key.offset -
4156 btrfs_file_extent_offset(leaf, fi);
4158 /* FIXME blocksize != 4096 */
4159 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4160 if (extent_start != 0) {
4162 if (test_bit(BTRFS_ROOT_REF_COWS,
4164 inode_sub_bytes(inode, num_dec);
4167 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4169 * we can't truncate inline items that have had
4173 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4174 btrfs_file_extent_other_encoding(leaf, fi) == 0 &&
4175 btrfs_file_extent_compression(leaf, fi) == 0) {
4176 u32 size = (u32)(new_size - found_key.offset);
4178 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4179 size = btrfs_file_extent_calc_inline_size(size);
4180 btrfs_truncate_item(path, size, 1);
4181 } else if (!del_item) {
4183 * We have to bail so the last_size is set to
4184 * just before this extent.
4186 ret = NEED_TRUNCATE_BLOCK;
4190 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4191 inode_sub_bytes(inode, item_end + 1 - new_size);
4195 last_size = found_key.offset;
4197 last_size = new_size;
4199 if (!pending_del_nr) {
4200 /* no pending yet, add ourselves */
4201 pending_del_slot = path->slots[0];
4203 } else if (pending_del_nr &&
4204 path->slots[0] + 1 == pending_del_slot) {
4205 /* hop on the pending chunk */
4207 pending_del_slot = path->slots[0];
4214 should_throttle = false;
4217 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4218 root == fs_info->tree_root)) {
4219 struct btrfs_ref ref = { 0 };
4221 btrfs_set_path_blocking(path);
4222 bytes_deleted += extent_num_bytes;
4224 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF,
4225 extent_start, extent_num_bytes, 0);
4226 ref.real_root = root->root_key.objectid;
4227 btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
4228 ino, extent_offset);
4229 ret = btrfs_free_extent(trans, &ref);
4231 btrfs_abort_transaction(trans, ret);
4235 if (btrfs_should_throttle_delayed_refs(trans))
4236 should_throttle = true;
4240 if (found_type == BTRFS_INODE_ITEM_KEY)
4243 if (path->slots[0] == 0 ||
4244 path->slots[0] != pending_del_slot ||
4246 if (pending_del_nr) {
4247 ret = btrfs_del_items(trans, root, path,
4251 btrfs_abort_transaction(trans, ret);
4256 btrfs_release_path(path);
4259 * We can generate a lot of delayed refs, so we need to
4260 * throttle every once and a while and make sure we're
4261 * adding enough space to keep up with the work we are
4262 * generating. Since we hold a transaction here we
4263 * can't flush, and we don't want to FLUSH_LIMIT because
4264 * we could have generated too many delayed refs to
4265 * actually allocate, so just bail if we're short and
4266 * let the normal reservation dance happen higher up.
4268 if (should_throttle) {
4269 ret = btrfs_delayed_refs_rsv_refill(fs_info,
4270 BTRFS_RESERVE_NO_FLUSH);
4282 if (ret >= 0 && pending_del_nr) {
4285 err = btrfs_del_items(trans, root, path, pending_del_slot,
4288 btrfs_abort_transaction(trans, err);
4292 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4293 ASSERT(last_size >= new_size);
4294 if (!ret && last_size > new_size)
4295 last_size = new_size;
4296 btrfs_ordered_update_i_size(inode, last_size, NULL);
4299 btrfs_free_path(path);
4304 * btrfs_truncate_block - read, zero a chunk and write a block
4305 * @inode - inode that we're zeroing
4306 * @from - the offset to start zeroing
4307 * @len - the length to zero, 0 to zero the entire range respective to the
4309 * @front - zero up to the offset instead of from the offset on
4311 * This will find the block for the "from" offset and cow the block and zero the
4312 * part we want to zero. This is used with truncate and hole punching.
4314 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4317 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4318 struct address_space *mapping = inode->i_mapping;
4319 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4320 struct btrfs_ordered_extent *ordered;
4321 struct extent_state *cached_state = NULL;
4322 struct extent_changeset *data_reserved = NULL;
4324 u32 blocksize = fs_info->sectorsize;
4325 pgoff_t index = from >> PAGE_SHIFT;
4326 unsigned offset = from & (blocksize - 1);
4328 gfp_t mask = btrfs_alloc_write_mask(mapping);
4333 if (IS_ALIGNED(offset, blocksize) &&
4334 (!len || IS_ALIGNED(len, blocksize)))
4337 block_start = round_down(from, blocksize);
4338 block_end = block_start + blocksize - 1;
4340 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
4341 block_start, blocksize);
4346 page = find_or_create_page(mapping, index, mask);
4348 btrfs_delalloc_release_space(inode, data_reserved,
4349 block_start, blocksize, true);
4350 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize);
4355 if (!PageUptodate(page)) {
4356 ret = btrfs_readpage(NULL, page);
4358 if (page->mapping != mapping) {
4363 if (!PageUptodate(page)) {
4368 wait_on_page_writeback(page);
4370 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4371 set_page_extent_mapped(page);
4373 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4375 unlock_extent_cached(io_tree, block_start, block_end,
4379 btrfs_start_ordered_extent(inode, ordered, 1);
4380 btrfs_put_ordered_extent(ordered);
4384 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4385 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4386 0, 0, &cached_state);
4388 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4391 unlock_extent_cached(io_tree, block_start, block_end,
4396 if (offset != blocksize) {
4398 len = blocksize - offset;
4401 memset(kaddr + (block_start - page_offset(page)),
4404 memset(kaddr + (block_start - page_offset(page)) + offset,
4406 flush_dcache_page(page);
4409 ClearPageChecked(page);
4410 set_page_dirty(page);
4411 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4415 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4417 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize);
4421 extent_changeset_free(data_reserved);
4425 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4426 u64 offset, u64 len)
4428 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4429 struct btrfs_trans_handle *trans;
4433 * Still need to make sure the inode looks like it's been updated so
4434 * that any holes get logged if we fsync.
4436 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4437 BTRFS_I(inode)->last_trans = fs_info->generation;
4438 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4439 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4444 * 1 - for the one we're dropping
4445 * 1 - for the one we're adding
4446 * 1 - for updating the inode.
4448 trans = btrfs_start_transaction(root, 3);
4450 return PTR_ERR(trans);
4452 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4454 btrfs_abort_transaction(trans, ret);
4455 btrfs_end_transaction(trans);
4459 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)),
4460 offset, 0, 0, len, 0, len, 0, 0, 0);
4462 btrfs_abort_transaction(trans, ret);
4464 btrfs_update_inode(trans, root, inode);
4465 btrfs_end_transaction(trans);
4470 * This function puts in dummy file extents for the area we're creating a hole
4471 * for. So if we are truncating this file to a larger size we need to insert
4472 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4473 * the range between oldsize and size
4475 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4477 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4478 struct btrfs_root *root = BTRFS_I(inode)->root;
4479 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4480 struct extent_map *em = NULL;
4481 struct extent_state *cached_state = NULL;
4482 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4483 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4484 u64 block_end = ALIGN(size, fs_info->sectorsize);
4491 * If our size started in the middle of a block we need to zero out the
4492 * rest of the block before we expand the i_size, otherwise we could
4493 * expose stale data.
4495 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4499 if (size <= hole_start)
4502 btrfs_lock_and_flush_ordered_range(io_tree, BTRFS_I(inode), hole_start,
4503 block_end - 1, &cached_state);
4504 cur_offset = hole_start;
4506 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
4507 block_end - cur_offset);
4513 last_byte = min(extent_map_end(em), block_end);
4514 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4515 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4516 struct extent_map *hole_em;
4517 hole_size = last_byte - cur_offset;
4519 err = maybe_insert_hole(root, inode, cur_offset,
4523 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
4524 cur_offset + hole_size - 1, 0);
4525 hole_em = alloc_extent_map();
4527 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4528 &BTRFS_I(inode)->runtime_flags);
4531 hole_em->start = cur_offset;
4532 hole_em->len = hole_size;
4533 hole_em->orig_start = cur_offset;
4535 hole_em->block_start = EXTENT_MAP_HOLE;
4536 hole_em->block_len = 0;
4537 hole_em->orig_block_len = 0;
4538 hole_em->ram_bytes = hole_size;
4539 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4540 hole_em->generation = fs_info->generation;
4543 write_lock(&em_tree->lock);
4544 err = add_extent_mapping(em_tree, hole_em, 1);
4545 write_unlock(&em_tree->lock);
4548 btrfs_drop_extent_cache(BTRFS_I(inode),
4553 free_extent_map(hole_em);
4556 free_extent_map(em);
4558 cur_offset = last_byte;
4559 if (cur_offset >= block_end)
4562 free_extent_map(em);
4563 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
4567 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4569 struct btrfs_root *root = BTRFS_I(inode)->root;
4570 struct btrfs_trans_handle *trans;
4571 loff_t oldsize = i_size_read(inode);
4572 loff_t newsize = attr->ia_size;
4573 int mask = attr->ia_valid;
4577 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4578 * special case where we need to update the times despite not having
4579 * these flags set. For all other operations the VFS set these flags
4580 * explicitly if it wants a timestamp update.
4582 if (newsize != oldsize) {
4583 inode_inc_iversion(inode);
4584 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4585 inode->i_ctime = inode->i_mtime =
4586 current_time(inode);
4589 if (newsize > oldsize) {
4591 * Don't do an expanding truncate while snapshotting is ongoing.
4592 * This is to ensure the snapshot captures a fully consistent
4593 * state of this file - if the snapshot captures this expanding
4594 * truncation, it must capture all writes that happened before
4597 btrfs_wait_for_snapshot_creation(root);
4598 ret = btrfs_cont_expand(inode, oldsize, newsize);
4600 btrfs_end_write_no_snapshotting(root);
4604 trans = btrfs_start_transaction(root, 1);
4605 if (IS_ERR(trans)) {
4606 btrfs_end_write_no_snapshotting(root);
4607 return PTR_ERR(trans);
4610 i_size_write(inode, newsize);
4611 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4612 pagecache_isize_extended(inode, oldsize, newsize);
4613 ret = btrfs_update_inode(trans, root, inode);
4614 btrfs_end_write_no_snapshotting(root);
4615 btrfs_end_transaction(trans);
4619 * We're truncating a file that used to have good data down to
4620 * zero. Make sure it gets into the ordered flush list so that
4621 * any new writes get down to disk quickly.
4624 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4625 &BTRFS_I(inode)->runtime_flags);
4627 truncate_setsize(inode, newsize);
4629 /* Disable nonlocked read DIO to avoid the endless truncate */
4630 btrfs_inode_block_unlocked_dio(BTRFS_I(inode));
4631 inode_dio_wait(inode);
4632 btrfs_inode_resume_unlocked_dio(BTRFS_I(inode));
4634 ret = btrfs_truncate(inode, newsize == oldsize);
4635 if (ret && inode->i_nlink) {
4639 * Truncate failed, so fix up the in-memory size. We
4640 * adjusted disk_i_size down as we removed extents, so
4641 * wait for disk_i_size to be stable and then update the
4642 * in-memory size to match.
4644 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
4647 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4654 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4656 struct inode *inode = d_inode(dentry);
4657 struct btrfs_root *root = BTRFS_I(inode)->root;
4660 if (btrfs_root_readonly(root))
4663 err = setattr_prepare(dentry, attr);
4667 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4668 err = btrfs_setsize(inode, attr);
4673 if (attr->ia_valid) {
4674 setattr_copy(inode, attr);
4675 inode_inc_iversion(inode);
4676 err = btrfs_dirty_inode(inode);
4678 if (!err && attr->ia_valid & ATTR_MODE)
4679 err = posix_acl_chmod(inode, inode->i_mode);
4686 * While truncating the inode pages during eviction, we get the VFS calling
4687 * btrfs_invalidatepage() against each page of the inode. This is slow because
4688 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4689 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4690 * extent_state structures over and over, wasting lots of time.
4692 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4693 * those expensive operations on a per page basis and do only the ordered io
4694 * finishing, while we release here the extent_map and extent_state structures,
4695 * without the excessive merging and splitting.
4697 static void evict_inode_truncate_pages(struct inode *inode)
4699 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4700 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
4701 struct rb_node *node;
4703 ASSERT(inode->i_state & I_FREEING);
4704 truncate_inode_pages_final(&inode->i_data);
4706 write_lock(&map_tree->lock);
4707 while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) {
4708 struct extent_map *em;
4710 node = rb_first_cached(&map_tree->map);
4711 em = rb_entry(node, struct extent_map, rb_node);
4712 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
4713 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
4714 remove_extent_mapping(map_tree, em);
4715 free_extent_map(em);
4716 if (need_resched()) {
4717 write_unlock(&map_tree->lock);
4719 write_lock(&map_tree->lock);
4722 write_unlock(&map_tree->lock);
4725 * Keep looping until we have no more ranges in the io tree.
4726 * We can have ongoing bios started by readpages (called from readahead)
4727 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
4728 * still in progress (unlocked the pages in the bio but did not yet
4729 * unlocked the ranges in the io tree). Therefore this means some
4730 * ranges can still be locked and eviction started because before
4731 * submitting those bios, which are executed by a separate task (work
4732 * queue kthread), inode references (inode->i_count) were not taken
4733 * (which would be dropped in the end io callback of each bio).
4734 * Therefore here we effectively end up waiting for those bios and
4735 * anyone else holding locked ranges without having bumped the inode's
4736 * reference count - if we don't do it, when they access the inode's
4737 * io_tree to unlock a range it may be too late, leading to an
4738 * use-after-free issue.
4740 spin_lock(&io_tree->lock);
4741 while (!RB_EMPTY_ROOT(&io_tree->state)) {
4742 struct extent_state *state;
4743 struct extent_state *cached_state = NULL;
4746 unsigned state_flags;
4748 node = rb_first(&io_tree->state);
4749 state = rb_entry(node, struct extent_state, rb_node);
4750 start = state->start;
4752 state_flags = state->state;
4753 spin_unlock(&io_tree->lock);
4755 lock_extent_bits(io_tree, start, end, &cached_state);
4758 * If still has DELALLOC flag, the extent didn't reach disk,
4759 * and its reserved space won't be freed by delayed_ref.
4760 * So we need to free its reserved space here.
4761 * (Refer to comment in btrfs_invalidatepage, case 2)
4763 * Note, end is the bytenr of last byte, so we need + 1 here.
4765 if (state_flags & EXTENT_DELALLOC)
4766 btrfs_qgroup_free_data(inode, NULL, start, end - start + 1);
4768 clear_extent_bit(io_tree, start, end,
4769 EXTENT_LOCKED | EXTENT_DELALLOC |
4770 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
4774 spin_lock(&io_tree->lock);
4776 spin_unlock(&io_tree->lock);
4779 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
4780 struct btrfs_block_rsv *rsv)
4782 struct btrfs_fs_info *fs_info = root->fs_info;
4783 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4784 struct btrfs_trans_handle *trans;
4785 u64 delayed_refs_extra = btrfs_calc_insert_metadata_size(fs_info, 1);
4789 * Eviction should be taking place at some place safe because of our
4790 * delayed iputs. However the normal flushing code will run delayed
4791 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
4793 * We reserve the delayed_refs_extra here again because we can't use
4794 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
4795 * above. We reserve our extra bit here because we generate a ton of
4796 * delayed refs activity by truncating.
4798 * If we cannot make our reservation we'll attempt to steal from the
4799 * global reserve, because we really want to be able to free up space.
4801 ret = btrfs_block_rsv_refill(root, rsv, rsv->size + delayed_refs_extra,
4802 BTRFS_RESERVE_FLUSH_EVICT);
4805 * Try to steal from the global reserve if there is space for
4808 if (btrfs_check_space_for_delayed_refs(fs_info) ||
4809 btrfs_block_rsv_migrate(global_rsv, rsv, rsv->size, 0)) {
4811 "could not allocate space for delete; will truncate on mount");
4812 return ERR_PTR(-ENOSPC);
4814 delayed_refs_extra = 0;
4817 trans = btrfs_join_transaction(root);
4821 if (delayed_refs_extra) {
4822 trans->block_rsv = &fs_info->trans_block_rsv;
4823 trans->bytes_reserved = delayed_refs_extra;
4824 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
4825 delayed_refs_extra, 1);
4830 void btrfs_evict_inode(struct inode *inode)
4832 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4833 struct btrfs_trans_handle *trans;
4834 struct btrfs_root *root = BTRFS_I(inode)->root;
4835 struct btrfs_block_rsv *rsv;
4838 trace_btrfs_inode_evict(inode);
4845 evict_inode_truncate_pages(inode);
4847 if (inode->i_nlink &&
4848 ((btrfs_root_refs(&root->root_item) != 0 &&
4849 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
4850 btrfs_is_free_space_inode(BTRFS_I(inode))))
4853 if (is_bad_inode(inode))
4856 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
4858 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
4861 if (inode->i_nlink > 0) {
4862 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
4863 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
4867 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
4871 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
4874 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
4877 btrfs_i_size_write(BTRFS_I(inode), 0);
4880 trans = evict_refill_and_join(root, rsv);
4884 trans->block_rsv = rsv;
4886 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4887 trans->block_rsv = &fs_info->trans_block_rsv;
4888 btrfs_end_transaction(trans);
4889 btrfs_btree_balance_dirty(fs_info);
4890 if (ret && ret != -ENOSPC && ret != -EAGAIN)
4897 * Errors here aren't a big deal, it just means we leave orphan items in
4898 * the tree. They will be cleaned up on the next mount. If the inode
4899 * number gets reused, cleanup deletes the orphan item without doing
4900 * anything, and unlink reuses the existing orphan item.
4902 * If it turns out that we are dropping too many of these, we might want
4903 * to add a mechanism for retrying these after a commit.
4905 trans = evict_refill_and_join(root, rsv);
4906 if (!IS_ERR(trans)) {
4907 trans->block_rsv = rsv;
4908 btrfs_orphan_del(trans, BTRFS_I(inode));
4909 trans->block_rsv = &fs_info->trans_block_rsv;
4910 btrfs_end_transaction(trans);
4913 if (!(root == fs_info->tree_root ||
4914 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4915 btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
4918 btrfs_free_block_rsv(fs_info, rsv);
4921 * If we didn't successfully delete, the orphan item will still be in
4922 * the tree and we'll retry on the next mount. Again, we might also want
4923 * to retry these periodically in the future.
4925 btrfs_remove_delayed_node(BTRFS_I(inode));
4930 * Return the key found in the dir entry in the location pointer, fill @type
4931 * with BTRFS_FT_*, and return 0.
4933 * If no dir entries were found, returns -ENOENT.
4934 * If found a corrupted location in dir entry, returns -EUCLEAN.
4936 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4937 struct btrfs_key *location, u8 *type)
4939 const char *name = dentry->d_name.name;
4940 int namelen = dentry->d_name.len;
4941 struct btrfs_dir_item *di;
4942 struct btrfs_path *path;
4943 struct btrfs_root *root = BTRFS_I(dir)->root;
4946 path = btrfs_alloc_path();
4950 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
4952 if (IS_ERR_OR_NULL(di)) {
4953 ret = di ? PTR_ERR(di) : -ENOENT;
4957 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4958 if (location->type != BTRFS_INODE_ITEM_KEY &&
4959 location->type != BTRFS_ROOT_ITEM_KEY) {
4961 btrfs_warn(root->fs_info,
4962 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
4963 __func__, name, btrfs_ino(BTRFS_I(dir)),
4964 location->objectid, location->type, location->offset);
4967 *type = btrfs_dir_type(path->nodes[0], di);
4969 btrfs_free_path(path);
4974 * when we hit a tree root in a directory, the btrfs part of the inode
4975 * needs to be changed to reflect the root directory of the tree root. This
4976 * is kind of like crossing a mount point.
4978 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
4980 struct dentry *dentry,
4981 struct btrfs_key *location,
4982 struct btrfs_root **sub_root)
4984 struct btrfs_path *path;
4985 struct btrfs_root *new_root;
4986 struct btrfs_root_ref *ref;
4987 struct extent_buffer *leaf;
4988 struct btrfs_key key;
4992 path = btrfs_alloc_path();
4999 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5000 key.type = BTRFS_ROOT_REF_KEY;
5001 key.offset = location->objectid;
5003 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5010 leaf = path->nodes[0];
5011 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5012 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5013 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5016 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5017 (unsigned long)(ref + 1),
5018 dentry->d_name.len);
5022 btrfs_release_path(path);
5024 new_root = btrfs_read_fs_root_no_name(fs_info, location);
5025 if (IS_ERR(new_root)) {
5026 err = PTR_ERR(new_root);
5030 *sub_root = new_root;
5031 location->objectid = btrfs_root_dirid(&new_root->root_item);
5032 location->type = BTRFS_INODE_ITEM_KEY;
5033 location->offset = 0;
5036 btrfs_free_path(path);
5040 static void inode_tree_add(struct inode *inode)
5042 struct btrfs_root *root = BTRFS_I(inode)->root;
5043 struct btrfs_inode *entry;
5045 struct rb_node *parent;
5046 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5047 u64 ino = btrfs_ino(BTRFS_I(inode));
5049 if (inode_unhashed(inode))
5052 spin_lock(&root->inode_lock);
5053 p = &root->inode_tree.rb_node;
5056 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5058 if (ino < btrfs_ino(entry))
5059 p = &parent->rb_left;
5060 else if (ino > btrfs_ino(entry))
5061 p = &parent->rb_right;
5063 WARN_ON(!(entry->vfs_inode.i_state &
5064 (I_WILL_FREE | I_FREEING)));
5065 rb_replace_node(parent, new, &root->inode_tree);
5066 RB_CLEAR_NODE(parent);
5067 spin_unlock(&root->inode_lock);
5071 rb_link_node(new, parent, p);
5072 rb_insert_color(new, &root->inode_tree);
5073 spin_unlock(&root->inode_lock);
5076 static void inode_tree_del(struct inode *inode)
5078 struct btrfs_root *root = BTRFS_I(inode)->root;
5081 spin_lock(&root->inode_lock);
5082 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5083 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5084 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5085 empty = RB_EMPTY_ROOT(&root->inode_tree);
5087 spin_unlock(&root->inode_lock);
5089 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5090 spin_lock(&root->inode_lock);
5091 empty = RB_EMPTY_ROOT(&root->inode_tree);
5092 spin_unlock(&root->inode_lock);
5094 btrfs_add_dead_root(root);
5099 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5101 struct btrfs_iget_args *args = p;
5102 inode->i_ino = args->location->objectid;
5103 memcpy(&BTRFS_I(inode)->location, args->location,
5104 sizeof(*args->location));
5105 BTRFS_I(inode)->root = args->root;
5109 static int btrfs_find_actor(struct inode *inode, void *opaque)
5111 struct btrfs_iget_args *args = opaque;
5112 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5113 args->root == BTRFS_I(inode)->root;
5116 static struct inode *btrfs_iget_locked(struct super_block *s,
5117 struct btrfs_key *location,
5118 struct btrfs_root *root)
5120 struct inode *inode;
5121 struct btrfs_iget_args args;
5122 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5124 args.location = location;
5127 inode = iget5_locked(s, hashval, btrfs_find_actor,
5128 btrfs_init_locked_inode,
5134 * Get an inode object given its location and corresponding root.
5135 * Path can be preallocated to prevent recursing back to iget through
5136 * allocator. NULL is also valid but may require an additional allocation
5139 struct inode *btrfs_iget_path(struct super_block *s, struct btrfs_key *location,
5140 struct btrfs_root *root, struct btrfs_path *path)
5142 struct inode *inode;
5144 inode = btrfs_iget_locked(s, location, root);
5146 return ERR_PTR(-ENOMEM);
5148 if (inode->i_state & I_NEW) {
5151 ret = btrfs_read_locked_inode(inode, path);
5153 inode_tree_add(inode);
5154 unlock_new_inode(inode);
5158 * ret > 0 can come from btrfs_search_slot called by
5159 * btrfs_read_locked_inode, this means the inode item
5164 inode = ERR_PTR(ret);
5171 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5172 struct btrfs_root *root)
5174 return btrfs_iget_path(s, location, root, NULL);
5177 static struct inode *new_simple_dir(struct super_block *s,
5178 struct btrfs_key *key,
5179 struct btrfs_root *root)
5181 struct inode *inode = new_inode(s);
5184 return ERR_PTR(-ENOMEM);
5186 BTRFS_I(inode)->root = root;
5187 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5188 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5190 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5192 * We only need lookup, the rest is read-only and there's no inode
5193 * associated with the dentry
5195 inode->i_op = &simple_dir_inode_operations;
5196 inode->i_opflags &= ~IOP_XATTR;
5197 inode->i_fop = &simple_dir_operations;
5198 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5199 inode->i_mtime = current_time(inode);
5200 inode->i_atime = inode->i_mtime;
5201 inode->i_ctime = inode->i_mtime;
5202 BTRFS_I(inode)->i_otime = inode->i_mtime;
5207 static inline u8 btrfs_inode_type(struct inode *inode)
5210 * Compile-time asserts that generic FT_* types still match
5213 BUILD_BUG_ON(BTRFS_FT_UNKNOWN != FT_UNKNOWN);
5214 BUILD_BUG_ON(BTRFS_FT_REG_FILE != FT_REG_FILE);
5215 BUILD_BUG_ON(BTRFS_FT_DIR != FT_DIR);
5216 BUILD_BUG_ON(BTRFS_FT_CHRDEV != FT_CHRDEV);
5217 BUILD_BUG_ON(BTRFS_FT_BLKDEV != FT_BLKDEV);
5218 BUILD_BUG_ON(BTRFS_FT_FIFO != FT_FIFO);
5219 BUILD_BUG_ON(BTRFS_FT_SOCK != FT_SOCK);
5220 BUILD_BUG_ON(BTRFS_FT_SYMLINK != FT_SYMLINK);
5222 return fs_umode_to_ftype(inode->i_mode);
5225 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5227 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5228 struct inode *inode;
5229 struct btrfs_root *root = BTRFS_I(dir)->root;
5230 struct btrfs_root *sub_root = root;
5231 struct btrfs_key location;
5236 if (dentry->d_name.len > BTRFS_NAME_LEN)
5237 return ERR_PTR(-ENAMETOOLONG);
5239 ret = btrfs_inode_by_name(dir, dentry, &location, &di_type);
5241 return ERR_PTR(ret);
5243 if (location.type == BTRFS_INODE_ITEM_KEY) {
5244 inode = btrfs_iget(dir->i_sb, &location, root);
5248 /* Do extra check against inode mode with di_type */
5249 if (btrfs_inode_type(inode) != di_type) {
5251 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5252 inode->i_mode, btrfs_inode_type(inode),
5255 return ERR_PTR(-EUCLEAN);
5260 index = srcu_read_lock(&fs_info->subvol_srcu);
5261 ret = fixup_tree_root_location(fs_info, dir, dentry,
5262 &location, &sub_root);
5265 inode = ERR_PTR(ret);
5267 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5269 inode = btrfs_iget(dir->i_sb, &location, sub_root);
5271 srcu_read_unlock(&fs_info->subvol_srcu, index);
5273 if (!IS_ERR(inode) && root != sub_root) {
5274 down_read(&fs_info->cleanup_work_sem);
5275 if (!sb_rdonly(inode->i_sb))
5276 ret = btrfs_orphan_cleanup(sub_root);
5277 up_read(&fs_info->cleanup_work_sem);
5280 inode = ERR_PTR(ret);
5287 static int btrfs_dentry_delete(const struct dentry *dentry)
5289 struct btrfs_root *root;
5290 struct inode *inode = d_inode(dentry);
5292 if (!inode && !IS_ROOT(dentry))
5293 inode = d_inode(dentry->d_parent);
5296 root = BTRFS_I(inode)->root;
5297 if (btrfs_root_refs(&root->root_item) == 0)
5300 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5306 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5309 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5311 if (inode == ERR_PTR(-ENOENT))
5313 return d_splice_alias(inode, dentry);
5317 * All this infrastructure exists because dir_emit can fault, and we are holding
5318 * the tree lock when doing readdir. For now just allocate a buffer and copy
5319 * our information into that, and then dir_emit from the buffer. This is
5320 * similar to what NFS does, only we don't keep the buffer around in pagecache
5321 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5322 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5325 static int btrfs_opendir(struct inode *inode, struct file *file)
5327 struct btrfs_file_private *private;
5329 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5332 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5333 if (!private->filldir_buf) {
5337 file->private_data = private;
5348 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5351 struct dir_entry *entry = addr;
5352 char *name = (char *)(entry + 1);
5354 ctx->pos = get_unaligned(&entry->offset);
5355 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5356 get_unaligned(&entry->ino),
5357 get_unaligned(&entry->type)))
5359 addr += sizeof(struct dir_entry) +
5360 get_unaligned(&entry->name_len);
5366 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5368 struct inode *inode = file_inode(file);
5369 struct btrfs_root *root = BTRFS_I(inode)->root;
5370 struct btrfs_file_private *private = file->private_data;
5371 struct btrfs_dir_item *di;
5372 struct btrfs_key key;
5373 struct btrfs_key found_key;
5374 struct btrfs_path *path;
5376 struct list_head ins_list;
5377 struct list_head del_list;
5379 struct extent_buffer *leaf;
5386 struct btrfs_key location;
5388 if (!dir_emit_dots(file, ctx))
5391 path = btrfs_alloc_path();
5395 addr = private->filldir_buf;
5396 path->reada = READA_FORWARD;
5398 INIT_LIST_HEAD(&ins_list);
5399 INIT_LIST_HEAD(&del_list);
5400 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5403 key.type = BTRFS_DIR_INDEX_KEY;
5404 key.offset = ctx->pos;
5405 key.objectid = btrfs_ino(BTRFS_I(inode));
5407 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5412 struct dir_entry *entry;
5414 leaf = path->nodes[0];
5415 slot = path->slots[0];
5416 if (slot >= btrfs_header_nritems(leaf)) {
5417 ret = btrfs_next_leaf(root, path);
5425 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5427 if (found_key.objectid != key.objectid)
5429 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5431 if (found_key.offset < ctx->pos)
5433 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5435 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5436 name_len = btrfs_dir_name_len(leaf, di);
5437 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5439 btrfs_release_path(path);
5440 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5443 addr = private->filldir_buf;
5450 put_unaligned(name_len, &entry->name_len);
5451 name_ptr = (char *)(entry + 1);
5452 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5454 put_unaligned(fs_ftype_to_dtype(btrfs_dir_type(leaf, di)),
5456 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5457 put_unaligned(location.objectid, &entry->ino);
5458 put_unaligned(found_key.offset, &entry->offset);
5460 addr += sizeof(struct dir_entry) + name_len;
5461 total_len += sizeof(struct dir_entry) + name_len;
5465 btrfs_release_path(path);
5467 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5471 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5476 * Stop new entries from being returned after we return the last
5479 * New directory entries are assigned a strictly increasing
5480 * offset. This means that new entries created during readdir
5481 * are *guaranteed* to be seen in the future by that readdir.
5482 * This has broken buggy programs which operate on names as
5483 * they're returned by readdir. Until we re-use freed offsets
5484 * we have this hack to stop new entries from being returned
5485 * under the assumption that they'll never reach this huge
5488 * This is being careful not to overflow 32bit loff_t unless the
5489 * last entry requires it because doing so has broken 32bit apps
5492 if (ctx->pos >= INT_MAX)
5493 ctx->pos = LLONG_MAX;
5500 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5501 btrfs_free_path(path);
5506 * This is somewhat expensive, updating the tree every time the
5507 * inode changes. But, it is most likely to find the inode in cache.
5508 * FIXME, needs more benchmarking...there are no reasons other than performance
5509 * to keep or drop this code.
5511 static int btrfs_dirty_inode(struct inode *inode)
5513 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5514 struct btrfs_root *root = BTRFS_I(inode)->root;
5515 struct btrfs_trans_handle *trans;
5518 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5521 trans = btrfs_join_transaction(root);
5523 return PTR_ERR(trans);
5525 ret = btrfs_update_inode(trans, root, inode);
5526 if (ret && ret == -ENOSPC) {
5527 /* whoops, lets try again with the full transaction */
5528 btrfs_end_transaction(trans);
5529 trans = btrfs_start_transaction(root, 1);
5531 return PTR_ERR(trans);
5533 ret = btrfs_update_inode(trans, root, inode);
5535 btrfs_end_transaction(trans);
5536 if (BTRFS_I(inode)->delayed_node)
5537 btrfs_balance_delayed_items(fs_info);
5543 * This is a copy of file_update_time. We need this so we can return error on
5544 * ENOSPC for updating the inode in the case of file write and mmap writes.
5546 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
5549 struct btrfs_root *root = BTRFS_I(inode)->root;
5550 bool dirty = flags & ~S_VERSION;
5552 if (btrfs_root_readonly(root))
5555 if (flags & S_VERSION)
5556 dirty |= inode_maybe_inc_iversion(inode, dirty);
5557 if (flags & S_CTIME)
5558 inode->i_ctime = *now;
5559 if (flags & S_MTIME)
5560 inode->i_mtime = *now;
5561 if (flags & S_ATIME)
5562 inode->i_atime = *now;
5563 return dirty ? btrfs_dirty_inode(inode) : 0;
5567 * find the highest existing sequence number in a directory
5568 * and then set the in-memory index_cnt variable to reflect
5569 * free sequence numbers
5571 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
5573 struct btrfs_root *root = inode->root;
5574 struct btrfs_key key, found_key;
5575 struct btrfs_path *path;
5576 struct extent_buffer *leaf;
5579 key.objectid = btrfs_ino(inode);
5580 key.type = BTRFS_DIR_INDEX_KEY;
5581 key.offset = (u64)-1;
5583 path = btrfs_alloc_path();
5587 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5590 /* FIXME: we should be able to handle this */
5596 * MAGIC NUMBER EXPLANATION:
5597 * since we search a directory based on f_pos we have to start at 2
5598 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5599 * else has to start at 2
5601 if (path->slots[0] == 0) {
5602 inode->index_cnt = 2;
5608 leaf = path->nodes[0];
5609 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5611 if (found_key.objectid != btrfs_ino(inode) ||
5612 found_key.type != BTRFS_DIR_INDEX_KEY) {
5613 inode->index_cnt = 2;
5617 inode->index_cnt = found_key.offset + 1;
5619 btrfs_free_path(path);
5624 * helper to find a free sequence number in a given directory. This current
5625 * code is very simple, later versions will do smarter things in the btree
5627 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
5631 if (dir->index_cnt == (u64)-1) {
5632 ret = btrfs_inode_delayed_dir_index_count(dir);
5634 ret = btrfs_set_inode_index_count(dir);
5640 *index = dir->index_cnt;
5646 static int btrfs_insert_inode_locked(struct inode *inode)
5648 struct btrfs_iget_args args;
5649 args.location = &BTRFS_I(inode)->location;
5650 args.root = BTRFS_I(inode)->root;
5652 return insert_inode_locked4(inode,
5653 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
5654 btrfs_find_actor, &args);
5658 * Inherit flags from the parent inode.
5660 * Currently only the compression flags and the cow flags are inherited.
5662 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
5669 flags = BTRFS_I(dir)->flags;
5671 if (flags & BTRFS_INODE_NOCOMPRESS) {
5672 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
5673 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
5674 } else if (flags & BTRFS_INODE_COMPRESS) {
5675 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
5676 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
5679 if (flags & BTRFS_INODE_NODATACOW) {
5680 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
5681 if (S_ISREG(inode->i_mode))
5682 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5685 btrfs_sync_inode_flags_to_i_flags(inode);
5688 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5689 struct btrfs_root *root,
5691 const char *name, int name_len,
5692 u64 ref_objectid, u64 objectid,
5693 umode_t mode, u64 *index)
5695 struct btrfs_fs_info *fs_info = root->fs_info;
5696 struct inode *inode;
5697 struct btrfs_inode_item *inode_item;
5698 struct btrfs_key *location;
5699 struct btrfs_path *path;
5700 struct btrfs_inode_ref *ref;
5701 struct btrfs_key key[2];
5703 int nitems = name ? 2 : 1;
5705 unsigned int nofs_flag;
5708 path = btrfs_alloc_path();
5710 return ERR_PTR(-ENOMEM);
5712 nofs_flag = memalloc_nofs_save();
5713 inode = new_inode(fs_info->sb);
5714 memalloc_nofs_restore(nofs_flag);
5716 btrfs_free_path(path);
5717 return ERR_PTR(-ENOMEM);
5721 * O_TMPFILE, set link count to 0, so that after this point,
5722 * we fill in an inode item with the correct link count.
5725 set_nlink(inode, 0);
5728 * we have to initialize this early, so we can reclaim the inode
5729 * number if we fail afterwards in this function.
5731 inode->i_ino = objectid;
5734 trace_btrfs_inode_request(dir);
5736 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
5738 btrfs_free_path(path);
5740 return ERR_PTR(ret);
5746 * index_cnt is ignored for everything but a dir,
5747 * btrfs_set_inode_index_count has an explanation for the magic
5750 BTRFS_I(inode)->index_cnt = 2;
5751 BTRFS_I(inode)->dir_index = *index;
5752 BTRFS_I(inode)->root = root;
5753 BTRFS_I(inode)->generation = trans->transid;
5754 inode->i_generation = BTRFS_I(inode)->generation;
5757 * We could have gotten an inode number from somebody who was fsynced
5758 * and then removed in this same transaction, so let's just set full
5759 * sync since it will be a full sync anyway and this will blow away the
5760 * old info in the log.
5762 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5764 key[0].objectid = objectid;
5765 key[0].type = BTRFS_INODE_ITEM_KEY;
5768 sizes[0] = sizeof(struct btrfs_inode_item);
5772 * Start new inodes with an inode_ref. This is slightly more
5773 * efficient for small numbers of hard links since they will
5774 * be packed into one item. Extended refs will kick in if we
5775 * add more hard links than can fit in the ref item.
5777 key[1].objectid = objectid;
5778 key[1].type = BTRFS_INODE_REF_KEY;
5779 key[1].offset = ref_objectid;
5781 sizes[1] = name_len + sizeof(*ref);
5784 location = &BTRFS_I(inode)->location;
5785 location->objectid = objectid;
5786 location->offset = 0;
5787 location->type = BTRFS_INODE_ITEM_KEY;
5789 ret = btrfs_insert_inode_locked(inode);
5795 path->leave_spinning = 1;
5796 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
5800 inode_init_owner(inode, dir, mode);
5801 inode_set_bytes(inode, 0);
5803 inode->i_mtime = current_time(inode);
5804 inode->i_atime = inode->i_mtime;
5805 inode->i_ctime = inode->i_mtime;
5806 BTRFS_I(inode)->i_otime = inode->i_mtime;
5808 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5809 struct btrfs_inode_item);
5810 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
5811 sizeof(*inode_item));
5812 fill_inode_item(trans, path->nodes[0], inode_item, inode);
5815 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5816 struct btrfs_inode_ref);
5817 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5818 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5819 ptr = (unsigned long)(ref + 1);
5820 write_extent_buffer(path->nodes[0], name, ptr, name_len);
5823 btrfs_mark_buffer_dirty(path->nodes[0]);
5824 btrfs_free_path(path);
5826 btrfs_inherit_iflags(inode, dir);
5828 if (S_ISREG(mode)) {
5829 if (btrfs_test_opt(fs_info, NODATASUM))
5830 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5831 if (btrfs_test_opt(fs_info, NODATACOW))
5832 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5833 BTRFS_INODE_NODATASUM;
5836 inode_tree_add(inode);
5838 trace_btrfs_inode_new(inode);
5839 btrfs_set_inode_last_trans(trans, inode);
5841 btrfs_update_root_times(trans, root);
5843 ret = btrfs_inode_inherit_props(trans, inode, dir);
5846 "error inheriting props for ino %llu (root %llu): %d",
5847 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
5852 discard_new_inode(inode);
5855 BTRFS_I(dir)->index_cnt--;
5856 btrfs_free_path(path);
5857 return ERR_PTR(ret);
5861 * utility function to add 'inode' into 'parent_inode' with
5862 * a give name and a given sequence number.
5863 * if 'add_backref' is true, also insert a backref from the
5864 * inode to the parent directory.
5866 int btrfs_add_link(struct btrfs_trans_handle *trans,
5867 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
5868 const char *name, int name_len, int add_backref, u64 index)
5871 struct btrfs_key key;
5872 struct btrfs_root *root = parent_inode->root;
5873 u64 ino = btrfs_ino(inode);
5874 u64 parent_ino = btrfs_ino(parent_inode);
5876 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5877 memcpy(&key, &inode->root->root_key, sizeof(key));
5880 key.type = BTRFS_INODE_ITEM_KEY;
5884 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5885 ret = btrfs_add_root_ref(trans, key.objectid,
5886 root->root_key.objectid, parent_ino,
5887 index, name, name_len);
5888 } else if (add_backref) {
5889 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5893 /* Nothing to clean up yet */
5897 ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key,
5898 btrfs_inode_type(&inode->vfs_inode), index);
5899 if (ret == -EEXIST || ret == -EOVERFLOW)
5902 btrfs_abort_transaction(trans, ret);
5906 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
5908 inode_inc_iversion(&parent_inode->vfs_inode);
5910 * If we are replaying a log tree, we do not want to update the mtime
5911 * and ctime of the parent directory with the current time, since the
5912 * log replay procedure is responsible for setting them to their correct
5913 * values (the ones it had when the fsync was done).
5915 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
5916 struct timespec64 now = current_time(&parent_inode->vfs_inode);
5918 parent_inode->vfs_inode.i_mtime = now;
5919 parent_inode->vfs_inode.i_ctime = now;
5921 ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode);
5923 btrfs_abort_transaction(trans, ret);
5927 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5930 err = btrfs_del_root_ref(trans, key.objectid,
5931 root->root_key.objectid, parent_ino,
5932 &local_index, name, name_len);
5934 btrfs_abort_transaction(trans, err);
5935 } else if (add_backref) {
5939 err = btrfs_del_inode_ref(trans, root, name, name_len,
5940 ino, parent_ino, &local_index);
5942 btrfs_abort_transaction(trans, err);
5945 /* Return the original error code */
5949 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5950 struct btrfs_inode *dir, struct dentry *dentry,
5951 struct btrfs_inode *inode, int backref, u64 index)
5953 int err = btrfs_add_link(trans, dir, inode,
5954 dentry->d_name.name, dentry->d_name.len,
5961 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5962 umode_t mode, dev_t rdev)
5964 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5965 struct btrfs_trans_handle *trans;
5966 struct btrfs_root *root = BTRFS_I(dir)->root;
5967 struct inode *inode = NULL;
5973 * 2 for inode item and ref
5975 * 1 for xattr if selinux is on
5977 trans = btrfs_start_transaction(root, 5);
5979 return PTR_ERR(trans);
5981 err = btrfs_find_free_ino(root, &objectid);
5985 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5986 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
5988 if (IS_ERR(inode)) {
5989 err = PTR_ERR(inode);
5995 * If the active LSM wants to access the inode during
5996 * d_instantiate it needs these. Smack checks to see
5997 * if the filesystem supports xattrs by looking at the
6000 inode->i_op = &btrfs_special_inode_operations;
6001 init_special_inode(inode, inode->i_mode, rdev);
6003 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6007 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6012 btrfs_update_inode(trans, root, inode);
6013 d_instantiate_new(dentry, inode);
6016 btrfs_end_transaction(trans);
6017 btrfs_btree_balance_dirty(fs_info);
6019 inode_dec_link_count(inode);
6020 discard_new_inode(inode);
6025 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6026 umode_t mode, bool excl)
6028 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6029 struct btrfs_trans_handle *trans;
6030 struct btrfs_root *root = BTRFS_I(dir)->root;
6031 struct inode *inode = NULL;
6037 * 2 for inode item and ref
6039 * 1 for xattr if selinux is on
6041 trans = btrfs_start_transaction(root, 5);
6043 return PTR_ERR(trans);
6045 err = btrfs_find_free_ino(root, &objectid);
6049 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6050 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6052 if (IS_ERR(inode)) {
6053 err = PTR_ERR(inode);
6058 * If the active LSM wants to access the inode during
6059 * d_instantiate it needs these. Smack checks to see
6060 * if the filesystem supports xattrs by looking at the
6063 inode->i_fop = &btrfs_file_operations;
6064 inode->i_op = &btrfs_file_inode_operations;
6065 inode->i_mapping->a_ops = &btrfs_aops;
6067 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6071 err = btrfs_update_inode(trans, root, inode);
6075 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6080 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6081 d_instantiate_new(dentry, inode);
6084 btrfs_end_transaction(trans);
6086 inode_dec_link_count(inode);
6087 discard_new_inode(inode);
6089 btrfs_btree_balance_dirty(fs_info);
6093 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6094 struct dentry *dentry)
6096 struct btrfs_trans_handle *trans = NULL;
6097 struct btrfs_root *root = BTRFS_I(dir)->root;
6098 struct inode *inode = d_inode(old_dentry);
6099 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6104 /* do not allow sys_link's with other subvols of the same device */
6105 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6108 if (inode->i_nlink >= BTRFS_LINK_MAX)
6111 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6116 * 2 items for inode and inode ref
6117 * 2 items for dir items
6118 * 1 item for parent inode
6119 * 1 item for orphan item deletion if O_TMPFILE
6121 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6122 if (IS_ERR(trans)) {
6123 err = PTR_ERR(trans);
6128 /* There are several dir indexes for this inode, clear the cache. */
6129 BTRFS_I(inode)->dir_index = 0ULL;
6131 inode_inc_iversion(inode);
6132 inode->i_ctime = current_time(inode);
6134 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6136 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6142 struct dentry *parent = dentry->d_parent;
6145 err = btrfs_update_inode(trans, root, inode);
6148 if (inode->i_nlink == 1) {
6150 * If new hard link count is 1, it's a file created
6151 * with open(2) O_TMPFILE flag.
6153 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6157 d_instantiate(dentry, inode);
6158 ret = btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent,
6160 if (ret == BTRFS_NEED_TRANS_COMMIT) {
6161 err = btrfs_commit_transaction(trans);
6168 btrfs_end_transaction(trans);
6170 inode_dec_link_count(inode);
6173 btrfs_btree_balance_dirty(fs_info);
6177 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6179 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6180 struct inode *inode = NULL;
6181 struct btrfs_trans_handle *trans;
6182 struct btrfs_root *root = BTRFS_I(dir)->root;
6188 * 2 items for inode and ref
6189 * 2 items for dir items
6190 * 1 for xattr if selinux is on
6192 trans = btrfs_start_transaction(root, 5);
6194 return PTR_ERR(trans);
6196 err = btrfs_find_free_ino(root, &objectid);
6200 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6201 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6202 S_IFDIR | mode, &index);
6203 if (IS_ERR(inode)) {
6204 err = PTR_ERR(inode);
6209 /* these must be set before we unlock the inode */
6210 inode->i_op = &btrfs_dir_inode_operations;
6211 inode->i_fop = &btrfs_dir_file_operations;
6213 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6217 btrfs_i_size_write(BTRFS_I(inode), 0);
6218 err = btrfs_update_inode(trans, root, inode);
6222 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6223 dentry->d_name.name,
6224 dentry->d_name.len, 0, index);
6228 d_instantiate_new(dentry, inode);
6231 btrfs_end_transaction(trans);
6233 inode_dec_link_count(inode);
6234 discard_new_inode(inode);
6236 btrfs_btree_balance_dirty(fs_info);
6240 static noinline int uncompress_inline(struct btrfs_path *path,
6242 size_t pg_offset, u64 extent_offset,
6243 struct btrfs_file_extent_item *item)
6246 struct extent_buffer *leaf = path->nodes[0];
6249 unsigned long inline_size;
6253 WARN_ON(pg_offset != 0);
6254 compress_type = btrfs_file_extent_compression(leaf, item);
6255 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6256 inline_size = btrfs_file_extent_inline_item_len(leaf,
6257 btrfs_item_nr(path->slots[0]));
6258 tmp = kmalloc(inline_size, GFP_NOFS);
6261 ptr = btrfs_file_extent_inline_start(item);
6263 read_extent_buffer(leaf, tmp, ptr, inline_size);
6265 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6266 ret = btrfs_decompress(compress_type, tmp, page,
6267 extent_offset, inline_size, max_size);
6270 * decompression code contains a memset to fill in any space between the end
6271 * of the uncompressed data and the end of max_size in case the decompressed
6272 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6273 * the end of an inline extent and the beginning of the next block, so we
6274 * cover that region here.
6277 if (max_size + pg_offset < PAGE_SIZE) {
6278 char *map = kmap(page);
6279 memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6287 * btrfs_get_extent - Lookup the first extent overlapping a range in a file.
6288 * @inode: file to search in
6289 * @page: page to read extent data into if the extent is inline
6290 * @pg_offset: offset into @page to copy to
6291 * @start: file offset
6292 * @len: length of range starting at @start
6294 * This returns the first &struct extent_map which overlaps with the given
6295 * range, reading it from the B-tree and caching it if necessary. Note that
6296 * there may be more extents which overlap the given range after the returned
6299 * If @page is not NULL and the extent is inline, this also reads the extent
6300 * data directly into the page and marks the extent up to date in the io_tree.
6302 * Return: ERR_PTR on error, non-NULL extent_map on success.
6304 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6305 struct page *page, size_t pg_offset,
6308 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6311 u64 extent_start = 0;
6313 u64 objectid = btrfs_ino(inode);
6314 int extent_type = -1;
6315 struct btrfs_path *path = NULL;
6316 struct btrfs_root *root = inode->root;
6317 struct btrfs_file_extent_item *item;
6318 struct extent_buffer *leaf;
6319 struct btrfs_key found_key;
6320 struct extent_map *em = NULL;
6321 struct extent_map_tree *em_tree = &inode->extent_tree;
6322 struct extent_io_tree *io_tree = &inode->io_tree;
6324 read_lock(&em_tree->lock);
6325 em = lookup_extent_mapping(em_tree, start, len);
6326 read_unlock(&em_tree->lock);
6329 if (em->start > start || em->start + em->len <= start)
6330 free_extent_map(em);
6331 else if (em->block_start == EXTENT_MAP_INLINE && page)
6332 free_extent_map(em);
6336 em = alloc_extent_map();
6341 em->start = EXTENT_MAP_HOLE;
6342 em->orig_start = EXTENT_MAP_HOLE;
6344 em->block_len = (u64)-1;
6346 path = btrfs_alloc_path();
6352 /* Chances are we'll be called again, so go ahead and do readahead */
6353 path->reada = READA_FORWARD;
6356 * Unless we're going to uncompress the inline extent, no sleep would
6359 path->leave_spinning = 1;
6361 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6365 } else if (ret > 0) {
6366 if (path->slots[0] == 0)
6371 leaf = path->nodes[0];
6372 item = btrfs_item_ptr(leaf, path->slots[0],
6373 struct btrfs_file_extent_item);
6374 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6375 if (found_key.objectid != objectid ||
6376 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6378 * If we backup past the first extent we want to move forward
6379 * and see if there is an extent in front of us, otherwise we'll
6380 * say there is a hole for our whole search range which can
6387 extent_type = btrfs_file_extent_type(leaf, item);
6388 extent_start = found_key.offset;
6389 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6390 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6391 /* Only regular file could have regular/prealloc extent */
6392 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6395 "regular/prealloc extent found for non-regular inode %llu",
6399 extent_end = extent_start +
6400 btrfs_file_extent_num_bytes(leaf, item);
6402 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6404 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6407 size = btrfs_file_extent_ram_bytes(leaf, item);
6408 extent_end = ALIGN(extent_start + size,
6409 fs_info->sectorsize);
6411 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6416 if (start >= extent_end) {
6418 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6419 ret = btrfs_next_leaf(root, path);
6423 } else if (ret > 0) {
6426 leaf = path->nodes[0];
6428 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6429 if (found_key.objectid != objectid ||
6430 found_key.type != BTRFS_EXTENT_DATA_KEY)
6432 if (start + len <= found_key.offset)
6434 if (start > found_key.offset)
6437 /* New extent overlaps with existing one */
6439 em->orig_start = start;
6440 em->len = found_key.offset - start;
6441 em->block_start = EXTENT_MAP_HOLE;
6445 btrfs_extent_item_to_extent_map(inode, path, item, !page, em);
6447 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6448 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6450 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6454 size_t extent_offset;
6460 size = btrfs_file_extent_ram_bytes(leaf, item);
6461 extent_offset = page_offset(page) + pg_offset - extent_start;
6462 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6463 size - extent_offset);
6464 em->start = extent_start + extent_offset;
6465 em->len = ALIGN(copy_size, fs_info->sectorsize);
6466 em->orig_block_len = em->len;
6467 em->orig_start = em->start;
6468 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6470 btrfs_set_path_blocking(path);
6471 if (!PageUptodate(page)) {
6472 if (btrfs_file_extent_compression(leaf, item) !=
6473 BTRFS_COMPRESS_NONE) {
6474 ret = uncompress_inline(path, page, pg_offset,
6475 extent_offset, item);
6482 read_extent_buffer(leaf, map + pg_offset, ptr,
6484 if (pg_offset + copy_size < PAGE_SIZE) {
6485 memset(map + pg_offset + copy_size, 0,
6486 PAGE_SIZE - pg_offset -
6491 flush_dcache_page(page);
6493 set_extent_uptodate(io_tree, em->start,
6494 extent_map_end(em) - 1, NULL, GFP_NOFS);
6499 em->orig_start = start;
6501 em->block_start = EXTENT_MAP_HOLE;
6503 btrfs_release_path(path);
6504 if (em->start > start || extent_map_end(em) <= start) {
6506 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6507 em->start, em->len, start, len);
6513 write_lock(&em_tree->lock);
6514 err = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6515 write_unlock(&em_tree->lock);
6517 btrfs_free_path(path);
6519 trace_btrfs_get_extent(root, inode, em);
6522 free_extent_map(em);
6523 return ERR_PTR(err);
6525 BUG_ON(!em); /* Error is always set */
6529 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
6532 struct extent_map *em;
6533 struct extent_map *hole_em = NULL;
6534 u64 delalloc_start = start;
6540 em = btrfs_get_extent(inode, NULL, 0, start, len);
6544 * If our em maps to:
6546 * - a pre-alloc extent,
6547 * there might actually be delalloc bytes behind it.
6549 if (em->block_start != EXTENT_MAP_HOLE &&
6550 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6555 /* check to see if we've wrapped (len == -1 or similar) */
6564 /* ok, we didn't find anything, lets look for delalloc */
6565 delalloc_len = count_range_bits(&inode->io_tree, &delalloc_start,
6566 end, len, EXTENT_DELALLOC, 1);
6567 delalloc_end = delalloc_start + delalloc_len;
6568 if (delalloc_end < delalloc_start)
6569 delalloc_end = (u64)-1;
6572 * We didn't find anything useful, return the original results from
6575 if (delalloc_start > end || delalloc_end <= start) {
6582 * Adjust the delalloc_start to make sure it doesn't go backwards from
6583 * the start they passed in
6585 delalloc_start = max(start, delalloc_start);
6586 delalloc_len = delalloc_end - delalloc_start;
6588 if (delalloc_len > 0) {
6591 const u64 hole_end = extent_map_end(hole_em);
6593 em = alloc_extent_map();
6601 * When btrfs_get_extent can't find anything it returns one
6604 * Make sure what it found really fits our range, and adjust to
6605 * make sure it is based on the start from the caller
6607 if (hole_end <= start || hole_em->start > end) {
6608 free_extent_map(hole_em);
6611 hole_start = max(hole_em->start, start);
6612 hole_len = hole_end - hole_start;
6615 if (hole_em && delalloc_start > hole_start) {
6617 * Our hole starts before our delalloc, so we have to
6618 * return just the parts of the hole that go until the
6621 em->len = min(hole_len, delalloc_start - hole_start);
6622 em->start = hole_start;
6623 em->orig_start = hole_start;
6625 * Don't adjust block start at all, it is fixed at
6628 em->block_start = hole_em->block_start;
6629 em->block_len = hole_len;
6630 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6631 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6634 * Hole is out of passed range or it starts after
6637 em->start = delalloc_start;
6638 em->len = delalloc_len;
6639 em->orig_start = delalloc_start;
6640 em->block_start = EXTENT_MAP_DELALLOC;
6641 em->block_len = delalloc_len;
6648 free_extent_map(hole_em);
6650 free_extent_map(em);
6651 return ERR_PTR(err);
6656 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
6659 const u64 orig_start,
6660 const u64 block_start,
6661 const u64 block_len,
6662 const u64 orig_block_len,
6663 const u64 ram_bytes,
6666 struct extent_map *em = NULL;
6669 if (type != BTRFS_ORDERED_NOCOW) {
6670 em = create_io_em(inode, start, len, orig_start,
6671 block_start, block_len, orig_block_len,
6673 BTRFS_COMPRESS_NONE, /* compress_type */
6678 ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
6679 len, block_len, type);
6682 free_extent_map(em);
6683 btrfs_drop_extent_cache(BTRFS_I(inode), start,
6684 start + len - 1, 0);
6693 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6696 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6697 struct btrfs_root *root = BTRFS_I(inode)->root;
6698 struct extent_map *em;
6699 struct btrfs_key ins;
6703 alloc_hint = get_extent_allocation_hint(inode, start, len);
6704 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
6705 0, alloc_hint, &ins, 1, 1);
6707 return ERR_PTR(ret);
6709 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
6710 ins.objectid, ins.offset, ins.offset,
6711 ins.offset, BTRFS_ORDERED_REGULAR);
6712 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
6714 btrfs_free_reserved_extent(fs_info, ins.objectid,
6721 * returns 1 when the nocow is safe, < 1 on error, 0 if the
6722 * block must be cow'd
6724 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
6725 u64 *orig_start, u64 *orig_block_len,
6728 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6729 struct btrfs_path *path;
6731 struct extent_buffer *leaf;
6732 struct btrfs_root *root = BTRFS_I(inode)->root;
6733 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6734 struct btrfs_file_extent_item *fi;
6735 struct btrfs_key key;
6742 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
6744 path = btrfs_alloc_path();
6748 ret = btrfs_lookup_file_extent(NULL, root, path,
6749 btrfs_ino(BTRFS_I(inode)), offset, 0);
6753 slot = path->slots[0];
6756 /* can't find the item, must cow */
6763 leaf = path->nodes[0];
6764 btrfs_item_key_to_cpu(leaf, &key, slot);
6765 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
6766 key.type != BTRFS_EXTENT_DATA_KEY) {
6767 /* not our file or wrong item type, must cow */
6771 if (key.offset > offset) {
6772 /* Wrong offset, must cow */
6776 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6777 found_type = btrfs_file_extent_type(leaf, fi);
6778 if (found_type != BTRFS_FILE_EXTENT_REG &&
6779 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6780 /* not a regular extent, must cow */
6784 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
6787 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6788 if (extent_end <= offset)
6791 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6792 if (disk_bytenr == 0)
6795 if (btrfs_file_extent_compression(leaf, fi) ||
6796 btrfs_file_extent_encryption(leaf, fi) ||
6797 btrfs_file_extent_other_encoding(leaf, fi))
6801 * Do the same check as in btrfs_cross_ref_exist but without the
6802 * unnecessary search.
6804 if (btrfs_file_extent_generation(leaf, fi) <=
6805 btrfs_root_last_snapshot(&root->root_item))
6808 backref_offset = btrfs_file_extent_offset(leaf, fi);
6811 *orig_start = key.offset - backref_offset;
6812 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6813 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6816 if (btrfs_extent_readonly(fs_info, disk_bytenr))
6819 num_bytes = min(offset + *len, extent_end) - offset;
6820 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6823 range_end = round_up(offset + num_bytes,
6824 root->fs_info->sectorsize) - 1;
6825 ret = test_range_bit(io_tree, offset, range_end,
6826 EXTENT_DELALLOC, 0, NULL);
6833 btrfs_release_path(path);
6836 * look for other files referencing this extent, if we
6837 * find any we must cow
6840 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
6841 key.offset - backref_offset, disk_bytenr);
6848 * adjust disk_bytenr and num_bytes to cover just the bytes
6849 * in this extent we are about to write. If there
6850 * are any csums in that range we have to cow in order
6851 * to keep the csums correct
6853 disk_bytenr += backref_offset;
6854 disk_bytenr += offset - key.offset;
6855 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
6858 * all of the above have passed, it is safe to overwrite this extent
6864 btrfs_free_path(path);
6868 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6869 struct extent_state **cached_state, int writing)
6871 struct btrfs_ordered_extent *ordered;
6875 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6878 * We're concerned with the entire range that we're going to be
6879 * doing DIO to, so we need to make sure there's no ordered
6880 * extents in this range.
6882 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
6883 lockend - lockstart + 1);
6886 * We need to make sure there are no buffered pages in this
6887 * range either, we could have raced between the invalidate in
6888 * generic_file_direct_write and locking the extent. The
6889 * invalidate needs to happen so that reads after a write do not
6893 (!writing || !filemap_range_has_page(inode->i_mapping,
6894 lockstart, lockend)))
6897 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6902 * If we are doing a DIO read and the ordered extent we
6903 * found is for a buffered write, we can not wait for it
6904 * to complete and retry, because if we do so we can
6905 * deadlock with concurrent buffered writes on page
6906 * locks. This happens only if our DIO read covers more
6907 * than one extent map, if at this point has already
6908 * created an ordered extent for a previous extent map
6909 * and locked its range in the inode's io tree, and a
6910 * concurrent write against that previous extent map's
6911 * range and this range started (we unlock the ranges
6912 * in the io tree only when the bios complete and
6913 * buffered writes always lock pages before attempting
6914 * to lock range in the io tree).
6917 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
6918 btrfs_start_ordered_extent(inode, ordered, 1);
6921 btrfs_put_ordered_extent(ordered);
6924 * We could trigger writeback for this range (and wait
6925 * for it to complete) and then invalidate the pages for
6926 * this range (through invalidate_inode_pages2_range()),
6927 * but that can lead us to a deadlock with a concurrent
6928 * call to readpages() (a buffered read or a defrag call
6929 * triggered a readahead) on a page lock due to an
6930 * ordered dio extent we created before but did not have
6931 * yet a corresponding bio submitted (whence it can not
6932 * complete), which makes readpages() wait for that
6933 * ordered extent to complete while holding a lock on
6948 /* The callers of this must take lock_extent() */
6949 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
6950 u64 orig_start, u64 block_start,
6951 u64 block_len, u64 orig_block_len,
6952 u64 ram_bytes, int compress_type,
6955 struct extent_map_tree *em_tree;
6956 struct extent_map *em;
6959 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
6960 type == BTRFS_ORDERED_COMPRESSED ||
6961 type == BTRFS_ORDERED_NOCOW ||
6962 type == BTRFS_ORDERED_REGULAR);
6964 em_tree = &BTRFS_I(inode)->extent_tree;
6965 em = alloc_extent_map();
6967 return ERR_PTR(-ENOMEM);
6970 em->orig_start = orig_start;
6972 em->block_len = block_len;
6973 em->block_start = block_start;
6974 em->orig_block_len = orig_block_len;
6975 em->ram_bytes = ram_bytes;
6976 em->generation = -1;
6977 set_bit(EXTENT_FLAG_PINNED, &em->flags);
6978 if (type == BTRFS_ORDERED_PREALLOC) {
6979 set_bit(EXTENT_FLAG_FILLING, &em->flags);
6980 } else if (type == BTRFS_ORDERED_COMPRESSED) {
6981 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6982 em->compress_type = compress_type;
6986 btrfs_drop_extent_cache(BTRFS_I(inode), em->start,
6987 em->start + em->len - 1, 0);
6988 write_lock(&em_tree->lock);
6989 ret = add_extent_mapping(em_tree, em, 1);
6990 write_unlock(&em_tree->lock);
6992 * The caller has taken lock_extent(), who could race with us
6995 } while (ret == -EEXIST);
6998 free_extent_map(em);
6999 return ERR_PTR(ret);
7002 /* em got 2 refs now, callers needs to do free_extent_map once. */
7007 static int btrfs_get_blocks_direct_read(struct extent_map *em,
7008 struct buffer_head *bh_result,
7009 struct inode *inode,
7012 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7014 if (em->block_start == EXTENT_MAP_HOLE ||
7015 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7018 len = min(len, em->len - (start - em->start));
7020 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7022 bh_result->b_size = len;
7023 bh_result->b_bdev = fs_info->fs_devices->latest_bdev;
7024 set_buffer_mapped(bh_result);
7029 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7030 struct buffer_head *bh_result,
7031 struct inode *inode,
7032 struct btrfs_dio_data *dio_data,
7035 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7036 struct extent_map *em = *map;
7040 * We don't allocate a new extent in the following cases
7042 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7044 * 2) The extent is marked as PREALLOC. We're good to go here and can
7045 * just use the extent.
7048 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7049 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7050 em->block_start != EXTENT_MAP_HOLE)) {
7052 u64 block_start, orig_start, orig_block_len, ram_bytes;
7054 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7055 type = BTRFS_ORDERED_PREALLOC;
7057 type = BTRFS_ORDERED_NOCOW;
7058 len = min(len, em->len - (start - em->start));
7059 block_start = em->block_start + (start - em->start);
7061 if (can_nocow_extent(inode, start, &len, &orig_start,
7062 &orig_block_len, &ram_bytes) == 1 &&
7063 btrfs_inc_nocow_writers(fs_info, block_start)) {
7064 struct extent_map *em2;
7066 em2 = btrfs_create_dio_extent(inode, start, len,
7067 orig_start, block_start,
7068 len, orig_block_len,
7070 btrfs_dec_nocow_writers(fs_info, block_start);
7071 if (type == BTRFS_ORDERED_PREALLOC) {
7072 free_extent_map(em);
7076 if (em2 && IS_ERR(em2)) {
7081 * For inode marked NODATACOW or extent marked PREALLOC,
7082 * use the existing or preallocated extent, so does not
7083 * need to adjust btrfs_space_info's bytes_may_use.
7085 btrfs_free_reserved_data_space_noquota(inode, start,
7091 /* this will cow the extent */
7092 len = bh_result->b_size;
7093 free_extent_map(em);
7094 *map = em = btrfs_new_extent_direct(inode, start, len);
7100 len = min(len, em->len - (start - em->start));
7103 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7105 bh_result->b_size = len;
7106 bh_result->b_bdev = fs_info->fs_devices->latest_bdev;
7107 set_buffer_mapped(bh_result);
7109 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7110 set_buffer_new(bh_result);
7113 * Need to update the i_size under the extent lock so buffered
7114 * readers will get the updated i_size when we unlock.
7116 if (!dio_data->overwrite && start + len > i_size_read(inode))
7117 i_size_write(inode, start + len);
7119 WARN_ON(dio_data->reserve < len);
7120 dio_data->reserve -= len;
7121 dio_data->unsubmitted_oe_range_end = start + len;
7122 current->journal_info = dio_data;
7127 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7128 struct buffer_head *bh_result, int create)
7130 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7131 struct extent_map *em;
7132 struct extent_state *cached_state = NULL;
7133 struct btrfs_dio_data *dio_data = NULL;
7134 u64 start = iblock << inode->i_blkbits;
7135 u64 lockstart, lockend;
7136 u64 len = bh_result->b_size;
7140 len = min_t(u64, len, fs_info->sectorsize);
7143 lockend = start + len - 1;
7145 if (current->journal_info) {
7147 * Need to pull our outstanding extents and set journal_info to NULL so
7148 * that anything that needs to check if there's a transaction doesn't get
7151 dio_data = current->journal_info;
7152 current->journal_info = NULL;
7156 * If this errors out it's because we couldn't invalidate pagecache for
7157 * this range and we need to fallback to buffered.
7159 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7165 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7172 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7173 * io. INLINE is special, and we could probably kludge it in here, but
7174 * it's still buffered so for safety lets just fall back to the generic
7177 * For COMPRESSED we _have_ to read the entire extent in so we can
7178 * decompress it, so there will be buffering required no matter what we
7179 * do, so go ahead and fallback to buffered.
7181 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7182 * to buffered IO. Don't blame me, this is the price we pay for using
7185 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7186 em->block_start == EXTENT_MAP_INLINE) {
7187 free_extent_map(em);
7193 ret = btrfs_get_blocks_direct_write(&em, bh_result, inode,
7194 dio_data, start, len);
7198 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
7199 lockend, &cached_state);
7201 ret = btrfs_get_blocks_direct_read(em, bh_result, inode,
7203 /* Can be negative only if we read from a hole */
7206 free_extent_map(em);
7210 * We need to unlock only the end area that we aren't using.
7211 * The rest is going to be unlocked by the endio routine.
7213 lockstart = start + bh_result->b_size;
7214 if (lockstart < lockend) {
7215 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
7216 lockstart, lockend, &cached_state);
7218 free_extent_state(cached_state);
7222 free_extent_map(em);
7227 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7231 current->journal_info = dio_data;
7235 static inline blk_status_t submit_dio_repair_bio(struct inode *inode,
7239 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7242 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7244 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR);
7248 ret = btrfs_map_bio(fs_info, bio, mirror_num);
7253 static int btrfs_check_dio_repairable(struct inode *inode,
7254 struct bio *failed_bio,
7255 struct io_failure_record *failrec,
7258 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7261 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
7262 if (num_copies == 1) {
7264 * we only have a single copy of the data, so don't bother with
7265 * all the retry and error correction code that follows. no
7266 * matter what the error is, it is very likely to persist.
7268 btrfs_debug(fs_info,
7269 "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
7270 num_copies, failrec->this_mirror, failed_mirror);
7274 failrec->failed_mirror = failed_mirror;
7275 failrec->this_mirror++;
7276 if (failrec->this_mirror == failed_mirror)
7277 failrec->this_mirror++;
7279 if (failrec->this_mirror > num_copies) {
7280 btrfs_debug(fs_info,
7281 "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
7282 num_copies, failrec->this_mirror, failed_mirror);
7289 static blk_status_t dio_read_error(struct inode *inode, struct bio *failed_bio,
7290 struct page *page, unsigned int pgoff,
7291 u64 start, u64 end, int failed_mirror,
7292 bio_end_io_t *repair_endio, void *repair_arg)
7294 struct io_failure_record *failrec;
7295 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7296 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7299 unsigned int read_mode = 0;
7302 blk_status_t status;
7303 struct bio_vec bvec;
7305 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
7307 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7309 return errno_to_blk_status(ret);
7311 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7314 free_io_failure(failure_tree, io_tree, failrec);
7315 return BLK_STS_IOERR;
7318 segs = bio_segments(failed_bio);
7319 bio_get_first_bvec(failed_bio, &bvec);
7321 (bvec.bv_len > btrfs_inode_sectorsize(inode)))
7322 read_mode |= REQ_FAILFAST_DEV;
7324 isector = start - btrfs_io_bio(failed_bio)->logical;
7325 isector >>= inode->i_sb->s_blocksize_bits;
7326 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7327 pgoff, isector, repair_endio, repair_arg);
7328 bio->bi_opf = REQ_OP_READ | read_mode;
7330 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7331 "repair DIO read error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d",
7332 read_mode, failrec->this_mirror, failrec->in_validation);
7334 status = submit_dio_repair_bio(inode, bio, failrec->this_mirror);
7336 free_io_failure(failure_tree, io_tree, failrec);
7343 struct btrfs_retry_complete {
7344 struct completion done;
7345 struct inode *inode;
7350 static void btrfs_retry_endio_nocsum(struct bio *bio)
7352 struct btrfs_retry_complete *done = bio->bi_private;
7353 struct inode *inode = done->inode;
7354 struct bio_vec *bvec;
7355 struct extent_io_tree *io_tree, *failure_tree;
7356 struct bvec_iter_all iter_all;
7361 ASSERT(bio->bi_vcnt == 1);
7362 io_tree = &BTRFS_I(inode)->io_tree;
7363 failure_tree = &BTRFS_I(inode)->io_failure_tree;
7364 ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(inode));
7367 ASSERT(!bio_flagged(bio, BIO_CLONED));
7368 bio_for_each_segment_all(bvec, bio, iter_all)
7369 clean_io_failure(BTRFS_I(inode)->root->fs_info, failure_tree,
7370 io_tree, done->start, bvec->bv_page,
7371 btrfs_ino(BTRFS_I(inode)), 0);
7373 complete(&done->done);
7377 static blk_status_t __btrfs_correct_data_nocsum(struct inode *inode,
7378 struct btrfs_io_bio *io_bio)
7380 struct btrfs_fs_info *fs_info;
7381 struct bio_vec bvec;
7382 struct bvec_iter iter;
7383 struct btrfs_retry_complete done;
7389 blk_status_t err = BLK_STS_OK;
7391 fs_info = BTRFS_I(inode)->root->fs_info;
7392 sectorsize = fs_info->sectorsize;
7394 start = io_bio->logical;
7396 io_bio->bio.bi_iter = io_bio->iter;
7398 bio_for_each_segment(bvec, &io_bio->bio, iter) {
7399 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7400 pgoff = bvec.bv_offset;
7402 next_block_or_try_again:
7405 init_completion(&done.done);
7407 ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
7408 pgoff, start, start + sectorsize - 1,
7410 btrfs_retry_endio_nocsum, &done);
7416 wait_for_completion_io(&done.done);
7418 if (!done.uptodate) {
7419 /* We might have another mirror, so try again */
7420 goto next_block_or_try_again;
7424 start += sectorsize;
7428 pgoff += sectorsize;
7429 ASSERT(pgoff < PAGE_SIZE);
7430 goto next_block_or_try_again;
7437 static void btrfs_retry_endio(struct bio *bio)
7439 struct btrfs_retry_complete *done = bio->bi_private;
7440 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7441 struct extent_io_tree *io_tree, *failure_tree;
7442 struct inode *inode = done->inode;
7443 struct bio_vec *bvec;
7447 struct bvec_iter_all iter_all;
7454 ASSERT(bio->bi_vcnt == 1);
7455 ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(done->inode));
7457 io_tree = &BTRFS_I(inode)->io_tree;
7458 failure_tree = &BTRFS_I(inode)->io_failure_tree;
7460 ASSERT(!bio_flagged(bio, BIO_CLONED));
7461 bio_for_each_segment_all(bvec, bio, iter_all) {
7462 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
7463 bvec->bv_offset, done->start,
7466 clean_io_failure(BTRFS_I(inode)->root->fs_info,
7467 failure_tree, io_tree, done->start,
7469 btrfs_ino(BTRFS_I(inode)),
7476 done->uptodate = uptodate;
7478 complete(&done->done);
7482 static blk_status_t __btrfs_subio_endio_read(struct inode *inode,
7483 struct btrfs_io_bio *io_bio, blk_status_t err)
7485 struct btrfs_fs_info *fs_info;
7486 struct bio_vec bvec;
7487 struct bvec_iter iter;
7488 struct btrfs_retry_complete done;
7495 bool uptodate = (err == 0);
7497 blk_status_t status;
7499 fs_info = BTRFS_I(inode)->root->fs_info;
7500 sectorsize = fs_info->sectorsize;
7503 start = io_bio->logical;
7505 io_bio->bio.bi_iter = io_bio->iter;
7507 bio_for_each_segment(bvec, &io_bio->bio, iter) {
7508 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7510 pgoff = bvec.bv_offset;
7513 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
7514 ret = __readpage_endio_check(inode, io_bio, csum_pos,
7515 bvec.bv_page, pgoff, start, sectorsize);
7522 init_completion(&done.done);
7524 status = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
7525 pgoff, start, start + sectorsize - 1,
7526 io_bio->mirror_num, btrfs_retry_endio,
7533 wait_for_completion_io(&done.done);
7535 if (!done.uptodate) {
7536 /* We might have another mirror, so try again */
7540 offset += sectorsize;
7541 start += sectorsize;
7547 pgoff += sectorsize;
7548 ASSERT(pgoff < PAGE_SIZE);
7556 static blk_status_t btrfs_subio_endio_read(struct inode *inode,
7557 struct btrfs_io_bio *io_bio, blk_status_t err)
7559 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7563 return __btrfs_correct_data_nocsum(inode, io_bio);
7567 return __btrfs_subio_endio_read(inode, io_bio, err);
7571 static void btrfs_endio_direct_read(struct bio *bio)
7573 struct btrfs_dio_private *dip = bio->bi_private;
7574 struct inode *inode = dip->inode;
7575 struct bio *dio_bio;
7576 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7577 blk_status_t err = bio->bi_status;
7579 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
7580 err = btrfs_subio_endio_read(inode, io_bio, err);
7582 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
7583 dip->logical_offset + dip->bytes - 1);
7584 dio_bio = dip->dio_bio;
7588 dio_bio->bi_status = err;
7589 dio_end_io(dio_bio);
7590 btrfs_io_bio_free_csum(io_bio);
7594 static void __endio_write_update_ordered(struct inode *inode,
7595 const u64 offset, const u64 bytes,
7596 const bool uptodate)
7598 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7599 struct btrfs_ordered_extent *ordered = NULL;
7600 struct btrfs_workqueue *wq;
7601 u64 ordered_offset = offset;
7602 u64 ordered_bytes = bytes;
7605 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
7606 wq = fs_info->endio_freespace_worker;
7608 wq = fs_info->endio_write_workers;
7610 while (ordered_offset < offset + bytes) {
7611 last_offset = ordered_offset;
7612 if (btrfs_dec_test_first_ordered_pending(inode, &ordered,
7616 btrfs_init_work(&ordered->work, finish_ordered_fn, NULL,
7618 btrfs_queue_work(wq, &ordered->work);
7621 * If btrfs_dec_test_ordered_pending does not find any ordered
7622 * extent in the range, we can exit.
7624 if (ordered_offset == last_offset)
7627 * Our bio might span multiple ordered extents. In this case
7628 * we keep going until we have accounted the whole dio.
7630 if (ordered_offset < offset + bytes) {
7631 ordered_bytes = offset + bytes - ordered_offset;
7637 static void btrfs_endio_direct_write(struct bio *bio)
7639 struct btrfs_dio_private *dip = bio->bi_private;
7640 struct bio *dio_bio = dip->dio_bio;
7642 __endio_write_update_ordered(dip->inode, dip->logical_offset,
7643 dip->bytes, !bio->bi_status);
7647 dio_bio->bi_status = bio->bi_status;
7648 dio_end_io(dio_bio);
7652 static blk_status_t btrfs_submit_bio_start_direct_io(void *private_data,
7653 struct bio *bio, u64 offset)
7655 struct inode *inode = private_data;
7657 ret = btrfs_csum_one_bio(inode, bio, offset, 1);
7658 BUG_ON(ret); /* -ENOMEM */
7662 static void btrfs_end_dio_bio(struct bio *bio)
7664 struct btrfs_dio_private *dip = bio->bi_private;
7665 blk_status_t err = bio->bi_status;
7668 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
7669 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
7670 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
7672 (unsigned long long)bio->bi_iter.bi_sector,
7673 bio->bi_iter.bi_size, err);
7675 if (dip->subio_endio)
7676 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
7680 * We want to perceive the errors flag being set before
7681 * decrementing the reference count. We don't need a barrier
7682 * since atomic operations with a return value are fully
7683 * ordered as per atomic_t.txt
7688 /* if there are more bios still pending for this dio, just exit */
7689 if (!atomic_dec_and_test(&dip->pending_bios))
7693 bio_io_error(dip->orig_bio);
7695 dip->dio_bio->bi_status = BLK_STS_OK;
7696 bio_endio(dip->orig_bio);
7702 static inline blk_status_t btrfs_lookup_and_bind_dio_csum(struct inode *inode,
7703 struct btrfs_dio_private *dip,
7707 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7708 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
7712 * We load all the csum data we need when we submit
7713 * the first bio to reduce the csum tree search and
7716 if (dip->logical_offset == file_offset) {
7717 ret = btrfs_lookup_bio_sums(inode, dip->orig_bio, file_offset,
7723 if (bio == dip->orig_bio)
7726 file_offset -= dip->logical_offset;
7727 file_offset >>= inode->i_sb->s_blocksize_bits;
7728 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
7733 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
7734 struct inode *inode, u64 file_offset, int async_submit)
7736 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7737 struct btrfs_dio_private *dip = bio->bi_private;
7738 bool write = bio_op(bio) == REQ_OP_WRITE;
7741 /* Check btrfs_submit_bio_hook() for rules about async submit. */
7743 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7746 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
7751 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
7754 if (write && async_submit) {
7755 ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0,
7757 btrfs_submit_bio_start_direct_io);
7761 * If we aren't doing async submit, calculate the csum of the
7764 ret = btrfs_csum_one_bio(inode, bio, file_offset, 1);
7768 ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio,
7774 ret = btrfs_map_bio(fs_info, bio, 0);
7779 static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip)
7781 struct inode *inode = dip->inode;
7782 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7784 struct bio *orig_bio = dip->orig_bio;
7785 u64 start_sector = orig_bio->bi_iter.bi_sector;
7786 u64 file_offset = dip->logical_offset;
7787 int async_submit = 0;
7789 int clone_offset = 0;
7792 blk_status_t status;
7793 struct btrfs_io_geometry geom;
7795 submit_len = orig_bio->bi_iter.bi_size;
7796 ret = btrfs_get_io_geometry(fs_info, btrfs_op(orig_bio),
7797 start_sector << 9, submit_len, &geom);
7801 if (geom.len >= submit_len) {
7803 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
7807 /* async crcs make it difficult to collect full stripe writes. */
7808 if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK)
7814 ASSERT(geom.len <= INT_MAX);
7815 atomic_inc(&dip->pending_bios);
7817 clone_len = min_t(int, submit_len, geom.len);
7820 * This will never fail as it's passing GPF_NOFS and
7821 * the allocation is backed by btrfs_bioset.
7823 bio = btrfs_bio_clone_partial(orig_bio, clone_offset,
7825 bio->bi_private = dip;
7826 bio->bi_end_io = btrfs_end_dio_bio;
7827 btrfs_io_bio(bio)->logical = file_offset;
7829 ASSERT(submit_len >= clone_len);
7830 submit_len -= clone_len;
7831 if (submit_len == 0)
7835 * Increase the count before we submit the bio so we know
7836 * the end IO handler won't happen before we increase the
7837 * count. Otherwise, the dip might get freed before we're
7838 * done setting it up.
7840 atomic_inc(&dip->pending_bios);
7842 status = btrfs_submit_dio_bio(bio, inode, file_offset,
7846 atomic_dec(&dip->pending_bios);
7850 clone_offset += clone_len;
7851 start_sector += clone_len >> 9;
7852 file_offset += clone_len;
7854 ret = btrfs_get_io_geometry(fs_info, btrfs_op(orig_bio),
7855 start_sector << 9, submit_len, &geom);
7858 } while (submit_len > 0);
7861 status = btrfs_submit_dio_bio(bio, inode, file_offset, async_submit);
7869 * Before atomic variable goto zero, we must make sure dip->errors is
7870 * perceived to be set. This ordering is ensured by the fact that an
7871 * atomic operations with a return value are fully ordered as per
7874 if (atomic_dec_and_test(&dip->pending_bios))
7875 bio_io_error(dip->orig_bio);
7877 /* bio_end_io() will handle error, so we needn't return it */
7881 static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode,
7884 struct btrfs_dio_private *dip = NULL;
7885 struct bio *bio = NULL;
7886 struct btrfs_io_bio *io_bio;
7887 bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
7890 bio = btrfs_bio_clone(dio_bio);
7892 dip = kzalloc(sizeof(*dip), GFP_NOFS);
7898 dip->private = dio_bio->bi_private;
7900 dip->logical_offset = file_offset;
7901 dip->bytes = dio_bio->bi_iter.bi_size;
7902 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
7903 bio->bi_private = dip;
7904 dip->orig_bio = bio;
7905 dip->dio_bio = dio_bio;
7906 atomic_set(&dip->pending_bios, 0);
7907 io_bio = btrfs_io_bio(bio);
7908 io_bio->logical = file_offset;
7911 bio->bi_end_io = btrfs_endio_direct_write;
7913 bio->bi_end_io = btrfs_endio_direct_read;
7914 dip->subio_endio = btrfs_subio_endio_read;
7918 * Reset the range for unsubmitted ordered extents (to a 0 length range)
7919 * even if we fail to submit a bio, because in such case we do the
7920 * corresponding error handling below and it must not be done a second
7921 * time by btrfs_direct_IO().
7924 struct btrfs_dio_data *dio_data = current->journal_info;
7926 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
7928 dio_data->unsubmitted_oe_range_start =
7929 dio_data->unsubmitted_oe_range_end;
7932 ret = btrfs_submit_direct_hook(dip);
7936 btrfs_io_bio_free_csum(io_bio);
7940 * If we arrived here it means either we failed to submit the dip
7941 * or we either failed to clone the dio_bio or failed to allocate the
7942 * dip. If we cloned the dio_bio and allocated the dip, we can just
7943 * call bio_endio against our io_bio so that we get proper resource
7944 * cleanup if we fail to submit the dip, otherwise, we must do the
7945 * same as btrfs_endio_direct_[write|read] because we can't call these
7946 * callbacks - they require an allocated dip and a clone of dio_bio.
7951 * The end io callbacks free our dip, do the final put on bio
7952 * and all the cleanup and final put for dio_bio (through
7959 __endio_write_update_ordered(inode,
7961 dio_bio->bi_iter.bi_size,
7964 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
7965 file_offset + dio_bio->bi_iter.bi_size - 1);
7967 dio_bio->bi_status = BLK_STS_IOERR;
7969 * Releases and cleans up our dio_bio, no need to bio_put()
7970 * nor bio_endio()/bio_io_error() against dio_bio.
7972 dio_end_io(dio_bio);
7979 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
7980 const struct iov_iter *iter, loff_t offset)
7984 unsigned int blocksize_mask = fs_info->sectorsize - 1;
7985 ssize_t retval = -EINVAL;
7987 if (offset & blocksize_mask)
7990 if (iov_iter_alignment(iter) & blocksize_mask)
7993 /* If this is a write we don't need to check anymore */
7994 if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter))
7997 * Check to make sure we don't have duplicate iov_base's in this
7998 * iovec, if so return EINVAL, otherwise we'll get csum errors
7999 * when reading back.
8001 for (seg = 0; seg < iter->nr_segs; seg++) {
8002 for (i = seg + 1; i < iter->nr_segs; i++) {
8003 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8012 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8014 struct file *file = iocb->ki_filp;
8015 struct inode *inode = file->f_mapping->host;
8016 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8017 struct btrfs_dio_data dio_data = { 0 };
8018 struct extent_changeset *data_reserved = NULL;
8019 loff_t offset = iocb->ki_pos;
8023 bool relock = false;
8026 if (check_direct_IO(fs_info, iter, offset))
8029 inode_dio_begin(inode);
8032 * The generic stuff only does filemap_write_and_wait_range, which
8033 * isn't enough if we've written compressed pages to this area, so
8034 * we need to flush the dirty pages again to make absolutely sure
8035 * that any outstanding dirty pages are on disk.
8037 count = iov_iter_count(iter);
8038 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8039 &BTRFS_I(inode)->runtime_flags))
8040 filemap_fdatawrite_range(inode->i_mapping, offset,
8041 offset + count - 1);
8043 if (iov_iter_rw(iter) == WRITE) {
8045 * If the write DIO is beyond the EOF, we need update
8046 * the isize, but it is protected by i_mutex. So we can
8047 * not unlock the i_mutex at this case.
8049 if (offset + count <= inode->i_size) {
8050 dio_data.overwrite = 1;
8051 inode_unlock(inode);
8053 } else if (iocb->ki_flags & IOCB_NOWAIT) {
8057 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
8063 * We need to know how many extents we reserved so that we can
8064 * do the accounting properly if we go over the number we
8065 * originally calculated. Abuse current->journal_info for this.
8067 dio_data.reserve = round_up(count,
8068 fs_info->sectorsize);
8069 dio_data.unsubmitted_oe_range_start = (u64)offset;
8070 dio_data.unsubmitted_oe_range_end = (u64)offset;
8071 current->journal_info = &dio_data;
8072 down_read(&BTRFS_I(inode)->dio_sem);
8073 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8074 &BTRFS_I(inode)->runtime_flags)) {
8075 inode_dio_end(inode);
8076 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8080 ret = __blockdev_direct_IO(iocb, inode,
8081 fs_info->fs_devices->latest_bdev,
8082 iter, btrfs_get_blocks_direct, NULL,
8083 btrfs_submit_direct, flags);
8084 if (iov_iter_rw(iter) == WRITE) {
8085 up_read(&BTRFS_I(inode)->dio_sem);
8086 current->journal_info = NULL;
8087 if (ret < 0 && ret != -EIOCBQUEUED) {
8088 if (dio_data.reserve)
8089 btrfs_delalloc_release_space(inode, data_reserved,
8090 offset, dio_data.reserve, true);
8092 * On error we might have left some ordered extents
8093 * without submitting corresponding bios for them, so
8094 * cleanup them up to avoid other tasks getting them
8095 * and waiting for them to complete forever.
8097 if (dio_data.unsubmitted_oe_range_start <
8098 dio_data.unsubmitted_oe_range_end)
8099 __endio_write_update_ordered(inode,
8100 dio_data.unsubmitted_oe_range_start,
8101 dio_data.unsubmitted_oe_range_end -
8102 dio_data.unsubmitted_oe_range_start,
8104 } else if (ret >= 0 && (size_t)ret < count)
8105 btrfs_delalloc_release_space(inode, data_reserved,
8106 offset, count - (size_t)ret, true);
8107 btrfs_delalloc_release_extents(BTRFS_I(inode), count);
8111 inode_dio_end(inode);
8115 extent_changeset_free(data_reserved);
8119 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8121 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8122 __u64 start, __u64 len)
8126 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8130 return extent_fiemap(inode, fieinfo, start, len);
8133 int btrfs_readpage(struct file *file, struct page *page)
8135 struct extent_io_tree *tree;
8136 tree = &BTRFS_I(page->mapping->host)->io_tree;
8137 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8140 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8142 struct inode *inode = page->mapping->host;
8145 if (current->flags & PF_MEMALLOC) {
8146 redirty_page_for_writepage(wbc, page);
8152 * If we are under memory pressure we will call this directly from the
8153 * VM, we need to make sure we have the inode referenced for the ordered
8154 * extent. If not just return like we didn't do anything.
8156 if (!igrab(inode)) {
8157 redirty_page_for_writepage(wbc, page);
8158 return AOP_WRITEPAGE_ACTIVATE;
8160 ret = extent_write_full_page(page, wbc);
8161 btrfs_add_delayed_iput(inode);
8165 static int btrfs_writepages(struct address_space *mapping,
8166 struct writeback_control *wbc)
8168 return extent_writepages(mapping, wbc);
8172 btrfs_readpages(struct file *file, struct address_space *mapping,
8173 struct list_head *pages, unsigned nr_pages)
8175 return extent_readpages(mapping, pages, nr_pages);
8178 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8180 int ret = try_release_extent_mapping(page, gfp_flags);
8182 ClearPagePrivate(page);
8183 set_page_private(page, 0);
8189 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8191 if (PageWriteback(page) || PageDirty(page))
8193 return __btrfs_releasepage(page, gfp_flags);
8196 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8197 unsigned int length)
8199 struct inode *inode = page->mapping->host;
8200 struct extent_io_tree *tree;
8201 struct btrfs_ordered_extent *ordered;
8202 struct extent_state *cached_state = NULL;
8203 u64 page_start = page_offset(page);
8204 u64 page_end = page_start + PAGE_SIZE - 1;
8207 int inode_evicting = inode->i_state & I_FREEING;
8210 * we have the page locked, so new writeback can't start,
8211 * and the dirty bit won't be cleared while we are here.
8213 * Wait for IO on this page so that we can safely clear
8214 * the PagePrivate2 bit and do ordered accounting
8216 wait_on_page_writeback(page);
8218 tree = &BTRFS_I(inode)->io_tree;
8220 btrfs_releasepage(page, GFP_NOFS);
8224 if (!inode_evicting)
8225 lock_extent_bits(tree, page_start, page_end, &cached_state);
8228 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
8229 page_end - start + 1);
8232 ordered->file_offset + ordered->num_bytes - 1);
8234 * IO on this page will never be started, so we need
8235 * to account for any ordered extents now
8237 if (!inode_evicting)
8238 clear_extent_bit(tree, start, end,
8239 EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
8240 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8241 EXTENT_DEFRAG, 1, 0, &cached_state);
8243 * whoever cleared the private bit is responsible
8244 * for the finish_ordered_io
8246 if (TestClearPagePrivate2(page)) {
8247 struct btrfs_ordered_inode_tree *tree;
8250 tree = &BTRFS_I(inode)->ordered_tree;
8252 spin_lock_irq(&tree->lock);
8253 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8254 new_len = start - ordered->file_offset;
8255 if (new_len < ordered->truncated_len)
8256 ordered->truncated_len = new_len;
8257 spin_unlock_irq(&tree->lock);
8259 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8261 end - start + 1, 1))
8262 btrfs_finish_ordered_io(ordered);
8264 btrfs_put_ordered_extent(ordered);
8265 if (!inode_evicting) {
8266 cached_state = NULL;
8267 lock_extent_bits(tree, start, end,
8272 if (start < page_end)
8277 * Qgroup reserved space handler
8278 * Page here will be either
8279 * 1) Already written to disk
8280 * In this case, its reserved space is released from data rsv map
8281 * and will be freed by delayed_ref handler finally.
8282 * So even we call qgroup_free_data(), it won't decrease reserved
8284 * 2) Not written to disk
8285 * This means the reserved space should be freed here. However,
8286 * if a truncate invalidates the page (by clearing PageDirty)
8287 * and the page is accounted for while allocating extent
8288 * in btrfs_check_data_free_space() we let delayed_ref to
8289 * free the entire extent.
8291 if (PageDirty(page))
8292 btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
8293 if (!inode_evicting) {
8294 clear_extent_bit(tree, page_start, page_end, EXTENT_LOCKED |
8295 EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
8296 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
8299 __btrfs_releasepage(page, GFP_NOFS);
8302 ClearPageChecked(page);
8303 if (PagePrivate(page)) {
8304 ClearPagePrivate(page);
8305 set_page_private(page, 0);
8311 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8312 * called from a page fault handler when a page is first dirtied. Hence we must
8313 * be careful to check for EOF conditions here. We set the page up correctly
8314 * for a written page which means we get ENOSPC checking when writing into
8315 * holes and correct delalloc and unwritten extent mapping on filesystems that
8316 * support these features.
8318 * We are not allowed to take the i_mutex here so we have to play games to
8319 * protect against truncate races as the page could now be beyond EOF. Because
8320 * truncate_setsize() writes the inode size before removing pages, once we have
8321 * the page lock we can determine safely if the page is beyond EOF. If it is not
8322 * beyond EOF, then the page is guaranteed safe against truncation until we
8325 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8327 struct page *page = vmf->page;
8328 struct inode *inode = file_inode(vmf->vma->vm_file);
8329 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8330 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8331 struct btrfs_ordered_extent *ordered;
8332 struct extent_state *cached_state = NULL;
8333 struct extent_changeset *data_reserved = NULL;
8335 unsigned long zero_start;
8345 reserved_space = PAGE_SIZE;
8347 sb_start_pagefault(inode->i_sb);
8348 page_start = page_offset(page);
8349 page_end = page_start + PAGE_SIZE - 1;
8353 * Reserving delalloc space after obtaining the page lock can lead to
8354 * deadlock. For example, if a dirty page is locked by this function
8355 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8356 * dirty page write out, then the btrfs_writepage() function could
8357 * end up waiting indefinitely to get a lock on the page currently
8358 * being processed by btrfs_page_mkwrite() function.
8360 ret2 = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
8363 ret2 = file_update_time(vmf->vma->vm_file);
8367 ret = vmf_error(ret2);
8373 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8376 size = i_size_read(inode);
8378 if ((page->mapping != inode->i_mapping) ||
8379 (page_start >= size)) {
8380 /* page got truncated out from underneath us */
8383 wait_on_page_writeback(page);
8385 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8386 set_page_extent_mapped(page);
8389 * we can't set the delalloc bits if there are pending ordered
8390 * extents. Drop our locks and wait for them to finish
8392 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8395 unlock_extent_cached(io_tree, page_start, page_end,
8398 btrfs_start_ordered_extent(inode, ordered, 1);
8399 btrfs_put_ordered_extent(ordered);
8403 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8404 reserved_space = round_up(size - page_start,
8405 fs_info->sectorsize);
8406 if (reserved_space < PAGE_SIZE) {
8407 end = page_start + reserved_space - 1;
8408 btrfs_delalloc_release_space(inode, data_reserved,
8409 page_start, PAGE_SIZE - reserved_space,
8415 * page_mkwrite gets called when the page is firstly dirtied after it's
8416 * faulted in, but write(2) could also dirty a page and set delalloc
8417 * bits, thus in this case for space account reason, we still need to
8418 * clear any delalloc bits within this page range since we have to
8419 * reserve data&meta space before lock_page() (see above comments).
8421 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8422 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8423 EXTENT_DEFRAG, 0, 0, &cached_state);
8425 ret2 = btrfs_set_extent_delalloc(inode, page_start, end, 0,
8428 unlock_extent_cached(io_tree, page_start, page_end,
8430 ret = VM_FAULT_SIGBUS;
8434 /* page is wholly or partially inside EOF */
8435 if (page_start + PAGE_SIZE > size)
8436 zero_start = offset_in_page(size);
8438 zero_start = PAGE_SIZE;
8440 if (zero_start != PAGE_SIZE) {
8442 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
8443 flush_dcache_page(page);
8446 ClearPageChecked(page);
8447 set_page_dirty(page);
8448 SetPageUptodate(page);
8450 BTRFS_I(inode)->last_trans = fs_info->generation;
8451 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8452 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8454 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8456 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8457 sb_end_pagefault(inode->i_sb);
8458 extent_changeset_free(data_reserved);
8459 return VM_FAULT_LOCKED;
8464 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8465 btrfs_delalloc_release_space(inode, data_reserved, page_start,
8466 reserved_space, (ret != 0));
8468 sb_end_pagefault(inode->i_sb);
8469 extent_changeset_free(data_reserved);
8473 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
8475 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8476 struct btrfs_root *root = BTRFS_I(inode)->root;
8477 struct btrfs_block_rsv *rsv;
8479 struct btrfs_trans_handle *trans;
8480 u64 mask = fs_info->sectorsize - 1;
8481 u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8483 if (!skip_writeback) {
8484 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8491 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8492 * things going on here:
8494 * 1) We need to reserve space to update our inode.
8496 * 2) We need to have something to cache all the space that is going to
8497 * be free'd up by the truncate operation, but also have some slack
8498 * space reserved in case it uses space during the truncate (thank you
8499 * very much snapshotting).
8501 * And we need these to be separate. The fact is we can use a lot of
8502 * space doing the truncate, and we have no earthly idea how much space
8503 * we will use, so we need the truncate reservation to be separate so it
8504 * doesn't end up using space reserved for updating the inode. We also
8505 * need to be able to stop the transaction and start a new one, which
8506 * means we need to be able to update the inode several times, and we
8507 * have no idea of knowing how many times that will be, so we can't just
8508 * reserve 1 item for the entirety of the operation, so that has to be
8509 * done separately as well.
8511 * So that leaves us with
8513 * 1) rsv - for the truncate reservation, which we will steal from the
8514 * transaction reservation.
8515 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8516 * updating the inode.
8518 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8521 rsv->size = min_size;
8525 * 1 for the truncate slack space
8526 * 1 for updating the inode.
8528 trans = btrfs_start_transaction(root, 2);
8529 if (IS_ERR(trans)) {
8530 ret = PTR_ERR(trans);
8534 /* Migrate the slack space for the truncate to our reserve */
8535 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8540 * So if we truncate and then write and fsync we normally would just
8541 * write the extents that changed, which is a problem if we need to
8542 * first truncate that entire inode. So set this flag so we write out
8543 * all of the extents in the inode to the sync log so we're completely
8546 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
8547 trans->block_rsv = rsv;
8550 ret = btrfs_truncate_inode_items(trans, root, inode,
8552 BTRFS_EXTENT_DATA_KEY);
8553 trans->block_rsv = &fs_info->trans_block_rsv;
8554 if (ret != -ENOSPC && ret != -EAGAIN)
8557 ret = btrfs_update_inode(trans, root, inode);
8561 btrfs_end_transaction(trans);
8562 btrfs_btree_balance_dirty(fs_info);
8564 trans = btrfs_start_transaction(root, 2);
8565 if (IS_ERR(trans)) {
8566 ret = PTR_ERR(trans);
8571 btrfs_block_rsv_release(fs_info, rsv, -1);
8572 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8573 rsv, min_size, false);
8574 BUG_ON(ret); /* shouldn't happen */
8575 trans->block_rsv = rsv;
8579 * We can't call btrfs_truncate_block inside a trans handle as we could
8580 * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
8581 * we've truncated everything except the last little bit, and can do
8582 * btrfs_truncate_block and then update the disk_i_size.
8584 if (ret == NEED_TRUNCATE_BLOCK) {
8585 btrfs_end_transaction(trans);
8586 btrfs_btree_balance_dirty(fs_info);
8588 ret = btrfs_truncate_block(inode, inode->i_size, 0, 0);
8591 trans = btrfs_start_transaction(root, 1);
8592 if (IS_ERR(trans)) {
8593 ret = PTR_ERR(trans);
8596 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
8602 trans->block_rsv = &fs_info->trans_block_rsv;
8603 ret2 = btrfs_update_inode(trans, root, inode);
8607 ret2 = btrfs_end_transaction(trans);
8610 btrfs_btree_balance_dirty(fs_info);
8613 btrfs_free_block_rsv(fs_info, rsv);
8619 * create a new subvolume directory/inode (helper for the ioctl).
8621 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8622 struct btrfs_root *new_root,
8623 struct btrfs_root *parent_root,
8626 struct inode *inode;
8630 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
8631 new_dirid, new_dirid,
8632 S_IFDIR | (~current_umask() & S_IRWXUGO),
8635 return PTR_ERR(inode);
8636 inode->i_op = &btrfs_dir_inode_operations;
8637 inode->i_fop = &btrfs_dir_file_operations;
8639 set_nlink(inode, 1);
8640 btrfs_i_size_write(BTRFS_I(inode), 0);
8641 unlock_new_inode(inode);
8643 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8645 btrfs_err(new_root->fs_info,
8646 "error inheriting subvolume %llu properties: %d",
8647 new_root->root_key.objectid, err);
8649 err = btrfs_update_inode(trans, new_root, inode);
8655 struct inode *btrfs_alloc_inode(struct super_block *sb)
8657 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8658 struct btrfs_inode *ei;
8659 struct inode *inode;
8661 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL);
8668 ei->last_sub_trans = 0;
8669 ei->logged_trans = 0;
8670 ei->delalloc_bytes = 0;
8671 ei->new_delalloc_bytes = 0;
8672 ei->defrag_bytes = 0;
8673 ei->disk_i_size = 0;
8676 ei->index_cnt = (u64)-1;
8678 ei->last_unlink_trans = 0;
8679 ei->last_log_commit = 0;
8681 spin_lock_init(&ei->lock);
8682 ei->outstanding_extents = 0;
8683 if (sb->s_magic != BTRFS_TEST_MAGIC)
8684 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8685 BTRFS_BLOCK_RSV_DELALLOC);
8686 ei->runtime_flags = 0;
8687 ei->prop_compress = BTRFS_COMPRESS_NONE;
8688 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8690 ei->delayed_node = NULL;
8692 ei->i_otime.tv_sec = 0;
8693 ei->i_otime.tv_nsec = 0;
8695 inode = &ei->vfs_inode;
8696 extent_map_tree_init(&ei->extent_tree);
8697 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO, inode);
8698 extent_io_tree_init(fs_info, &ei->io_failure_tree,
8699 IO_TREE_INODE_IO_FAILURE, inode);
8700 ei->io_tree.track_uptodate = true;
8701 ei->io_failure_tree.track_uptodate = true;
8702 atomic_set(&ei->sync_writers, 0);
8703 mutex_init(&ei->log_mutex);
8704 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8705 INIT_LIST_HEAD(&ei->delalloc_inodes);
8706 INIT_LIST_HEAD(&ei->delayed_iput);
8707 RB_CLEAR_NODE(&ei->rb_node);
8708 init_rwsem(&ei->dio_sem);
8713 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8714 void btrfs_test_destroy_inode(struct inode *inode)
8716 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
8717 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8721 void btrfs_free_inode(struct inode *inode)
8723 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8726 void btrfs_destroy_inode(struct inode *inode)
8728 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8729 struct btrfs_ordered_extent *ordered;
8730 struct btrfs_root *root = BTRFS_I(inode)->root;
8732 WARN_ON(!hlist_empty(&inode->i_dentry));
8733 WARN_ON(inode->i_data.nrpages);
8734 WARN_ON(BTRFS_I(inode)->block_rsv.reserved);
8735 WARN_ON(BTRFS_I(inode)->block_rsv.size);
8736 WARN_ON(BTRFS_I(inode)->outstanding_extents);
8737 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
8738 WARN_ON(BTRFS_I(inode)->new_delalloc_bytes);
8739 WARN_ON(BTRFS_I(inode)->csum_bytes);
8740 WARN_ON(BTRFS_I(inode)->defrag_bytes);
8743 * This can happen where we create an inode, but somebody else also
8744 * created the same inode and we need to destroy the one we already
8751 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8756 "found ordered extent %llu %llu on inode cleanup",
8757 ordered->file_offset, ordered->num_bytes);
8758 btrfs_remove_ordered_extent(inode, ordered);
8759 btrfs_put_ordered_extent(ordered);
8760 btrfs_put_ordered_extent(ordered);
8763 btrfs_qgroup_check_reserved_leak(inode);
8764 inode_tree_del(inode);
8765 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
8768 int btrfs_drop_inode(struct inode *inode)
8770 struct btrfs_root *root = BTRFS_I(inode)->root;
8775 /* the snap/subvol tree is on deleting */
8776 if (btrfs_root_refs(&root->root_item) == 0)
8779 return generic_drop_inode(inode);
8782 static void init_once(void *foo)
8784 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8786 inode_init_once(&ei->vfs_inode);
8789 void __cold btrfs_destroy_cachep(void)
8792 * Make sure all delayed rcu free inodes are flushed before we
8796 kmem_cache_destroy(btrfs_inode_cachep);
8797 kmem_cache_destroy(btrfs_trans_handle_cachep);
8798 kmem_cache_destroy(btrfs_path_cachep);
8799 kmem_cache_destroy(btrfs_free_space_cachep);
8800 kmem_cache_destroy(btrfs_free_space_bitmap_cachep);
8803 int __init btrfs_init_cachep(void)
8805 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8806 sizeof(struct btrfs_inode), 0,
8807 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
8809 if (!btrfs_inode_cachep)
8812 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8813 sizeof(struct btrfs_trans_handle), 0,
8814 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
8815 if (!btrfs_trans_handle_cachep)
8818 btrfs_path_cachep = kmem_cache_create("btrfs_path",
8819 sizeof(struct btrfs_path), 0,
8820 SLAB_MEM_SPREAD, NULL);
8821 if (!btrfs_path_cachep)
8824 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8825 sizeof(struct btrfs_free_space), 0,
8826 SLAB_MEM_SPREAD, NULL);
8827 if (!btrfs_free_space_cachep)
8830 btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap",
8831 PAGE_SIZE, PAGE_SIZE,
8832 SLAB_RED_ZONE, NULL);
8833 if (!btrfs_free_space_bitmap_cachep)
8838 btrfs_destroy_cachep();
8842 static int btrfs_getattr(const struct path *path, struct kstat *stat,
8843 u32 request_mask, unsigned int flags)
8846 struct inode *inode = d_inode(path->dentry);
8847 u32 blocksize = inode->i_sb->s_blocksize;
8848 u32 bi_flags = BTRFS_I(inode)->flags;
8850 stat->result_mask |= STATX_BTIME;
8851 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
8852 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
8853 if (bi_flags & BTRFS_INODE_APPEND)
8854 stat->attributes |= STATX_ATTR_APPEND;
8855 if (bi_flags & BTRFS_INODE_COMPRESS)
8856 stat->attributes |= STATX_ATTR_COMPRESSED;
8857 if (bi_flags & BTRFS_INODE_IMMUTABLE)
8858 stat->attributes |= STATX_ATTR_IMMUTABLE;
8859 if (bi_flags & BTRFS_INODE_NODUMP)
8860 stat->attributes |= STATX_ATTR_NODUMP;
8862 stat->attributes_mask |= (STATX_ATTR_APPEND |
8863 STATX_ATTR_COMPRESSED |
8864 STATX_ATTR_IMMUTABLE |
8867 generic_fillattr(inode, stat);
8868 stat->dev = BTRFS_I(inode)->root->anon_dev;
8870 spin_lock(&BTRFS_I(inode)->lock);
8871 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
8872 spin_unlock(&BTRFS_I(inode)->lock);
8873 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8874 ALIGN(delalloc_bytes, blocksize)) >> 9;
8878 static int btrfs_rename_exchange(struct inode *old_dir,
8879 struct dentry *old_dentry,
8880 struct inode *new_dir,
8881 struct dentry *new_dentry)
8883 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
8884 struct btrfs_trans_handle *trans;
8885 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8886 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8887 struct inode *new_inode = new_dentry->d_inode;
8888 struct inode *old_inode = old_dentry->d_inode;
8889 struct timespec64 ctime = current_time(old_inode);
8890 struct dentry *parent;
8891 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
8892 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
8896 bool root_log_pinned = false;
8897 bool dest_log_pinned = false;
8898 struct btrfs_log_ctx ctx_root;
8899 struct btrfs_log_ctx ctx_dest;
8900 bool sync_log_root = false;
8901 bool sync_log_dest = false;
8902 bool commit_transaction = false;
8904 /* we only allow rename subvolume link between subvolumes */
8905 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8908 btrfs_init_log_ctx(&ctx_root, old_inode);
8909 btrfs_init_log_ctx(&ctx_dest, new_inode);
8911 /* close the race window with snapshot create/destroy ioctl */
8912 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
8913 new_ino == BTRFS_FIRST_FREE_OBJECTID)
8914 down_read(&fs_info->subvol_sem);
8917 * We want to reserve the absolute worst case amount of items. So if
8918 * both inodes are subvols and we need to unlink them then that would
8919 * require 4 item modifications, but if they are both normal inodes it
8920 * would require 5 item modifications, so we'll assume their normal
8921 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
8922 * should cover the worst case number of items we'll modify.
8924 trans = btrfs_start_transaction(root, 12);
8925 if (IS_ERR(trans)) {
8926 ret = PTR_ERR(trans);
8931 btrfs_record_root_in_trans(trans, dest);
8934 * We need to find a free sequence number both in the source and
8935 * in the destination directory for the exchange.
8937 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
8940 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
8944 BTRFS_I(old_inode)->dir_index = 0ULL;
8945 BTRFS_I(new_inode)->dir_index = 0ULL;
8947 /* Reference for the source. */
8948 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8949 /* force full log commit if subvolume involved. */
8950 btrfs_set_log_full_commit(trans);
8952 btrfs_pin_log_trans(root);
8953 root_log_pinned = true;
8954 ret = btrfs_insert_inode_ref(trans, dest,
8955 new_dentry->d_name.name,
8956 new_dentry->d_name.len,
8958 btrfs_ino(BTRFS_I(new_dir)),
8964 /* And now for the dest. */
8965 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8966 /* force full log commit if subvolume involved. */
8967 btrfs_set_log_full_commit(trans);
8969 btrfs_pin_log_trans(dest);
8970 dest_log_pinned = true;
8971 ret = btrfs_insert_inode_ref(trans, root,
8972 old_dentry->d_name.name,
8973 old_dentry->d_name.len,
8975 btrfs_ino(BTRFS_I(old_dir)),
8981 /* Update inode version and ctime/mtime. */
8982 inode_inc_iversion(old_dir);
8983 inode_inc_iversion(new_dir);
8984 inode_inc_iversion(old_inode);
8985 inode_inc_iversion(new_inode);
8986 old_dir->i_ctime = old_dir->i_mtime = ctime;
8987 new_dir->i_ctime = new_dir->i_mtime = ctime;
8988 old_inode->i_ctime = ctime;
8989 new_inode->i_ctime = ctime;
8991 if (old_dentry->d_parent != new_dentry->d_parent) {
8992 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
8993 BTRFS_I(old_inode), 1);
8994 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
8995 BTRFS_I(new_inode), 1);
8998 /* src is a subvolume */
8999 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9000 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9001 } else { /* src is an inode */
9002 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9003 BTRFS_I(old_dentry->d_inode),
9004 old_dentry->d_name.name,
9005 old_dentry->d_name.len);
9007 ret = btrfs_update_inode(trans, root, old_inode);
9010 btrfs_abort_transaction(trans, ret);
9014 /* dest is a subvolume */
9015 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9016 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9017 } else { /* dest is an inode */
9018 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9019 BTRFS_I(new_dentry->d_inode),
9020 new_dentry->d_name.name,
9021 new_dentry->d_name.len);
9023 ret = btrfs_update_inode(trans, dest, new_inode);
9026 btrfs_abort_transaction(trans, ret);
9030 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9031 new_dentry->d_name.name,
9032 new_dentry->d_name.len, 0, old_idx);
9034 btrfs_abort_transaction(trans, ret);
9038 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9039 old_dentry->d_name.name,
9040 old_dentry->d_name.len, 0, new_idx);
9042 btrfs_abort_transaction(trans, ret);
9046 if (old_inode->i_nlink == 1)
9047 BTRFS_I(old_inode)->dir_index = old_idx;
9048 if (new_inode->i_nlink == 1)
9049 BTRFS_I(new_inode)->dir_index = new_idx;
9051 if (root_log_pinned) {
9052 parent = new_dentry->d_parent;
9053 ret = btrfs_log_new_name(trans, BTRFS_I(old_inode),
9054 BTRFS_I(old_dir), parent,
9056 if (ret == BTRFS_NEED_LOG_SYNC)
9057 sync_log_root = true;
9058 else if (ret == BTRFS_NEED_TRANS_COMMIT)
9059 commit_transaction = true;
9061 btrfs_end_log_trans(root);
9062 root_log_pinned = false;
9064 if (dest_log_pinned) {
9065 if (!commit_transaction) {
9066 parent = old_dentry->d_parent;
9067 ret = btrfs_log_new_name(trans, BTRFS_I(new_inode),
9068 BTRFS_I(new_dir), parent,
9070 if (ret == BTRFS_NEED_LOG_SYNC)
9071 sync_log_dest = true;
9072 else if (ret == BTRFS_NEED_TRANS_COMMIT)
9073 commit_transaction = true;
9076 btrfs_end_log_trans(dest);
9077 dest_log_pinned = false;
9081 * If we have pinned a log and an error happened, we unpin tasks
9082 * trying to sync the log and force them to fallback to a transaction
9083 * commit if the log currently contains any of the inodes involved in
9084 * this rename operation (to ensure we do not persist a log with an
9085 * inconsistent state for any of these inodes or leading to any
9086 * inconsistencies when replayed). If the transaction was aborted, the
9087 * abortion reason is propagated to userspace when attempting to commit
9088 * the transaction. If the log does not contain any of these inodes, we
9089 * allow the tasks to sync it.
9091 if (ret && (root_log_pinned || dest_log_pinned)) {
9092 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9093 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9094 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9096 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9097 btrfs_set_log_full_commit(trans);
9099 if (root_log_pinned) {
9100 btrfs_end_log_trans(root);
9101 root_log_pinned = false;
9103 if (dest_log_pinned) {
9104 btrfs_end_log_trans(dest);
9105 dest_log_pinned = false;
9108 if (!ret && sync_log_root && !commit_transaction) {
9109 ret = btrfs_sync_log(trans, BTRFS_I(old_inode)->root,
9112 commit_transaction = true;
9114 if (!ret && sync_log_dest && !commit_transaction) {
9115 ret = btrfs_sync_log(trans, BTRFS_I(new_inode)->root,
9118 commit_transaction = true;
9120 if (commit_transaction) {
9122 * We may have set commit_transaction when logging the new name
9123 * in the destination root, in which case we left the source
9124 * root context in the list of log contextes. So make sure we
9125 * remove it to avoid invalid memory accesses, since the context
9126 * was allocated in our stack frame.
9128 if (sync_log_root) {
9129 mutex_lock(&root->log_mutex);
9130 list_del_init(&ctx_root.list);
9131 mutex_unlock(&root->log_mutex);
9133 ret = btrfs_commit_transaction(trans);
9137 ret2 = btrfs_end_transaction(trans);
9138 ret = ret ? ret : ret2;
9141 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
9142 old_ino == BTRFS_FIRST_FREE_OBJECTID)
9143 up_read(&fs_info->subvol_sem);
9145 ASSERT(list_empty(&ctx_root.list));
9146 ASSERT(list_empty(&ctx_dest.list));
9151 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9152 struct btrfs_root *root,
9154 struct dentry *dentry)
9157 struct inode *inode;
9161 ret = btrfs_find_free_ino(root, &objectid);
9165 inode = btrfs_new_inode(trans, root, dir,
9166 dentry->d_name.name,
9168 btrfs_ino(BTRFS_I(dir)),
9170 S_IFCHR | WHITEOUT_MODE,
9173 if (IS_ERR(inode)) {
9174 ret = PTR_ERR(inode);
9178 inode->i_op = &btrfs_special_inode_operations;
9179 init_special_inode(inode, inode->i_mode,
9182 ret = btrfs_init_inode_security(trans, inode, dir,
9187 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9188 BTRFS_I(inode), 0, index);
9192 ret = btrfs_update_inode(trans, root, inode);
9194 unlock_new_inode(inode);
9196 inode_dec_link_count(inode);
9202 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9203 struct inode *new_dir, struct dentry *new_dentry,
9206 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9207 struct btrfs_trans_handle *trans;
9208 unsigned int trans_num_items;
9209 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9210 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9211 struct inode *new_inode = d_inode(new_dentry);
9212 struct inode *old_inode = d_inode(old_dentry);
9215 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9216 bool log_pinned = false;
9217 struct btrfs_log_ctx ctx;
9218 bool sync_log = false;
9219 bool commit_transaction = false;
9221 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9224 /* we only allow rename subvolume link between subvolumes */
9225 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9228 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9229 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9232 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9233 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9237 /* check for collisions, even if the name isn't there */
9238 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9239 new_dentry->d_name.name,
9240 new_dentry->d_name.len);
9243 if (ret == -EEXIST) {
9245 * eexist without a new_inode */
9246 if (WARN_ON(!new_inode)) {
9250 /* maybe -EOVERFLOW */
9257 * we're using rename to replace one file with another. Start IO on it
9258 * now so we don't add too much work to the end of the transaction
9260 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9261 filemap_flush(old_inode->i_mapping);
9263 /* close the racy window with snapshot create/destroy ioctl */
9264 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9265 down_read(&fs_info->subvol_sem);
9267 * We want to reserve the absolute worst case amount of items. So if
9268 * both inodes are subvols and we need to unlink them then that would
9269 * require 4 item modifications, but if they are both normal inodes it
9270 * would require 5 item modifications, so we'll assume they are normal
9271 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9272 * should cover the worst case number of items we'll modify.
9273 * If our rename has the whiteout flag, we need more 5 units for the
9274 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9275 * when selinux is enabled).
9277 trans_num_items = 11;
9278 if (flags & RENAME_WHITEOUT)
9279 trans_num_items += 5;
9280 trans = btrfs_start_transaction(root, trans_num_items);
9281 if (IS_ERR(trans)) {
9282 ret = PTR_ERR(trans);
9287 btrfs_record_root_in_trans(trans, dest);
9289 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9293 BTRFS_I(old_inode)->dir_index = 0ULL;
9294 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9295 /* force full log commit if subvolume involved. */
9296 btrfs_set_log_full_commit(trans);
9298 btrfs_pin_log_trans(root);
9300 ret = btrfs_insert_inode_ref(trans, dest,
9301 new_dentry->d_name.name,
9302 new_dentry->d_name.len,
9304 btrfs_ino(BTRFS_I(new_dir)), index);
9309 inode_inc_iversion(old_dir);
9310 inode_inc_iversion(new_dir);
9311 inode_inc_iversion(old_inode);
9312 old_dir->i_ctime = old_dir->i_mtime =
9313 new_dir->i_ctime = new_dir->i_mtime =
9314 old_inode->i_ctime = current_time(old_dir);
9316 if (old_dentry->d_parent != new_dentry->d_parent)
9317 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9318 BTRFS_I(old_inode), 1);
9320 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9321 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9323 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9324 BTRFS_I(d_inode(old_dentry)),
9325 old_dentry->d_name.name,
9326 old_dentry->d_name.len);
9328 ret = btrfs_update_inode(trans, root, old_inode);
9331 btrfs_abort_transaction(trans, ret);
9336 inode_inc_iversion(new_inode);
9337 new_inode->i_ctime = current_time(new_inode);
9338 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9339 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9340 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9341 BUG_ON(new_inode->i_nlink == 0);
9343 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9344 BTRFS_I(d_inode(new_dentry)),
9345 new_dentry->d_name.name,
9346 new_dentry->d_name.len);
9348 if (!ret && new_inode->i_nlink == 0)
9349 ret = btrfs_orphan_add(trans,
9350 BTRFS_I(d_inode(new_dentry)));
9352 btrfs_abort_transaction(trans, ret);
9357 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9358 new_dentry->d_name.name,
9359 new_dentry->d_name.len, 0, index);
9361 btrfs_abort_transaction(trans, ret);
9365 if (old_inode->i_nlink == 1)
9366 BTRFS_I(old_inode)->dir_index = index;
9369 struct dentry *parent = new_dentry->d_parent;
9371 btrfs_init_log_ctx(&ctx, old_inode);
9372 ret = btrfs_log_new_name(trans, BTRFS_I(old_inode),
9373 BTRFS_I(old_dir), parent,
9375 if (ret == BTRFS_NEED_LOG_SYNC)
9377 else if (ret == BTRFS_NEED_TRANS_COMMIT)
9378 commit_transaction = true;
9380 btrfs_end_log_trans(root);
9384 if (flags & RENAME_WHITEOUT) {
9385 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9389 btrfs_abort_transaction(trans, ret);
9395 * If we have pinned the log and an error happened, we unpin tasks
9396 * trying to sync the log and force them to fallback to a transaction
9397 * commit if the log currently contains any of the inodes involved in
9398 * this rename operation (to ensure we do not persist a log with an
9399 * inconsistent state for any of these inodes or leading to any
9400 * inconsistencies when replayed). If the transaction was aborted, the
9401 * abortion reason is propagated to userspace when attempting to commit
9402 * the transaction. If the log does not contain any of these inodes, we
9403 * allow the tasks to sync it.
9405 if (ret && log_pinned) {
9406 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9407 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9408 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9410 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9411 btrfs_set_log_full_commit(trans);
9413 btrfs_end_log_trans(root);
9416 if (!ret && sync_log) {
9417 ret = btrfs_sync_log(trans, BTRFS_I(old_inode)->root, &ctx);
9419 commit_transaction = true;
9421 if (commit_transaction) {
9422 ret = btrfs_commit_transaction(trans);
9426 ret2 = btrfs_end_transaction(trans);
9427 ret = ret ? ret : ret2;
9430 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9431 up_read(&fs_info->subvol_sem);
9436 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9437 struct inode *new_dir, struct dentry *new_dentry,
9440 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9443 if (flags & RENAME_EXCHANGE)
9444 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9447 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9450 struct btrfs_delalloc_work {
9451 struct inode *inode;
9452 struct completion completion;
9453 struct list_head list;
9454 struct btrfs_work work;
9457 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9459 struct btrfs_delalloc_work *delalloc_work;
9460 struct inode *inode;
9462 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9464 inode = delalloc_work->inode;
9465 filemap_flush(inode->i_mapping);
9466 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9467 &BTRFS_I(inode)->runtime_flags))
9468 filemap_flush(inode->i_mapping);
9471 complete(&delalloc_work->completion);
9474 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9476 struct btrfs_delalloc_work *work;
9478 work = kmalloc(sizeof(*work), GFP_NOFS);
9482 init_completion(&work->completion);
9483 INIT_LIST_HEAD(&work->list);
9484 work->inode = inode;
9485 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9491 * some fairly slow code that needs optimization. This walks the list
9492 * of all the inodes with pending delalloc and forces them to disk.
9494 static int start_delalloc_inodes(struct btrfs_root *root, int nr, bool snapshot)
9496 struct btrfs_inode *binode;
9497 struct inode *inode;
9498 struct btrfs_delalloc_work *work, *next;
9499 struct list_head works;
9500 struct list_head splice;
9503 INIT_LIST_HEAD(&works);
9504 INIT_LIST_HEAD(&splice);
9506 mutex_lock(&root->delalloc_mutex);
9507 spin_lock(&root->delalloc_lock);
9508 list_splice_init(&root->delalloc_inodes, &splice);
9509 while (!list_empty(&splice)) {
9510 binode = list_entry(splice.next, struct btrfs_inode,
9513 list_move_tail(&binode->delalloc_inodes,
9514 &root->delalloc_inodes);
9515 inode = igrab(&binode->vfs_inode);
9517 cond_resched_lock(&root->delalloc_lock);
9520 spin_unlock(&root->delalloc_lock);
9523 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9524 &binode->runtime_flags);
9525 work = btrfs_alloc_delalloc_work(inode);
9531 list_add_tail(&work->list, &works);
9532 btrfs_queue_work(root->fs_info->flush_workers,
9535 if (nr != -1 && ret >= nr)
9538 spin_lock(&root->delalloc_lock);
9540 spin_unlock(&root->delalloc_lock);
9543 list_for_each_entry_safe(work, next, &works, list) {
9544 list_del_init(&work->list);
9545 wait_for_completion(&work->completion);
9549 if (!list_empty(&splice)) {
9550 spin_lock(&root->delalloc_lock);
9551 list_splice_tail(&splice, &root->delalloc_inodes);
9552 spin_unlock(&root->delalloc_lock);
9554 mutex_unlock(&root->delalloc_mutex);
9558 int btrfs_start_delalloc_snapshot(struct btrfs_root *root)
9560 struct btrfs_fs_info *fs_info = root->fs_info;
9563 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9566 ret = start_delalloc_inodes(root, -1, true);
9572 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int nr)
9574 struct btrfs_root *root;
9575 struct list_head splice;
9578 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9581 INIT_LIST_HEAD(&splice);
9583 mutex_lock(&fs_info->delalloc_root_mutex);
9584 spin_lock(&fs_info->delalloc_root_lock);
9585 list_splice_init(&fs_info->delalloc_roots, &splice);
9586 while (!list_empty(&splice) && nr) {
9587 root = list_first_entry(&splice, struct btrfs_root,
9589 root = btrfs_grab_fs_root(root);
9591 list_move_tail(&root->delalloc_root,
9592 &fs_info->delalloc_roots);
9593 spin_unlock(&fs_info->delalloc_root_lock);
9595 ret = start_delalloc_inodes(root, nr, false);
9596 btrfs_put_fs_root(root);
9604 spin_lock(&fs_info->delalloc_root_lock);
9606 spin_unlock(&fs_info->delalloc_root_lock);
9610 if (!list_empty(&splice)) {
9611 spin_lock(&fs_info->delalloc_root_lock);
9612 list_splice_tail(&splice, &fs_info->delalloc_roots);
9613 spin_unlock(&fs_info->delalloc_root_lock);
9615 mutex_unlock(&fs_info->delalloc_root_mutex);
9619 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9620 const char *symname)
9622 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9623 struct btrfs_trans_handle *trans;
9624 struct btrfs_root *root = BTRFS_I(dir)->root;
9625 struct btrfs_path *path;
9626 struct btrfs_key key;
9627 struct inode *inode = NULL;
9634 struct btrfs_file_extent_item *ei;
9635 struct extent_buffer *leaf;
9637 name_len = strlen(symname);
9638 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9639 return -ENAMETOOLONG;
9642 * 2 items for inode item and ref
9643 * 2 items for dir items
9644 * 1 item for updating parent inode item
9645 * 1 item for the inline extent item
9646 * 1 item for xattr if selinux is on
9648 trans = btrfs_start_transaction(root, 7);
9650 return PTR_ERR(trans);
9652 err = btrfs_find_free_ino(root, &objectid);
9656 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9657 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
9658 objectid, S_IFLNK|S_IRWXUGO, &index);
9659 if (IS_ERR(inode)) {
9660 err = PTR_ERR(inode);
9666 * If the active LSM wants to access the inode during
9667 * d_instantiate it needs these. Smack checks to see
9668 * if the filesystem supports xattrs by looking at the
9671 inode->i_fop = &btrfs_file_operations;
9672 inode->i_op = &btrfs_file_inode_operations;
9673 inode->i_mapping->a_ops = &btrfs_aops;
9674 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9676 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9680 path = btrfs_alloc_path();
9685 key.objectid = btrfs_ino(BTRFS_I(inode));
9687 key.type = BTRFS_EXTENT_DATA_KEY;
9688 datasize = btrfs_file_extent_calc_inline_size(name_len);
9689 err = btrfs_insert_empty_item(trans, root, path, &key,
9692 btrfs_free_path(path);
9695 leaf = path->nodes[0];
9696 ei = btrfs_item_ptr(leaf, path->slots[0],
9697 struct btrfs_file_extent_item);
9698 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9699 btrfs_set_file_extent_type(leaf, ei,
9700 BTRFS_FILE_EXTENT_INLINE);
9701 btrfs_set_file_extent_encryption(leaf, ei, 0);
9702 btrfs_set_file_extent_compression(leaf, ei, 0);
9703 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9704 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9706 ptr = btrfs_file_extent_inline_start(ei);
9707 write_extent_buffer(leaf, symname, ptr, name_len);
9708 btrfs_mark_buffer_dirty(leaf);
9709 btrfs_free_path(path);
9711 inode->i_op = &btrfs_symlink_inode_operations;
9712 inode_nohighmem(inode);
9713 inode_set_bytes(inode, name_len);
9714 btrfs_i_size_write(BTRFS_I(inode), name_len);
9715 err = btrfs_update_inode(trans, root, inode);
9717 * Last step, add directory indexes for our symlink inode. This is the
9718 * last step to avoid extra cleanup of these indexes if an error happens
9722 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9723 BTRFS_I(inode), 0, index);
9727 d_instantiate_new(dentry, inode);
9730 btrfs_end_transaction(trans);
9732 inode_dec_link_count(inode);
9733 discard_new_inode(inode);
9735 btrfs_btree_balance_dirty(fs_info);
9739 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9740 u64 start, u64 num_bytes, u64 min_size,
9741 loff_t actual_len, u64 *alloc_hint,
9742 struct btrfs_trans_handle *trans)
9744 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9745 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9746 struct extent_map *em;
9747 struct btrfs_root *root = BTRFS_I(inode)->root;
9748 struct btrfs_key ins;
9749 u64 cur_offset = start;
9752 u64 last_alloc = (u64)-1;
9754 bool own_trans = true;
9755 u64 end = start + num_bytes - 1;
9759 while (num_bytes > 0) {
9761 trans = btrfs_start_transaction(root, 3);
9762 if (IS_ERR(trans)) {
9763 ret = PTR_ERR(trans);
9768 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9769 cur_bytes = max(cur_bytes, min_size);
9771 * If we are severely fragmented we could end up with really
9772 * small allocations, so if the allocator is returning small
9773 * chunks lets make its job easier by only searching for those
9776 cur_bytes = min(cur_bytes, last_alloc);
9777 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9778 min_size, 0, *alloc_hint, &ins, 1, 0);
9781 btrfs_end_transaction(trans);
9784 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
9786 last_alloc = ins.offset;
9787 ret = insert_reserved_file_extent(trans, inode,
9788 cur_offset, ins.objectid,
9789 ins.offset, ins.offset,
9790 ins.offset, 0, 0, 0,
9791 BTRFS_FILE_EXTENT_PREALLOC);
9793 btrfs_free_reserved_extent(fs_info, ins.objectid,
9795 btrfs_abort_transaction(trans, ret);
9797 btrfs_end_transaction(trans);
9801 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
9802 cur_offset + ins.offset -1, 0);
9804 em = alloc_extent_map();
9806 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9807 &BTRFS_I(inode)->runtime_flags);
9811 em->start = cur_offset;
9812 em->orig_start = cur_offset;
9813 em->len = ins.offset;
9814 em->block_start = ins.objectid;
9815 em->block_len = ins.offset;
9816 em->orig_block_len = ins.offset;
9817 em->ram_bytes = ins.offset;
9818 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9819 em->generation = trans->transid;
9822 write_lock(&em_tree->lock);
9823 ret = add_extent_mapping(em_tree, em, 1);
9824 write_unlock(&em_tree->lock);
9827 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
9828 cur_offset + ins.offset - 1,
9831 free_extent_map(em);
9833 num_bytes -= ins.offset;
9834 cur_offset += ins.offset;
9835 *alloc_hint = ins.objectid + ins.offset;
9837 inode_inc_iversion(inode);
9838 inode->i_ctime = current_time(inode);
9839 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9840 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9841 (actual_len > inode->i_size) &&
9842 (cur_offset > inode->i_size)) {
9843 if (cur_offset > actual_len)
9844 i_size = actual_len;
9846 i_size = cur_offset;
9847 i_size_write(inode, i_size);
9848 btrfs_ordered_update_i_size(inode, i_size, NULL);
9851 ret = btrfs_update_inode(trans, root, inode);
9854 btrfs_abort_transaction(trans, ret);
9856 btrfs_end_transaction(trans);
9861 btrfs_end_transaction(trans);
9863 if (cur_offset < end)
9864 btrfs_free_reserved_data_space(inode, NULL, cur_offset,
9865 end - cur_offset + 1);
9869 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9870 u64 start, u64 num_bytes, u64 min_size,
9871 loff_t actual_len, u64 *alloc_hint)
9873 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9874 min_size, actual_len, alloc_hint,
9878 int btrfs_prealloc_file_range_trans(struct inode *inode,
9879 struct btrfs_trans_handle *trans, int mode,
9880 u64 start, u64 num_bytes, u64 min_size,
9881 loff_t actual_len, u64 *alloc_hint)
9883 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9884 min_size, actual_len, alloc_hint, trans);
9887 static int btrfs_set_page_dirty(struct page *page)
9889 return __set_page_dirty_nobuffers(page);
9892 static int btrfs_permission(struct inode *inode, int mask)
9894 struct btrfs_root *root = BTRFS_I(inode)->root;
9895 umode_t mode = inode->i_mode;
9897 if (mask & MAY_WRITE &&
9898 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9899 if (btrfs_root_readonly(root))
9901 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9904 return generic_permission(inode, mask);
9907 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
9909 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9910 struct btrfs_trans_handle *trans;
9911 struct btrfs_root *root = BTRFS_I(dir)->root;
9912 struct inode *inode = NULL;
9918 * 5 units required for adding orphan entry
9920 trans = btrfs_start_transaction(root, 5);
9922 return PTR_ERR(trans);
9924 ret = btrfs_find_free_ino(root, &objectid);
9928 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
9929 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
9930 if (IS_ERR(inode)) {
9931 ret = PTR_ERR(inode);
9936 inode->i_fop = &btrfs_file_operations;
9937 inode->i_op = &btrfs_file_inode_operations;
9939 inode->i_mapping->a_ops = &btrfs_aops;
9940 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9942 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
9946 ret = btrfs_update_inode(trans, root, inode);
9949 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
9954 * We set number of links to 0 in btrfs_new_inode(), and here we set
9955 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9958 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9960 set_nlink(inode, 1);
9961 d_tmpfile(dentry, inode);
9962 unlock_new_inode(inode);
9963 mark_inode_dirty(inode);
9965 btrfs_end_transaction(trans);
9967 discard_new_inode(inode);
9968 btrfs_btree_balance_dirty(fs_info);
9972 void btrfs_set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
9974 struct inode *inode = tree->private_data;
9975 unsigned long index = start >> PAGE_SHIFT;
9976 unsigned long end_index = end >> PAGE_SHIFT;
9979 while (index <= end_index) {
9980 page = find_get_page(inode->i_mapping, index);
9981 ASSERT(page); /* Pages should be in the extent_io_tree */
9982 set_page_writeback(page);
9990 * Add an entry indicating a block group or device which is pinned by a
9991 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
9992 * negative errno on failure.
9994 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
9995 bool is_block_group)
9997 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
9998 struct btrfs_swapfile_pin *sp, *entry;
10000 struct rb_node *parent = NULL;
10002 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10007 sp->is_block_group = is_block_group;
10009 spin_lock(&fs_info->swapfile_pins_lock);
10010 p = &fs_info->swapfile_pins.rb_node;
10013 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10014 if (sp->ptr < entry->ptr ||
10015 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10016 p = &(*p)->rb_left;
10017 } else if (sp->ptr > entry->ptr ||
10018 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10019 p = &(*p)->rb_right;
10021 spin_unlock(&fs_info->swapfile_pins_lock);
10026 rb_link_node(&sp->node, parent, p);
10027 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10028 spin_unlock(&fs_info->swapfile_pins_lock);
10032 /* Free all of the entries pinned by this swapfile. */
10033 static void btrfs_free_swapfile_pins(struct inode *inode)
10035 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10036 struct btrfs_swapfile_pin *sp;
10037 struct rb_node *node, *next;
10039 spin_lock(&fs_info->swapfile_pins_lock);
10040 node = rb_first(&fs_info->swapfile_pins);
10042 next = rb_next(node);
10043 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10044 if (sp->inode == inode) {
10045 rb_erase(&sp->node, &fs_info->swapfile_pins);
10046 if (sp->is_block_group)
10047 btrfs_put_block_group(sp->ptr);
10052 spin_unlock(&fs_info->swapfile_pins_lock);
10055 struct btrfs_swap_info {
10061 unsigned long nr_pages;
10065 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10066 struct btrfs_swap_info *bsi)
10068 unsigned long nr_pages;
10069 u64 first_ppage, first_ppage_reported, next_ppage;
10072 first_ppage = ALIGN(bsi->block_start, PAGE_SIZE) >> PAGE_SHIFT;
10073 next_ppage = ALIGN_DOWN(bsi->block_start + bsi->block_len,
10074 PAGE_SIZE) >> PAGE_SHIFT;
10076 if (first_ppage >= next_ppage)
10078 nr_pages = next_ppage - first_ppage;
10080 first_ppage_reported = first_ppage;
10081 if (bsi->start == 0)
10082 first_ppage_reported++;
10083 if (bsi->lowest_ppage > first_ppage_reported)
10084 bsi->lowest_ppage = first_ppage_reported;
10085 if (bsi->highest_ppage < (next_ppage - 1))
10086 bsi->highest_ppage = next_ppage - 1;
10088 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10091 bsi->nr_extents += ret;
10092 bsi->nr_pages += nr_pages;
10096 static void btrfs_swap_deactivate(struct file *file)
10098 struct inode *inode = file_inode(file);
10100 btrfs_free_swapfile_pins(inode);
10101 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10104 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10107 struct inode *inode = file_inode(file);
10108 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10109 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10110 struct extent_state *cached_state = NULL;
10111 struct extent_map *em = NULL;
10112 struct btrfs_device *device = NULL;
10113 struct btrfs_swap_info bsi = {
10114 .lowest_ppage = (sector_t)-1ULL,
10121 * If the swap file was just created, make sure delalloc is done. If the
10122 * file changes again after this, the user is doing something stupid and
10123 * we don't really care.
10125 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10130 * The inode is locked, so these flags won't change after we check them.
10132 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10133 btrfs_warn(fs_info, "swapfile must not be compressed");
10136 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10137 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10140 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10141 btrfs_warn(fs_info, "swapfile must not be checksummed");
10146 * Balance or device remove/replace/resize can move stuff around from
10147 * under us. The EXCL_OP flag makes sure they aren't running/won't run
10148 * concurrently while we are mapping the swap extents, and
10149 * fs_info->swapfile_pins prevents them from running while the swap file
10150 * is active and moving the extents. Note that this also prevents a
10151 * concurrent device add which isn't actually necessary, but it's not
10152 * really worth the trouble to allow it.
10154 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
10155 btrfs_warn(fs_info,
10156 "cannot activate swapfile while exclusive operation is running");
10160 * Snapshots can create extents which require COW even if NODATACOW is
10161 * set. We use this counter to prevent snapshots. We must increment it
10162 * before walking the extents because we don't want a concurrent
10163 * snapshot to run after we've already checked the extents.
10165 atomic_inc(&BTRFS_I(inode)->root->nr_swapfiles);
10167 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10169 lock_extent_bits(io_tree, 0, isize - 1, &cached_state);
10171 while (start < isize) {
10172 u64 logical_block_start, physical_block_start;
10173 struct btrfs_block_group *bg;
10174 u64 len = isize - start;
10176 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
10182 if (em->block_start == EXTENT_MAP_HOLE) {
10183 btrfs_warn(fs_info, "swapfile must not have holes");
10187 if (em->block_start == EXTENT_MAP_INLINE) {
10189 * It's unlikely we'll ever actually find ourselves
10190 * here, as a file small enough to fit inline won't be
10191 * big enough to store more than the swap header, but in
10192 * case something changes in the future, let's catch it
10193 * here rather than later.
10195 btrfs_warn(fs_info, "swapfile must not be inline");
10199 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10200 btrfs_warn(fs_info, "swapfile must not be compressed");
10205 logical_block_start = em->block_start + (start - em->start);
10206 len = min(len, em->len - (start - em->start));
10207 free_extent_map(em);
10210 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL);
10216 btrfs_warn(fs_info,
10217 "swapfile must not be copy-on-write");
10222 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10228 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10229 btrfs_warn(fs_info,
10230 "swapfile must have single data profile");
10235 if (device == NULL) {
10236 device = em->map_lookup->stripes[0].dev;
10237 ret = btrfs_add_swapfile_pin(inode, device, false);
10242 } else if (device != em->map_lookup->stripes[0].dev) {
10243 btrfs_warn(fs_info, "swapfile must be on one device");
10248 physical_block_start = (em->map_lookup->stripes[0].physical +
10249 (logical_block_start - em->start));
10250 len = min(len, em->len - (logical_block_start - em->start));
10251 free_extent_map(em);
10254 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10256 btrfs_warn(fs_info,
10257 "could not find block group containing swapfile");
10262 ret = btrfs_add_swapfile_pin(inode, bg, true);
10264 btrfs_put_block_group(bg);
10271 if (bsi.block_len &&
10272 bsi.block_start + bsi.block_len == physical_block_start) {
10273 bsi.block_len += len;
10275 if (bsi.block_len) {
10276 ret = btrfs_add_swap_extent(sis, &bsi);
10281 bsi.block_start = physical_block_start;
10282 bsi.block_len = len;
10289 ret = btrfs_add_swap_extent(sis, &bsi);
10292 if (!IS_ERR_OR_NULL(em))
10293 free_extent_map(em);
10295 unlock_extent_cached(io_tree, 0, isize - 1, &cached_state);
10298 btrfs_swap_deactivate(file);
10300 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
10306 sis->bdev = device->bdev;
10307 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10308 sis->max = bsi.nr_pages;
10309 sis->pages = bsi.nr_pages - 1;
10310 sis->highest_bit = bsi.nr_pages - 1;
10311 return bsi.nr_extents;
10314 static void btrfs_swap_deactivate(struct file *file)
10318 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10321 return -EOPNOTSUPP;
10325 static const struct inode_operations btrfs_dir_inode_operations = {
10326 .getattr = btrfs_getattr,
10327 .lookup = btrfs_lookup,
10328 .create = btrfs_create,
10329 .unlink = btrfs_unlink,
10330 .link = btrfs_link,
10331 .mkdir = btrfs_mkdir,
10332 .rmdir = btrfs_rmdir,
10333 .rename = btrfs_rename2,
10334 .symlink = btrfs_symlink,
10335 .setattr = btrfs_setattr,
10336 .mknod = btrfs_mknod,
10337 .listxattr = btrfs_listxattr,
10338 .permission = btrfs_permission,
10339 .get_acl = btrfs_get_acl,
10340 .set_acl = btrfs_set_acl,
10341 .update_time = btrfs_update_time,
10342 .tmpfile = btrfs_tmpfile,
10345 static const struct file_operations btrfs_dir_file_operations = {
10346 .llseek = generic_file_llseek,
10347 .read = generic_read_dir,
10348 .iterate_shared = btrfs_real_readdir,
10349 .open = btrfs_opendir,
10350 .unlocked_ioctl = btrfs_ioctl,
10351 #ifdef CONFIG_COMPAT
10352 .compat_ioctl = btrfs_compat_ioctl,
10354 .release = btrfs_release_file,
10355 .fsync = btrfs_sync_file,
10358 static const struct extent_io_ops btrfs_extent_io_ops = {
10359 /* mandatory callbacks */
10360 .submit_bio_hook = btrfs_submit_bio_hook,
10361 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10365 * btrfs doesn't support the bmap operation because swapfiles
10366 * use bmap to make a mapping of extents in the file. They assume
10367 * these extents won't change over the life of the file and they
10368 * use the bmap result to do IO directly to the drive.
10370 * the btrfs bmap call would return logical addresses that aren't
10371 * suitable for IO and they also will change frequently as COW
10372 * operations happen. So, swapfile + btrfs == corruption.
10374 * For now we're avoiding this by dropping bmap.
10376 static const struct address_space_operations btrfs_aops = {
10377 .readpage = btrfs_readpage,
10378 .writepage = btrfs_writepage,
10379 .writepages = btrfs_writepages,
10380 .readpages = btrfs_readpages,
10381 .direct_IO = btrfs_direct_IO,
10382 .invalidatepage = btrfs_invalidatepage,
10383 .releasepage = btrfs_releasepage,
10384 .set_page_dirty = btrfs_set_page_dirty,
10385 .error_remove_page = generic_error_remove_page,
10386 .swap_activate = btrfs_swap_activate,
10387 .swap_deactivate = btrfs_swap_deactivate,
10390 static const struct inode_operations btrfs_file_inode_operations = {
10391 .getattr = btrfs_getattr,
10392 .setattr = btrfs_setattr,
10393 .listxattr = btrfs_listxattr,
10394 .permission = btrfs_permission,
10395 .fiemap = btrfs_fiemap,
10396 .get_acl = btrfs_get_acl,
10397 .set_acl = btrfs_set_acl,
10398 .update_time = btrfs_update_time,
10400 static const struct inode_operations btrfs_special_inode_operations = {
10401 .getattr = btrfs_getattr,
10402 .setattr = btrfs_setattr,
10403 .permission = btrfs_permission,
10404 .listxattr = btrfs_listxattr,
10405 .get_acl = btrfs_get_acl,
10406 .set_acl = btrfs_set_acl,
10407 .update_time = btrfs_update_time,
10409 static const struct inode_operations btrfs_symlink_inode_operations = {
10410 .get_link = page_get_link,
10411 .getattr = btrfs_getattr,
10412 .setattr = btrfs_setattr,
10413 .permission = btrfs_permission,
10414 .listxattr = btrfs_listxattr,
10415 .update_time = btrfs_update_time,
10418 const struct dentry_operations btrfs_dentry_operations = {
10419 .d_delete = btrfs_dentry_delete,