1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <crypto/hash.h>
7 #include <linux/kernel.h>
9 #include <linux/blk-cgroup.h>
10 #include <linux/file.h>
12 #include <linux/pagemap.h>
13 #include <linux/highmem.h>
14 #include <linux/time.h>
15 #include <linux/init.h>
16 #include <linux/string.h>
17 #include <linux/backing-dev.h>
18 #include <linux/writeback.h>
19 #include <linux/compat.h>
20 #include <linux/xattr.h>
21 #include <linux/posix_acl.h>
22 #include <linux/falloc.h>
23 #include <linux/slab.h>
24 #include <linux/ratelimit.h>
25 #include <linux/btrfs.h>
26 #include <linux/blkdev.h>
27 #include <linux/posix_acl_xattr.h>
28 #include <linux/uio.h>
29 #include <linux/magic.h>
30 #include <linux/iversion.h>
31 #include <linux/swap.h>
32 #include <linux/migrate.h>
33 #include <linux/sched/mm.h>
34 #include <linux/iomap.h>
35 #include <asm/unaligned.h>
36 #include <linux/fsverity.h>
40 #include "transaction.h"
41 #include "btrfs_inode.h"
42 #include "ordered-data.h"
46 #include "compression.h"
50 #include "delalloc-space.h"
51 #include "block-group.h"
52 #include "space-info.h"
55 #include "inode-item.h"
57 #include "accessors.h"
58 #include "extent-tree.h"
59 #include "root-tree.h"
62 #include "file-item.h"
63 #include "uuid-tree.h"
67 #include "relocation.h"
72 #include "raid-stripe-tree.h"
74 struct btrfs_iget_args {
76 struct btrfs_root *root;
79 struct btrfs_dio_data {
81 struct extent_changeset *data_reserved;
82 struct btrfs_ordered_extent *ordered;
83 bool data_space_reserved;
87 struct btrfs_dio_private {
92 /* This must be last */
93 struct btrfs_bio bbio;
96 static struct bio_set btrfs_dio_bioset;
98 struct btrfs_rename_ctx {
99 /* Output field. Stores the index number of the old directory entry. */
104 * Used by data_reloc_print_warning_inode() to pass needed info for filename
105 * resolution and output of error message.
107 struct data_reloc_warn {
108 struct btrfs_path path;
109 struct btrfs_fs_info *fs_info;
110 u64 extent_item_size;
116 * For the file_extent_tree, we want to hold the inode lock when we lookup and
117 * update the disk_i_size, but lockdep will complain because our io_tree we hold
118 * the tree lock and get the inode lock when setting delalloc. These two things
119 * are unrelated, so make a class for the file_extent_tree so we don't get the
120 * two locking patterns mixed up.
122 static struct lock_class_key file_extent_tree_class;
124 static const struct inode_operations btrfs_dir_inode_operations;
125 static const struct inode_operations btrfs_symlink_inode_operations;
126 static const struct inode_operations btrfs_special_inode_operations;
127 static const struct inode_operations btrfs_file_inode_operations;
128 static const struct address_space_operations btrfs_aops;
129 static const struct file_operations btrfs_dir_file_operations;
131 static struct kmem_cache *btrfs_inode_cachep;
133 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
134 static int btrfs_truncate(struct btrfs_inode *inode, bool skip_writeback);
136 static noinline int run_delalloc_cow(struct btrfs_inode *inode,
137 struct page *locked_page, u64 start,
138 u64 end, struct writeback_control *wbc,
140 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
141 u64 len, u64 orig_start, u64 block_start,
142 u64 block_len, u64 orig_block_len,
143 u64 ram_bytes, int compress_type,
146 static int data_reloc_print_warning_inode(u64 inum, u64 offset, u64 num_bytes,
147 u64 root, void *warn_ctx)
149 struct data_reloc_warn *warn = warn_ctx;
150 struct btrfs_fs_info *fs_info = warn->fs_info;
151 struct extent_buffer *eb;
152 struct btrfs_inode_item *inode_item;
153 struct inode_fs_paths *ipath = NULL;
154 struct btrfs_root *local_root;
155 struct btrfs_key key;
156 unsigned int nofs_flag;
160 local_root = btrfs_get_fs_root(fs_info, root, true);
161 if (IS_ERR(local_root)) {
162 ret = PTR_ERR(local_root);
166 /* This makes the path point to (inum INODE_ITEM ioff). */
168 key.type = BTRFS_INODE_ITEM_KEY;
171 ret = btrfs_search_slot(NULL, local_root, &key, &warn->path, 0, 0);
173 btrfs_put_root(local_root);
174 btrfs_release_path(&warn->path);
178 eb = warn->path.nodes[0];
179 inode_item = btrfs_item_ptr(eb, warn->path.slots[0], struct btrfs_inode_item);
180 nlink = btrfs_inode_nlink(eb, inode_item);
181 btrfs_release_path(&warn->path);
183 nofs_flag = memalloc_nofs_save();
184 ipath = init_ipath(4096, local_root, &warn->path);
185 memalloc_nofs_restore(nofs_flag);
187 btrfs_put_root(local_root);
188 ret = PTR_ERR(ipath);
191 * -ENOMEM, not a critical error, just output an generic error
195 "checksum error at logical %llu mirror %u root %llu, inode %llu offset %llu",
196 warn->logical, warn->mirror_num, root, inum, offset);
199 ret = paths_from_inode(inum, ipath);
204 * We deliberately ignore the bit ipath might have been too small to
205 * hold all of the paths here
207 for (int i = 0; i < ipath->fspath->elem_cnt; i++) {
209 "checksum error at logical %llu mirror %u root %llu inode %llu offset %llu length %u links %u (path: %s)",
210 warn->logical, warn->mirror_num, root, inum, offset,
211 fs_info->sectorsize, nlink,
212 (char *)(unsigned long)ipath->fspath->val[i]);
215 btrfs_put_root(local_root);
221 "checksum error at logical %llu mirror %u root %llu inode %llu offset %llu, path resolving failed with ret=%d",
222 warn->logical, warn->mirror_num, root, inum, offset, ret);
229 * Do extra user-friendly error output (e.g. lookup all the affected files).
231 * Return true if we succeeded doing the backref lookup.
232 * Return false if such lookup failed, and has to fallback to the old error message.
234 static void print_data_reloc_error(const struct btrfs_inode *inode, u64 file_off,
235 const u8 *csum, const u8 *csum_expected,
238 struct btrfs_fs_info *fs_info = inode->root->fs_info;
239 struct btrfs_path path = { 0 };
240 struct btrfs_key found_key = { 0 };
241 struct extent_buffer *eb;
242 struct btrfs_extent_item *ei;
243 const u32 csum_size = fs_info->csum_size;
249 mutex_lock(&fs_info->reloc_mutex);
250 logical = btrfs_get_reloc_bg_bytenr(fs_info);
251 mutex_unlock(&fs_info->reloc_mutex);
253 if (logical == U64_MAX) {
254 btrfs_warn_rl(fs_info, "has data reloc tree but no running relocation");
255 btrfs_warn_rl(fs_info,
256 "csum failed root %lld ino %llu off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
257 inode->root->root_key.objectid, btrfs_ino(inode), file_off,
258 CSUM_FMT_VALUE(csum_size, csum),
259 CSUM_FMT_VALUE(csum_size, csum_expected),
265 btrfs_warn_rl(fs_info,
266 "csum failed root %lld ino %llu off %llu logical %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
267 inode->root->root_key.objectid,
268 btrfs_ino(inode), file_off, logical,
269 CSUM_FMT_VALUE(csum_size, csum),
270 CSUM_FMT_VALUE(csum_size, csum_expected),
273 ret = extent_from_logical(fs_info, logical, &path, &found_key, &flags);
275 btrfs_err_rl(fs_info, "failed to lookup extent item for logical %llu: %d",
280 ei = btrfs_item_ptr(eb, path.slots[0], struct btrfs_extent_item);
281 item_size = btrfs_item_size(eb, path.slots[0]);
282 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
283 unsigned long ptr = 0;
288 ret = tree_backref_for_extent(&ptr, eb, &found_key, ei,
289 item_size, &ref_root,
292 btrfs_warn_rl(fs_info,
293 "failed to resolve tree backref for logical %llu: %d",
300 btrfs_warn_rl(fs_info,
301 "csum error at logical %llu mirror %u: metadata %s (level %d) in tree %llu",
303 (ref_level ? "node" : "leaf"),
304 ref_level, ref_root);
306 btrfs_release_path(&path);
308 struct btrfs_backref_walk_ctx ctx = { 0 };
309 struct data_reloc_warn reloc_warn = { 0 };
311 btrfs_release_path(&path);
313 ctx.bytenr = found_key.objectid;
314 ctx.extent_item_pos = logical - found_key.objectid;
315 ctx.fs_info = fs_info;
317 reloc_warn.logical = logical;
318 reloc_warn.extent_item_size = found_key.offset;
319 reloc_warn.mirror_num = mirror_num;
320 reloc_warn.fs_info = fs_info;
322 iterate_extent_inodes(&ctx, true,
323 data_reloc_print_warning_inode, &reloc_warn);
327 static void __cold btrfs_print_data_csum_error(struct btrfs_inode *inode,
328 u64 logical_start, u8 *csum, u8 *csum_expected, int mirror_num)
330 struct btrfs_root *root = inode->root;
331 const u32 csum_size = root->fs_info->csum_size;
333 /* For data reloc tree, it's better to do a backref lookup instead. */
334 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
335 return print_data_reloc_error(inode, logical_start, csum,
336 csum_expected, mirror_num);
338 /* Output without objectid, which is more meaningful */
339 if (root->root_key.objectid >= BTRFS_LAST_FREE_OBJECTID) {
340 btrfs_warn_rl(root->fs_info,
341 "csum failed root %lld ino %lld off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
342 root->root_key.objectid, btrfs_ino(inode),
344 CSUM_FMT_VALUE(csum_size, csum),
345 CSUM_FMT_VALUE(csum_size, csum_expected),
348 btrfs_warn_rl(root->fs_info,
349 "csum failed root %llu ino %llu off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
350 root->root_key.objectid, btrfs_ino(inode),
352 CSUM_FMT_VALUE(csum_size, csum),
353 CSUM_FMT_VALUE(csum_size, csum_expected),
359 * Lock inode i_rwsem based on arguments passed.
361 * ilock_flags can have the following bit set:
363 * BTRFS_ILOCK_SHARED - acquire a shared lock on the inode
364 * BTRFS_ILOCK_TRY - try to acquire the lock, if fails on first attempt
366 * BTRFS_ILOCK_MMAP - acquire a write lock on the i_mmap_lock
368 int btrfs_inode_lock(struct btrfs_inode *inode, unsigned int ilock_flags)
370 if (ilock_flags & BTRFS_ILOCK_SHARED) {
371 if (ilock_flags & BTRFS_ILOCK_TRY) {
372 if (!inode_trylock_shared(&inode->vfs_inode))
377 inode_lock_shared(&inode->vfs_inode);
379 if (ilock_flags & BTRFS_ILOCK_TRY) {
380 if (!inode_trylock(&inode->vfs_inode))
385 inode_lock(&inode->vfs_inode);
387 if (ilock_flags & BTRFS_ILOCK_MMAP)
388 down_write(&inode->i_mmap_lock);
393 * Unock inode i_rwsem.
395 * ilock_flags should contain the same bits set as passed to btrfs_inode_lock()
396 * to decide whether the lock acquired is shared or exclusive.
398 void btrfs_inode_unlock(struct btrfs_inode *inode, unsigned int ilock_flags)
400 if (ilock_flags & BTRFS_ILOCK_MMAP)
401 up_write(&inode->i_mmap_lock);
402 if (ilock_flags & BTRFS_ILOCK_SHARED)
403 inode_unlock_shared(&inode->vfs_inode);
405 inode_unlock(&inode->vfs_inode);
409 * Cleanup all submitted ordered extents in specified range to handle errors
410 * from the btrfs_run_delalloc_range() callback.
412 * NOTE: caller must ensure that when an error happens, it can not call
413 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
414 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
415 * to be released, which we want to happen only when finishing the ordered
416 * extent (btrfs_finish_ordered_io()).
418 static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode,
419 struct page *locked_page,
420 u64 offset, u64 bytes)
422 unsigned long index = offset >> PAGE_SHIFT;
423 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
424 u64 page_start = 0, page_end = 0;
428 page_start = page_offset(locked_page);
429 page_end = page_start + PAGE_SIZE - 1;
432 while (index <= end_index) {
434 * For locked page, we will call btrfs_mark_ordered_io_finished
435 * through btrfs_mark_ordered_io_finished() on it
436 * in run_delalloc_range() for the error handling, which will
437 * clear page Ordered and run the ordered extent accounting.
439 * Here we can't just clear the Ordered bit, or
440 * btrfs_mark_ordered_io_finished() would skip the accounting
441 * for the page range, and the ordered extent will never finish.
443 if (locked_page && index == (page_start >> PAGE_SHIFT)) {
447 page = find_get_page(inode->vfs_inode.i_mapping, index);
453 * Here we just clear all Ordered bits for every page in the
454 * range, then btrfs_mark_ordered_io_finished() will handle
455 * the ordered extent accounting for the range.
457 btrfs_folio_clamp_clear_ordered(inode->root->fs_info,
458 page_folio(page), offset, bytes);
463 /* The locked page covers the full range, nothing needs to be done */
464 if (bytes + offset <= page_start + PAGE_SIZE)
467 * In case this page belongs to the delalloc range being
468 * instantiated then skip it, since the first page of a range is
469 * going to be properly cleaned up by the caller of
472 if (page_start >= offset && page_end <= (offset + bytes - 1)) {
473 bytes = offset + bytes - page_offset(locked_page) - PAGE_SIZE;
474 offset = page_offset(locked_page) + PAGE_SIZE;
478 return btrfs_mark_ordered_io_finished(inode, NULL, offset, bytes, false);
481 static int btrfs_dirty_inode(struct btrfs_inode *inode);
483 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
484 struct btrfs_new_inode_args *args)
488 if (args->default_acl) {
489 err = __btrfs_set_acl(trans, args->inode, args->default_acl,
495 err = __btrfs_set_acl(trans, args->inode, args->acl, ACL_TYPE_ACCESS);
499 if (!args->default_acl && !args->acl)
500 cache_no_acl(args->inode);
501 return btrfs_xattr_security_init(trans, args->inode, args->dir,
502 &args->dentry->d_name);
506 * this does all the hard work for inserting an inline extent into
507 * the btree. The caller should have done a btrfs_drop_extents so that
508 * no overlapping inline items exist in the btree
510 static int insert_inline_extent(struct btrfs_trans_handle *trans,
511 struct btrfs_path *path,
512 struct btrfs_inode *inode, bool extent_inserted,
513 size_t size, size_t compressed_size,
515 struct page **compressed_pages,
518 struct btrfs_root *root = inode->root;
519 struct extent_buffer *leaf;
520 struct page *page = NULL;
523 struct btrfs_file_extent_item *ei;
525 size_t cur_size = size;
528 ASSERT((compressed_size > 0 && compressed_pages) ||
529 (compressed_size == 0 && !compressed_pages));
531 if (compressed_size && compressed_pages)
532 cur_size = compressed_size;
534 if (!extent_inserted) {
535 struct btrfs_key key;
538 key.objectid = btrfs_ino(inode);
540 key.type = BTRFS_EXTENT_DATA_KEY;
542 datasize = btrfs_file_extent_calc_inline_size(cur_size);
543 ret = btrfs_insert_empty_item(trans, root, path, &key,
548 leaf = path->nodes[0];
549 ei = btrfs_item_ptr(leaf, path->slots[0],
550 struct btrfs_file_extent_item);
551 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
552 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
553 btrfs_set_file_extent_encryption(leaf, ei, 0);
554 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
555 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
556 ptr = btrfs_file_extent_inline_start(ei);
558 if (compress_type != BTRFS_COMPRESS_NONE) {
561 while (compressed_size > 0) {
562 cpage = compressed_pages[i];
563 cur_size = min_t(unsigned long, compressed_size,
566 kaddr = kmap_local_page(cpage);
567 write_extent_buffer(leaf, kaddr, ptr, cur_size);
572 compressed_size -= cur_size;
574 btrfs_set_file_extent_compression(leaf, ei,
577 page = find_get_page(inode->vfs_inode.i_mapping, 0);
578 btrfs_set_file_extent_compression(leaf, ei, 0);
579 kaddr = kmap_local_page(page);
580 write_extent_buffer(leaf, kaddr, ptr, size);
584 btrfs_mark_buffer_dirty(trans, leaf);
585 btrfs_release_path(path);
588 * We align size to sectorsize for inline extents just for simplicity
591 ret = btrfs_inode_set_file_extent_range(inode, 0,
592 ALIGN(size, root->fs_info->sectorsize));
597 * We're an inline extent, so nobody can extend the file past i_size
598 * without locking a page we already have locked.
600 * We must do any i_size and inode updates before we unlock the pages.
601 * Otherwise we could end up racing with unlink.
603 i_size = i_size_read(&inode->vfs_inode);
604 if (update_i_size && size > i_size) {
605 i_size_write(&inode->vfs_inode, size);
608 inode->disk_i_size = i_size;
616 * conditionally insert an inline extent into the file. This
617 * does the checks required to make sure the data is small enough
618 * to fit as an inline extent.
620 static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 size,
621 size_t compressed_size,
623 struct page **compressed_pages,
626 struct btrfs_drop_extents_args drop_args = { 0 };
627 struct btrfs_root *root = inode->root;
628 struct btrfs_fs_info *fs_info = root->fs_info;
629 struct btrfs_trans_handle *trans;
630 u64 data_len = (compressed_size ?: size);
632 struct btrfs_path *path;
635 * We can create an inline extent if it ends at or beyond the current
636 * i_size, is no larger than a sector (decompressed), and the (possibly
637 * compressed) data fits in a leaf and the configured maximum inline
640 if (size < i_size_read(&inode->vfs_inode) ||
641 size > fs_info->sectorsize ||
642 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
643 data_len > fs_info->max_inline)
646 path = btrfs_alloc_path();
650 trans = btrfs_join_transaction(root);
652 btrfs_free_path(path);
653 return PTR_ERR(trans);
655 trans->block_rsv = &inode->block_rsv;
657 drop_args.path = path;
659 drop_args.end = fs_info->sectorsize;
660 drop_args.drop_cache = true;
661 drop_args.replace_extent = true;
662 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(data_len);
663 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
665 btrfs_abort_transaction(trans, ret);
669 ret = insert_inline_extent(trans, path, inode, drop_args.extent_inserted,
670 size, compressed_size, compress_type,
671 compressed_pages, update_i_size);
672 if (ret && ret != -ENOSPC) {
673 btrfs_abort_transaction(trans, ret);
675 } else if (ret == -ENOSPC) {
680 btrfs_update_inode_bytes(inode, size, drop_args.bytes_found);
681 ret = btrfs_update_inode(trans, inode);
682 if (ret && ret != -ENOSPC) {
683 btrfs_abort_transaction(trans, ret);
685 } else if (ret == -ENOSPC) {
690 btrfs_set_inode_full_sync(inode);
693 * Don't forget to free the reserved space, as for inlined extent
694 * it won't count as data extent, free them directly here.
695 * And at reserve time, it's always aligned to page size, so
696 * just free one page here.
698 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE, NULL);
699 btrfs_free_path(path);
700 btrfs_end_transaction(trans);
704 struct async_extent {
709 unsigned long nr_pages;
711 struct list_head list;
715 struct btrfs_inode *inode;
716 struct page *locked_page;
719 blk_opf_t write_flags;
720 struct list_head extents;
721 struct cgroup_subsys_state *blkcg_css;
722 struct btrfs_work work;
723 struct async_cow *async_cow;
728 struct async_chunk chunks[];
731 static noinline int add_async_extent(struct async_chunk *cow,
732 u64 start, u64 ram_size,
735 unsigned long nr_pages,
738 struct async_extent *async_extent;
740 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
743 async_extent->start = start;
744 async_extent->ram_size = ram_size;
745 async_extent->compressed_size = compressed_size;
746 async_extent->pages = pages;
747 async_extent->nr_pages = nr_pages;
748 async_extent->compress_type = compress_type;
749 list_add_tail(&async_extent->list, &cow->extents);
754 * Check if the inode needs to be submitted to compression, based on mount
755 * options, defragmentation, properties or heuristics.
757 static inline int inode_need_compress(struct btrfs_inode *inode, u64 start,
760 struct btrfs_fs_info *fs_info = inode->root->fs_info;
762 if (!btrfs_inode_can_compress(inode)) {
763 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
764 KERN_ERR "BTRFS: unexpected compression for ino %llu\n",
769 * Special check for subpage.
771 * We lock the full page then run each delalloc range in the page, thus
772 * for the following case, we will hit some subpage specific corner case:
775 * | |///////| |///////|
778 * In above case, both range A and range B will try to unlock the full
779 * page [0, 64K), causing the one finished later will have page
780 * unlocked already, triggering various page lock requirement BUG_ON()s.
782 * So here we add an artificial limit that subpage compression can only
783 * if the range is fully page aligned.
785 * In theory we only need to ensure the first page is fully covered, but
786 * the tailing partial page will be locked until the full compression
787 * finishes, delaying the write of other range.
789 * TODO: Make btrfs_run_delalloc_range() to lock all delalloc range
790 * first to prevent any submitted async extent to unlock the full page.
791 * By this, we can ensure for subpage case that only the last async_cow
792 * will unlock the full page.
794 if (fs_info->sectorsize < PAGE_SIZE) {
795 if (!PAGE_ALIGNED(start) ||
796 !PAGE_ALIGNED(end + 1))
801 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
804 if (inode->defrag_compress)
806 /* bad compression ratios */
807 if (inode->flags & BTRFS_INODE_NOCOMPRESS)
809 if (btrfs_test_opt(fs_info, COMPRESS) ||
810 inode->flags & BTRFS_INODE_COMPRESS ||
811 inode->prop_compress)
812 return btrfs_compress_heuristic(&inode->vfs_inode, start, end);
816 static inline void inode_should_defrag(struct btrfs_inode *inode,
817 u64 start, u64 end, u64 num_bytes, u32 small_write)
819 /* If this is a small write inside eof, kick off a defrag */
820 if (num_bytes < small_write &&
821 (start > 0 || end + 1 < inode->disk_i_size))
822 btrfs_add_inode_defrag(NULL, inode, small_write);
826 * Work queue call back to started compression on a file and pages.
828 * This is done inside an ordered work queue, and the compression is spread
829 * across many cpus. The actual IO submission is step two, and the ordered work
830 * queue takes care of making sure that happens in the same order things were
831 * put onto the queue by writepages and friends.
833 * If this code finds it can't get good compression, it puts an entry onto the
834 * work queue to write the uncompressed bytes. This makes sure that both
835 * compressed inodes and uncompressed inodes are written in the same order that
836 * the flusher thread sent them down.
838 static void compress_file_range(struct btrfs_work *work)
840 struct async_chunk *async_chunk =
841 container_of(work, struct async_chunk, work);
842 struct btrfs_inode *inode = async_chunk->inode;
843 struct btrfs_fs_info *fs_info = inode->root->fs_info;
844 struct address_space *mapping = inode->vfs_inode.i_mapping;
845 u64 blocksize = fs_info->sectorsize;
846 u64 start = async_chunk->start;
847 u64 end = async_chunk->end;
852 unsigned long nr_pages;
853 unsigned long total_compressed = 0;
854 unsigned long total_in = 0;
857 int compress_type = fs_info->compress_type;
859 inode_should_defrag(inode, start, end, end - start + 1, SZ_16K);
862 * We need to call clear_page_dirty_for_io on each page in the range.
863 * Otherwise applications with the file mmap'd can wander in and change
864 * the page contents while we are compressing them.
866 extent_range_clear_dirty_for_io(&inode->vfs_inode, start, end);
869 * We need to save i_size before now because it could change in between
870 * us evaluating the size and assigning it. This is because we lock and
871 * unlock the page in truncate and fallocate, and then modify the i_size
874 * The barriers are to emulate READ_ONCE, remove that once i_size_read
878 i_size = i_size_read(&inode->vfs_inode);
880 actual_end = min_t(u64, i_size, end + 1);
883 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
884 nr_pages = min_t(unsigned long, nr_pages, BTRFS_MAX_COMPRESSED_PAGES);
887 * we don't want to send crud past the end of i_size through
888 * compression, that's just a waste of CPU time. So, if the
889 * end of the file is before the start of our current
890 * requested range of bytes, we bail out to the uncompressed
891 * cleanup code that can deal with all of this.
893 * It isn't really the fastest way to fix things, but this is a
894 * very uncommon corner.
896 if (actual_end <= start)
897 goto cleanup_and_bail_uncompressed;
899 total_compressed = actual_end - start;
902 * Skip compression for a small file range(<=blocksize) that
903 * isn't an inline extent, since it doesn't save disk space at all.
905 if (total_compressed <= blocksize &&
906 (start > 0 || end + 1 < inode->disk_i_size))
907 goto cleanup_and_bail_uncompressed;
910 * For subpage case, we require full page alignment for the sector
912 * Thus we must also check against @actual_end, not just @end.
914 if (blocksize < PAGE_SIZE) {
915 if (!PAGE_ALIGNED(start) ||
916 !PAGE_ALIGNED(round_up(actual_end, blocksize)))
917 goto cleanup_and_bail_uncompressed;
920 total_compressed = min_t(unsigned long, total_compressed,
921 BTRFS_MAX_UNCOMPRESSED);
926 * We do compression for mount -o compress and when the inode has not
927 * been flagged as NOCOMPRESS. This flag can change at any time if we
928 * discover bad compression ratios.
930 if (!inode_need_compress(inode, start, end))
931 goto cleanup_and_bail_uncompressed;
933 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
936 * Memory allocation failure is not a fatal error, we can fall
937 * back to uncompressed code.
939 goto cleanup_and_bail_uncompressed;
942 if (inode->defrag_compress)
943 compress_type = inode->defrag_compress;
944 else if (inode->prop_compress)
945 compress_type = inode->prop_compress;
947 /* Compression level is applied here. */
948 ret = btrfs_compress_pages(compress_type | (fs_info->compress_level << 4),
949 mapping, start, pages, &nr_pages, &total_in,
952 goto mark_incompressible;
955 * Zero the tail end of the last page, as we might be sending it down
958 poff = offset_in_page(total_compressed);
960 memzero_page(pages[nr_pages - 1], poff, PAGE_SIZE - poff);
963 * Try to create an inline extent.
965 * If we didn't compress the entire range, try to create an uncompressed
966 * inline extent, else a compressed one.
968 * Check cow_file_range() for why we don't even try to create inline
969 * extent for the subpage case.
971 if (start == 0 && fs_info->sectorsize == PAGE_SIZE) {
972 if (total_in < actual_end) {
973 ret = cow_file_range_inline(inode, actual_end, 0,
974 BTRFS_COMPRESS_NONE, NULL,
977 ret = cow_file_range_inline(inode, actual_end,
979 compress_type, pages,
983 unsigned long clear_flags = EXTENT_DELALLOC |
984 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
985 EXTENT_DO_ACCOUNTING;
988 mapping_set_error(mapping, -EIO);
991 * inline extent creation worked or returned error,
992 * we don't need to create any more async work items.
993 * Unlock and free up our temp pages.
995 * We use DO_ACCOUNTING here because we need the
996 * delalloc_release_metadata to be done _after_ we drop
997 * our outstanding extent for clearing delalloc for this
1000 extent_clear_unlock_delalloc(inode, start, end,
1004 PAGE_START_WRITEBACK |
1005 PAGE_END_WRITEBACK);
1011 * We aren't doing an inline extent. Round the compressed size up to a
1012 * block size boundary so the allocator does sane things.
1014 total_compressed = ALIGN(total_compressed, blocksize);
1017 * One last check to make sure the compression is really a win, compare
1018 * the page count read with the blocks on disk, compression must free at
1021 total_in = round_up(total_in, fs_info->sectorsize);
1022 if (total_compressed + blocksize > total_in)
1023 goto mark_incompressible;
1026 * The async work queues will take care of doing actual allocation on
1027 * disk for these compressed pages, and will submit the bios.
1029 ret = add_async_extent(async_chunk, start, total_in, total_compressed, pages,
1030 nr_pages, compress_type);
1032 if (start + total_in < end) {
1039 mark_incompressible:
1040 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) && !inode->prop_compress)
1041 inode->flags |= BTRFS_INODE_NOCOMPRESS;
1042 cleanup_and_bail_uncompressed:
1043 ret = add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
1044 BTRFS_COMPRESS_NONE);
1048 for (i = 0; i < nr_pages; i++) {
1049 WARN_ON(pages[i]->mapping);
1050 btrfs_free_compr_page(pages[i]);
1056 static void free_async_extent_pages(struct async_extent *async_extent)
1060 if (!async_extent->pages)
1063 for (i = 0; i < async_extent->nr_pages; i++) {
1064 WARN_ON(async_extent->pages[i]->mapping);
1065 btrfs_free_compr_page(async_extent->pages[i]);
1067 kfree(async_extent->pages);
1068 async_extent->nr_pages = 0;
1069 async_extent->pages = NULL;
1072 static void submit_uncompressed_range(struct btrfs_inode *inode,
1073 struct async_extent *async_extent,
1074 struct page *locked_page)
1076 u64 start = async_extent->start;
1077 u64 end = async_extent->start + async_extent->ram_size - 1;
1079 struct writeback_control wbc = {
1080 .sync_mode = WB_SYNC_ALL,
1081 .range_start = start,
1083 .no_cgroup_owner = 1,
1086 wbc_attach_fdatawrite_inode(&wbc, &inode->vfs_inode);
1087 ret = run_delalloc_cow(inode, locked_page, start, end, &wbc, false);
1088 wbc_detach_inode(&wbc);
1090 btrfs_cleanup_ordered_extents(inode, locked_page, start, end - start + 1);
1092 const u64 page_start = page_offset(locked_page);
1094 set_page_writeback(locked_page);
1095 end_page_writeback(locked_page);
1096 btrfs_mark_ordered_io_finished(inode, locked_page,
1097 page_start, PAGE_SIZE,
1099 mapping_set_error(locked_page->mapping, ret);
1100 unlock_page(locked_page);
1105 static void submit_one_async_extent(struct async_chunk *async_chunk,
1106 struct async_extent *async_extent,
1109 struct btrfs_inode *inode = async_chunk->inode;
1110 struct extent_io_tree *io_tree = &inode->io_tree;
1111 struct btrfs_root *root = inode->root;
1112 struct btrfs_fs_info *fs_info = root->fs_info;
1113 struct btrfs_ordered_extent *ordered;
1114 struct btrfs_key ins;
1115 struct page *locked_page = NULL;
1116 struct extent_map *em;
1118 u64 start = async_extent->start;
1119 u64 end = async_extent->start + async_extent->ram_size - 1;
1121 if (async_chunk->blkcg_css)
1122 kthread_associate_blkcg(async_chunk->blkcg_css);
1125 * If async_chunk->locked_page is in the async_extent range, we need to
1128 if (async_chunk->locked_page) {
1129 u64 locked_page_start = page_offset(async_chunk->locked_page);
1130 u64 locked_page_end = locked_page_start + PAGE_SIZE - 1;
1132 if (!(start >= locked_page_end || end <= locked_page_start))
1133 locked_page = async_chunk->locked_page;
1135 lock_extent(io_tree, start, end, NULL);
1137 if (async_extent->compress_type == BTRFS_COMPRESS_NONE) {
1138 submit_uncompressed_range(inode, async_extent, locked_page);
1142 ret = btrfs_reserve_extent(root, async_extent->ram_size,
1143 async_extent->compressed_size,
1144 async_extent->compressed_size,
1145 0, *alloc_hint, &ins, 1, 1);
1148 * Here we used to try again by going back to non-compressed
1149 * path for ENOSPC. But we can't reserve space even for
1150 * compressed size, how could it work for uncompressed size
1151 * which requires larger size? So here we directly go error
1157 /* Here we're doing allocation and writeback of the compressed pages */
1158 em = create_io_em(inode, start,
1159 async_extent->ram_size, /* len */
1160 start, /* orig_start */
1161 ins.objectid, /* block_start */
1162 ins.offset, /* block_len */
1163 ins.offset, /* orig_block_len */
1164 async_extent->ram_size, /* ram_bytes */
1165 async_extent->compress_type,
1166 BTRFS_ORDERED_COMPRESSED);
1169 goto out_free_reserve;
1171 free_extent_map(em);
1173 ordered = btrfs_alloc_ordered_extent(inode, start, /* file_offset */
1174 async_extent->ram_size, /* num_bytes */
1175 async_extent->ram_size, /* ram_bytes */
1176 ins.objectid, /* disk_bytenr */
1177 ins.offset, /* disk_num_bytes */
1179 1 << BTRFS_ORDERED_COMPRESSED,
1180 async_extent->compress_type);
1181 if (IS_ERR(ordered)) {
1182 btrfs_drop_extent_map_range(inode, start, end, false);
1183 ret = PTR_ERR(ordered);
1184 goto out_free_reserve;
1186 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1188 /* Clear dirty, set writeback and unlock the pages. */
1189 extent_clear_unlock_delalloc(inode, start, end,
1190 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
1191 PAGE_UNLOCK | PAGE_START_WRITEBACK);
1192 btrfs_submit_compressed_write(ordered,
1193 async_extent->pages, /* compressed_pages */
1194 async_extent->nr_pages,
1195 async_chunk->write_flags, true);
1196 *alloc_hint = ins.objectid + ins.offset;
1198 if (async_chunk->blkcg_css)
1199 kthread_associate_blkcg(NULL);
1200 kfree(async_extent);
1204 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1205 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1207 mapping_set_error(inode->vfs_inode.i_mapping, -EIO);
1208 extent_clear_unlock_delalloc(inode, start, end,
1209 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
1210 EXTENT_DELALLOC_NEW |
1211 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
1212 PAGE_UNLOCK | PAGE_START_WRITEBACK |
1213 PAGE_END_WRITEBACK);
1214 free_async_extent_pages(async_extent);
1215 if (async_chunk->blkcg_css)
1216 kthread_associate_blkcg(NULL);
1217 btrfs_debug(fs_info,
1218 "async extent submission failed root=%lld inode=%llu start=%llu len=%llu ret=%d",
1219 root->root_key.objectid, btrfs_ino(inode), start,
1220 async_extent->ram_size, ret);
1221 kfree(async_extent);
1224 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
1227 struct extent_map_tree *em_tree = &inode->extent_tree;
1228 struct extent_map *em;
1231 read_lock(&em_tree->lock);
1232 em = search_extent_mapping(em_tree, start, num_bytes);
1235 * if block start isn't an actual block number then find the
1236 * first block in this inode and use that as a hint. If that
1237 * block is also bogus then just don't worry about it.
1239 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1240 free_extent_map(em);
1241 em = search_extent_mapping(em_tree, 0, 0);
1242 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
1243 alloc_hint = em->block_start;
1245 free_extent_map(em);
1247 alloc_hint = em->block_start;
1248 free_extent_map(em);
1251 read_unlock(&em_tree->lock);
1257 * when extent_io.c finds a delayed allocation range in the file,
1258 * the call backs end up in this code. The basic idea is to
1259 * allocate extents on disk for the range, and create ordered data structs
1260 * in ram to track those extents.
1262 * locked_page is the page that writepage had locked already. We use
1263 * it to make sure we don't do extra locks or unlocks.
1265 * When this function fails, it unlocks all pages except @locked_page.
1267 * When this function successfully creates an inline extent, it returns 1 and
1268 * unlocks all pages including locked_page and starts I/O on them.
1269 * (In reality inline extents are limited to a single page, so locked_page is
1270 * the only page handled anyway).
1272 * When this function succeed and creates a normal extent, the page locking
1273 * status depends on the passed in flags:
1275 * - If @keep_locked is set, all pages are kept locked.
1276 * - Else all pages except for @locked_page are unlocked.
1278 * When a failure happens in the second or later iteration of the
1279 * while-loop, the ordered extents created in previous iterations are kept
1280 * intact. So, the caller must clean them up by calling
1281 * btrfs_cleanup_ordered_extents(). See btrfs_run_delalloc_range() for
1284 static noinline int cow_file_range(struct btrfs_inode *inode,
1285 struct page *locked_page, u64 start, u64 end,
1287 bool keep_locked, bool no_inline)
1289 struct btrfs_root *root = inode->root;
1290 struct btrfs_fs_info *fs_info = root->fs_info;
1292 u64 orig_start = start;
1294 unsigned long ram_size;
1295 u64 cur_alloc_size = 0;
1297 u64 blocksize = fs_info->sectorsize;
1298 struct btrfs_key ins;
1299 struct extent_map *em;
1300 unsigned clear_bits;
1301 unsigned long page_ops;
1302 bool extent_reserved = false;
1305 if (btrfs_is_free_space_inode(inode)) {
1310 num_bytes = ALIGN(end - start + 1, blocksize);
1311 num_bytes = max(blocksize, num_bytes);
1312 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1314 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1317 * Due to the page size limit, for subpage we can only trigger the
1318 * writeback for the dirty sectors of page, that means data writeback
1319 * is doing more writeback than what we want.
1321 * This is especially unexpected for some call sites like fallocate,
1322 * where we only increase i_size after everything is done.
1323 * This means we can trigger inline extent even if we didn't want to.
1324 * So here we skip inline extent creation completely.
1326 if (start == 0 && fs_info->sectorsize == PAGE_SIZE && !no_inline) {
1327 u64 actual_end = min_t(u64, i_size_read(&inode->vfs_inode),
1330 /* lets try to make an inline extent */
1331 ret = cow_file_range_inline(inode, actual_end, 0,
1332 BTRFS_COMPRESS_NONE, NULL, false);
1335 * We use DO_ACCOUNTING here because we need the
1336 * delalloc_release_metadata to be run _after_ we drop
1337 * our outstanding extent for clearing delalloc for this
1340 extent_clear_unlock_delalloc(inode, start, end,
1342 EXTENT_LOCKED | EXTENT_DELALLOC |
1343 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1344 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1345 PAGE_START_WRITEBACK | PAGE_END_WRITEBACK);
1347 * locked_page is locked by the caller of
1348 * writepage_delalloc(), not locked by
1349 * __process_pages_contig().
1351 * We can't let __process_pages_contig() to unlock it,
1352 * as it doesn't have any subpage::writers recorded.
1354 * Here we manually unlock the page, since the caller
1355 * can't determine if it's an inline extent or a
1356 * compressed extent.
1358 unlock_page(locked_page);
1361 } else if (ret < 0) {
1366 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1369 * Relocation relies on the relocated extents to have exactly the same
1370 * size as the original extents. Normally writeback for relocation data
1371 * extents follows a NOCOW path because relocation preallocates the
1372 * extents. However, due to an operation such as scrub turning a block
1373 * group to RO mode, it may fallback to COW mode, so we must make sure
1374 * an extent allocated during COW has exactly the requested size and can
1375 * not be split into smaller extents, otherwise relocation breaks and
1376 * fails during the stage where it updates the bytenr of file extent
1379 if (btrfs_is_data_reloc_root(root))
1380 min_alloc_size = num_bytes;
1382 min_alloc_size = fs_info->sectorsize;
1384 while (num_bytes > 0) {
1385 struct btrfs_ordered_extent *ordered;
1387 cur_alloc_size = num_bytes;
1388 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1389 min_alloc_size, 0, alloc_hint,
1391 if (ret == -EAGAIN) {
1393 * btrfs_reserve_extent only returns -EAGAIN for zoned
1394 * file systems, which is an indication that there are
1395 * no active zones to allocate from at the moment.
1397 * If this is the first loop iteration, wait for at
1398 * least one zone to finish before retrying the
1399 * allocation. Otherwise ask the caller to write out
1400 * the already allocated blocks before coming back to
1401 * us, or return -ENOSPC if it can't handle retries.
1403 ASSERT(btrfs_is_zoned(fs_info));
1404 if (start == orig_start) {
1405 wait_on_bit_io(&inode->root->fs_info->flags,
1406 BTRFS_FS_NEED_ZONE_FINISH,
1407 TASK_UNINTERRUPTIBLE);
1411 *done_offset = start - 1;
1418 cur_alloc_size = ins.offset;
1419 extent_reserved = true;
1421 ram_size = ins.offset;
1422 em = create_io_em(inode, start, ins.offset, /* len */
1423 start, /* orig_start */
1424 ins.objectid, /* block_start */
1425 ins.offset, /* block_len */
1426 ins.offset, /* orig_block_len */
1427 ram_size, /* ram_bytes */
1428 BTRFS_COMPRESS_NONE, /* compress_type */
1429 BTRFS_ORDERED_REGULAR /* type */);
1434 free_extent_map(em);
1436 ordered = btrfs_alloc_ordered_extent(inode, start, ram_size,
1437 ram_size, ins.objectid, cur_alloc_size,
1438 0, 1 << BTRFS_ORDERED_REGULAR,
1439 BTRFS_COMPRESS_NONE);
1440 if (IS_ERR(ordered)) {
1441 ret = PTR_ERR(ordered);
1442 goto out_drop_extent_cache;
1445 if (btrfs_is_data_reloc_root(root)) {
1446 ret = btrfs_reloc_clone_csums(ordered);
1449 * Only drop cache here, and process as normal.
1451 * We must not allow extent_clear_unlock_delalloc()
1452 * at out_unlock label to free meta of this ordered
1453 * extent, as its meta should be freed by
1454 * btrfs_finish_ordered_io().
1456 * So we must continue until @start is increased to
1457 * skip current ordered extent.
1460 btrfs_drop_extent_map_range(inode, start,
1461 start + ram_size - 1,
1464 btrfs_put_ordered_extent(ordered);
1466 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1469 * We're not doing compressed IO, don't unlock the first page
1470 * (which the caller expects to stay locked), don't clear any
1471 * dirty bits and don't set any writeback bits
1473 * Do set the Ordered (Private2) bit so we know this page was
1474 * properly setup for writepage.
1476 page_ops = (keep_locked ? 0 : PAGE_UNLOCK);
1477 page_ops |= PAGE_SET_ORDERED;
1479 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1481 EXTENT_LOCKED | EXTENT_DELALLOC,
1483 if (num_bytes < cur_alloc_size)
1486 num_bytes -= cur_alloc_size;
1487 alloc_hint = ins.objectid + ins.offset;
1488 start += cur_alloc_size;
1489 extent_reserved = false;
1492 * btrfs_reloc_clone_csums() error, since start is increased
1493 * extent_clear_unlock_delalloc() at out_unlock label won't
1494 * free metadata of current ordered extent, we're OK to exit.
1504 out_drop_extent_cache:
1505 btrfs_drop_extent_map_range(inode, start, start + ram_size - 1, false);
1507 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1508 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1511 * Now, we have three regions to clean up:
1513 * |-------(1)----|---(2)---|-------------(3)----------|
1514 * `- orig_start `- start `- start + cur_alloc_size `- end
1516 * We process each region below.
1519 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1520 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1521 page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK;
1524 * For the range (1). We have already instantiated the ordered extents
1525 * for this region. They are cleaned up by
1526 * btrfs_cleanup_ordered_extents() in e.g,
1527 * btrfs_run_delalloc_range(). EXTENT_LOCKED | EXTENT_DELALLOC are
1528 * already cleared in the above loop. And, EXTENT_DELALLOC_NEW |
1529 * EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV are handled by the cleanup
1532 * However, in case of @keep_locked, we still need to unlock the pages
1533 * (except @locked_page) to ensure all the pages are unlocked.
1535 if (keep_locked && orig_start < start) {
1537 mapping_set_error(inode->vfs_inode.i_mapping, ret);
1538 extent_clear_unlock_delalloc(inode, orig_start, start - 1,
1539 locked_page, 0, page_ops);
1543 * For the range (2). If we reserved an extent for our delalloc range
1544 * (or a subrange) and failed to create the respective ordered extent,
1545 * then it means that when we reserved the extent we decremented the
1546 * extent's size from the data space_info's bytes_may_use counter and
1547 * incremented the space_info's bytes_reserved counter by the same
1548 * amount. We must make sure extent_clear_unlock_delalloc() does not try
1549 * to decrement again the data space_info's bytes_may_use counter,
1550 * therefore we do not pass it the flag EXTENT_CLEAR_DATA_RESV.
1552 if (extent_reserved) {
1553 extent_clear_unlock_delalloc(inode, start,
1554 start + cur_alloc_size - 1,
1558 start += cur_alloc_size;
1562 * For the range (3). We never touched the region. In addition to the
1563 * clear_bits above, we add EXTENT_CLEAR_DATA_RESV to release the data
1564 * space_info's bytes_may_use counter, reserved in
1565 * btrfs_check_data_free_space().
1568 clear_bits |= EXTENT_CLEAR_DATA_RESV;
1569 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1570 clear_bits, page_ops);
1576 * Phase two of compressed writeback. This is the ordered portion of the code,
1577 * which only gets called in the order the work was queued. We walk all the
1578 * async extents created by compress_file_range and send them down to the disk.
1580 * If called with @do_free == true then it'll try to finish the work and free
1581 * the work struct eventually.
1583 static noinline void submit_compressed_extents(struct btrfs_work *work, bool do_free)
1585 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1587 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1588 struct async_extent *async_extent;
1589 unsigned long nr_pages;
1593 struct async_chunk *async_chunk;
1594 struct async_cow *async_cow;
1596 async_chunk = container_of(work, struct async_chunk, work);
1597 btrfs_add_delayed_iput(async_chunk->inode);
1598 if (async_chunk->blkcg_css)
1599 css_put(async_chunk->blkcg_css);
1601 async_cow = async_chunk->async_cow;
1602 if (atomic_dec_and_test(&async_cow->num_chunks))
1607 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1610 while (!list_empty(&async_chunk->extents)) {
1611 async_extent = list_entry(async_chunk->extents.next,
1612 struct async_extent, list);
1613 list_del(&async_extent->list);
1614 submit_one_async_extent(async_chunk, async_extent, &alloc_hint);
1617 /* atomic_sub_return implies a barrier */
1618 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1620 cond_wake_up_nomb(&fs_info->async_submit_wait);
1623 static bool run_delalloc_compressed(struct btrfs_inode *inode,
1624 struct page *locked_page, u64 start,
1625 u64 end, struct writeback_control *wbc)
1627 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1628 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1629 struct async_cow *ctx;
1630 struct async_chunk *async_chunk;
1631 unsigned long nr_pages;
1632 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1635 const blk_opf_t write_flags = wbc_to_write_flags(wbc);
1637 nofs_flag = memalloc_nofs_save();
1638 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1639 memalloc_nofs_restore(nofs_flag);
1643 unlock_extent(&inode->io_tree, start, end, NULL);
1644 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
1646 async_chunk = ctx->chunks;
1647 atomic_set(&ctx->num_chunks, num_chunks);
1649 for (i = 0; i < num_chunks; i++) {
1650 u64 cur_end = min(end, start + SZ_512K - 1);
1653 * igrab is called higher up in the call chain, take only the
1654 * lightweight reference for the callback lifetime
1656 ihold(&inode->vfs_inode);
1657 async_chunk[i].async_cow = ctx;
1658 async_chunk[i].inode = inode;
1659 async_chunk[i].start = start;
1660 async_chunk[i].end = cur_end;
1661 async_chunk[i].write_flags = write_flags;
1662 INIT_LIST_HEAD(&async_chunk[i].extents);
1665 * The locked_page comes all the way from writepage and its
1666 * the original page we were actually given. As we spread
1667 * this large delalloc region across multiple async_chunk
1668 * structs, only the first struct needs a pointer to locked_page
1670 * This way we don't need racey decisions about who is supposed
1675 * Depending on the compressibility, the pages might or
1676 * might not go through async. We want all of them to
1677 * be accounted against wbc once. Let's do it here
1678 * before the paths diverge. wbc accounting is used
1679 * only for foreign writeback detection and doesn't
1680 * need full accuracy. Just account the whole thing
1681 * against the first page.
1683 wbc_account_cgroup_owner(wbc, locked_page,
1685 async_chunk[i].locked_page = locked_page;
1688 async_chunk[i].locked_page = NULL;
1691 if (blkcg_css != blkcg_root_css) {
1693 async_chunk[i].blkcg_css = blkcg_css;
1694 async_chunk[i].write_flags |= REQ_BTRFS_CGROUP_PUNT;
1696 async_chunk[i].blkcg_css = NULL;
1699 btrfs_init_work(&async_chunk[i].work, compress_file_range,
1700 submit_compressed_extents);
1702 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1703 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1705 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1707 start = cur_end + 1;
1713 * Run the delalloc range from start to end, and write back any dirty pages
1714 * covered by the range.
1716 static noinline int run_delalloc_cow(struct btrfs_inode *inode,
1717 struct page *locked_page, u64 start,
1718 u64 end, struct writeback_control *wbc,
1721 u64 done_offset = end;
1724 while (start <= end) {
1725 ret = cow_file_range(inode, locked_page, start, end, &done_offset,
1729 extent_write_locked_range(&inode->vfs_inode, locked_page, start,
1730 done_offset, wbc, pages_dirty);
1731 start = done_offset + 1;
1737 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1738 u64 bytenr, u64 num_bytes, bool nowait)
1740 struct btrfs_root *csum_root = btrfs_csum_root(fs_info, bytenr);
1741 struct btrfs_ordered_sum *sums;
1745 ret = btrfs_lookup_csums_list(csum_root, bytenr, bytenr + num_bytes - 1,
1747 if (ret == 0 && list_empty(&list))
1750 while (!list_empty(&list)) {
1751 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1752 list_del(&sums->list);
1760 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1761 const u64 start, const u64 end)
1763 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1764 const bool is_reloc_ino = btrfs_is_data_reloc_root(inode->root);
1765 const u64 range_bytes = end + 1 - start;
1766 struct extent_io_tree *io_tree = &inode->io_tree;
1767 u64 range_start = start;
1772 * If EXTENT_NORESERVE is set it means that when the buffered write was
1773 * made we had not enough available data space and therefore we did not
1774 * reserve data space for it, since we though we could do NOCOW for the
1775 * respective file range (either there is prealloc extent or the inode
1776 * has the NOCOW bit set).
1778 * However when we need to fallback to COW mode (because for example the
1779 * block group for the corresponding extent was turned to RO mode by a
1780 * scrub or relocation) we need to do the following:
1782 * 1) We increment the bytes_may_use counter of the data space info.
1783 * If COW succeeds, it allocates a new data extent and after doing
1784 * that it decrements the space info's bytes_may_use counter and
1785 * increments its bytes_reserved counter by the same amount (we do
1786 * this at btrfs_add_reserved_bytes()). So we need to increment the
1787 * bytes_may_use counter to compensate (when space is reserved at
1788 * buffered write time, the bytes_may_use counter is incremented);
1790 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1791 * that if the COW path fails for any reason, it decrements (through
1792 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1793 * data space info, which we incremented in the step above.
1795 * If we need to fallback to cow and the inode corresponds to a free
1796 * space cache inode or an inode of the data relocation tree, we must
1797 * also increment bytes_may_use of the data space_info for the same
1798 * reason. Space caches and relocated data extents always get a prealloc
1799 * extent for them, however scrub or balance may have set the block
1800 * group that contains that extent to RO mode and therefore force COW
1801 * when starting writeback.
1803 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1804 EXTENT_NORESERVE, 0, NULL);
1805 if (count > 0 || is_space_ino || is_reloc_ino) {
1807 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1808 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1810 if (is_space_ino || is_reloc_ino)
1811 bytes = range_bytes;
1813 spin_lock(&sinfo->lock);
1814 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1815 spin_unlock(&sinfo->lock);
1818 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1823 * Don't try to create inline extents, as a mix of inline extent that
1824 * is written out and unlocked directly and a normal NOCOW extent
1827 ret = cow_file_range(inode, locked_page, start, end, NULL, false, true);
1832 struct can_nocow_file_extent_args {
1835 /* Start file offset of the range we want to NOCOW. */
1837 /* End file offset (inclusive) of the range we want to NOCOW. */
1839 bool writeback_path;
1842 * Free the path passed to can_nocow_file_extent() once it's not needed
1847 /* Output fields. Only set when can_nocow_file_extent() returns 1. */
1852 /* Number of bytes that can be written to in NOCOW mode. */
1857 * Check if we can NOCOW the file extent that the path points to.
1858 * This function may return with the path released, so the caller should check
1859 * if path->nodes[0] is NULL or not if it needs to use the path afterwards.
1861 * Returns: < 0 on error
1862 * 0 if we can not NOCOW
1865 static int can_nocow_file_extent(struct btrfs_path *path,
1866 struct btrfs_key *key,
1867 struct btrfs_inode *inode,
1868 struct can_nocow_file_extent_args *args)
1870 const bool is_freespace_inode = btrfs_is_free_space_inode(inode);
1871 struct extent_buffer *leaf = path->nodes[0];
1872 struct btrfs_root *root = inode->root;
1873 struct btrfs_file_extent_item *fi;
1878 bool nowait = path->nowait;
1880 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
1881 extent_type = btrfs_file_extent_type(leaf, fi);
1883 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1886 /* Can't access these fields unless we know it's not an inline extent. */
1887 args->disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1888 args->disk_num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1889 args->extent_offset = btrfs_file_extent_offset(leaf, fi);
1891 if (!(inode->flags & BTRFS_INODE_NODATACOW) &&
1892 extent_type == BTRFS_FILE_EXTENT_REG)
1896 * If the extent was created before the generation where the last snapshot
1897 * for its subvolume was created, then this implies the extent is shared,
1898 * hence we must COW.
1900 if (!args->strict &&
1901 btrfs_file_extent_generation(leaf, fi) <=
1902 btrfs_root_last_snapshot(&root->root_item))
1905 /* An explicit hole, must COW. */
1906 if (args->disk_bytenr == 0)
1909 /* Compressed/encrypted/encoded extents must be COWed. */
1910 if (btrfs_file_extent_compression(leaf, fi) ||
1911 btrfs_file_extent_encryption(leaf, fi) ||
1912 btrfs_file_extent_other_encoding(leaf, fi))
1915 extent_end = btrfs_file_extent_end(path);
1918 * The following checks can be expensive, as they need to take other
1919 * locks and do btree or rbtree searches, so release the path to avoid
1920 * blocking other tasks for too long.
1922 btrfs_release_path(path);
1924 ret = btrfs_cross_ref_exist(root, btrfs_ino(inode),
1925 key->offset - args->extent_offset,
1926 args->disk_bytenr, args->strict, path);
1927 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
1931 if (args->free_path) {
1933 * We don't need the path anymore, plus through the
1934 * csum_exist_in_range() call below we will end up allocating
1935 * another path. So free the path to avoid unnecessary extra
1938 btrfs_free_path(path);
1942 /* If there are pending snapshots for this root, we must COW. */
1943 if (args->writeback_path && !is_freespace_inode &&
1944 atomic_read(&root->snapshot_force_cow))
1947 args->disk_bytenr += args->extent_offset;
1948 args->disk_bytenr += args->start - key->offset;
1949 args->num_bytes = min(args->end + 1, extent_end) - args->start;
1952 * Force COW if csums exist in the range. This ensures that csums for a
1953 * given extent are either valid or do not exist.
1955 ret = csum_exist_in_range(root->fs_info, args->disk_bytenr, args->num_bytes,
1957 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
1963 if (args->free_path && path)
1964 btrfs_free_path(path);
1966 return ret < 0 ? ret : can_nocow;
1970 * when nowcow writeback call back. This checks for snapshots or COW copies
1971 * of the extents that exist in the file, and COWs the file as required.
1973 * If no cow copies or snapshots exist, we write directly to the existing
1976 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
1977 struct page *locked_page,
1978 const u64 start, const u64 end)
1980 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1981 struct btrfs_root *root = inode->root;
1982 struct btrfs_path *path;
1983 u64 cow_start = (u64)-1;
1984 u64 cur_offset = start;
1986 bool check_prev = true;
1987 u64 ino = btrfs_ino(inode);
1988 struct can_nocow_file_extent_args nocow_args = { 0 };
1991 * Normally on a zoned device we're only doing COW writes, but in case
1992 * of relocation on a zoned filesystem serializes I/O so that we're only
1993 * writing sequentially and can end up here as well.
1995 ASSERT(!btrfs_is_zoned(fs_info) || btrfs_is_data_reloc_root(root));
1997 path = btrfs_alloc_path();
2003 nocow_args.end = end;
2004 nocow_args.writeback_path = true;
2007 struct btrfs_block_group *nocow_bg = NULL;
2008 struct btrfs_ordered_extent *ordered;
2009 struct btrfs_key found_key;
2010 struct btrfs_file_extent_item *fi;
2011 struct extent_buffer *leaf;
2018 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
2024 * If there is no extent for our range when doing the initial
2025 * search, then go back to the previous slot as it will be the
2026 * one containing the search offset
2028 if (ret > 0 && path->slots[0] > 0 && check_prev) {
2029 leaf = path->nodes[0];
2030 btrfs_item_key_to_cpu(leaf, &found_key,
2031 path->slots[0] - 1);
2032 if (found_key.objectid == ino &&
2033 found_key.type == BTRFS_EXTENT_DATA_KEY)
2038 /* Go to next leaf if we have exhausted the current one */
2039 leaf = path->nodes[0];
2040 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2041 ret = btrfs_next_leaf(root, path);
2046 leaf = path->nodes[0];
2049 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2051 /* Didn't find anything for our INO */
2052 if (found_key.objectid > ino)
2055 * Keep searching until we find an EXTENT_ITEM or there are no
2056 * more extents for this inode
2058 if (WARN_ON_ONCE(found_key.objectid < ino) ||
2059 found_key.type < BTRFS_EXTENT_DATA_KEY) {
2064 /* Found key is not EXTENT_DATA_KEY or starts after req range */
2065 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
2066 found_key.offset > end)
2070 * If the found extent starts after requested offset, then
2071 * adjust extent_end to be right before this extent begins
2073 if (found_key.offset > cur_offset) {
2074 extent_end = found_key.offset;
2080 * Found extent which begins before our range and potentially
2083 fi = btrfs_item_ptr(leaf, path->slots[0],
2084 struct btrfs_file_extent_item);
2085 extent_type = btrfs_file_extent_type(leaf, fi);
2086 /* If this is triggered then we have a memory corruption. */
2087 ASSERT(extent_type < BTRFS_NR_FILE_EXTENT_TYPES);
2088 if (WARN_ON(extent_type >= BTRFS_NR_FILE_EXTENT_TYPES)) {
2092 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
2093 extent_end = btrfs_file_extent_end(path);
2096 * If the extent we got ends before our current offset, skip to
2099 if (extent_end <= cur_offset) {
2104 nocow_args.start = cur_offset;
2105 ret = can_nocow_file_extent(path, &found_key, inode, &nocow_args);
2112 nocow_bg = btrfs_inc_nocow_writers(fs_info, nocow_args.disk_bytenr);
2116 * If we can't perform NOCOW writeback for the range,
2117 * then record the beginning of the range that needs to
2118 * be COWed. It will be written out before the next
2119 * NOCOW range if we find one, or when exiting this
2122 if (cow_start == (u64)-1)
2123 cow_start = cur_offset;
2124 cur_offset = extent_end;
2125 if (cur_offset > end)
2127 if (!path->nodes[0])
2134 * COW range from cow_start to found_key.offset - 1. As the key
2135 * will contain the beginning of the first extent that can be
2136 * NOCOW, following one which needs to be COW'ed
2138 if (cow_start != (u64)-1) {
2139 ret = fallback_to_cow(inode, locked_page,
2140 cow_start, found_key.offset - 1);
2141 cow_start = (u64)-1;
2143 btrfs_dec_nocow_writers(nocow_bg);
2148 nocow_end = cur_offset + nocow_args.num_bytes - 1;
2149 is_prealloc = extent_type == BTRFS_FILE_EXTENT_PREALLOC;
2151 u64 orig_start = found_key.offset - nocow_args.extent_offset;
2152 struct extent_map *em;
2154 em = create_io_em(inode, cur_offset, nocow_args.num_bytes,
2156 nocow_args.disk_bytenr, /* block_start */
2157 nocow_args.num_bytes, /* block_len */
2158 nocow_args.disk_num_bytes, /* orig_block_len */
2159 ram_bytes, BTRFS_COMPRESS_NONE,
2160 BTRFS_ORDERED_PREALLOC);
2162 btrfs_dec_nocow_writers(nocow_bg);
2166 free_extent_map(em);
2169 ordered = btrfs_alloc_ordered_extent(inode, cur_offset,
2170 nocow_args.num_bytes, nocow_args.num_bytes,
2171 nocow_args.disk_bytenr, nocow_args.num_bytes, 0,
2173 ? (1 << BTRFS_ORDERED_PREALLOC)
2174 : (1 << BTRFS_ORDERED_NOCOW),
2175 BTRFS_COMPRESS_NONE);
2176 btrfs_dec_nocow_writers(nocow_bg);
2177 if (IS_ERR(ordered)) {
2179 btrfs_drop_extent_map_range(inode, cur_offset,
2182 ret = PTR_ERR(ordered);
2186 if (btrfs_is_data_reloc_root(root))
2188 * Error handled later, as we must prevent
2189 * extent_clear_unlock_delalloc() in error handler
2190 * from freeing metadata of created ordered extent.
2192 ret = btrfs_reloc_clone_csums(ordered);
2193 btrfs_put_ordered_extent(ordered);
2195 extent_clear_unlock_delalloc(inode, cur_offset, nocow_end,
2196 locked_page, EXTENT_LOCKED |
2198 EXTENT_CLEAR_DATA_RESV,
2199 PAGE_UNLOCK | PAGE_SET_ORDERED);
2201 cur_offset = extent_end;
2204 * btrfs_reloc_clone_csums() error, now we're OK to call error
2205 * handler, as metadata for created ordered extent will only
2206 * be freed by btrfs_finish_ordered_io().
2210 if (cur_offset > end)
2213 btrfs_release_path(path);
2215 if (cur_offset <= end && cow_start == (u64)-1)
2216 cow_start = cur_offset;
2218 if (cow_start != (u64)-1) {
2220 ret = fallback_to_cow(inode, locked_page, cow_start, end);
2221 cow_start = (u64)-1;
2226 btrfs_free_path(path);
2231 * If an error happened while a COW region is outstanding, cur_offset
2232 * needs to be reset to cow_start to ensure the COW region is unlocked
2235 if (cow_start != (u64)-1)
2236 cur_offset = cow_start;
2237 if (cur_offset < end)
2238 extent_clear_unlock_delalloc(inode, cur_offset, end,
2239 locked_page, EXTENT_LOCKED |
2240 EXTENT_DELALLOC | EXTENT_DEFRAG |
2241 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
2242 PAGE_START_WRITEBACK |
2243 PAGE_END_WRITEBACK);
2244 btrfs_free_path(path);
2248 static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end)
2250 if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) {
2251 if (inode->defrag_bytes &&
2252 test_range_bit_exists(&inode->io_tree, start, end, EXTENT_DEFRAG))
2260 * Function to process delayed allocation (create CoW) for ranges which are
2261 * being touched for the first time.
2263 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
2264 u64 start, u64 end, struct writeback_control *wbc)
2266 const bool zoned = btrfs_is_zoned(inode->root->fs_info);
2270 * The range must cover part of the @locked_page, or a return of 1
2271 * can confuse the caller.
2273 ASSERT(!(end <= page_offset(locked_page) ||
2274 start >= page_offset(locked_page) + PAGE_SIZE));
2276 if (should_nocow(inode, start, end)) {
2277 ret = run_delalloc_nocow(inode, locked_page, start, end);
2281 if (btrfs_inode_can_compress(inode) &&
2282 inode_need_compress(inode, start, end) &&
2283 run_delalloc_compressed(inode, locked_page, start, end, wbc))
2287 ret = run_delalloc_cow(inode, locked_page, start, end, wbc,
2290 ret = cow_file_range(inode, locked_page, start, end, NULL,
2295 btrfs_cleanup_ordered_extents(inode, locked_page, start,
2300 void btrfs_split_delalloc_extent(struct btrfs_inode *inode,
2301 struct extent_state *orig, u64 split)
2303 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2306 lockdep_assert_held(&inode->io_tree.lock);
2308 /* not delalloc, ignore it */
2309 if (!(orig->state & EXTENT_DELALLOC))
2312 size = orig->end - orig->start + 1;
2313 if (size > fs_info->max_extent_size) {
2318 * See the explanation in btrfs_merge_delalloc_extent, the same
2319 * applies here, just in reverse.
2321 new_size = orig->end - split + 1;
2322 num_extents = count_max_extents(fs_info, new_size);
2323 new_size = split - orig->start;
2324 num_extents += count_max_extents(fs_info, new_size);
2325 if (count_max_extents(fs_info, size) >= num_extents)
2329 spin_lock(&inode->lock);
2330 btrfs_mod_outstanding_extents(inode, 1);
2331 spin_unlock(&inode->lock);
2335 * Handle merged delayed allocation extents so we can keep track of new extents
2336 * that are just merged onto old extents, such as when we are doing sequential
2337 * writes, so we can properly account for the metadata space we'll need.
2339 void btrfs_merge_delalloc_extent(struct btrfs_inode *inode, struct extent_state *new,
2340 struct extent_state *other)
2342 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2343 u64 new_size, old_size;
2346 lockdep_assert_held(&inode->io_tree.lock);
2348 /* not delalloc, ignore it */
2349 if (!(other->state & EXTENT_DELALLOC))
2352 if (new->start > other->start)
2353 new_size = new->end - other->start + 1;
2355 new_size = other->end - new->start + 1;
2357 /* we're not bigger than the max, unreserve the space and go */
2358 if (new_size <= fs_info->max_extent_size) {
2359 spin_lock(&inode->lock);
2360 btrfs_mod_outstanding_extents(inode, -1);
2361 spin_unlock(&inode->lock);
2366 * We have to add up either side to figure out how many extents were
2367 * accounted for before we merged into one big extent. If the number of
2368 * extents we accounted for is <= the amount we need for the new range
2369 * then we can return, otherwise drop. Think of it like this
2373 * So we've grown the extent by a MAX_SIZE extent, this would mean we
2374 * need 2 outstanding extents, on one side we have 1 and the other side
2375 * we have 1 so they are == and we can return. But in this case
2377 * [MAX_SIZE+4k][MAX_SIZE+4k]
2379 * Each range on their own accounts for 2 extents, but merged together
2380 * they are only 3 extents worth of accounting, so we need to drop in
2383 old_size = other->end - other->start + 1;
2384 num_extents = count_max_extents(fs_info, old_size);
2385 old_size = new->end - new->start + 1;
2386 num_extents += count_max_extents(fs_info, old_size);
2387 if (count_max_extents(fs_info, new_size) >= num_extents)
2390 spin_lock(&inode->lock);
2391 btrfs_mod_outstanding_extents(inode, -1);
2392 spin_unlock(&inode->lock);
2395 static void btrfs_add_delalloc_inode(struct btrfs_inode *inode)
2397 struct btrfs_root *root = inode->root;
2398 struct btrfs_fs_info *fs_info = root->fs_info;
2400 spin_lock(&root->delalloc_lock);
2401 ASSERT(list_empty(&inode->delalloc_inodes));
2402 list_add_tail(&inode->delalloc_inodes, &root->delalloc_inodes);
2403 root->nr_delalloc_inodes++;
2404 if (root->nr_delalloc_inodes == 1) {
2405 spin_lock(&fs_info->delalloc_root_lock);
2406 ASSERT(list_empty(&root->delalloc_root));
2407 list_add_tail(&root->delalloc_root, &fs_info->delalloc_roots);
2408 spin_unlock(&fs_info->delalloc_root_lock);
2410 spin_unlock(&root->delalloc_lock);
2413 void __btrfs_del_delalloc_inode(struct btrfs_inode *inode)
2415 struct btrfs_root *root = inode->root;
2416 struct btrfs_fs_info *fs_info = root->fs_info;
2418 lockdep_assert_held(&root->delalloc_lock);
2421 * We may be called after the inode was already deleted from the list,
2422 * namely in the transaction abort path btrfs_destroy_delalloc_inodes(),
2423 * and then later through btrfs_clear_delalloc_extent() while the inode
2424 * still has ->delalloc_bytes > 0.
2426 if (!list_empty(&inode->delalloc_inodes)) {
2427 list_del_init(&inode->delalloc_inodes);
2428 root->nr_delalloc_inodes--;
2429 if (!root->nr_delalloc_inodes) {
2430 ASSERT(list_empty(&root->delalloc_inodes));
2431 spin_lock(&fs_info->delalloc_root_lock);
2432 ASSERT(!list_empty(&root->delalloc_root));
2433 list_del_init(&root->delalloc_root);
2434 spin_unlock(&fs_info->delalloc_root_lock);
2439 static void btrfs_del_delalloc_inode(struct btrfs_inode *inode)
2441 spin_lock(&inode->root->delalloc_lock);
2442 __btrfs_del_delalloc_inode(inode);
2443 spin_unlock(&inode->root->delalloc_lock);
2447 * Properly track delayed allocation bytes in the inode and to maintain the
2448 * list of inodes that have pending delalloc work to be done.
2450 void btrfs_set_delalloc_extent(struct btrfs_inode *inode, struct extent_state *state,
2453 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2455 lockdep_assert_held(&inode->io_tree.lock);
2457 if ((bits & EXTENT_DEFRAG) && !(bits & EXTENT_DELALLOC))
2460 * set_bit and clear bit hooks normally require _irqsave/restore
2461 * but in this case, we are only testing for the DELALLOC
2462 * bit, which is only set or cleared with irqs on
2464 if (!(state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2465 u64 len = state->end + 1 - state->start;
2466 u64 prev_delalloc_bytes;
2467 u32 num_extents = count_max_extents(fs_info, len);
2469 spin_lock(&inode->lock);
2470 btrfs_mod_outstanding_extents(inode, num_extents);
2471 spin_unlock(&inode->lock);
2473 /* For sanity tests */
2474 if (btrfs_is_testing(fs_info))
2477 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2478 fs_info->delalloc_batch);
2479 spin_lock(&inode->lock);
2480 prev_delalloc_bytes = inode->delalloc_bytes;
2481 inode->delalloc_bytes += len;
2482 if (bits & EXTENT_DEFRAG)
2483 inode->defrag_bytes += len;
2484 spin_unlock(&inode->lock);
2487 * We don't need to be under the protection of the inode's lock,
2488 * because we are called while holding the inode's io_tree lock
2489 * and are therefore protected against concurrent calls of this
2490 * function and btrfs_clear_delalloc_extent().
2492 if (!btrfs_is_free_space_inode(inode) && prev_delalloc_bytes == 0)
2493 btrfs_add_delalloc_inode(inode);
2496 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2497 (bits & EXTENT_DELALLOC_NEW)) {
2498 spin_lock(&inode->lock);
2499 inode->new_delalloc_bytes += state->end + 1 - state->start;
2500 spin_unlock(&inode->lock);
2505 * Once a range is no longer delalloc this function ensures that proper
2506 * accounting happens.
2508 void btrfs_clear_delalloc_extent(struct btrfs_inode *inode,
2509 struct extent_state *state, u32 bits)
2511 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2512 u64 len = state->end + 1 - state->start;
2513 u32 num_extents = count_max_extents(fs_info, len);
2515 lockdep_assert_held(&inode->io_tree.lock);
2517 if ((state->state & EXTENT_DEFRAG) && (bits & EXTENT_DEFRAG)) {
2518 spin_lock(&inode->lock);
2519 inode->defrag_bytes -= len;
2520 spin_unlock(&inode->lock);
2524 * set_bit and clear bit hooks normally require _irqsave/restore
2525 * but in this case, we are only testing for the DELALLOC
2526 * bit, which is only set or cleared with irqs on
2528 if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2529 struct btrfs_root *root = inode->root;
2530 u64 new_delalloc_bytes;
2532 spin_lock(&inode->lock);
2533 btrfs_mod_outstanding_extents(inode, -num_extents);
2534 spin_unlock(&inode->lock);
2537 * We don't reserve metadata space for space cache inodes so we
2538 * don't need to call delalloc_release_metadata if there is an
2541 if (bits & EXTENT_CLEAR_META_RESV &&
2542 root != fs_info->tree_root)
2543 btrfs_delalloc_release_metadata(inode, len, false);
2545 /* For sanity tests. */
2546 if (btrfs_is_testing(fs_info))
2549 if (!btrfs_is_data_reloc_root(root) &&
2550 !btrfs_is_free_space_inode(inode) &&
2551 !(state->state & EXTENT_NORESERVE) &&
2552 (bits & EXTENT_CLEAR_DATA_RESV))
2553 btrfs_free_reserved_data_space_noquota(fs_info, len);
2555 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2556 fs_info->delalloc_batch);
2557 spin_lock(&inode->lock);
2558 inode->delalloc_bytes -= len;
2559 new_delalloc_bytes = inode->delalloc_bytes;
2560 spin_unlock(&inode->lock);
2563 * We don't need to be under the protection of the inode's lock,
2564 * because we are called while holding the inode's io_tree lock
2565 * and are therefore protected against concurrent calls of this
2566 * function and btrfs_set_delalloc_extent().
2568 if (!btrfs_is_free_space_inode(inode) && new_delalloc_bytes == 0)
2569 btrfs_del_delalloc_inode(inode);
2572 if ((state->state & EXTENT_DELALLOC_NEW) &&
2573 (bits & EXTENT_DELALLOC_NEW)) {
2574 spin_lock(&inode->lock);
2575 ASSERT(inode->new_delalloc_bytes >= len);
2576 inode->new_delalloc_bytes -= len;
2577 if (bits & EXTENT_ADD_INODE_BYTES)
2578 inode_add_bytes(&inode->vfs_inode, len);
2579 spin_unlock(&inode->lock);
2583 static int btrfs_extract_ordered_extent(struct btrfs_bio *bbio,
2584 struct btrfs_ordered_extent *ordered)
2586 u64 start = (u64)bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT;
2587 u64 len = bbio->bio.bi_iter.bi_size;
2588 struct btrfs_ordered_extent *new;
2591 /* Must always be called for the beginning of an ordered extent. */
2592 if (WARN_ON_ONCE(start != ordered->disk_bytenr))
2595 /* No need to split if the ordered extent covers the entire bio. */
2596 if (ordered->disk_num_bytes == len) {
2597 refcount_inc(&ordered->refs);
2598 bbio->ordered = ordered;
2603 * Don't split the extent_map for NOCOW extents, as we're writing into
2604 * a pre-existing one.
2606 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
2607 ret = split_extent_map(bbio->inode, bbio->file_offset,
2608 ordered->num_bytes, len,
2609 ordered->disk_bytenr);
2614 new = btrfs_split_ordered_extent(ordered, len);
2616 return PTR_ERR(new);
2617 bbio->ordered = new;
2622 * given a list of ordered sums record them in the inode. This happens
2623 * at IO completion time based on sums calculated at bio submission time.
2625 static int add_pending_csums(struct btrfs_trans_handle *trans,
2626 struct list_head *list)
2628 struct btrfs_ordered_sum *sum;
2629 struct btrfs_root *csum_root = NULL;
2632 list_for_each_entry(sum, list, list) {
2633 trans->adding_csums = true;
2635 csum_root = btrfs_csum_root(trans->fs_info,
2637 ret = btrfs_csum_file_blocks(trans, csum_root, sum);
2638 trans->adding_csums = false;
2645 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2648 struct extent_state **cached_state)
2650 u64 search_start = start;
2651 const u64 end = start + len - 1;
2653 while (search_start < end) {
2654 const u64 search_len = end - search_start + 1;
2655 struct extent_map *em;
2659 em = btrfs_get_extent(inode, NULL, search_start, search_len);
2663 if (em->block_start != EXTENT_MAP_HOLE)
2667 if (em->start < search_start)
2668 em_len -= search_start - em->start;
2669 if (em_len > search_len)
2670 em_len = search_len;
2672 ret = set_extent_bit(&inode->io_tree, search_start,
2673 search_start + em_len - 1,
2674 EXTENT_DELALLOC_NEW, cached_state);
2676 search_start = extent_map_end(em);
2677 free_extent_map(em);
2684 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2685 unsigned int extra_bits,
2686 struct extent_state **cached_state)
2688 WARN_ON(PAGE_ALIGNED(end));
2690 if (start >= i_size_read(&inode->vfs_inode) &&
2691 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2693 * There can't be any extents following eof in this case so just
2694 * set the delalloc new bit for the range directly.
2696 extra_bits |= EXTENT_DELALLOC_NEW;
2700 ret = btrfs_find_new_delalloc_bytes(inode, start,
2707 return set_extent_bit(&inode->io_tree, start, end,
2708 EXTENT_DELALLOC | extra_bits, cached_state);
2711 /* see btrfs_writepage_start_hook for details on why this is required */
2712 struct btrfs_writepage_fixup {
2714 struct btrfs_inode *inode;
2715 struct btrfs_work work;
2718 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2720 struct btrfs_writepage_fixup *fixup =
2721 container_of(work, struct btrfs_writepage_fixup, work);
2722 struct btrfs_ordered_extent *ordered;
2723 struct extent_state *cached_state = NULL;
2724 struct extent_changeset *data_reserved = NULL;
2725 struct page *page = fixup->page;
2726 struct btrfs_inode *inode = fixup->inode;
2727 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2728 u64 page_start = page_offset(page);
2729 u64 page_end = page_offset(page) + PAGE_SIZE - 1;
2731 bool free_delalloc_space = true;
2734 * This is similar to page_mkwrite, we need to reserve the space before
2735 * we take the page lock.
2737 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2743 * Before we queued this fixup, we took a reference on the page.
2744 * page->mapping may go NULL, but it shouldn't be moved to a different
2747 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2749 * Unfortunately this is a little tricky, either
2751 * 1) We got here and our page had already been dealt with and
2752 * we reserved our space, thus ret == 0, so we need to just
2753 * drop our space reservation and bail. This can happen the
2754 * first time we come into the fixup worker, or could happen
2755 * while waiting for the ordered extent.
2756 * 2) Our page was already dealt with, but we happened to get an
2757 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2758 * this case we obviously don't have anything to release, but
2759 * because the page was already dealt with we don't want to
2760 * mark the page with an error, so make sure we're resetting
2761 * ret to 0. This is why we have this check _before_ the ret
2762 * check, because we do not want to have a surprise ENOSPC
2763 * when the page was already properly dealt with.
2766 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2767 btrfs_delalloc_release_space(inode, data_reserved,
2768 page_start, PAGE_SIZE,
2776 * We can't mess with the page state unless it is locked, so now that
2777 * it is locked bail if we failed to make our space reservation.
2782 lock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2784 /* already ordered? We're done */
2785 if (PageOrdered(page))
2788 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2790 unlock_extent(&inode->io_tree, page_start, page_end,
2793 btrfs_start_ordered_extent(ordered);
2794 btrfs_put_ordered_extent(ordered);
2798 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2804 * Everything went as planned, we're now the owner of a dirty page with
2805 * delayed allocation bits set and space reserved for our COW
2808 * The page was dirty when we started, nothing should have cleaned it.
2810 BUG_ON(!PageDirty(page));
2811 free_delalloc_space = false;
2813 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2814 if (free_delalloc_space)
2815 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2817 unlock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2821 * We hit ENOSPC or other errors. Update the mapping and page
2822 * to reflect the errors and clean the page.
2824 mapping_set_error(page->mapping, ret);
2825 btrfs_mark_ordered_io_finished(inode, page, page_start,
2827 clear_page_dirty_for_io(page);
2829 btrfs_folio_clear_checked(fs_info, page_folio(page), page_start, PAGE_SIZE);
2833 extent_changeset_free(data_reserved);
2835 * As a precaution, do a delayed iput in case it would be the last iput
2836 * that could need flushing space. Recursing back to fixup worker would
2839 btrfs_add_delayed_iput(inode);
2843 * There are a few paths in the higher layers of the kernel that directly
2844 * set the page dirty bit without asking the filesystem if it is a
2845 * good idea. This causes problems because we want to make sure COW
2846 * properly happens and the data=ordered rules are followed.
2848 * In our case any range that doesn't have the ORDERED bit set
2849 * hasn't been properly setup for IO. We kick off an async process
2850 * to fix it up. The async helper will wait for ordered extents, set
2851 * the delalloc bit and make it safe to write the page.
2853 int btrfs_writepage_cow_fixup(struct page *page)
2855 struct inode *inode = page->mapping->host;
2856 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
2857 struct btrfs_writepage_fixup *fixup;
2859 /* This page has ordered extent covering it already */
2860 if (PageOrdered(page))
2864 * PageChecked is set below when we create a fixup worker for this page,
2865 * don't try to create another one if we're already PageChecked()
2867 * The extent_io writepage code will redirty the page if we send back
2870 if (PageChecked(page))
2873 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2878 * We are already holding a reference to this inode from
2879 * write_cache_pages. We need to hold it because the space reservation
2880 * takes place outside of the page lock, and we can't trust
2881 * page->mapping outside of the page lock.
2884 btrfs_folio_set_checked(fs_info, page_folio(page), page_offset(page), PAGE_SIZE);
2886 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL);
2888 fixup->inode = BTRFS_I(inode);
2889 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2894 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2895 struct btrfs_inode *inode, u64 file_pos,
2896 struct btrfs_file_extent_item *stack_fi,
2897 const bool update_inode_bytes,
2898 u64 qgroup_reserved)
2900 struct btrfs_root *root = inode->root;
2901 const u64 sectorsize = root->fs_info->sectorsize;
2902 struct btrfs_path *path;
2903 struct extent_buffer *leaf;
2904 struct btrfs_key ins;
2905 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
2906 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
2907 u64 offset = btrfs_stack_file_extent_offset(stack_fi);
2908 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
2909 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
2910 struct btrfs_drop_extents_args drop_args = { 0 };
2913 path = btrfs_alloc_path();
2918 * we may be replacing one extent in the tree with another.
2919 * The new extent is pinned in the extent map, and we don't want
2920 * to drop it from the cache until it is completely in the btree.
2922 * So, tell btrfs_drop_extents to leave this extent in the cache.
2923 * the caller is expected to unpin it and allow it to be merged
2926 drop_args.path = path;
2927 drop_args.start = file_pos;
2928 drop_args.end = file_pos + num_bytes;
2929 drop_args.replace_extent = true;
2930 drop_args.extent_item_size = sizeof(*stack_fi);
2931 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2935 if (!drop_args.extent_inserted) {
2936 ins.objectid = btrfs_ino(inode);
2937 ins.offset = file_pos;
2938 ins.type = BTRFS_EXTENT_DATA_KEY;
2940 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2945 leaf = path->nodes[0];
2946 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
2947 write_extent_buffer(leaf, stack_fi,
2948 btrfs_item_ptr_offset(leaf, path->slots[0]),
2949 sizeof(struct btrfs_file_extent_item));
2951 btrfs_mark_buffer_dirty(trans, leaf);
2952 btrfs_release_path(path);
2955 * If we dropped an inline extent here, we know the range where it is
2956 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
2957 * number of bytes only for that range containing the inline extent.
2958 * The remaining of the range will be processed when clearning the
2959 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
2961 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
2962 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
2964 inline_size = drop_args.bytes_found - inline_size;
2965 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
2966 drop_args.bytes_found -= inline_size;
2967 num_bytes -= sectorsize;
2970 if (update_inode_bytes)
2971 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
2973 ins.objectid = disk_bytenr;
2974 ins.offset = disk_num_bytes;
2975 ins.type = BTRFS_EXTENT_ITEM_KEY;
2977 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
2981 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
2983 qgroup_reserved, &ins);
2985 btrfs_free_path(path);
2990 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2993 struct btrfs_block_group *cache;
2995 cache = btrfs_lookup_block_group(fs_info, start);
2998 spin_lock(&cache->lock);
2999 cache->delalloc_bytes -= len;
3000 spin_unlock(&cache->lock);
3002 btrfs_put_block_group(cache);
3005 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
3006 struct btrfs_ordered_extent *oe)
3008 struct btrfs_file_extent_item stack_fi;
3009 bool update_inode_bytes;
3010 u64 num_bytes = oe->num_bytes;
3011 u64 ram_bytes = oe->ram_bytes;
3013 memset(&stack_fi, 0, sizeof(stack_fi));
3014 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
3015 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
3016 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
3017 oe->disk_num_bytes);
3018 btrfs_set_stack_file_extent_offset(&stack_fi, oe->offset);
3019 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags)) {
3020 num_bytes = oe->truncated_len;
3021 ram_bytes = num_bytes;
3023 btrfs_set_stack_file_extent_num_bytes(&stack_fi, num_bytes);
3024 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, ram_bytes);
3025 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
3026 /* Encryption and other encoding is reserved and all 0 */
3029 * For delalloc, when completing an ordered extent we update the inode's
3030 * bytes when clearing the range in the inode's io tree, so pass false
3031 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
3032 * except if the ordered extent was truncated.
3034 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
3035 test_bit(BTRFS_ORDERED_ENCODED, &oe->flags) ||
3036 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
3038 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
3039 oe->file_offset, &stack_fi,
3040 update_inode_bytes, oe->qgroup_rsv);
3044 * As ordered data IO finishes, this gets called so we can finish
3045 * an ordered extent if the range of bytes in the file it covers are
3048 int btrfs_finish_one_ordered(struct btrfs_ordered_extent *ordered_extent)
3050 struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
3051 struct btrfs_root *root = inode->root;
3052 struct btrfs_fs_info *fs_info = root->fs_info;
3053 struct btrfs_trans_handle *trans = NULL;
3054 struct extent_io_tree *io_tree = &inode->io_tree;
3055 struct extent_state *cached_state = NULL;
3057 int compress_type = 0;
3059 u64 logical_len = ordered_extent->num_bytes;
3060 bool freespace_inode;
3061 bool truncated = false;
3062 bool clear_reserved_extent = true;
3063 unsigned int clear_bits = EXTENT_DEFRAG;
3065 start = ordered_extent->file_offset;
3066 end = start + ordered_extent->num_bytes - 1;
3068 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3069 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
3070 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags) &&
3071 !test_bit(BTRFS_ORDERED_ENCODED, &ordered_extent->flags))
3072 clear_bits |= EXTENT_DELALLOC_NEW;
3074 freespace_inode = btrfs_is_free_space_inode(inode);
3075 if (!freespace_inode)
3076 btrfs_lockdep_acquire(fs_info, btrfs_ordered_extent);
3078 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
3083 if (btrfs_is_zoned(fs_info))
3084 btrfs_zone_finish_endio(fs_info, ordered_extent->disk_bytenr,
3085 ordered_extent->disk_num_bytes);
3087 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
3089 logical_len = ordered_extent->truncated_len;
3090 /* Truncated the entire extent, don't bother adding */
3095 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
3096 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
3098 btrfs_inode_safe_disk_i_size_write(inode, 0);
3099 if (freespace_inode)
3100 trans = btrfs_join_transaction_spacecache(root);
3102 trans = btrfs_join_transaction(root);
3103 if (IS_ERR(trans)) {
3104 ret = PTR_ERR(trans);
3108 trans->block_rsv = &inode->block_rsv;
3109 ret = btrfs_update_inode_fallback(trans, inode);
3110 if (ret) /* -ENOMEM or corruption */
3111 btrfs_abort_transaction(trans, ret);
3115 clear_bits |= EXTENT_LOCKED;
3116 lock_extent(io_tree, start, end, &cached_state);
3118 if (freespace_inode)
3119 trans = btrfs_join_transaction_spacecache(root);
3121 trans = btrfs_join_transaction(root);
3122 if (IS_ERR(trans)) {
3123 ret = PTR_ERR(trans);
3128 trans->block_rsv = &inode->block_rsv;
3130 ret = btrfs_insert_raid_extent(trans, ordered_extent);
3134 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3135 compress_type = ordered_extent->compress_type;
3136 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3137 BUG_ON(compress_type);
3138 ret = btrfs_mark_extent_written(trans, inode,
3139 ordered_extent->file_offset,
3140 ordered_extent->file_offset +
3142 btrfs_zoned_release_data_reloc_bg(fs_info, ordered_extent->disk_bytenr,
3143 ordered_extent->disk_num_bytes);
3145 BUG_ON(root == fs_info->tree_root);
3146 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
3148 clear_reserved_extent = false;
3149 btrfs_release_delalloc_bytes(fs_info,
3150 ordered_extent->disk_bytenr,
3151 ordered_extent->disk_num_bytes);
3155 btrfs_abort_transaction(trans, ret);
3159 ret = unpin_extent_cache(inode, ordered_extent->file_offset,
3160 ordered_extent->num_bytes, trans->transid);
3162 btrfs_abort_transaction(trans, ret);
3166 ret = add_pending_csums(trans, &ordered_extent->list);
3168 btrfs_abort_transaction(trans, ret);
3173 * If this is a new delalloc range, clear its new delalloc flag to
3174 * update the inode's number of bytes. This needs to be done first
3175 * before updating the inode item.
3177 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
3178 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
3179 clear_extent_bit(&inode->io_tree, start, end,
3180 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
3183 btrfs_inode_safe_disk_i_size_write(inode, 0);
3184 ret = btrfs_update_inode_fallback(trans, inode);
3185 if (ret) { /* -ENOMEM or corruption */
3186 btrfs_abort_transaction(trans, ret);
3191 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
3195 btrfs_end_transaction(trans);
3197 if (ret || truncated) {
3198 u64 unwritten_start = start;
3201 * If we failed to finish this ordered extent for any reason we
3202 * need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3203 * extent, and mark the inode with the error if it wasn't
3204 * already set. Any error during writeback would have already
3205 * set the mapping error, so we need to set it if we're the ones
3206 * marking this ordered extent as failed.
3208 if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR,
3209 &ordered_extent->flags))
3210 mapping_set_error(ordered_extent->inode->i_mapping, -EIO);
3213 unwritten_start += logical_len;
3214 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
3217 * Drop extent maps for the part of the extent we didn't write.
3219 * We have an exception here for the free_space_inode, this is
3220 * because when we do btrfs_get_extent() on the free space inode
3221 * we will search the commit root. If this is a new block group
3222 * we won't find anything, and we will trip over the assert in
3223 * writepage where we do ASSERT(em->block_start !=
3226 * Theoretically we could also skip this for any NOCOW extent as
3227 * we don't mess with the extent map tree in the NOCOW case, but
3228 * for now simply skip this if we are the free space inode.
3230 if (!btrfs_is_free_space_inode(inode))
3231 btrfs_drop_extent_map_range(inode, unwritten_start,
3235 * If the ordered extent had an IOERR or something else went
3236 * wrong we need to return the space for this ordered extent
3237 * back to the allocator. We only free the extent in the
3238 * truncated case if we didn't write out the extent at all.
3240 * If we made it past insert_reserved_file_extent before we
3241 * errored out then we don't need to do this as the accounting
3242 * has already been done.
3244 if ((ret || !logical_len) &&
3245 clear_reserved_extent &&
3246 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3247 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3249 * Discard the range before returning it back to the
3252 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3253 btrfs_discard_extent(fs_info,
3254 ordered_extent->disk_bytenr,
3255 ordered_extent->disk_num_bytes,
3257 btrfs_free_reserved_extent(fs_info,
3258 ordered_extent->disk_bytenr,
3259 ordered_extent->disk_num_bytes, 1);
3261 * Actually free the qgroup rsv which was released when
3262 * the ordered extent was created.
3264 btrfs_qgroup_free_refroot(fs_info, inode->root->root_key.objectid,
3265 ordered_extent->qgroup_rsv,
3266 BTRFS_QGROUP_RSV_DATA);
3271 * This needs to be done to make sure anybody waiting knows we are done
3272 * updating everything for this ordered extent.
3274 btrfs_remove_ordered_extent(inode, ordered_extent);
3277 btrfs_put_ordered_extent(ordered_extent);
3278 /* once for the tree */
3279 btrfs_put_ordered_extent(ordered_extent);
3284 int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered)
3286 if (btrfs_is_zoned(inode_to_fs_info(ordered->inode)) &&
3287 !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags) &&
3288 list_empty(&ordered->bioc_list))
3289 btrfs_finish_ordered_zoned(ordered);
3290 return btrfs_finish_one_ordered(ordered);
3294 * Verify the checksum for a single sector without any extra action that depend
3295 * on the type of I/O.
3297 int btrfs_check_sector_csum(struct btrfs_fs_info *fs_info, struct page *page,
3298 u32 pgoff, u8 *csum, const u8 * const csum_expected)
3300 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3303 ASSERT(pgoff + fs_info->sectorsize <= PAGE_SIZE);
3305 shash->tfm = fs_info->csum_shash;
3307 kaddr = kmap_local_page(page) + pgoff;
3308 crypto_shash_digest(shash, kaddr, fs_info->sectorsize, csum);
3309 kunmap_local(kaddr);
3311 if (memcmp(csum, csum_expected, fs_info->csum_size))
3317 * Verify the checksum of a single data sector.
3319 * @bbio: btrfs_io_bio which contains the csum
3320 * @dev: device the sector is on
3321 * @bio_offset: offset to the beginning of the bio (in bytes)
3322 * @bv: bio_vec to check
3324 * Check if the checksum on a data block is valid. When a checksum mismatch is
3325 * detected, report the error and fill the corrupted range with zero.
3327 * Return %true if the sector is ok or had no checksum to start with, else %false.
3329 bool btrfs_data_csum_ok(struct btrfs_bio *bbio, struct btrfs_device *dev,
3330 u32 bio_offset, struct bio_vec *bv)
3332 struct btrfs_inode *inode = bbio->inode;
3333 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3334 u64 file_offset = bbio->file_offset + bio_offset;
3335 u64 end = file_offset + bv->bv_len - 1;
3337 u8 csum[BTRFS_CSUM_SIZE];
3339 ASSERT(bv->bv_len == fs_info->sectorsize);
3344 if (btrfs_is_data_reloc_root(inode->root) &&
3345 test_range_bit(&inode->io_tree, file_offset, end, EXTENT_NODATASUM,
3347 /* Skip the range without csum for data reloc inode */
3348 clear_extent_bits(&inode->io_tree, file_offset, end,
3353 csum_expected = bbio->csum + (bio_offset >> fs_info->sectorsize_bits) *
3355 if (btrfs_check_sector_csum(fs_info, bv->bv_page, bv->bv_offset, csum,
3361 btrfs_print_data_csum_error(inode, file_offset, csum, csum_expected,
3364 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS);
3370 * Perform a delayed iput on @inode.
3372 * @inode: The inode we want to perform iput on
3374 * This function uses the generic vfs_inode::i_count to track whether we should
3375 * just decrement it (in case it's > 1) or if this is the last iput then link
3376 * the inode to the delayed iput machinery. Delayed iputs are processed at
3377 * transaction commit time/superblock commit/cleaner kthread.
3379 void btrfs_add_delayed_iput(struct btrfs_inode *inode)
3381 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3382 unsigned long flags;
3384 if (atomic_add_unless(&inode->vfs_inode.i_count, -1, 1))
3387 atomic_inc(&fs_info->nr_delayed_iputs);
3389 * Need to be irq safe here because we can be called from either an irq
3390 * context (see bio.c and btrfs_put_ordered_extent()) or a non-irq
3393 spin_lock_irqsave(&fs_info->delayed_iput_lock, flags);
3394 ASSERT(list_empty(&inode->delayed_iput));
3395 list_add_tail(&inode->delayed_iput, &fs_info->delayed_iputs);
3396 spin_unlock_irqrestore(&fs_info->delayed_iput_lock, flags);
3397 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3398 wake_up_process(fs_info->cleaner_kthread);
3401 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3402 struct btrfs_inode *inode)
3404 list_del_init(&inode->delayed_iput);
3405 spin_unlock_irq(&fs_info->delayed_iput_lock);
3406 iput(&inode->vfs_inode);
3407 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3408 wake_up(&fs_info->delayed_iputs_wait);
3409 spin_lock_irq(&fs_info->delayed_iput_lock);
3412 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3413 struct btrfs_inode *inode)
3415 if (!list_empty(&inode->delayed_iput)) {
3416 spin_lock_irq(&fs_info->delayed_iput_lock);
3417 if (!list_empty(&inode->delayed_iput))
3418 run_delayed_iput_locked(fs_info, inode);
3419 spin_unlock_irq(&fs_info->delayed_iput_lock);
3423 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3426 * btrfs_put_ordered_extent() can run in irq context (see bio.c), which
3427 * calls btrfs_add_delayed_iput() and that needs to lock
3428 * fs_info->delayed_iput_lock. So we need to disable irqs here to
3429 * prevent a deadlock.
3431 spin_lock_irq(&fs_info->delayed_iput_lock);
3432 while (!list_empty(&fs_info->delayed_iputs)) {
3433 struct btrfs_inode *inode;
3435 inode = list_first_entry(&fs_info->delayed_iputs,
3436 struct btrfs_inode, delayed_iput);
3437 run_delayed_iput_locked(fs_info, inode);
3438 if (need_resched()) {
3439 spin_unlock_irq(&fs_info->delayed_iput_lock);
3441 spin_lock_irq(&fs_info->delayed_iput_lock);
3444 spin_unlock_irq(&fs_info->delayed_iput_lock);
3448 * Wait for flushing all delayed iputs
3450 * @fs_info: the filesystem
3452 * This will wait on any delayed iputs that are currently running with KILLABLE
3453 * set. Once they are all done running we will return, unless we are killed in
3454 * which case we return EINTR. This helps in user operations like fallocate etc
3455 * that might get blocked on the iputs.
3457 * Return EINTR if we were killed, 0 if nothing's pending
3459 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3461 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3462 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3469 * This creates an orphan entry for the given inode in case something goes wrong
3470 * in the middle of an unlink.
3472 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3473 struct btrfs_inode *inode)
3477 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3478 if (ret && ret != -EEXIST) {
3479 btrfs_abort_transaction(trans, ret);
3487 * We have done the delete so we can go ahead and remove the orphan item for
3488 * this particular inode.
3490 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3491 struct btrfs_inode *inode)
3493 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3497 * this cleans up any orphans that may be left on the list from the last use
3500 int btrfs_orphan_cleanup(struct btrfs_root *root)
3502 struct btrfs_fs_info *fs_info = root->fs_info;
3503 struct btrfs_path *path;
3504 struct extent_buffer *leaf;
3505 struct btrfs_key key, found_key;
3506 struct btrfs_trans_handle *trans;
3507 struct inode *inode;
3508 u64 last_objectid = 0;
3509 int ret = 0, nr_unlink = 0;
3511 if (test_and_set_bit(BTRFS_ROOT_ORPHAN_CLEANUP, &root->state))
3514 path = btrfs_alloc_path();
3519 path->reada = READA_BACK;
3521 key.objectid = BTRFS_ORPHAN_OBJECTID;
3522 key.type = BTRFS_ORPHAN_ITEM_KEY;
3523 key.offset = (u64)-1;
3526 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3531 * if ret == 0 means we found what we were searching for, which
3532 * is weird, but possible, so only screw with path if we didn't
3533 * find the key and see if we have stuff that matches
3537 if (path->slots[0] == 0)
3542 /* pull out the item */
3543 leaf = path->nodes[0];
3544 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3546 /* make sure the item matches what we want */
3547 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3549 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3552 /* release the path since we're done with it */
3553 btrfs_release_path(path);
3556 * this is where we are basically btrfs_lookup, without the
3557 * crossing root thing. we store the inode number in the
3558 * offset of the orphan item.
3561 if (found_key.offset == last_objectid) {
3563 * We found the same inode as before. This means we were
3564 * not able to remove its items via eviction triggered
3565 * by an iput(). A transaction abort may have happened,
3566 * due to -ENOSPC for example, so try to grab the error
3567 * that lead to a transaction abort, if any.
3570 "Error removing orphan entry, stopping orphan cleanup");
3571 ret = BTRFS_FS_ERROR(fs_info) ?: -EINVAL;
3575 last_objectid = found_key.offset;
3577 found_key.objectid = found_key.offset;
3578 found_key.type = BTRFS_INODE_ITEM_KEY;
3579 found_key.offset = 0;
3580 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3581 if (IS_ERR(inode)) {
3582 ret = PTR_ERR(inode);
3588 if (!inode && root == fs_info->tree_root) {
3589 struct btrfs_root *dead_root;
3590 int is_dead_root = 0;
3593 * This is an orphan in the tree root. Currently these
3594 * could come from 2 sources:
3595 * a) a root (snapshot/subvolume) deletion in progress
3596 * b) a free space cache inode
3597 * We need to distinguish those two, as the orphan item
3598 * for a root must not get deleted before the deletion
3599 * of the snapshot/subvolume's tree completes.
3601 * btrfs_find_orphan_roots() ran before us, which has
3602 * found all deleted roots and loaded them into
3603 * fs_info->fs_roots_radix. So here we can find if an
3604 * orphan item corresponds to a deleted root by looking
3605 * up the root from that radix tree.
3608 spin_lock(&fs_info->fs_roots_radix_lock);
3609 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3610 (unsigned long)found_key.objectid);
3611 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3613 spin_unlock(&fs_info->fs_roots_radix_lock);
3616 /* prevent this orphan from being found again */
3617 key.offset = found_key.objectid - 1;
3624 * If we have an inode with links, there are a couple of
3627 * 1. We were halfway through creating fsverity metadata for the
3628 * file. In that case, the orphan item represents incomplete
3629 * fsverity metadata which must be cleaned up with
3630 * btrfs_drop_verity_items and deleting the orphan item.
3632 * 2. Old kernels (before v3.12) used to create an
3633 * orphan item for truncate indicating that there were possibly
3634 * extent items past i_size that needed to be deleted. In v3.12,
3635 * truncate was changed to update i_size in sync with the extent
3636 * items, but the (useless) orphan item was still created. Since
3637 * v4.18, we don't create the orphan item for truncate at all.
3639 * So, this item could mean that we need to do a truncate, but
3640 * only if this filesystem was last used on a pre-v3.12 kernel
3641 * and was not cleanly unmounted. The odds of that are quite
3642 * slim, and it's a pain to do the truncate now, so just delete
3645 * It's also possible that this orphan item was supposed to be
3646 * deleted but wasn't. The inode number may have been reused,
3647 * but either way, we can delete the orphan item.
3649 if (!inode || inode->i_nlink) {
3651 ret = btrfs_drop_verity_items(BTRFS_I(inode));
3657 trans = btrfs_start_transaction(root, 1);
3658 if (IS_ERR(trans)) {
3659 ret = PTR_ERR(trans);
3662 btrfs_debug(fs_info, "auto deleting %Lu",
3663 found_key.objectid);
3664 ret = btrfs_del_orphan_item(trans, root,
3665 found_key.objectid);
3666 btrfs_end_transaction(trans);
3674 /* this will do delete_inode and everything for us */
3677 /* release the path since we're done with it */
3678 btrfs_release_path(path);
3680 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3681 trans = btrfs_join_transaction(root);
3683 btrfs_end_transaction(trans);
3687 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3691 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3692 btrfs_free_path(path);
3697 * very simple check to peek ahead in the leaf looking for xattrs. If we
3698 * don't find any xattrs, we know there can't be any acls.
3700 * slot is the slot the inode is in, objectid is the objectid of the inode
3702 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3703 int slot, u64 objectid,
3704 int *first_xattr_slot)
3706 u32 nritems = btrfs_header_nritems(leaf);
3707 struct btrfs_key found_key;
3708 static u64 xattr_access = 0;
3709 static u64 xattr_default = 0;
3712 if (!xattr_access) {
3713 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3714 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3715 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3716 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3720 *first_xattr_slot = -1;
3721 while (slot < nritems) {
3722 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3724 /* we found a different objectid, there must not be acls */
3725 if (found_key.objectid != objectid)
3728 /* we found an xattr, assume we've got an acl */
3729 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3730 if (*first_xattr_slot == -1)
3731 *first_xattr_slot = slot;
3732 if (found_key.offset == xattr_access ||
3733 found_key.offset == xattr_default)
3738 * we found a key greater than an xattr key, there can't
3739 * be any acls later on
3741 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3748 * it goes inode, inode backrefs, xattrs, extents,
3749 * so if there are a ton of hard links to an inode there can
3750 * be a lot of backrefs. Don't waste time searching too hard,
3751 * this is just an optimization
3756 /* we hit the end of the leaf before we found an xattr or
3757 * something larger than an xattr. We have to assume the inode
3760 if (*first_xattr_slot == -1)
3761 *first_xattr_slot = slot;
3766 * read an inode from the btree into the in-memory inode
3768 static int btrfs_read_locked_inode(struct inode *inode,
3769 struct btrfs_path *in_path)
3771 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
3772 struct btrfs_path *path = in_path;
3773 struct extent_buffer *leaf;
3774 struct btrfs_inode_item *inode_item;
3775 struct btrfs_root *root = BTRFS_I(inode)->root;
3776 struct btrfs_key location;
3781 bool filled = false;
3782 int first_xattr_slot;
3784 ret = btrfs_fill_inode(inode, &rdev);
3789 path = btrfs_alloc_path();
3794 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3796 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3798 if (path != in_path)
3799 btrfs_free_path(path);
3803 leaf = path->nodes[0];
3808 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3809 struct btrfs_inode_item);
3810 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3811 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3812 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3813 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3814 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3815 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3816 round_up(i_size_read(inode), fs_info->sectorsize));
3818 inode_set_atime(inode, btrfs_timespec_sec(leaf, &inode_item->atime),
3819 btrfs_timespec_nsec(leaf, &inode_item->atime));
3821 inode_set_mtime(inode, btrfs_timespec_sec(leaf, &inode_item->mtime),
3822 btrfs_timespec_nsec(leaf, &inode_item->mtime));
3824 inode_set_ctime(inode, btrfs_timespec_sec(leaf, &inode_item->ctime),
3825 btrfs_timespec_nsec(leaf, &inode_item->ctime));
3827 BTRFS_I(inode)->i_otime_sec = btrfs_timespec_sec(leaf, &inode_item->otime);
3828 BTRFS_I(inode)->i_otime_nsec = btrfs_timespec_nsec(leaf, &inode_item->otime);
3830 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3831 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3832 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3834 inode_set_iversion_queried(inode,
3835 btrfs_inode_sequence(leaf, inode_item));
3836 inode->i_generation = BTRFS_I(inode)->generation;
3838 rdev = btrfs_inode_rdev(leaf, inode_item);
3840 BTRFS_I(inode)->index_cnt = (u64)-1;
3841 btrfs_inode_split_flags(btrfs_inode_flags(leaf, inode_item),
3842 &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags);
3846 * If we were modified in the current generation and evicted from memory
3847 * and then re-read we need to do a full sync since we don't have any
3848 * idea about which extents were modified before we were evicted from
3851 * This is required for both inode re-read from disk and delayed inode
3852 * in the delayed_nodes xarray.
3854 if (BTRFS_I(inode)->last_trans == btrfs_get_fs_generation(fs_info))
3855 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3856 &BTRFS_I(inode)->runtime_flags);
3859 * We don't persist the id of the transaction where an unlink operation
3860 * against the inode was last made. So here we assume the inode might
3861 * have been evicted, and therefore the exact value of last_unlink_trans
3862 * lost, and set it to last_trans to avoid metadata inconsistencies
3863 * between the inode and its parent if the inode is fsync'ed and the log
3864 * replayed. For example, in the scenario:
3867 * ln mydir/foo mydir/bar
3870 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3871 * xfs_io -c fsync mydir/foo
3873 * mount fs, triggers fsync log replay
3875 * We must make sure that when we fsync our inode foo we also log its
3876 * parent inode, otherwise after log replay the parent still has the
3877 * dentry with the "bar" name but our inode foo has a link count of 1
3878 * and doesn't have an inode ref with the name "bar" anymore.
3880 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3881 * but it guarantees correctness at the expense of occasional full
3882 * transaction commits on fsync if our inode is a directory, or if our
3883 * inode is not a directory, logging its parent unnecessarily.
3885 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3888 * Same logic as for last_unlink_trans. We don't persist the generation
3889 * of the last transaction where this inode was used for a reflink
3890 * operation, so after eviction and reloading the inode we must be
3891 * pessimistic and assume the last transaction that modified the inode.
3893 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3896 if (inode->i_nlink != 1 ||
3897 path->slots[0] >= btrfs_header_nritems(leaf))
3900 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3901 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3904 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3905 if (location.type == BTRFS_INODE_REF_KEY) {
3906 struct btrfs_inode_ref *ref;
3908 ref = (struct btrfs_inode_ref *)ptr;
3909 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3910 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3911 struct btrfs_inode_extref *extref;
3913 extref = (struct btrfs_inode_extref *)ptr;
3914 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3919 * try to precache a NULL acl entry for files that don't have
3920 * any xattrs or acls
3922 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3923 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3924 if (first_xattr_slot != -1) {
3925 path->slots[0] = first_xattr_slot;
3926 ret = btrfs_load_inode_props(inode, path);
3929 "error loading props for ino %llu (root %llu): %d",
3930 btrfs_ino(BTRFS_I(inode)),
3931 root->root_key.objectid, ret);
3933 if (path != in_path)
3934 btrfs_free_path(path);
3937 cache_no_acl(inode);
3939 switch (inode->i_mode & S_IFMT) {
3941 inode->i_mapping->a_ops = &btrfs_aops;
3942 inode->i_fop = &btrfs_file_operations;
3943 inode->i_op = &btrfs_file_inode_operations;
3946 inode->i_fop = &btrfs_dir_file_operations;
3947 inode->i_op = &btrfs_dir_inode_operations;
3950 inode->i_op = &btrfs_symlink_inode_operations;
3951 inode_nohighmem(inode);
3952 inode->i_mapping->a_ops = &btrfs_aops;
3955 inode->i_op = &btrfs_special_inode_operations;
3956 init_special_inode(inode, inode->i_mode, rdev);
3960 btrfs_sync_inode_flags_to_i_flags(inode);
3965 * given a leaf and an inode, copy the inode fields into the leaf
3967 static void fill_inode_item(struct btrfs_trans_handle *trans,
3968 struct extent_buffer *leaf,
3969 struct btrfs_inode_item *item,
3970 struct inode *inode)
3972 struct btrfs_map_token token;
3975 btrfs_init_map_token(&token, leaf);
3977 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3978 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3979 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
3980 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3981 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3983 btrfs_set_token_timespec_sec(&token, &item->atime,
3984 inode_get_atime_sec(inode));
3985 btrfs_set_token_timespec_nsec(&token, &item->atime,
3986 inode_get_atime_nsec(inode));
3988 btrfs_set_token_timespec_sec(&token, &item->mtime,
3989 inode_get_mtime_sec(inode));
3990 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3991 inode_get_mtime_nsec(inode));
3993 btrfs_set_token_timespec_sec(&token, &item->ctime,
3994 inode_get_ctime_sec(inode));
3995 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3996 inode_get_ctime_nsec(inode));
3998 btrfs_set_token_timespec_sec(&token, &item->otime, BTRFS_I(inode)->i_otime_sec);
3999 btrfs_set_token_timespec_nsec(&token, &item->otime, BTRFS_I(inode)->i_otime_nsec);
4001 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
4002 btrfs_set_token_inode_generation(&token, item,
4003 BTRFS_I(inode)->generation);
4004 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
4005 btrfs_set_token_inode_transid(&token, item, trans->transid);
4006 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
4007 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
4008 BTRFS_I(inode)->ro_flags);
4009 btrfs_set_token_inode_flags(&token, item, flags);
4010 btrfs_set_token_inode_block_group(&token, item, 0);
4014 * copy everything in the in-memory inode into the btree.
4016 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
4017 struct btrfs_inode *inode)
4019 struct btrfs_inode_item *inode_item;
4020 struct btrfs_path *path;
4021 struct extent_buffer *leaf;
4024 path = btrfs_alloc_path();
4028 ret = btrfs_lookup_inode(trans, inode->root, path, &inode->location, 1);
4035 leaf = path->nodes[0];
4036 inode_item = btrfs_item_ptr(leaf, path->slots[0],
4037 struct btrfs_inode_item);
4039 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
4040 btrfs_mark_buffer_dirty(trans, leaf);
4041 btrfs_set_inode_last_trans(trans, inode);
4044 btrfs_free_path(path);
4049 * copy everything in the in-memory inode into the btree.
4051 int btrfs_update_inode(struct btrfs_trans_handle *trans,
4052 struct btrfs_inode *inode)
4054 struct btrfs_root *root = inode->root;
4055 struct btrfs_fs_info *fs_info = root->fs_info;
4059 * If the inode is a free space inode, we can deadlock during commit
4060 * if we put it into the delayed code.
4062 * The data relocation inode should also be directly updated
4065 if (!btrfs_is_free_space_inode(inode)
4066 && !btrfs_is_data_reloc_root(root)
4067 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4068 btrfs_update_root_times(trans, root);
4070 ret = btrfs_delayed_update_inode(trans, inode);
4072 btrfs_set_inode_last_trans(trans, inode);
4076 return btrfs_update_inode_item(trans, inode);
4079 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4080 struct btrfs_inode *inode)
4084 ret = btrfs_update_inode(trans, inode);
4086 return btrfs_update_inode_item(trans, inode);
4091 * unlink helper that gets used here in inode.c and in the tree logging
4092 * recovery code. It remove a link in a directory with a given name, and
4093 * also drops the back refs in the inode to the directory
4095 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4096 struct btrfs_inode *dir,
4097 struct btrfs_inode *inode,
4098 const struct fscrypt_str *name,
4099 struct btrfs_rename_ctx *rename_ctx)
4101 struct btrfs_root *root = dir->root;
4102 struct btrfs_fs_info *fs_info = root->fs_info;
4103 struct btrfs_path *path;
4105 struct btrfs_dir_item *di;
4107 u64 ino = btrfs_ino(inode);
4108 u64 dir_ino = btrfs_ino(dir);
4110 path = btrfs_alloc_path();
4116 di = btrfs_lookup_dir_item(trans, root, path, dir_ino, name, -1);
4117 if (IS_ERR_OR_NULL(di)) {
4118 ret = di ? PTR_ERR(di) : -ENOENT;
4121 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4124 btrfs_release_path(path);
4127 * If we don't have dir index, we have to get it by looking up
4128 * the inode ref, since we get the inode ref, remove it directly,
4129 * it is unnecessary to do delayed deletion.
4131 * But if we have dir index, needn't search inode ref to get it.
4132 * Since the inode ref is close to the inode item, it is better
4133 * that we delay to delete it, and just do this deletion when
4134 * we update the inode item.
4136 if (inode->dir_index) {
4137 ret = btrfs_delayed_delete_inode_ref(inode);
4139 index = inode->dir_index;
4144 ret = btrfs_del_inode_ref(trans, root, name, ino, dir_ino, &index);
4147 "failed to delete reference to %.*s, inode %llu parent %llu",
4148 name->len, name->name, ino, dir_ino);
4149 btrfs_abort_transaction(trans, ret);
4154 rename_ctx->index = index;
4156 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4158 btrfs_abort_transaction(trans, ret);
4163 * If we are in a rename context, we don't need to update anything in the
4164 * log. That will be done later during the rename by btrfs_log_new_name().
4165 * Besides that, doing it here would only cause extra unnecessary btree
4166 * operations on the log tree, increasing latency for applications.
4169 btrfs_del_inode_ref_in_log(trans, root, name, inode, dir_ino);
4170 btrfs_del_dir_entries_in_log(trans, root, name, dir, index);
4174 * If we have a pending delayed iput we could end up with the final iput
4175 * being run in btrfs-cleaner context. If we have enough of these built
4176 * up we can end up burning a lot of time in btrfs-cleaner without any
4177 * way to throttle the unlinks. Since we're currently holding a ref on
4178 * the inode we can run the delayed iput here without any issues as the
4179 * final iput won't be done until after we drop the ref we're currently
4182 btrfs_run_delayed_iput(fs_info, inode);
4184 btrfs_free_path(path);
4188 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name->len * 2);
4189 inode_inc_iversion(&inode->vfs_inode);
4190 inode_inc_iversion(&dir->vfs_inode);
4191 inode_set_mtime_to_ts(&dir->vfs_inode, inode_set_ctime_current(&dir->vfs_inode));
4192 ret = btrfs_update_inode(trans, dir);
4197 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4198 struct btrfs_inode *dir, struct btrfs_inode *inode,
4199 const struct fscrypt_str *name)
4203 ret = __btrfs_unlink_inode(trans, dir, inode, name, NULL);
4205 drop_nlink(&inode->vfs_inode);
4206 ret = btrfs_update_inode(trans, inode);
4212 * helper to start transaction for unlink and rmdir.
4214 * unlink and rmdir are special in btrfs, they do not always free space, so
4215 * if we cannot make our reservations the normal way try and see if there is
4216 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4217 * allow the unlink to occur.
4219 static struct btrfs_trans_handle *__unlink_start_trans(struct btrfs_inode *dir)
4221 struct btrfs_root *root = dir->root;
4223 return btrfs_start_transaction_fallback_global_rsv(root,
4224 BTRFS_UNLINK_METADATA_UNITS);
4227 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4229 struct btrfs_trans_handle *trans;
4230 struct inode *inode = d_inode(dentry);
4232 struct fscrypt_name fname;
4234 ret = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4238 /* This needs to handle no-key deletions later on */
4240 trans = __unlink_start_trans(BTRFS_I(dir));
4241 if (IS_ERR(trans)) {
4242 ret = PTR_ERR(trans);
4246 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4249 ret = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4254 if (inode->i_nlink == 0) {
4255 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4261 btrfs_end_transaction(trans);
4262 btrfs_btree_balance_dirty(BTRFS_I(dir)->root->fs_info);
4264 fscrypt_free_filename(&fname);
4268 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4269 struct btrfs_inode *dir, struct dentry *dentry)
4271 struct btrfs_root *root = dir->root;
4272 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4273 struct btrfs_path *path;
4274 struct extent_buffer *leaf;
4275 struct btrfs_dir_item *di;
4276 struct btrfs_key key;
4280 u64 dir_ino = btrfs_ino(dir);
4281 struct fscrypt_name fname;
4283 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
4287 /* This needs to handle no-key deletions later on */
4289 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4290 objectid = inode->root->root_key.objectid;
4291 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4292 objectid = inode->location.objectid;
4295 fscrypt_free_filename(&fname);
4299 path = btrfs_alloc_path();
4305 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4306 &fname.disk_name, -1);
4307 if (IS_ERR_OR_NULL(di)) {
4308 ret = di ? PTR_ERR(di) : -ENOENT;
4312 leaf = path->nodes[0];
4313 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4314 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4315 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4317 btrfs_abort_transaction(trans, ret);
4320 btrfs_release_path(path);
4323 * This is a placeholder inode for a subvolume we didn't have a
4324 * reference to at the time of the snapshot creation. In the meantime
4325 * we could have renamed the real subvol link into our snapshot, so
4326 * depending on btrfs_del_root_ref to return -ENOENT here is incorrect.
4327 * Instead simply lookup the dir_index_item for this entry so we can
4328 * remove it. Otherwise we know we have a ref to the root and we can
4329 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4331 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4332 di = btrfs_search_dir_index_item(root, path, dir_ino, &fname.disk_name);
4333 if (IS_ERR_OR_NULL(di)) {
4338 btrfs_abort_transaction(trans, ret);
4342 leaf = path->nodes[0];
4343 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4345 btrfs_release_path(path);
4347 ret = btrfs_del_root_ref(trans, objectid,
4348 root->root_key.objectid, dir_ino,
4349 &index, &fname.disk_name);
4351 btrfs_abort_transaction(trans, ret);
4356 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4358 btrfs_abort_transaction(trans, ret);
4362 btrfs_i_size_write(dir, dir->vfs_inode.i_size - fname.disk_name.len * 2);
4363 inode_inc_iversion(&dir->vfs_inode);
4364 inode_set_mtime_to_ts(&dir->vfs_inode, inode_set_ctime_current(&dir->vfs_inode));
4365 ret = btrfs_update_inode_fallback(trans, dir);
4367 btrfs_abort_transaction(trans, ret);
4369 btrfs_free_path(path);
4370 fscrypt_free_filename(&fname);
4375 * Helper to check if the subvolume references other subvolumes or if it's
4378 static noinline int may_destroy_subvol(struct btrfs_root *root)
4380 struct btrfs_fs_info *fs_info = root->fs_info;
4381 struct btrfs_path *path;
4382 struct btrfs_dir_item *di;
4383 struct btrfs_key key;
4384 struct fscrypt_str name = FSTR_INIT("default", 7);
4388 path = btrfs_alloc_path();
4392 /* Make sure this root isn't set as the default subvol */
4393 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4394 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4396 if (di && !IS_ERR(di)) {
4397 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4398 if (key.objectid == root->root_key.objectid) {
4401 "deleting default subvolume %llu is not allowed",
4405 btrfs_release_path(path);
4408 key.objectid = root->root_key.objectid;
4409 key.type = BTRFS_ROOT_REF_KEY;
4410 key.offset = (u64)-1;
4412 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4417 * Key with offset -1 found, there would have to exist a root
4418 * with such id, but this is out of valid range.
4425 if (path->slots[0] > 0) {
4427 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4428 if (key.objectid == root->root_key.objectid &&
4429 key.type == BTRFS_ROOT_REF_KEY)
4433 btrfs_free_path(path);
4437 /* Delete all dentries for inodes belonging to the root */
4438 static void btrfs_prune_dentries(struct btrfs_root *root)
4440 struct btrfs_fs_info *fs_info = root->fs_info;
4441 struct rb_node *node;
4442 struct rb_node *prev;
4443 struct btrfs_inode *entry;
4444 struct inode *inode;
4447 if (!BTRFS_FS_ERROR(fs_info))
4448 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4450 spin_lock(&root->inode_lock);
4452 node = root->inode_tree.rb_node;
4456 entry = rb_entry(node, struct btrfs_inode, rb_node);
4458 if (objectid < btrfs_ino(entry))
4459 node = node->rb_left;
4460 else if (objectid > btrfs_ino(entry))
4461 node = node->rb_right;
4467 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4468 if (objectid <= btrfs_ino(entry)) {
4472 prev = rb_next(prev);
4476 entry = rb_entry(node, struct btrfs_inode, rb_node);
4477 objectid = btrfs_ino(entry) + 1;
4478 inode = igrab(&entry->vfs_inode);
4480 spin_unlock(&root->inode_lock);
4481 if (atomic_read(&inode->i_count) > 1)
4482 d_prune_aliases(inode);
4484 * btrfs_drop_inode will have it removed from the inode
4485 * cache when its usage count hits zero.
4489 spin_lock(&root->inode_lock);
4493 if (cond_resched_lock(&root->inode_lock))
4496 node = rb_next(node);
4498 spin_unlock(&root->inode_lock);
4501 int btrfs_delete_subvolume(struct btrfs_inode *dir, struct dentry *dentry)
4503 struct btrfs_root *root = dir->root;
4504 struct btrfs_fs_info *fs_info = root->fs_info;
4505 struct inode *inode = d_inode(dentry);
4506 struct btrfs_root *dest = BTRFS_I(inode)->root;
4507 struct btrfs_trans_handle *trans;
4508 struct btrfs_block_rsv block_rsv;
4512 down_write(&fs_info->subvol_sem);
4515 * Don't allow to delete a subvolume with send in progress. This is
4516 * inside the inode lock so the error handling that has to drop the bit
4517 * again is not run concurrently.
4519 spin_lock(&dest->root_item_lock);
4520 if (dest->send_in_progress) {
4521 spin_unlock(&dest->root_item_lock);
4523 "attempt to delete subvolume %llu during send",
4524 dest->root_key.objectid);
4528 if (atomic_read(&dest->nr_swapfiles)) {
4529 spin_unlock(&dest->root_item_lock);
4531 "attempt to delete subvolume %llu with active swapfile",
4532 root->root_key.objectid);
4536 root_flags = btrfs_root_flags(&dest->root_item);
4537 btrfs_set_root_flags(&dest->root_item,
4538 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4539 spin_unlock(&dest->root_item_lock);
4541 ret = may_destroy_subvol(dest);
4545 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4547 * One for dir inode,
4548 * two for dir entries,
4549 * two for root ref/backref.
4551 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4555 trans = btrfs_start_transaction(root, 0);
4556 if (IS_ERR(trans)) {
4557 ret = PTR_ERR(trans);
4560 trans->block_rsv = &block_rsv;
4561 trans->bytes_reserved = block_rsv.size;
4563 btrfs_record_snapshot_destroy(trans, dir);
4565 ret = btrfs_unlink_subvol(trans, dir, dentry);
4567 btrfs_abort_transaction(trans, ret);
4571 ret = btrfs_record_root_in_trans(trans, dest);
4573 btrfs_abort_transaction(trans, ret);
4577 memset(&dest->root_item.drop_progress, 0,
4578 sizeof(dest->root_item.drop_progress));
4579 btrfs_set_root_drop_level(&dest->root_item, 0);
4580 btrfs_set_root_refs(&dest->root_item, 0);
4582 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4583 ret = btrfs_insert_orphan_item(trans,
4585 dest->root_key.objectid);
4587 btrfs_abort_transaction(trans, ret);
4592 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4593 BTRFS_UUID_KEY_SUBVOL,
4594 dest->root_key.objectid);
4595 if (ret && ret != -ENOENT) {
4596 btrfs_abort_transaction(trans, ret);
4599 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4600 ret = btrfs_uuid_tree_remove(trans,
4601 dest->root_item.received_uuid,
4602 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4603 dest->root_key.objectid);
4604 if (ret && ret != -ENOENT) {
4605 btrfs_abort_transaction(trans, ret);
4610 free_anon_bdev(dest->anon_dev);
4613 trans->block_rsv = NULL;
4614 trans->bytes_reserved = 0;
4615 ret = btrfs_end_transaction(trans);
4616 inode->i_flags |= S_DEAD;
4618 btrfs_subvolume_release_metadata(root, &block_rsv);
4621 spin_lock(&dest->root_item_lock);
4622 root_flags = btrfs_root_flags(&dest->root_item);
4623 btrfs_set_root_flags(&dest->root_item,
4624 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4625 spin_unlock(&dest->root_item_lock);
4628 up_write(&fs_info->subvol_sem);
4630 d_invalidate(dentry);
4631 btrfs_prune_dentries(dest);
4632 ASSERT(dest->send_in_progress == 0);
4638 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4640 struct inode *inode = d_inode(dentry);
4641 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
4643 struct btrfs_trans_handle *trans;
4644 u64 last_unlink_trans;
4645 struct fscrypt_name fname;
4647 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4649 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) {
4650 if (unlikely(btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))) {
4652 "extent tree v2 doesn't support snapshot deletion yet");
4655 return btrfs_delete_subvolume(BTRFS_I(dir), dentry);
4658 err = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4662 /* This needs to handle no-key deletions later on */
4664 trans = __unlink_start_trans(BTRFS_I(dir));
4665 if (IS_ERR(trans)) {
4666 err = PTR_ERR(trans);
4670 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4671 err = btrfs_unlink_subvol(trans, BTRFS_I(dir), dentry);
4675 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4679 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4681 /* now the directory is empty */
4682 err = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4685 btrfs_i_size_write(BTRFS_I(inode), 0);
4687 * Propagate the last_unlink_trans value of the deleted dir to
4688 * its parent directory. This is to prevent an unrecoverable
4689 * log tree in the case we do something like this:
4691 * 2) create snapshot under dir foo
4692 * 3) delete the snapshot
4695 * 6) fsync foo or some file inside foo
4697 if (last_unlink_trans >= trans->transid)
4698 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4701 btrfs_end_transaction(trans);
4703 btrfs_btree_balance_dirty(fs_info);
4704 fscrypt_free_filename(&fname);
4710 * Read, zero a chunk and write a block.
4712 * @inode - inode that we're zeroing
4713 * @from - the offset to start zeroing
4714 * @len - the length to zero, 0 to zero the entire range respective to the
4716 * @front - zero up to the offset instead of from the offset on
4718 * This will find the block for the "from" offset and cow the block and zero the
4719 * part we want to zero. This is used with truncate and hole punching.
4721 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4724 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4725 struct address_space *mapping = inode->vfs_inode.i_mapping;
4726 struct extent_io_tree *io_tree = &inode->io_tree;
4727 struct btrfs_ordered_extent *ordered;
4728 struct extent_state *cached_state = NULL;
4729 struct extent_changeset *data_reserved = NULL;
4730 bool only_release_metadata = false;
4731 u32 blocksize = fs_info->sectorsize;
4732 pgoff_t index = from >> PAGE_SHIFT;
4733 unsigned offset = from & (blocksize - 1);
4734 struct folio *folio;
4735 gfp_t mask = btrfs_alloc_write_mask(mapping);
4736 size_t write_bytes = blocksize;
4741 if (IS_ALIGNED(offset, blocksize) &&
4742 (!len || IS_ALIGNED(len, blocksize)))
4745 block_start = round_down(from, blocksize);
4746 block_end = block_start + blocksize - 1;
4748 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4751 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes, false) > 0) {
4752 /* For nocow case, no need to reserve data space */
4753 only_release_metadata = true;
4758 ret = btrfs_delalloc_reserve_metadata(inode, blocksize, blocksize, false);
4760 if (!only_release_metadata)
4761 btrfs_free_reserved_data_space(inode, data_reserved,
4762 block_start, blocksize);
4766 folio = __filemap_get_folio(mapping, index,
4767 FGP_LOCK | FGP_ACCESSED | FGP_CREAT, mask);
4768 if (IS_ERR(folio)) {
4769 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4771 btrfs_delalloc_release_extents(inode, blocksize);
4776 if (!folio_test_uptodate(folio)) {
4777 ret = btrfs_read_folio(NULL, folio);
4779 if (folio->mapping != mapping) {
4780 folio_unlock(folio);
4784 if (!folio_test_uptodate(folio)) {
4791 * We unlock the page after the io is completed and then re-lock it
4792 * above. release_folio() could have come in between that and cleared
4793 * folio private, but left the page in the mapping. Set the page mapped
4794 * here to make sure it's properly set for the subpage stuff.
4796 ret = set_folio_extent_mapped(folio);
4800 folio_wait_writeback(folio);
4802 lock_extent(io_tree, block_start, block_end, &cached_state);
4804 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4806 unlock_extent(io_tree, block_start, block_end, &cached_state);
4807 folio_unlock(folio);
4809 btrfs_start_ordered_extent(ordered);
4810 btrfs_put_ordered_extent(ordered);
4814 clear_extent_bit(&inode->io_tree, block_start, block_end,
4815 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4818 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4821 unlock_extent(io_tree, block_start, block_end, &cached_state);
4825 if (offset != blocksize) {
4827 len = blocksize - offset;
4829 folio_zero_range(folio, block_start - folio_pos(folio),
4832 folio_zero_range(folio,
4833 (block_start - folio_pos(folio)) + offset,
4836 btrfs_folio_clear_checked(fs_info, folio, block_start,
4837 block_end + 1 - block_start);
4838 btrfs_folio_set_dirty(fs_info, folio, block_start,
4839 block_end + 1 - block_start);
4840 unlock_extent(io_tree, block_start, block_end, &cached_state);
4842 if (only_release_metadata)
4843 set_extent_bit(&inode->io_tree, block_start, block_end,
4844 EXTENT_NORESERVE, NULL);
4848 if (only_release_metadata)
4849 btrfs_delalloc_release_metadata(inode, blocksize, true);
4851 btrfs_delalloc_release_space(inode, data_reserved,
4852 block_start, blocksize, true);
4854 btrfs_delalloc_release_extents(inode, blocksize);
4855 folio_unlock(folio);
4858 if (only_release_metadata)
4859 btrfs_check_nocow_unlock(inode);
4860 extent_changeset_free(data_reserved);
4864 static int maybe_insert_hole(struct btrfs_inode *inode, u64 offset, u64 len)
4866 struct btrfs_root *root = inode->root;
4867 struct btrfs_fs_info *fs_info = root->fs_info;
4868 struct btrfs_trans_handle *trans;
4869 struct btrfs_drop_extents_args drop_args = { 0 };
4873 * If NO_HOLES is enabled, we don't need to do anything.
4874 * Later, up in the call chain, either btrfs_set_inode_last_sub_trans()
4875 * or btrfs_update_inode() will be called, which guarantee that the next
4876 * fsync will know this inode was changed and needs to be logged.
4878 if (btrfs_fs_incompat(fs_info, NO_HOLES))
4882 * 1 - for the one we're dropping
4883 * 1 - for the one we're adding
4884 * 1 - for updating the inode.
4886 trans = btrfs_start_transaction(root, 3);
4888 return PTR_ERR(trans);
4890 drop_args.start = offset;
4891 drop_args.end = offset + len;
4892 drop_args.drop_cache = true;
4894 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
4896 btrfs_abort_transaction(trans, ret);
4897 btrfs_end_transaction(trans);
4901 ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset, len);
4903 btrfs_abort_transaction(trans, ret);
4905 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
4906 btrfs_update_inode(trans, inode);
4908 btrfs_end_transaction(trans);
4913 * This function puts in dummy file extents for the area we're creating a hole
4914 * for. So if we are truncating this file to a larger size we need to insert
4915 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4916 * the range between oldsize and size
4918 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
4920 struct btrfs_root *root = inode->root;
4921 struct btrfs_fs_info *fs_info = root->fs_info;
4922 struct extent_io_tree *io_tree = &inode->io_tree;
4923 struct extent_map *em = NULL;
4924 struct extent_state *cached_state = NULL;
4925 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4926 u64 block_end = ALIGN(size, fs_info->sectorsize);
4933 * If our size started in the middle of a block we need to zero out the
4934 * rest of the block before we expand the i_size, otherwise we could
4935 * expose stale data.
4937 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4941 if (size <= hole_start)
4944 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
4946 cur_offset = hole_start;
4948 em = btrfs_get_extent(inode, NULL, cur_offset, block_end - cur_offset);
4954 last_byte = min(extent_map_end(em), block_end);
4955 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4956 hole_size = last_byte - cur_offset;
4958 if (!(em->flags & EXTENT_FLAG_PREALLOC)) {
4959 struct extent_map *hole_em;
4961 err = maybe_insert_hole(inode, cur_offset, hole_size);
4965 err = btrfs_inode_set_file_extent_range(inode,
4966 cur_offset, hole_size);
4970 hole_em = alloc_extent_map();
4972 btrfs_drop_extent_map_range(inode, cur_offset,
4973 cur_offset + hole_size - 1,
4975 btrfs_set_inode_full_sync(inode);
4978 hole_em->start = cur_offset;
4979 hole_em->len = hole_size;
4980 hole_em->orig_start = cur_offset;
4982 hole_em->block_start = EXTENT_MAP_HOLE;
4983 hole_em->block_len = 0;
4984 hole_em->orig_block_len = 0;
4985 hole_em->ram_bytes = hole_size;
4986 hole_em->generation = btrfs_get_fs_generation(fs_info);
4988 err = btrfs_replace_extent_map_range(inode, hole_em, true);
4989 free_extent_map(hole_em);
4991 err = btrfs_inode_set_file_extent_range(inode,
4992 cur_offset, hole_size);
4997 free_extent_map(em);
4999 cur_offset = last_byte;
5000 if (cur_offset >= block_end)
5003 free_extent_map(em);
5004 unlock_extent(io_tree, hole_start, block_end - 1, &cached_state);
5008 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5010 struct btrfs_root *root = BTRFS_I(inode)->root;
5011 struct btrfs_trans_handle *trans;
5012 loff_t oldsize = i_size_read(inode);
5013 loff_t newsize = attr->ia_size;
5014 int mask = attr->ia_valid;
5018 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5019 * special case where we need to update the times despite not having
5020 * these flags set. For all other operations the VFS set these flags
5021 * explicitly if it wants a timestamp update.
5023 if (newsize != oldsize) {
5024 inode_inc_iversion(inode);
5025 if (!(mask & (ATTR_CTIME | ATTR_MTIME))) {
5026 inode_set_mtime_to_ts(inode,
5027 inode_set_ctime_current(inode));
5031 if (newsize > oldsize) {
5033 * Don't do an expanding truncate while snapshotting is ongoing.
5034 * This is to ensure the snapshot captures a fully consistent
5035 * state of this file - if the snapshot captures this expanding
5036 * truncation, it must capture all writes that happened before
5039 btrfs_drew_write_lock(&root->snapshot_lock);
5040 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
5042 btrfs_drew_write_unlock(&root->snapshot_lock);
5046 trans = btrfs_start_transaction(root, 1);
5047 if (IS_ERR(trans)) {
5048 btrfs_drew_write_unlock(&root->snapshot_lock);
5049 return PTR_ERR(trans);
5052 i_size_write(inode, newsize);
5053 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
5054 pagecache_isize_extended(inode, oldsize, newsize);
5055 ret = btrfs_update_inode(trans, BTRFS_I(inode));
5056 btrfs_drew_write_unlock(&root->snapshot_lock);
5057 btrfs_end_transaction(trans);
5059 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
5061 if (btrfs_is_zoned(fs_info)) {
5062 ret = btrfs_wait_ordered_range(inode,
5063 ALIGN(newsize, fs_info->sectorsize),
5070 * We're truncating a file that used to have good data down to
5071 * zero. Make sure any new writes to the file get on disk
5075 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5076 &BTRFS_I(inode)->runtime_flags);
5078 truncate_setsize(inode, newsize);
5080 inode_dio_wait(inode);
5082 ret = btrfs_truncate(BTRFS_I(inode), newsize == oldsize);
5083 if (ret && inode->i_nlink) {
5087 * Truncate failed, so fix up the in-memory size. We
5088 * adjusted disk_i_size down as we removed extents, so
5089 * wait for disk_i_size to be stable and then update the
5090 * in-memory size to match.
5092 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5095 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5102 static int btrfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
5105 struct inode *inode = d_inode(dentry);
5106 struct btrfs_root *root = BTRFS_I(inode)->root;
5109 if (btrfs_root_readonly(root))
5112 err = setattr_prepare(idmap, dentry, attr);
5116 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5117 err = btrfs_setsize(inode, attr);
5122 if (attr->ia_valid) {
5123 setattr_copy(idmap, inode, attr);
5124 inode_inc_iversion(inode);
5125 err = btrfs_dirty_inode(BTRFS_I(inode));
5127 if (!err && attr->ia_valid & ATTR_MODE)
5128 err = posix_acl_chmod(idmap, dentry, inode->i_mode);
5135 * While truncating the inode pages during eviction, we get the VFS
5136 * calling btrfs_invalidate_folio() against each folio of the inode. This
5137 * is slow because the calls to btrfs_invalidate_folio() result in a
5138 * huge amount of calls to lock_extent() and clear_extent_bit(),
5139 * which keep merging and splitting extent_state structures over and over,
5140 * wasting lots of time.
5142 * Therefore if the inode is being evicted, let btrfs_invalidate_folio()
5143 * skip all those expensive operations on a per folio basis and do only
5144 * the ordered io finishing, while we release here the extent_map and
5145 * extent_state structures, without the excessive merging and splitting.
5147 static void evict_inode_truncate_pages(struct inode *inode)
5149 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5150 struct rb_node *node;
5152 ASSERT(inode->i_state & I_FREEING);
5153 truncate_inode_pages_final(&inode->i_data);
5155 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
5158 * Keep looping until we have no more ranges in the io tree.
5159 * We can have ongoing bios started by readahead that have
5160 * their endio callback (extent_io.c:end_bio_extent_readpage)
5161 * still in progress (unlocked the pages in the bio but did not yet
5162 * unlocked the ranges in the io tree). Therefore this means some
5163 * ranges can still be locked and eviction started because before
5164 * submitting those bios, which are executed by a separate task (work
5165 * queue kthread), inode references (inode->i_count) were not taken
5166 * (which would be dropped in the end io callback of each bio).
5167 * Therefore here we effectively end up waiting for those bios and
5168 * anyone else holding locked ranges without having bumped the inode's
5169 * reference count - if we don't do it, when they access the inode's
5170 * io_tree to unlock a range it may be too late, leading to an
5171 * use-after-free issue.
5173 spin_lock(&io_tree->lock);
5174 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5175 struct extent_state *state;
5176 struct extent_state *cached_state = NULL;
5179 unsigned state_flags;
5181 node = rb_first(&io_tree->state);
5182 state = rb_entry(node, struct extent_state, rb_node);
5183 start = state->start;
5185 state_flags = state->state;
5186 spin_unlock(&io_tree->lock);
5188 lock_extent(io_tree, start, end, &cached_state);
5191 * If still has DELALLOC flag, the extent didn't reach disk,
5192 * and its reserved space won't be freed by delayed_ref.
5193 * So we need to free its reserved space here.
5194 * (Refer to comment in btrfs_invalidate_folio, case 2)
5196 * Note, end is the bytenr of last byte, so we need + 1 here.
5198 if (state_flags & EXTENT_DELALLOC)
5199 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5200 end - start + 1, NULL);
5202 clear_extent_bit(io_tree, start, end,
5203 EXTENT_CLEAR_ALL_BITS | EXTENT_DO_ACCOUNTING,
5207 spin_lock(&io_tree->lock);
5209 spin_unlock(&io_tree->lock);
5212 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5213 struct btrfs_block_rsv *rsv)
5215 struct btrfs_fs_info *fs_info = root->fs_info;
5216 struct btrfs_trans_handle *trans;
5217 u64 delayed_refs_extra = btrfs_calc_delayed_ref_bytes(fs_info, 1);
5221 * Eviction should be taking place at some place safe because of our
5222 * delayed iputs. However the normal flushing code will run delayed
5223 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5225 * We reserve the delayed_refs_extra here again because we can't use
5226 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5227 * above. We reserve our extra bit here because we generate a ton of
5228 * delayed refs activity by truncating.
5230 * BTRFS_RESERVE_FLUSH_EVICT will steal from the global_rsv if it can,
5231 * if we fail to make this reservation we can re-try without the
5232 * delayed_refs_extra so we can make some forward progress.
5234 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size + delayed_refs_extra,
5235 BTRFS_RESERVE_FLUSH_EVICT);
5237 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size,
5238 BTRFS_RESERVE_FLUSH_EVICT);
5241 "could not allocate space for delete; will truncate on mount");
5242 return ERR_PTR(-ENOSPC);
5244 delayed_refs_extra = 0;
5247 trans = btrfs_join_transaction(root);
5251 if (delayed_refs_extra) {
5252 trans->block_rsv = &fs_info->trans_block_rsv;
5253 trans->bytes_reserved = delayed_refs_extra;
5254 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5255 delayed_refs_extra, true);
5260 void btrfs_evict_inode(struct inode *inode)
5262 struct btrfs_fs_info *fs_info;
5263 struct btrfs_trans_handle *trans;
5264 struct btrfs_root *root = BTRFS_I(inode)->root;
5265 struct btrfs_block_rsv *rsv = NULL;
5268 trace_btrfs_inode_evict(inode);
5271 fsverity_cleanup_inode(inode);
5276 fs_info = inode_to_fs_info(inode);
5277 evict_inode_truncate_pages(inode);
5279 if (inode->i_nlink &&
5280 ((btrfs_root_refs(&root->root_item) != 0 &&
5281 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5282 btrfs_is_free_space_inode(BTRFS_I(inode))))
5285 if (is_bad_inode(inode))
5288 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5291 if (inode->i_nlink > 0) {
5292 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5293 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5298 * This makes sure the inode item in tree is uptodate and the space for
5299 * the inode update is released.
5301 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5306 * This drops any pending insert or delete operations we have for this
5307 * inode. We could have a delayed dir index deletion queued up, but
5308 * we're removing the inode completely so that'll be taken care of in
5311 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
5313 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5316 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5317 rsv->failfast = true;
5319 btrfs_i_size_write(BTRFS_I(inode), 0);
5322 struct btrfs_truncate_control control = {
5323 .inode = BTRFS_I(inode),
5324 .ino = btrfs_ino(BTRFS_I(inode)),
5329 trans = evict_refill_and_join(root, rsv);
5333 trans->block_rsv = rsv;
5335 ret = btrfs_truncate_inode_items(trans, root, &control);
5336 trans->block_rsv = &fs_info->trans_block_rsv;
5337 btrfs_end_transaction(trans);
5339 * We have not added new delayed items for our inode after we
5340 * have flushed its delayed items, so no need to throttle on
5341 * delayed items. However we have modified extent buffers.
5343 btrfs_btree_balance_dirty_nodelay(fs_info);
5344 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5351 * Errors here aren't a big deal, it just means we leave orphan items in
5352 * the tree. They will be cleaned up on the next mount. If the inode
5353 * number gets reused, cleanup deletes the orphan item without doing
5354 * anything, and unlink reuses the existing orphan item.
5356 * If it turns out that we are dropping too many of these, we might want
5357 * to add a mechanism for retrying these after a commit.
5359 trans = evict_refill_and_join(root, rsv);
5360 if (!IS_ERR(trans)) {
5361 trans->block_rsv = rsv;
5362 btrfs_orphan_del(trans, BTRFS_I(inode));
5363 trans->block_rsv = &fs_info->trans_block_rsv;
5364 btrfs_end_transaction(trans);
5368 btrfs_free_block_rsv(fs_info, rsv);
5370 * If we didn't successfully delete, the orphan item will still be in
5371 * the tree and we'll retry on the next mount. Again, we might also want
5372 * to retry these periodically in the future.
5374 btrfs_remove_delayed_node(BTRFS_I(inode));
5375 fsverity_cleanup_inode(inode);
5380 * Return the key found in the dir entry in the location pointer, fill @type
5381 * with BTRFS_FT_*, and return 0.
5383 * If no dir entries were found, returns -ENOENT.
5384 * If found a corrupted location in dir entry, returns -EUCLEAN.
5386 static int btrfs_inode_by_name(struct btrfs_inode *dir, struct dentry *dentry,
5387 struct btrfs_key *location, u8 *type)
5389 struct btrfs_dir_item *di;
5390 struct btrfs_path *path;
5391 struct btrfs_root *root = dir->root;
5393 struct fscrypt_name fname;
5395 path = btrfs_alloc_path();
5399 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
5403 * fscrypt_setup_filename() should never return a positive value, but
5404 * gcc on sparc/parisc thinks it can, so assert that doesn't happen.
5408 /* This needs to handle no-key deletions later on */
5410 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir),
5411 &fname.disk_name, 0);
5412 if (IS_ERR_OR_NULL(di)) {
5413 ret = di ? PTR_ERR(di) : -ENOENT;
5417 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5418 if (location->type != BTRFS_INODE_ITEM_KEY &&
5419 location->type != BTRFS_ROOT_ITEM_KEY) {
5421 btrfs_warn(root->fs_info,
5422 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5423 __func__, fname.disk_name.name, btrfs_ino(dir),
5424 location->objectid, location->type, location->offset);
5427 *type = btrfs_dir_ftype(path->nodes[0], di);
5429 fscrypt_free_filename(&fname);
5430 btrfs_free_path(path);
5435 * when we hit a tree root in a directory, the btrfs part of the inode
5436 * needs to be changed to reflect the root directory of the tree root. This
5437 * is kind of like crossing a mount point.
5439 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5440 struct btrfs_inode *dir,
5441 struct dentry *dentry,
5442 struct btrfs_key *location,
5443 struct btrfs_root **sub_root)
5445 struct btrfs_path *path;
5446 struct btrfs_root *new_root;
5447 struct btrfs_root_ref *ref;
5448 struct extent_buffer *leaf;
5449 struct btrfs_key key;
5452 struct fscrypt_name fname;
5454 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 0, &fname);
5458 path = btrfs_alloc_path();
5465 key.objectid = dir->root->root_key.objectid;
5466 key.type = BTRFS_ROOT_REF_KEY;
5467 key.offset = location->objectid;
5469 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5476 leaf = path->nodes[0];
5477 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5478 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5479 btrfs_root_ref_name_len(leaf, ref) != fname.disk_name.len)
5482 ret = memcmp_extent_buffer(leaf, fname.disk_name.name,
5483 (unsigned long)(ref + 1), fname.disk_name.len);
5487 btrfs_release_path(path);
5489 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5490 if (IS_ERR(new_root)) {
5491 err = PTR_ERR(new_root);
5495 *sub_root = new_root;
5496 location->objectid = btrfs_root_dirid(&new_root->root_item);
5497 location->type = BTRFS_INODE_ITEM_KEY;
5498 location->offset = 0;
5501 btrfs_free_path(path);
5502 fscrypt_free_filename(&fname);
5506 static void inode_tree_add(struct btrfs_inode *inode)
5508 struct btrfs_root *root = inode->root;
5509 struct btrfs_inode *entry;
5511 struct rb_node *parent;
5512 struct rb_node *new = &inode->rb_node;
5513 u64 ino = btrfs_ino(inode);
5515 if (inode_unhashed(&inode->vfs_inode))
5518 spin_lock(&root->inode_lock);
5519 p = &root->inode_tree.rb_node;
5522 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5524 if (ino < btrfs_ino(entry))
5525 p = &parent->rb_left;
5526 else if (ino > btrfs_ino(entry))
5527 p = &parent->rb_right;
5529 WARN_ON(!(entry->vfs_inode.i_state &
5530 (I_WILL_FREE | I_FREEING)));
5531 rb_replace_node(parent, new, &root->inode_tree);
5532 RB_CLEAR_NODE(parent);
5533 spin_unlock(&root->inode_lock);
5537 rb_link_node(new, parent, p);
5538 rb_insert_color(new, &root->inode_tree);
5539 spin_unlock(&root->inode_lock);
5542 static void inode_tree_del(struct btrfs_inode *inode)
5544 struct btrfs_root *root = inode->root;
5547 spin_lock(&root->inode_lock);
5548 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5549 rb_erase(&inode->rb_node, &root->inode_tree);
5550 RB_CLEAR_NODE(&inode->rb_node);
5551 empty = RB_EMPTY_ROOT(&root->inode_tree);
5553 spin_unlock(&root->inode_lock);
5555 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5556 spin_lock(&root->inode_lock);
5557 empty = RB_EMPTY_ROOT(&root->inode_tree);
5558 spin_unlock(&root->inode_lock);
5560 btrfs_add_dead_root(root);
5565 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5567 struct btrfs_iget_args *args = p;
5569 inode->i_ino = args->ino;
5570 BTRFS_I(inode)->location.objectid = args->ino;
5571 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5572 BTRFS_I(inode)->location.offset = 0;
5573 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5575 if (args->root && args->root == args->root->fs_info->tree_root &&
5576 args->ino != BTRFS_BTREE_INODE_OBJECTID)
5577 set_bit(BTRFS_INODE_FREE_SPACE_INODE,
5578 &BTRFS_I(inode)->runtime_flags);
5582 static int btrfs_find_actor(struct inode *inode, void *opaque)
5584 struct btrfs_iget_args *args = opaque;
5586 return args->ino == BTRFS_I(inode)->location.objectid &&
5587 args->root == BTRFS_I(inode)->root;
5590 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5591 struct btrfs_root *root)
5593 struct inode *inode;
5594 struct btrfs_iget_args args;
5595 unsigned long hashval = btrfs_inode_hash(ino, root);
5600 inode = iget5_locked(s, hashval, btrfs_find_actor,
5601 btrfs_init_locked_inode,
5607 * Get an inode object given its inode number and corresponding root.
5608 * Path can be preallocated to prevent recursing back to iget through
5609 * allocator. NULL is also valid but may require an additional allocation
5612 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5613 struct btrfs_root *root, struct btrfs_path *path)
5615 struct inode *inode;
5617 inode = btrfs_iget_locked(s, ino, root);
5619 return ERR_PTR(-ENOMEM);
5621 if (inode->i_state & I_NEW) {
5624 ret = btrfs_read_locked_inode(inode, path);
5626 inode_tree_add(BTRFS_I(inode));
5627 unlock_new_inode(inode);
5631 * ret > 0 can come from btrfs_search_slot called by
5632 * btrfs_read_locked_inode, this means the inode item
5637 inode = ERR_PTR(ret);
5644 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5646 return btrfs_iget_path(s, ino, root, NULL);
5649 static struct inode *new_simple_dir(struct inode *dir,
5650 struct btrfs_key *key,
5651 struct btrfs_root *root)
5653 struct timespec64 ts;
5654 struct inode *inode = new_inode(dir->i_sb);
5657 return ERR_PTR(-ENOMEM);
5659 BTRFS_I(inode)->root = btrfs_grab_root(root);
5660 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5661 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5663 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5665 * We only need lookup, the rest is read-only and there's no inode
5666 * associated with the dentry
5668 inode->i_op = &simple_dir_inode_operations;
5669 inode->i_opflags &= ~IOP_XATTR;
5670 inode->i_fop = &simple_dir_operations;
5671 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5673 ts = inode_set_ctime_current(inode);
5674 inode_set_mtime_to_ts(inode, ts);
5675 inode_set_atime_to_ts(inode, inode_get_atime(dir));
5676 BTRFS_I(inode)->i_otime_sec = ts.tv_sec;
5677 BTRFS_I(inode)->i_otime_nsec = ts.tv_nsec;
5679 inode->i_uid = dir->i_uid;
5680 inode->i_gid = dir->i_gid;
5685 static_assert(BTRFS_FT_UNKNOWN == FT_UNKNOWN);
5686 static_assert(BTRFS_FT_REG_FILE == FT_REG_FILE);
5687 static_assert(BTRFS_FT_DIR == FT_DIR);
5688 static_assert(BTRFS_FT_CHRDEV == FT_CHRDEV);
5689 static_assert(BTRFS_FT_BLKDEV == FT_BLKDEV);
5690 static_assert(BTRFS_FT_FIFO == FT_FIFO);
5691 static_assert(BTRFS_FT_SOCK == FT_SOCK);
5692 static_assert(BTRFS_FT_SYMLINK == FT_SYMLINK);
5694 static inline u8 btrfs_inode_type(struct inode *inode)
5696 return fs_umode_to_ftype(inode->i_mode);
5699 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5701 struct btrfs_fs_info *fs_info = inode_to_fs_info(dir);
5702 struct inode *inode;
5703 struct btrfs_root *root = BTRFS_I(dir)->root;
5704 struct btrfs_root *sub_root = root;
5705 struct btrfs_key location;
5709 if (dentry->d_name.len > BTRFS_NAME_LEN)
5710 return ERR_PTR(-ENAMETOOLONG);
5712 ret = btrfs_inode_by_name(BTRFS_I(dir), dentry, &location, &di_type);
5714 return ERR_PTR(ret);
5716 if (location.type == BTRFS_INODE_ITEM_KEY) {
5717 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5721 /* Do extra check against inode mode with di_type */
5722 if (btrfs_inode_type(inode) != di_type) {
5724 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5725 inode->i_mode, btrfs_inode_type(inode),
5728 return ERR_PTR(-EUCLEAN);
5733 ret = fixup_tree_root_location(fs_info, BTRFS_I(dir), dentry,
5734 &location, &sub_root);
5737 inode = ERR_PTR(ret);
5739 inode = new_simple_dir(dir, &location, root);
5741 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5742 btrfs_put_root(sub_root);
5747 down_read(&fs_info->cleanup_work_sem);
5748 if (!sb_rdonly(inode->i_sb))
5749 ret = btrfs_orphan_cleanup(sub_root);
5750 up_read(&fs_info->cleanup_work_sem);
5753 inode = ERR_PTR(ret);
5760 static int btrfs_dentry_delete(const struct dentry *dentry)
5762 struct btrfs_root *root;
5763 struct inode *inode = d_inode(dentry);
5765 if (!inode && !IS_ROOT(dentry))
5766 inode = d_inode(dentry->d_parent);
5769 root = BTRFS_I(inode)->root;
5770 if (btrfs_root_refs(&root->root_item) == 0)
5773 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5779 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5782 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5784 if (inode == ERR_PTR(-ENOENT))
5786 return d_splice_alias(inode, dentry);
5790 * Find the highest existing sequence number in a directory and then set the
5791 * in-memory index_cnt variable to the first free sequence number.
5793 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
5795 struct btrfs_root *root = inode->root;
5796 struct btrfs_key key, found_key;
5797 struct btrfs_path *path;
5798 struct extent_buffer *leaf;
5801 key.objectid = btrfs_ino(inode);
5802 key.type = BTRFS_DIR_INDEX_KEY;
5803 key.offset = (u64)-1;
5805 path = btrfs_alloc_path();
5809 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5812 /* FIXME: we should be able to handle this */
5817 if (path->slots[0] == 0) {
5818 inode->index_cnt = BTRFS_DIR_START_INDEX;
5824 leaf = path->nodes[0];
5825 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5827 if (found_key.objectid != btrfs_ino(inode) ||
5828 found_key.type != BTRFS_DIR_INDEX_KEY) {
5829 inode->index_cnt = BTRFS_DIR_START_INDEX;
5833 inode->index_cnt = found_key.offset + 1;
5835 btrfs_free_path(path);
5839 static int btrfs_get_dir_last_index(struct btrfs_inode *dir, u64 *index)
5843 btrfs_inode_lock(dir, 0);
5844 if (dir->index_cnt == (u64)-1) {
5845 ret = btrfs_inode_delayed_dir_index_count(dir);
5847 ret = btrfs_set_inode_index_count(dir);
5853 /* index_cnt is the index number of next new entry, so decrement it. */
5854 *index = dir->index_cnt - 1;
5856 btrfs_inode_unlock(dir, 0);
5862 * All this infrastructure exists because dir_emit can fault, and we are holding
5863 * the tree lock when doing readdir. For now just allocate a buffer and copy
5864 * our information into that, and then dir_emit from the buffer. This is
5865 * similar to what NFS does, only we don't keep the buffer around in pagecache
5866 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5867 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5870 static int btrfs_opendir(struct inode *inode, struct file *file)
5872 struct btrfs_file_private *private;
5876 ret = btrfs_get_dir_last_index(BTRFS_I(inode), &last_index);
5880 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5883 private->last_index = last_index;
5884 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5885 if (!private->filldir_buf) {
5889 file->private_data = private;
5893 static loff_t btrfs_dir_llseek(struct file *file, loff_t offset, int whence)
5895 struct btrfs_file_private *private = file->private_data;
5898 ret = btrfs_get_dir_last_index(BTRFS_I(file_inode(file)),
5899 &private->last_index);
5903 return generic_file_llseek(file, offset, whence);
5913 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5916 struct dir_entry *entry = addr;
5917 char *name = (char *)(entry + 1);
5919 ctx->pos = get_unaligned(&entry->offset);
5920 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5921 get_unaligned(&entry->ino),
5922 get_unaligned(&entry->type)))
5924 addr += sizeof(struct dir_entry) +
5925 get_unaligned(&entry->name_len);
5931 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5933 struct inode *inode = file_inode(file);
5934 struct btrfs_root *root = BTRFS_I(inode)->root;
5935 struct btrfs_file_private *private = file->private_data;
5936 struct btrfs_dir_item *di;
5937 struct btrfs_key key;
5938 struct btrfs_key found_key;
5939 struct btrfs_path *path;
5941 LIST_HEAD(ins_list);
5942 LIST_HEAD(del_list);
5949 struct btrfs_key location;
5951 if (!dir_emit_dots(file, ctx))
5954 path = btrfs_alloc_path();
5958 addr = private->filldir_buf;
5959 path->reada = READA_FORWARD;
5961 put = btrfs_readdir_get_delayed_items(inode, private->last_index,
5962 &ins_list, &del_list);
5965 key.type = BTRFS_DIR_INDEX_KEY;
5966 key.offset = ctx->pos;
5967 key.objectid = btrfs_ino(BTRFS_I(inode));
5969 btrfs_for_each_slot(root, &key, &found_key, path, ret) {
5970 struct dir_entry *entry;
5971 struct extent_buffer *leaf = path->nodes[0];
5974 if (found_key.objectid != key.objectid)
5976 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5978 if (found_key.offset < ctx->pos)
5980 if (found_key.offset > private->last_index)
5982 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5984 di = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
5985 name_len = btrfs_dir_name_len(leaf, di);
5986 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5988 btrfs_release_path(path);
5989 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5992 addr = private->filldir_buf;
5998 ftype = btrfs_dir_flags_to_ftype(btrfs_dir_flags(leaf, di));
6000 name_ptr = (char *)(entry + 1);
6001 read_extent_buffer(leaf, name_ptr,
6002 (unsigned long)(di + 1), name_len);
6003 put_unaligned(name_len, &entry->name_len);
6004 put_unaligned(fs_ftype_to_dtype(ftype), &entry->type);
6005 btrfs_dir_item_key_to_cpu(leaf, di, &location);
6006 put_unaligned(location.objectid, &entry->ino);
6007 put_unaligned(found_key.offset, &entry->offset);
6009 addr += sizeof(struct dir_entry) + name_len;
6010 total_len += sizeof(struct dir_entry) + name_len;
6012 /* Catch error encountered during iteration */
6016 btrfs_release_path(path);
6018 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6022 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
6027 * Stop new entries from being returned after we return the last
6030 * New directory entries are assigned a strictly increasing
6031 * offset. This means that new entries created during readdir
6032 * are *guaranteed* to be seen in the future by that readdir.
6033 * This has broken buggy programs which operate on names as
6034 * they're returned by readdir. Until we re-use freed offsets
6035 * we have this hack to stop new entries from being returned
6036 * under the assumption that they'll never reach this huge
6039 * This is being careful not to overflow 32bit loff_t unless the
6040 * last entry requires it because doing so has broken 32bit apps
6043 if (ctx->pos >= INT_MAX)
6044 ctx->pos = LLONG_MAX;
6051 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6052 btrfs_free_path(path);
6057 * This is somewhat expensive, updating the tree every time the
6058 * inode changes. But, it is most likely to find the inode in cache.
6059 * FIXME, needs more benchmarking...there are no reasons other than performance
6060 * to keep or drop this code.
6062 static int btrfs_dirty_inode(struct btrfs_inode *inode)
6064 struct btrfs_root *root = inode->root;
6065 struct btrfs_fs_info *fs_info = root->fs_info;
6066 struct btrfs_trans_handle *trans;
6069 if (test_bit(BTRFS_INODE_DUMMY, &inode->runtime_flags))
6072 trans = btrfs_join_transaction(root);
6074 return PTR_ERR(trans);
6076 ret = btrfs_update_inode(trans, inode);
6077 if (ret == -ENOSPC || ret == -EDQUOT) {
6078 /* whoops, lets try again with the full transaction */
6079 btrfs_end_transaction(trans);
6080 trans = btrfs_start_transaction(root, 1);
6082 return PTR_ERR(trans);
6084 ret = btrfs_update_inode(trans, inode);
6086 btrfs_end_transaction(trans);
6087 if (inode->delayed_node)
6088 btrfs_balance_delayed_items(fs_info);
6094 * This is a copy of file_update_time. We need this so we can return error on
6095 * ENOSPC for updating the inode in the case of file write and mmap writes.
6097 static int btrfs_update_time(struct inode *inode, int flags)
6099 struct btrfs_root *root = BTRFS_I(inode)->root;
6102 if (btrfs_root_readonly(root))
6105 dirty = inode_update_timestamps(inode, flags);
6106 return dirty ? btrfs_dirty_inode(BTRFS_I(inode)) : 0;
6110 * helper to find a free sequence number in a given directory. This current
6111 * code is very simple, later versions will do smarter things in the btree
6113 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6117 if (dir->index_cnt == (u64)-1) {
6118 ret = btrfs_inode_delayed_dir_index_count(dir);
6120 ret = btrfs_set_inode_index_count(dir);
6126 *index = dir->index_cnt;
6132 static int btrfs_insert_inode_locked(struct inode *inode)
6134 struct btrfs_iget_args args;
6136 args.ino = BTRFS_I(inode)->location.objectid;
6137 args.root = BTRFS_I(inode)->root;
6139 return insert_inode_locked4(inode,
6140 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6141 btrfs_find_actor, &args);
6144 int btrfs_new_inode_prepare(struct btrfs_new_inode_args *args,
6145 unsigned int *trans_num_items)
6147 struct inode *dir = args->dir;
6148 struct inode *inode = args->inode;
6151 if (!args->orphan) {
6152 ret = fscrypt_setup_filename(dir, &args->dentry->d_name, 0,
6158 ret = posix_acl_create(dir, &inode->i_mode, &args->default_acl, &args->acl);
6160 fscrypt_free_filename(&args->fname);
6164 /* 1 to add inode item */
6165 *trans_num_items = 1;
6166 /* 1 to add compression property */
6167 if (BTRFS_I(dir)->prop_compress)
6168 (*trans_num_items)++;
6169 /* 1 to add default ACL xattr */
6170 if (args->default_acl)
6171 (*trans_num_items)++;
6172 /* 1 to add access ACL xattr */
6174 (*trans_num_items)++;
6175 #ifdef CONFIG_SECURITY
6176 /* 1 to add LSM xattr */
6177 if (dir->i_security)
6178 (*trans_num_items)++;
6181 /* 1 to add orphan item */
6182 (*trans_num_items)++;
6186 * 1 to add dir index
6187 * 1 to update parent inode item
6189 * No need for 1 unit for the inode ref item because it is
6190 * inserted in a batch together with the inode item at
6191 * btrfs_create_new_inode().
6193 *trans_num_items += 3;
6198 void btrfs_new_inode_args_destroy(struct btrfs_new_inode_args *args)
6200 posix_acl_release(args->acl);
6201 posix_acl_release(args->default_acl);
6202 fscrypt_free_filename(&args->fname);
6206 * Inherit flags from the parent inode.
6208 * Currently only the compression flags and the cow flags are inherited.
6210 static void btrfs_inherit_iflags(struct btrfs_inode *inode, struct btrfs_inode *dir)
6216 if (flags & BTRFS_INODE_NOCOMPRESS) {
6217 inode->flags &= ~BTRFS_INODE_COMPRESS;
6218 inode->flags |= BTRFS_INODE_NOCOMPRESS;
6219 } else if (flags & BTRFS_INODE_COMPRESS) {
6220 inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
6221 inode->flags |= BTRFS_INODE_COMPRESS;
6224 if (flags & BTRFS_INODE_NODATACOW) {
6225 inode->flags |= BTRFS_INODE_NODATACOW;
6226 if (S_ISREG(inode->vfs_inode.i_mode))
6227 inode->flags |= BTRFS_INODE_NODATASUM;
6230 btrfs_sync_inode_flags_to_i_flags(&inode->vfs_inode);
6233 int btrfs_create_new_inode(struct btrfs_trans_handle *trans,
6234 struct btrfs_new_inode_args *args)
6236 struct timespec64 ts;
6237 struct inode *dir = args->dir;
6238 struct inode *inode = args->inode;
6239 const struct fscrypt_str *name = args->orphan ? NULL : &args->fname.disk_name;
6240 struct btrfs_fs_info *fs_info = inode_to_fs_info(dir);
6241 struct btrfs_root *root;
6242 struct btrfs_inode_item *inode_item;
6243 struct btrfs_key *location;
6244 struct btrfs_path *path;
6246 struct btrfs_inode_ref *ref;
6247 struct btrfs_key key[2];
6249 struct btrfs_item_batch batch;
6253 path = btrfs_alloc_path();
6258 BTRFS_I(inode)->root = btrfs_grab_root(BTRFS_I(dir)->root);
6259 root = BTRFS_I(inode)->root;
6261 ret = btrfs_get_free_objectid(root, &objectid);
6264 inode->i_ino = objectid;
6268 * O_TMPFILE, set link count to 0, so that after this point, we
6269 * fill in an inode item with the correct link count.
6271 set_nlink(inode, 0);
6273 trace_btrfs_inode_request(dir);
6275 ret = btrfs_set_inode_index(BTRFS_I(dir), &BTRFS_I(inode)->dir_index);
6279 /* index_cnt is ignored for everything but a dir. */
6280 BTRFS_I(inode)->index_cnt = BTRFS_DIR_START_INDEX;
6281 BTRFS_I(inode)->generation = trans->transid;
6282 inode->i_generation = BTRFS_I(inode)->generation;
6285 * We don't have any capability xattrs set here yet, shortcut any
6286 * queries for the xattrs here. If we add them later via the inode
6287 * security init path or any other path this flag will be cleared.
6289 set_bit(BTRFS_INODE_NO_CAP_XATTR, &BTRFS_I(inode)->runtime_flags);
6292 * Subvolumes don't inherit flags from their parent directory.
6293 * Originally this was probably by accident, but we probably can't
6294 * change it now without compatibility issues.
6297 btrfs_inherit_iflags(BTRFS_I(inode), BTRFS_I(dir));
6299 if (S_ISREG(inode->i_mode)) {
6300 if (btrfs_test_opt(fs_info, NODATASUM))
6301 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6302 if (btrfs_test_opt(fs_info, NODATACOW))
6303 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6304 BTRFS_INODE_NODATASUM;
6307 location = &BTRFS_I(inode)->location;
6308 location->objectid = objectid;
6309 location->offset = 0;
6310 location->type = BTRFS_INODE_ITEM_KEY;
6312 ret = btrfs_insert_inode_locked(inode);
6315 BTRFS_I(dir)->index_cnt--;
6320 * We could have gotten an inode number from somebody who was fsynced
6321 * and then removed in this same transaction, so let's just set full
6322 * sync since it will be a full sync anyway and this will blow away the
6323 * old info in the log.
6325 btrfs_set_inode_full_sync(BTRFS_I(inode));
6327 key[0].objectid = objectid;
6328 key[0].type = BTRFS_INODE_ITEM_KEY;
6331 sizes[0] = sizeof(struct btrfs_inode_item);
6333 if (!args->orphan) {
6335 * Start new inodes with an inode_ref. This is slightly more
6336 * efficient for small numbers of hard links since they will
6337 * be packed into one item. Extended refs will kick in if we
6338 * add more hard links than can fit in the ref item.
6340 key[1].objectid = objectid;
6341 key[1].type = BTRFS_INODE_REF_KEY;
6343 key[1].offset = objectid;
6344 sizes[1] = 2 + sizeof(*ref);
6346 key[1].offset = btrfs_ino(BTRFS_I(dir));
6347 sizes[1] = name->len + sizeof(*ref);
6351 batch.keys = &key[0];
6352 batch.data_sizes = &sizes[0];
6353 batch.total_data_size = sizes[0] + (args->orphan ? 0 : sizes[1]);
6354 batch.nr = args->orphan ? 1 : 2;
6355 ret = btrfs_insert_empty_items(trans, root, path, &batch);
6357 btrfs_abort_transaction(trans, ret);
6361 ts = simple_inode_init_ts(inode);
6362 BTRFS_I(inode)->i_otime_sec = ts.tv_sec;
6363 BTRFS_I(inode)->i_otime_nsec = ts.tv_nsec;
6366 * We're going to fill the inode item now, so at this point the inode
6367 * must be fully initialized.
6370 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6371 struct btrfs_inode_item);
6372 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6373 sizeof(*inode_item));
6374 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6376 if (!args->orphan) {
6377 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6378 struct btrfs_inode_ref);
6379 ptr = (unsigned long)(ref + 1);
6381 btrfs_set_inode_ref_name_len(path->nodes[0], ref, 2);
6382 btrfs_set_inode_ref_index(path->nodes[0], ref, 0);
6383 write_extent_buffer(path->nodes[0], "..", ptr, 2);
6385 btrfs_set_inode_ref_name_len(path->nodes[0], ref,
6387 btrfs_set_inode_ref_index(path->nodes[0], ref,
6388 BTRFS_I(inode)->dir_index);
6389 write_extent_buffer(path->nodes[0], name->name, ptr,
6394 btrfs_mark_buffer_dirty(trans, path->nodes[0]);
6396 * We don't need the path anymore, plus inheriting properties, adding
6397 * ACLs, security xattrs, orphan item or adding the link, will result in
6398 * allocating yet another path. So just free our path.
6400 btrfs_free_path(path);
6404 struct inode *parent;
6407 * Subvolumes inherit properties from their parent subvolume,
6408 * not the directory they were created in.
6410 parent = btrfs_iget(fs_info->sb, BTRFS_FIRST_FREE_OBJECTID,
6411 BTRFS_I(dir)->root);
6412 if (IS_ERR(parent)) {
6413 ret = PTR_ERR(parent);
6415 ret = btrfs_inode_inherit_props(trans, inode, parent);
6419 ret = btrfs_inode_inherit_props(trans, inode, dir);
6423 "error inheriting props for ino %llu (root %llu): %d",
6424 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid,
6429 * Subvolumes don't inherit ACLs or get passed to the LSM. This is
6432 if (!args->subvol) {
6433 ret = btrfs_init_inode_security(trans, args);
6435 btrfs_abort_transaction(trans, ret);
6440 inode_tree_add(BTRFS_I(inode));
6442 trace_btrfs_inode_new(inode);
6443 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6445 btrfs_update_root_times(trans, root);
6448 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
6450 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
6451 0, BTRFS_I(inode)->dir_index);
6454 btrfs_abort_transaction(trans, ret);
6462 * discard_new_inode() calls iput(), but the caller owns the reference
6466 discard_new_inode(inode);
6468 btrfs_free_path(path);
6473 * utility function to add 'inode' into 'parent_inode' with
6474 * a give name and a given sequence number.
6475 * if 'add_backref' is true, also insert a backref from the
6476 * inode to the parent directory.
6478 int btrfs_add_link(struct btrfs_trans_handle *trans,
6479 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6480 const struct fscrypt_str *name, int add_backref, u64 index)
6483 struct btrfs_key key;
6484 struct btrfs_root *root = parent_inode->root;
6485 u64 ino = btrfs_ino(inode);
6486 u64 parent_ino = btrfs_ino(parent_inode);
6488 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6489 memcpy(&key, &inode->root->root_key, sizeof(key));
6492 key.type = BTRFS_INODE_ITEM_KEY;
6496 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6497 ret = btrfs_add_root_ref(trans, key.objectid,
6498 root->root_key.objectid, parent_ino,
6500 } else if (add_backref) {
6501 ret = btrfs_insert_inode_ref(trans, root, name,
6502 ino, parent_ino, index);
6505 /* Nothing to clean up yet */
6509 ret = btrfs_insert_dir_item(trans, name, parent_inode, &key,
6510 btrfs_inode_type(&inode->vfs_inode), index);
6511 if (ret == -EEXIST || ret == -EOVERFLOW)
6514 btrfs_abort_transaction(trans, ret);
6518 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6520 inode_inc_iversion(&parent_inode->vfs_inode);
6522 * If we are replaying a log tree, we do not want to update the mtime
6523 * and ctime of the parent directory with the current time, since the
6524 * log replay procedure is responsible for setting them to their correct
6525 * values (the ones it had when the fsync was done).
6527 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags))
6528 inode_set_mtime_to_ts(&parent_inode->vfs_inode,
6529 inode_set_ctime_current(&parent_inode->vfs_inode));
6531 ret = btrfs_update_inode(trans, parent_inode);
6533 btrfs_abort_transaction(trans, ret);
6537 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6540 err = btrfs_del_root_ref(trans, key.objectid,
6541 root->root_key.objectid, parent_ino,
6542 &local_index, name);
6544 btrfs_abort_transaction(trans, err);
6545 } else if (add_backref) {
6549 err = btrfs_del_inode_ref(trans, root, name, ino, parent_ino,
6552 btrfs_abort_transaction(trans, err);
6555 /* Return the original error code */
6559 static int btrfs_create_common(struct inode *dir, struct dentry *dentry,
6560 struct inode *inode)
6562 struct btrfs_fs_info *fs_info = inode_to_fs_info(dir);
6563 struct btrfs_root *root = BTRFS_I(dir)->root;
6564 struct btrfs_new_inode_args new_inode_args = {
6569 unsigned int trans_num_items;
6570 struct btrfs_trans_handle *trans;
6573 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
6577 trans = btrfs_start_transaction(root, trans_num_items);
6578 if (IS_ERR(trans)) {
6579 err = PTR_ERR(trans);
6580 goto out_new_inode_args;
6583 err = btrfs_create_new_inode(trans, &new_inode_args);
6585 d_instantiate_new(dentry, inode);
6587 btrfs_end_transaction(trans);
6588 btrfs_btree_balance_dirty(fs_info);
6590 btrfs_new_inode_args_destroy(&new_inode_args);
6597 static int btrfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
6598 struct dentry *dentry, umode_t mode, dev_t rdev)
6600 struct inode *inode;
6602 inode = new_inode(dir->i_sb);
6605 inode_init_owner(idmap, inode, dir, mode);
6606 inode->i_op = &btrfs_special_inode_operations;
6607 init_special_inode(inode, inode->i_mode, rdev);
6608 return btrfs_create_common(dir, dentry, inode);
6611 static int btrfs_create(struct mnt_idmap *idmap, struct inode *dir,
6612 struct dentry *dentry, umode_t mode, bool excl)
6614 struct inode *inode;
6616 inode = new_inode(dir->i_sb);
6619 inode_init_owner(idmap, inode, dir, mode);
6620 inode->i_fop = &btrfs_file_operations;
6621 inode->i_op = &btrfs_file_inode_operations;
6622 inode->i_mapping->a_ops = &btrfs_aops;
6623 return btrfs_create_common(dir, dentry, inode);
6626 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6627 struct dentry *dentry)
6629 struct btrfs_trans_handle *trans = NULL;
6630 struct btrfs_root *root = BTRFS_I(dir)->root;
6631 struct inode *inode = d_inode(old_dentry);
6632 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
6633 struct fscrypt_name fname;
6638 /* do not allow sys_link's with other subvols of the same device */
6639 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6642 if (inode->i_nlink >= BTRFS_LINK_MAX)
6645 err = fscrypt_setup_filename(dir, &dentry->d_name, 0, &fname);
6649 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6654 * 2 items for inode and inode ref
6655 * 2 items for dir items
6656 * 1 item for parent inode
6657 * 1 item for orphan item deletion if O_TMPFILE
6659 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6660 if (IS_ERR(trans)) {
6661 err = PTR_ERR(trans);
6666 /* There are several dir indexes for this inode, clear the cache. */
6667 BTRFS_I(inode)->dir_index = 0ULL;
6669 inode_inc_iversion(inode);
6670 inode_set_ctime_current(inode);
6672 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6674 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6675 &fname.disk_name, 1, index);
6680 struct dentry *parent = dentry->d_parent;
6682 err = btrfs_update_inode(trans, BTRFS_I(inode));
6685 if (inode->i_nlink == 1) {
6687 * If new hard link count is 1, it's a file created
6688 * with open(2) O_TMPFILE flag.
6690 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6694 d_instantiate(dentry, inode);
6695 btrfs_log_new_name(trans, old_dentry, NULL, 0, parent);
6699 fscrypt_free_filename(&fname);
6701 btrfs_end_transaction(trans);
6703 inode_dec_link_count(inode);
6706 btrfs_btree_balance_dirty(fs_info);
6710 static int btrfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
6711 struct dentry *dentry, umode_t mode)
6713 struct inode *inode;
6715 inode = new_inode(dir->i_sb);
6718 inode_init_owner(idmap, inode, dir, S_IFDIR | mode);
6719 inode->i_op = &btrfs_dir_inode_operations;
6720 inode->i_fop = &btrfs_dir_file_operations;
6721 return btrfs_create_common(dir, dentry, inode);
6724 static noinline int uncompress_inline(struct btrfs_path *path,
6726 struct btrfs_file_extent_item *item)
6729 struct extent_buffer *leaf = path->nodes[0];
6732 unsigned long inline_size;
6736 compress_type = btrfs_file_extent_compression(leaf, item);
6737 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6738 inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]);
6739 tmp = kmalloc(inline_size, GFP_NOFS);
6742 ptr = btrfs_file_extent_inline_start(item);
6744 read_extent_buffer(leaf, tmp, ptr, inline_size);
6746 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6747 ret = btrfs_decompress(compress_type, tmp, page, 0, inline_size, max_size);
6750 * decompression code contains a memset to fill in any space between the end
6751 * of the uncompressed data and the end of max_size in case the decompressed
6752 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6753 * the end of an inline extent and the beginning of the next block, so we
6754 * cover that region here.
6757 if (max_size < PAGE_SIZE)
6758 memzero_page(page, max_size, PAGE_SIZE - max_size);
6763 static int read_inline_extent(struct btrfs_inode *inode, struct btrfs_path *path,
6766 struct btrfs_file_extent_item *fi;
6770 if (!page || PageUptodate(page))
6773 ASSERT(page_offset(page) == 0);
6775 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
6776 struct btrfs_file_extent_item);
6777 if (btrfs_file_extent_compression(path->nodes[0], fi) != BTRFS_COMPRESS_NONE)
6778 return uncompress_inline(path, page, fi);
6780 copy_size = min_t(u64, PAGE_SIZE,
6781 btrfs_file_extent_ram_bytes(path->nodes[0], fi));
6782 kaddr = kmap_local_page(page);
6783 read_extent_buffer(path->nodes[0], kaddr,
6784 btrfs_file_extent_inline_start(fi), copy_size);
6785 kunmap_local(kaddr);
6786 if (copy_size < PAGE_SIZE)
6787 memzero_page(page, copy_size, PAGE_SIZE - copy_size);
6792 * Lookup the first extent overlapping a range in a file.
6794 * @inode: file to search in
6795 * @page: page to read extent data into if the extent is inline
6796 * @start: file offset
6797 * @len: length of range starting at @start
6799 * Return the first &struct extent_map which overlaps the given range, reading
6800 * it from the B-tree and caching it if necessary. Note that there may be more
6801 * extents which overlap the given range after the returned extent_map.
6803 * If @page is not NULL and the extent is inline, this also reads the extent
6804 * data directly into the page and marks the extent up to date in the io_tree.
6806 * Return: ERR_PTR on error, non-NULL extent_map on success.
6808 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6809 struct page *page, u64 start, u64 len)
6811 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6813 u64 extent_start = 0;
6815 u64 objectid = btrfs_ino(inode);
6816 int extent_type = -1;
6817 struct btrfs_path *path = NULL;
6818 struct btrfs_root *root = inode->root;
6819 struct btrfs_file_extent_item *item;
6820 struct extent_buffer *leaf;
6821 struct btrfs_key found_key;
6822 struct extent_map *em = NULL;
6823 struct extent_map_tree *em_tree = &inode->extent_tree;
6825 read_lock(&em_tree->lock);
6826 em = lookup_extent_mapping(em_tree, start, len);
6827 read_unlock(&em_tree->lock);
6830 if (em->start > start || em->start + em->len <= start)
6831 free_extent_map(em);
6832 else if (em->block_start == EXTENT_MAP_INLINE && page)
6833 free_extent_map(em);
6837 em = alloc_extent_map();
6842 em->start = EXTENT_MAP_HOLE;
6843 em->orig_start = EXTENT_MAP_HOLE;
6845 em->block_len = (u64)-1;
6847 path = btrfs_alloc_path();
6853 /* Chances are we'll be called again, so go ahead and do readahead */
6854 path->reada = READA_FORWARD;
6857 * The same explanation in load_free_space_cache applies here as well,
6858 * we only read when we're loading the free space cache, and at that
6859 * point the commit_root has everything we need.
6861 if (btrfs_is_free_space_inode(inode)) {
6862 path->search_commit_root = 1;
6863 path->skip_locking = 1;
6866 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6869 } else if (ret > 0) {
6870 if (path->slots[0] == 0)
6876 leaf = path->nodes[0];
6877 item = btrfs_item_ptr(leaf, path->slots[0],
6878 struct btrfs_file_extent_item);
6879 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6880 if (found_key.objectid != objectid ||
6881 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6883 * If we backup past the first extent we want to move forward
6884 * and see if there is an extent in front of us, otherwise we'll
6885 * say there is a hole for our whole search range which can
6892 extent_type = btrfs_file_extent_type(leaf, item);
6893 extent_start = found_key.offset;
6894 extent_end = btrfs_file_extent_end(path);
6895 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6896 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6897 /* Only regular file could have regular/prealloc extent */
6898 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6901 "regular/prealloc extent found for non-regular inode %llu",
6905 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6907 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6908 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6913 if (start >= extent_end) {
6915 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6916 ret = btrfs_next_leaf(root, path);
6922 leaf = path->nodes[0];
6924 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6925 if (found_key.objectid != objectid ||
6926 found_key.type != BTRFS_EXTENT_DATA_KEY)
6928 if (start + len <= found_key.offset)
6930 if (start > found_key.offset)
6933 /* New extent overlaps with existing one */
6935 em->orig_start = start;
6936 em->len = found_key.offset - start;
6937 em->block_start = EXTENT_MAP_HOLE;
6941 btrfs_extent_item_to_extent_map(inode, path, item, em);
6943 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6944 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6946 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6948 * Inline extent can only exist at file offset 0. This is
6949 * ensured by tree-checker and inline extent creation path.
6950 * Thus all members representing file offsets should be zero.
6952 ASSERT(extent_start == 0);
6953 ASSERT(em->start == 0);
6956 * btrfs_extent_item_to_extent_map() should have properly
6957 * initialized em members already.
6959 * Other members are not utilized for inline extents.
6961 ASSERT(em->block_start == EXTENT_MAP_INLINE);
6962 ASSERT(em->len == fs_info->sectorsize);
6964 ret = read_inline_extent(inode, path, page);
6971 em->orig_start = start;
6973 em->block_start = EXTENT_MAP_HOLE;
6976 btrfs_release_path(path);
6977 if (em->start > start || extent_map_end(em) <= start) {
6979 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6980 em->start, em->len, start, len);
6985 write_lock(&em_tree->lock);
6986 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6987 write_unlock(&em_tree->lock);
6989 btrfs_free_path(path);
6991 trace_btrfs_get_extent(root, inode, em);
6994 free_extent_map(em);
6995 return ERR_PTR(ret);
7000 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
7001 struct btrfs_dio_data *dio_data,
7004 const u64 orig_start,
7005 const u64 block_start,
7006 const u64 block_len,
7007 const u64 orig_block_len,
7008 const u64 ram_bytes,
7011 struct extent_map *em = NULL;
7012 struct btrfs_ordered_extent *ordered;
7014 if (type != BTRFS_ORDERED_NOCOW) {
7015 em = create_io_em(inode, start, len, orig_start, block_start,
7016 block_len, orig_block_len, ram_bytes,
7017 BTRFS_COMPRESS_NONE, /* compress_type */
7022 ordered = btrfs_alloc_ordered_extent(inode, start, len, len,
7023 block_start, block_len, 0,
7025 (1 << BTRFS_ORDERED_DIRECT),
7026 BTRFS_COMPRESS_NONE);
7027 if (IS_ERR(ordered)) {
7029 free_extent_map(em);
7030 btrfs_drop_extent_map_range(inode, start,
7031 start + len - 1, false);
7033 em = ERR_CAST(ordered);
7035 ASSERT(!dio_data->ordered);
7036 dio_data->ordered = ordered;
7043 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
7044 struct btrfs_dio_data *dio_data,
7047 struct btrfs_root *root = inode->root;
7048 struct btrfs_fs_info *fs_info = root->fs_info;
7049 struct extent_map *em;
7050 struct btrfs_key ins;
7054 alloc_hint = get_extent_allocation_hint(inode, start, len);
7056 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7057 0, alloc_hint, &ins, 1, 1);
7058 if (ret == -EAGAIN) {
7059 ASSERT(btrfs_is_zoned(fs_info));
7060 wait_on_bit_io(&inode->root->fs_info->flags, BTRFS_FS_NEED_ZONE_FINISH,
7061 TASK_UNINTERRUPTIBLE);
7065 return ERR_PTR(ret);
7067 em = btrfs_create_dio_extent(inode, dio_data, start, ins.offset, start,
7068 ins.objectid, ins.offset, ins.offset,
7069 ins.offset, BTRFS_ORDERED_REGULAR);
7070 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7072 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7078 static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
7080 struct btrfs_block_group *block_group;
7081 bool readonly = false;
7083 block_group = btrfs_lookup_block_group(fs_info, bytenr);
7084 if (!block_group || block_group->ro)
7087 btrfs_put_block_group(block_group);
7092 * Check if we can do nocow write into the range [@offset, @offset + @len)
7094 * @offset: File offset
7095 * @len: The length to write, will be updated to the nocow writeable
7097 * @orig_start: (optional) Return the original file offset of the file extent
7098 * @orig_len: (optional) Return the original on-disk length of the file extent
7099 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7100 * @strict: if true, omit optimizations that might force us into unnecessary
7101 * cow. e.g., don't trust generation number.
7104 * >0 and update @len if we can do nocow write
7105 * 0 if we can't do nocow write
7106 * <0 if error happened
7108 * NOTE: This only checks the file extents, caller is responsible to wait for
7109 * any ordered extents.
7111 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7112 u64 *orig_start, u64 *orig_block_len,
7113 u64 *ram_bytes, bool nowait, bool strict)
7115 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
7116 struct can_nocow_file_extent_args nocow_args = { 0 };
7117 struct btrfs_path *path;
7119 struct extent_buffer *leaf;
7120 struct btrfs_root *root = BTRFS_I(inode)->root;
7121 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7122 struct btrfs_file_extent_item *fi;
7123 struct btrfs_key key;
7126 path = btrfs_alloc_path();
7129 path->nowait = nowait;
7131 ret = btrfs_lookup_file_extent(NULL, root, path,
7132 btrfs_ino(BTRFS_I(inode)), offset, 0);
7137 if (path->slots[0] == 0) {
7138 /* can't find the item, must cow */
7145 leaf = path->nodes[0];
7146 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7147 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7148 key.type != BTRFS_EXTENT_DATA_KEY) {
7149 /* not our file or wrong item type, must cow */
7153 if (key.offset > offset) {
7154 /* Wrong offset, must cow */
7158 if (btrfs_file_extent_end(path) <= offset)
7161 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
7162 found_type = btrfs_file_extent_type(leaf, fi);
7164 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7166 nocow_args.start = offset;
7167 nocow_args.end = offset + *len - 1;
7168 nocow_args.strict = strict;
7169 nocow_args.free_path = true;
7171 ret = can_nocow_file_extent(path, &key, BTRFS_I(inode), &nocow_args);
7172 /* can_nocow_file_extent() has freed the path. */
7176 /* Treat errors as not being able to NOCOW. */
7182 if (btrfs_extent_readonly(fs_info, nocow_args.disk_bytenr))
7185 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7186 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7189 range_end = round_up(offset + nocow_args.num_bytes,
7190 root->fs_info->sectorsize) - 1;
7191 ret = test_range_bit_exists(io_tree, offset, range_end, EXTENT_DELALLOC);
7199 *orig_start = key.offset - nocow_args.extent_offset;
7201 *orig_block_len = nocow_args.disk_num_bytes;
7203 *len = nocow_args.num_bytes;
7206 btrfs_free_path(path);
7210 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7211 struct extent_state **cached_state,
7212 unsigned int iomap_flags)
7214 const bool writing = (iomap_flags & IOMAP_WRITE);
7215 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7216 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7217 struct btrfs_ordered_extent *ordered;
7222 if (!try_lock_extent(io_tree, lockstart, lockend,
7226 lock_extent(io_tree, lockstart, lockend, cached_state);
7229 * We're concerned with the entire range that we're going to be
7230 * doing DIO to, so we need to make sure there's no ordered
7231 * extents in this range.
7233 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7234 lockend - lockstart + 1);
7237 * We need to make sure there are no buffered pages in this
7238 * range either, we could have raced between the invalidate in
7239 * generic_file_direct_write and locking the extent. The
7240 * invalidate needs to happen so that reads after a write do not
7244 (!writing || !filemap_range_has_page(inode->i_mapping,
7245 lockstart, lockend)))
7248 unlock_extent(io_tree, lockstart, lockend, cached_state);
7252 btrfs_put_ordered_extent(ordered);
7257 * If we are doing a DIO read and the ordered extent we
7258 * found is for a buffered write, we can not wait for it
7259 * to complete and retry, because if we do so we can
7260 * deadlock with concurrent buffered writes on page
7261 * locks. This happens only if our DIO read covers more
7262 * than one extent map, if at this point has already
7263 * created an ordered extent for a previous extent map
7264 * and locked its range in the inode's io tree, and a
7265 * concurrent write against that previous extent map's
7266 * range and this range started (we unlock the ranges
7267 * in the io tree only when the bios complete and
7268 * buffered writes always lock pages before attempting
7269 * to lock range in the io tree).
7272 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7273 btrfs_start_ordered_extent(ordered);
7275 ret = nowait ? -EAGAIN : -ENOTBLK;
7276 btrfs_put_ordered_extent(ordered);
7279 * We could trigger writeback for this range (and wait
7280 * for it to complete) and then invalidate the pages for
7281 * this range (through invalidate_inode_pages2_range()),
7282 * but that can lead us to a deadlock with a concurrent
7283 * call to readahead (a buffered read or a defrag call
7284 * triggered a readahead) on a page lock due to an
7285 * ordered dio extent we created before but did not have
7286 * yet a corresponding bio submitted (whence it can not
7287 * complete), which makes readahead wait for that
7288 * ordered extent to complete while holding a lock on
7291 ret = nowait ? -EAGAIN : -ENOTBLK;
7303 /* The callers of this must take lock_extent() */
7304 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7305 u64 len, u64 orig_start, u64 block_start,
7306 u64 block_len, u64 orig_block_len,
7307 u64 ram_bytes, int compress_type,
7310 struct extent_map *em;
7313 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7314 type == BTRFS_ORDERED_COMPRESSED ||
7315 type == BTRFS_ORDERED_NOCOW ||
7316 type == BTRFS_ORDERED_REGULAR);
7318 em = alloc_extent_map();
7320 return ERR_PTR(-ENOMEM);
7323 em->orig_start = orig_start;
7325 em->block_len = block_len;
7326 em->block_start = block_start;
7327 em->orig_block_len = orig_block_len;
7328 em->ram_bytes = ram_bytes;
7329 em->generation = -1;
7330 em->flags |= EXTENT_FLAG_PINNED;
7331 if (type == BTRFS_ORDERED_PREALLOC)
7332 em->flags |= EXTENT_FLAG_FILLING;
7333 else if (type == BTRFS_ORDERED_COMPRESSED)
7334 extent_map_set_compression(em, compress_type);
7336 ret = btrfs_replace_extent_map_range(inode, em, true);
7338 free_extent_map(em);
7339 return ERR_PTR(ret);
7342 /* em got 2 refs now, callers needs to do free_extent_map once. */
7347 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7348 struct inode *inode,
7349 struct btrfs_dio_data *dio_data,
7350 u64 start, u64 *lenp,
7351 unsigned int iomap_flags)
7353 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7354 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
7355 struct extent_map *em = *map;
7357 u64 block_start, orig_start, orig_block_len, ram_bytes;
7358 struct btrfs_block_group *bg;
7359 bool can_nocow = false;
7360 bool space_reserved = false;
7366 * We don't allocate a new extent in the following cases
7368 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7370 * 2) The extent is marked as PREALLOC. We're good to go here and can
7371 * just use the extent.
7374 if ((em->flags & EXTENT_FLAG_PREALLOC) ||
7375 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7376 em->block_start != EXTENT_MAP_HOLE)) {
7377 if (em->flags & EXTENT_FLAG_PREALLOC)
7378 type = BTRFS_ORDERED_PREALLOC;
7380 type = BTRFS_ORDERED_NOCOW;
7381 len = min(len, em->len - (start - em->start));
7382 block_start = em->block_start + (start - em->start);
7384 if (can_nocow_extent(inode, start, &len, &orig_start,
7385 &orig_block_len, &ram_bytes, false, false) == 1) {
7386 bg = btrfs_inc_nocow_writers(fs_info, block_start);
7394 struct extent_map *em2;
7396 /* We can NOCOW, so only need to reserve metadata space. */
7397 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7400 /* Our caller expects us to free the input extent map. */
7401 free_extent_map(em);
7403 btrfs_dec_nocow_writers(bg);
7404 if (nowait && (ret == -ENOSPC || ret == -EDQUOT))
7408 space_reserved = true;
7410 em2 = btrfs_create_dio_extent(BTRFS_I(inode), dio_data, start, len,
7411 orig_start, block_start,
7412 len, orig_block_len,
7414 btrfs_dec_nocow_writers(bg);
7415 if (type == BTRFS_ORDERED_PREALLOC) {
7416 free_extent_map(em);
7426 dio_data->nocow_done = true;
7428 /* Our caller expects us to free the input extent map. */
7429 free_extent_map(em);
7438 * If we could not allocate data space before locking the file
7439 * range and we can't do a NOCOW write, then we have to fail.
7441 if (!dio_data->data_space_reserved) {
7447 * We have to COW and we have already reserved data space before,
7448 * so now we reserve only metadata.
7450 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7454 space_reserved = true;
7456 em = btrfs_new_extent_direct(BTRFS_I(inode), dio_data, start, len);
7462 len = min(len, em->len - (start - em->start));
7464 btrfs_delalloc_release_metadata(BTRFS_I(inode),
7465 prev_len - len, true);
7469 * We have created our ordered extent, so we can now release our reservation
7470 * for an outstanding extent.
7472 btrfs_delalloc_release_extents(BTRFS_I(inode), prev_len);
7475 * Need to update the i_size under the extent lock so buffered
7476 * readers will get the updated i_size when we unlock.
7478 if (start + len > i_size_read(inode))
7479 i_size_write(inode, start + len);
7481 if (ret && space_reserved) {
7482 btrfs_delalloc_release_extents(BTRFS_I(inode), len);
7483 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, true);
7489 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7490 loff_t length, unsigned int flags, struct iomap *iomap,
7491 struct iomap *srcmap)
7493 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7494 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
7495 struct extent_map *em;
7496 struct extent_state *cached_state = NULL;
7497 struct btrfs_dio_data *dio_data = iter->private;
7498 u64 lockstart, lockend;
7499 const bool write = !!(flags & IOMAP_WRITE);
7502 const u64 data_alloc_len = length;
7503 bool unlock_extents = false;
7506 * We could potentially fault if we have a buffer > PAGE_SIZE, and if
7507 * we're NOWAIT we may submit a bio for a partial range and return
7508 * EIOCBQUEUED, which would result in an errant short read.
7510 * The best way to handle this would be to allow for partial completions
7511 * of iocb's, so we could submit the partial bio, return and fault in
7512 * the rest of the pages, and then submit the io for the rest of the
7513 * range. However we don't have that currently, so simply return
7514 * -EAGAIN at this point so that the normal path is used.
7516 if (!write && (flags & IOMAP_NOWAIT) && length > PAGE_SIZE)
7520 * Cap the size of reads to that usually seen in buffered I/O as we need
7521 * to allocate a contiguous array for the checksums.
7524 len = min_t(u64, len, fs_info->sectorsize * BTRFS_MAX_BIO_SECTORS);
7527 lockend = start + len - 1;
7530 * iomap_dio_rw() only does filemap_write_and_wait_range(), which isn't
7531 * enough if we've written compressed pages to this area, so we need to
7532 * flush the dirty pages again to make absolutely sure that any
7533 * outstanding dirty pages are on disk - the first flush only starts
7534 * compression on the data, while keeping the pages locked, so by the
7535 * time the second flush returns we know bios for the compressed pages
7536 * were submitted and finished, and the pages no longer under writeback.
7538 * If we have a NOWAIT request and we have any pages in the range that
7539 * are locked, likely due to compression still in progress, we don't want
7540 * to block on page locks. We also don't want to block on pages marked as
7541 * dirty or under writeback (same as for the non-compression case).
7542 * iomap_dio_rw() did the same check, but after that and before we got
7543 * here, mmap'ed writes may have happened or buffered reads started
7544 * (readpage() and readahead(), which lock pages), as we haven't locked
7545 * the file range yet.
7547 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7548 &BTRFS_I(inode)->runtime_flags)) {
7549 if (flags & IOMAP_NOWAIT) {
7550 if (filemap_range_needs_writeback(inode->i_mapping,
7551 lockstart, lockend))
7554 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7555 start + length - 1);
7561 memset(dio_data, 0, sizeof(*dio_data));
7564 * We always try to allocate data space and must do it before locking
7565 * the file range, to avoid deadlocks with concurrent writes to the same
7566 * range if the range has several extents and the writes don't expand the
7567 * current i_size (the inode lock is taken in shared mode). If we fail to
7568 * allocate data space here we continue and later, after locking the
7569 * file range, we fail with ENOSPC only if we figure out we can not do a
7572 if (write && !(flags & IOMAP_NOWAIT)) {
7573 ret = btrfs_check_data_free_space(BTRFS_I(inode),
7574 &dio_data->data_reserved,
7575 start, data_alloc_len, false);
7577 dio_data->data_space_reserved = true;
7578 else if (ret && !(BTRFS_I(inode)->flags &
7579 (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
7584 * If this errors out it's because we couldn't invalidate pagecache for
7585 * this range and we need to fallback to buffered IO, or we are doing a
7586 * NOWAIT read/write and we need to block.
7588 ret = lock_extent_direct(inode, lockstart, lockend, &cached_state, flags);
7592 em = btrfs_get_extent(BTRFS_I(inode), NULL, start, len);
7599 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7600 * io. INLINE is special, and we could probably kludge it in here, but
7601 * it's still buffered so for safety lets just fall back to the generic
7604 * For COMPRESSED we _have_ to read the entire extent in so we can
7605 * decompress it, so there will be buffering required no matter what we
7606 * do, so go ahead and fallback to buffered.
7608 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7609 * to buffered IO. Don't blame me, this is the price we pay for using
7612 if (extent_map_is_compressed(em) ||
7613 em->block_start == EXTENT_MAP_INLINE) {
7614 free_extent_map(em);
7616 * If we are in a NOWAIT context, return -EAGAIN in order to
7617 * fallback to buffered IO. This is not only because we can
7618 * block with buffered IO (no support for NOWAIT semantics at
7619 * the moment) but also to avoid returning short reads to user
7620 * space - this happens if we were able to read some data from
7621 * previous non-compressed extents and then when we fallback to
7622 * buffered IO, at btrfs_file_read_iter() by calling
7623 * filemap_read(), we fail to fault in pages for the read buffer,
7624 * in which case filemap_read() returns a short read (the number
7625 * of bytes previously read is > 0, so it does not return -EFAULT).
7627 ret = (flags & IOMAP_NOWAIT) ? -EAGAIN : -ENOTBLK;
7631 len = min(len, em->len - (start - em->start));
7634 * If we have a NOWAIT request and the range contains multiple extents
7635 * (or a mix of extents and holes), then we return -EAGAIN to make the
7636 * caller fallback to a context where it can do a blocking (without
7637 * NOWAIT) request. This way we avoid doing partial IO and returning
7638 * success to the caller, which is not optimal for writes and for reads
7639 * it can result in unexpected behaviour for an application.
7641 * When doing a read, because we use IOMAP_DIO_PARTIAL when calling
7642 * iomap_dio_rw(), we can end up returning less data then what the caller
7643 * asked for, resulting in an unexpected, and incorrect, short read.
7644 * That is, the caller asked to read N bytes and we return less than that,
7645 * which is wrong unless we are crossing EOF. This happens if we get a
7646 * page fault error when trying to fault in pages for the buffer that is
7647 * associated to the struct iov_iter passed to iomap_dio_rw(), and we
7648 * have previously submitted bios for other extents in the range, in
7649 * which case iomap_dio_rw() may return us EIOCBQUEUED if not all of
7650 * those bios have completed by the time we get the page fault error,
7651 * which we return back to our caller - we should only return EIOCBQUEUED
7652 * after we have submitted bios for all the extents in the range.
7654 if ((flags & IOMAP_NOWAIT) && len < length) {
7655 free_extent_map(em);
7661 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7662 start, &len, flags);
7665 unlock_extents = true;
7666 /* Recalc len in case the new em is smaller than requested */
7667 len = min(len, em->len - (start - em->start));
7668 if (dio_data->data_space_reserved) {
7670 u64 release_len = 0;
7672 if (dio_data->nocow_done) {
7673 release_offset = start;
7674 release_len = data_alloc_len;
7675 } else if (len < data_alloc_len) {
7676 release_offset = start + len;
7677 release_len = data_alloc_len - len;
7680 if (release_len > 0)
7681 btrfs_free_reserved_data_space(BTRFS_I(inode),
7682 dio_data->data_reserved,
7688 * We need to unlock only the end area that we aren't using.
7689 * The rest is going to be unlocked by the endio routine.
7691 lockstart = start + len;
7692 if (lockstart < lockend)
7693 unlock_extents = true;
7697 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7700 free_extent_state(cached_state);
7703 * Translate extent map information to iomap.
7704 * We trim the extents (and move the addr) even though iomap code does
7705 * that, since we have locked only the parts we are performing I/O in.
7707 if ((em->block_start == EXTENT_MAP_HOLE) ||
7708 ((em->flags & EXTENT_FLAG_PREALLOC) && !write)) {
7709 iomap->addr = IOMAP_NULL_ADDR;
7710 iomap->type = IOMAP_HOLE;
7712 iomap->addr = em->block_start + (start - em->start);
7713 iomap->type = IOMAP_MAPPED;
7715 iomap->offset = start;
7716 iomap->bdev = fs_info->fs_devices->latest_dev->bdev;
7717 iomap->length = len;
7718 free_extent_map(em);
7723 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7726 if (dio_data->data_space_reserved) {
7727 btrfs_free_reserved_data_space(BTRFS_I(inode),
7728 dio_data->data_reserved,
7729 start, data_alloc_len);
7730 extent_changeset_free(dio_data->data_reserved);
7736 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7737 ssize_t written, unsigned int flags, struct iomap *iomap)
7739 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7740 struct btrfs_dio_data *dio_data = iter->private;
7741 size_t submitted = dio_data->submitted;
7742 const bool write = !!(flags & IOMAP_WRITE);
7745 if (!write && (iomap->type == IOMAP_HOLE)) {
7746 /* If reading from a hole, unlock and return */
7747 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1,
7752 if (submitted < length) {
7754 length -= submitted;
7756 btrfs_finish_ordered_extent(dio_data->ordered, NULL,
7757 pos, length, false);
7759 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7760 pos + length - 1, NULL);
7764 btrfs_put_ordered_extent(dio_data->ordered);
7765 dio_data->ordered = NULL;
7769 extent_changeset_free(dio_data->data_reserved);
7773 static void btrfs_dio_end_io(struct btrfs_bio *bbio)
7775 struct btrfs_dio_private *dip =
7776 container_of(bbio, struct btrfs_dio_private, bbio);
7777 struct btrfs_inode *inode = bbio->inode;
7778 struct bio *bio = &bbio->bio;
7780 if (bio->bi_status) {
7781 btrfs_warn(inode->root->fs_info,
7782 "direct IO failed ino %llu op 0x%0x offset %#llx len %u err no %d",
7783 btrfs_ino(inode), bio->bi_opf,
7784 dip->file_offset, dip->bytes, bio->bi_status);
7787 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
7788 btrfs_finish_ordered_extent(bbio->ordered, NULL,
7789 dip->file_offset, dip->bytes,
7792 unlock_extent(&inode->io_tree, dip->file_offset,
7793 dip->file_offset + dip->bytes - 1, NULL);
7796 bbio->bio.bi_private = bbio->private;
7797 iomap_dio_bio_end_io(bio);
7800 static void btrfs_dio_submit_io(const struct iomap_iter *iter, struct bio *bio,
7803 struct btrfs_bio *bbio = btrfs_bio(bio);
7804 struct btrfs_dio_private *dip =
7805 container_of(bbio, struct btrfs_dio_private, bbio);
7806 struct btrfs_dio_data *dio_data = iter->private;
7808 btrfs_bio_init(bbio, BTRFS_I(iter->inode)->root->fs_info,
7809 btrfs_dio_end_io, bio->bi_private);
7810 bbio->inode = BTRFS_I(iter->inode);
7811 bbio->file_offset = file_offset;
7813 dip->file_offset = file_offset;
7814 dip->bytes = bio->bi_iter.bi_size;
7816 dio_data->submitted += bio->bi_iter.bi_size;
7819 * Check if we are doing a partial write. If we are, we need to split
7820 * the ordered extent to match the submitted bio. Hang on to the
7821 * remaining unfinishable ordered_extent in dio_data so that it can be
7822 * cancelled in iomap_end to avoid a deadlock wherein faulting the
7823 * remaining pages is blocked on the outstanding ordered extent.
7825 if (iter->flags & IOMAP_WRITE) {
7828 ret = btrfs_extract_ordered_extent(bbio, dio_data->ordered);
7830 btrfs_finish_ordered_extent(dio_data->ordered, NULL,
7831 file_offset, dip->bytes,
7833 bio->bi_status = errno_to_blk_status(ret);
7834 iomap_dio_bio_end_io(bio);
7839 btrfs_submit_bio(bbio, 0);
7842 static const struct iomap_ops btrfs_dio_iomap_ops = {
7843 .iomap_begin = btrfs_dio_iomap_begin,
7844 .iomap_end = btrfs_dio_iomap_end,
7847 static const struct iomap_dio_ops btrfs_dio_ops = {
7848 .submit_io = btrfs_dio_submit_io,
7849 .bio_set = &btrfs_dio_bioset,
7852 ssize_t btrfs_dio_read(struct kiocb *iocb, struct iov_iter *iter, size_t done_before)
7854 struct btrfs_dio_data data = { 0 };
7856 return iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7857 IOMAP_DIO_PARTIAL, &data, done_before);
7860 struct iomap_dio *btrfs_dio_write(struct kiocb *iocb, struct iov_iter *iter,
7863 struct btrfs_dio_data data = { 0 };
7865 return __iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7866 IOMAP_DIO_PARTIAL, &data, done_before);
7869 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7872 struct btrfs_inode *btrfs_inode = BTRFS_I(inode);
7875 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
7880 * fiemap_prep() called filemap_write_and_wait() for the whole possible
7881 * file range (0 to LLONG_MAX), but that is not enough if we have
7882 * compression enabled. The first filemap_fdatawrite_range() only kicks
7883 * in the compression of data (in an async thread) and will return
7884 * before the compression is done and writeback is started. A second
7885 * filemap_fdatawrite_range() is needed to wait for the compression to
7886 * complete and writeback to start. We also need to wait for ordered
7887 * extents to complete, because our fiemap implementation uses mainly
7888 * file extent items to list the extents, searching for extent maps
7889 * only for file ranges with holes or prealloc extents to figure out
7890 * if we have delalloc in those ranges.
7892 if (fieinfo->fi_flags & FIEMAP_FLAG_SYNC) {
7893 ret = btrfs_wait_ordered_range(inode, 0, LLONG_MAX);
7898 btrfs_inode_lock(btrfs_inode, BTRFS_ILOCK_SHARED);
7901 * We did an initial flush to avoid holding the inode's lock while
7902 * triggering writeback and waiting for the completion of IO and ordered
7903 * extents. Now after we locked the inode we do it again, because it's
7904 * possible a new write may have happened in between those two steps.
7906 if (fieinfo->fi_flags & FIEMAP_FLAG_SYNC) {
7907 ret = btrfs_wait_ordered_range(inode, 0, LLONG_MAX);
7909 btrfs_inode_unlock(btrfs_inode, BTRFS_ILOCK_SHARED);
7914 ret = extent_fiemap(btrfs_inode, fieinfo, start, len);
7915 btrfs_inode_unlock(btrfs_inode, BTRFS_ILOCK_SHARED);
7920 static int btrfs_writepages(struct address_space *mapping,
7921 struct writeback_control *wbc)
7923 return extent_writepages(mapping, wbc);
7926 static void btrfs_readahead(struct readahead_control *rac)
7928 extent_readahead(rac);
7932 * For release_folio() and invalidate_folio() we have a race window where
7933 * folio_end_writeback() is called but the subpage spinlock is not yet released.
7934 * If we continue to release/invalidate the page, we could cause use-after-free
7935 * for subpage spinlock. So this function is to spin and wait for subpage
7938 static void wait_subpage_spinlock(struct page *page)
7940 struct btrfs_fs_info *fs_info = page_to_fs_info(page);
7941 struct folio *folio = page_folio(page);
7942 struct btrfs_subpage *subpage;
7944 if (!btrfs_is_subpage(fs_info, page->mapping))
7947 ASSERT(folio_test_private(folio) && folio_get_private(folio));
7948 subpage = folio_get_private(folio);
7951 * This may look insane as we just acquire the spinlock and release it,
7952 * without doing anything. But we just want to make sure no one is
7953 * still holding the subpage spinlock.
7954 * And since the page is not dirty nor writeback, and we have page
7955 * locked, the only possible way to hold a spinlock is from the endio
7956 * function to clear page writeback.
7958 * Here we just acquire the spinlock so that all existing callers
7959 * should exit and we're safe to release/invalidate the page.
7961 spin_lock_irq(&subpage->lock);
7962 spin_unlock_irq(&subpage->lock);
7965 static bool __btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
7967 int ret = try_release_extent_mapping(&folio->page, gfp_flags);
7970 wait_subpage_spinlock(&folio->page);
7971 clear_page_extent_mapped(&folio->page);
7976 static bool btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
7978 if (folio_test_writeback(folio) || folio_test_dirty(folio))
7980 return __btrfs_release_folio(folio, gfp_flags);
7983 #ifdef CONFIG_MIGRATION
7984 static int btrfs_migrate_folio(struct address_space *mapping,
7985 struct folio *dst, struct folio *src,
7986 enum migrate_mode mode)
7988 int ret = filemap_migrate_folio(mapping, dst, src, mode);
7990 if (ret != MIGRATEPAGE_SUCCESS)
7993 if (folio_test_ordered(src)) {
7994 folio_clear_ordered(src);
7995 folio_set_ordered(dst);
7998 return MIGRATEPAGE_SUCCESS;
8001 #define btrfs_migrate_folio NULL
8004 static void btrfs_invalidate_folio(struct folio *folio, size_t offset,
8007 struct btrfs_inode *inode = folio_to_inode(folio);
8008 struct btrfs_fs_info *fs_info = inode->root->fs_info;
8009 struct extent_io_tree *tree = &inode->io_tree;
8010 struct extent_state *cached_state = NULL;
8011 u64 page_start = folio_pos(folio);
8012 u64 page_end = page_start + folio_size(folio) - 1;
8014 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
8017 * We have folio locked so no new ordered extent can be created on this
8018 * page, nor bio can be submitted for this folio.
8020 * But already submitted bio can still be finished on this folio.
8021 * Furthermore, endio function won't skip folio which has Ordered
8022 * (Private2) already cleared, so it's possible for endio and
8023 * invalidate_folio to do the same ordered extent accounting twice
8026 * So here we wait for any submitted bios to finish, so that we won't
8027 * do double ordered extent accounting on the same folio.
8029 folio_wait_writeback(folio);
8030 wait_subpage_spinlock(&folio->page);
8033 * For subpage case, we have call sites like
8034 * btrfs_punch_hole_lock_range() which passes range not aligned to
8036 * If the range doesn't cover the full folio, we don't need to and
8037 * shouldn't clear page extent mapped, as folio->private can still
8038 * record subpage dirty bits for other part of the range.
8040 * For cases that invalidate the full folio even the range doesn't
8041 * cover the full folio, like invalidating the last folio, we're
8042 * still safe to wait for ordered extent to finish.
8044 if (!(offset == 0 && length == folio_size(folio))) {
8045 btrfs_release_folio(folio, GFP_NOFS);
8049 if (!inode_evicting)
8050 lock_extent(tree, page_start, page_end, &cached_state);
8053 while (cur < page_end) {
8054 struct btrfs_ordered_extent *ordered;
8057 u32 extra_flags = 0;
8059 ordered = btrfs_lookup_first_ordered_range(inode, cur,
8060 page_end + 1 - cur);
8062 range_end = page_end;
8064 * No ordered extent covering this range, we are safe
8065 * to delete all extent states in the range.
8067 extra_flags = EXTENT_CLEAR_ALL_BITS;
8070 if (ordered->file_offset > cur) {
8072 * There is a range between [cur, oe->file_offset) not
8073 * covered by any ordered extent.
8074 * We are safe to delete all extent states, and handle
8075 * the ordered extent in the next iteration.
8077 range_end = ordered->file_offset - 1;
8078 extra_flags = EXTENT_CLEAR_ALL_BITS;
8082 range_end = min(ordered->file_offset + ordered->num_bytes - 1,
8084 ASSERT(range_end + 1 - cur < U32_MAX);
8085 range_len = range_end + 1 - cur;
8086 if (!btrfs_folio_test_ordered(fs_info, folio, cur, range_len)) {
8088 * If Ordered (Private2) is cleared, it means endio has
8089 * already been executed for the range.
8090 * We can't delete the extent states as
8091 * btrfs_finish_ordered_io() may still use some of them.
8095 btrfs_folio_clear_ordered(fs_info, folio, cur, range_len);
8098 * IO on this page will never be started, so we need to account
8099 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8100 * here, must leave that up for the ordered extent completion.
8102 * This will also unlock the range for incoming
8103 * btrfs_finish_ordered_io().
8105 if (!inode_evicting)
8106 clear_extent_bit(tree, cur, range_end,
8108 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8109 EXTENT_DEFRAG, &cached_state);
8111 spin_lock_irq(&inode->ordered_tree_lock);
8112 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8113 ordered->truncated_len = min(ordered->truncated_len,
8114 cur - ordered->file_offset);
8115 spin_unlock_irq(&inode->ordered_tree_lock);
8118 * If the ordered extent has finished, we're safe to delete all
8119 * the extent states of the range, otherwise
8120 * btrfs_finish_ordered_io() will get executed by endio for
8121 * other pages, so we can't delete extent states.
8123 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8124 cur, range_end + 1 - cur)) {
8125 btrfs_finish_ordered_io(ordered);
8127 * The ordered extent has finished, now we're again
8128 * safe to delete all extent states of the range.
8130 extra_flags = EXTENT_CLEAR_ALL_BITS;
8134 btrfs_put_ordered_extent(ordered);
8136 * Qgroup reserved space handler
8137 * Sector(s) here will be either:
8139 * 1) Already written to disk or bio already finished
8140 * Then its QGROUP_RESERVED bit in io_tree is already cleared.
8141 * Qgroup will be handled by its qgroup_record then.
8142 * btrfs_qgroup_free_data() call will do nothing here.
8144 * 2) Not written to disk yet
8145 * Then btrfs_qgroup_free_data() call will clear the
8146 * QGROUP_RESERVED bit of its io_tree, and free the qgroup
8147 * reserved data space.
8148 * Since the IO will never happen for this page.
8150 btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur, NULL);
8151 if (!inode_evicting) {
8152 clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED |
8153 EXTENT_DELALLOC | EXTENT_UPTODATE |
8154 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG |
8155 extra_flags, &cached_state);
8157 cur = range_end + 1;
8160 * We have iterated through all ordered extents of the page, the page
8161 * should not have Ordered (Private2) anymore, or the above iteration
8162 * did something wrong.
8164 ASSERT(!folio_test_ordered(folio));
8165 btrfs_folio_clear_checked(fs_info, folio, folio_pos(folio), folio_size(folio));
8166 if (!inode_evicting)
8167 __btrfs_release_folio(folio, GFP_NOFS);
8168 clear_page_extent_mapped(&folio->page);
8172 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8173 * called from a page fault handler when a page is first dirtied. Hence we must
8174 * be careful to check for EOF conditions here. We set the page up correctly
8175 * for a written page which means we get ENOSPC checking when writing into
8176 * holes and correct delalloc and unwritten extent mapping on filesystems that
8177 * support these features.
8179 * We are not allowed to take the i_mutex here so we have to play games to
8180 * protect against truncate races as the page could now be beyond EOF. Because
8181 * truncate_setsize() writes the inode size before removing pages, once we have
8182 * the page lock we can determine safely if the page is beyond EOF. If it is not
8183 * beyond EOF, then the page is guaranteed safe against truncation until we
8186 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8188 struct page *page = vmf->page;
8189 struct folio *folio = page_folio(page);
8190 struct inode *inode = file_inode(vmf->vma->vm_file);
8191 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
8192 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8193 struct btrfs_ordered_extent *ordered;
8194 struct extent_state *cached_state = NULL;
8195 struct extent_changeset *data_reserved = NULL;
8196 unsigned long zero_start;
8206 ASSERT(folio_order(folio) == 0);
8208 reserved_space = PAGE_SIZE;
8210 sb_start_pagefault(inode->i_sb);
8211 page_start = page_offset(page);
8212 page_end = page_start + PAGE_SIZE - 1;
8216 * Reserving delalloc space after obtaining the page lock can lead to
8217 * deadlock. For example, if a dirty page is locked by this function
8218 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8219 * dirty page write out, then the btrfs_writepages() function could
8220 * end up waiting indefinitely to get a lock on the page currently
8221 * being processed by btrfs_page_mkwrite() function.
8223 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8224 page_start, reserved_space);
8226 ret2 = file_update_time(vmf->vma->vm_file);
8230 ret = vmf_error(ret2);
8236 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8238 down_read(&BTRFS_I(inode)->i_mmap_lock);
8240 size = i_size_read(inode);
8242 if ((page->mapping != inode->i_mapping) ||
8243 (page_start >= size)) {
8244 /* page got truncated out from underneath us */
8247 wait_on_page_writeback(page);
8249 lock_extent(io_tree, page_start, page_end, &cached_state);
8250 ret2 = set_page_extent_mapped(page);
8252 ret = vmf_error(ret2);
8253 unlock_extent(io_tree, page_start, page_end, &cached_state);
8258 * we can't set the delalloc bits if there are pending ordered
8259 * extents. Drop our locks and wait for them to finish
8261 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8264 unlock_extent(io_tree, page_start, page_end, &cached_state);
8266 up_read(&BTRFS_I(inode)->i_mmap_lock);
8267 btrfs_start_ordered_extent(ordered);
8268 btrfs_put_ordered_extent(ordered);
8272 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8273 reserved_space = round_up(size - page_start,
8274 fs_info->sectorsize);
8275 if (reserved_space < PAGE_SIZE) {
8276 end = page_start + reserved_space - 1;
8277 btrfs_delalloc_release_space(BTRFS_I(inode),
8278 data_reserved, page_start,
8279 PAGE_SIZE - reserved_space, true);
8284 * page_mkwrite gets called when the page is firstly dirtied after it's
8285 * faulted in, but write(2) could also dirty a page and set delalloc
8286 * bits, thus in this case for space account reason, we still need to
8287 * clear any delalloc bits within this page range since we have to
8288 * reserve data&meta space before lock_page() (see above comments).
8290 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8291 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8292 EXTENT_DEFRAG, &cached_state);
8294 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8297 unlock_extent(io_tree, page_start, page_end, &cached_state);
8298 ret = VM_FAULT_SIGBUS;
8302 /* page is wholly or partially inside EOF */
8303 if (page_start + PAGE_SIZE > size)
8304 zero_start = offset_in_page(size);
8306 zero_start = PAGE_SIZE;
8308 if (zero_start != PAGE_SIZE)
8309 memzero_page(page, zero_start, PAGE_SIZE - zero_start);
8311 btrfs_folio_clear_checked(fs_info, folio, page_start, PAGE_SIZE);
8312 btrfs_folio_set_dirty(fs_info, folio, page_start, end + 1 - page_start);
8313 btrfs_folio_set_uptodate(fs_info, folio, page_start, end + 1 - page_start);
8315 btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
8317 unlock_extent(io_tree, page_start, page_end, &cached_state);
8318 up_read(&BTRFS_I(inode)->i_mmap_lock);
8320 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8321 sb_end_pagefault(inode->i_sb);
8322 extent_changeset_free(data_reserved);
8323 return VM_FAULT_LOCKED;
8327 up_read(&BTRFS_I(inode)->i_mmap_lock);
8329 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8330 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8331 reserved_space, (ret != 0));
8333 sb_end_pagefault(inode->i_sb);
8334 extent_changeset_free(data_reserved);
8338 static int btrfs_truncate(struct btrfs_inode *inode, bool skip_writeback)
8340 struct btrfs_truncate_control control = {
8342 .ino = btrfs_ino(inode),
8343 .min_type = BTRFS_EXTENT_DATA_KEY,
8344 .clear_extent_range = true,
8346 struct btrfs_root *root = inode->root;
8347 struct btrfs_fs_info *fs_info = root->fs_info;
8348 struct btrfs_block_rsv *rsv;
8350 struct btrfs_trans_handle *trans;
8351 u64 mask = fs_info->sectorsize - 1;
8352 const u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8354 if (!skip_writeback) {
8355 ret = btrfs_wait_ordered_range(&inode->vfs_inode,
8356 inode->vfs_inode.i_size & (~mask),
8363 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8364 * things going on here:
8366 * 1) We need to reserve space to update our inode.
8368 * 2) We need to have something to cache all the space that is going to
8369 * be free'd up by the truncate operation, but also have some slack
8370 * space reserved in case it uses space during the truncate (thank you
8371 * very much snapshotting).
8373 * And we need these to be separate. The fact is we can use a lot of
8374 * space doing the truncate, and we have no earthly idea how much space
8375 * we will use, so we need the truncate reservation to be separate so it
8376 * doesn't end up using space reserved for updating the inode. We also
8377 * need to be able to stop the transaction and start a new one, which
8378 * means we need to be able to update the inode several times, and we
8379 * have no idea of knowing how many times that will be, so we can't just
8380 * reserve 1 item for the entirety of the operation, so that has to be
8381 * done separately as well.
8383 * So that leaves us with
8385 * 1) rsv - for the truncate reservation, which we will steal from the
8386 * transaction reservation.
8387 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8388 * updating the inode.
8390 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8393 rsv->size = min_size;
8394 rsv->failfast = true;
8397 * 1 for the truncate slack space
8398 * 1 for updating the inode.
8400 trans = btrfs_start_transaction(root, 2);
8401 if (IS_ERR(trans)) {
8402 ret = PTR_ERR(trans);
8406 /* Migrate the slack space for the truncate to our reserve */
8407 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8410 * We have reserved 2 metadata units when we started the transaction and
8411 * min_size matches 1 unit, so this should never fail, but if it does,
8412 * it's not critical we just fail truncation.
8415 btrfs_end_transaction(trans);
8419 trans->block_rsv = rsv;
8422 struct extent_state *cached_state = NULL;
8423 const u64 new_size = inode->vfs_inode.i_size;
8424 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
8426 control.new_size = new_size;
8427 lock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8429 * We want to drop from the next block forward in case this new
8430 * size is not block aligned since we will be keeping the last
8431 * block of the extent just the way it is.
8433 btrfs_drop_extent_map_range(inode,
8434 ALIGN(new_size, fs_info->sectorsize),
8437 ret = btrfs_truncate_inode_items(trans, root, &control);
8439 inode_sub_bytes(&inode->vfs_inode, control.sub_bytes);
8440 btrfs_inode_safe_disk_i_size_write(inode, control.last_size);
8442 unlock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8444 trans->block_rsv = &fs_info->trans_block_rsv;
8445 if (ret != -ENOSPC && ret != -EAGAIN)
8448 ret = btrfs_update_inode(trans, inode);
8452 btrfs_end_transaction(trans);
8453 btrfs_btree_balance_dirty(fs_info);
8455 trans = btrfs_start_transaction(root, 2);
8456 if (IS_ERR(trans)) {
8457 ret = PTR_ERR(trans);
8462 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8463 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8464 rsv, min_size, false);
8466 * We have reserved 2 metadata units when we started the
8467 * transaction and min_size matches 1 unit, so this should never
8468 * fail, but if it does, it's not critical we just fail truncation.
8473 trans->block_rsv = rsv;
8477 * We can't call btrfs_truncate_block inside a trans handle as we could
8478 * deadlock with freeze, if we got BTRFS_NEED_TRUNCATE_BLOCK then we
8479 * know we've truncated everything except the last little bit, and can
8480 * do btrfs_truncate_block and then update the disk_i_size.
8482 if (ret == BTRFS_NEED_TRUNCATE_BLOCK) {
8483 btrfs_end_transaction(trans);
8484 btrfs_btree_balance_dirty(fs_info);
8486 ret = btrfs_truncate_block(inode, inode->vfs_inode.i_size, 0, 0);
8489 trans = btrfs_start_transaction(root, 1);
8490 if (IS_ERR(trans)) {
8491 ret = PTR_ERR(trans);
8494 btrfs_inode_safe_disk_i_size_write(inode, 0);
8500 trans->block_rsv = &fs_info->trans_block_rsv;
8501 ret2 = btrfs_update_inode(trans, inode);
8505 ret2 = btrfs_end_transaction(trans);
8508 btrfs_btree_balance_dirty(fs_info);
8511 btrfs_free_block_rsv(fs_info, rsv);
8513 * So if we truncate and then write and fsync we normally would just
8514 * write the extents that changed, which is a problem if we need to
8515 * first truncate that entire inode. So set this flag so we write out
8516 * all of the extents in the inode to the sync log so we're completely
8519 * If no extents were dropped or trimmed we don't need to force the next
8520 * fsync to truncate all the inode's items from the log and re-log them
8521 * all. This means the truncate operation did not change the file size,
8522 * or changed it to a smaller size but there was only an implicit hole
8523 * between the old i_size and the new i_size, and there were no prealloc
8524 * extents beyond i_size to drop.
8526 if (control.extents_found > 0)
8527 btrfs_set_inode_full_sync(inode);
8532 struct inode *btrfs_new_subvol_inode(struct mnt_idmap *idmap,
8535 struct inode *inode;
8537 inode = new_inode(dir->i_sb);
8540 * Subvolumes don't inherit the sgid bit or the parent's gid if
8541 * the parent's sgid bit is set. This is probably a bug.
8543 inode_init_owner(idmap, inode, NULL,
8544 S_IFDIR | (~current_umask() & S_IRWXUGO));
8545 inode->i_op = &btrfs_dir_inode_operations;
8546 inode->i_fop = &btrfs_dir_file_operations;
8551 struct inode *btrfs_alloc_inode(struct super_block *sb)
8553 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8554 struct btrfs_inode *ei;
8555 struct inode *inode;
8556 struct extent_io_tree *file_extent_tree = NULL;
8558 /* Self tests may pass a NULL fs_info. */
8559 if (fs_info && !btrfs_fs_incompat(fs_info, NO_HOLES)) {
8560 file_extent_tree = kmalloc(sizeof(struct extent_io_tree), GFP_KERNEL);
8561 if (!file_extent_tree)
8565 ei = alloc_inode_sb(sb, btrfs_inode_cachep, GFP_KERNEL);
8567 kfree(file_extent_tree);
8574 ei->last_sub_trans = 0;
8575 ei->logged_trans = 0;
8576 ei->delalloc_bytes = 0;
8577 ei->new_delalloc_bytes = 0;
8578 ei->defrag_bytes = 0;
8579 ei->disk_i_size = 0;
8583 ei->index_cnt = (u64)-1;
8585 ei->last_unlink_trans = 0;
8586 ei->last_reflink_trans = 0;
8587 ei->last_log_commit = 0;
8589 spin_lock_init(&ei->lock);
8590 ei->outstanding_extents = 0;
8591 if (sb->s_magic != BTRFS_TEST_MAGIC)
8592 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8593 BTRFS_BLOCK_RSV_DELALLOC);
8594 ei->runtime_flags = 0;
8595 ei->prop_compress = BTRFS_COMPRESS_NONE;
8596 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8598 ei->delayed_node = NULL;
8600 ei->i_otime_sec = 0;
8601 ei->i_otime_nsec = 0;
8603 inode = &ei->vfs_inode;
8604 extent_map_tree_init(&ei->extent_tree);
8606 /* This io tree sets the valid inode. */
8607 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO);
8608 ei->io_tree.inode = ei;
8610 ei->file_extent_tree = file_extent_tree;
8611 if (file_extent_tree) {
8612 extent_io_tree_init(fs_info, ei->file_extent_tree,
8613 IO_TREE_INODE_FILE_EXTENT);
8614 /* Lockdep class is set only for the file extent tree. */
8615 lockdep_set_class(&ei->file_extent_tree->lock, &file_extent_tree_class);
8617 mutex_init(&ei->log_mutex);
8618 spin_lock_init(&ei->ordered_tree_lock);
8619 ei->ordered_tree = RB_ROOT;
8620 ei->ordered_tree_last = NULL;
8621 INIT_LIST_HEAD(&ei->delalloc_inodes);
8622 INIT_LIST_HEAD(&ei->delayed_iput);
8623 RB_CLEAR_NODE(&ei->rb_node);
8624 init_rwsem(&ei->i_mmap_lock);
8629 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8630 void btrfs_test_destroy_inode(struct inode *inode)
8632 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
8633 kfree(BTRFS_I(inode)->file_extent_tree);
8634 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8638 void btrfs_free_inode(struct inode *inode)
8640 kfree(BTRFS_I(inode)->file_extent_tree);
8641 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8644 void btrfs_destroy_inode(struct inode *vfs_inode)
8646 struct btrfs_ordered_extent *ordered;
8647 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8648 struct btrfs_root *root = inode->root;
8649 bool freespace_inode;
8651 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8652 WARN_ON(vfs_inode->i_data.nrpages);
8653 WARN_ON(inode->block_rsv.reserved);
8654 WARN_ON(inode->block_rsv.size);
8655 WARN_ON(inode->outstanding_extents);
8656 if (!S_ISDIR(vfs_inode->i_mode)) {
8657 WARN_ON(inode->delalloc_bytes);
8658 WARN_ON(inode->new_delalloc_bytes);
8660 WARN_ON(inode->csum_bytes);
8661 WARN_ON(inode->defrag_bytes);
8664 * This can happen where we create an inode, but somebody else also
8665 * created the same inode and we need to destroy the one we already
8672 * If this is a free space inode do not take the ordered extents lockdep
8675 freespace_inode = btrfs_is_free_space_inode(inode);
8678 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8682 btrfs_err(root->fs_info,
8683 "found ordered extent %llu %llu on inode cleanup",
8684 ordered->file_offset, ordered->num_bytes);
8686 if (!freespace_inode)
8687 btrfs_lockdep_acquire(root->fs_info, btrfs_ordered_extent);
8689 btrfs_remove_ordered_extent(inode, ordered);
8690 btrfs_put_ordered_extent(ordered);
8691 btrfs_put_ordered_extent(ordered);
8694 btrfs_qgroup_check_reserved_leak(inode);
8695 inode_tree_del(inode);
8696 btrfs_drop_extent_map_range(inode, 0, (u64)-1, false);
8697 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
8698 btrfs_put_root(inode->root);
8701 int btrfs_drop_inode(struct inode *inode)
8703 struct btrfs_root *root = BTRFS_I(inode)->root;
8708 /* the snap/subvol tree is on deleting */
8709 if (btrfs_root_refs(&root->root_item) == 0)
8712 return generic_drop_inode(inode);
8715 static void init_once(void *foo)
8717 struct btrfs_inode *ei = foo;
8719 inode_init_once(&ei->vfs_inode);
8722 void __cold btrfs_destroy_cachep(void)
8725 * Make sure all delayed rcu free inodes are flushed before we
8729 bioset_exit(&btrfs_dio_bioset);
8730 kmem_cache_destroy(btrfs_inode_cachep);
8733 int __init btrfs_init_cachep(void)
8735 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8736 sizeof(struct btrfs_inode), 0,
8737 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
8739 if (!btrfs_inode_cachep)
8742 if (bioset_init(&btrfs_dio_bioset, BIO_POOL_SIZE,
8743 offsetof(struct btrfs_dio_private, bbio.bio),
8749 btrfs_destroy_cachep();
8753 static int btrfs_getattr(struct mnt_idmap *idmap,
8754 const struct path *path, struct kstat *stat,
8755 u32 request_mask, unsigned int flags)
8759 struct inode *inode = d_inode(path->dentry);
8760 u32 blocksize = btrfs_sb(inode->i_sb)->sectorsize;
8761 u32 bi_flags = BTRFS_I(inode)->flags;
8762 u32 bi_ro_flags = BTRFS_I(inode)->ro_flags;
8764 stat->result_mask |= STATX_BTIME;
8765 stat->btime.tv_sec = BTRFS_I(inode)->i_otime_sec;
8766 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime_nsec;
8767 if (bi_flags & BTRFS_INODE_APPEND)
8768 stat->attributes |= STATX_ATTR_APPEND;
8769 if (bi_flags & BTRFS_INODE_COMPRESS)
8770 stat->attributes |= STATX_ATTR_COMPRESSED;
8771 if (bi_flags & BTRFS_INODE_IMMUTABLE)
8772 stat->attributes |= STATX_ATTR_IMMUTABLE;
8773 if (bi_flags & BTRFS_INODE_NODUMP)
8774 stat->attributes |= STATX_ATTR_NODUMP;
8775 if (bi_ro_flags & BTRFS_INODE_RO_VERITY)
8776 stat->attributes |= STATX_ATTR_VERITY;
8778 stat->attributes_mask |= (STATX_ATTR_APPEND |
8779 STATX_ATTR_COMPRESSED |
8780 STATX_ATTR_IMMUTABLE |
8783 generic_fillattr(idmap, request_mask, inode, stat);
8784 stat->dev = BTRFS_I(inode)->root->anon_dev;
8786 spin_lock(&BTRFS_I(inode)->lock);
8787 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
8788 inode_bytes = inode_get_bytes(inode);
8789 spin_unlock(&BTRFS_I(inode)->lock);
8790 stat->blocks = (ALIGN(inode_bytes, blocksize) +
8791 ALIGN(delalloc_bytes, blocksize)) >> SECTOR_SHIFT;
8795 static int btrfs_rename_exchange(struct inode *old_dir,
8796 struct dentry *old_dentry,
8797 struct inode *new_dir,
8798 struct dentry *new_dentry)
8800 struct btrfs_fs_info *fs_info = inode_to_fs_info(old_dir);
8801 struct btrfs_trans_handle *trans;
8802 unsigned int trans_num_items;
8803 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8804 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8805 struct inode *new_inode = new_dentry->d_inode;
8806 struct inode *old_inode = old_dentry->d_inode;
8807 struct btrfs_rename_ctx old_rename_ctx;
8808 struct btrfs_rename_ctx new_rename_ctx;
8809 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
8810 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
8815 bool need_abort = false;
8816 struct fscrypt_name old_fname, new_fname;
8817 struct fscrypt_str *old_name, *new_name;
8820 * For non-subvolumes allow exchange only within one subvolume, in the
8821 * same inode namespace. Two subvolumes (represented as directory) can
8822 * be exchanged as they're a logical link and have a fixed inode number.
8825 (old_ino != BTRFS_FIRST_FREE_OBJECTID ||
8826 new_ino != BTRFS_FIRST_FREE_OBJECTID))
8829 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
8833 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
8835 fscrypt_free_filename(&old_fname);
8839 old_name = &old_fname.disk_name;
8840 new_name = &new_fname.disk_name;
8842 /* close the race window with snapshot create/destroy ioctl */
8843 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
8844 new_ino == BTRFS_FIRST_FREE_OBJECTID)
8845 down_read(&fs_info->subvol_sem);
8849 * 1 to remove old dir item
8850 * 1 to remove old dir index
8851 * 1 to add new dir item
8852 * 1 to add new dir index
8853 * 1 to update parent inode
8855 * If the parents are the same, we only need to account for one
8857 trans_num_items = (old_dir == new_dir ? 9 : 10);
8858 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8860 * 1 to remove old root ref
8861 * 1 to remove old root backref
8862 * 1 to add new root ref
8863 * 1 to add new root backref
8865 trans_num_items += 4;
8868 * 1 to update inode item
8869 * 1 to remove old inode ref
8870 * 1 to add new inode ref
8872 trans_num_items += 3;
8874 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
8875 trans_num_items += 4;
8877 trans_num_items += 3;
8878 trans = btrfs_start_transaction(root, trans_num_items);
8879 if (IS_ERR(trans)) {
8880 ret = PTR_ERR(trans);
8885 ret = btrfs_record_root_in_trans(trans, dest);
8891 * We need to find a free sequence number both in the source and
8892 * in the destination directory for the exchange.
8894 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
8897 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
8901 BTRFS_I(old_inode)->dir_index = 0ULL;
8902 BTRFS_I(new_inode)->dir_index = 0ULL;
8904 /* Reference for the source. */
8905 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8906 /* force full log commit if subvolume involved. */
8907 btrfs_set_log_full_commit(trans);
8909 ret = btrfs_insert_inode_ref(trans, dest, new_name, old_ino,
8910 btrfs_ino(BTRFS_I(new_dir)),
8917 /* And now for the dest. */
8918 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8919 /* force full log commit if subvolume involved. */
8920 btrfs_set_log_full_commit(trans);
8922 ret = btrfs_insert_inode_ref(trans, root, old_name, new_ino,
8923 btrfs_ino(BTRFS_I(old_dir)),
8927 btrfs_abort_transaction(trans, ret);
8932 /* Update inode version and ctime/mtime. */
8933 inode_inc_iversion(old_dir);
8934 inode_inc_iversion(new_dir);
8935 inode_inc_iversion(old_inode);
8936 inode_inc_iversion(new_inode);
8937 simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
8939 if (old_dentry->d_parent != new_dentry->d_parent) {
8940 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
8941 BTRFS_I(old_inode), true);
8942 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
8943 BTRFS_I(new_inode), true);
8946 /* src is a subvolume */
8947 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8948 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
8949 } else { /* src is an inode */
8950 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
8951 BTRFS_I(old_dentry->d_inode),
8952 old_name, &old_rename_ctx);
8954 ret = btrfs_update_inode(trans, BTRFS_I(old_inode));
8957 btrfs_abort_transaction(trans, ret);
8961 /* dest is a subvolume */
8962 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8963 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
8964 } else { /* dest is an inode */
8965 ret = __btrfs_unlink_inode(trans, BTRFS_I(new_dir),
8966 BTRFS_I(new_dentry->d_inode),
8967 new_name, &new_rename_ctx);
8969 ret = btrfs_update_inode(trans, BTRFS_I(new_inode));
8972 btrfs_abort_transaction(trans, ret);
8976 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
8977 new_name, 0, old_idx);
8979 btrfs_abort_transaction(trans, ret);
8983 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
8984 old_name, 0, new_idx);
8986 btrfs_abort_transaction(trans, ret);
8990 if (old_inode->i_nlink == 1)
8991 BTRFS_I(old_inode)->dir_index = old_idx;
8992 if (new_inode->i_nlink == 1)
8993 BTRFS_I(new_inode)->dir_index = new_idx;
8996 * Now pin the logs of the roots. We do it to ensure that no other task
8997 * can sync the logs while we are in progress with the rename, because
8998 * that could result in an inconsistency in case any of the inodes that
8999 * are part of this rename operation were logged before.
9001 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9002 btrfs_pin_log_trans(root);
9003 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9004 btrfs_pin_log_trans(dest);
9006 /* Do the log updates for all inodes. */
9007 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9008 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9009 old_rename_ctx.index, new_dentry->d_parent);
9010 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9011 btrfs_log_new_name(trans, new_dentry, BTRFS_I(new_dir),
9012 new_rename_ctx.index, old_dentry->d_parent);
9014 /* Now unpin the logs. */
9015 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9016 btrfs_end_log_trans(root);
9017 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9018 btrfs_end_log_trans(dest);
9020 ret2 = btrfs_end_transaction(trans);
9021 ret = ret ? ret : ret2;
9023 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
9024 old_ino == BTRFS_FIRST_FREE_OBJECTID)
9025 up_read(&fs_info->subvol_sem);
9027 fscrypt_free_filename(&new_fname);
9028 fscrypt_free_filename(&old_fname);
9032 static struct inode *new_whiteout_inode(struct mnt_idmap *idmap,
9035 struct inode *inode;
9037 inode = new_inode(dir->i_sb);
9039 inode_init_owner(idmap, inode, dir,
9040 S_IFCHR | WHITEOUT_MODE);
9041 inode->i_op = &btrfs_special_inode_operations;
9042 init_special_inode(inode, inode->i_mode, WHITEOUT_DEV);
9047 static int btrfs_rename(struct mnt_idmap *idmap,
9048 struct inode *old_dir, struct dentry *old_dentry,
9049 struct inode *new_dir, struct dentry *new_dentry,
9052 struct btrfs_fs_info *fs_info = inode_to_fs_info(old_dir);
9053 struct btrfs_new_inode_args whiteout_args = {
9055 .dentry = old_dentry,
9057 struct btrfs_trans_handle *trans;
9058 unsigned int trans_num_items;
9059 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9060 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9061 struct inode *new_inode = d_inode(new_dentry);
9062 struct inode *old_inode = d_inode(old_dentry);
9063 struct btrfs_rename_ctx rename_ctx;
9067 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9068 struct fscrypt_name old_fname, new_fname;
9070 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9073 /* we only allow rename subvolume link between subvolumes */
9074 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9077 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9078 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9081 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9082 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9085 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
9089 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
9091 fscrypt_free_filename(&old_fname);
9095 /* check for collisions, even if the name isn't there */
9096 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, &new_fname.disk_name);
9098 if (ret == -EEXIST) {
9100 * eexist without a new_inode */
9101 if (WARN_ON(!new_inode)) {
9102 goto out_fscrypt_names;
9105 /* maybe -EOVERFLOW */
9106 goto out_fscrypt_names;
9112 * we're using rename to replace one file with another. Start IO on it
9113 * now so we don't add too much work to the end of the transaction
9115 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9116 filemap_flush(old_inode->i_mapping);
9118 if (flags & RENAME_WHITEOUT) {
9119 whiteout_args.inode = new_whiteout_inode(idmap, old_dir);
9120 if (!whiteout_args.inode) {
9122 goto out_fscrypt_names;
9124 ret = btrfs_new_inode_prepare(&whiteout_args, &trans_num_items);
9126 goto out_whiteout_inode;
9128 /* 1 to update the old parent inode. */
9129 trans_num_items = 1;
9132 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9133 /* Close the race window with snapshot create/destroy ioctl */
9134 down_read(&fs_info->subvol_sem);
9136 * 1 to remove old root ref
9137 * 1 to remove old root backref
9138 * 1 to add new root ref
9139 * 1 to add new root backref
9141 trans_num_items += 4;
9145 * 1 to remove old inode ref
9146 * 1 to add new inode ref
9148 trans_num_items += 3;
9151 * 1 to remove old dir item
9152 * 1 to remove old dir index
9153 * 1 to add new dir item
9154 * 1 to add new dir index
9156 trans_num_items += 4;
9157 /* 1 to update new parent inode if it's not the same as the old parent */
9158 if (new_dir != old_dir)
9163 * 1 to remove inode ref
9164 * 1 to remove dir item
9165 * 1 to remove dir index
9166 * 1 to possibly add orphan item
9168 trans_num_items += 5;
9170 trans = btrfs_start_transaction(root, trans_num_items);
9171 if (IS_ERR(trans)) {
9172 ret = PTR_ERR(trans);
9177 ret = btrfs_record_root_in_trans(trans, dest);
9182 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9186 BTRFS_I(old_inode)->dir_index = 0ULL;
9187 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9188 /* force full log commit if subvolume involved. */
9189 btrfs_set_log_full_commit(trans);
9191 ret = btrfs_insert_inode_ref(trans, dest, &new_fname.disk_name,
9192 old_ino, btrfs_ino(BTRFS_I(new_dir)),
9198 inode_inc_iversion(old_dir);
9199 inode_inc_iversion(new_dir);
9200 inode_inc_iversion(old_inode);
9201 simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
9203 if (old_dentry->d_parent != new_dentry->d_parent)
9204 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9205 BTRFS_I(old_inode), true);
9207 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9208 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
9210 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9211 BTRFS_I(d_inode(old_dentry)),
9212 &old_fname.disk_name, &rename_ctx);
9214 ret = btrfs_update_inode(trans, BTRFS_I(old_inode));
9217 btrfs_abort_transaction(trans, ret);
9222 inode_inc_iversion(new_inode);
9223 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9224 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9225 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
9226 BUG_ON(new_inode->i_nlink == 0);
9228 ret = btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9229 BTRFS_I(d_inode(new_dentry)),
9230 &new_fname.disk_name);
9232 if (!ret && new_inode->i_nlink == 0)
9233 ret = btrfs_orphan_add(trans,
9234 BTRFS_I(d_inode(new_dentry)));
9236 btrfs_abort_transaction(trans, ret);
9241 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9242 &new_fname.disk_name, 0, index);
9244 btrfs_abort_transaction(trans, ret);
9248 if (old_inode->i_nlink == 1)
9249 BTRFS_I(old_inode)->dir_index = index;
9251 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9252 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9253 rename_ctx.index, new_dentry->d_parent);
9255 if (flags & RENAME_WHITEOUT) {
9256 ret = btrfs_create_new_inode(trans, &whiteout_args);
9258 btrfs_abort_transaction(trans, ret);
9261 unlock_new_inode(whiteout_args.inode);
9262 iput(whiteout_args.inode);
9263 whiteout_args.inode = NULL;
9267 ret2 = btrfs_end_transaction(trans);
9268 ret = ret ? ret : ret2;
9270 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9271 up_read(&fs_info->subvol_sem);
9272 if (flags & RENAME_WHITEOUT)
9273 btrfs_new_inode_args_destroy(&whiteout_args);
9275 if (flags & RENAME_WHITEOUT)
9276 iput(whiteout_args.inode);
9278 fscrypt_free_filename(&old_fname);
9279 fscrypt_free_filename(&new_fname);
9283 static int btrfs_rename2(struct mnt_idmap *idmap, struct inode *old_dir,
9284 struct dentry *old_dentry, struct inode *new_dir,
9285 struct dentry *new_dentry, unsigned int flags)
9289 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9292 if (flags & RENAME_EXCHANGE)
9293 ret = btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9296 ret = btrfs_rename(idmap, old_dir, old_dentry, new_dir,
9299 btrfs_btree_balance_dirty(BTRFS_I(new_dir)->root->fs_info);
9304 struct btrfs_delalloc_work {
9305 struct inode *inode;
9306 struct completion completion;
9307 struct list_head list;
9308 struct btrfs_work work;
9311 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9313 struct btrfs_delalloc_work *delalloc_work;
9314 struct inode *inode;
9316 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9318 inode = delalloc_work->inode;
9319 filemap_flush(inode->i_mapping);
9320 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9321 &BTRFS_I(inode)->runtime_flags))
9322 filemap_flush(inode->i_mapping);
9325 complete(&delalloc_work->completion);
9328 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9330 struct btrfs_delalloc_work *work;
9332 work = kmalloc(sizeof(*work), GFP_NOFS);
9336 init_completion(&work->completion);
9337 INIT_LIST_HEAD(&work->list);
9338 work->inode = inode;
9339 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL);
9345 * some fairly slow code that needs optimization. This walks the list
9346 * of all the inodes with pending delalloc and forces them to disk.
9348 static int start_delalloc_inodes(struct btrfs_root *root,
9349 struct writeback_control *wbc, bool snapshot,
9350 bool in_reclaim_context)
9352 struct btrfs_inode *binode;
9353 struct inode *inode;
9354 struct btrfs_delalloc_work *work, *next;
9358 bool full_flush = wbc->nr_to_write == LONG_MAX;
9360 mutex_lock(&root->delalloc_mutex);
9361 spin_lock(&root->delalloc_lock);
9362 list_splice_init(&root->delalloc_inodes, &splice);
9363 while (!list_empty(&splice)) {
9364 binode = list_entry(splice.next, struct btrfs_inode,
9367 list_move_tail(&binode->delalloc_inodes,
9368 &root->delalloc_inodes);
9370 if (in_reclaim_context &&
9371 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9374 inode = igrab(&binode->vfs_inode);
9376 cond_resched_lock(&root->delalloc_lock);
9379 spin_unlock(&root->delalloc_lock);
9382 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9383 &binode->runtime_flags);
9385 work = btrfs_alloc_delalloc_work(inode);
9391 list_add_tail(&work->list, &works);
9392 btrfs_queue_work(root->fs_info->flush_workers,
9395 ret = filemap_fdatawrite_wbc(inode->i_mapping, wbc);
9396 btrfs_add_delayed_iput(BTRFS_I(inode));
9397 if (ret || wbc->nr_to_write <= 0)
9401 spin_lock(&root->delalloc_lock);
9403 spin_unlock(&root->delalloc_lock);
9406 list_for_each_entry_safe(work, next, &works, list) {
9407 list_del_init(&work->list);
9408 wait_for_completion(&work->completion);
9412 if (!list_empty(&splice)) {
9413 spin_lock(&root->delalloc_lock);
9414 list_splice_tail(&splice, &root->delalloc_inodes);
9415 spin_unlock(&root->delalloc_lock);
9417 mutex_unlock(&root->delalloc_mutex);
9421 int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9423 struct writeback_control wbc = {
9424 .nr_to_write = LONG_MAX,
9425 .sync_mode = WB_SYNC_NONE,
9427 .range_end = LLONG_MAX,
9429 struct btrfs_fs_info *fs_info = root->fs_info;
9431 if (BTRFS_FS_ERROR(fs_info))
9434 return start_delalloc_inodes(root, &wbc, true, in_reclaim_context);
9437 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9438 bool in_reclaim_context)
9440 struct writeback_control wbc = {
9442 .sync_mode = WB_SYNC_NONE,
9444 .range_end = LLONG_MAX,
9446 struct btrfs_root *root;
9450 if (BTRFS_FS_ERROR(fs_info))
9453 mutex_lock(&fs_info->delalloc_root_mutex);
9454 spin_lock(&fs_info->delalloc_root_lock);
9455 list_splice_init(&fs_info->delalloc_roots, &splice);
9456 while (!list_empty(&splice)) {
9458 * Reset nr_to_write here so we know that we're doing a full
9462 wbc.nr_to_write = LONG_MAX;
9464 root = list_first_entry(&splice, struct btrfs_root,
9466 root = btrfs_grab_root(root);
9468 list_move_tail(&root->delalloc_root,
9469 &fs_info->delalloc_roots);
9470 spin_unlock(&fs_info->delalloc_root_lock);
9472 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9473 btrfs_put_root(root);
9474 if (ret < 0 || wbc.nr_to_write <= 0)
9476 spin_lock(&fs_info->delalloc_root_lock);
9478 spin_unlock(&fs_info->delalloc_root_lock);
9482 if (!list_empty(&splice)) {
9483 spin_lock(&fs_info->delalloc_root_lock);
9484 list_splice_tail(&splice, &fs_info->delalloc_roots);
9485 spin_unlock(&fs_info->delalloc_root_lock);
9487 mutex_unlock(&fs_info->delalloc_root_mutex);
9491 static int btrfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
9492 struct dentry *dentry, const char *symname)
9494 struct btrfs_fs_info *fs_info = inode_to_fs_info(dir);
9495 struct btrfs_trans_handle *trans;
9496 struct btrfs_root *root = BTRFS_I(dir)->root;
9497 struct btrfs_path *path;
9498 struct btrfs_key key;
9499 struct inode *inode;
9500 struct btrfs_new_inode_args new_inode_args = {
9504 unsigned int trans_num_items;
9509 struct btrfs_file_extent_item *ei;
9510 struct extent_buffer *leaf;
9512 name_len = strlen(symname);
9513 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9514 return -ENAMETOOLONG;
9516 inode = new_inode(dir->i_sb);
9519 inode_init_owner(idmap, inode, dir, S_IFLNK | S_IRWXUGO);
9520 inode->i_op = &btrfs_symlink_inode_operations;
9521 inode_nohighmem(inode);
9522 inode->i_mapping->a_ops = &btrfs_aops;
9523 btrfs_i_size_write(BTRFS_I(inode), name_len);
9524 inode_set_bytes(inode, name_len);
9526 new_inode_args.inode = inode;
9527 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9530 /* 1 additional item for the inline extent */
9533 trans = btrfs_start_transaction(root, trans_num_items);
9534 if (IS_ERR(trans)) {
9535 err = PTR_ERR(trans);
9536 goto out_new_inode_args;
9539 err = btrfs_create_new_inode(trans, &new_inode_args);
9543 path = btrfs_alloc_path();
9546 btrfs_abort_transaction(trans, err);
9547 discard_new_inode(inode);
9551 key.objectid = btrfs_ino(BTRFS_I(inode));
9553 key.type = BTRFS_EXTENT_DATA_KEY;
9554 datasize = btrfs_file_extent_calc_inline_size(name_len);
9555 err = btrfs_insert_empty_item(trans, root, path, &key,
9558 btrfs_abort_transaction(trans, err);
9559 btrfs_free_path(path);
9560 discard_new_inode(inode);
9564 leaf = path->nodes[0];
9565 ei = btrfs_item_ptr(leaf, path->slots[0],
9566 struct btrfs_file_extent_item);
9567 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9568 btrfs_set_file_extent_type(leaf, ei,
9569 BTRFS_FILE_EXTENT_INLINE);
9570 btrfs_set_file_extent_encryption(leaf, ei, 0);
9571 btrfs_set_file_extent_compression(leaf, ei, 0);
9572 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9573 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9575 ptr = btrfs_file_extent_inline_start(ei);
9576 write_extent_buffer(leaf, symname, ptr, name_len);
9577 btrfs_mark_buffer_dirty(trans, leaf);
9578 btrfs_free_path(path);
9580 d_instantiate_new(dentry, inode);
9583 btrfs_end_transaction(trans);
9584 btrfs_btree_balance_dirty(fs_info);
9586 btrfs_new_inode_args_destroy(&new_inode_args);
9593 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9594 struct btrfs_trans_handle *trans_in,
9595 struct btrfs_inode *inode,
9596 struct btrfs_key *ins,
9599 struct btrfs_file_extent_item stack_fi;
9600 struct btrfs_replace_extent_info extent_info;
9601 struct btrfs_trans_handle *trans = trans_in;
9602 struct btrfs_path *path;
9603 u64 start = ins->objectid;
9604 u64 len = ins->offset;
9605 u64 qgroup_released = 0;
9608 memset(&stack_fi, 0, sizeof(stack_fi));
9610 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9611 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9612 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9613 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9614 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9615 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9616 /* Encryption and other encoding is reserved and all 0 */
9618 ret = btrfs_qgroup_release_data(inode, file_offset, len, &qgroup_released);
9620 return ERR_PTR(ret);
9623 ret = insert_reserved_file_extent(trans, inode,
9624 file_offset, &stack_fi,
9625 true, qgroup_released);
9631 extent_info.disk_offset = start;
9632 extent_info.disk_len = len;
9633 extent_info.data_offset = 0;
9634 extent_info.data_len = len;
9635 extent_info.file_offset = file_offset;
9636 extent_info.extent_buf = (char *)&stack_fi;
9637 extent_info.is_new_extent = true;
9638 extent_info.update_times = true;
9639 extent_info.qgroup_reserved = qgroup_released;
9640 extent_info.insertions = 0;
9642 path = btrfs_alloc_path();
9648 ret = btrfs_replace_file_extents(inode, path, file_offset,
9649 file_offset + len - 1, &extent_info,
9651 btrfs_free_path(path);
9658 * We have released qgroup data range at the beginning of the function,
9659 * and normally qgroup_released bytes will be freed when committing
9661 * But if we error out early, we have to free what we have released
9662 * or we leak qgroup data reservation.
9664 btrfs_qgroup_free_refroot(inode->root->fs_info,
9665 inode->root->root_key.objectid, qgroup_released,
9666 BTRFS_QGROUP_RSV_DATA);
9667 return ERR_PTR(ret);
9670 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9671 u64 start, u64 num_bytes, u64 min_size,
9672 loff_t actual_len, u64 *alloc_hint,
9673 struct btrfs_trans_handle *trans)
9675 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
9676 struct extent_map *em;
9677 struct btrfs_root *root = BTRFS_I(inode)->root;
9678 struct btrfs_key ins;
9679 u64 cur_offset = start;
9680 u64 clear_offset = start;
9683 u64 last_alloc = (u64)-1;
9685 bool own_trans = true;
9686 u64 end = start + num_bytes - 1;
9690 while (num_bytes > 0) {
9691 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9692 cur_bytes = max(cur_bytes, min_size);
9694 * If we are severely fragmented we could end up with really
9695 * small allocations, so if the allocator is returning small
9696 * chunks lets make its job easier by only searching for those
9699 cur_bytes = min(cur_bytes, last_alloc);
9700 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9701 min_size, 0, *alloc_hint, &ins, 1, 0);
9706 * We've reserved this space, and thus converted it from
9707 * ->bytes_may_use to ->bytes_reserved. Any error that happens
9708 * from here on out we will only need to clear our reservation
9709 * for the remaining unreserved area, so advance our
9710 * clear_offset by our extent size.
9712 clear_offset += ins.offset;
9714 last_alloc = ins.offset;
9715 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
9718 * Now that we inserted the prealloc extent we can finally
9719 * decrement the number of reservations in the block group.
9720 * If we did it before, we could race with relocation and have
9721 * relocation miss the reserved extent, making it fail later.
9723 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
9724 if (IS_ERR(trans)) {
9725 ret = PTR_ERR(trans);
9726 btrfs_free_reserved_extent(fs_info, ins.objectid,
9731 em = alloc_extent_map();
9733 btrfs_drop_extent_map_range(BTRFS_I(inode), cur_offset,
9734 cur_offset + ins.offset - 1, false);
9735 btrfs_set_inode_full_sync(BTRFS_I(inode));
9739 em->start = cur_offset;
9740 em->orig_start = cur_offset;
9741 em->len = ins.offset;
9742 em->block_start = ins.objectid;
9743 em->block_len = ins.offset;
9744 em->orig_block_len = ins.offset;
9745 em->ram_bytes = ins.offset;
9746 em->flags |= EXTENT_FLAG_PREALLOC;
9747 em->generation = trans->transid;
9749 ret = btrfs_replace_extent_map_range(BTRFS_I(inode), em, true);
9750 free_extent_map(em);
9752 num_bytes -= ins.offset;
9753 cur_offset += ins.offset;
9754 *alloc_hint = ins.objectid + ins.offset;
9756 inode_inc_iversion(inode);
9757 inode_set_ctime_current(inode);
9758 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9759 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9760 (actual_len > inode->i_size) &&
9761 (cur_offset > inode->i_size)) {
9762 if (cur_offset > actual_len)
9763 i_size = actual_len;
9765 i_size = cur_offset;
9766 i_size_write(inode, i_size);
9767 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
9770 ret = btrfs_update_inode(trans, BTRFS_I(inode));
9773 btrfs_abort_transaction(trans, ret);
9775 btrfs_end_transaction(trans);
9780 btrfs_end_transaction(trans);
9784 if (clear_offset < end)
9785 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
9786 end - clear_offset + 1);
9790 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9791 u64 start, u64 num_bytes, u64 min_size,
9792 loff_t actual_len, u64 *alloc_hint)
9794 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9795 min_size, actual_len, alloc_hint,
9799 int btrfs_prealloc_file_range_trans(struct inode *inode,
9800 struct btrfs_trans_handle *trans, int mode,
9801 u64 start, u64 num_bytes, u64 min_size,
9802 loff_t actual_len, u64 *alloc_hint)
9804 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9805 min_size, actual_len, alloc_hint, trans);
9808 static int btrfs_permission(struct mnt_idmap *idmap,
9809 struct inode *inode, int mask)
9811 struct btrfs_root *root = BTRFS_I(inode)->root;
9812 umode_t mode = inode->i_mode;
9814 if (mask & MAY_WRITE &&
9815 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9816 if (btrfs_root_readonly(root))
9818 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9821 return generic_permission(idmap, inode, mask);
9824 static int btrfs_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
9825 struct file *file, umode_t mode)
9827 struct btrfs_fs_info *fs_info = inode_to_fs_info(dir);
9828 struct btrfs_trans_handle *trans;
9829 struct btrfs_root *root = BTRFS_I(dir)->root;
9830 struct inode *inode;
9831 struct btrfs_new_inode_args new_inode_args = {
9833 .dentry = file->f_path.dentry,
9836 unsigned int trans_num_items;
9839 inode = new_inode(dir->i_sb);
9842 inode_init_owner(idmap, inode, dir, mode);
9843 inode->i_fop = &btrfs_file_operations;
9844 inode->i_op = &btrfs_file_inode_operations;
9845 inode->i_mapping->a_ops = &btrfs_aops;
9847 new_inode_args.inode = inode;
9848 ret = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9852 trans = btrfs_start_transaction(root, trans_num_items);
9853 if (IS_ERR(trans)) {
9854 ret = PTR_ERR(trans);
9855 goto out_new_inode_args;
9858 ret = btrfs_create_new_inode(trans, &new_inode_args);
9861 * We set number of links to 0 in btrfs_create_new_inode(), and here we
9862 * set it to 1 because d_tmpfile() will issue a warning if the count is
9865 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9867 set_nlink(inode, 1);
9870 d_tmpfile(file, inode);
9871 unlock_new_inode(inode);
9872 mark_inode_dirty(inode);
9875 btrfs_end_transaction(trans);
9876 btrfs_btree_balance_dirty(fs_info);
9878 btrfs_new_inode_args_destroy(&new_inode_args);
9882 return finish_open_simple(file, ret);
9885 void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end)
9887 struct btrfs_fs_info *fs_info = inode->root->fs_info;
9888 unsigned long index = start >> PAGE_SHIFT;
9889 unsigned long end_index = end >> PAGE_SHIFT;
9893 ASSERT(end + 1 - start <= U32_MAX);
9894 len = end + 1 - start;
9895 while (index <= end_index) {
9896 page = find_get_page(inode->vfs_inode.i_mapping, index);
9897 ASSERT(page); /* Pages should be in the extent_io_tree */
9899 /* This is for data, which doesn't yet support larger folio. */
9900 ASSERT(folio_order(page_folio(page)) == 0);
9901 btrfs_folio_set_writeback(fs_info, page_folio(page), start, len);
9907 int btrfs_encoded_io_compression_from_extent(struct btrfs_fs_info *fs_info,
9910 switch (compress_type) {
9911 case BTRFS_COMPRESS_NONE:
9912 return BTRFS_ENCODED_IO_COMPRESSION_NONE;
9913 case BTRFS_COMPRESS_ZLIB:
9914 return BTRFS_ENCODED_IO_COMPRESSION_ZLIB;
9915 case BTRFS_COMPRESS_LZO:
9917 * The LZO format depends on the sector size. 64K is the maximum
9918 * sector size that we support.
9920 if (fs_info->sectorsize < SZ_4K || fs_info->sectorsize > SZ_64K)
9922 return BTRFS_ENCODED_IO_COMPRESSION_LZO_4K +
9923 (fs_info->sectorsize_bits - 12);
9924 case BTRFS_COMPRESS_ZSTD:
9925 return BTRFS_ENCODED_IO_COMPRESSION_ZSTD;
9931 static ssize_t btrfs_encoded_read_inline(
9933 struct iov_iter *iter, u64 start,
9935 struct extent_state **cached_state,
9936 u64 extent_start, size_t count,
9937 struct btrfs_ioctl_encoded_io_args *encoded,
9940 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
9941 struct btrfs_root *root = inode->root;
9942 struct btrfs_fs_info *fs_info = root->fs_info;
9943 struct extent_io_tree *io_tree = &inode->io_tree;
9944 struct btrfs_path *path;
9945 struct extent_buffer *leaf;
9946 struct btrfs_file_extent_item *item;
9952 path = btrfs_alloc_path();
9957 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
9961 /* The extent item disappeared? */
9966 leaf = path->nodes[0];
9967 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
9969 ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
9970 ptr = btrfs_file_extent_inline_start(item);
9972 encoded->len = min_t(u64, extent_start + ram_bytes,
9973 inode->vfs_inode.i_size) - iocb->ki_pos;
9974 ret = btrfs_encoded_io_compression_from_extent(fs_info,
9975 btrfs_file_extent_compression(leaf, item));
9978 encoded->compression = ret;
9979 if (encoded->compression) {
9982 inline_size = btrfs_file_extent_inline_item_len(leaf,
9984 if (inline_size > count) {
9988 count = inline_size;
9989 encoded->unencoded_len = ram_bytes;
9990 encoded->unencoded_offset = iocb->ki_pos - extent_start;
9992 count = min_t(u64, count, encoded->len);
9993 encoded->len = count;
9994 encoded->unencoded_len = count;
9995 ptr += iocb->ki_pos - extent_start;
9998 tmp = kmalloc(count, GFP_NOFS);
10003 read_extent_buffer(leaf, tmp, ptr, count);
10004 btrfs_release_path(path);
10005 unlock_extent(io_tree, start, lockend, cached_state);
10006 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10009 ret = copy_to_iter(tmp, count, iter);
10014 btrfs_free_path(path);
10018 struct btrfs_encoded_read_private {
10019 wait_queue_head_t wait;
10021 blk_status_t status;
10024 static void btrfs_encoded_read_endio(struct btrfs_bio *bbio)
10026 struct btrfs_encoded_read_private *priv = bbio->private;
10028 if (bbio->bio.bi_status) {
10030 * The memory barrier implied by the atomic_dec_return() here
10031 * pairs with the memory barrier implied by the
10032 * atomic_dec_return() or io_wait_event() in
10033 * btrfs_encoded_read_regular_fill_pages() to ensure that this
10034 * write is observed before the load of status in
10035 * btrfs_encoded_read_regular_fill_pages().
10037 WRITE_ONCE(priv->status, bbio->bio.bi_status);
10039 if (!atomic_dec_return(&priv->pending))
10040 wake_up(&priv->wait);
10041 bio_put(&bbio->bio);
10044 int btrfs_encoded_read_regular_fill_pages(struct btrfs_inode *inode,
10045 u64 file_offset, u64 disk_bytenr,
10046 u64 disk_io_size, struct page **pages)
10048 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10049 struct btrfs_encoded_read_private priv = {
10050 .pending = ATOMIC_INIT(1),
10052 unsigned long i = 0;
10053 struct btrfs_bio *bbio;
10055 init_waitqueue_head(&priv.wait);
10057 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
10058 btrfs_encoded_read_endio, &priv);
10059 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
10060 bbio->inode = inode;
10063 size_t bytes = min_t(u64, disk_io_size, PAGE_SIZE);
10065 if (bio_add_page(&bbio->bio, pages[i], bytes, 0) < bytes) {
10066 atomic_inc(&priv.pending);
10067 btrfs_submit_bio(bbio, 0);
10069 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
10070 btrfs_encoded_read_endio, &priv);
10071 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
10072 bbio->inode = inode;
10077 disk_bytenr += bytes;
10078 disk_io_size -= bytes;
10079 } while (disk_io_size);
10081 atomic_inc(&priv.pending);
10082 btrfs_submit_bio(bbio, 0);
10084 if (atomic_dec_return(&priv.pending))
10085 io_wait_event(priv.wait, !atomic_read(&priv.pending));
10086 /* See btrfs_encoded_read_endio() for ordering. */
10087 return blk_status_to_errno(READ_ONCE(priv.status));
10090 static ssize_t btrfs_encoded_read_regular(struct kiocb *iocb,
10091 struct iov_iter *iter,
10092 u64 start, u64 lockend,
10093 struct extent_state **cached_state,
10094 u64 disk_bytenr, u64 disk_io_size,
10095 size_t count, bool compressed,
10098 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10099 struct extent_io_tree *io_tree = &inode->io_tree;
10100 struct page **pages;
10101 unsigned long nr_pages, i;
10103 size_t page_offset;
10106 nr_pages = DIV_ROUND_UP(disk_io_size, PAGE_SIZE);
10107 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
10110 ret = btrfs_alloc_page_array(nr_pages, pages, 0);
10116 ret = btrfs_encoded_read_regular_fill_pages(inode, start, disk_bytenr,
10117 disk_io_size, pages);
10121 unlock_extent(io_tree, start, lockend, cached_state);
10122 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10129 i = (iocb->ki_pos - start) >> PAGE_SHIFT;
10130 page_offset = (iocb->ki_pos - start) & (PAGE_SIZE - 1);
10133 while (cur < count) {
10134 size_t bytes = min_t(size_t, count - cur,
10135 PAGE_SIZE - page_offset);
10137 if (copy_page_to_iter(pages[i], page_offset, bytes,
10148 for (i = 0; i < nr_pages; i++) {
10150 __free_page(pages[i]);
10156 ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter,
10157 struct btrfs_ioctl_encoded_io_args *encoded)
10159 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10160 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10161 struct extent_io_tree *io_tree = &inode->io_tree;
10163 size_t count = iov_iter_count(iter);
10164 u64 start, lockend, disk_bytenr, disk_io_size;
10165 struct extent_state *cached_state = NULL;
10166 struct extent_map *em;
10167 bool unlocked = false;
10169 file_accessed(iocb->ki_filp);
10171 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
10173 if (iocb->ki_pos >= inode->vfs_inode.i_size) {
10174 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10177 start = ALIGN_DOWN(iocb->ki_pos, fs_info->sectorsize);
10179 * We don't know how long the extent containing iocb->ki_pos is, but if
10180 * it's compressed we know that it won't be longer than this.
10182 lockend = start + BTRFS_MAX_UNCOMPRESSED - 1;
10185 struct btrfs_ordered_extent *ordered;
10187 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start,
10188 lockend - start + 1);
10190 goto out_unlock_inode;
10191 lock_extent(io_tree, start, lockend, &cached_state);
10192 ordered = btrfs_lookup_ordered_range(inode, start,
10193 lockend - start + 1);
10196 btrfs_put_ordered_extent(ordered);
10197 unlock_extent(io_tree, start, lockend, &cached_state);
10201 em = btrfs_get_extent(inode, NULL, start, lockend - start + 1);
10204 goto out_unlock_extent;
10207 if (em->block_start == EXTENT_MAP_INLINE) {
10208 u64 extent_start = em->start;
10211 * For inline extents we get everything we need out of the
10214 free_extent_map(em);
10216 ret = btrfs_encoded_read_inline(iocb, iter, start, lockend,
10217 &cached_state, extent_start,
10218 count, encoded, &unlocked);
10223 * We only want to return up to EOF even if the extent extends beyond
10226 encoded->len = min_t(u64, extent_map_end(em),
10227 inode->vfs_inode.i_size) - iocb->ki_pos;
10228 if (em->block_start == EXTENT_MAP_HOLE ||
10229 (em->flags & EXTENT_FLAG_PREALLOC)) {
10230 disk_bytenr = EXTENT_MAP_HOLE;
10231 count = min_t(u64, count, encoded->len);
10232 encoded->len = count;
10233 encoded->unencoded_len = count;
10234 } else if (extent_map_is_compressed(em)) {
10235 disk_bytenr = em->block_start;
10237 * Bail if the buffer isn't large enough to return the whole
10238 * compressed extent.
10240 if (em->block_len > count) {
10244 disk_io_size = em->block_len;
10245 count = em->block_len;
10246 encoded->unencoded_len = em->ram_bytes;
10247 encoded->unencoded_offset = iocb->ki_pos - em->orig_start;
10248 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10249 extent_map_compression(em));
10252 encoded->compression = ret;
10254 disk_bytenr = em->block_start + (start - em->start);
10255 if (encoded->len > count)
10256 encoded->len = count;
10258 * Don't read beyond what we locked. This also limits the page
10259 * allocations that we'll do.
10261 disk_io_size = min(lockend + 1, iocb->ki_pos + encoded->len) - start;
10262 count = start + disk_io_size - iocb->ki_pos;
10263 encoded->len = count;
10264 encoded->unencoded_len = count;
10265 disk_io_size = ALIGN(disk_io_size, fs_info->sectorsize);
10267 free_extent_map(em);
10270 if (disk_bytenr == EXTENT_MAP_HOLE) {
10271 unlock_extent(io_tree, start, lockend, &cached_state);
10272 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10274 ret = iov_iter_zero(count, iter);
10278 ret = btrfs_encoded_read_regular(iocb, iter, start, lockend,
10279 &cached_state, disk_bytenr,
10280 disk_io_size, count,
10281 encoded->compression,
10287 iocb->ki_pos += encoded->len;
10289 free_extent_map(em);
10292 unlock_extent(io_tree, start, lockend, &cached_state);
10295 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10299 ssize_t btrfs_do_encoded_write(struct kiocb *iocb, struct iov_iter *from,
10300 const struct btrfs_ioctl_encoded_io_args *encoded)
10302 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10303 struct btrfs_root *root = inode->root;
10304 struct btrfs_fs_info *fs_info = root->fs_info;
10305 struct extent_io_tree *io_tree = &inode->io_tree;
10306 struct extent_changeset *data_reserved = NULL;
10307 struct extent_state *cached_state = NULL;
10308 struct btrfs_ordered_extent *ordered;
10312 u64 num_bytes, ram_bytes, disk_num_bytes;
10313 unsigned long nr_pages, i;
10314 struct page **pages;
10315 struct btrfs_key ins;
10316 bool extent_reserved = false;
10317 struct extent_map *em;
10320 switch (encoded->compression) {
10321 case BTRFS_ENCODED_IO_COMPRESSION_ZLIB:
10322 compression = BTRFS_COMPRESS_ZLIB;
10324 case BTRFS_ENCODED_IO_COMPRESSION_ZSTD:
10325 compression = BTRFS_COMPRESS_ZSTD;
10327 case BTRFS_ENCODED_IO_COMPRESSION_LZO_4K:
10328 case BTRFS_ENCODED_IO_COMPRESSION_LZO_8K:
10329 case BTRFS_ENCODED_IO_COMPRESSION_LZO_16K:
10330 case BTRFS_ENCODED_IO_COMPRESSION_LZO_32K:
10331 case BTRFS_ENCODED_IO_COMPRESSION_LZO_64K:
10332 /* The sector size must match for LZO. */
10333 if (encoded->compression -
10334 BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + 12 !=
10335 fs_info->sectorsize_bits)
10337 compression = BTRFS_COMPRESS_LZO;
10342 if (encoded->encryption != BTRFS_ENCODED_IO_ENCRYPTION_NONE)
10346 * Compressed extents should always have checksums, so error out if we
10347 * have a NOCOW file or inode was created while mounted with NODATASUM.
10349 if (inode->flags & BTRFS_INODE_NODATASUM)
10352 orig_count = iov_iter_count(from);
10354 /* The extent size must be sane. */
10355 if (encoded->unencoded_len > BTRFS_MAX_UNCOMPRESSED ||
10356 orig_count > BTRFS_MAX_COMPRESSED || orig_count == 0)
10360 * The compressed data must be smaller than the decompressed data.
10362 * It's of course possible for data to compress to larger or the same
10363 * size, but the buffered I/O path falls back to no compression for such
10364 * data, and we don't want to break any assumptions by creating these
10367 * Note that this is less strict than the current check we have that the
10368 * compressed data must be at least one sector smaller than the
10369 * decompressed data. We only want to enforce the weaker requirement
10370 * from old kernels that it is at least one byte smaller.
10372 if (orig_count >= encoded->unencoded_len)
10375 /* The extent must start on a sector boundary. */
10376 start = iocb->ki_pos;
10377 if (!IS_ALIGNED(start, fs_info->sectorsize))
10381 * The extent must end on a sector boundary. However, we allow a write
10382 * which ends at or extends i_size to have an unaligned length; we round
10383 * up the extent size and set i_size to the unaligned end.
10385 if (start + encoded->len < inode->vfs_inode.i_size &&
10386 !IS_ALIGNED(start + encoded->len, fs_info->sectorsize))
10389 /* Finally, the offset in the unencoded data must be sector-aligned. */
10390 if (!IS_ALIGNED(encoded->unencoded_offset, fs_info->sectorsize))
10393 num_bytes = ALIGN(encoded->len, fs_info->sectorsize);
10394 ram_bytes = ALIGN(encoded->unencoded_len, fs_info->sectorsize);
10395 end = start + num_bytes - 1;
10398 * If the extent cannot be inline, the compressed data on disk must be
10399 * sector-aligned. For convenience, we extend it with zeroes if it
10402 disk_num_bytes = ALIGN(orig_count, fs_info->sectorsize);
10403 nr_pages = DIV_ROUND_UP(disk_num_bytes, PAGE_SIZE);
10404 pages = kvcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL_ACCOUNT);
10407 for (i = 0; i < nr_pages; i++) {
10408 size_t bytes = min_t(size_t, PAGE_SIZE, iov_iter_count(from));
10411 pages[i] = alloc_page(GFP_KERNEL_ACCOUNT);
10416 kaddr = kmap_local_page(pages[i]);
10417 if (copy_from_iter(kaddr, bytes, from) != bytes) {
10418 kunmap_local(kaddr);
10422 if (bytes < PAGE_SIZE)
10423 memset(kaddr + bytes, 0, PAGE_SIZE - bytes);
10424 kunmap_local(kaddr);
10428 struct btrfs_ordered_extent *ordered;
10430 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start, num_bytes);
10433 ret = invalidate_inode_pages2_range(inode->vfs_inode.i_mapping,
10434 start >> PAGE_SHIFT,
10435 end >> PAGE_SHIFT);
10438 lock_extent(io_tree, start, end, &cached_state);
10439 ordered = btrfs_lookup_ordered_range(inode, start, num_bytes);
10441 !filemap_range_has_page(inode->vfs_inode.i_mapping, start, end))
10444 btrfs_put_ordered_extent(ordered);
10445 unlock_extent(io_tree, start, end, &cached_state);
10450 * We don't use the higher-level delalloc space functions because our
10451 * num_bytes and disk_num_bytes are different.
10453 ret = btrfs_alloc_data_chunk_ondemand(inode, disk_num_bytes);
10456 ret = btrfs_qgroup_reserve_data(inode, &data_reserved, start, num_bytes);
10458 goto out_free_data_space;
10459 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes, disk_num_bytes,
10462 goto out_qgroup_free_data;
10464 /* Try an inline extent first. */
10465 if (start == 0 && encoded->unencoded_len == encoded->len &&
10466 encoded->unencoded_offset == 0) {
10467 ret = cow_file_range_inline(inode, encoded->len, orig_count,
10468 compression, pages, true);
10472 goto out_delalloc_release;
10476 ret = btrfs_reserve_extent(root, disk_num_bytes, disk_num_bytes,
10477 disk_num_bytes, 0, 0, &ins, 1, 1);
10479 goto out_delalloc_release;
10480 extent_reserved = true;
10482 em = create_io_em(inode, start, num_bytes,
10483 start - encoded->unencoded_offset, ins.objectid,
10484 ins.offset, ins.offset, ram_bytes, compression,
10485 BTRFS_ORDERED_COMPRESSED);
10488 goto out_free_reserved;
10490 free_extent_map(em);
10492 ordered = btrfs_alloc_ordered_extent(inode, start, num_bytes, ram_bytes,
10493 ins.objectid, ins.offset,
10494 encoded->unencoded_offset,
10495 (1 << BTRFS_ORDERED_ENCODED) |
10496 (1 << BTRFS_ORDERED_COMPRESSED),
10498 if (IS_ERR(ordered)) {
10499 btrfs_drop_extent_map_range(inode, start, end, false);
10500 ret = PTR_ERR(ordered);
10501 goto out_free_reserved;
10503 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10505 if (start + encoded->len > inode->vfs_inode.i_size)
10506 i_size_write(&inode->vfs_inode, start + encoded->len);
10508 unlock_extent(io_tree, start, end, &cached_state);
10510 btrfs_delalloc_release_extents(inode, num_bytes);
10512 btrfs_submit_compressed_write(ordered, pages, nr_pages, 0, false);
10517 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10518 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
10519 out_delalloc_release:
10520 btrfs_delalloc_release_extents(inode, num_bytes);
10521 btrfs_delalloc_release_metadata(inode, disk_num_bytes, ret < 0);
10522 out_qgroup_free_data:
10524 btrfs_qgroup_free_data(inode, data_reserved, start, num_bytes, NULL);
10525 out_free_data_space:
10527 * If btrfs_reserve_extent() succeeded, then we already decremented
10530 if (!extent_reserved)
10531 btrfs_free_reserved_data_space_noquota(fs_info, disk_num_bytes);
10533 unlock_extent(io_tree, start, end, &cached_state);
10535 for (i = 0; i < nr_pages; i++) {
10537 __free_page(pages[i]);
10542 iocb->ki_pos += encoded->len;
10548 * Add an entry indicating a block group or device which is pinned by a
10549 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10550 * negative errno on failure.
10552 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10553 bool is_block_group)
10555 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10556 struct btrfs_swapfile_pin *sp, *entry;
10557 struct rb_node **p;
10558 struct rb_node *parent = NULL;
10560 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10565 sp->is_block_group = is_block_group;
10566 sp->bg_extent_count = 1;
10568 spin_lock(&fs_info->swapfile_pins_lock);
10569 p = &fs_info->swapfile_pins.rb_node;
10572 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10573 if (sp->ptr < entry->ptr ||
10574 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10575 p = &(*p)->rb_left;
10576 } else if (sp->ptr > entry->ptr ||
10577 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10578 p = &(*p)->rb_right;
10580 if (is_block_group)
10581 entry->bg_extent_count++;
10582 spin_unlock(&fs_info->swapfile_pins_lock);
10587 rb_link_node(&sp->node, parent, p);
10588 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10589 spin_unlock(&fs_info->swapfile_pins_lock);
10593 /* Free all of the entries pinned by this swapfile. */
10594 static void btrfs_free_swapfile_pins(struct inode *inode)
10596 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10597 struct btrfs_swapfile_pin *sp;
10598 struct rb_node *node, *next;
10600 spin_lock(&fs_info->swapfile_pins_lock);
10601 node = rb_first(&fs_info->swapfile_pins);
10603 next = rb_next(node);
10604 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10605 if (sp->inode == inode) {
10606 rb_erase(&sp->node, &fs_info->swapfile_pins);
10607 if (sp->is_block_group) {
10608 btrfs_dec_block_group_swap_extents(sp->ptr,
10609 sp->bg_extent_count);
10610 btrfs_put_block_group(sp->ptr);
10616 spin_unlock(&fs_info->swapfile_pins_lock);
10619 struct btrfs_swap_info {
10625 unsigned long nr_pages;
10629 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10630 struct btrfs_swap_info *bsi)
10632 unsigned long nr_pages;
10633 unsigned long max_pages;
10634 u64 first_ppage, first_ppage_reported, next_ppage;
10638 * Our swapfile may have had its size extended after the swap header was
10639 * written. In that case activating the swapfile should not go beyond
10640 * the max size set in the swap header.
10642 if (bsi->nr_pages >= sis->max)
10645 max_pages = sis->max - bsi->nr_pages;
10646 first_ppage = PAGE_ALIGN(bsi->block_start) >> PAGE_SHIFT;
10647 next_ppage = PAGE_ALIGN_DOWN(bsi->block_start + bsi->block_len) >> PAGE_SHIFT;
10649 if (first_ppage >= next_ppage)
10651 nr_pages = next_ppage - first_ppage;
10652 nr_pages = min(nr_pages, max_pages);
10654 first_ppage_reported = first_ppage;
10655 if (bsi->start == 0)
10656 first_ppage_reported++;
10657 if (bsi->lowest_ppage > first_ppage_reported)
10658 bsi->lowest_ppage = first_ppage_reported;
10659 if (bsi->highest_ppage < (next_ppage - 1))
10660 bsi->highest_ppage = next_ppage - 1;
10662 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10665 bsi->nr_extents += ret;
10666 bsi->nr_pages += nr_pages;
10670 static void btrfs_swap_deactivate(struct file *file)
10672 struct inode *inode = file_inode(file);
10674 btrfs_free_swapfile_pins(inode);
10675 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10678 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10681 struct inode *inode = file_inode(file);
10682 struct btrfs_root *root = BTRFS_I(inode)->root;
10683 struct btrfs_fs_info *fs_info = root->fs_info;
10684 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10685 struct extent_state *cached_state = NULL;
10686 struct extent_map *em = NULL;
10687 struct btrfs_chunk_map *map = NULL;
10688 struct btrfs_device *device = NULL;
10689 struct btrfs_swap_info bsi = {
10690 .lowest_ppage = (sector_t)-1ULL,
10697 * If the swap file was just created, make sure delalloc is done. If the
10698 * file changes again after this, the user is doing something stupid and
10699 * we don't really care.
10701 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10706 * The inode is locked, so these flags won't change after we check them.
10708 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10709 btrfs_warn(fs_info, "swapfile must not be compressed");
10712 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10713 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10716 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10717 btrfs_warn(fs_info, "swapfile must not be checksummed");
10722 * Balance or device remove/replace/resize can move stuff around from
10723 * under us. The exclop protection makes sure they aren't running/won't
10724 * run concurrently while we are mapping the swap extents, and
10725 * fs_info->swapfile_pins prevents them from running while the swap
10726 * file is active and moving the extents. Note that this also prevents
10727 * a concurrent device add which isn't actually necessary, but it's not
10728 * really worth the trouble to allow it.
10730 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10731 btrfs_warn(fs_info,
10732 "cannot activate swapfile while exclusive operation is running");
10737 * Prevent snapshot creation while we are activating the swap file.
10738 * We do not want to race with snapshot creation. If snapshot creation
10739 * already started before we bumped nr_swapfiles from 0 to 1 and
10740 * completes before the first write into the swap file after it is
10741 * activated, than that write would fallback to COW.
10743 if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) {
10744 btrfs_exclop_finish(fs_info);
10745 btrfs_warn(fs_info,
10746 "cannot activate swapfile because snapshot creation is in progress");
10750 * Snapshots can create extents which require COW even if NODATACOW is
10751 * set. We use this counter to prevent snapshots. We must increment it
10752 * before walking the extents because we don't want a concurrent
10753 * snapshot to run after we've already checked the extents.
10755 * It is possible that subvolume is marked for deletion but still not
10756 * removed yet. To prevent this race, we check the root status before
10757 * activating the swapfile.
10759 spin_lock(&root->root_item_lock);
10760 if (btrfs_root_dead(root)) {
10761 spin_unlock(&root->root_item_lock);
10763 btrfs_exclop_finish(fs_info);
10764 btrfs_warn(fs_info,
10765 "cannot activate swapfile because subvolume %llu is being deleted",
10766 root->root_key.objectid);
10769 atomic_inc(&root->nr_swapfiles);
10770 spin_unlock(&root->root_item_lock);
10772 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10774 lock_extent(io_tree, 0, isize - 1, &cached_state);
10776 while (start < isize) {
10777 u64 logical_block_start, physical_block_start;
10778 struct btrfs_block_group *bg;
10779 u64 len = isize - start;
10781 em = btrfs_get_extent(BTRFS_I(inode), NULL, start, len);
10787 if (em->block_start == EXTENT_MAP_HOLE) {
10788 btrfs_warn(fs_info, "swapfile must not have holes");
10792 if (em->block_start == EXTENT_MAP_INLINE) {
10794 * It's unlikely we'll ever actually find ourselves
10795 * here, as a file small enough to fit inline won't be
10796 * big enough to store more than the swap header, but in
10797 * case something changes in the future, let's catch it
10798 * here rather than later.
10800 btrfs_warn(fs_info, "swapfile must not be inline");
10804 if (extent_map_is_compressed(em)) {
10805 btrfs_warn(fs_info, "swapfile must not be compressed");
10810 logical_block_start = em->block_start + (start - em->start);
10811 len = min(len, em->len - (start - em->start));
10812 free_extent_map(em);
10815 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, false, true);
10821 btrfs_warn(fs_info,
10822 "swapfile must not be copy-on-write");
10827 map = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10829 ret = PTR_ERR(map);
10833 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10834 btrfs_warn(fs_info,
10835 "swapfile must have single data profile");
10840 if (device == NULL) {
10841 device = map->stripes[0].dev;
10842 ret = btrfs_add_swapfile_pin(inode, device, false);
10847 } else if (device != map->stripes[0].dev) {
10848 btrfs_warn(fs_info, "swapfile must be on one device");
10853 physical_block_start = (map->stripes[0].physical +
10854 (logical_block_start - map->start));
10855 len = min(len, map->chunk_len - (logical_block_start - map->start));
10856 btrfs_free_chunk_map(map);
10859 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10861 btrfs_warn(fs_info,
10862 "could not find block group containing swapfile");
10867 if (!btrfs_inc_block_group_swap_extents(bg)) {
10868 btrfs_warn(fs_info,
10869 "block group for swapfile at %llu is read-only%s",
10871 atomic_read(&fs_info->scrubs_running) ?
10872 " (scrub running)" : "");
10873 btrfs_put_block_group(bg);
10878 ret = btrfs_add_swapfile_pin(inode, bg, true);
10880 btrfs_put_block_group(bg);
10887 if (bsi.block_len &&
10888 bsi.block_start + bsi.block_len == physical_block_start) {
10889 bsi.block_len += len;
10891 if (bsi.block_len) {
10892 ret = btrfs_add_swap_extent(sis, &bsi);
10897 bsi.block_start = physical_block_start;
10898 bsi.block_len = len;
10905 ret = btrfs_add_swap_extent(sis, &bsi);
10908 if (!IS_ERR_OR_NULL(em))
10909 free_extent_map(em);
10910 if (!IS_ERR_OR_NULL(map))
10911 btrfs_free_chunk_map(map);
10913 unlock_extent(io_tree, 0, isize - 1, &cached_state);
10916 btrfs_swap_deactivate(file);
10918 btrfs_drew_write_unlock(&root->snapshot_lock);
10920 btrfs_exclop_finish(fs_info);
10926 sis->bdev = device->bdev;
10927 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10928 sis->max = bsi.nr_pages;
10929 sis->pages = bsi.nr_pages - 1;
10930 sis->highest_bit = bsi.nr_pages - 1;
10931 return bsi.nr_extents;
10934 static void btrfs_swap_deactivate(struct file *file)
10938 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10941 return -EOPNOTSUPP;
10946 * Update the number of bytes used in the VFS' inode. When we replace extents in
10947 * a range (clone, dedupe, fallocate's zero range), we must update the number of
10948 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
10949 * always get a correct value.
10951 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
10952 const u64 add_bytes,
10953 const u64 del_bytes)
10955 if (add_bytes == del_bytes)
10958 spin_lock(&inode->lock);
10960 inode_sub_bytes(&inode->vfs_inode, del_bytes);
10962 inode_add_bytes(&inode->vfs_inode, add_bytes);
10963 spin_unlock(&inode->lock);
10967 * Verify that there are no ordered extents for a given file range.
10969 * @inode: The target inode.
10970 * @start: Start offset of the file range, should be sector size aligned.
10971 * @end: End offset (inclusive) of the file range, its value +1 should be
10972 * sector size aligned.
10974 * This should typically be used for cases where we locked an inode's VFS lock in
10975 * exclusive mode, we have also locked the inode's i_mmap_lock in exclusive mode,
10976 * we have flushed all delalloc in the range, we have waited for all ordered
10977 * extents in the range to complete and finally we have locked the file range in
10978 * the inode's io_tree.
10980 void btrfs_assert_inode_range_clean(struct btrfs_inode *inode, u64 start, u64 end)
10982 struct btrfs_root *root = inode->root;
10983 struct btrfs_ordered_extent *ordered;
10985 if (!IS_ENABLED(CONFIG_BTRFS_ASSERT))
10988 ordered = btrfs_lookup_first_ordered_range(inode, start, end + 1 - start);
10990 btrfs_err(root->fs_info,
10991 "found unexpected ordered extent in file range [%llu, %llu] for inode %llu root %llu (ordered range [%llu, %llu])",
10992 start, end, btrfs_ino(inode), root->root_key.objectid,
10993 ordered->file_offset,
10994 ordered->file_offset + ordered->num_bytes - 1);
10995 btrfs_put_ordered_extent(ordered);
10998 ASSERT(ordered == NULL);
11001 static const struct inode_operations btrfs_dir_inode_operations = {
11002 .getattr = btrfs_getattr,
11003 .lookup = btrfs_lookup,
11004 .create = btrfs_create,
11005 .unlink = btrfs_unlink,
11006 .link = btrfs_link,
11007 .mkdir = btrfs_mkdir,
11008 .rmdir = btrfs_rmdir,
11009 .rename = btrfs_rename2,
11010 .symlink = btrfs_symlink,
11011 .setattr = btrfs_setattr,
11012 .mknod = btrfs_mknod,
11013 .listxattr = btrfs_listxattr,
11014 .permission = btrfs_permission,
11015 .get_inode_acl = btrfs_get_acl,
11016 .set_acl = btrfs_set_acl,
11017 .update_time = btrfs_update_time,
11018 .tmpfile = btrfs_tmpfile,
11019 .fileattr_get = btrfs_fileattr_get,
11020 .fileattr_set = btrfs_fileattr_set,
11023 static const struct file_operations btrfs_dir_file_operations = {
11024 .llseek = btrfs_dir_llseek,
11025 .read = generic_read_dir,
11026 .iterate_shared = btrfs_real_readdir,
11027 .open = btrfs_opendir,
11028 .unlocked_ioctl = btrfs_ioctl,
11029 #ifdef CONFIG_COMPAT
11030 .compat_ioctl = btrfs_compat_ioctl,
11032 .release = btrfs_release_file,
11033 .fsync = btrfs_sync_file,
11037 * btrfs doesn't support the bmap operation because swapfiles
11038 * use bmap to make a mapping of extents in the file. They assume
11039 * these extents won't change over the life of the file and they
11040 * use the bmap result to do IO directly to the drive.
11042 * the btrfs bmap call would return logical addresses that aren't
11043 * suitable for IO and they also will change frequently as COW
11044 * operations happen. So, swapfile + btrfs == corruption.
11046 * For now we're avoiding this by dropping bmap.
11048 static const struct address_space_operations btrfs_aops = {
11049 .read_folio = btrfs_read_folio,
11050 .writepages = btrfs_writepages,
11051 .readahead = btrfs_readahead,
11052 .invalidate_folio = btrfs_invalidate_folio,
11053 .release_folio = btrfs_release_folio,
11054 .migrate_folio = btrfs_migrate_folio,
11055 .dirty_folio = filemap_dirty_folio,
11056 .error_remove_folio = generic_error_remove_folio,
11057 .swap_activate = btrfs_swap_activate,
11058 .swap_deactivate = btrfs_swap_deactivate,
11061 static const struct inode_operations btrfs_file_inode_operations = {
11062 .getattr = btrfs_getattr,
11063 .setattr = btrfs_setattr,
11064 .listxattr = btrfs_listxattr,
11065 .permission = btrfs_permission,
11066 .fiemap = btrfs_fiemap,
11067 .get_inode_acl = btrfs_get_acl,
11068 .set_acl = btrfs_set_acl,
11069 .update_time = btrfs_update_time,
11070 .fileattr_get = btrfs_fileattr_get,
11071 .fileattr_set = btrfs_fileattr_set,
11073 static const struct inode_operations btrfs_special_inode_operations = {
11074 .getattr = btrfs_getattr,
11075 .setattr = btrfs_setattr,
11076 .permission = btrfs_permission,
11077 .listxattr = btrfs_listxattr,
11078 .get_inode_acl = btrfs_get_acl,
11079 .set_acl = btrfs_set_acl,
11080 .update_time = btrfs_update_time,
11082 static const struct inode_operations btrfs_symlink_inode_operations = {
11083 .get_link = page_get_link,
11084 .getattr = btrfs_getattr,
11085 .setattr = btrfs_setattr,
11086 .permission = btrfs_permission,
11087 .listxattr = btrfs_listxattr,
11088 .update_time = btrfs_update_time,
11091 const struct dentry_operations btrfs_dentry_operations = {
11092 .d_delete = btrfs_dentry_delete,