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 * We can't reserve contiguous space for the compressed size.
1149 * Unlikely, but it's possible that we could have enough
1150 * non-contiguous space for the uncompressed size instead. So
1151 * fall back to uncompressed.
1153 submit_uncompressed_range(inode, async_extent, locked_page);
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);
1206 mapping_set_error(inode->vfs_inode.i_mapping, -EIO);
1207 extent_clear_unlock_delalloc(inode, start, end,
1208 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
1209 EXTENT_DELALLOC_NEW |
1210 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
1211 PAGE_UNLOCK | PAGE_START_WRITEBACK |
1212 PAGE_END_WRITEBACK);
1213 free_async_extent_pages(async_extent);
1214 if (async_chunk->blkcg_css)
1215 kthread_associate_blkcg(NULL);
1216 btrfs_debug(fs_info,
1217 "async extent submission failed root=%lld inode=%llu start=%llu len=%llu ret=%d",
1218 root->root_key.objectid, btrfs_ino(inode), start,
1219 async_extent->ram_size, ret);
1220 kfree(async_extent);
1223 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
1226 struct extent_map_tree *em_tree = &inode->extent_tree;
1227 struct extent_map *em;
1230 read_lock(&em_tree->lock);
1231 em = search_extent_mapping(em_tree, start, num_bytes);
1234 * if block start isn't an actual block number then find the
1235 * first block in this inode and use that as a hint. If that
1236 * block is also bogus then just don't worry about it.
1238 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1239 free_extent_map(em);
1240 em = search_extent_mapping(em_tree, 0, 0);
1241 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
1242 alloc_hint = em->block_start;
1244 free_extent_map(em);
1246 alloc_hint = em->block_start;
1247 free_extent_map(em);
1250 read_unlock(&em_tree->lock);
1256 * when extent_io.c finds a delayed allocation range in the file,
1257 * the call backs end up in this code. The basic idea is to
1258 * allocate extents on disk for the range, and create ordered data structs
1259 * in ram to track those extents.
1261 * locked_page is the page that writepage had locked already. We use
1262 * it to make sure we don't do extra locks or unlocks.
1264 * When this function fails, it unlocks all pages except @locked_page.
1266 * When this function successfully creates an inline extent, it returns 1 and
1267 * unlocks all pages including locked_page and starts I/O on them.
1268 * (In reality inline extents are limited to a single page, so locked_page is
1269 * the only page handled anyway).
1271 * When this function succeed and creates a normal extent, the page locking
1272 * status depends on the passed in flags:
1274 * - If @keep_locked is set, all pages are kept locked.
1275 * - Else all pages except for @locked_page are unlocked.
1277 * When a failure happens in the second or later iteration of the
1278 * while-loop, the ordered extents created in previous iterations are kept
1279 * intact. So, the caller must clean them up by calling
1280 * btrfs_cleanup_ordered_extents(). See btrfs_run_delalloc_range() for
1283 static noinline int cow_file_range(struct btrfs_inode *inode,
1284 struct page *locked_page, u64 start, u64 end,
1286 bool keep_locked, bool no_inline)
1288 struct btrfs_root *root = inode->root;
1289 struct btrfs_fs_info *fs_info = root->fs_info;
1291 u64 orig_start = start;
1293 unsigned long ram_size;
1294 u64 cur_alloc_size = 0;
1296 u64 blocksize = fs_info->sectorsize;
1297 struct btrfs_key ins;
1298 struct extent_map *em;
1299 unsigned clear_bits;
1300 unsigned long page_ops;
1301 bool extent_reserved = false;
1304 if (btrfs_is_free_space_inode(inode)) {
1309 num_bytes = ALIGN(end - start + 1, blocksize);
1310 num_bytes = max(blocksize, num_bytes);
1311 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1313 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1316 * Due to the page size limit, for subpage we can only trigger the
1317 * writeback for the dirty sectors of page, that means data writeback
1318 * is doing more writeback than what we want.
1320 * This is especially unexpected for some call sites like fallocate,
1321 * where we only increase i_size after everything is done.
1322 * This means we can trigger inline extent even if we didn't want to.
1323 * So here we skip inline extent creation completely.
1325 if (start == 0 && fs_info->sectorsize == PAGE_SIZE && !no_inline) {
1326 u64 actual_end = min_t(u64, i_size_read(&inode->vfs_inode),
1329 /* lets try to make an inline extent */
1330 ret = cow_file_range_inline(inode, actual_end, 0,
1331 BTRFS_COMPRESS_NONE, NULL, false);
1334 * We use DO_ACCOUNTING here because we need the
1335 * delalloc_release_metadata to be run _after_ we drop
1336 * our outstanding extent for clearing delalloc for this
1339 extent_clear_unlock_delalloc(inode, start, end,
1341 EXTENT_LOCKED | EXTENT_DELALLOC |
1342 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1343 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1344 PAGE_START_WRITEBACK | PAGE_END_WRITEBACK);
1346 * locked_page is locked by the caller of
1347 * writepage_delalloc(), not locked by
1348 * __process_pages_contig().
1350 * We can't let __process_pages_contig() to unlock it,
1351 * as it doesn't have any subpage::writers recorded.
1353 * Here we manually unlock the page, since the caller
1354 * can't determine if it's an inline extent or a
1355 * compressed extent.
1357 unlock_page(locked_page);
1360 } else if (ret < 0) {
1365 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1368 * Relocation relies on the relocated extents to have exactly the same
1369 * size as the original extents. Normally writeback for relocation data
1370 * extents follows a NOCOW path because relocation preallocates the
1371 * extents. However, due to an operation such as scrub turning a block
1372 * group to RO mode, it may fallback to COW mode, so we must make sure
1373 * an extent allocated during COW has exactly the requested size and can
1374 * not be split into smaller extents, otherwise relocation breaks and
1375 * fails during the stage where it updates the bytenr of file extent
1378 if (btrfs_is_data_reloc_root(root))
1379 min_alloc_size = num_bytes;
1381 min_alloc_size = fs_info->sectorsize;
1383 while (num_bytes > 0) {
1384 struct btrfs_ordered_extent *ordered;
1386 cur_alloc_size = num_bytes;
1387 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1388 min_alloc_size, 0, alloc_hint,
1390 if (ret == -EAGAIN) {
1392 * btrfs_reserve_extent only returns -EAGAIN for zoned
1393 * file systems, which is an indication that there are
1394 * no active zones to allocate from at the moment.
1396 * If this is the first loop iteration, wait for at
1397 * least one zone to finish before retrying the
1398 * allocation. Otherwise ask the caller to write out
1399 * the already allocated blocks before coming back to
1400 * us, or return -ENOSPC if it can't handle retries.
1402 ASSERT(btrfs_is_zoned(fs_info));
1403 if (start == orig_start) {
1404 wait_on_bit_io(&inode->root->fs_info->flags,
1405 BTRFS_FS_NEED_ZONE_FINISH,
1406 TASK_UNINTERRUPTIBLE);
1410 *done_offset = start - 1;
1417 cur_alloc_size = ins.offset;
1418 extent_reserved = true;
1420 ram_size = ins.offset;
1421 em = create_io_em(inode, start, ins.offset, /* len */
1422 start, /* orig_start */
1423 ins.objectid, /* block_start */
1424 ins.offset, /* block_len */
1425 ins.offset, /* orig_block_len */
1426 ram_size, /* ram_bytes */
1427 BTRFS_COMPRESS_NONE, /* compress_type */
1428 BTRFS_ORDERED_REGULAR /* type */);
1433 free_extent_map(em);
1435 ordered = btrfs_alloc_ordered_extent(inode, start, ram_size,
1436 ram_size, ins.objectid, cur_alloc_size,
1437 0, 1 << BTRFS_ORDERED_REGULAR,
1438 BTRFS_COMPRESS_NONE);
1439 if (IS_ERR(ordered)) {
1440 ret = PTR_ERR(ordered);
1441 goto out_drop_extent_cache;
1444 if (btrfs_is_data_reloc_root(root)) {
1445 ret = btrfs_reloc_clone_csums(ordered);
1448 * Only drop cache here, and process as normal.
1450 * We must not allow extent_clear_unlock_delalloc()
1451 * at out_unlock label to free meta of this ordered
1452 * extent, as its meta should be freed by
1453 * btrfs_finish_ordered_io().
1455 * So we must continue until @start is increased to
1456 * skip current ordered extent.
1459 btrfs_drop_extent_map_range(inode, start,
1460 start + ram_size - 1,
1463 btrfs_put_ordered_extent(ordered);
1465 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1468 * We're not doing compressed IO, don't unlock the first page
1469 * (which the caller expects to stay locked), don't clear any
1470 * dirty bits and don't set any writeback bits
1472 * Do set the Ordered (Private2) bit so we know this page was
1473 * properly setup for writepage.
1475 page_ops = (keep_locked ? 0 : PAGE_UNLOCK);
1476 page_ops |= PAGE_SET_ORDERED;
1478 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1480 EXTENT_LOCKED | EXTENT_DELALLOC,
1482 if (num_bytes < cur_alloc_size)
1485 num_bytes -= cur_alloc_size;
1486 alloc_hint = ins.objectid + ins.offset;
1487 start += cur_alloc_size;
1488 extent_reserved = false;
1491 * btrfs_reloc_clone_csums() error, since start is increased
1492 * extent_clear_unlock_delalloc() at out_unlock label won't
1493 * free metadata of current ordered extent, we're OK to exit.
1503 out_drop_extent_cache:
1504 btrfs_drop_extent_map_range(inode, start, start + ram_size - 1, false);
1506 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1507 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1510 * Now, we have three regions to clean up:
1512 * |-------(1)----|---(2)---|-------------(3)----------|
1513 * `- orig_start `- start `- start + cur_alloc_size `- end
1515 * We process each region below.
1518 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1519 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1520 page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK;
1523 * For the range (1). We have already instantiated the ordered extents
1524 * for this region. They are cleaned up by
1525 * btrfs_cleanup_ordered_extents() in e.g,
1526 * btrfs_run_delalloc_range(). EXTENT_LOCKED | EXTENT_DELALLOC are
1527 * already cleared in the above loop. And, EXTENT_DELALLOC_NEW |
1528 * EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV are handled by the cleanup
1531 * However, in case of @keep_locked, we still need to unlock the pages
1532 * (except @locked_page) to ensure all the pages are unlocked.
1534 if (keep_locked && orig_start < start) {
1536 mapping_set_error(inode->vfs_inode.i_mapping, ret);
1537 extent_clear_unlock_delalloc(inode, orig_start, start - 1,
1538 locked_page, 0, page_ops);
1542 * For the range (2). If we reserved an extent for our delalloc range
1543 * (or a subrange) and failed to create the respective ordered extent,
1544 * then it means that when we reserved the extent we decremented the
1545 * extent's size from the data space_info's bytes_may_use counter and
1546 * incremented the space_info's bytes_reserved counter by the same
1547 * amount. We must make sure extent_clear_unlock_delalloc() does not try
1548 * to decrement again the data space_info's bytes_may_use counter,
1549 * therefore we do not pass it the flag EXTENT_CLEAR_DATA_RESV.
1551 if (extent_reserved) {
1552 extent_clear_unlock_delalloc(inode, start,
1553 start + cur_alloc_size - 1,
1557 start += cur_alloc_size;
1561 * For the range (3). We never touched the region. In addition to the
1562 * clear_bits above, we add EXTENT_CLEAR_DATA_RESV to release the data
1563 * space_info's bytes_may_use counter, reserved in
1564 * btrfs_check_data_free_space().
1567 clear_bits |= EXTENT_CLEAR_DATA_RESV;
1568 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1569 clear_bits, page_ops);
1575 * Phase two of compressed writeback. This is the ordered portion of the code,
1576 * which only gets called in the order the work was queued. We walk all the
1577 * async extents created by compress_file_range and send them down to the disk.
1579 * If called with @do_free == true then it'll try to finish the work and free
1580 * the work struct eventually.
1582 static noinline void submit_compressed_extents(struct btrfs_work *work, bool do_free)
1584 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1586 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1587 struct async_extent *async_extent;
1588 unsigned long nr_pages;
1592 struct async_chunk *async_chunk;
1593 struct async_cow *async_cow;
1595 async_chunk = container_of(work, struct async_chunk, work);
1596 btrfs_add_delayed_iput(async_chunk->inode);
1597 if (async_chunk->blkcg_css)
1598 css_put(async_chunk->blkcg_css);
1600 async_cow = async_chunk->async_cow;
1601 if (atomic_dec_and_test(&async_cow->num_chunks))
1606 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1609 while (!list_empty(&async_chunk->extents)) {
1610 async_extent = list_entry(async_chunk->extents.next,
1611 struct async_extent, list);
1612 list_del(&async_extent->list);
1613 submit_one_async_extent(async_chunk, async_extent, &alloc_hint);
1616 /* atomic_sub_return implies a barrier */
1617 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1619 cond_wake_up_nomb(&fs_info->async_submit_wait);
1622 static bool run_delalloc_compressed(struct btrfs_inode *inode,
1623 struct page *locked_page, u64 start,
1624 u64 end, struct writeback_control *wbc)
1626 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1627 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1628 struct async_cow *ctx;
1629 struct async_chunk *async_chunk;
1630 unsigned long nr_pages;
1631 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1634 const blk_opf_t write_flags = wbc_to_write_flags(wbc);
1636 nofs_flag = memalloc_nofs_save();
1637 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1638 memalloc_nofs_restore(nofs_flag);
1642 unlock_extent(&inode->io_tree, start, end, NULL);
1643 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
1645 async_chunk = ctx->chunks;
1646 atomic_set(&ctx->num_chunks, num_chunks);
1648 for (i = 0; i < num_chunks; i++) {
1649 u64 cur_end = min(end, start + SZ_512K - 1);
1652 * igrab is called higher up in the call chain, take only the
1653 * lightweight reference for the callback lifetime
1655 ihold(&inode->vfs_inode);
1656 async_chunk[i].async_cow = ctx;
1657 async_chunk[i].inode = inode;
1658 async_chunk[i].start = start;
1659 async_chunk[i].end = cur_end;
1660 async_chunk[i].write_flags = write_flags;
1661 INIT_LIST_HEAD(&async_chunk[i].extents);
1664 * The locked_page comes all the way from writepage and its
1665 * the original page we were actually given. As we spread
1666 * this large delalloc region across multiple async_chunk
1667 * structs, only the first struct needs a pointer to locked_page
1669 * This way we don't need racey decisions about who is supposed
1674 * Depending on the compressibility, the pages might or
1675 * might not go through async. We want all of them to
1676 * be accounted against wbc once. Let's do it here
1677 * before the paths diverge. wbc accounting is used
1678 * only for foreign writeback detection and doesn't
1679 * need full accuracy. Just account the whole thing
1680 * against the first page.
1682 wbc_account_cgroup_owner(wbc, locked_page,
1684 async_chunk[i].locked_page = locked_page;
1687 async_chunk[i].locked_page = NULL;
1690 if (blkcg_css != blkcg_root_css) {
1692 async_chunk[i].blkcg_css = blkcg_css;
1693 async_chunk[i].write_flags |= REQ_BTRFS_CGROUP_PUNT;
1695 async_chunk[i].blkcg_css = NULL;
1698 btrfs_init_work(&async_chunk[i].work, compress_file_range,
1699 submit_compressed_extents);
1701 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1702 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1704 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1706 start = cur_end + 1;
1712 * Run the delalloc range from start to end, and write back any dirty pages
1713 * covered by the range.
1715 static noinline int run_delalloc_cow(struct btrfs_inode *inode,
1716 struct page *locked_page, u64 start,
1717 u64 end, struct writeback_control *wbc,
1720 u64 done_offset = end;
1723 while (start <= end) {
1724 ret = cow_file_range(inode, locked_page, start, end, &done_offset,
1728 extent_write_locked_range(&inode->vfs_inode, locked_page, start,
1729 done_offset, wbc, pages_dirty);
1730 start = done_offset + 1;
1736 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1737 u64 bytenr, u64 num_bytes, bool nowait)
1739 struct btrfs_root *csum_root = btrfs_csum_root(fs_info, bytenr);
1740 struct btrfs_ordered_sum *sums;
1744 ret = btrfs_lookup_csums_list(csum_root, bytenr, bytenr + num_bytes - 1,
1746 if (ret == 0 && list_empty(&list))
1749 while (!list_empty(&list)) {
1750 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1751 list_del(&sums->list);
1759 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1760 const u64 start, const u64 end)
1762 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1763 const bool is_reloc_ino = btrfs_is_data_reloc_root(inode->root);
1764 const u64 range_bytes = end + 1 - start;
1765 struct extent_io_tree *io_tree = &inode->io_tree;
1766 u64 range_start = start;
1771 * If EXTENT_NORESERVE is set it means that when the buffered write was
1772 * made we had not enough available data space and therefore we did not
1773 * reserve data space for it, since we though we could do NOCOW for the
1774 * respective file range (either there is prealloc extent or the inode
1775 * has the NOCOW bit set).
1777 * However when we need to fallback to COW mode (because for example the
1778 * block group for the corresponding extent was turned to RO mode by a
1779 * scrub or relocation) we need to do the following:
1781 * 1) We increment the bytes_may_use counter of the data space info.
1782 * If COW succeeds, it allocates a new data extent and after doing
1783 * that it decrements the space info's bytes_may_use counter and
1784 * increments its bytes_reserved counter by the same amount (we do
1785 * this at btrfs_add_reserved_bytes()). So we need to increment the
1786 * bytes_may_use counter to compensate (when space is reserved at
1787 * buffered write time, the bytes_may_use counter is incremented);
1789 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1790 * that if the COW path fails for any reason, it decrements (through
1791 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1792 * data space info, which we incremented in the step above.
1794 * If we need to fallback to cow and the inode corresponds to a free
1795 * space cache inode or an inode of the data relocation tree, we must
1796 * also increment bytes_may_use of the data space_info for the same
1797 * reason. Space caches and relocated data extents always get a prealloc
1798 * extent for them, however scrub or balance may have set the block
1799 * group that contains that extent to RO mode and therefore force COW
1800 * when starting writeback.
1802 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1803 EXTENT_NORESERVE, 0, NULL);
1804 if (count > 0 || is_space_ino || is_reloc_ino) {
1806 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1807 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1809 if (is_space_ino || is_reloc_ino)
1810 bytes = range_bytes;
1812 spin_lock(&sinfo->lock);
1813 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1814 spin_unlock(&sinfo->lock);
1817 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1822 * Don't try to create inline extents, as a mix of inline extent that
1823 * is written out and unlocked directly and a normal NOCOW extent
1826 ret = cow_file_range(inode, locked_page, start, end, NULL, false, true);
1831 struct can_nocow_file_extent_args {
1834 /* Start file offset of the range we want to NOCOW. */
1836 /* End file offset (inclusive) of the range we want to NOCOW. */
1838 bool writeback_path;
1841 * Free the path passed to can_nocow_file_extent() once it's not needed
1846 /* Output fields. Only set when can_nocow_file_extent() returns 1. */
1851 /* Number of bytes that can be written to in NOCOW mode. */
1856 * Check if we can NOCOW the file extent that the path points to.
1857 * This function may return with the path released, so the caller should check
1858 * if path->nodes[0] is NULL or not if it needs to use the path afterwards.
1860 * Returns: < 0 on error
1861 * 0 if we can not NOCOW
1864 static int can_nocow_file_extent(struct btrfs_path *path,
1865 struct btrfs_key *key,
1866 struct btrfs_inode *inode,
1867 struct can_nocow_file_extent_args *args)
1869 const bool is_freespace_inode = btrfs_is_free_space_inode(inode);
1870 struct extent_buffer *leaf = path->nodes[0];
1871 struct btrfs_root *root = inode->root;
1872 struct btrfs_file_extent_item *fi;
1877 bool nowait = path->nowait;
1879 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
1880 extent_type = btrfs_file_extent_type(leaf, fi);
1882 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1885 /* Can't access these fields unless we know it's not an inline extent. */
1886 args->disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1887 args->disk_num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1888 args->extent_offset = btrfs_file_extent_offset(leaf, fi);
1890 if (!(inode->flags & BTRFS_INODE_NODATACOW) &&
1891 extent_type == BTRFS_FILE_EXTENT_REG)
1895 * If the extent was created before the generation where the last snapshot
1896 * for its subvolume was created, then this implies the extent is shared,
1897 * hence we must COW.
1899 if (!args->strict &&
1900 btrfs_file_extent_generation(leaf, fi) <=
1901 btrfs_root_last_snapshot(&root->root_item))
1904 /* An explicit hole, must COW. */
1905 if (args->disk_bytenr == 0)
1908 /* Compressed/encrypted/encoded extents must be COWed. */
1909 if (btrfs_file_extent_compression(leaf, fi) ||
1910 btrfs_file_extent_encryption(leaf, fi) ||
1911 btrfs_file_extent_other_encoding(leaf, fi))
1914 extent_end = btrfs_file_extent_end(path);
1917 * The following checks can be expensive, as they need to take other
1918 * locks and do btree or rbtree searches, so release the path to avoid
1919 * blocking other tasks for too long.
1921 btrfs_release_path(path);
1923 ret = btrfs_cross_ref_exist(root, btrfs_ino(inode),
1924 key->offset - args->extent_offset,
1925 args->disk_bytenr, args->strict, path);
1926 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
1930 if (args->free_path) {
1932 * We don't need the path anymore, plus through the
1933 * csum_exist_in_range() call below we will end up allocating
1934 * another path. So free the path to avoid unnecessary extra
1937 btrfs_free_path(path);
1941 /* If there are pending snapshots for this root, we must COW. */
1942 if (args->writeback_path && !is_freespace_inode &&
1943 atomic_read(&root->snapshot_force_cow))
1946 args->disk_bytenr += args->extent_offset;
1947 args->disk_bytenr += args->start - key->offset;
1948 args->num_bytes = min(args->end + 1, extent_end) - args->start;
1951 * Force COW if csums exist in the range. This ensures that csums for a
1952 * given extent are either valid or do not exist.
1954 ret = csum_exist_in_range(root->fs_info, args->disk_bytenr, args->num_bytes,
1956 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
1962 if (args->free_path && path)
1963 btrfs_free_path(path);
1965 return ret < 0 ? ret : can_nocow;
1969 * when nowcow writeback call back. This checks for snapshots or COW copies
1970 * of the extents that exist in the file, and COWs the file as required.
1972 * If no cow copies or snapshots exist, we write directly to the existing
1975 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
1976 struct page *locked_page,
1977 const u64 start, const u64 end)
1979 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1980 struct btrfs_root *root = inode->root;
1981 struct btrfs_path *path;
1982 u64 cow_start = (u64)-1;
1983 u64 cur_offset = start;
1985 bool check_prev = true;
1986 u64 ino = btrfs_ino(inode);
1987 struct can_nocow_file_extent_args nocow_args = { 0 };
1990 * Normally on a zoned device we're only doing COW writes, but in case
1991 * of relocation on a zoned filesystem serializes I/O so that we're only
1992 * writing sequentially and can end up here as well.
1994 ASSERT(!btrfs_is_zoned(fs_info) || btrfs_is_data_reloc_root(root));
1996 path = btrfs_alloc_path();
2002 nocow_args.end = end;
2003 nocow_args.writeback_path = true;
2006 struct btrfs_block_group *nocow_bg = NULL;
2007 struct btrfs_ordered_extent *ordered;
2008 struct btrfs_key found_key;
2009 struct btrfs_file_extent_item *fi;
2010 struct extent_buffer *leaf;
2017 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
2023 * If there is no extent for our range when doing the initial
2024 * search, then go back to the previous slot as it will be the
2025 * one containing the search offset
2027 if (ret > 0 && path->slots[0] > 0 && check_prev) {
2028 leaf = path->nodes[0];
2029 btrfs_item_key_to_cpu(leaf, &found_key,
2030 path->slots[0] - 1);
2031 if (found_key.objectid == ino &&
2032 found_key.type == BTRFS_EXTENT_DATA_KEY)
2037 /* Go to next leaf if we have exhausted the current one */
2038 leaf = path->nodes[0];
2039 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2040 ret = btrfs_next_leaf(root, path);
2045 leaf = path->nodes[0];
2048 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2050 /* Didn't find anything for our INO */
2051 if (found_key.objectid > ino)
2054 * Keep searching until we find an EXTENT_ITEM or there are no
2055 * more extents for this inode
2057 if (WARN_ON_ONCE(found_key.objectid < ino) ||
2058 found_key.type < BTRFS_EXTENT_DATA_KEY) {
2063 /* Found key is not EXTENT_DATA_KEY or starts after req range */
2064 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
2065 found_key.offset > end)
2069 * If the found extent starts after requested offset, then
2070 * adjust extent_end to be right before this extent begins
2072 if (found_key.offset > cur_offset) {
2073 extent_end = found_key.offset;
2079 * Found extent which begins before our range and potentially
2082 fi = btrfs_item_ptr(leaf, path->slots[0],
2083 struct btrfs_file_extent_item);
2084 extent_type = btrfs_file_extent_type(leaf, fi);
2085 /* If this is triggered then we have a memory corruption. */
2086 ASSERT(extent_type < BTRFS_NR_FILE_EXTENT_TYPES);
2087 if (WARN_ON(extent_type >= BTRFS_NR_FILE_EXTENT_TYPES)) {
2091 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
2092 extent_end = btrfs_file_extent_end(path);
2095 * If the extent we got ends before our current offset, skip to
2098 if (extent_end <= cur_offset) {
2103 nocow_args.start = cur_offset;
2104 ret = can_nocow_file_extent(path, &found_key, inode, &nocow_args);
2111 nocow_bg = btrfs_inc_nocow_writers(fs_info, nocow_args.disk_bytenr);
2115 * If we can't perform NOCOW writeback for the range,
2116 * then record the beginning of the range that needs to
2117 * be COWed. It will be written out before the next
2118 * NOCOW range if we find one, or when exiting this
2121 if (cow_start == (u64)-1)
2122 cow_start = cur_offset;
2123 cur_offset = extent_end;
2124 if (cur_offset > end)
2126 if (!path->nodes[0])
2133 * COW range from cow_start to found_key.offset - 1. As the key
2134 * will contain the beginning of the first extent that can be
2135 * NOCOW, following one which needs to be COW'ed
2137 if (cow_start != (u64)-1) {
2138 ret = fallback_to_cow(inode, locked_page,
2139 cow_start, found_key.offset - 1);
2140 cow_start = (u64)-1;
2142 btrfs_dec_nocow_writers(nocow_bg);
2147 nocow_end = cur_offset + nocow_args.num_bytes - 1;
2148 is_prealloc = extent_type == BTRFS_FILE_EXTENT_PREALLOC;
2150 u64 orig_start = found_key.offset - nocow_args.extent_offset;
2151 struct extent_map *em;
2153 em = create_io_em(inode, cur_offset, nocow_args.num_bytes,
2155 nocow_args.disk_bytenr, /* block_start */
2156 nocow_args.num_bytes, /* block_len */
2157 nocow_args.disk_num_bytes, /* orig_block_len */
2158 ram_bytes, BTRFS_COMPRESS_NONE,
2159 BTRFS_ORDERED_PREALLOC);
2161 btrfs_dec_nocow_writers(nocow_bg);
2165 free_extent_map(em);
2168 ordered = btrfs_alloc_ordered_extent(inode, cur_offset,
2169 nocow_args.num_bytes, nocow_args.num_bytes,
2170 nocow_args.disk_bytenr, nocow_args.num_bytes, 0,
2172 ? (1 << BTRFS_ORDERED_PREALLOC)
2173 : (1 << BTRFS_ORDERED_NOCOW),
2174 BTRFS_COMPRESS_NONE);
2175 btrfs_dec_nocow_writers(nocow_bg);
2176 if (IS_ERR(ordered)) {
2178 btrfs_drop_extent_map_range(inode, cur_offset,
2181 ret = PTR_ERR(ordered);
2185 if (btrfs_is_data_reloc_root(root))
2187 * Error handled later, as we must prevent
2188 * extent_clear_unlock_delalloc() in error handler
2189 * from freeing metadata of created ordered extent.
2191 ret = btrfs_reloc_clone_csums(ordered);
2192 btrfs_put_ordered_extent(ordered);
2194 extent_clear_unlock_delalloc(inode, cur_offset, nocow_end,
2195 locked_page, EXTENT_LOCKED |
2197 EXTENT_CLEAR_DATA_RESV,
2198 PAGE_UNLOCK | PAGE_SET_ORDERED);
2200 cur_offset = extent_end;
2203 * btrfs_reloc_clone_csums() error, now we're OK to call error
2204 * handler, as metadata for created ordered extent will only
2205 * be freed by btrfs_finish_ordered_io().
2209 if (cur_offset > end)
2212 btrfs_release_path(path);
2214 if (cur_offset <= end && cow_start == (u64)-1)
2215 cow_start = cur_offset;
2217 if (cow_start != (u64)-1) {
2219 ret = fallback_to_cow(inode, locked_page, cow_start, end);
2220 cow_start = (u64)-1;
2225 btrfs_free_path(path);
2230 * If an error happened while a COW region is outstanding, cur_offset
2231 * needs to be reset to cow_start to ensure the COW region is unlocked
2234 if (cow_start != (u64)-1)
2235 cur_offset = cow_start;
2236 if (cur_offset < end)
2237 extent_clear_unlock_delalloc(inode, cur_offset, end,
2238 locked_page, EXTENT_LOCKED |
2239 EXTENT_DELALLOC | EXTENT_DEFRAG |
2240 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
2241 PAGE_START_WRITEBACK |
2242 PAGE_END_WRITEBACK);
2243 btrfs_free_path(path);
2247 static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end)
2249 if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) {
2250 if (inode->defrag_bytes &&
2251 test_range_bit_exists(&inode->io_tree, start, end, EXTENT_DEFRAG))
2259 * Function to process delayed allocation (create CoW) for ranges which are
2260 * being touched for the first time.
2262 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
2263 u64 start, u64 end, struct writeback_control *wbc)
2265 const bool zoned = btrfs_is_zoned(inode->root->fs_info);
2269 * The range must cover part of the @locked_page, or a return of 1
2270 * can confuse the caller.
2272 ASSERT(!(end <= page_offset(locked_page) ||
2273 start >= page_offset(locked_page) + PAGE_SIZE));
2275 if (should_nocow(inode, start, end)) {
2276 ret = run_delalloc_nocow(inode, locked_page, start, end);
2280 if (btrfs_inode_can_compress(inode) &&
2281 inode_need_compress(inode, start, end) &&
2282 run_delalloc_compressed(inode, locked_page, start, end, wbc))
2286 ret = run_delalloc_cow(inode, locked_page, start, end, wbc,
2289 ret = cow_file_range(inode, locked_page, start, end, NULL,
2294 btrfs_cleanup_ordered_extents(inode, locked_page, start,
2299 void btrfs_split_delalloc_extent(struct btrfs_inode *inode,
2300 struct extent_state *orig, u64 split)
2302 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2305 lockdep_assert_held(&inode->io_tree.lock);
2307 /* not delalloc, ignore it */
2308 if (!(orig->state & EXTENT_DELALLOC))
2311 size = orig->end - orig->start + 1;
2312 if (size > fs_info->max_extent_size) {
2317 * See the explanation in btrfs_merge_delalloc_extent, the same
2318 * applies here, just in reverse.
2320 new_size = orig->end - split + 1;
2321 num_extents = count_max_extents(fs_info, new_size);
2322 new_size = split - orig->start;
2323 num_extents += count_max_extents(fs_info, new_size);
2324 if (count_max_extents(fs_info, size) >= num_extents)
2328 spin_lock(&inode->lock);
2329 btrfs_mod_outstanding_extents(inode, 1);
2330 spin_unlock(&inode->lock);
2334 * Handle merged delayed allocation extents so we can keep track of new extents
2335 * that are just merged onto old extents, such as when we are doing sequential
2336 * writes, so we can properly account for the metadata space we'll need.
2338 void btrfs_merge_delalloc_extent(struct btrfs_inode *inode, struct extent_state *new,
2339 struct extent_state *other)
2341 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2342 u64 new_size, old_size;
2345 lockdep_assert_held(&inode->io_tree.lock);
2347 /* not delalloc, ignore it */
2348 if (!(other->state & EXTENT_DELALLOC))
2351 if (new->start > other->start)
2352 new_size = new->end - other->start + 1;
2354 new_size = other->end - new->start + 1;
2356 /* we're not bigger than the max, unreserve the space and go */
2357 if (new_size <= fs_info->max_extent_size) {
2358 spin_lock(&inode->lock);
2359 btrfs_mod_outstanding_extents(inode, -1);
2360 spin_unlock(&inode->lock);
2365 * We have to add up either side to figure out how many extents were
2366 * accounted for before we merged into one big extent. If the number of
2367 * extents we accounted for is <= the amount we need for the new range
2368 * then we can return, otherwise drop. Think of it like this
2372 * So we've grown the extent by a MAX_SIZE extent, this would mean we
2373 * need 2 outstanding extents, on one side we have 1 and the other side
2374 * we have 1 so they are == and we can return. But in this case
2376 * [MAX_SIZE+4k][MAX_SIZE+4k]
2378 * Each range on their own accounts for 2 extents, but merged together
2379 * they are only 3 extents worth of accounting, so we need to drop in
2382 old_size = other->end - other->start + 1;
2383 num_extents = count_max_extents(fs_info, old_size);
2384 old_size = new->end - new->start + 1;
2385 num_extents += count_max_extents(fs_info, old_size);
2386 if (count_max_extents(fs_info, new_size) >= num_extents)
2389 spin_lock(&inode->lock);
2390 btrfs_mod_outstanding_extents(inode, -1);
2391 spin_unlock(&inode->lock);
2394 static void btrfs_add_delalloc_inode(struct btrfs_inode *inode)
2396 struct btrfs_root *root = inode->root;
2397 struct btrfs_fs_info *fs_info = root->fs_info;
2399 spin_lock(&root->delalloc_lock);
2400 ASSERT(list_empty(&inode->delalloc_inodes));
2401 list_add_tail(&inode->delalloc_inodes, &root->delalloc_inodes);
2402 root->nr_delalloc_inodes++;
2403 if (root->nr_delalloc_inodes == 1) {
2404 spin_lock(&fs_info->delalloc_root_lock);
2405 ASSERT(list_empty(&root->delalloc_root));
2406 list_add_tail(&root->delalloc_root, &fs_info->delalloc_roots);
2407 spin_unlock(&fs_info->delalloc_root_lock);
2409 spin_unlock(&root->delalloc_lock);
2412 void btrfs_del_delalloc_inode(struct btrfs_inode *inode)
2414 struct btrfs_root *root = inode->root;
2415 struct btrfs_fs_info *fs_info = root->fs_info;
2417 lockdep_assert_held(&root->delalloc_lock);
2420 * We may be called after the inode was already deleted from the list,
2421 * namely in the transaction abort path btrfs_destroy_delalloc_inodes(),
2422 * and then later through btrfs_clear_delalloc_extent() while the inode
2423 * still has ->delalloc_bytes > 0.
2425 if (!list_empty(&inode->delalloc_inodes)) {
2426 list_del_init(&inode->delalloc_inodes);
2427 root->nr_delalloc_inodes--;
2428 if (!root->nr_delalloc_inodes) {
2429 ASSERT(list_empty(&root->delalloc_inodes));
2430 spin_lock(&fs_info->delalloc_root_lock);
2431 ASSERT(!list_empty(&root->delalloc_root));
2432 list_del_init(&root->delalloc_root);
2433 spin_unlock(&fs_info->delalloc_root_lock);
2439 * Properly track delayed allocation bytes in the inode and to maintain the
2440 * list of inodes that have pending delalloc work to be done.
2442 void btrfs_set_delalloc_extent(struct btrfs_inode *inode, struct extent_state *state,
2445 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2447 lockdep_assert_held(&inode->io_tree.lock);
2449 if ((bits & EXTENT_DEFRAG) && !(bits & EXTENT_DELALLOC))
2452 * set_bit and clear bit hooks normally require _irqsave/restore
2453 * but in this case, we are only testing for the DELALLOC
2454 * bit, which is only set or cleared with irqs on
2456 if (!(state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2457 u64 len = state->end + 1 - state->start;
2458 u64 prev_delalloc_bytes;
2459 u32 num_extents = count_max_extents(fs_info, len);
2461 spin_lock(&inode->lock);
2462 btrfs_mod_outstanding_extents(inode, num_extents);
2463 spin_unlock(&inode->lock);
2465 /* For sanity tests */
2466 if (btrfs_is_testing(fs_info))
2469 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2470 fs_info->delalloc_batch);
2471 spin_lock(&inode->lock);
2472 prev_delalloc_bytes = inode->delalloc_bytes;
2473 inode->delalloc_bytes += len;
2474 if (bits & EXTENT_DEFRAG)
2475 inode->defrag_bytes += len;
2476 spin_unlock(&inode->lock);
2479 * We don't need to be under the protection of the inode's lock,
2480 * because we are called while holding the inode's io_tree lock
2481 * and are therefore protected against concurrent calls of this
2482 * function and btrfs_clear_delalloc_extent().
2484 if (!btrfs_is_free_space_inode(inode) && prev_delalloc_bytes == 0)
2485 btrfs_add_delalloc_inode(inode);
2488 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2489 (bits & EXTENT_DELALLOC_NEW)) {
2490 spin_lock(&inode->lock);
2491 inode->new_delalloc_bytes += state->end + 1 - state->start;
2492 spin_unlock(&inode->lock);
2497 * Once a range is no longer delalloc this function ensures that proper
2498 * accounting happens.
2500 void btrfs_clear_delalloc_extent(struct btrfs_inode *inode,
2501 struct extent_state *state, u32 bits)
2503 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2504 u64 len = state->end + 1 - state->start;
2505 u32 num_extents = count_max_extents(fs_info, len);
2507 lockdep_assert_held(&inode->io_tree.lock);
2509 if ((state->state & EXTENT_DEFRAG) && (bits & EXTENT_DEFRAG)) {
2510 spin_lock(&inode->lock);
2511 inode->defrag_bytes -= len;
2512 spin_unlock(&inode->lock);
2516 * set_bit and clear bit hooks normally require _irqsave/restore
2517 * but in this case, we are only testing for the DELALLOC
2518 * bit, which is only set or cleared with irqs on
2520 if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2521 struct btrfs_root *root = inode->root;
2522 u64 new_delalloc_bytes;
2524 spin_lock(&inode->lock);
2525 btrfs_mod_outstanding_extents(inode, -num_extents);
2526 spin_unlock(&inode->lock);
2529 * We don't reserve metadata space for space cache inodes so we
2530 * don't need to call delalloc_release_metadata if there is an
2533 if (bits & EXTENT_CLEAR_META_RESV &&
2534 root != fs_info->tree_root)
2535 btrfs_delalloc_release_metadata(inode, len, true);
2537 /* For sanity tests. */
2538 if (btrfs_is_testing(fs_info))
2541 if (!btrfs_is_data_reloc_root(root) &&
2542 !btrfs_is_free_space_inode(inode) &&
2543 !(state->state & EXTENT_NORESERVE) &&
2544 (bits & EXTENT_CLEAR_DATA_RESV))
2545 btrfs_free_reserved_data_space_noquota(fs_info, len);
2547 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2548 fs_info->delalloc_batch);
2549 spin_lock(&inode->lock);
2550 inode->delalloc_bytes -= len;
2551 new_delalloc_bytes = inode->delalloc_bytes;
2552 spin_unlock(&inode->lock);
2555 * We don't need to be under the protection of the inode's lock,
2556 * because we are called while holding the inode's io_tree lock
2557 * and are therefore protected against concurrent calls of this
2558 * function and btrfs_set_delalloc_extent().
2560 if (!btrfs_is_free_space_inode(inode) && new_delalloc_bytes == 0) {
2561 spin_lock(&root->delalloc_lock);
2562 btrfs_del_delalloc_inode(inode);
2563 spin_unlock(&root->delalloc_lock);
2567 if ((state->state & EXTENT_DELALLOC_NEW) &&
2568 (bits & EXTENT_DELALLOC_NEW)) {
2569 spin_lock(&inode->lock);
2570 ASSERT(inode->new_delalloc_bytes >= len);
2571 inode->new_delalloc_bytes -= len;
2572 if (bits & EXTENT_ADD_INODE_BYTES)
2573 inode_add_bytes(&inode->vfs_inode, len);
2574 spin_unlock(&inode->lock);
2578 static int btrfs_extract_ordered_extent(struct btrfs_bio *bbio,
2579 struct btrfs_ordered_extent *ordered)
2581 u64 start = (u64)bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT;
2582 u64 len = bbio->bio.bi_iter.bi_size;
2583 struct btrfs_ordered_extent *new;
2586 /* Must always be called for the beginning of an ordered extent. */
2587 if (WARN_ON_ONCE(start != ordered->disk_bytenr))
2590 /* No need to split if the ordered extent covers the entire bio. */
2591 if (ordered->disk_num_bytes == len) {
2592 refcount_inc(&ordered->refs);
2593 bbio->ordered = ordered;
2598 * Don't split the extent_map for NOCOW extents, as we're writing into
2599 * a pre-existing one.
2601 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
2602 ret = split_extent_map(bbio->inode, bbio->file_offset,
2603 ordered->num_bytes, len,
2604 ordered->disk_bytenr);
2609 new = btrfs_split_ordered_extent(ordered, len);
2611 return PTR_ERR(new);
2612 bbio->ordered = new;
2617 * given a list of ordered sums record them in the inode. This happens
2618 * at IO completion time based on sums calculated at bio submission time.
2620 static int add_pending_csums(struct btrfs_trans_handle *trans,
2621 struct list_head *list)
2623 struct btrfs_ordered_sum *sum;
2624 struct btrfs_root *csum_root = NULL;
2627 list_for_each_entry(sum, list, list) {
2628 trans->adding_csums = true;
2630 csum_root = btrfs_csum_root(trans->fs_info,
2632 ret = btrfs_csum_file_blocks(trans, csum_root, sum);
2633 trans->adding_csums = false;
2640 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2643 struct extent_state **cached_state)
2645 u64 search_start = start;
2646 const u64 end = start + len - 1;
2648 while (search_start < end) {
2649 const u64 search_len = end - search_start + 1;
2650 struct extent_map *em;
2654 em = btrfs_get_extent(inode, NULL, search_start, search_len);
2658 if (em->block_start != EXTENT_MAP_HOLE)
2662 if (em->start < search_start)
2663 em_len -= search_start - em->start;
2664 if (em_len > search_len)
2665 em_len = search_len;
2667 ret = set_extent_bit(&inode->io_tree, search_start,
2668 search_start + em_len - 1,
2669 EXTENT_DELALLOC_NEW, cached_state);
2671 search_start = extent_map_end(em);
2672 free_extent_map(em);
2679 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2680 unsigned int extra_bits,
2681 struct extent_state **cached_state)
2683 WARN_ON(PAGE_ALIGNED(end));
2685 if (start >= i_size_read(&inode->vfs_inode) &&
2686 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2688 * There can't be any extents following eof in this case so just
2689 * set the delalloc new bit for the range directly.
2691 extra_bits |= EXTENT_DELALLOC_NEW;
2695 ret = btrfs_find_new_delalloc_bytes(inode, start,
2702 return set_extent_bit(&inode->io_tree, start, end,
2703 EXTENT_DELALLOC | extra_bits, cached_state);
2706 /* see btrfs_writepage_start_hook for details on why this is required */
2707 struct btrfs_writepage_fixup {
2709 struct btrfs_inode *inode;
2710 struct btrfs_work work;
2713 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2715 struct btrfs_writepage_fixup *fixup =
2716 container_of(work, struct btrfs_writepage_fixup, work);
2717 struct btrfs_ordered_extent *ordered;
2718 struct extent_state *cached_state = NULL;
2719 struct extent_changeset *data_reserved = NULL;
2720 struct page *page = fixup->page;
2721 struct btrfs_inode *inode = fixup->inode;
2722 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2723 u64 page_start = page_offset(page);
2724 u64 page_end = page_offset(page) + PAGE_SIZE - 1;
2726 bool free_delalloc_space = true;
2729 * This is similar to page_mkwrite, we need to reserve the space before
2730 * we take the page lock.
2732 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2738 * Before we queued this fixup, we took a reference on the page.
2739 * page->mapping may go NULL, but it shouldn't be moved to a different
2742 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2744 * Unfortunately this is a little tricky, either
2746 * 1) We got here and our page had already been dealt with and
2747 * we reserved our space, thus ret == 0, so we need to just
2748 * drop our space reservation and bail. This can happen the
2749 * first time we come into the fixup worker, or could happen
2750 * while waiting for the ordered extent.
2751 * 2) Our page was already dealt with, but we happened to get an
2752 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2753 * this case we obviously don't have anything to release, but
2754 * because the page was already dealt with we don't want to
2755 * mark the page with an error, so make sure we're resetting
2756 * ret to 0. This is why we have this check _before_ the ret
2757 * check, because we do not want to have a surprise ENOSPC
2758 * when the page was already properly dealt with.
2761 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2762 btrfs_delalloc_release_space(inode, data_reserved,
2763 page_start, PAGE_SIZE,
2771 * We can't mess with the page state unless it is locked, so now that
2772 * it is locked bail if we failed to make our space reservation.
2777 lock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2779 /* already ordered? We're done */
2780 if (PageOrdered(page))
2783 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2785 unlock_extent(&inode->io_tree, page_start, page_end,
2788 btrfs_start_ordered_extent(ordered);
2789 btrfs_put_ordered_extent(ordered);
2793 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2799 * Everything went as planned, we're now the owner of a dirty page with
2800 * delayed allocation bits set and space reserved for our COW
2803 * The page was dirty when we started, nothing should have cleaned it.
2805 BUG_ON(!PageDirty(page));
2806 free_delalloc_space = false;
2808 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2809 if (free_delalloc_space)
2810 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2812 unlock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2816 * We hit ENOSPC or other errors. Update the mapping and page
2817 * to reflect the errors and clean the page.
2819 mapping_set_error(page->mapping, ret);
2820 btrfs_mark_ordered_io_finished(inode, page, page_start,
2822 clear_page_dirty_for_io(page);
2824 btrfs_folio_clear_checked(fs_info, page_folio(page), page_start, PAGE_SIZE);
2828 extent_changeset_free(data_reserved);
2830 * As a precaution, do a delayed iput in case it would be the last iput
2831 * that could need flushing space. Recursing back to fixup worker would
2834 btrfs_add_delayed_iput(inode);
2838 * There are a few paths in the higher layers of the kernel that directly
2839 * set the page dirty bit without asking the filesystem if it is a
2840 * good idea. This causes problems because we want to make sure COW
2841 * properly happens and the data=ordered rules are followed.
2843 * In our case any range that doesn't have the ORDERED bit set
2844 * hasn't been properly setup for IO. We kick off an async process
2845 * to fix it up. The async helper will wait for ordered extents, set
2846 * the delalloc bit and make it safe to write the page.
2848 int btrfs_writepage_cow_fixup(struct page *page)
2850 struct inode *inode = page->mapping->host;
2851 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
2852 struct btrfs_writepage_fixup *fixup;
2854 /* This page has ordered extent covering it already */
2855 if (PageOrdered(page))
2859 * PageChecked is set below when we create a fixup worker for this page,
2860 * don't try to create another one if we're already PageChecked()
2862 * The extent_io writepage code will redirty the page if we send back
2865 if (PageChecked(page))
2868 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2873 * We are already holding a reference to this inode from
2874 * write_cache_pages. We need to hold it because the space reservation
2875 * takes place outside of the page lock, and we can't trust
2876 * page->mapping outside of the page lock.
2879 btrfs_folio_set_checked(fs_info, page_folio(page), page_offset(page), PAGE_SIZE);
2881 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL);
2883 fixup->inode = BTRFS_I(inode);
2884 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2889 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2890 struct btrfs_inode *inode, u64 file_pos,
2891 struct btrfs_file_extent_item *stack_fi,
2892 const bool update_inode_bytes,
2893 u64 qgroup_reserved)
2895 struct btrfs_root *root = inode->root;
2896 const u64 sectorsize = root->fs_info->sectorsize;
2897 struct btrfs_path *path;
2898 struct extent_buffer *leaf;
2899 struct btrfs_key ins;
2900 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
2901 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
2902 u64 offset = btrfs_stack_file_extent_offset(stack_fi);
2903 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
2904 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
2905 struct btrfs_drop_extents_args drop_args = { 0 };
2908 path = btrfs_alloc_path();
2913 * we may be replacing one extent in the tree with another.
2914 * The new extent is pinned in the extent map, and we don't want
2915 * to drop it from the cache until it is completely in the btree.
2917 * So, tell btrfs_drop_extents to leave this extent in the cache.
2918 * the caller is expected to unpin it and allow it to be merged
2921 drop_args.path = path;
2922 drop_args.start = file_pos;
2923 drop_args.end = file_pos + num_bytes;
2924 drop_args.replace_extent = true;
2925 drop_args.extent_item_size = sizeof(*stack_fi);
2926 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2930 if (!drop_args.extent_inserted) {
2931 ins.objectid = btrfs_ino(inode);
2932 ins.offset = file_pos;
2933 ins.type = BTRFS_EXTENT_DATA_KEY;
2935 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2940 leaf = path->nodes[0];
2941 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
2942 write_extent_buffer(leaf, stack_fi,
2943 btrfs_item_ptr_offset(leaf, path->slots[0]),
2944 sizeof(struct btrfs_file_extent_item));
2946 btrfs_mark_buffer_dirty(trans, leaf);
2947 btrfs_release_path(path);
2950 * If we dropped an inline extent here, we know the range where it is
2951 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
2952 * number of bytes only for that range containing the inline extent.
2953 * The remaining of the range will be processed when clearning the
2954 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
2956 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
2957 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
2959 inline_size = drop_args.bytes_found - inline_size;
2960 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
2961 drop_args.bytes_found -= inline_size;
2962 num_bytes -= sectorsize;
2965 if (update_inode_bytes)
2966 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
2968 ins.objectid = disk_bytenr;
2969 ins.offset = disk_num_bytes;
2970 ins.type = BTRFS_EXTENT_ITEM_KEY;
2972 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
2976 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
2978 qgroup_reserved, &ins);
2980 btrfs_free_path(path);
2985 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2988 struct btrfs_block_group *cache;
2990 cache = btrfs_lookup_block_group(fs_info, start);
2993 spin_lock(&cache->lock);
2994 cache->delalloc_bytes -= len;
2995 spin_unlock(&cache->lock);
2997 btrfs_put_block_group(cache);
3000 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
3001 struct btrfs_ordered_extent *oe)
3003 struct btrfs_file_extent_item stack_fi;
3004 bool update_inode_bytes;
3005 u64 num_bytes = oe->num_bytes;
3006 u64 ram_bytes = oe->ram_bytes;
3008 memset(&stack_fi, 0, sizeof(stack_fi));
3009 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
3010 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
3011 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
3012 oe->disk_num_bytes);
3013 btrfs_set_stack_file_extent_offset(&stack_fi, oe->offset);
3014 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags)) {
3015 num_bytes = oe->truncated_len;
3016 ram_bytes = num_bytes;
3018 btrfs_set_stack_file_extent_num_bytes(&stack_fi, num_bytes);
3019 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, ram_bytes);
3020 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
3021 /* Encryption and other encoding is reserved and all 0 */
3024 * For delalloc, when completing an ordered extent we update the inode's
3025 * bytes when clearing the range in the inode's io tree, so pass false
3026 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
3027 * except if the ordered extent was truncated.
3029 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
3030 test_bit(BTRFS_ORDERED_ENCODED, &oe->flags) ||
3031 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
3033 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
3034 oe->file_offset, &stack_fi,
3035 update_inode_bytes, oe->qgroup_rsv);
3039 * As ordered data IO finishes, this gets called so we can finish
3040 * an ordered extent if the range of bytes in the file it covers are
3043 int btrfs_finish_one_ordered(struct btrfs_ordered_extent *ordered_extent)
3045 struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
3046 struct btrfs_root *root = inode->root;
3047 struct btrfs_fs_info *fs_info = root->fs_info;
3048 struct btrfs_trans_handle *trans = NULL;
3049 struct extent_io_tree *io_tree = &inode->io_tree;
3050 struct extent_state *cached_state = NULL;
3052 int compress_type = 0;
3054 u64 logical_len = ordered_extent->num_bytes;
3055 bool freespace_inode;
3056 bool truncated = false;
3057 bool clear_reserved_extent = true;
3058 unsigned int clear_bits = EXTENT_DEFRAG;
3060 start = ordered_extent->file_offset;
3061 end = start + ordered_extent->num_bytes - 1;
3063 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3064 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
3065 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags) &&
3066 !test_bit(BTRFS_ORDERED_ENCODED, &ordered_extent->flags))
3067 clear_bits |= EXTENT_DELALLOC_NEW;
3069 freespace_inode = btrfs_is_free_space_inode(inode);
3070 if (!freespace_inode)
3071 btrfs_lockdep_acquire(fs_info, btrfs_ordered_extent);
3073 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
3078 if (btrfs_is_zoned(fs_info))
3079 btrfs_zone_finish_endio(fs_info, ordered_extent->disk_bytenr,
3080 ordered_extent->disk_num_bytes);
3082 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
3084 logical_len = ordered_extent->truncated_len;
3085 /* Truncated the entire extent, don't bother adding */
3090 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
3091 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
3093 btrfs_inode_safe_disk_i_size_write(inode, 0);
3094 if (freespace_inode)
3095 trans = btrfs_join_transaction_spacecache(root);
3097 trans = btrfs_join_transaction(root);
3098 if (IS_ERR(trans)) {
3099 ret = PTR_ERR(trans);
3103 trans->block_rsv = &inode->block_rsv;
3104 ret = btrfs_update_inode_fallback(trans, inode);
3105 if (ret) /* -ENOMEM or corruption */
3106 btrfs_abort_transaction(trans, ret);
3110 clear_bits |= EXTENT_LOCKED;
3111 lock_extent(io_tree, start, end, &cached_state);
3113 if (freespace_inode)
3114 trans = btrfs_join_transaction_spacecache(root);
3116 trans = btrfs_join_transaction(root);
3117 if (IS_ERR(trans)) {
3118 ret = PTR_ERR(trans);
3123 trans->block_rsv = &inode->block_rsv;
3125 ret = btrfs_insert_raid_extent(trans, ordered_extent);
3129 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3130 compress_type = ordered_extent->compress_type;
3131 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3132 BUG_ON(compress_type);
3133 ret = btrfs_mark_extent_written(trans, inode,
3134 ordered_extent->file_offset,
3135 ordered_extent->file_offset +
3137 btrfs_zoned_release_data_reloc_bg(fs_info, ordered_extent->disk_bytenr,
3138 ordered_extent->disk_num_bytes);
3140 BUG_ON(root == fs_info->tree_root);
3141 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
3143 clear_reserved_extent = false;
3144 btrfs_release_delalloc_bytes(fs_info,
3145 ordered_extent->disk_bytenr,
3146 ordered_extent->disk_num_bytes);
3150 btrfs_abort_transaction(trans, ret);
3154 ret = unpin_extent_cache(inode, ordered_extent->file_offset,
3155 ordered_extent->num_bytes, trans->transid);
3157 btrfs_abort_transaction(trans, ret);
3161 ret = add_pending_csums(trans, &ordered_extent->list);
3163 btrfs_abort_transaction(trans, ret);
3168 * If this is a new delalloc range, clear its new delalloc flag to
3169 * update the inode's number of bytes. This needs to be done first
3170 * before updating the inode item.
3172 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
3173 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
3174 clear_extent_bit(&inode->io_tree, start, end,
3175 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
3178 btrfs_inode_safe_disk_i_size_write(inode, 0);
3179 ret = btrfs_update_inode_fallback(trans, inode);
3180 if (ret) { /* -ENOMEM or corruption */
3181 btrfs_abort_transaction(trans, ret);
3186 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
3190 btrfs_end_transaction(trans);
3192 if (ret || truncated) {
3193 u64 unwritten_start = start;
3196 * If we failed to finish this ordered extent for any reason we
3197 * need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3198 * extent, and mark the inode with the error if it wasn't
3199 * already set. Any error during writeback would have already
3200 * set the mapping error, so we need to set it if we're the ones
3201 * marking this ordered extent as failed.
3203 if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR,
3204 &ordered_extent->flags))
3205 mapping_set_error(ordered_extent->inode->i_mapping, -EIO);
3208 unwritten_start += logical_len;
3209 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
3212 * Drop extent maps for the part of the extent we didn't write.
3214 * We have an exception here for the free_space_inode, this is
3215 * because when we do btrfs_get_extent() on the free space inode
3216 * we will search the commit root. If this is a new block group
3217 * we won't find anything, and we will trip over the assert in
3218 * writepage where we do ASSERT(em->block_start !=
3221 * Theoretically we could also skip this for any NOCOW extent as
3222 * we don't mess with the extent map tree in the NOCOW case, but
3223 * for now simply skip this if we are the free space inode.
3225 if (!btrfs_is_free_space_inode(inode))
3226 btrfs_drop_extent_map_range(inode, unwritten_start,
3230 * If the ordered extent had an IOERR or something else went
3231 * wrong we need to return the space for this ordered extent
3232 * back to the allocator. We only free the extent in the
3233 * truncated case if we didn't write out the extent at all.
3235 * If we made it past insert_reserved_file_extent before we
3236 * errored out then we don't need to do this as the accounting
3237 * has already been done.
3239 if ((ret || !logical_len) &&
3240 clear_reserved_extent &&
3241 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3242 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3244 * Discard the range before returning it back to the
3247 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3248 btrfs_discard_extent(fs_info,
3249 ordered_extent->disk_bytenr,
3250 ordered_extent->disk_num_bytes,
3252 btrfs_free_reserved_extent(fs_info,
3253 ordered_extent->disk_bytenr,
3254 ordered_extent->disk_num_bytes, 1);
3256 * Actually free the qgroup rsv which was released when
3257 * the ordered extent was created.
3259 btrfs_qgroup_free_refroot(fs_info, inode->root->root_key.objectid,
3260 ordered_extent->qgroup_rsv,
3261 BTRFS_QGROUP_RSV_DATA);
3266 * This needs to be done to make sure anybody waiting knows we are done
3267 * updating everything for this ordered extent.
3269 btrfs_remove_ordered_extent(inode, ordered_extent);
3272 btrfs_put_ordered_extent(ordered_extent);
3273 /* once for the tree */
3274 btrfs_put_ordered_extent(ordered_extent);
3279 int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered)
3281 if (btrfs_is_zoned(inode_to_fs_info(ordered->inode)) &&
3282 !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags) &&
3283 list_empty(&ordered->bioc_list))
3284 btrfs_finish_ordered_zoned(ordered);
3285 return btrfs_finish_one_ordered(ordered);
3289 * Verify the checksum for a single sector without any extra action that depend
3290 * on the type of I/O.
3292 int btrfs_check_sector_csum(struct btrfs_fs_info *fs_info, struct page *page,
3293 u32 pgoff, u8 *csum, const u8 * const csum_expected)
3295 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3298 ASSERT(pgoff + fs_info->sectorsize <= PAGE_SIZE);
3300 shash->tfm = fs_info->csum_shash;
3302 kaddr = kmap_local_page(page) + pgoff;
3303 crypto_shash_digest(shash, kaddr, fs_info->sectorsize, csum);
3304 kunmap_local(kaddr);
3306 if (memcmp(csum, csum_expected, fs_info->csum_size))
3312 * Verify the checksum of a single data sector.
3314 * @bbio: btrfs_io_bio which contains the csum
3315 * @dev: device the sector is on
3316 * @bio_offset: offset to the beginning of the bio (in bytes)
3317 * @bv: bio_vec to check
3319 * Check if the checksum on a data block is valid. When a checksum mismatch is
3320 * detected, report the error and fill the corrupted range with zero.
3322 * Return %true if the sector is ok or had no checksum to start with, else %false.
3324 bool btrfs_data_csum_ok(struct btrfs_bio *bbio, struct btrfs_device *dev,
3325 u32 bio_offset, struct bio_vec *bv)
3327 struct btrfs_inode *inode = bbio->inode;
3328 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3329 u64 file_offset = bbio->file_offset + bio_offset;
3330 u64 end = file_offset + bv->bv_len - 1;
3332 u8 csum[BTRFS_CSUM_SIZE];
3334 ASSERT(bv->bv_len == fs_info->sectorsize);
3339 if (btrfs_is_data_reloc_root(inode->root) &&
3340 test_range_bit(&inode->io_tree, file_offset, end, EXTENT_NODATASUM,
3342 /* Skip the range without csum for data reloc inode */
3343 clear_extent_bits(&inode->io_tree, file_offset, end,
3348 csum_expected = bbio->csum + (bio_offset >> fs_info->sectorsize_bits) *
3350 if (btrfs_check_sector_csum(fs_info, bv->bv_page, bv->bv_offset, csum,
3356 btrfs_print_data_csum_error(inode, file_offset, csum, csum_expected,
3359 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS);
3365 * Perform a delayed iput on @inode.
3367 * @inode: The inode we want to perform iput on
3369 * This function uses the generic vfs_inode::i_count to track whether we should
3370 * just decrement it (in case it's > 1) or if this is the last iput then link
3371 * the inode to the delayed iput machinery. Delayed iputs are processed at
3372 * transaction commit time/superblock commit/cleaner kthread.
3374 void btrfs_add_delayed_iput(struct btrfs_inode *inode)
3376 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3377 unsigned long flags;
3379 if (atomic_add_unless(&inode->vfs_inode.i_count, -1, 1))
3382 atomic_inc(&fs_info->nr_delayed_iputs);
3384 * Need to be irq safe here because we can be called from either an irq
3385 * context (see bio.c and btrfs_put_ordered_extent()) or a non-irq
3388 spin_lock_irqsave(&fs_info->delayed_iput_lock, flags);
3389 ASSERT(list_empty(&inode->delayed_iput));
3390 list_add_tail(&inode->delayed_iput, &fs_info->delayed_iputs);
3391 spin_unlock_irqrestore(&fs_info->delayed_iput_lock, flags);
3392 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3393 wake_up_process(fs_info->cleaner_kthread);
3396 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3397 struct btrfs_inode *inode)
3399 list_del_init(&inode->delayed_iput);
3400 spin_unlock_irq(&fs_info->delayed_iput_lock);
3401 iput(&inode->vfs_inode);
3402 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3403 wake_up(&fs_info->delayed_iputs_wait);
3404 spin_lock_irq(&fs_info->delayed_iput_lock);
3407 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3408 struct btrfs_inode *inode)
3410 if (!list_empty(&inode->delayed_iput)) {
3411 spin_lock_irq(&fs_info->delayed_iput_lock);
3412 if (!list_empty(&inode->delayed_iput))
3413 run_delayed_iput_locked(fs_info, inode);
3414 spin_unlock_irq(&fs_info->delayed_iput_lock);
3418 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3421 * btrfs_put_ordered_extent() can run in irq context (see bio.c), which
3422 * calls btrfs_add_delayed_iput() and that needs to lock
3423 * fs_info->delayed_iput_lock. So we need to disable irqs here to
3424 * prevent a deadlock.
3426 spin_lock_irq(&fs_info->delayed_iput_lock);
3427 while (!list_empty(&fs_info->delayed_iputs)) {
3428 struct btrfs_inode *inode;
3430 inode = list_first_entry(&fs_info->delayed_iputs,
3431 struct btrfs_inode, delayed_iput);
3432 run_delayed_iput_locked(fs_info, inode);
3433 if (need_resched()) {
3434 spin_unlock_irq(&fs_info->delayed_iput_lock);
3436 spin_lock_irq(&fs_info->delayed_iput_lock);
3439 spin_unlock_irq(&fs_info->delayed_iput_lock);
3443 * Wait for flushing all delayed iputs
3445 * @fs_info: the filesystem
3447 * This will wait on any delayed iputs that are currently running with KILLABLE
3448 * set. Once they are all done running we will return, unless we are killed in
3449 * which case we return EINTR. This helps in user operations like fallocate etc
3450 * that might get blocked on the iputs.
3452 * Return EINTR if we were killed, 0 if nothing's pending
3454 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3456 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3457 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3464 * This creates an orphan entry for the given inode in case something goes wrong
3465 * in the middle of an unlink.
3467 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3468 struct btrfs_inode *inode)
3472 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3473 if (ret && ret != -EEXIST) {
3474 btrfs_abort_transaction(trans, ret);
3482 * We have done the delete so we can go ahead and remove the orphan item for
3483 * this particular inode.
3485 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3486 struct btrfs_inode *inode)
3488 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3492 * this cleans up any orphans that may be left on the list from the last use
3495 int btrfs_orphan_cleanup(struct btrfs_root *root)
3497 struct btrfs_fs_info *fs_info = root->fs_info;
3498 struct btrfs_path *path;
3499 struct extent_buffer *leaf;
3500 struct btrfs_key key, found_key;
3501 struct btrfs_trans_handle *trans;
3502 struct inode *inode;
3503 u64 last_objectid = 0;
3504 int ret = 0, nr_unlink = 0;
3506 if (test_and_set_bit(BTRFS_ROOT_ORPHAN_CLEANUP, &root->state))
3509 path = btrfs_alloc_path();
3514 path->reada = READA_BACK;
3516 key.objectid = BTRFS_ORPHAN_OBJECTID;
3517 key.type = BTRFS_ORPHAN_ITEM_KEY;
3518 key.offset = (u64)-1;
3521 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3526 * if ret == 0 means we found what we were searching for, which
3527 * is weird, but possible, so only screw with path if we didn't
3528 * find the key and see if we have stuff that matches
3532 if (path->slots[0] == 0)
3537 /* pull out the item */
3538 leaf = path->nodes[0];
3539 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3541 /* make sure the item matches what we want */
3542 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3544 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3547 /* release the path since we're done with it */
3548 btrfs_release_path(path);
3551 * this is where we are basically btrfs_lookup, without the
3552 * crossing root thing. we store the inode number in the
3553 * offset of the orphan item.
3556 if (found_key.offset == last_objectid) {
3558 * We found the same inode as before. This means we were
3559 * not able to remove its items via eviction triggered
3560 * by an iput(). A transaction abort may have happened,
3561 * due to -ENOSPC for example, so try to grab the error
3562 * that lead to a transaction abort, if any.
3565 "Error removing orphan entry, stopping orphan cleanup");
3566 ret = BTRFS_FS_ERROR(fs_info) ?: -EINVAL;
3570 last_objectid = found_key.offset;
3572 found_key.objectid = found_key.offset;
3573 found_key.type = BTRFS_INODE_ITEM_KEY;
3574 found_key.offset = 0;
3575 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3576 if (IS_ERR(inode)) {
3577 ret = PTR_ERR(inode);
3583 if (!inode && root == fs_info->tree_root) {
3584 struct btrfs_root *dead_root;
3585 int is_dead_root = 0;
3588 * This is an orphan in the tree root. Currently these
3589 * could come from 2 sources:
3590 * a) a root (snapshot/subvolume) deletion in progress
3591 * b) a free space cache inode
3592 * We need to distinguish those two, as the orphan item
3593 * for a root must not get deleted before the deletion
3594 * of the snapshot/subvolume's tree completes.
3596 * btrfs_find_orphan_roots() ran before us, which has
3597 * found all deleted roots and loaded them into
3598 * fs_info->fs_roots_radix. So here we can find if an
3599 * orphan item corresponds to a deleted root by looking
3600 * up the root from that radix tree.
3603 spin_lock(&fs_info->fs_roots_radix_lock);
3604 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3605 (unsigned long)found_key.objectid);
3606 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3608 spin_unlock(&fs_info->fs_roots_radix_lock);
3611 /* prevent this orphan from being found again */
3612 key.offset = found_key.objectid - 1;
3619 * If we have an inode with links, there are a couple of
3622 * 1. We were halfway through creating fsverity metadata for the
3623 * file. In that case, the orphan item represents incomplete
3624 * fsverity metadata which must be cleaned up with
3625 * btrfs_drop_verity_items and deleting the orphan item.
3627 * 2. Old kernels (before v3.12) used to create an
3628 * orphan item for truncate indicating that there were possibly
3629 * extent items past i_size that needed to be deleted. In v3.12,
3630 * truncate was changed to update i_size in sync with the extent
3631 * items, but the (useless) orphan item was still created. Since
3632 * v4.18, we don't create the orphan item for truncate at all.
3634 * So, this item could mean that we need to do a truncate, but
3635 * only if this filesystem was last used on a pre-v3.12 kernel
3636 * and was not cleanly unmounted. The odds of that are quite
3637 * slim, and it's a pain to do the truncate now, so just delete
3640 * It's also possible that this orphan item was supposed to be
3641 * deleted but wasn't. The inode number may have been reused,
3642 * but either way, we can delete the orphan item.
3644 if (!inode || inode->i_nlink) {
3646 ret = btrfs_drop_verity_items(BTRFS_I(inode));
3652 trans = btrfs_start_transaction(root, 1);
3653 if (IS_ERR(trans)) {
3654 ret = PTR_ERR(trans);
3657 btrfs_debug(fs_info, "auto deleting %Lu",
3658 found_key.objectid);
3659 ret = btrfs_del_orphan_item(trans, root,
3660 found_key.objectid);
3661 btrfs_end_transaction(trans);
3669 /* this will do delete_inode and everything for us */
3672 /* release the path since we're done with it */
3673 btrfs_release_path(path);
3675 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3676 trans = btrfs_join_transaction(root);
3678 btrfs_end_transaction(trans);
3682 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3686 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3687 btrfs_free_path(path);
3692 * very simple check to peek ahead in the leaf looking for xattrs. If we
3693 * don't find any xattrs, we know there can't be any acls.
3695 * slot is the slot the inode is in, objectid is the objectid of the inode
3697 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3698 int slot, u64 objectid,
3699 int *first_xattr_slot)
3701 u32 nritems = btrfs_header_nritems(leaf);
3702 struct btrfs_key found_key;
3703 static u64 xattr_access = 0;
3704 static u64 xattr_default = 0;
3707 if (!xattr_access) {
3708 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3709 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3710 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3711 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3715 *first_xattr_slot = -1;
3716 while (slot < nritems) {
3717 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3719 /* we found a different objectid, there must not be acls */
3720 if (found_key.objectid != objectid)
3723 /* we found an xattr, assume we've got an acl */
3724 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3725 if (*first_xattr_slot == -1)
3726 *first_xattr_slot = slot;
3727 if (found_key.offset == xattr_access ||
3728 found_key.offset == xattr_default)
3733 * we found a key greater than an xattr key, there can't
3734 * be any acls later on
3736 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3743 * it goes inode, inode backrefs, xattrs, extents,
3744 * so if there are a ton of hard links to an inode there can
3745 * be a lot of backrefs. Don't waste time searching too hard,
3746 * this is just an optimization
3751 /* we hit the end of the leaf before we found an xattr or
3752 * something larger than an xattr. We have to assume the inode
3755 if (*first_xattr_slot == -1)
3756 *first_xattr_slot = slot;
3761 * read an inode from the btree into the in-memory inode
3763 static int btrfs_read_locked_inode(struct inode *inode,
3764 struct btrfs_path *in_path)
3766 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
3767 struct btrfs_path *path = in_path;
3768 struct extent_buffer *leaf;
3769 struct btrfs_inode_item *inode_item;
3770 struct btrfs_root *root = BTRFS_I(inode)->root;
3771 struct btrfs_key location;
3776 bool filled = false;
3777 int first_xattr_slot;
3779 ret = btrfs_fill_inode(inode, &rdev);
3784 path = btrfs_alloc_path();
3789 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3791 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3793 if (path != in_path)
3794 btrfs_free_path(path);
3798 leaf = path->nodes[0];
3803 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3804 struct btrfs_inode_item);
3805 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3806 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3807 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3808 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3809 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3810 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3811 round_up(i_size_read(inode), fs_info->sectorsize));
3813 inode_set_atime(inode, btrfs_timespec_sec(leaf, &inode_item->atime),
3814 btrfs_timespec_nsec(leaf, &inode_item->atime));
3816 inode_set_mtime(inode, btrfs_timespec_sec(leaf, &inode_item->mtime),
3817 btrfs_timespec_nsec(leaf, &inode_item->mtime));
3819 inode_set_ctime(inode, btrfs_timespec_sec(leaf, &inode_item->ctime),
3820 btrfs_timespec_nsec(leaf, &inode_item->ctime));
3822 BTRFS_I(inode)->i_otime_sec = btrfs_timespec_sec(leaf, &inode_item->otime);
3823 BTRFS_I(inode)->i_otime_nsec = btrfs_timespec_nsec(leaf, &inode_item->otime);
3825 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3826 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3827 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3829 inode_set_iversion_queried(inode,
3830 btrfs_inode_sequence(leaf, inode_item));
3831 inode->i_generation = BTRFS_I(inode)->generation;
3833 rdev = btrfs_inode_rdev(leaf, inode_item);
3835 BTRFS_I(inode)->index_cnt = (u64)-1;
3836 btrfs_inode_split_flags(btrfs_inode_flags(leaf, inode_item),
3837 &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags);
3841 * If we were modified in the current generation and evicted from memory
3842 * and then re-read we need to do a full sync since we don't have any
3843 * idea about which extents were modified before we were evicted from
3846 * This is required for both inode re-read from disk and delayed inode
3847 * in the delayed_nodes xarray.
3849 if (BTRFS_I(inode)->last_trans == btrfs_get_fs_generation(fs_info))
3850 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3851 &BTRFS_I(inode)->runtime_flags);
3854 * We don't persist the id of the transaction where an unlink operation
3855 * against the inode was last made. So here we assume the inode might
3856 * have been evicted, and therefore the exact value of last_unlink_trans
3857 * lost, and set it to last_trans to avoid metadata inconsistencies
3858 * between the inode and its parent if the inode is fsync'ed and the log
3859 * replayed. For example, in the scenario:
3862 * ln mydir/foo mydir/bar
3865 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3866 * xfs_io -c fsync mydir/foo
3868 * mount fs, triggers fsync log replay
3870 * We must make sure that when we fsync our inode foo we also log its
3871 * parent inode, otherwise after log replay the parent still has the
3872 * dentry with the "bar" name but our inode foo has a link count of 1
3873 * and doesn't have an inode ref with the name "bar" anymore.
3875 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3876 * but it guarantees correctness at the expense of occasional full
3877 * transaction commits on fsync if our inode is a directory, or if our
3878 * inode is not a directory, logging its parent unnecessarily.
3880 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3883 * Same logic as for last_unlink_trans. We don't persist the generation
3884 * of the last transaction where this inode was used for a reflink
3885 * operation, so after eviction and reloading the inode we must be
3886 * pessimistic and assume the last transaction that modified the inode.
3888 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3891 if (inode->i_nlink != 1 ||
3892 path->slots[0] >= btrfs_header_nritems(leaf))
3895 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3896 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3899 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3900 if (location.type == BTRFS_INODE_REF_KEY) {
3901 struct btrfs_inode_ref *ref;
3903 ref = (struct btrfs_inode_ref *)ptr;
3904 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3905 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3906 struct btrfs_inode_extref *extref;
3908 extref = (struct btrfs_inode_extref *)ptr;
3909 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3914 * try to precache a NULL acl entry for files that don't have
3915 * any xattrs or acls
3917 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3918 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3919 if (first_xattr_slot != -1) {
3920 path->slots[0] = first_xattr_slot;
3921 ret = btrfs_load_inode_props(inode, path);
3924 "error loading props for ino %llu (root %llu): %d",
3925 btrfs_ino(BTRFS_I(inode)),
3926 root->root_key.objectid, ret);
3928 if (path != in_path)
3929 btrfs_free_path(path);
3932 cache_no_acl(inode);
3934 switch (inode->i_mode & S_IFMT) {
3936 inode->i_mapping->a_ops = &btrfs_aops;
3937 inode->i_fop = &btrfs_file_operations;
3938 inode->i_op = &btrfs_file_inode_operations;
3941 inode->i_fop = &btrfs_dir_file_operations;
3942 inode->i_op = &btrfs_dir_inode_operations;
3945 inode->i_op = &btrfs_symlink_inode_operations;
3946 inode_nohighmem(inode);
3947 inode->i_mapping->a_ops = &btrfs_aops;
3950 inode->i_op = &btrfs_special_inode_operations;
3951 init_special_inode(inode, inode->i_mode, rdev);
3955 btrfs_sync_inode_flags_to_i_flags(inode);
3960 * given a leaf and an inode, copy the inode fields into the leaf
3962 static void fill_inode_item(struct btrfs_trans_handle *trans,
3963 struct extent_buffer *leaf,
3964 struct btrfs_inode_item *item,
3965 struct inode *inode)
3967 struct btrfs_map_token token;
3970 btrfs_init_map_token(&token, leaf);
3972 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3973 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3974 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
3975 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3976 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3978 btrfs_set_token_timespec_sec(&token, &item->atime,
3979 inode_get_atime_sec(inode));
3980 btrfs_set_token_timespec_nsec(&token, &item->atime,
3981 inode_get_atime_nsec(inode));
3983 btrfs_set_token_timespec_sec(&token, &item->mtime,
3984 inode_get_mtime_sec(inode));
3985 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3986 inode_get_mtime_nsec(inode));
3988 btrfs_set_token_timespec_sec(&token, &item->ctime,
3989 inode_get_ctime_sec(inode));
3990 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3991 inode_get_ctime_nsec(inode));
3993 btrfs_set_token_timespec_sec(&token, &item->otime, BTRFS_I(inode)->i_otime_sec);
3994 btrfs_set_token_timespec_nsec(&token, &item->otime, BTRFS_I(inode)->i_otime_nsec);
3996 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
3997 btrfs_set_token_inode_generation(&token, item,
3998 BTRFS_I(inode)->generation);
3999 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
4000 btrfs_set_token_inode_transid(&token, item, trans->transid);
4001 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
4002 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
4003 BTRFS_I(inode)->ro_flags);
4004 btrfs_set_token_inode_flags(&token, item, flags);
4005 btrfs_set_token_inode_block_group(&token, item, 0);
4009 * copy everything in the in-memory inode into the btree.
4011 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
4012 struct btrfs_inode *inode)
4014 struct btrfs_inode_item *inode_item;
4015 struct btrfs_path *path;
4016 struct extent_buffer *leaf;
4019 path = btrfs_alloc_path();
4023 ret = btrfs_lookup_inode(trans, inode->root, path, &inode->location, 1);
4030 leaf = path->nodes[0];
4031 inode_item = btrfs_item_ptr(leaf, path->slots[0],
4032 struct btrfs_inode_item);
4034 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
4035 btrfs_mark_buffer_dirty(trans, leaf);
4036 btrfs_set_inode_last_trans(trans, inode);
4039 btrfs_free_path(path);
4044 * copy everything in the in-memory inode into the btree.
4046 int btrfs_update_inode(struct btrfs_trans_handle *trans,
4047 struct btrfs_inode *inode)
4049 struct btrfs_root *root = inode->root;
4050 struct btrfs_fs_info *fs_info = root->fs_info;
4054 * If the inode is a free space inode, we can deadlock during commit
4055 * if we put it into the delayed code.
4057 * The data relocation inode should also be directly updated
4060 if (!btrfs_is_free_space_inode(inode)
4061 && !btrfs_is_data_reloc_root(root)
4062 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4063 btrfs_update_root_times(trans, root);
4065 ret = btrfs_delayed_update_inode(trans, inode);
4067 btrfs_set_inode_last_trans(trans, inode);
4071 return btrfs_update_inode_item(trans, inode);
4074 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4075 struct btrfs_inode *inode)
4079 ret = btrfs_update_inode(trans, inode);
4081 return btrfs_update_inode_item(trans, inode);
4086 * unlink helper that gets used here in inode.c and in the tree logging
4087 * recovery code. It remove a link in a directory with a given name, and
4088 * also drops the back refs in the inode to the directory
4090 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4091 struct btrfs_inode *dir,
4092 struct btrfs_inode *inode,
4093 const struct fscrypt_str *name,
4094 struct btrfs_rename_ctx *rename_ctx)
4096 struct btrfs_root *root = dir->root;
4097 struct btrfs_fs_info *fs_info = root->fs_info;
4098 struct btrfs_path *path;
4100 struct btrfs_dir_item *di;
4102 u64 ino = btrfs_ino(inode);
4103 u64 dir_ino = btrfs_ino(dir);
4105 path = btrfs_alloc_path();
4111 di = btrfs_lookup_dir_item(trans, root, path, dir_ino, name, -1);
4112 if (IS_ERR_OR_NULL(di)) {
4113 ret = di ? PTR_ERR(di) : -ENOENT;
4116 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4119 btrfs_release_path(path);
4122 * If we don't have dir index, we have to get it by looking up
4123 * the inode ref, since we get the inode ref, remove it directly,
4124 * it is unnecessary to do delayed deletion.
4126 * But if we have dir index, needn't search inode ref to get it.
4127 * Since the inode ref is close to the inode item, it is better
4128 * that we delay to delete it, and just do this deletion when
4129 * we update the inode item.
4131 if (inode->dir_index) {
4132 ret = btrfs_delayed_delete_inode_ref(inode);
4134 index = inode->dir_index;
4139 ret = btrfs_del_inode_ref(trans, root, name, ino, dir_ino, &index);
4142 "failed to delete reference to %.*s, inode %llu parent %llu",
4143 name->len, name->name, ino, dir_ino);
4144 btrfs_abort_transaction(trans, ret);
4149 rename_ctx->index = index;
4151 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4153 btrfs_abort_transaction(trans, ret);
4158 * If we are in a rename context, we don't need to update anything in the
4159 * log. That will be done later during the rename by btrfs_log_new_name().
4160 * Besides that, doing it here would only cause extra unnecessary btree
4161 * operations on the log tree, increasing latency for applications.
4164 btrfs_del_inode_ref_in_log(trans, root, name, inode, dir_ino);
4165 btrfs_del_dir_entries_in_log(trans, root, name, dir, index);
4169 * If we have a pending delayed iput we could end up with the final iput
4170 * being run in btrfs-cleaner context. If we have enough of these built
4171 * up we can end up burning a lot of time in btrfs-cleaner without any
4172 * way to throttle the unlinks. Since we're currently holding a ref on
4173 * the inode we can run the delayed iput here without any issues as the
4174 * final iput won't be done until after we drop the ref we're currently
4177 btrfs_run_delayed_iput(fs_info, inode);
4179 btrfs_free_path(path);
4183 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name->len * 2);
4184 inode_inc_iversion(&inode->vfs_inode);
4185 inode_inc_iversion(&dir->vfs_inode);
4186 inode_set_mtime_to_ts(&dir->vfs_inode, inode_set_ctime_current(&dir->vfs_inode));
4187 ret = btrfs_update_inode(trans, dir);
4192 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4193 struct btrfs_inode *dir, struct btrfs_inode *inode,
4194 const struct fscrypt_str *name)
4198 ret = __btrfs_unlink_inode(trans, dir, inode, name, NULL);
4200 drop_nlink(&inode->vfs_inode);
4201 ret = btrfs_update_inode(trans, inode);
4207 * helper to start transaction for unlink and rmdir.
4209 * unlink and rmdir are special in btrfs, they do not always free space, so
4210 * if we cannot make our reservations the normal way try and see if there is
4211 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4212 * allow the unlink to occur.
4214 static struct btrfs_trans_handle *__unlink_start_trans(struct btrfs_inode *dir)
4216 struct btrfs_root *root = dir->root;
4218 return btrfs_start_transaction_fallback_global_rsv(root,
4219 BTRFS_UNLINK_METADATA_UNITS);
4222 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4224 struct btrfs_trans_handle *trans;
4225 struct inode *inode = d_inode(dentry);
4227 struct fscrypt_name fname;
4229 ret = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4233 /* This needs to handle no-key deletions later on */
4235 trans = __unlink_start_trans(BTRFS_I(dir));
4236 if (IS_ERR(trans)) {
4237 ret = PTR_ERR(trans);
4241 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4244 ret = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4249 if (inode->i_nlink == 0) {
4250 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4256 btrfs_end_transaction(trans);
4257 btrfs_btree_balance_dirty(BTRFS_I(dir)->root->fs_info);
4259 fscrypt_free_filename(&fname);
4263 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4264 struct btrfs_inode *dir, struct dentry *dentry)
4266 struct btrfs_root *root = dir->root;
4267 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4268 struct btrfs_path *path;
4269 struct extent_buffer *leaf;
4270 struct btrfs_dir_item *di;
4271 struct btrfs_key key;
4275 u64 dir_ino = btrfs_ino(dir);
4276 struct fscrypt_name fname;
4278 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
4282 /* This needs to handle no-key deletions later on */
4284 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4285 objectid = inode->root->root_key.objectid;
4286 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4287 objectid = inode->location.objectid;
4290 fscrypt_free_filename(&fname);
4294 path = btrfs_alloc_path();
4300 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4301 &fname.disk_name, -1);
4302 if (IS_ERR_OR_NULL(di)) {
4303 ret = di ? PTR_ERR(di) : -ENOENT;
4307 leaf = path->nodes[0];
4308 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4309 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4310 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4312 btrfs_abort_transaction(trans, ret);
4315 btrfs_release_path(path);
4318 * This is a placeholder inode for a subvolume we didn't have a
4319 * reference to at the time of the snapshot creation. In the meantime
4320 * we could have renamed the real subvol link into our snapshot, so
4321 * depending on btrfs_del_root_ref to return -ENOENT here is incorrect.
4322 * Instead simply lookup the dir_index_item for this entry so we can
4323 * remove it. Otherwise we know we have a ref to the root and we can
4324 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4326 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4327 di = btrfs_search_dir_index_item(root, path, dir_ino, &fname.disk_name);
4328 if (IS_ERR_OR_NULL(di)) {
4333 btrfs_abort_transaction(trans, ret);
4337 leaf = path->nodes[0];
4338 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4340 btrfs_release_path(path);
4342 ret = btrfs_del_root_ref(trans, objectid,
4343 root->root_key.objectid, dir_ino,
4344 &index, &fname.disk_name);
4346 btrfs_abort_transaction(trans, ret);
4351 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4353 btrfs_abort_transaction(trans, ret);
4357 btrfs_i_size_write(dir, dir->vfs_inode.i_size - fname.disk_name.len * 2);
4358 inode_inc_iversion(&dir->vfs_inode);
4359 inode_set_mtime_to_ts(&dir->vfs_inode, inode_set_ctime_current(&dir->vfs_inode));
4360 ret = btrfs_update_inode_fallback(trans, dir);
4362 btrfs_abort_transaction(trans, ret);
4364 btrfs_free_path(path);
4365 fscrypt_free_filename(&fname);
4370 * Helper to check if the subvolume references other subvolumes or if it's
4373 static noinline int may_destroy_subvol(struct btrfs_root *root)
4375 struct btrfs_fs_info *fs_info = root->fs_info;
4376 struct btrfs_path *path;
4377 struct btrfs_dir_item *di;
4378 struct btrfs_key key;
4379 struct fscrypt_str name = FSTR_INIT("default", 7);
4383 path = btrfs_alloc_path();
4387 /* Make sure this root isn't set as the default subvol */
4388 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4389 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4391 if (di && !IS_ERR(di)) {
4392 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4393 if (key.objectid == root->root_key.objectid) {
4396 "deleting default subvolume %llu is not allowed",
4400 btrfs_release_path(path);
4403 key.objectid = root->root_key.objectid;
4404 key.type = BTRFS_ROOT_REF_KEY;
4405 key.offset = (u64)-1;
4407 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4412 * Key with offset -1 found, there would have to exist a root
4413 * with such id, but this is out of valid range.
4420 if (path->slots[0] > 0) {
4422 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4423 if (key.objectid == root->root_key.objectid &&
4424 key.type == BTRFS_ROOT_REF_KEY)
4428 btrfs_free_path(path);
4432 /* Delete all dentries for inodes belonging to the root */
4433 static void btrfs_prune_dentries(struct btrfs_root *root)
4435 struct btrfs_fs_info *fs_info = root->fs_info;
4436 struct rb_node *node;
4437 struct rb_node *prev;
4438 struct btrfs_inode *entry;
4439 struct inode *inode;
4442 if (!BTRFS_FS_ERROR(fs_info))
4443 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4445 spin_lock(&root->inode_lock);
4447 node = root->inode_tree.rb_node;
4451 entry = rb_entry(node, struct btrfs_inode, rb_node);
4453 if (objectid < btrfs_ino(entry))
4454 node = node->rb_left;
4455 else if (objectid > btrfs_ino(entry))
4456 node = node->rb_right;
4462 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4463 if (objectid <= btrfs_ino(entry)) {
4467 prev = rb_next(prev);
4471 entry = rb_entry(node, struct btrfs_inode, rb_node);
4472 objectid = btrfs_ino(entry) + 1;
4473 inode = igrab(&entry->vfs_inode);
4475 spin_unlock(&root->inode_lock);
4476 if (atomic_read(&inode->i_count) > 1)
4477 d_prune_aliases(inode);
4479 * btrfs_drop_inode will have it removed from the inode
4480 * cache when its usage count hits zero.
4484 spin_lock(&root->inode_lock);
4488 if (cond_resched_lock(&root->inode_lock))
4491 node = rb_next(node);
4493 spin_unlock(&root->inode_lock);
4496 int btrfs_delete_subvolume(struct btrfs_inode *dir, struct dentry *dentry)
4498 struct btrfs_root *root = dir->root;
4499 struct btrfs_fs_info *fs_info = root->fs_info;
4500 struct inode *inode = d_inode(dentry);
4501 struct btrfs_root *dest = BTRFS_I(inode)->root;
4502 struct btrfs_trans_handle *trans;
4503 struct btrfs_block_rsv block_rsv;
4505 u64 qgroup_reserved = 0;
4508 down_write(&fs_info->subvol_sem);
4511 * Don't allow to delete a subvolume with send in progress. This is
4512 * inside the inode lock so the error handling that has to drop the bit
4513 * again is not run concurrently.
4515 spin_lock(&dest->root_item_lock);
4516 if (dest->send_in_progress) {
4517 spin_unlock(&dest->root_item_lock);
4519 "attempt to delete subvolume %llu during send",
4520 dest->root_key.objectid);
4524 if (atomic_read(&dest->nr_swapfiles)) {
4525 spin_unlock(&dest->root_item_lock);
4527 "attempt to delete subvolume %llu with active swapfile",
4528 root->root_key.objectid);
4532 root_flags = btrfs_root_flags(&dest->root_item);
4533 btrfs_set_root_flags(&dest->root_item,
4534 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4535 spin_unlock(&dest->root_item_lock);
4537 ret = may_destroy_subvol(dest);
4541 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4543 * One for dir inode,
4544 * two for dir entries,
4545 * two for root ref/backref.
4547 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4550 qgroup_reserved = block_rsv.qgroup_rsv_reserved;
4552 trans = btrfs_start_transaction(root, 0);
4553 if (IS_ERR(trans)) {
4554 ret = PTR_ERR(trans);
4557 ret = btrfs_record_root_in_trans(trans, root);
4559 btrfs_abort_transaction(trans, ret);
4562 btrfs_qgroup_convert_reserved_meta(root, qgroup_reserved);
4563 qgroup_reserved = 0;
4564 trans->block_rsv = &block_rsv;
4565 trans->bytes_reserved = block_rsv.size;
4567 btrfs_record_snapshot_destroy(trans, dir);
4569 ret = btrfs_unlink_subvol(trans, dir, dentry);
4571 btrfs_abort_transaction(trans, ret);
4575 ret = btrfs_record_root_in_trans(trans, dest);
4577 btrfs_abort_transaction(trans, ret);
4581 memset(&dest->root_item.drop_progress, 0,
4582 sizeof(dest->root_item.drop_progress));
4583 btrfs_set_root_drop_level(&dest->root_item, 0);
4584 btrfs_set_root_refs(&dest->root_item, 0);
4586 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4587 ret = btrfs_insert_orphan_item(trans,
4589 dest->root_key.objectid);
4591 btrfs_abort_transaction(trans, ret);
4596 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4597 BTRFS_UUID_KEY_SUBVOL,
4598 dest->root_key.objectid);
4599 if (ret && ret != -ENOENT) {
4600 btrfs_abort_transaction(trans, ret);
4603 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4604 ret = btrfs_uuid_tree_remove(trans,
4605 dest->root_item.received_uuid,
4606 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4607 dest->root_key.objectid);
4608 if (ret && ret != -ENOENT) {
4609 btrfs_abort_transaction(trans, ret);
4614 free_anon_bdev(dest->anon_dev);
4617 trans->block_rsv = NULL;
4618 trans->bytes_reserved = 0;
4619 ret = btrfs_end_transaction(trans);
4620 inode->i_flags |= S_DEAD;
4622 btrfs_block_rsv_release(fs_info, &block_rsv, (u64)-1, NULL);
4623 if (qgroup_reserved)
4624 btrfs_qgroup_free_meta_prealloc(root, qgroup_reserved);
4627 spin_lock(&dest->root_item_lock);
4628 root_flags = btrfs_root_flags(&dest->root_item);
4629 btrfs_set_root_flags(&dest->root_item,
4630 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4631 spin_unlock(&dest->root_item_lock);
4634 up_write(&fs_info->subvol_sem);
4636 d_invalidate(dentry);
4637 btrfs_prune_dentries(dest);
4638 ASSERT(dest->send_in_progress == 0);
4644 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4646 struct inode *inode = d_inode(dentry);
4647 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
4649 struct btrfs_trans_handle *trans;
4650 u64 last_unlink_trans;
4651 struct fscrypt_name fname;
4653 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4655 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) {
4656 if (unlikely(btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))) {
4658 "extent tree v2 doesn't support snapshot deletion yet");
4661 return btrfs_delete_subvolume(BTRFS_I(dir), dentry);
4664 err = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4668 /* This needs to handle no-key deletions later on */
4670 trans = __unlink_start_trans(BTRFS_I(dir));
4671 if (IS_ERR(trans)) {
4672 err = PTR_ERR(trans);
4676 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4677 err = btrfs_unlink_subvol(trans, BTRFS_I(dir), dentry);
4681 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4685 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4687 /* now the directory is empty */
4688 err = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4691 btrfs_i_size_write(BTRFS_I(inode), 0);
4693 * Propagate the last_unlink_trans value of the deleted dir to
4694 * its parent directory. This is to prevent an unrecoverable
4695 * log tree in the case we do something like this:
4697 * 2) create snapshot under dir foo
4698 * 3) delete the snapshot
4701 * 6) fsync foo or some file inside foo
4703 if (last_unlink_trans >= trans->transid)
4704 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4707 btrfs_end_transaction(trans);
4709 btrfs_btree_balance_dirty(fs_info);
4710 fscrypt_free_filename(&fname);
4716 * Read, zero a chunk and write a block.
4718 * @inode - inode that we're zeroing
4719 * @from - the offset to start zeroing
4720 * @len - the length to zero, 0 to zero the entire range respective to the
4722 * @front - zero up to the offset instead of from the offset on
4724 * This will find the block for the "from" offset and cow the block and zero the
4725 * part we want to zero. This is used with truncate and hole punching.
4727 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4730 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4731 struct address_space *mapping = inode->vfs_inode.i_mapping;
4732 struct extent_io_tree *io_tree = &inode->io_tree;
4733 struct btrfs_ordered_extent *ordered;
4734 struct extent_state *cached_state = NULL;
4735 struct extent_changeset *data_reserved = NULL;
4736 bool only_release_metadata = false;
4737 u32 blocksize = fs_info->sectorsize;
4738 pgoff_t index = from >> PAGE_SHIFT;
4739 unsigned offset = from & (blocksize - 1);
4740 struct folio *folio;
4741 gfp_t mask = btrfs_alloc_write_mask(mapping);
4742 size_t write_bytes = blocksize;
4747 if (IS_ALIGNED(offset, blocksize) &&
4748 (!len || IS_ALIGNED(len, blocksize)))
4751 block_start = round_down(from, blocksize);
4752 block_end = block_start + blocksize - 1;
4754 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4757 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes, false) > 0) {
4758 /* For nocow case, no need to reserve data space */
4759 only_release_metadata = true;
4764 ret = btrfs_delalloc_reserve_metadata(inode, blocksize, blocksize, false);
4766 if (!only_release_metadata)
4767 btrfs_free_reserved_data_space(inode, data_reserved,
4768 block_start, blocksize);
4772 folio = __filemap_get_folio(mapping, index,
4773 FGP_LOCK | FGP_ACCESSED | FGP_CREAT, mask);
4774 if (IS_ERR(folio)) {
4775 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4777 btrfs_delalloc_release_extents(inode, blocksize);
4782 if (!folio_test_uptodate(folio)) {
4783 ret = btrfs_read_folio(NULL, folio);
4785 if (folio->mapping != mapping) {
4786 folio_unlock(folio);
4790 if (!folio_test_uptodate(folio)) {
4797 * We unlock the page after the io is completed and then re-lock it
4798 * above. release_folio() could have come in between that and cleared
4799 * folio private, but left the page in the mapping. Set the page mapped
4800 * here to make sure it's properly set for the subpage stuff.
4802 ret = set_folio_extent_mapped(folio);
4806 folio_wait_writeback(folio);
4808 lock_extent(io_tree, block_start, block_end, &cached_state);
4810 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4812 unlock_extent(io_tree, block_start, block_end, &cached_state);
4813 folio_unlock(folio);
4815 btrfs_start_ordered_extent(ordered);
4816 btrfs_put_ordered_extent(ordered);
4820 clear_extent_bit(&inode->io_tree, block_start, block_end,
4821 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4824 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4827 unlock_extent(io_tree, block_start, block_end, &cached_state);
4831 if (offset != blocksize) {
4833 len = blocksize - offset;
4835 folio_zero_range(folio, block_start - folio_pos(folio),
4838 folio_zero_range(folio,
4839 (block_start - folio_pos(folio)) + offset,
4842 btrfs_folio_clear_checked(fs_info, folio, block_start,
4843 block_end + 1 - block_start);
4844 btrfs_folio_set_dirty(fs_info, folio, block_start,
4845 block_end + 1 - block_start);
4846 unlock_extent(io_tree, block_start, block_end, &cached_state);
4848 if (only_release_metadata)
4849 set_extent_bit(&inode->io_tree, block_start, block_end,
4850 EXTENT_NORESERVE, NULL);
4854 if (only_release_metadata)
4855 btrfs_delalloc_release_metadata(inode, blocksize, true);
4857 btrfs_delalloc_release_space(inode, data_reserved,
4858 block_start, blocksize, true);
4860 btrfs_delalloc_release_extents(inode, blocksize);
4861 folio_unlock(folio);
4864 if (only_release_metadata)
4865 btrfs_check_nocow_unlock(inode);
4866 extent_changeset_free(data_reserved);
4870 static int maybe_insert_hole(struct btrfs_inode *inode, u64 offset, u64 len)
4872 struct btrfs_root *root = inode->root;
4873 struct btrfs_fs_info *fs_info = root->fs_info;
4874 struct btrfs_trans_handle *trans;
4875 struct btrfs_drop_extents_args drop_args = { 0 };
4879 * If NO_HOLES is enabled, we don't need to do anything.
4880 * Later, up in the call chain, either btrfs_set_inode_last_sub_trans()
4881 * or btrfs_update_inode() will be called, which guarantee that the next
4882 * fsync will know this inode was changed and needs to be logged.
4884 if (btrfs_fs_incompat(fs_info, NO_HOLES))
4888 * 1 - for the one we're dropping
4889 * 1 - for the one we're adding
4890 * 1 - for updating the inode.
4892 trans = btrfs_start_transaction(root, 3);
4894 return PTR_ERR(trans);
4896 drop_args.start = offset;
4897 drop_args.end = offset + len;
4898 drop_args.drop_cache = true;
4900 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
4902 btrfs_abort_transaction(trans, ret);
4903 btrfs_end_transaction(trans);
4907 ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset, len);
4909 btrfs_abort_transaction(trans, ret);
4911 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
4912 btrfs_update_inode(trans, inode);
4914 btrfs_end_transaction(trans);
4919 * This function puts in dummy file extents for the area we're creating a hole
4920 * for. So if we are truncating this file to a larger size we need to insert
4921 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4922 * the range between oldsize and size
4924 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
4926 struct btrfs_root *root = inode->root;
4927 struct btrfs_fs_info *fs_info = root->fs_info;
4928 struct extent_io_tree *io_tree = &inode->io_tree;
4929 struct extent_map *em = NULL;
4930 struct extent_state *cached_state = NULL;
4931 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4932 u64 block_end = ALIGN(size, fs_info->sectorsize);
4939 * If our size started in the middle of a block we need to zero out the
4940 * rest of the block before we expand the i_size, otherwise we could
4941 * expose stale data.
4943 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4947 if (size <= hole_start)
4950 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
4952 cur_offset = hole_start;
4954 em = btrfs_get_extent(inode, NULL, cur_offset, block_end - cur_offset);
4960 last_byte = min(extent_map_end(em), block_end);
4961 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4962 hole_size = last_byte - cur_offset;
4964 if (!(em->flags & EXTENT_FLAG_PREALLOC)) {
4965 struct extent_map *hole_em;
4967 err = maybe_insert_hole(inode, cur_offset, hole_size);
4971 err = btrfs_inode_set_file_extent_range(inode,
4972 cur_offset, hole_size);
4976 hole_em = alloc_extent_map();
4978 btrfs_drop_extent_map_range(inode, cur_offset,
4979 cur_offset + hole_size - 1,
4981 btrfs_set_inode_full_sync(inode);
4984 hole_em->start = cur_offset;
4985 hole_em->len = hole_size;
4986 hole_em->orig_start = cur_offset;
4988 hole_em->block_start = EXTENT_MAP_HOLE;
4989 hole_em->block_len = 0;
4990 hole_em->orig_block_len = 0;
4991 hole_em->ram_bytes = hole_size;
4992 hole_em->generation = btrfs_get_fs_generation(fs_info);
4994 err = btrfs_replace_extent_map_range(inode, hole_em, true);
4995 free_extent_map(hole_em);
4997 err = btrfs_inode_set_file_extent_range(inode,
4998 cur_offset, hole_size);
5003 free_extent_map(em);
5005 cur_offset = last_byte;
5006 if (cur_offset >= block_end)
5009 free_extent_map(em);
5010 unlock_extent(io_tree, hole_start, block_end - 1, &cached_state);
5014 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5016 struct btrfs_root *root = BTRFS_I(inode)->root;
5017 struct btrfs_trans_handle *trans;
5018 loff_t oldsize = i_size_read(inode);
5019 loff_t newsize = attr->ia_size;
5020 int mask = attr->ia_valid;
5024 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5025 * special case where we need to update the times despite not having
5026 * these flags set. For all other operations the VFS set these flags
5027 * explicitly if it wants a timestamp update.
5029 if (newsize != oldsize) {
5030 inode_inc_iversion(inode);
5031 if (!(mask & (ATTR_CTIME | ATTR_MTIME))) {
5032 inode_set_mtime_to_ts(inode,
5033 inode_set_ctime_current(inode));
5037 if (newsize > oldsize) {
5039 * Don't do an expanding truncate while snapshotting is ongoing.
5040 * This is to ensure the snapshot captures a fully consistent
5041 * state of this file - if the snapshot captures this expanding
5042 * truncation, it must capture all writes that happened before
5045 btrfs_drew_write_lock(&root->snapshot_lock);
5046 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
5048 btrfs_drew_write_unlock(&root->snapshot_lock);
5052 trans = btrfs_start_transaction(root, 1);
5053 if (IS_ERR(trans)) {
5054 btrfs_drew_write_unlock(&root->snapshot_lock);
5055 return PTR_ERR(trans);
5058 i_size_write(inode, newsize);
5059 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
5060 pagecache_isize_extended(inode, oldsize, newsize);
5061 ret = btrfs_update_inode(trans, BTRFS_I(inode));
5062 btrfs_drew_write_unlock(&root->snapshot_lock);
5063 btrfs_end_transaction(trans);
5065 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
5067 if (btrfs_is_zoned(fs_info)) {
5068 ret = btrfs_wait_ordered_range(inode,
5069 ALIGN(newsize, fs_info->sectorsize),
5076 * We're truncating a file that used to have good data down to
5077 * zero. Make sure any new writes to the file get on disk
5081 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5082 &BTRFS_I(inode)->runtime_flags);
5084 truncate_setsize(inode, newsize);
5086 inode_dio_wait(inode);
5088 ret = btrfs_truncate(BTRFS_I(inode), newsize == oldsize);
5089 if (ret && inode->i_nlink) {
5093 * Truncate failed, so fix up the in-memory size. We
5094 * adjusted disk_i_size down as we removed extents, so
5095 * wait for disk_i_size to be stable and then update the
5096 * in-memory size to match.
5098 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5101 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5108 static int btrfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
5111 struct inode *inode = d_inode(dentry);
5112 struct btrfs_root *root = BTRFS_I(inode)->root;
5115 if (btrfs_root_readonly(root))
5118 err = setattr_prepare(idmap, dentry, attr);
5122 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5123 err = btrfs_setsize(inode, attr);
5128 if (attr->ia_valid) {
5129 setattr_copy(idmap, inode, attr);
5130 inode_inc_iversion(inode);
5131 err = btrfs_dirty_inode(BTRFS_I(inode));
5133 if (!err && attr->ia_valid & ATTR_MODE)
5134 err = posix_acl_chmod(idmap, dentry, inode->i_mode);
5141 * While truncating the inode pages during eviction, we get the VFS
5142 * calling btrfs_invalidate_folio() against each folio of the inode. This
5143 * is slow because the calls to btrfs_invalidate_folio() result in a
5144 * huge amount of calls to lock_extent() and clear_extent_bit(),
5145 * which keep merging and splitting extent_state structures over and over,
5146 * wasting lots of time.
5148 * Therefore if the inode is being evicted, let btrfs_invalidate_folio()
5149 * skip all those expensive operations on a per folio basis and do only
5150 * the ordered io finishing, while we release here the extent_map and
5151 * extent_state structures, without the excessive merging and splitting.
5153 static void evict_inode_truncate_pages(struct inode *inode)
5155 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5156 struct rb_node *node;
5158 ASSERT(inode->i_state & I_FREEING);
5159 truncate_inode_pages_final(&inode->i_data);
5161 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
5164 * Keep looping until we have no more ranges in the io tree.
5165 * We can have ongoing bios started by readahead that have
5166 * their endio callback (extent_io.c:end_bio_extent_readpage)
5167 * still in progress (unlocked the pages in the bio but did not yet
5168 * unlocked the ranges in the io tree). Therefore this means some
5169 * ranges can still be locked and eviction started because before
5170 * submitting those bios, which are executed by a separate task (work
5171 * queue kthread), inode references (inode->i_count) were not taken
5172 * (which would be dropped in the end io callback of each bio).
5173 * Therefore here we effectively end up waiting for those bios and
5174 * anyone else holding locked ranges without having bumped the inode's
5175 * reference count - if we don't do it, when they access the inode's
5176 * io_tree to unlock a range it may be too late, leading to an
5177 * use-after-free issue.
5179 spin_lock(&io_tree->lock);
5180 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5181 struct extent_state *state;
5182 struct extent_state *cached_state = NULL;
5185 unsigned state_flags;
5187 node = rb_first(&io_tree->state);
5188 state = rb_entry(node, struct extent_state, rb_node);
5189 start = state->start;
5191 state_flags = state->state;
5192 spin_unlock(&io_tree->lock);
5194 lock_extent(io_tree, start, end, &cached_state);
5197 * If still has DELALLOC flag, the extent didn't reach disk,
5198 * and its reserved space won't be freed by delayed_ref.
5199 * So we need to free its reserved space here.
5200 * (Refer to comment in btrfs_invalidate_folio, case 2)
5202 * Note, end is the bytenr of last byte, so we need + 1 here.
5204 if (state_flags & EXTENT_DELALLOC)
5205 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5206 end - start + 1, NULL);
5208 clear_extent_bit(io_tree, start, end,
5209 EXTENT_CLEAR_ALL_BITS | EXTENT_DO_ACCOUNTING,
5213 spin_lock(&io_tree->lock);
5215 spin_unlock(&io_tree->lock);
5218 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5219 struct btrfs_block_rsv *rsv)
5221 struct btrfs_fs_info *fs_info = root->fs_info;
5222 struct btrfs_trans_handle *trans;
5223 u64 delayed_refs_extra = btrfs_calc_delayed_ref_bytes(fs_info, 1);
5227 * Eviction should be taking place at some place safe because of our
5228 * delayed iputs. However the normal flushing code will run delayed
5229 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5231 * We reserve the delayed_refs_extra here again because we can't use
5232 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5233 * above. We reserve our extra bit here because we generate a ton of
5234 * delayed refs activity by truncating.
5236 * BTRFS_RESERVE_FLUSH_EVICT will steal from the global_rsv if it can,
5237 * if we fail to make this reservation we can re-try without the
5238 * delayed_refs_extra so we can make some forward progress.
5240 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size + delayed_refs_extra,
5241 BTRFS_RESERVE_FLUSH_EVICT);
5243 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size,
5244 BTRFS_RESERVE_FLUSH_EVICT);
5247 "could not allocate space for delete; will truncate on mount");
5248 return ERR_PTR(-ENOSPC);
5250 delayed_refs_extra = 0;
5253 trans = btrfs_join_transaction(root);
5257 if (delayed_refs_extra) {
5258 trans->block_rsv = &fs_info->trans_block_rsv;
5259 trans->bytes_reserved = delayed_refs_extra;
5260 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5261 delayed_refs_extra, true);
5266 void btrfs_evict_inode(struct inode *inode)
5268 struct btrfs_fs_info *fs_info;
5269 struct btrfs_trans_handle *trans;
5270 struct btrfs_root *root = BTRFS_I(inode)->root;
5271 struct btrfs_block_rsv *rsv = NULL;
5274 trace_btrfs_inode_evict(inode);
5277 fsverity_cleanup_inode(inode);
5282 fs_info = inode_to_fs_info(inode);
5283 evict_inode_truncate_pages(inode);
5285 if (inode->i_nlink &&
5286 ((btrfs_root_refs(&root->root_item) != 0 &&
5287 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5288 btrfs_is_free_space_inode(BTRFS_I(inode))))
5291 if (is_bad_inode(inode))
5294 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5297 if (inode->i_nlink > 0) {
5298 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5299 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5304 * This makes sure the inode item in tree is uptodate and the space for
5305 * the inode update is released.
5307 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5312 * This drops any pending insert or delete operations we have for this
5313 * inode. We could have a delayed dir index deletion queued up, but
5314 * we're removing the inode completely so that'll be taken care of in
5317 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
5319 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5322 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5323 rsv->failfast = true;
5325 btrfs_i_size_write(BTRFS_I(inode), 0);
5328 struct btrfs_truncate_control control = {
5329 .inode = BTRFS_I(inode),
5330 .ino = btrfs_ino(BTRFS_I(inode)),
5335 trans = evict_refill_and_join(root, rsv);
5339 trans->block_rsv = rsv;
5341 ret = btrfs_truncate_inode_items(trans, root, &control);
5342 trans->block_rsv = &fs_info->trans_block_rsv;
5343 btrfs_end_transaction(trans);
5345 * We have not added new delayed items for our inode after we
5346 * have flushed its delayed items, so no need to throttle on
5347 * delayed items. However we have modified extent buffers.
5349 btrfs_btree_balance_dirty_nodelay(fs_info);
5350 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5357 * Errors here aren't a big deal, it just means we leave orphan items in
5358 * the tree. They will be cleaned up on the next mount. If the inode
5359 * number gets reused, cleanup deletes the orphan item without doing
5360 * anything, and unlink reuses the existing orphan item.
5362 * If it turns out that we are dropping too many of these, we might want
5363 * to add a mechanism for retrying these after a commit.
5365 trans = evict_refill_and_join(root, rsv);
5366 if (!IS_ERR(trans)) {
5367 trans->block_rsv = rsv;
5368 btrfs_orphan_del(trans, BTRFS_I(inode));
5369 trans->block_rsv = &fs_info->trans_block_rsv;
5370 btrfs_end_transaction(trans);
5374 btrfs_free_block_rsv(fs_info, rsv);
5376 * If we didn't successfully delete, the orphan item will still be in
5377 * the tree and we'll retry on the next mount. Again, we might also want
5378 * to retry these periodically in the future.
5380 btrfs_remove_delayed_node(BTRFS_I(inode));
5381 fsverity_cleanup_inode(inode);
5386 * Return the key found in the dir entry in the location pointer, fill @type
5387 * with BTRFS_FT_*, and return 0.
5389 * If no dir entries were found, returns -ENOENT.
5390 * If found a corrupted location in dir entry, returns -EUCLEAN.
5392 static int btrfs_inode_by_name(struct btrfs_inode *dir, struct dentry *dentry,
5393 struct btrfs_key *location, u8 *type)
5395 struct btrfs_dir_item *di;
5396 struct btrfs_path *path;
5397 struct btrfs_root *root = dir->root;
5399 struct fscrypt_name fname;
5401 path = btrfs_alloc_path();
5405 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
5409 * fscrypt_setup_filename() should never return a positive value, but
5410 * gcc on sparc/parisc thinks it can, so assert that doesn't happen.
5414 /* This needs to handle no-key deletions later on */
5416 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir),
5417 &fname.disk_name, 0);
5418 if (IS_ERR_OR_NULL(di)) {
5419 ret = di ? PTR_ERR(di) : -ENOENT;
5423 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5424 if (location->type != BTRFS_INODE_ITEM_KEY &&
5425 location->type != BTRFS_ROOT_ITEM_KEY) {
5427 btrfs_warn(root->fs_info,
5428 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5429 __func__, fname.disk_name.name, btrfs_ino(dir),
5430 location->objectid, location->type, location->offset);
5433 *type = btrfs_dir_ftype(path->nodes[0], di);
5435 fscrypt_free_filename(&fname);
5436 btrfs_free_path(path);
5441 * when we hit a tree root in a directory, the btrfs part of the inode
5442 * needs to be changed to reflect the root directory of the tree root. This
5443 * is kind of like crossing a mount point.
5445 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5446 struct btrfs_inode *dir,
5447 struct dentry *dentry,
5448 struct btrfs_key *location,
5449 struct btrfs_root **sub_root)
5451 struct btrfs_path *path;
5452 struct btrfs_root *new_root;
5453 struct btrfs_root_ref *ref;
5454 struct extent_buffer *leaf;
5455 struct btrfs_key key;
5458 struct fscrypt_name fname;
5460 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 0, &fname);
5464 path = btrfs_alloc_path();
5471 key.objectid = dir->root->root_key.objectid;
5472 key.type = BTRFS_ROOT_REF_KEY;
5473 key.offset = location->objectid;
5475 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5482 leaf = path->nodes[0];
5483 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5484 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5485 btrfs_root_ref_name_len(leaf, ref) != fname.disk_name.len)
5488 ret = memcmp_extent_buffer(leaf, fname.disk_name.name,
5489 (unsigned long)(ref + 1), fname.disk_name.len);
5493 btrfs_release_path(path);
5495 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5496 if (IS_ERR(new_root)) {
5497 err = PTR_ERR(new_root);
5501 *sub_root = new_root;
5502 location->objectid = btrfs_root_dirid(&new_root->root_item);
5503 location->type = BTRFS_INODE_ITEM_KEY;
5504 location->offset = 0;
5507 btrfs_free_path(path);
5508 fscrypt_free_filename(&fname);
5512 static void inode_tree_add(struct btrfs_inode *inode)
5514 struct btrfs_root *root = inode->root;
5515 struct btrfs_inode *entry;
5517 struct rb_node *parent;
5518 struct rb_node *new = &inode->rb_node;
5519 u64 ino = btrfs_ino(inode);
5521 if (inode_unhashed(&inode->vfs_inode))
5524 spin_lock(&root->inode_lock);
5525 p = &root->inode_tree.rb_node;
5528 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5530 if (ino < btrfs_ino(entry))
5531 p = &parent->rb_left;
5532 else if (ino > btrfs_ino(entry))
5533 p = &parent->rb_right;
5535 WARN_ON(!(entry->vfs_inode.i_state &
5536 (I_WILL_FREE | I_FREEING)));
5537 rb_replace_node(parent, new, &root->inode_tree);
5538 RB_CLEAR_NODE(parent);
5539 spin_unlock(&root->inode_lock);
5543 rb_link_node(new, parent, p);
5544 rb_insert_color(new, &root->inode_tree);
5545 spin_unlock(&root->inode_lock);
5548 static void inode_tree_del(struct btrfs_inode *inode)
5550 struct btrfs_root *root = inode->root;
5553 spin_lock(&root->inode_lock);
5554 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5555 rb_erase(&inode->rb_node, &root->inode_tree);
5556 RB_CLEAR_NODE(&inode->rb_node);
5557 empty = RB_EMPTY_ROOT(&root->inode_tree);
5559 spin_unlock(&root->inode_lock);
5561 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5562 spin_lock(&root->inode_lock);
5563 empty = RB_EMPTY_ROOT(&root->inode_tree);
5564 spin_unlock(&root->inode_lock);
5566 btrfs_add_dead_root(root);
5571 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5573 struct btrfs_iget_args *args = p;
5575 inode->i_ino = args->ino;
5576 BTRFS_I(inode)->location.objectid = args->ino;
5577 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5578 BTRFS_I(inode)->location.offset = 0;
5579 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5581 if (args->root && args->root == args->root->fs_info->tree_root &&
5582 args->ino != BTRFS_BTREE_INODE_OBJECTID)
5583 set_bit(BTRFS_INODE_FREE_SPACE_INODE,
5584 &BTRFS_I(inode)->runtime_flags);
5588 static int btrfs_find_actor(struct inode *inode, void *opaque)
5590 struct btrfs_iget_args *args = opaque;
5592 return args->ino == BTRFS_I(inode)->location.objectid &&
5593 args->root == BTRFS_I(inode)->root;
5596 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5597 struct btrfs_root *root)
5599 struct inode *inode;
5600 struct btrfs_iget_args args;
5601 unsigned long hashval = btrfs_inode_hash(ino, root);
5606 inode = iget5_locked(s, hashval, btrfs_find_actor,
5607 btrfs_init_locked_inode,
5613 * Get an inode object given its inode number and corresponding root.
5614 * Path can be preallocated to prevent recursing back to iget through
5615 * allocator. NULL is also valid but may require an additional allocation
5618 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5619 struct btrfs_root *root, struct btrfs_path *path)
5621 struct inode *inode;
5623 inode = btrfs_iget_locked(s, ino, root);
5625 return ERR_PTR(-ENOMEM);
5627 if (inode->i_state & I_NEW) {
5630 ret = btrfs_read_locked_inode(inode, path);
5632 inode_tree_add(BTRFS_I(inode));
5633 unlock_new_inode(inode);
5637 * ret > 0 can come from btrfs_search_slot called by
5638 * btrfs_read_locked_inode, this means the inode item
5643 inode = ERR_PTR(ret);
5650 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5652 return btrfs_iget_path(s, ino, root, NULL);
5655 static struct inode *new_simple_dir(struct inode *dir,
5656 struct btrfs_key *key,
5657 struct btrfs_root *root)
5659 struct timespec64 ts;
5660 struct inode *inode = new_inode(dir->i_sb);
5663 return ERR_PTR(-ENOMEM);
5665 BTRFS_I(inode)->root = btrfs_grab_root(root);
5666 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5667 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5669 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5671 * We only need lookup, the rest is read-only and there's no inode
5672 * associated with the dentry
5674 inode->i_op = &simple_dir_inode_operations;
5675 inode->i_opflags &= ~IOP_XATTR;
5676 inode->i_fop = &simple_dir_operations;
5677 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5679 ts = inode_set_ctime_current(inode);
5680 inode_set_mtime_to_ts(inode, ts);
5681 inode_set_atime_to_ts(inode, inode_get_atime(dir));
5682 BTRFS_I(inode)->i_otime_sec = ts.tv_sec;
5683 BTRFS_I(inode)->i_otime_nsec = ts.tv_nsec;
5685 inode->i_uid = dir->i_uid;
5686 inode->i_gid = dir->i_gid;
5691 static_assert(BTRFS_FT_UNKNOWN == FT_UNKNOWN);
5692 static_assert(BTRFS_FT_REG_FILE == FT_REG_FILE);
5693 static_assert(BTRFS_FT_DIR == FT_DIR);
5694 static_assert(BTRFS_FT_CHRDEV == FT_CHRDEV);
5695 static_assert(BTRFS_FT_BLKDEV == FT_BLKDEV);
5696 static_assert(BTRFS_FT_FIFO == FT_FIFO);
5697 static_assert(BTRFS_FT_SOCK == FT_SOCK);
5698 static_assert(BTRFS_FT_SYMLINK == FT_SYMLINK);
5700 static inline u8 btrfs_inode_type(struct inode *inode)
5702 return fs_umode_to_ftype(inode->i_mode);
5705 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5707 struct btrfs_fs_info *fs_info = inode_to_fs_info(dir);
5708 struct inode *inode;
5709 struct btrfs_root *root = BTRFS_I(dir)->root;
5710 struct btrfs_root *sub_root = root;
5711 struct btrfs_key location;
5715 if (dentry->d_name.len > BTRFS_NAME_LEN)
5716 return ERR_PTR(-ENAMETOOLONG);
5718 ret = btrfs_inode_by_name(BTRFS_I(dir), dentry, &location, &di_type);
5720 return ERR_PTR(ret);
5722 if (location.type == BTRFS_INODE_ITEM_KEY) {
5723 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5727 /* Do extra check against inode mode with di_type */
5728 if (btrfs_inode_type(inode) != di_type) {
5730 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5731 inode->i_mode, btrfs_inode_type(inode),
5734 return ERR_PTR(-EUCLEAN);
5739 ret = fixup_tree_root_location(fs_info, BTRFS_I(dir), dentry,
5740 &location, &sub_root);
5743 inode = ERR_PTR(ret);
5745 inode = new_simple_dir(dir, &location, root);
5747 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5748 btrfs_put_root(sub_root);
5753 down_read(&fs_info->cleanup_work_sem);
5754 if (!sb_rdonly(inode->i_sb))
5755 ret = btrfs_orphan_cleanup(sub_root);
5756 up_read(&fs_info->cleanup_work_sem);
5759 inode = ERR_PTR(ret);
5766 static int btrfs_dentry_delete(const struct dentry *dentry)
5768 struct btrfs_root *root;
5769 struct inode *inode = d_inode(dentry);
5771 if (!inode && !IS_ROOT(dentry))
5772 inode = d_inode(dentry->d_parent);
5775 root = BTRFS_I(inode)->root;
5776 if (btrfs_root_refs(&root->root_item) == 0)
5779 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5785 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5788 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5790 if (inode == ERR_PTR(-ENOENT))
5792 return d_splice_alias(inode, dentry);
5796 * Find the highest existing sequence number in a directory and then set the
5797 * in-memory index_cnt variable to the first free sequence number.
5799 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
5801 struct btrfs_root *root = inode->root;
5802 struct btrfs_key key, found_key;
5803 struct btrfs_path *path;
5804 struct extent_buffer *leaf;
5807 key.objectid = btrfs_ino(inode);
5808 key.type = BTRFS_DIR_INDEX_KEY;
5809 key.offset = (u64)-1;
5811 path = btrfs_alloc_path();
5815 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5818 /* FIXME: we should be able to handle this */
5823 if (path->slots[0] == 0) {
5824 inode->index_cnt = BTRFS_DIR_START_INDEX;
5830 leaf = path->nodes[0];
5831 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5833 if (found_key.objectid != btrfs_ino(inode) ||
5834 found_key.type != BTRFS_DIR_INDEX_KEY) {
5835 inode->index_cnt = BTRFS_DIR_START_INDEX;
5839 inode->index_cnt = found_key.offset + 1;
5841 btrfs_free_path(path);
5845 static int btrfs_get_dir_last_index(struct btrfs_inode *dir, u64 *index)
5849 btrfs_inode_lock(dir, 0);
5850 if (dir->index_cnt == (u64)-1) {
5851 ret = btrfs_inode_delayed_dir_index_count(dir);
5853 ret = btrfs_set_inode_index_count(dir);
5859 /* index_cnt is the index number of next new entry, so decrement it. */
5860 *index = dir->index_cnt - 1;
5862 btrfs_inode_unlock(dir, 0);
5868 * All this infrastructure exists because dir_emit can fault, and we are holding
5869 * the tree lock when doing readdir. For now just allocate a buffer and copy
5870 * our information into that, and then dir_emit from the buffer. This is
5871 * similar to what NFS does, only we don't keep the buffer around in pagecache
5872 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5873 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5876 static int btrfs_opendir(struct inode *inode, struct file *file)
5878 struct btrfs_file_private *private;
5882 ret = btrfs_get_dir_last_index(BTRFS_I(inode), &last_index);
5886 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5889 private->last_index = last_index;
5890 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5891 if (!private->filldir_buf) {
5895 file->private_data = private;
5899 static loff_t btrfs_dir_llseek(struct file *file, loff_t offset, int whence)
5901 struct btrfs_file_private *private = file->private_data;
5904 ret = btrfs_get_dir_last_index(BTRFS_I(file_inode(file)),
5905 &private->last_index);
5909 return generic_file_llseek(file, offset, whence);
5919 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5922 struct dir_entry *entry = addr;
5923 char *name = (char *)(entry + 1);
5925 ctx->pos = get_unaligned(&entry->offset);
5926 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5927 get_unaligned(&entry->ino),
5928 get_unaligned(&entry->type)))
5930 addr += sizeof(struct dir_entry) +
5931 get_unaligned(&entry->name_len);
5937 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5939 struct inode *inode = file_inode(file);
5940 struct btrfs_root *root = BTRFS_I(inode)->root;
5941 struct btrfs_file_private *private = file->private_data;
5942 struct btrfs_dir_item *di;
5943 struct btrfs_key key;
5944 struct btrfs_key found_key;
5945 struct btrfs_path *path;
5947 LIST_HEAD(ins_list);
5948 LIST_HEAD(del_list);
5955 struct btrfs_key location;
5957 if (!dir_emit_dots(file, ctx))
5960 path = btrfs_alloc_path();
5964 addr = private->filldir_buf;
5965 path->reada = READA_FORWARD;
5967 put = btrfs_readdir_get_delayed_items(inode, private->last_index,
5968 &ins_list, &del_list);
5971 key.type = BTRFS_DIR_INDEX_KEY;
5972 key.offset = ctx->pos;
5973 key.objectid = btrfs_ino(BTRFS_I(inode));
5975 btrfs_for_each_slot(root, &key, &found_key, path, ret) {
5976 struct dir_entry *entry;
5977 struct extent_buffer *leaf = path->nodes[0];
5980 if (found_key.objectid != key.objectid)
5982 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5984 if (found_key.offset < ctx->pos)
5986 if (found_key.offset > private->last_index)
5988 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5990 di = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
5991 name_len = btrfs_dir_name_len(leaf, di);
5992 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5994 btrfs_release_path(path);
5995 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5998 addr = private->filldir_buf;
6004 ftype = btrfs_dir_flags_to_ftype(btrfs_dir_flags(leaf, di));
6006 name_ptr = (char *)(entry + 1);
6007 read_extent_buffer(leaf, name_ptr,
6008 (unsigned long)(di + 1), name_len);
6009 put_unaligned(name_len, &entry->name_len);
6010 put_unaligned(fs_ftype_to_dtype(ftype), &entry->type);
6011 btrfs_dir_item_key_to_cpu(leaf, di, &location);
6012 put_unaligned(location.objectid, &entry->ino);
6013 put_unaligned(found_key.offset, &entry->offset);
6015 addr += sizeof(struct dir_entry) + name_len;
6016 total_len += sizeof(struct dir_entry) + name_len;
6018 /* Catch error encountered during iteration */
6022 btrfs_release_path(path);
6024 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6028 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
6033 * Stop new entries from being returned after we return the last
6036 * New directory entries are assigned a strictly increasing
6037 * offset. This means that new entries created during readdir
6038 * are *guaranteed* to be seen in the future by that readdir.
6039 * This has broken buggy programs which operate on names as
6040 * they're returned by readdir. Until we re-use freed offsets
6041 * we have this hack to stop new entries from being returned
6042 * under the assumption that they'll never reach this huge
6045 * This is being careful not to overflow 32bit loff_t unless the
6046 * last entry requires it because doing so has broken 32bit apps
6049 if (ctx->pos >= INT_MAX)
6050 ctx->pos = LLONG_MAX;
6057 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6058 btrfs_free_path(path);
6063 * This is somewhat expensive, updating the tree every time the
6064 * inode changes. But, it is most likely to find the inode in cache.
6065 * FIXME, needs more benchmarking...there are no reasons other than performance
6066 * to keep or drop this code.
6068 static int btrfs_dirty_inode(struct btrfs_inode *inode)
6070 struct btrfs_root *root = inode->root;
6071 struct btrfs_fs_info *fs_info = root->fs_info;
6072 struct btrfs_trans_handle *trans;
6075 if (test_bit(BTRFS_INODE_DUMMY, &inode->runtime_flags))
6078 trans = btrfs_join_transaction(root);
6080 return PTR_ERR(trans);
6082 ret = btrfs_update_inode(trans, inode);
6083 if (ret == -ENOSPC || ret == -EDQUOT) {
6084 /* whoops, lets try again with the full transaction */
6085 btrfs_end_transaction(trans);
6086 trans = btrfs_start_transaction(root, 1);
6088 return PTR_ERR(trans);
6090 ret = btrfs_update_inode(trans, inode);
6092 btrfs_end_transaction(trans);
6093 if (inode->delayed_node)
6094 btrfs_balance_delayed_items(fs_info);
6100 * This is a copy of file_update_time. We need this so we can return error on
6101 * ENOSPC for updating the inode in the case of file write and mmap writes.
6103 static int btrfs_update_time(struct inode *inode, int flags)
6105 struct btrfs_root *root = BTRFS_I(inode)->root;
6108 if (btrfs_root_readonly(root))
6111 dirty = inode_update_timestamps(inode, flags);
6112 return dirty ? btrfs_dirty_inode(BTRFS_I(inode)) : 0;
6116 * helper to find a free sequence number in a given directory. This current
6117 * code is very simple, later versions will do smarter things in the btree
6119 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6123 if (dir->index_cnt == (u64)-1) {
6124 ret = btrfs_inode_delayed_dir_index_count(dir);
6126 ret = btrfs_set_inode_index_count(dir);
6132 *index = dir->index_cnt;
6138 static int btrfs_insert_inode_locked(struct inode *inode)
6140 struct btrfs_iget_args args;
6142 args.ino = BTRFS_I(inode)->location.objectid;
6143 args.root = BTRFS_I(inode)->root;
6145 return insert_inode_locked4(inode,
6146 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6147 btrfs_find_actor, &args);
6150 int btrfs_new_inode_prepare(struct btrfs_new_inode_args *args,
6151 unsigned int *trans_num_items)
6153 struct inode *dir = args->dir;
6154 struct inode *inode = args->inode;
6157 if (!args->orphan) {
6158 ret = fscrypt_setup_filename(dir, &args->dentry->d_name, 0,
6164 ret = posix_acl_create(dir, &inode->i_mode, &args->default_acl, &args->acl);
6166 fscrypt_free_filename(&args->fname);
6170 /* 1 to add inode item */
6171 *trans_num_items = 1;
6172 /* 1 to add compression property */
6173 if (BTRFS_I(dir)->prop_compress)
6174 (*trans_num_items)++;
6175 /* 1 to add default ACL xattr */
6176 if (args->default_acl)
6177 (*trans_num_items)++;
6178 /* 1 to add access ACL xattr */
6180 (*trans_num_items)++;
6181 #ifdef CONFIG_SECURITY
6182 /* 1 to add LSM xattr */
6183 if (dir->i_security)
6184 (*trans_num_items)++;
6187 /* 1 to add orphan item */
6188 (*trans_num_items)++;
6192 * 1 to add dir index
6193 * 1 to update parent inode item
6195 * No need for 1 unit for the inode ref item because it is
6196 * inserted in a batch together with the inode item at
6197 * btrfs_create_new_inode().
6199 *trans_num_items += 3;
6204 void btrfs_new_inode_args_destroy(struct btrfs_new_inode_args *args)
6206 posix_acl_release(args->acl);
6207 posix_acl_release(args->default_acl);
6208 fscrypt_free_filename(&args->fname);
6212 * Inherit flags from the parent inode.
6214 * Currently only the compression flags and the cow flags are inherited.
6216 static void btrfs_inherit_iflags(struct btrfs_inode *inode, struct btrfs_inode *dir)
6222 if (flags & BTRFS_INODE_NOCOMPRESS) {
6223 inode->flags &= ~BTRFS_INODE_COMPRESS;
6224 inode->flags |= BTRFS_INODE_NOCOMPRESS;
6225 } else if (flags & BTRFS_INODE_COMPRESS) {
6226 inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
6227 inode->flags |= BTRFS_INODE_COMPRESS;
6230 if (flags & BTRFS_INODE_NODATACOW) {
6231 inode->flags |= BTRFS_INODE_NODATACOW;
6232 if (S_ISREG(inode->vfs_inode.i_mode))
6233 inode->flags |= BTRFS_INODE_NODATASUM;
6236 btrfs_sync_inode_flags_to_i_flags(&inode->vfs_inode);
6239 int btrfs_create_new_inode(struct btrfs_trans_handle *trans,
6240 struct btrfs_new_inode_args *args)
6242 struct timespec64 ts;
6243 struct inode *dir = args->dir;
6244 struct inode *inode = args->inode;
6245 const struct fscrypt_str *name = args->orphan ? NULL : &args->fname.disk_name;
6246 struct btrfs_fs_info *fs_info = inode_to_fs_info(dir);
6247 struct btrfs_root *root;
6248 struct btrfs_inode_item *inode_item;
6249 struct btrfs_key *location;
6250 struct btrfs_path *path;
6252 struct btrfs_inode_ref *ref;
6253 struct btrfs_key key[2];
6255 struct btrfs_item_batch batch;
6259 path = btrfs_alloc_path();
6264 BTRFS_I(inode)->root = btrfs_grab_root(BTRFS_I(dir)->root);
6265 root = BTRFS_I(inode)->root;
6267 ret = btrfs_get_free_objectid(root, &objectid);
6270 inode->i_ino = objectid;
6274 * O_TMPFILE, set link count to 0, so that after this point, we
6275 * fill in an inode item with the correct link count.
6277 set_nlink(inode, 0);
6279 trace_btrfs_inode_request(dir);
6281 ret = btrfs_set_inode_index(BTRFS_I(dir), &BTRFS_I(inode)->dir_index);
6285 /* index_cnt is ignored for everything but a dir. */
6286 BTRFS_I(inode)->index_cnt = BTRFS_DIR_START_INDEX;
6287 BTRFS_I(inode)->generation = trans->transid;
6288 inode->i_generation = BTRFS_I(inode)->generation;
6291 * We don't have any capability xattrs set here yet, shortcut any
6292 * queries for the xattrs here. If we add them later via the inode
6293 * security init path or any other path this flag will be cleared.
6295 set_bit(BTRFS_INODE_NO_CAP_XATTR, &BTRFS_I(inode)->runtime_flags);
6298 * Subvolumes don't inherit flags from their parent directory.
6299 * Originally this was probably by accident, but we probably can't
6300 * change it now without compatibility issues.
6303 btrfs_inherit_iflags(BTRFS_I(inode), BTRFS_I(dir));
6305 if (S_ISREG(inode->i_mode)) {
6306 if (btrfs_test_opt(fs_info, NODATASUM))
6307 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6308 if (btrfs_test_opt(fs_info, NODATACOW))
6309 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6310 BTRFS_INODE_NODATASUM;
6313 location = &BTRFS_I(inode)->location;
6314 location->objectid = objectid;
6315 location->offset = 0;
6316 location->type = BTRFS_INODE_ITEM_KEY;
6318 ret = btrfs_insert_inode_locked(inode);
6321 BTRFS_I(dir)->index_cnt--;
6326 * We could have gotten an inode number from somebody who was fsynced
6327 * and then removed in this same transaction, so let's just set full
6328 * sync since it will be a full sync anyway and this will blow away the
6329 * old info in the log.
6331 btrfs_set_inode_full_sync(BTRFS_I(inode));
6333 key[0].objectid = objectid;
6334 key[0].type = BTRFS_INODE_ITEM_KEY;
6337 sizes[0] = sizeof(struct btrfs_inode_item);
6339 if (!args->orphan) {
6341 * Start new inodes with an inode_ref. This is slightly more
6342 * efficient for small numbers of hard links since they will
6343 * be packed into one item. Extended refs will kick in if we
6344 * add more hard links than can fit in the ref item.
6346 key[1].objectid = objectid;
6347 key[1].type = BTRFS_INODE_REF_KEY;
6349 key[1].offset = objectid;
6350 sizes[1] = 2 + sizeof(*ref);
6352 key[1].offset = btrfs_ino(BTRFS_I(dir));
6353 sizes[1] = name->len + sizeof(*ref);
6357 batch.keys = &key[0];
6358 batch.data_sizes = &sizes[0];
6359 batch.total_data_size = sizes[0] + (args->orphan ? 0 : sizes[1]);
6360 batch.nr = args->orphan ? 1 : 2;
6361 ret = btrfs_insert_empty_items(trans, root, path, &batch);
6363 btrfs_abort_transaction(trans, ret);
6367 ts = simple_inode_init_ts(inode);
6368 BTRFS_I(inode)->i_otime_sec = ts.tv_sec;
6369 BTRFS_I(inode)->i_otime_nsec = ts.tv_nsec;
6372 * We're going to fill the inode item now, so at this point the inode
6373 * must be fully initialized.
6376 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6377 struct btrfs_inode_item);
6378 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6379 sizeof(*inode_item));
6380 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6382 if (!args->orphan) {
6383 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6384 struct btrfs_inode_ref);
6385 ptr = (unsigned long)(ref + 1);
6387 btrfs_set_inode_ref_name_len(path->nodes[0], ref, 2);
6388 btrfs_set_inode_ref_index(path->nodes[0], ref, 0);
6389 write_extent_buffer(path->nodes[0], "..", ptr, 2);
6391 btrfs_set_inode_ref_name_len(path->nodes[0], ref,
6393 btrfs_set_inode_ref_index(path->nodes[0], ref,
6394 BTRFS_I(inode)->dir_index);
6395 write_extent_buffer(path->nodes[0], name->name, ptr,
6400 btrfs_mark_buffer_dirty(trans, path->nodes[0]);
6402 * We don't need the path anymore, plus inheriting properties, adding
6403 * ACLs, security xattrs, orphan item or adding the link, will result in
6404 * allocating yet another path. So just free our path.
6406 btrfs_free_path(path);
6410 struct inode *parent;
6413 * Subvolumes inherit properties from their parent subvolume,
6414 * not the directory they were created in.
6416 parent = btrfs_iget(fs_info->sb, BTRFS_FIRST_FREE_OBJECTID,
6417 BTRFS_I(dir)->root);
6418 if (IS_ERR(parent)) {
6419 ret = PTR_ERR(parent);
6421 ret = btrfs_inode_inherit_props(trans, inode, parent);
6425 ret = btrfs_inode_inherit_props(trans, inode, dir);
6429 "error inheriting props for ino %llu (root %llu): %d",
6430 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid,
6435 * Subvolumes don't inherit ACLs or get passed to the LSM. This is
6438 if (!args->subvol) {
6439 ret = btrfs_init_inode_security(trans, args);
6441 btrfs_abort_transaction(trans, ret);
6446 inode_tree_add(BTRFS_I(inode));
6448 trace_btrfs_inode_new(inode);
6449 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6451 btrfs_update_root_times(trans, root);
6454 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
6456 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
6457 0, BTRFS_I(inode)->dir_index);
6460 btrfs_abort_transaction(trans, ret);
6468 * discard_new_inode() calls iput(), but the caller owns the reference
6472 discard_new_inode(inode);
6474 btrfs_free_path(path);
6479 * utility function to add 'inode' into 'parent_inode' with
6480 * a give name and a given sequence number.
6481 * if 'add_backref' is true, also insert a backref from the
6482 * inode to the parent directory.
6484 int btrfs_add_link(struct btrfs_trans_handle *trans,
6485 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6486 const struct fscrypt_str *name, int add_backref, u64 index)
6489 struct btrfs_key key;
6490 struct btrfs_root *root = parent_inode->root;
6491 u64 ino = btrfs_ino(inode);
6492 u64 parent_ino = btrfs_ino(parent_inode);
6494 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6495 memcpy(&key, &inode->root->root_key, sizeof(key));
6498 key.type = BTRFS_INODE_ITEM_KEY;
6502 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6503 ret = btrfs_add_root_ref(trans, key.objectid,
6504 root->root_key.objectid, parent_ino,
6506 } else if (add_backref) {
6507 ret = btrfs_insert_inode_ref(trans, root, name,
6508 ino, parent_ino, index);
6511 /* Nothing to clean up yet */
6515 ret = btrfs_insert_dir_item(trans, name, parent_inode, &key,
6516 btrfs_inode_type(&inode->vfs_inode), index);
6517 if (ret == -EEXIST || ret == -EOVERFLOW)
6520 btrfs_abort_transaction(trans, ret);
6524 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6526 inode_inc_iversion(&parent_inode->vfs_inode);
6528 * If we are replaying a log tree, we do not want to update the mtime
6529 * and ctime of the parent directory with the current time, since the
6530 * log replay procedure is responsible for setting them to their correct
6531 * values (the ones it had when the fsync was done).
6533 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags))
6534 inode_set_mtime_to_ts(&parent_inode->vfs_inode,
6535 inode_set_ctime_current(&parent_inode->vfs_inode));
6537 ret = btrfs_update_inode(trans, parent_inode);
6539 btrfs_abort_transaction(trans, ret);
6543 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6546 err = btrfs_del_root_ref(trans, key.objectid,
6547 root->root_key.objectid, parent_ino,
6548 &local_index, name);
6550 btrfs_abort_transaction(trans, err);
6551 } else if (add_backref) {
6555 err = btrfs_del_inode_ref(trans, root, name, ino, parent_ino,
6558 btrfs_abort_transaction(trans, err);
6561 /* Return the original error code */
6565 static int btrfs_create_common(struct inode *dir, struct dentry *dentry,
6566 struct inode *inode)
6568 struct btrfs_fs_info *fs_info = inode_to_fs_info(dir);
6569 struct btrfs_root *root = BTRFS_I(dir)->root;
6570 struct btrfs_new_inode_args new_inode_args = {
6575 unsigned int trans_num_items;
6576 struct btrfs_trans_handle *trans;
6579 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
6583 trans = btrfs_start_transaction(root, trans_num_items);
6584 if (IS_ERR(trans)) {
6585 err = PTR_ERR(trans);
6586 goto out_new_inode_args;
6589 err = btrfs_create_new_inode(trans, &new_inode_args);
6591 d_instantiate_new(dentry, inode);
6593 btrfs_end_transaction(trans);
6594 btrfs_btree_balance_dirty(fs_info);
6596 btrfs_new_inode_args_destroy(&new_inode_args);
6603 static int btrfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
6604 struct dentry *dentry, umode_t mode, dev_t rdev)
6606 struct inode *inode;
6608 inode = new_inode(dir->i_sb);
6611 inode_init_owner(idmap, inode, dir, mode);
6612 inode->i_op = &btrfs_special_inode_operations;
6613 init_special_inode(inode, inode->i_mode, rdev);
6614 return btrfs_create_common(dir, dentry, inode);
6617 static int btrfs_create(struct mnt_idmap *idmap, struct inode *dir,
6618 struct dentry *dentry, umode_t mode, bool excl)
6620 struct inode *inode;
6622 inode = new_inode(dir->i_sb);
6625 inode_init_owner(idmap, inode, dir, mode);
6626 inode->i_fop = &btrfs_file_operations;
6627 inode->i_op = &btrfs_file_inode_operations;
6628 inode->i_mapping->a_ops = &btrfs_aops;
6629 return btrfs_create_common(dir, dentry, inode);
6632 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6633 struct dentry *dentry)
6635 struct btrfs_trans_handle *trans = NULL;
6636 struct btrfs_root *root = BTRFS_I(dir)->root;
6637 struct inode *inode = d_inode(old_dentry);
6638 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
6639 struct fscrypt_name fname;
6644 /* do not allow sys_link's with other subvols of the same device */
6645 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6648 if (inode->i_nlink >= BTRFS_LINK_MAX)
6651 err = fscrypt_setup_filename(dir, &dentry->d_name, 0, &fname);
6655 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6660 * 2 items for inode and inode ref
6661 * 2 items for dir items
6662 * 1 item for parent inode
6663 * 1 item for orphan item deletion if O_TMPFILE
6665 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6666 if (IS_ERR(trans)) {
6667 err = PTR_ERR(trans);
6672 /* There are several dir indexes for this inode, clear the cache. */
6673 BTRFS_I(inode)->dir_index = 0ULL;
6675 inode_inc_iversion(inode);
6676 inode_set_ctime_current(inode);
6678 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6680 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6681 &fname.disk_name, 1, index);
6686 struct dentry *parent = dentry->d_parent;
6688 err = btrfs_update_inode(trans, BTRFS_I(inode));
6691 if (inode->i_nlink == 1) {
6693 * If new hard link count is 1, it's a file created
6694 * with open(2) O_TMPFILE flag.
6696 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6700 d_instantiate(dentry, inode);
6701 btrfs_log_new_name(trans, old_dentry, NULL, 0, parent);
6705 fscrypt_free_filename(&fname);
6707 btrfs_end_transaction(trans);
6709 inode_dec_link_count(inode);
6712 btrfs_btree_balance_dirty(fs_info);
6716 static int btrfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
6717 struct dentry *dentry, umode_t mode)
6719 struct inode *inode;
6721 inode = new_inode(dir->i_sb);
6724 inode_init_owner(idmap, inode, dir, S_IFDIR | mode);
6725 inode->i_op = &btrfs_dir_inode_operations;
6726 inode->i_fop = &btrfs_dir_file_operations;
6727 return btrfs_create_common(dir, dentry, inode);
6730 static noinline int uncompress_inline(struct btrfs_path *path,
6732 struct btrfs_file_extent_item *item)
6735 struct extent_buffer *leaf = path->nodes[0];
6738 unsigned long inline_size;
6742 compress_type = btrfs_file_extent_compression(leaf, item);
6743 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6744 inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]);
6745 tmp = kmalloc(inline_size, GFP_NOFS);
6748 ptr = btrfs_file_extent_inline_start(item);
6750 read_extent_buffer(leaf, tmp, ptr, inline_size);
6752 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6753 ret = btrfs_decompress(compress_type, tmp, page, 0, inline_size, max_size);
6756 * decompression code contains a memset to fill in any space between the end
6757 * of the uncompressed data and the end of max_size in case the decompressed
6758 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6759 * the end of an inline extent and the beginning of the next block, so we
6760 * cover that region here.
6763 if (max_size < PAGE_SIZE)
6764 memzero_page(page, max_size, PAGE_SIZE - max_size);
6769 static int read_inline_extent(struct btrfs_inode *inode, struct btrfs_path *path,
6772 struct btrfs_file_extent_item *fi;
6776 if (!page || PageUptodate(page))
6779 ASSERT(page_offset(page) == 0);
6781 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
6782 struct btrfs_file_extent_item);
6783 if (btrfs_file_extent_compression(path->nodes[0], fi) != BTRFS_COMPRESS_NONE)
6784 return uncompress_inline(path, page, fi);
6786 copy_size = min_t(u64, PAGE_SIZE,
6787 btrfs_file_extent_ram_bytes(path->nodes[0], fi));
6788 kaddr = kmap_local_page(page);
6789 read_extent_buffer(path->nodes[0], kaddr,
6790 btrfs_file_extent_inline_start(fi), copy_size);
6791 kunmap_local(kaddr);
6792 if (copy_size < PAGE_SIZE)
6793 memzero_page(page, copy_size, PAGE_SIZE - copy_size);
6798 * Lookup the first extent overlapping a range in a file.
6800 * @inode: file to search in
6801 * @page: page to read extent data into if the extent is inline
6802 * @start: file offset
6803 * @len: length of range starting at @start
6805 * Return the first &struct extent_map which overlaps the given range, reading
6806 * it from the B-tree and caching it if necessary. Note that there may be more
6807 * extents which overlap the given range after the returned extent_map.
6809 * If @page is not NULL and the extent is inline, this also reads the extent
6810 * data directly into the page and marks the extent up to date in the io_tree.
6812 * Return: ERR_PTR on error, non-NULL extent_map on success.
6814 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6815 struct page *page, u64 start, u64 len)
6817 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6819 u64 extent_start = 0;
6821 u64 objectid = btrfs_ino(inode);
6822 int extent_type = -1;
6823 struct btrfs_path *path = NULL;
6824 struct btrfs_root *root = inode->root;
6825 struct btrfs_file_extent_item *item;
6826 struct extent_buffer *leaf;
6827 struct btrfs_key found_key;
6828 struct extent_map *em = NULL;
6829 struct extent_map_tree *em_tree = &inode->extent_tree;
6831 read_lock(&em_tree->lock);
6832 em = lookup_extent_mapping(em_tree, start, len);
6833 read_unlock(&em_tree->lock);
6836 if (em->start > start || em->start + em->len <= start)
6837 free_extent_map(em);
6838 else if (em->block_start == EXTENT_MAP_INLINE && page)
6839 free_extent_map(em);
6843 em = alloc_extent_map();
6848 em->start = EXTENT_MAP_HOLE;
6849 em->orig_start = EXTENT_MAP_HOLE;
6851 em->block_len = (u64)-1;
6853 path = btrfs_alloc_path();
6859 /* Chances are we'll be called again, so go ahead and do readahead */
6860 path->reada = READA_FORWARD;
6863 * The same explanation in load_free_space_cache applies here as well,
6864 * we only read when we're loading the free space cache, and at that
6865 * point the commit_root has everything we need.
6867 if (btrfs_is_free_space_inode(inode)) {
6868 path->search_commit_root = 1;
6869 path->skip_locking = 1;
6872 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6875 } else if (ret > 0) {
6876 if (path->slots[0] == 0)
6882 leaf = path->nodes[0];
6883 item = btrfs_item_ptr(leaf, path->slots[0],
6884 struct btrfs_file_extent_item);
6885 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6886 if (found_key.objectid != objectid ||
6887 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6889 * If we backup past the first extent we want to move forward
6890 * and see if there is an extent in front of us, otherwise we'll
6891 * say there is a hole for our whole search range which can
6898 extent_type = btrfs_file_extent_type(leaf, item);
6899 extent_start = found_key.offset;
6900 extent_end = btrfs_file_extent_end(path);
6901 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6902 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6903 /* Only regular file could have regular/prealloc extent */
6904 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6907 "regular/prealloc extent found for non-regular inode %llu",
6911 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6913 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6914 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6919 if (start >= extent_end) {
6921 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6922 ret = btrfs_next_leaf(root, path);
6928 leaf = path->nodes[0];
6930 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6931 if (found_key.objectid != objectid ||
6932 found_key.type != BTRFS_EXTENT_DATA_KEY)
6934 if (start + len <= found_key.offset)
6936 if (start > found_key.offset)
6939 /* New extent overlaps with existing one */
6941 em->orig_start = start;
6942 em->len = found_key.offset - start;
6943 em->block_start = EXTENT_MAP_HOLE;
6947 btrfs_extent_item_to_extent_map(inode, path, item, em);
6949 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6950 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6952 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6954 * Inline extent can only exist at file offset 0. This is
6955 * ensured by tree-checker and inline extent creation path.
6956 * Thus all members representing file offsets should be zero.
6958 ASSERT(extent_start == 0);
6959 ASSERT(em->start == 0);
6962 * btrfs_extent_item_to_extent_map() should have properly
6963 * initialized em members already.
6965 * Other members are not utilized for inline extents.
6967 ASSERT(em->block_start == EXTENT_MAP_INLINE);
6968 ASSERT(em->len == fs_info->sectorsize);
6970 ret = read_inline_extent(inode, path, page);
6977 em->orig_start = start;
6979 em->block_start = EXTENT_MAP_HOLE;
6982 btrfs_release_path(path);
6983 if (em->start > start || extent_map_end(em) <= start) {
6985 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6986 em->start, em->len, start, len);
6991 write_lock(&em_tree->lock);
6992 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6993 write_unlock(&em_tree->lock);
6995 btrfs_free_path(path);
6997 trace_btrfs_get_extent(root, inode, em);
7000 free_extent_map(em);
7001 return ERR_PTR(ret);
7006 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
7007 struct btrfs_dio_data *dio_data,
7010 const u64 orig_start,
7011 const u64 block_start,
7012 const u64 block_len,
7013 const u64 orig_block_len,
7014 const u64 ram_bytes,
7017 struct extent_map *em = NULL;
7018 struct btrfs_ordered_extent *ordered;
7020 if (type != BTRFS_ORDERED_NOCOW) {
7021 em = create_io_em(inode, start, len, orig_start, block_start,
7022 block_len, orig_block_len, ram_bytes,
7023 BTRFS_COMPRESS_NONE, /* compress_type */
7028 ordered = btrfs_alloc_ordered_extent(inode, start, len, len,
7029 block_start, block_len, 0,
7031 (1 << BTRFS_ORDERED_DIRECT),
7032 BTRFS_COMPRESS_NONE);
7033 if (IS_ERR(ordered)) {
7035 free_extent_map(em);
7036 btrfs_drop_extent_map_range(inode, start,
7037 start + len - 1, false);
7039 em = ERR_CAST(ordered);
7041 ASSERT(!dio_data->ordered);
7042 dio_data->ordered = ordered;
7049 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
7050 struct btrfs_dio_data *dio_data,
7053 struct btrfs_root *root = inode->root;
7054 struct btrfs_fs_info *fs_info = root->fs_info;
7055 struct extent_map *em;
7056 struct btrfs_key ins;
7060 alloc_hint = get_extent_allocation_hint(inode, start, len);
7062 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7063 0, alloc_hint, &ins, 1, 1);
7064 if (ret == -EAGAIN) {
7065 ASSERT(btrfs_is_zoned(fs_info));
7066 wait_on_bit_io(&inode->root->fs_info->flags, BTRFS_FS_NEED_ZONE_FINISH,
7067 TASK_UNINTERRUPTIBLE);
7071 return ERR_PTR(ret);
7073 em = btrfs_create_dio_extent(inode, dio_data, start, ins.offset, start,
7074 ins.objectid, ins.offset, ins.offset,
7075 ins.offset, BTRFS_ORDERED_REGULAR);
7076 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7078 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7084 static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
7086 struct btrfs_block_group *block_group;
7087 bool readonly = false;
7089 block_group = btrfs_lookup_block_group(fs_info, bytenr);
7090 if (!block_group || block_group->ro)
7093 btrfs_put_block_group(block_group);
7098 * Check if we can do nocow write into the range [@offset, @offset + @len)
7100 * @offset: File offset
7101 * @len: The length to write, will be updated to the nocow writeable
7103 * @orig_start: (optional) Return the original file offset of the file extent
7104 * @orig_len: (optional) Return the original on-disk length of the file extent
7105 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7106 * @strict: if true, omit optimizations that might force us into unnecessary
7107 * cow. e.g., don't trust generation number.
7110 * >0 and update @len if we can do nocow write
7111 * 0 if we can't do nocow write
7112 * <0 if error happened
7114 * NOTE: This only checks the file extents, caller is responsible to wait for
7115 * any ordered extents.
7117 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7118 u64 *orig_start, u64 *orig_block_len,
7119 u64 *ram_bytes, bool nowait, bool strict)
7121 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
7122 struct can_nocow_file_extent_args nocow_args = { 0 };
7123 struct btrfs_path *path;
7125 struct extent_buffer *leaf;
7126 struct btrfs_root *root = BTRFS_I(inode)->root;
7127 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7128 struct btrfs_file_extent_item *fi;
7129 struct btrfs_key key;
7132 path = btrfs_alloc_path();
7135 path->nowait = nowait;
7137 ret = btrfs_lookup_file_extent(NULL, root, path,
7138 btrfs_ino(BTRFS_I(inode)), offset, 0);
7143 if (path->slots[0] == 0) {
7144 /* can't find the item, must cow */
7151 leaf = path->nodes[0];
7152 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7153 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7154 key.type != BTRFS_EXTENT_DATA_KEY) {
7155 /* not our file or wrong item type, must cow */
7159 if (key.offset > offset) {
7160 /* Wrong offset, must cow */
7164 if (btrfs_file_extent_end(path) <= offset)
7167 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
7168 found_type = btrfs_file_extent_type(leaf, fi);
7170 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7172 nocow_args.start = offset;
7173 nocow_args.end = offset + *len - 1;
7174 nocow_args.strict = strict;
7175 nocow_args.free_path = true;
7177 ret = can_nocow_file_extent(path, &key, BTRFS_I(inode), &nocow_args);
7178 /* can_nocow_file_extent() has freed the path. */
7182 /* Treat errors as not being able to NOCOW. */
7188 if (btrfs_extent_readonly(fs_info, nocow_args.disk_bytenr))
7191 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7192 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7195 range_end = round_up(offset + nocow_args.num_bytes,
7196 root->fs_info->sectorsize) - 1;
7197 ret = test_range_bit_exists(io_tree, offset, range_end, EXTENT_DELALLOC);
7205 *orig_start = key.offset - nocow_args.extent_offset;
7207 *orig_block_len = nocow_args.disk_num_bytes;
7209 *len = nocow_args.num_bytes;
7212 btrfs_free_path(path);
7216 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7217 struct extent_state **cached_state,
7218 unsigned int iomap_flags)
7220 const bool writing = (iomap_flags & IOMAP_WRITE);
7221 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7222 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7223 struct btrfs_ordered_extent *ordered;
7228 if (!try_lock_extent(io_tree, lockstart, lockend,
7232 lock_extent(io_tree, lockstart, lockend, cached_state);
7235 * We're concerned with the entire range that we're going to be
7236 * doing DIO to, so we need to make sure there's no ordered
7237 * extents in this range.
7239 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7240 lockend - lockstart + 1);
7243 * We need to make sure there are no buffered pages in this
7244 * range either, we could have raced between the invalidate in
7245 * generic_file_direct_write and locking the extent. The
7246 * invalidate needs to happen so that reads after a write do not
7250 (!writing || !filemap_range_has_page(inode->i_mapping,
7251 lockstart, lockend)))
7254 unlock_extent(io_tree, lockstart, lockend, cached_state);
7258 btrfs_put_ordered_extent(ordered);
7263 * If we are doing a DIO read and the ordered extent we
7264 * found is for a buffered write, we can not wait for it
7265 * to complete and retry, because if we do so we can
7266 * deadlock with concurrent buffered writes on page
7267 * locks. This happens only if our DIO read covers more
7268 * than one extent map, if at this point has already
7269 * created an ordered extent for a previous extent map
7270 * and locked its range in the inode's io tree, and a
7271 * concurrent write against that previous extent map's
7272 * range and this range started (we unlock the ranges
7273 * in the io tree only when the bios complete and
7274 * buffered writes always lock pages before attempting
7275 * to lock range in the io tree).
7278 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7279 btrfs_start_ordered_extent(ordered);
7281 ret = nowait ? -EAGAIN : -ENOTBLK;
7282 btrfs_put_ordered_extent(ordered);
7285 * We could trigger writeback for this range (and wait
7286 * for it to complete) and then invalidate the pages for
7287 * this range (through invalidate_inode_pages2_range()),
7288 * but that can lead us to a deadlock with a concurrent
7289 * call to readahead (a buffered read or a defrag call
7290 * triggered a readahead) on a page lock due to an
7291 * ordered dio extent we created before but did not have
7292 * yet a corresponding bio submitted (whence it can not
7293 * complete), which makes readahead wait for that
7294 * ordered extent to complete while holding a lock on
7297 ret = nowait ? -EAGAIN : -ENOTBLK;
7309 /* The callers of this must take lock_extent() */
7310 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7311 u64 len, u64 orig_start, u64 block_start,
7312 u64 block_len, u64 orig_block_len,
7313 u64 ram_bytes, int compress_type,
7316 struct extent_map *em;
7319 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7320 type == BTRFS_ORDERED_COMPRESSED ||
7321 type == BTRFS_ORDERED_NOCOW ||
7322 type == BTRFS_ORDERED_REGULAR);
7324 em = alloc_extent_map();
7326 return ERR_PTR(-ENOMEM);
7329 em->orig_start = orig_start;
7331 em->block_len = block_len;
7332 em->block_start = block_start;
7333 em->orig_block_len = orig_block_len;
7334 em->ram_bytes = ram_bytes;
7335 em->generation = -1;
7336 em->flags |= EXTENT_FLAG_PINNED;
7337 if (type == BTRFS_ORDERED_PREALLOC)
7338 em->flags |= EXTENT_FLAG_FILLING;
7339 else if (type == BTRFS_ORDERED_COMPRESSED)
7340 extent_map_set_compression(em, compress_type);
7342 ret = btrfs_replace_extent_map_range(inode, em, true);
7344 free_extent_map(em);
7345 return ERR_PTR(ret);
7348 /* em got 2 refs now, callers needs to do free_extent_map once. */
7353 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7354 struct inode *inode,
7355 struct btrfs_dio_data *dio_data,
7356 u64 start, u64 *lenp,
7357 unsigned int iomap_flags)
7359 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7360 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
7361 struct extent_map *em = *map;
7363 u64 block_start, orig_start, orig_block_len, ram_bytes;
7364 struct btrfs_block_group *bg;
7365 bool can_nocow = false;
7366 bool space_reserved = false;
7372 * We don't allocate a new extent in the following cases
7374 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7376 * 2) The extent is marked as PREALLOC. We're good to go here and can
7377 * just use the extent.
7380 if ((em->flags & EXTENT_FLAG_PREALLOC) ||
7381 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7382 em->block_start != EXTENT_MAP_HOLE)) {
7383 if (em->flags & EXTENT_FLAG_PREALLOC)
7384 type = BTRFS_ORDERED_PREALLOC;
7386 type = BTRFS_ORDERED_NOCOW;
7387 len = min(len, em->len - (start - em->start));
7388 block_start = em->block_start + (start - em->start);
7390 if (can_nocow_extent(inode, start, &len, &orig_start,
7391 &orig_block_len, &ram_bytes, false, false) == 1) {
7392 bg = btrfs_inc_nocow_writers(fs_info, block_start);
7400 struct extent_map *em2;
7402 /* We can NOCOW, so only need to reserve metadata space. */
7403 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7406 /* Our caller expects us to free the input extent map. */
7407 free_extent_map(em);
7409 btrfs_dec_nocow_writers(bg);
7410 if (nowait && (ret == -ENOSPC || ret == -EDQUOT))
7414 space_reserved = true;
7416 em2 = btrfs_create_dio_extent(BTRFS_I(inode), dio_data, start, len,
7417 orig_start, block_start,
7418 len, orig_block_len,
7420 btrfs_dec_nocow_writers(bg);
7421 if (type == BTRFS_ORDERED_PREALLOC) {
7422 free_extent_map(em);
7432 dio_data->nocow_done = true;
7434 /* Our caller expects us to free the input extent map. */
7435 free_extent_map(em);
7444 * If we could not allocate data space before locking the file
7445 * range and we can't do a NOCOW write, then we have to fail.
7447 if (!dio_data->data_space_reserved) {
7453 * We have to COW and we have already reserved data space before,
7454 * so now we reserve only metadata.
7456 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7460 space_reserved = true;
7462 em = btrfs_new_extent_direct(BTRFS_I(inode), dio_data, start, len);
7468 len = min(len, em->len - (start - em->start));
7470 btrfs_delalloc_release_metadata(BTRFS_I(inode),
7471 prev_len - len, true);
7475 * We have created our ordered extent, so we can now release our reservation
7476 * for an outstanding extent.
7478 btrfs_delalloc_release_extents(BTRFS_I(inode), prev_len);
7481 * Need to update the i_size under the extent lock so buffered
7482 * readers will get the updated i_size when we unlock.
7484 if (start + len > i_size_read(inode))
7485 i_size_write(inode, start + len);
7487 if (ret && space_reserved) {
7488 btrfs_delalloc_release_extents(BTRFS_I(inode), len);
7489 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, true);
7495 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7496 loff_t length, unsigned int flags, struct iomap *iomap,
7497 struct iomap *srcmap)
7499 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7500 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
7501 struct extent_map *em;
7502 struct extent_state *cached_state = NULL;
7503 struct btrfs_dio_data *dio_data = iter->private;
7504 u64 lockstart, lockend;
7505 const bool write = !!(flags & IOMAP_WRITE);
7508 const u64 data_alloc_len = length;
7509 bool unlock_extents = false;
7512 * We could potentially fault if we have a buffer > PAGE_SIZE, and if
7513 * we're NOWAIT we may submit a bio for a partial range and return
7514 * EIOCBQUEUED, which would result in an errant short read.
7516 * The best way to handle this would be to allow for partial completions
7517 * of iocb's, so we could submit the partial bio, return and fault in
7518 * the rest of the pages, and then submit the io for the rest of the
7519 * range. However we don't have that currently, so simply return
7520 * -EAGAIN at this point so that the normal path is used.
7522 if (!write && (flags & IOMAP_NOWAIT) && length > PAGE_SIZE)
7526 * Cap the size of reads to that usually seen in buffered I/O as we need
7527 * to allocate a contiguous array for the checksums.
7530 len = min_t(u64, len, fs_info->sectorsize * BTRFS_MAX_BIO_SECTORS);
7533 lockend = start + len - 1;
7536 * iomap_dio_rw() only does filemap_write_and_wait_range(), which isn't
7537 * enough if we've written compressed pages to this area, so we need to
7538 * flush the dirty pages again to make absolutely sure that any
7539 * outstanding dirty pages are on disk - the first flush only starts
7540 * compression on the data, while keeping the pages locked, so by the
7541 * time the second flush returns we know bios for the compressed pages
7542 * were submitted and finished, and the pages no longer under writeback.
7544 * If we have a NOWAIT request and we have any pages in the range that
7545 * are locked, likely due to compression still in progress, we don't want
7546 * to block on page locks. We also don't want to block on pages marked as
7547 * dirty or under writeback (same as for the non-compression case).
7548 * iomap_dio_rw() did the same check, but after that and before we got
7549 * here, mmap'ed writes may have happened or buffered reads started
7550 * (readpage() and readahead(), which lock pages), as we haven't locked
7551 * the file range yet.
7553 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7554 &BTRFS_I(inode)->runtime_flags)) {
7555 if (flags & IOMAP_NOWAIT) {
7556 if (filemap_range_needs_writeback(inode->i_mapping,
7557 lockstart, lockend))
7560 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7561 start + length - 1);
7567 memset(dio_data, 0, sizeof(*dio_data));
7570 * We always try to allocate data space and must do it before locking
7571 * the file range, to avoid deadlocks with concurrent writes to the same
7572 * range if the range has several extents and the writes don't expand the
7573 * current i_size (the inode lock is taken in shared mode). If we fail to
7574 * allocate data space here we continue and later, after locking the
7575 * file range, we fail with ENOSPC only if we figure out we can not do a
7578 if (write && !(flags & IOMAP_NOWAIT)) {
7579 ret = btrfs_check_data_free_space(BTRFS_I(inode),
7580 &dio_data->data_reserved,
7581 start, data_alloc_len, false);
7583 dio_data->data_space_reserved = true;
7584 else if (ret && !(BTRFS_I(inode)->flags &
7585 (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
7590 * If this errors out it's because we couldn't invalidate pagecache for
7591 * this range and we need to fallback to buffered IO, or we are doing a
7592 * NOWAIT read/write and we need to block.
7594 ret = lock_extent_direct(inode, lockstart, lockend, &cached_state, flags);
7598 em = btrfs_get_extent(BTRFS_I(inode), NULL, start, len);
7605 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7606 * io. INLINE is special, and we could probably kludge it in here, but
7607 * it's still buffered so for safety lets just fall back to the generic
7610 * For COMPRESSED we _have_ to read the entire extent in so we can
7611 * decompress it, so there will be buffering required no matter what we
7612 * do, so go ahead and fallback to buffered.
7614 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7615 * to buffered IO. Don't blame me, this is the price we pay for using
7618 if (extent_map_is_compressed(em) ||
7619 em->block_start == EXTENT_MAP_INLINE) {
7620 free_extent_map(em);
7622 * If we are in a NOWAIT context, return -EAGAIN in order to
7623 * fallback to buffered IO. This is not only because we can
7624 * block with buffered IO (no support for NOWAIT semantics at
7625 * the moment) but also to avoid returning short reads to user
7626 * space - this happens if we were able to read some data from
7627 * previous non-compressed extents and then when we fallback to
7628 * buffered IO, at btrfs_file_read_iter() by calling
7629 * filemap_read(), we fail to fault in pages for the read buffer,
7630 * in which case filemap_read() returns a short read (the number
7631 * of bytes previously read is > 0, so it does not return -EFAULT).
7633 ret = (flags & IOMAP_NOWAIT) ? -EAGAIN : -ENOTBLK;
7637 len = min(len, em->len - (start - em->start));
7640 * If we have a NOWAIT request and the range contains multiple extents
7641 * (or a mix of extents and holes), then we return -EAGAIN to make the
7642 * caller fallback to a context where it can do a blocking (without
7643 * NOWAIT) request. This way we avoid doing partial IO and returning
7644 * success to the caller, which is not optimal for writes and for reads
7645 * it can result in unexpected behaviour for an application.
7647 * When doing a read, because we use IOMAP_DIO_PARTIAL when calling
7648 * iomap_dio_rw(), we can end up returning less data then what the caller
7649 * asked for, resulting in an unexpected, and incorrect, short read.
7650 * That is, the caller asked to read N bytes and we return less than that,
7651 * which is wrong unless we are crossing EOF. This happens if we get a
7652 * page fault error when trying to fault in pages for the buffer that is
7653 * associated to the struct iov_iter passed to iomap_dio_rw(), and we
7654 * have previously submitted bios for other extents in the range, in
7655 * which case iomap_dio_rw() may return us EIOCBQUEUED if not all of
7656 * those bios have completed by the time we get the page fault error,
7657 * which we return back to our caller - we should only return EIOCBQUEUED
7658 * after we have submitted bios for all the extents in the range.
7660 if ((flags & IOMAP_NOWAIT) && len < length) {
7661 free_extent_map(em);
7667 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7668 start, &len, flags);
7671 unlock_extents = true;
7672 /* Recalc len in case the new em is smaller than requested */
7673 len = min(len, em->len - (start - em->start));
7674 if (dio_data->data_space_reserved) {
7676 u64 release_len = 0;
7678 if (dio_data->nocow_done) {
7679 release_offset = start;
7680 release_len = data_alloc_len;
7681 } else if (len < data_alloc_len) {
7682 release_offset = start + len;
7683 release_len = data_alloc_len - len;
7686 if (release_len > 0)
7687 btrfs_free_reserved_data_space(BTRFS_I(inode),
7688 dio_data->data_reserved,
7694 * We need to unlock only the end area that we aren't using.
7695 * The rest is going to be unlocked by the endio routine.
7697 lockstart = start + len;
7698 if (lockstart < lockend)
7699 unlock_extents = true;
7703 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7706 free_extent_state(cached_state);
7709 * Translate extent map information to iomap.
7710 * We trim the extents (and move the addr) even though iomap code does
7711 * that, since we have locked only the parts we are performing I/O in.
7713 if ((em->block_start == EXTENT_MAP_HOLE) ||
7714 ((em->flags & EXTENT_FLAG_PREALLOC) && !write)) {
7715 iomap->addr = IOMAP_NULL_ADDR;
7716 iomap->type = IOMAP_HOLE;
7718 iomap->addr = em->block_start + (start - em->start);
7719 iomap->type = IOMAP_MAPPED;
7721 iomap->offset = start;
7722 iomap->bdev = fs_info->fs_devices->latest_dev->bdev;
7723 iomap->length = len;
7724 free_extent_map(em);
7729 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7732 if (dio_data->data_space_reserved) {
7733 btrfs_free_reserved_data_space(BTRFS_I(inode),
7734 dio_data->data_reserved,
7735 start, data_alloc_len);
7736 extent_changeset_free(dio_data->data_reserved);
7742 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7743 ssize_t written, unsigned int flags, struct iomap *iomap)
7745 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7746 struct btrfs_dio_data *dio_data = iter->private;
7747 size_t submitted = dio_data->submitted;
7748 const bool write = !!(flags & IOMAP_WRITE);
7751 if (!write && (iomap->type == IOMAP_HOLE)) {
7752 /* If reading from a hole, unlock and return */
7753 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1,
7758 if (submitted < length) {
7760 length -= submitted;
7762 btrfs_finish_ordered_extent(dio_data->ordered, NULL,
7763 pos, length, false);
7765 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7766 pos + length - 1, NULL);
7770 btrfs_put_ordered_extent(dio_data->ordered);
7771 dio_data->ordered = NULL;
7775 extent_changeset_free(dio_data->data_reserved);
7779 static void btrfs_dio_end_io(struct btrfs_bio *bbio)
7781 struct btrfs_dio_private *dip =
7782 container_of(bbio, struct btrfs_dio_private, bbio);
7783 struct btrfs_inode *inode = bbio->inode;
7784 struct bio *bio = &bbio->bio;
7786 if (bio->bi_status) {
7787 btrfs_warn(inode->root->fs_info,
7788 "direct IO failed ino %llu op 0x%0x offset %#llx len %u err no %d",
7789 btrfs_ino(inode), bio->bi_opf,
7790 dip->file_offset, dip->bytes, bio->bi_status);
7793 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
7794 btrfs_finish_ordered_extent(bbio->ordered, NULL,
7795 dip->file_offset, dip->bytes,
7798 unlock_extent(&inode->io_tree, dip->file_offset,
7799 dip->file_offset + dip->bytes - 1, NULL);
7802 bbio->bio.bi_private = bbio->private;
7803 iomap_dio_bio_end_io(bio);
7806 static void btrfs_dio_submit_io(const struct iomap_iter *iter, struct bio *bio,
7809 struct btrfs_bio *bbio = btrfs_bio(bio);
7810 struct btrfs_dio_private *dip =
7811 container_of(bbio, struct btrfs_dio_private, bbio);
7812 struct btrfs_dio_data *dio_data = iter->private;
7814 btrfs_bio_init(bbio, BTRFS_I(iter->inode)->root->fs_info,
7815 btrfs_dio_end_io, bio->bi_private);
7816 bbio->inode = BTRFS_I(iter->inode);
7817 bbio->file_offset = file_offset;
7819 dip->file_offset = file_offset;
7820 dip->bytes = bio->bi_iter.bi_size;
7822 dio_data->submitted += bio->bi_iter.bi_size;
7825 * Check if we are doing a partial write. If we are, we need to split
7826 * the ordered extent to match the submitted bio. Hang on to the
7827 * remaining unfinishable ordered_extent in dio_data so that it can be
7828 * cancelled in iomap_end to avoid a deadlock wherein faulting the
7829 * remaining pages is blocked on the outstanding ordered extent.
7831 if (iter->flags & IOMAP_WRITE) {
7834 ret = btrfs_extract_ordered_extent(bbio, dio_data->ordered);
7836 btrfs_finish_ordered_extent(dio_data->ordered, NULL,
7837 file_offset, dip->bytes,
7839 bio->bi_status = errno_to_blk_status(ret);
7840 iomap_dio_bio_end_io(bio);
7845 btrfs_submit_bio(bbio, 0);
7848 static const struct iomap_ops btrfs_dio_iomap_ops = {
7849 .iomap_begin = btrfs_dio_iomap_begin,
7850 .iomap_end = btrfs_dio_iomap_end,
7853 static const struct iomap_dio_ops btrfs_dio_ops = {
7854 .submit_io = btrfs_dio_submit_io,
7855 .bio_set = &btrfs_dio_bioset,
7858 ssize_t btrfs_dio_read(struct kiocb *iocb, struct iov_iter *iter, size_t done_before)
7860 struct btrfs_dio_data data = { 0 };
7862 return iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7863 IOMAP_DIO_PARTIAL, &data, done_before);
7866 struct iomap_dio *btrfs_dio_write(struct kiocb *iocb, struct iov_iter *iter,
7869 struct btrfs_dio_data data = { 0 };
7871 return __iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7872 IOMAP_DIO_PARTIAL, &data, done_before);
7875 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7878 struct btrfs_inode *btrfs_inode = BTRFS_I(inode);
7881 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
7886 * fiemap_prep() called filemap_write_and_wait() for the whole possible
7887 * file range (0 to LLONG_MAX), but that is not enough if we have
7888 * compression enabled. The first filemap_fdatawrite_range() only kicks
7889 * in the compression of data (in an async thread) and will return
7890 * before the compression is done and writeback is started. A second
7891 * filemap_fdatawrite_range() is needed to wait for the compression to
7892 * complete and writeback to start. We also need to wait for ordered
7893 * extents to complete, because our fiemap implementation uses mainly
7894 * file extent items to list the extents, searching for extent maps
7895 * only for file ranges with holes or prealloc extents to figure out
7896 * if we have delalloc in those ranges.
7898 if (fieinfo->fi_flags & FIEMAP_FLAG_SYNC) {
7899 ret = btrfs_wait_ordered_range(inode, 0, LLONG_MAX);
7904 btrfs_inode_lock(btrfs_inode, BTRFS_ILOCK_SHARED);
7907 * We did an initial flush to avoid holding the inode's lock while
7908 * triggering writeback and waiting for the completion of IO and ordered
7909 * extents. Now after we locked the inode we do it again, because it's
7910 * possible a new write may have happened in between those two steps.
7912 if (fieinfo->fi_flags & FIEMAP_FLAG_SYNC) {
7913 ret = btrfs_wait_ordered_range(inode, 0, LLONG_MAX);
7915 btrfs_inode_unlock(btrfs_inode, BTRFS_ILOCK_SHARED);
7920 ret = extent_fiemap(btrfs_inode, fieinfo, start, len);
7921 btrfs_inode_unlock(btrfs_inode, BTRFS_ILOCK_SHARED);
7926 static int btrfs_writepages(struct address_space *mapping,
7927 struct writeback_control *wbc)
7929 return extent_writepages(mapping, wbc);
7932 static void btrfs_readahead(struct readahead_control *rac)
7934 extent_readahead(rac);
7938 * For release_folio() and invalidate_folio() we have a race window where
7939 * folio_end_writeback() is called but the subpage spinlock is not yet released.
7940 * If we continue to release/invalidate the page, we could cause use-after-free
7941 * for subpage spinlock. So this function is to spin and wait for subpage
7944 static void wait_subpage_spinlock(struct page *page)
7946 struct btrfs_fs_info *fs_info = page_to_fs_info(page);
7947 struct folio *folio = page_folio(page);
7948 struct btrfs_subpage *subpage;
7950 if (!btrfs_is_subpage(fs_info, page->mapping))
7953 ASSERT(folio_test_private(folio) && folio_get_private(folio));
7954 subpage = folio_get_private(folio);
7957 * This may look insane as we just acquire the spinlock and release it,
7958 * without doing anything. But we just want to make sure no one is
7959 * still holding the subpage spinlock.
7960 * And since the page is not dirty nor writeback, and we have page
7961 * locked, the only possible way to hold a spinlock is from the endio
7962 * function to clear page writeback.
7964 * Here we just acquire the spinlock so that all existing callers
7965 * should exit and we're safe to release/invalidate the page.
7967 spin_lock_irq(&subpage->lock);
7968 spin_unlock_irq(&subpage->lock);
7971 static bool __btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
7973 int ret = try_release_extent_mapping(&folio->page, gfp_flags);
7976 wait_subpage_spinlock(&folio->page);
7977 clear_page_extent_mapped(&folio->page);
7982 static bool btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
7984 if (folio_test_writeback(folio) || folio_test_dirty(folio))
7986 return __btrfs_release_folio(folio, gfp_flags);
7989 #ifdef CONFIG_MIGRATION
7990 static int btrfs_migrate_folio(struct address_space *mapping,
7991 struct folio *dst, struct folio *src,
7992 enum migrate_mode mode)
7994 int ret = filemap_migrate_folio(mapping, dst, src, mode);
7996 if (ret != MIGRATEPAGE_SUCCESS)
7999 if (folio_test_ordered(src)) {
8000 folio_clear_ordered(src);
8001 folio_set_ordered(dst);
8004 return MIGRATEPAGE_SUCCESS;
8007 #define btrfs_migrate_folio NULL
8010 static void btrfs_invalidate_folio(struct folio *folio, size_t offset,
8013 struct btrfs_inode *inode = folio_to_inode(folio);
8014 struct btrfs_fs_info *fs_info = inode->root->fs_info;
8015 struct extent_io_tree *tree = &inode->io_tree;
8016 struct extent_state *cached_state = NULL;
8017 u64 page_start = folio_pos(folio);
8018 u64 page_end = page_start + folio_size(folio) - 1;
8020 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
8023 * We have folio locked so no new ordered extent can be created on this
8024 * page, nor bio can be submitted for this folio.
8026 * But already submitted bio can still be finished on this folio.
8027 * Furthermore, endio function won't skip folio which has Ordered
8028 * (Private2) already cleared, so it's possible for endio and
8029 * invalidate_folio to do the same ordered extent accounting twice
8032 * So here we wait for any submitted bios to finish, so that we won't
8033 * do double ordered extent accounting on the same folio.
8035 folio_wait_writeback(folio);
8036 wait_subpage_spinlock(&folio->page);
8039 * For subpage case, we have call sites like
8040 * btrfs_punch_hole_lock_range() which passes range not aligned to
8042 * If the range doesn't cover the full folio, we don't need to and
8043 * shouldn't clear page extent mapped, as folio->private can still
8044 * record subpage dirty bits for other part of the range.
8046 * For cases that invalidate the full folio even the range doesn't
8047 * cover the full folio, like invalidating the last folio, we're
8048 * still safe to wait for ordered extent to finish.
8050 if (!(offset == 0 && length == folio_size(folio))) {
8051 btrfs_release_folio(folio, GFP_NOFS);
8055 if (!inode_evicting)
8056 lock_extent(tree, page_start, page_end, &cached_state);
8059 while (cur < page_end) {
8060 struct btrfs_ordered_extent *ordered;
8063 u32 extra_flags = 0;
8065 ordered = btrfs_lookup_first_ordered_range(inode, cur,
8066 page_end + 1 - cur);
8068 range_end = page_end;
8070 * No ordered extent covering this range, we are safe
8071 * to delete all extent states in the range.
8073 extra_flags = EXTENT_CLEAR_ALL_BITS;
8076 if (ordered->file_offset > cur) {
8078 * There is a range between [cur, oe->file_offset) not
8079 * covered by any ordered extent.
8080 * We are safe to delete all extent states, and handle
8081 * the ordered extent in the next iteration.
8083 range_end = ordered->file_offset - 1;
8084 extra_flags = EXTENT_CLEAR_ALL_BITS;
8088 range_end = min(ordered->file_offset + ordered->num_bytes - 1,
8090 ASSERT(range_end + 1 - cur < U32_MAX);
8091 range_len = range_end + 1 - cur;
8092 if (!btrfs_folio_test_ordered(fs_info, folio, cur, range_len)) {
8094 * If Ordered (Private2) is cleared, it means endio has
8095 * already been executed for the range.
8096 * We can't delete the extent states as
8097 * btrfs_finish_ordered_io() may still use some of them.
8101 btrfs_folio_clear_ordered(fs_info, folio, cur, range_len);
8104 * IO on this page will never be started, so we need to account
8105 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8106 * here, must leave that up for the ordered extent completion.
8108 * This will also unlock the range for incoming
8109 * btrfs_finish_ordered_io().
8111 if (!inode_evicting)
8112 clear_extent_bit(tree, cur, range_end,
8114 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8115 EXTENT_DEFRAG, &cached_state);
8117 spin_lock_irq(&inode->ordered_tree_lock);
8118 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8119 ordered->truncated_len = min(ordered->truncated_len,
8120 cur - ordered->file_offset);
8121 spin_unlock_irq(&inode->ordered_tree_lock);
8124 * If the ordered extent has finished, we're safe to delete all
8125 * the extent states of the range, otherwise
8126 * btrfs_finish_ordered_io() will get executed by endio for
8127 * other pages, so we can't delete extent states.
8129 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8130 cur, range_end + 1 - cur)) {
8131 btrfs_finish_ordered_io(ordered);
8133 * The ordered extent has finished, now we're again
8134 * safe to delete all extent states of the range.
8136 extra_flags = EXTENT_CLEAR_ALL_BITS;
8140 btrfs_put_ordered_extent(ordered);
8142 * Qgroup reserved space handler
8143 * Sector(s) here will be either:
8145 * 1) Already written to disk or bio already finished
8146 * Then its QGROUP_RESERVED bit in io_tree is already cleared.
8147 * Qgroup will be handled by its qgroup_record then.
8148 * btrfs_qgroup_free_data() call will do nothing here.
8150 * 2) Not written to disk yet
8151 * Then btrfs_qgroup_free_data() call will clear the
8152 * QGROUP_RESERVED bit of its io_tree, and free the qgroup
8153 * reserved data space.
8154 * Since the IO will never happen for this page.
8156 btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur, NULL);
8157 if (!inode_evicting) {
8158 clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED |
8159 EXTENT_DELALLOC | EXTENT_UPTODATE |
8160 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG |
8161 extra_flags, &cached_state);
8163 cur = range_end + 1;
8166 * We have iterated through all ordered extents of the page, the page
8167 * should not have Ordered (Private2) anymore, or the above iteration
8168 * did something wrong.
8170 ASSERT(!folio_test_ordered(folio));
8171 btrfs_folio_clear_checked(fs_info, folio, folio_pos(folio), folio_size(folio));
8172 if (!inode_evicting)
8173 __btrfs_release_folio(folio, GFP_NOFS);
8174 clear_page_extent_mapped(&folio->page);
8178 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8179 * called from a page fault handler when a page is first dirtied. Hence we must
8180 * be careful to check for EOF conditions here. We set the page up correctly
8181 * for a written page which means we get ENOSPC checking when writing into
8182 * holes and correct delalloc and unwritten extent mapping on filesystems that
8183 * support these features.
8185 * We are not allowed to take the i_mutex here so we have to play games to
8186 * protect against truncate races as the page could now be beyond EOF. Because
8187 * truncate_setsize() writes the inode size before removing pages, once we have
8188 * the page lock we can determine safely if the page is beyond EOF. If it is not
8189 * beyond EOF, then the page is guaranteed safe against truncation until we
8192 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8194 struct page *page = vmf->page;
8195 struct folio *folio = page_folio(page);
8196 struct inode *inode = file_inode(vmf->vma->vm_file);
8197 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
8198 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8199 struct btrfs_ordered_extent *ordered;
8200 struct extent_state *cached_state = NULL;
8201 struct extent_changeset *data_reserved = NULL;
8202 unsigned long zero_start;
8212 ASSERT(folio_order(folio) == 0);
8214 reserved_space = PAGE_SIZE;
8216 sb_start_pagefault(inode->i_sb);
8217 page_start = page_offset(page);
8218 page_end = page_start + PAGE_SIZE - 1;
8222 * Reserving delalloc space after obtaining the page lock can lead to
8223 * deadlock. For example, if a dirty page is locked by this function
8224 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8225 * dirty page write out, then the btrfs_writepages() function could
8226 * end up waiting indefinitely to get a lock on the page currently
8227 * being processed by btrfs_page_mkwrite() function.
8229 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8230 page_start, reserved_space);
8232 ret2 = file_update_time(vmf->vma->vm_file);
8236 ret = vmf_error(ret2);
8242 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8244 down_read(&BTRFS_I(inode)->i_mmap_lock);
8246 size = i_size_read(inode);
8248 if ((page->mapping != inode->i_mapping) ||
8249 (page_start >= size)) {
8250 /* page got truncated out from underneath us */
8253 wait_on_page_writeback(page);
8255 lock_extent(io_tree, page_start, page_end, &cached_state);
8256 ret2 = set_page_extent_mapped(page);
8258 ret = vmf_error(ret2);
8259 unlock_extent(io_tree, page_start, page_end, &cached_state);
8264 * we can't set the delalloc bits if there are pending ordered
8265 * extents. Drop our locks and wait for them to finish
8267 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8270 unlock_extent(io_tree, page_start, page_end, &cached_state);
8272 up_read(&BTRFS_I(inode)->i_mmap_lock);
8273 btrfs_start_ordered_extent(ordered);
8274 btrfs_put_ordered_extent(ordered);
8278 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8279 reserved_space = round_up(size - page_start,
8280 fs_info->sectorsize);
8281 if (reserved_space < PAGE_SIZE) {
8282 end = page_start + reserved_space - 1;
8283 btrfs_delalloc_release_space(BTRFS_I(inode),
8284 data_reserved, page_start,
8285 PAGE_SIZE - reserved_space, true);
8290 * page_mkwrite gets called when the page is firstly dirtied after it's
8291 * faulted in, but write(2) could also dirty a page and set delalloc
8292 * bits, thus in this case for space account reason, we still need to
8293 * clear any delalloc bits within this page range since we have to
8294 * reserve data&meta space before lock_page() (see above comments).
8296 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8297 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8298 EXTENT_DEFRAG, &cached_state);
8300 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8303 unlock_extent(io_tree, page_start, page_end, &cached_state);
8304 ret = VM_FAULT_SIGBUS;
8308 /* page is wholly or partially inside EOF */
8309 if (page_start + PAGE_SIZE > size)
8310 zero_start = offset_in_page(size);
8312 zero_start = PAGE_SIZE;
8314 if (zero_start != PAGE_SIZE)
8315 memzero_page(page, zero_start, PAGE_SIZE - zero_start);
8317 btrfs_folio_clear_checked(fs_info, folio, page_start, PAGE_SIZE);
8318 btrfs_folio_set_dirty(fs_info, folio, page_start, end + 1 - page_start);
8319 btrfs_folio_set_uptodate(fs_info, folio, page_start, end + 1 - page_start);
8321 btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
8323 unlock_extent(io_tree, page_start, page_end, &cached_state);
8324 up_read(&BTRFS_I(inode)->i_mmap_lock);
8326 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8327 sb_end_pagefault(inode->i_sb);
8328 extent_changeset_free(data_reserved);
8329 return VM_FAULT_LOCKED;
8333 up_read(&BTRFS_I(inode)->i_mmap_lock);
8335 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8336 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8337 reserved_space, (ret != 0));
8339 sb_end_pagefault(inode->i_sb);
8340 extent_changeset_free(data_reserved);
8344 static int btrfs_truncate(struct btrfs_inode *inode, bool skip_writeback)
8346 struct btrfs_truncate_control control = {
8348 .ino = btrfs_ino(inode),
8349 .min_type = BTRFS_EXTENT_DATA_KEY,
8350 .clear_extent_range = true,
8352 struct btrfs_root *root = inode->root;
8353 struct btrfs_fs_info *fs_info = root->fs_info;
8354 struct btrfs_block_rsv *rsv;
8356 struct btrfs_trans_handle *trans;
8357 u64 mask = fs_info->sectorsize - 1;
8358 const u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8360 if (!skip_writeback) {
8361 ret = btrfs_wait_ordered_range(&inode->vfs_inode,
8362 inode->vfs_inode.i_size & (~mask),
8369 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8370 * things going on here:
8372 * 1) We need to reserve space to update our inode.
8374 * 2) We need to have something to cache all the space that is going to
8375 * be free'd up by the truncate operation, but also have some slack
8376 * space reserved in case it uses space during the truncate (thank you
8377 * very much snapshotting).
8379 * And we need these to be separate. The fact is we can use a lot of
8380 * space doing the truncate, and we have no earthly idea how much space
8381 * we will use, so we need the truncate reservation to be separate so it
8382 * doesn't end up using space reserved for updating the inode. We also
8383 * need to be able to stop the transaction and start a new one, which
8384 * means we need to be able to update the inode several times, and we
8385 * have no idea of knowing how many times that will be, so we can't just
8386 * reserve 1 item for the entirety of the operation, so that has to be
8387 * done separately as well.
8389 * So that leaves us with
8391 * 1) rsv - for the truncate reservation, which we will steal from the
8392 * transaction reservation.
8393 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8394 * updating the inode.
8396 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8399 rsv->size = min_size;
8400 rsv->failfast = true;
8403 * 1 for the truncate slack space
8404 * 1 for updating the inode.
8406 trans = btrfs_start_transaction(root, 2);
8407 if (IS_ERR(trans)) {
8408 ret = PTR_ERR(trans);
8412 /* Migrate the slack space for the truncate to our reserve */
8413 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8416 * We have reserved 2 metadata units when we started the transaction and
8417 * min_size matches 1 unit, so this should never fail, but if it does,
8418 * it's not critical we just fail truncation.
8421 btrfs_end_transaction(trans);
8425 trans->block_rsv = rsv;
8428 struct extent_state *cached_state = NULL;
8429 const u64 new_size = inode->vfs_inode.i_size;
8430 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
8432 control.new_size = new_size;
8433 lock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8435 * We want to drop from the next block forward in case this new
8436 * size is not block aligned since we will be keeping the last
8437 * block of the extent just the way it is.
8439 btrfs_drop_extent_map_range(inode,
8440 ALIGN(new_size, fs_info->sectorsize),
8443 ret = btrfs_truncate_inode_items(trans, root, &control);
8445 inode_sub_bytes(&inode->vfs_inode, control.sub_bytes);
8446 btrfs_inode_safe_disk_i_size_write(inode, control.last_size);
8448 unlock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8450 trans->block_rsv = &fs_info->trans_block_rsv;
8451 if (ret != -ENOSPC && ret != -EAGAIN)
8454 ret = btrfs_update_inode(trans, inode);
8458 btrfs_end_transaction(trans);
8459 btrfs_btree_balance_dirty(fs_info);
8461 trans = btrfs_start_transaction(root, 2);
8462 if (IS_ERR(trans)) {
8463 ret = PTR_ERR(trans);
8468 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8469 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8470 rsv, min_size, false);
8472 * We have reserved 2 metadata units when we started the
8473 * transaction and min_size matches 1 unit, so this should never
8474 * fail, but if it does, it's not critical we just fail truncation.
8479 trans->block_rsv = rsv;
8483 * We can't call btrfs_truncate_block inside a trans handle as we could
8484 * deadlock with freeze, if we got BTRFS_NEED_TRUNCATE_BLOCK then we
8485 * know we've truncated everything except the last little bit, and can
8486 * do btrfs_truncate_block and then update the disk_i_size.
8488 if (ret == BTRFS_NEED_TRUNCATE_BLOCK) {
8489 btrfs_end_transaction(trans);
8490 btrfs_btree_balance_dirty(fs_info);
8492 ret = btrfs_truncate_block(inode, inode->vfs_inode.i_size, 0, 0);
8495 trans = btrfs_start_transaction(root, 1);
8496 if (IS_ERR(trans)) {
8497 ret = PTR_ERR(trans);
8500 btrfs_inode_safe_disk_i_size_write(inode, 0);
8506 trans->block_rsv = &fs_info->trans_block_rsv;
8507 ret2 = btrfs_update_inode(trans, inode);
8511 ret2 = btrfs_end_transaction(trans);
8514 btrfs_btree_balance_dirty(fs_info);
8517 btrfs_free_block_rsv(fs_info, rsv);
8519 * So if we truncate and then write and fsync we normally would just
8520 * write the extents that changed, which is a problem if we need to
8521 * first truncate that entire inode. So set this flag so we write out
8522 * all of the extents in the inode to the sync log so we're completely
8525 * If no extents were dropped or trimmed we don't need to force the next
8526 * fsync to truncate all the inode's items from the log and re-log them
8527 * all. This means the truncate operation did not change the file size,
8528 * or changed it to a smaller size but there was only an implicit hole
8529 * between the old i_size and the new i_size, and there were no prealloc
8530 * extents beyond i_size to drop.
8532 if (control.extents_found > 0)
8533 btrfs_set_inode_full_sync(inode);
8538 struct inode *btrfs_new_subvol_inode(struct mnt_idmap *idmap,
8541 struct inode *inode;
8543 inode = new_inode(dir->i_sb);
8546 * Subvolumes don't inherit the sgid bit or the parent's gid if
8547 * the parent's sgid bit is set. This is probably a bug.
8549 inode_init_owner(idmap, inode, NULL,
8550 S_IFDIR | (~current_umask() & S_IRWXUGO));
8551 inode->i_op = &btrfs_dir_inode_operations;
8552 inode->i_fop = &btrfs_dir_file_operations;
8557 struct inode *btrfs_alloc_inode(struct super_block *sb)
8559 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8560 struct btrfs_inode *ei;
8561 struct inode *inode;
8562 struct extent_io_tree *file_extent_tree = NULL;
8564 /* Self tests may pass a NULL fs_info. */
8565 if (fs_info && !btrfs_fs_incompat(fs_info, NO_HOLES)) {
8566 file_extent_tree = kmalloc(sizeof(struct extent_io_tree), GFP_KERNEL);
8567 if (!file_extent_tree)
8571 ei = alloc_inode_sb(sb, btrfs_inode_cachep, GFP_KERNEL);
8573 kfree(file_extent_tree);
8580 ei->last_sub_trans = 0;
8581 ei->logged_trans = 0;
8582 ei->delalloc_bytes = 0;
8583 ei->new_delalloc_bytes = 0;
8584 ei->defrag_bytes = 0;
8585 ei->disk_i_size = 0;
8589 ei->index_cnt = (u64)-1;
8591 ei->last_unlink_trans = 0;
8592 ei->last_reflink_trans = 0;
8593 ei->last_log_commit = 0;
8595 spin_lock_init(&ei->lock);
8596 ei->outstanding_extents = 0;
8597 if (sb->s_magic != BTRFS_TEST_MAGIC)
8598 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8599 BTRFS_BLOCK_RSV_DELALLOC);
8600 ei->runtime_flags = 0;
8601 ei->prop_compress = BTRFS_COMPRESS_NONE;
8602 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8604 ei->delayed_node = NULL;
8606 ei->i_otime_sec = 0;
8607 ei->i_otime_nsec = 0;
8609 inode = &ei->vfs_inode;
8610 extent_map_tree_init(&ei->extent_tree);
8612 /* This io tree sets the valid inode. */
8613 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO);
8614 ei->io_tree.inode = ei;
8616 ei->file_extent_tree = file_extent_tree;
8617 if (file_extent_tree) {
8618 extent_io_tree_init(fs_info, ei->file_extent_tree,
8619 IO_TREE_INODE_FILE_EXTENT);
8620 /* Lockdep class is set only for the file extent tree. */
8621 lockdep_set_class(&ei->file_extent_tree->lock, &file_extent_tree_class);
8623 mutex_init(&ei->log_mutex);
8624 spin_lock_init(&ei->ordered_tree_lock);
8625 ei->ordered_tree = RB_ROOT;
8626 ei->ordered_tree_last = NULL;
8627 INIT_LIST_HEAD(&ei->delalloc_inodes);
8628 INIT_LIST_HEAD(&ei->delayed_iput);
8629 RB_CLEAR_NODE(&ei->rb_node);
8630 init_rwsem(&ei->i_mmap_lock);
8635 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8636 void btrfs_test_destroy_inode(struct inode *inode)
8638 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
8639 kfree(BTRFS_I(inode)->file_extent_tree);
8640 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8644 void btrfs_free_inode(struct inode *inode)
8646 kfree(BTRFS_I(inode)->file_extent_tree);
8647 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8650 void btrfs_destroy_inode(struct inode *vfs_inode)
8652 struct btrfs_ordered_extent *ordered;
8653 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8654 struct btrfs_root *root = inode->root;
8655 bool freespace_inode;
8657 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8658 WARN_ON(vfs_inode->i_data.nrpages);
8659 WARN_ON(inode->block_rsv.reserved);
8660 WARN_ON(inode->block_rsv.size);
8661 WARN_ON(inode->outstanding_extents);
8662 if (!S_ISDIR(vfs_inode->i_mode)) {
8663 WARN_ON(inode->delalloc_bytes);
8664 WARN_ON(inode->new_delalloc_bytes);
8666 WARN_ON(inode->csum_bytes);
8667 WARN_ON(inode->defrag_bytes);
8670 * This can happen where we create an inode, but somebody else also
8671 * created the same inode and we need to destroy the one we already
8678 * If this is a free space inode do not take the ordered extents lockdep
8681 freespace_inode = btrfs_is_free_space_inode(inode);
8684 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8688 btrfs_err(root->fs_info,
8689 "found ordered extent %llu %llu on inode cleanup",
8690 ordered->file_offset, ordered->num_bytes);
8692 if (!freespace_inode)
8693 btrfs_lockdep_acquire(root->fs_info, btrfs_ordered_extent);
8695 btrfs_remove_ordered_extent(inode, ordered);
8696 btrfs_put_ordered_extent(ordered);
8697 btrfs_put_ordered_extent(ordered);
8700 btrfs_qgroup_check_reserved_leak(inode);
8701 inode_tree_del(inode);
8702 btrfs_drop_extent_map_range(inode, 0, (u64)-1, false);
8703 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
8704 btrfs_put_root(inode->root);
8707 int btrfs_drop_inode(struct inode *inode)
8709 struct btrfs_root *root = BTRFS_I(inode)->root;
8714 /* the snap/subvol tree is on deleting */
8715 if (btrfs_root_refs(&root->root_item) == 0)
8718 return generic_drop_inode(inode);
8721 static void init_once(void *foo)
8723 struct btrfs_inode *ei = foo;
8725 inode_init_once(&ei->vfs_inode);
8728 void __cold btrfs_destroy_cachep(void)
8731 * Make sure all delayed rcu free inodes are flushed before we
8735 bioset_exit(&btrfs_dio_bioset);
8736 kmem_cache_destroy(btrfs_inode_cachep);
8739 int __init btrfs_init_cachep(void)
8741 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8742 sizeof(struct btrfs_inode), 0,
8743 SLAB_RECLAIM_ACCOUNT | SLAB_ACCOUNT,
8745 if (!btrfs_inode_cachep)
8748 if (bioset_init(&btrfs_dio_bioset, BIO_POOL_SIZE,
8749 offsetof(struct btrfs_dio_private, bbio.bio),
8755 btrfs_destroy_cachep();
8759 static int btrfs_getattr(struct mnt_idmap *idmap,
8760 const struct path *path, struct kstat *stat,
8761 u32 request_mask, unsigned int flags)
8765 struct inode *inode = d_inode(path->dentry);
8766 u32 blocksize = btrfs_sb(inode->i_sb)->sectorsize;
8767 u32 bi_flags = BTRFS_I(inode)->flags;
8768 u32 bi_ro_flags = BTRFS_I(inode)->ro_flags;
8770 stat->result_mask |= STATX_BTIME;
8771 stat->btime.tv_sec = BTRFS_I(inode)->i_otime_sec;
8772 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime_nsec;
8773 if (bi_flags & BTRFS_INODE_APPEND)
8774 stat->attributes |= STATX_ATTR_APPEND;
8775 if (bi_flags & BTRFS_INODE_COMPRESS)
8776 stat->attributes |= STATX_ATTR_COMPRESSED;
8777 if (bi_flags & BTRFS_INODE_IMMUTABLE)
8778 stat->attributes |= STATX_ATTR_IMMUTABLE;
8779 if (bi_flags & BTRFS_INODE_NODUMP)
8780 stat->attributes |= STATX_ATTR_NODUMP;
8781 if (bi_ro_flags & BTRFS_INODE_RO_VERITY)
8782 stat->attributes |= STATX_ATTR_VERITY;
8784 stat->attributes_mask |= (STATX_ATTR_APPEND |
8785 STATX_ATTR_COMPRESSED |
8786 STATX_ATTR_IMMUTABLE |
8789 generic_fillattr(idmap, request_mask, inode, stat);
8790 stat->dev = BTRFS_I(inode)->root->anon_dev;
8792 spin_lock(&BTRFS_I(inode)->lock);
8793 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
8794 inode_bytes = inode_get_bytes(inode);
8795 spin_unlock(&BTRFS_I(inode)->lock);
8796 stat->blocks = (ALIGN(inode_bytes, blocksize) +
8797 ALIGN(delalloc_bytes, blocksize)) >> SECTOR_SHIFT;
8801 static int btrfs_rename_exchange(struct inode *old_dir,
8802 struct dentry *old_dentry,
8803 struct inode *new_dir,
8804 struct dentry *new_dentry)
8806 struct btrfs_fs_info *fs_info = inode_to_fs_info(old_dir);
8807 struct btrfs_trans_handle *trans;
8808 unsigned int trans_num_items;
8809 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8810 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8811 struct inode *new_inode = new_dentry->d_inode;
8812 struct inode *old_inode = old_dentry->d_inode;
8813 struct btrfs_rename_ctx old_rename_ctx;
8814 struct btrfs_rename_ctx new_rename_ctx;
8815 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
8816 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
8821 bool need_abort = false;
8822 struct fscrypt_name old_fname, new_fname;
8823 struct fscrypt_str *old_name, *new_name;
8826 * For non-subvolumes allow exchange only within one subvolume, in the
8827 * same inode namespace. Two subvolumes (represented as directory) can
8828 * be exchanged as they're a logical link and have a fixed inode number.
8831 (old_ino != BTRFS_FIRST_FREE_OBJECTID ||
8832 new_ino != BTRFS_FIRST_FREE_OBJECTID))
8835 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
8839 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
8841 fscrypt_free_filename(&old_fname);
8845 old_name = &old_fname.disk_name;
8846 new_name = &new_fname.disk_name;
8848 /* close the race window with snapshot create/destroy ioctl */
8849 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
8850 new_ino == BTRFS_FIRST_FREE_OBJECTID)
8851 down_read(&fs_info->subvol_sem);
8855 * 1 to remove old dir item
8856 * 1 to remove old dir index
8857 * 1 to add new dir item
8858 * 1 to add new dir index
8859 * 1 to update parent inode
8861 * If the parents are the same, we only need to account for one
8863 trans_num_items = (old_dir == new_dir ? 9 : 10);
8864 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8866 * 1 to remove old root ref
8867 * 1 to remove old root backref
8868 * 1 to add new root ref
8869 * 1 to add new root backref
8871 trans_num_items += 4;
8874 * 1 to update inode item
8875 * 1 to remove old inode ref
8876 * 1 to add new inode ref
8878 trans_num_items += 3;
8880 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
8881 trans_num_items += 4;
8883 trans_num_items += 3;
8884 trans = btrfs_start_transaction(root, trans_num_items);
8885 if (IS_ERR(trans)) {
8886 ret = PTR_ERR(trans);
8891 ret = btrfs_record_root_in_trans(trans, dest);
8897 * We need to find a free sequence number both in the source and
8898 * in the destination directory for the exchange.
8900 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
8903 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
8907 BTRFS_I(old_inode)->dir_index = 0ULL;
8908 BTRFS_I(new_inode)->dir_index = 0ULL;
8910 /* Reference for the source. */
8911 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8912 /* force full log commit if subvolume involved. */
8913 btrfs_set_log_full_commit(trans);
8915 ret = btrfs_insert_inode_ref(trans, dest, new_name, old_ino,
8916 btrfs_ino(BTRFS_I(new_dir)),
8923 /* And now for the dest. */
8924 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8925 /* force full log commit if subvolume involved. */
8926 btrfs_set_log_full_commit(trans);
8928 ret = btrfs_insert_inode_ref(trans, root, old_name, new_ino,
8929 btrfs_ino(BTRFS_I(old_dir)),
8933 btrfs_abort_transaction(trans, ret);
8938 /* Update inode version and ctime/mtime. */
8939 inode_inc_iversion(old_dir);
8940 inode_inc_iversion(new_dir);
8941 inode_inc_iversion(old_inode);
8942 inode_inc_iversion(new_inode);
8943 simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
8945 if (old_dentry->d_parent != new_dentry->d_parent) {
8946 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
8947 BTRFS_I(old_inode), true);
8948 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
8949 BTRFS_I(new_inode), true);
8952 /* src is a subvolume */
8953 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8954 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
8955 } else { /* src is an inode */
8956 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
8957 BTRFS_I(old_dentry->d_inode),
8958 old_name, &old_rename_ctx);
8960 ret = btrfs_update_inode(trans, BTRFS_I(old_inode));
8963 btrfs_abort_transaction(trans, ret);
8967 /* dest is a subvolume */
8968 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8969 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
8970 } else { /* dest is an inode */
8971 ret = __btrfs_unlink_inode(trans, BTRFS_I(new_dir),
8972 BTRFS_I(new_dentry->d_inode),
8973 new_name, &new_rename_ctx);
8975 ret = btrfs_update_inode(trans, BTRFS_I(new_inode));
8978 btrfs_abort_transaction(trans, ret);
8982 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
8983 new_name, 0, old_idx);
8985 btrfs_abort_transaction(trans, ret);
8989 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
8990 old_name, 0, new_idx);
8992 btrfs_abort_transaction(trans, ret);
8996 if (old_inode->i_nlink == 1)
8997 BTRFS_I(old_inode)->dir_index = old_idx;
8998 if (new_inode->i_nlink == 1)
8999 BTRFS_I(new_inode)->dir_index = new_idx;
9002 * Now pin the logs of the roots. We do it to ensure that no other task
9003 * can sync the logs while we are in progress with the rename, because
9004 * that could result in an inconsistency in case any of the inodes that
9005 * are part of this rename operation were logged before.
9007 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9008 btrfs_pin_log_trans(root);
9009 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9010 btrfs_pin_log_trans(dest);
9012 /* Do the log updates for all inodes. */
9013 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9014 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9015 old_rename_ctx.index, new_dentry->d_parent);
9016 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9017 btrfs_log_new_name(trans, new_dentry, BTRFS_I(new_dir),
9018 new_rename_ctx.index, old_dentry->d_parent);
9020 /* Now unpin the logs. */
9021 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9022 btrfs_end_log_trans(root);
9023 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9024 btrfs_end_log_trans(dest);
9026 ret2 = btrfs_end_transaction(trans);
9027 ret = ret ? ret : ret2;
9029 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
9030 old_ino == BTRFS_FIRST_FREE_OBJECTID)
9031 up_read(&fs_info->subvol_sem);
9033 fscrypt_free_filename(&new_fname);
9034 fscrypt_free_filename(&old_fname);
9038 static struct inode *new_whiteout_inode(struct mnt_idmap *idmap,
9041 struct inode *inode;
9043 inode = new_inode(dir->i_sb);
9045 inode_init_owner(idmap, inode, dir,
9046 S_IFCHR | WHITEOUT_MODE);
9047 inode->i_op = &btrfs_special_inode_operations;
9048 init_special_inode(inode, inode->i_mode, WHITEOUT_DEV);
9053 static int btrfs_rename(struct mnt_idmap *idmap,
9054 struct inode *old_dir, struct dentry *old_dentry,
9055 struct inode *new_dir, struct dentry *new_dentry,
9058 struct btrfs_fs_info *fs_info = inode_to_fs_info(old_dir);
9059 struct btrfs_new_inode_args whiteout_args = {
9061 .dentry = old_dentry,
9063 struct btrfs_trans_handle *trans;
9064 unsigned int trans_num_items;
9065 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9066 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9067 struct inode *new_inode = d_inode(new_dentry);
9068 struct inode *old_inode = d_inode(old_dentry);
9069 struct btrfs_rename_ctx rename_ctx;
9073 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9074 struct fscrypt_name old_fname, new_fname;
9076 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9079 /* we only allow rename subvolume link between subvolumes */
9080 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9083 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9084 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9087 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9088 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9091 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
9095 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
9097 fscrypt_free_filename(&old_fname);
9101 /* check for collisions, even if the name isn't there */
9102 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, &new_fname.disk_name);
9104 if (ret == -EEXIST) {
9106 * eexist without a new_inode */
9107 if (WARN_ON(!new_inode)) {
9108 goto out_fscrypt_names;
9111 /* maybe -EOVERFLOW */
9112 goto out_fscrypt_names;
9118 * we're using rename to replace one file with another. Start IO on it
9119 * now so we don't add too much work to the end of the transaction
9121 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9122 filemap_flush(old_inode->i_mapping);
9124 if (flags & RENAME_WHITEOUT) {
9125 whiteout_args.inode = new_whiteout_inode(idmap, old_dir);
9126 if (!whiteout_args.inode) {
9128 goto out_fscrypt_names;
9130 ret = btrfs_new_inode_prepare(&whiteout_args, &trans_num_items);
9132 goto out_whiteout_inode;
9134 /* 1 to update the old parent inode. */
9135 trans_num_items = 1;
9138 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9139 /* Close the race window with snapshot create/destroy ioctl */
9140 down_read(&fs_info->subvol_sem);
9142 * 1 to remove old root ref
9143 * 1 to remove old root backref
9144 * 1 to add new root ref
9145 * 1 to add new root backref
9147 trans_num_items += 4;
9151 * 1 to remove old inode ref
9152 * 1 to add new inode ref
9154 trans_num_items += 3;
9157 * 1 to remove old dir item
9158 * 1 to remove old dir index
9159 * 1 to add new dir item
9160 * 1 to add new dir index
9162 trans_num_items += 4;
9163 /* 1 to update new parent inode if it's not the same as the old parent */
9164 if (new_dir != old_dir)
9169 * 1 to remove inode ref
9170 * 1 to remove dir item
9171 * 1 to remove dir index
9172 * 1 to possibly add orphan item
9174 trans_num_items += 5;
9176 trans = btrfs_start_transaction(root, trans_num_items);
9177 if (IS_ERR(trans)) {
9178 ret = PTR_ERR(trans);
9183 ret = btrfs_record_root_in_trans(trans, dest);
9188 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9192 BTRFS_I(old_inode)->dir_index = 0ULL;
9193 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9194 /* force full log commit if subvolume involved. */
9195 btrfs_set_log_full_commit(trans);
9197 ret = btrfs_insert_inode_ref(trans, dest, &new_fname.disk_name,
9198 old_ino, btrfs_ino(BTRFS_I(new_dir)),
9204 inode_inc_iversion(old_dir);
9205 inode_inc_iversion(new_dir);
9206 inode_inc_iversion(old_inode);
9207 simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
9209 if (old_dentry->d_parent != new_dentry->d_parent)
9210 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9211 BTRFS_I(old_inode), true);
9213 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9214 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
9216 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9217 BTRFS_I(d_inode(old_dentry)),
9218 &old_fname.disk_name, &rename_ctx);
9220 ret = btrfs_update_inode(trans, BTRFS_I(old_inode));
9223 btrfs_abort_transaction(trans, ret);
9228 inode_inc_iversion(new_inode);
9229 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9230 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9231 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
9232 BUG_ON(new_inode->i_nlink == 0);
9234 ret = btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9235 BTRFS_I(d_inode(new_dentry)),
9236 &new_fname.disk_name);
9238 if (!ret && new_inode->i_nlink == 0)
9239 ret = btrfs_orphan_add(trans,
9240 BTRFS_I(d_inode(new_dentry)));
9242 btrfs_abort_transaction(trans, ret);
9247 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9248 &new_fname.disk_name, 0, index);
9250 btrfs_abort_transaction(trans, ret);
9254 if (old_inode->i_nlink == 1)
9255 BTRFS_I(old_inode)->dir_index = index;
9257 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9258 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9259 rename_ctx.index, new_dentry->d_parent);
9261 if (flags & RENAME_WHITEOUT) {
9262 ret = btrfs_create_new_inode(trans, &whiteout_args);
9264 btrfs_abort_transaction(trans, ret);
9267 unlock_new_inode(whiteout_args.inode);
9268 iput(whiteout_args.inode);
9269 whiteout_args.inode = NULL;
9273 ret2 = btrfs_end_transaction(trans);
9274 ret = ret ? ret : ret2;
9276 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9277 up_read(&fs_info->subvol_sem);
9278 if (flags & RENAME_WHITEOUT)
9279 btrfs_new_inode_args_destroy(&whiteout_args);
9281 if (flags & RENAME_WHITEOUT)
9282 iput(whiteout_args.inode);
9284 fscrypt_free_filename(&old_fname);
9285 fscrypt_free_filename(&new_fname);
9289 static int btrfs_rename2(struct mnt_idmap *idmap, struct inode *old_dir,
9290 struct dentry *old_dentry, struct inode *new_dir,
9291 struct dentry *new_dentry, unsigned int flags)
9295 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9298 if (flags & RENAME_EXCHANGE)
9299 ret = btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9302 ret = btrfs_rename(idmap, old_dir, old_dentry, new_dir,
9305 btrfs_btree_balance_dirty(BTRFS_I(new_dir)->root->fs_info);
9310 struct btrfs_delalloc_work {
9311 struct inode *inode;
9312 struct completion completion;
9313 struct list_head list;
9314 struct btrfs_work work;
9317 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9319 struct btrfs_delalloc_work *delalloc_work;
9320 struct inode *inode;
9322 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9324 inode = delalloc_work->inode;
9325 filemap_flush(inode->i_mapping);
9326 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9327 &BTRFS_I(inode)->runtime_flags))
9328 filemap_flush(inode->i_mapping);
9331 complete(&delalloc_work->completion);
9334 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9336 struct btrfs_delalloc_work *work;
9338 work = kmalloc(sizeof(*work), GFP_NOFS);
9342 init_completion(&work->completion);
9343 INIT_LIST_HEAD(&work->list);
9344 work->inode = inode;
9345 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL);
9351 * some fairly slow code that needs optimization. This walks the list
9352 * of all the inodes with pending delalloc and forces them to disk.
9354 static int start_delalloc_inodes(struct btrfs_root *root,
9355 struct writeback_control *wbc, bool snapshot,
9356 bool in_reclaim_context)
9358 struct btrfs_inode *binode;
9359 struct inode *inode;
9360 struct btrfs_delalloc_work *work, *next;
9364 bool full_flush = wbc->nr_to_write == LONG_MAX;
9366 mutex_lock(&root->delalloc_mutex);
9367 spin_lock(&root->delalloc_lock);
9368 list_splice_init(&root->delalloc_inodes, &splice);
9369 while (!list_empty(&splice)) {
9370 binode = list_entry(splice.next, struct btrfs_inode,
9373 list_move_tail(&binode->delalloc_inodes,
9374 &root->delalloc_inodes);
9376 if (in_reclaim_context &&
9377 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9380 inode = igrab(&binode->vfs_inode);
9382 cond_resched_lock(&root->delalloc_lock);
9385 spin_unlock(&root->delalloc_lock);
9388 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9389 &binode->runtime_flags);
9391 work = btrfs_alloc_delalloc_work(inode);
9397 list_add_tail(&work->list, &works);
9398 btrfs_queue_work(root->fs_info->flush_workers,
9401 ret = filemap_fdatawrite_wbc(inode->i_mapping, wbc);
9402 btrfs_add_delayed_iput(BTRFS_I(inode));
9403 if (ret || wbc->nr_to_write <= 0)
9407 spin_lock(&root->delalloc_lock);
9409 spin_unlock(&root->delalloc_lock);
9412 list_for_each_entry_safe(work, next, &works, list) {
9413 list_del_init(&work->list);
9414 wait_for_completion(&work->completion);
9418 if (!list_empty(&splice)) {
9419 spin_lock(&root->delalloc_lock);
9420 list_splice_tail(&splice, &root->delalloc_inodes);
9421 spin_unlock(&root->delalloc_lock);
9423 mutex_unlock(&root->delalloc_mutex);
9427 int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9429 struct writeback_control wbc = {
9430 .nr_to_write = LONG_MAX,
9431 .sync_mode = WB_SYNC_NONE,
9433 .range_end = LLONG_MAX,
9435 struct btrfs_fs_info *fs_info = root->fs_info;
9437 if (BTRFS_FS_ERROR(fs_info))
9440 return start_delalloc_inodes(root, &wbc, true, in_reclaim_context);
9443 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9444 bool in_reclaim_context)
9446 struct writeback_control wbc = {
9448 .sync_mode = WB_SYNC_NONE,
9450 .range_end = LLONG_MAX,
9452 struct btrfs_root *root;
9456 if (BTRFS_FS_ERROR(fs_info))
9459 mutex_lock(&fs_info->delalloc_root_mutex);
9460 spin_lock(&fs_info->delalloc_root_lock);
9461 list_splice_init(&fs_info->delalloc_roots, &splice);
9462 while (!list_empty(&splice)) {
9464 * Reset nr_to_write here so we know that we're doing a full
9468 wbc.nr_to_write = LONG_MAX;
9470 root = list_first_entry(&splice, struct btrfs_root,
9472 root = btrfs_grab_root(root);
9474 list_move_tail(&root->delalloc_root,
9475 &fs_info->delalloc_roots);
9476 spin_unlock(&fs_info->delalloc_root_lock);
9478 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9479 btrfs_put_root(root);
9480 if (ret < 0 || wbc.nr_to_write <= 0)
9482 spin_lock(&fs_info->delalloc_root_lock);
9484 spin_unlock(&fs_info->delalloc_root_lock);
9488 if (!list_empty(&splice)) {
9489 spin_lock(&fs_info->delalloc_root_lock);
9490 list_splice_tail(&splice, &fs_info->delalloc_roots);
9491 spin_unlock(&fs_info->delalloc_root_lock);
9493 mutex_unlock(&fs_info->delalloc_root_mutex);
9497 static int btrfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
9498 struct dentry *dentry, const char *symname)
9500 struct btrfs_fs_info *fs_info = inode_to_fs_info(dir);
9501 struct btrfs_trans_handle *trans;
9502 struct btrfs_root *root = BTRFS_I(dir)->root;
9503 struct btrfs_path *path;
9504 struct btrfs_key key;
9505 struct inode *inode;
9506 struct btrfs_new_inode_args new_inode_args = {
9510 unsigned int trans_num_items;
9515 struct btrfs_file_extent_item *ei;
9516 struct extent_buffer *leaf;
9518 name_len = strlen(symname);
9519 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9520 return -ENAMETOOLONG;
9522 inode = new_inode(dir->i_sb);
9525 inode_init_owner(idmap, inode, dir, S_IFLNK | S_IRWXUGO);
9526 inode->i_op = &btrfs_symlink_inode_operations;
9527 inode_nohighmem(inode);
9528 inode->i_mapping->a_ops = &btrfs_aops;
9529 btrfs_i_size_write(BTRFS_I(inode), name_len);
9530 inode_set_bytes(inode, name_len);
9532 new_inode_args.inode = inode;
9533 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9536 /* 1 additional item for the inline extent */
9539 trans = btrfs_start_transaction(root, trans_num_items);
9540 if (IS_ERR(trans)) {
9541 err = PTR_ERR(trans);
9542 goto out_new_inode_args;
9545 err = btrfs_create_new_inode(trans, &new_inode_args);
9549 path = btrfs_alloc_path();
9552 btrfs_abort_transaction(trans, err);
9553 discard_new_inode(inode);
9557 key.objectid = btrfs_ino(BTRFS_I(inode));
9559 key.type = BTRFS_EXTENT_DATA_KEY;
9560 datasize = btrfs_file_extent_calc_inline_size(name_len);
9561 err = btrfs_insert_empty_item(trans, root, path, &key,
9564 btrfs_abort_transaction(trans, err);
9565 btrfs_free_path(path);
9566 discard_new_inode(inode);
9570 leaf = path->nodes[0];
9571 ei = btrfs_item_ptr(leaf, path->slots[0],
9572 struct btrfs_file_extent_item);
9573 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9574 btrfs_set_file_extent_type(leaf, ei,
9575 BTRFS_FILE_EXTENT_INLINE);
9576 btrfs_set_file_extent_encryption(leaf, ei, 0);
9577 btrfs_set_file_extent_compression(leaf, ei, 0);
9578 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9579 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9581 ptr = btrfs_file_extent_inline_start(ei);
9582 write_extent_buffer(leaf, symname, ptr, name_len);
9583 btrfs_mark_buffer_dirty(trans, leaf);
9584 btrfs_free_path(path);
9586 d_instantiate_new(dentry, inode);
9589 btrfs_end_transaction(trans);
9590 btrfs_btree_balance_dirty(fs_info);
9592 btrfs_new_inode_args_destroy(&new_inode_args);
9599 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9600 struct btrfs_trans_handle *trans_in,
9601 struct btrfs_inode *inode,
9602 struct btrfs_key *ins,
9605 struct btrfs_file_extent_item stack_fi;
9606 struct btrfs_replace_extent_info extent_info;
9607 struct btrfs_trans_handle *trans = trans_in;
9608 struct btrfs_path *path;
9609 u64 start = ins->objectid;
9610 u64 len = ins->offset;
9611 u64 qgroup_released = 0;
9614 memset(&stack_fi, 0, sizeof(stack_fi));
9616 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9617 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9618 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9619 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9620 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9621 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9622 /* Encryption and other encoding is reserved and all 0 */
9624 ret = btrfs_qgroup_release_data(inode, file_offset, len, &qgroup_released);
9626 return ERR_PTR(ret);
9629 ret = insert_reserved_file_extent(trans, inode,
9630 file_offset, &stack_fi,
9631 true, qgroup_released);
9637 extent_info.disk_offset = start;
9638 extent_info.disk_len = len;
9639 extent_info.data_offset = 0;
9640 extent_info.data_len = len;
9641 extent_info.file_offset = file_offset;
9642 extent_info.extent_buf = (char *)&stack_fi;
9643 extent_info.is_new_extent = true;
9644 extent_info.update_times = true;
9645 extent_info.qgroup_reserved = qgroup_released;
9646 extent_info.insertions = 0;
9648 path = btrfs_alloc_path();
9654 ret = btrfs_replace_file_extents(inode, path, file_offset,
9655 file_offset + len - 1, &extent_info,
9657 btrfs_free_path(path);
9664 * We have released qgroup data range at the beginning of the function,
9665 * and normally qgroup_released bytes will be freed when committing
9667 * But if we error out early, we have to free what we have released
9668 * or we leak qgroup data reservation.
9670 btrfs_qgroup_free_refroot(inode->root->fs_info,
9671 inode->root->root_key.objectid, qgroup_released,
9672 BTRFS_QGROUP_RSV_DATA);
9673 return ERR_PTR(ret);
9676 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9677 u64 start, u64 num_bytes, u64 min_size,
9678 loff_t actual_len, u64 *alloc_hint,
9679 struct btrfs_trans_handle *trans)
9681 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
9682 struct extent_map *em;
9683 struct btrfs_root *root = BTRFS_I(inode)->root;
9684 struct btrfs_key ins;
9685 u64 cur_offset = start;
9686 u64 clear_offset = start;
9689 u64 last_alloc = (u64)-1;
9691 bool own_trans = true;
9692 u64 end = start + num_bytes - 1;
9696 while (num_bytes > 0) {
9697 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9698 cur_bytes = max(cur_bytes, min_size);
9700 * If we are severely fragmented we could end up with really
9701 * small allocations, so if the allocator is returning small
9702 * chunks lets make its job easier by only searching for those
9705 cur_bytes = min(cur_bytes, last_alloc);
9706 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9707 min_size, 0, *alloc_hint, &ins, 1, 0);
9712 * We've reserved this space, and thus converted it from
9713 * ->bytes_may_use to ->bytes_reserved. Any error that happens
9714 * from here on out we will only need to clear our reservation
9715 * for the remaining unreserved area, so advance our
9716 * clear_offset by our extent size.
9718 clear_offset += ins.offset;
9720 last_alloc = ins.offset;
9721 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
9724 * Now that we inserted the prealloc extent we can finally
9725 * decrement the number of reservations in the block group.
9726 * If we did it before, we could race with relocation and have
9727 * relocation miss the reserved extent, making it fail later.
9729 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
9730 if (IS_ERR(trans)) {
9731 ret = PTR_ERR(trans);
9732 btrfs_free_reserved_extent(fs_info, ins.objectid,
9737 em = alloc_extent_map();
9739 btrfs_drop_extent_map_range(BTRFS_I(inode), cur_offset,
9740 cur_offset + ins.offset - 1, false);
9741 btrfs_set_inode_full_sync(BTRFS_I(inode));
9745 em->start = cur_offset;
9746 em->orig_start = cur_offset;
9747 em->len = ins.offset;
9748 em->block_start = ins.objectid;
9749 em->block_len = ins.offset;
9750 em->orig_block_len = ins.offset;
9751 em->ram_bytes = ins.offset;
9752 em->flags |= EXTENT_FLAG_PREALLOC;
9753 em->generation = trans->transid;
9755 ret = btrfs_replace_extent_map_range(BTRFS_I(inode), em, true);
9756 free_extent_map(em);
9758 num_bytes -= ins.offset;
9759 cur_offset += ins.offset;
9760 *alloc_hint = ins.objectid + ins.offset;
9762 inode_inc_iversion(inode);
9763 inode_set_ctime_current(inode);
9764 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9765 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9766 (actual_len > inode->i_size) &&
9767 (cur_offset > inode->i_size)) {
9768 if (cur_offset > actual_len)
9769 i_size = actual_len;
9771 i_size = cur_offset;
9772 i_size_write(inode, i_size);
9773 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
9776 ret = btrfs_update_inode(trans, BTRFS_I(inode));
9779 btrfs_abort_transaction(trans, ret);
9781 btrfs_end_transaction(trans);
9786 btrfs_end_transaction(trans);
9790 if (clear_offset < end)
9791 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
9792 end - clear_offset + 1);
9796 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9797 u64 start, u64 num_bytes, u64 min_size,
9798 loff_t actual_len, u64 *alloc_hint)
9800 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9801 min_size, actual_len, alloc_hint,
9805 int btrfs_prealloc_file_range_trans(struct inode *inode,
9806 struct btrfs_trans_handle *trans, int mode,
9807 u64 start, u64 num_bytes, u64 min_size,
9808 loff_t actual_len, u64 *alloc_hint)
9810 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9811 min_size, actual_len, alloc_hint, trans);
9814 static int btrfs_permission(struct mnt_idmap *idmap,
9815 struct inode *inode, int mask)
9817 struct btrfs_root *root = BTRFS_I(inode)->root;
9818 umode_t mode = inode->i_mode;
9820 if (mask & MAY_WRITE &&
9821 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9822 if (btrfs_root_readonly(root))
9824 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9827 return generic_permission(idmap, inode, mask);
9830 static int btrfs_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
9831 struct file *file, umode_t mode)
9833 struct btrfs_fs_info *fs_info = inode_to_fs_info(dir);
9834 struct btrfs_trans_handle *trans;
9835 struct btrfs_root *root = BTRFS_I(dir)->root;
9836 struct inode *inode;
9837 struct btrfs_new_inode_args new_inode_args = {
9839 .dentry = file->f_path.dentry,
9842 unsigned int trans_num_items;
9845 inode = new_inode(dir->i_sb);
9848 inode_init_owner(idmap, inode, dir, mode);
9849 inode->i_fop = &btrfs_file_operations;
9850 inode->i_op = &btrfs_file_inode_operations;
9851 inode->i_mapping->a_ops = &btrfs_aops;
9853 new_inode_args.inode = inode;
9854 ret = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9858 trans = btrfs_start_transaction(root, trans_num_items);
9859 if (IS_ERR(trans)) {
9860 ret = PTR_ERR(trans);
9861 goto out_new_inode_args;
9864 ret = btrfs_create_new_inode(trans, &new_inode_args);
9867 * We set number of links to 0 in btrfs_create_new_inode(), and here we
9868 * set it to 1 because d_tmpfile() will issue a warning if the count is
9871 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9873 set_nlink(inode, 1);
9876 d_tmpfile(file, inode);
9877 unlock_new_inode(inode);
9878 mark_inode_dirty(inode);
9881 btrfs_end_transaction(trans);
9882 btrfs_btree_balance_dirty(fs_info);
9884 btrfs_new_inode_args_destroy(&new_inode_args);
9888 return finish_open_simple(file, ret);
9891 void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end)
9893 struct btrfs_fs_info *fs_info = inode->root->fs_info;
9894 unsigned long index = start >> PAGE_SHIFT;
9895 unsigned long end_index = end >> PAGE_SHIFT;
9899 ASSERT(end + 1 - start <= U32_MAX);
9900 len = end + 1 - start;
9901 while (index <= end_index) {
9902 page = find_get_page(inode->vfs_inode.i_mapping, index);
9903 ASSERT(page); /* Pages should be in the extent_io_tree */
9905 /* This is for data, which doesn't yet support larger folio. */
9906 ASSERT(folio_order(page_folio(page)) == 0);
9907 btrfs_folio_set_writeback(fs_info, page_folio(page), start, len);
9913 int btrfs_encoded_io_compression_from_extent(struct btrfs_fs_info *fs_info,
9916 switch (compress_type) {
9917 case BTRFS_COMPRESS_NONE:
9918 return BTRFS_ENCODED_IO_COMPRESSION_NONE;
9919 case BTRFS_COMPRESS_ZLIB:
9920 return BTRFS_ENCODED_IO_COMPRESSION_ZLIB;
9921 case BTRFS_COMPRESS_LZO:
9923 * The LZO format depends on the sector size. 64K is the maximum
9924 * sector size that we support.
9926 if (fs_info->sectorsize < SZ_4K || fs_info->sectorsize > SZ_64K)
9928 return BTRFS_ENCODED_IO_COMPRESSION_LZO_4K +
9929 (fs_info->sectorsize_bits - 12);
9930 case BTRFS_COMPRESS_ZSTD:
9931 return BTRFS_ENCODED_IO_COMPRESSION_ZSTD;
9937 static ssize_t btrfs_encoded_read_inline(
9939 struct iov_iter *iter, u64 start,
9941 struct extent_state **cached_state,
9942 u64 extent_start, size_t count,
9943 struct btrfs_ioctl_encoded_io_args *encoded,
9946 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
9947 struct btrfs_root *root = inode->root;
9948 struct btrfs_fs_info *fs_info = root->fs_info;
9949 struct extent_io_tree *io_tree = &inode->io_tree;
9950 struct btrfs_path *path;
9951 struct extent_buffer *leaf;
9952 struct btrfs_file_extent_item *item;
9958 path = btrfs_alloc_path();
9963 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
9967 /* The extent item disappeared? */
9972 leaf = path->nodes[0];
9973 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
9975 ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
9976 ptr = btrfs_file_extent_inline_start(item);
9978 encoded->len = min_t(u64, extent_start + ram_bytes,
9979 inode->vfs_inode.i_size) - iocb->ki_pos;
9980 ret = btrfs_encoded_io_compression_from_extent(fs_info,
9981 btrfs_file_extent_compression(leaf, item));
9984 encoded->compression = ret;
9985 if (encoded->compression) {
9988 inline_size = btrfs_file_extent_inline_item_len(leaf,
9990 if (inline_size > count) {
9994 count = inline_size;
9995 encoded->unencoded_len = ram_bytes;
9996 encoded->unencoded_offset = iocb->ki_pos - extent_start;
9998 count = min_t(u64, count, encoded->len);
9999 encoded->len = count;
10000 encoded->unencoded_len = count;
10001 ptr += iocb->ki_pos - extent_start;
10004 tmp = kmalloc(count, GFP_NOFS);
10009 read_extent_buffer(leaf, tmp, ptr, count);
10010 btrfs_release_path(path);
10011 unlock_extent(io_tree, start, lockend, cached_state);
10012 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10015 ret = copy_to_iter(tmp, count, iter);
10020 btrfs_free_path(path);
10024 struct btrfs_encoded_read_private {
10025 wait_queue_head_t wait;
10027 blk_status_t status;
10030 static void btrfs_encoded_read_endio(struct btrfs_bio *bbio)
10032 struct btrfs_encoded_read_private *priv = bbio->private;
10034 if (bbio->bio.bi_status) {
10036 * The memory barrier implied by the atomic_dec_return() here
10037 * pairs with the memory barrier implied by the
10038 * atomic_dec_return() or io_wait_event() in
10039 * btrfs_encoded_read_regular_fill_pages() to ensure that this
10040 * write is observed before the load of status in
10041 * btrfs_encoded_read_regular_fill_pages().
10043 WRITE_ONCE(priv->status, bbio->bio.bi_status);
10045 if (!atomic_dec_return(&priv->pending))
10046 wake_up(&priv->wait);
10047 bio_put(&bbio->bio);
10050 int btrfs_encoded_read_regular_fill_pages(struct btrfs_inode *inode,
10051 u64 file_offset, u64 disk_bytenr,
10052 u64 disk_io_size, struct page **pages)
10054 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10055 struct btrfs_encoded_read_private priv = {
10056 .pending = ATOMIC_INIT(1),
10058 unsigned long i = 0;
10059 struct btrfs_bio *bbio;
10061 init_waitqueue_head(&priv.wait);
10063 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
10064 btrfs_encoded_read_endio, &priv);
10065 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
10066 bbio->inode = inode;
10069 size_t bytes = min_t(u64, disk_io_size, PAGE_SIZE);
10071 if (bio_add_page(&bbio->bio, pages[i], bytes, 0) < bytes) {
10072 atomic_inc(&priv.pending);
10073 btrfs_submit_bio(bbio, 0);
10075 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
10076 btrfs_encoded_read_endio, &priv);
10077 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
10078 bbio->inode = inode;
10083 disk_bytenr += bytes;
10084 disk_io_size -= bytes;
10085 } while (disk_io_size);
10087 atomic_inc(&priv.pending);
10088 btrfs_submit_bio(bbio, 0);
10090 if (atomic_dec_return(&priv.pending))
10091 io_wait_event(priv.wait, !atomic_read(&priv.pending));
10092 /* See btrfs_encoded_read_endio() for ordering. */
10093 return blk_status_to_errno(READ_ONCE(priv.status));
10096 static ssize_t btrfs_encoded_read_regular(struct kiocb *iocb,
10097 struct iov_iter *iter,
10098 u64 start, u64 lockend,
10099 struct extent_state **cached_state,
10100 u64 disk_bytenr, u64 disk_io_size,
10101 size_t count, bool compressed,
10104 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10105 struct extent_io_tree *io_tree = &inode->io_tree;
10106 struct page **pages;
10107 unsigned long nr_pages, i;
10109 size_t page_offset;
10112 nr_pages = DIV_ROUND_UP(disk_io_size, PAGE_SIZE);
10113 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
10116 ret = btrfs_alloc_page_array(nr_pages, pages, 0);
10122 ret = btrfs_encoded_read_regular_fill_pages(inode, start, disk_bytenr,
10123 disk_io_size, pages);
10127 unlock_extent(io_tree, start, lockend, cached_state);
10128 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10135 i = (iocb->ki_pos - start) >> PAGE_SHIFT;
10136 page_offset = (iocb->ki_pos - start) & (PAGE_SIZE - 1);
10139 while (cur < count) {
10140 size_t bytes = min_t(size_t, count - cur,
10141 PAGE_SIZE - page_offset);
10143 if (copy_page_to_iter(pages[i], page_offset, bytes,
10154 for (i = 0; i < nr_pages; i++) {
10156 __free_page(pages[i]);
10162 ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter,
10163 struct btrfs_ioctl_encoded_io_args *encoded)
10165 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10166 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10167 struct extent_io_tree *io_tree = &inode->io_tree;
10169 size_t count = iov_iter_count(iter);
10170 u64 start, lockend, disk_bytenr, disk_io_size;
10171 struct extent_state *cached_state = NULL;
10172 struct extent_map *em;
10173 bool unlocked = false;
10175 file_accessed(iocb->ki_filp);
10177 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
10179 if (iocb->ki_pos >= inode->vfs_inode.i_size) {
10180 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10183 start = ALIGN_DOWN(iocb->ki_pos, fs_info->sectorsize);
10185 * We don't know how long the extent containing iocb->ki_pos is, but if
10186 * it's compressed we know that it won't be longer than this.
10188 lockend = start + BTRFS_MAX_UNCOMPRESSED - 1;
10191 struct btrfs_ordered_extent *ordered;
10193 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start,
10194 lockend - start + 1);
10196 goto out_unlock_inode;
10197 lock_extent(io_tree, start, lockend, &cached_state);
10198 ordered = btrfs_lookup_ordered_range(inode, start,
10199 lockend - start + 1);
10202 btrfs_put_ordered_extent(ordered);
10203 unlock_extent(io_tree, start, lockend, &cached_state);
10207 em = btrfs_get_extent(inode, NULL, start, lockend - start + 1);
10210 goto out_unlock_extent;
10213 if (em->block_start == EXTENT_MAP_INLINE) {
10214 u64 extent_start = em->start;
10217 * For inline extents we get everything we need out of the
10220 free_extent_map(em);
10222 ret = btrfs_encoded_read_inline(iocb, iter, start, lockend,
10223 &cached_state, extent_start,
10224 count, encoded, &unlocked);
10229 * We only want to return up to EOF even if the extent extends beyond
10232 encoded->len = min_t(u64, extent_map_end(em),
10233 inode->vfs_inode.i_size) - iocb->ki_pos;
10234 if (em->block_start == EXTENT_MAP_HOLE ||
10235 (em->flags & EXTENT_FLAG_PREALLOC)) {
10236 disk_bytenr = EXTENT_MAP_HOLE;
10237 count = min_t(u64, count, encoded->len);
10238 encoded->len = count;
10239 encoded->unencoded_len = count;
10240 } else if (extent_map_is_compressed(em)) {
10241 disk_bytenr = em->block_start;
10243 * Bail if the buffer isn't large enough to return the whole
10244 * compressed extent.
10246 if (em->block_len > count) {
10250 disk_io_size = em->block_len;
10251 count = em->block_len;
10252 encoded->unencoded_len = em->ram_bytes;
10253 encoded->unencoded_offset = iocb->ki_pos - em->orig_start;
10254 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10255 extent_map_compression(em));
10258 encoded->compression = ret;
10260 disk_bytenr = em->block_start + (start - em->start);
10261 if (encoded->len > count)
10262 encoded->len = count;
10264 * Don't read beyond what we locked. This also limits the page
10265 * allocations that we'll do.
10267 disk_io_size = min(lockend + 1, iocb->ki_pos + encoded->len) - start;
10268 count = start + disk_io_size - iocb->ki_pos;
10269 encoded->len = count;
10270 encoded->unencoded_len = count;
10271 disk_io_size = ALIGN(disk_io_size, fs_info->sectorsize);
10273 free_extent_map(em);
10276 if (disk_bytenr == EXTENT_MAP_HOLE) {
10277 unlock_extent(io_tree, start, lockend, &cached_state);
10278 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10280 ret = iov_iter_zero(count, iter);
10284 ret = btrfs_encoded_read_regular(iocb, iter, start, lockend,
10285 &cached_state, disk_bytenr,
10286 disk_io_size, count,
10287 encoded->compression,
10293 iocb->ki_pos += encoded->len;
10295 free_extent_map(em);
10298 unlock_extent(io_tree, start, lockend, &cached_state);
10301 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10305 ssize_t btrfs_do_encoded_write(struct kiocb *iocb, struct iov_iter *from,
10306 const struct btrfs_ioctl_encoded_io_args *encoded)
10308 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10309 struct btrfs_root *root = inode->root;
10310 struct btrfs_fs_info *fs_info = root->fs_info;
10311 struct extent_io_tree *io_tree = &inode->io_tree;
10312 struct extent_changeset *data_reserved = NULL;
10313 struct extent_state *cached_state = NULL;
10314 struct btrfs_ordered_extent *ordered;
10318 u64 num_bytes, ram_bytes, disk_num_bytes;
10319 unsigned long nr_pages, i;
10320 struct page **pages;
10321 struct btrfs_key ins;
10322 bool extent_reserved = false;
10323 struct extent_map *em;
10326 switch (encoded->compression) {
10327 case BTRFS_ENCODED_IO_COMPRESSION_ZLIB:
10328 compression = BTRFS_COMPRESS_ZLIB;
10330 case BTRFS_ENCODED_IO_COMPRESSION_ZSTD:
10331 compression = BTRFS_COMPRESS_ZSTD;
10333 case BTRFS_ENCODED_IO_COMPRESSION_LZO_4K:
10334 case BTRFS_ENCODED_IO_COMPRESSION_LZO_8K:
10335 case BTRFS_ENCODED_IO_COMPRESSION_LZO_16K:
10336 case BTRFS_ENCODED_IO_COMPRESSION_LZO_32K:
10337 case BTRFS_ENCODED_IO_COMPRESSION_LZO_64K:
10338 /* The sector size must match for LZO. */
10339 if (encoded->compression -
10340 BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + 12 !=
10341 fs_info->sectorsize_bits)
10343 compression = BTRFS_COMPRESS_LZO;
10348 if (encoded->encryption != BTRFS_ENCODED_IO_ENCRYPTION_NONE)
10352 * Compressed extents should always have checksums, so error out if we
10353 * have a NOCOW file or inode was created while mounted with NODATASUM.
10355 if (inode->flags & BTRFS_INODE_NODATASUM)
10358 orig_count = iov_iter_count(from);
10360 /* The extent size must be sane. */
10361 if (encoded->unencoded_len > BTRFS_MAX_UNCOMPRESSED ||
10362 orig_count > BTRFS_MAX_COMPRESSED || orig_count == 0)
10366 * The compressed data must be smaller than the decompressed data.
10368 * It's of course possible for data to compress to larger or the same
10369 * size, but the buffered I/O path falls back to no compression for such
10370 * data, and we don't want to break any assumptions by creating these
10373 * Note that this is less strict than the current check we have that the
10374 * compressed data must be at least one sector smaller than the
10375 * decompressed data. We only want to enforce the weaker requirement
10376 * from old kernels that it is at least one byte smaller.
10378 if (orig_count >= encoded->unencoded_len)
10381 /* The extent must start on a sector boundary. */
10382 start = iocb->ki_pos;
10383 if (!IS_ALIGNED(start, fs_info->sectorsize))
10387 * The extent must end on a sector boundary. However, we allow a write
10388 * which ends at or extends i_size to have an unaligned length; we round
10389 * up the extent size and set i_size to the unaligned end.
10391 if (start + encoded->len < inode->vfs_inode.i_size &&
10392 !IS_ALIGNED(start + encoded->len, fs_info->sectorsize))
10395 /* Finally, the offset in the unencoded data must be sector-aligned. */
10396 if (!IS_ALIGNED(encoded->unencoded_offset, fs_info->sectorsize))
10399 num_bytes = ALIGN(encoded->len, fs_info->sectorsize);
10400 ram_bytes = ALIGN(encoded->unencoded_len, fs_info->sectorsize);
10401 end = start + num_bytes - 1;
10404 * If the extent cannot be inline, the compressed data on disk must be
10405 * sector-aligned. For convenience, we extend it with zeroes if it
10408 disk_num_bytes = ALIGN(orig_count, fs_info->sectorsize);
10409 nr_pages = DIV_ROUND_UP(disk_num_bytes, PAGE_SIZE);
10410 pages = kvcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL_ACCOUNT);
10413 for (i = 0; i < nr_pages; i++) {
10414 size_t bytes = min_t(size_t, PAGE_SIZE, iov_iter_count(from));
10417 pages[i] = alloc_page(GFP_KERNEL_ACCOUNT);
10422 kaddr = kmap_local_page(pages[i]);
10423 if (copy_from_iter(kaddr, bytes, from) != bytes) {
10424 kunmap_local(kaddr);
10428 if (bytes < PAGE_SIZE)
10429 memset(kaddr + bytes, 0, PAGE_SIZE - bytes);
10430 kunmap_local(kaddr);
10434 struct btrfs_ordered_extent *ordered;
10436 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start, num_bytes);
10439 ret = invalidate_inode_pages2_range(inode->vfs_inode.i_mapping,
10440 start >> PAGE_SHIFT,
10441 end >> PAGE_SHIFT);
10444 lock_extent(io_tree, start, end, &cached_state);
10445 ordered = btrfs_lookup_ordered_range(inode, start, num_bytes);
10447 !filemap_range_has_page(inode->vfs_inode.i_mapping, start, end))
10450 btrfs_put_ordered_extent(ordered);
10451 unlock_extent(io_tree, start, end, &cached_state);
10456 * We don't use the higher-level delalloc space functions because our
10457 * num_bytes and disk_num_bytes are different.
10459 ret = btrfs_alloc_data_chunk_ondemand(inode, disk_num_bytes);
10462 ret = btrfs_qgroup_reserve_data(inode, &data_reserved, start, num_bytes);
10464 goto out_free_data_space;
10465 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes, disk_num_bytes,
10468 goto out_qgroup_free_data;
10470 /* Try an inline extent first. */
10471 if (start == 0 && encoded->unencoded_len == encoded->len &&
10472 encoded->unencoded_offset == 0) {
10473 ret = cow_file_range_inline(inode, encoded->len, orig_count,
10474 compression, pages, true);
10478 goto out_delalloc_release;
10482 ret = btrfs_reserve_extent(root, disk_num_bytes, disk_num_bytes,
10483 disk_num_bytes, 0, 0, &ins, 1, 1);
10485 goto out_delalloc_release;
10486 extent_reserved = true;
10488 em = create_io_em(inode, start, num_bytes,
10489 start - encoded->unencoded_offset, ins.objectid,
10490 ins.offset, ins.offset, ram_bytes, compression,
10491 BTRFS_ORDERED_COMPRESSED);
10494 goto out_free_reserved;
10496 free_extent_map(em);
10498 ordered = btrfs_alloc_ordered_extent(inode, start, num_bytes, ram_bytes,
10499 ins.objectid, ins.offset,
10500 encoded->unencoded_offset,
10501 (1 << BTRFS_ORDERED_ENCODED) |
10502 (1 << BTRFS_ORDERED_COMPRESSED),
10504 if (IS_ERR(ordered)) {
10505 btrfs_drop_extent_map_range(inode, start, end, false);
10506 ret = PTR_ERR(ordered);
10507 goto out_free_reserved;
10509 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10511 if (start + encoded->len > inode->vfs_inode.i_size)
10512 i_size_write(&inode->vfs_inode, start + encoded->len);
10514 unlock_extent(io_tree, start, end, &cached_state);
10516 btrfs_delalloc_release_extents(inode, num_bytes);
10518 btrfs_submit_compressed_write(ordered, pages, nr_pages, 0, false);
10523 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10524 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
10525 out_delalloc_release:
10526 btrfs_delalloc_release_extents(inode, num_bytes);
10527 btrfs_delalloc_release_metadata(inode, disk_num_bytes, ret < 0);
10528 out_qgroup_free_data:
10530 btrfs_qgroup_free_data(inode, data_reserved, start, num_bytes, NULL);
10531 out_free_data_space:
10533 * If btrfs_reserve_extent() succeeded, then we already decremented
10536 if (!extent_reserved)
10537 btrfs_free_reserved_data_space_noquota(fs_info, disk_num_bytes);
10539 unlock_extent(io_tree, start, end, &cached_state);
10541 for (i = 0; i < nr_pages; i++) {
10543 __free_page(pages[i]);
10548 iocb->ki_pos += encoded->len;
10554 * Add an entry indicating a block group or device which is pinned by a
10555 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10556 * negative errno on failure.
10558 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10559 bool is_block_group)
10561 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10562 struct btrfs_swapfile_pin *sp, *entry;
10563 struct rb_node **p;
10564 struct rb_node *parent = NULL;
10566 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10571 sp->is_block_group = is_block_group;
10572 sp->bg_extent_count = 1;
10574 spin_lock(&fs_info->swapfile_pins_lock);
10575 p = &fs_info->swapfile_pins.rb_node;
10578 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10579 if (sp->ptr < entry->ptr ||
10580 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10581 p = &(*p)->rb_left;
10582 } else if (sp->ptr > entry->ptr ||
10583 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10584 p = &(*p)->rb_right;
10586 if (is_block_group)
10587 entry->bg_extent_count++;
10588 spin_unlock(&fs_info->swapfile_pins_lock);
10593 rb_link_node(&sp->node, parent, p);
10594 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10595 spin_unlock(&fs_info->swapfile_pins_lock);
10599 /* Free all of the entries pinned by this swapfile. */
10600 static void btrfs_free_swapfile_pins(struct inode *inode)
10602 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10603 struct btrfs_swapfile_pin *sp;
10604 struct rb_node *node, *next;
10606 spin_lock(&fs_info->swapfile_pins_lock);
10607 node = rb_first(&fs_info->swapfile_pins);
10609 next = rb_next(node);
10610 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10611 if (sp->inode == inode) {
10612 rb_erase(&sp->node, &fs_info->swapfile_pins);
10613 if (sp->is_block_group) {
10614 btrfs_dec_block_group_swap_extents(sp->ptr,
10615 sp->bg_extent_count);
10616 btrfs_put_block_group(sp->ptr);
10622 spin_unlock(&fs_info->swapfile_pins_lock);
10625 struct btrfs_swap_info {
10631 unsigned long nr_pages;
10635 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10636 struct btrfs_swap_info *bsi)
10638 unsigned long nr_pages;
10639 unsigned long max_pages;
10640 u64 first_ppage, first_ppage_reported, next_ppage;
10644 * Our swapfile may have had its size extended after the swap header was
10645 * written. In that case activating the swapfile should not go beyond
10646 * the max size set in the swap header.
10648 if (bsi->nr_pages >= sis->max)
10651 max_pages = sis->max - bsi->nr_pages;
10652 first_ppage = PAGE_ALIGN(bsi->block_start) >> PAGE_SHIFT;
10653 next_ppage = PAGE_ALIGN_DOWN(bsi->block_start + bsi->block_len) >> PAGE_SHIFT;
10655 if (first_ppage >= next_ppage)
10657 nr_pages = next_ppage - first_ppage;
10658 nr_pages = min(nr_pages, max_pages);
10660 first_ppage_reported = first_ppage;
10661 if (bsi->start == 0)
10662 first_ppage_reported++;
10663 if (bsi->lowest_ppage > first_ppage_reported)
10664 bsi->lowest_ppage = first_ppage_reported;
10665 if (bsi->highest_ppage < (next_ppage - 1))
10666 bsi->highest_ppage = next_ppage - 1;
10668 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10671 bsi->nr_extents += ret;
10672 bsi->nr_pages += nr_pages;
10676 static void btrfs_swap_deactivate(struct file *file)
10678 struct inode *inode = file_inode(file);
10680 btrfs_free_swapfile_pins(inode);
10681 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10684 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10687 struct inode *inode = file_inode(file);
10688 struct btrfs_root *root = BTRFS_I(inode)->root;
10689 struct btrfs_fs_info *fs_info = root->fs_info;
10690 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10691 struct extent_state *cached_state = NULL;
10692 struct extent_map *em = NULL;
10693 struct btrfs_chunk_map *map = NULL;
10694 struct btrfs_device *device = NULL;
10695 struct btrfs_swap_info bsi = {
10696 .lowest_ppage = (sector_t)-1ULL,
10703 * If the swap file was just created, make sure delalloc is done. If the
10704 * file changes again after this, the user is doing something stupid and
10705 * we don't really care.
10707 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10712 * The inode is locked, so these flags won't change after we check them.
10714 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10715 btrfs_warn(fs_info, "swapfile must not be compressed");
10718 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10719 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10722 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10723 btrfs_warn(fs_info, "swapfile must not be checksummed");
10728 * Balance or device remove/replace/resize can move stuff around from
10729 * under us. The exclop protection makes sure they aren't running/won't
10730 * run concurrently while we are mapping the swap extents, and
10731 * fs_info->swapfile_pins prevents them from running while the swap
10732 * file is active and moving the extents. Note that this also prevents
10733 * a concurrent device add which isn't actually necessary, but it's not
10734 * really worth the trouble to allow it.
10736 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10737 btrfs_warn(fs_info,
10738 "cannot activate swapfile while exclusive operation is running");
10743 * Prevent snapshot creation while we are activating the swap file.
10744 * We do not want to race with snapshot creation. If snapshot creation
10745 * already started before we bumped nr_swapfiles from 0 to 1 and
10746 * completes before the first write into the swap file after it is
10747 * activated, than that write would fallback to COW.
10749 if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) {
10750 btrfs_exclop_finish(fs_info);
10751 btrfs_warn(fs_info,
10752 "cannot activate swapfile because snapshot creation is in progress");
10756 * Snapshots can create extents which require COW even if NODATACOW is
10757 * set. We use this counter to prevent snapshots. We must increment it
10758 * before walking the extents because we don't want a concurrent
10759 * snapshot to run after we've already checked the extents.
10761 * It is possible that subvolume is marked for deletion but still not
10762 * removed yet. To prevent this race, we check the root status before
10763 * activating the swapfile.
10765 spin_lock(&root->root_item_lock);
10766 if (btrfs_root_dead(root)) {
10767 spin_unlock(&root->root_item_lock);
10769 btrfs_exclop_finish(fs_info);
10770 btrfs_warn(fs_info,
10771 "cannot activate swapfile because subvolume %llu is being deleted",
10772 root->root_key.objectid);
10775 atomic_inc(&root->nr_swapfiles);
10776 spin_unlock(&root->root_item_lock);
10778 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10780 lock_extent(io_tree, 0, isize - 1, &cached_state);
10782 while (start < isize) {
10783 u64 logical_block_start, physical_block_start;
10784 struct btrfs_block_group *bg;
10785 u64 len = isize - start;
10787 em = btrfs_get_extent(BTRFS_I(inode), NULL, start, len);
10793 if (em->block_start == EXTENT_MAP_HOLE) {
10794 btrfs_warn(fs_info, "swapfile must not have holes");
10798 if (em->block_start == EXTENT_MAP_INLINE) {
10800 * It's unlikely we'll ever actually find ourselves
10801 * here, as a file small enough to fit inline won't be
10802 * big enough to store more than the swap header, but in
10803 * case something changes in the future, let's catch it
10804 * here rather than later.
10806 btrfs_warn(fs_info, "swapfile must not be inline");
10810 if (extent_map_is_compressed(em)) {
10811 btrfs_warn(fs_info, "swapfile must not be compressed");
10816 logical_block_start = em->block_start + (start - em->start);
10817 len = min(len, em->len - (start - em->start));
10818 free_extent_map(em);
10821 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, false, true);
10827 btrfs_warn(fs_info,
10828 "swapfile must not be copy-on-write");
10833 map = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10835 ret = PTR_ERR(map);
10839 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10840 btrfs_warn(fs_info,
10841 "swapfile must have single data profile");
10846 if (device == NULL) {
10847 device = map->stripes[0].dev;
10848 ret = btrfs_add_swapfile_pin(inode, device, false);
10853 } else if (device != map->stripes[0].dev) {
10854 btrfs_warn(fs_info, "swapfile must be on one device");
10859 physical_block_start = (map->stripes[0].physical +
10860 (logical_block_start - map->start));
10861 len = min(len, map->chunk_len - (logical_block_start - map->start));
10862 btrfs_free_chunk_map(map);
10865 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10867 btrfs_warn(fs_info,
10868 "could not find block group containing swapfile");
10873 if (!btrfs_inc_block_group_swap_extents(bg)) {
10874 btrfs_warn(fs_info,
10875 "block group for swapfile at %llu is read-only%s",
10877 atomic_read(&fs_info->scrubs_running) ?
10878 " (scrub running)" : "");
10879 btrfs_put_block_group(bg);
10884 ret = btrfs_add_swapfile_pin(inode, bg, true);
10886 btrfs_put_block_group(bg);
10893 if (bsi.block_len &&
10894 bsi.block_start + bsi.block_len == physical_block_start) {
10895 bsi.block_len += len;
10897 if (bsi.block_len) {
10898 ret = btrfs_add_swap_extent(sis, &bsi);
10903 bsi.block_start = physical_block_start;
10904 bsi.block_len = len;
10911 ret = btrfs_add_swap_extent(sis, &bsi);
10914 if (!IS_ERR_OR_NULL(em))
10915 free_extent_map(em);
10916 if (!IS_ERR_OR_NULL(map))
10917 btrfs_free_chunk_map(map);
10919 unlock_extent(io_tree, 0, isize - 1, &cached_state);
10922 btrfs_swap_deactivate(file);
10924 btrfs_drew_write_unlock(&root->snapshot_lock);
10926 btrfs_exclop_finish(fs_info);
10932 sis->bdev = device->bdev;
10933 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10934 sis->max = bsi.nr_pages;
10935 sis->pages = bsi.nr_pages - 1;
10936 sis->highest_bit = bsi.nr_pages - 1;
10937 return bsi.nr_extents;
10940 static void btrfs_swap_deactivate(struct file *file)
10944 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10947 return -EOPNOTSUPP;
10952 * Update the number of bytes used in the VFS' inode. When we replace extents in
10953 * a range (clone, dedupe, fallocate's zero range), we must update the number of
10954 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
10955 * always get a correct value.
10957 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
10958 const u64 add_bytes,
10959 const u64 del_bytes)
10961 if (add_bytes == del_bytes)
10964 spin_lock(&inode->lock);
10966 inode_sub_bytes(&inode->vfs_inode, del_bytes);
10968 inode_add_bytes(&inode->vfs_inode, add_bytes);
10969 spin_unlock(&inode->lock);
10973 * Verify that there are no ordered extents for a given file range.
10975 * @inode: The target inode.
10976 * @start: Start offset of the file range, should be sector size aligned.
10977 * @end: End offset (inclusive) of the file range, its value +1 should be
10978 * sector size aligned.
10980 * This should typically be used for cases where we locked an inode's VFS lock in
10981 * exclusive mode, we have also locked the inode's i_mmap_lock in exclusive mode,
10982 * we have flushed all delalloc in the range, we have waited for all ordered
10983 * extents in the range to complete and finally we have locked the file range in
10984 * the inode's io_tree.
10986 void btrfs_assert_inode_range_clean(struct btrfs_inode *inode, u64 start, u64 end)
10988 struct btrfs_root *root = inode->root;
10989 struct btrfs_ordered_extent *ordered;
10991 if (!IS_ENABLED(CONFIG_BTRFS_ASSERT))
10994 ordered = btrfs_lookup_first_ordered_range(inode, start, end + 1 - start);
10996 btrfs_err(root->fs_info,
10997 "found unexpected ordered extent in file range [%llu, %llu] for inode %llu root %llu (ordered range [%llu, %llu])",
10998 start, end, btrfs_ino(inode), root->root_key.objectid,
10999 ordered->file_offset,
11000 ordered->file_offset + ordered->num_bytes - 1);
11001 btrfs_put_ordered_extent(ordered);
11004 ASSERT(ordered == NULL);
11007 static const struct inode_operations btrfs_dir_inode_operations = {
11008 .getattr = btrfs_getattr,
11009 .lookup = btrfs_lookup,
11010 .create = btrfs_create,
11011 .unlink = btrfs_unlink,
11012 .link = btrfs_link,
11013 .mkdir = btrfs_mkdir,
11014 .rmdir = btrfs_rmdir,
11015 .rename = btrfs_rename2,
11016 .symlink = btrfs_symlink,
11017 .setattr = btrfs_setattr,
11018 .mknod = btrfs_mknod,
11019 .listxattr = btrfs_listxattr,
11020 .permission = btrfs_permission,
11021 .get_inode_acl = btrfs_get_acl,
11022 .set_acl = btrfs_set_acl,
11023 .update_time = btrfs_update_time,
11024 .tmpfile = btrfs_tmpfile,
11025 .fileattr_get = btrfs_fileattr_get,
11026 .fileattr_set = btrfs_fileattr_set,
11029 static const struct file_operations btrfs_dir_file_operations = {
11030 .llseek = btrfs_dir_llseek,
11031 .read = generic_read_dir,
11032 .iterate_shared = btrfs_real_readdir,
11033 .open = btrfs_opendir,
11034 .unlocked_ioctl = btrfs_ioctl,
11035 #ifdef CONFIG_COMPAT
11036 .compat_ioctl = btrfs_compat_ioctl,
11038 .release = btrfs_release_file,
11039 .fsync = btrfs_sync_file,
11043 * btrfs doesn't support the bmap operation because swapfiles
11044 * use bmap to make a mapping of extents in the file. They assume
11045 * these extents won't change over the life of the file and they
11046 * use the bmap result to do IO directly to the drive.
11048 * the btrfs bmap call would return logical addresses that aren't
11049 * suitable for IO and they also will change frequently as COW
11050 * operations happen. So, swapfile + btrfs == corruption.
11052 * For now we're avoiding this by dropping bmap.
11054 static const struct address_space_operations btrfs_aops = {
11055 .read_folio = btrfs_read_folio,
11056 .writepages = btrfs_writepages,
11057 .readahead = btrfs_readahead,
11058 .invalidate_folio = btrfs_invalidate_folio,
11059 .release_folio = btrfs_release_folio,
11060 .migrate_folio = btrfs_migrate_folio,
11061 .dirty_folio = filemap_dirty_folio,
11062 .error_remove_folio = generic_error_remove_folio,
11063 .swap_activate = btrfs_swap_activate,
11064 .swap_deactivate = btrfs_swap_deactivate,
11067 static const struct inode_operations btrfs_file_inode_operations = {
11068 .getattr = btrfs_getattr,
11069 .setattr = btrfs_setattr,
11070 .listxattr = btrfs_listxattr,
11071 .permission = btrfs_permission,
11072 .fiemap = btrfs_fiemap,
11073 .get_inode_acl = btrfs_get_acl,
11074 .set_acl = btrfs_set_acl,
11075 .update_time = btrfs_update_time,
11076 .fileattr_get = btrfs_fileattr_get,
11077 .fileattr_set = btrfs_fileattr_set,
11079 static const struct inode_operations btrfs_special_inode_operations = {
11080 .getattr = btrfs_getattr,
11081 .setattr = btrfs_setattr,
11082 .permission = btrfs_permission,
11083 .listxattr = btrfs_listxattr,
11084 .get_inode_acl = btrfs_get_acl,
11085 .set_acl = btrfs_set_acl,
11086 .update_time = btrfs_update_time,
11088 static const struct inode_operations btrfs_symlink_inode_operations = {
11089 .get_link = page_get_link,
11090 .getattr = btrfs_getattr,
11091 .setattr = btrfs_setattr,
11092 .permission = btrfs_permission,
11093 .listxattr = btrfs_listxattr,
11094 .update_time = btrfs_update_time,
11097 const struct dentry_operations btrfs_dentry_operations = {
11098 .d_delete = btrfs_dentry_delete,