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 "print-tree.h"
43 #include "ordered-data.h"
47 #include "compression.h"
49 #include "free-space-cache.h"
52 #include "delalloc-space.h"
53 #include "block-group.h"
54 #include "space-info.h"
57 #include "inode-item.h"
59 #include "accessors.h"
60 #include "extent-tree.h"
61 #include "root-tree.h"
64 #include "file-item.h"
65 #include "uuid-tree.h"
69 #include "relocation.h"
75 struct btrfs_iget_args {
77 struct btrfs_root *root;
80 struct btrfs_dio_data {
82 struct extent_changeset *data_reserved;
83 struct btrfs_ordered_extent *ordered;
84 bool data_space_reserved;
88 struct btrfs_dio_private {
93 /* This must be last */
94 struct btrfs_bio bbio;
97 static struct bio_set btrfs_dio_bioset;
99 struct btrfs_rename_ctx {
100 /* Output field. Stores the index number of the old directory entry. */
105 * Used by data_reloc_print_warning_inode() to pass needed info for filename
106 * resolution and output of error message.
108 struct data_reloc_warn {
109 struct btrfs_path path;
110 struct btrfs_fs_info *fs_info;
111 u64 extent_item_size;
116 static const struct inode_operations btrfs_dir_inode_operations;
117 static const struct inode_operations btrfs_symlink_inode_operations;
118 static const struct inode_operations btrfs_special_inode_operations;
119 static const struct inode_operations btrfs_file_inode_operations;
120 static const struct address_space_operations btrfs_aops;
121 static const struct file_operations btrfs_dir_file_operations;
123 static struct kmem_cache *btrfs_inode_cachep;
125 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
126 static int btrfs_truncate(struct btrfs_inode *inode, bool skip_writeback);
128 static noinline int cow_file_range(struct btrfs_inode *inode,
129 struct page *locked_page,
130 u64 start, u64 end, u64 *done_offset,
131 bool keep_locked, bool no_inline);
132 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
133 u64 len, u64 orig_start, u64 block_start,
134 u64 block_len, u64 orig_block_len,
135 u64 ram_bytes, int compress_type,
138 static int data_reloc_print_warning_inode(u64 inum, u64 offset, u64 num_bytes,
139 u64 root, void *warn_ctx)
141 struct data_reloc_warn *warn = warn_ctx;
142 struct btrfs_fs_info *fs_info = warn->fs_info;
143 struct extent_buffer *eb;
144 struct btrfs_inode_item *inode_item;
145 struct inode_fs_paths *ipath = NULL;
146 struct btrfs_root *local_root;
147 struct btrfs_key key;
148 unsigned int nofs_flag;
152 local_root = btrfs_get_fs_root(fs_info, root, true);
153 if (IS_ERR(local_root)) {
154 ret = PTR_ERR(local_root);
158 /* This makes the path point to (inum INODE_ITEM ioff). */
160 key.type = BTRFS_INODE_ITEM_KEY;
163 ret = btrfs_search_slot(NULL, local_root, &key, &warn->path, 0, 0);
165 btrfs_put_root(local_root);
166 btrfs_release_path(&warn->path);
170 eb = warn->path.nodes[0];
171 inode_item = btrfs_item_ptr(eb, warn->path.slots[0], struct btrfs_inode_item);
172 nlink = btrfs_inode_nlink(eb, inode_item);
173 btrfs_release_path(&warn->path);
175 nofs_flag = memalloc_nofs_save();
176 ipath = init_ipath(4096, local_root, &warn->path);
177 memalloc_nofs_restore(nofs_flag);
179 btrfs_put_root(local_root);
180 ret = PTR_ERR(ipath);
183 * -ENOMEM, not a critical error, just output an generic error
187 "checksum error at logical %llu mirror %u root %llu, inode %llu offset %llu",
188 warn->logical, warn->mirror_num, root, inum, offset);
191 ret = paths_from_inode(inum, ipath);
196 * We deliberately ignore the bit ipath might have been too small to
197 * hold all of the paths here
199 for (int i = 0; i < ipath->fspath->elem_cnt; i++) {
201 "checksum error at logical %llu mirror %u root %llu inode %llu offset %llu length %u links %u (path: %s)",
202 warn->logical, warn->mirror_num, root, inum, offset,
203 fs_info->sectorsize, nlink,
204 (char *)(unsigned long)ipath->fspath->val[i]);
207 btrfs_put_root(local_root);
213 "checksum error at logical %llu mirror %u root %llu inode %llu offset %llu, path resolving failed with ret=%d",
214 warn->logical, warn->mirror_num, root, inum, offset, ret);
221 * Do extra user-friendly error output (e.g. lookup all the affected files).
223 * Return true if we succeeded doing the backref lookup.
224 * Return false if such lookup failed, and has to fallback to the old error message.
226 static void print_data_reloc_error(const struct btrfs_inode *inode, u64 file_off,
227 const u8 *csum, const u8 *csum_expected,
230 struct btrfs_fs_info *fs_info = inode->root->fs_info;
231 struct btrfs_path path = { 0 };
232 struct btrfs_key found_key = { 0 };
233 struct extent_buffer *eb;
234 struct btrfs_extent_item *ei;
235 const u32 csum_size = fs_info->csum_size;
241 mutex_lock(&fs_info->reloc_mutex);
242 logical = btrfs_get_reloc_bg_bytenr(fs_info);
243 mutex_unlock(&fs_info->reloc_mutex);
245 if (logical == U64_MAX) {
246 btrfs_warn_rl(fs_info, "has data reloc tree but no running relocation");
247 btrfs_warn_rl(fs_info,
248 "csum failed root %lld ino %llu off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
249 inode->root->root_key.objectid, btrfs_ino(inode), file_off,
250 CSUM_FMT_VALUE(csum_size, csum),
251 CSUM_FMT_VALUE(csum_size, csum_expected),
257 btrfs_warn_rl(fs_info,
258 "csum failed root %lld ino %llu off %llu logical %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
259 inode->root->root_key.objectid,
260 btrfs_ino(inode), file_off, logical,
261 CSUM_FMT_VALUE(csum_size, csum),
262 CSUM_FMT_VALUE(csum_size, csum_expected),
265 ret = extent_from_logical(fs_info, logical, &path, &found_key, &flags);
267 btrfs_err_rl(fs_info, "failed to lookup extent item for logical %llu: %d",
272 ei = btrfs_item_ptr(eb, path.slots[0], struct btrfs_extent_item);
273 item_size = btrfs_item_size(eb, path.slots[0]);
274 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
275 unsigned long ptr = 0;
280 ret = tree_backref_for_extent(&ptr, eb, &found_key, ei,
281 item_size, &ref_root,
284 btrfs_warn_rl(fs_info,
285 "failed to resolve tree backref for logical %llu: %d",
292 btrfs_warn_rl(fs_info,
293 "csum error at logical %llu mirror %u: metadata %s (level %d) in tree %llu",
295 (ref_level ? "node" : "leaf"),
296 ref_level, ref_root);
298 btrfs_release_path(&path);
300 struct btrfs_backref_walk_ctx ctx = { 0 };
301 struct data_reloc_warn reloc_warn = { 0 };
303 btrfs_release_path(&path);
305 ctx.bytenr = found_key.objectid;
306 ctx.extent_item_pos = logical - found_key.objectid;
307 ctx.fs_info = fs_info;
309 reloc_warn.logical = logical;
310 reloc_warn.extent_item_size = found_key.offset;
311 reloc_warn.mirror_num = mirror_num;
312 reloc_warn.fs_info = fs_info;
314 iterate_extent_inodes(&ctx, true,
315 data_reloc_print_warning_inode, &reloc_warn);
319 static void __cold btrfs_print_data_csum_error(struct btrfs_inode *inode,
320 u64 logical_start, u8 *csum, u8 *csum_expected, int mirror_num)
322 struct btrfs_root *root = inode->root;
323 const u32 csum_size = root->fs_info->csum_size;
325 /* For data reloc tree, it's better to do a backref lookup instead. */
326 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
327 return print_data_reloc_error(inode, logical_start, csum,
328 csum_expected, mirror_num);
330 /* Output without objectid, which is more meaningful */
331 if (root->root_key.objectid >= BTRFS_LAST_FREE_OBJECTID) {
332 btrfs_warn_rl(root->fs_info,
333 "csum failed root %lld ino %lld off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
334 root->root_key.objectid, btrfs_ino(inode),
336 CSUM_FMT_VALUE(csum_size, csum),
337 CSUM_FMT_VALUE(csum_size, csum_expected),
340 btrfs_warn_rl(root->fs_info,
341 "csum failed root %llu ino %llu 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),
351 * btrfs_inode_lock - lock inode i_rwsem based on arguments passed
353 * ilock_flags can have the following bit set:
355 * BTRFS_ILOCK_SHARED - acquire a shared lock on the inode
356 * BTRFS_ILOCK_TRY - try to acquire the lock, if fails on first attempt
358 * BTRFS_ILOCK_MMAP - acquire a write lock on the i_mmap_lock
360 int btrfs_inode_lock(struct btrfs_inode *inode, unsigned int ilock_flags)
362 if (ilock_flags & BTRFS_ILOCK_SHARED) {
363 if (ilock_flags & BTRFS_ILOCK_TRY) {
364 if (!inode_trylock_shared(&inode->vfs_inode))
369 inode_lock_shared(&inode->vfs_inode);
371 if (ilock_flags & BTRFS_ILOCK_TRY) {
372 if (!inode_trylock(&inode->vfs_inode))
377 inode_lock(&inode->vfs_inode);
379 if (ilock_flags & BTRFS_ILOCK_MMAP)
380 down_write(&inode->i_mmap_lock);
385 * btrfs_inode_unlock - unock inode i_rwsem
387 * ilock_flags should contain the same bits set as passed to btrfs_inode_lock()
388 * to decide whether the lock acquired is shared or exclusive.
390 void btrfs_inode_unlock(struct btrfs_inode *inode, unsigned int ilock_flags)
392 if (ilock_flags & BTRFS_ILOCK_MMAP)
393 up_write(&inode->i_mmap_lock);
394 if (ilock_flags & BTRFS_ILOCK_SHARED)
395 inode_unlock_shared(&inode->vfs_inode);
397 inode_unlock(&inode->vfs_inode);
401 * Cleanup all submitted ordered extents in specified range to handle errors
402 * from the btrfs_run_delalloc_range() callback.
404 * NOTE: caller must ensure that when an error happens, it can not call
405 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
406 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
407 * to be released, which we want to happen only when finishing the ordered
408 * extent (btrfs_finish_ordered_io()).
410 static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode,
411 struct page *locked_page,
412 u64 offset, u64 bytes)
414 unsigned long index = offset >> PAGE_SHIFT;
415 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
416 u64 page_start = 0, page_end = 0;
420 page_start = page_offset(locked_page);
421 page_end = page_start + PAGE_SIZE - 1;
424 while (index <= end_index) {
426 * For locked page, we will call btrfs_mark_ordered_io_finished
427 * through btrfs_mark_ordered_io_finished() on it
428 * in run_delalloc_range() for the error handling, which will
429 * clear page Ordered and run the ordered extent accounting.
431 * Here we can't just clear the Ordered bit, or
432 * btrfs_mark_ordered_io_finished() would skip the accounting
433 * for the page range, and the ordered extent will never finish.
435 if (locked_page && index == (page_start >> PAGE_SHIFT)) {
439 page = find_get_page(inode->vfs_inode.i_mapping, index);
445 * Here we just clear all Ordered bits for every page in the
446 * range, then btrfs_mark_ordered_io_finished() will handle
447 * the ordered extent accounting for the range.
449 btrfs_page_clamp_clear_ordered(inode->root->fs_info, page,
455 /* The locked page covers the full range, nothing needs to be done */
456 if (bytes + offset <= page_start + PAGE_SIZE)
459 * In case this page belongs to the delalloc range being
460 * instantiated then skip it, since the first page of a range is
461 * going to be properly cleaned up by the caller of
464 if (page_start >= offset && page_end <= (offset + bytes - 1)) {
465 bytes = offset + bytes - page_offset(locked_page) - PAGE_SIZE;
466 offset = page_offset(locked_page) + PAGE_SIZE;
470 return btrfs_mark_ordered_io_finished(inode, NULL, offset, bytes, false);
473 static int btrfs_dirty_inode(struct btrfs_inode *inode);
475 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
476 struct btrfs_new_inode_args *args)
480 if (args->default_acl) {
481 err = __btrfs_set_acl(trans, args->inode, args->default_acl,
487 err = __btrfs_set_acl(trans, args->inode, args->acl, ACL_TYPE_ACCESS);
491 if (!args->default_acl && !args->acl)
492 cache_no_acl(args->inode);
493 return btrfs_xattr_security_init(trans, args->inode, args->dir,
494 &args->dentry->d_name);
498 * this does all the hard work for inserting an inline extent into
499 * the btree. The caller should have done a btrfs_drop_extents so that
500 * no overlapping inline items exist in the btree
502 static int insert_inline_extent(struct btrfs_trans_handle *trans,
503 struct btrfs_path *path,
504 struct btrfs_inode *inode, bool extent_inserted,
505 size_t size, size_t compressed_size,
507 struct page **compressed_pages,
510 struct btrfs_root *root = inode->root;
511 struct extent_buffer *leaf;
512 struct page *page = NULL;
515 struct btrfs_file_extent_item *ei;
517 size_t cur_size = size;
520 ASSERT((compressed_size > 0 && compressed_pages) ||
521 (compressed_size == 0 && !compressed_pages));
523 if (compressed_size && compressed_pages)
524 cur_size = compressed_size;
526 if (!extent_inserted) {
527 struct btrfs_key key;
530 key.objectid = btrfs_ino(inode);
532 key.type = BTRFS_EXTENT_DATA_KEY;
534 datasize = btrfs_file_extent_calc_inline_size(cur_size);
535 ret = btrfs_insert_empty_item(trans, root, path, &key,
540 leaf = path->nodes[0];
541 ei = btrfs_item_ptr(leaf, path->slots[0],
542 struct btrfs_file_extent_item);
543 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
544 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
545 btrfs_set_file_extent_encryption(leaf, ei, 0);
546 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
547 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
548 ptr = btrfs_file_extent_inline_start(ei);
550 if (compress_type != BTRFS_COMPRESS_NONE) {
553 while (compressed_size > 0) {
554 cpage = compressed_pages[i];
555 cur_size = min_t(unsigned long, compressed_size,
558 kaddr = kmap_local_page(cpage);
559 write_extent_buffer(leaf, kaddr, ptr, cur_size);
564 compressed_size -= cur_size;
566 btrfs_set_file_extent_compression(leaf, ei,
569 page = find_get_page(inode->vfs_inode.i_mapping, 0);
570 btrfs_set_file_extent_compression(leaf, ei, 0);
571 kaddr = kmap_local_page(page);
572 write_extent_buffer(leaf, kaddr, ptr, size);
576 btrfs_mark_buffer_dirty(leaf);
577 btrfs_release_path(path);
580 * We align size to sectorsize for inline extents just for simplicity
583 ret = btrfs_inode_set_file_extent_range(inode, 0,
584 ALIGN(size, root->fs_info->sectorsize));
589 * We're an inline extent, so nobody can extend the file past i_size
590 * without locking a page we already have locked.
592 * We must do any i_size and inode updates before we unlock the pages.
593 * Otherwise we could end up racing with unlink.
595 i_size = i_size_read(&inode->vfs_inode);
596 if (update_i_size && size > i_size) {
597 i_size_write(&inode->vfs_inode, size);
600 inode->disk_i_size = i_size;
608 * conditionally insert an inline extent into the file. This
609 * does the checks required to make sure the data is small enough
610 * to fit as an inline extent.
612 static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 size,
613 size_t compressed_size,
615 struct page **compressed_pages,
618 struct btrfs_drop_extents_args drop_args = { 0 };
619 struct btrfs_root *root = inode->root;
620 struct btrfs_fs_info *fs_info = root->fs_info;
621 struct btrfs_trans_handle *trans;
622 u64 data_len = (compressed_size ?: size);
624 struct btrfs_path *path;
627 * We can create an inline extent if it ends at or beyond the current
628 * i_size, is no larger than a sector (decompressed), and the (possibly
629 * compressed) data fits in a leaf and the configured maximum inline
632 if (size < i_size_read(&inode->vfs_inode) ||
633 size > fs_info->sectorsize ||
634 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
635 data_len > fs_info->max_inline)
638 path = btrfs_alloc_path();
642 trans = btrfs_join_transaction(root);
644 btrfs_free_path(path);
645 return PTR_ERR(trans);
647 trans->block_rsv = &inode->block_rsv;
649 drop_args.path = path;
651 drop_args.end = fs_info->sectorsize;
652 drop_args.drop_cache = true;
653 drop_args.replace_extent = true;
654 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(data_len);
655 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
657 btrfs_abort_transaction(trans, ret);
661 ret = insert_inline_extent(trans, path, inode, drop_args.extent_inserted,
662 size, compressed_size, compress_type,
663 compressed_pages, update_i_size);
664 if (ret && ret != -ENOSPC) {
665 btrfs_abort_transaction(trans, ret);
667 } else if (ret == -ENOSPC) {
672 btrfs_update_inode_bytes(inode, size, drop_args.bytes_found);
673 ret = btrfs_update_inode(trans, root, inode);
674 if (ret && ret != -ENOSPC) {
675 btrfs_abort_transaction(trans, ret);
677 } else if (ret == -ENOSPC) {
682 btrfs_set_inode_full_sync(inode);
685 * Don't forget to free the reserved space, as for inlined extent
686 * it won't count as data extent, free them directly here.
687 * And at reserve time, it's always aligned to page size, so
688 * just free one page here.
690 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
691 btrfs_free_path(path);
692 btrfs_end_transaction(trans);
696 struct async_extent {
701 unsigned long nr_pages;
703 struct list_head list;
707 struct btrfs_inode *inode;
708 struct page *locked_page;
711 blk_opf_t write_flags;
712 struct list_head extents;
713 struct cgroup_subsys_state *blkcg_css;
714 struct btrfs_work work;
715 struct async_cow *async_cow;
720 struct async_chunk chunks[];
723 static noinline int add_async_extent(struct async_chunk *cow,
724 u64 start, u64 ram_size,
727 unsigned long nr_pages,
730 struct async_extent *async_extent;
732 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
733 BUG_ON(!async_extent); /* -ENOMEM */
734 async_extent->start = start;
735 async_extent->ram_size = ram_size;
736 async_extent->compressed_size = compressed_size;
737 async_extent->pages = pages;
738 async_extent->nr_pages = nr_pages;
739 async_extent->compress_type = compress_type;
740 list_add_tail(&async_extent->list, &cow->extents);
745 * Check if the inode needs to be submitted to compression, based on mount
746 * options, defragmentation, properties or heuristics.
748 static inline int inode_need_compress(struct btrfs_inode *inode, u64 start,
751 struct btrfs_fs_info *fs_info = inode->root->fs_info;
753 if (!btrfs_inode_can_compress(inode)) {
754 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
755 KERN_ERR "BTRFS: unexpected compression for ino %llu\n",
760 * Special check for subpage.
762 * We lock the full page then run each delalloc range in the page, thus
763 * for the following case, we will hit some subpage specific corner case:
766 * | |///////| |///////|
769 * In above case, both range A and range B will try to unlock the full
770 * page [0, 64K), causing the one finished later will have page
771 * unlocked already, triggering various page lock requirement BUG_ON()s.
773 * So here we add an artificial limit that subpage compression can only
774 * if the range is fully page aligned.
776 * In theory we only need to ensure the first page is fully covered, but
777 * the tailing partial page will be locked until the full compression
778 * finishes, delaying the write of other range.
780 * TODO: Make btrfs_run_delalloc_range() to lock all delalloc range
781 * first to prevent any submitted async extent to unlock the full page.
782 * By this, we can ensure for subpage case that only the last async_cow
783 * will unlock the full page.
785 if (fs_info->sectorsize < PAGE_SIZE) {
786 if (!PAGE_ALIGNED(start) ||
787 !PAGE_ALIGNED(end + 1))
792 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
795 if (inode->defrag_compress)
797 /* bad compression ratios */
798 if (inode->flags & BTRFS_INODE_NOCOMPRESS)
800 if (btrfs_test_opt(fs_info, COMPRESS) ||
801 inode->flags & BTRFS_INODE_COMPRESS ||
802 inode->prop_compress)
803 return btrfs_compress_heuristic(&inode->vfs_inode, start, end);
807 static inline void inode_should_defrag(struct btrfs_inode *inode,
808 u64 start, u64 end, u64 num_bytes, u32 small_write)
810 /* If this is a small write inside eof, kick off a defrag */
811 if (num_bytes < small_write &&
812 (start > 0 || end + 1 < inode->disk_i_size))
813 btrfs_add_inode_defrag(NULL, inode, small_write);
817 * Work queue call back to started compression on a file and pages.
819 * This is done inside an ordered work queue, and the compression is spread
820 * across many cpus. The actual IO submission is step two, and the ordered work
821 * queue takes care of making sure that happens in the same order things were
822 * put onto the queue by writepages and friends.
824 * If this code finds it can't get good compression, it puts an entry onto the
825 * work queue to write the uncompressed bytes. This makes sure that both
826 * compressed inodes and uncompressed inodes are written in the same order that
827 * the flusher thread sent them down.
829 static void compress_file_range(struct btrfs_work *work)
831 struct async_chunk *async_chunk =
832 container_of(work, struct async_chunk, work);
833 struct btrfs_inode *inode = async_chunk->inode;
834 struct btrfs_fs_info *fs_info = inode->root->fs_info;
835 struct address_space *mapping = inode->vfs_inode.i_mapping;
836 u64 blocksize = fs_info->sectorsize;
837 u64 start = async_chunk->start;
838 u64 end = async_chunk->end;
843 unsigned long nr_pages;
844 unsigned long total_compressed = 0;
845 unsigned long total_in = 0;
848 int compress_type = fs_info->compress_type;
851 inode_should_defrag(inode, start, end, end - start + 1, SZ_16K);
854 * We need to save i_size before now because it could change in between
855 * us evaluating the size and assigning it. This is because we lock and
856 * unlock the page in truncate and fallocate, and then modify the i_size
859 * The barriers are to emulate READ_ONCE, remove that once i_size_read
863 i_size = i_size_read(&inode->vfs_inode);
865 actual_end = min_t(u64, i_size, end + 1);
868 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
869 nr_pages = min_t(unsigned long, nr_pages, BTRFS_MAX_COMPRESSED_PAGES);
872 * we don't want to send crud past the end of i_size through
873 * compression, that's just a waste of CPU time. So, if the
874 * end of the file is before the start of our current
875 * requested range of bytes, we bail out to the uncompressed
876 * cleanup code that can deal with all of this.
878 * It isn't really the fastest way to fix things, but this is a
879 * very uncommon corner.
881 if (actual_end <= start)
882 goto cleanup_and_bail_uncompressed;
884 total_compressed = actual_end - start;
887 * Skip compression for a small file range(<=blocksize) that
888 * isn't an inline extent, since it doesn't save disk space at all.
890 if (total_compressed <= blocksize &&
891 (start > 0 || end + 1 < inode->disk_i_size))
892 goto cleanup_and_bail_uncompressed;
895 * For subpage case, we require full page alignment for the sector
897 * Thus we must also check against @actual_end, not just @end.
899 if (blocksize < PAGE_SIZE) {
900 if (!PAGE_ALIGNED(start) ||
901 !PAGE_ALIGNED(round_up(actual_end, blocksize)))
902 goto cleanup_and_bail_uncompressed;
905 total_compressed = min_t(unsigned long, total_compressed,
906 BTRFS_MAX_UNCOMPRESSED);
911 * We do compression for mount -o compress and when the inode has not
912 * been flagged as NOCOMPRESS. This flag can change at any time if we
913 * discover bad compression ratios.
915 if (!inode_need_compress(inode, start, end))
916 goto cleanup_and_bail_uncompressed;
918 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
921 * Memory allocation failure is not a fatal error, we can fall
922 * back to uncompressed code.
924 goto cleanup_and_bail_uncompressed;
927 if (inode->defrag_compress)
928 compress_type = inode->defrag_compress;
929 else if (inode->prop_compress)
930 compress_type = inode->prop_compress;
933 * We need to call clear_page_dirty_for_io on each page in the range.
934 * Otherwise applications with the file mmap'd can wander in and change
935 * the page contents while we are compressing them.
937 * If the compression fails for any reason, we set the pages dirty again
940 * Note that the remaining part is redirtied, the start pointer has
941 * moved, the end is the original one.
944 extent_range_clear_dirty_for_io(&inode->vfs_inode, start, end);
948 /* Compression level is applied here. */
949 ret = btrfs_compress_pages(compress_type | (fs_info->compress_level << 4),
950 mapping, start, pages, &nr_pages, &total_in,
953 goto cleanup_and_bail_uncompressed;
956 * Zero the tail end of the last page, as we might be sending it down
959 poff = offset_in_page(total_compressed);
961 memzero_page(pages[nr_pages - 1], poff, PAGE_SIZE - poff);
964 * Try to create an inline extent.
966 * If we didn't compress the entire range, try to create an uncompressed
967 * inline extent, else a compressed one.
969 * Check cow_file_range() for why we don't even try to create inline
970 * extent for the subpage case.
972 if (start == 0 && fs_info->sectorsize == PAGE_SIZE) {
973 if (total_in < actual_end) {
974 ret = cow_file_range_inline(inode, actual_end, 0,
975 BTRFS_COMPRESS_NONE, NULL,
978 ret = cow_file_range_inline(inode, actual_end,
980 compress_type, pages,
984 unsigned long clear_flags = EXTENT_DELALLOC |
985 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
986 EXTENT_DO_ACCOUNTING;
989 mapping_set_error(mapping, -EIO);
992 * inline extent creation worked or returned error,
993 * we don't need to create any more async work items.
994 * Unlock and free up our temp pages.
996 * We use DO_ACCOUNTING here because we need the
997 * delalloc_release_metadata to be done _after_ we drop
998 * our outstanding extent for clearing delalloc for this
1001 extent_clear_unlock_delalloc(inode, start, end,
1005 PAGE_START_WRITEBACK |
1006 PAGE_END_WRITEBACK);
1007 for (i = 0; i < nr_pages; i++) {
1008 WARN_ON(pages[i]->mapping);
1017 * We aren't doing an inline extent. Round the compressed size up to a
1018 * block size boundary so the allocator does sane things.
1020 total_compressed = ALIGN(total_compressed, blocksize);
1023 * One last check to make sure the compression is really a win, compare
1024 * the page count read with the blocks on disk, compression must free at
1027 total_in = round_up(total_in, fs_info->sectorsize);
1028 if (total_compressed + blocksize > total_in)
1029 goto cleanup_and_bail_uncompressed;
1032 * The async work queues will take care of doing actual allocation on
1033 * disk for these compressed pages, and will submit the bios.
1035 add_async_extent(async_chunk, start, total_in, total_compressed, pages,
1036 nr_pages, compress_type);
1037 if (start + total_in < end) {
1044 cleanup_and_bail_uncompressed:
1047 * the compression code ran but failed to make things smaller,
1048 * free any pages it allocated and our page pointer array
1050 for (i = 0; i < nr_pages; i++) {
1051 WARN_ON(pages[i]->mapping);
1056 total_compressed = 0;
1059 /* flag the file so we don't compress in the future */
1060 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
1061 !(inode->prop_compress)) {
1062 inode->flags |= BTRFS_INODE_NOCOMPRESS;
1067 * No compression, but we still need to write the pages in the file
1068 * we've been given so far. redirty the locked page if it corresponds
1069 * to our extent and set things up for the async work queue to run
1070 * cow_file_range to do the normal delalloc dance.
1072 if (async_chunk->locked_page &&
1073 (page_offset(async_chunk->locked_page) >= start &&
1074 page_offset(async_chunk->locked_page)) <= end) {
1075 __set_page_dirty_nobuffers(async_chunk->locked_page);
1076 /* unlocked later on in the async handlers */
1080 extent_range_redirty_for_io(&inode->vfs_inode, start, end);
1081 add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
1082 BTRFS_COMPRESS_NONE);
1085 static void free_async_extent_pages(struct async_extent *async_extent)
1089 if (!async_extent->pages)
1092 for (i = 0; i < async_extent->nr_pages; i++) {
1093 WARN_ON(async_extent->pages[i]->mapping);
1094 put_page(async_extent->pages[i]);
1096 kfree(async_extent->pages);
1097 async_extent->nr_pages = 0;
1098 async_extent->pages = NULL;
1101 static void submit_uncompressed_range(struct btrfs_inode *inode,
1102 struct async_extent *async_extent,
1103 struct page *locked_page)
1105 u64 start = async_extent->start;
1106 u64 end = async_extent->start + async_extent->ram_size - 1;
1108 struct writeback_control wbc = {
1109 .sync_mode = WB_SYNC_ALL,
1110 .range_start = start,
1112 .no_cgroup_owner = 1,
1116 * Call cow_file_range() to run the delalloc range directly, since we
1117 * won't go to NOCOW or async path again.
1119 * Also we call cow_file_range() with @unlock_page == 0, so that we
1120 * can directly submit them without interruption.
1122 ret = cow_file_range(inode, locked_page, start, end, NULL, true, false);
1123 /* Inline extent inserted, page gets unlocked and everything is done */
1128 btrfs_cleanup_ordered_extents(inode, locked_page, start, end - start + 1);
1130 const u64 page_start = page_offset(locked_page);
1132 set_page_writeback(locked_page);
1133 end_page_writeback(locked_page);
1134 btrfs_mark_ordered_io_finished(inode, locked_page,
1135 page_start, PAGE_SIZE,
1137 btrfs_page_clear_uptodate(inode->root->fs_info,
1138 locked_page, page_start,
1140 mapping_set_error(locked_page->mapping, ret);
1141 unlock_page(locked_page);
1146 /* All pages will be unlocked, including @locked_page */
1147 wbc_attach_fdatawrite_inode(&wbc, &inode->vfs_inode);
1148 extent_write_locked_range(&inode->vfs_inode, start, end, &wbc);
1149 wbc_detach_inode(&wbc);
1152 static void submit_one_async_extent(struct async_chunk *async_chunk,
1153 struct async_extent *async_extent,
1156 struct btrfs_inode *inode = async_chunk->inode;
1157 struct extent_io_tree *io_tree = &inode->io_tree;
1158 struct btrfs_root *root = inode->root;
1159 struct btrfs_fs_info *fs_info = root->fs_info;
1160 struct btrfs_ordered_extent *ordered;
1161 struct btrfs_key ins;
1162 struct page *locked_page = NULL;
1163 struct extent_map *em;
1165 u64 start = async_extent->start;
1166 u64 end = async_extent->start + async_extent->ram_size - 1;
1168 if (async_chunk->blkcg_css)
1169 kthread_associate_blkcg(async_chunk->blkcg_css);
1172 * If async_chunk->locked_page is in the async_extent range, we need to
1175 if (async_chunk->locked_page) {
1176 u64 locked_page_start = page_offset(async_chunk->locked_page);
1177 u64 locked_page_end = locked_page_start + PAGE_SIZE - 1;
1179 if (!(start >= locked_page_end || end <= locked_page_start))
1180 locked_page = async_chunk->locked_page;
1182 lock_extent(io_tree, start, end, NULL);
1184 if (async_extent->compress_type == BTRFS_COMPRESS_NONE) {
1185 submit_uncompressed_range(inode, async_extent, locked_page);
1189 ret = btrfs_reserve_extent(root, async_extent->ram_size,
1190 async_extent->compressed_size,
1191 async_extent->compressed_size,
1192 0, *alloc_hint, &ins, 1, 1);
1195 * Here we used to try again by going back to non-compressed
1196 * path for ENOSPC. But we can't reserve space even for
1197 * compressed size, how could it work for uncompressed size
1198 * which requires larger size? So here we directly go error
1204 /* Here we're doing allocation and writeback of the compressed pages */
1205 em = create_io_em(inode, start,
1206 async_extent->ram_size, /* len */
1207 start, /* orig_start */
1208 ins.objectid, /* block_start */
1209 ins.offset, /* block_len */
1210 ins.offset, /* orig_block_len */
1211 async_extent->ram_size, /* ram_bytes */
1212 async_extent->compress_type,
1213 BTRFS_ORDERED_COMPRESSED);
1216 goto out_free_reserve;
1218 free_extent_map(em);
1220 ordered = btrfs_alloc_ordered_extent(inode, start, /* file_offset */
1221 async_extent->ram_size, /* num_bytes */
1222 async_extent->ram_size, /* ram_bytes */
1223 ins.objectid, /* disk_bytenr */
1224 ins.offset, /* disk_num_bytes */
1226 1 << BTRFS_ORDERED_COMPRESSED,
1227 async_extent->compress_type);
1228 if (IS_ERR(ordered)) {
1229 btrfs_drop_extent_map_range(inode, start, end, false);
1230 ret = PTR_ERR(ordered);
1231 goto out_free_reserve;
1233 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1235 /* Clear dirty, set writeback and unlock the pages. */
1236 extent_clear_unlock_delalloc(inode, start, end,
1237 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
1238 PAGE_UNLOCK | PAGE_START_WRITEBACK);
1239 btrfs_submit_compressed_write(ordered,
1240 async_extent->pages, /* compressed_pages */
1241 async_extent->nr_pages,
1242 async_chunk->write_flags, true);
1243 *alloc_hint = ins.objectid + ins.offset;
1245 if (async_chunk->blkcg_css)
1246 kthread_associate_blkcg(NULL);
1247 kfree(async_extent);
1251 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1252 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1254 mapping_set_error(inode->vfs_inode.i_mapping, -EIO);
1255 extent_clear_unlock_delalloc(inode, start, end,
1256 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
1257 EXTENT_DELALLOC_NEW |
1258 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
1259 PAGE_UNLOCK | PAGE_START_WRITEBACK |
1260 PAGE_END_WRITEBACK);
1261 free_async_extent_pages(async_extent);
1262 if (async_chunk->blkcg_css)
1263 kthread_associate_blkcg(NULL);
1264 btrfs_debug(fs_info,
1265 "async extent submission failed root=%lld inode=%llu start=%llu len=%llu ret=%d",
1266 root->root_key.objectid, btrfs_ino(inode), start,
1267 async_extent->ram_size, ret);
1268 kfree(async_extent);
1271 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
1274 struct extent_map_tree *em_tree = &inode->extent_tree;
1275 struct extent_map *em;
1278 read_lock(&em_tree->lock);
1279 em = search_extent_mapping(em_tree, start, num_bytes);
1282 * if block start isn't an actual block number then find the
1283 * first block in this inode and use that as a hint. If that
1284 * block is also bogus then just don't worry about it.
1286 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1287 free_extent_map(em);
1288 em = search_extent_mapping(em_tree, 0, 0);
1289 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
1290 alloc_hint = em->block_start;
1292 free_extent_map(em);
1294 alloc_hint = em->block_start;
1295 free_extent_map(em);
1298 read_unlock(&em_tree->lock);
1304 * when extent_io.c finds a delayed allocation range in the file,
1305 * the call backs end up in this code. The basic idea is to
1306 * allocate extents on disk for the range, and create ordered data structs
1307 * in ram to track those extents.
1309 * locked_page is the page that writepage had locked already. We use
1310 * it to make sure we don't do extra locks or unlocks.
1312 * When this function fails, it unlocks all pages except @locked_page.
1314 * When this function successfully creates an inline extent, it returns 1 and
1315 * unlocks all pages including locked_page and starts I/O on them.
1316 * (In reality inline extents are limited to a single page, so locked_page is
1317 * the only page handled anyway).
1319 * When this function succeed and creates a normal extent, the page locking
1320 * status depends on the passed in flags:
1322 * - If @keep_locked is set, all pages are kept locked.
1323 * - Else all pages except for @locked_page are unlocked.
1325 * When a failure happens in the second or later iteration of the
1326 * while-loop, the ordered extents created in previous iterations are kept
1327 * intact. So, the caller must clean them up by calling
1328 * btrfs_cleanup_ordered_extents(). See btrfs_run_delalloc_range() for
1331 static noinline int cow_file_range(struct btrfs_inode *inode,
1332 struct page *locked_page, u64 start, u64 end,
1334 bool keep_locked, bool no_inline)
1336 struct btrfs_root *root = inode->root;
1337 struct btrfs_fs_info *fs_info = root->fs_info;
1339 u64 orig_start = start;
1341 unsigned long ram_size;
1342 u64 cur_alloc_size = 0;
1344 u64 blocksize = fs_info->sectorsize;
1345 struct btrfs_key ins;
1346 struct extent_map *em;
1347 unsigned clear_bits;
1348 unsigned long page_ops;
1349 bool extent_reserved = false;
1352 if (btrfs_is_free_space_inode(inode)) {
1357 num_bytes = ALIGN(end - start + 1, blocksize);
1358 num_bytes = max(blocksize, num_bytes);
1359 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1361 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1364 * Due to the page size limit, for subpage we can only trigger the
1365 * writeback for the dirty sectors of page, that means data writeback
1366 * is doing more writeback than what we want.
1368 * This is especially unexpected for some call sites like fallocate,
1369 * where we only increase i_size after everything is done.
1370 * This means we can trigger inline extent even if we didn't want to.
1371 * So here we skip inline extent creation completely.
1373 if (start == 0 && fs_info->sectorsize == PAGE_SIZE && !no_inline) {
1374 u64 actual_end = min_t(u64, i_size_read(&inode->vfs_inode),
1377 /* lets try to make an inline extent */
1378 ret = cow_file_range_inline(inode, actual_end, 0,
1379 BTRFS_COMPRESS_NONE, NULL, false);
1382 * We use DO_ACCOUNTING here because we need the
1383 * delalloc_release_metadata to be run _after_ we drop
1384 * our outstanding extent for clearing delalloc for this
1387 extent_clear_unlock_delalloc(inode, start, end,
1389 EXTENT_LOCKED | EXTENT_DELALLOC |
1390 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1391 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1392 PAGE_START_WRITEBACK | PAGE_END_WRITEBACK);
1394 * locked_page is locked by the caller of
1395 * writepage_delalloc(), not locked by
1396 * __process_pages_contig().
1398 * We can't let __process_pages_contig() to unlock it,
1399 * as it doesn't have any subpage::writers recorded.
1401 * Here we manually unlock the page, since the caller
1402 * can't determine if it's an inline extent or a
1403 * compressed extent.
1405 unlock_page(locked_page);
1407 } else if (ret < 0) {
1412 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1415 * Relocation relies on the relocated extents to have exactly the same
1416 * size as the original extents. Normally writeback for relocation data
1417 * extents follows a NOCOW path because relocation preallocates the
1418 * extents. However, due to an operation such as scrub turning a block
1419 * group to RO mode, it may fallback to COW mode, so we must make sure
1420 * an extent allocated during COW has exactly the requested size and can
1421 * not be split into smaller extents, otherwise relocation breaks and
1422 * fails during the stage where it updates the bytenr of file extent
1425 if (btrfs_is_data_reloc_root(root))
1426 min_alloc_size = num_bytes;
1428 min_alloc_size = fs_info->sectorsize;
1430 while (num_bytes > 0) {
1431 struct btrfs_ordered_extent *ordered;
1433 cur_alloc_size = num_bytes;
1434 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1435 min_alloc_size, 0, alloc_hint,
1439 cur_alloc_size = ins.offset;
1440 extent_reserved = true;
1442 ram_size = ins.offset;
1443 em = create_io_em(inode, start, ins.offset, /* len */
1444 start, /* orig_start */
1445 ins.objectid, /* block_start */
1446 ins.offset, /* block_len */
1447 ins.offset, /* orig_block_len */
1448 ram_size, /* ram_bytes */
1449 BTRFS_COMPRESS_NONE, /* compress_type */
1450 BTRFS_ORDERED_REGULAR /* type */);
1455 free_extent_map(em);
1457 ordered = btrfs_alloc_ordered_extent(inode, start, ram_size,
1458 ram_size, ins.objectid, cur_alloc_size,
1459 0, 1 << BTRFS_ORDERED_REGULAR,
1460 BTRFS_COMPRESS_NONE);
1461 if (IS_ERR(ordered)) {
1462 ret = PTR_ERR(ordered);
1463 goto out_drop_extent_cache;
1466 if (btrfs_is_data_reloc_root(root)) {
1467 ret = btrfs_reloc_clone_csums(ordered);
1470 * Only drop cache here, and process as normal.
1472 * We must not allow extent_clear_unlock_delalloc()
1473 * at out_unlock label to free meta of this ordered
1474 * extent, as its meta should be freed by
1475 * btrfs_finish_ordered_io().
1477 * So we must continue until @start is increased to
1478 * skip current ordered extent.
1481 btrfs_drop_extent_map_range(inode, start,
1482 start + ram_size - 1,
1485 btrfs_put_ordered_extent(ordered);
1487 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1490 * We're not doing compressed IO, don't unlock the first page
1491 * (which the caller expects to stay locked), don't clear any
1492 * dirty bits and don't set any writeback bits
1494 * Do set the Ordered (Private2) bit so we know this page was
1495 * properly setup for writepage.
1497 page_ops = (keep_locked ? 0 : PAGE_UNLOCK);
1498 page_ops |= PAGE_SET_ORDERED;
1500 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1502 EXTENT_LOCKED | EXTENT_DELALLOC,
1504 if (num_bytes < cur_alloc_size)
1507 num_bytes -= cur_alloc_size;
1508 alloc_hint = ins.objectid + ins.offset;
1509 start += cur_alloc_size;
1510 extent_reserved = false;
1513 * btrfs_reloc_clone_csums() error, since start is increased
1514 * extent_clear_unlock_delalloc() at out_unlock label won't
1515 * free metadata of current ordered extent, we're OK to exit.
1522 out_drop_extent_cache:
1523 btrfs_drop_extent_map_range(inode, start, start + ram_size - 1, false);
1525 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1526 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1529 * If done_offset is non-NULL and ret == -EAGAIN, we expect the
1530 * caller to write out the successfully allocated region and retry.
1532 if (done_offset && ret == -EAGAIN) {
1533 if (orig_start < start)
1534 *done_offset = start - 1;
1536 *done_offset = start;
1538 } else if (ret == -EAGAIN) {
1539 /* Convert to -ENOSPC since the caller cannot retry. */
1544 * Now, we have three regions to clean up:
1546 * |-------(1)----|---(2)---|-------------(3)----------|
1547 * `- orig_start `- start `- start + cur_alloc_size `- end
1549 * We process each region below.
1552 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1553 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1554 page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK;
1557 * For the range (1). We have already instantiated the ordered extents
1558 * for this region. They are cleaned up by
1559 * btrfs_cleanup_ordered_extents() in e.g,
1560 * btrfs_run_delalloc_range(). EXTENT_LOCKED | EXTENT_DELALLOC are
1561 * already cleared in the above loop. And, EXTENT_DELALLOC_NEW |
1562 * EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV are handled by the cleanup
1565 * However, in case of @keep_locked, we still need to unlock the pages
1566 * (except @locked_page) to ensure all the pages are unlocked.
1568 if (keep_locked && orig_start < start) {
1570 mapping_set_error(inode->vfs_inode.i_mapping, ret);
1571 extent_clear_unlock_delalloc(inode, orig_start, start - 1,
1572 locked_page, 0, page_ops);
1576 * For the range (2). If we reserved an extent for our delalloc range
1577 * (or a subrange) and failed to create the respective ordered extent,
1578 * then it means that when we reserved the extent we decremented the
1579 * extent's size from the data space_info's bytes_may_use counter and
1580 * incremented the space_info's bytes_reserved counter by the same
1581 * amount. We must make sure extent_clear_unlock_delalloc() does not try
1582 * to decrement again the data space_info's bytes_may_use counter,
1583 * therefore we do not pass it the flag EXTENT_CLEAR_DATA_RESV.
1585 if (extent_reserved) {
1586 extent_clear_unlock_delalloc(inode, start,
1587 start + cur_alloc_size - 1,
1591 start += cur_alloc_size;
1595 * For the range (3). We never touched the region. In addition to the
1596 * clear_bits above, we add EXTENT_CLEAR_DATA_RESV to release the data
1597 * space_info's bytes_may_use counter, reserved in
1598 * btrfs_check_data_free_space().
1601 clear_bits |= EXTENT_CLEAR_DATA_RESV;
1602 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1603 clear_bits, page_ops);
1609 * Phase two of compressed writeback. This is the ordered portion of the code,
1610 * which only gets called in the order the work was queued. We walk all the
1611 * async extents created by compress_file_range and send them down to the disk.
1613 static noinline void submit_compressed_extents(struct btrfs_work *work)
1615 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1617 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1618 struct async_extent *async_extent;
1619 unsigned long nr_pages;
1622 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1625 while (!list_empty(&async_chunk->extents)) {
1626 async_extent = list_entry(async_chunk->extents.next,
1627 struct async_extent, list);
1628 list_del(&async_extent->list);
1629 submit_one_async_extent(async_chunk, async_extent, &alloc_hint);
1632 /* atomic_sub_return implies a barrier */
1633 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1635 cond_wake_up_nomb(&fs_info->async_submit_wait);
1638 static noinline void async_cow_free(struct btrfs_work *work)
1640 struct async_chunk *async_chunk;
1641 struct async_cow *async_cow;
1643 async_chunk = container_of(work, struct async_chunk, work);
1644 btrfs_add_delayed_iput(async_chunk->inode);
1645 if (async_chunk->blkcg_css)
1646 css_put(async_chunk->blkcg_css);
1648 async_cow = async_chunk->async_cow;
1649 if (atomic_dec_and_test(&async_cow->num_chunks))
1653 static bool run_delalloc_compressed(struct btrfs_inode *inode,
1654 struct page *locked_page, u64 start,
1655 u64 end, struct writeback_control *wbc)
1657 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1658 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1659 struct async_cow *ctx;
1660 struct async_chunk *async_chunk;
1661 unsigned long nr_pages;
1662 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1665 const blk_opf_t write_flags = wbc_to_write_flags(wbc);
1667 nofs_flag = memalloc_nofs_save();
1668 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1669 memalloc_nofs_restore(nofs_flag);
1673 unlock_extent(&inode->io_tree, start, end, NULL);
1674 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
1676 async_chunk = ctx->chunks;
1677 atomic_set(&ctx->num_chunks, num_chunks);
1679 for (i = 0; i < num_chunks; i++) {
1680 u64 cur_end = min(end, start + SZ_512K - 1);
1683 * igrab is called higher up in the call chain, take only the
1684 * lightweight reference for the callback lifetime
1686 ihold(&inode->vfs_inode);
1687 async_chunk[i].async_cow = ctx;
1688 async_chunk[i].inode = inode;
1689 async_chunk[i].start = start;
1690 async_chunk[i].end = cur_end;
1691 async_chunk[i].write_flags = write_flags;
1692 INIT_LIST_HEAD(&async_chunk[i].extents);
1695 * The locked_page comes all the way from writepage and its
1696 * the original page we were actually given. As we spread
1697 * this large delalloc region across multiple async_chunk
1698 * structs, only the first struct needs a pointer to locked_page
1700 * This way we don't need racey decisions about who is supposed
1705 * Depending on the compressibility, the pages might or
1706 * might not go through async. We want all of them to
1707 * be accounted against wbc once. Let's do it here
1708 * before the paths diverge. wbc accounting is used
1709 * only for foreign writeback detection and doesn't
1710 * need full accuracy. Just account the whole thing
1711 * against the first page.
1713 wbc_account_cgroup_owner(wbc, locked_page,
1715 async_chunk[i].locked_page = locked_page;
1718 async_chunk[i].locked_page = NULL;
1721 if (blkcg_css != blkcg_root_css) {
1723 async_chunk[i].blkcg_css = blkcg_css;
1724 async_chunk[i].write_flags |= REQ_BTRFS_CGROUP_PUNT;
1726 async_chunk[i].blkcg_css = NULL;
1729 btrfs_init_work(&async_chunk[i].work, compress_file_range,
1730 submit_compressed_extents, async_cow_free);
1732 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1733 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1735 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1737 start = cur_end + 1;
1742 static noinline int run_delalloc_zoned(struct btrfs_inode *inode,
1743 struct page *locked_page, u64 start,
1744 u64 end, struct writeback_control *wbc)
1746 u64 done_offset = end;
1748 bool locked_page_done = false;
1750 while (start <= end) {
1751 ret = cow_file_range(inode, locked_page, start, end, &done_offset,
1753 if (ret && ret != -EAGAIN)
1759 if (done_offset == start) {
1760 wait_on_bit_io(&inode->root->fs_info->flags,
1761 BTRFS_FS_NEED_ZONE_FINISH,
1762 TASK_UNINTERRUPTIBLE);
1766 if (!locked_page_done) {
1767 __set_page_dirty_nobuffers(locked_page);
1768 account_page_redirty(locked_page);
1770 locked_page_done = true;
1771 extent_write_locked_range(&inode->vfs_inode, start, done_offset,
1773 start = done_offset + 1;
1779 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1780 u64 bytenr, u64 num_bytes, bool nowait)
1782 struct btrfs_root *csum_root = btrfs_csum_root(fs_info, bytenr);
1783 struct btrfs_ordered_sum *sums;
1787 ret = btrfs_lookup_csums_list(csum_root, bytenr, bytenr + num_bytes - 1,
1789 if (ret == 0 && list_empty(&list))
1792 while (!list_empty(&list)) {
1793 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1794 list_del(&sums->list);
1802 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1803 const u64 start, const u64 end)
1805 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1806 const bool is_reloc_ino = btrfs_is_data_reloc_root(inode->root);
1807 const u64 range_bytes = end + 1 - start;
1808 struct extent_io_tree *io_tree = &inode->io_tree;
1809 u64 range_start = start;
1814 * If EXTENT_NORESERVE is set it means that when the buffered write was
1815 * made we had not enough available data space and therefore we did not
1816 * reserve data space for it, since we though we could do NOCOW for the
1817 * respective file range (either there is prealloc extent or the inode
1818 * has the NOCOW bit set).
1820 * However when we need to fallback to COW mode (because for example the
1821 * block group for the corresponding extent was turned to RO mode by a
1822 * scrub or relocation) we need to do the following:
1824 * 1) We increment the bytes_may_use counter of the data space info.
1825 * If COW succeeds, it allocates a new data extent and after doing
1826 * that it decrements the space info's bytes_may_use counter and
1827 * increments its bytes_reserved counter by the same amount (we do
1828 * this at btrfs_add_reserved_bytes()). So we need to increment the
1829 * bytes_may_use counter to compensate (when space is reserved at
1830 * buffered write time, the bytes_may_use counter is incremented);
1832 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1833 * that if the COW path fails for any reason, it decrements (through
1834 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1835 * data space info, which we incremented in the step above.
1837 * If we need to fallback to cow and the inode corresponds to a free
1838 * space cache inode or an inode of the data relocation tree, we must
1839 * also increment bytes_may_use of the data space_info for the same
1840 * reason. Space caches and relocated data extents always get a prealloc
1841 * extent for them, however scrub or balance may have set the block
1842 * group that contains that extent to RO mode and therefore force COW
1843 * when starting writeback.
1845 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1846 EXTENT_NORESERVE, 0, NULL);
1847 if (count > 0 || is_space_ino || is_reloc_ino) {
1849 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1850 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1852 if (is_space_ino || is_reloc_ino)
1853 bytes = range_bytes;
1855 spin_lock(&sinfo->lock);
1856 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1857 spin_unlock(&sinfo->lock);
1860 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1865 * Don't try to create inline extents, as a mix of inline extent that
1866 * is written out and unlocked directly and a normal NOCOW extent
1869 ret = cow_file_range(inode, locked_page, start, end, NULL, false, true);
1874 struct can_nocow_file_extent_args {
1877 /* Start file offset of the range we want to NOCOW. */
1879 /* End file offset (inclusive) of the range we want to NOCOW. */
1881 bool writeback_path;
1884 * Free the path passed to can_nocow_file_extent() once it's not needed
1889 /* Output fields. Only set when can_nocow_file_extent() returns 1. */
1894 /* Number of bytes that can be written to in NOCOW mode. */
1899 * Check if we can NOCOW the file extent that the path points to.
1900 * This function may return with the path released, so the caller should check
1901 * if path->nodes[0] is NULL or not if it needs to use the path afterwards.
1903 * Returns: < 0 on error
1904 * 0 if we can not NOCOW
1907 static int can_nocow_file_extent(struct btrfs_path *path,
1908 struct btrfs_key *key,
1909 struct btrfs_inode *inode,
1910 struct can_nocow_file_extent_args *args)
1912 const bool is_freespace_inode = btrfs_is_free_space_inode(inode);
1913 struct extent_buffer *leaf = path->nodes[0];
1914 struct btrfs_root *root = inode->root;
1915 struct btrfs_file_extent_item *fi;
1920 bool nowait = path->nowait;
1922 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
1923 extent_type = btrfs_file_extent_type(leaf, fi);
1925 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1928 /* Can't access these fields unless we know it's not an inline extent. */
1929 args->disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1930 args->disk_num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1931 args->extent_offset = btrfs_file_extent_offset(leaf, fi);
1933 if (!(inode->flags & BTRFS_INODE_NODATACOW) &&
1934 extent_type == BTRFS_FILE_EXTENT_REG)
1938 * If the extent was created before the generation where the last snapshot
1939 * for its subvolume was created, then this implies the extent is shared,
1940 * hence we must COW.
1942 if (!args->strict &&
1943 btrfs_file_extent_generation(leaf, fi) <=
1944 btrfs_root_last_snapshot(&root->root_item))
1947 /* An explicit hole, must COW. */
1948 if (args->disk_bytenr == 0)
1951 /* Compressed/encrypted/encoded extents must be COWed. */
1952 if (btrfs_file_extent_compression(leaf, fi) ||
1953 btrfs_file_extent_encryption(leaf, fi) ||
1954 btrfs_file_extent_other_encoding(leaf, fi))
1957 extent_end = btrfs_file_extent_end(path);
1960 * The following checks can be expensive, as they need to take other
1961 * locks and do btree or rbtree searches, so release the path to avoid
1962 * blocking other tasks for too long.
1964 btrfs_release_path(path);
1966 ret = btrfs_cross_ref_exist(root, btrfs_ino(inode),
1967 key->offset - args->extent_offset,
1968 args->disk_bytenr, args->strict, path);
1969 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
1973 if (args->free_path) {
1975 * We don't need the path anymore, plus through the
1976 * csum_exist_in_range() call below we will end up allocating
1977 * another path. So free the path to avoid unnecessary extra
1980 btrfs_free_path(path);
1984 /* If there are pending snapshots for this root, we must COW. */
1985 if (args->writeback_path && !is_freespace_inode &&
1986 atomic_read(&root->snapshot_force_cow))
1989 args->disk_bytenr += args->extent_offset;
1990 args->disk_bytenr += args->start - key->offset;
1991 args->num_bytes = min(args->end + 1, extent_end) - args->start;
1994 * Force COW if csums exist in the range. This ensures that csums for a
1995 * given extent are either valid or do not exist.
1997 ret = csum_exist_in_range(root->fs_info, args->disk_bytenr, args->num_bytes,
1999 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
2005 if (args->free_path && path)
2006 btrfs_free_path(path);
2008 return ret < 0 ? ret : can_nocow;
2012 * when nowcow writeback call back. This checks for snapshots or COW copies
2013 * of the extents that exist in the file, and COWs the file as required.
2015 * If no cow copies or snapshots exist, we write directly to the existing
2018 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
2019 struct page *locked_page,
2020 const u64 start, const u64 end)
2022 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2023 struct btrfs_root *root = inode->root;
2024 struct btrfs_path *path;
2025 u64 cow_start = (u64)-1;
2026 u64 cur_offset = start;
2028 bool check_prev = true;
2029 u64 ino = btrfs_ino(inode);
2030 struct btrfs_block_group *bg;
2032 struct can_nocow_file_extent_args nocow_args = { 0 };
2034 path = btrfs_alloc_path();
2036 extent_clear_unlock_delalloc(inode, start, end, locked_page,
2037 EXTENT_LOCKED | EXTENT_DELALLOC |
2038 EXTENT_DO_ACCOUNTING |
2039 EXTENT_DEFRAG, PAGE_UNLOCK |
2040 PAGE_START_WRITEBACK |
2041 PAGE_END_WRITEBACK);
2045 nocow_args.end = end;
2046 nocow_args.writeback_path = true;
2049 struct btrfs_ordered_extent *ordered;
2050 struct btrfs_key found_key;
2051 struct btrfs_file_extent_item *fi;
2052 struct extent_buffer *leaf;
2061 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
2067 * If there is no extent for our range when doing the initial
2068 * search, then go back to the previous slot as it will be the
2069 * one containing the search offset
2071 if (ret > 0 && path->slots[0] > 0 && check_prev) {
2072 leaf = path->nodes[0];
2073 btrfs_item_key_to_cpu(leaf, &found_key,
2074 path->slots[0] - 1);
2075 if (found_key.objectid == ino &&
2076 found_key.type == BTRFS_EXTENT_DATA_KEY)
2081 /* Go to next leaf if we have exhausted the current one */
2082 leaf = path->nodes[0];
2083 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2084 ret = btrfs_next_leaf(root, path);
2086 if (cow_start != (u64)-1)
2087 cur_offset = cow_start;
2092 leaf = path->nodes[0];
2095 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2097 /* Didn't find anything for our INO */
2098 if (found_key.objectid > ino)
2101 * Keep searching until we find an EXTENT_ITEM or there are no
2102 * more extents for this inode
2104 if (WARN_ON_ONCE(found_key.objectid < ino) ||
2105 found_key.type < BTRFS_EXTENT_DATA_KEY) {
2110 /* Found key is not EXTENT_DATA_KEY or starts after req range */
2111 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
2112 found_key.offset > end)
2116 * If the found extent starts after requested offset, then
2117 * adjust extent_end to be right before this extent begins
2119 if (found_key.offset > cur_offset) {
2120 extent_end = found_key.offset;
2126 * Found extent which begins before our range and potentially
2129 fi = btrfs_item_ptr(leaf, path->slots[0],
2130 struct btrfs_file_extent_item);
2131 extent_type = btrfs_file_extent_type(leaf, fi);
2132 /* If this is triggered then we have a memory corruption. */
2133 ASSERT(extent_type < BTRFS_NR_FILE_EXTENT_TYPES);
2134 if (WARN_ON(extent_type >= BTRFS_NR_FILE_EXTENT_TYPES)) {
2138 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
2139 extent_end = btrfs_file_extent_end(path);
2142 * If the extent we got ends before our current offset, skip to
2145 if (extent_end <= cur_offset) {
2150 nocow_args.start = cur_offset;
2151 ret = can_nocow_file_extent(path, &found_key, inode, &nocow_args);
2153 if (cow_start != (u64)-1)
2154 cur_offset = cow_start;
2156 } else if (ret == 0) {
2161 bg = btrfs_inc_nocow_writers(fs_info, nocow_args.disk_bytenr);
2166 * If nocow is false then record the beginning of the range
2167 * that needs to be COWed
2170 if (cow_start == (u64)-1)
2171 cow_start = cur_offset;
2172 cur_offset = extent_end;
2173 if (cur_offset > end)
2175 if (!path->nodes[0])
2182 * COW range from cow_start to found_key.offset - 1. As the key
2183 * will contain the beginning of the first extent that can be
2184 * NOCOW, following one which needs to be COW'ed
2186 if (cow_start != (u64)-1) {
2187 ret = fallback_to_cow(inode, locked_page,
2188 cow_start, found_key.offset - 1);
2191 cow_start = (u64)-1;
2194 nocow_end = cur_offset + nocow_args.num_bytes - 1;
2195 is_prealloc = extent_type == BTRFS_FILE_EXTENT_PREALLOC;
2197 u64 orig_start = found_key.offset - nocow_args.extent_offset;
2198 struct extent_map *em;
2200 em = create_io_em(inode, cur_offset, nocow_args.num_bytes,
2202 nocow_args.disk_bytenr, /* block_start */
2203 nocow_args.num_bytes, /* block_len */
2204 nocow_args.disk_num_bytes, /* orig_block_len */
2205 ram_bytes, BTRFS_COMPRESS_NONE,
2206 BTRFS_ORDERED_PREALLOC);
2211 free_extent_map(em);
2214 ordered = btrfs_alloc_ordered_extent(inode, cur_offset,
2215 nocow_args.num_bytes, nocow_args.num_bytes,
2216 nocow_args.disk_bytenr, nocow_args.num_bytes, 0,
2218 ? (1 << BTRFS_ORDERED_PREALLOC)
2219 : (1 << BTRFS_ORDERED_NOCOW),
2220 BTRFS_COMPRESS_NONE);
2221 if (IS_ERR(ordered)) {
2223 btrfs_drop_extent_map_range(inode, cur_offset,
2226 ret = PTR_ERR(ordered);
2231 btrfs_dec_nocow_writers(bg);
2235 if (btrfs_is_data_reloc_root(root))
2237 * Error handled later, as we must prevent
2238 * extent_clear_unlock_delalloc() in error handler
2239 * from freeing metadata of created ordered extent.
2241 ret = btrfs_reloc_clone_csums(ordered);
2242 btrfs_put_ordered_extent(ordered);
2244 extent_clear_unlock_delalloc(inode, cur_offset, nocow_end,
2245 locked_page, EXTENT_LOCKED |
2247 EXTENT_CLEAR_DATA_RESV,
2248 PAGE_UNLOCK | PAGE_SET_ORDERED);
2250 cur_offset = extent_end;
2253 * btrfs_reloc_clone_csums() error, now we're OK to call error
2254 * handler, as metadata for created ordered extent will only
2255 * be freed by btrfs_finish_ordered_io().
2259 if (cur_offset > end)
2262 btrfs_release_path(path);
2264 if (cur_offset <= end && cow_start == (u64)-1)
2265 cow_start = cur_offset;
2267 if (cow_start != (u64)-1) {
2269 ret = fallback_to_cow(inode, locked_page, cow_start, end);
2276 btrfs_dec_nocow_writers(bg);
2278 if (ret && cur_offset < end)
2279 extent_clear_unlock_delalloc(inode, cur_offset, end,
2280 locked_page, EXTENT_LOCKED |
2281 EXTENT_DELALLOC | EXTENT_DEFRAG |
2282 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
2283 PAGE_START_WRITEBACK |
2284 PAGE_END_WRITEBACK);
2285 btrfs_free_path(path);
2289 static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end)
2291 if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) {
2292 if (inode->defrag_bytes &&
2293 test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG,
2302 * Function to process delayed allocation (create CoW) for ranges which are
2303 * being touched for the first time.
2305 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
2306 u64 start, u64 end, struct writeback_control *wbc)
2308 const bool zoned = btrfs_is_zoned(inode->root->fs_info);
2312 * The range must cover part of the @locked_page, or a return of 1
2313 * can confuse the caller.
2315 ASSERT(!(end <= page_offset(locked_page) ||
2316 start >= page_offset(locked_page) + PAGE_SIZE));
2318 if (should_nocow(inode, start, end)) {
2320 * Normally on a zoned device we're only doing COW writes, but
2321 * in case of relocation on a zoned filesystem we have taken
2322 * precaution, that we're only writing sequentially. It's safe
2323 * to use run_delalloc_nocow() here, like for regular
2324 * preallocated inodes.
2326 ASSERT(!zoned || btrfs_is_data_reloc_root(inode->root));
2327 ret = run_delalloc_nocow(inode, locked_page, start, end);
2331 if (btrfs_inode_can_compress(inode) &&
2332 inode_need_compress(inode, start, end) &&
2333 run_delalloc_compressed(inode, locked_page, start, end, wbc))
2337 ret = run_delalloc_zoned(inode, locked_page, start, end, wbc);
2339 ret = cow_file_range(inode, locked_page, start, end, NULL,
2344 btrfs_cleanup_ordered_extents(inode, locked_page, start,
2349 void btrfs_split_delalloc_extent(struct btrfs_inode *inode,
2350 struct extent_state *orig, u64 split)
2352 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2355 /* not delalloc, ignore it */
2356 if (!(orig->state & EXTENT_DELALLOC))
2359 size = orig->end - orig->start + 1;
2360 if (size > fs_info->max_extent_size) {
2365 * See the explanation in btrfs_merge_delalloc_extent, the same
2366 * applies here, just in reverse.
2368 new_size = orig->end - split + 1;
2369 num_extents = count_max_extents(fs_info, new_size);
2370 new_size = split - orig->start;
2371 num_extents += count_max_extents(fs_info, new_size);
2372 if (count_max_extents(fs_info, size) >= num_extents)
2376 spin_lock(&inode->lock);
2377 btrfs_mod_outstanding_extents(inode, 1);
2378 spin_unlock(&inode->lock);
2382 * Handle merged delayed allocation extents so we can keep track of new extents
2383 * that are just merged onto old extents, such as when we are doing sequential
2384 * writes, so we can properly account for the metadata space we'll need.
2386 void btrfs_merge_delalloc_extent(struct btrfs_inode *inode, struct extent_state *new,
2387 struct extent_state *other)
2389 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2390 u64 new_size, old_size;
2393 /* not delalloc, ignore it */
2394 if (!(other->state & EXTENT_DELALLOC))
2397 if (new->start > other->start)
2398 new_size = new->end - other->start + 1;
2400 new_size = other->end - new->start + 1;
2402 /* we're not bigger than the max, unreserve the space and go */
2403 if (new_size <= fs_info->max_extent_size) {
2404 spin_lock(&inode->lock);
2405 btrfs_mod_outstanding_extents(inode, -1);
2406 spin_unlock(&inode->lock);
2411 * We have to add up either side to figure out how many extents were
2412 * accounted for before we merged into one big extent. If the number of
2413 * extents we accounted for is <= the amount we need for the new range
2414 * then we can return, otherwise drop. Think of it like this
2418 * So we've grown the extent by a MAX_SIZE extent, this would mean we
2419 * need 2 outstanding extents, on one side we have 1 and the other side
2420 * we have 1 so they are == and we can return. But in this case
2422 * [MAX_SIZE+4k][MAX_SIZE+4k]
2424 * Each range on their own accounts for 2 extents, but merged together
2425 * they are only 3 extents worth of accounting, so we need to drop in
2428 old_size = other->end - other->start + 1;
2429 num_extents = count_max_extents(fs_info, old_size);
2430 old_size = new->end - new->start + 1;
2431 num_extents += count_max_extents(fs_info, old_size);
2432 if (count_max_extents(fs_info, new_size) >= num_extents)
2435 spin_lock(&inode->lock);
2436 btrfs_mod_outstanding_extents(inode, -1);
2437 spin_unlock(&inode->lock);
2440 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
2441 struct btrfs_inode *inode)
2443 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2445 spin_lock(&root->delalloc_lock);
2446 if (list_empty(&inode->delalloc_inodes)) {
2447 list_add_tail(&inode->delalloc_inodes, &root->delalloc_inodes);
2448 set_bit(BTRFS_INODE_IN_DELALLOC_LIST, &inode->runtime_flags);
2449 root->nr_delalloc_inodes++;
2450 if (root->nr_delalloc_inodes == 1) {
2451 spin_lock(&fs_info->delalloc_root_lock);
2452 BUG_ON(!list_empty(&root->delalloc_root));
2453 list_add_tail(&root->delalloc_root,
2454 &fs_info->delalloc_roots);
2455 spin_unlock(&fs_info->delalloc_root_lock);
2458 spin_unlock(&root->delalloc_lock);
2461 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
2462 struct btrfs_inode *inode)
2464 struct btrfs_fs_info *fs_info = root->fs_info;
2466 if (!list_empty(&inode->delalloc_inodes)) {
2467 list_del_init(&inode->delalloc_inodes);
2468 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2469 &inode->runtime_flags);
2470 root->nr_delalloc_inodes--;
2471 if (!root->nr_delalloc_inodes) {
2472 ASSERT(list_empty(&root->delalloc_inodes));
2473 spin_lock(&fs_info->delalloc_root_lock);
2474 BUG_ON(list_empty(&root->delalloc_root));
2475 list_del_init(&root->delalloc_root);
2476 spin_unlock(&fs_info->delalloc_root_lock);
2481 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
2482 struct btrfs_inode *inode)
2484 spin_lock(&root->delalloc_lock);
2485 __btrfs_del_delalloc_inode(root, inode);
2486 spin_unlock(&root->delalloc_lock);
2490 * Properly track delayed allocation bytes in the inode and to maintain the
2491 * list of inodes that have pending delalloc work to be done.
2493 void btrfs_set_delalloc_extent(struct btrfs_inode *inode, struct extent_state *state,
2496 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2498 if ((bits & EXTENT_DEFRAG) && !(bits & EXTENT_DELALLOC))
2501 * set_bit and clear bit hooks normally require _irqsave/restore
2502 * but in this case, we are only testing for the DELALLOC
2503 * bit, which is only set or cleared with irqs on
2505 if (!(state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2506 struct btrfs_root *root = inode->root;
2507 u64 len = state->end + 1 - state->start;
2508 u32 num_extents = count_max_extents(fs_info, len);
2509 bool do_list = !btrfs_is_free_space_inode(inode);
2511 spin_lock(&inode->lock);
2512 btrfs_mod_outstanding_extents(inode, num_extents);
2513 spin_unlock(&inode->lock);
2515 /* For sanity tests */
2516 if (btrfs_is_testing(fs_info))
2519 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2520 fs_info->delalloc_batch);
2521 spin_lock(&inode->lock);
2522 inode->delalloc_bytes += len;
2523 if (bits & EXTENT_DEFRAG)
2524 inode->defrag_bytes += len;
2525 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2526 &inode->runtime_flags))
2527 btrfs_add_delalloc_inodes(root, inode);
2528 spin_unlock(&inode->lock);
2531 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2532 (bits & EXTENT_DELALLOC_NEW)) {
2533 spin_lock(&inode->lock);
2534 inode->new_delalloc_bytes += state->end + 1 - state->start;
2535 spin_unlock(&inode->lock);
2540 * Once a range is no longer delalloc this function ensures that proper
2541 * accounting happens.
2543 void btrfs_clear_delalloc_extent(struct btrfs_inode *inode,
2544 struct extent_state *state, u32 bits)
2546 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2547 u64 len = state->end + 1 - state->start;
2548 u32 num_extents = count_max_extents(fs_info, len);
2550 if ((state->state & EXTENT_DEFRAG) && (bits & EXTENT_DEFRAG)) {
2551 spin_lock(&inode->lock);
2552 inode->defrag_bytes -= len;
2553 spin_unlock(&inode->lock);
2557 * set_bit and clear bit hooks normally require _irqsave/restore
2558 * but in this case, we are only testing for the DELALLOC
2559 * bit, which is only set or cleared with irqs on
2561 if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2562 struct btrfs_root *root = inode->root;
2563 bool do_list = !btrfs_is_free_space_inode(inode);
2565 spin_lock(&inode->lock);
2566 btrfs_mod_outstanding_extents(inode, -num_extents);
2567 spin_unlock(&inode->lock);
2570 * We don't reserve metadata space for space cache inodes so we
2571 * don't need to call delalloc_release_metadata if there is an
2574 if (bits & EXTENT_CLEAR_META_RESV &&
2575 root != fs_info->tree_root)
2576 btrfs_delalloc_release_metadata(inode, len, false);
2578 /* For sanity tests. */
2579 if (btrfs_is_testing(fs_info))
2582 if (!btrfs_is_data_reloc_root(root) &&
2583 do_list && !(state->state & EXTENT_NORESERVE) &&
2584 (bits & EXTENT_CLEAR_DATA_RESV))
2585 btrfs_free_reserved_data_space_noquota(fs_info, len);
2587 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2588 fs_info->delalloc_batch);
2589 spin_lock(&inode->lock);
2590 inode->delalloc_bytes -= len;
2591 if (do_list && inode->delalloc_bytes == 0 &&
2592 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2593 &inode->runtime_flags))
2594 btrfs_del_delalloc_inode(root, inode);
2595 spin_unlock(&inode->lock);
2598 if ((state->state & EXTENT_DELALLOC_NEW) &&
2599 (bits & EXTENT_DELALLOC_NEW)) {
2600 spin_lock(&inode->lock);
2601 ASSERT(inode->new_delalloc_bytes >= len);
2602 inode->new_delalloc_bytes -= len;
2603 if (bits & EXTENT_ADD_INODE_BYTES)
2604 inode_add_bytes(&inode->vfs_inode, len);
2605 spin_unlock(&inode->lock);
2609 static int btrfs_extract_ordered_extent(struct btrfs_bio *bbio,
2610 struct btrfs_ordered_extent *ordered)
2612 u64 start = (u64)bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT;
2613 u64 len = bbio->bio.bi_iter.bi_size;
2614 struct btrfs_ordered_extent *new;
2617 /* Must always be called for the beginning of an ordered extent. */
2618 if (WARN_ON_ONCE(start != ordered->disk_bytenr))
2621 /* No need to split if the ordered extent covers the entire bio. */
2622 if (ordered->disk_num_bytes == len) {
2623 refcount_inc(&ordered->refs);
2624 bbio->ordered = ordered;
2629 * Don't split the extent_map for NOCOW extents, as we're writing into
2630 * a pre-existing one.
2632 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
2633 ret = split_extent_map(bbio->inode, bbio->file_offset,
2634 ordered->num_bytes, len,
2635 ordered->disk_bytenr);
2640 new = btrfs_split_ordered_extent(ordered, len);
2642 return PTR_ERR(new);
2643 bbio->ordered = new;
2648 * given a list of ordered sums record them in the inode. This happens
2649 * at IO completion time based on sums calculated at bio submission time.
2651 static int add_pending_csums(struct btrfs_trans_handle *trans,
2652 struct list_head *list)
2654 struct btrfs_ordered_sum *sum;
2655 struct btrfs_root *csum_root = NULL;
2658 list_for_each_entry(sum, list, list) {
2659 trans->adding_csums = true;
2661 csum_root = btrfs_csum_root(trans->fs_info,
2663 ret = btrfs_csum_file_blocks(trans, csum_root, sum);
2664 trans->adding_csums = false;
2671 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2674 struct extent_state **cached_state)
2676 u64 search_start = start;
2677 const u64 end = start + len - 1;
2679 while (search_start < end) {
2680 const u64 search_len = end - search_start + 1;
2681 struct extent_map *em;
2685 em = btrfs_get_extent(inode, NULL, 0, search_start, search_len);
2689 if (em->block_start != EXTENT_MAP_HOLE)
2693 if (em->start < search_start)
2694 em_len -= search_start - em->start;
2695 if (em_len > search_len)
2696 em_len = search_len;
2698 ret = set_extent_bit(&inode->io_tree, search_start,
2699 search_start + em_len - 1,
2700 EXTENT_DELALLOC_NEW, cached_state);
2702 search_start = extent_map_end(em);
2703 free_extent_map(em);
2710 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2711 unsigned int extra_bits,
2712 struct extent_state **cached_state)
2714 WARN_ON(PAGE_ALIGNED(end));
2716 if (start >= i_size_read(&inode->vfs_inode) &&
2717 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2719 * There can't be any extents following eof in this case so just
2720 * set the delalloc new bit for the range directly.
2722 extra_bits |= EXTENT_DELALLOC_NEW;
2726 ret = btrfs_find_new_delalloc_bytes(inode, start,
2733 return set_extent_bit(&inode->io_tree, start, end,
2734 EXTENT_DELALLOC | extra_bits, cached_state);
2737 /* see btrfs_writepage_start_hook for details on why this is required */
2738 struct btrfs_writepage_fixup {
2740 struct btrfs_inode *inode;
2741 struct btrfs_work work;
2744 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2746 struct btrfs_writepage_fixup *fixup =
2747 container_of(work, struct btrfs_writepage_fixup, work);
2748 struct btrfs_ordered_extent *ordered;
2749 struct extent_state *cached_state = NULL;
2750 struct extent_changeset *data_reserved = NULL;
2751 struct page *page = fixup->page;
2752 struct btrfs_inode *inode = fixup->inode;
2753 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2754 u64 page_start = page_offset(page);
2755 u64 page_end = page_offset(page) + PAGE_SIZE - 1;
2757 bool free_delalloc_space = true;
2760 * This is similar to page_mkwrite, we need to reserve the space before
2761 * we take the page lock.
2763 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2769 * Before we queued this fixup, we took a reference on the page.
2770 * page->mapping may go NULL, but it shouldn't be moved to a different
2773 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2775 * Unfortunately this is a little tricky, either
2777 * 1) We got here and our page had already been dealt with and
2778 * we reserved our space, thus ret == 0, so we need to just
2779 * drop our space reservation and bail. This can happen the
2780 * first time we come into the fixup worker, or could happen
2781 * while waiting for the ordered extent.
2782 * 2) Our page was already dealt with, but we happened to get an
2783 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2784 * this case we obviously don't have anything to release, but
2785 * because the page was already dealt with we don't want to
2786 * mark the page with an error, so make sure we're resetting
2787 * ret to 0. This is why we have this check _before_ the ret
2788 * check, because we do not want to have a surprise ENOSPC
2789 * when the page was already properly dealt with.
2792 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2793 btrfs_delalloc_release_space(inode, data_reserved,
2794 page_start, PAGE_SIZE,
2802 * We can't mess with the page state unless it is locked, so now that
2803 * it is locked bail if we failed to make our space reservation.
2808 lock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2810 /* already ordered? We're done */
2811 if (PageOrdered(page))
2814 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2816 unlock_extent(&inode->io_tree, page_start, page_end,
2819 btrfs_start_ordered_extent(ordered);
2820 btrfs_put_ordered_extent(ordered);
2824 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2830 * Everything went as planned, we're now the owner of a dirty page with
2831 * delayed allocation bits set and space reserved for our COW
2834 * The page was dirty when we started, nothing should have cleaned it.
2836 BUG_ON(!PageDirty(page));
2837 free_delalloc_space = false;
2839 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2840 if (free_delalloc_space)
2841 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2843 unlock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2847 * We hit ENOSPC or other errors. Update the mapping and page
2848 * to reflect the errors and clean the page.
2850 mapping_set_error(page->mapping, ret);
2851 btrfs_mark_ordered_io_finished(inode, page, page_start,
2853 btrfs_page_clear_uptodate(fs_info, page, page_start, PAGE_SIZE);
2854 clear_page_dirty_for_io(page);
2856 btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE);
2860 extent_changeset_free(data_reserved);
2862 * As a precaution, do a delayed iput in case it would be the last iput
2863 * that could need flushing space. Recursing back to fixup worker would
2866 btrfs_add_delayed_iput(inode);
2870 * There are a few paths in the higher layers of the kernel that directly
2871 * set the page dirty bit without asking the filesystem if it is a
2872 * good idea. This causes problems because we want to make sure COW
2873 * properly happens and the data=ordered rules are followed.
2875 * In our case any range that doesn't have the ORDERED bit set
2876 * hasn't been properly setup for IO. We kick off an async process
2877 * to fix it up. The async helper will wait for ordered extents, set
2878 * the delalloc bit and make it safe to write the page.
2880 int btrfs_writepage_cow_fixup(struct page *page)
2882 struct inode *inode = page->mapping->host;
2883 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2884 struct btrfs_writepage_fixup *fixup;
2886 /* This page has ordered extent covering it already */
2887 if (PageOrdered(page))
2891 * PageChecked is set below when we create a fixup worker for this page,
2892 * don't try to create another one if we're already PageChecked()
2894 * The extent_io writepage code will redirty the page if we send back
2897 if (PageChecked(page))
2900 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2905 * We are already holding a reference to this inode from
2906 * write_cache_pages. We need to hold it because the space reservation
2907 * takes place outside of the page lock, and we can't trust
2908 * page->mapping outside of the page lock.
2911 btrfs_page_set_checked(fs_info, page, page_offset(page), PAGE_SIZE);
2913 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
2915 fixup->inode = BTRFS_I(inode);
2916 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2921 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2922 struct btrfs_inode *inode, u64 file_pos,
2923 struct btrfs_file_extent_item *stack_fi,
2924 const bool update_inode_bytes,
2925 u64 qgroup_reserved)
2927 struct btrfs_root *root = inode->root;
2928 const u64 sectorsize = root->fs_info->sectorsize;
2929 struct btrfs_path *path;
2930 struct extent_buffer *leaf;
2931 struct btrfs_key ins;
2932 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
2933 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
2934 u64 offset = btrfs_stack_file_extent_offset(stack_fi);
2935 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
2936 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
2937 struct btrfs_drop_extents_args drop_args = { 0 };
2940 path = btrfs_alloc_path();
2945 * we may be replacing one extent in the tree with another.
2946 * The new extent is pinned in the extent map, and we don't want
2947 * to drop it from the cache until it is completely in the btree.
2949 * So, tell btrfs_drop_extents to leave this extent in the cache.
2950 * the caller is expected to unpin it and allow it to be merged
2953 drop_args.path = path;
2954 drop_args.start = file_pos;
2955 drop_args.end = file_pos + num_bytes;
2956 drop_args.replace_extent = true;
2957 drop_args.extent_item_size = sizeof(*stack_fi);
2958 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2962 if (!drop_args.extent_inserted) {
2963 ins.objectid = btrfs_ino(inode);
2964 ins.offset = file_pos;
2965 ins.type = BTRFS_EXTENT_DATA_KEY;
2967 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2972 leaf = path->nodes[0];
2973 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
2974 write_extent_buffer(leaf, stack_fi,
2975 btrfs_item_ptr_offset(leaf, path->slots[0]),
2976 sizeof(struct btrfs_file_extent_item));
2978 btrfs_mark_buffer_dirty(leaf);
2979 btrfs_release_path(path);
2982 * If we dropped an inline extent here, we know the range where it is
2983 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
2984 * number of bytes only for that range containing the inline extent.
2985 * The remaining of the range will be processed when clearning the
2986 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
2988 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
2989 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
2991 inline_size = drop_args.bytes_found - inline_size;
2992 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
2993 drop_args.bytes_found -= inline_size;
2994 num_bytes -= sectorsize;
2997 if (update_inode_bytes)
2998 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
3000 ins.objectid = disk_bytenr;
3001 ins.offset = disk_num_bytes;
3002 ins.type = BTRFS_EXTENT_ITEM_KEY;
3004 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
3008 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
3010 qgroup_reserved, &ins);
3012 btrfs_free_path(path);
3017 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
3020 struct btrfs_block_group *cache;
3022 cache = btrfs_lookup_block_group(fs_info, start);
3025 spin_lock(&cache->lock);
3026 cache->delalloc_bytes -= len;
3027 spin_unlock(&cache->lock);
3029 btrfs_put_block_group(cache);
3032 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
3033 struct btrfs_ordered_extent *oe)
3035 struct btrfs_file_extent_item stack_fi;
3036 bool update_inode_bytes;
3037 u64 num_bytes = oe->num_bytes;
3038 u64 ram_bytes = oe->ram_bytes;
3040 memset(&stack_fi, 0, sizeof(stack_fi));
3041 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
3042 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
3043 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
3044 oe->disk_num_bytes);
3045 btrfs_set_stack_file_extent_offset(&stack_fi, oe->offset);
3046 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags)) {
3047 num_bytes = oe->truncated_len;
3048 ram_bytes = num_bytes;
3050 btrfs_set_stack_file_extent_num_bytes(&stack_fi, num_bytes);
3051 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, ram_bytes);
3052 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
3053 /* Encryption and other encoding is reserved and all 0 */
3056 * For delalloc, when completing an ordered extent we update the inode's
3057 * bytes when clearing the range in the inode's io tree, so pass false
3058 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
3059 * except if the ordered extent was truncated.
3061 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
3062 test_bit(BTRFS_ORDERED_ENCODED, &oe->flags) ||
3063 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
3065 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
3066 oe->file_offset, &stack_fi,
3067 update_inode_bytes, oe->qgroup_rsv);
3071 * As ordered data IO finishes, this gets called so we can finish
3072 * an ordered extent if the range of bytes in the file it covers are
3075 int btrfs_finish_one_ordered(struct btrfs_ordered_extent *ordered_extent)
3077 struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
3078 struct btrfs_root *root = inode->root;
3079 struct btrfs_fs_info *fs_info = root->fs_info;
3080 struct btrfs_trans_handle *trans = NULL;
3081 struct extent_io_tree *io_tree = &inode->io_tree;
3082 struct extent_state *cached_state = NULL;
3084 int compress_type = 0;
3086 u64 logical_len = ordered_extent->num_bytes;
3087 bool freespace_inode;
3088 bool truncated = false;
3089 bool clear_reserved_extent = true;
3090 unsigned int clear_bits = EXTENT_DEFRAG;
3092 start = ordered_extent->file_offset;
3093 end = start + ordered_extent->num_bytes - 1;
3095 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3096 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
3097 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags) &&
3098 !test_bit(BTRFS_ORDERED_ENCODED, &ordered_extent->flags))
3099 clear_bits |= EXTENT_DELALLOC_NEW;
3101 freespace_inode = btrfs_is_free_space_inode(inode);
3102 if (!freespace_inode)
3103 btrfs_lockdep_acquire(fs_info, btrfs_ordered_extent);
3105 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
3110 if (btrfs_is_zoned(fs_info))
3111 btrfs_zone_finish_endio(fs_info, ordered_extent->disk_bytenr,
3112 ordered_extent->disk_num_bytes);
3114 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
3116 logical_len = ordered_extent->truncated_len;
3117 /* Truncated the entire extent, don't bother adding */
3122 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
3123 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
3125 btrfs_inode_safe_disk_i_size_write(inode, 0);
3126 if (freespace_inode)
3127 trans = btrfs_join_transaction_spacecache(root);
3129 trans = btrfs_join_transaction(root);
3130 if (IS_ERR(trans)) {
3131 ret = PTR_ERR(trans);
3135 trans->block_rsv = &inode->block_rsv;
3136 ret = btrfs_update_inode_fallback(trans, root, inode);
3137 if (ret) /* -ENOMEM or corruption */
3138 btrfs_abort_transaction(trans, ret);
3142 clear_bits |= EXTENT_LOCKED;
3143 lock_extent(io_tree, start, end, &cached_state);
3145 if (freespace_inode)
3146 trans = btrfs_join_transaction_spacecache(root);
3148 trans = btrfs_join_transaction(root);
3149 if (IS_ERR(trans)) {
3150 ret = PTR_ERR(trans);
3155 trans->block_rsv = &inode->block_rsv;
3157 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3158 compress_type = ordered_extent->compress_type;
3159 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3160 BUG_ON(compress_type);
3161 ret = btrfs_mark_extent_written(trans, inode,
3162 ordered_extent->file_offset,
3163 ordered_extent->file_offset +
3165 btrfs_zoned_release_data_reloc_bg(fs_info, ordered_extent->disk_bytenr,
3166 ordered_extent->disk_num_bytes);
3168 BUG_ON(root == fs_info->tree_root);
3169 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
3171 clear_reserved_extent = false;
3172 btrfs_release_delalloc_bytes(fs_info,
3173 ordered_extent->disk_bytenr,
3174 ordered_extent->disk_num_bytes);
3177 unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset,
3178 ordered_extent->num_bytes, trans->transid);
3180 btrfs_abort_transaction(trans, ret);
3184 ret = add_pending_csums(trans, &ordered_extent->list);
3186 btrfs_abort_transaction(trans, ret);
3191 * If this is a new delalloc range, clear its new delalloc flag to
3192 * update the inode's number of bytes. This needs to be done first
3193 * before updating the inode item.
3195 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
3196 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
3197 clear_extent_bit(&inode->io_tree, start, end,
3198 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
3201 btrfs_inode_safe_disk_i_size_write(inode, 0);
3202 ret = btrfs_update_inode_fallback(trans, root, inode);
3203 if (ret) { /* -ENOMEM or corruption */
3204 btrfs_abort_transaction(trans, ret);
3209 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
3213 btrfs_end_transaction(trans);
3215 if (ret || truncated) {
3216 u64 unwritten_start = start;
3219 * If we failed to finish this ordered extent for any reason we
3220 * need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3221 * extent, and mark the inode with the error if it wasn't
3222 * already set. Any error during writeback would have already
3223 * set the mapping error, so we need to set it if we're the ones
3224 * marking this ordered extent as failed.
3226 if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR,
3227 &ordered_extent->flags))
3228 mapping_set_error(ordered_extent->inode->i_mapping, -EIO);
3231 unwritten_start += logical_len;
3232 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
3234 /* Drop extent maps for the part of the extent we didn't write. */
3235 btrfs_drop_extent_map_range(inode, unwritten_start, end, false);
3238 * If the ordered extent had an IOERR or something else went
3239 * wrong we need to return the space for this ordered extent
3240 * back to the allocator. We only free the extent in the
3241 * truncated case if we didn't write out the extent at all.
3243 * If we made it past insert_reserved_file_extent before we
3244 * errored out then we don't need to do this as the accounting
3245 * has already been done.
3247 if ((ret || !logical_len) &&
3248 clear_reserved_extent &&
3249 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3250 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3252 * Discard the range before returning it back to the
3255 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3256 btrfs_discard_extent(fs_info,
3257 ordered_extent->disk_bytenr,
3258 ordered_extent->disk_num_bytes,
3260 btrfs_free_reserved_extent(fs_info,
3261 ordered_extent->disk_bytenr,
3262 ordered_extent->disk_num_bytes, 1);
3264 * Actually free the qgroup rsv which was released when
3265 * the ordered extent was created.
3267 btrfs_qgroup_free_refroot(fs_info, inode->root->root_key.objectid,
3268 ordered_extent->qgroup_rsv,
3269 BTRFS_QGROUP_RSV_DATA);
3274 * This needs to be done to make sure anybody waiting knows we are done
3275 * updating everything for this ordered extent.
3277 btrfs_remove_ordered_extent(inode, ordered_extent);
3280 btrfs_put_ordered_extent(ordered_extent);
3281 /* once for the tree */
3282 btrfs_put_ordered_extent(ordered_extent);
3287 int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered)
3289 if (btrfs_is_zoned(btrfs_sb(ordered->inode->i_sb)) &&
3290 !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
3291 btrfs_finish_ordered_zoned(ordered);
3292 return btrfs_finish_one_ordered(ordered);
3296 * Verify the checksum for a single sector without any extra action that depend
3297 * on the type of I/O.
3299 int btrfs_check_sector_csum(struct btrfs_fs_info *fs_info, struct page *page,
3300 u32 pgoff, u8 *csum, const u8 * const csum_expected)
3302 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3305 ASSERT(pgoff + fs_info->sectorsize <= PAGE_SIZE);
3307 shash->tfm = fs_info->csum_shash;
3309 kaddr = kmap_local_page(page) + pgoff;
3310 crypto_shash_digest(shash, kaddr, fs_info->sectorsize, csum);
3311 kunmap_local(kaddr);
3313 if (memcmp(csum, csum_expected, fs_info->csum_size))
3319 * Verify the checksum of a single data sector.
3321 * @bbio: btrfs_io_bio which contains the csum
3322 * @dev: device the sector is on
3323 * @bio_offset: offset to the beginning of the bio (in bytes)
3324 * @bv: bio_vec to check
3326 * Check if the checksum on a data block is valid. When a checksum mismatch is
3327 * detected, report the error and fill the corrupted range with zero.
3329 * Return %true if the sector is ok or had no checksum to start with, else %false.
3331 bool btrfs_data_csum_ok(struct btrfs_bio *bbio, struct btrfs_device *dev,
3332 u32 bio_offset, struct bio_vec *bv)
3334 struct btrfs_inode *inode = bbio->inode;
3335 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3336 u64 file_offset = bbio->file_offset + bio_offset;
3337 u64 end = file_offset + bv->bv_len - 1;
3339 u8 csum[BTRFS_CSUM_SIZE];
3341 ASSERT(bv->bv_len == fs_info->sectorsize);
3346 if (btrfs_is_data_reloc_root(inode->root) &&
3347 test_range_bit(&inode->io_tree, file_offset, end, EXTENT_NODATASUM,
3349 /* Skip the range without csum for data reloc inode */
3350 clear_extent_bits(&inode->io_tree, file_offset, end,
3355 csum_expected = bbio->csum + (bio_offset >> fs_info->sectorsize_bits) *
3357 if (btrfs_check_sector_csum(fs_info, bv->bv_page, bv->bv_offset, csum,
3363 btrfs_print_data_csum_error(inode, file_offset, csum, csum_expected,
3366 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS);
3372 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3374 * @inode: The inode we want to perform iput on
3376 * This function uses the generic vfs_inode::i_count to track whether we should
3377 * just decrement it (in case it's > 1) or if this is the last iput then link
3378 * the inode to the delayed iput machinery. Delayed iputs are processed at
3379 * transaction commit time/superblock commit/cleaner kthread.
3381 void btrfs_add_delayed_iput(struct btrfs_inode *inode)
3383 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3384 unsigned long flags;
3386 if (atomic_add_unless(&inode->vfs_inode.i_count, -1, 1))
3389 atomic_inc(&fs_info->nr_delayed_iputs);
3391 * Need to be irq safe here because we can be called from either an irq
3392 * context (see bio.c and btrfs_put_ordered_extent()) or a non-irq
3395 spin_lock_irqsave(&fs_info->delayed_iput_lock, flags);
3396 ASSERT(list_empty(&inode->delayed_iput));
3397 list_add_tail(&inode->delayed_iput, &fs_info->delayed_iputs);
3398 spin_unlock_irqrestore(&fs_info->delayed_iput_lock, flags);
3399 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3400 wake_up_process(fs_info->cleaner_kthread);
3403 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3404 struct btrfs_inode *inode)
3406 list_del_init(&inode->delayed_iput);
3407 spin_unlock_irq(&fs_info->delayed_iput_lock);
3408 iput(&inode->vfs_inode);
3409 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3410 wake_up(&fs_info->delayed_iputs_wait);
3411 spin_lock_irq(&fs_info->delayed_iput_lock);
3414 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3415 struct btrfs_inode *inode)
3417 if (!list_empty(&inode->delayed_iput)) {
3418 spin_lock_irq(&fs_info->delayed_iput_lock);
3419 if (!list_empty(&inode->delayed_iput))
3420 run_delayed_iput_locked(fs_info, inode);
3421 spin_unlock_irq(&fs_info->delayed_iput_lock);
3425 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3428 * btrfs_put_ordered_extent() can run in irq context (see bio.c), which
3429 * calls btrfs_add_delayed_iput() and that needs to lock
3430 * fs_info->delayed_iput_lock. So we need to disable irqs here to
3431 * prevent a deadlock.
3433 spin_lock_irq(&fs_info->delayed_iput_lock);
3434 while (!list_empty(&fs_info->delayed_iputs)) {
3435 struct btrfs_inode *inode;
3437 inode = list_first_entry(&fs_info->delayed_iputs,
3438 struct btrfs_inode, delayed_iput);
3439 run_delayed_iput_locked(fs_info, inode);
3440 if (need_resched()) {
3441 spin_unlock_irq(&fs_info->delayed_iput_lock);
3443 spin_lock_irq(&fs_info->delayed_iput_lock);
3446 spin_unlock_irq(&fs_info->delayed_iput_lock);
3450 * Wait for flushing all delayed iputs
3452 * @fs_info: the filesystem
3454 * This will wait on any delayed iputs that are currently running with KILLABLE
3455 * set. Once they are all done running we will return, unless we are killed in
3456 * which case we return EINTR. This helps in user operations like fallocate etc
3457 * that might get blocked on the iputs.
3459 * Return EINTR if we were killed, 0 if nothing's pending
3461 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3463 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3464 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3471 * This creates an orphan entry for the given inode in case something goes wrong
3472 * in the middle of an unlink.
3474 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3475 struct btrfs_inode *inode)
3479 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3480 if (ret && ret != -EEXIST) {
3481 btrfs_abort_transaction(trans, ret);
3489 * We have done the delete so we can go ahead and remove the orphan item for
3490 * this particular inode.
3492 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3493 struct btrfs_inode *inode)
3495 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3499 * this cleans up any orphans that may be left on the list from the last use
3502 int btrfs_orphan_cleanup(struct btrfs_root *root)
3504 struct btrfs_fs_info *fs_info = root->fs_info;
3505 struct btrfs_path *path;
3506 struct extent_buffer *leaf;
3507 struct btrfs_key key, found_key;
3508 struct btrfs_trans_handle *trans;
3509 struct inode *inode;
3510 u64 last_objectid = 0;
3511 int ret = 0, nr_unlink = 0;
3513 if (test_and_set_bit(BTRFS_ROOT_ORPHAN_CLEANUP, &root->state))
3516 path = btrfs_alloc_path();
3521 path->reada = READA_BACK;
3523 key.objectid = BTRFS_ORPHAN_OBJECTID;
3524 key.type = BTRFS_ORPHAN_ITEM_KEY;
3525 key.offset = (u64)-1;
3528 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3533 * if ret == 0 means we found what we were searching for, which
3534 * is weird, but possible, so only screw with path if we didn't
3535 * find the key and see if we have stuff that matches
3539 if (path->slots[0] == 0)
3544 /* pull out the item */
3545 leaf = path->nodes[0];
3546 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3548 /* make sure the item matches what we want */
3549 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3551 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3554 /* release the path since we're done with it */
3555 btrfs_release_path(path);
3558 * this is where we are basically btrfs_lookup, without the
3559 * crossing root thing. we store the inode number in the
3560 * offset of the orphan item.
3563 if (found_key.offset == last_objectid) {
3565 "Error removing orphan entry, stopping orphan cleanup");
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 = btrfs_sb(inode->i_sb);
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->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3814 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3816 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3817 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3819 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3820 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3822 BTRFS_I(inode)->i_otime.tv_sec =
3823 btrfs_timespec_sec(leaf, &inode_item->otime);
3824 BTRFS_I(inode)->i_otime.tv_nsec =
3825 btrfs_timespec_nsec(leaf, &inode_item->otime);
3827 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3828 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3829 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3831 inode_set_iversion_queried(inode,
3832 btrfs_inode_sequence(leaf, inode_item));
3833 inode->i_generation = BTRFS_I(inode)->generation;
3835 rdev = btrfs_inode_rdev(leaf, inode_item);
3837 BTRFS_I(inode)->index_cnt = (u64)-1;
3838 btrfs_inode_split_flags(btrfs_inode_flags(leaf, inode_item),
3839 &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags);
3843 * If we were modified in the current generation and evicted from memory
3844 * and then re-read we need to do a full sync since we don't have any
3845 * idea about which extents were modified before we were evicted from
3848 * This is required for both inode re-read from disk and delayed inode
3849 * in delayed_nodes_tree.
3851 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3852 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3853 &BTRFS_I(inode)->runtime_flags);
3856 * We don't persist the id of the transaction where an unlink operation
3857 * against the inode was last made. So here we assume the inode might
3858 * have been evicted, and therefore the exact value of last_unlink_trans
3859 * lost, and set it to last_trans to avoid metadata inconsistencies
3860 * between the inode and its parent if the inode is fsync'ed and the log
3861 * replayed. For example, in the scenario:
3864 * ln mydir/foo mydir/bar
3867 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3868 * xfs_io -c fsync mydir/foo
3870 * mount fs, triggers fsync log replay
3872 * We must make sure that when we fsync our inode foo we also log its
3873 * parent inode, otherwise after log replay the parent still has the
3874 * dentry with the "bar" name but our inode foo has a link count of 1
3875 * and doesn't have an inode ref with the name "bar" anymore.
3877 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3878 * but it guarantees correctness at the expense of occasional full
3879 * transaction commits on fsync if our inode is a directory, or if our
3880 * inode is not a directory, logging its parent unnecessarily.
3882 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3885 * Same logic as for last_unlink_trans. We don't persist the generation
3886 * of the last transaction where this inode was used for a reflink
3887 * operation, so after eviction and reloading the inode we must be
3888 * pessimistic and assume the last transaction that modified the inode.
3890 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3893 if (inode->i_nlink != 1 ||
3894 path->slots[0] >= btrfs_header_nritems(leaf))
3897 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3898 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3901 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3902 if (location.type == BTRFS_INODE_REF_KEY) {
3903 struct btrfs_inode_ref *ref;
3905 ref = (struct btrfs_inode_ref *)ptr;
3906 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3907 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3908 struct btrfs_inode_extref *extref;
3910 extref = (struct btrfs_inode_extref *)ptr;
3911 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3916 * try to precache a NULL acl entry for files that don't have
3917 * any xattrs or acls
3919 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3920 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3921 if (first_xattr_slot != -1) {
3922 path->slots[0] = first_xattr_slot;
3923 ret = btrfs_load_inode_props(inode, path);
3926 "error loading props for ino %llu (root %llu): %d",
3927 btrfs_ino(BTRFS_I(inode)),
3928 root->root_key.objectid, ret);
3930 if (path != in_path)
3931 btrfs_free_path(path);
3934 cache_no_acl(inode);
3936 switch (inode->i_mode & S_IFMT) {
3938 inode->i_mapping->a_ops = &btrfs_aops;
3939 inode->i_fop = &btrfs_file_operations;
3940 inode->i_op = &btrfs_file_inode_operations;
3943 inode->i_fop = &btrfs_dir_file_operations;
3944 inode->i_op = &btrfs_dir_inode_operations;
3947 inode->i_op = &btrfs_symlink_inode_operations;
3948 inode_nohighmem(inode);
3949 inode->i_mapping->a_ops = &btrfs_aops;
3952 inode->i_op = &btrfs_special_inode_operations;
3953 init_special_inode(inode, inode->i_mode, rdev);
3957 btrfs_sync_inode_flags_to_i_flags(inode);
3962 * given a leaf and an inode, copy the inode fields into the leaf
3964 static void fill_inode_item(struct btrfs_trans_handle *trans,
3965 struct extent_buffer *leaf,
3966 struct btrfs_inode_item *item,
3967 struct inode *inode)
3969 struct btrfs_map_token token;
3972 btrfs_init_map_token(&token, leaf);
3974 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3975 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3976 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
3977 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3978 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3980 btrfs_set_token_timespec_sec(&token, &item->atime,
3981 inode->i_atime.tv_sec);
3982 btrfs_set_token_timespec_nsec(&token, &item->atime,
3983 inode->i_atime.tv_nsec);
3985 btrfs_set_token_timespec_sec(&token, &item->mtime,
3986 inode->i_mtime.tv_sec);
3987 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3988 inode->i_mtime.tv_nsec);
3990 btrfs_set_token_timespec_sec(&token, &item->ctime,
3991 inode->i_ctime.tv_sec);
3992 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3993 inode->i_ctime.tv_nsec);
3995 btrfs_set_token_timespec_sec(&token, &item->otime,
3996 BTRFS_I(inode)->i_otime.tv_sec);
3997 btrfs_set_token_timespec_nsec(&token, &item->otime,
3998 BTRFS_I(inode)->i_otime.tv_nsec);
4000 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
4001 btrfs_set_token_inode_generation(&token, item,
4002 BTRFS_I(inode)->generation);
4003 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
4004 btrfs_set_token_inode_transid(&token, item, trans->transid);
4005 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
4006 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
4007 BTRFS_I(inode)->ro_flags);
4008 btrfs_set_token_inode_flags(&token, item, flags);
4009 btrfs_set_token_inode_block_group(&token, item, 0);
4013 * copy everything in the in-memory inode into the btree.
4015 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
4016 struct btrfs_root *root,
4017 struct btrfs_inode *inode)
4019 struct btrfs_inode_item *inode_item;
4020 struct btrfs_path *path;
4021 struct extent_buffer *leaf;
4024 path = btrfs_alloc_path();
4028 ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1);
4035 leaf = path->nodes[0];
4036 inode_item = btrfs_item_ptr(leaf, path->slots[0],
4037 struct btrfs_inode_item);
4039 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
4040 btrfs_mark_buffer_dirty(leaf);
4041 btrfs_set_inode_last_trans(trans, inode);
4044 btrfs_free_path(path);
4049 * copy everything in the in-memory inode into the btree.
4051 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
4052 struct btrfs_root *root,
4053 struct btrfs_inode *inode)
4055 struct btrfs_fs_info *fs_info = root->fs_info;
4059 * If the inode is a free space inode, we can deadlock during commit
4060 * if we put it into the delayed code.
4062 * The data relocation inode should also be directly updated
4065 if (!btrfs_is_free_space_inode(inode)
4066 && !btrfs_is_data_reloc_root(root)
4067 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4068 btrfs_update_root_times(trans, root);
4070 ret = btrfs_delayed_update_inode(trans, root, inode);
4072 btrfs_set_inode_last_trans(trans, inode);
4076 return btrfs_update_inode_item(trans, root, inode);
4079 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4080 struct btrfs_root *root, struct btrfs_inode *inode)
4084 ret = btrfs_update_inode(trans, root, inode);
4086 return btrfs_update_inode_item(trans, root, inode);
4091 * unlink helper that gets used here in inode.c and in the tree logging
4092 * recovery code. It remove a link in a directory with a given name, and
4093 * also drops the back refs in the inode to the directory
4095 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4096 struct btrfs_inode *dir,
4097 struct btrfs_inode *inode,
4098 const struct fscrypt_str *name,
4099 struct btrfs_rename_ctx *rename_ctx)
4101 struct btrfs_root *root = dir->root;
4102 struct btrfs_fs_info *fs_info = root->fs_info;
4103 struct btrfs_path *path;
4105 struct btrfs_dir_item *di;
4107 u64 ino = btrfs_ino(inode);
4108 u64 dir_ino = btrfs_ino(dir);
4110 path = btrfs_alloc_path();
4116 di = btrfs_lookup_dir_item(trans, root, path, dir_ino, name, -1);
4117 if (IS_ERR_OR_NULL(di)) {
4118 ret = di ? PTR_ERR(di) : -ENOENT;
4121 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4124 btrfs_release_path(path);
4127 * If we don't have dir index, we have to get it by looking up
4128 * the inode ref, since we get the inode ref, remove it directly,
4129 * it is unnecessary to do delayed deletion.
4131 * But if we have dir index, needn't search inode ref to get it.
4132 * Since the inode ref is close to the inode item, it is better
4133 * that we delay to delete it, and just do this deletion when
4134 * we update the inode item.
4136 if (inode->dir_index) {
4137 ret = btrfs_delayed_delete_inode_ref(inode);
4139 index = inode->dir_index;
4144 ret = btrfs_del_inode_ref(trans, root, name, ino, dir_ino, &index);
4147 "failed to delete reference to %.*s, inode %llu parent %llu",
4148 name->len, name->name, ino, dir_ino);
4149 btrfs_abort_transaction(trans, ret);
4154 rename_ctx->index = index;
4156 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4158 btrfs_abort_transaction(trans, ret);
4163 * If we are in a rename context, we don't need to update anything in the
4164 * log. That will be done later during the rename by btrfs_log_new_name().
4165 * Besides that, doing it here would only cause extra unnecessary btree
4166 * operations on the log tree, increasing latency for applications.
4169 btrfs_del_inode_ref_in_log(trans, root, name, inode, dir_ino);
4170 btrfs_del_dir_entries_in_log(trans, root, name, dir, index);
4174 * If we have a pending delayed iput we could end up with the final iput
4175 * being run in btrfs-cleaner context. If we have enough of these built
4176 * up we can end up burning a lot of time in btrfs-cleaner without any
4177 * way to throttle the unlinks. Since we're currently holding a ref on
4178 * the inode we can run the delayed iput here without any issues as the
4179 * final iput won't be done until after we drop the ref we're currently
4182 btrfs_run_delayed_iput(fs_info, inode);
4184 btrfs_free_path(path);
4188 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name->len * 2);
4189 inode_inc_iversion(&inode->vfs_inode);
4190 inode_inc_iversion(&dir->vfs_inode);
4191 inode->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4192 dir->vfs_inode.i_mtime = inode->vfs_inode.i_ctime;
4193 dir->vfs_inode.i_ctime = inode->vfs_inode.i_ctime;
4194 ret = btrfs_update_inode(trans, root, dir);
4199 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4200 struct btrfs_inode *dir, struct btrfs_inode *inode,
4201 const struct fscrypt_str *name)
4205 ret = __btrfs_unlink_inode(trans, dir, inode, name, NULL);
4207 drop_nlink(&inode->vfs_inode);
4208 ret = btrfs_update_inode(trans, inode->root, inode);
4214 * helper to start transaction for unlink and rmdir.
4216 * unlink and rmdir are special in btrfs, they do not always free space, so
4217 * if we cannot make our reservations the normal way try and see if there is
4218 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4219 * allow the unlink to occur.
4221 static struct btrfs_trans_handle *__unlink_start_trans(struct btrfs_inode *dir)
4223 struct btrfs_root *root = dir->root;
4225 return btrfs_start_transaction_fallback_global_rsv(root,
4226 BTRFS_UNLINK_METADATA_UNITS);
4229 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4231 struct btrfs_trans_handle *trans;
4232 struct inode *inode = d_inode(dentry);
4234 struct fscrypt_name fname;
4236 ret = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4240 /* This needs to handle no-key deletions later on */
4242 trans = __unlink_start_trans(BTRFS_I(dir));
4243 if (IS_ERR(trans)) {
4244 ret = PTR_ERR(trans);
4248 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4251 ret = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4256 if (inode->i_nlink == 0) {
4257 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4263 btrfs_end_transaction(trans);
4264 btrfs_btree_balance_dirty(BTRFS_I(dir)->root->fs_info);
4266 fscrypt_free_filename(&fname);
4270 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4271 struct btrfs_inode *dir, struct dentry *dentry)
4273 struct btrfs_root *root = dir->root;
4274 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4275 struct btrfs_path *path;
4276 struct extent_buffer *leaf;
4277 struct btrfs_dir_item *di;
4278 struct btrfs_key key;
4282 u64 dir_ino = btrfs_ino(dir);
4283 struct fscrypt_name fname;
4285 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
4289 /* This needs to handle no-key deletions later on */
4291 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4292 objectid = inode->root->root_key.objectid;
4293 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4294 objectid = inode->location.objectid;
4297 fscrypt_free_filename(&fname);
4301 path = btrfs_alloc_path();
4307 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4308 &fname.disk_name, -1);
4309 if (IS_ERR_OR_NULL(di)) {
4310 ret = di ? PTR_ERR(di) : -ENOENT;
4314 leaf = path->nodes[0];
4315 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4316 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4317 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4319 btrfs_abort_transaction(trans, ret);
4322 btrfs_release_path(path);
4325 * This is a placeholder inode for a subvolume we didn't have a
4326 * reference to at the time of the snapshot creation. In the meantime
4327 * we could have renamed the real subvol link into our snapshot, so
4328 * depending on btrfs_del_root_ref to return -ENOENT here is incorrect.
4329 * Instead simply lookup the dir_index_item for this entry so we can
4330 * remove it. Otherwise we know we have a ref to the root and we can
4331 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4333 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4334 di = btrfs_search_dir_index_item(root, path, dir_ino, &fname.disk_name);
4335 if (IS_ERR_OR_NULL(di)) {
4340 btrfs_abort_transaction(trans, ret);
4344 leaf = path->nodes[0];
4345 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4347 btrfs_release_path(path);
4349 ret = btrfs_del_root_ref(trans, objectid,
4350 root->root_key.objectid, dir_ino,
4351 &index, &fname.disk_name);
4353 btrfs_abort_transaction(trans, ret);
4358 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4360 btrfs_abort_transaction(trans, ret);
4364 btrfs_i_size_write(dir, dir->vfs_inode.i_size - fname.disk_name.len * 2);
4365 inode_inc_iversion(&dir->vfs_inode);
4366 dir->vfs_inode.i_mtime = current_time(&dir->vfs_inode);
4367 dir->vfs_inode.i_ctime = dir->vfs_inode.i_mtime;
4368 ret = btrfs_update_inode_fallback(trans, root, dir);
4370 btrfs_abort_transaction(trans, ret);
4372 btrfs_free_path(path);
4373 fscrypt_free_filename(&fname);
4378 * Helper to check if the subvolume references other subvolumes or if it's
4381 static noinline int may_destroy_subvol(struct btrfs_root *root)
4383 struct btrfs_fs_info *fs_info = root->fs_info;
4384 struct btrfs_path *path;
4385 struct btrfs_dir_item *di;
4386 struct btrfs_key key;
4387 struct fscrypt_str name = FSTR_INIT("default", 7);
4391 path = btrfs_alloc_path();
4395 /* Make sure this root isn't set as the default subvol */
4396 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4397 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4399 if (di && !IS_ERR(di)) {
4400 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4401 if (key.objectid == root->root_key.objectid) {
4404 "deleting default subvolume %llu is not allowed",
4408 btrfs_release_path(path);
4411 key.objectid = root->root_key.objectid;
4412 key.type = BTRFS_ROOT_REF_KEY;
4413 key.offset = (u64)-1;
4415 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4421 if (path->slots[0] > 0) {
4423 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4424 if (key.objectid == root->root_key.objectid &&
4425 key.type == BTRFS_ROOT_REF_KEY)
4429 btrfs_free_path(path);
4433 /* Delete all dentries for inodes belonging to the root */
4434 static void btrfs_prune_dentries(struct btrfs_root *root)
4436 struct btrfs_fs_info *fs_info = root->fs_info;
4437 struct rb_node *node;
4438 struct rb_node *prev;
4439 struct btrfs_inode *entry;
4440 struct inode *inode;
4443 if (!BTRFS_FS_ERROR(fs_info))
4444 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4446 spin_lock(&root->inode_lock);
4448 node = root->inode_tree.rb_node;
4452 entry = rb_entry(node, struct btrfs_inode, rb_node);
4454 if (objectid < btrfs_ino(entry))
4455 node = node->rb_left;
4456 else if (objectid > btrfs_ino(entry))
4457 node = node->rb_right;
4463 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4464 if (objectid <= btrfs_ino(entry)) {
4468 prev = rb_next(prev);
4472 entry = rb_entry(node, struct btrfs_inode, rb_node);
4473 objectid = btrfs_ino(entry) + 1;
4474 inode = igrab(&entry->vfs_inode);
4476 spin_unlock(&root->inode_lock);
4477 if (atomic_read(&inode->i_count) > 1)
4478 d_prune_aliases(inode);
4480 * btrfs_drop_inode will have it removed from the inode
4481 * cache when its usage count hits zero.
4485 spin_lock(&root->inode_lock);
4489 if (cond_resched_lock(&root->inode_lock))
4492 node = rb_next(node);
4494 spin_unlock(&root->inode_lock);
4497 int btrfs_delete_subvolume(struct btrfs_inode *dir, struct dentry *dentry)
4499 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4500 struct btrfs_root *root = dir->root;
4501 struct inode *inode = d_inode(dentry);
4502 struct btrfs_root *dest = BTRFS_I(inode)->root;
4503 struct btrfs_trans_handle *trans;
4504 struct btrfs_block_rsv block_rsv;
4509 * Don't allow to delete a subvolume with send in progress. This is
4510 * inside the inode lock so the error handling that has to drop the bit
4511 * again is not run concurrently.
4513 spin_lock(&dest->root_item_lock);
4514 if (dest->send_in_progress) {
4515 spin_unlock(&dest->root_item_lock);
4517 "attempt to delete subvolume %llu during send",
4518 dest->root_key.objectid);
4521 if (atomic_read(&dest->nr_swapfiles)) {
4522 spin_unlock(&dest->root_item_lock);
4524 "attempt to delete subvolume %llu with active swapfile",
4525 root->root_key.objectid);
4528 root_flags = btrfs_root_flags(&dest->root_item);
4529 btrfs_set_root_flags(&dest->root_item,
4530 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4531 spin_unlock(&dest->root_item_lock);
4533 down_write(&fs_info->subvol_sem);
4535 ret = may_destroy_subvol(dest);
4539 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4541 * One for dir inode,
4542 * two for dir entries,
4543 * two for root ref/backref.
4545 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4549 trans = btrfs_start_transaction(root, 0);
4550 if (IS_ERR(trans)) {
4551 ret = PTR_ERR(trans);
4554 trans->block_rsv = &block_rsv;
4555 trans->bytes_reserved = block_rsv.size;
4557 btrfs_record_snapshot_destroy(trans, dir);
4559 ret = btrfs_unlink_subvol(trans, dir, dentry);
4561 btrfs_abort_transaction(trans, ret);
4565 ret = btrfs_record_root_in_trans(trans, dest);
4567 btrfs_abort_transaction(trans, ret);
4571 memset(&dest->root_item.drop_progress, 0,
4572 sizeof(dest->root_item.drop_progress));
4573 btrfs_set_root_drop_level(&dest->root_item, 0);
4574 btrfs_set_root_refs(&dest->root_item, 0);
4576 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4577 ret = btrfs_insert_orphan_item(trans,
4579 dest->root_key.objectid);
4581 btrfs_abort_transaction(trans, ret);
4586 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4587 BTRFS_UUID_KEY_SUBVOL,
4588 dest->root_key.objectid);
4589 if (ret && ret != -ENOENT) {
4590 btrfs_abort_transaction(trans, ret);
4593 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4594 ret = btrfs_uuid_tree_remove(trans,
4595 dest->root_item.received_uuid,
4596 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4597 dest->root_key.objectid);
4598 if (ret && ret != -ENOENT) {
4599 btrfs_abort_transaction(trans, ret);
4604 free_anon_bdev(dest->anon_dev);
4607 trans->block_rsv = NULL;
4608 trans->bytes_reserved = 0;
4609 ret = btrfs_end_transaction(trans);
4610 inode->i_flags |= S_DEAD;
4612 btrfs_subvolume_release_metadata(root, &block_rsv);
4614 up_write(&fs_info->subvol_sem);
4616 spin_lock(&dest->root_item_lock);
4617 root_flags = btrfs_root_flags(&dest->root_item);
4618 btrfs_set_root_flags(&dest->root_item,
4619 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4620 spin_unlock(&dest->root_item_lock);
4622 d_invalidate(dentry);
4623 btrfs_prune_dentries(dest);
4624 ASSERT(dest->send_in_progress == 0);
4630 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4632 struct inode *inode = d_inode(dentry);
4633 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
4635 struct btrfs_trans_handle *trans;
4636 u64 last_unlink_trans;
4637 struct fscrypt_name fname;
4639 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4641 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) {
4642 if (unlikely(btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))) {
4644 "extent tree v2 doesn't support snapshot deletion yet");
4647 return btrfs_delete_subvolume(BTRFS_I(dir), dentry);
4650 err = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4654 /* This needs to handle no-key deletions later on */
4656 trans = __unlink_start_trans(BTRFS_I(dir));
4657 if (IS_ERR(trans)) {
4658 err = PTR_ERR(trans);
4662 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4663 err = btrfs_unlink_subvol(trans, BTRFS_I(dir), dentry);
4667 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4671 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4673 /* now the directory is empty */
4674 err = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4677 btrfs_i_size_write(BTRFS_I(inode), 0);
4679 * Propagate the last_unlink_trans value of the deleted dir to
4680 * its parent directory. This is to prevent an unrecoverable
4681 * log tree in the case we do something like this:
4683 * 2) create snapshot under dir foo
4684 * 3) delete the snapshot
4687 * 6) fsync foo or some file inside foo
4689 if (last_unlink_trans >= trans->transid)
4690 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4693 btrfs_end_transaction(trans);
4695 btrfs_btree_balance_dirty(fs_info);
4696 fscrypt_free_filename(&fname);
4702 * btrfs_truncate_block - read, zero a chunk and write a block
4703 * @inode - inode that we're zeroing
4704 * @from - the offset to start zeroing
4705 * @len - the length to zero, 0 to zero the entire range respective to the
4707 * @front - zero up to the offset instead of from the offset on
4709 * This will find the block for the "from" offset and cow the block and zero the
4710 * part we want to zero. This is used with truncate and hole punching.
4712 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4715 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4716 struct address_space *mapping = inode->vfs_inode.i_mapping;
4717 struct extent_io_tree *io_tree = &inode->io_tree;
4718 struct btrfs_ordered_extent *ordered;
4719 struct extent_state *cached_state = NULL;
4720 struct extent_changeset *data_reserved = NULL;
4721 bool only_release_metadata = false;
4722 u32 blocksize = fs_info->sectorsize;
4723 pgoff_t index = from >> PAGE_SHIFT;
4724 unsigned offset = from & (blocksize - 1);
4726 gfp_t mask = btrfs_alloc_write_mask(mapping);
4727 size_t write_bytes = blocksize;
4732 if (IS_ALIGNED(offset, blocksize) &&
4733 (!len || IS_ALIGNED(len, blocksize)))
4736 block_start = round_down(from, blocksize);
4737 block_end = block_start + blocksize - 1;
4739 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4742 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes, false) > 0) {
4743 /* For nocow case, no need to reserve data space */
4744 only_release_metadata = true;
4749 ret = btrfs_delalloc_reserve_metadata(inode, blocksize, blocksize, false);
4751 if (!only_release_metadata)
4752 btrfs_free_reserved_data_space(inode, data_reserved,
4753 block_start, blocksize);
4757 page = find_or_create_page(mapping, index, mask);
4759 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4761 btrfs_delalloc_release_extents(inode, blocksize);
4766 if (!PageUptodate(page)) {
4767 ret = btrfs_read_folio(NULL, page_folio(page));
4769 if (page->mapping != mapping) {
4774 if (!PageUptodate(page)) {
4781 * We unlock the page after the io is completed and then re-lock it
4782 * above. release_folio() could have come in between that and cleared
4783 * PagePrivate(), but left the page in the mapping. Set the page mapped
4784 * here to make sure it's properly set for the subpage stuff.
4786 ret = set_page_extent_mapped(page);
4790 wait_on_page_writeback(page);
4792 lock_extent(io_tree, block_start, block_end, &cached_state);
4794 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4796 unlock_extent(io_tree, block_start, block_end, &cached_state);
4799 btrfs_start_ordered_extent(ordered);
4800 btrfs_put_ordered_extent(ordered);
4804 clear_extent_bit(&inode->io_tree, block_start, block_end,
4805 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4808 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4811 unlock_extent(io_tree, block_start, block_end, &cached_state);
4815 if (offset != blocksize) {
4817 len = blocksize - offset;
4819 memzero_page(page, (block_start - page_offset(page)),
4822 memzero_page(page, (block_start - page_offset(page)) + offset,
4825 btrfs_page_clear_checked(fs_info, page, block_start,
4826 block_end + 1 - block_start);
4827 btrfs_page_set_dirty(fs_info, page, block_start, block_end + 1 - block_start);
4828 unlock_extent(io_tree, block_start, block_end, &cached_state);
4830 if (only_release_metadata)
4831 set_extent_bit(&inode->io_tree, block_start, block_end,
4832 EXTENT_NORESERVE, NULL);
4836 if (only_release_metadata)
4837 btrfs_delalloc_release_metadata(inode, blocksize, true);
4839 btrfs_delalloc_release_space(inode, data_reserved,
4840 block_start, blocksize, true);
4842 btrfs_delalloc_release_extents(inode, blocksize);
4846 if (only_release_metadata)
4847 btrfs_check_nocow_unlock(inode);
4848 extent_changeset_free(data_reserved);
4852 static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode,
4853 u64 offset, u64 len)
4855 struct btrfs_fs_info *fs_info = root->fs_info;
4856 struct btrfs_trans_handle *trans;
4857 struct btrfs_drop_extents_args drop_args = { 0 };
4861 * If NO_HOLES is enabled, we don't need to do anything.
4862 * Later, up in the call chain, either btrfs_set_inode_last_sub_trans()
4863 * or btrfs_update_inode() will be called, which guarantee that the next
4864 * fsync will know this inode was changed and needs to be logged.
4866 if (btrfs_fs_incompat(fs_info, NO_HOLES))
4870 * 1 - for the one we're dropping
4871 * 1 - for the one we're adding
4872 * 1 - for updating the inode.
4874 trans = btrfs_start_transaction(root, 3);
4876 return PTR_ERR(trans);
4878 drop_args.start = offset;
4879 drop_args.end = offset + len;
4880 drop_args.drop_cache = true;
4882 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
4884 btrfs_abort_transaction(trans, ret);
4885 btrfs_end_transaction(trans);
4889 ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset, len);
4891 btrfs_abort_transaction(trans, ret);
4893 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
4894 btrfs_update_inode(trans, root, inode);
4896 btrfs_end_transaction(trans);
4901 * This function puts in dummy file extents for the area we're creating a hole
4902 * for. So if we are truncating this file to a larger size we need to insert
4903 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4904 * the range between oldsize and size
4906 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
4908 struct btrfs_root *root = inode->root;
4909 struct btrfs_fs_info *fs_info = root->fs_info;
4910 struct extent_io_tree *io_tree = &inode->io_tree;
4911 struct extent_map *em = NULL;
4912 struct extent_state *cached_state = NULL;
4913 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4914 u64 block_end = ALIGN(size, fs_info->sectorsize);
4921 * If our size started in the middle of a block we need to zero out the
4922 * rest of the block before we expand the i_size, otherwise we could
4923 * expose stale data.
4925 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4929 if (size <= hole_start)
4932 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
4934 cur_offset = hole_start;
4936 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4937 block_end - cur_offset);
4943 last_byte = min(extent_map_end(em), block_end);
4944 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4945 hole_size = last_byte - cur_offset;
4947 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4948 struct extent_map *hole_em;
4950 err = maybe_insert_hole(root, inode, cur_offset,
4955 err = btrfs_inode_set_file_extent_range(inode,
4956 cur_offset, hole_size);
4960 hole_em = alloc_extent_map();
4962 btrfs_drop_extent_map_range(inode, cur_offset,
4963 cur_offset + hole_size - 1,
4965 btrfs_set_inode_full_sync(inode);
4968 hole_em->start = cur_offset;
4969 hole_em->len = hole_size;
4970 hole_em->orig_start = cur_offset;
4972 hole_em->block_start = EXTENT_MAP_HOLE;
4973 hole_em->block_len = 0;
4974 hole_em->orig_block_len = 0;
4975 hole_em->ram_bytes = hole_size;
4976 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4977 hole_em->generation = fs_info->generation;
4979 err = btrfs_replace_extent_map_range(inode, hole_em, true);
4980 free_extent_map(hole_em);
4982 err = btrfs_inode_set_file_extent_range(inode,
4983 cur_offset, hole_size);
4988 free_extent_map(em);
4990 cur_offset = last_byte;
4991 if (cur_offset >= block_end)
4994 free_extent_map(em);
4995 unlock_extent(io_tree, hole_start, block_end - 1, &cached_state);
4999 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5001 struct btrfs_root *root = BTRFS_I(inode)->root;
5002 struct btrfs_trans_handle *trans;
5003 loff_t oldsize = i_size_read(inode);
5004 loff_t newsize = attr->ia_size;
5005 int mask = attr->ia_valid;
5009 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5010 * special case where we need to update the times despite not having
5011 * these flags set. For all other operations the VFS set these flags
5012 * explicitly if it wants a timestamp update.
5014 if (newsize != oldsize) {
5015 inode_inc_iversion(inode);
5016 if (!(mask & (ATTR_CTIME | ATTR_MTIME))) {
5017 inode->i_mtime = current_time(inode);
5018 inode->i_ctime = inode->i_mtime;
5022 if (newsize > oldsize) {
5024 * Don't do an expanding truncate while snapshotting is ongoing.
5025 * This is to ensure the snapshot captures a fully consistent
5026 * state of this file - if the snapshot captures this expanding
5027 * truncation, it must capture all writes that happened before
5030 btrfs_drew_write_lock(&root->snapshot_lock);
5031 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
5033 btrfs_drew_write_unlock(&root->snapshot_lock);
5037 trans = btrfs_start_transaction(root, 1);
5038 if (IS_ERR(trans)) {
5039 btrfs_drew_write_unlock(&root->snapshot_lock);
5040 return PTR_ERR(trans);
5043 i_size_write(inode, newsize);
5044 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
5045 pagecache_isize_extended(inode, oldsize, newsize);
5046 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5047 btrfs_drew_write_unlock(&root->snapshot_lock);
5048 btrfs_end_transaction(trans);
5050 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5052 if (btrfs_is_zoned(fs_info)) {
5053 ret = btrfs_wait_ordered_range(inode,
5054 ALIGN(newsize, fs_info->sectorsize),
5061 * We're truncating a file that used to have good data down to
5062 * zero. Make sure any new writes to the file get on disk
5066 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5067 &BTRFS_I(inode)->runtime_flags);
5069 truncate_setsize(inode, newsize);
5071 inode_dio_wait(inode);
5073 ret = btrfs_truncate(BTRFS_I(inode), newsize == oldsize);
5074 if (ret && inode->i_nlink) {
5078 * Truncate failed, so fix up the in-memory size. We
5079 * adjusted disk_i_size down as we removed extents, so
5080 * wait for disk_i_size to be stable and then update the
5081 * in-memory size to match.
5083 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5086 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5093 static int btrfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
5096 struct inode *inode = d_inode(dentry);
5097 struct btrfs_root *root = BTRFS_I(inode)->root;
5100 if (btrfs_root_readonly(root))
5103 err = setattr_prepare(idmap, dentry, attr);
5107 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5108 err = btrfs_setsize(inode, attr);
5113 if (attr->ia_valid) {
5114 setattr_copy(idmap, inode, attr);
5115 inode_inc_iversion(inode);
5116 err = btrfs_dirty_inode(BTRFS_I(inode));
5118 if (!err && attr->ia_valid & ATTR_MODE)
5119 err = posix_acl_chmod(idmap, dentry, inode->i_mode);
5126 * While truncating the inode pages during eviction, we get the VFS
5127 * calling btrfs_invalidate_folio() against each folio of the inode. This
5128 * is slow because the calls to btrfs_invalidate_folio() result in a
5129 * huge amount of calls to lock_extent() and clear_extent_bit(),
5130 * which keep merging and splitting extent_state structures over and over,
5131 * wasting lots of time.
5133 * Therefore if the inode is being evicted, let btrfs_invalidate_folio()
5134 * skip all those expensive operations on a per folio basis and do only
5135 * the ordered io finishing, while we release here the extent_map and
5136 * extent_state structures, without the excessive merging and splitting.
5138 static void evict_inode_truncate_pages(struct inode *inode)
5140 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5141 struct rb_node *node;
5143 ASSERT(inode->i_state & I_FREEING);
5144 truncate_inode_pages_final(&inode->i_data);
5146 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
5149 * Keep looping until we have no more ranges in the io tree.
5150 * We can have ongoing bios started by readahead that have
5151 * their endio callback (extent_io.c:end_bio_extent_readpage)
5152 * still in progress (unlocked the pages in the bio but did not yet
5153 * unlocked the ranges in the io tree). Therefore this means some
5154 * ranges can still be locked and eviction started because before
5155 * submitting those bios, which are executed by a separate task (work
5156 * queue kthread), inode references (inode->i_count) were not taken
5157 * (which would be dropped in the end io callback of each bio).
5158 * Therefore here we effectively end up waiting for those bios and
5159 * anyone else holding locked ranges without having bumped the inode's
5160 * reference count - if we don't do it, when they access the inode's
5161 * io_tree to unlock a range it may be too late, leading to an
5162 * use-after-free issue.
5164 spin_lock(&io_tree->lock);
5165 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5166 struct extent_state *state;
5167 struct extent_state *cached_state = NULL;
5170 unsigned state_flags;
5172 node = rb_first(&io_tree->state);
5173 state = rb_entry(node, struct extent_state, rb_node);
5174 start = state->start;
5176 state_flags = state->state;
5177 spin_unlock(&io_tree->lock);
5179 lock_extent(io_tree, start, end, &cached_state);
5182 * If still has DELALLOC flag, the extent didn't reach disk,
5183 * and its reserved space won't be freed by delayed_ref.
5184 * So we need to free its reserved space here.
5185 * (Refer to comment in btrfs_invalidate_folio, case 2)
5187 * Note, end is the bytenr of last byte, so we need + 1 here.
5189 if (state_flags & EXTENT_DELALLOC)
5190 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5193 clear_extent_bit(io_tree, start, end,
5194 EXTENT_CLEAR_ALL_BITS | EXTENT_DO_ACCOUNTING,
5198 spin_lock(&io_tree->lock);
5200 spin_unlock(&io_tree->lock);
5203 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5204 struct btrfs_block_rsv *rsv)
5206 struct btrfs_fs_info *fs_info = root->fs_info;
5207 struct btrfs_trans_handle *trans;
5208 u64 delayed_refs_extra = btrfs_calc_delayed_ref_bytes(fs_info, 1);
5212 * Eviction should be taking place at some place safe because of our
5213 * delayed iputs. However the normal flushing code will run delayed
5214 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5216 * We reserve the delayed_refs_extra here again because we can't use
5217 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5218 * above. We reserve our extra bit here because we generate a ton of
5219 * delayed refs activity by truncating.
5221 * BTRFS_RESERVE_FLUSH_EVICT will steal from the global_rsv if it can,
5222 * if we fail to make this reservation we can re-try without the
5223 * delayed_refs_extra so we can make some forward progress.
5225 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size + delayed_refs_extra,
5226 BTRFS_RESERVE_FLUSH_EVICT);
5228 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size,
5229 BTRFS_RESERVE_FLUSH_EVICT);
5232 "could not allocate space for delete; will truncate on mount");
5233 return ERR_PTR(-ENOSPC);
5235 delayed_refs_extra = 0;
5238 trans = btrfs_join_transaction(root);
5242 if (delayed_refs_extra) {
5243 trans->block_rsv = &fs_info->trans_block_rsv;
5244 trans->bytes_reserved = delayed_refs_extra;
5245 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5246 delayed_refs_extra, true);
5251 void btrfs_evict_inode(struct inode *inode)
5253 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5254 struct btrfs_trans_handle *trans;
5255 struct btrfs_root *root = BTRFS_I(inode)->root;
5256 struct btrfs_block_rsv *rsv = NULL;
5259 trace_btrfs_inode_evict(inode);
5262 fsverity_cleanup_inode(inode);
5267 evict_inode_truncate_pages(inode);
5269 if (inode->i_nlink &&
5270 ((btrfs_root_refs(&root->root_item) != 0 &&
5271 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5272 btrfs_is_free_space_inode(BTRFS_I(inode))))
5275 if (is_bad_inode(inode))
5278 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5281 if (inode->i_nlink > 0) {
5282 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5283 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5288 * This makes sure the inode item in tree is uptodate and the space for
5289 * the inode update is released.
5291 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5296 * This drops any pending insert or delete operations we have for this
5297 * inode. We could have a delayed dir index deletion queued up, but
5298 * we're removing the inode completely so that'll be taken care of in
5301 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
5303 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5306 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5307 rsv->failfast = true;
5309 btrfs_i_size_write(BTRFS_I(inode), 0);
5312 struct btrfs_truncate_control control = {
5313 .inode = BTRFS_I(inode),
5314 .ino = btrfs_ino(BTRFS_I(inode)),
5319 trans = evict_refill_and_join(root, rsv);
5323 trans->block_rsv = rsv;
5325 ret = btrfs_truncate_inode_items(trans, root, &control);
5326 trans->block_rsv = &fs_info->trans_block_rsv;
5327 btrfs_end_transaction(trans);
5329 * We have not added new delayed items for our inode after we
5330 * have flushed its delayed items, so no need to throttle on
5331 * delayed items. However we have modified extent buffers.
5333 btrfs_btree_balance_dirty_nodelay(fs_info);
5334 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5341 * Errors here aren't a big deal, it just means we leave orphan items in
5342 * the tree. They will be cleaned up on the next mount. If the inode
5343 * number gets reused, cleanup deletes the orphan item without doing
5344 * anything, and unlink reuses the existing orphan item.
5346 * If it turns out that we are dropping too many of these, we might want
5347 * to add a mechanism for retrying these after a commit.
5349 trans = evict_refill_and_join(root, rsv);
5350 if (!IS_ERR(trans)) {
5351 trans->block_rsv = rsv;
5352 btrfs_orphan_del(trans, BTRFS_I(inode));
5353 trans->block_rsv = &fs_info->trans_block_rsv;
5354 btrfs_end_transaction(trans);
5358 btrfs_free_block_rsv(fs_info, rsv);
5360 * If we didn't successfully delete, the orphan item will still be in
5361 * the tree and we'll retry on the next mount. Again, we might also want
5362 * to retry these periodically in the future.
5364 btrfs_remove_delayed_node(BTRFS_I(inode));
5365 fsverity_cleanup_inode(inode);
5370 * Return the key found in the dir entry in the location pointer, fill @type
5371 * with BTRFS_FT_*, and return 0.
5373 * If no dir entries were found, returns -ENOENT.
5374 * If found a corrupted location in dir entry, returns -EUCLEAN.
5376 static int btrfs_inode_by_name(struct btrfs_inode *dir, struct dentry *dentry,
5377 struct btrfs_key *location, u8 *type)
5379 struct btrfs_dir_item *di;
5380 struct btrfs_path *path;
5381 struct btrfs_root *root = dir->root;
5383 struct fscrypt_name fname;
5385 path = btrfs_alloc_path();
5389 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
5393 * fscrypt_setup_filename() should never return a positive value, but
5394 * gcc on sparc/parisc thinks it can, so assert that doesn't happen.
5398 /* This needs to handle no-key deletions later on */
5400 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir),
5401 &fname.disk_name, 0);
5402 if (IS_ERR_OR_NULL(di)) {
5403 ret = di ? PTR_ERR(di) : -ENOENT;
5407 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5408 if (location->type != BTRFS_INODE_ITEM_KEY &&
5409 location->type != BTRFS_ROOT_ITEM_KEY) {
5411 btrfs_warn(root->fs_info,
5412 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5413 __func__, fname.disk_name.name, btrfs_ino(dir),
5414 location->objectid, location->type, location->offset);
5417 *type = btrfs_dir_ftype(path->nodes[0], di);
5419 fscrypt_free_filename(&fname);
5420 btrfs_free_path(path);
5425 * when we hit a tree root in a directory, the btrfs part of the inode
5426 * needs to be changed to reflect the root directory of the tree root. This
5427 * is kind of like crossing a mount point.
5429 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5430 struct btrfs_inode *dir,
5431 struct dentry *dentry,
5432 struct btrfs_key *location,
5433 struct btrfs_root **sub_root)
5435 struct btrfs_path *path;
5436 struct btrfs_root *new_root;
5437 struct btrfs_root_ref *ref;
5438 struct extent_buffer *leaf;
5439 struct btrfs_key key;
5442 struct fscrypt_name fname;
5444 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 0, &fname);
5448 path = btrfs_alloc_path();
5455 key.objectid = dir->root->root_key.objectid;
5456 key.type = BTRFS_ROOT_REF_KEY;
5457 key.offset = location->objectid;
5459 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5466 leaf = path->nodes[0];
5467 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5468 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5469 btrfs_root_ref_name_len(leaf, ref) != fname.disk_name.len)
5472 ret = memcmp_extent_buffer(leaf, fname.disk_name.name,
5473 (unsigned long)(ref + 1), fname.disk_name.len);
5477 btrfs_release_path(path);
5479 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5480 if (IS_ERR(new_root)) {
5481 err = PTR_ERR(new_root);
5485 *sub_root = new_root;
5486 location->objectid = btrfs_root_dirid(&new_root->root_item);
5487 location->type = BTRFS_INODE_ITEM_KEY;
5488 location->offset = 0;
5491 btrfs_free_path(path);
5492 fscrypt_free_filename(&fname);
5496 static void inode_tree_add(struct btrfs_inode *inode)
5498 struct btrfs_root *root = inode->root;
5499 struct btrfs_inode *entry;
5501 struct rb_node *parent;
5502 struct rb_node *new = &inode->rb_node;
5503 u64 ino = btrfs_ino(inode);
5505 if (inode_unhashed(&inode->vfs_inode))
5508 spin_lock(&root->inode_lock);
5509 p = &root->inode_tree.rb_node;
5512 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5514 if (ino < btrfs_ino(entry))
5515 p = &parent->rb_left;
5516 else if (ino > btrfs_ino(entry))
5517 p = &parent->rb_right;
5519 WARN_ON(!(entry->vfs_inode.i_state &
5520 (I_WILL_FREE | I_FREEING)));
5521 rb_replace_node(parent, new, &root->inode_tree);
5522 RB_CLEAR_NODE(parent);
5523 spin_unlock(&root->inode_lock);
5527 rb_link_node(new, parent, p);
5528 rb_insert_color(new, &root->inode_tree);
5529 spin_unlock(&root->inode_lock);
5532 static void inode_tree_del(struct btrfs_inode *inode)
5534 struct btrfs_root *root = inode->root;
5537 spin_lock(&root->inode_lock);
5538 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5539 rb_erase(&inode->rb_node, &root->inode_tree);
5540 RB_CLEAR_NODE(&inode->rb_node);
5541 empty = RB_EMPTY_ROOT(&root->inode_tree);
5543 spin_unlock(&root->inode_lock);
5545 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5546 spin_lock(&root->inode_lock);
5547 empty = RB_EMPTY_ROOT(&root->inode_tree);
5548 spin_unlock(&root->inode_lock);
5550 btrfs_add_dead_root(root);
5555 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5557 struct btrfs_iget_args *args = p;
5559 inode->i_ino = args->ino;
5560 BTRFS_I(inode)->location.objectid = args->ino;
5561 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5562 BTRFS_I(inode)->location.offset = 0;
5563 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5564 BUG_ON(args->root && !BTRFS_I(inode)->root);
5566 if (args->root && args->root == args->root->fs_info->tree_root &&
5567 args->ino != BTRFS_BTREE_INODE_OBJECTID)
5568 set_bit(BTRFS_INODE_FREE_SPACE_INODE,
5569 &BTRFS_I(inode)->runtime_flags);
5573 static int btrfs_find_actor(struct inode *inode, void *opaque)
5575 struct btrfs_iget_args *args = opaque;
5577 return args->ino == BTRFS_I(inode)->location.objectid &&
5578 args->root == BTRFS_I(inode)->root;
5581 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5582 struct btrfs_root *root)
5584 struct inode *inode;
5585 struct btrfs_iget_args args;
5586 unsigned long hashval = btrfs_inode_hash(ino, root);
5591 inode = iget5_locked(s, hashval, btrfs_find_actor,
5592 btrfs_init_locked_inode,
5598 * Get an inode object given its inode number and corresponding root.
5599 * Path can be preallocated to prevent recursing back to iget through
5600 * allocator. NULL is also valid but may require an additional allocation
5603 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5604 struct btrfs_root *root, struct btrfs_path *path)
5606 struct inode *inode;
5608 inode = btrfs_iget_locked(s, ino, root);
5610 return ERR_PTR(-ENOMEM);
5612 if (inode->i_state & I_NEW) {
5615 ret = btrfs_read_locked_inode(inode, path);
5617 inode_tree_add(BTRFS_I(inode));
5618 unlock_new_inode(inode);
5622 * ret > 0 can come from btrfs_search_slot called by
5623 * btrfs_read_locked_inode, this means the inode item
5628 inode = ERR_PTR(ret);
5635 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5637 return btrfs_iget_path(s, ino, root, NULL);
5640 static struct inode *new_simple_dir(struct super_block *s,
5641 struct btrfs_key *key,
5642 struct btrfs_root *root)
5644 struct inode *inode = new_inode(s);
5647 return ERR_PTR(-ENOMEM);
5649 BTRFS_I(inode)->root = btrfs_grab_root(root);
5650 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5651 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5653 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5655 * We only need lookup, the rest is read-only and there's no inode
5656 * associated with the dentry
5658 inode->i_op = &simple_dir_inode_operations;
5659 inode->i_opflags &= ~IOP_XATTR;
5660 inode->i_fop = &simple_dir_operations;
5661 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5662 inode->i_mtime = current_time(inode);
5663 inode->i_atime = inode->i_mtime;
5664 inode->i_ctime = inode->i_mtime;
5665 BTRFS_I(inode)->i_otime = inode->i_mtime;
5670 static_assert(BTRFS_FT_UNKNOWN == FT_UNKNOWN);
5671 static_assert(BTRFS_FT_REG_FILE == FT_REG_FILE);
5672 static_assert(BTRFS_FT_DIR == FT_DIR);
5673 static_assert(BTRFS_FT_CHRDEV == FT_CHRDEV);
5674 static_assert(BTRFS_FT_BLKDEV == FT_BLKDEV);
5675 static_assert(BTRFS_FT_FIFO == FT_FIFO);
5676 static_assert(BTRFS_FT_SOCK == FT_SOCK);
5677 static_assert(BTRFS_FT_SYMLINK == FT_SYMLINK);
5679 static inline u8 btrfs_inode_type(struct inode *inode)
5681 return fs_umode_to_ftype(inode->i_mode);
5684 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5686 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5687 struct inode *inode;
5688 struct btrfs_root *root = BTRFS_I(dir)->root;
5689 struct btrfs_root *sub_root = root;
5690 struct btrfs_key location;
5694 if (dentry->d_name.len > BTRFS_NAME_LEN)
5695 return ERR_PTR(-ENAMETOOLONG);
5697 ret = btrfs_inode_by_name(BTRFS_I(dir), dentry, &location, &di_type);
5699 return ERR_PTR(ret);
5701 if (location.type == BTRFS_INODE_ITEM_KEY) {
5702 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5706 /* Do extra check against inode mode with di_type */
5707 if (btrfs_inode_type(inode) != di_type) {
5709 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5710 inode->i_mode, btrfs_inode_type(inode),
5713 return ERR_PTR(-EUCLEAN);
5718 ret = fixup_tree_root_location(fs_info, BTRFS_I(dir), dentry,
5719 &location, &sub_root);
5722 inode = ERR_PTR(ret);
5724 inode = new_simple_dir(dir->i_sb, &location, root);
5726 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5727 btrfs_put_root(sub_root);
5732 down_read(&fs_info->cleanup_work_sem);
5733 if (!sb_rdonly(inode->i_sb))
5734 ret = btrfs_orphan_cleanup(sub_root);
5735 up_read(&fs_info->cleanup_work_sem);
5738 inode = ERR_PTR(ret);
5745 static int btrfs_dentry_delete(const struct dentry *dentry)
5747 struct btrfs_root *root;
5748 struct inode *inode = d_inode(dentry);
5750 if (!inode && !IS_ROOT(dentry))
5751 inode = d_inode(dentry->d_parent);
5754 root = BTRFS_I(inode)->root;
5755 if (btrfs_root_refs(&root->root_item) == 0)
5758 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5764 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5767 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5769 if (inode == ERR_PTR(-ENOENT))
5771 return d_splice_alias(inode, dentry);
5775 * Find the highest existing sequence number in a directory and then set the
5776 * in-memory index_cnt variable to the first free sequence number.
5778 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
5780 struct btrfs_root *root = inode->root;
5781 struct btrfs_key key, found_key;
5782 struct btrfs_path *path;
5783 struct extent_buffer *leaf;
5786 key.objectid = btrfs_ino(inode);
5787 key.type = BTRFS_DIR_INDEX_KEY;
5788 key.offset = (u64)-1;
5790 path = btrfs_alloc_path();
5794 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5797 /* FIXME: we should be able to handle this */
5802 if (path->slots[0] == 0) {
5803 inode->index_cnt = BTRFS_DIR_START_INDEX;
5809 leaf = path->nodes[0];
5810 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5812 if (found_key.objectid != btrfs_ino(inode) ||
5813 found_key.type != BTRFS_DIR_INDEX_KEY) {
5814 inode->index_cnt = BTRFS_DIR_START_INDEX;
5818 inode->index_cnt = found_key.offset + 1;
5820 btrfs_free_path(path);
5824 static int btrfs_get_dir_last_index(struct btrfs_inode *dir, u64 *index)
5826 if (dir->index_cnt == (u64)-1) {
5829 ret = btrfs_inode_delayed_dir_index_count(dir);
5831 ret = btrfs_set_inode_index_count(dir);
5837 *index = dir->index_cnt;
5843 * All this infrastructure exists because dir_emit can fault, and we are holding
5844 * the tree lock when doing readdir. For now just allocate a buffer and copy
5845 * our information into that, and then dir_emit from the buffer. This is
5846 * similar to what NFS does, only we don't keep the buffer around in pagecache
5847 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5848 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5851 static int btrfs_opendir(struct inode *inode, struct file *file)
5853 struct btrfs_file_private *private;
5857 ret = btrfs_get_dir_last_index(BTRFS_I(inode), &last_index);
5861 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5864 private->last_index = last_index;
5865 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5866 if (!private->filldir_buf) {
5870 file->private_data = private;
5881 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5884 struct dir_entry *entry = addr;
5885 char *name = (char *)(entry + 1);
5887 ctx->pos = get_unaligned(&entry->offset);
5888 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5889 get_unaligned(&entry->ino),
5890 get_unaligned(&entry->type)))
5892 addr += sizeof(struct dir_entry) +
5893 get_unaligned(&entry->name_len);
5899 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5901 struct inode *inode = file_inode(file);
5902 struct btrfs_root *root = BTRFS_I(inode)->root;
5903 struct btrfs_file_private *private = file->private_data;
5904 struct btrfs_dir_item *di;
5905 struct btrfs_key key;
5906 struct btrfs_key found_key;
5907 struct btrfs_path *path;
5909 struct list_head ins_list;
5910 struct list_head del_list;
5917 struct btrfs_key location;
5919 if (!dir_emit_dots(file, ctx))
5922 path = btrfs_alloc_path();
5926 addr = private->filldir_buf;
5927 path->reada = READA_FORWARD;
5929 INIT_LIST_HEAD(&ins_list);
5930 INIT_LIST_HEAD(&del_list);
5931 put = btrfs_readdir_get_delayed_items(inode, private->last_index,
5932 &ins_list, &del_list);
5935 key.type = BTRFS_DIR_INDEX_KEY;
5936 key.offset = ctx->pos;
5937 key.objectid = btrfs_ino(BTRFS_I(inode));
5939 btrfs_for_each_slot(root, &key, &found_key, path, ret) {
5940 struct dir_entry *entry;
5941 struct extent_buffer *leaf = path->nodes[0];
5944 if (found_key.objectid != key.objectid)
5946 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5948 if (found_key.offset < ctx->pos)
5950 if (found_key.offset > private->last_index)
5952 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5954 di = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
5955 name_len = btrfs_dir_name_len(leaf, di);
5956 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5958 btrfs_release_path(path);
5959 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5962 addr = private->filldir_buf;
5968 ftype = btrfs_dir_flags_to_ftype(btrfs_dir_flags(leaf, di));
5970 name_ptr = (char *)(entry + 1);
5971 read_extent_buffer(leaf, name_ptr,
5972 (unsigned long)(di + 1), name_len);
5973 put_unaligned(name_len, &entry->name_len);
5974 put_unaligned(fs_ftype_to_dtype(ftype), &entry->type);
5975 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5976 put_unaligned(location.objectid, &entry->ino);
5977 put_unaligned(found_key.offset, &entry->offset);
5979 addr += sizeof(struct dir_entry) + name_len;
5980 total_len += sizeof(struct dir_entry) + name_len;
5982 /* Catch error encountered during iteration */
5986 btrfs_release_path(path);
5988 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5992 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5997 * Stop new entries from being returned after we return the last
6000 * New directory entries are assigned a strictly increasing
6001 * offset. This means that new entries created during readdir
6002 * are *guaranteed* to be seen in the future by that readdir.
6003 * This has broken buggy programs which operate on names as
6004 * they're returned by readdir. Until we re-use freed offsets
6005 * we have this hack to stop new entries from being returned
6006 * under the assumption that they'll never reach this huge
6009 * This is being careful not to overflow 32bit loff_t unless the
6010 * last entry requires it because doing so has broken 32bit apps
6013 if (ctx->pos >= INT_MAX)
6014 ctx->pos = LLONG_MAX;
6021 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6022 btrfs_free_path(path);
6027 * This is somewhat expensive, updating the tree every time the
6028 * inode changes. But, it is most likely to find the inode in cache.
6029 * FIXME, needs more benchmarking...there are no reasons other than performance
6030 * to keep or drop this code.
6032 static int btrfs_dirty_inode(struct btrfs_inode *inode)
6034 struct btrfs_root *root = inode->root;
6035 struct btrfs_fs_info *fs_info = root->fs_info;
6036 struct btrfs_trans_handle *trans;
6039 if (test_bit(BTRFS_INODE_DUMMY, &inode->runtime_flags))
6042 trans = btrfs_join_transaction(root);
6044 return PTR_ERR(trans);
6046 ret = btrfs_update_inode(trans, root, inode);
6047 if (ret && (ret == -ENOSPC || ret == -EDQUOT)) {
6048 /* whoops, lets try again with the full transaction */
6049 btrfs_end_transaction(trans);
6050 trans = btrfs_start_transaction(root, 1);
6052 return PTR_ERR(trans);
6054 ret = btrfs_update_inode(trans, root, inode);
6056 btrfs_end_transaction(trans);
6057 if (inode->delayed_node)
6058 btrfs_balance_delayed_items(fs_info);
6064 * This is a copy of file_update_time. We need this so we can return error on
6065 * ENOSPC for updating the inode in the case of file write and mmap writes.
6067 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
6070 struct btrfs_root *root = BTRFS_I(inode)->root;
6071 bool dirty = flags & ~S_VERSION;
6073 if (btrfs_root_readonly(root))
6076 if (flags & S_VERSION)
6077 dirty |= inode_maybe_inc_iversion(inode, dirty);
6078 if (flags & S_CTIME)
6079 inode->i_ctime = *now;
6080 if (flags & S_MTIME)
6081 inode->i_mtime = *now;
6082 if (flags & S_ATIME)
6083 inode->i_atime = *now;
6084 return dirty ? btrfs_dirty_inode(BTRFS_I(inode)) : 0;
6088 * helper to find a free sequence number in a given directory. This current
6089 * code is very simple, later versions will do smarter things in the btree
6091 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6095 if (dir->index_cnt == (u64)-1) {
6096 ret = btrfs_inode_delayed_dir_index_count(dir);
6098 ret = btrfs_set_inode_index_count(dir);
6104 *index = dir->index_cnt;
6110 static int btrfs_insert_inode_locked(struct inode *inode)
6112 struct btrfs_iget_args args;
6114 args.ino = BTRFS_I(inode)->location.objectid;
6115 args.root = BTRFS_I(inode)->root;
6117 return insert_inode_locked4(inode,
6118 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6119 btrfs_find_actor, &args);
6122 int btrfs_new_inode_prepare(struct btrfs_new_inode_args *args,
6123 unsigned int *trans_num_items)
6125 struct inode *dir = args->dir;
6126 struct inode *inode = args->inode;
6129 if (!args->orphan) {
6130 ret = fscrypt_setup_filename(dir, &args->dentry->d_name, 0,
6136 ret = posix_acl_create(dir, &inode->i_mode, &args->default_acl, &args->acl);
6138 fscrypt_free_filename(&args->fname);
6142 /* 1 to add inode item */
6143 *trans_num_items = 1;
6144 /* 1 to add compression property */
6145 if (BTRFS_I(dir)->prop_compress)
6146 (*trans_num_items)++;
6147 /* 1 to add default ACL xattr */
6148 if (args->default_acl)
6149 (*trans_num_items)++;
6150 /* 1 to add access ACL xattr */
6152 (*trans_num_items)++;
6153 #ifdef CONFIG_SECURITY
6154 /* 1 to add LSM xattr */
6155 if (dir->i_security)
6156 (*trans_num_items)++;
6159 /* 1 to add orphan item */
6160 (*trans_num_items)++;
6164 * 1 to add dir index
6165 * 1 to update parent inode item
6167 * No need for 1 unit for the inode ref item because it is
6168 * inserted in a batch together with the inode item at
6169 * btrfs_create_new_inode().
6171 *trans_num_items += 3;
6176 void btrfs_new_inode_args_destroy(struct btrfs_new_inode_args *args)
6178 posix_acl_release(args->acl);
6179 posix_acl_release(args->default_acl);
6180 fscrypt_free_filename(&args->fname);
6184 * Inherit flags from the parent inode.
6186 * Currently only the compression flags and the cow flags are inherited.
6188 static void btrfs_inherit_iflags(struct btrfs_inode *inode, struct btrfs_inode *dir)
6194 if (flags & BTRFS_INODE_NOCOMPRESS) {
6195 inode->flags &= ~BTRFS_INODE_COMPRESS;
6196 inode->flags |= BTRFS_INODE_NOCOMPRESS;
6197 } else if (flags & BTRFS_INODE_COMPRESS) {
6198 inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
6199 inode->flags |= BTRFS_INODE_COMPRESS;
6202 if (flags & BTRFS_INODE_NODATACOW) {
6203 inode->flags |= BTRFS_INODE_NODATACOW;
6204 if (S_ISREG(inode->vfs_inode.i_mode))
6205 inode->flags |= BTRFS_INODE_NODATASUM;
6208 btrfs_sync_inode_flags_to_i_flags(&inode->vfs_inode);
6211 int btrfs_create_new_inode(struct btrfs_trans_handle *trans,
6212 struct btrfs_new_inode_args *args)
6214 struct inode *dir = args->dir;
6215 struct inode *inode = args->inode;
6216 const struct fscrypt_str *name = args->orphan ? NULL : &args->fname.disk_name;
6217 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6218 struct btrfs_root *root;
6219 struct btrfs_inode_item *inode_item;
6220 struct btrfs_key *location;
6221 struct btrfs_path *path;
6223 struct btrfs_inode_ref *ref;
6224 struct btrfs_key key[2];
6226 struct btrfs_item_batch batch;
6230 path = btrfs_alloc_path();
6235 BTRFS_I(inode)->root = btrfs_grab_root(BTRFS_I(dir)->root);
6236 root = BTRFS_I(inode)->root;
6238 ret = btrfs_get_free_objectid(root, &objectid);
6241 inode->i_ino = objectid;
6245 * O_TMPFILE, set link count to 0, so that after this point, we
6246 * fill in an inode item with the correct link count.
6248 set_nlink(inode, 0);
6250 trace_btrfs_inode_request(dir);
6252 ret = btrfs_set_inode_index(BTRFS_I(dir), &BTRFS_I(inode)->dir_index);
6256 /* index_cnt is ignored for everything but a dir. */
6257 BTRFS_I(inode)->index_cnt = BTRFS_DIR_START_INDEX;
6258 BTRFS_I(inode)->generation = trans->transid;
6259 inode->i_generation = BTRFS_I(inode)->generation;
6262 * Subvolumes don't inherit flags from their parent directory.
6263 * Originally this was probably by accident, but we probably can't
6264 * change it now without compatibility issues.
6267 btrfs_inherit_iflags(BTRFS_I(inode), BTRFS_I(dir));
6269 if (S_ISREG(inode->i_mode)) {
6270 if (btrfs_test_opt(fs_info, NODATASUM))
6271 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6272 if (btrfs_test_opt(fs_info, NODATACOW))
6273 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6274 BTRFS_INODE_NODATASUM;
6277 location = &BTRFS_I(inode)->location;
6278 location->objectid = objectid;
6279 location->offset = 0;
6280 location->type = BTRFS_INODE_ITEM_KEY;
6282 ret = btrfs_insert_inode_locked(inode);
6285 BTRFS_I(dir)->index_cnt--;
6290 * We could have gotten an inode number from somebody who was fsynced
6291 * and then removed in this same transaction, so let's just set full
6292 * sync since it will be a full sync anyway and this will blow away the
6293 * old info in the log.
6295 btrfs_set_inode_full_sync(BTRFS_I(inode));
6297 key[0].objectid = objectid;
6298 key[0].type = BTRFS_INODE_ITEM_KEY;
6301 sizes[0] = sizeof(struct btrfs_inode_item);
6303 if (!args->orphan) {
6305 * Start new inodes with an inode_ref. This is slightly more
6306 * efficient for small numbers of hard links since they will
6307 * be packed into one item. Extended refs will kick in if we
6308 * add more hard links than can fit in the ref item.
6310 key[1].objectid = objectid;
6311 key[1].type = BTRFS_INODE_REF_KEY;
6313 key[1].offset = objectid;
6314 sizes[1] = 2 + sizeof(*ref);
6316 key[1].offset = btrfs_ino(BTRFS_I(dir));
6317 sizes[1] = name->len + sizeof(*ref);
6321 batch.keys = &key[0];
6322 batch.data_sizes = &sizes[0];
6323 batch.total_data_size = sizes[0] + (args->orphan ? 0 : sizes[1]);
6324 batch.nr = args->orphan ? 1 : 2;
6325 ret = btrfs_insert_empty_items(trans, root, path, &batch);
6327 btrfs_abort_transaction(trans, ret);
6331 inode->i_mtime = current_time(inode);
6332 inode->i_atime = inode->i_mtime;
6333 inode->i_ctime = inode->i_mtime;
6334 BTRFS_I(inode)->i_otime = inode->i_mtime;
6337 * We're going to fill the inode item now, so at this point the inode
6338 * must be fully initialized.
6341 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6342 struct btrfs_inode_item);
6343 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6344 sizeof(*inode_item));
6345 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6347 if (!args->orphan) {
6348 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6349 struct btrfs_inode_ref);
6350 ptr = (unsigned long)(ref + 1);
6352 btrfs_set_inode_ref_name_len(path->nodes[0], ref, 2);
6353 btrfs_set_inode_ref_index(path->nodes[0], ref, 0);
6354 write_extent_buffer(path->nodes[0], "..", ptr, 2);
6356 btrfs_set_inode_ref_name_len(path->nodes[0], ref,
6358 btrfs_set_inode_ref_index(path->nodes[0], ref,
6359 BTRFS_I(inode)->dir_index);
6360 write_extent_buffer(path->nodes[0], name->name, ptr,
6365 btrfs_mark_buffer_dirty(path->nodes[0]);
6367 * We don't need the path anymore, plus inheriting properties, adding
6368 * ACLs, security xattrs, orphan item or adding the link, will result in
6369 * allocating yet another path. So just free our path.
6371 btrfs_free_path(path);
6375 struct inode *parent;
6378 * Subvolumes inherit properties from their parent subvolume,
6379 * not the directory they were created in.
6381 parent = btrfs_iget(fs_info->sb, BTRFS_FIRST_FREE_OBJECTID,
6382 BTRFS_I(dir)->root);
6383 if (IS_ERR(parent)) {
6384 ret = PTR_ERR(parent);
6386 ret = btrfs_inode_inherit_props(trans, inode, parent);
6390 ret = btrfs_inode_inherit_props(trans, inode, dir);
6394 "error inheriting props for ino %llu (root %llu): %d",
6395 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid,
6400 * Subvolumes don't inherit ACLs or get passed to the LSM. This is
6403 if (!args->subvol) {
6404 ret = btrfs_init_inode_security(trans, args);
6406 btrfs_abort_transaction(trans, ret);
6411 inode_tree_add(BTRFS_I(inode));
6413 trace_btrfs_inode_new(inode);
6414 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6416 btrfs_update_root_times(trans, root);
6419 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
6421 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
6422 0, BTRFS_I(inode)->dir_index);
6425 btrfs_abort_transaction(trans, ret);
6433 * discard_new_inode() calls iput(), but the caller owns the reference
6437 discard_new_inode(inode);
6439 btrfs_free_path(path);
6444 * utility function to add 'inode' into 'parent_inode' with
6445 * a give name and a given sequence number.
6446 * if 'add_backref' is true, also insert a backref from the
6447 * inode to the parent directory.
6449 int btrfs_add_link(struct btrfs_trans_handle *trans,
6450 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6451 const struct fscrypt_str *name, int add_backref, u64 index)
6454 struct btrfs_key key;
6455 struct btrfs_root *root = parent_inode->root;
6456 u64 ino = btrfs_ino(inode);
6457 u64 parent_ino = btrfs_ino(parent_inode);
6459 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6460 memcpy(&key, &inode->root->root_key, sizeof(key));
6463 key.type = BTRFS_INODE_ITEM_KEY;
6467 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6468 ret = btrfs_add_root_ref(trans, key.objectid,
6469 root->root_key.objectid, parent_ino,
6471 } else if (add_backref) {
6472 ret = btrfs_insert_inode_ref(trans, root, name,
6473 ino, parent_ino, index);
6476 /* Nothing to clean up yet */
6480 ret = btrfs_insert_dir_item(trans, name, parent_inode, &key,
6481 btrfs_inode_type(&inode->vfs_inode), index);
6482 if (ret == -EEXIST || ret == -EOVERFLOW)
6485 btrfs_abort_transaction(trans, ret);
6489 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6491 inode_inc_iversion(&parent_inode->vfs_inode);
6493 * If we are replaying a log tree, we do not want to update the mtime
6494 * and ctime of the parent directory with the current time, since the
6495 * log replay procedure is responsible for setting them to their correct
6496 * values (the ones it had when the fsync was done).
6498 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
6499 struct timespec64 now = current_time(&parent_inode->vfs_inode);
6501 parent_inode->vfs_inode.i_mtime = now;
6502 parent_inode->vfs_inode.i_ctime = now;
6504 ret = btrfs_update_inode(trans, root, parent_inode);
6506 btrfs_abort_transaction(trans, ret);
6510 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6513 err = btrfs_del_root_ref(trans, key.objectid,
6514 root->root_key.objectid, parent_ino,
6515 &local_index, name);
6517 btrfs_abort_transaction(trans, err);
6518 } else if (add_backref) {
6522 err = btrfs_del_inode_ref(trans, root, name, ino, parent_ino,
6525 btrfs_abort_transaction(trans, err);
6528 /* Return the original error code */
6532 static int btrfs_create_common(struct inode *dir, struct dentry *dentry,
6533 struct inode *inode)
6535 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6536 struct btrfs_root *root = BTRFS_I(dir)->root;
6537 struct btrfs_new_inode_args new_inode_args = {
6542 unsigned int trans_num_items;
6543 struct btrfs_trans_handle *trans;
6546 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
6550 trans = btrfs_start_transaction(root, trans_num_items);
6551 if (IS_ERR(trans)) {
6552 err = PTR_ERR(trans);
6553 goto out_new_inode_args;
6556 err = btrfs_create_new_inode(trans, &new_inode_args);
6558 d_instantiate_new(dentry, inode);
6560 btrfs_end_transaction(trans);
6561 btrfs_btree_balance_dirty(fs_info);
6563 btrfs_new_inode_args_destroy(&new_inode_args);
6570 static int btrfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
6571 struct dentry *dentry, umode_t mode, dev_t rdev)
6573 struct inode *inode;
6575 inode = new_inode(dir->i_sb);
6578 inode_init_owner(idmap, inode, dir, mode);
6579 inode->i_op = &btrfs_special_inode_operations;
6580 init_special_inode(inode, inode->i_mode, rdev);
6581 return btrfs_create_common(dir, dentry, inode);
6584 static int btrfs_create(struct mnt_idmap *idmap, struct inode *dir,
6585 struct dentry *dentry, umode_t mode, bool excl)
6587 struct inode *inode;
6589 inode = new_inode(dir->i_sb);
6592 inode_init_owner(idmap, inode, dir, mode);
6593 inode->i_fop = &btrfs_file_operations;
6594 inode->i_op = &btrfs_file_inode_operations;
6595 inode->i_mapping->a_ops = &btrfs_aops;
6596 return btrfs_create_common(dir, dentry, inode);
6599 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6600 struct dentry *dentry)
6602 struct btrfs_trans_handle *trans = NULL;
6603 struct btrfs_root *root = BTRFS_I(dir)->root;
6604 struct inode *inode = d_inode(old_dentry);
6605 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6606 struct fscrypt_name fname;
6611 /* do not allow sys_link's with other subvols of the same device */
6612 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6615 if (inode->i_nlink >= BTRFS_LINK_MAX)
6618 err = fscrypt_setup_filename(dir, &dentry->d_name, 0, &fname);
6622 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6627 * 2 items for inode and inode ref
6628 * 2 items for dir items
6629 * 1 item for parent inode
6630 * 1 item for orphan item deletion if O_TMPFILE
6632 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6633 if (IS_ERR(trans)) {
6634 err = PTR_ERR(trans);
6639 /* There are several dir indexes for this inode, clear the cache. */
6640 BTRFS_I(inode)->dir_index = 0ULL;
6642 inode_inc_iversion(inode);
6643 inode->i_ctime = current_time(inode);
6645 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6647 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6648 &fname.disk_name, 1, index);
6653 struct dentry *parent = dentry->d_parent;
6655 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6658 if (inode->i_nlink == 1) {
6660 * If new hard link count is 1, it's a file created
6661 * with open(2) O_TMPFILE flag.
6663 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6667 d_instantiate(dentry, inode);
6668 btrfs_log_new_name(trans, old_dentry, NULL, 0, parent);
6672 fscrypt_free_filename(&fname);
6674 btrfs_end_transaction(trans);
6676 inode_dec_link_count(inode);
6679 btrfs_btree_balance_dirty(fs_info);
6683 static int btrfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
6684 struct dentry *dentry, umode_t mode)
6686 struct inode *inode;
6688 inode = new_inode(dir->i_sb);
6691 inode_init_owner(idmap, inode, dir, S_IFDIR | mode);
6692 inode->i_op = &btrfs_dir_inode_operations;
6693 inode->i_fop = &btrfs_dir_file_operations;
6694 return btrfs_create_common(dir, dentry, inode);
6697 static noinline int uncompress_inline(struct btrfs_path *path,
6699 struct btrfs_file_extent_item *item)
6702 struct extent_buffer *leaf = path->nodes[0];
6705 unsigned long inline_size;
6709 compress_type = btrfs_file_extent_compression(leaf, item);
6710 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6711 inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]);
6712 tmp = kmalloc(inline_size, GFP_NOFS);
6715 ptr = btrfs_file_extent_inline_start(item);
6717 read_extent_buffer(leaf, tmp, ptr, inline_size);
6719 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6720 ret = btrfs_decompress(compress_type, tmp, page, 0, inline_size, max_size);
6723 * decompression code contains a memset to fill in any space between the end
6724 * of the uncompressed data and the end of max_size in case the decompressed
6725 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6726 * the end of an inline extent and the beginning of the next block, so we
6727 * cover that region here.
6730 if (max_size < PAGE_SIZE)
6731 memzero_page(page, max_size, PAGE_SIZE - max_size);
6736 static int read_inline_extent(struct btrfs_inode *inode, struct btrfs_path *path,
6739 struct btrfs_file_extent_item *fi;
6743 if (!page || PageUptodate(page))
6746 ASSERT(page_offset(page) == 0);
6748 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
6749 struct btrfs_file_extent_item);
6750 if (btrfs_file_extent_compression(path->nodes[0], fi) != BTRFS_COMPRESS_NONE)
6751 return uncompress_inline(path, page, fi);
6753 copy_size = min_t(u64, PAGE_SIZE,
6754 btrfs_file_extent_ram_bytes(path->nodes[0], fi));
6755 kaddr = kmap_local_page(page);
6756 read_extent_buffer(path->nodes[0], kaddr,
6757 btrfs_file_extent_inline_start(fi), copy_size);
6758 kunmap_local(kaddr);
6759 if (copy_size < PAGE_SIZE)
6760 memzero_page(page, copy_size, PAGE_SIZE - copy_size);
6765 * Lookup the first extent overlapping a range in a file.
6767 * @inode: file to search in
6768 * @page: page to read extent data into if the extent is inline
6769 * @pg_offset: offset into @page to copy to
6770 * @start: file offset
6771 * @len: length of range starting at @start
6773 * Return the first &struct extent_map which overlaps the given range, reading
6774 * it from the B-tree and caching it if necessary. Note that there may be more
6775 * extents which overlap the given range after the returned extent_map.
6777 * If @page is not NULL and the extent is inline, this also reads the extent
6778 * data directly into the page and marks the extent up to date in the io_tree.
6780 * Return: ERR_PTR on error, non-NULL extent_map on success.
6782 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6783 struct page *page, size_t pg_offset,
6786 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6788 u64 extent_start = 0;
6790 u64 objectid = btrfs_ino(inode);
6791 int extent_type = -1;
6792 struct btrfs_path *path = NULL;
6793 struct btrfs_root *root = inode->root;
6794 struct btrfs_file_extent_item *item;
6795 struct extent_buffer *leaf;
6796 struct btrfs_key found_key;
6797 struct extent_map *em = NULL;
6798 struct extent_map_tree *em_tree = &inode->extent_tree;
6800 read_lock(&em_tree->lock);
6801 em = lookup_extent_mapping(em_tree, start, len);
6802 read_unlock(&em_tree->lock);
6805 if (em->start > start || em->start + em->len <= start)
6806 free_extent_map(em);
6807 else if (em->block_start == EXTENT_MAP_INLINE && page)
6808 free_extent_map(em);
6812 em = alloc_extent_map();
6817 em->start = EXTENT_MAP_HOLE;
6818 em->orig_start = EXTENT_MAP_HOLE;
6820 em->block_len = (u64)-1;
6822 path = btrfs_alloc_path();
6828 /* Chances are we'll be called again, so go ahead and do readahead */
6829 path->reada = READA_FORWARD;
6832 * The same explanation in load_free_space_cache applies here as well,
6833 * we only read when we're loading the free space cache, and at that
6834 * point the commit_root has everything we need.
6836 if (btrfs_is_free_space_inode(inode)) {
6837 path->search_commit_root = 1;
6838 path->skip_locking = 1;
6841 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6844 } else if (ret > 0) {
6845 if (path->slots[0] == 0)
6851 leaf = path->nodes[0];
6852 item = btrfs_item_ptr(leaf, path->slots[0],
6853 struct btrfs_file_extent_item);
6854 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6855 if (found_key.objectid != objectid ||
6856 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6858 * If we backup past the first extent we want to move forward
6859 * and see if there is an extent in front of us, otherwise we'll
6860 * say there is a hole for our whole search range which can
6867 extent_type = btrfs_file_extent_type(leaf, item);
6868 extent_start = found_key.offset;
6869 extent_end = btrfs_file_extent_end(path);
6870 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6871 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6872 /* Only regular file could have regular/prealloc extent */
6873 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6876 "regular/prealloc extent found for non-regular inode %llu",
6880 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6882 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6883 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6888 if (start >= extent_end) {
6890 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6891 ret = btrfs_next_leaf(root, path);
6897 leaf = path->nodes[0];
6899 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6900 if (found_key.objectid != objectid ||
6901 found_key.type != BTRFS_EXTENT_DATA_KEY)
6903 if (start + len <= found_key.offset)
6905 if (start > found_key.offset)
6908 /* New extent overlaps with existing one */
6910 em->orig_start = start;
6911 em->len = found_key.offset - start;
6912 em->block_start = EXTENT_MAP_HOLE;
6916 btrfs_extent_item_to_extent_map(inode, path, item, em);
6918 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6919 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6921 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6923 * Inline extent can only exist at file offset 0. This is
6924 * ensured by tree-checker and inline extent creation path.
6925 * Thus all members representing file offsets should be zero.
6927 ASSERT(pg_offset == 0);
6928 ASSERT(extent_start == 0);
6929 ASSERT(em->start == 0);
6932 * btrfs_extent_item_to_extent_map() should have properly
6933 * initialized em members already.
6935 * Other members are not utilized for inline extents.
6937 ASSERT(em->block_start == EXTENT_MAP_INLINE);
6938 ASSERT(em->len == fs_info->sectorsize);
6940 ret = read_inline_extent(inode, path, page);
6947 em->orig_start = start;
6949 em->block_start = EXTENT_MAP_HOLE;
6952 btrfs_release_path(path);
6953 if (em->start > start || extent_map_end(em) <= start) {
6955 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6956 em->start, em->len, start, len);
6961 write_lock(&em_tree->lock);
6962 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6963 write_unlock(&em_tree->lock);
6965 btrfs_free_path(path);
6967 trace_btrfs_get_extent(root, inode, em);
6970 free_extent_map(em);
6971 return ERR_PTR(ret);
6976 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
6977 struct btrfs_dio_data *dio_data,
6980 const u64 orig_start,
6981 const u64 block_start,
6982 const u64 block_len,
6983 const u64 orig_block_len,
6984 const u64 ram_bytes,
6987 struct extent_map *em = NULL;
6988 struct btrfs_ordered_extent *ordered;
6990 if (type != BTRFS_ORDERED_NOCOW) {
6991 em = create_io_em(inode, start, len, orig_start, block_start,
6992 block_len, orig_block_len, ram_bytes,
6993 BTRFS_COMPRESS_NONE, /* compress_type */
6998 ordered = btrfs_alloc_ordered_extent(inode, start, len, len,
6999 block_start, block_len, 0,
7001 (1 << BTRFS_ORDERED_DIRECT),
7002 BTRFS_COMPRESS_NONE);
7003 if (IS_ERR(ordered)) {
7005 free_extent_map(em);
7006 btrfs_drop_extent_map_range(inode, start,
7007 start + len - 1, false);
7009 em = ERR_CAST(ordered);
7011 ASSERT(!dio_data->ordered);
7012 dio_data->ordered = ordered;
7019 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
7020 struct btrfs_dio_data *dio_data,
7023 struct btrfs_root *root = inode->root;
7024 struct btrfs_fs_info *fs_info = root->fs_info;
7025 struct extent_map *em;
7026 struct btrfs_key ins;
7030 alloc_hint = get_extent_allocation_hint(inode, start, len);
7031 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7032 0, alloc_hint, &ins, 1, 1);
7034 return ERR_PTR(ret);
7036 em = btrfs_create_dio_extent(inode, dio_data, start, ins.offset, start,
7037 ins.objectid, ins.offset, ins.offset,
7038 ins.offset, BTRFS_ORDERED_REGULAR);
7039 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7041 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7047 static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
7049 struct btrfs_block_group *block_group;
7050 bool readonly = false;
7052 block_group = btrfs_lookup_block_group(fs_info, bytenr);
7053 if (!block_group || block_group->ro)
7056 btrfs_put_block_group(block_group);
7061 * Check if we can do nocow write into the range [@offset, @offset + @len)
7063 * @offset: File offset
7064 * @len: The length to write, will be updated to the nocow writeable
7066 * @orig_start: (optional) Return the original file offset of the file extent
7067 * @orig_len: (optional) Return the original on-disk length of the file extent
7068 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7069 * @strict: if true, omit optimizations that might force us into unnecessary
7070 * cow. e.g., don't trust generation number.
7073 * >0 and update @len if we can do nocow write
7074 * 0 if we can't do nocow write
7075 * <0 if error happened
7077 * NOTE: This only checks the file extents, caller is responsible to wait for
7078 * any ordered extents.
7080 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7081 u64 *orig_start, u64 *orig_block_len,
7082 u64 *ram_bytes, bool nowait, bool strict)
7084 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7085 struct can_nocow_file_extent_args nocow_args = { 0 };
7086 struct btrfs_path *path;
7088 struct extent_buffer *leaf;
7089 struct btrfs_root *root = BTRFS_I(inode)->root;
7090 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7091 struct btrfs_file_extent_item *fi;
7092 struct btrfs_key key;
7095 path = btrfs_alloc_path();
7098 path->nowait = nowait;
7100 ret = btrfs_lookup_file_extent(NULL, root, path,
7101 btrfs_ino(BTRFS_I(inode)), offset, 0);
7106 if (path->slots[0] == 0) {
7107 /* can't find the item, must cow */
7114 leaf = path->nodes[0];
7115 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7116 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7117 key.type != BTRFS_EXTENT_DATA_KEY) {
7118 /* not our file or wrong item type, must cow */
7122 if (key.offset > offset) {
7123 /* Wrong offset, must cow */
7127 if (btrfs_file_extent_end(path) <= offset)
7130 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
7131 found_type = btrfs_file_extent_type(leaf, fi);
7133 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7135 nocow_args.start = offset;
7136 nocow_args.end = offset + *len - 1;
7137 nocow_args.strict = strict;
7138 nocow_args.free_path = true;
7140 ret = can_nocow_file_extent(path, &key, BTRFS_I(inode), &nocow_args);
7141 /* can_nocow_file_extent() has freed the path. */
7145 /* Treat errors as not being able to NOCOW. */
7151 if (btrfs_extent_readonly(fs_info, nocow_args.disk_bytenr))
7154 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7155 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7158 range_end = round_up(offset + nocow_args.num_bytes,
7159 root->fs_info->sectorsize) - 1;
7160 ret = test_range_bit(io_tree, offset, range_end,
7161 EXTENT_DELALLOC, 0, NULL);
7169 *orig_start = key.offset - nocow_args.extent_offset;
7171 *orig_block_len = nocow_args.disk_num_bytes;
7173 *len = nocow_args.num_bytes;
7176 btrfs_free_path(path);
7180 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7181 struct extent_state **cached_state,
7182 unsigned int iomap_flags)
7184 const bool writing = (iomap_flags & IOMAP_WRITE);
7185 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7186 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7187 struct btrfs_ordered_extent *ordered;
7192 if (!try_lock_extent(io_tree, lockstart, lockend,
7196 lock_extent(io_tree, lockstart, lockend, cached_state);
7199 * We're concerned with the entire range that we're going to be
7200 * doing DIO to, so we need to make sure there's no ordered
7201 * extents in this range.
7203 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7204 lockend - lockstart + 1);
7207 * We need to make sure there are no buffered pages in this
7208 * range either, we could have raced between the invalidate in
7209 * generic_file_direct_write and locking the extent. The
7210 * invalidate needs to happen so that reads after a write do not
7214 (!writing || !filemap_range_has_page(inode->i_mapping,
7215 lockstart, lockend)))
7218 unlock_extent(io_tree, lockstart, lockend, cached_state);
7222 btrfs_put_ordered_extent(ordered);
7227 * If we are doing a DIO read and the ordered extent we
7228 * found is for a buffered write, we can not wait for it
7229 * to complete and retry, because if we do so we can
7230 * deadlock with concurrent buffered writes on page
7231 * locks. This happens only if our DIO read covers more
7232 * than one extent map, if at this point has already
7233 * created an ordered extent for a previous extent map
7234 * and locked its range in the inode's io tree, and a
7235 * concurrent write against that previous extent map's
7236 * range and this range started (we unlock the ranges
7237 * in the io tree only when the bios complete and
7238 * buffered writes always lock pages before attempting
7239 * to lock range in the io tree).
7242 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7243 btrfs_start_ordered_extent(ordered);
7245 ret = nowait ? -EAGAIN : -ENOTBLK;
7246 btrfs_put_ordered_extent(ordered);
7249 * We could trigger writeback for this range (and wait
7250 * for it to complete) and then invalidate the pages for
7251 * this range (through invalidate_inode_pages2_range()),
7252 * but that can lead us to a deadlock with a concurrent
7253 * call to readahead (a buffered read or a defrag call
7254 * triggered a readahead) on a page lock due to an
7255 * ordered dio extent we created before but did not have
7256 * yet a corresponding bio submitted (whence it can not
7257 * complete), which makes readahead wait for that
7258 * ordered extent to complete while holding a lock on
7261 ret = nowait ? -EAGAIN : -ENOTBLK;
7273 /* The callers of this must take lock_extent() */
7274 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7275 u64 len, u64 orig_start, u64 block_start,
7276 u64 block_len, u64 orig_block_len,
7277 u64 ram_bytes, int compress_type,
7280 struct extent_map *em;
7283 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7284 type == BTRFS_ORDERED_COMPRESSED ||
7285 type == BTRFS_ORDERED_NOCOW ||
7286 type == BTRFS_ORDERED_REGULAR);
7288 em = alloc_extent_map();
7290 return ERR_PTR(-ENOMEM);
7293 em->orig_start = orig_start;
7295 em->block_len = block_len;
7296 em->block_start = block_start;
7297 em->orig_block_len = orig_block_len;
7298 em->ram_bytes = ram_bytes;
7299 em->generation = -1;
7300 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7301 if (type == BTRFS_ORDERED_PREALLOC) {
7302 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7303 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7304 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7305 em->compress_type = compress_type;
7308 ret = btrfs_replace_extent_map_range(inode, em, true);
7310 free_extent_map(em);
7311 return ERR_PTR(ret);
7314 /* em got 2 refs now, callers needs to do free_extent_map once. */
7319 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7320 struct inode *inode,
7321 struct btrfs_dio_data *dio_data,
7322 u64 start, u64 *lenp,
7323 unsigned int iomap_flags)
7325 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7326 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7327 struct extent_map *em = *map;
7329 u64 block_start, orig_start, orig_block_len, ram_bytes;
7330 struct btrfs_block_group *bg;
7331 bool can_nocow = false;
7332 bool space_reserved = false;
7338 * We don't allocate a new extent in the following cases
7340 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7342 * 2) The extent is marked as PREALLOC. We're good to go here and can
7343 * just use the extent.
7346 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7347 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7348 em->block_start != EXTENT_MAP_HOLE)) {
7349 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7350 type = BTRFS_ORDERED_PREALLOC;
7352 type = BTRFS_ORDERED_NOCOW;
7353 len = min(len, em->len - (start - em->start));
7354 block_start = em->block_start + (start - em->start);
7356 if (can_nocow_extent(inode, start, &len, &orig_start,
7357 &orig_block_len, &ram_bytes, false, false) == 1) {
7358 bg = btrfs_inc_nocow_writers(fs_info, block_start);
7366 struct extent_map *em2;
7368 /* We can NOCOW, so only need to reserve metadata space. */
7369 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7372 /* Our caller expects us to free the input extent map. */
7373 free_extent_map(em);
7375 btrfs_dec_nocow_writers(bg);
7376 if (nowait && (ret == -ENOSPC || ret == -EDQUOT))
7380 space_reserved = true;
7382 em2 = btrfs_create_dio_extent(BTRFS_I(inode), dio_data, start, len,
7383 orig_start, block_start,
7384 len, orig_block_len,
7386 btrfs_dec_nocow_writers(bg);
7387 if (type == BTRFS_ORDERED_PREALLOC) {
7388 free_extent_map(em);
7398 dio_data->nocow_done = true;
7400 /* Our caller expects us to free the input extent map. */
7401 free_extent_map(em);
7410 * If we could not allocate data space before locking the file
7411 * range and we can't do a NOCOW write, then we have to fail.
7413 if (!dio_data->data_space_reserved) {
7419 * We have to COW and we have already reserved data space before,
7420 * so now we reserve only metadata.
7422 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7426 space_reserved = true;
7428 em = btrfs_new_extent_direct(BTRFS_I(inode), dio_data, start, len);
7434 len = min(len, em->len - (start - em->start));
7436 btrfs_delalloc_release_metadata(BTRFS_I(inode),
7437 prev_len - len, true);
7441 * We have created our ordered extent, so we can now release our reservation
7442 * for an outstanding extent.
7444 btrfs_delalloc_release_extents(BTRFS_I(inode), prev_len);
7447 * Need to update the i_size under the extent lock so buffered
7448 * readers will get the updated i_size when we unlock.
7450 if (start + len > i_size_read(inode))
7451 i_size_write(inode, start + len);
7453 if (ret && space_reserved) {
7454 btrfs_delalloc_release_extents(BTRFS_I(inode), len);
7455 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, true);
7461 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7462 loff_t length, unsigned int flags, struct iomap *iomap,
7463 struct iomap *srcmap)
7465 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7466 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7467 struct extent_map *em;
7468 struct extent_state *cached_state = NULL;
7469 struct btrfs_dio_data *dio_data = iter->private;
7470 u64 lockstart, lockend;
7471 const bool write = !!(flags & IOMAP_WRITE);
7474 const u64 data_alloc_len = length;
7475 bool unlock_extents = false;
7478 * We could potentially fault if we have a buffer > PAGE_SIZE, and if
7479 * we're NOWAIT we may submit a bio for a partial range and return
7480 * EIOCBQUEUED, which would result in an errant short read.
7482 * The best way to handle this would be to allow for partial completions
7483 * of iocb's, so we could submit the partial bio, return and fault in
7484 * the rest of the pages, and then submit the io for the rest of the
7485 * range. However we don't have that currently, so simply return
7486 * -EAGAIN at this point so that the normal path is used.
7488 if (!write && (flags & IOMAP_NOWAIT) && length > PAGE_SIZE)
7492 * Cap the size of reads to that usually seen in buffered I/O as we need
7493 * to allocate a contiguous array for the checksums.
7496 len = min_t(u64, len, fs_info->sectorsize * BTRFS_MAX_BIO_SECTORS);
7499 lockend = start + len - 1;
7502 * iomap_dio_rw() only does filemap_write_and_wait_range(), which isn't
7503 * enough if we've written compressed pages to this area, so we need to
7504 * flush the dirty pages again to make absolutely sure that any
7505 * outstanding dirty pages are on disk - the first flush only starts
7506 * compression on the data, while keeping the pages locked, so by the
7507 * time the second flush returns we know bios for the compressed pages
7508 * were submitted and finished, and the pages no longer under writeback.
7510 * If we have a NOWAIT request and we have any pages in the range that
7511 * are locked, likely due to compression still in progress, we don't want
7512 * to block on page locks. We also don't want to block on pages marked as
7513 * dirty or under writeback (same as for the non-compression case).
7514 * iomap_dio_rw() did the same check, but after that and before we got
7515 * here, mmap'ed writes may have happened or buffered reads started
7516 * (readpage() and readahead(), which lock pages), as we haven't locked
7517 * the file range yet.
7519 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7520 &BTRFS_I(inode)->runtime_flags)) {
7521 if (flags & IOMAP_NOWAIT) {
7522 if (filemap_range_needs_writeback(inode->i_mapping,
7523 lockstart, lockend))
7526 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7527 start + length - 1);
7533 memset(dio_data, 0, sizeof(*dio_data));
7536 * We always try to allocate data space and must do it before locking
7537 * the file range, to avoid deadlocks with concurrent writes to the same
7538 * range if the range has several extents and the writes don't expand the
7539 * current i_size (the inode lock is taken in shared mode). If we fail to
7540 * allocate data space here we continue and later, after locking the
7541 * file range, we fail with ENOSPC only if we figure out we can not do a
7544 if (write && !(flags & IOMAP_NOWAIT)) {
7545 ret = btrfs_check_data_free_space(BTRFS_I(inode),
7546 &dio_data->data_reserved,
7547 start, data_alloc_len, false);
7549 dio_data->data_space_reserved = true;
7550 else if (ret && !(BTRFS_I(inode)->flags &
7551 (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
7556 * If this errors out it's because we couldn't invalidate pagecache for
7557 * this range and we need to fallback to buffered IO, or we are doing a
7558 * NOWAIT read/write and we need to block.
7560 ret = lock_extent_direct(inode, lockstart, lockend, &cached_state, flags);
7564 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7571 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7572 * io. INLINE is special, and we could probably kludge it in here, but
7573 * it's still buffered so for safety lets just fall back to the generic
7576 * For COMPRESSED we _have_ to read the entire extent in so we can
7577 * decompress it, so there will be buffering required no matter what we
7578 * do, so go ahead and fallback to buffered.
7580 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7581 * to buffered IO. Don't blame me, this is the price we pay for using
7584 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7585 em->block_start == EXTENT_MAP_INLINE) {
7586 free_extent_map(em);
7588 * If we are in a NOWAIT context, return -EAGAIN in order to
7589 * fallback to buffered IO. This is not only because we can
7590 * block with buffered IO (no support for NOWAIT semantics at
7591 * the moment) but also to avoid returning short reads to user
7592 * space - this happens if we were able to read some data from
7593 * previous non-compressed extents and then when we fallback to
7594 * buffered IO, at btrfs_file_read_iter() by calling
7595 * filemap_read(), we fail to fault in pages for the read buffer,
7596 * in which case filemap_read() returns a short read (the number
7597 * of bytes previously read is > 0, so it does not return -EFAULT).
7599 ret = (flags & IOMAP_NOWAIT) ? -EAGAIN : -ENOTBLK;
7603 len = min(len, em->len - (start - em->start));
7606 * If we have a NOWAIT request and the range contains multiple extents
7607 * (or a mix of extents and holes), then we return -EAGAIN to make the
7608 * caller fallback to a context where it can do a blocking (without
7609 * NOWAIT) request. This way we avoid doing partial IO and returning
7610 * success to the caller, which is not optimal for writes and for reads
7611 * it can result in unexpected behaviour for an application.
7613 * When doing a read, because we use IOMAP_DIO_PARTIAL when calling
7614 * iomap_dio_rw(), we can end up returning less data then what the caller
7615 * asked for, resulting in an unexpected, and incorrect, short read.
7616 * That is, the caller asked to read N bytes and we return less than that,
7617 * which is wrong unless we are crossing EOF. This happens if we get a
7618 * page fault error when trying to fault in pages for the buffer that is
7619 * associated to the struct iov_iter passed to iomap_dio_rw(), and we
7620 * have previously submitted bios for other extents in the range, in
7621 * which case iomap_dio_rw() may return us EIOCBQUEUED if not all of
7622 * those bios have completed by the time we get the page fault error,
7623 * which we return back to our caller - we should only return EIOCBQUEUED
7624 * after we have submitted bios for all the extents in the range.
7626 if ((flags & IOMAP_NOWAIT) && len < length) {
7627 free_extent_map(em);
7633 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7634 start, &len, flags);
7637 unlock_extents = true;
7638 /* Recalc len in case the new em is smaller than requested */
7639 len = min(len, em->len - (start - em->start));
7640 if (dio_data->data_space_reserved) {
7642 u64 release_len = 0;
7644 if (dio_data->nocow_done) {
7645 release_offset = start;
7646 release_len = data_alloc_len;
7647 } else if (len < data_alloc_len) {
7648 release_offset = start + len;
7649 release_len = data_alloc_len - len;
7652 if (release_len > 0)
7653 btrfs_free_reserved_data_space(BTRFS_I(inode),
7654 dio_data->data_reserved,
7660 * We need to unlock only the end area that we aren't using.
7661 * The rest is going to be unlocked by the endio routine.
7663 lockstart = start + len;
7664 if (lockstart < lockend)
7665 unlock_extents = true;
7669 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7672 free_extent_state(cached_state);
7675 * Translate extent map information to iomap.
7676 * We trim the extents (and move the addr) even though iomap code does
7677 * that, since we have locked only the parts we are performing I/O in.
7679 if ((em->block_start == EXTENT_MAP_HOLE) ||
7680 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7681 iomap->addr = IOMAP_NULL_ADDR;
7682 iomap->type = IOMAP_HOLE;
7684 iomap->addr = em->block_start + (start - em->start);
7685 iomap->type = IOMAP_MAPPED;
7687 iomap->offset = start;
7688 iomap->bdev = fs_info->fs_devices->latest_dev->bdev;
7689 iomap->length = len;
7690 free_extent_map(em);
7695 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7698 if (dio_data->data_space_reserved) {
7699 btrfs_free_reserved_data_space(BTRFS_I(inode),
7700 dio_data->data_reserved,
7701 start, data_alloc_len);
7702 extent_changeset_free(dio_data->data_reserved);
7708 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7709 ssize_t written, unsigned int flags, struct iomap *iomap)
7711 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7712 struct btrfs_dio_data *dio_data = iter->private;
7713 size_t submitted = dio_data->submitted;
7714 const bool write = !!(flags & IOMAP_WRITE);
7717 if (!write && (iomap->type == IOMAP_HOLE)) {
7718 /* If reading from a hole, unlock and return */
7719 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1,
7724 if (submitted < length) {
7726 length -= submitted;
7728 btrfs_finish_ordered_extent(dio_data->ordered, NULL,
7729 pos, length, false);
7731 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7732 pos + length - 1, NULL);
7736 btrfs_put_ordered_extent(dio_data->ordered);
7737 dio_data->ordered = NULL;
7741 extent_changeset_free(dio_data->data_reserved);
7745 static void btrfs_dio_end_io(struct btrfs_bio *bbio)
7747 struct btrfs_dio_private *dip =
7748 container_of(bbio, struct btrfs_dio_private, bbio);
7749 struct btrfs_inode *inode = bbio->inode;
7750 struct bio *bio = &bbio->bio;
7752 if (bio->bi_status) {
7753 btrfs_warn(inode->root->fs_info,
7754 "direct IO failed ino %llu op 0x%0x offset %#llx len %u err no %d",
7755 btrfs_ino(inode), bio->bi_opf,
7756 dip->file_offset, dip->bytes, bio->bi_status);
7759 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
7760 btrfs_finish_ordered_extent(bbio->ordered, NULL,
7761 dip->file_offset, dip->bytes,
7764 unlock_extent(&inode->io_tree, dip->file_offset,
7765 dip->file_offset + dip->bytes - 1, NULL);
7768 bbio->bio.bi_private = bbio->private;
7769 iomap_dio_bio_end_io(bio);
7772 static void btrfs_dio_submit_io(const struct iomap_iter *iter, struct bio *bio,
7775 struct btrfs_bio *bbio = btrfs_bio(bio);
7776 struct btrfs_dio_private *dip =
7777 container_of(bbio, struct btrfs_dio_private, bbio);
7778 struct btrfs_dio_data *dio_data = iter->private;
7780 btrfs_bio_init(bbio, BTRFS_I(iter->inode)->root->fs_info,
7781 btrfs_dio_end_io, bio->bi_private);
7782 bbio->inode = BTRFS_I(iter->inode);
7783 bbio->file_offset = file_offset;
7785 dip->file_offset = file_offset;
7786 dip->bytes = bio->bi_iter.bi_size;
7788 dio_data->submitted += bio->bi_iter.bi_size;
7791 * Check if we are doing a partial write. If we are, we need to split
7792 * the ordered extent to match the submitted bio. Hang on to the
7793 * remaining unfinishable ordered_extent in dio_data so that it can be
7794 * cancelled in iomap_end to avoid a deadlock wherein faulting the
7795 * remaining pages is blocked on the outstanding ordered extent.
7797 if (iter->flags & IOMAP_WRITE) {
7800 ret = btrfs_extract_ordered_extent(bbio, dio_data->ordered);
7802 btrfs_finish_ordered_extent(dio_data->ordered, NULL,
7803 file_offset, dip->bytes,
7805 bio->bi_status = errno_to_blk_status(ret);
7806 iomap_dio_bio_end_io(bio);
7811 btrfs_submit_bio(bbio, 0);
7814 static const struct iomap_ops btrfs_dio_iomap_ops = {
7815 .iomap_begin = btrfs_dio_iomap_begin,
7816 .iomap_end = btrfs_dio_iomap_end,
7819 static const struct iomap_dio_ops btrfs_dio_ops = {
7820 .submit_io = btrfs_dio_submit_io,
7821 .bio_set = &btrfs_dio_bioset,
7824 ssize_t btrfs_dio_read(struct kiocb *iocb, struct iov_iter *iter, size_t done_before)
7826 struct btrfs_dio_data data = { 0 };
7828 return iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7829 IOMAP_DIO_PARTIAL, &data, done_before);
7832 struct iomap_dio *btrfs_dio_write(struct kiocb *iocb, struct iov_iter *iter,
7835 struct btrfs_dio_data data = { 0 };
7837 return __iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7838 IOMAP_DIO_PARTIAL, &data, done_before);
7841 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7846 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
7851 * fiemap_prep() called filemap_write_and_wait() for the whole possible
7852 * file range (0 to LLONG_MAX), but that is not enough if we have
7853 * compression enabled. The first filemap_fdatawrite_range() only kicks
7854 * in the compression of data (in an async thread) and will return
7855 * before the compression is done and writeback is started. A second
7856 * filemap_fdatawrite_range() is needed to wait for the compression to
7857 * complete and writeback to start. We also need to wait for ordered
7858 * extents to complete, because our fiemap implementation uses mainly
7859 * file extent items to list the extents, searching for extent maps
7860 * only for file ranges with holes or prealloc extents to figure out
7861 * if we have delalloc in those ranges.
7863 if (fieinfo->fi_flags & FIEMAP_FLAG_SYNC) {
7864 ret = btrfs_wait_ordered_range(inode, 0, LLONG_MAX);
7869 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
7872 static int btrfs_writepages(struct address_space *mapping,
7873 struct writeback_control *wbc)
7875 return extent_writepages(mapping, wbc);
7878 static void btrfs_readahead(struct readahead_control *rac)
7880 extent_readahead(rac);
7884 * For release_folio() and invalidate_folio() we have a race window where
7885 * folio_end_writeback() is called but the subpage spinlock is not yet released.
7886 * If we continue to release/invalidate the page, we could cause use-after-free
7887 * for subpage spinlock. So this function is to spin and wait for subpage
7890 static void wait_subpage_spinlock(struct page *page)
7892 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
7893 struct btrfs_subpage *subpage;
7895 if (!btrfs_is_subpage(fs_info, page))
7898 ASSERT(PagePrivate(page) && page->private);
7899 subpage = (struct btrfs_subpage *)page->private;
7902 * This may look insane as we just acquire the spinlock and release it,
7903 * without doing anything. But we just want to make sure no one is
7904 * still holding the subpage spinlock.
7905 * And since the page is not dirty nor writeback, and we have page
7906 * locked, the only possible way to hold a spinlock is from the endio
7907 * function to clear page writeback.
7909 * Here we just acquire the spinlock so that all existing callers
7910 * should exit and we're safe to release/invalidate the page.
7912 spin_lock_irq(&subpage->lock);
7913 spin_unlock_irq(&subpage->lock);
7916 static bool __btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
7918 int ret = try_release_extent_mapping(&folio->page, gfp_flags);
7921 wait_subpage_spinlock(&folio->page);
7922 clear_page_extent_mapped(&folio->page);
7927 static bool btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
7929 if (folio_test_writeback(folio) || folio_test_dirty(folio))
7931 return __btrfs_release_folio(folio, gfp_flags);
7934 #ifdef CONFIG_MIGRATION
7935 static int btrfs_migrate_folio(struct address_space *mapping,
7936 struct folio *dst, struct folio *src,
7937 enum migrate_mode mode)
7939 int ret = filemap_migrate_folio(mapping, dst, src, mode);
7941 if (ret != MIGRATEPAGE_SUCCESS)
7944 if (folio_test_ordered(src)) {
7945 folio_clear_ordered(src);
7946 folio_set_ordered(dst);
7949 return MIGRATEPAGE_SUCCESS;
7952 #define btrfs_migrate_folio NULL
7955 static void btrfs_invalidate_folio(struct folio *folio, size_t offset,
7958 struct btrfs_inode *inode = BTRFS_I(folio->mapping->host);
7959 struct btrfs_fs_info *fs_info = inode->root->fs_info;
7960 struct extent_io_tree *tree = &inode->io_tree;
7961 struct extent_state *cached_state = NULL;
7962 u64 page_start = folio_pos(folio);
7963 u64 page_end = page_start + folio_size(folio) - 1;
7965 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
7968 * We have folio locked so no new ordered extent can be created on this
7969 * page, nor bio can be submitted for this folio.
7971 * But already submitted bio can still be finished on this folio.
7972 * Furthermore, endio function won't skip folio which has Ordered
7973 * (Private2) already cleared, so it's possible for endio and
7974 * invalidate_folio to do the same ordered extent accounting twice
7977 * So here we wait for any submitted bios to finish, so that we won't
7978 * do double ordered extent accounting on the same folio.
7980 folio_wait_writeback(folio);
7981 wait_subpage_spinlock(&folio->page);
7984 * For subpage case, we have call sites like
7985 * btrfs_punch_hole_lock_range() which passes range not aligned to
7987 * If the range doesn't cover the full folio, we don't need to and
7988 * shouldn't clear page extent mapped, as folio->private can still
7989 * record subpage dirty bits for other part of the range.
7991 * For cases that invalidate the full folio even the range doesn't
7992 * cover the full folio, like invalidating the last folio, we're
7993 * still safe to wait for ordered extent to finish.
7995 if (!(offset == 0 && length == folio_size(folio))) {
7996 btrfs_release_folio(folio, GFP_NOFS);
8000 if (!inode_evicting)
8001 lock_extent(tree, page_start, page_end, &cached_state);
8004 while (cur < page_end) {
8005 struct btrfs_ordered_extent *ordered;
8008 u32 extra_flags = 0;
8010 ordered = btrfs_lookup_first_ordered_range(inode, cur,
8011 page_end + 1 - cur);
8013 range_end = page_end;
8015 * No ordered extent covering this range, we are safe
8016 * to delete all extent states in the range.
8018 extra_flags = EXTENT_CLEAR_ALL_BITS;
8021 if (ordered->file_offset > cur) {
8023 * There is a range between [cur, oe->file_offset) not
8024 * covered by any ordered extent.
8025 * We are safe to delete all extent states, and handle
8026 * the ordered extent in the next iteration.
8028 range_end = ordered->file_offset - 1;
8029 extra_flags = EXTENT_CLEAR_ALL_BITS;
8033 range_end = min(ordered->file_offset + ordered->num_bytes - 1,
8035 ASSERT(range_end + 1 - cur < U32_MAX);
8036 range_len = range_end + 1 - cur;
8037 if (!btrfs_page_test_ordered(fs_info, &folio->page, cur, range_len)) {
8039 * If Ordered (Private2) is cleared, it means endio has
8040 * already been executed for the range.
8041 * We can't delete the extent states as
8042 * btrfs_finish_ordered_io() may still use some of them.
8046 btrfs_page_clear_ordered(fs_info, &folio->page, cur, range_len);
8049 * IO on this page will never be started, so we need to account
8050 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8051 * here, must leave that up for the ordered extent completion.
8053 * This will also unlock the range for incoming
8054 * btrfs_finish_ordered_io().
8056 if (!inode_evicting)
8057 clear_extent_bit(tree, cur, range_end,
8059 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8060 EXTENT_DEFRAG, &cached_state);
8062 spin_lock_irq(&inode->ordered_tree.lock);
8063 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8064 ordered->truncated_len = min(ordered->truncated_len,
8065 cur - ordered->file_offset);
8066 spin_unlock_irq(&inode->ordered_tree.lock);
8069 * If the ordered extent has finished, we're safe to delete all
8070 * the extent states of the range, otherwise
8071 * btrfs_finish_ordered_io() will get executed by endio for
8072 * other pages, so we can't delete extent states.
8074 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8075 cur, range_end + 1 - cur)) {
8076 btrfs_finish_ordered_io(ordered);
8078 * The ordered extent has finished, now we're again
8079 * safe to delete all extent states of the range.
8081 extra_flags = EXTENT_CLEAR_ALL_BITS;
8085 btrfs_put_ordered_extent(ordered);
8087 * Qgroup reserved space handler
8088 * Sector(s) here will be either:
8090 * 1) Already written to disk or bio already finished
8091 * Then its QGROUP_RESERVED bit in io_tree is already cleared.
8092 * Qgroup will be handled by its qgroup_record then.
8093 * btrfs_qgroup_free_data() call will do nothing here.
8095 * 2) Not written to disk yet
8096 * Then btrfs_qgroup_free_data() call will clear the
8097 * QGROUP_RESERVED bit of its io_tree, and free the qgroup
8098 * reserved data space.
8099 * Since the IO will never happen for this page.
8101 btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur);
8102 if (!inode_evicting) {
8103 clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED |
8104 EXTENT_DELALLOC | EXTENT_UPTODATE |
8105 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG |
8106 extra_flags, &cached_state);
8108 cur = range_end + 1;
8111 * We have iterated through all ordered extents of the page, the page
8112 * should not have Ordered (Private2) anymore, or the above iteration
8113 * did something wrong.
8115 ASSERT(!folio_test_ordered(folio));
8116 btrfs_page_clear_checked(fs_info, &folio->page, folio_pos(folio), folio_size(folio));
8117 if (!inode_evicting)
8118 __btrfs_release_folio(folio, GFP_NOFS);
8119 clear_page_extent_mapped(&folio->page);
8123 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8124 * called from a page fault handler when a page is first dirtied. Hence we must
8125 * be careful to check for EOF conditions here. We set the page up correctly
8126 * for a written page which means we get ENOSPC checking when writing into
8127 * holes and correct delalloc and unwritten extent mapping on filesystems that
8128 * support these features.
8130 * We are not allowed to take the i_mutex here so we have to play games to
8131 * protect against truncate races as the page could now be beyond EOF. Because
8132 * truncate_setsize() writes the inode size before removing pages, once we have
8133 * the page lock we can determine safely if the page is beyond EOF. If it is not
8134 * beyond EOF, then the page is guaranteed safe against truncation until we
8137 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8139 struct page *page = vmf->page;
8140 struct inode *inode = file_inode(vmf->vma->vm_file);
8141 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8142 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8143 struct btrfs_ordered_extent *ordered;
8144 struct extent_state *cached_state = NULL;
8145 struct extent_changeset *data_reserved = NULL;
8146 unsigned long zero_start;
8156 reserved_space = PAGE_SIZE;
8158 sb_start_pagefault(inode->i_sb);
8159 page_start = page_offset(page);
8160 page_end = page_start + PAGE_SIZE - 1;
8164 * Reserving delalloc space after obtaining the page lock can lead to
8165 * deadlock. For example, if a dirty page is locked by this function
8166 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8167 * dirty page write out, then the btrfs_writepages() function could
8168 * end up waiting indefinitely to get a lock on the page currently
8169 * being processed by btrfs_page_mkwrite() function.
8171 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8172 page_start, reserved_space);
8174 ret2 = file_update_time(vmf->vma->vm_file);
8178 ret = vmf_error(ret2);
8184 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8186 down_read(&BTRFS_I(inode)->i_mmap_lock);
8188 size = i_size_read(inode);
8190 if ((page->mapping != inode->i_mapping) ||
8191 (page_start >= size)) {
8192 /* page got truncated out from underneath us */
8195 wait_on_page_writeback(page);
8197 lock_extent(io_tree, page_start, page_end, &cached_state);
8198 ret2 = set_page_extent_mapped(page);
8200 ret = vmf_error(ret2);
8201 unlock_extent(io_tree, page_start, page_end, &cached_state);
8206 * we can't set the delalloc bits if there are pending ordered
8207 * extents. Drop our locks and wait for them to finish
8209 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8212 unlock_extent(io_tree, page_start, page_end, &cached_state);
8214 up_read(&BTRFS_I(inode)->i_mmap_lock);
8215 btrfs_start_ordered_extent(ordered);
8216 btrfs_put_ordered_extent(ordered);
8220 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8221 reserved_space = round_up(size - page_start,
8222 fs_info->sectorsize);
8223 if (reserved_space < PAGE_SIZE) {
8224 end = page_start + reserved_space - 1;
8225 btrfs_delalloc_release_space(BTRFS_I(inode),
8226 data_reserved, page_start,
8227 PAGE_SIZE - reserved_space, true);
8232 * page_mkwrite gets called when the page is firstly dirtied after it's
8233 * faulted in, but write(2) could also dirty a page and set delalloc
8234 * bits, thus in this case for space account reason, we still need to
8235 * clear any delalloc bits within this page range since we have to
8236 * reserve data&meta space before lock_page() (see above comments).
8238 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8239 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8240 EXTENT_DEFRAG, &cached_state);
8242 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8245 unlock_extent(io_tree, page_start, page_end, &cached_state);
8246 ret = VM_FAULT_SIGBUS;
8250 /* page is wholly or partially inside EOF */
8251 if (page_start + PAGE_SIZE > size)
8252 zero_start = offset_in_page(size);
8254 zero_start = PAGE_SIZE;
8256 if (zero_start != PAGE_SIZE)
8257 memzero_page(page, zero_start, PAGE_SIZE - zero_start);
8259 btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE);
8260 btrfs_page_set_dirty(fs_info, page, page_start, end + 1 - page_start);
8261 btrfs_page_set_uptodate(fs_info, page, page_start, end + 1 - page_start);
8263 btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
8265 unlock_extent(io_tree, page_start, page_end, &cached_state);
8266 up_read(&BTRFS_I(inode)->i_mmap_lock);
8268 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8269 sb_end_pagefault(inode->i_sb);
8270 extent_changeset_free(data_reserved);
8271 return VM_FAULT_LOCKED;
8275 up_read(&BTRFS_I(inode)->i_mmap_lock);
8277 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8278 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8279 reserved_space, (ret != 0));
8281 sb_end_pagefault(inode->i_sb);
8282 extent_changeset_free(data_reserved);
8286 static int btrfs_truncate(struct btrfs_inode *inode, bool skip_writeback)
8288 struct btrfs_truncate_control control = {
8290 .ino = btrfs_ino(inode),
8291 .min_type = BTRFS_EXTENT_DATA_KEY,
8292 .clear_extent_range = true,
8294 struct btrfs_root *root = inode->root;
8295 struct btrfs_fs_info *fs_info = root->fs_info;
8296 struct btrfs_block_rsv *rsv;
8298 struct btrfs_trans_handle *trans;
8299 u64 mask = fs_info->sectorsize - 1;
8300 const u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8302 if (!skip_writeback) {
8303 ret = btrfs_wait_ordered_range(&inode->vfs_inode,
8304 inode->vfs_inode.i_size & (~mask),
8311 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8312 * things going on here:
8314 * 1) We need to reserve space to update our inode.
8316 * 2) We need to have something to cache all the space that is going to
8317 * be free'd up by the truncate operation, but also have some slack
8318 * space reserved in case it uses space during the truncate (thank you
8319 * very much snapshotting).
8321 * And we need these to be separate. The fact is we can use a lot of
8322 * space doing the truncate, and we have no earthly idea how much space
8323 * we will use, so we need the truncate reservation to be separate so it
8324 * doesn't end up using space reserved for updating the inode. We also
8325 * need to be able to stop the transaction and start a new one, which
8326 * means we need to be able to update the inode several times, and we
8327 * have no idea of knowing how many times that will be, so we can't just
8328 * reserve 1 item for the entirety of the operation, so that has to be
8329 * done separately as well.
8331 * So that leaves us with
8333 * 1) rsv - for the truncate reservation, which we will steal from the
8334 * transaction reservation.
8335 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8336 * updating the inode.
8338 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8341 rsv->size = min_size;
8342 rsv->failfast = true;
8345 * 1 for the truncate slack space
8346 * 1 for updating the inode.
8348 trans = btrfs_start_transaction(root, 2);
8349 if (IS_ERR(trans)) {
8350 ret = PTR_ERR(trans);
8354 /* Migrate the slack space for the truncate to our reserve */
8355 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8358 * We have reserved 2 metadata units when we started the transaction and
8359 * min_size matches 1 unit, so this should never fail, but if it does,
8360 * it's not critical we just fail truncation.
8363 btrfs_end_transaction(trans);
8367 trans->block_rsv = rsv;
8370 struct extent_state *cached_state = NULL;
8371 const u64 new_size = inode->vfs_inode.i_size;
8372 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
8374 control.new_size = new_size;
8375 lock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8377 * We want to drop from the next block forward in case this new
8378 * size is not block aligned since we will be keeping the last
8379 * block of the extent just the way it is.
8381 btrfs_drop_extent_map_range(inode,
8382 ALIGN(new_size, fs_info->sectorsize),
8385 ret = btrfs_truncate_inode_items(trans, root, &control);
8387 inode_sub_bytes(&inode->vfs_inode, control.sub_bytes);
8388 btrfs_inode_safe_disk_i_size_write(inode, control.last_size);
8390 unlock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8392 trans->block_rsv = &fs_info->trans_block_rsv;
8393 if (ret != -ENOSPC && ret != -EAGAIN)
8396 ret = btrfs_update_inode(trans, root, inode);
8400 btrfs_end_transaction(trans);
8401 btrfs_btree_balance_dirty(fs_info);
8403 trans = btrfs_start_transaction(root, 2);
8404 if (IS_ERR(trans)) {
8405 ret = PTR_ERR(trans);
8410 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8411 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8412 rsv, min_size, false);
8414 * We have reserved 2 metadata units when we started the
8415 * transaction and min_size matches 1 unit, so this should never
8416 * fail, but if it does, it's not critical we just fail truncation.
8421 trans->block_rsv = rsv;
8425 * We can't call btrfs_truncate_block inside a trans handle as we could
8426 * deadlock with freeze, if we got BTRFS_NEED_TRUNCATE_BLOCK then we
8427 * know we've truncated everything except the last little bit, and can
8428 * do btrfs_truncate_block and then update the disk_i_size.
8430 if (ret == BTRFS_NEED_TRUNCATE_BLOCK) {
8431 btrfs_end_transaction(trans);
8432 btrfs_btree_balance_dirty(fs_info);
8434 ret = btrfs_truncate_block(inode, inode->vfs_inode.i_size, 0, 0);
8437 trans = btrfs_start_transaction(root, 1);
8438 if (IS_ERR(trans)) {
8439 ret = PTR_ERR(trans);
8442 btrfs_inode_safe_disk_i_size_write(inode, 0);
8448 trans->block_rsv = &fs_info->trans_block_rsv;
8449 ret2 = btrfs_update_inode(trans, root, inode);
8453 ret2 = btrfs_end_transaction(trans);
8456 btrfs_btree_balance_dirty(fs_info);
8459 btrfs_free_block_rsv(fs_info, rsv);
8461 * So if we truncate and then write and fsync we normally would just
8462 * write the extents that changed, which is a problem if we need to
8463 * first truncate that entire inode. So set this flag so we write out
8464 * all of the extents in the inode to the sync log so we're completely
8467 * If no extents were dropped or trimmed we don't need to force the next
8468 * fsync to truncate all the inode's items from the log and re-log them
8469 * all. This means the truncate operation did not change the file size,
8470 * or changed it to a smaller size but there was only an implicit hole
8471 * between the old i_size and the new i_size, and there were no prealloc
8472 * extents beyond i_size to drop.
8474 if (control.extents_found > 0)
8475 btrfs_set_inode_full_sync(inode);
8480 struct inode *btrfs_new_subvol_inode(struct mnt_idmap *idmap,
8483 struct inode *inode;
8485 inode = new_inode(dir->i_sb);
8488 * Subvolumes don't inherit the sgid bit or the parent's gid if
8489 * the parent's sgid bit is set. This is probably a bug.
8491 inode_init_owner(idmap, inode, NULL,
8492 S_IFDIR | (~current_umask() & S_IRWXUGO));
8493 inode->i_op = &btrfs_dir_inode_operations;
8494 inode->i_fop = &btrfs_dir_file_operations;
8499 struct inode *btrfs_alloc_inode(struct super_block *sb)
8501 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8502 struct btrfs_inode *ei;
8503 struct inode *inode;
8505 ei = alloc_inode_sb(sb, btrfs_inode_cachep, GFP_KERNEL);
8512 ei->last_sub_trans = 0;
8513 ei->logged_trans = 0;
8514 ei->delalloc_bytes = 0;
8515 ei->new_delalloc_bytes = 0;
8516 ei->defrag_bytes = 0;
8517 ei->disk_i_size = 0;
8521 ei->index_cnt = (u64)-1;
8523 ei->last_unlink_trans = 0;
8524 ei->last_reflink_trans = 0;
8525 ei->last_log_commit = 0;
8527 spin_lock_init(&ei->lock);
8528 ei->outstanding_extents = 0;
8529 if (sb->s_magic != BTRFS_TEST_MAGIC)
8530 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8531 BTRFS_BLOCK_RSV_DELALLOC);
8532 ei->runtime_flags = 0;
8533 ei->prop_compress = BTRFS_COMPRESS_NONE;
8534 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8536 ei->delayed_node = NULL;
8538 ei->i_otime.tv_sec = 0;
8539 ei->i_otime.tv_nsec = 0;
8541 inode = &ei->vfs_inode;
8542 extent_map_tree_init(&ei->extent_tree);
8543 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO);
8544 ei->io_tree.inode = ei;
8545 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8546 IO_TREE_INODE_FILE_EXTENT);
8547 mutex_init(&ei->log_mutex);
8548 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8549 INIT_LIST_HEAD(&ei->delalloc_inodes);
8550 INIT_LIST_HEAD(&ei->delayed_iput);
8551 RB_CLEAR_NODE(&ei->rb_node);
8552 init_rwsem(&ei->i_mmap_lock);
8557 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8558 void btrfs_test_destroy_inode(struct inode *inode)
8560 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
8561 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8565 void btrfs_free_inode(struct inode *inode)
8567 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8570 void btrfs_destroy_inode(struct inode *vfs_inode)
8572 struct btrfs_ordered_extent *ordered;
8573 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8574 struct btrfs_root *root = inode->root;
8575 bool freespace_inode;
8577 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8578 WARN_ON(vfs_inode->i_data.nrpages);
8579 WARN_ON(inode->block_rsv.reserved);
8580 WARN_ON(inode->block_rsv.size);
8581 WARN_ON(inode->outstanding_extents);
8582 if (!S_ISDIR(vfs_inode->i_mode)) {
8583 WARN_ON(inode->delalloc_bytes);
8584 WARN_ON(inode->new_delalloc_bytes);
8586 WARN_ON(inode->csum_bytes);
8587 WARN_ON(inode->defrag_bytes);
8590 * This can happen where we create an inode, but somebody else also
8591 * created the same inode and we need to destroy the one we already
8598 * If this is a free space inode do not take the ordered extents lockdep
8601 freespace_inode = btrfs_is_free_space_inode(inode);
8604 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8608 btrfs_err(root->fs_info,
8609 "found ordered extent %llu %llu on inode cleanup",
8610 ordered->file_offset, ordered->num_bytes);
8612 if (!freespace_inode)
8613 btrfs_lockdep_acquire(root->fs_info, btrfs_ordered_extent);
8615 btrfs_remove_ordered_extent(inode, ordered);
8616 btrfs_put_ordered_extent(ordered);
8617 btrfs_put_ordered_extent(ordered);
8620 btrfs_qgroup_check_reserved_leak(inode);
8621 inode_tree_del(inode);
8622 btrfs_drop_extent_map_range(inode, 0, (u64)-1, false);
8623 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
8624 btrfs_put_root(inode->root);
8627 int btrfs_drop_inode(struct inode *inode)
8629 struct btrfs_root *root = BTRFS_I(inode)->root;
8634 /* the snap/subvol tree is on deleting */
8635 if (btrfs_root_refs(&root->root_item) == 0)
8638 return generic_drop_inode(inode);
8641 static void init_once(void *foo)
8643 struct btrfs_inode *ei = foo;
8645 inode_init_once(&ei->vfs_inode);
8648 void __cold btrfs_destroy_cachep(void)
8651 * Make sure all delayed rcu free inodes are flushed before we
8655 bioset_exit(&btrfs_dio_bioset);
8656 kmem_cache_destroy(btrfs_inode_cachep);
8659 int __init btrfs_init_cachep(void)
8661 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8662 sizeof(struct btrfs_inode), 0,
8663 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
8665 if (!btrfs_inode_cachep)
8668 if (bioset_init(&btrfs_dio_bioset, BIO_POOL_SIZE,
8669 offsetof(struct btrfs_dio_private, bbio.bio),
8675 btrfs_destroy_cachep();
8679 static int btrfs_getattr(struct mnt_idmap *idmap,
8680 const struct path *path, struct kstat *stat,
8681 u32 request_mask, unsigned int flags)
8685 struct inode *inode = d_inode(path->dentry);
8686 u32 blocksize = inode->i_sb->s_blocksize;
8687 u32 bi_flags = BTRFS_I(inode)->flags;
8688 u32 bi_ro_flags = BTRFS_I(inode)->ro_flags;
8690 stat->result_mask |= STATX_BTIME;
8691 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
8692 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
8693 if (bi_flags & BTRFS_INODE_APPEND)
8694 stat->attributes |= STATX_ATTR_APPEND;
8695 if (bi_flags & BTRFS_INODE_COMPRESS)
8696 stat->attributes |= STATX_ATTR_COMPRESSED;
8697 if (bi_flags & BTRFS_INODE_IMMUTABLE)
8698 stat->attributes |= STATX_ATTR_IMMUTABLE;
8699 if (bi_flags & BTRFS_INODE_NODUMP)
8700 stat->attributes |= STATX_ATTR_NODUMP;
8701 if (bi_ro_flags & BTRFS_INODE_RO_VERITY)
8702 stat->attributes |= STATX_ATTR_VERITY;
8704 stat->attributes_mask |= (STATX_ATTR_APPEND |
8705 STATX_ATTR_COMPRESSED |
8706 STATX_ATTR_IMMUTABLE |
8709 generic_fillattr(idmap, inode, stat);
8710 stat->dev = BTRFS_I(inode)->root->anon_dev;
8712 spin_lock(&BTRFS_I(inode)->lock);
8713 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
8714 inode_bytes = inode_get_bytes(inode);
8715 spin_unlock(&BTRFS_I(inode)->lock);
8716 stat->blocks = (ALIGN(inode_bytes, blocksize) +
8717 ALIGN(delalloc_bytes, blocksize)) >> SECTOR_SHIFT;
8721 static int btrfs_rename_exchange(struct inode *old_dir,
8722 struct dentry *old_dentry,
8723 struct inode *new_dir,
8724 struct dentry *new_dentry)
8726 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
8727 struct btrfs_trans_handle *trans;
8728 unsigned int trans_num_items;
8729 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8730 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8731 struct inode *new_inode = new_dentry->d_inode;
8732 struct inode *old_inode = old_dentry->d_inode;
8733 struct timespec64 ctime = current_time(old_inode);
8734 struct btrfs_rename_ctx old_rename_ctx;
8735 struct btrfs_rename_ctx new_rename_ctx;
8736 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
8737 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
8742 bool need_abort = false;
8743 struct fscrypt_name old_fname, new_fname;
8744 struct fscrypt_str *old_name, *new_name;
8747 * For non-subvolumes allow exchange only within one subvolume, in the
8748 * same inode namespace. Two subvolumes (represented as directory) can
8749 * be exchanged as they're a logical link and have a fixed inode number.
8752 (old_ino != BTRFS_FIRST_FREE_OBJECTID ||
8753 new_ino != BTRFS_FIRST_FREE_OBJECTID))
8756 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
8760 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
8762 fscrypt_free_filename(&old_fname);
8766 old_name = &old_fname.disk_name;
8767 new_name = &new_fname.disk_name;
8769 /* close the race window with snapshot create/destroy ioctl */
8770 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
8771 new_ino == BTRFS_FIRST_FREE_OBJECTID)
8772 down_read(&fs_info->subvol_sem);
8776 * 1 to remove old dir item
8777 * 1 to remove old dir index
8778 * 1 to add new dir item
8779 * 1 to add new dir index
8780 * 1 to update parent inode
8782 * If the parents are the same, we only need to account for one
8784 trans_num_items = (old_dir == new_dir ? 9 : 10);
8785 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8787 * 1 to remove old root ref
8788 * 1 to remove old root backref
8789 * 1 to add new root ref
8790 * 1 to add new root backref
8792 trans_num_items += 4;
8795 * 1 to update inode item
8796 * 1 to remove old inode ref
8797 * 1 to add new inode ref
8799 trans_num_items += 3;
8801 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
8802 trans_num_items += 4;
8804 trans_num_items += 3;
8805 trans = btrfs_start_transaction(root, trans_num_items);
8806 if (IS_ERR(trans)) {
8807 ret = PTR_ERR(trans);
8812 ret = btrfs_record_root_in_trans(trans, dest);
8818 * We need to find a free sequence number both in the source and
8819 * in the destination directory for the exchange.
8821 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
8824 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
8828 BTRFS_I(old_inode)->dir_index = 0ULL;
8829 BTRFS_I(new_inode)->dir_index = 0ULL;
8831 /* Reference for the source. */
8832 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8833 /* force full log commit if subvolume involved. */
8834 btrfs_set_log_full_commit(trans);
8836 ret = btrfs_insert_inode_ref(trans, dest, new_name, old_ino,
8837 btrfs_ino(BTRFS_I(new_dir)),
8844 /* And now for the dest. */
8845 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8846 /* force full log commit if subvolume involved. */
8847 btrfs_set_log_full_commit(trans);
8849 ret = btrfs_insert_inode_ref(trans, root, old_name, new_ino,
8850 btrfs_ino(BTRFS_I(old_dir)),
8854 btrfs_abort_transaction(trans, ret);
8859 /* Update inode version and ctime/mtime. */
8860 inode_inc_iversion(old_dir);
8861 inode_inc_iversion(new_dir);
8862 inode_inc_iversion(old_inode);
8863 inode_inc_iversion(new_inode);
8864 old_dir->i_mtime = ctime;
8865 old_dir->i_ctime = ctime;
8866 new_dir->i_mtime = ctime;
8867 new_dir->i_ctime = ctime;
8868 old_inode->i_ctime = ctime;
8869 new_inode->i_ctime = ctime;
8871 if (old_dentry->d_parent != new_dentry->d_parent) {
8872 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
8873 BTRFS_I(old_inode), true);
8874 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
8875 BTRFS_I(new_inode), true);
8878 /* src is a subvolume */
8879 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8880 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
8881 } else { /* src is an inode */
8882 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
8883 BTRFS_I(old_dentry->d_inode),
8884 old_name, &old_rename_ctx);
8886 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
8889 btrfs_abort_transaction(trans, ret);
8893 /* dest is a subvolume */
8894 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8895 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
8896 } else { /* dest is an inode */
8897 ret = __btrfs_unlink_inode(trans, BTRFS_I(new_dir),
8898 BTRFS_I(new_dentry->d_inode),
8899 new_name, &new_rename_ctx);
8901 ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode));
8904 btrfs_abort_transaction(trans, ret);
8908 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
8909 new_name, 0, old_idx);
8911 btrfs_abort_transaction(trans, ret);
8915 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
8916 old_name, 0, new_idx);
8918 btrfs_abort_transaction(trans, ret);
8922 if (old_inode->i_nlink == 1)
8923 BTRFS_I(old_inode)->dir_index = old_idx;
8924 if (new_inode->i_nlink == 1)
8925 BTRFS_I(new_inode)->dir_index = new_idx;
8928 * Now pin the logs of the roots. We do it to ensure that no other task
8929 * can sync the logs while we are in progress with the rename, because
8930 * that could result in an inconsistency in case any of the inodes that
8931 * are part of this rename operation were logged before.
8933 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8934 btrfs_pin_log_trans(root);
8935 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8936 btrfs_pin_log_trans(dest);
8938 /* Do the log updates for all inodes. */
8939 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8940 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
8941 old_rename_ctx.index, new_dentry->d_parent);
8942 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8943 btrfs_log_new_name(trans, new_dentry, BTRFS_I(new_dir),
8944 new_rename_ctx.index, old_dentry->d_parent);
8946 /* Now unpin the logs. */
8947 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8948 btrfs_end_log_trans(root);
8949 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8950 btrfs_end_log_trans(dest);
8952 ret2 = btrfs_end_transaction(trans);
8953 ret = ret ? ret : ret2;
8955 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
8956 old_ino == BTRFS_FIRST_FREE_OBJECTID)
8957 up_read(&fs_info->subvol_sem);
8959 fscrypt_free_filename(&new_fname);
8960 fscrypt_free_filename(&old_fname);
8964 static struct inode *new_whiteout_inode(struct mnt_idmap *idmap,
8967 struct inode *inode;
8969 inode = new_inode(dir->i_sb);
8971 inode_init_owner(idmap, inode, dir,
8972 S_IFCHR | WHITEOUT_MODE);
8973 inode->i_op = &btrfs_special_inode_operations;
8974 init_special_inode(inode, inode->i_mode, WHITEOUT_DEV);
8979 static int btrfs_rename(struct mnt_idmap *idmap,
8980 struct inode *old_dir, struct dentry *old_dentry,
8981 struct inode *new_dir, struct dentry *new_dentry,
8984 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
8985 struct btrfs_new_inode_args whiteout_args = {
8987 .dentry = old_dentry,
8989 struct btrfs_trans_handle *trans;
8990 unsigned int trans_num_items;
8991 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8992 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8993 struct inode *new_inode = d_inode(new_dentry);
8994 struct inode *old_inode = d_inode(old_dentry);
8995 struct btrfs_rename_ctx rename_ctx;
8999 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9000 struct fscrypt_name old_fname, new_fname;
9002 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9005 /* we only allow rename subvolume link between subvolumes */
9006 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9009 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9010 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9013 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9014 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9017 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
9021 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
9023 fscrypt_free_filename(&old_fname);
9027 /* check for collisions, even if the name isn't there */
9028 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, &new_fname.disk_name);
9030 if (ret == -EEXIST) {
9032 * eexist without a new_inode */
9033 if (WARN_ON(!new_inode)) {
9034 goto out_fscrypt_names;
9037 /* maybe -EOVERFLOW */
9038 goto out_fscrypt_names;
9044 * we're using rename to replace one file with another. Start IO on it
9045 * now so we don't add too much work to the end of the transaction
9047 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9048 filemap_flush(old_inode->i_mapping);
9050 if (flags & RENAME_WHITEOUT) {
9051 whiteout_args.inode = new_whiteout_inode(idmap, old_dir);
9052 if (!whiteout_args.inode) {
9054 goto out_fscrypt_names;
9056 ret = btrfs_new_inode_prepare(&whiteout_args, &trans_num_items);
9058 goto out_whiteout_inode;
9060 /* 1 to update the old parent inode. */
9061 trans_num_items = 1;
9064 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9065 /* Close the race window with snapshot create/destroy ioctl */
9066 down_read(&fs_info->subvol_sem);
9068 * 1 to remove old root ref
9069 * 1 to remove old root backref
9070 * 1 to add new root ref
9071 * 1 to add new root backref
9073 trans_num_items += 4;
9077 * 1 to remove old inode ref
9078 * 1 to add new inode ref
9080 trans_num_items += 3;
9083 * 1 to remove old dir item
9084 * 1 to remove old dir index
9085 * 1 to add new dir item
9086 * 1 to add new dir index
9088 trans_num_items += 4;
9089 /* 1 to update new parent inode if it's not the same as the old parent */
9090 if (new_dir != old_dir)
9095 * 1 to remove inode ref
9096 * 1 to remove dir item
9097 * 1 to remove dir index
9098 * 1 to possibly add orphan item
9100 trans_num_items += 5;
9102 trans = btrfs_start_transaction(root, trans_num_items);
9103 if (IS_ERR(trans)) {
9104 ret = PTR_ERR(trans);
9109 ret = btrfs_record_root_in_trans(trans, dest);
9114 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9118 BTRFS_I(old_inode)->dir_index = 0ULL;
9119 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9120 /* force full log commit if subvolume involved. */
9121 btrfs_set_log_full_commit(trans);
9123 ret = btrfs_insert_inode_ref(trans, dest, &new_fname.disk_name,
9124 old_ino, btrfs_ino(BTRFS_I(new_dir)),
9130 inode_inc_iversion(old_dir);
9131 inode_inc_iversion(new_dir);
9132 inode_inc_iversion(old_inode);
9133 old_dir->i_mtime = current_time(old_dir);
9134 old_dir->i_ctime = old_dir->i_mtime;
9135 new_dir->i_mtime = old_dir->i_mtime;
9136 new_dir->i_ctime = old_dir->i_mtime;
9137 old_inode->i_ctime = old_dir->i_mtime;
9139 if (old_dentry->d_parent != new_dentry->d_parent)
9140 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9141 BTRFS_I(old_inode), true);
9143 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9144 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
9146 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9147 BTRFS_I(d_inode(old_dentry)),
9148 &old_fname.disk_name, &rename_ctx);
9150 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9153 btrfs_abort_transaction(trans, ret);
9158 inode_inc_iversion(new_inode);
9159 new_inode->i_ctime = current_time(new_inode);
9160 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9161 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9162 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
9163 BUG_ON(new_inode->i_nlink == 0);
9165 ret = btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9166 BTRFS_I(d_inode(new_dentry)),
9167 &new_fname.disk_name);
9169 if (!ret && new_inode->i_nlink == 0)
9170 ret = btrfs_orphan_add(trans,
9171 BTRFS_I(d_inode(new_dentry)));
9173 btrfs_abort_transaction(trans, ret);
9178 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9179 &new_fname.disk_name, 0, index);
9181 btrfs_abort_transaction(trans, ret);
9185 if (old_inode->i_nlink == 1)
9186 BTRFS_I(old_inode)->dir_index = index;
9188 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9189 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9190 rename_ctx.index, new_dentry->d_parent);
9192 if (flags & RENAME_WHITEOUT) {
9193 ret = btrfs_create_new_inode(trans, &whiteout_args);
9195 btrfs_abort_transaction(trans, ret);
9198 unlock_new_inode(whiteout_args.inode);
9199 iput(whiteout_args.inode);
9200 whiteout_args.inode = NULL;
9204 ret2 = btrfs_end_transaction(trans);
9205 ret = ret ? ret : ret2;
9207 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9208 up_read(&fs_info->subvol_sem);
9209 if (flags & RENAME_WHITEOUT)
9210 btrfs_new_inode_args_destroy(&whiteout_args);
9212 if (flags & RENAME_WHITEOUT)
9213 iput(whiteout_args.inode);
9215 fscrypt_free_filename(&old_fname);
9216 fscrypt_free_filename(&new_fname);
9220 static int btrfs_rename2(struct mnt_idmap *idmap, struct inode *old_dir,
9221 struct dentry *old_dentry, struct inode *new_dir,
9222 struct dentry *new_dentry, unsigned int flags)
9226 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9229 if (flags & RENAME_EXCHANGE)
9230 ret = btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9233 ret = btrfs_rename(idmap, old_dir, old_dentry, new_dir,
9236 btrfs_btree_balance_dirty(BTRFS_I(new_dir)->root->fs_info);
9241 struct btrfs_delalloc_work {
9242 struct inode *inode;
9243 struct completion completion;
9244 struct list_head list;
9245 struct btrfs_work work;
9248 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9250 struct btrfs_delalloc_work *delalloc_work;
9251 struct inode *inode;
9253 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9255 inode = delalloc_work->inode;
9256 filemap_flush(inode->i_mapping);
9257 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9258 &BTRFS_I(inode)->runtime_flags))
9259 filemap_flush(inode->i_mapping);
9262 complete(&delalloc_work->completion);
9265 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9267 struct btrfs_delalloc_work *work;
9269 work = kmalloc(sizeof(*work), GFP_NOFS);
9273 init_completion(&work->completion);
9274 INIT_LIST_HEAD(&work->list);
9275 work->inode = inode;
9276 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9282 * some fairly slow code that needs optimization. This walks the list
9283 * of all the inodes with pending delalloc and forces them to disk.
9285 static int start_delalloc_inodes(struct btrfs_root *root,
9286 struct writeback_control *wbc, bool snapshot,
9287 bool in_reclaim_context)
9289 struct btrfs_inode *binode;
9290 struct inode *inode;
9291 struct btrfs_delalloc_work *work, *next;
9292 struct list_head works;
9293 struct list_head splice;
9295 bool full_flush = wbc->nr_to_write == LONG_MAX;
9297 INIT_LIST_HEAD(&works);
9298 INIT_LIST_HEAD(&splice);
9300 mutex_lock(&root->delalloc_mutex);
9301 spin_lock(&root->delalloc_lock);
9302 list_splice_init(&root->delalloc_inodes, &splice);
9303 while (!list_empty(&splice)) {
9304 binode = list_entry(splice.next, struct btrfs_inode,
9307 list_move_tail(&binode->delalloc_inodes,
9308 &root->delalloc_inodes);
9310 if (in_reclaim_context &&
9311 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9314 inode = igrab(&binode->vfs_inode);
9316 cond_resched_lock(&root->delalloc_lock);
9319 spin_unlock(&root->delalloc_lock);
9322 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9323 &binode->runtime_flags);
9325 work = btrfs_alloc_delalloc_work(inode);
9331 list_add_tail(&work->list, &works);
9332 btrfs_queue_work(root->fs_info->flush_workers,
9335 ret = filemap_fdatawrite_wbc(inode->i_mapping, wbc);
9336 btrfs_add_delayed_iput(BTRFS_I(inode));
9337 if (ret || wbc->nr_to_write <= 0)
9341 spin_lock(&root->delalloc_lock);
9343 spin_unlock(&root->delalloc_lock);
9346 list_for_each_entry_safe(work, next, &works, list) {
9347 list_del_init(&work->list);
9348 wait_for_completion(&work->completion);
9352 if (!list_empty(&splice)) {
9353 spin_lock(&root->delalloc_lock);
9354 list_splice_tail(&splice, &root->delalloc_inodes);
9355 spin_unlock(&root->delalloc_lock);
9357 mutex_unlock(&root->delalloc_mutex);
9361 int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9363 struct writeback_control wbc = {
9364 .nr_to_write = LONG_MAX,
9365 .sync_mode = WB_SYNC_NONE,
9367 .range_end = LLONG_MAX,
9369 struct btrfs_fs_info *fs_info = root->fs_info;
9371 if (BTRFS_FS_ERROR(fs_info))
9374 return start_delalloc_inodes(root, &wbc, true, in_reclaim_context);
9377 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9378 bool in_reclaim_context)
9380 struct writeback_control wbc = {
9382 .sync_mode = WB_SYNC_NONE,
9384 .range_end = LLONG_MAX,
9386 struct btrfs_root *root;
9387 struct list_head splice;
9390 if (BTRFS_FS_ERROR(fs_info))
9393 INIT_LIST_HEAD(&splice);
9395 mutex_lock(&fs_info->delalloc_root_mutex);
9396 spin_lock(&fs_info->delalloc_root_lock);
9397 list_splice_init(&fs_info->delalloc_roots, &splice);
9398 while (!list_empty(&splice)) {
9400 * Reset nr_to_write here so we know that we're doing a full
9404 wbc.nr_to_write = LONG_MAX;
9406 root = list_first_entry(&splice, struct btrfs_root,
9408 root = btrfs_grab_root(root);
9410 list_move_tail(&root->delalloc_root,
9411 &fs_info->delalloc_roots);
9412 spin_unlock(&fs_info->delalloc_root_lock);
9414 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9415 btrfs_put_root(root);
9416 if (ret < 0 || wbc.nr_to_write <= 0)
9418 spin_lock(&fs_info->delalloc_root_lock);
9420 spin_unlock(&fs_info->delalloc_root_lock);
9424 if (!list_empty(&splice)) {
9425 spin_lock(&fs_info->delalloc_root_lock);
9426 list_splice_tail(&splice, &fs_info->delalloc_roots);
9427 spin_unlock(&fs_info->delalloc_root_lock);
9429 mutex_unlock(&fs_info->delalloc_root_mutex);
9433 static int btrfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
9434 struct dentry *dentry, const char *symname)
9436 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9437 struct btrfs_trans_handle *trans;
9438 struct btrfs_root *root = BTRFS_I(dir)->root;
9439 struct btrfs_path *path;
9440 struct btrfs_key key;
9441 struct inode *inode;
9442 struct btrfs_new_inode_args new_inode_args = {
9446 unsigned int trans_num_items;
9451 struct btrfs_file_extent_item *ei;
9452 struct extent_buffer *leaf;
9454 name_len = strlen(symname);
9455 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9456 return -ENAMETOOLONG;
9458 inode = new_inode(dir->i_sb);
9461 inode_init_owner(idmap, inode, dir, S_IFLNK | S_IRWXUGO);
9462 inode->i_op = &btrfs_symlink_inode_operations;
9463 inode_nohighmem(inode);
9464 inode->i_mapping->a_ops = &btrfs_aops;
9465 btrfs_i_size_write(BTRFS_I(inode), name_len);
9466 inode_set_bytes(inode, name_len);
9468 new_inode_args.inode = inode;
9469 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9472 /* 1 additional item for the inline extent */
9475 trans = btrfs_start_transaction(root, trans_num_items);
9476 if (IS_ERR(trans)) {
9477 err = PTR_ERR(trans);
9478 goto out_new_inode_args;
9481 err = btrfs_create_new_inode(trans, &new_inode_args);
9485 path = btrfs_alloc_path();
9488 btrfs_abort_transaction(trans, err);
9489 discard_new_inode(inode);
9493 key.objectid = btrfs_ino(BTRFS_I(inode));
9495 key.type = BTRFS_EXTENT_DATA_KEY;
9496 datasize = btrfs_file_extent_calc_inline_size(name_len);
9497 err = btrfs_insert_empty_item(trans, root, path, &key,
9500 btrfs_abort_transaction(trans, err);
9501 btrfs_free_path(path);
9502 discard_new_inode(inode);
9506 leaf = path->nodes[0];
9507 ei = btrfs_item_ptr(leaf, path->slots[0],
9508 struct btrfs_file_extent_item);
9509 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9510 btrfs_set_file_extent_type(leaf, ei,
9511 BTRFS_FILE_EXTENT_INLINE);
9512 btrfs_set_file_extent_encryption(leaf, ei, 0);
9513 btrfs_set_file_extent_compression(leaf, ei, 0);
9514 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9515 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9517 ptr = btrfs_file_extent_inline_start(ei);
9518 write_extent_buffer(leaf, symname, ptr, name_len);
9519 btrfs_mark_buffer_dirty(leaf);
9520 btrfs_free_path(path);
9522 d_instantiate_new(dentry, inode);
9525 btrfs_end_transaction(trans);
9526 btrfs_btree_balance_dirty(fs_info);
9528 btrfs_new_inode_args_destroy(&new_inode_args);
9535 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9536 struct btrfs_trans_handle *trans_in,
9537 struct btrfs_inode *inode,
9538 struct btrfs_key *ins,
9541 struct btrfs_file_extent_item stack_fi;
9542 struct btrfs_replace_extent_info extent_info;
9543 struct btrfs_trans_handle *trans = trans_in;
9544 struct btrfs_path *path;
9545 u64 start = ins->objectid;
9546 u64 len = ins->offset;
9547 int qgroup_released;
9550 memset(&stack_fi, 0, sizeof(stack_fi));
9552 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9553 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9554 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9555 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9556 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9557 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9558 /* Encryption and other encoding is reserved and all 0 */
9560 qgroup_released = btrfs_qgroup_release_data(inode, file_offset, len);
9561 if (qgroup_released < 0)
9562 return ERR_PTR(qgroup_released);
9565 ret = insert_reserved_file_extent(trans, inode,
9566 file_offset, &stack_fi,
9567 true, qgroup_released);
9573 extent_info.disk_offset = start;
9574 extent_info.disk_len = len;
9575 extent_info.data_offset = 0;
9576 extent_info.data_len = len;
9577 extent_info.file_offset = file_offset;
9578 extent_info.extent_buf = (char *)&stack_fi;
9579 extent_info.is_new_extent = true;
9580 extent_info.update_times = true;
9581 extent_info.qgroup_reserved = qgroup_released;
9582 extent_info.insertions = 0;
9584 path = btrfs_alloc_path();
9590 ret = btrfs_replace_file_extents(inode, path, file_offset,
9591 file_offset + len - 1, &extent_info,
9593 btrfs_free_path(path);
9600 * We have released qgroup data range at the beginning of the function,
9601 * and normally qgroup_released bytes will be freed when committing
9603 * But if we error out early, we have to free what we have released
9604 * or we leak qgroup data reservation.
9606 btrfs_qgroup_free_refroot(inode->root->fs_info,
9607 inode->root->root_key.objectid, qgroup_released,
9608 BTRFS_QGROUP_RSV_DATA);
9609 return ERR_PTR(ret);
9612 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9613 u64 start, u64 num_bytes, u64 min_size,
9614 loff_t actual_len, u64 *alloc_hint,
9615 struct btrfs_trans_handle *trans)
9617 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9618 struct extent_map *em;
9619 struct btrfs_root *root = BTRFS_I(inode)->root;
9620 struct btrfs_key ins;
9621 u64 cur_offset = start;
9622 u64 clear_offset = start;
9625 u64 last_alloc = (u64)-1;
9627 bool own_trans = true;
9628 u64 end = start + num_bytes - 1;
9632 while (num_bytes > 0) {
9633 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9634 cur_bytes = max(cur_bytes, min_size);
9636 * If we are severely fragmented we could end up with really
9637 * small allocations, so if the allocator is returning small
9638 * chunks lets make its job easier by only searching for those
9641 cur_bytes = min(cur_bytes, last_alloc);
9642 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9643 min_size, 0, *alloc_hint, &ins, 1, 0);
9648 * We've reserved this space, and thus converted it from
9649 * ->bytes_may_use to ->bytes_reserved. Any error that happens
9650 * from here on out we will only need to clear our reservation
9651 * for the remaining unreserved area, so advance our
9652 * clear_offset by our extent size.
9654 clear_offset += ins.offset;
9656 last_alloc = ins.offset;
9657 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
9660 * Now that we inserted the prealloc extent we can finally
9661 * decrement the number of reservations in the block group.
9662 * If we did it before, we could race with relocation and have
9663 * relocation miss the reserved extent, making it fail later.
9665 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
9666 if (IS_ERR(trans)) {
9667 ret = PTR_ERR(trans);
9668 btrfs_free_reserved_extent(fs_info, ins.objectid,
9673 em = alloc_extent_map();
9675 btrfs_drop_extent_map_range(BTRFS_I(inode), cur_offset,
9676 cur_offset + ins.offset - 1, false);
9677 btrfs_set_inode_full_sync(BTRFS_I(inode));
9681 em->start = cur_offset;
9682 em->orig_start = cur_offset;
9683 em->len = ins.offset;
9684 em->block_start = ins.objectid;
9685 em->block_len = ins.offset;
9686 em->orig_block_len = ins.offset;
9687 em->ram_bytes = ins.offset;
9688 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9689 em->generation = trans->transid;
9691 ret = btrfs_replace_extent_map_range(BTRFS_I(inode), em, true);
9692 free_extent_map(em);
9694 num_bytes -= ins.offset;
9695 cur_offset += ins.offset;
9696 *alloc_hint = ins.objectid + ins.offset;
9698 inode_inc_iversion(inode);
9699 inode->i_ctime = current_time(inode);
9700 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9701 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9702 (actual_len > inode->i_size) &&
9703 (cur_offset > inode->i_size)) {
9704 if (cur_offset > actual_len)
9705 i_size = actual_len;
9707 i_size = cur_offset;
9708 i_size_write(inode, i_size);
9709 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
9712 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
9715 btrfs_abort_transaction(trans, ret);
9717 btrfs_end_transaction(trans);
9722 btrfs_end_transaction(trans);
9726 if (clear_offset < end)
9727 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
9728 end - clear_offset + 1);
9732 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9733 u64 start, u64 num_bytes, u64 min_size,
9734 loff_t actual_len, u64 *alloc_hint)
9736 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9737 min_size, actual_len, alloc_hint,
9741 int btrfs_prealloc_file_range_trans(struct inode *inode,
9742 struct btrfs_trans_handle *trans, int mode,
9743 u64 start, u64 num_bytes, u64 min_size,
9744 loff_t actual_len, u64 *alloc_hint)
9746 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9747 min_size, actual_len, alloc_hint, trans);
9750 static int btrfs_permission(struct mnt_idmap *idmap,
9751 struct inode *inode, int mask)
9753 struct btrfs_root *root = BTRFS_I(inode)->root;
9754 umode_t mode = inode->i_mode;
9756 if (mask & MAY_WRITE &&
9757 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9758 if (btrfs_root_readonly(root))
9760 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9763 return generic_permission(idmap, inode, mask);
9766 static int btrfs_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
9767 struct file *file, umode_t mode)
9769 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9770 struct btrfs_trans_handle *trans;
9771 struct btrfs_root *root = BTRFS_I(dir)->root;
9772 struct inode *inode;
9773 struct btrfs_new_inode_args new_inode_args = {
9775 .dentry = file->f_path.dentry,
9778 unsigned int trans_num_items;
9781 inode = new_inode(dir->i_sb);
9784 inode_init_owner(idmap, inode, dir, mode);
9785 inode->i_fop = &btrfs_file_operations;
9786 inode->i_op = &btrfs_file_inode_operations;
9787 inode->i_mapping->a_ops = &btrfs_aops;
9789 new_inode_args.inode = inode;
9790 ret = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9794 trans = btrfs_start_transaction(root, trans_num_items);
9795 if (IS_ERR(trans)) {
9796 ret = PTR_ERR(trans);
9797 goto out_new_inode_args;
9800 ret = btrfs_create_new_inode(trans, &new_inode_args);
9803 * We set number of links to 0 in btrfs_create_new_inode(), and here we
9804 * set it to 1 because d_tmpfile() will issue a warning if the count is
9807 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9809 set_nlink(inode, 1);
9812 d_tmpfile(file, inode);
9813 unlock_new_inode(inode);
9814 mark_inode_dirty(inode);
9817 btrfs_end_transaction(trans);
9818 btrfs_btree_balance_dirty(fs_info);
9820 btrfs_new_inode_args_destroy(&new_inode_args);
9824 return finish_open_simple(file, ret);
9827 void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end)
9829 struct btrfs_fs_info *fs_info = inode->root->fs_info;
9830 unsigned long index = start >> PAGE_SHIFT;
9831 unsigned long end_index = end >> PAGE_SHIFT;
9835 ASSERT(end + 1 - start <= U32_MAX);
9836 len = end + 1 - start;
9837 while (index <= end_index) {
9838 page = find_get_page(inode->vfs_inode.i_mapping, index);
9839 ASSERT(page); /* Pages should be in the extent_io_tree */
9841 btrfs_page_set_writeback(fs_info, page, start, len);
9847 int btrfs_encoded_io_compression_from_extent(struct btrfs_fs_info *fs_info,
9850 switch (compress_type) {
9851 case BTRFS_COMPRESS_NONE:
9852 return BTRFS_ENCODED_IO_COMPRESSION_NONE;
9853 case BTRFS_COMPRESS_ZLIB:
9854 return BTRFS_ENCODED_IO_COMPRESSION_ZLIB;
9855 case BTRFS_COMPRESS_LZO:
9857 * The LZO format depends on the sector size. 64K is the maximum
9858 * sector size that we support.
9860 if (fs_info->sectorsize < SZ_4K || fs_info->sectorsize > SZ_64K)
9862 return BTRFS_ENCODED_IO_COMPRESSION_LZO_4K +
9863 (fs_info->sectorsize_bits - 12);
9864 case BTRFS_COMPRESS_ZSTD:
9865 return BTRFS_ENCODED_IO_COMPRESSION_ZSTD;
9871 static ssize_t btrfs_encoded_read_inline(
9873 struct iov_iter *iter, u64 start,
9875 struct extent_state **cached_state,
9876 u64 extent_start, size_t count,
9877 struct btrfs_ioctl_encoded_io_args *encoded,
9880 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
9881 struct btrfs_root *root = inode->root;
9882 struct btrfs_fs_info *fs_info = root->fs_info;
9883 struct extent_io_tree *io_tree = &inode->io_tree;
9884 struct btrfs_path *path;
9885 struct extent_buffer *leaf;
9886 struct btrfs_file_extent_item *item;
9892 path = btrfs_alloc_path();
9897 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
9901 /* The extent item disappeared? */
9906 leaf = path->nodes[0];
9907 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
9909 ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
9910 ptr = btrfs_file_extent_inline_start(item);
9912 encoded->len = min_t(u64, extent_start + ram_bytes,
9913 inode->vfs_inode.i_size) - iocb->ki_pos;
9914 ret = btrfs_encoded_io_compression_from_extent(fs_info,
9915 btrfs_file_extent_compression(leaf, item));
9918 encoded->compression = ret;
9919 if (encoded->compression) {
9922 inline_size = btrfs_file_extent_inline_item_len(leaf,
9924 if (inline_size > count) {
9928 count = inline_size;
9929 encoded->unencoded_len = ram_bytes;
9930 encoded->unencoded_offset = iocb->ki_pos - extent_start;
9932 count = min_t(u64, count, encoded->len);
9933 encoded->len = count;
9934 encoded->unencoded_len = count;
9935 ptr += iocb->ki_pos - extent_start;
9938 tmp = kmalloc(count, GFP_NOFS);
9943 read_extent_buffer(leaf, tmp, ptr, count);
9944 btrfs_release_path(path);
9945 unlock_extent(io_tree, start, lockend, cached_state);
9946 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
9949 ret = copy_to_iter(tmp, count, iter);
9954 btrfs_free_path(path);
9958 struct btrfs_encoded_read_private {
9959 wait_queue_head_t wait;
9961 blk_status_t status;
9964 static void btrfs_encoded_read_endio(struct btrfs_bio *bbio)
9966 struct btrfs_encoded_read_private *priv = bbio->private;
9968 if (bbio->bio.bi_status) {
9970 * The memory barrier implied by the atomic_dec_return() here
9971 * pairs with the memory barrier implied by the
9972 * atomic_dec_return() or io_wait_event() in
9973 * btrfs_encoded_read_regular_fill_pages() to ensure that this
9974 * write is observed before the load of status in
9975 * btrfs_encoded_read_regular_fill_pages().
9977 WRITE_ONCE(priv->status, bbio->bio.bi_status);
9979 if (!atomic_dec_return(&priv->pending))
9980 wake_up(&priv->wait);
9981 bio_put(&bbio->bio);
9984 int btrfs_encoded_read_regular_fill_pages(struct btrfs_inode *inode,
9985 u64 file_offset, u64 disk_bytenr,
9986 u64 disk_io_size, struct page **pages)
9988 struct btrfs_fs_info *fs_info = inode->root->fs_info;
9989 struct btrfs_encoded_read_private priv = {
9990 .pending = ATOMIC_INIT(1),
9992 unsigned long i = 0;
9993 struct btrfs_bio *bbio;
9995 init_waitqueue_head(&priv.wait);
9997 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
9998 btrfs_encoded_read_endio, &priv);
9999 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
10000 bbio->inode = inode;
10003 size_t bytes = min_t(u64, disk_io_size, PAGE_SIZE);
10005 if (bio_add_page(&bbio->bio, pages[i], bytes, 0) < bytes) {
10006 atomic_inc(&priv.pending);
10007 btrfs_submit_bio(bbio, 0);
10009 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
10010 btrfs_encoded_read_endio, &priv);
10011 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
10012 bbio->inode = inode;
10017 disk_bytenr += bytes;
10018 disk_io_size -= bytes;
10019 } while (disk_io_size);
10021 atomic_inc(&priv.pending);
10022 btrfs_submit_bio(bbio, 0);
10024 if (atomic_dec_return(&priv.pending))
10025 io_wait_event(priv.wait, !atomic_read(&priv.pending));
10026 /* See btrfs_encoded_read_endio() for ordering. */
10027 return blk_status_to_errno(READ_ONCE(priv.status));
10030 static ssize_t btrfs_encoded_read_regular(struct kiocb *iocb,
10031 struct iov_iter *iter,
10032 u64 start, u64 lockend,
10033 struct extent_state **cached_state,
10034 u64 disk_bytenr, u64 disk_io_size,
10035 size_t count, bool compressed,
10038 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10039 struct extent_io_tree *io_tree = &inode->io_tree;
10040 struct page **pages;
10041 unsigned long nr_pages, i;
10043 size_t page_offset;
10046 nr_pages = DIV_ROUND_UP(disk_io_size, PAGE_SIZE);
10047 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
10050 ret = btrfs_alloc_page_array(nr_pages, pages);
10056 ret = btrfs_encoded_read_regular_fill_pages(inode, start, disk_bytenr,
10057 disk_io_size, pages);
10061 unlock_extent(io_tree, start, lockend, cached_state);
10062 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10069 i = (iocb->ki_pos - start) >> PAGE_SHIFT;
10070 page_offset = (iocb->ki_pos - start) & (PAGE_SIZE - 1);
10073 while (cur < count) {
10074 size_t bytes = min_t(size_t, count - cur,
10075 PAGE_SIZE - page_offset);
10077 if (copy_page_to_iter(pages[i], page_offset, bytes,
10088 for (i = 0; i < nr_pages; i++) {
10090 __free_page(pages[i]);
10096 ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter,
10097 struct btrfs_ioctl_encoded_io_args *encoded)
10099 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10100 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10101 struct extent_io_tree *io_tree = &inode->io_tree;
10103 size_t count = iov_iter_count(iter);
10104 u64 start, lockend, disk_bytenr, disk_io_size;
10105 struct extent_state *cached_state = NULL;
10106 struct extent_map *em;
10107 bool unlocked = false;
10109 file_accessed(iocb->ki_filp);
10111 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
10113 if (iocb->ki_pos >= inode->vfs_inode.i_size) {
10114 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10117 start = ALIGN_DOWN(iocb->ki_pos, fs_info->sectorsize);
10119 * We don't know how long the extent containing iocb->ki_pos is, but if
10120 * it's compressed we know that it won't be longer than this.
10122 lockend = start + BTRFS_MAX_UNCOMPRESSED - 1;
10125 struct btrfs_ordered_extent *ordered;
10127 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start,
10128 lockend - start + 1);
10130 goto out_unlock_inode;
10131 lock_extent(io_tree, start, lockend, &cached_state);
10132 ordered = btrfs_lookup_ordered_range(inode, start,
10133 lockend - start + 1);
10136 btrfs_put_ordered_extent(ordered);
10137 unlock_extent(io_tree, start, lockend, &cached_state);
10141 em = btrfs_get_extent(inode, NULL, 0, start, lockend - start + 1);
10144 goto out_unlock_extent;
10147 if (em->block_start == EXTENT_MAP_INLINE) {
10148 u64 extent_start = em->start;
10151 * For inline extents we get everything we need out of the
10154 free_extent_map(em);
10156 ret = btrfs_encoded_read_inline(iocb, iter, start, lockend,
10157 &cached_state, extent_start,
10158 count, encoded, &unlocked);
10163 * We only want to return up to EOF even if the extent extends beyond
10166 encoded->len = min_t(u64, extent_map_end(em),
10167 inode->vfs_inode.i_size) - iocb->ki_pos;
10168 if (em->block_start == EXTENT_MAP_HOLE ||
10169 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
10170 disk_bytenr = EXTENT_MAP_HOLE;
10171 count = min_t(u64, count, encoded->len);
10172 encoded->len = count;
10173 encoded->unencoded_len = count;
10174 } else if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10175 disk_bytenr = em->block_start;
10177 * Bail if the buffer isn't large enough to return the whole
10178 * compressed extent.
10180 if (em->block_len > count) {
10184 disk_io_size = em->block_len;
10185 count = em->block_len;
10186 encoded->unencoded_len = em->ram_bytes;
10187 encoded->unencoded_offset = iocb->ki_pos - em->orig_start;
10188 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10189 em->compress_type);
10192 encoded->compression = ret;
10194 disk_bytenr = em->block_start + (start - em->start);
10195 if (encoded->len > count)
10196 encoded->len = count;
10198 * Don't read beyond what we locked. This also limits the page
10199 * allocations that we'll do.
10201 disk_io_size = min(lockend + 1, iocb->ki_pos + encoded->len) - start;
10202 count = start + disk_io_size - iocb->ki_pos;
10203 encoded->len = count;
10204 encoded->unencoded_len = count;
10205 disk_io_size = ALIGN(disk_io_size, fs_info->sectorsize);
10207 free_extent_map(em);
10210 if (disk_bytenr == EXTENT_MAP_HOLE) {
10211 unlock_extent(io_tree, start, lockend, &cached_state);
10212 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10214 ret = iov_iter_zero(count, iter);
10218 ret = btrfs_encoded_read_regular(iocb, iter, start, lockend,
10219 &cached_state, disk_bytenr,
10220 disk_io_size, count,
10221 encoded->compression,
10227 iocb->ki_pos += encoded->len;
10229 free_extent_map(em);
10232 unlock_extent(io_tree, start, lockend, &cached_state);
10235 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10239 ssize_t btrfs_do_encoded_write(struct kiocb *iocb, struct iov_iter *from,
10240 const struct btrfs_ioctl_encoded_io_args *encoded)
10242 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10243 struct btrfs_root *root = inode->root;
10244 struct btrfs_fs_info *fs_info = root->fs_info;
10245 struct extent_io_tree *io_tree = &inode->io_tree;
10246 struct extent_changeset *data_reserved = NULL;
10247 struct extent_state *cached_state = NULL;
10248 struct btrfs_ordered_extent *ordered;
10252 u64 num_bytes, ram_bytes, disk_num_bytes;
10253 unsigned long nr_pages, i;
10254 struct page **pages;
10255 struct btrfs_key ins;
10256 bool extent_reserved = false;
10257 struct extent_map *em;
10260 switch (encoded->compression) {
10261 case BTRFS_ENCODED_IO_COMPRESSION_ZLIB:
10262 compression = BTRFS_COMPRESS_ZLIB;
10264 case BTRFS_ENCODED_IO_COMPRESSION_ZSTD:
10265 compression = BTRFS_COMPRESS_ZSTD;
10267 case BTRFS_ENCODED_IO_COMPRESSION_LZO_4K:
10268 case BTRFS_ENCODED_IO_COMPRESSION_LZO_8K:
10269 case BTRFS_ENCODED_IO_COMPRESSION_LZO_16K:
10270 case BTRFS_ENCODED_IO_COMPRESSION_LZO_32K:
10271 case BTRFS_ENCODED_IO_COMPRESSION_LZO_64K:
10272 /* The sector size must match for LZO. */
10273 if (encoded->compression -
10274 BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + 12 !=
10275 fs_info->sectorsize_bits)
10277 compression = BTRFS_COMPRESS_LZO;
10282 if (encoded->encryption != BTRFS_ENCODED_IO_ENCRYPTION_NONE)
10285 orig_count = iov_iter_count(from);
10287 /* The extent size must be sane. */
10288 if (encoded->unencoded_len > BTRFS_MAX_UNCOMPRESSED ||
10289 orig_count > BTRFS_MAX_COMPRESSED || orig_count == 0)
10293 * The compressed data must be smaller than the decompressed data.
10295 * It's of course possible for data to compress to larger or the same
10296 * size, but the buffered I/O path falls back to no compression for such
10297 * data, and we don't want to break any assumptions by creating these
10300 * Note that this is less strict than the current check we have that the
10301 * compressed data must be at least one sector smaller than the
10302 * decompressed data. We only want to enforce the weaker requirement
10303 * from old kernels that it is at least one byte smaller.
10305 if (orig_count >= encoded->unencoded_len)
10308 /* The extent must start on a sector boundary. */
10309 start = iocb->ki_pos;
10310 if (!IS_ALIGNED(start, fs_info->sectorsize))
10314 * The extent must end on a sector boundary. However, we allow a write
10315 * which ends at or extends i_size to have an unaligned length; we round
10316 * up the extent size and set i_size to the unaligned end.
10318 if (start + encoded->len < inode->vfs_inode.i_size &&
10319 !IS_ALIGNED(start + encoded->len, fs_info->sectorsize))
10322 /* Finally, the offset in the unencoded data must be sector-aligned. */
10323 if (!IS_ALIGNED(encoded->unencoded_offset, fs_info->sectorsize))
10326 num_bytes = ALIGN(encoded->len, fs_info->sectorsize);
10327 ram_bytes = ALIGN(encoded->unencoded_len, fs_info->sectorsize);
10328 end = start + num_bytes - 1;
10331 * If the extent cannot be inline, the compressed data on disk must be
10332 * sector-aligned. For convenience, we extend it with zeroes if it
10335 disk_num_bytes = ALIGN(orig_count, fs_info->sectorsize);
10336 nr_pages = DIV_ROUND_UP(disk_num_bytes, PAGE_SIZE);
10337 pages = kvcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL_ACCOUNT);
10340 for (i = 0; i < nr_pages; i++) {
10341 size_t bytes = min_t(size_t, PAGE_SIZE, iov_iter_count(from));
10344 pages[i] = alloc_page(GFP_KERNEL_ACCOUNT);
10349 kaddr = kmap_local_page(pages[i]);
10350 if (copy_from_iter(kaddr, bytes, from) != bytes) {
10351 kunmap_local(kaddr);
10355 if (bytes < PAGE_SIZE)
10356 memset(kaddr + bytes, 0, PAGE_SIZE - bytes);
10357 kunmap_local(kaddr);
10361 struct btrfs_ordered_extent *ordered;
10363 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start, num_bytes);
10366 ret = invalidate_inode_pages2_range(inode->vfs_inode.i_mapping,
10367 start >> PAGE_SHIFT,
10368 end >> PAGE_SHIFT);
10371 lock_extent(io_tree, start, end, &cached_state);
10372 ordered = btrfs_lookup_ordered_range(inode, start, num_bytes);
10374 !filemap_range_has_page(inode->vfs_inode.i_mapping, start, end))
10377 btrfs_put_ordered_extent(ordered);
10378 unlock_extent(io_tree, start, end, &cached_state);
10383 * We don't use the higher-level delalloc space functions because our
10384 * num_bytes and disk_num_bytes are different.
10386 ret = btrfs_alloc_data_chunk_ondemand(inode, disk_num_bytes);
10389 ret = btrfs_qgroup_reserve_data(inode, &data_reserved, start, num_bytes);
10391 goto out_free_data_space;
10392 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes, disk_num_bytes,
10395 goto out_qgroup_free_data;
10397 /* Try an inline extent first. */
10398 if (start == 0 && encoded->unencoded_len == encoded->len &&
10399 encoded->unencoded_offset == 0) {
10400 ret = cow_file_range_inline(inode, encoded->len, orig_count,
10401 compression, pages, true);
10405 goto out_delalloc_release;
10409 ret = btrfs_reserve_extent(root, disk_num_bytes, disk_num_bytes,
10410 disk_num_bytes, 0, 0, &ins, 1, 1);
10412 goto out_delalloc_release;
10413 extent_reserved = true;
10415 em = create_io_em(inode, start, num_bytes,
10416 start - encoded->unencoded_offset, ins.objectid,
10417 ins.offset, ins.offset, ram_bytes, compression,
10418 BTRFS_ORDERED_COMPRESSED);
10421 goto out_free_reserved;
10423 free_extent_map(em);
10425 ordered = btrfs_alloc_ordered_extent(inode, start, num_bytes, ram_bytes,
10426 ins.objectid, ins.offset,
10427 encoded->unencoded_offset,
10428 (1 << BTRFS_ORDERED_ENCODED) |
10429 (1 << BTRFS_ORDERED_COMPRESSED),
10431 if (IS_ERR(ordered)) {
10432 btrfs_drop_extent_map_range(inode, start, end, false);
10433 ret = PTR_ERR(ordered);
10434 goto out_free_reserved;
10436 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10438 if (start + encoded->len > inode->vfs_inode.i_size)
10439 i_size_write(&inode->vfs_inode, start + encoded->len);
10441 unlock_extent(io_tree, start, end, &cached_state);
10443 btrfs_delalloc_release_extents(inode, num_bytes);
10445 btrfs_submit_compressed_write(ordered, pages, nr_pages, 0, false);
10450 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10451 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
10452 out_delalloc_release:
10453 btrfs_delalloc_release_extents(inode, num_bytes);
10454 btrfs_delalloc_release_metadata(inode, disk_num_bytes, ret < 0);
10455 out_qgroup_free_data:
10457 btrfs_qgroup_free_data(inode, data_reserved, start, num_bytes);
10458 out_free_data_space:
10460 * If btrfs_reserve_extent() succeeded, then we already decremented
10463 if (!extent_reserved)
10464 btrfs_free_reserved_data_space_noquota(fs_info, disk_num_bytes);
10466 unlock_extent(io_tree, start, end, &cached_state);
10468 for (i = 0; i < nr_pages; i++) {
10470 __free_page(pages[i]);
10475 iocb->ki_pos += encoded->len;
10481 * Add an entry indicating a block group or device which is pinned by a
10482 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10483 * negative errno on failure.
10485 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10486 bool is_block_group)
10488 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10489 struct btrfs_swapfile_pin *sp, *entry;
10490 struct rb_node **p;
10491 struct rb_node *parent = NULL;
10493 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10498 sp->is_block_group = is_block_group;
10499 sp->bg_extent_count = 1;
10501 spin_lock(&fs_info->swapfile_pins_lock);
10502 p = &fs_info->swapfile_pins.rb_node;
10505 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10506 if (sp->ptr < entry->ptr ||
10507 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10508 p = &(*p)->rb_left;
10509 } else if (sp->ptr > entry->ptr ||
10510 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10511 p = &(*p)->rb_right;
10513 if (is_block_group)
10514 entry->bg_extent_count++;
10515 spin_unlock(&fs_info->swapfile_pins_lock);
10520 rb_link_node(&sp->node, parent, p);
10521 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10522 spin_unlock(&fs_info->swapfile_pins_lock);
10526 /* Free all of the entries pinned by this swapfile. */
10527 static void btrfs_free_swapfile_pins(struct inode *inode)
10529 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10530 struct btrfs_swapfile_pin *sp;
10531 struct rb_node *node, *next;
10533 spin_lock(&fs_info->swapfile_pins_lock);
10534 node = rb_first(&fs_info->swapfile_pins);
10536 next = rb_next(node);
10537 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10538 if (sp->inode == inode) {
10539 rb_erase(&sp->node, &fs_info->swapfile_pins);
10540 if (sp->is_block_group) {
10541 btrfs_dec_block_group_swap_extents(sp->ptr,
10542 sp->bg_extent_count);
10543 btrfs_put_block_group(sp->ptr);
10549 spin_unlock(&fs_info->swapfile_pins_lock);
10552 struct btrfs_swap_info {
10558 unsigned long nr_pages;
10562 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10563 struct btrfs_swap_info *bsi)
10565 unsigned long nr_pages;
10566 unsigned long max_pages;
10567 u64 first_ppage, first_ppage_reported, next_ppage;
10571 * Our swapfile may have had its size extended after the swap header was
10572 * written. In that case activating the swapfile should not go beyond
10573 * the max size set in the swap header.
10575 if (bsi->nr_pages >= sis->max)
10578 max_pages = sis->max - bsi->nr_pages;
10579 first_ppage = PAGE_ALIGN(bsi->block_start) >> PAGE_SHIFT;
10580 next_ppage = PAGE_ALIGN_DOWN(bsi->block_start + bsi->block_len) >> PAGE_SHIFT;
10582 if (first_ppage >= next_ppage)
10584 nr_pages = next_ppage - first_ppage;
10585 nr_pages = min(nr_pages, max_pages);
10587 first_ppage_reported = first_ppage;
10588 if (bsi->start == 0)
10589 first_ppage_reported++;
10590 if (bsi->lowest_ppage > first_ppage_reported)
10591 bsi->lowest_ppage = first_ppage_reported;
10592 if (bsi->highest_ppage < (next_ppage - 1))
10593 bsi->highest_ppage = next_ppage - 1;
10595 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10598 bsi->nr_extents += ret;
10599 bsi->nr_pages += nr_pages;
10603 static void btrfs_swap_deactivate(struct file *file)
10605 struct inode *inode = file_inode(file);
10607 btrfs_free_swapfile_pins(inode);
10608 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10611 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10614 struct inode *inode = file_inode(file);
10615 struct btrfs_root *root = BTRFS_I(inode)->root;
10616 struct btrfs_fs_info *fs_info = root->fs_info;
10617 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10618 struct extent_state *cached_state = NULL;
10619 struct extent_map *em = NULL;
10620 struct btrfs_device *device = NULL;
10621 struct btrfs_swap_info bsi = {
10622 .lowest_ppage = (sector_t)-1ULL,
10629 * If the swap file was just created, make sure delalloc is done. If the
10630 * file changes again after this, the user is doing something stupid and
10631 * we don't really care.
10633 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10638 * The inode is locked, so these flags won't change after we check them.
10640 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10641 btrfs_warn(fs_info, "swapfile must not be compressed");
10644 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10645 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10648 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10649 btrfs_warn(fs_info, "swapfile must not be checksummed");
10654 * Balance or device remove/replace/resize can move stuff around from
10655 * under us. The exclop protection makes sure they aren't running/won't
10656 * run concurrently while we are mapping the swap extents, and
10657 * fs_info->swapfile_pins prevents them from running while the swap
10658 * file is active and moving the extents. Note that this also prevents
10659 * a concurrent device add which isn't actually necessary, but it's not
10660 * really worth the trouble to allow it.
10662 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10663 btrfs_warn(fs_info,
10664 "cannot activate swapfile while exclusive operation is running");
10669 * Prevent snapshot creation while we are activating the swap file.
10670 * We do not want to race with snapshot creation. If snapshot creation
10671 * already started before we bumped nr_swapfiles from 0 to 1 and
10672 * completes before the first write into the swap file after it is
10673 * activated, than that write would fallback to COW.
10675 if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) {
10676 btrfs_exclop_finish(fs_info);
10677 btrfs_warn(fs_info,
10678 "cannot activate swapfile because snapshot creation is in progress");
10682 * Snapshots can create extents which require COW even if NODATACOW is
10683 * set. We use this counter to prevent snapshots. We must increment it
10684 * before walking the extents because we don't want a concurrent
10685 * snapshot to run after we've already checked the extents.
10687 * It is possible that subvolume is marked for deletion but still not
10688 * removed yet. To prevent this race, we check the root status before
10689 * activating the swapfile.
10691 spin_lock(&root->root_item_lock);
10692 if (btrfs_root_dead(root)) {
10693 spin_unlock(&root->root_item_lock);
10695 btrfs_exclop_finish(fs_info);
10696 btrfs_warn(fs_info,
10697 "cannot activate swapfile because subvolume %llu is being deleted",
10698 root->root_key.objectid);
10701 atomic_inc(&root->nr_swapfiles);
10702 spin_unlock(&root->root_item_lock);
10704 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10706 lock_extent(io_tree, 0, isize - 1, &cached_state);
10708 while (start < isize) {
10709 u64 logical_block_start, physical_block_start;
10710 struct btrfs_block_group *bg;
10711 u64 len = isize - start;
10713 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
10719 if (em->block_start == EXTENT_MAP_HOLE) {
10720 btrfs_warn(fs_info, "swapfile must not have holes");
10724 if (em->block_start == EXTENT_MAP_INLINE) {
10726 * It's unlikely we'll ever actually find ourselves
10727 * here, as a file small enough to fit inline won't be
10728 * big enough to store more than the swap header, but in
10729 * case something changes in the future, let's catch it
10730 * here rather than later.
10732 btrfs_warn(fs_info, "swapfile must not be inline");
10736 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10737 btrfs_warn(fs_info, "swapfile must not be compressed");
10742 logical_block_start = em->block_start + (start - em->start);
10743 len = min(len, em->len - (start - em->start));
10744 free_extent_map(em);
10747 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, false, true);
10753 btrfs_warn(fs_info,
10754 "swapfile must not be copy-on-write");
10759 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10765 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10766 btrfs_warn(fs_info,
10767 "swapfile must have single data profile");
10772 if (device == NULL) {
10773 device = em->map_lookup->stripes[0].dev;
10774 ret = btrfs_add_swapfile_pin(inode, device, false);
10779 } else if (device != em->map_lookup->stripes[0].dev) {
10780 btrfs_warn(fs_info, "swapfile must be on one device");
10785 physical_block_start = (em->map_lookup->stripes[0].physical +
10786 (logical_block_start - em->start));
10787 len = min(len, em->len - (logical_block_start - em->start));
10788 free_extent_map(em);
10791 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10793 btrfs_warn(fs_info,
10794 "could not find block group containing swapfile");
10799 if (!btrfs_inc_block_group_swap_extents(bg)) {
10800 btrfs_warn(fs_info,
10801 "block group for swapfile at %llu is read-only%s",
10803 atomic_read(&fs_info->scrubs_running) ?
10804 " (scrub running)" : "");
10805 btrfs_put_block_group(bg);
10810 ret = btrfs_add_swapfile_pin(inode, bg, true);
10812 btrfs_put_block_group(bg);
10819 if (bsi.block_len &&
10820 bsi.block_start + bsi.block_len == physical_block_start) {
10821 bsi.block_len += len;
10823 if (bsi.block_len) {
10824 ret = btrfs_add_swap_extent(sis, &bsi);
10829 bsi.block_start = physical_block_start;
10830 bsi.block_len = len;
10837 ret = btrfs_add_swap_extent(sis, &bsi);
10840 if (!IS_ERR_OR_NULL(em))
10841 free_extent_map(em);
10843 unlock_extent(io_tree, 0, isize - 1, &cached_state);
10846 btrfs_swap_deactivate(file);
10848 btrfs_drew_write_unlock(&root->snapshot_lock);
10850 btrfs_exclop_finish(fs_info);
10856 sis->bdev = device->bdev;
10857 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10858 sis->max = bsi.nr_pages;
10859 sis->pages = bsi.nr_pages - 1;
10860 sis->highest_bit = bsi.nr_pages - 1;
10861 return bsi.nr_extents;
10864 static void btrfs_swap_deactivate(struct file *file)
10868 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10871 return -EOPNOTSUPP;
10876 * Update the number of bytes used in the VFS' inode. When we replace extents in
10877 * a range (clone, dedupe, fallocate's zero range), we must update the number of
10878 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
10879 * always get a correct value.
10881 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
10882 const u64 add_bytes,
10883 const u64 del_bytes)
10885 if (add_bytes == del_bytes)
10888 spin_lock(&inode->lock);
10890 inode_sub_bytes(&inode->vfs_inode, del_bytes);
10892 inode_add_bytes(&inode->vfs_inode, add_bytes);
10893 spin_unlock(&inode->lock);
10897 * Verify that there are no ordered extents for a given file range.
10899 * @inode: The target inode.
10900 * @start: Start offset of the file range, should be sector size aligned.
10901 * @end: End offset (inclusive) of the file range, its value +1 should be
10902 * sector size aligned.
10904 * This should typically be used for cases where we locked an inode's VFS lock in
10905 * exclusive mode, we have also locked the inode's i_mmap_lock in exclusive mode,
10906 * we have flushed all delalloc in the range, we have waited for all ordered
10907 * extents in the range to complete and finally we have locked the file range in
10908 * the inode's io_tree.
10910 void btrfs_assert_inode_range_clean(struct btrfs_inode *inode, u64 start, u64 end)
10912 struct btrfs_root *root = inode->root;
10913 struct btrfs_ordered_extent *ordered;
10915 if (!IS_ENABLED(CONFIG_BTRFS_ASSERT))
10918 ordered = btrfs_lookup_first_ordered_range(inode, start, end + 1 - start);
10920 btrfs_err(root->fs_info,
10921 "found unexpected ordered extent in file range [%llu, %llu] for inode %llu root %llu (ordered range [%llu, %llu])",
10922 start, end, btrfs_ino(inode), root->root_key.objectid,
10923 ordered->file_offset,
10924 ordered->file_offset + ordered->num_bytes - 1);
10925 btrfs_put_ordered_extent(ordered);
10928 ASSERT(ordered == NULL);
10931 static const struct inode_operations btrfs_dir_inode_operations = {
10932 .getattr = btrfs_getattr,
10933 .lookup = btrfs_lookup,
10934 .create = btrfs_create,
10935 .unlink = btrfs_unlink,
10936 .link = btrfs_link,
10937 .mkdir = btrfs_mkdir,
10938 .rmdir = btrfs_rmdir,
10939 .rename = btrfs_rename2,
10940 .symlink = btrfs_symlink,
10941 .setattr = btrfs_setattr,
10942 .mknod = btrfs_mknod,
10943 .listxattr = btrfs_listxattr,
10944 .permission = btrfs_permission,
10945 .get_inode_acl = btrfs_get_acl,
10946 .set_acl = btrfs_set_acl,
10947 .update_time = btrfs_update_time,
10948 .tmpfile = btrfs_tmpfile,
10949 .fileattr_get = btrfs_fileattr_get,
10950 .fileattr_set = btrfs_fileattr_set,
10953 static const struct file_operations btrfs_dir_file_operations = {
10954 .llseek = generic_file_llseek,
10955 .read = generic_read_dir,
10956 .iterate_shared = btrfs_real_readdir,
10957 .open = btrfs_opendir,
10958 .unlocked_ioctl = btrfs_ioctl,
10959 #ifdef CONFIG_COMPAT
10960 .compat_ioctl = btrfs_compat_ioctl,
10962 .release = btrfs_release_file,
10963 .fsync = btrfs_sync_file,
10967 * btrfs doesn't support the bmap operation because swapfiles
10968 * use bmap to make a mapping of extents in the file. They assume
10969 * these extents won't change over the life of the file and they
10970 * use the bmap result to do IO directly to the drive.
10972 * the btrfs bmap call would return logical addresses that aren't
10973 * suitable for IO and they also will change frequently as COW
10974 * operations happen. So, swapfile + btrfs == corruption.
10976 * For now we're avoiding this by dropping bmap.
10978 static const struct address_space_operations btrfs_aops = {
10979 .read_folio = btrfs_read_folio,
10980 .writepages = btrfs_writepages,
10981 .readahead = btrfs_readahead,
10982 .invalidate_folio = btrfs_invalidate_folio,
10983 .release_folio = btrfs_release_folio,
10984 .migrate_folio = btrfs_migrate_folio,
10985 .dirty_folio = filemap_dirty_folio,
10986 .error_remove_page = generic_error_remove_page,
10987 .swap_activate = btrfs_swap_activate,
10988 .swap_deactivate = btrfs_swap_deactivate,
10991 static const struct inode_operations btrfs_file_inode_operations = {
10992 .getattr = btrfs_getattr,
10993 .setattr = btrfs_setattr,
10994 .listxattr = btrfs_listxattr,
10995 .permission = btrfs_permission,
10996 .fiemap = btrfs_fiemap,
10997 .get_inode_acl = btrfs_get_acl,
10998 .set_acl = btrfs_set_acl,
10999 .update_time = btrfs_update_time,
11000 .fileattr_get = btrfs_fileattr_get,
11001 .fileattr_set = btrfs_fileattr_set,
11003 static const struct inode_operations btrfs_special_inode_operations = {
11004 .getattr = btrfs_getattr,
11005 .setattr = btrfs_setattr,
11006 .permission = btrfs_permission,
11007 .listxattr = btrfs_listxattr,
11008 .get_inode_acl = btrfs_get_acl,
11009 .set_acl = btrfs_set_acl,
11010 .update_time = btrfs_update_time,
11012 static const struct inode_operations btrfs_symlink_inode_operations = {
11013 .get_link = page_get_link,
11014 .getattr = btrfs_getattr,
11015 .setattr = btrfs_setattr,
11016 .permission = btrfs_permission,
11017 .listxattr = btrfs_listxattr,
11018 .update_time = btrfs_update_time,
11021 const struct dentry_operations btrfs_dentry_operations = {
11022 .d_delete = btrfs_dentry_delete,