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 run_delalloc_cow(struct btrfs_inode *inode,
129 struct page *locked_page, u64 start,
130 u64 end, struct writeback_control *wbc,
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;
850 inode_should_defrag(inode, start, end, end - start + 1, SZ_16K);
853 * We need to call clear_page_dirty_for_io on each page in the range.
854 * Otherwise applications with the file mmap'd can wander in and change
855 * the page contents while we are compressing them.
857 extent_range_clear_dirty_for_io(&inode->vfs_inode, start, end);
860 * We need to save i_size before now because it could change in between
861 * us evaluating the size and assigning it. This is because we lock and
862 * unlock the page in truncate and fallocate, and then modify the i_size
865 * The barriers are to emulate READ_ONCE, remove that once i_size_read
869 i_size = i_size_read(&inode->vfs_inode);
871 actual_end = min_t(u64, i_size, end + 1);
874 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
875 nr_pages = min_t(unsigned long, nr_pages, BTRFS_MAX_COMPRESSED_PAGES);
878 * we don't want to send crud past the end of i_size through
879 * compression, that's just a waste of CPU time. So, if the
880 * end of the file is before the start of our current
881 * requested range of bytes, we bail out to the uncompressed
882 * cleanup code that can deal with all of this.
884 * It isn't really the fastest way to fix things, but this is a
885 * very uncommon corner.
887 if (actual_end <= start)
888 goto cleanup_and_bail_uncompressed;
890 total_compressed = actual_end - start;
893 * Skip compression for a small file range(<=blocksize) that
894 * isn't an inline extent, since it doesn't save disk space at all.
896 if (total_compressed <= blocksize &&
897 (start > 0 || end + 1 < inode->disk_i_size))
898 goto cleanup_and_bail_uncompressed;
901 * For subpage case, we require full page alignment for the sector
903 * Thus we must also check against @actual_end, not just @end.
905 if (blocksize < PAGE_SIZE) {
906 if (!PAGE_ALIGNED(start) ||
907 !PAGE_ALIGNED(round_up(actual_end, blocksize)))
908 goto cleanup_and_bail_uncompressed;
911 total_compressed = min_t(unsigned long, total_compressed,
912 BTRFS_MAX_UNCOMPRESSED);
917 * We do compression for mount -o compress and when the inode has not
918 * been flagged as NOCOMPRESS. This flag can change at any time if we
919 * discover bad compression ratios.
921 if (!inode_need_compress(inode, start, end))
922 goto cleanup_and_bail_uncompressed;
924 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
927 * Memory allocation failure is not a fatal error, we can fall
928 * back to uncompressed code.
930 goto cleanup_and_bail_uncompressed;
933 if (inode->defrag_compress)
934 compress_type = inode->defrag_compress;
935 else if (inode->prop_compress)
936 compress_type = inode->prop_compress;
938 /* Compression level is applied here. */
939 ret = btrfs_compress_pages(compress_type | (fs_info->compress_level << 4),
940 mapping, start, pages, &nr_pages, &total_in,
943 goto mark_incompressible;
946 * Zero the tail end of the last page, as we might be sending it down
949 poff = offset_in_page(total_compressed);
951 memzero_page(pages[nr_pages - 1], poff, PAGE_SIZE - poff);
954 * Try to create an inline extent.
956 * If we didn't compress the entire range, try to create an uncompressed
957 * inline extent, else a compressed one.
959 * Check cow_file_range() for why we don't even try to create inline
960 * extent for the subpage case.
962 if (start == 0 && fs_info->sectorsize == PAGE_SIZE) {
963 if (total_in < actual_end) {
964 ret = cow_file_range_inline(inode, actual_end, 0,
965 BTRFS_COMPRESS_NONE, NULL,
968 ret = cow_file_range_inline(inode, actual_end,
970 compress_type, pages,
974 unsigned long clear_flags = EXTENT_DELALLOC |
975 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
976 EXTENT_DO_ACCOUNTING;
979 mapping_set_error(mapping, -EIO);
982 * inline extent creation worked or returned error,
983 * we don't need to create any more async work items.
984 * Unlock and free up our temp pages.
986 * We use DO_ACCOUNTING here because we need the
987 * delalloc_release_metadata to be done _after_ we drop
988 * our outstanding extent for clearing delalloc for this
991 extent_clear_unlock_delalloc(inode, start, end,
995 PAGE_START_WRITEBACK |
1002 * We aren't doing an inline extent. Round the compressed size up to a
1003 * block size boundary so the allocator does sane things.
1005 total_compressed = ALIGN(total_compressed, blocksize);
1008 * One last check to make sure the compression is really a win, compare
1009 * the page count read with the blocks on disk, compression must free at
1012 total_in = round_up(total_in, fs_info->sectorsize);
1013 if (total_compressed + blocksize > total_in)
1014 goto mark_incompressible;
1017 * The async work queues will take care of doing actual allocation on
1018 * disk for these compressed pages, and will submit the bios.
1020 add_async_extent(async_chunk, start, total_in, total_compressed, pages,
1021 nr_pages, compress_type);
1022 if (start + total_in < end) {
1029 mark_incompressible:
1030 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) && !inode->prop_compress)
1031 inode->flags |= BTRFS_INODE_NOCOMPRESS;
1032 cleanup_and_bail_uncompressed:
1033 add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
1034 BTRFS_COMPRESS_NONE);
1037 for (i = 0; i < nr_pages; i++) {
1038 WARN_ON(pages[i]->mapping);
1045 static void free_async_extent_pages(struct async_extent *async_extent)
1049 if (!async_extent->pages)
1052 for (i = 0; i < async_extent->nr_pages; i++) {
1053 WARN_ON(async_extent->pages[i]->mapping);
1054 put_page(async_extent->pages[i]);
1056 kfree(async_extent->pages);
1057 async_extent->nr_pages = 0;
1058 async_extent->pages = NULL;
1061 static void submit_uncompressed_range(struct btrfs_inode *inode,
1062 struct async_extent *async_extent,
1063 struct page *locked_page)
1065 u64 start = async_extent->start;
1066 u64 end = async_extent->start + async_extent->ram_size - 1;
1068 struct writeback_control wbc = {
1069 .sync_mode = WB_SYNC_ALL,
1070 .range_start = start,
1072 .no_cgroup_owner = 1,
1075 wbc_attach_fdatawrite_inode(&wbc, &inode->vfs_inode);
1076 ret = run_delalloc_cow(inode, locked_page, start, end, &wbc, false);
1077 wbc_detach_inode(&wbc);
1079 btrfs_cleanup_ordered_extents(inode, locked_page, start, end - start + 1);
1081 const u64 page_start = page_offset(locked_page);
1083 set_page_writeback(locked_page);
1084 end_page_writeback(locked_page);
1085 btrfs_mark_ordered_io_finished(inode, locked_page,
1086 page_start, PAGE_SIZE,
1088 btrfs_page_clear_uptodate(inode->root->fs_info,
1089 locked_page, page_start,
1091 mapping_set_error(locked_page->mapping, ret);
1092 unlock_page(locked_page);
1097 static void submit_one_async_extent(struct async_chunk *async_chunk,
1098 struct async_extent *async_extent,
1101 struct btrfs_inode *inode = async_chunk->inode;
1102 struct extent_io_tree *io_tree = &inode->io_tree;
1103 struct btrfs_root *root = inode->root;
1104 struct btrfs_fs_info *fs_info = root->fs_info;
1105 struct btrfs_ordered_extent *ordered;
1106 struct btrfs_key ins;
1107 struct page *locked_page = NULL;
1108 struct extent_map *em;
1110 u64 start = async_extent->start;
1111 u64 end = async_extent->start + async_extent->ram_size - 1;
1113 if (async_chunk->blkcg_css)
1114 kthread_associate_blkcg(async_chunk->blkcg_css);
1117 * If async_chunk->locked_page is in the async_extent range, we need to
1120 if (async_chunk->locked_page) {
1121 u64 locked_page_start = page_offset(async_chunk->locked_page);
1122 u64 locked_page_end = locked_page_start + PAGE_SIZE - 1;
1124 if (!(start >= locked_page_end || end <= locked_page_start))
1125 locked_page = async_chunk->locked_page;
1127 lock_extent(io_tree, start, end, NULL);
1129 if (async_extent->compress_type == BTRFS_COMPRESS_NONE) {
1130 submit_uncompressed_range(inode, async_extent, locked_page);
1134 ret = btrfs_reserve_extent(root, async_extent->ram_size,
1135 async_extent->compressed_size,
1136 async_extent->compressed_size,
1137 0, *alloc_hint, &ins, 1, 1);
1140 * Here we used to try again by going back to non-compressed
1141 * path for ENOSPC. But we can't reserve space even for
1142 * compressed size, how could it work for uncompressed size
1143 * which requires larger size? So here we directly go error
1149 /* Here we're doing allocation and writeback of the compressed pages */
1150 em = create_io_em(inode, start,
1151 async_extent->ram_size, /* len */
1152 start, /* orig_start */
1153 ins.objectid, /* block_start */
1154 ins.offset, /* block_len */
1155 ins.offset, /* orig_block_len */
1156 async_extent->ram_size, /* ram_bytes */
1157 async_extent->compress_type,
1158 BTRFS_ORDERED_COMPRESSED);
1161 goto out_free_reserve;
1163 free_extent_map(em);
1165 ordered = btrfs_alloc_ordered_extent(inode, start, /* file_offset */
1166 async_extent->ram_size, /* num_bytes */
1167 async_extent->ram_size, /* ram_bytes */
1168 ins.objectid, /* disk_bytenr */
1169 ins.offset, /* disk_num_bytes */
1171 1 << BTRFS_ORDERED_COMPRESSED,
1172 async_extent->compress_type);
1173 if (IS_ERR(ordered)) {
1174 btrfs_drop_extent_map_range(inode, start, end, false);
1175 ret = PTR_ERR(ordered);
1176 goto out_free_reserve;
1178 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1180 /* Clear dirty, set writeback and unlock the pages. */
1181 extent_clear_unlock_delalloc(inode, start, end,
1182 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
1183 PAGE_UNLOCK | PAGE_START_WRITEBACK);
1184 btrfs_submit_compressed_write(ordered,
1185 async_extent->pages, /* compressed_pages */
1186 async_extent->nr_pages,
1187 async_chunk->write_flags, true);
1188 *alloc_hint = ins.objectid + ins.offset;
1190 if (async_chunk->blkcg_css)
1191 kthread_associate_blkcg(NULL);
1192 kfree(async_extent);
1196 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1197 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1199 mapping_set_error(inode->vfs_inode.i_mapping, -EIO);
1200 extent_clear_unlock_delalloc(inode, start, end,
1201 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
1202 EXTENT_DELALLOC_NEW |
1203 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
1204 PAGE_UNLOCK | PAGE_START_WRITEBACK |
1205 PAGE_END_WRITEBACK);
1206 free_async_extent_pages(async_extent);
1207 if (async_chunk->blkcg_css)
1208 kthread_associate_blkcg(NULL);
1209 btrfs_debug(fs_info,
1210 "async extent submission failed root=%lld inode=%llu start=%llu len=%llu ret=%d",
1211 root->root_key.objectid, btrfs_ino(inode), start,
1212 async_extent->ram_size, ret);
1213 kfree(async_extent);
1216 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
1219 struct extent_map_tree *em_tree = &inode->extent_tree;
1220 struct extent_map *em;
1223 read_lock(&em_tree->lock);
1224 em = search_extent_mapping(em_tree, start, num_bytes);
1227 * if block start isn't an actual block number then find the
1228 * first block in this inode and use that as a hint. If that
1229 * block is also bogus then just don't worry about it.
1231 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1232 free_extent_map(em);
1233 em = search_extent_mapping(em_tree, 0, 0);
1234 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
1235 alloc_hint = em->block_start;
1237 free_extent_map(em);
1239 alloc_hint = em->block_start;
1240 free_extent_map(em);
1243 read_unlock(&em_tree->lock);
1249 * when extent_io.c finds a delayed allocation range in the file,
1250 * the call backs end up in this code. The basic idea is to
1251 * allocate extents on disk for the range, and create ordered data structs
1252 * in ram to track those extents.
1254 * locked_page is the page that writepage had locked already. We use
1255 * it to make sure we don't do extra locks or unlocks.
1257 * When this function fails, it unlocks all pages except @locked_page.
1259 * When this function successfully creates an inline extent, it returns 1 and
1260 * unlocks all pages including locked_page and starts I/O on them.
1261 * (In reality inline extents are limited to a single page, so locked_page is
1262 * the only page handled anyway).
1264 * When this function succeed and creates a normal extent, the page locking
1265 * status depends on the passed in flags:
1267 * - If @keep_locked is set, all pages are kept locked.
1268 * - Else all pages except for @locked_page are unlocked.
1270 * When a failure happens in the second or later iteration of the
1271 * while-loop, the ordered extents created in previous iterations are kept
1272 * intact. So, the caller must clean them up by calling
1273 * btrfs_cleanup_ordered_extents(). See btrfs_run_delalloc_range() for
1276 static noinline int cow_file_range(struct btrfs_inode *inode,
1277 struct page *locked_page, u64 start, u64 end,
1279 bool keep_locked, bool no_inline)
1281 struct btrfs_root *root = inode->root;
1282 struct btrfs_fs_info *fs_info = root->fs_info;
1284 u64 orig_start = start;
1286 unsigned long ram_size;
1287 u64 cur_alloc_size = 0;
1289 u64 blocksize = fs_info->sectorsize;
1290 struct btrfs_key ins;
1291 struct extent_map *em;
1292 unsigned clear_bits;
1293 unsigned long page_ops;
1294 bool extent_reserved = false;
1297 if (btrfs_is_free_space_inode(inode)) {
1302 num_bytes = ALIGN(end - start + 1, blocksize);
1303 num_bytes = max(blocksize, num_bytes);
1304 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1306 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1309 * Due to the page size limit, for subpage we can only trigger the
1310 * writeback for the dirty sectors of page, that means data writeback
1311 * is doing more writeback than what we want.
1313 * This is especially unexpected for some call sites like fallocate,
1314 * where we only increase i_size after everything is done.
1315 * This means we can trigger inline extent even if we didn't want to.
1316 * So here we skip inline extent creation completely.
1318 if (start == 0 && fs_info->sectorsize == PAGE_SIZE && !no_inline) {
1319 u64 actual_end = min_t(u64, i_size_read(&inode->vfs_inode),
1322 /* lets try to make an inline extent */
1323 ret = cow_file_range_inline(inode, actual_end, 0,
1324 BTRFS_COMPRESS_NONE, NULL, false);
1327 * We use DO_ACCOUNTING here because we need the
1328 * delalloc_release_metadata to be run _after_ we drop
1329 * our outstanding extent for clearing delalloc for this
1332 extent_clear_unlock_delalloc(inode, start, end,
1334 EXTENT_LOCKED | EXTENT_DELALLOC |
1335 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1336 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1337 PAGE_START_WRITEBACK | PAGE_END_WRITEBACK);
1339 * locked_page is locked by the caller of
1340 * writepage_delalloc(), not locked by
1341 * __process_pages_contig().
1343 * We can't let __process_pages_contig() to unlock it,
1344 * as it doesn't have any subpage::writers recorded.
1346 * Here we manually unlock the page, since the caller
1347 * can't determine if it's an inline extent or a
1348 * compressed extent.
1350 unlock_page(locked_page);
1353 } else if (ret < 0) {
1358 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1361 * Relocation relies on the relocated extents to have exactly the same
1362 * size as the original extents. Normally writeback for relocation data
1363 * extents follows a NOCOW path because relocation preallocates the
1364 * extents. However, due to an operation such as scrub turning a block
1365 * group to RO mode, it may fallback to COW mode, so we must make sure
1366 * an extent allocated during COW has exactly the requested size and can
1367 * not be split into smaller extents, otherwise relocation breaks and
1368 * fails during the stage where it updates the bytenr of file extent
1371 if (btrfs_is_data_reloc_root(root))
1372 min_alloc_size = num_bytes;
1374 min_alloc_size = fs_info->sectorsize;
1376 while (num_bytes > 0) {
1377 struct btrfs_ordered_extent *ordered;
1379 cur_alloc_size = num_bytes;
1380 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1381 min_alloc_size, 0, alloc_hint,
1383 if (ret == -EAGAIN) {
1385 * btrfs_reserve_extent only returns -EAGAIN for zoned
1386 * file systems, which is an indication that there are
1387 * no active zones to allocate from at the moment.
1389 * If this is the first loop iteration, wait for at
1390 * least one zone to finish before retrying the
1391 * allocation. Otherwise ask the caller to write out
1392 * the already allocated blocks before coming back to
1393 * us, or return -ENOSPC if it can't handle retries.
1395 ASSERT(btrfs_is_zoned(fs_info));
1396 if (start == orig_start) {
1397 wait_on_bit_io(&inode->root->fs_info->flags,
1398 BTRFS_FS_NEED_ZONE_FINISH,
1399 TASK_UNINTERRUPTIBLE);
1403 *done_offset = start - 1;
1410 cur_alloc_size = ins.offset;
1411 extent_reserved = true;
1413 ram_size = ins.offset;
1414 em = create_io_em(inode, start, ins.offset, /* len */
1415 start, /* orig_start */
1416 ins.objectid, /* block_start */
1417 ins.offset, /* block_len */
1418 ins.offset, /* orig_block_len */
1419 ram_size, /* ram_bytes */
1420 BTRFS_COMPRESS_NONE, /* compress_type */
1421 BTRFS_ORDERED_REGULAR /* type */);
1426 free_extent_map(em);
1428 ordered = btrfs_alloc_ordered_extent(inode, start, ram_size,
1429 ram_size, ins.objectid, cur_alloc_size,
1430 0, 1 << BTRFS_ORDERED_REGULAR,
1431 BTRFS_COMPRESS_NONE);
1432 if (IS_ERR(ordered)) {
1433 ret = PTR_ERR(ordered);
1434 goto out_drop_extent_cache;
1437 if (btrfs_is_data_reloc_root(root)) {
1438 ret = btrfs_reloc_clone_csums(ordered);
1441 * Only drop cache here, and process as normal.
1443 * We must not allow extent_clear_unlock_delalloc()
1444 * at out_unlock label to free meta of this ordered
1445 * extent, as its meta should be freed by
1446 * btrfs_finish_ordered_io().
1448 * So we must continue until @start is increased to
1449 * skip current ordered extent.
1452 btrfs_drop_extent_map_range(inode, start,
1453 start + ram_size - 1,
1456 btrfs_put_ordered_extent(ordered);
1458 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1461 * We're not doing compressed IO, don't unlock the first page
1462 * (which the caller expects to stay locked), don't clear any
1463 * dirty bits and don't set any writeback bits
1465 * Do set the Ordered (Private2) bit so we know this page was
1466 * properly setup for writepage.
1468 page_ops = (keep_locked ? 0 : PAGE_UNLOCK);
1469 page_ops |= PAGE_SET_ORDERED;
1471 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1473 EXTENT_LOCKED | EXTENT_DELALLOC,
1475 if (num_bytes < cur_alloc_size)
1478 num_bytes -= cur_alloc_size;
1479 alloc_hint = ins.objectid + ins.offset;
1480 start += cur_alloc_size;
1481 extent_reserved = false;
1484 * btrfs_reloc_clone_csums() error, since start is increased
1485 * extent_clear_unlock_delalloc() at out_unlock label won't
1486 * free metadata of current ordered extent, we're OK to exit.
1496 out_drop_extent_cache:
1497 btrfs_drop_extent_map_range(inode, start, start + ram_size - 1, false);
1499 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1500 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1503 * Now, we have three regions to clean up:
1505 * |-------(1)----|---(2)---|-------------(3)----------|
1506 * `- orig_start `- start `- start + cur_alloc_size `- end
1508 * We process each region below.
1511 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1512 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1513 page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK;
1516 * For the range (1). We have already instantiated the ordered extents
1517 * for this region. They are cleaned up by
1518 * btrfs_cleanup_ordered_extents() in e.g,
1519 * btrfs_run_delalloc_range(). EXTENT_LOCKED | EXTENT_DELALLOC are
1520 * already cleared in the above loop. And, EXTENT_DELALLOC_NEW |
1521 * EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV are handled by the cleanup
1524 * However, in case of @keep_locked, we still need to unlock the pages
1525 * (except @locked_page) to ensure all the pages are unlocked.
1527 if (keep_locked && orig_start < start) {
1529 mapping_set_error(inode->vfs_inode.i_mapping, ret);
1530 extent_clear_unlock_delalloc(inode, orig_start, start - 1,
1531 locked_page, 0, page_ops);
1535 * For the range (2). If we reserved an extent for our delalloc range
1536 * (or a subrange) and failed to create the respective ordered extent,
1537 * then it means that when we reserved the extent we decremented the
1538 * extent's size from the data space_info's bytes_may_use counter and
1539 * incremented the space_info's bytes_reserved counter by the same
1540 * amount. We must make sure extent_clear_unlock_delalloc() does not try
1541 * to decrement again the data space_info's bytes_may_use counter,
1542 * therefore we do not pass it the flag EXTENT_CLEAR_DATA_RESV.
1544 if (extent_reserved) {
1545 extent_clear_unlock_delalloc(inode, start,
1546 start + cur_alloc_size - 1,
1550 start += cur_alloc_size;
1554 * For the range (3). We never touched the region. In addition to the
1555 * clear_bits above, we add EXTENT_CLEAR_DATA_RESV to release the data
1556 * space_info's bytes_may_use counter, reserved in
1557 * btrfs_check_data_free_space().
1560 clear_bits |= EXTENT_CLEAR_DATA_RESV;
1561 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1562 clear_bits, page_ops);
1568 * Phase two of compressed writeback. This is the ordered portion of the code,
1569 * which only gets called in the order the work was queued. We walk all the
1570 * async extents created by compress_file_range and send them down to the disk.
1572 static noinline void submit_compressed_extents(struct btrfs_work *work)
1574 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1576 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1577 struct async_extent *async_extent;
1578 unsigned long nr_pages;
1581 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1584 while (!list_empty(&async_chunk->extents)) {
1585 async_extent = list_entry(async_chunk->extents.next,
1586 struct async_extent, list);
1587 list_del(&async_extent->list);
1588 submit_one_async_extent(async_chunk, async_extent, &alloc_hint);
1591 /* atomic_sub_return implies a barrier */
1592 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1594 cond_wake_up_nomb(&fs_info->async_submit_wait);
1597 static noinline void async_cow_free(struct btrfs_work *work)
1599 struct async_chunk *async_chunk;
1600 struct async_cow *async_cow;
1602 async_chunk = container_of(work, struct async_chunk, work);
1603 btrfs_add_delayed_iput(async_chunk->inode);
1604 if (async_chunk->blkcg_css)
1605 css_put(async_chunk->blkcg_css);
1607 async_cow = async_chunk->async_cow;
1608 if (atomic_dec_and_test(&async_cow->num_chunks))
1612 static bool run_delalloc_compressed(struct btrfs_inode *inode,
1613 struct page *locked_page, u64 start,
1614 u64 end, struct writeback_control *wbc)
1616 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1617 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1618 struct async_cow *ctx;
1619 struct async_chunk *async_chunk;
1620 unsigned long nr_pages;
1621 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1624 const blk_opf_t write_flags = wbc_to_write_flags(wbc);
1626 nofs_flag = memalloc_nofs_save();
1627 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1628 memalloc_nofs_restore(nofs_flag);
1632 unlock_extent(&inode->io_tree, start, end, NULL);
1633 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
1635 async_chunk = ctx->chunks;
1636 atomic_set(&ctx->num_chunks, num_chunks);
1638 for (i = 0; i < num_chunks; i++) {
1639 u64 cur_end = min(end, start + SZ_512K - 1);
1642 * igrab is called higher up in the call chain, take only the
1643 * lightweight reference for the callback lifetime
1645 ihold(&inode->vfs_inode);
1646 async_chunk[i].async_cow = ctx;
1647 async_chunk[i].inode = inode;
1648 async_chunk[i].start = start;
1649 async_chunk[i].end = cur_end;
1650 async_chunk[i].write_flags = write_flags;
1651 INIT_LIST_HEAD(&async_chunk[i].extents);
1654 * The locked_page comes all the way from writepage and its
1655 * the original page we were actually given. As we spread
1656 * this large delalloc region across multiple async_chunk
1657 * structs, only the first struct needs a pointer to locked_page
1659 * This way we don't need racey decisions about who is supposed
1664 * Depending on the compressibility, the pages might or
1665 * might not go through async. We want all of them to
1666 * be accounted against wbc once. Let's do it here
1667 * before the paths diverge. wbc accounting is used
1668 * only for foreign writeback detection and doesn't
1669 * need full accuracy. Just account the whole thing
1670 * against the first page.
1672 wbc_account_cgroup_owner(wbc, locked_page,
1674 async_chunk[i].locked_page = locked_page;
1677 async_chunk[i].locked_page = NULL;
1680 if (blkcg_css != blkcg_root_css) {
1682 async_chunk[i].blkcg_css = blkcg_css;
1683 async_chunk[i].write_flags |= REQ_BTRFS_CGROUP_PUNT;
1685 async_chunk[i].blkcg_css = NULL;
1688 btrfs_init_work(&async_chunk[i].work, compress_file_range,
1689 submit_compressed_extents, async_cow_free);
1691 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1692 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1694 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1696 start = cur_end + 1;
1702 * Run the delalloc range from start to end, and write back any dirty pages
1703 * covered by the range.
1705 static noinline int run_delalloc_cow(struct btrfs_inode *inode,
1706 struct page *locked_page, u64 start,
1707 u64 end, struct writeback_control *wbc,
1710 u64 done_offset = end;
1713 while (start <= end) {
1714 ret = cow_file_range(inode, locked_page, start, end, &done_offset,
1718 extent_write_locked_range(&inode->vfs_inode, locked_page, start,
1719 done_offset, wbc, pages_dirty);
1720 start = done_offset + 1;
1726 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1727 u64 bytenr, u64 num_bytes, bool nowait)
1729 struct btrfs_root *csum_root = btrfs_csum_root(fs_info, bytenr);
1730 struct btrfs_ordered_sum *sums;
1734 ret = btrfs_lookup_csums_list(csum_root, bytenr, bytenr + num_bytes - 1,
1736 if (ret == 0 && list_empty(&list))
1739 while (!list_empty(&list)) {
1740 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1741 list_del(&sums->list);
1749 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1750 const u64 start, const u64 end)
1752 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1753 const bool is_reloc_ino = btrfs_is_data_reloc_root(inode->root);
1754 const u64 range_bytes = end + 1 - start;
1755 struct extent_io_tree *io_tree = &inode->io_tree;
1756 u64 range_start = start;
1761 * If EXTENT_NORESERVE is set it means that when the buffered write was
1762 * made we had not enough available data space and therefore we did not
1763 * reserve data space for it, since we though we could do NOCOW for the
1764 * respective file range (either there is prealloc extent or the inode
1765 * has the NOCOW bit set).
1767 * However when we need to fallback to COW mode (because for example the
1768 * block group for the corresponding extent was turned to RO mode by a
1769 * scrub or relocation) we need to do the following:
1771 * 1) We increment the bytes_may_use counter of the data space info.
1772 * If COW succeeds, it allocates a new data extent and after doing
1773 * that it decrements the space info's bytes_may_use counter and
1774 * increments its bytes_reserved counter by the same amount (we do
1775 * this at btrfs_add_reserved_bytes()). So we need to increment the
1776 * bytes_may_use counter to compensate (when space is reserved at
1777 * buffered write time, the bytes_may_use counter is incremented);
1779 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1780 * that if the COW path fails for any reason, it decrements (through
1781 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1782 * data space info, which we incremented in the step above.
1784 * If we need to fallback to cow and the inode corresponds to a free
1785 * space cache inode or an inode of the data relocation tree, we must
1786 * also increment bytes_may_use of the data space_info for the same
1787 * reason. Space caches and relocated data extents always get a prealloc
1788 * extent for them, however scrub or balance may have set the block
1789 * group that contains that extent to RO mode and therefore force COW
1790 * when starting writeback.
1792 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1793 EXTENT_NORESERVE, 0, NULL);
1794 if (count > 0 || is_space_ino || is_reloc_ino) {
1796 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1797 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1799 if (is_space_ino || is_reloc_ino)
1800 bytes = range_bytes;
1802 spin_lock(&sinfo->lock);
1803 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1804 spin_unlock(&sinfo->lock);
1807 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1812 * Don't try to create inline extents, as a mix of inline extent that
1813 * is written out and unlocked directly and a normal NOCOW extent
1816 ret = cow_file_range(inode, locked_page, start, end, NULL, false, true);
1821 struct can_nocow_file_extent_args {
1824 /* Start file offset of the range we want to NOCOW. */
1826 /* End file offset (inclusive) of the range we want to NOCOW. */
1828 bool writeback_path;
1831 * Free the path passed to can_nocow_file_extent() once it's not needed
1836 /* Output fields. Only set when can_nocow_file_extent() returns 1. */
1841 /* Number of bytes that can be written to in NOCOW mode. */
1846 * Check if we can NOCOW the file extent that the path points to.
1847 * This function may return with the path released, so the caller should check
1848 * if path->nodes[0] is NULL or not if it needs to use the path afterwards.
1850 * Returns: < 0 on error
1851 * 0 if we can not NOCOW
1854 static int can_nocow_file_extent(struct btrfs_path *path,
1855 struct btrfs_key *key,
1856 struct btrfs_inode *inode,
1857 struct can_nocow_file_extent_args *args)
1859 const bool is_freespace_inode = btrfs_is_free_space_inode(inode);
1860 struct extent_buffer *leaf = path->nodes[0];
1861 struct btrfs_root *root = inode->root;
1862 struct btrfs_file_extent_item *fi;
1867 bool nowait = path->nowait;
1869 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
1870 extent_type = btrfs_file_extent_type(leaf, fi);
1872 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1875 /* Can't access these fields unless we know it's not an inline extent. */
1876 args->disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1877 args->disk_num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1878 args->extent_offset = btrfs_file_extent_offset(leaf, fi);
1880 if (!(inode->flags & BTRFS_INODE_NODATACOW) &&
1881 extent_type == BTRFS_FILE_EXTENT_REG)
1885 * If the extent was created before the generation where the last snapshot
1886 * for its subvolume was created, then this implies the extent is shared,
1887 * hence we must COW.
1889 if (!args->strict &&
1890 btrfs_file_extent_generation(leaf, fi) <=
1891 btrfs_root_last_snapshot(&root->root_item))
1894 /* An explicit hole, must COW. */
1895 if (args->disk_bytenr == 0)
1898 /* Compressed/encrypted/encoded extents must be COWed. */
1899 if (btrfs_file_extent_compression(leaf, fi) ||
1900 btrfs_file_extent_encryption(leaf, fi) ||
1901 btrfs_file_extent_other_encoding(leaf, fi))
1904 extent_end = btrfs_file_extent_end(path);
1907 * The following checks can be expensive, as they need to take other
1908 * locks and do btree or rbtree searches, so release the path to avoid
1909 * blocking other tasks for too long.
1911 btrfs_release_path(path);
1913 ret = btrfs_cross_ref_exist(root, btrfs_ino(inode),
1914 key->offset - args->extent_offset,
1915 args->disk_bytenr, args->strict, path);
1916 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
1920 if (args->free_path) {
1922 * We don't need the path anymore, plus through the
1923 * csum_exist_in_range() call below we will end up allocating
1924 * another path. So free the path to avoid unnecessary extra
1927 btrfs_free_path(path);
1931 /* If there are pending snapshots for this root, we must COW. */
1932 if (args->writeback_path && !is_freespace_inode &&
1933 atomic_read(&root->snapshot_force_cow))
1936 args->disk_bytenr += args->extent_offset;
1937 args->disk_bytenr += args->start - key->offset;
1938 args->num_bytes = min(args->end + 1, extent_end) - args->start;
1941 * Force COW if csums exist in the range. This ensures that csums for a
1942 * given extent are either valid or do not exist.
1944 ret = csum_exist_in_range(root->fs_info, args->disk_bytenr, args->num_bytes,
1946 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
1952 if (args->free_path && path)
1953 btrfs_free_path(path);
1955 return ret < 0 ? ret : can_nocow;
1959 * when nowcow writeback call back. This checks for snapshots or COW copies
1960 * of the extents that exist in the file, and COWs the file as required.
1962 * If no cow copies or snapshots exist, we write directly to the existing
1965 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
1966 struct page *locked_page,
1967 const u64 start, const u64 end)
1969 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1970 struct btrfs_root *root = inode->root;
1971 struct btrfs_path *path;
1972 u64 cow_start = (u64)-1;
1973 u64 cur_offset = start;
1975 bool check_prev = true;
1976 u64 ino = btrfs_ino(inode);
1977 struct btrfs_block_group *bg;
1979 struct can_nocow_file_extent_args nocow_args = { 0 };
1981 path = btrfs_alloc_path();
1983 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1984 EXTENT_LOCKED | EXTENT_DELALLOC |
1985 EXTENT_DO_ACCOUNTING |
1986 EXTENT_DEFRAG, PAGE_UNLOCK |
1987 PAGE_START_WRITEBACK |
1988 PAGE_END_WRITEBACK);
1992 nocow_args.end = end;
1993 nocow_args.writeback_path = true;
1996 struct btrfs_ordered_extent *ordered;
1997 struct btrfs_key found_key;
1998 struct btrfs_file_extent_item *fi;
1999 struct extent_buffer *leaf;
2008 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
2014 * If there is no extent for our range when doing the initial
2015 * search, then go back to the previous slot as it will be the
2016 * one containing the search offset
2018 if (ret > 0 && path->slots[0] > 0 && check_prev) {
2019 leaf = path->nodes[0];
2020 btrfs_item_key_to_cpu(leaf, &found_key,
2021 path->slots[0] - 1);
2022 if (found_key.objectid == ino &&
2023 found_key.type == BTRFS_EXTENT_DATA_KEY)
2028 /* Go to next leaf if we have exhausted the current one */
2029 leaf = path->nodes[0];
2030 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2031 ret = btrfs_next_leaf(root, path);
2033 if (cow_start != (u64)-1)
2034 cur_offset = cow_start;
2039 leaf = path->nodes[0];
2042 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2044 /* Didn't find anything for our INO */
2045 if (found_key.objectid > ino)
2048 * Keep searching until we find an EXTENT_ITEM or there are no
2049 * more extents for this inode
2051 if (WARN_ON_ONCE(found_key.objectid < ino) ||
2052 found_key.type < BTRFS_EXTENT_DATA_KEY) {
2057 /* Found key is not EXTENT_DATA_KEY or starts after req range */
2058 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
2059 found_key.offset > end)
2063 * If the found extent starts after requested offset, then
2064 * adjust extent_end to be right before this extent begins
2066 if (found_key.offset > cur_offset) {
2067 extent_end = found_key.offset;
2073 * Found extent which begins before our range and potentially
2076 fi = btrfs_item_ptr(leaf, path->slots[0],
2077 struct btrfs_file_extent_item);
2078 extent_type = btrfs_file_extent_type(leaf, fi);
2079 /* If this is triggered then we have a memory corruption. */
2080 ASSERT(extent_type < BTRFS_NR_FILE_EXTENT_TYPES);
2081 if (WARN_ON(extent_type >= BTRFS_NR_FILE_EXTENT_TYPES)) {
2085 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
2086 extent_end = btrfs_file_extent_end(path);
2089 * If the extent we got ends before our current offset, skip to
2092 if (extent_end <= cur_offset) {
2097 nocow_args.start = cur_offset;
2098 ret = can_nocow_file_extent(path, &found_key, inode, &nocow_args);
2100 if (cow_start != (u64)-1)
2101 cur_offset = cow_start;
2103 } else if (ret == 0) {
2108 bg = btrfs_inc_nocow_writers(fs_info, nocow_args.disk_bytenr);
2113 * If nocow is false then record the beginning of the range
2114 * that needs to be COWed
2117 if (cow_start == (u64)-1)
2118 cow_start = cur_offset;
2119 cur_offset = extent_end;
2120 if (cur_offset > end)
2122 if (!path->nodes[0])
2129 * COW range from cow_start to found_key.offset - 1. As the key
2130 * will contain the beginning of the first extent that can be
2131 * NOCOW, following one which needs to be COW'ed
2133 if (cow_start != (u64)-1) {
2134 ret = fallback_to_cow(inode, locked_page,
2135 cow_start, found_key.offset - 1);
2138 cow_start = (u64)-1;
2141 nocow_end = cur_offset + nocow_args.num_bytes - 1;
2142 is_prealloc = extent_type == BTRFS_FILE_EXTENT_PREALLOC;
2144 u64 orig_start = found_key.offset - nocow_args.extent_offset;
2145 struct extent_map *em;
2147 em = create_io_em(inode, cur_offset, nocow_args.num_bytes,
2149 nocow_args.disk_bytenr, /* block_start */
2150 nocow_args.num_bytes, /* block_len */
2151 nocow_args.disk_num_bytes, /* orig_block_len */
2152 ram_bytes, BTRFS_COMPRESS_NONE,
2153 BTRFS_ORDERED_PREALLOC);
2158 free_extent_map(em);
2161 ordered = btrfs_alloc_ordered_extent(inode, cur_offset,
2162 nocow_args.num_bytes, nocow_args.num_bytes,
2163 nocow_args.disk_bytenr, nocow_args.num_bytes, 0,
2165 ? (1 << BTRFS_ORDERED_PREALLOC)
2166 : (1 << BTRFS_ORDERED_NOCOW),
2167 BTRFS_COMPRESS_NONE);
2168 if (IS_ERR(ordered)) {
2170 btrfs_drop_extent_map_range(inode, cur_offset,
2173 ret = PTR_ERR(ordered);
2178 btrfs_dec_nocow_writers(bg);
2182 if (btrfs_is_data_reloc_root(root))
2184 * Error handled later, as we must prevent
2185 * extent_clear_unlock_delalloc() in error handler
2186 * from freeing metadata of created ordered extent.
2188 ret = btrfs_reloc_clone_csums(ordered);
2189 btrfs_put_ordered_extent(ordered);
2191 extent_clear_unlock_delalloc(inode, cur_offset, nocow_end,
2192 locked_page, EXTENT_LOCKED |
2194 EXTENT_CLEAR_DATA_RESV,
2195 PAGE_UNLOCK | PAGE_SET_ORDERED);
2197 cur_offset = extent_end;
2200 * btrfs_reloc_clone_csums() error, now we're OK to call error
2201 * handler, as metadata for created ordered extent will only
2202 * be freed by btrfs_finish_ordered_io().
2206 if (cur_offset > end)
2209 btrfs_release_path(path);
2211 if (cur_offset <= end && cow_start == (u64)-1)
2212 cow_start = cur_offset;
2214 if (cow_start != (u64)-1) {
2216 ret = fallback_to_cow(inode, locked_page, cow_start, end);
2223 btrfs_dec_nocow_writers(bg);
2225 if (ret && cur_offset < end)
2226 extent_clear_unlock_delalloc(inode, cur_offset, end,
2227 locked_page, EXTENT_LOCKED |
2228 EXTENT_DELALLOC | EXTENT_DEFRAG |
2229 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
2230 PAGE_START_WRITEBACK |
2231 PAGE_END_WRITEBACK);
2232 btrfs_free_path(path);
2236 static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end)
2238 if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) {
2239 if (inode->defrag_bytes &&
2240 test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG,
2249 * Function to process delayed allocation (create CoW) for ranges which are
2250 * being touched for the first time.
2252 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
2253 u64 start, u64 end, struct writeback_control *wbc)
2255 const bool zoned = btrfs_is_zoned(inode->root->fs_info);
2259 * The range must cover part of the @locked_page, or a return of 1
2260 * can confuse the caller.
2262 ASSERT(!(end <= page_offset(locked_page) ||
2263 start >= page_offset(locked_page) + PAGE_SIZE));
2265 if (should_nocow(inode, start, end)) {
2267 * Normally on a zoned device we're only doing COW writes, but
2268 * in case of relocation on a zoned filesystem we have taken
2269 * precaution, that we're only writing sequentially. It's safe
2270 * to use run_delalloc_nocow() here, like for regular
2271 * preallocated inodes.
2273 ASSERT(!zoned || btrfs_is_data_reloc_root(inode->root));
2274 ret = run_delalloc_nocow(inode, locked_page, start, end);
2278 if (btrfs_inode_can_compress(inode) &&
2279 inode_need_compress(inode, start, end) &&
2280 run_delalloc_compressed(inode, locked_page, start, end, wbc))
2284 ret = run_delalloc_cow(inode, locked_page, start, end, wbc,
2287 ret = cow_file_range(inode, locked_page, start, end, NULL,
2292 btrfs_cleanup_ordered_extents(inode, locked_page, start,
2297 void btrfs_split_delalloc_extent(struct btrfs_inode *inode,
2298 struct extent_state *orig, u64 split)
2300 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2303 /* not delalloc, ignore it */
2304 if (!(orig->state & EXTENT_DELALLOC))
2307 size = orig->end - orig->start + 1;
2308 if (size > fs_info->max_extent_size) {
2313 * See the explanation in btrfs_merge_delalloc_extent, the same
2314 * applies here, just in reverse.
2316 new_size = orig->end - split + 1;
2317 num_extents = count_max_extents(fs_info, new_size);
2318 new_size = split - orig->start;
2319 num_extents += count_max_extents(fs_info, new_size);
2320 if (count_max_extents(fs_info, size) >= num_extents)
2324 spin_lock(&inode->lock);
2325 btrfs_mod_outstanding_extents(inode, 1);
2326 spin_unlock(&inode->lock);
2330 * Handle merged delayed allocation extents so we can keep track of new extents
2331 * that are just merged onto old extents, such as when we are doing sequential
2332 * writes, so we can properly account for the metadata space we'll need.
2334 void btrfs_merge_delalloc_extent(struct btrfs_inode *inode, struct extent_state *new,
2335 struct extent_state *other)
2337 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2338 u64 new_size, old_size;
2341 /* not delalloc, ignore it */
2342 if (!(other->state & EXTENT_DELALLOC))
2345 if (new->start > other->start)
2346 new_size = new->end - other->start + 1;
2348 new_size = other->end - new->start + 1;
2350 /* we're not bigger than the max, unreserve the space and go */
2351 if (new_size <= fs_info->max_extent_size) {
2352 spin_lock(&inode->lock);
2353 btrfs_mod_outstanding_extents(inode, -1);
2354 spin_unlock(&inode->lock);
2359 * We have to add up either side to figure out how many extents were
2360 * accounted for before we merged into one big extent. If the number of
2361 * extents we accounted for is <= the amount we need for the new range
2362 * then we can return, otherwise drop. Think of it like this
2366 * So we've grown the extent by a MAX_SIZE extent, this would mean we
2367 * need 2 outstanding extents, on one side we have 1 and the other side
2368 * we have 1 so they are == and we can return. But in this case
2370 * [MAX_SIZE+4k][MAX_SIZE+4k]
2372 * Each range on their own accounts for 2 extents, but merged together
2373 * they are only 3 extents worth of accounting, so we need to drop in
2376 old_size = other->end - other->start + 1;
2377 num_extents = count_max_extents(fs_info, old_size);
2378 old_size = new->end - new->start + 1;
2379 num_extents += count_max_extents(fs_info, old_size);
2380 if (count_max_extents(fs_info, new_size) >= num_extents)
2383 spin_lock(&inode->lock);
2384 btrfs_mod_outstanding_extents(inode, -1);
2385 spin_unlock(&inode->lock);
2388 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
2389 struct btrfs_inode *inode)
2391 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2393 spin_lock(&root->delalloc_lock);
2394 if (list_empty(&inode->delalloc_inodes)) {
2395 list_add_tail(&inode->delalloc_inodes, &root->delalloc_inodes);
2396 set_bit(BTRFS_INODE_IN_DELALLOC_LIST, &inode->runtime_flags);
2397 root->nr_delalloc_inodes++;
2398 if (root->nr_delalloc_inodes == 1) {
2399 spin_lock(&fs_info->delalloc_root_lock);
2400 BUG_ON(!list_empty(&root->delalloc_root));
2401 list_add_tail(&root->delalloc_root,
2402 &fs_info->delalloc_roots);
2403 spin_unlock(&fs_info->delalloc_root_lock);
2406 spin_unlock(&root->delalloc_lock);
2409 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
2410 struct btrfs_inode *inode)
2412 struct btrfs_fs_info *fs_info = root->fs_info;
2414 if (!list_empty(&inode->delalloc_inodes)) {
2415 list_del_init(&inode->delalloc_inodes);
2416 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2417 &inode->runtime_flags);
2418 root->nr_delalloc_inodes--;
2419 if (!root->nr_delalloc_inodes) {
2420 ASSERT(list_empty(&root->delalloc_inodes));
2421 spin_lock(&fs_info->delalloc_root_lock);
2422 BUG_ON(list_empty(&root->delalloc_root));
2423 list_del_init(&root->delalloc_root);
2424 spin_unlock(&fs_info->delalloc_root_lock);
2429 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
2430 struct btrfs_inode *inode)
2432 spin_lock(&root->delalloc_lock);
2433 __btrfs_del_delalloc_inode(root, inode);
2434 spin_unlock(&root->delalloc_lock);
2438 * Properly track delayed allocation bytes in the inode and to maintain the
2439 * list of inodes that have pending delalloc work to be done.
2441 void btrfs_set_delalloc_extent(struct btrfs_inode *inode, struct extent_state *state,
2444 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2446 if ((bits & EXTENT_DEFRAG) && !(bits & EXTENT_DELALLOC))
2449 * set_bit and clear bit hooks normally require _irqsave/restore
2450 * but in this case, we are only testing for the DELALLOC
2451 * bit, which is only set or cleared with irqs on
2453 if (!(state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2454 struct btrfs_root *root = inode->root;
2455 u64 len = state->end + 1 - state->start;
2456 u32 num_extents = count_max_extents(fs_info, len);
2457 bool do_list = !btrfs_is_free_space_inode(inode);
2459 spin_lock(&inode->lock);
2460 btrfs_mod_outstanding_extents(inode, num_extents);
2461 spin_unlock(&inode->lock);
2463 /* For sanity tests */
2464 if (btrfs_is_testing(fs_info))
2467 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2468 fs_info->delalloc_batch);
2469 spin_lock(&inode->lock);
2470 inode->delalloc_bytes += len;
2471 if (bits & EXTENT_DEFRAG)
2472 inode->defrag_bytes += len;
2473 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2474 &inode->runtime_flags))
2475 btrfs_add_delalloc_inodes(root, inode);
2476 spin_unlock(&inode->lock);
2479 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2480 (bits & EXTENT_DELALLOC_NEW)) {
2481 spin_lock(&inode->lock);
2482 inode->new_delalloc_bytes += state->end + 1 - state->start;
2483 spin_unlock(&inode->lock);
2488 * Once a range is no longer delalloc this function ensures that proper
2489 * accounting happens.
2491 void btrfs_clear_delalloc_extent(struct btrfs_inode *inode,
2492 struct extent_state *state, u32 bits)
2494 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2495 u64 len = state->end + 1 - state->start;
2496 u32 num_extents = count_max_extents(fs_info, len);
2498 if ((state->state & EXTENT_DEFRAG) && (bits & EXTENT_DEFRAG)) {
2499 spin_lock(&inode->lock);
2500 inode->defrag_bytes -= len;
2501 spin_unlock(&inode->lock);
2505 * set_bit and clear bit hooks normally require _irqsave/restore
2506 * but in this case, we are only testing for the DELALLOC
2507 * bit, which is only set or cleared with irqs on
2509 if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2510 struct btrfs_root *root = inode->root;
2511 bool do_list = !btrfs_is_free_space_inode(inode);
2513 spin_lock(&inode->lock);
2514 btrfs_mod_outstanding_extents(inode, -num_extents);
2515 spin_unlock(&inode->lock);
2518 * We don't reserve metadata space for space cache inodes so we
2519 * don't need to call delalloc_release_metadata if there is an
2522 if (bits & EXTENT_CLEAR_META_RESV &&
2523 root != fs_info->tree_root)
2524 btrfs_delalloc_release_metadata(inode, len, false);
2526 /* For sanity tests. */
2527 if (btrfs_is_testing(fs_info))
2530 if (!btrfs_is_data_reloc_root(root) &&
2531 do_list && !(state->state & EXTENT_NORESERVE) &&
2532 (bits & EXTENT_CLEAR_DATA_RESV))
2533 btrfs_free_reserved_data_space_noquota(fs_info, len);
2535 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2536 fs_info->delalloc_batch);
2537 spin_lock(&inode->lock);
2538 inode->delalloc_bytes -= len;
2539 if (do_list && inode->delalloc_bytes == 0 &&
2540 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2541 &inode->runtime_flags))
2542 btrfs_del_delalloc_inode(root, inode);
2543 spin_unlock(&inode->lock);
2546 if ((state->state & EXTENT_DELALLOC_NEW) &&
2547 (bits & EXTENT_DELALLOC_NEW)) {
2548 spin_lock(&inode->lock);
2549 ASSERT(inode->new_delalloc_bytes >= len);
2550 inode->new_delalloc_bytes -= len;
2551 if (bits & EXTENT_ADD_INODE_BYTES)
2552 inode_add_bytes(&inode->vfs_inode, len);
2553 spin_unlock(&inode->lock);
2557 static int btrfs_extract_ordered_extent(struct btrfs_bio *bbio,
2558 struct btrfs_ordered_extent *ordered)
2560 u64 start = (u64)bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT;
2561 u64 len = bbio->bio.bi_iter.bi_size;
2562 struct btrfs_ordered_extent *new;
2565 /* Must always be called for the beginning of an ordered extent. */
2566 if (WARN_ON_ONCE(start != ordered->disk_bytenr))
2569 /* No need to split if the ordered extent covers the entire bio. */
2570 if (ordered->disk_num_bytes == len) {
2571 refcount_inc(&ordered->refs);
2572 bbio->ordered = ordered;
2577 * Don't split the extent_map for NOCOW extents, as we're writing into
2578 * a pre-existing one.
2580 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
2581 ret = split_extent_map(bbio->inode, bbio->file_offset,
2582 ordered->num_bytes, len,
2583 ordered->disk_bytenr);
2588 new = btrfs_split_ordered_extent(ordered, len);
2590 return PTR_ERR(new);
2591 bbio->ordered = new;
2596 * given a list of ordered sums record them in the inode. This happens
2597 * at IO completion time based on sums calculated at bio submission time.
2599 static int add_pending_csums(struct btrfs_trans_handle *trans,
2600 struct list_head *list)
2602 struct btrfs_ordered_sum *sum;
2603 struct btrfs_root *csum_root = NULL;
2606 list_for_each_entry(sum, list, list) {
2607 trans->adding_csums = true;
2609 csum_root = btrfs_csum_root(trans->fs_info,
2611 ret = btrfs_csum_file_blocks(trans, csum_root, sum);
2612 trans->adding_csums = false;
2619 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2622 struct extent_state **cached_state)
2624 u64 search_start = start;
2625 const u64 end = start + len - 1;
2627 while (search_start < end) {
2628 const u64 search_len = end - search_start + 1;
2629 struct extent_map *em;
2633 em = btrfs_get_extent(inode, NULL, 0, search_start, search_len);
2637 if (em->block_start != EXTENT_MAP_HOLE)
2641 if (em->start < search_start)
2642 em_len -= search_start - em->start;
2643 if (em_len > search_len)
2644 em_len = search_len;
2646 ret = set_extent_bit(&inode->io_tree, search_start,
2647 search_start + em_len - 1,
2648 EXTENT_DELALLOC_NEW, cached_state);
2650 search_start = extent_map_end(em);
2651 free_extent_map(em);
2658 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2659 unsigned int extra_bits,
2660 struct extent_state **cached_state)
2662 WARN_ON(PAGE_ALIGNED(end));
2664 if (start >= i_size_read(&inode->vfs_inode) &&
2665 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2667 * There can't be any extents following eof in this case so just
2668 * set the delalloc new bit for the range directly.
2670 extra_bits |= EXTENT_DELALLOC_NEW;
2674 ret = btrfs_find_new_delalloc_bytes(inode, start,
2681 return set_extent_bit(&inode->io_tree, start, end,
2682 EXTENT_DELALLOC | extra_bits, cached_state);
2685 /* see btrfs_writepage_start_hook for details on why this is required */
2686 struct btrfs_writepage_fixup {
2688 struct btrfs_inode *inode;
2689 struct btrfs_work work;
2692 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2694 struct btrfs_writepage_fixup *fixup =
2695 container_of(work, struct btrfs_writepage_fixup, work);
2696 struct btrfs_ordered_extent *ordered;
2697 struct extent_state *cached_state = NULL;
2698 struct extent_changeset *data_reserved = NULL;
2699 struct page *page = fixup->page;
2700 struct btrfs_inode *inode = fixup->inode;
2701 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2702 u64 page_start = page_offset(page);
2703 u64 page_end = page_offset(page) + PAGE_SIZE - 1;
2705 bool free_delalloc_space = true;
2708 * This is similar to page_mkwrite, we need to reserve the space before
2709 * we take the page lock.
2711 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2717 * Before we queued this fixup, we took a reference on the page.
2718 * page->mapping may go NULL, but it shouldn't be moved to a different
2721 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2723 * Unfortunately this is a little tricky, either
2725 * 1) We got here and our page had already been dealt with and
2726 * we reserved our space, thus ret == 0, so we need to just
2727 * drop our space reservation and bail. This can happen the
2728 * first time we come into the fixup worker, or could happen
2729 * while waiting for the ordered extent.
2730 * 2) Our page was already dealt with, but we happened to get an
2731 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2732 * this case we obviously don't have anything to release, but
2733 * because the page was already dealt with we don't want to
2734 * mark the page with an error, so make sure we're resetting
2735 * ret to 0. This is why we have this check _before_ the ret
2736 * check, because we do not want to have a surprise ENOSPC
2737 * when the page was already properly dealt with.
2740 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2741 btrfs_delalloc_release_space(inode, data_reserved,
2742 page_start, PAGE_SIZE,
2750 * We can't mess with the page state unless it is locked, so now that
2751 * it is locked bail if we failed to make our space reservation.
2756 lock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2758 /* already ordered? We're done */
2759 if (PageOrdered(page))
2762 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2764 unlock_extent(&inode->io_tree, page_start, page_end,
2767 btrfs_start_ordered_extent(ordered);
2768 btrfs_put_ordered_extent(ordered);
2772 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2778 * Everything went as planned, we're now the owner of a dirty page with
2779 * delayed allocation bits set and space reserved for our COW
2782 * The page was dirty when we started, nothing should have cleaned it.
2784 BUG_ON(!PageDirty(page));
2785 free_delalloc_space = false;
2787 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2788 if (free_delalloc_space)
2789 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2791 unlock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2795 * We hit ENOSPC or other errors. Update the mapping and page
2796 * to reflect the errors and clean the page.
2798 mapping_set_error(page->mapping, ret);
2799 btrfs_mark_ordered_io_finished(inode, page, page_start,
2801 btrfs_page_clear_uptodate(fs_info, page, page_start, PAGE_SIZE);
2802 clear_page_dirty_for_io(page);
2804 btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE);
2808 extent_changeset_free(data_reserved);
2810 * As a precaution, do a delayed iput in case it would be the last iput
2811 * that could need flushing space. Recursing back to fixup worker would
2814 btrfs_add_delayed_iput(inode);
2818 * There are a few paths in the higher layers of the kernel that directly
2819 * set the page dirty bit without asking the filesystem if it is a
2820 * good idea. This causes problems because we want to make sure COW
2821 * properly happens and the data=ordered rules are followed.
2823 * In our case any range that doesn't have the ORDERED bit set
2824 * hasn't been properly setup for IO. We kick off an async process
2825 * to fix it up. The async helper will wait for ordered extents, set
2826 * the delalloc bit and make it safe to write the page.
2828 int btrfs_writepage_cow_fixup(struct page *page)
2830 struct inode *inode = page->mapping->host;
2831 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2832 struct btrfs_writepage_fixup *fixup;
2834 /* This page has ordered extent covering it already */
2835 if (PageOrdered(page))
2839 * PageChecked is set below when we create a fixup worker for this page,
2840 * don't try to create another one if we're already PageChecked()
2842 * The extent_io writepage code will redirty the page if we send back
2845 if (PageChecked(page))
2848 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2853 * We are already holding a reference to this inode from
2854 * write_cache_pages. We need to hold it because the space reservation
2855 * takes place outside of the page lock, and we can't trust
2856 * page->mapping outside of the page lock.
2859 btrfs_page_set_checked(fs_info, page, page_offset(page), PAGE_SIZE);
2861 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
2863 fixup->inode = BTRFS_I(inode);
2864 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2869 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2870 struct btrfs_inode *inode, u64 file_pos,
2871 struct btrfs_file_extent_item *stack_fi,
2872 const bool update_inode_bytes,
2873 u64 qgroup_reserved)
2875 struct btrfs_root *root = inode->root;
2876 const u64 sectorsize = root->fs_info->sectorsize;
2877 struct btrfs_path *path;
2878 struct extent_buffer *leaf;
2879 struct btrfs_key ins;
2880 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
2881 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
2882 u64 offset = btrfs_stack_file_extent_offset(stack_fi);
2883 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
2884 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
2885 struct btrfs_drop_extents_args drop_args = { 0 };
2888 path = btrfs_alloc_path();
2893 * we may be replacing one extent in the tree with another.
2894 * The new extent is pinned in the extent map, and we don't want
2895 * to drop it from the cache until it is completely in the btree.
2897 * So, tell btrfs_drop_extents to leave this extent in the cache.
2898 * the caller is expected to unpin it and allow it to be merged
2901 drop_args.path = path;
2902 drop_args.start = file_pos;
2903 drop_args.end = file_pos + num_bytes;
2904 drop_args.replace_extent = true;
2905 drop_args.extent_item_size = sizeof(*stack_fi);
2906 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2910 if (!drop_args.extent_inserted) {
2911 ins.objectid = btrfs_ino(inode);
2912 ins.offset = file_pos;
2913 ins.type = BTRFS_EXTENT_DATA_KEY;
2915 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2920 leaf = path->nodes[0];
2921 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
2922 write_extent_buffer(leaf, stack_fi,
2923 btrfs_item_ptr_offset(leaf, path->slots[0]),
2924 sizeof(struct btrfs_file_extent_item));
2926 btrfs_mark_buffer_dirty(leaf);
2927 btrfs_release_path(path);
2930 * If we dropped an inline extent here, we know the range where it is
2931 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
2932 * number of bytes only for that range containing the inline extent.
2933 * The remaining of the range will be processed when clearning the
2934 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
2936 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
2937 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
2939 inline_size = drop_args.bytes_found - inline_size;
2940 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
2941 drop_args.bytes_found -= inline_size;
2942 num_bytes -= sectorsize;
2945 if (update_inode_bytes)
2946 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
2948 ins.objectid = disk_bytenr;
2949 ins.offset = disk_num_bytes;
2950 ins.type = BTRFS_EXTENT_ITEM_KEY;
2952 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
2956 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
2958 qgroup_reserved, &ins);
2960 btrfs_free_path(path);
2965 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2968 struct btrfs_block_group *cache;
2970 cache = btrfs_lookup_block_group(fs_info, start);
2973 spin_lock(&cache->lock);
2974 cache->delalloc_bytes -= len;
2975 spin_unlock(&cache->lock);
2977 btrfs_put_block_group(cache);
2980 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
2981 struct btrfs_ordered_extent *oe)
2983 struct btrfs_file_extent_item stack_fi;
2984 bool update_inode_bytes;
2985 u64 num_bytes = oe->num_bytes;
2986 u64 ram_bytes = oe->ram_bytes;
2988 memset(&stack_fi, 0, sizeof(stack_fi));
2989 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
2990 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
2991 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
2992 oe->disk_num_bytes);
2993 btrfs_set_stack_file_extent_offset(&stack_fi, oe->offset);
2994 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags)) {
2995 num_bytes = oe->truncated_len;
2996 ram_bytes = num_bytes;
2998 btrfs_set_stack_file_extent_num_bytes(&stack_fi, num_bytes);
2999 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, ram_bytes);
3000 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
3001 /* Encryption and other encoding is reserved and all 0 */
3004 * For delalloc, when completing an ordered extent we update the inode's
3005 * bytes when clearing the range in the inode's io tree, so pass false
3006 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
3007 * except if the ordered extent was truncated.
3009 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
3010 test_bit(BTRFS_ORDERED_ENCODED, &oe->flags) ||
3011 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
3013 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
3014 oe->file_offset, &stack_fi,
3015 update_inode_bytes, oe->qgroup_rsv);
3019 * As ordered data IO finishes, this gets called so we can finish
3020 * an ordered extent if the range of bytes in the file it covers are
3023 int btrfs_finish_one_ordered(struct btrfs_ordered_extent *ordered_extent)
3025 struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
3026 struct btrfs_root *root = inode->root;
3027 struct btrfs_fs_info *fs_info = root->fs_info;
3028 struct btrfs_trans_handle *trans = NULL;
3029 struct extent_io_tree *io_tree = &inode->io_tree;
3030 struct extent_state *cached_state = NULL;
3032 int compress_type = 0;
3034 u64 logical_len = ordered_extent->num_bytes;
3035 bool freespace_inode;
3036 bool truncated = false;
3037 bool clear_reserved_extent = true;
3038 unsigned int clear_bits = EXTENT_DEFRAG;
3040 start = ordered_extent->file_offset;
3041 end = start + ordered_extent->num_bytes - 1;
3043 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3044 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
3045 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags) &&
3046 !test_bit(BTRFS_ORDERED_ENCODED, &ordered_extent->flags))
3047 clear_bits |= EXTENT_DELALLOC_NEW;
3049 freespace_inode = btrfs_is_free_space_inode(inode);
3050 if (!freespace_inode)
3051 btrfs_lockdep_acquire(fs_info, btrfs_ordered_extent);
3053 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
3058 if (btrfs_is_zoned(fs_info))
3059 btrfs_zone_finish_endio(fs_info, ordered_extent->disk_bytenr,
3060 ordered_extent->disk_num_bytes);
3062 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
3064 logical_len = ordered_extent->truncated_len;
3065 /* Truncated the entire extent, don't bother adding */
3070 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
3071 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
3073 btrfs_inode_safe_disk_i_size_write(inode, 0);
3074 if (freespace_inode)
3075 trans = btrfs_join_transaction_spacecache(root);
3077 trans = btrfs_join_transaction(root);
3078 if (IS_ERR(trans)) {
3079 ret = PTR_ERR(trans);
3083 trans->block_rsv = &inode->block_rsv;
3084 ret = btrfs_update_inode_fallback(trans, root, inode);
3085 if (ret) /* -ENOMEM or corruption */
3086 btrfs_abort_transaction(trans, ret);
3090 clear_bits |= EXTENT_LOCKED;
3091 lock_extent(io_tree, start, end, &cached_state);
3093 if (freespace_inode)
3094 trans = btrfs_join_transaction_spacecache(root);
3096 trans = btrfs_join_transaction(root);
3097 if (IS_ERR(trans)) {
3098 ret = PTR_ERR(trans);
3103 trans->block_rsv = &inode->block_rsv;
3105 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3106 compress_type = ordered_extent->compress_type;
3107 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3108 BUG_ON(compress_type);
3109 ret = btrfs_mark_extent_written(trans, inode,
3110 ordered_extent->file_offset,
3111 ordered_extent->file_offset +
3113 btrfs_zoned_release_data_reloc_bg(fs_info, ordered_extent->disk_bytenr,
3114 ordered_extent->disk_num_bytes);
3116 BUG_ON(root == fs_info->tree_root);
3117 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
3119 clear_reserved_extent = false;
3120 btrfs_release_delalloc_bytes(fs_info,
3121 ordered_extent->disk_bytenr,
3122 ordered_extent->disk_num_bytes);
3125 unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset,
3126 ordered_extent->num_bytes, trans->transid);
3128 btrfs_abort_transaction(trans, ret);
3132 ret = add_pending_csums(trans, &ordered_extent->list);
3134 btrfs_abort_transaction(trans, ret);
3139 * If this is a new delalloc range, clear its new delalloc flag to
3140 * update the inode's number of bytes. This needs to be done first
3141 * before updating the inode item.
3143 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
3144 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
3145 clear_extent_bit(&inode->io_tree, start, end,
3146 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
3149 btrfs_inode_safe_disk_i_size_write(inode, 0);
3150 ret = btrfs_update_inode_fallback(trans, root, inode);
3151 if (ret) { /* -ENOMEM or corruption */
3152 btrfs_abort_transaction(trans, ret);
3157 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
3161 btrfs_end_transaction(trans);
3163 if (ret || truncated) {
3164 u64 unwritten_start = start;
3167 * If we failed to finish this ordered extent for any reason we
3168 * need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3169 * extent, and mark the inode with the error if it wasn't
3170 * already set. Any error during writeback would have already
3171 * set the mapping error, so we need to set it if we're the ones
3172 * marking this ordered extent as failed.
3174 if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR,
3175 &ordered_extent->flags))
3176 mapping_set_error(ordered_extent->inode->i_mapping, -EIO);
3179 unwritten_start += logical_len;
3180 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
3182 /* Drop extent maps for the part of the extent we didn't write. */
3183 btrfs_drop_extent_map_range(inode, unwritten_start, end, false);
3186 * If the ordered extent had an IOERR or something else went
3187 * wrong we need to return the space for this ordered extent
3188 * back to the allocator. We only free the extent in the
3189 * truncated case if we didn't write out the extent at all.
3191 * If we made it past insert_reserved_file_extent before we
3192 * errored out then we don't need to do this as the accounting
3193 * has already been done.
3195 if ((ret || !logical_len) &&
3196 clear_reserved_extent &&
3197 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3198 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3200 * Discard the range before returning it back to the
3203 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3204 btrfs_discard_extent(fs_info,
3205 ordered_extent->disk_bytenr,
3206 ordered_extent->disk_num_bytes,
3208 btrfs_free_reserved_extent(fs_info,
3209 ordered_extent->disk_bytenr,
3210 ordered_extent->disk_num_bytes, 1);
3212 * Actually free the qgroup rsv which was released when
3213 * the ordered extent was created.
3215 btrfs_qgroup_free_refroot(fs_info, inode->root->root_key.objectid,
3216 ordered_extent->qgroup_rsv,
3217 BTRFS_QGROUP_RSV_DATA);
3222 * This needs to be done to make sure anybody waiting knows we are done
3223 * updating everything for this ordered extent.
3225 btrfs_remove_ordered_extent(inode, ordered_extent);
3228 btrfs_put_ordered_extent(ordered_extent);
3229 /* once for the tree */
3230 btrfs_put_ordered_extent(ordered_extent);
3235 int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered)
3237 if (btrfs_is_zoned(btrfs_sb(ordered->inode->i_sb)) &&
3238 !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
3239 btrfs_finish_ordered_zoned(ordered);
3240 return btrfs_finish_one_ordered(ordered);
3244 * Verify the checksum for a single sector without any extra action that depend
3245 * on the type of I/O.
3247 int btrfs_check_sector_csum(struct btrfs_fs_info *fs_info, struct page *page,
3248 u32 pgoff, u8 *csum, const u8 * const csum_expected)
3250 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3253 ASSERT(pgoff + fs_info->sectorsize <= PAGE_SIZE);
3255 shash->tfm = fs_info->csum_shash;
3257 kaddr = kmap_local_page(page) + pgoff;
3258 crypto_shash_digest(shash, kaddr, fs_info->sectorsize, csum);
3259 kunmap_local(kaddr);
3261 if (memcmp(csum, csum_expected, fs_info->csum_size))
3267 * Verify the checksum of a single data sector.
3269 * @bbio: btrfs_io_bio which contains the csum
3270 * @dev: device the sector is on
3271 * @bio_offset: offset to the beginning of the bio (in bytes)
3272 * @bv: bio_vec to check
3274 * Check if the checksum on a data block is valid. When a checksum mismatch is
3275 * detected, report the error and fill the corrupted range with zero.
3277 * Return %true if the sector is ok or had no checksum to start with, else %false.
3279 bool btrfs_data_csum_ok(struct btrfs_bio *bbio, struct btrfs_device *dev,
3280 u32 bio_offset, struct bio_vec *bv)
3282 struct btrfs_inode *inode = bbio->inode;
3283 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3284 u64 file_offset = bbio->file_offset + bio_offset;
3285 u64 end = file_offset + bv->bv_len - 1;
3287 u8 csum[BTRFS_CSUM_SIZE];
3289 ASSERT(bv->bv_len == fs_info->sectorsize);
3294 if (btrfs_is_data_reloc_root(inode->root) &&
3295 test_range_bit(&inode->io_tree, file_offset, end, EXTENT_NODATASUM,
3297 /* Skip the range without csum for data reloc inode */
3298 clear_extent_bits(&inode->io_tree, file_offset, end,
3303 csum_expected = bbio->csum + (bio_offset >> fs_info->sectorsize_bits) *
3305 if (btrfs_check_sector_csum(fs_info, bv->bv_page, bv->bv_offset, csum,
3311 btrfs_print_data_csum_error(inode, file_offset, csum, csum_expected,
3314 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS);
3320 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3322 * @inode: The inode we want to perform iput on
3324 * This function uses the generic vfs_inode::i_count to track whether we should
3325 * just decrement it (in case it's > 1) or if this is the last iput then link
3326 * the inode to the delayed iput machinery. Delayed iputs are processed at
3327 * transaction commit time/superblock commit/cleaner kthread.
3329 void btrfs_add_delayed_iput(struct btrfs_inode *inode)
3331 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3332 unsigned long flags;
3334 if (atomic_add_unless(&inode->vfs_inode.i_count, -1, 1))
3337 atomic_inc(&fs_info->nr_delayed_iputs);
3339 * Need to be irq safe here because we can be called from either an irq
3340 * context (see bio.c and btrfs_put_ordered_extent()) or a non-irq
3343 spin_lock_irqsave(&fs_info->delayed_iput_lock, flags);
3344 ASSERT(list_empty(&inode->delayed_iput));
3345 list_add_tail(&inode->delayed_iput, &fs_info->delayed_iputs);
3346 spin_unlock_irqrestore(&fs_info->delayed_iput_lock, flags);
3347 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3348 wake_up_process(fs_info->cleaner_kthread);
3351 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3352 struct btrfs_inode *inode)
3354 list_del_init(&inode->delayed_iput);
3355 spin_unlock_irq(&fs_info->delayed_iput_lock);
3356 iput(&inode->vfs_inode);
3357 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3358 wake_up(&fs_info->delayed_iputs_wait);
3359 spin_lock_irq(&fs_info->delayed_iput_lock);
3362 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3363 struct btrfs_inode *inode)
3365 if (!list_empty(&inode->delayed_iput)) {
3366 spin_lock_irq(&fs_info->delayed_iput_lock);
3367 if (!list_empty(&inode->delayed_iput))
3368 run_delayed_iput_locked(fs_info, inode);
3369 spin_unlock_irq(&fs_info->delayed_iput_lock);
3373 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3376 * btrfs_put_ordered_extent() can run in irq context (see bio.c), which
3377 * calls btrfs_add_delayed_iput() and that needs to lock
3378 * fs_info->delayed_iput_lock. So we need to disable irqs here to
3379 * prevent a deadlock.
3381 spin_lock_irq(&fs_info->delayed_iput_lock);
3382 while (!list_empty(&fs_info->delayed_iputs)) {
3383 struct btrfs_inode *inode;
3385 inode = list_first_entry(&fs_info->delayed_iputs,
3386 struct btrfs_inode, delayed_iput);
3387 run_delayed_iput_locked(fs_info, inode);
3388 if (need_resched()) {
3389 spin_unlock_irq(&fs_info->delayed_iput_lock);
3391 spin_lock_irq(&fs_info->delayed_iput_lock);
3394 spin_unlock_irq(&fs_info->delayed_iput_lock);
3398 * Wait for flushing all delayed iputs
3400 * @fs_info: the filesystem
3402 * This will wait on any delayed iputs that are currently running with KILLABLE
3403 * set. Once they are all done running we will return, unless we are killed in
3404 * which case we return EINTR. This helps in user operations like fallocate etc
3405 * that might get blocked on the iputs.
3407 * Return EINTR if we were killed, 0 if nothing's pending
3409 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3411 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3412 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3419 * This creates an orphan entry for the given inode in case something goes wrong
3420 * in the middle of an unlink.
3422 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3423 struct btrfs_inode *inode)
3427 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3428 if (ret && ret != -EEXIST) {
3429 btrfs_abort_transaction(trans, ret);
3437 * We have done the delete so we can go ahead and remove the orphan item for
3438 * this particular inode.
3440 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3441 struct btrfs_inode *inode)
3443 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3447 * this cleans up any orphans that may be left on the list from the last use
3450 int btrfs_orphan_cleanup(struct btrfs_root *root)
3452 struct btrfs_fs_info *fs_info = root->fs_info;
3453 struct btrfs_path *path;
3454 struct extent_buffer *leaf;
3455 struct btrfs_key key, found_key;
3456 struct btrfs_trans_handle *trans;
3457 struct inode *inode;
3458 u64 last_objectid = 0;
3459 int ret = 0, nr_unlink = 0;
3461 if (test_and_set_bit(BTRFS_ROOT_ORPHAN_CLEANUP, &root->state))
3464 path = btrfs_alloc_path();
3469 path->reada = READA_BACK;
3471 key.objectid = BTRFS_ORPHAN_OBJECTID;
3472 key.type = BTRFS_ORPHAN_ITEM_KEY;
3473 key.offset = (u64)-1;
3476 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3481 * if ret == 0 means we found what we were searching for, which
3482 * is weird, but possible, so only screw with path if we didn't
3483 * find the key and see if we have stuff that matches
3487 if (path->slots[0] == 0)
3492 /* pull out the item */
3493 leaf = path->nodes[0];
3494 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3496 /* make sure the item matches what we want */
3497 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3499 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3502 /* release the path since we're done with it */
3503 btrfs_release_path(path);
3506 * this is where we are basically btrfs_lookup, without the
3507 * crossing root thing. we store the inode number in the
3508 * offset of the orphan item.
3511 if (found_key.offset == last_objectid) {
3513 "Error removing orphan entry, stopping orphan cleanup");
3518 last_objectid = found_key.offset;
3520 found_key.objectid = found_key.offset;
3521 found_key.type = BTRFS_INODE_ITEM_KEY;
3522 found_key.offset = 0;
3523 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3524 if (IS_ERR(inode)) {
3525 ret = PTR_ERR(inode);
3531 if (!inode && root == fs_info->tree_root) {
3532 struct btrfs_root *dead_root;
3533 int is_dead_root = 0;
3536 * This is an orphan in the tree root. Currently these
3537 * could come from 2 sources:
3538 * a) a root (snapshot/subvolume) deletion in progress
3539 * b) a free space cache inode
3540 * We need to distinguish those two, as the orphan item
3541 * for a root must not get deleted before the deletion
3542 * of the snapshot/subvolume's tree completes.
3544 * btrfs_find_orphan_roots() ran before us, which has
3545 * found all deleted roots and loaded them into
3546 * fs_info->fs_roots_radix. So here we can find if an
3547 * orphan item corresponds to a deleted root by looking
3548 * up the root from that radix tree.
3551 spin_lock(&fs_info->fs_roots_radix_lock);
3552 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3553 (unsigned long)found_key.objectid);
3554 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3556 spin_unlock(&fs_info->fs_roots_radix_lock);
3559 /* prevent this orphan from being found again */
3560 key.offset = found_key.objectid - 1;
3567 * If we have an inode with links, there are a couple of
3570 * 1. We were halfway through creating fsverity metadata for the
3571 * file. In that case, the orphan item represents incomplete
3572 * fsverity metadata which must be cleaned up with
3573 * btrfs_drop_verity_items and deleting the orphan item.
3575 * 2. Old kernels (before v3.12) used to create an
3576 * orphan item for truncate indicating that there were possibly
3577 * extent items past i_size that needed to be deleted. In v3.12,
3578 * truncate was changed to update i_size in sync with the extent
3579 * items, but the (useless) orphan item was still created. Since
3580 * v4.18, we don't create the orphan item for truncate at all.
3582 * So, this item could mean that we need to do a truncate, but
3583 * only if this filesystem was last used on a pre-v3.12 kernel
3584 * and was not cleanly unmounted. The odds of that are quite
3585 * slim, and it's a pain to do the truncate now, so just delete
3588 * It's also possible that this orphan item was supposed to be
3589 * deleted but wasn't. The inode number may have been reused,
3590 * but either way, we can delete the orphan item.
3592 if (!inode || inode->i_nlink) {
3594 ret = btrfs_drop_verity_items(BTRFS_I(inode));
3600 trans = btrfs_start_transaction(root, 1);
3601 if (IS_ERR(trans)) {
3602 ret = PTR_ERR(trans);
3605 btrfs_debug(fs_info, "auto deleting %Lu",
3606 found_key.objectid);
3607 ret = btrfs_del_orphan_item(trans, root,
3608 found_key.objectid);
3609 btrfs_end_transaction(trans);
3617 /* this will do delete_inode and everything for us */
3620 /* release the path since we're done with it */
3621 btrfs_release_path(path);
3623 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3624 trans = btrfs_join_transaction(root);
3626 btrfs_end_transaction(trans);
3630 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3634 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3635 btrfs_free_path(path);
3640 * very simple check to peek ahead in the leaf looking for xattrs. If we
3641 * don't find any xattrs, we know there can't be any acls.
3643 * slot is the slot the inode is in, objectid is the objectid of the inode
3645 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3646 int slot, u64 objectid,
3647 int *first_xattr_slot)
3649 u32 nritems = btrfs_header_nritems(leaf);
3650 struct btrfs_key found_key;
3651 static u64 xattr_access = 0;
3652 static u64 xattr_default = 0;
3655 if (!xattr_access) {
3656 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3657 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3658 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3659 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3663 *first_xattr_slot = -1;
3664 while (slot < nritems) {
3665 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3667 /* we found a different objectid, there must not be acls */
3668 if (found_key.objectid != objectid)
3671 /* we found an xattr, assume we've got an acl */
3672 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3673 if (*first_xattr_slot == -1)
3674 *first_xattr_slot = slot;
3675 if (found_key.offset == xattr_access ||
3676 found_key.offset == xattr_default)
3681 * we found a key greater than an xattr key, there can't
3682 * be any acls later on
3684 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3691 * it goes inode, inode backrefs, xattrs, extents,
3692 * so if there are a ton of hard links to an inode there can
3693 * be a lot of backrefs. Don't waste time searching too hard,
3694 * this is just an optimization
3699 /* we hit the end of the leaf before we found an xattr or
3700 * something larger than an xattr. We have to assume the inode
3703 if (*first_xattr_slot == -1)
3704 *first_xattr_slot = slot;
3709 * read an inode from the btree into the in-memory inode
3711 static int btrfs_read_locked_inode(struct inode *inode,
3712 struct btrfs_path *in_path)
3714 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3715 struct btrfs_path *path = in_path;
3716 struct extent_buffer *leaf;
3717 struct btrfs_inode_item *inode_item;
3718 struct btrfs_root *root = BTRFS_I(inode)->root;
3719 struct btrfs_key location;
3724 bool filled = false;
3725 int first_xattr_slot;
3727 ret = btrfs_fill_inode(inode, &rdev);
3732 path = btrfs_alloc_path();
3737 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3739 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3741 if (path != in_path)
3742 btrfs_free_path(path);
3746 leaf = path->nodes[0];
3751 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3752 struct btrfs_inode_item);
3753 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3754 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3755 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3756 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3757 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3758 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3759 round_up(i_size_read(inode), fs_info->sectorsize));
3761 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3762 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3764 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3765 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3767 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3768 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3770 BTRFS_I(inode)->i_otime.tv_sec =
3771 btrfs_timespec_sec(leaf, &inode_item->otime);
3772 BTRFS_I(inode)->i_otime.tv_nsec =
3773 btrfs_timespec_nsec(leaf, &inode_item->otime);
3775 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3776 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3777 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3779 inode_set_iversion_queried(inode,
3780 btrfs_inode_sequence(leaf, inode_item));
3781 inode->i_generation = BTRFS_I(inode)->generation;
3783 rdev = btrfs_inode_rdev(leaf, inode_item);
3785 BTRFS_I(inode)->index_cnt = (u64)-1;
3786 btrfs_inode_split_flags(btrfs_inode_flags(leaf, inode_item),
3787 &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags);
3791 * If we were modified in the current generation and evicted from memory
3792 * and then re-read we need to do a full sync since we don't have any
3793 * idea about which extents were modified before we were evicted from
3796 * This is required for both inode re-read from disk and delayed inode
3797 * in delayed_nodes_tree.
3799 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3800 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3801 &BTRFS_I(inode)->runtime_flags);
3804 * We don't persist the id of the transaction where an unlink operation
3805 * against the inode was last made. So here we assume the inode might
3806 * have been evicted, and therefore the exact value of last_unlink_trans
3807 * lost, and set it to last_trans to avoid metadata inconsistencies
3808 * between the inode and its parent if the inode is fsync'ed and the log
3809 * replayed. For example, in the scenario:
3812 * ln mydir/foo mydir/bar
3815 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3816 * xfs_io -c fsync mydir/foo
3818 * mount fs, triggers fsync log replay
3820 * We must make sure that when we fsync our inode foo we also log its
3821 * parent inode, otherwise after log replay the parent still has the
3822 * dentry with the "bar" name but our inode foo has a link count of 1
3823 * and doesn't have an inode ref with the name "bar" anymore.
3825 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3826 * but it guarantees correctness at the expense of occasional full
3827 * transaction commits on fsync if our inode is a directory, or if our
3828 * inode is not a directory, logging its parent unnecessarily.
3830 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3833 * Same logic as for last_unlink_trans. We don't persist the generation
3834 * of the last transaction where this inode was used for a reflink
3835 * operation, so after eviction and reloading the inode we must be
3836 * pessimistic and assume the last transaction that modified the inode.
3838 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3841 if (inode->i_nlink != 1 ||
3842 path->slots[0] >= btrfs_header_nritems(leaf))
3845 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3846 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3849 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3850 if (location.type == BTRFS_INODE_REF_KEY) {
3851 struct btrfs_inode_ref *ref;
3853 ref = (struct btrfs_inode_ref *)ptr;
3854 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3855 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3856 struct btrfs_inode_extref *extref;
3858 extref = (struct btrfs_inode_extref *)ptr;
3859 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3864 * try to precache a NULL acl entry for files that don't have
3865 * any xattrs or acls
3867 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3868 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3869 if (first_xattr_slot != -1) {
3870 path->slots[0] = first_xattr_slot;
3871 ret = btrfs_load_inode_props(inode, path);
3874 "error loading props for ino %llu (root %llu): %d",
3875 btrfs_ino(BTRFS_I(inode)),
3876 root->root_key.objectid, ret);
3878 if (path != in_path)
3879 btrfs_free_path(path);
3882 cache_no_acl(inode);
3884 switch (inode->i_mode & S_IFMT) {
3886 inode->i_mapping->a_ops = &btrfs_aops;
3887 inode->i_fop = &btrfs_file_operations;
3888 inode->i_op = &btrfs_file_inode_operations;
3891 inode->i_fop = &btrfs_dir_file_operations;
3892 inode->i_op = &btrfs_dir_inode_operations;
3895 inode->i_op = &btrfs_symlink_inode_operations;
3896 inode_nohighmem(inode);
3897 inode->i_mapping->a_ops = &btrfs_aops;
3900 inode->i_op = &btrfs_special_inode_operations;
3901 init_special_inode(inode, inode->i_mode, rdev);
3905 btrfs_sync_inode_flags_to_i_flags(inode);
3910 * given a leaf and an inode, copy the inode fields into the leaf
3912 static void fill_inode_item(struct btrfs_trans_handle *trans,
3913 struct extent_buffer *leaf,
3914 struct btrfs_inode_item *item,
3915 struct inode *inode)
3917 struct btrfs_map_token token;
3920 btrfs_init_map_token(&token, leaf);
3922 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3923 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3924 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
3925 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3926 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3928 btrfs_set_token_timespec_sec(&token, &item->atime,
3929 inode->i_atime.tv_sec);
3930 btrfs_set_token_timespec_nsec(&token, &item->atime,
3931 inode->i_atime.tv_nsec);
3933 btrfs_set_token_timespec_sec(&token, &item->mtime,
3934 inode->i_mtime.tv_sec);
3935 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3936 inode->i_mtime.tv_nsec);
3938 btrfs_set_token_timespec_sec(&token, &item->ctime,
3939 inode->i_ctime.tv_sec);
3940 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3941 inode->i_ctime.tv_nsec);
3943 btrfs_set_token_timespec_sec(&token, &item->otime,
3944 BTRFS_I(inode)->i_otime.tv_sec);
3945 btrfs_set_token_timespec_nsec(&token, &item->otime,
3946 BTRFS_I(inode)->i_otime.tv_nsec);
3948 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
3949 btrfs_set_token_inode_generation(&token, item,
3950 BTRFS_I(inode)->generation);
3951 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
3952 btrfs_set_token_inode_transid(&token, item, trans->transid);
3953 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
3954 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
3955 BTRFS_I(inode)->ro_flags);
3956 btrfs_set_token_inode_flags(&token, item, flags);
3957 btrfs_set_token_inode_block_group(&token, item, 0);
3961 * copy everything in the in-memory inode into the btree.
3963 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3964 struct btrfs_root *root,
3965 struct btrfs_inode *inode)
3967 struct btrfs_inode_item *inode_item;
3968 struct btrfs_path *path;
3969 struct extent_buffer *leaf;
3972 path = btrfs_alloc_path();
3976 ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1);
3983 leaf = path->nodes[0];
3984 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3985 struct btrfs_inode_item);
3987 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
3988 btrfs_mark_buffer_dirty(leaf);
3989 btrfs_set_inode_last_trans(trans, inode);
3992 btrfs_free_path(path);
3997 * copy everything in the in-memory inode into the btree.
3999 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
4000 struct btrfs_root *root,
4001 struct btrfs_inode *inode)
4003 struct btrfs_fs_info *fs_info = root->fs_info;
4007 * If the inode is a free space inode, we can deadlock during commit
4008 * if we put it into the delayed code.
4010 * The data relocation inode should also be directly updated
4013 if (!btrfs_is_free_space_inode(inode)
4014 && !btrfs_is_data_reloc_root(root)
4015 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4016 btrfs_update_root_times(trans, root);
4018 ret = btrfs_delayed_update_inode(trans, root, inode);
4020 btrfs_set_inode_last_trans(trans, inode);
4024 return btrfs_update_inode_item(trans, root, inode);
4027 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4028 struct btrfs_root *root, struct btrfs_inode *inode)
4032 ret = btrfs_update_inode(trans, root, inode);
4034 return btrfs_update_inode_item(trans, root, inode);
4039 * unlink helper that gets used here in inode.c and in the tree logging
4040 * recovery code. It remove a link in a directory with a given name, and
4041 * also drops the back refs in the inode to the directory
4043 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4044 struct btrfs_inode *dir,
4045 struct btrfs_inode *inode,
4046 const struct fscrypt_str *name,
4047 struct btrfs_rename_ctx *rename_ctx)
4049 struct btrfs_root *root = dir->root;
4050 struct btrfs_fs_info *fs_info = root->fs_info;
4051 struct btrfs_path *path;
4053 struct btrfs_dir_item *di;
4055 u64 ino = btrfs_ino(inode);
4056 u64 dir_ino = btrfs_ino(dir);
4058 path = btrfs_alloc_path();
4064 di = btrfs_lookup_dir_item(trans, root, path, dir_ino, name, -1);
4065 if (IS_ERR_OR_NULL(di)) {
4066 ret = di ? PTR_ERR(di) : -ENOENT;
4069 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4072 btrfs_release_path(path);
4075 * If we don't have dir index, we have to get it by looking up
4076 * the inode ref, since we get the inode ref, remove it directly,
4077 * it is unnecessary to do delayed deletion.
4079 * But if we have dir index, needn't search inode ref to get it.
4080 * Since the inode ref is close to the inode item, it is better
4081 * that we delay to delete it, and just do this deletion when
4082 * we update the inode item.
4084 if (inode->dir_index) {
4085 ret = btrfs_delayed_delete_inode_ref(inode);
4087 index = inode->dir_index;
4092 ret = btrfs_del_inode_ref(trans, root, name, ino, dir_ino, &index);
4095 "failed to delete reference to %.*s, inode %llu parent %llu",
4096 name->len, name->name, ino, dir_ino);
4097 btrfs_abort_transaction(trans, ret);
4102 rename_ctx->index = index;
4104 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4106 btrfs_abort_transaction(trans, ret);
4111 * If we are in a rename context, we don't need to update anything in the
4112 * log. That will be done later during the rename by btrfs_log_new_name().
4113 * Besides that, doing it here would only cause extra unnecessary btree
4114 * operations on the log tree, increasing latency for applications.
4117 btrfs_del_inode_ref_in_log(trans, root, name, inode, dir_ino);
4118 btrfs_del_dir_entries_in_log(trans, root, name, dir, index);
4122 * If we have a pending delayed iput we could end up with the final iput
4123 * being run in btrfs-cleaner context. If we have enough of these built
4124 * up we can end up burning a lot of time in btrfs-cleaner without any
4125 * way to throttle the unlinks. Since we're currently holding a ref on
4126 * the inode we can run the delayed iput here without any issues as the
4127 * final iput won't be done until after we drop the ref we're currently
4130 btrfs_run_delayed_iput(fs_info, inode);
4132 btrfs_free_path(path);
4136 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name->len * 2);
4137 inode_inc_iversion(&inode->vfs_inode);
4138 inode_inc_iversion(&dir->vfs_inode);
4139 inode->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4140 dir->vfs_inode.i_mtime = inode->vfs_inode.i_ctime;
4141 dir->vfs_inode.i_ctime = inode->vfs_inode.i_ctime;
4142 ret = btrfs_update_inode(trans, root, dir);
4147 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4148 struct btrfs_inode *dir, struct btrfs_inode *inode,
4149 const struct fscrypt_str *name)
4153 ret = __btrfs_unlink_inode(trans, dir, inode, name, NULL);
4155 drop_nlink(&inode->vfs_inode);
4156 ret = btrfs_update_inode(trans, inode->root, inode);
4162 * helper to start transaction for unlink and rmdir.
4164 * unlink and rmdir are special in btrfs, they do not always free space, so
4165 * if we cannot make our reservations the normal way try and see if there is
4166 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4167 * allow the unlink to occur.
4169 static struct btrfs_trans_handle *__unlink_start_trans(struct btrfs_inode *dir)
4171 struct btrfs_root *root = dir->root;
4173 return btrfs_start_transaction_fallback_global_rsv(root,
4174 BTRFS_UNLINK_METADATA_UNITS);
4177 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4179 struct btrfs_trans_handle *trans;
4180 struct inode *inode = d_inode(dentry);
4182 struct fscrypt_name fname;
4184 ret = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4188 /* This needs to handle no-key deletions later on */
4190 trans = __unlink_start_trans(BTRFS_I(dir));
4191 if (IS_ERR(trans)) {
4192 ret = PTR_ERR(trans);
4196 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4199 ret = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4204 if (inode->i_nlink == 0) {
4205 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4211 btrfs_end_transaction(trans);
4212 btrfs_btree_balance_dirty(BTRFS_I(dir)->root->fs_info);
4214 fscrypt_free_filename(&fname);
4218 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4219 struct btrfs_inode *dir, struct dentry *dentry)
4221 struct btrfs_root *root = dir->root;
4222 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4223 struct btrfs_path *path;
4224 struct extent_buffer *leaf;
4225 struct btrfs_dir_item *di;
4226 struct btrfs_key key;
4230 u64 dir_ino = btrfs_ino(dir);
4231 struct fscrypt_name fname;
4233 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
4237 /* This needs to handle no-key deletions later on */
4239 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4240 objectid = inode->root->root_key.objectid;
4241 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4242 objectid = inode->location.objectid;
4245 fscrypt_free_filename(&fname);
4249 path = btrfs_alloc_path();
4255 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4256 &fname.disk_name, -1);
4257 if (IS_ERR_OR_NULL(di)) {
4258 ret = di ? PTR_ERR(di) : -ENOENT;
4262 leaf = path->nodes[0];
4263 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4264 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4265 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4267 btrfs_abort_transaction(trans, ret);
4270 btrfs_release_path(path);
4273 * This is a placeholder inode for a subvolume we didn't have a
4274 * reference to at the time of the snapshot creation. In the meantime
4275 * we could have renamed the real subvol link into our snapshot, so
4276 * depending on btrfs_del_root_ref to return -ENOENT here is incorrect.
4277 * Instead simply lookup the dir_index_item for this entry so we can
4278 * remove it. Otherwise we know we have a ref to the root and we can
4279 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4281 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4282 di = btrfs_search_dir_index_item(root, path, dir_ino, &fname.disk_name);
4283 if (IS_ERR_OR_NULL(di)) {
4288 btrfs_abort_transaction(trans, ret);
4292 leaf = path->nodes[0];
4293 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4295 btrfs_release_path(path);
4297 ret = btrfs_del_root_ref(trans, objectid,
4298 root->root_key.objectid, dir_ino,
4299 &index, &fname.disk_name);
4301 btrfs_abort_transaction(trans, ret);
4306 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4308 btrfs_abort_transaction(trans, ret);
4312 btrfs_i_size_write(dir, dir->vfs_inode.i_size - fname.disk_name.len * 2);
4313 inode_inc_iversion(&dir->vfs_inode);
4314 dir->vfs_inode.i_mtime = current_time(&dir->vfs_inode);
4315 dir->vfs_inode.i_ctime = dir->vfs_inode.i_mtime;
4316 ret = btrfs_update_inode_fallback(trans, root, dir);
4318 btrfs_abort_transaction(trans, ret);
4320 btrfs_free_path(path);
4321 fscrypt_free_filename(&fname);
4326 * Helper to check if the subvolume references other subvolumes or if it's
4329 static noinline int may_destroy_subvol(struct btrfs_root *root)
4331 struct btrfs_fs_info *fs_info = root->fs_info;
4332 struct btrfs_path *path;
4333 struct btrfs_dir_item *di;
4334 struct btrfs_key key;
4335 struct fscrypt_str name = FSTR_INIT("default", 7);
4339 path = btrfs_alloc_path();
4343 /* Make sure this root isn't set as the default subvol */
4344 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4345 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4347 if (di && !IS_ERR(di)) {
4348 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4349 if (key.objectid == root->root_key.objectid) {
4352 "deleting default subvolume %llu is not allowed",
4356 btrfs_release_path(path);
4359 key.objectid = root->root_key.objectid;
4360 key.type = BTRFS_ROOT_REF_KEY;
4361 key.offset = (u64)-1;
4363 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4369 if (path->slots[0] > 0) {
4371 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4372 if (key.objectid == root->root_key.objectid &&
4373 key.type == BTRFS_ROOT_REF_KEY)
4377 btrfs_free_path(path);
4381 /* Delete all dentries for inodes belonging to the root */
4382 static void btrfs_prune_dentries(struct btrfs_root *root)
4384 struct btrfs_fs_info *fs_info = root->fs_info;
4385 struct rb_node *node;
4386 struct rb_node *prev;
4387 struct btrfs_inode *entry;
4388 struct inode *inode;
4391 if (!BTRFS_FS_ERROR(fs_info))
4392 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4394 spin_lock(&root->inode_lock);
4396 node = root->inode_tree.rb_node;
4400 entry = rb_entry(node, struct btrfs_inode, rb_node);
4402 if (objectid < btrfs_ino(entry))
4403 node = node->rb_left;
4404 else if (objectid > btrfs_ino(entry))
4405 node = node->rb_right;
4411 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4412 if (objectid <= btrfs_ino(entry)) {
4416 prev = rb_next(prev);
4420 entry = rb_entry(node, struct btrfs_inode, rb_node);
4421 objectid = btrfs_ino(entry) + 1;
4422 inode = igrab(&entry->vfs_inode);
4424 spin_unlock(&root->inode_lock);
4425 if (atomic_read(&inode->i_count) > 1)
4426 d_prune_aliases(inode);
4428 * btrfs_drop_inode will have it removed from the inode
4429 * cache when its usage count hits zero.
4433 spin_lock(&root->inode_lock);
4437 if (cond_resched_lock(&root->inode_lock))
4440 node = rb_next(node);
4442 spin_unlock(&root->inode_lock);
4445 int btrfs_delete_subvolume(struct btrfs_inode *dir, struct dentry *dentry)
4447 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4448 struct btrfs_root *root = dir->root;
4449 struct inode *inode = d_inode(dentry);
4450 struct btrfs_root *dest = BTRFS_I(inode)->root;
4451 struct btrfs_trans_handle *trans;
4452 struct btrfs_block_rsv block_rsv;
4457 * Don't allow to delete a subvolume with send in progress. This is
4458 * inside the inode lock so the error handling that has to drop the bit
4459 * again is not run concurrently.
4461 spin_lock(&dest->root_item_lock);
4462 if (dest->send_in_progress) {
4463 spin_unlock(&dest->root_item_lock);
4465 "attempt to delete subvolume %llu during send",
4466 dest->root_key.objectid);
4469 if (atomic_read(&dest->nr_swapfiles)) {
4470 spin_unlock(&dest->root_item_lock);
4472 "attempt to delete subvolume %llu with active swapfile",
4473 root->root_key.objectid);
4476 root_flags = btrfs_root_flags(&dest->root_item);
4477 btrfs_set_root_flags(&dest->root_item,
4478 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4479 spin_unlock(&dest->root_item_lock);
4481 down_write(&fs_info->subvol_sem);
4483 ret = may_destroy_subvol(dest);
4487 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4489 * One for dir inode,
4490 * two for dir entries,
4491 * two for root ref/backref.
4493 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4497 trans = btrfs_start_transaction(root, 0);
4498 if (IS_ERR(trans)) {
4499 ret = PTR_ERR(trans);
4502 trans->block_rsv = &block_rsv;
4503 trans->bytes_reserved = block_rsv.size;
4505 btrfs_record_snapshot_destroy(trans, dir);
4507 ret = btrfs_unlink_subvol(trans, dir, dentry);
4509 btrfs_abort_transaction(trans, ret);
4513 ret = btrfs_record_root_in_trans(trans, dest);
4515 btrfs_abort_transaction(trans, ret);
4519 memset(&dest->root_item.drop_progress, 0,
4520 sizeof(dest->root_item.drop_progress));
4521 btrfs_set_root_drop_level(&dest->root_item, 0);
4522 btrfs_set_root_refs(&dest->root_item, 0);
4524 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4525 ret = btrfs_insert_orphan_item(trans,
4527 dest->root_key.objectid);
4529 btrfs_abort_transaction(trans, ret);
4534 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4535 BTRFS_UUID_KEY_SUBVOL,
4536 dest->root_key.objectid);
4537 if (ret && ret != -ENOENT) {
4538 btrfs_abort_transaction(trans, ret);
4541 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4542 ret = btrfs_uuid_tree_remove(trans,
4543 dest->root_item.received_uuid,
4544 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4545 dest->root_key.objectid);
4546 if (ret && ret != -ENOENT) {
4547 btrfs_abort_transaction(trans, ret);
4552 free_anon_bdev(dest->anon_dev);
4555 trans->block_rsv = NULL;
4556 trans->bytes_reserved = 0;
4557 ret = btrfs_end_transaction(trans);
4558 inode->i_flags |= S_DEAD;
4560 btrfs_subvolume_release_metadata(root, &block_rsv);
4562 up_write(&fs_info->subvol_sem);
4564 spin_lock(&dest->root_item_lock);
4565 root_flags = btrfs_root_flags(&dest->root_item);
4566 btrfs_set_root_flags(&dest->root_item,
4567 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4568 spin_unlock(&dest->root_item_lock);
4570 d_invalidate(dentry);
4571 btrfs_prune_dentries(dest);
4572 ASSERT(dest->send_in_progress == 0);
4578 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4580 struct inode *inode = d_inode(dentry);
4581 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
4583 struct btrfs_trans_handle *trans;
4584 u64 last_unlink_trans;
4585 struct fscrypt_name fname;
4587 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4589 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) {
4590 if (unlikely(btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))) {
4592 "extent tree v2 doesn't support snapshot deletion yet");
4595 return btrfs_delete_subvolume(BTRFS_I(dir), dentry);
4598 err = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4602 /* This needs to handle no-key deletions later on */
4604 trans = __unlink_start_trans(BTRFS_I(dir));
4605 if (IS_ERR(trans)) {
4606 err = PTR_ERR(trans);
4610 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4611 err = btrfs_unlink_subvol(trans, BTRFS_I(dir), dentry);
4615 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4619 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4621 /* now the directory is empty */
4622 err = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4625 btrfs_i_size_write(BTRFS_I(inode), 0);
4627 * Propagate the last_unlink_trans value of the deleted dir to
4628 * its parent directory. This is to prevent an unrecoverable
4629 * log tree in the case we do something like this:
4631 * 2) create snapshot under dir foo
4632 * 3) delete the snapshot
4635 * 6) fsync foo or some file inside foo
4637 if (last_unlink_trans >= trans->transid)
4638 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4641 btrfs_end_transaction(trans);
4643 btrfs_btree_balance_dirty(fs_info);
4644 fscrypt_free_filename(&fname);
4650 * btrfs_truncate_block - read, zero a chunk and write a block
4651 * @inode - inode that we're zeroing
4652 * @from - the offset to start zeroing
4653 * @len - the length to zero, 0 to zero the entire range respective to the
4655 * @front - zero up to the offset instead of from the offset on
4657 * This will find the block for the "from" offset and cow the block and zero the
4658 * part we want to zero. This is used with truncate and hole punching.
4660 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4663 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4664 struct address_space *mapping = inode->vfs_inode.i_mapping;
4665 struct extent_io_tree *io_tree = &inode->io_tree;
4666 struct btrfs_ordered_extent *ordered;
4667 struct extent_state *cached_state = NULL;
4668 struct extent_changeset *data_reserved = NULL;
4669 bool only_release_metadata = false;
4670 u32 blocksize = fs_info->sectorsize;
4671 pgoff_t index = from >> PAGE_SHIFT;
4672 unsigned offset = from & (blocksize - 1);
4674 gfp_t mask = btrfs_alloc_write_mask(mapping);
4675 size_t write_bytes = blocksize;
4680 if (IS_ALIGNED(offset, blocksize) &&
4681 (!len || IS_ALIGNED(len, blocksize)))
4684 block_start = round_down(from, blocksize);
4685 block_end = block_start + blocksize - 1;
4687 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4690 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes, false) > 0) {
4691 /* For nocow case, no need to reserve data space */
4692 only_release_metadata = true;
4697 ret = btrfs_delalloc_reserve_metadata(inode, blocksize, blocksize, false);
4699 if (!only_release_metadata)
4700 btrfs_free_reserved_data_space(inode, data_reserved,
4701 block_start, blocksize);
4705 page = find_or_create_page(mapping, index, mask);
4707 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4709 btrfs_delalloc_release_extents(inode, blocksize);
4714 if (!PageUptodate(page)) {
4715 ret = btrfs_read_folio(NULL, page_folio(page));
4717 if (page->mapping != mapping) {
4722 if (!PageUptodate(page)) {
4729 * We unlock the page after the io is completed and then re-lock it
4730 * above. release_folio() could have come in between that and cleared
4731 * PagePrivate(), but left the page in the mapping. Set the page mapped
4732 * here to make sure it's properly set for the subpage stuff.
4734 ret = set_page_extent_mapped(page);
4738 wait_on_page_writeback(page);
4740 lock_extent(io_tree, block_start, block_end, &cached_state);
4742 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4744 unlock_extent(io_tree, block_start, block_end, &cached_state);
4747 btrfs_start_ordered_extent(ordered);
4748 btrfs_put_ordered_extent(ordered);
4752 clear_extent_bit(&inode->io_tree, block_start, block_end,
4753 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4756 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4759 unlock_extent(io_tree, block_start, block_end, &cached_state);
4763 if (offset != blocksize) {
4765 len = blocksize - offset;
4767 memzero_page(page, (block_start - page_offset(page)),
4770 memzero_page(page, (block_start - page_offset(page)) + offset,
4773 btrfs_page_clear_checked(fs_info, page, block_start,
4774 block_end + 1 - block_start);
4775 btrfs_page_set_dirty(fs_info, page, block_start, block_end + 1 - block_start);
4776 unlock_extent(io_tree, block_start, block_end, &cached_state);
4778 if (only_release_metadata)
4779 set_extent_bit(&inode->io_tree, block_start, block_end,
4780 EXTENT_NORESERVE, NULL);
4784 if (only_release_metadata)
4785 btrfs_delalloc_release_metadata(inode, blocksize, true);
4787 btrfs_delalloc_release_space(inode, data_reserved,
4788 block_start, blocksize, true);
4790 btrfs_delalloc_release_extents(inode, blocksize);
4794 if (only_release_metadata)
4795 btrfs_check_nocow_unlock(inode);
4796 extent_changeset_free(data_reserved);
4800 static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode,
4801 u64 offset, u64 len)
4803 struct btrfs_fs_info *fs_info = root->fs_info;
4804 struct btrfs_trans_handle *trans;
4805 struct btrfs_drop_extents_args drop_args = { 0 };
4809 * If NO_HOLES is enabled, we don't need to do anything.
4810 * Later, up in the call chain, either btrfs_set_inode_last_sub_trans()
4811 * or btrfs_update_inode() will be called, which guarantee that the next
4812 * fsync will know this inode was changed and needs to be logged.
4814 if (btrfs_fs_incompat(fs_info, NO_HOLES))
4818 * 1 - for the one we're dropping
4819 * 1 - for the one we're adding
4820 * 1 - for updating the inode.
4822 trans = btrfs_start_transaction(root, 3);
4824 return PTR_ERR(trans);
4826 drop_args.start = offset;
4827 drop_args.end = offset + len;
4828 drop_args.drop_cache = true;
4830 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
4832 btrfs_abort_transaction(trans, ret);
4833 btrfs_end_transaction(trans);
4837 ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset, len);
4839 btrfs_abort_transaction(trans, ret);
4841 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
4842 btrfs_update_inode(trans, root, inode);
4844 btrfs_end_transaction(trans);
4849 * This function puts in dummy file extents for the area we're creating a hole
4850 * for. So if we are truncating this file to a larger size we need to insert
4851 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4852 * the range between oldsize and size
4854 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
4856 struct btrfs_root *root = inode->root;
4857 struct btrfs_fs_info *fs_info = root->fs_info;
4858 struct extent_io_tree *io_tree = &inode->io_tree;
4859 struct extent_map *em = NULL;
4860 struct extent_state *cached_state = NULL;
4861 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4862 u64 block_end = ALIGN(size, fs_info->sectorsize);
4869 * If our size started in the middle of a block we need to zero out the
4870 * rest of the block before we expand the i_size, otherwise we could
4871 * expose stale data.
4873 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4877 if (size <= hole_start)
4880 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
4882 cur_offset = hole_start;
4884 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4885 block_end - cur_offset);
4891 last_byte = min(extent_map_end(em), block_end);
4892 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4893 hole_size = last_byte - cur_offset;
4895 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4896 struct extent_map *hole_em;
4898 err = maybe_insert_hole(root, inode, cur_offset,
4903 err = btrfs_inode_set_file_extent_range(inode,
4904 cur_offset, hole_size);
4908 hole_em = alloc_extent_map();
4910 btrfs_drop_extent_map_range(inode, cur_offset,
4911 cur_offset + hole_size - 1,
4913 btrfs_set_inode_full_sync(inode);
4916 hole_em->start = cur_offset;
4917 hole_em->len = hole_size;
4918 hole_em->orig_start = cur_offset;
4920 hole_em->block_start = EXTENT_MAP_HOLE;
4921 hole_em->block_len = 0;
4922 hole_em->orig_block_len = 0;
4923 hole_em->ram_bytes = hole_size;
4924 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4925 hole_em->generation = fs_info->generation;
4927 err = btrfs_replace_extent_map_range(inode, hole_em, true);
4928 free_extent_map(hole_em);
4930 err = btrfs_inode_set_file_extent_range(inode,
4931 cur_offset, hole_size);
4936 free_extent_map(em);
4938 cur_offset = last_byte;
4939 if (cur_offset >= block_end)
4942 free_extent_map(em);
4943 unlock_extent(io_tree, hole_start, block_end - 1, &cached_state);
4947 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4949 struct btrfs_root *root = BTRFS_I(inode)->root;
4950 struct btrfs_trans_handle *trans;
4951 loff_t oldsize = i_size_read(inode);
4952 loff_t newsize = attr->ia_size;
4953 int mask = attr->ia_valid;
4957 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4958 * special case where we need to update the times despite not having
4959 * these flags set. For all other operations the VFS set these flags
4960 * explicitly if it wants a timestamp update.
4962 if (newsize != oldsize) {
4963 inode_inc_iversion(inode);
4964 if (!(mask & (ATTR_CTIME | ATTR_MTIME))) {
4965 inode->i_mtime = current_time(inode);
4966 inode->i_ctime = inode->i_mtime;
4970 if (newsize > oldsize) {
4972 * Don't do an expanding truncate while snapshotting is ongoing.
4973 * This is to ensure the snapshot captures a fully consistent
4974 * state of this file - if the snapshot captures this expanding
4975 * truncation, it must capture all writes that happened before
4978 btrfs_drew_write_lock(&root->snapshot_lock);
4979 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
4981 btrfs_drew_write_unlock(&root->snapshot_lock);
4985 trans = btrfs_start_transaction(root, 1);
4986 if (IS_ERR(trans)) {
4987 btrfs_drew_write_unlock(&root->snapshot_lock);
4988 return PTR_ERR(trans);
4991 i_size_write(inode, newsize);
4992 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
4993 pagecache_isize_extended(inode, oldsize, newsize);
4994 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
4995 btrfs_drew_write_unlock(&root->snapshot_lock);
4996 btrfs_end_transaction(trans);
4998 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5000 if (btrfs_is_zoned(fs_info)) {
5001 ret = btrfs_wait_ordered_range(inode,
5002 ALIGN(newsize, fs_info->sectorsize),
5009 * We're truncating a file that used to have good data down to
5010 * zero. Make sure any new writes to the file get on disk
5014 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5015 &BTRFS_I(inode)->runtime_flags);
5017 truncate_setsize(inode, newsize);
5019 inode_dio_wait(inode);
5021 ret = btrfs_truncate(BTRFS_I(inode), newsize == oldsize);
5022 if (ret && inode->i_nlink) {
5026 * Truncate failed, so fix up the in-memory size. We
5027 * adjusted disk_i_size down as we removed extents, so
5028 * wait for disk_i_size to be stable and then update the
5029 * in-memory size to match.
5031 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5034 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5041 static int btrfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
5044 struct inode *inode = d_inode(dentry);
5045 struct btrfs_root *root = BTRFS_I(inode)->root;
5048 if (btrfs_root_readonly(root))
5051 err = setattr_prepare(idmap, dentry, attr);
5055 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5056 err = btrfs_setsize(inode, attr);
5061 if (attr->ia_valid) {
5062 setattr_copy(idmap, inode, attr);
5063 inode_inc_iversion(inode);
5064 err = btrfs_dirty_inode(BTRFS_I(inode));
5066 if (!err && attr->ia_valid & ATTR_MODE)
5067 err = posix_acl_chmod(idmap, dentry, inode->i_mode);
5074 * While truncating the inode pages during eviction, we get the VFS
5075 * calling btrfs_invalidate_folio() against each folio of the inode. This
5076 * is slow because the calls to btrfs_invalidate_folio() result in a
5077 * huge amount of calls to lock_extent() and clear_extent_bit(),
5078 * which keep merging and splitting extent_state structures over and over,
5079 * wasting lots of time.
5081 * Therefore if the inode is being evicted, let btrfs_invalidate_folio()
5082 * skip all those expensive operations on a per folio basis and do only
5083 * the ordered io finishing, while we release here the extent_map and
5084 * extent_state structures, without the excessive merging and splitting.
5086 static void evict_inode_truncate_pages(struct inode *inode)
5088 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5089 struct rb_node *node;
5091 ASSERT(inode->i_state & I_FREEING);
5092 truncate_inode_pages_final(&inode->i_data);
5094 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
5097 * Keep looping until we have no more ranges in the io tree.
5098 * We can have ongoing bios started by readahead that have
5099 * their endio callback (extent_io.c:end_bio_extent_readpage)
5100 * still in progress (unlocked the pages in the bio but did not yet
5101 * unlocked the ranges in the io tree). Therefore this means some
5102 * ranges can still be locked and eviction started because before
5103 * submitting those bios, which are executed by a separate task (work
5104 * queue kthread), inode references (inode->i_count) were not taken
5105 * (which would be dropped in the end io callback of each bio).
5106 * Therefore here we effectively end up waiting for those bios and
5107 * anyone else holding locked ranges without having bumped the inode's
5108 * reference count - if we don't do it, when they access the inode's
5109 * io_tree to unlock a range it may be too late, leading to an
5110 * use-after-free issue.
5112 spin_lock(&io_tree->lock);
5113 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5114 struct extent_state *state;
5115 struct extent_state *cached_state = NULL;
5118 unsigned state_flags;
5120 node = rb_first(&io_tree->state);
5121 state = rb_entry(node, struct extent_state, rb_node);
5122 start = state->start;
5124 state_flags = state->state;
5125 spin_unlock(&io_tree->lock);
5127 lock_extent(io_tree, start, end, &cached_state);
5130 * If still has DELALLOC flag, the extent didn't reach disk,
5131 * and its reserved space won't be freed by delayed_ref.
5132 * So we need to free its reserved space here.
5133 * (Refer to comment in btrfs_invalidate_folio, case 2)
5135 * Note, end is the bytenr of last byte, so we need + 1 here.
5137 if (state_flags & EXTENT_DELALLOC)
5138 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5141 clear_extent_bit(io_tree, start, end,
5142 EXTENT_CLEAR_ALL_BITS | EXTENT_DO_ACCOUNTING,
5146 spin_lock(&io_tree->lock);
5148 spin_unlock(&io_tree->lock);
5151 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5152 struct btrfs_block_rsv *rsv)
5154 struct btrfs_fs_info *fs_info = root->fs_info;
5155 struct btrfs_trans_handle *trans;
5156 u64 delayed_refs_extra = btrfs_calc_delayed_ref_bytes(fs_info, 1);
5160 * Eviction should be taking place at some place safe because of our
5161 * delayed iputs. However the normal flushing code will run delayed
5162 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5164 * We reserve the delayed_refs_extra here again because we can't use
5165 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5166 * above. We reserve our extra bit here because we generate a ton of
5167 * delayed refs activity by truncating.
5169 * BTRFS_RESERVE_FLUSH_EVICT will steal from the global_rsv if it can,
5170 * if we fail to make this reservation we can re-try without the
5171 * delayed_refs_extra so we can make some forward progress.
5173 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size + delayed_refs_extra,
5174 BTRFS_RESERVE_FLUSH_EVICT);
5176 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size,
5177 BTRFS_RESERVE_FLUSH_EVICT);
5180 "could not allocate space for delete; will truncate on mount");
5181 return ERR_PTR(-ENOSPC);
5183 delayed_refs_extra = 0;
5186 trans = btrfs_join_transaction(root);
5190 if (delayed_refs_extra) {
5191 trans->block_rsv = &fs_info->trans_block_rsv;
5192 trans->bytes_reserved = delayed_refs_extra;
5193 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5194 delayed_refs_extra, true);
5199 void btrfs_evict_inode(struct inode *inode)
5201 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5202 struct btrfs_trans_handle *trans;
5203 struct btrfs_root *root = BTRFS_I(inode)->root;
5204 struct btrfs_block_rsv *rsv = NULL;
5207 trace_btrfs_inode_evict(inode);
5210 fsverity_cleanup_inode(inode);
5215 evict_inode_truncate_pages(inode);
5217 if (inode->i_nlink &&
5218 ((btrfs_root_refs(&root->root_item) != 0 &&
5219 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5220 btrfs_is_free_space_inode(BTRFS_I(inode))))
5223 if (is_bad_inode(inode))
5226 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5229 if (inode->i_nlink > 0) {
5230 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5231 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5236 * This makes sure the inode item in tree is uptodate and the space for
5237 * the inode update is released.
5239 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5244 * This drops any pending insert or delete operations we have for this
5245 * inode. We could have a delayed dir index deletion queued up, but
5246 * we're removing the inode completely so that'll be taken care of in
5249 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
5251 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5254 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5255 rsv->failfast = true;
5257 btrfs_i_size_write(BTRFS_I(inode), 0);
5260 struct btrfs_truncate_control control = {
5261 .inode = BTRFS_I(inode),
5262 .ino = btrfs_ino(BTRFS_I(inode)),
5267 trans = evict_refill_and_join(root, rsv);
5271 trans->block_rsv = rsv;
5273 ret = btrfs_truncate_inode_items(trans, root, &control);
5274 trans->block_rsv = &fs_info->trans_block_rsv;
5275 btrfs_end_transaction(trans);
5277 * We have not added new delayed items for our inode after we
5278 * have flushed its delayed items, so no need to throttle on
5279 * delayed items. However we have modified extent buffers.
5281 btrfs_btree_balance_dirty_nodelay(fs_info);
5282 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5289 * Errors here aren't a big deal, it just means we leave orphan items in
5290 * the tree. They will be cleaned up on the next mount. If the inode
5291 * number gets reused, cleanup deletes the orphan item without doing
5292 * anything, and unlink reuses the existing orphan item.
5294 * If it turns out that we are dropping too many of these, we might want
5295 * to add a mechanism for retrying these after a commit.
5297 trans = evict_refill_and_join(root, rsv);
5298 if (!IS_ERR(trans)) {
5299 trans->block_rsv = rsv;
5300 btrfs_orphan_del(trans, BTRFS_I(inode));
5301 trans->block_rsv = &fs_info->trans_block_rsv;
5302 btrfs_end_transaction(trans);
5306 btrfs_free_block_rsv(fs_info, rsv);
5308 * If we didn't successfully delete, the orphan item will still be in
5309 * the tree and we'll retry on the next mount. Again, we might also want
5310 * to retry these periodically in the future.
5312 btrfs_remove_delayed_node(BTRFS_I(inode));
5313 fsverity_cleanup_inode(inode);
5318 * Return the key found in the dir entry in the location pointer, fill @type
5319 * with BTRFS_FT_*, and return 0.
5321 * If no dir entries were found, returns -ENOENT.
5322 * If found a corrupted location in dir entry, returns -EUCLEAN.
5324 static int btrfs_inode_by_name(struct btrfs_inode *dir, struct dentry *dentry,
5325 struct btrfs_key *location, u8 *type)
5327 struct btrfs_dir_item *di;
5328 struct btrfs_path *path;
5329 struct btrfs_root *root = dir->root;
5331 struct fscrypt_name fname;
5333 path = btrfs_alloc_path();
5337 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
5341 * fscrypt_setup_filename() should never return a positive value, but
5342 * gcc on sparc/parisc thinks it can, so assert that doesn't happen.
5346 /* This needs to handle no-key deletions later on */
5348 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir),
5349 &fname.disk_name, 0);
5350 if (IS_ERR_OR_NULL(di)) {
5351 ret = di ? PTR_ERR(di) : -ENOENT;
5355 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5356 if (location->type != BTRFS_INODE_ITEM_KEY &&
5357 location->type != BTRFS_ROOT_ITEM_KEY) {
5359 btrfs_warn(root->fs_info,
5360 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5361 __func__, fname.disk_name.name, btrfs_ino(dir),
5362 location->objectid, location->type, location->offset);
5365 *type = btrfs_dir_ftype(path->nodes[0], di);
5367 fscrypt_free_filename(&fname);
5368 btrfs_free_path(path);
5373 * when we hit a tree root in a directory, the btrfs part of the inode
5374 * needs to be changed to reflect the root directory of the tree root. This
5375 * is kind of like crossing a mount point.
5377 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5378 struct btrfs_inode *dir,
5379 struct dentry *dentry,
5380 struct btrfs_key *location,
5381 struct btrfs_root **sub_root)
5383 struct btrfs_path *path;
5384 struct btrfs_root *new_root;
5385 struct btrfs_root_ref *ref;
5386 struct extent_buffer *leaf;
5387 struct btrfs_key key;
5390 struct fscrypt_name fname;
5392 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 0, &fname);
5396 path = btrfs_alloc_path();
5403 key.objectid = dir->root->root_key.objectid;
5404 key.type = BTRFS_ROOT_REF_KEY;
5405 key.offset = location->objectid;
5407 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5414 leaf = path->nodes[0];
5415 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5416 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5417 btrfs_root_ref_name_len(leaf, ref) != fname.disk_name.len)
5420 ret = memcmp_extent_buffer(leaf, fname.disk_name.name,
5421 (unsigned long)(ref + 1), fname.disk_name.len);
5425 btrfs_release_path(path);
5427 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5428 if (IS_ERR(new_root)) {
5429 err = PTR_ERR(new_root);
5433 *sub_root = new_root;
5434 location->objectid = btrfs_root_dirid(&new_root->root_item);
5435 location->type = BTRFS_INODE_ITEM_KEY;
5436 location->offset = 0;
5439 btrfs_free_path(path);
5440 fscrypt_free_filename(&fname);
5444 static void inode_tree_add(struct btrfs_inode *inode)
5446 struct btrfs_root *root = inode->root;
5447 struct btrfs_inode *entry;
5449 struct rb_node *parent;
5450 struct rb_node *new = &inode->rb_node;
5451 u64 ino = btrfs_ino(inode);
5453 if (inode_unhashed(&inode->vfs_inode))
5456 spin_lock(&root->inode_lock);
5457 p = &root->inode_tree.rb_node;
5460 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5462 if (ino < btrfs_ino(entry))
5463 p = &parent->rb_left;
5464 else if (ino > btrfs_ino(entry))
5465 p = &parent->rb_right;
5467 WARN_ON(!(entry->vfs_inode.i_state &
5468 (I_WILL_FREE | I_FREEING)));
5469 rb_replace_node(parent, new, &root->inode_tree);
5470 RB_CLEAR_NODE(parent);
5471 spin_unlock(&root->inode_lock);
5475 rb_link_node(new, parent, p);
5476 rb_insert_color(new, &root->inode_tree);
5477 spin_unlock(&root->inode_lock);
5480 static void inode_tree_del(struct btrfs_inode *inode)
5482 struct btrfs_root *root = inode->root;
5485 spin_lock(&root->inode_lock);
5486 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5487 rb_erase(&inode->rb_node, &root->inode_tree);
5488 RB_CLEAR_NODE(&inode->rb_node);
5489 empty = RB_EMPTY_ROOT(&root->inode_tree);
5491 spin_unlock(&root->inode_lock);
5493 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5494 spin_lock(&root->inode_lock);
5495 empty = RB_EMPTY_ROOT(&root->inode_tree);
5496 spin_unlock(&root->inode_lock);
5498 btrfs_add_dead_root(root);
5503 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5505 struct btrfs_iget_args *args = p;
5507 inode->i_ino = args->ino;
5508 BTRFS_I(inode)->location.objectid = args->ino;
5509 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5510 BTRFS_I(inode)->location.offset = 0;
5511 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5512 BUG_ON(args->root && !BTRFS_I(inode)->root);
5514 if (args->root && args->root == args->root->fs_info->tree_root &&
5515 args->ino != BTRFS_BTREE_INODE_OBJECTID)
5516 set_bit(BTRFS_INODE_FREE_SPACE_INODE,
5517 &BTRFS_I(inode)->runtime_flags);
5521 static int btrfs_find_actor(struct inode *inode, void *opaque)
5523 struct btrfs_iget_args *args = opaque;
5525 return args->ino == BTRFS_I(inode)->location.objectid &&
5526 args->root == BTRFS_I(inode)->root;
5529 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5530 struct btrfs_root *root)
5532 struct inode *inode;
5533 struct btrfs_iget_args args;
5534 unsigned long hashval = btrfs_inode_hash(ino, root);
5539 inode = iget5_locked(s, hashval, btrfs_find_actor,
5540 btrfs_init_locked_inode,
5546 * Get an inode object given its inode number and corresponding root.
5547 * Path can be preallocated to prevent recursing back to iget through
5548 * allocator. NULL is also valid but may require an additional allocation
5551 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5552 struct btrfs_root *root, struct btrfs_path *path)
5554 struct inode *inode;
5556 inode = btrfs_iget_locked(s, ino, root);
5558 return ERR_PTR(-ENOMEM);
5560 if (inode->i_state & I_NEW) {
5563 ret = btrfs_read_locked_inode(inode, path);
5565 inode_tree_add(BTRFS_I(inode));
5566 unlock_new_inode(inode);
5570 * ret > 0 can come from btrfs_search_slot called by
5571 * btrfs_read_locked_inode, this means the inode item
5576 inode = ERR_PTR(ret);
5583 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5585 return btrfs_iget_path(s, ino, root, NULL);
5588 static struct inode *new_simple_dir(struct super_block *s,
5589 struct btrfs_key *key,
5590 struct btrfs_root *root)
5592 struct inode *inode = new_inode(s);
5595 return ERR_PTR(-ENOMEM);
5597 BTRFS_I(inode)->root = btrfs_grab_root(root);
5598 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5599 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5601 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5603 * We only need lookup, the rest is read-only and there's no inode
5604 * associated with the dentry
5606 inode->i_op = &simple_dir_inode_operations;
5607 inode->i_opflags &= ~IOP_XATTR;
5608 inode->i_fop = &simple_dir_operations;
5609 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5610 inode->i_mtime = current_time(inode);
5611 inode->i_atime = inode->i_mtime;
5612 inode->i_ctime = inode->i_mtime;
5613 BTRFS_I(inode)->i_otime = inode->i_mtime;
5618 static_assert(BTRFS_FT_UNKNOWN == FT_UNKNOWN);
5619 static_assert(BTRFS_FT_REG_FILE == FT_REG_FILE);
5620 static_assert(BTRFS_FT_DIR == FT_DIR);
5621 static_assert(BTRFS_FT_CHRDEV == FT_CHRDEV);
5622 static_assert(BTRFS_FT_BLKDEV == FT_BLKDEV);
5623 static_assert(BTRFS_FT_FIFO == FT_FIFO);
5624 static_assert(BTRFS_FT_SOCK == FT_SOCK);
5625 static_assert(BTRFS_FT_SYMLINK == FT_SYMLINK);
5627 static inline u8 btrfs_inode_type(struct inode *inode)
5629 return fs_umode_to_ftype(inode->i_mode);
5632 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5634 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5635 struct inode *inode;
5636 struct btrfs_root *root = BTRFS_I(dir)->root;
5637 struct btrfs_root *sub_root = root;
5638 struct btrfs_key location;
5642 if (dentry->d_name.len > BTRFS_NAME_LEN)
5643 return ERR_PTR(-ENAMETOOLONG);
5645 ret = btrfs_inode_by_name(BTRFS_I(dir), dentry, &location, &di_type);
5647 return ERR_PTR(ret);
5649 if (location.type == BTRFS_INODE_ITEM_KEY) {
5650 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5654 /* Do extra check against inode mode with di_type */
5655 if (btrfs_inode_type(inode) != di_type) {
5657 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5658 inode->i_mode, btrfs_inode_type(inode),
5661 return ERR_PTR(-EUCLEAN);
5666 ret = fixup_tree_root_location(fs_info, BTRFS_I(dir), dentry,
5667 &location, &sub_root);
5670 inode = ERR_PTR(ret);
5672 inode = new_simple_dir(dir->i_sb, &location, root);
5674 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5675 btrfs_put_root(sub_root);
5680 down_read(&fs_info->cleanup_work_sem);
5681 if (!sb_rdonly(inode->i_sb))
5682 ret = btrfs_orphan_cleanup(sub_root);
5683 up_read(&fs_info->cleanup_work_sem);
5686 inode = ERR_PTR(ret);
5693 static int btrfs_dentry_delete(const struct dentry *dentry)
5695 struct btrfs_root *root;
5696 struct inode *inode = d_inode(dentry);
5698 if (!inode && !IS_ROOT(dentry))
5699 inode = d_inode(dentry->d_parent);
5702 root = BTRFS_I(inode)->root;
5703 if (btrfs_root_refs(&root->root_item) == 0)
5706 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5712 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5715 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5717 if (inode == ERR_PTR(-ENOENT))
5719 return d_splice_alias(inode, dentry);
5723 * Find the highest existing sequence number in a directory and then set the
5724 * in-memory index_cnt variable to the first free sequence number.
5726 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
5728 struct btrfs_root *root = inode->root;
5729 struct btrfs_key key, found_key;
5730 struct btrfs_path *path;
5731 struct extent_buffer *leaf;
5734 key.objectid = btrfs_ino(inode);
5735 key.type = BTRFS_DIR_INDEX_KEY;
5736 key.offset = (u64)-1;
5738 path = btrfs_alloc_path();
5742 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5745 /* FIXME: we should be able to handle this */
5750 if (path->slots[0] == 0) {
5751 inode->index_cnt = BTRFS_DIR_START_INDEX;
5757 leaf = path->nodes[0];
5758 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5760 if (found_key.objectid != btrfs_ino(inode) ||
5761 found_key.type != BTRFS_DIR_INDEX_KEY) {
5762 inode->index_cnt = BTRFS_DIR_START_INDEX;
5766 inode->index_cnt = found_key.offset + 1;
5768 btrfs_free_path(path);
5772 static int btrfs_get_dir_last_index(struct btrfs_inode *dir, u64 *index)
5774 if (dir->index_cnt == (u64)-1) {
5777 ret = btrfs_inode_delayed_dir_index_count(dir);
5779 ret = btrfs_set_inode_index_count(dir);
5785 *index = dir->index_cnt;
5791 * All this infrastructure exists because dir_emit can fault, and we are holding
5792 * the tree lock when doing readdir. For now just allocate a buffer and copy
5793 * our information into that, and then dir_emit from the buffer. This is
5794 * similar to what NFS does, only we don't keep the buffer around in pagecache
5795 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5796 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5799 static int btrfs_opendir(struct inode *inode, struct file *file)
5801 struct btrfs_file_private *private;
5805 ret = btrfs_get_dir_last_index(BTRFS_I(inode), &last_index);
5809 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5812 private->last_index = last_index;
5813 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5814 if (!private->filldir_buf) {
5818 file->private_data = private;
5829 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5832 struct dir_entry *entry = addr;
5833 char *name = (char *)(entry + 1);
5835 ctx->pos = get_unaligned(&entry->offset);
5836 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5837 get_unaligned(&entry->ino),
5838 get_unaligned(&entry->type)))
5840 addr += sizeof(struct dir_entry) +
5841 get_unaligned(&entry->name_len);
5847 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5849 struct inode *inode = file_inode(file);
5850 struct btrfs_root *root = BTRFS_I(inode)->root;
5851 struct btrfs_file_private *private = file->private_data;
5852 struct btrfs_dir_item *di;
5853 struct btrfs_key key;
5854 struct btrfs_key found_key;
5855 struct btrfs_path *path;
5857 struct list_head ins_list;
5858 struct list_head del_list;
5865 struct btrfs_key location;
5867 if (!dir_emit_dots(file, ctx))
5870 path = btrfs_alloc_path();
5874 addr = private->filldir_buf;
5875 path->reada = READA_FORWARD;
5877 INIT_LIST_HEAD(&ins_list);
5878 INIT_LIST_HEAD(&del_list);
5879 put = btrfs_readdir_get_delayed_items(inode, private->last_index,
5880 &ins_list, &del_list);
5883 key.type = BTRFS_DIR_INDEX_KEY;
5884 key.offset = ctx->pos;
5885 key.objectid = btrfs_ino(BTRFS_I(inode));
5887 btrfs_for_each_slot(root, &key, &found_key, path, ret) {
5888 struct dir_entry *entry;
5889 struct extent_buffer *leaf = path->nodes[0];
5892 if (found_key.objectid != key.objectid)
5894 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5896 if (found_key.offset < ctx->pos)
5898 if (found_key.offset > private->last_index)
5900 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5902 di = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
5903 name_len = btrfs_dir_name_len(leaf, di);
5904 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5906 btrfs_release_path(path);
5907 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5910 addr = private->filldir_buf;
5916 ftype = btrfs_dir_flags_to_ftype(btrfs_dir_flags(leaf, di));
5918 name_ptr = (char *)(entry + 1);
5919 read_extent_buffer(leaf, name_ptr,
5920 (unsigned long)(di + 1), name_len);
5921 put_unaligned(name_len, &entry->name_len);
5922 put_unaligned(fs_ftype_to_dtype(ftype), &entry->type);
5923 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5924 put_unaligned(location.objectid, &entry->ino);
5925 put_unaligned(found_key.offset, &entry->offset);
5927 addr += sizeof(struct dir_entry) + name_len;
5928 total_len += sizeof(struct dir_entry) + name_len;
5930 /* Catch error encountered during iteration */
5934 btrfs_release_path(path);
5936 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5940 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5945 * Stop new entries from being returned after we return the last
5948 * New directory entries are assigned a strictly increasing
5949 * offset. This means that new entries created during readdir
5950 * are *guaranteed* to be seen in the future by that readdir.
5951 * This has broken buggy programs which operate on names as
5952 * they're returned by readdir. Until we re-use freed offsets
5953 * we have this hack to stop new entries from being returned
5954 * under the assumption that they'll never reach this huge
5957 * This is being careful not to overflow 32bit loff_t unless the
5958 * last entry requires it because doing so has broken 32bit apps
5961 if (ctx->pos >= INT_MAX)
5962 ctx->pos = LLONG_MAX;
5969 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5970 btrfs_free_path(path);
5975 * This is somewhat expensive, updating the tree every time the
5976 * inode changes. But, it is most likely to find the inode in cache.
5977 * FIXME, needs more benchmarking...there are no reasons other than performance
5978 * to keep or drop this code.
5980 static int btrfs_dirty_inode(struct btrfs_inode *inode)
5982 struct btrfs_root *root = inode->root;
5983 struct btrfs_fs_info *fs_info = root->fs_info;
5984 struct btrfs_trans_handle *trans;
5987 if (test_bit(BTRFS_INODE_DUMMY, &inode->runtime_flags))
5990 trans = btrfs_join_transaction(root);
5992 return PTR_ERR(trans);
5994 ret = btrfs_update_inode(trans, root, inode);
5995 if (ret && (ret == -ENOSPC || ret == -EDQUOT)) {
5996 /* whoops, lets try again with the full transaction */
5997 btrfs_end_transaction(trans);
5998 trans = btrfs_start_transaction(root, 1);
6000 return PTR_ERR(trans);
6002 ret = btrfs_update_inode(trans, root, inode);
6004 btrfs_end_transaction(trans);
6005 if (inode->delayed_node)
6006 btrfs_balance_delayed_items(fs_info);
6012 * This is a copy of file_update_time. We need this so we can return error on
6013 * ENOSPC for updating the inode in the case of file write and mmap writes.
6015 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
6018 struct btrfs_root *root = BTRFS_I(inode)->root;
6019 bool dirty = flags & ~S_VERSION;
6021 if (btrfs_root_readonly(root))
6024 if (flags & S_VERSION)
6025 dirty |= inode_maybe_inc_iversion(inode, dirty);
6026 if (flags & S_CTIME)
6027 inode->i_ctime = *now;
6028 if (flags & S_MTIME)
6029 inode->i_mtime = *now;
6030 if (flags & S_ATIME)
6031 inode->i_atime = *now;
6032 return dirty ? btrfs_dirty_inode(BTRFS_I(inode)) : 0;
6036 * helper to find a free sequence number in a given directory. This current
6037 * code is very simple, later versions will do smarter things in the btree
6039 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6043 if (dir->index_cnt == (u64)-1) {
6044 ret = btrfs_inode_delayed_dir_index_count(dir);
6046 ret = btrfs_set_inode_index_count(dir);
6052 *index = dir->index_cnt;
6058 static int btrfs_insert_inode_locked(struct inode *inode)
6060 struct btrfs_iget_args args;
6062 args.ino = BTRFS_I(inode)->location.objectid;
6063 args.root = BTRFS_I(inode)->root;
6065 return insert_inode_locked4(inode,
6066 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6067 btrfs_find_actor, &args);
6070 int btrfs_new_inode_prepare(struct btrfs_new_inode_args *args,
6071 unsigned int *trans_num_items)
6073 struct inode *dir = args->dir;
6074 struct inode *inode = args->inode;
6077 if (!args->orphan) {
6078 ret = fscrypt_setup_filename(dir, &args->dentry->d_name, 0,
6084 ret = posix_acl_create(dir, &inode->i_mode, &args->default_acl, &args->acl);
6086 fscrypt_free_filename(&args->fname);
6090 /* 1 to add inode item */
6091 *trans_num_items = 1;
6092 /* 1 to add compression property */
6093 if (BTRFS_I(dir)->prop_compress)
6094 (*trans_num_items)++;
6095 /* 1 to add default ACL xattr */
6096 if (args->default_acl)
6097 (*trans_num_items)++;
6098 /* 1 to add access ACL xattr */
6100 (*trans_num_items)++;
6101 #ifdef CONFIG_SECURITY
6102 /* 1 to add LSM xattr */
6103 if (dir->i_security)
6104 (*trans_num_items)++;
6107 /* 1 to add orphan item */
6108 (*trans_num_items)++;
6112 * 1 to add dir index
6113 * 1 to update parent inode item
6115 * No need for 1 unit for the inode ref item because it is
6116 * inserted in a batch together with the inode item at
6117 * btrfs_create_new_inode().
6119 *trans_num_items += 3;
6124 void btrfs_new_inode_args_destroy(struct btrfs_new_inode_args *args)
6126 posix_acl_release(args->acl);
6127 posix_acl_release(args->default_acl);
6128 fscrypt_free_filename(&args->fname);
6132 * Inherit flags from the parent inode.
6134 * Currently only the compression flags and the cow flags are inherited.
6136 static void btrfs_inherit_iflags(struct btrfs_inode *inode, struct btrfs_inode *dir)
6142 if (flags & BTRFS_INODE_NOCOMPRESS) {
6143 inode->flags &= ~BTRFS_INODE_COMPRESS;
6144 inode->flags |= BTRFS_INODE_NOCOMPRESS;
6145 } else if (flags & BTRFS_INODE_COMPRESS) {
6146 inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
6147 inode->flags |= BTRFS_INODE_COMPRESS;
6150 if (flags & BTRFS_INODE_NODATACOW) {
6151 inode->flags |= BTRFS_INODE_NODATACOW;
6152 if (S_ISREG(inode->vfs_inode.i_mode))
6153 inode->flags |= BTRFS_INODE_NODATASUM;
6156 btrfs_sync_inode_flags_to_i_flags(&inode->vfs_inode);
6159 int btrfs_create_new_inode(struct btrfs_trans_handle *trans,
6160 struct btrfs_new_inode_args *args)
6162 struct inode *dir = args->dir;
6163 struct inode *inode = args->inode;
6164 const struct fscrypt_str *name = args->orphan ? NULL : &args->fname.disk_name;
6165 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6166 struct btrfs_root *root;
6167 struct btrfs_inode_item *inode_item;
6168 struct btrfs_key *location;
6169 struct btrfs_path *path;
6171 struct btrfs_inode_ref *ref;
6172 struct btrfs_key key[2];
6174 struct btrfs_item_batch batch;
6178 path = btrfs_alloc_path();
6183 BTRFS_I(inode)->root = btrfs_grab_root(BTRFS_I(dir)->root);
6184 root = BTRFS_I(inode)->root;
6186 ret = btrfs_get_free_objectid(root, &objectid);
6189 inode->i_ino = objectid;
6193 * O_TMPFILE, set link count to 0, so that after this point, we
6194 * fill in an inode item with the correct link count.
6196 set_nlink(inode, 0);
6198 trace_btrfs_inode_request(dir);
6200 ret = btrfs_set_inode_index(BTRFS_I(dir), &BTRFS_I(inode)->dir_index);
6204 /* index_cnt is ignored for everything but a dir. */
6205 BTRFS_I(inode)->index_cnt = BTRFS_DIR_START_INDEX;
6206 BTRFS_I(inode)->generation = trans->transid;
6207 inode->i_generation = BTRFS_I(inode)->generation;
6210 * Subvolumes don't inherit flags from their parent directory.
6211 * Originally this was probably by accident, but we probably can't
6212 * change it now without compatibility issues.
6215 btrfs_inherit_iflags(BTRFS_I(inode), BTRFS_I(dir));
6217 if (S_ISREG(inode->i_mode)) {
6218 if (btrfs_test_opt(fs_info, NODATASUM))
6219 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6220 if (btrfs_test_opt(fs_info, NODATACOW))
6221 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6222 BTRFS_INODE_NODATASUM;
6225 location = &BTRFS_I(inode)->location;
6226 location->objectid = objectid;
6227 location->offset = 0;
6228 location->type = BTRFS_INODE_ITEM_KEY;
6230 ret = btrfs_insert_inode_locked(inode);
6233 BTRFS_I(dir)->index_cnt--;
6238 * We could have gotten an inode number from somebody who was fsynced
6239 * and then removed in this same transaction, so let's just set full
6240 * sync since it will be a full sync anyway and this will blow away the
6241 * old info in the log.
6243 btrfs_set_inode_full_sync(BTRFS_I(inode));
6245 key[0].objectid = objectid;
6246 key[0].type = BTRFS_INODE_ITEM_KEY;
6249 sizes[0] = sizeof(struct btrfs_inode_item);
6251 if (!args->orphan) {
6253 * Start new inodes with an inode_ref. This is slightly more
6254 * efficient for small numbers of hard links since they will
6255 * be packed into one item. Extended refs will kick in if we
6256 * add more hard links than can fit in the ref item.
6258 key[1].objectid = objectid;
6259 key[1].type = BTRFS_INODE_REF_KEY;
6261 key[1].offset = objectid;
6262 sizes[1] = 2 + sizeof(*ref);
6264 key[1].offset = btrfs_ino(BTRFS_I(dir));
6265 sizes[1] = name->len + sizeof(*ref);
6269 batch.keys = &key[0];
6270 batch.data_sizes = &sizes[0];
6271 batch.total_data_size = sizes[0] + (args->orphan ? 0 : sizes[1]);
6272 batch.nr = args->orphan ? 1 : 2;
6273 ret = btrfs_insert_empty_items(trans, root, path, &batch);
6275 btrfs_abort_transaction(trans, ret);
6279 inode->i_mtime = current_time(inode);
6280 inode->i_atime = inode->i_mtime;
6281 inode->i_ctime = inode->i_mtime;
6282 BTRFS_I(inode)->i_otime = inode->i_mtime;
6285 * We're going to fill the inode item now, so at this point the inode
6286 * must be fully initialized.
6289 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6290 struct btrfs_inode_item);
6291 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6292 sizeof(*inode_item));
6293 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6295 if (!args->orphan) {
6296 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6297 struct btrfs_inode_ref);
6298 ptr = (unsigned long)(ref + 1);
6300 btrfs_set_inode_ref_name_len(path->nodes[0], ref, 2);
6301 btrfs_set_inode_ref_index(path->nodes[0], ref, 0);
6302 write_extent_buffer(path->nodes[0], "..", ptr, 2);
6304 btrfs_set_inode_ref_name_len(path->nodes[0], ref,
6306 btrfs_set_inode_ref_index(path->nodes[0], ref,
6307 BTRFS_I(inode)->dir_index);
6308 write_extent_buffer(path->nodes[0], name->name, ptr,
6313 btrfs_mark_buffer_dirty(path->nodes[0]);
6315 * We don't need the path anymore, plus inheriting properties, adding
6316 * ACLs, security xattrs, orphan item or adding the link, will result in
6317 * allocating yet another path. So just free our path.
6319 btrfs_free_path(path);
6323 struct inode *parent;
6326 * Subvolumes inherit properties from their parent subvolume,
6327 * not the directory they were created in.
6329 parent = btrfs_iget(fs_info->sb, BTRFS_FIRST_FREE_OBJECTID,
6330 BTRFS_I(dir)->root);
6331 if (IS_ERR(parent)) {
6332 ret = PTR_ERR(parent);
6334 ret = btrfs_inode_inherit_props(trans, inode, parent);
6338 ret = btrfs_inode_inherit_props(trans, inode, dir);
6342 "error inheriting props for ino %llu (root %llu): %d",
6343 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid,
6348 * Subvolumes don't inherit ACLs or get passed to the LSM. This is
6351 if (!args->subvol) {
6352 ret = btrfs_init_inode_security(trans, args);
6354 btrfs_abort_transaction(trans, ret);
6359 inode_tree_add(BTRFS_I(inode));
6361 trace_btrfs_inode_new(inode);
6362 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6364 btrfs_update_root_times(trans, root);
6367 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
6369 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
6370 0, BTRFS_I(inode)->dir_index);
6373 btrfs_abort_transaction(trans, ret);
6381 * discard_new_inode() calls iput(), but the caller owns the reference
6385 discard_new_inode(inode);
6387 btrfs_free_path(path);
6392 * utility function to add 'inode' into 'parent_inode' with
6393 * a give name and a given sequence number.
6394 * if 'add_backref' is true, also insert a backref from the
6395 * inode to the parent directory.
6397 int btrfs_add_link(struct btrfs_trans_handle *trans,
6398 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6399 const struct fscrypt_str *name, int add_backref, u64 index)
6402 struct btrfs_key key;
6403 struct btrfs_root *root = parent_inode->root;
6404 u64 ino = btrfs_ino(inode);
6405 u64 parent_ino = btrfs_ino(parent_inode);
6407 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6408 memcpy(&key, &inode->root->root_key, sizeof(key));
6411 key.type = BTRFS_INODE_ITEM_KEY;
6415 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6416 ret = btrfs_add_root_ref(trans, key.objectid,
6417 root->root_key.objectid, parent_ino,
6419 } else if (add_backref) {
6420 ret = btrfs_insert_inode_ref(trans, root, name,
6421 ino, parent_ino, index);
6424 /* Nothing to clean up yet */
6428 ret = btrfs_insert_dir_item(trans, name, parent_inode, &key,
6429 btrfs_inode_type(&inode->vfs_inode), index);
6430 if (ret == -EEXIST || ret == -EOVERFLOW)
6433 btrfs_abort_transaction(trans, ret);
6437 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6439 inode_inc_iversion(&parent_inode->vfs_inode);
6441 * If we are replaying a log tree, we do not want to update the mtime
6442 * and ctime of the parent directory with the current time, since the
6443 * log replay procedure is responsible for setting them to their correct
6444 * values (the ones it had when the fsync was done).
6446 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
6447 struct timespec64 now = current_time(&parent_inode->vfs_inode);
6449 parent_inode->vfs_inode.i_mtime = now;
6450 parent_inode->vfs_inode.i_ctime = now;
6452 ret = btrfs_update_inode(trans, root, parent_inode);
6454 btrfs_abort_transaction(trans, ret);
6458 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6461 err = btrfs_del_root_ref(trans, key.objectid,
6462 root->root_key.objectid, parent_ino,
6463 &local_index, name);
6465 btrfs_abort_transaction(trans, err);
6466 } else if (add_backref) {
6470 err = btrfs_del_inode_ref(trans, root, name, ino, parent_ino,
6473 btrfs_abort_transaction(trans, err);
6476 /* Return the original error code */
6480 static int btrfs_create_common(struct inode *dir, struct dentry *dentry,
6481 struct inode *inode)
6483 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6484 struct btrfs_root *root = BTRFS_I(dir)->root;
6485 struct btrfs_new_inode_args new_inode_args = {
6490 unsigned int trans_num_items;
6491 struct btrfs_trans_handle *trans;
6494 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
6498 trans = btrfs_start_transaction(root, trans_num_items);
6499 if (IS_ERR(trans)) {
6500 err = PTR_ERR(trans);
6501 goto out_new_inode_args;
6504 err = btrfs_create_new_inode(trans, &new_inode_args);
6506 d_instantiate_new(dentry, inode);
6508 btrfs_end_transaction(trans);
6509 btrfs_btree_balance_dirty(fs_info);
6511 btrfs_new_inode_args_destroy(&new_inode_args);
6518 static int btrfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
6519 struct dentry *dentry, umode_t mode, dev_t rdev)
6521 struct inode *inode;
6523 inode = new_inode(dir->i_sb);
6526 inode_init_owner(idmap, inode, dir, mode);
6527 inode->i_op = &btrfs_special_inode_operations;
6528 init_special_inode(inode, inode->i_mode, rdev);
6529 return btrfs_create_common(dir, dentry, inode);
6532 static int btrfs_create(struct mnt_idmap *idmap, struct inode *dir,
6533 struct dentry *dentry, umode_t mode, bool excl)
6535 struct inode *inode;
6537 inode = new_inode(dir->i_sb);
6540 inode_init_owner(idmap, inode, dir, mode);
6541 inode->i_fop = &btrfs_file_operations;
6542 inode->i_op = &btrfs_file_inode_operations;
6543 inode->i_mapping->a_ops = &btrfs_aops;
6544 return btrfs_create_common(dir, dentry, inode);
6547 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6548 struct dentry *dentry)
6550 struct btrfs_trans_handle *trans = NULL;
6551 struct btrfs_root *root = BTRFS_I(dir)->root;
6552 struct inode *inode = d_inode(old_dentry);
6553 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6554 struct fscrypt_name fname;
6559 /* do not allow sys_link's with other subvols of the same device */
6560 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6563 if (inode->i_nlink >= BTRFS_LINK_MAX)
6566 err = fscrypt_setup_filename(dir, &dentry->d_name, 0, &fname);
6570 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6575 * 2 items for inode and inode ref
6576 * 2 items for dir items
6577 * 1 item for parent inode
6578 * 1 item for orphan item deletion if O_TMPFILE
6580 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6581 if (IS_ERR(trans)) {
6582 err = PTR_ERR(trans);
6587 /* There are several dir indexes for this inode, clear the cache. */
6588 BTRFS_I(inode)->dir_index = 0ULL;
6590 inode_inc_iversion(inode);
6591 inode->i_ctime = current_time(inode);
6593 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6595 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6596 &fname.disk_name, 1, index);
6601 struct dentry *parent = dentry->d_parent;
6603 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6606 if (inode->i_nlink == 1) {
6608 * If new hard link count is 1, it's a file created
6609 * with open(2) O_TMPFILE flag.
6611 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6615 d_instantiate(dentry, inode);
6616 btrfs_log_new_name(trans, old_dentry, NULL, 0, parent);
6620 fscrypt_free_filename(&fname);
6622 btrfs_end_transaction(trans);
6624 inode_dec_link_count(inode);
6627 btrfs_btree_balance_dirty(fs_info);
6631 static int btrfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
6632 struct dentry *dentry, umode_t mode)
6634 struct inode *inode;
6636 inode = new_inode(dir->i_sb);
6639 inode_init_owner(idmap, inode, dir, S_IFDIR | mode);
6640 inode->i_op = &btrfs_dir_inode_operations;
6641 inode->i_fop = &btrfs_dir_file_operations;
6642 return btrfs_create_common(dir, dentry, inode);
6645 static noinline int uncompress_inline(struct btrfs_path *path,
6647 struct btrfs_file_extent_item *item)
6650 struct extent_buffer *leaf = path->nodes[0];
6653 unsigned long inline_size;
6657 compress_type = btrfs_file_extent_compression(leaf, item);
6658 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6659 inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]);
6660 tmp = kmalloc(inline_size, GFP_NOFS);
6663 ptr = btrfs_file_extent_inline_start(item);
6665 read_extent_buffer(leaf, tmp, ptr, inline_size);
6667 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6668 ret = btrfs_decompress(compress_type, tmp, page, 0, inline_size, max_size);
6671 * decompression code contains a memset to fill in any space between the end
6672 * of the uncompressed data and the end of max_size in case the decompressed
6673 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6674 * the end of an inline extent and the beginning of the next block, so we
6675 * cover that region here.
6678 if (max_size < PAGE_SIZE)
6679 memzero_page(page, max_size, PAGE_SIZE - max_size);
6684 static int read_inline_extent(struct btrfs_inode *inode, struct btrfs_path *path,
6687 struct btrfs_file_extent_item *fi;
6691 if (!page || PageUptodate(page))
6694 ASSERT(page_offset(page) == 0);
6696 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
6697 struct btrfs_file_extent_item);
6698 if (btrfs_file_extent_compression(path->nodes[0], fi) != BTRFS_COMPRESS_NONE)
6699 return uncompress_inline(path, page, fi);
6701 copy_size = min_t(u64, PAGE_SIZE,
6702 btrfs_file_extent_ram_bytes(path->nodes[0], fi));
6703 kaddr = kmap_local_page(page);
6704 read_extent_buffer(path->nodes[0], kaddr,
6705 btrfs_file_extent_inline_start(fi), copy_size);
6706 kunmap_local(kaddr);
6707 if (copy_size < PAGE_SIZE)
6708 memzero_page(page, copy_size, PAGE_SIZE - copy_size);
6713 * Lookup the first extent overlapping a range in a file.
6715 * @inode: file to search in
6716 * @page: page to read extent data into if the extent is inline
6717 * @pg_offset: offset into @page to copy to
6718 * @start: file offset
6719 * @len: length of range starting at @start
6721 * Return the first &struct extent_map which overlaps the given range, reading
6722 * it from the B-tree and caching it if necessary. Note that there may be more
6723 * extents which overlap the given range after the returned extent_map.
6725 * If @page is not NULL and the extent is inline, this also reads the extent
6726 * data directly into the page and marks the extent up to date in the io_tree.
6728 * Return: ERR_PTR on error, non-NULL extent_map on success.
6730 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6731 struct page *page, size_t pg_offset,
6734 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6736 u64 extent_start = 0;
6738 u64 objectid = btrfs_ino(inode);
6739 int extent_type = -1;
6740 struct btrfs_path *path = NULL;
6741 struct btrfs_root *root = inode->root;
6742 struct btrfs_file_extent_item *item;
6743 struct extent_buffer *leaf;
6744 struct btrfs_key found_key;
6745 struct extent_map *em = NULL;
6746 struct extent_map_tree *em_tree = &inode->extent_tree;
6748 read_lock(&em_tree->lock);
6749 em = lookup_extent_mapping(em_tree, start, len);
6750 read_unlock(&em_tree->lock);
6753 if (em->start > start || em->start + em->len <= start)
6754 free_extent_map(em);
6755 else if (em->block_start == EXTENT_MAP_INLINE && page)
6756 free_extent_map(em);
6760 em = alloc_extent_map();
6765 em->start = EXTENT_MAP_HOLE;
6766 em->orig_start = EXTENT_MAP_HOLE;
6768 em->block_len = (u64)-1;
6770 path = btrfs_alloc_path();
6776 /* Chances are we'll be called again, so go ahead and do readahead */
6777 path->reada = READA_FORWARD;
6780 * The same explanation in load_free_space_cache applies here as well,
6781 * we only read when we're loading the free space cache, and at that
6782 * point the commit_root has everything we need.
6784 if (btrfs_is_free_space_inode(inode)) {
6785 path->search_commit_root = 1;
6786 path->skip_locking = 1;
6789 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6792 } else if (ret > 0) {
6793 if (path->slots[0] == 0)
6799 leaf = path->nodes[0];
6800 item = btrfs_item_ptr(leaf, path->slots[0],
6801 struct btrfs_file_extent_item);
6802 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6803 if (found_key.objectid != objectid ||
6804 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6806 * If we backup past the first extent we want to move forward
6807 * and see if there is an extent in front of us, otherwise we'll
6808 * say there is a hole for our whole search range which can
6815 extent_type = btrfs_file_extent_type(leaf, item);
6816 extent_start = found_key.offset;
6817 extent_end = btrfs_file_extent_end(path);
6818 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6819 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6820 /* Only regular file could have regular/prealloc extent */
6821 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6824 "regular/prealloc extent found for non-regular inode %llu",
6828 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6830 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6831 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6836 if (start >= extent_end) {
6838 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6839 ret = btrfs_next_leaf(root, path);
6845 leaf = path->nodes[0];
6847 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6848 if (found_key.objectid != objectid ||
6849 found_key.type != BTRFS_EXTENT_DATA_KEY)
6851 if (start + len <= found_key.offset)
6853 if (start > found_key.offset)
6856 /* New extent overlaps with existing one */
6858 em->orig_start = start;
6859 em->len = found_key.offset - start;
6860 em->block_start = EXTENT_MAP_HOLE;
6864 btrfs_extent_item_to_extent_map(inode, path, item, em);
6866 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6867 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6869 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6871 * Inline extent can only exist at file offset 0. This is
6872 * ensured by tree-checker and inline extent creation path.
6873 * Thus all members representing file offsets should be zero.
6875 ASSERT(pg_offset == 0);
6876 ASSERT(extent_start == 0);
6877 ASSERT(em->start == 0);
6880 * btrfs_extent_item_to_extent_map() should have properly
6881 * initialized em members already.
6883 * Other members are not utilized for inline extents.
6885 ASSERT(em->block_start == EXTENT_MAP_INLINE);
6886 ASSERT(em->len == fs_info->sectorsize);
6888 ret = read_inline_extent(inode, path, page);
6895 em->orig_start = start;
6897 em->block_start = EXTENT_MAP_HOLE;
6900 btrfs_release_path(path);
6901 if (em->start > start || extent_map_end(em) <= start) {
6903 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6904 em->start, em->len, start, len);
6909 write_lock(&em_tree->lock);
6910 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6911 write_unlock(&em_tree->lock);
6913 btrfs_free_path(path);
6915 trace_btrfs_get_extent(root, inode, em);
6918 free_extent_map(em);
6919 return ERR_PTR(ret);
6924 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
6925 struct btrfs_dio_data *dio_data,
6928 const u64 orig_start,
6929 const u64 block_start,
6930 const u64 block_len,
6931 const u64 orig_block_len,
6932 const u64 ram_bytes,
6935 struct extent_map *em = NULL;
6936 struct btrfs_ordered_extent *ordered;
6938 if (type != BTRFS_ORDERED_NOCOW) {
6939 em = create_io_em(inode, start, len, orig_start, block_start,
6940 block_len, orig_block_len, ram_bytes,
6941 BTRFS_COMPRESS_NONE, /* compress_type */
6946 ordered = btrfs_alloc_ordered_extent(inode, start, len, len,
6947 block_start, block_len, 0,
6949 (1 << BTRFS_ORDERED_DIRECT),
6950 BTRFS_COMPRESS_NONE);
6951 if (IS_ERR(ordered)) {
6953 free_extent_map(em);
6954 btrfs_drop_extent_map_range(inode, start,
6955 start + len - 1, false);
6957 em = ERR_CAST(ordered);
6959 ASSERT(!dio_data->ordered);
6960 dio_data->ordered = ordered;
6967 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
6968 struct btrfs_dio_data *dio_data,
6971 struct btrfs_root *root = inode->root;
6972 struct btrfs_fs_info *fs_info = root->fs_info;
6973 struct extent_map *em;
6974 struct btrfs_key ins;
6978 alloc_hint = get_extent_allocation_hint(inode, start, len);
6979 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
6980 0, alloc_hint, &ins, 1, 1);
6982 return ERR_PTR(ret);
6984 em = btrfs_create_dio_extent(inode, dio_data, start, ins.offset, start,
6985 ins.objectid, ins.offset, ins.offset,
6986 ins.offset, BTRFS_ORDERED_REGULAR);
6987 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
6989 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
6995 static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
6997 struct btrfs_block_group *block_group;
6998 bool readonly = false;
7000 block_group = btrfs_lookup_block_group(fs_info, bytenr);
7001 if (!block_group || block_group->ro)
7004 btrfs_put_block_group(block_group);
7009 * Check if we can do nocow write into the range [@offset, @offset + @len)
7011 * @offset: File offset
7012 * @len: The length to write, will be updated to the nocow writeable
7014 * @orig_start: (optional) Return the original file offset of the file extent
7015 * @orig_len: (optional) Return the original on-disk length of the file extent
7016 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7017 * @strict: if true, omit optimizations that might force us into unnecessary
7018 * cow. e.g., don't trust generation number.
7021 * >0 and update @len if we can do nocow write
7022 * 0 if we can't do nocow write
7023 * <0 if error happened
7025 * NOTE: This only checks the file extents, caller is responsible to wait for
7026 * any ordered extents.
7028 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7029 u64 *orig_start, u64 *orig_block_len,
7030 u64 *ram_bytes, bool nowait, bool strict)
7032 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7033 struct can_nocow_file_extent_args nocow_args = { 0 };
7034 struct btrfs_path *path;
7036 struct extent_buffer *leaf;
7037 struct btrfs_root *root = BTRFS_I(inode)->root;
7038 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7039 struct btrfs_file_extent_item *fi;
7040 struct btrfs_key key;
7043 path = btrfs_alloc_path();
7046 path->nowait = nowait;
7048 ret = btrfs_lookup_file_extent(NULL, root, path,
7049 btrfs_ino(BTRFS_I(inode)), offset, 0);
7054 if (path->slots[0] == 0) {
7055 /* can't find the item, must cow */
7062 leaf = path->nodes[0];
7063 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7064 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7065 key.type != BTRFS_EXTENT_DATA_KEY) {
7066 /* not our file or wrong item type, must cow */
7070 if (key.offset > offset) {
7071 /* Wrong offset, must cow */
7075 if (btrfs_file_extent_end(path) <= offset)
7078 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
7079 found_type = btrfs_file_extent_type(leaf, fi);
7081 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7083 nocow_args.start = offset;
7084 nocow_args.end = offset + *len - 1;
7085 nocow_args.strict = strict;
7086 nocow_args.free_path = true;
7088 ret = can_nocow_file_extent(path, &key, BTRFS_I(inode), &nocow_args);
7089 /* can_nocow_file_extent() has freed the path. */
7093 /* Treat errors as not being able to NOCOW. */
7099 if (btrfs_extent_readonly(fs_info, nocow_args.disk_bytenr))
7102 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7103 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7106 range_end = round_up(offset + nocow_args.num_bytes,
7107 root->fs_info->sectorsize) - 1;
7108 ret = test_range_bit(io_tree, offset, range_end,
7109 EXTENT_DELALLOC, 0, NULL);
7117 *orig_start = key.offset - nocow_args.extent_offset;
7119 *orig_block_len = nocow_args.disk_num_bytes;
7121 *len = nocow_args.num_bytes;
7124 btrfs_free_path(path);
7128 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7129 struct extent_state **cached_state,
7130 unsigned int iomap_flags)
7132 const bool writing = (iomap_flags & IOMAP_WRITE);
7133 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7134 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7135 struct btrfs_ordered_extent *ordered;
7140 if (!try_lock_extent(io_tree, lockstart, lockend,
7144 lock_extent(io_tree, lockstart, lockend, cached_state);
7147 * We're concerned with the entire range that we're going to be
7148 * doing DIO to, so we need to make sure there's no ordered
7149 * extents in this range.
7151 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7152 lockend - lockstart + 1);
7155 * We need to make sure there are no buffered pages in this
7156 * range either, we could have raced between the invalidate in
7157 * generic_file_direct_write and locking the extent. The
7158 * invalidate needs to happen so that reads after a write do not
7162 (!writing || !filemap_range_has_page(inode->i_mapping,
7163 lockstart, lockend)))
7166 unlock_extent(io_tree, lockstart, lockend, cached_state);
7170 btrfs_put_ordered_extent(ordered);
7175 * If we are doing a DIO read and the ordered extent we
7176 * found is for a buffered write, we can not wait for it
7177 * to complete and retry, because if we do so we can
7178 * deadlock with concurrent buffered writes on page
7179 * locks. This happens only if our DIO read covers more
7180 * than one extent map, if at this point has already
7181 * created an ordered extent for a previous extent map
7182 * and locked its range in the inode's io tree, and a
7183 * concurrent write against that previous extent map's
7184 * range and this range started (we unlock the ranges
7185 * in the io tree only when the bios complete and
7186 * buffered writes always lock pages before attempting
7187 * to lock range in the io tree).
7190 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7191 btrfs_start_ordered_extent(ordered);
7193 ret = nowait ? -EAGAIN : -ENOTBLK;
7194 btrfs_put_ordered_extent(ordered);
7197 * We could trigger writeback for this range (and wait
7198 * for it to complete) and then invalidate the pages for
7199 * this range (through invalidate_inode_pages2_range()),
7200 * but that can lead us to a deadlock with a concurrent
7201 * call to readahead (a buffered read or a defrag call
7202 * triggered a readahead) on a page lock due to an
7203 * ordered dio extent we created before but did not have
7204 * yet a corresponding bio submitted (whence it can not
7205 * complete), which makes readahead wait for that
7206 * ordered extent to complete while holding a lock on
7209 ret = nowait ? -EAGAIN : -ENOTBLK;
7221 /* The callers of this must take lock_extent() */
7222 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7223 u64 len, u64 orig_start, u64 block_start,
7224 u64 block_len, u64 orig_block_len,
7225 u64 ram_bytes, int compress_type,
7228 struct extent_map *em;
7231 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7232 type == BTRFS_ORDERED_COMPRESSED ||
7233 type == BTRFS_ORDERED_NOCOW ||
7234 type == BTRFS_ORDERED_REGULAR);
7236 em = alloc_extent_map();
7238 return ERR_PTR(-ENOMEM);
7241 em->orig_start = orig_start;
7243 em->block_len = block_len;
7244 em->block_start = block_start;
7245 em->orig_block_len = orig_block_len;
7246 em->ram_bytes = ram_bytes;
7247 em->generation = -1;
7248 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7249 if (type == BTRFS_ORDERED_PREALLOC) {
7250 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7251 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7252 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7253 em->compress_type = compress_type;
7256 ret = btrfs_replace_extent_map_range(inode, em, true);
7258 free_extent_map(em);
7259 return ERR_PTR(ret);
7262 /* em got 2 refs now, callers needs to do free_extent_map once. */
7267 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7268 struct inode *inode,
7269 struct btrfs_dio_data *dio_data,
7270 u64 start, u64 *lenp,
7271 unsigned int iomap_flags)
7273 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7274 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7275 struct extent_map *em = *map;
7277 u64 block_start, orig_start, orig_block_len, ram_bytes;
7278 struct btrfs_block_group *bg;
7279 bool can_nocow = false;
7280 bool space_reserved = false;
7286 * We don't allocate a new extent in the following cases
7288 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7290 * 2) The extent is marked as PREALLOC. We're good to go here and can
7291 * just use the extent.
7294 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7295 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7296 em->block_start != EXTENT_MAP_HOLE)) {
7297 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7298 type = BTRFS_ORDERED_PREALLOC;
7300 type = BTRFS_ORDERED_NOCOW;
7301 len = min(len, em->len - (start - em->start));
7302 block_start = em->block_start + (start - em->start);
7304 if (can_nocow_extent(inode, start, &len, &orig_start,
7305 &orig_block_len, &ram_bytes, false, false) == 1) {
7306 bg = btrfs_inc_nocow_writers(fs_info, block_start);
7314 struct extent_map *em2;
7316 /* We can NOCOW, so only need to reserve metadata space. */
7317 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7320 /* Our caller expects us to free the input extent map. */
7321 free_extent_map(em);
7323 btrfs_dec_nocow_writers(bg);
7324 if (nowait && (ret == -ENOSPC || ret == -EDQUOT))
7328 space_reserved = true;
7330 em2 = btrfs_create_dio_extent(BTRFS_I(inode), dio_data, start, len,
7331 orig_start, block_start,
7332 len, orig_block_len,
7334 btrfs_dec_nocow_writers(bg);
7335 if (type == BTRFS_ORDERED_PREALLOC) {
7336 free_extent_map(em);
7346 dio_data->nocow_done = true;
7348 /* Our caller expects us to free the input extent map. */
7349 free_extent_map(em);
7358 * If we could not allocate data space before locking the file
7359 * range and we can't do a NOCOW write, then we have to fail.
7361 if (!dio_data->data_space_reserved) {
7367 * We have to COW and we have already reserved data space before,
7368 * so now we reserve only metadata.
7370 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7374 space_reserved = true;
7376 em = btrfs_new_extent_direct(BTRFS_I(inode), dio_data, start, len);
7382 len = min(len, em->len - (start - em->start));
7384 btrfs_delalloc_release_metadata(BTRFS_I(inode),
7385 prev_len - len, true);
7389 * We have created our ordered extent, so we can now release our reservation
7390 * for an outstanding extent.
7392 btrfs_delalloc_release_extents(BTRFS_I(inode), prev_len);
7395 * Need to update the i_size under the extent lock so buffered
7396 * readers will get the updated i_size when we unlock.
7398 if (start + len > i_size_read(inode))
7399 i_size_write(inode, start + len);
7401 if (ret && space_reserved) {
7402 btrfs_delalloc_release_extents(BTRFS_I(inode), len);
7403 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, true);
7409 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7410 loff_t length, unsigned int flags, struct iomap *iomap,
7411 struct iomap *srcmap)
7413 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7414 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7415 struct extent_map *em;
7416 struct extent_state *cached_state = NULL;
7417 struct btrfs_dio_data *dio_data = iter->private;
7418 u64 lockstart, lockend;
7419 const bool write = !!(flags & IOMAP_WRITE);
7422 const u64 data_alloc_len = length;
7423 bool unlock_extents = false;
7426 * We could potentially fault if we have a buffer > PAGE_SIZE, and if
7427 * we're NOWAIT we may submit a bio for a partial range and return
7428 * EIOCBQUEUED, which would result in an errant short read.
7430 * The best way to handle this would be to allow for partial completions
7431 * of iocb's, so we could submit the partial bio, return and fault in
7432 * the rest of the pages, and then submit the io for the rest of the
7433 * range. However we don't have that currently, so simply return
7434 * -EAGAIN at this point so that the normal path is used.
7436 if (!write && (flags & IOMAP_NOWAIT) && length > PAGE_SIZE)
7440 * Cap the size of reads to that usually seen in buffered I/O as we need
7441 * to allocate a contiguous array for the checksums.
7444 len = min_t(u64, len, fs_info->sectorsize * BTRFS_MAX_BIO_SECTORS);
7447 lockend = start + len - 1;
7450 * iomap_dio_rw() only does filemap_write_and_wait_range(), which isn't
7451 * enough if we've written compressed pages to this area, so we need to
7452 * flush the dirty pages again to make absolutely sure that any
7453 * outstanding dirty pages are on disk - the first flush only starts
7454 * compression on the data, while keeping the pages locked, so by the
7455 * time the second flush returns we know bios for the compressed pages
7456 * were submitted and finished, and the pages no longer under writeback.
7458 * If we have a NOWAIT request and we have any pages in the range that
7459 * are locked, likely due to compression still in progress, we don't want
7460 * to block on page locks. We also don't want to block on pages marked as
7461 * dirty or under writeback (same as for the non-compression case).
7462 * iomap_dio_rw() did the same check, but after that and before we got
7463 * here, mmap'ed writes may have happened or buffered reads started
7464 * (readpage() and readahead(), which lock pages), as we haven't locked
7465 * the file range yet.
7467 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7468 &BTRFS_I(inode)->runtime_flags)) {
7469 if (flags & IOMAP_NOWAIT) {
7470 if (filemap_range_needs_writeback(inode->i_mapping,
7471 lockstart, lockend))
7474 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7475 start + length - 1);
7481 memset(dio_data, 0, sizeof(*dio_data));
7484 * We always try to allocate data space and must do it before locking
7485 * the file range, to avoid deadlocks with concurrent writes to the same
7486 * range if the range has several extents and the writes don't expand the
7487 * current i_size (the inode lock is taken in shared mode). If we fail to
7488 * allocate data space here we continue and later, after locking the
7489 * file range, we fail with ENOSPC only if we figure out we can not do a
7492 if (write && !(flags & IOMAP_NOWAIT)) {
7493 ret = btrfs_check_data_free_space(BTRFS_I(inode),
7494 &dio_data->data_reserved,
7495 start, data_alloc_len, false);
7497 dio_data->data_space_reserved = true;
7498 else if (ret && !(BTRFS_I(inode)->flags &
7499 (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
7504 * If this errors out it's because we couldn't invalidate pagecache for
7505 * this range and we need to fallback to buffered IO, or we are doing a
7506 * NOWAIT read/write and we need to block.
7508 ret = lock_extent_direct(inode, lockstart, lockend, &cached_state, flags);
7512 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7519 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7520 * io. INLINE is special, and we could probably kludge it in here, but
7521 * it's still buffered so for safety lets just fall back to the generic
7524 * For COMPRESSED we _have_ to read the entire extent in so we can
7525 * decompress it, so there will be buffering required no matter what we
7526 * do, so go ahead and fallback to buffered.
7528 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7529 * to buffered IO. Don't blame me, this is the price we pay for using
7532 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7533 em->block_start == EXTENT_MAP_INLINE) {
7534 free_extent_map(em);
7536 * If we are in a NOWAIT context, return -EAGAIN in order to
7537 * fallback to buffered IO. This is not only because we can
7538 * block with buffered IO (no support for NOWAIT semantics at
7539 * the moment) but also to avoid returning short reads to user
7540 * space - this happens if we were able to read some data from
7541 * previous non-compressed extents and then when we fallback to
7542 * buffered IO, at btrfs_file_read_iter() by calling
7543 * filemap_read(), we fail to fault in pages for the read buffer,
7544 * in which case filemap_read() returns a short read (the number
7545 * of bytes previously read is > 0, so it does not return -EFAULT).
7547 ret = (flags & IOMAP_NOWAIT) ? -EAGAIN : -ENOTBLK;
7551 len = min(len, em->len - (start - em->start));
7554 * If we have a NOWAIT request and the range contains multiple extents
7555 * (or a mix of extents and holes), then we return -EAGAIN to make the
7556 * caller fallback to a context where it can do a blocking (without
7557 * NOWAIT) request. This way we avoid doing partial IO and returning
7558 * success to the caller, which is not optimal for writes and for reads
7559 * it can result in unexpected behaviour for an application.
7561 * When doing a read, because we use IOMAP_DIO_PARTIAL when calling
7562 * iomap_dio_rw(), we can end up returning less data then what the caller
7563 * asked for, resulting in an unexpected, and incorrect, short read.
7564 * That is, the caller asked to read N bytes and we return less than that,
7565 * which is wrong unless we are crossing EOF. This happens if we get a
7566 * page fault error when trying to fault in pages for the buffer that is
7567 * associated to the struct iov_iter passed to iomap_dio_rw(), and we
7568 * have previously submitted bios for other extents in the range, in
7569 * which case iomap_dio_rw() may return us EIOCBQUEUED if not all of
7570 * those bios have completed by the time we get the page fault error,
7571 * which we return back to our caller - we should only return EIOCBQUEUED
7572 * after we have submitted bios for all the extents in the range.
7574 if ((flags & IOMAP_NOWAIT) && len < length) {
7575 free_extent_map(em);
7581 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7582 start, &len, flags);
7585 unlock_extents = true;
7586 /* Recalc len in case the new em is smaller than requested */
7587 len = min(len, em->len - (start - em->start));
7588 if (dio_data->data_space_reserved) {
7590 u64 release_len = 0;
7592 if (dio_data->nocow_done) {
7593 release_offset = start;
7594 release_len = data_alloc_len;
7595 } else if (len < data_alloc_len) {
7596 release_offset = start + len;
7597 release_len = data_alloc_len - len;
7600 if (release_len > 0)
7601 btrfs_free_reserved_data_space(BTRFS_I(inode),
7602 dio_data->data_reserved,
7608 * We need to unlock only the end area that we aren't using.
7609 * The rest is going to be unlocked by the endio routine.
7611 lockstart = start + len;
7612 if (lockstart < lockend)
7613 unlock_extents = true;
7617 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7620 free_extent_state(cached_state);
7623 * Translate extent map information to iomap.
7624 * We trim the extents (and move the addr) even though iomap code does
7625 * that, since we have locked only the parts we are performing I/O in.
7627 if ((em->block_start == EXTENT_MAP_HOLE) ||
7628 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7629 iomap->addr = IOMAP_NULL_ADDR;
7630 iomap->type = IOMAP_HOLE;
7632 iomap->addr = em->block_start + (start - em->start);
7633 iomap->type = IOMAP_MAPPED;
7635 iomap->offset = start;
7636 iomap->bdev = fs_info->fs_devices->latest_dev->bdev;
7637 iomap->length = len;
7638 free_extent_map(em);
7643 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7646 if (dio_data->data_space_reserved) {
7647 btrfs_free_reserved_data_space(BTRFS_I(inode),
7648 dio_data->data_reserved,
7649 start, data_alloc_len);
7650 extent_changeset_free(dio_data->data_reserved);
7656 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7657 ssize_t written, unsigned int flags, struct iomap *iomap)
7659 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7660 struct btrfs_dio_data *dio_data = iter->private;
7661 size_t submitted = dio_data->submitted;
7662 const bool write = !!(flags & IOMAP_WRITE);
7665 if (!write && (iomap->type == IOMAP_HOLE)) {
7666 /* If reading from a hole, unlock and return */
7667 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1,
7672 if (submitted < length) {
7674 length -= submitted;
7676 btrfs_finish_ordered_extent(dio_data->ordered, NULL,
7677 pos, length, false);
7679 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7680 pos + length - 1, NULL);
7684 btrfs_put_ordered_extent(dio_data->ordered);
7685 dio_data->ordered = NULL;
7689 extent_changeset_free(dio_data->data_reserved);
7693 static void btrfs_dio_end_io(struct btrfs_bio *bbio)
7695 struct btrfs_dio_private *dip =
7696 container_of(bbio, struct btrfs_dio_private, bbio);
7697 struct btrfs_inode *inode = bbio->inode;
7698 struct bio *bio = &bbio->bio;
7700 if (bio->bi_status) {
7701 btrfs_warn(inode->root->fs_info,
7702 "direct IO failed ino %llu op 0x%0x offset %#llx len %u err no %d",
7703 btrfs_ino(inode), bio->bi_opf,
7704 dip->file_offset, dip->bytes, bio->bi_status);
7707 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
7708 btrfs_finish_ordered_extent(bbio->ordered, NULL,
7709 dip->file_offset, dip->bytes,
7712 unlock_extent(&inode->io_tree, dip->file_offset,
7713 dip->file_offset + dip->bytes - 1, NULL);
7716 bbio->bio.bi_private = bbio->private;
7717 iomap_dio_bio_end_io(bio);
7720 static void btrfs_dio_submit_io(const struct iomap_iter *iter, struct bio *bio,
7723 struct btrfs_bio *bbio = btrfs_bio(bio);
7724 struct btrfs_dio_private *dip =
7725 container_of(bbio, struct btrfs_dio_private, bbio);
7726 struct btrfs_dio_data *dio_data = iter->private;
7728 btrfs_bio_init(bbio, BTRFS_I(iter->inode)->root->fs_info,
7729 btrfs_dio_end_io, bio->bi_private);
7730 bbio->inode = BTRFS_I(iter->inode);
7731 bbio->file_offset = file_offset;
7733 dip->file_offset = file_offset;
7734 dip->bytes = bio->bi_iter.bi_size;
7736 dio_data->submitted += bio->bi_iter.bi_size;
7739 * Check if we are doing a partial write. If we are, we need to split
7740 * the ordered extent to match the submitted bio. Hang on to the
7741 * remaining unfinishable ordered_extent in dio_data so that it can be
7742 * cancelled in iomap_end to avoid a deadlock wherein faulting the
7743 * remaining pages is blocked on the outstanding ordered extent.
7745 if (iter->flags & IOMAP_WRITE) {
7748 ret = btrfs_extract_ordered_extent(bbio, dio_data->ordered);
7750 btrfs_finish_ordered_extent(dio_data->ordered, NULL,
7751 file_offset, dip->bytes,
7753 bio->bi_status = errno_to_blk_status(ret);
7754 iomap_dio_bio_end_io(bio);
7759 btrfs_submit_bio(bbio, 0);
7762 static const struct iomap_ops btrfs_dio_iomap_ops = {
7763 .iomap_begin = btrfs_dio_iomap_begin,
7764 .iomap_end = btrfs_dio_iomap_end,
7767 static const struct iomap_dio_ops btrfs_dio_ops = {
7768 .submit_io = btrfs_dio_submit_io,
7769 .bio_set = &btrfs_dio_bioset,
7772 ssize_t btrfs_dio_read(struct kiocb *iocb, struct iov_iter *iter, size_t done_before)
7774 struct btrfs_dio_data data = { 0 };
7776 return iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7777 IOMAP_DIO_PARTIAL, &data, done_before);
7780 struct iomap_dio *btrfs_dio_write(struct kiocb *iocb, struct iov_iter *iter,
7783 struct btrfs_dio_data data = { 0 };
7785 return __iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7786 IOMAP_DIO_PARTIAL, &data, done_before);
7789 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7794 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
7799 * fiemap_prep() called filemap_write_and_wait() for the whole possible
7800 * file range (0 to LLONG_MAX), but that is not enough if we have
7801 * compression enabled. The first filemap_fdatawrite_range() only kicks
7802 * in the compression of data (in an async thread) and will return
7803 * before the compression is done and writeback is started. A second
7804 * filemap_fdatawrite_range() is needed to wait for the compression to
7805 * complete and writeback to start. We also need to wait for ordered
7806 * extents to complete, because our fiemap implementation uses mainly
7807 * file extent items to list the extents, searching for extent maps
7808 * only for file ranges with holes or prealloc extents to figure out
7809 * if we have delalloc in those ranges.
7811 if (fieinfo->fi_flags & FIEMAP_FLAG_SYNC) {
7812 ret = btrfs_wait_ordered_range(inode, 0, LLONG_MAX);
7817 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
7820 static int btrfs_writepages(struct address_space *mapping,
7821 struct writeback_control *wbc)
7823 return extent_writepages(mapping, wbc);
7826 static void btrfs_readahead(struct readahead_control *rac)
7828 extent_readahead(rac);
7832 * For release_folio() and invalidate_folio() we have a race window where
7833 * folio_end_writeback() is called but the subpage spinlock is not yet released.
7834 * If we continue to release/invalidate the page, we could cause use-after-free
7835 * for subpage spinlock. So this function is to spin and wait for subpage
7838 static void wait_subpage_spinlock(struct page *page)
7840 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
7841 struct btrfs_subpage *subpage;
7843 if (!btrfs_is_subpage(fs_info, page))
7846 ASSERT(PagePrivate(page) && page->private);
7847 subpage = (struct btrfs_subpage *)page->private;
7850 * This may look insane as we just acquire the spinlock and release it,
7851 * without doing anything. But we just want to make sure no one is
7852 * still holding the subpage spinlock.
7853 * And since the page is not dirty nor writeback, and we have page
7854 * locked, the only possible way to hold a spinlock is from the endio
7855 * function to clear page writeback.
7857 * Here we just acquire the spinlock so that all existing callers
7858 * should exit and we're safe to release/invalidate the page.
7860 spin_lock_irq(&subpage->lock);
7861 spin_unlock_irq(&subpage->lock);
7864 static bool __btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
7866 int ret = try_release_extent_mapping(&folio->page, gfp_flags);
7869 wait_subpage_spinlock(&folio->page);
7870 clear_page_extent_mapped(&folio->page);
7875 static bool btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
7877 if (folio_test_writeback(folio) || folio_test_dirty(folio))
7879 return __btrfs_release_folio(folio, gfp_flags);
7882 #ifdef CONFIG_MIGRATION
7883 static int btrfs_migrate_folio(struct address_space *mapping,
7884 struct folio *dst, struct folio *src,
7885 enum migrate_mode mode)
7887 int ret = filemap_migrate_folio(mapping, dst, src, mode);
7889 if (ret != MIGRATEPAGE_SUCCESS)
7892 if (folio_test_ordered(src)) {
7893 folio_clear_ordered(src);
7894 folio_set_ordered(dst);
7897 return MIGRATEPAGE_SUCCESS;
7900 #define btrfs_migrate_folio NULL
7903 static void btrfs_invalidate_folio(struct folio *folio, size_t offset,
7906 struct btrfs_inode *inode = BTRFS_I(folio->mapping->host);
7907 struct btrfs_fs_info *fs_info = inode->root->fs_info;
7908 struct extent_io_tree *tree = &inode->io_tree;
7909 struct extent_state *cached_state = NULL;
7910 u64 page_start = folio_pos(folio);
7911 u64 page_end = page_start + folio_size(folio) - 1;
7913 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
7916 * We have folio locked so no new ordered extent can be created on this
7917 * page, nor bio can be submitted for this folio.
7919 * But already submitted bio can still be finished on this folio.
7920 * Furthermore, endio function won't skip folio which has Ordered
7921 * (Private2) already cleared, so it's possible for endio and
7922 * invalidate_folio to do the same ordered extent accounting twice
7925 * So here we wait for any submitted bios to finish, so that we won't
7926 * do double ordered extent accounting on the same folio.
7928 folio_wait_writeback(folio);
7929 wait_subpage_spinlock(&folio->page);
7932 * For subpage case, we have call sites like
7933 * btrfs_punch_hole_lock_range() which passes range not aligned to
7935 * If the range doesn't cover the full folio, we don't need to and
7936 * shouldn't clear page extent mapped, as folio->private can still
7937 * record subpage dirty bits for other part of the range.
7939 * For cases that invalidate the full folio even the range doesn't
7940 * cover the full folio, like invalidating the last folio, we're
7941 * still safe to wait for ordered extent to finish.
7943 if (!(offset == 0 && length == folio_size(folio))) {
7944 btrfs_release_folio(folio, GFP_NOFS);
7948 if (!inode_evicting)
7949 lock_extent(tree, page_start, page_end, &cached_state);
7952 while (cur < page_end) {
7953 struct btrfs_ordered_extent *ordered;
7956 u32 extra_flags = 0;
7958 ordered = btrfs_lookup_first_ordered_range(inode, cur,
7959 page_end + 1 - cur);
7961 range_end = page_end;
7963 * No ordered extent covering this range, we are safe
7964 * to delete all extent states in the range.
7966 extra_flags = EXTENT_CLEAR_ALL_BITS;
7969 if (ordered->file_offset > cur) {
7971 * There is a range between [cur, oe->file_offset) not
7972 * covered by any ordered extent.
7973 * We are safe to delete all extent states, and handle
7974 * the ordered extent in the next iteration.
7976 range_end = ordered->file_offset - 1;
7977 extra_flags = EXTENT_CLEAR_ALL_BITS;
7981 range_end = min(ordered->file_offset + ordered->num_bytes - 1,
7983 ASSERT(range_end + 1 - cur < U32_MAX);
7984 range_len = range_end + 1 - cur;
7985 if (!btrfs_page_test_ordered(fs_info, &folio->page, cur, range_len)) {
7987 * If Ordered (Private2) is cleared, it means endio has
7988 * already been executed for the range.
7989 * We can't delete the extent states as
7990 * btrfs_finish_ordered_io() may still use some of them.
7994 btrfs_page_clear_ordered(fs_info, &folio->page, cur, range_len);
7997 * IO on this page will never be started, so we need to account
7998 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
7999 * here, must leave that up for the ordered extent completion.
8001 * This will also unlock the range for incoming
8002 * btrfs_finish_ordered_io().
8004 if (!inode_evicting)
8005 clear_extent_bit(tree, cur, range_end,
8007 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8008 EXTENT_DEFRAG, &cached_state);
8010 spin_lock_irq(&inode->ordered_tree.lock);
8011 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8012 ordered->truncated_len = min(ordered->truncated_len,
8013 cur - ordered->file_offset);
8014 spin_unlock_irq(&inode->ordered_tree.lock);
8017 * If the ordered extent has finished, we're safe to delete all
8018 * the extent states of the range, otherwise
8019 * btrfs_finish_ordered_io() will get executed by endio for
8020 * other pages, so we can't delete extent states.
8022 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8023 cur, range_end + 1 - cur)) {
8024 btrfs_finish_ordered_io(ordered);
8026 * The ordered extent has finished, now we're again
8027 * safe to delete all extent states of the range.
8029 extra_flags = EXTENT_CLEAR_ALL_BITS;
8033 btrfs_put_ordered_extent(ordered);
8035 * Qgroup reserved space handler
8036 * Sector(s) here will be either:
8038 * 1) Already written to disk or bio already finished
8039 * Then its QGROUP_RESERVED bit in io_tree is already cleared.
8040 * Qgroup will be handled by its qgroup_record then.
8041 * btrfs_qgroup_free_data() call will do nothing here.
8043 * 2) Not written to disk yet
8044 * Then btrfs_qgroup_free_data() call will clear the
8045 * QGROUP_RESERVED bit of its io_tree, and free the qgroup
8046 * reserved data space.
8047 * Since the IO will never happen for this page.
8049 btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur);
8050 if (!inode_evicting) {
8051 clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED |
8052 EXTENT_DELALLOC | EXTENT_UPTODATE |
8053 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG |
8054 extra_flags, &cached_state);
8056 cur = range_end + 1;
8059 * We have iterated through all ordered extents of the page, the page
8060 * should not have Ordered (Private2) anymore, or the above iteration
8061 * did something wrong.
8063 ASSERT(!folio_test_ordered(folio));
8064 btrfs_page_clear_checked(fs_info, &folio->page, folio_pos(folio), folio_size(folio));
8065 if (!inode_evicting)
8066 __btrfs_release_folio(folio, GFP_NOFS);
8067 clear_page_extent_mapped(&folio->page);
8071 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8072 * called from a page fault handler when a page is first dirtied. Hence we must
8073 * be careful to check for EOF conditions here. We set the page up correctly
8074 * for a written page which means we get ENOSPC checking when writing into
8075 * holes and correct delalloc and unwritten extent mapping on filesystems that
8076 * support these features.
8078 * We are not allowed to take the i_mutex here so we have to play games to
8079 * protect against truncate races as the page could now be beyond EOF. Because
8080 * truncate_setsize() writes the inode size before removing pages, once we have
8081 * the page lock we can determine safely if the page is beyond EOF. If it is not
8082 * beyond EOF, then the page is guaranteed safe against truncation until we
8085 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8087 struct page *page = vmf->page;
8088 struct inode *inode = file_inode(vmf->vma->vm_file);
8089 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8090 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8091 struct btrfs_ordered_extent *ordered;
8092 struct extent_state *cached_state = NULL;
8093 struct extent_changeset *data_reserved = NULL;
8094 unsigned long zero_start;
8104 reserved_space = PAGE_SIZE;
8106 sb_start_pagefault(inode->i_sb);
8107 page_start = page_offset(page);
8108 page_end = page_start + PAGE_SIZE - 1;
8112 * Reserving delalloc space after obtaining the page lock can lead to
8113 * deadlock. For example, if a dirty page is locked by this function
8114 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8115 * dirty page write out, then the btrfs_writepages() function could
8116 * end up waiting indefinitely to get a lock on the page currently
8117 * being processed by btrfs_page_mkwrite() function.
8119 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8120 page_start, reserved_space);
8122 ret2 = file_update_time(vmf->vma->vm_file);
8126 ret = vmf_error(ret2);
8132 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8134 down_read(&BTRFS_I(inode)->i_mmap_lock);
8136 size = i_size_read(inode);
8138 if ((page->mapping != inode->i_mapping) ||
8139 (page_start >= size)) {
8140 /* page got truncated out from underneath us */
8143 wait_on_page_writeback(page);
8145 lock_extent(io_tree, page_start, page_end, &cached_state);
8146 ret2 = set_page_extent_mapped(page);
8148 ret = vmf_error(ret2);
8149 unlock_extent(io_tree, page_start, page_end, &cached_state);
8154 * we can't set the delalloc bits if there are pending ordered
8155 * extents. Drop our locks and wait for them to finish
8157 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8160 unlock_extent(io_tree, page_start, page_end, &cached_state);
8162 up_read(&BTRFS_I(inode)->i_mmap_lock);
8163 btrfs_start_ordered_extent(ordered);
8164 btrfs_put_ordered_extent(ordered);
8168 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8169 reserved_space = round_up(size - page_start,
8170 fs_info->sectorsize);
8171 if (reserved_space < PAGE_SIZE) {
8172 end = page_start + reserved_space - 1;
8173 btrfs_delalloc_release_space(BTRFS_I(inode),
8174 data_reserved, page_start,
8175 PAGE_SIZE - reserved_space, true);
8180 * page_mkwrite gets called when the page is firstly dirtied after it's
8181 * faulted in, but write(2) could also dirty a page and set delalloc
8182 * bits, thus in this case for space account reason, we still need to
8183 * clear any delalloc bits within this page range since we have to
8184 * reserve data&meta space before lock_page() (see above comments).
8186 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8187 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8188 EXTENT_DEFRAG, &cached_state);
8190 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8193 unlock_extent(io_tree, page_start, page_end, &cached_state);
8194 ret = VM_FAULT_SIGBUS;
8198 /* page is wholly or partially inside EOF */
8199 if (page_start + PAGE_SIZE > size)
8200 zero_start = offset_in_page(size);
8202 zero_start = PAGE_SIZE;
8204 if (zero_start != PAGE_SIZE)
8205 memzero_page(page, zero_start, PAGE_SIZE - zero_start);
8207 btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE);
8208 btrfs_page_set_dirty(fs_info, page, page_start, end + 1 - page_start);
8209 btrfs_page_set_uptodate(fs_info, page, page_start, end + 1 - page_start);
8211 btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
8213 unlock_extent(io_tree, page_start, page_end, &cached_state);
8214 up_read(&BTRFS_I(inode)->i_mmap_lock);
8216 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8217 sb_end_pagefault(inode->i_sb);
8218 extent_changeset_free(data_reserved);
8219 return VM_FAULT_LOCKED;
8223 up_read(&BTRFS_I(inode)->i_mmap_lock);
8225 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8226 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8227 reserved_space, (ret != 0));
8229 sb_end_pagefault(inode->i_sb);
8230 extent_changeset_free(data_reserved);
8234 static int btrfs_truncate(struct btrfs_inode *inode, bool skip_writeback)
8236 struct btrfs_truncate_control control = {
8238 .ino = btrfs_ino(inode),
8239 .min_type = BTRFS_EXTENT_DATA_KEY,
8240 .clear_extent_range = true,
8242 struct btrfs_root *root = inode->root;
8243 struct btrfs_fs_info *fs_info = root->fs_info;
8244 struct btrfs_block_rsv *rsv;
8246 struct btrfs_trans_handle *trans;
8247 u64 mask = fs_info->sectorsize - 1;
8248 const u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8250 if (!skip_writeback) {
8251 ret = btrfs_wait_ordered_range(&inode->vfs_inode,
8252 inode->vfs_inode.i_size & (~mask),
8259 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8260 * things going on here:
8262 * 1) We need to reserve space to update our inode.
8264 * 2) We need to have something to cache all the space that is going to
8265 * be free'd up by the truncate operation, but also have some slack
8266 * space reserved in case it uses space during the truncate (thank you
8267 * very much snapshotting).
8269 * And we need these to be separate. The fact is we can use a lot of
8270 * space doing the truncate, and we have no earthly idea how much space
8271 * we will use, so we need the truncate reservation to be separate so it
8272 * doesn't end up using space reserved for updating the inode. We also
8273 * need to be able to stop the transaction and start a new one, which
8274 * means we need to be able to update the inode several times, and we
8275 * have no idea of knowing how many times that will be, so we can't just
8276 * reserve 1 item for the entirety of the operation, so that has to be
8277 * done separately as well.
8279 * So that leaves us with
8281 * 1) rsv - for the truncate reservation, which we will steal from the
8282 * transaction reservation.
8283 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8284 * updating the inode.
8286 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8289 rsv->size = min_size;
8290 rsv->failfast = true;
8293 * 1 for the truncate slack space
8294 * 1 for updating the inode.
8296 trans = btrfs_start_transaction(root, 2);
8297 if (IS_ERR(trans)) {
8298 ret = PTR_ERR(trans);
8302 /* Migrate the slack space for the truncate to our reserve */
8303 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8306 * We have reserved 2 metadata units when we started the transaction and
8307 * min_size matches 1 unit, so this should never fail, but if it does,
8308 * it's not critical we just fail truncation.
8311 btrfs_end_transaction(trans);
8315 trans->block_rsv = rsv;
8318 struct extent_state *cached_state = NULL;
8319 const u64 new_size = inode->vfs_inode.i_size;
8320 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
8322 control.new_size = new_size;
8323 lock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8325 * We want to drop from the next block forward in case this new
8326 * size is not block aligned since we will be keeping the last
8327 * block of the extent just the way it is.
8329 btrfs_drop_extent_map_range(inode,
8330 ALIGN(new_size, fs_info->sectorsize),
8333 ret = btrfs_truncate_inode_items(trans, root, &control);
8335 inode_sub_bytes(&inode->vfs_inode, control.sub_bytes);
8336 btrfs_inode_safe_disk_i_size_write(inode, control.last_size);
8338 unlock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8340 trans->block_rsv = &fs_info->trans_block_rsv;
8341 if (ret != -ENOSPC && ret != -EAGAIN)
8344 ret = btrfs_update_inode(trans, root, inode);
8348 btrfs_end_transaction(trans);
8349 btrfs_btree_balance_dirty(fs_info);
8351 trans = btrfs_start_transaction(root, 2);
8352 if (IS_ERR(trans)) {
8353 ret = PTR_ERR(trans);
8358 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8359 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8360 rsv, min_size, false);
8362 * We have reserved 2 metadata units when we started the
8363 * transaction and min_size matches 1 unit, so this should never
8364 * fail, but if it does, it's not critical we just fail truncation.
8369 trans->block_rsv = rsv;
8373 * We can't call btrfs_truncate_block inside a trans handle as we could
8374 * deadlock with freeze, if we got BTRFS_NEED_TRUNCATE_BLOCK then we
8375 * know we've truncated everything except the last little bit, and can
8376 * do btrfs_truncate_block and then update the disk_i_size.
8378 if (ret == BTRFS_NEED_TRUNCATE_BLOCK) {
8379 btrfs_end_transaction(trans);
8380 btrfs_btree_balance_dirty(fs_info);
8382 ret = btrfs_truncate_block(inode, inode->vfs_inode.i_size, 0, 0);
8385 trans = btrfs_start_transaction(root, 1);
8386 if (IS_ERR(trans)) {
8387 ret = PTR_ERR(trans);
8390 btrfs_inode_safe_disk_i_size_write(inode, 0);
8396 trans->block_rsv = &fs_info->trans_block_rsv;
8397 ret2 = btrfs_update_inode(trans, root, inode);
8401 ret2 = btrfs_end_transaction(trans);
8404 btrfs_btree_balance_dirty(fs_info);
8407 btrfs_free_block_rsv(fs_info, rsv);
8409 * So if we truncate and then write and fsync we normally would just
8410 * write the extents that changed, which is a problem if we need to
8411 * first truncate that entire inode. So set this flag so we write out
8412 * all of the extents in the inode to the sync log so we're completely
8415 * If no extents were dropped or trimmed we don't need to force the next
8416 * fsync to truncate all the inode's items from the log and re-log them
8417 * all. This means the truncate operation did not change the file size,
8418 * or changed it to a smaller size but there was only an implicit hole
8419 * between the old i_size and the new i_size, and there were no prealloc
8420 * extents beyond i_size to drop.
8422 if (control.extents_found > 0)
8423 btrfs_set_inode_full_sync(inode);
8428 struct inode *btrfs_new_subvol_inode(struct mnt_idmap *idmap,
8431 struct inode *inode;
8433 inode = new_inode(dir->i_sb);
8436 * Subvolumes don't inherit the sgid bit or the parent's gid if
8437 * the parent's sgid bit is set. This is probably a bug.
8439 inode_init_owner(idmap, inode, NULL,
8440 S_IFDIR | (~current_umask() & S_IRWXUGO));
8441 inode->i_op = &btrfs_dir_inode_operations;
8442 inode->i_fop = &btrfs_dir_file_operations;
8447 struct inode *btrfs_alloc_inode(struct super_block *sb)
8449 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8450 struct btrfs_inode *ei;
8451 struct inode *inode;
8453 ei = alloc_inode_sb(sb, btrfs_inode_cachep, GFP_KERNEL);
8460 ei->last_sub_trans = 0;
8461 ei->logged_trans = 0;
8462 ei->delalloc_bytes = 0;
8463 ei->new_delalloc_bytes = 0;
8464 ei->defrag_bytes = 0;
8465 ei->disk_i_size = 0;
8469 ei->index_cnt = (u64)-1;
8471 ei->last_unlink_trans = 0;
8472 ei->last_reflink_trans = 0;
8473 ei->last_log_commit = 0;
8475 spin_lock_init(&ei->lock);
8476 ei->outstanding_extents = 0;
8477 if (sb->s_magic != BTRFS_TEST_MAGIC)
8478 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8479 BTRFS_BLOCK_RSV_DELALLOC);
8480 ei->runtime_flags = 0;
8481 ei->prop_compress = BTRFS_COMPRESS_NONE;
8482 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8484 ei->delayed_node = NULL;
8486 ei->i_otime.tv_sec = 0;
8487 ei->i_otime.tv_nsec = 0;
8489 inode = &ei->vfs_inode;
8490 extent_map_tree_init(&ei->extent_tree);
8491 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO);
8492 ei->io_tree.inode = ei;
8493 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8494 IO_TREE_INODE_FILE_EXTENT);
8495 mutex_init(&ei->log_mutex);
8496 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8497 INIT_LIST_HEAD(&ei->delalloc_inodes);
8498 INIT_LIST_HEAD(&ei->delayed_iput);
8499 RB_CLEAR_NODE(&ei->rb_node);
8500 init_rwsem(&ei->i_mmap_lock);
8505 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8506 void btrfs_test_destroy_inode(struct inode *inode)
8508 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
8509 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8513 void btrfs_free_inode(struct inode *inode)
8515 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8518 void btrfs_destroy_inode(struct inode *vfs_inode)
8520 struct btrfs_ordered_extent *ordered;
8521 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8522 struct btrfs_root *root = inode->root;
8523 bool freespace_inode;
8525 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8526 WARN_ON(vfs_inode->i_data.nrpages);
8527 WARN_ON(inode->block_rsv.reserved);
8528 WARN_ON(inode->block_rsv.size);
8529 WARN_ON(inode->outstanding_extents);
8530 if (!S_ISDIR(vfs_inode->i_mode)) {
8531 WARN_ON(inode->delalloc_bytes);
8532 WARN_ON(inode->new_delalloc_bytes);
8534 WARN_ON(inode->csum_bytes);
8535 WARN_ON(inode->defrag_bytes);
8538 * This can happen where we create an inode, but somebody else also
8539 * created the same inode and we need to destroy the one we already
8546 * If this is a free space inode do not take the ordered extents lockdep
8549 freespace_inode = btrfs_is_free_space_inode(inode);
8552 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8556 btrfs_err(root->fs_info,
8557 "found ordered extent %llu %llu on inode cleanup",
8558 ordered->file_offset, ordered->num_bytes);
8560 if (!freespace_inode)
8561 btrfs_lockdep_acquire(root->fs_info, btrfs_ordered_extent);
8563 btrfs_remove_ordered_extent(inode, ordered);
8564 btrfs_put_ordered_extent(ordered);
8565 btrfs_put_ordered_extent(ordered);
8568 btrfs_qgroup_check_reserved_leak(inode);
8569 inode_tree_del(inode);
8570 btrfs_drop_extent_map_range(inode, 0, (u64)-1, false);
8571 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
8572 btrfs_put_root(inode->root);
8575 int btrfs_drop_inode(struct inode *inode)
8577 struct btrfs_root *root = BTRFS_I(inode)->root;
8582 /* the snap/subvol tree is on deleting */
8583 if (btrfs_root_refs(&root->root_item) == 0)
8586 return generic_drop_inode(inode);
8589 static void init_once(void *foo)
8591 struct btrfs_inode *ei = foo;
8593 inode_init_once(&ei->vfs_inode);
8596 void __cold btrfs_destroy_cachep(void)
8599 * Make sure all delayed rcu free inodes are flushed before we
8603 bioset_exit(&btrfs_dio_bioset);
8604 kmem_cache_destroy(btrfs_inode_cachep);
8607 int __init btrfs_init_cachep(void)
8609 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8610 sizeof(struct btrfs_inode), 0,
8611 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
8613 if (!btrfs_inode_cachep)
8616 if (bioset_init(&btrfs_dio_bioset, BIO_POOL_SIZE,
8617 offsetof(struct btrfs_dio_private, bbio.bio),
8623 btrfs_destroy_cachep();
8627 static int btrfs_getattr(struct mnt_idmap *idmap,
8628 const struct path *path, struct kstat *stat,
8629 u32 request_mask, unsigned int flags)
8633 struct inode *inode = d_inode(path->dentry);
8634 u32 blocksize = inode->i_sb->s_blocksize;
8635 u32 bi_flags = BTRFS_I(inode)->flags;
8636 u32 bi_ro_flags = BTRFS_I(inode)->ro_flags;
8638 stat->result_mask |= STATX_BTIME;
8639 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
8640 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
8641 if (bi_flags & BTRFS_INODE_APPEND)
8642 stat->attributes |= STATX_ATTR_APPEND;
8643 if (bi_flags & BTRFS_INODE_COMPRESS)
8644 stat->attributes |= STATX_ATTR_COMPRESSED;
8645 if (bi_flags & BTRFS_INODE_IMMUTABLE)
8646 stat->attributes |= STATX_ATTR_IMMUTABLE;
8647 if (bi_flags & BTRFS_INODE_NODUMP)
8648 stat->attributes |= STATX_ATTR_NODUMP;
8649 if (bi_ro_flags & BTRFS_INODE_RO_VERITY)
8650 stat->attributes |= STATX_ATTR_VERITY;
8652 stat->attributes_mask |= (STATX_ATTR_APPEND |
8653 STATX_ATTR_COMPRESSED |
8654 STATX_ATTR_IMMUTABLE |
8657 generic_fillattr(idmap, inode, stat);
8658 stat->dev = BTRFS_I(inode)->root->anon_dev;
8660 spin_lock(&BTRFS_I(inode)->lock);
8661 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
8662 inode_bytes = inode_get_bytes(inode);
8663 spin_unlock(&BTRFS_I(inode)->lock);
8664 stat->blocks = (ALIGN(inode_bytes, blocksize) +
8665 ALIGN(delalloc_bytes, blocksize)) >> SECTOR_SHIFT;
8669 static int btrfs_rename_exchange(struct inode *old_dir,
8670 struct dentry *old_dentry,
8671 struct inode *new_dir,
8672 struct dentry *new_dentry)
8674 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
8675 struct btrfs_trans_handle *trans;
8676 unsigned int trans_num_items;
8677 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8678 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8679 struct inode *new_inode = new_dentry->d_inode;
8680 struct inode *old_inode = old_dentry->d_inode;
8681 struct timespec64 ctime = current_time(old_inode);
8682 struct btrfs_rename_ctx old_rename_ctx;
8683 struct btrfs_rename_ctx new_rename_ctx;
8684 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
8685 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
8690 bool need_abort = false;
8691 struct fscrypt_name old_fname, new_fname;
8692 struct fscrypt_str *old_name, *new_name;
8695 * For non-subvolumes allow exchange only within one subvolume, in the
8696 * same inode namespace. Two subvolumes (represented as directory) can
8697 * be exchanged as they're a logical link and have a fixed inode number.
8700 (old_ino != BTRFS_FIRST_FREE_OBJECTID ||
8701 new_ino != BTRFS_FIRST_FREE_OBJECTID))
8704 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
8708 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
8710 fscrypt_free_filename(&old_fname);
8714 old_name = &old_fname.disk_name;
8715 new_name = &new_fname.disk_name;
8717 /* close the race window with snapshot create/destroy ioctl */
8718 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
8719 new_ino == BTRFS_FIRST_FREE_OBJECTID)
8720 down_read(&fs_info->subvol_sem);
8724 * 1 to remove old dir item
8725 * 1 to remove old dir index
8726 * 1 to add new dir item
8727 * 1 to add new dir index
8728 * 1 to update parent inode
8730 * If the parents are the same, we only need to account for one
8732 trans_num_items = (old_dir == new_dir ? 9 : 10);
8733 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8735 * 1 to remove old root ref
8736 * 1 to remove old root backref
8737 * 1 to add new root ref
8738 * 1 to add new root backref
8740 trans_num_items += 4;
8743 * 1 to update inode item
8744 * 1 to remove old inode ref
8745 * 1 to add new inode ref
8747 trans_num_items += 3;
8749 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
8750 trans_num_items += 4;
8752 trans_num_items += 3;
8753 trans = btrfs_start_transaction(root, trans_num_items);
8754 if (IS_ERR(trans)) {
8755 ret = PTR_ERR(trans);
8760 ret = btrfs_record_root_in_trans(trans, dest);
8766 * We need to find a free sequence number both in the source and
8767 * in the destination directory for the exchange.
8769 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
8772 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
8776 BTRFS_I(old_inode)->dir_index = 0ULL;
8777 BTRFS_I(new_inode)->dir_index = 0ULL;
8779 /* Reference for the source. */
8780 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8781 /* force full log commit if subvolume involved. */
8782 btrfs_set_log_full_commit(trans);
8784 ret = btrfs_insert_inode_ref(trans, dest, new_name, old_ino,
8785 btrfs_ino(BTRFS_I(new_dir)),
8792 /* And now for the dest. */
8793 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8794 /* force full log commit if subvolume involved. */
8795 btrfs_set_log_full_commit(trans);
8797 ret = btrfs_insert_inode_ref(trans, root, old_name, new_ino,
8798 btrfs_ino(BTRFS_I(old_dir)),
8802 btrfs_abort_transaction(trans, ret);
8807 /* Update inode version and ctime/mtime. */
8808 inode_inc_iversion(old_dir);
8809 inode_inc_iversion(new_dir);
8810 inode_inc_iversion(old_inode);
8811 inode_inc_iversion(new_inode);
8812 old_dir->i_mtime = ctime;
8813 old_dir->i_ctime = ctime;
8814 new_dir->i_mtime = ctime;
8815 new_dir->i_ctime = ctime;
8816 old_inode->i_ctime = ctime;
8817 new_inode->i_ctime = ctime;
8819 if (old_dentry->d_parent != new_dentry->d_parent) {
8820 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
8821 BTRFS_I(old_inode), true);
8822 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
8823 BTRFS_I(new_inode), true);
8826 /* src is a subvolume */
8827 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8828 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
8829 } else { /* src is an inode */
8830 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
8831 BTRFS_I(old_dentry->d_inode),
8832 old_name, &old_rename_ctx);
8834 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
8837 btrfs_abort_transaction(trans, ret);
8841 /* dest is a subvolume */
8842 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8843 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
8844 } else { /* dest is an inode */
8845 ret = __btrfs_unlink_inode(trans, BTRFS_I(new_dir),
8846 BTRFS_I(new_dentry->d_inode),
8847 new_name, &new_rename_ctx);
8849 ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode));
8852 btrfs_abort_transaction(trans, ret);
8856 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
8857 new_name, 0, old_idx);
8859 btrfs_abort_transaction(trans, ret);
8863 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
8864 old_name, 0, new_idx);
8866 btrfs_abort_transaction(trans, ret);
8870 if (old_inode->i_nlink == 1)
8871 BTRFS_I(old_inode)->dir_index = old_idx;
8872 if (new_inode->i_nlink == 1)
8873 BTRFS_I(new_inode)->dir_index = new_idx;
8876 * Now pin the logs of the roots. We do it to ensure that no other task
8877 * can sync the logs while we are in progress with the rename, because
8878 * that could result in an inconsistency in case any of the inodes that
8879 * are part of this rename operation were logged before.
8881 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8882 btrfs_pin_log_trans(root);
8883 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8884 btrfs_pin_log_trans(dest);
8886 /* Do the log updates for all inodes. */
8887 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8888 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
8889 old_rename_ctx.index, new_dentry->d_parent);
8890 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8891 btrfs_log_new_name(trans, new_dentry, BTRFS_I(new_dir),
8892 new_rename_ctx.index, old_dentry->d_parent);
8894 /* Now unpin the logs. */
8895 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8896 btrfs_end_log_trans(root);
8897 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8898 btrfs_end_log_trans(dest);
8900 ret2 = btrfs_end_transaction(trans);
8901 ret = ret ? ret : ret2;
8903 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
8904 old_ino == BTRFS_FIRST_FREE_OBJECTID)
8905 up_read(&fs_info->subvol_sem);
8907 fscrypt_free_filename(&new_fname);
8908 fscrypt_free_filename(&old_fname);
8912 static struct inode *new_whiteout_inode(struct mnt_idmap *idmap,
8915 struct inode *inode;
8917 inode = new_inode(dir->i_sb);
8919 inode_init_owner(idmap, inode, dir,
8920 S_IFCHR | WHITEOUT_MODE);
8921 inode->i_op = &btrfs_special_inode_operations;
8922 init_special_inode(inode, inode->i_mode, WHITEOUT_DEV);
8927 static int btrfs_rename(struct mnt_idmap *idmap,
8928 struct inode *old_dir, struct dentry *old_dentry,
8929 struct inode *new_dir, struct dentry *new_dentry,
8932 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
8933 struct btrfs_new_inode_args whiteout_args = {
8935 .dentry = old_dentry,
8937 struct btrfs_trans_handle *trans;
8938 unsigned int trans_num_items;
8939 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8940 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8941 struct inode *new_inode = d_inode(new_dentry);
8942 struct inode *old_inode = d_inode(old_dentry);
8943 struct btrfs_rename_ctx rename_ctx;
8947 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
8948 struct fscrypt_name old_fname, new_fname;
8950 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8953 /* we only allow rename subvolume link between subvolumes */
8954 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8957 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8958 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
8961 if (S_ISDIR(old_inode->i_mode) && new_inode &&
8962 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8965 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
8969 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
8971 fscrypt_free_filename(&old_fname);
8975 /* check for collisions, even if the name isn't there */
8976 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, &new_fname.disk_name);
8978 if (ret == -EEXIST) {
8980 * eexist without a new_inode */
8981 if (WARN_ON(!new_inode)) {
8982 goto out_fscrypt_names;
8985 /* maybe -EOVERFLOW */
8986 goto out_fscrypt_names;
8992 * we're using rename to replace one file with another. Start IO on it
8993 * now so we don't add too much work to the end of the transaction
8995 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
8996 filemap_flush(old_inode->i_mapping);
8998 if (flags & RENAME_WHITEOUT) {
8999 whiteout_args.inode = new_whiteout_inode(idmap, old_dir);
9000 if (!whiteout_args.inode) {
9002 goto out_fscrypt_names;
9004 ret = btrfs_new_inode_prepare(&whiteout_args, &trans_num_items);
9006 goto out_whiteout_inode;
9008 /* 1 to update the old parent inode. */
9009 trans_num_items = 1;
9012 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9013 /* Close the race window with snapshot create/destroy ioctl */
9014 down_read(&fs_info->subvol_sem);
9016 * 1 to remove old root ref
9017 * 1 to remove old root backref
9018 * 1 to add new root ref
9019 * 1 to add new root backref
9021 trans_num_items += 4;
9025 * 1 to remove old inode ref
9026 * 1 to add new inode ref
9028 trans_num_items += 3;
9031 * 1 to remove old dir item
9032 * 1 to remove old dir index
9033 * 1 to add new dir item
9034 * 1 to add new dir index
9036 trans_num_items += 4;
9037 /* 1 to update new parent inode if it's not the same as the old parent */
9038 if (new_dir != old_dir)
9043 * 1 to remove inode ref
9044 * 1 to remove dir item
9045 * 1 to remove dir index
9046 * 1 to possibly add orphan item
9048 trans_num_items += 5;
9050 trans = btrfs_start_transaction(root, trans_num_items);
9051 if (IS_ERR(trans)) {
9052 ret = PTR_ERR(trans);
9057 ret = btrfs_record_root_in_trans(trans, dest);
9062 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9066 BTRFS_I(old_inode)->dir_index = 0ULL;
9067 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9068 /* force full log commit if subvolume involved. */
9069 btrfs_set_log_full_commit(trans);
9071 ret = btrfs_insert_inode_ref(trans, dest, &new_fname.disk_name,
9072 old_ino, btrfs_ino(BTRFS_I(new_dir)),
9078 inode_inc_iversion(old_dir);
9079 inode_inc_iversion(new_dir);
9080 inode_inc_iversion(old_inode);
9081 old_dir->i_mtime = current_time(old_dir);
9082 old_dir->i_ctime = old_dir->i_mtime;
9083 new_dir->i_mtime = old_dir->i_mtime;
9084 new_dir->i_ctime = old_dir->i_mtime;
9085 old_inode->i_ctime = old_dir->i_mtime;
9087 if (old_dentry->d_parent != new_dentry->d_parent)
9088 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9089 BTRFS_I(old_inode), true);
9091 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9092 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
9094 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9095 BTRFS_I(d_inode(old_dentry)),
9096 &old_fname.disk_name, &rename_ctx);
9098 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9101 btrfs_abort_transaction(trans, ret);
9106 inode_inc_iversion(new_inode);
9107 new_inode->i_ctime = current_time(new_inode);
9108 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9109 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9110 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
9111 BUG_ON(new_inode->i_nlink == 0);
9113 ret = btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9114 BTRFS_I(d_inode(new_dentry)),
9115 &new_fname.disk_name);
9117 if (!ret && new_inode->i_nlink == 0)
9118 ret = btrfs_orphan_add(trans,
9119 BTRFS_I(d_inode(new_dentry)));
9121 btrfs_abort_transaction(trans, ret);
9126 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9127 &new_fname.disk_name, 0, index);
9129 btrfs_abort_transaction(trans, ret);
9133 if (old_inode->i_nlink == 1)
9134 BTRFS_I(old_inode)->dir_index = index;
9136 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9137 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9138 rename_ctx.index, new_dentry->d_parent);
9140 if (flags & RENAME_WHITEOUT) {
9141 ret = btrfs_create_new_inode(trans, &whiteout_args);
9143 btrfs_abort_transaction(trans, ret);
9146 unlock_new_inode(whiteout_args.inode);
9147 iput(whiteout_args.inode);
9148 whiteout_args.inode = NULL;
9152 ret2 = btrfs_end_transaction(trans);
9153 ret = ret ? ret : ret2;
9155 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9156 up_read(&fs_info->subvol_sem);
9157 if (flags & RENAME_WHITEOUT)
9158 btrfs_new_inode_args_destroy(&whiteout_args);
9160 if (flags & RENAME_WHITEOUT)
9161 iput(whiteout_args.inode);
9163 fscrypt_free_filename(&old_fname);
9164 fscrypt_free_filename(&new_fname);
9168 static int btrfs_rename2(struct mnt_idmap *idmap, struct inode *old_dir,
9169 struct dentry *old_dentry, struct inode *new_dir,
9170 struct dentry *new_dentry, unsigned int flags)
9174 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9177 if (flags & RENAME_EXCHANGE)
9178 ret = btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9181 ret = btrfs_rename(idmap, old_dir, old_dentry, new_dir,
9184 btrfs_btree_balance_dirty(BTRFS_I(new_dir)->root->fs_info);
9189 struct btrfs_delalloc_work {
9190 struct inode *inode;
9191 struct completion completion;
9192 struct list_head list;
9193 struct btrfs_work work;
9196 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9198 struct btrfs_delalloc_work *delalloc_work;
9199 struct inode *inode;
9201 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9203 inode = delalloc_work->inode;
9204 filemap_flush(inode->i_mapping);
9205 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9206 &BTRFS_I(inode)->runtime_flags))
9207 filemap_flush(inode->i_mapping);
9210 complete(&delalloc_work->completion);
9213 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9215 struct btrfs_delalloc_work *work;
9217 work = kmalloc(sizeof(*work), GFP_NOFS);
9221 init_completion(&work->completion);
9222 INIT_LIST_HEAD(&work->list);
9223 work->inode = inode;
9224 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9230 * some fairly slow code that needs optimization. This walks the list
9231 * of all the inodes with pending delalloc and forces them to disk.
9233 static int start_delalloc_inodes(struct btrfs_root *root,
9234 struct writeback_control *wbc, bool snapshot,
9235 bool in_reclaim_context)
9237 struct btrfs_inode *binode;
9238 struct inode *inode;
9239 struct btrfs_delalloc_work *work, *next;
9240 struct list_head works;
9241 struct list_head splice;
9243 bool full_flush = wbc->nr_to_write == LONG_MAX;
9245 INIT_LIST_HEAD(&works);
9246 INIT_LIST_HEAD(&splice);
9248 mutex_lock(&root->delalloc_mutex);
9249 spin_lock(&root->delalloc_lock);
9250 list_splice_init(&root->delalloc_inodes, &splice);
9251 while (!list_empty(&splice)) {
9252 binode = list_entry(splice.next, struct btrfs_inode,
9255 list_move_tail(&binode->delalloc_inodes,
9256 &root->delalloc_inodes);
9258 if (in_reclaim_context &&
9259 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9262 inode = igrab(&binode->vfs_inode);
9264 cond_resched_lock(&root->delalloc_lock);
9267 spin_unlock(&root->delalloc_lock);
9270 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9271 &binode->runtime_flags);
9273 work = btrfs_alloc_delalloc_work(inode);
9279 list_add_tail(&work->list, &works);
9280 btrfs_queue_work(root->fs_info->flush_workers,
9283 ret = filemap_fdatawrite_wbc(inode->i_mapping, wbc);
9284 btrfs_add_delayed_iput(BTRFS_I(inode));
9285 if (ret || wbc->nr_to_write <= 0)
9289 spin_lock(&root->delalloc_lock);
9291 spin_unlock(&root->delalloc_lock);
9294 list_for_each_entry_safe(work, next, &works, list) {
9295 list_del_init(&work->list);
9296 wait_for_completion(&work->completion);
9300 if (!list_empty(&splice)) {
9301 spin_lock(&root->delalloc_lock);
9302 list_splice_tail(&splice, &root->delalloc_inodes);
9303 spin_unlock(&root->delalloc_lock);
9305 mutex_unlock(&root->delalloc_mutex);
9309 int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9311 struct writeback_control wbc = {
9312 .nr_to_write = LONG_MAX,
9313 .sync_mode = WB_SYNC_NONE,
9315 .range_end = LLONG_MAX,
9317 struct btrfs_fs_info *fs_info = root->fs_info;
9319 if (BTRFS_FS_ERROR(fs_info))
9322 return start_delalloc_inodes(root, &wbc, true, in_reclaim_context);
9325 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9326 bool in_reclaim_context)
9328 struct writeback_control wbc = {
9330 .sync_mode = WB_SYNC_NONE,
9332 .range_end = LLONG_MAX,
9334 struct btrfs_root *root;
9335 struct list_head splice;
9338 if (BTRFS_FS_ERROR(fs_info))
9341 INIT_LIST_HEAD(&splice);
9343 mutex_lock(&fs_info->delalloc_root_mutex);
9344 spin_lock(&fs_info->delalloc_root_lock);
9345 list_splice_init(&fs_info->delalloc_roots, &splice);
9346 while (!list_empty(&splice)) {
9348 * Reset nr_to_write here so we know that we're doing a full
9352 wbc.nr_to_write = LONG_MAX;
9354 root = list_first_entry(&splice, struct btrfs_root,
9356 root = btrfs_grab_root(root);
9358 list_move_tail(&root->delalloc_root,
9359 &fs_info->delalloc_roots);
9360 spin_unlock(&fs_info->delalloc_root_lock);
9362 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9363 btrfs_put_root(root);
9364 if (ret < 0 || wbc.nr_to_write <= 0)
9366 spin_lock(&fs_info->delalloc_root_lock);
9368 spin_unlock(&fs_info->delalloc_root_lock);
9372 if (!list_empty(&splice)) {
9373 spin_lock(&fs_info->delalloc_root_lock);
9374 list_splice_tail(&splice, &fs_info->delalloc_roots);
9375 spin_unlock(&fs_info->delalloc_root_lock);
9377 mutex_unlock(&fs_info->delalloc_root_mutex);
9381 static int btrfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
9382 struct dentry *dentry, const char *symname)
9384 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9385 struct btrfs_trans_handle *trans;
9386 struct btrfs_root *root = BTRFS_I(dir)->root;
9387 struct btrfs_path *path;
9388 struct btrfs_key key;
9389 struct inode *inode;
9390 struct btrfs_new_inode_args new_inode_args = {
9394 unsigned int trans_num_items;
9399 struct btrfs_file_extent_item *ei;
9400 struct extent_buffer *leaf;
9402 name_len = strlen(symname);
9403 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9404 return -ENAMETOOLONG;
9406 inode = new_inode(dir->i_sb);
9409 inode_init_owner(idmap, inode, dir, S_IFLNK | S_IRWXUGO);
9410 inode->i_op = &btrfs_symlink_inode_operations;
9411 inode_nohighmem(inode);
9412 inode->i_mapping->a_ops = &btrfs_aops;
9413 btrfs_i_size_write(BTRFS_I(inode), name_len);
9414 inode_set_bytes(inode, name_len);
9416 new_inode_args.inode = inode;
9417 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9420 /* 1 additional item for the inline extent */
9423 trans = btrfs_start_transaction(root, trans_num_items);
9424 if (IS_ERR(trans)) {
9425 err = PTR_ERR(trans);
9426 goto out_new_inode_args;
9429 err = btrfs_create_new_inode(trans, &new_inode_args);
9433 path = btrfs_alloc_path();
9436 btrfs_abort_transaction(trans, err);
9437 discard_new_inode(inode);
9441 key.objectid = btrfs_ino(BTRFS_I(inode));
9443 key.type = BTRFS_EXTENT_DATA_KEY;
9444 datasize = btrfs_file_extent_calc_inline_size(name_len);
9445 err = btrfs_insert_empty_item(trans, root, path, &key,
9448 btrfs_abort_transaction(trans, err);
9449 btrfs_free_path(path);
9450 discard_new_inode(inode);
9454 leaf = path->nodes[0];
9455 ei = btrfs_item_ptr(leaf, path->slots[0],
9456 struct btrfs_file_extent_item);
9457 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9458 btrfs_set_file_extent_type(leaf, ei,
9459 BTRFS_FILE_EXTENT_INLINE);
9460 btrfs_set_file_extent_encryption(leaf, ei, 0);
9461 btrfs_set_file_extent_compression(leaf, ei, 0);
9462 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9463 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9465 ptr = btrfs_file_extent_inline_start(ei);
9466 write_extent_buffer(leaf, symname, ptr, name_len);
9467 btrfs_mark_buffer_dirty(leaf);
9468 btrfs_free_path(path);
9470 d_instantiate_new(dentry, inode);
9473 btrfs_end_transaction(trans);
9474 btrfs_btree_balance_dirty(fs_info);
9476 btrfs_new_inode_args_destroy(&new_inode_args);
9483 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9484 struct btrfs_trans_handle *trans_in,
9485 struct btrfs_inode *inode,
9486 struct btrfs_key *ins,
9489 struct btrfs_file_extent_item stack_fi;
9490 struct btrfs_replace_extent_info extent_info;
9491 struct btrfs_trans_handle *trans = trans_in;
9492 struct btrfs_path *path;
9493 u64 start = ins->objectid;
9494 u64 len = ins->offset;
9495 int qgroup_released;
9498 memset(&stack_fi, 0, sizeof(stack_fi));
9500 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9501 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9502 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9503 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9504 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9505 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9506 /* Encryption and other encoding is reserved and all 0 */
9508 qgroup_released = btrfs_qgroup_release_data(inode, file_offset, len);
9509 if (qgroup_released < 0)
9510 return ERR_PTR(qgroup_released);
9513 ret = insert_reserved_file_extent(trans, inode,
9514 file_offset, &stack_fi,
9515 true, qgroup_released);
9521 extent_info.disk_offset = start;
9522 extent_info.disk_len = len;
9523 extent_info.data_offset = 0;
9524 extent_info.data_len = len;
9525 extent_info.file_offset = file_offset;
9526 extent_info.extent_buf = (char *)&stack_fi;
9527 extent_info.is_new_extent = true;
9528 extent_info.update_times = true;
9529 extent_info.qgroup_reserved = qgroup_released;
9530 extent_info.insertions = 0;
9532 path = btrfs_alloc_path();
9538 ret = btrfs_replace_file_extents(inode, path, file_offset,
9539 file_offset + len - 1, &extent_info,
9541 btrfs_free_path(path);
9548 * We have released qgroup data range at the beginning of the function,
9549 * and normally qgroup_released bytes will be freed when committing
9551 * But if we error out early, we have to free what we have released
9552 * or we leak qgroup data reservation.
9554 btrfs_qgroup_free_refroot(inode->root->fs_info,
9555 inode->root->root_key.objectid, qgroup_released,
9556 BTRFS_QGROUP_RSV_DATA);
9557 return ERR_PTR(ret);
9560 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9561 u64 start, u64 num_bytes, u64 min_size,
9562 loff_t actual_len, u64 *alloc_hint,
9563 struct btrfs_trans_handle *trans)
9565 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9566 struct extent_map *em;
9567 struct btrfs_root *root = BTRFS_I(inode)->root;
9568 struct btrfs_key ins;
9569 u64 cur_offset = start;
9570 u64 clear_offset = start;
9573 u64 last_alloc = (u64)-1;
9575 bool own_trans = true;
9576 u64 end = start + num_bytes - 1;
9580 while (num_bytes > 0) {
9581 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9582 cur_bytes = max(cur_bytes, min_size);
9584 * If we are severely fragmented we could end up with really
9585 * small allocations, so if the allocator is returning small
9586 * chunks lets make its job easier by only searching for those
9589 cur_bytes = min(cur_bytes, last_alloc);
9590 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9591 min_size, 0, *alloc_hint, &ins, 1, 0);
9596 * We've reserved this space, and thus converted it from
9597 * ->bytes_may_use to ->bytes_reserved. Any error that happens
9598 * from here on out we will only need to clear our reservation
9599 * for the remaining unreserved area, so advance our
9600 * clear_offset by our extent size.
9602 clear_offset += ins.offset;
9604 last_alloc = ins.offset;
9605 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
9608 * Now that we inserted the prealloc extent we can finally
9609 * decrement the number of reservations in the block group.
9610 * If we did it before, we could race with relocation and have
9611 * relocation miss the reserved extent, making it fail later.
9613 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
9614 if (IS_ERR(trans)) {
9615 ret = PTR_ERR(trans);
9616 btrfs_free_reserved_extent(fs_info, ins.objectid,
9621 em = alloc_extent_map();
9623 btrfs_drop_extent_map_range(BTRFS_I(inode), cur_offset,
9624 cur_offset + ins.offset - 1, false);
9625 btrfs_set_inode_full_sync(BTRFS_I(inode));
9629 em->start = cur_offset;
9630 em->orig_start = cur_offset;
9631 em->len = ins.offset;
9632 em->block_start = ins.objectid;
9633 em->block_len = ins.offset;
9634 em->orig_block_len = ins.offset;
9635 em->ram_bytes = ins.offset;
9636 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9637 em->generation = trans->transid;
9639 ret = btrfs_replace_extent_map_range(BTRFS_I(inode), em, true);
9640 free_extent_map(em);
9642 num_bytes -= ins.offset;
9643 cur_offset += ins.offset;
9644 *alloc_hint = ins.objectid + ins.offset;
9646 inode_inc_iversion(inode);
9647 inode->i_ctime = current_time(inode);
9648 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9649 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9650 (actual_len > inode->i_size) &&
9651 (cur_offset > inode->i_size)) {
9652 if (cur_offset > actual_len)
9653 i_size = actual_len;
9655 i_size = cur_offset;
9656 i_size_write(inode, i_size);
9657 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
9660 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
9663 btrfs_abort_transaction(trans, ret);
9665 btrfs_end_transaction(trans);
9670 btrfs_end_transaction(trans);
9674 if (clear_offset < end)
9675 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
9676 end - clear_offset + 1);
9680 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9681 u64 start, u64 num_bytes, u64 min_size,
9682 loff_t actual_len, u64 *alloc_hint)
9684 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9685 min_size, actual_len, alloc_hint,
9689 int btrfs_prealloc_file_range_trans(struct inode *inode,
9690 struct btrfs_trans_handle *trans, int mode,
9691 u64 start, u64 num_bytes, u64 min_size,
9692 loff_t actual_len, u64 *alloc_hint)
9694 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9695 min_size, actual_len, alloc_hint, trans);
9698 static int btrfs_permission(struct mnt_idmap *idmap,
9699 struct inode *inode, int mask)
9701 struct btrfs_root *root = BTRFS_I(inode)->root;
9702 umode_t mode = inode->i_mode;
9704 if (mask & MAY_WRITE &&
9705 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9706 if (btrfs_root_readonly(root))
9708 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9711 return generic_permission(idmap, inode, mask);
9714 static int btrfs_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
9715 struct file *file, umode_t mode)
9717 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9718 struct btrfs_trans_handle *trans;
9719 struct btrfs_root *root = BTRFS_I(dir)->root;
9720 struct inode *inode;
9721 struct btrfs_new_inode_args new_inode_args = {
9723 .dentry = file->f_path.dentry,
9726 unsigned int trans_num_items;
9729 inode = new_inode(dir->i_sb);
9732 inode_init_owner(idmap, inode, dir, mode);
9733 inode->i_fop = &btrfs_file_operations;
9734 inode->i_op = &btrfs_file_inode_operations;
9735 inode->i_mapping->a_ops = &btrfs_aops;
9737 new_inode_args.inode = inode;
9738 ret = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9742 trans = btrfs_start_transaction(root, trans_num_items);
9743 if (IS_ERR(trans)) {
9744 ret = PTR_ERR(trans);
9745 goto out_new_inode_args;
9748 ret = btrfs_create_new_inode(trans, &new_inode_args);
9751 * We set number of links to 0 in btrfs_create_new_inode(), and here we
9752 * set it to 1 because d_tmpfile() will issue a warning if the count is
9755 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9757 set_nlink(inode, 1);
9760 d_tmpfile(file, inode);
9761 unlock_new_inode(inode);
9762 mark_inode_dirty(inode);
9765 btrfs_end_transaction(trans);
9766 btrfs_btree_balance_dirty(fs_info);
9768 btrfs_new_inode_args_destroy(&new_inode_args);
9772 return finish_open_simple(file, ret);
9775 void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end)
9777 struct btrfs_fs_info *fs_info = inode->root->fs_info;
9778 unsigned long index = start >> PAGE_SHIFT;
9779 unsigned long end_index = end >> PAGE_SHIFT;
9783 ASSERT(end + 1 - start <= U32_MAX);
9784 len = end + 1 - start;
9785 while (index <= end_index) {
9786 page = find_get_page(inode->vfs_inode.i_mapping, index);
9787 ASSERT(page); /* Pages should be in the extent_io_tree */
9789 btrfs_page_set_writeback(fs_info, page, start, len);
9795 int btrfs_encoded_io_compression_from_extent(struct btrfs_fs_info *fs_info,
9798 switch (compress_type) {
9799 case BTRFS_COMPRESS_NONE:
9800 return BTRFS_ENCODED_IO_COMPRESSION_NONE;
9801 case BTRFS_COMPRESS_ZLIB:
9802 return BTRFS_ENCODED_IO_COMPRESSION_ZLIB;
9803 case BTRFS_COMPRESS_LZO:
9805 * The LZO format depends on the sector size. 64K is the maximum
9806 * sector size that we support.
9808 if (fs_info->sectorsize < SZ_4K || fs_info->sectorsize > SZ_64K)
9810 return BTRFS_ENCODED_IO_COMPRESSION_LZO_4K +
9811 (fs_info->sectorsize_bits - 12);
9812 case BTRFS_COMPRESS_ZSTD:
9813 return BTRFS_ENCODED_IO_COMPRESSION_ZSTD;
9819 static ssize_t btrfs_encoded_read_inline(
9821 struct iov_iter *iter, u64 start,
9823 struct extent_state **cached_state,
9824 u64 extent_start, size_t count,
9825 struct btrfs_ioctl_encoded_io_args *encoded,
9828 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
9829 struct btrfs_root *root = inode->root;
9830 struct btrfs_fs_info *fs_info = root->fs_info;
9831 struct extent_io_tree *io_tree = &inode->io_tree;
9832 struct btrfs_path *path;
9833 struct extent_buffer *leaf;
9834 struct btrfs_file_extent_item *item;
9840 path = btrfs_alloc_path();
9845 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
9849 /* The extent item disappeared? */
9854 leaf = path->nodes[0];
9855 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
9857 ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
9858 ptr = btrfs_file_extent_inline_start(item);
9860 encoded->len = min_t(u64, extent_start + ram_bytes,
9861 inode->vfs_inode.i_size) - iocb->ki_pos;
9862 ret = btrfs_encoded_io_compression_from_extent(fs_info,
9863 btrfs_file_extent_compression(leaf, item));
9866 encoded->compression = ret;
9867 if (encoded->compression) {
9870 inline_size = btrfs_file_extent_inline_item_len(leaf,
9872 if (inline_size > count) {
9876 count = inline_size;
9877 encoded->unencoded_len = ram_bytes;
9878 encoded->unencoded_offset = iocb->ki_pos - extent_start;
9880 count = min_t(u64, count, encoded->len);
9881 encoded->len = count;
9882 encoded->unencoded_len = count;
9883 ptr += iocb->ki_pos - extent_start;
9886 tmp = kmalloc(count, GFP_NOFS);
9891 read_extent_buffer(leaf, tmp, ptr, count);
9892 btrfs_release_path(path);
9893 unlock_extent(io_tree, start, lockend, cached_state);
9894 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
9897 ret = copy_to_iter(tmp, count, iter);
9902 btrfs_free_path(path);
9906 struct btrfs_encoded_read_private {
9907 wait_queue_head_t wait;
9909 blk_status_t status;
9912 static void btrfs_encoded_read_endio(struct btrfs_bio *bbio)
9914 struct btrfs_encoded_read_private *priv = bbio->private;
9916 if (bbio->bio.bi_status) {
9918 * The memory barrier implied by the atomic_dec_return() here
9919 * pairs with the memory barrier implied by the
9920 * atomic_dec_return() or io_wait_event() in
9921 * btrfs_encoded_read_regular_fill_pages() to ensure that this
9922 * write is observed before the load of status in
9923 * btrfs_encoded_read_regular_fill_pages().
9925 WRITE_ONCE(priv->status, bbio->bio.bi_status);
9927 if (!atomic_dec_return(&priv->pending))
9928 wake_up(&priv->wait);
9929 bio_put(&bbio->bio);
9932 int btrfs_encoded_read_regular_fill_pages(struct btrfs_inode *inode,
9933 u64 file_offset, u64 disk_bytenr,
9934 u64 disk_io_size, struct page **pages)
9936 struct btrfs_fs_info *fs_info = inode->root->fs_info;
9937 struct btrfs_encoded_read_private priv = {
9938 .pending = ATOMIC_INIT(1),
9940 unsigned long i = 0;
9941 struct btrfs_bio *bbio;
9943 init_waitqueue_head(&priv.wait);
9945 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
9946 btrfs_encoded_read_endio, &priv);
9947 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
9948 bbio->inode = inode;
9951 size_t bytes = min_t(u64, disk_io_size, PAGE_SIZE);
9953 if (bio_add_page(&bbio->bio, pages[i], bytes, 0) < bytes) {
9954 atomic_inc(&priv.pending);
9955 btrfs_submit_bio(bbio, 0);
9957 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
9958 btrfs_encoded_read_endio, &priv);
9959 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
9960 bbio->inode = inode;
9965 disk_bytenr += bytes;
9966 disk_io_size -= bytes;
9967 } while (disk_io_size);
9969 atomic_inc(&priv.pending);
9970 btrfs_submit_bio(bbio, 0);
9972 if (atomic_dec_return(&priv.pending))
9973 io_wait_event(priv.wait, !atomic_read(&priv.pending));
9974 /* See btrfs_encoded_read_endio() for ordering. */
9975 return blk_status_to_errno(READ_ONCE(priv.status));
9978 static ssize_t btrfs_encoded_read_regular(struct kiocb *iocb,
9979 struct iov_iter *iter,
9980 u64 start, u64 lockend,
9981 struct extent_state **cached_state,
9982 u64 disk_bytenr, u64 disk_io_size,
9983 size_t count, bool compressed,
9986 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
9987 struct extent_io_tree *io_tree = &inode->io_tree;
9988 struct page **pages;
9989 unsigned long nr_pages, i;
9994 nr_pages = DIV_ROUND_UP(disk_io_size, PAGE_SIZE);
9995 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
9998 ret = btrfs_alloc_page_array(nr_pages, pages);
10004 ret = btrfs_encoded_read_regular_fill_pages(inode, start, disk_bytenr,
10005 disk_io_size, pages);
10009 unlock_extent(io_tree, start, lockend, cached_state);
10010 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10017 i = (iocb->ki_pos - start) >> PAGE_SHIFT;
10018 page_offset = (iocb->ki_pos - start) & (PAGE_SIZE - 1);
10021 while (cur < count) {
10022 size_t bytes = min_t(size_t, count - cur,
10023 PAGE_SIZE - page_offset);
10025 if (copy_page_to_iter(pages[i], page_offset, bytes,
10036 for (i = 0; i < nr_pages; i++) {
10038 __free_page(pages[i]);
10044 ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter,
10045 struct btrfs_ioctl_encoded_io_args *encoded)
10047 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10048 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10049 struct extent_io_tree *io_tree = &inode->io_tree;
10051 size_t count = iov_iter_count(iter);
10052 u64 start, lockend, disk_bytenr, disk_io_size;
10053 struct extent_state *cached_state = NULL;
10054 struct extent_map *em;
10055 bool unlocked = false;
10057 file_accessed(iocb->ki_filp);
10059 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
10061 if (iocb->ki_pos >= inode->vfs_inode.i_size) {
10062 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10065 start = ALIGN_DOWN(iocb->ki_pos, fs_info->sectorsize);
10067 * We don't know how long the extent containing iocb->ki_pos is, but if
10068 * it's compressed we know that it won't be longer than this.
10070 lockend = start + BTRFS_MAX_UNCOMPRESSED - 1;
10073 struct btrfs_ordered_extent *ordered;
10075 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start,
10076 lockend - start + 1);
10078 goto out_unlock_inode;
10079 lock_extent(io_tree, start, lockend, &cached_state);
10080 ordered = btrfs_lookup_ordered_range(inode, start,
10081 lockend - start + 1);
10084 btrfs_put_ordered_extent(ordered);
10085 unlock_extent(io_tree, start, lockend, &cached_state);
10089 em = btrfs_get_extent(inode, NULL, 0, start, lockend - start + 1);
10092 goto out_unlock_extent;
10095 if (em->block_start == EXTENT_MAP_INLINE) {
10096 u64 extent_start = em->start;
10099 * For inline extents we get everything we need out of the
10102 free_extent_map(em);
10104 ret = btrfs_encoded_read_inline(iocb, iter, start, lockend,
10105 &cached_state, extent_start,
10106 count, encoded, &unlocked);
10111 * We only want to return up to EOF even if the extent extends beyond
10114 encoded->len = min_t(u64, extent_map_end(em),
10115 inode->vfs_inode.i_size) - iocb->ki_pos;
10116 if (em->block_start == EXTENT_MAP_HOLE ||
10117 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
10118 disk_bytenr = EXTENT_MAP_HOLE;
10119 count = min_t(u64, count, encoded->len);
10120 encoded->len = count;
10121 encoded->unencoded_len = count;
10122 } else if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10123 disk_bytenr = em->block_start;
10125 * Bail if the buffer isn't large enough to return the whole
10126 * compressed extent.
10128 if (em->block_len > count) {
10132 disk_io_size = em->block_len;
10133 count = em->block_len;
10134 encoded->unencoded_len = em->ram_bytes;
10135 encoded->unencoded_offset = iocb->ki_pos - em->orig_start;
10136 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10137 em->compress_type);
10140 encoded->compression = ret;
10142 disk_bytenr = em->block_start + (start - em->start);
10143 if (encoded->len > count)
10144 encoded->len = count;
10146 * Don't read beyond what we locked. This also limits the page
10147 * allocations that we'll do.
10149 disk_io_size = min(lockend + 1, iocb->ki_pos + encoded->len) - start;
10150 count = start + disk_io_size - iocb->ki_pos;
10151 encoded->len = count;
10152 encoded->unencoded_len = count;
10153 disk_io_size = ALIGN(disk_io_size, fs_info->sectorsize);
10155 free_extent_map(em);
10158 if (disk_bytenr == EXTENT_MAP_HOLE) {
10159 unlock_extent(io_tree, start, lockend, &cached_state);
10160 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10162 ret = iov_iter_zero(count, iter);
10166 ret = btrfs_encoded_read_regular(iocb, iter, start, lockend,
10167 &cached_state, disk_bytenr,
10168 disk_io_size, count,
10169 encoded->compression,
10175 iocb->ki_pos += encoded->len;
10177 free_extent_map(em);
10180 unlock_extent(io_tree, start, lockend, &cached_state);
10183 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10187 ssize_t btrfs_do_encoded_write(struct kiocb *iocb, struct iov_iter *from,
10188 const struct btrfs_ioctl_encoded_io_args *encoded)
10190 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10191 struct btrfs_root *root = inode->root;
10192 struct btrfs_fs_info *fs_info = root->fs_info;
10193 struct extent_io_tree *io_tree = &inode->io_tree;
10194 struct extent_changeset *data_reserved = NULL;
10195 struct extent_state *cached_state = NULL;
10196 struct btrfs_ordered_extent *ordered;
10200 u64 num_bytes, ram_bytes, disk_num_bytes;
10201 unsigned long nr_pages, i;
10202 struct page **pages;
10203 struct btrfs_key ins;
10204 bool extent_reserved = false;
10205 struct extent_map *em;
10208 switch (encoded->compression) {
10209 case BTRFS_ENCODED_IO_COMPRESSION_ZLIB:
10210 compression = BTRFS_COMPRESS_ZLIB;
10212 case BTRFS_ENCODED_IO_COMPRESSION_ZSTD:
10213 compression = BTRFS_COMPRESS_ZSTD;
10215 case BTRFS_ENCODED_IO_COMPRESSION_LZO_4K:
10216 case BTRFS_ENCODED_IO_COMPRESSION_LZO_8K:
10217 case BTRFS_ENCODED_IO_COMPRESSION_LZO_16K:
10218 case BTRFS_ENCODED_IO_COMPRESSION_LZO_32K:
10219 case BTRFS_ENCODED_IO_COMPRESSION_LZO_64K:
10220 /* The sector size must match for LZO. */
10221 if (encoded->compression -
10222 BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + 12 !=
10223 fs_info->sectorsize_bits)
10225 compression = BTRFS_COMPRESS_LZO;
10230 if (encoded->encryption != BTRFS_ENCODED_IO_ENCRYPTION_NONE)
10233 orig_count = iov_iter_count(from);
10235 /* The extent size must be sane. */
10236 if (encoded->unencoded_len > BTRFS_MAX_UNCOMPRESSED ||
10237 orig_count > BTRFS_MAX_COMPRESSED || orig_count == 0)
10241 * The compressed data must be smaller than the decompressed data.
10243 * It's of course possible for data to compress to larger or the same
10244 * size, but the buffered I/O path falls back to no compression for such
10245 * data, and we don't want to break any assumptions by creating these
10248 * Note that this is less strict than the current check we have that the
10249 * compressed data must be at least one sector smaller than the
10250 * decompressed data. We only want to enforce the weaker requirement
10251 * from old kernels that it is at least one byte smaller.
10253 if (orig_count >= encoded->unencoded_len)
10256 /* The extent must start on a sector boundary. */
10257 start = iocb->ki_pos;
10258 if (!IS_ALIGNED(start, fs_info->sectorsize))
10262 * The extent must end on a sector boundary. However, we allow a write
10263 * which ends at or extends i_size to have an unaligned length; we round
10264 * up the extent size and set i_size to the unaligned end.
10266 if (start + encoded->len < inode->vfs_inode.i_size &&
10267 !IS_ALIGNED(start + encoded->len, fs_info->sectorsize))
10270 /* Finally, the offset in the unencoded data must be sector-aligned. */
10271 if (!IS_ALIGNED(encoded->unencoded_offset, fs_info->sectorsize))
10274 num_bytes = ALIGN(encoded->len, fs_info->sectorsize);
10275 ram_bytes = ALIGN(encoded->unencoded_len, fs_info->sectorsize);
10276 end = start + num_bytes - 1;
10279 * If the extent cannot be inline, the compressed data on disk must be
10280 * sector-aligned. For convenience, we extend it with zeroes if it
10283 disk_num_bytes = ALIGN(orig_count, fs_info->sectorsize);
10284 nr_pages = DIV_ROUND_UP(disk_num_bytes, PAGE_SIZE);
10285 pages = kvcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL_ACCOUNT);
10288 for (i = 0; i < nr_pages; i++) {
10289 size_t bytes = min_t(size_t, PAGE_SIZE, iov_iter_count(from));
10292 pages[i] = alloc_page(GFP_KERNEL_ACCOUNT);
10297 kaddr = kmap_local_page(pages[i]);
10298 if (copy_from_iter(kaddr, bytes, from) != bytes) {
10299 kunmap_local(kaddr);
10303 if (bytes < PAGE_SIZE)
10304 memset(kaddr + bytes, 0, PAGE_SIZE - bytes);
10305 kunmap_local(kaddr);
10309 struct btrfs_ordered_extent *ordered;
10311 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start, num_bytes);
10314 ret = invalidate_inode_pages2_range(inode->vfs_inode.i_mapping,
10315 start >> PAGE_SHIFT,
10316 end >> PAGE_SHIFT);
10319 lock_extent(io_tree, start, end, &cached_state);
10320 ordered = btrfs_lookup_ordered_range(inode, start, num_bytes);
10322 !filemap_range_has_page(inode->vfs_inode.i_mapping, start, end))
10325 btrfs_put_ordered_extent(ordered);
10326 unlock_extent(io_tree, start, end, &cached_state);
10331 * We don't use the higher-level delalloc space functions because our
10332 * num_bytes and disk_num_bytes are different.
10334 ret = btrfs_alloc_data_chunk_ondemand(inode, disk_num_bytes);
10337 ret = btrfs_qgroup_reserve_data(inode, &data_reserved, start, num_bytes);
10339 goto out_free_data_space;
10340 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes, disk_num_bytes,
10343 goto out_qgroup_free_data;
10345 /* Try an inline extent first. */
10346 if (start == 0 && encoded->unencoded_len == encoded->len &&
10347 encoded->unencoded_offset == 0) {
10348 ret = cow_file_range_inline(inode, encoded->len, orig_count,
10349 compression, pages, true);
10353 goto out_delalloc_release;
10357 ret = btrfs_reserve_extent(root, disk_num_bytes, disk_num_bytes,
10358 disk_num_bytes, 0, 0, &ins, 1, 1);
10360 goto out_delalloc_release;
10361 extent_reserved = true;
10363 em = create_io_em(inode, start, num_bytes,
10364 start - encoded->unencoded_offset, ins.objectid,
10365 ins.offset, ins.offset, ram_bytes, compression,
10366 BTRFS_ORDERED_COMPRESSED);
10369 goto out_free_reserved;
10371 free_extent_map(em);
10373 ordered = btrfs_alloc_ordered_extent(inode, start, num_bytes, ram_bytes,
10374 ins.objectid, ins.offset,
10375 encoded->unencoded_offset,
10376 (1 << BTRFS_ORDERED_ENCODED) |
10377 (1 << BTRFS_ORDERED_COMPRESSED),
10379 if (IS_ERR(ordered)) {
10380 btrfs_drop_extent_map_range(inode, start, end, false);
10381 ret = PTR_ERR(ordered);
10382 goto out_free_reserved;
10384 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10386 if (start + encoded->len > inode->vfs_inode.i_size)
10387 i_size_write(&inode->vfs_inode, start + encoded->len);
10389 unlock_extent(io_tree, start, end, &cached_state);
10391 btrfs_delalloc_release_extents(inode, num_bytes);
10393 btrfs_submit_compressed_write(ordered, pages, nr_pages, 0, false);
10398 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10399 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
10400 out_delalloc_release:
10401 btrfs_delalloc_release_extents(inode, num_bytes);
10402 btrfs_delalloc_release_metadata(inode, disk_num_bytes, ret < 0);
10403 out_qgroup_free_data:
10405 btrfs_qgroup_free_data(inode, data_reserved, start, num_bytes);
10406 out_free_data_space:
10408 * If btrfs_reserve_extent() succeeded, then we already decremented
10411 if (!extent_reserved)
10412 btrfs_free_reserved_data_space_noquota(fs_info, disk_num_bytes);
10414 unlock_extent(io_tree, start, end, &cached_state);
10416 for (i = 0; i < nr_pages; i++) {
10418 __free_page(pages[i]);
10423 iocb->ki_pos += encoded->len;
10429 * Add an entry indicating a block group or device which is pinned by a
10430 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10431 * negative errno on failure.
10433 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10434 bool is_block_group)
10436 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10437 struct btrfs_swapfile_pin *sp, *entry;
10438 struct rb_node **p;
10439 struct rb_node *parent = NULL;
10441 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10446 sp->is_block_group = is_block_group;
10447 sp->bg_extent_count = 1;
10449 spin_lock(&fs_info->swapfile_pins_lock);
10450 p = &fs_info->swapfile_pins.rb_node;
10453 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10454 if (sp->ptr < entry->ptr ||
10455 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10456 p = &(*p)->rb_left;
10457 } else if (sp->ptr > entry->ptr ||
10458 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10459 p = &(*p)->rb_right;
10461 if (is_block_group)
10462 entry->bg_extent_count++;
10463 spin_unlock(&fs_info->swapfile_pins_lock);
10468 rb_link_node(&sp->node, parent, p);
10469 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10470 spin_unlock(&fs_info->swapfile_pins_lock);
10474 /* Free all of the entries pinned by this swapfile. */
10475 static void btrfs_free_swapfile_pins(struct inode *inode)
10477 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10478 struct btrfs_swapfile_pin *sp;
10479 struct rb_node *node, *next;
10481 spin_lock(&fs_info->swapfile_pins_lock);
10482 node = rb_first(&fs_info->swapfile_pins);
10484 next = rb_next(node);
10485 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10486 if (sp->inode == inode) {
10487 rb_erase(&sp->node, &fs_info->swapfile_pins);
10488 if (sp->is_block_group) {
10489 btrfs_dec_block_group_swap_extents(sp->ptr,
10490 sp->bg_extent_count);
10491 btrfs_put_block_group(sp->ptr);
10497 spin_unlock(&fs_info->swapfile_pins_lock);
10500 struct btrfs_swap_info {
10506 unsigned long nr_pages;
10510 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10511 struct btrfs_swap_info *bsi)
10513 unsigned long nr_pages;
10514 unsigned long max_pages;
10515 u64 first_ppage, first_ppage_reported, next_ppage;
10519 * Our swapfile may have had its size extended after the swap header was
10520 * written. In that case activating the swapfile should not go beyond
10521 * the max size set in the swap header.
10523 if (bsi->nr_pages >= sis->max)
10526 max_pages = sis->max - bsi->nr_pages;
10527 first_ppage = PAGE_ALIGN(bsi->block_start) >> PAGE_SHIFT;
10528 next_ppage = PAGE_ALIGN_DOWN(bsi->block_start + bsi->block_len) >> PAGE_SHIFT;
10530 if (first_ppage >= next_ppage)
10532 nr_pages = next_ppage - first_ppage;
10533 nr_pages = min(nr_pages, max_pages);
10535 first_ppage_reported = first_ppage;
10536 if (bsi->start == 0)
10537 first_ppage_reported++;
10538 if (bsi->lowest_ppage > first_ppage_reported)
10539 bsi->lowest_ppage = first_ppage_reported;
10540 if (bsi->highest_ppage < (next_ppage - 1))
10541 bsi->highest_ppage = next_ppage - 1;
10543 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10546 bsi->nr_extents += ret;
10547 bsi->nr_pages += nr_pages;
10551 static void btrfs_swap_deactivate(struct file *file)
10553 struct inode *inode = file_inode(file);
10555 btrfs_free_swapfile_pins(inode);
10556 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10559 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10562 struct inode *inode = file_inode(file);
10563 struct btrfs_root *root = BTRFS_I(inode)->root;
10564 struct btrfs_fs_info *fs_info = root->fs_info;
10565 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10566 struct extent_state *cached_state = NULL;
10567 struct extent_map *em = NULL;
10568 struct btrfs_device *device = NULL;
10569 struct btrfs_swap_info bsi = {
10570 .lowest_ppage = (sector_t)-1ULL,
10577 * If the swap file was just created, make sure delalloc is done. If the
10578 * file changes again after this, the user is doing something stupid and
10579 * we don't really care.
10581 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10586 * The inode is locked, so these flags won't change after we check them.
10588 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10589 btrfs_warn(fs_info, "swapfile must not be compressed");
10592 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10593 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10596 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10597 btrfs_warn(fs_info, "swapfile must not be checksummed");
10602 * Balance or device remove/replace/resize can move stuff around from
10603 * under us. The exclop protection makes sure they aren't running/won't
10604 * run concurrently while we are mapping the swap extents, and
10605 * fs_info->swapfile_pins prevents them from running while the swap
10606 * file is active and moving the extents. Note that this also prevents
10607 * a concurrent device add which isn't actually necessary, but it's not
10608 * really worth the trouble to allow it.
10610 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10611 btrfs_warn(fs_info,
10612 "cannot activate swapfile while exclusive operation is running");
10617 * Prevent snapshot creation while we are activating the swap file.
10618 * We do not want to race with snapshot creation. If snapshot creation
10619 * already started before we bumped nr_swapfiles from 0 to 1 and
10620 * completes before the first write into the swap file after it is
10621 * activated, than that write would fallback to COW.
10623 if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) {
10624 btrfs_exclop_finish(fs_info);
10625 btrfs_warn(fs_info,
10626 "cannot activate swapfile because snapshot creation is in progress");
10630 * Snapshots can create extents which require COW even if NODATACOW is
10631 * set. We use this counter to prevent snapshots. We must increment it
10632 * before walking the extents because we don't want a concurrent
10633 * snapshot to run after we've already checked the extents.
10635 * It is possible that subvolume is marked for deletion but still not
10636 * removed yet. To prevent this race, we check the root status before
10637 * activating the swapfile.
10639 spin_lock(&root->root_item_lock);
10640 if (btrfs_root_dead(root)) {
10641 spin_unlock(&root->root_item_lock);
10643 btrfs_exclop_finish(fs_info);
10644 btrfs_warn(fs_info,
10645 "cannot activate swapfile because subvolume %llu is being deleted",
10646 root->root_key.objectid);
10649 atomic_inc(&root->nr_swapfiles);
10650 spin_unlock(&root->root_item_lock);
10652 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10654 lock_extent(io_tree, 0, isize - 1, &cached_state);
10656 while (start < isize) {
10657 u64 logical_block_start, physical_block_start;
10658 struct btrfs_block_group *bg;
10659 u64 len = isize - start;
10661 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
10667 if (em->block_start == EXTENT_MAP_HOLE) {
10668 btrfs_warn(fs_info, "swapfile must not have holes");
10672 if (em->block_start == EXTENT_MAP_INLINE) {
10674 * It's unlikely we'll ever actually find ourselves
10675 * here, as a file small enough to fit inline won't be
10676 * big enough to store more than the swap header, but in
10677 * case something changes in the future, let's catch it
10678 * here rather than later.
10680 btrfs_warn(fs_info, "swapfile must not be inline");
10684 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10685 btrfs_warn(fs_info, "swapfile must not be compressed");
10690 logical_block_start = em->block_start + (start - em->start);
10691 len = min(len, em->len - (start - em->start));
10692 free_extent_map(em);
10695 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, false, true);
10701 btrfs_warn(fs_info,
10702 "swapfile must not be copy-on-write");
10707 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10713 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10714 btrfs_warn(fs_info,
10715 "swapfile must have single data profile");
10720 if (device == NULL) {
10721 device = em->map_lookup->stripes[0].dev;
10722 ret = btrfs_add_swapfile_pin(inode, device, false);
10727 } else if (device != em->map_lookup->stripes[0].dev) {
10728 btrfs_warn(fs_info, "swapfile must be on one device");
10733 physical_block_start = (em->map_lookup->stripes[0].physical +
10734 (logical_block_start - em->start));
10735 len = min(len, em->len - (logical_block_start - em->start));
10736 free_extent_map(em);
10739 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10741 btrfs_warn(fs_info,
10742 "could not find block group containing swapfile");
10747 if (!btrfs_inc_block_group_swap_extents(bg)) {
10748 btrfs_warn(fs_info,
10749 "block group for swapfile at %llu is read-only%s",
10751 atomic_read(&fs_info->scrubs_running) ?
10752 " (scrub running)" : "");
10753 btrfs_put_block_group(bg);
10758 ret = btrfs_add_swapfile_pin(inode, bg, true);
10760 btrfs_put_block_group(bg);
10767 if (bsi.block_len &&
10768 bsi.block_start + bsi.block_len == physical_block_start) {
10769 bsi.block_len += len;
10771 if (bsi.block_len) {
10772 ret = btrfs_add_swap_extent(sis, &bsi);
10777 bsi.block_start = physical_block_start;
10778 bsi.block_len = len;
10785 ret = btrfs_add_swap_extent(sis, &bsi);
10788 if (!IS_ERR_OR_NULL(em))
10789 free_extent_map(em);
10791 unlock_extent(io_tree, 0, isize - 1, &cached_state);
10794 btrfs_swap_deactivate(file);
10796 btrfs_drew_write_unlock(&root->snapshot_lock);
10798 btrfs_exclop_finish(fs_info);
10804 sis->bdev = device->bdev;
10805 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10806 sis->max = bsi.nr_pages;
10807 sis->pages = bsi.nr_pages - 1;
10808 sis->highest_bit = bsi.nr_pages - 1;
10809 return bsi.nr_extents;
10812 static void btrfs_swap_deactivate(struct file *file)
10816 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10819 return -EOPNOTSUPP;
10824 * Update the number of bytes used in the VFS' inode. When we replace extents in
10825 * a range (clone, dedupe, fallocate's zero range), we must update the number of
10826 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
10827 * always get a correct value.
10829 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
10830 const u64 add_bytes,
10831 const u64 del_bytes)
10833 if (add_bytes == del_bytes)
10836 spin_lock(&inode->lock);
10838 inode_sub_bytes(&inode->vfs_inode, del_bytes);
10840 inode_add_bytes(&inode->vfs_inode, add_bytes);
10841 spin_unlock(&inode->lock);
10845 * Verify that there are no ordered extents for a given file range.
10847 * @inode: The target inode.
10848 * @start: Start offset of the file range, should be sector size aligned.
10849 * @end: End offset (inclusive) of the file range, its value +1 should be
10850 * sector size aligned.
10852 * This should typically be used for cases where we locked an inode's VFS lock in
10853 * exclusive mode, we have also locked the inode's i_mmap_lock in exclusive mode,
10854 * we have flushed all delalloc in the range, we have waited for all ordered
10855 * extents in the range to complete and finally we have locked the file range in
10856 * the inode's io_tree.
10858 void btrfs_assert_inode_range_clean(struct btrfs_inode *inode, u64 start, u64 end)
10860 struct btrfs_root *root = inode->root;
10861 struct btrfs_ordered_extent *ordered;
10863 if (!IS_ENABLED(CONFIG_BTRFS_ASSERT))
10866 ordered = btrfs_lookup_first_ordered_range(inode, start, end + 1 - start);
10868 btrfs_err(root->fs_info,
10869 "found unexpected ordered extent in file range [%llu, %llu] for inode %llu root %llu (ordered range [%llu, %llu])",
10870 start, end, btrfs_ino(inode), root->root_key.objectid,
10871 ordered->file_offset,
10872 ordered->file_offset + ordered->num_bytes - 1);
10873 btrfs_put_ordered_extent(ordered);
10876 ASSERT(ordered == NULL);
10879 static const struct inode_operations btrfs_dir_inode_operations = {
10880 .getattr = btrfs_getattr,
10881 .lookup = btrfs_lookup,
10882 .create = btrfs_create,
10883 .unlink = btrfs_unlink,
10884 .link = btrfs_link,
10885 .mkdir = btrfs_mkdir,
10886 .rmdir = btrfs_rmdir,
10887 .rename = btrfs_rename2,
10888 .symlink = btrfs_symlink,
10889 .setattr = btrfs_setattr,
10890 .mknod = btrfs_mknod,
10891 .listxattr = btrfs_listxattr,
10892 .permission = btrfs_permission,
10893 .get_inode_acl = btrfs_get_acl,
10894 .set_acl = btrfs_set_acl,
10895 .update_time = btrfs_update_time,
10896 .tmpfile = btrfs_tmpfile,
10897 .fileattr_get = btrfs_fileattr_get,
10898 .fileattr_set = btrfs_fileattr_set,
10901 static const struct file_operations btrfs_dir_file_operations = {
10902 .llseek = generic_file_llseek,
10903 .read = generic_read_dir,
10904 .iterate_shared = btrfs_real_readdir,
10905 .open = btrfs_opendir,
10906 .unlocked_ioctl = btrfs_ioctl,
10907 #ifdef CONFIG_COMPAT
10908 .compat_ioctl = btrfs_compat_ioctl,
10910 .release = btrfs_release_file,
10911 .fsync = btrfs_sync_file,
10915 * btrfs doesn't support the bmap operation because swapfiles
10916 * use bmap to make a mapping of extents in the file. They assume
10917 * these extents won't change over the life of the file and they
10918 * use the bmap result to do IO directly to the drive.
10920 * the btrfs bmap call would return logical addresses that aren't
10921 * suitable for IO and they also will change frequently as COW
10922 * operations happen. So, swapfile + btrfs == corruption.
10924 * For now we're avoiding this by dropping bmap.
10926 static const struct address_space_operations btrfs_aops = {
10927 .read_folio = btrfs_read_folio,
10928 .writepages = btrfs_writepages,
10929 .readahead = btrfs_readahead,
10930 .invalidate_folio = btrfs_invalidate_folio,
10931 .release_folio = btrfs_release_folio,
10932 .migrate_folio = btrfs_migrate_folio,
10933 .dirty_folio = filemap_dirty_folio,
10934 .error_remove_page = generic_error_remove_page,
10935 .swap_activate = btrfs_swap_activate,
10936 .swap_deactivate = btrfs_swap_deactivate,
10939 static const struct inode_operations btrfs_file_inode_operations = {
10940 .getattr = btrfs_getattr,
10941 .setattr = btrfs_setattr,
10942 .listxattr = btrfs_listxattr,
10943 .permission = btrfs_permission,
10944 .fiemap = btrfs_fiemap,
10945 .get_inode_acl = btrfs_get_acl,
10946 .set_acl = btrfs_set_acl,
10947 .update_time = btrfs_update_time,
10948 .fileattr_get = btrfs_fileattr_get,
10949 .fileattr_set = btrfs_fileattr_set,
10951 static const struct inode_operations btrfs_special_inode_operations = {
10952 .getattr = btrfs_getattr,
10953 .setattr = btrfs_setattr,
10954 .permission = btrfs_permission,
10955 .listxattr = btrfs_listxattr,
10956 .get_inode_acl = btrfs_get_acl,
10957 .set_acl = btrfs_set_acl,
10958 .update_time = btrfs_update_time,
10960 static const struct inode_operations btrfs_symlink_inode_operations = {
10961 .get_link = page_get_link,
10962 .getattr = btrfs_getattr,
10963 .setattr = btrfs_setattr,
10964 .permission = btrfs_permission,
10965 .listxattr = btrfs_listxattr,
10966 .update_time = btrfs_update_time,
10969 const struct dentry_operations btrfs_dentry_operations = {
10970 .d_delete = btrfs_dentry_delete,