1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
7 #include <linux/pagemap.h>
8 #include <linux/time.h>
9 #include <linux/init.h>
10 #include <linux/string.h>
11 #include <linux/backing-dev.h>
12 #include <linux/falloc.h>
13 #include <linux/writeback.h>
14 #include <linux/compat.h>
15 #include <linux/slab.h>
16 #include <linux/btrfs.h>
17 #include <linux/uio.h>
18 #include <linux/iversion.h>
19 #include <linux/fsverity.h>
22 #include "transaction.h"
23 #include "btrfs_inode.h"
24 #include "print-tree.h"
29 #include "compression.h"
30 #include "delalloc-space.h"
34 static struct kmem_cache *btrfs_inode_defrag_cachep;
36 * when auto defrag is enabled we
37 * queue up these defrag structs to remember which
38 * inodes need defragging passes
41 struct rb_node rb_node;
45 * transid where the defrag was added, we search for
46 * extents newer than this
53 /* last offset we were able to defrag */
56 /* if we've wrapped around back to zero once already */
60 static int __compare_inode_defrag(struct inode_defrag *defrag1,
61 struct inode_defrag *defrag2)
63 if (defrag1->root > defrag2->root)
65 else if (defrag1->root < defrag2->root)
67 else if (defrag1->ino > defrag2->ino)
69 else if (defrag1->ino < defrag2->ino)
75 /* pop a record for an inode into the defrag tree. The lock
76 * must be held already
78 * If you're inserting a record for an older transid than an
79 * existing record, the transid already in the tree is lowered
81 * If an existing record is found the defrag item you
84 static int __btrfs_add_inode_defrag(struct btrfs_inode *inode,
85 struct inode_defrag *defrag)
87 struct btrfs_fs_info *fs_info = inode->root->fs_info;
88 struct inode_defrag *entry;
90 struct rb_node *parent = NULL;
93 p = &fs_info->defrag_inodes.rb_node;
96 entry = rb_entry(parent, struct inode_defrag, rb_node);
98 ret = __compare_inode_defrag(defrag, entry);
100 p = &parent->rb_left;
102 p = &parent->rb_right;
104 /* if we're reinserting an entry for
105 * an old defrag run, make sure to
106 * lower the transid of our existing record
108 if (defrag->transid < entry->transid)
109 entry->transid = defrag->transid;
110 if (defrag->last_offset > entry->last_offset)
111 entry->last_offset = defrag->last_offset;
115 set_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags);
116 rb_link_node(&defrag->rb_node, parent, p);
117 rb_insert_color(&defrag->rb_node, &fs_info->defrag_inodes);
121 static inline int __need_auto_defrag(struct btrfs_fs_info *fs_info)
123 if (!btrfs_test_opt(fs_info, AUTO_DEFRAG))
126 if (btrfs_fs_closing(fs_info))
133 * insert a defrag record for this inode if auto defrag is
136 int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans,
137 struct btrfs_inode *inode)
139 struct btrfs_root *root = inode->root;
140 struct btrfs_fs_info *fs_info = root->fs_info;
141 struct inode_defrag *defrag;
145 if (!__need_auto_defrag(fs_info))
148 if (test_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags))
152 transid = trans->transid;
154 transid = inode->root->last_trans;
156 defrag = kmem_cache_zalloc(btrfs_inode_defrag_cachep, GFP_NOFS);
160 defrag->ino = btrfs_ino(inode);
161 defrag->transid = transid;
162 defrag->root = root->root_key.objectid;
164 spin_lock(&fs_info->defrag_inodes_lock);
165 if (!test_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags)) {
167 * If we set IN_DEFRAG flag and evict the inode from memory,
168 * and then re-read this inode, this new inode doesn't have
169 * IN_DEFRAG flag. At the case, we may find the existed defrag.
171 ret = __btrfs_add_inode_defrag(inode, defrag);
173 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
175 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
177 spin_unlock(&fs_info->defrag_inodes_lock);
182 * Requeue the defrag object. If there is a defrag object that points to
183 * the same inode in the tree, we will merge them together (by
184 * __btrfs_add_inode_defrag()) and free the one that we want to requeue.
186 static void btrfs_requeue_inode_defrag(struct btrfs_inode *inode,
187 struct inode_defrag *defrag)
189 struct btrfs_fs_info *fs_info = inode->root->fs_info;
192 if (!__need_auto_defrag(fs_info))
196 * Here we don't check the IN_DEFRAG flag, because we need merge
199 spin_lock(&fs_info->defrag_inodes_lock);
200 ret = __btrfs_add_inode_defrag(inode, defrag);
201 spin_unlock(&fs_info->defrag_inodes_lock);
206 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
210 * pick the defragable inode that we want, if it doesn't exist, we will get
213 static struct inode_defrag *
214 btrfs_pick_defrag_inode(struct btrfs_fs_info *fs_info, u64 root, u64 ino)
216 struct inode_defrag *entry = NULL;
217 struct inode_defrag tmp;
219 struct rb_node *parent = NULL;
225 spin_lock(&fs_info->defrag_inodes_lock);
226 p = fs_info->defrag_inodes.rb_node;
229 entry = rb_entry(parent, struct inode_defrag, rb_node);
231 ret = __compare_inode_defrag(&tmp, entry);
235 p = parent->rb_right;
240 if (parent && __compare_inode_defrag(&tmp, entry) > 0) {
241 parent = rb_next(parent);
243 entry = rb_entry(parent, struct inode_defrag, rb_node);
249 rb_erase(parent, &fs_info->defrag_inodes);
250 spin_unlock(&fs_info->defrag_inodes_lock);
254 void btrfs_cleanup_defrag_inodes(struct btrfs_fs_info *fs_info)
256 struct inode_defrag *defrag;
257 struct rb_node *node;
259 spin_lock(&fs_info->defrag_inodes_lock);
260 node = rb_first(&fs_info->defrag_inodes);
262 rb_erase(node, &fs_info->defrag_inodes);
263 defrag = rb_entry(node, struct inode_defrag, rb_node);
264 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
266 cond_resched_lock(&fs_info->defrag_inodes_lock);
268 node = rb_first(&fs_info->defrag_inodes);
270 spin_unlock(&fs_info->defrag_inodes_lock);
273 #define BTRFS_DEFRAG_BATCH 1024
275 static int __btrfs_run_defrag_inode(struct btrfs_fs_info *fs_info,
276 struct inode_defrag *defrag)
278 struct btrfs_root *inode_root;
280 struct btrfs_ioctl_defrag_range_args range;
285 inode_root = btrfs_get_fs_root(fs_info, defrag->root, true);
286 if (IS_ERR(inode_root)) {
287 ret = PTR_ERR(inode_root);
291 inode = btrfs_iget(fs_info->sb, defrag->ino, inode_root);
292 btrfs_put_root(inode_root);
294 ret = PTR_ERR(inode);
298 /* do a chunk of defrag */
299 clear_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
300 memset(&range, 0, sizeof(range));
302 range.start = defrag->last_offset;
304 sb_start_write(fs_info->sb);
305 num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid,
307 sb_end_write(fs_info->sb);
309 * if we filled the whole defrag batch, there
310 * must be more work to do. Queue this defrag
313 if (num_defrag == BTRFS_DEFRAG_BATCH) {
314 defrag->last_offset = range.start;
315 btrfs_requeue_inode_defrag(BTRFS_I(inode), defrag);
316 } else if (defrag->last_offset && !defrag->cycled) {
318 * we didn't fill our defrag batch, but
319 * we didn't start at zero. Make sure we loop
320 * around to the start of the file.
322 defrag->last_offset = 0;
324 btrfs_requeue_inode_defrag(BTRFS_I(inode), defrag);
326 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
332 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
337 * run through the list of inodes in the FS that need
340 int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info)
342 struct inode_defrag *defrag;
344 u64 root_objectid = 0;
346 atomic_inc(&fs_info->defrag_running);
348 /* Pause the auto defragger. */
349 if (test_bit(BTRFS_FS_STATE_REMOUNTING,
353 if (!__need_auto_defrag(fs_info))
356 /* find an inode to defrag */
357 defrag = btrfs_pick_defrag_inode(fs_info, root_objectid,
360 if (root_objectid || first_ino) {
369 first_ino = defrag->ino + 1;
370 root_objectid = defrag->root;
372 __btrfs_run_defrag_inode(fs_info, defrag);
374 atomic_dec(&fs_info->defrag_running);
377 * during unmount, we use the transaction_wait queue to
378 * wait for the defragger to stop
380 wake_up(&fs_info->transaction_wait);
384 /* simple helper to fault in pages and copy. This should go away
385 * and be replaced with calls into generic code.
387 static noinline int btrfs_copy_from_user(loff_t pos, size_t write_bytes,
388 struct page **prepared_pages,
392 size_t total_copied = 0;
394 int offset = offset_in_page(pos);
396 while (write_bytes > 0) {
397 size_t count = min_t(size_t,
398 PAGE_SIZE - offset, write_bytes);
399 struct page *page = prepared_pages[pg];
401 * Copy data from userspace to the current page
403 copied = copy_page_from_iter_atomic(page, offset, count, i);
405 /* Flush processor's dcache for this page */
406 flush_dcache_page(page);
409 * if we get a partial write, we can end up with
410 * partially up to date pages. These add
411 * a lot of complexity, so make sure they don't
412 * happen by forcing this copy to be retried.
414 * The rest of the btrfs_file_write code will fall
415 * back to page at a time copies after we return 0.
417 if (unlikely(copied < count)) {
418 if (!PageUptodate(page)) {
419 iov_iter_revert(i, copied);
426 write_bytes -= copied;
427 total_copied += copied;
429 if (offset == PAGE_SIZE) {
438 * unlocks pages after btrfs_file_write is done with them
440 static void btrfs_drop_pages(struct btrfs_fs_info *fs_info,
441 struct page **pages, size_t num_pages,
445 u64 block_start = round_down(pos, fs_info->sectorsize);
446 u64 block_len = round_up(pos + copied, fs_info->sectorsize) - block_start;
448 ASSERT(block_len <= U32_MAX);
449 for (i = 0; i < num_pages; i++) {
450 /* page checked is some magic around finding pages that
451 * have been modified without going through btrfs_set_page_dirty
452 * clear it here. There should be no need to mark the pages
453 * accessed as prepare_pages should have marked them accessed
454 * in prepare_pages via find_or_create_page()
456 btrfs_page_clamp_clear_checked(fs_info, pages[i], block_start,
458 unlock_page(pages[i]);
464 * After btrfs_copy_from_user(), update the following things for delalloc:
465 * - Mark newly dirtied pages as DELALLOC in the io tree.
466 * Used to advise which range is to be written back.
467 * - Mark modified pages as Uptodate/Dirty and not needing COW fixup
468 * - Update inode size for past EOF write
470 int btrfs_dirty_pages(struct btrfs_inode *inode, struct page **pages,
471 size_t num_pages, loff_t pos, size_t write_bytes,
472 struct extent_state **cached, bool noreserve)
474 struct btrfs_fs_info *fs_info = inode->root->fs_info;
479 u64 end_of_last_block;
480 u64 end_pos = pos + write_bytes;
481 loff_t isize = i_size_read(&inode->vfs_inode);
482 unsigned int extra_bits = 0;
484 if (write_bytes == 0)
488 extra_bits |= EXTENT_NORESERVE;
490 start_pos = round_down(pos, fs_info->sectorsize);
491 num_bytes = round_up(write_bytes + pos - start_pos,
492 fs_info->sectorsize);
493 ASSERT(num_bytes <= U32_MAX);
495 end_of_last_block = start_pos + num_bytes - 1;
498 * The pages may have already been dirty, clear out old accounting so
499 * we can set things up properly
501 clear_extent_bit(&inode->io_tree, start_pos, end_of_last_block,
502 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
505 err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
510 for (i = 0; i < num_pages; i++) {
511 struct page *p = pages[i];
513 btrfs_page_clamp_set_uptodate(fs_info, p, start_pos, num_bytes);
514 btrfs_page_clamp_clear_checked(fs_info, p, start_pos, num_bytes);
515 btrfs_page_clamp_set_dirty(fs_info, p, start_pos, num_bytes);
519 * we've only changed i_size in ram, and we haven't updated
520 * the disk i_size. There is no need to log the inode
524 i_size_write(&inode->vfs_inode, end_pos);
529 * this drops all the extents in the cache that intersect the range
530 * [start, end]. Existing extents are split as required.
532 void btrfs_drop_extent_cache(struct btrfs_inode *inode, u64 start, u64 end,
535 struct extent_map *em;
536 struct extent_map *split = NULL;
537 struct extent_map *split2 = NULL;
538 struct extent_map_tree *em_tree = &inode->extent_tree;
539 u64 len = end - start + 1;
547 WARN_ON(end < start);
548 if (end == (u64)-1) {
557 split = alloc_extent_map();
559 split2 = alloc_extent_map();
560 if (!split || !split2)
563 write_lock(&em_tree->lock);
564 em = lookup_extent_mapping(em_tree, start, len);
566 write_unlock(&em_tree->lock);
570 gen = em->generation;
571 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
572 if (testend && em->start + em->len >= start + len) {
574 write_unlock(&em_tree->lock);
577 start = em->start + em->len;
579 len = start + len - (em->start + em->len);
581 write_unlock(&em_tree->lock);
584 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
585 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
586 clear_bit(EXTENT_FLAG_LOGGING, &flags);
587 modified = !list_empty(&em->list);
591 if (em->start < start) {
592 split->start = em->start;
593 split->len = start - em->start;
595 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
596 split->orig_start = em->orig_start;
597 split->block_start = em->block_start;
600 split->block_len = em->block_len;
602 split->block_len = split->len;
603 split->orig_block_len = max(split->block_len,
605 split->ram_bytes = em->ram_bytes;
607 split->orig_start = split->start;
608 split->block_len = 0;
609 split->block_start = em->block_start;
610 split->orig_block_len = 0;
611 split->ram_bytes = split->len;
614 split->generation = gen;
615 split->flags = flags;
616 split->compress_type = em->compress_type;
617 replace_extent_mapping(em_tree, em, split, modified);
618 free_extent_map(split);
622 if (testend && em->start + em->len > start + len) {
623 u64 diff = start + len - em->start;
625 split->start = start + len;
626 split->len = em->start + em->len - (start + len);
627 split->flags = flags;
628 split->compress_type = em->compress_type;
629 split->generation = gen;
631 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
632 split->orig_block_len = max(em->block_len,
635 split->ram_bytes = em->ram_bytes;
637 split->block_len = em->block_len;
638 split->block_start = em->block_start;
639 split->orig_start = em->orig_start;
641 split->block_len = split->len;
642 split->block_start = em->block_start
644 split->orig_start = em->orig_start;
647 split->ram_bytes = split->len;
648 split->orig_start = split->start;
649 split->block_len = 0;
650 split->block_start = em->block_start;
651 split->orig_block_len = 0;
654 if (extent_map_in_tree(em)) {
655 replace_extent_mapping(em_tree, em, split,
658 ret = add_extent_mapping(em_tree, split,
660 ASSERT(ret == 0); /* Logic error */
662 free_extent_map(split);
666 if (extent_map_in_tree(em))
667 remove_extent_mapping(em_tree, em);
668 write_unlock(&em_tree->lock);
672 /* once for the tree*/
676 free_extent_map(split);
678 free_extent_map(split2);
682 * this is very complex, but the basic idea is to drop all extents
683 * in the range start - end. hint_block is filled in with a block number
684 * that would be a good hint to the block allocator for this file.
686 * If an extent intersects the range but is not entirely inside the range
687 * it is either truncated or split. Anything entirely inside the range
688 * is deleted from the tree.
690 * Note: the VFS' inode number of bytes is not updated, it's up to the caller
691 * to deal with that. We set the field 'bytes_found' of the arguments structure
692 * with the number of allocated bytes found in the target range, so that the
693 * caller can update the inode's number of bytes in an atomic way when
694 * replacing extents in a range to avoid races with stat(2).
696 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
697 struct btrfs_root *root, struct btrfs_inode *inode,
698 struct btrfs_drop_extents_args *args)
700 struct btrfs_fs_info *fs_info = root->fs_info;
701 struct extent_buffer *leaf;
702 struct btrfs_file_extent_item *fi;
703 struct btrfs_ref ref = { 0 };
704 struct btrfs_key key;
705 struct btrfs_key new_key;
706 u64 ino = btrfs_ino(inode);
707 u64 search_start = args->start;
710 u64 extent_offset = 0;
712 u64 last_end = args->start;
718 int modify_tree = -1;
721 int leafs_visited = 0;
722 struct btrfs_path *path = args->path;
724 args->bytes_found = 0;
725 args->extent_inserted = false;
727 /* Must always have a path if ->replace_extent is true */
728 ASSERT(!(args->replace_extent && !args->path));
731 path = btrfs_alloc_path();
738 if (args->drop_cache)
739 btrfs_drop_extent_cache(inode, args->start, args->end - 1, 0);
741 if (args->start >= inode->disk_i_size && !args->replace_extent)
744 update_refs = (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID);
747 ret = btrfs_lookup_file_extent(trans, root, path, ino,
748 search_start, modify_tree);
751 if (ret > 0 && path->slots[0] > 0 && search_start == args->start) {
752 leaf = path->nodes[0];
753 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
754 if (key.objectid == ino &&
755 key.type == BTRFS_EXTENT_DATA_KEY)
761 leaf = path->nodes[0];
762 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
764 ret = btrfs_next_leaf(root, path);
772 leaf = path->nodes[0];
776 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
778 if (key.objectid > ino)
780 if (WARN_ON_ONCE(key.objectid < ino) ||
781 key.type < BTRFS_EXTENT_DATA_KEY) {
786 if (key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= args->end)
789 fi = btrfs_item_ptr(leaf, path->slots[0],
790 struct btrfs_file_extent_item);
791 extent_type = btrfs_file_extent_type(leaf, fi);
793 if (extent_type == BTRFS_FILE_EXTENT_REG ||
794 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
795 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
796 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
797 extent_offset = btrfs_file_extent_offset(leaf, fi);
798 extent_end = key.offset +
799 btrfs_file_extent_num_bytes(leaf, fi);
800 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
801 extent_end = key.offset +
802 btrfs_file_extent_ram_bytes(leaf, fi);
809 * Don't skip extent items representing 0 byte lengths. They
810 * used to be created (bug) if while punching holes we hit
811 * -ENOSPC condition. So if we find one here, just ensure we
812 * delete it, otherwise we would insert a new file extent item
813 * with the same key (offset) as that 0 bytes length file
814 * extent item in the call to setup_items_for_insert() later
817 if (extent_end == key.offset && extent_end >= search_start) {
818 last_end = extent_end;
819 goto delete_extent_item;
822 if (extent_end <= search_start) {
828 search_start = max(key.offset, args->start);
829 if (recow || !modify_tree) {
831 btrfs_release_path(path);
836 * | - range to drop - |
837 * | -------- extent -------- |
839 if (args->start > key.offset && args->end < extent_end) {
841 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
846 memcpy(&new_key, &key, sizeof(new_key));
847 new_key.offset = args->start;
848 ret = btrfs_duplicate_item(trans, root, path,
850 if (ret == -EAGAIN) {
851 btrfs_release_path(path);
857 leaf = path->nodes[0];
858 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
859 struct btrfs_file_extent_item);
860 btrfs_set_file_extent_num_bytes(leaf, fi,
861 args->start - key.offset);
863 fi = btrfs_item_ptr(leaf, path->slots[0],
864 struct btrfs_file_extent_item);
866 extent_offset += args->start - key.offset;
867 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
868 btrfs_set_file_extent_num_bytes(leaf, fi,
869 extent_end - args->start);
870 btrfs_mark_buffer_dirty(leaf);
872 if (update_refs && disk_bytenr > 0) {
873 btrfs_init_generic_ref(&ref,
874 BTRFS_ADD_DELAYED_REF,
875 disk_bytenr, num_bytes, 0);
876 btrfs_init_data_ref(&ref,
877 root->root_key.objectid,
879 args->start - extent_offset,
881 ret = btrfs_inc_extent_ref(trans, &ref);
882 BUG_ON(ret); /* -ENOMEM */
884 key.offset = args->start;
887 * From here on out we will have actually dropped something, so
888 * last_end can be updated.
890 last_end = extent_end;
893 * | ---- range to drop ----- |
894 * | -------- extent -------- |
896 if (args->start <= key.offset && args->end < extent_end) {
897 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
902 memcpy(&new_key, &key, sizeof(new_key));
903 new_key.offset = args->end;
904 btrfs_set_item_key_safe(fs_info, path, &new_key);
906 extent_offset += args->end - key.offset;
907 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
908 btrfs_set_file_extent_num_bytes(leaf, fi,
909 extent_end - args->end);
910 btrfs_mark_buffer_dirty(leaf);
911 if (update_refs && disk_bytenr > 0)
912 args->bytes_found += args->end - key.offset;
916 search_start = extent_end;
918 * | ---- range to drop ----- |
919 * | -------- extent -------- |
921 if (args->start > key.offset && args->end >= extent_end) {
923 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
928 btrfs_set_file_extent_num_bytes(leaf, fi,
929 args->start - key.offset);
930 btrfs_mark_buffer_dirty(leaf);
931 if (update_refs && disk_bytenr > 0)
932 args->bytes_found += extent_end - args->start;
933 if (args->end == extent_end)
941 * | ---- range to drop ----- |
942 * | ------ extent ------ |
944 if (args->start <= key.offset && args->end >= extent_end) {
947 del_slot = path->slots[0];
950 BUG_ON(del_slot + del_nr != path->slots[0]);
955 extent_type == BTRFS_FILE_EXTENT_INLINE) {
956 args->bytes_found += extent_end - key.offset;
957 extent_end = ALIGN(extent_end,
958 fs_info->sectorsize);
959 } else if (update_refs && disk_bytenr > 0) {
960 btrfs_init_generic_ref(&ref,
961 BTRFS_DROP_DELAYED_REF,
962 disk_bytenr, num_bytes, 0);
963 btrfs_init_data_ref(&ref,
964 root->root_key.objectid,
966 key.offset - extent_offset, 0,
968 ret = btrfs_free_extent(trans, &ref);
969 BUG_ON(ret); /* -ENOMEM */
970 args->bytes_found += extent_end - key.offset;
973 if (args->end == extent_end)
976 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
981 ret = btrfs_del_items(trans, root, path, del_slot,
984 btrfs_abort_transaction(trans, ret);
991 btrfs_release_path(path);
998 if (!ret && del_nr > 0) {
1000 * Set path->slots[0] to first slot, so that after the delete
1001 * if items are move off from our leaf to its immediate left or
1002 * right neighbor leafs, we end up with a correct and adjusted
1003 * path->slots[0] for our insertion (if args->replace_extent).
1005 path->slots[0] = del_slot;
1006 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1008 btrfs_abort_transaction(trans, ret);
1011 leaf = path->nodes[0];
1013 * If btrfs_del_items() was called, it might have deleted a leaf, in
1014 * which case it unlocked our path, so check path->locks[0] matches a
1017 if (!ret && args->replace_extent && leafs_visited == 1 &&
1018 path->locks[0] == BTRFS_WRITE_LOCK &&
1019 btrfs_leaf_free_space(leaf) >=
1020 sizeof(struct btrfs_item) + args->extent_item_size) {
1023 key.type = BTRFS_EXTENT_DATA_KEY;
1024 key.offset = args->start;
1025 if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) {
1026 struct btrfs_key slot_key;
1028 btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]);
1029 if (btrfs_comp_cpu_keys(&key, &slot_key) > 0)
1032 btrfs_setup_item_for_insert(root, path, &key, args->extent_item_size);
1033 args->extent_inserted = true;
1037 btrfs_free_path(path);
1038 else if (!args->extent_inserted)
1039 btrfs_release_path(path);
1041 args->drop_end = found ? min(args->end, last_end) : args->end;
1046 static int extent_mergeable(struct extent_buffer *leaf, int slot,
1047 u64 objectid, u64 bytenr, u64 orig_offset,
1048 u64 *start, u64 *end)
1050 struct btrfs_file_extent_item *fi;
1051 struct btrfs_key key;
1054 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
1057 btrfs_item_key_to_cpu(leaf, &key, slot);
1058 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
1061 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1062 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
1063 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
1064 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
1065 btrfs_file_extent_compression(leaf, fi) ||
1066 btrfs_file_extent_encryption(leaf, fi) ||
1067 btrfs_file_extent_other_encoding(leaf, fi))
1070 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1071 if ((*start && *start != key.offset) || (*end && *end != extent_end))
1074 *start = key.offset;
1080 * Mark extent in the range start - end as written.
1082 * This changes extent type from 'pre-allocated' to 'regular'. If only
1083 * part of extent is marked as written, the extent will be split into
1086 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
1087 struct btrfs_inode *inode, u64 start, u64 end)
1089 struct btrfs_fs_info *fs_info = trans->fs_info;
1090 struct btrfs_root *root = inode->root;
1091 struct extent_buffer *leaf;
1092 struct btrfs_path *path;
1093 struct btrfs_file_extent_item *fi;
1094 struct btrfs_ref ref = { 0 };
1095 struct btrfs_key key;
1096 struct btrfs_key new_key;
1108 u64 ino = btrfs_ino(inode);
1110 path = btrfs_alloc_path();
1117 key.type = BTRFS_EXTENT_DATA_KEY;
1120 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1123 if (ret > 0 && path->slots[0] > 0)
1126 leaf = path->nodes[0];
1127 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1128 if (key.objectid != ino ||
1129 key.type != BTRFS_EXTENT_DATA_KEY) {
1131 btrfs_abort_transaction(trans, ret);
1134 fi = btrfs_item_ptr(leaf, path->slots[0],
1135 struct btrfs_file_extent_item);
1136 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_PREALLOC) {
1138 btrfs_abort_transaction(trans, ret);
1141 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1142 if (key.offset > start || extent_end < end) {
1144 btrfs_abort_transaction(trans, ret);
1148 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1149 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1150 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
1151 memcpy(&new_key, &key, sizeof(new_key));
1153 if (start == key.offset && end < extent_end) {
1156 if (extent_mergeable(leaf, path->slots[0] - 1,
1157 ino, bytenr, orig_offset,
1158 &other_start, &other_end)) {
1159 new_key.offset = end;
1160 btrfs_set_item_key_safe(fs_info, path, &new_key);
1161 fi = btrfs_item_ptr(leaf, path->slots[0],
1162 struct btrfs_file_extent_item);
1163 btrfs_set_file_extent_generation(leaf, fi,
1165 btrfs_set_file_extent_num_bytes(leaf, fi,
1167 btrfs_set_file_extent_offset(leaf, fi,
1169 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1170 struct btrfs_file_extent_item);
1171 btrfs_set_file_extent_generation(leaf, fi,
1173 btrfs_set_file_extent_num_bytes(leaf, fi,
1175 btrfs_mark_buffer_dirty(leaf);
1180 if (start > key.offset && end == extent_end) {
1183 if (extent_mergeable(leaf, path->slots[0] + 1,
1184 ino, bytenr, orig_offset,
1185 &other_start, &other_end)) {
1186 fi = btrfs_item_ptr(leaf, path->slots[0],
1187 struct btrfs_file_extent_item);
1188 btrfs_set_file_extent_num_bytes(leaf, fi,
1189 start - key.offset);
1190 btrfs_set_file_extent_generation(leaf, fi,
1193 new_key.offset = start;
1194 btrfs_set_item_key_safe(fs_info, path, &new_key);
1196 fi = btrfs_item_ptr(leaf, path->slots[0],
1197 struct btrfs_file_extent_item);
1198 btrfs_set_file_extent_generation(leaf, fi,
1200 btrfs_set_file_extent_num_bytes(leaf, fi,
1202 btrfs_set_file_extent_offset(leaf, fi,
1203 start - orig_offset);
1204 btrfs_mark_buffer_dirty(leaf);
1209 while (start > key.offset || end < extent_end) {
1210 if (key.offset == start)
1213 new_key.offset = split;
1214 ret = btrfs_duplicate_item(trans, root, path, &new_key);
1215 if (ret == -EAGAIN) {
1216 btrfs_release_path(path);
1220 btrfs_abort_transaction(trans, ret);
1224 leaf = path->nodes[0];
1225 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1226 struct btrfs_file_extent_item);
1227 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1228 btrfs_set_file_extent_num_bytes(leaf, fi,
1229 split - key.offset);
1231 fi = btrfs_item_ptr(leaf, path->slots[0],
1232 struct btrfs_file_extent_item);
1234 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1235 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
1236 btrfs_set_file_extent_num_bytes(leaf, fi,
1237 extent_end - split);
1238 btrfs_mark_buffer_dirty(leaf);
1240 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, bytenr,
1242 btrfs_init_data_ref(&ref, root->root_key.objectid, ino,
1243 orig_offset, 0, false);
1244 ret = btrfs_inc_extent_ref(trans, &ref);
1246 btrfs_abort_transaction(trans, ret);
1250 if (split == start) {
1253 if (start != key.offset) {
1255 btrfs_abort_transaction(trans, ret);
1266 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
1268 btrfs_init_data_ref(&ref, root->root_key.objectid, ino, orig_offset,
1270 if (extent_mergeable(leaf, path->slots[0] + 1,
1271 ino, bytenr, orig_offset,
1272 &other_start, &other_end)) {
1274 btrfs_release_path(path);
1277 extent_end = other_end;
1278 del_slot = path->slots[0] + 1;
1280 ret = btrfs_free_extent(trans, &ref);
1282 btrfs_abort_transaction(trans, ret);
1288 if (extent_mergeable(leaf, path->slots[0] - 1,
1289 ino, bytenr, orig_offset,
1290 &other_start, &other_end)) {
1292 btrfs_release_path(path);
1295 key.offset = other_start;
1296 del_slot = path->slots[0];
1298 ret = btrfs_free_extent(trans, &ref);
1300 btrfs_abort_transaction(trans, ret);
1305 fi = btrfs_item_ptr(leaf, path->slots[0],
1306 struct btrfs_file_extent_item);
1307 btrfs_set_file_extent_type(leaf, fi,
1308 BTRFS_FILE_EXTENT_REG);
1309 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1310 btrfs_mark_buffer_dirty(leaf);
1312 fi = btrfs_item_ptr(leaf, del_slot - 1,
1313 struct btrfs_file_extent_item);
1314 btrfs_set_file_extent_type(leaf, fi,
1315 BTRFS_FILE_EXTENT_REG);
1316 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1317 btrfs_set_file_extent_num_bytes(leaf, fi,
1318 extent_end - key.offset);
1319 btrfs_mark_buffer_dirty(leaf);
1321 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1323 btrfs_abort_transaction(trans, ret);
1328 btrfs_free_path(path);
1333 * on error we return an unlocked page and the error value
1334 * on success we return a locked page and 0
1336 static int prepare_uptodate_page(struct inode *inode,
1337 struct page *page, u64 pos,
1338 bool force_uptodate)
1342 if (((pos & (PAGE_SIZE - 1)) || force_uptodate) &&
1343 !PageUptodate(page)) {
1344 ret = btrfs_readpage(NULL, page);
1348 if (!PageUptodate(page)) {
1354 * Since btrfs_readpage() will unlock the page before it
1355 * returns, there is a window where btrfs_releasepage() can be
1356 * called to release the page. Here we check both inode
1357 * mapping and PagePrivate() to make sure the page was not
1360 * The private flag check is essential for subpage as we need
1361 * to store extra bitmap using page->private.
1363 if (page->mapping != inode->i_mapping || !PagePrivate(page)) {
1372 * this just gets pages into the page cache and locks them down.
1374 static noinline int prepare_pages(struct inode *inode, struct page **pages,
1375 size_t num_pages, loff_t pos,
1376 size_t write_bytes, bool force_uptodate)
1379 unsigned long index = pos >> PAGE_SHIFT;
1380 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1384 for (i = 0; i < num_pages; i++) {
1386 pages[i] = find_or_create_page(inode->i_mapping, index + i,
1387 mask | __GFP_WRITE);
1394 err = set_page_extent_mapped(pages[i]);
1401 err = prepare_uptodate_page(inode, pages[i], pos,
1403 if (!err && i == num_pages - 1)
1404 err = prepare_uptodate_page(inode, pages[i],
1405 pos + write_bytes, false);
1408 if (err == -EAGAIN) {
1415 wait_on_page_writeback(pages[i]);
1420 while (faili >= 0) {
1421 unlock_page(pages[faili]);
1422 put_page(pages[faili]);
1430 * This function locks the extent and properly waits for data=ordered extents
1431 * to finish before allowing the pages to be modified if need.
1434 * 1 - the extent is locked
1435 * 0 - the extent is not locked, and everything is OK
1436 * -EAGAIN - need re-prepare the pages
1437 * the other < 0 number - Something wrong happens
1440 lock_and_cleanup_extent_if_need(struct btrfs_inode *inode, struct page **pages,
1441 size_t num_pages, loff_t pos,
1443 u64 *lockstart, u64 *lockend,
1444 struct extent_state **cached_state)
1446 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1452 start_pos = round_down(pos, fs_info->sectorsize);
1453 last_pos = round_up(pos + write_bytes, fs_info->sectorsize) - 1;
1455 if (start_pos < inode->vfs_inode.i_size) {
1456 struct btrfs_ordered_extent *ordered;
1458 lock_extent_bits(&inode->io_tree, start_pos, last_pos,
1460 ordered = btrfs_lookup_ordered_range(inode, start_pos,
1461 last_pos - start_pos + 1);
1463 ordered->file_offset + ordered->num_bytes > start_pos &&
1464 ordered->file_offset <= last_pos) {
1465 unlock_extent_cached(&inode->io_tree, start_pos,
1466 last_pos, cached_state);
1467 for (i = 0; i < num_pages; i++) {
1468 unlock_page(pages[i]);
1471 btrfs_start_ordered_extent(ordered, 1);
1472 btrfs_put_ordered_extent(ordered);
1476 btrfs_put_ordered_extent(ordered);
1478 *lockstart = start_pos;
1479 *lockend = last_pos;
1484 * We should be called after prepare_pages() which should have locked
1485 * all pages in the range.
1487 for (i = 0; i < num_pages; i++)
1488 WARN_ON(!PageLocked(pages[i]));
1493 static int check_can_nocow(struct btrfs_inode *inode, loff_t pos,
1494 size_t *write_bytes, bool nowait)
1496 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1497 struct btrfs_root *root = inode->root;
1498 u64 lockstart, lockend;
1502 if (!(inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
1505 if (!nowait && !btrfs_drew_try_write_lock(&root->snapshot_lock))
1508 lockstart = round_down(pos, fs_info->sectorsize);
1509 lockend = round_up(pos + *write_bytes,
1510 fs_info->sectorsize) - 1;
1511 num_bytes = lockend - lockstart + 1;
1514 struct btrfs_ordered_extent *ordered;
1516 if (!try_lock_extent(&inode->io_tree, lockstart, lockend))
1519 ordered = btrfs_lookup_ordered_range(inode, lockstart,
1522 btrfs_put_ordered_extent(ordered);
1527 btrfs_lock_and_flush_ordered_range(inode, lockstart,
1531 ret = can_nocow_extent(&inode->vfs_inode, lockstart, &num_bytes,
1532 NULL, NULL, NULL, false);
1536 btrfs_drew_write_unlock(&root->snapshot_lock);
1538 *write_bytes = min_t(size_t, *write_bytes ,
1539 num_bytes - pos + lockstart);
1542 unlock_extent(&inode->io_tree, lockstart, lockend);
1547 static int check_nocow_nolock(struct btrfs_inode *inode, loff_t pos,
1548 size_t *write_bytes)
1550 return check_can_nocow(inode, pos, write_bytes, true);
1554 * Check if we can do nocow write into the range [@pos, @pos + @write_bytes)
1557 * @write_bytes: The length to write, will be updated to the nocow writeable
1560 * This function will flush ordered extents in the range to ensure proper
1564 * >0 and update @write_bytes if we can do nocow write
1565 * 0 if we can't do nocow write
1566 * -EAGAIN if we can't get the needed lock or there are ordered extents
1567 * for * (nowait == true) case
1568 * <0 if other error happened
1570 * NOTE: Callers need to release the lock by btrfs_check_nocow_unlock().
1572 int btrfs_check_nocow_lock(struct btrfs_inode *inode, loff_t pos,
1573 size_t *write_bytes)
1575 return check_can_nocow(inode, pos, write_bytes, false);
1578 void btrfs_check_nocow_unlock(struct btrfs_inode *inode)
1580 btrfs_drew_write_unlock(&inode->root->snapshot_lock);
1583 static void update_time_for_write(struct inode *inode)
1585 struct timespec64 now;
1587 if (IS_NOCMTIME(inode))
1590 now = current_time(inode);
1591 if (!timespec64_equal(&inode->i_mtime, &now))
1592 inode->i_mtime = now;
1594 if (!timespec64_equal(&inode->i_ctime, &now))
1595 inode->i_ctime = now;
1597 if (IS_I_VERSION(inode))
1598 inode_inc_iversion(inode);
1601 static int btrfs_write_check(struct kiocb *iocb, struct iov_iter *from,
1604 struct file *file = iocb->ki_filp;
1605 struct inode *inode = file_inode(file);
1606 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1607 loff_t pos = iocb->ki_pos;
1612 if (iocb->ki_flags & IOCB_NOWAIT) {
1613 size_t nocow_bytes = count;
1615 /* We will allocate space in case nodatacow is not set, so bail */
1616 if (check_nocow_nolock(BTRFS_I(inode), pos, &nocow_bytes) <= 0)
1619 * There are holes in the range or parts of the range that must
1620 * be COWed (shared extents, RO block groups, etc), so just bail
1623 if (nocow_bytes < count)
1627 current->backing_dev_info = inode_to_bdi(inode);
1628 ret = file_remove_privs(file);
1633 * We reserve space for updating the inode when we reserve space for the
1634 * extent we are going to write, so we will enospc out there. We don't
1635 * need to start yet another transaction to update the inode as we will
1636 * update the inode when we finish writing whatever data we write.
1638 update_time_for_write(inode);
1640 start_pos = round_down(pos, fs_info->sectorsize);
1641 oldsize = i_size_read(inode);
1642 if (start_pos > oldsize) {
1643 /* Expand hole size to cover write data, preventing empty gap */
1644 loff_t end_pos = round_up(pos + count, fs_info->sectorsize);
1646 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, end_pos);
1648 current->backing_dev_info = NULL;
1656 static noinline ssize_t btrfs_buffered_write(struct kiocb *iocb,
1659 struct file *file = iocb->ki_filp;
1661 struct inode *inode = file_inode(file);
1662 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1663 struct page **pages = NULL;
1664 struct extent_changeset *data_reserved = NULL;
1665 u64 release_bytes = 0;
1668 size_t num_written = 0;
1671 bool only_release_metadata = false;
1672 bool force_page_uptodate = false;
1673 loff_t old_isize = i_size_read(inode);
1674 unsigned int ilock_flags = 0;
1676 if (iocb->ki_flags & IOCB_NOWAIT)
1677 ilock_flags |= BTRFS_ILOCK_TRY;
1679 ret = btrfs_inode_lock(inode, ilock_flags);
1683 ret = generic_write_checks(iocb, i);
1687 ret = btrfs_write_check(iocb, i, ret);
1692 nrptrs = min(DIV_ROUND_UP(iov_iter_count(i), PAGE_SIZE),
1693 PAGE_SIZE / (sizeof(struct page *)));
1694 nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1695 nrptrs = max(nrptrs, 8);
1696 pages = kmalloc_array(nrptrs, sizeof(struct page *), GFP_KERNEL);
1702 while (iov_iter_count(i) > 0) {
1703 struct extent_state *cached_state = NULL;
1704 size_t offset = offset_in_page(pos);
1705 size_t sector_offset;
1706 size_t write_bytes = min(iov_iter_count(i),
1707 nrptrs * (size_t)PAGE_SIZE -
1710 size_t reserve_bytes;
1713 size_t dirty_sectors;
1718 * Fault pages before locking them in prepare_pages
1719 * to avoid recursive lock
1721 if (unlikely(fault_in_iov_iter_readable(i, write_bytes))) {
1726 only_release_metadata = false;
1727 sector_offset = pos & (fs_info->sectorsize - 1);
1729 extent_changeset_release(data_reserved);
1730 ret = btrfs_check_data_free_space(BTRFS_I(inode),
1731 &data_reserved, pos,
1735 * If we don't have to COW at the offset, reserve
1736 * metadata only. write_bytes may get smaller than
1739 if (btrfs_check_nocow_lock(BTRFS_I(inode), pos,
1741 only_release_metadata = true;
1746 num_pages = DIV_ROUND_UP(write_bytes + offset, PAGE_SIZE);
1747 WARN_ON(num_pages > nrptrs);
1748 reserve_bytes = round_up(write_bytes + sector_offset,
1749 fs_info->sectorsize);
1750 WARN_ON(reserve_bytes == 0);
1751 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
1754 if (!only_release_metadata)
1755 btrfs_free_reserved_data_space(BTRFS_I(inode),
1759 btrfs_check_nocow_unlock(BTRFS_I(inode));
1763 release_bytes = reserve_bytes;
1766 * This is going to setup the pages array with the number of
1767 * pages we want, so we don't really need to worry about the
1768 * contents of pages from loop to loop
1770 ret = prepare_pages(inode, pages, num_pages,
1772 force_page_uptodate);
1774 btrfs_delalloc_release_extents(BTRFS_I(inode),
1779 extents_locked = lock_and_cleanup_extent_if_need(
1780 BTRFS_I(inode), pages,
1781 num_pages, pos, write_bytes, &lockstart,
1782 &lockend, &cached_state);
1783 if (extents_locked < 0) {
1784 if (extents_locked == -EAGAIN)
1786 btrfs_delalloc_release_extents(BTRFS_I(inode),
1788 ret = extents_locked;
1792 copied = btrfs_copy_from_user(pos, write_bytes, pages, i);
1794 num_sectors = BTRFS_BYTES_TO_BLKS(fs_info, reserve_bytes);
1795 dirty_sectors = round_up(copied + sector_offset,
1796 fs_info->sectorsize);
1797 dirty_sectors = BTRFS_BYTES_TO_BLKS(fs_info, dirty_sectors);
1800 * if we have trouble faulting in the pages, fall
1801 * back to one page at a time
1803 if (copied < write_bytes)
1807 force_page_uptodate = true;
1811 force_page_uptodate = false;
1812 dirty_pages = DIV_ROUND_UP(copied + offset,
1816 if (num_sectors > dirty_sectors) {
1817 /* release everything except the sectors we dirtied */
1818 release_bytes -= dirty_sectors << fs_info->sectorsize_bits;
1819 if (only_release_metadata) {
1820 btrfs_delalloc_release_metadata(BTRFS_I(inode),
1821 release_bytes, true);
1825 __pos = round_down(pos,
1826 fs_info->sectorsize) +
1827 (dirty_pages << PAGE_SHIFT);
1828 btrfs_delalloc_release_space(BTRFS_I(inode),
1829 data_reserved, __pos,
1830 release_bytes, true);
1834 release_bytes = round_up(copied + sector_offset,
1835 fs_info->sectorsize);
1837 ret = btrfs_dirty_pages(BTRFS_I(inode), pages,
1838 dirty_pages, pos, copied,
1839 &cached_state, only_release_metadata);
1842 * If we have not locked the extent range, because the range's
1843 * start offset is >= i_size, we might still have a non-NULL
1844 * cached extent state, acquired while marking the extent range
1845 * as delalloc through btrfs_dirty_pages(). Therefore free any
1846 * possible cached extent state to avoid a memory leak.
1849 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1850 lockstart, lockend, &cached_state);
1852 free_extent_state(cached_state);
1854 btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes);
1856 btrfs_drop_pages(fs_info, pages, num_pages, pos, copied);
1861 if (only_release_metadata)
1862 btrfs_check_nocow_unlock(BTRFS_I(inode));
1864 btrfs_drop_pages(fs_info, pages, num_pages, pos, copied);
1868 balance_dirty_pages_ratelimited(inode->i_mapping);
1871 num_written += copied;
1876 if (release_bytes) {
1877 if (only_release_metadata) {
1878 btrfs_check_nocow_unlock(BTRFS_I(inode));
1879 btrfs_delalloc_release_metadata(BTRFS_I(inode),
1880 release_bytes, true);
1882 btrfs_delalloc_release_space(BTRFS_I(inode),
1884 round_down(pos, fs_info->sectorsize),
1885 release_bytes, true);
1889 extent_changeset_free(data_reserved);
1890 if (num_written > 0) {
1891 pagecache_isize_extended(inode, old_isize, iocb->ki_pos);
1892 iocb->ki_pos += num_written;
1895 btrfs_inode_unlock(inode, ilock_flags);
1896 return num_written ? num_written : ret;
1899 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
1900 const struct iov_iter *iter, loff_t offset)
1902 const u32 blocksize_mask = fs_info->sectorsize - 1;
1904 if (offset & blocksize_mask)
1907 if (iov_iter_alignment(iter) & blocksize_mask)
1913 static ssize_t btrfs_direct_write(struct kiocb *iocb, struct iov_iter *from)
1915 struct file *file = iocb->ki_filp;
1916 struct inode *inode = file_inode(file);
1917 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1919 ssize_t written = 0;
1920 ssize_t written_buffered;
1923 unsigned int ilock_flags = 0;
1924 struct iomap_dio *dio = NULL;
1926 if (iocb->ki_flags & IOCB_NOWAIT)
1927 ilock_flags |= BTRFS_ILOCK_TRY;
1929 /* If the write DIO is within EOF, use a shared lock */
1930 if (iocb->ki_pos + iov_iter_count(from) <= i_size_read(inode))
1931 ilock_flags |= BTRFS_ILOCK_SHARED;
1934 err = btrfs_inode_lock(inode, ilock_flags);
1938 err = generic_write_checks(iocb, from);
1940 btrfs_inode_unlock(inode, ilock_flags);
1944 err = btrfs_write_check(iocb, from, err);
1946 btrfs_inode_unlock(inode, ilock_flags);
1952 * Re-check since file size may have changed just before taking the
1953 * lock or pos may have changed because of O_APPEND in generic_write_check()
1955 if ((ilock_flags & BTRFS_ILOCK_SHARED) &&
1956 pos + iov_iter_count(from) > i_size_read(inode)) {
1957 btrfs_inode_unlock(inode, ilock_flags);
1958 ilock_flags &= ~BTRFS_ILOCK_SHARED;
1962 if (check_direct_IO(fs_info, from, pos)) {
1963 btrfs_inode_unlock(inode, ilock_flags);
1967 dio = __iomap_dio_rw(iocb, from, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
1970 btrfs_inode_unlock(inode, ilock_flags);
1972 if (IS_ERR_OR_NULL(dio)) {
1973 err = PTR_ERR_OR_ZERO(dio);
1974 if (err < 0 && err != -ENOTBLK)
1977 written = iomap_dio_complete(dio);
1980 if (written < 0 || !iov_iter_count(from)) {
1987 written_buffered = btrfs_buffered_write(iocb, from);
1988 if (written_buffered < 0) {
1989 err = written_buffered;
1993 * Ensure all data is persisted. We want the next direct IO read to be
1994 * able to read what was just written.
1996 endbyte = pos + written_buffered - 1;
1997 err = btrfs_fdatawrite_range(inode, pos, endbyte);
2000 err = filemap_fdatawait_range(inode->i_mapping, pos, endbyte);
2003 written += written_buffered;
2004 iocb->ki_pos = pos + written_buffered;
2005 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_SHIFT,
2006 endbyte >> PAGE_SHIFT);
2008 return written ? written : err;
2011 static ssize_t btrfs_file_write_iter(struct kiocb *iocb,
2012 struct iov_iter *from)
2014 struct file *file = iocb->ki_filp;
2015 struct btrfs_inode *inode = BTRFS_I(file_inode(file));
2016 ssize_t num_written = 0;
2017 const bool sync = iocb->ki_flags & IOCB_DSYNC;
2020 * If the fs flips readonly due to some impossible error, although we
2021 * have opened a file as writable, we have to stop this write operation
2022 * to ensure consistency.
2024 if (BTRFS_FS_ERROR(inode->root->fs_info))
2027 if (!(iocb->ki_flags & IOCB_DIRECT) &&
2028 (iocb->ki_flags & IOCB_NOWAIT))
2032 atomic_inc(&inode->sync_writers);
2034 if (iocb->ki_flags & IOCB_DIRECT)
2035 num_written = btrfs_direct_write(iocb, from);
2037 num_written = btrfs_buffered_write(iocb, from);
2039 btrfs_set_inode_last_sub_trans(inode);
2041 if (num_written > 0)
2042 num_written = generic_write_sync(iocb, num_written);
2045 atomic_dec(&inode->sync_writers);
2047 current->backing_dev_info = NULL;
2051 int btrfs_release_file(struct inode *inode, struct file *filp)
2053 struct btrfs_file_private *private = filp->private_data;
2055 if (private && private->filldir_buf)
2056 kfree(private->filldir_buf);
2058 filp->private_data = NULL;
2061 * Set by setattr when we are about to truncate a file from a non-zero
2062 * size to a zero size. This tries to flush down new bytes that may
2063 * have been written if the application were using truncate to replace
2066 if (test_and_clear_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
2067 &BTRFS_I(inode)->runtime_flags))
2068 filemap_flush(inode->i_mapping);
2072 static int start_ordered_ops(struct inode *inode, loff_t start, loff_t end)
2075 struct blk_plug plug;
2078 * This is only called in fsync, which would do synchronous writes, so
2079 * a plug can merge adjacent IOs as much as possible. Esp. in case of
2080 * multiple disks using raid profile, a large IO can be split to
2081 * several segments of stripe length (currently 64K).
2083 blk_start_plug(&plug);
2084 atomic_inc(&BTRFS_I(inode)->sync_writers);
2085 ret = btrfs_fdatawrite_range(inode, start, end);
2086 atomic_dec(&BTRFS_I(inode)->sync_writers);
2087 blk_finish_plug(&plug);
2092 static inline bool skip_inode_logging(const struct btrfs_log_ctx *ctx)
2094 struct btrfs_inode *inode = BTRFS_I(ctx->inode);
2095 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2097 if (btrfs_inode_in_log(inode, fs_info->generation) &&
2098 list_empty(&ctx->ordered_extents))
2102 * If we are doing a fast fsync we can not bail out if the inode's
2103 * last_trans is <= then the last committed transaction, because we only
2104 * update the last_trans of the inode during ordered extent completion,
2105 * and for a fast fsync we don't wait for that, we only wait for the
2106 * writeback to complete.
2108 if (inode->last_trans <= fs_info->last_trans_committed &&
2109 (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags) ||
2110 list_empty(&ctx->ordered_extents)))
2117 * fsync call for both files and directories. This logs the inode into
2118 * the tree log instead of forcing full commits whenever possible.
2120 * It needs to call filemap_fdatawait so that all ordered extent updates are
2121 * in the metadata btree are up to date for copying to the log.
2123 * It drops the inode mutex before doing the tree log commit. This is an
2124 * important optimization for directories because holding the mutex prevents
2125 * new operations on the dir while we write to disk.
2127 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
2129 struct dentry *dentry = file_dentry(file);
2130 struct inode *inode = d_inode(dentry);
2131 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2132 struct btrfs_root *root = BTRFS_I(inode)->root;
2133 struct btrfs_trans_handle *trans;
2134 struct btrfs_log_ctx ctx;
2139 trace_btrfs_sync_file(file, datasync);
2141 btrfs_init_log_ctx(&ctx, inode);
2144 * Always set the range to a full range, otherwise we can get into
2145 * several problems, from missing file extent items to represent holes
2146 * when not using the NO_HOLES feature, to log tree corruption due to
2147 * races between hole detection during logging and completion of ordered
2148 * extents outside the range, to missing checksums due to ordered extents
2149 * for which we flushed only a subset of their pages.
2153 len = (u64)LLONG_MAX + 1;
2156 * We write the dirty pages in the range and wait until they complete
2157 * out of the ->i_mutex. If so, we can flush the dirty pages by
2158 * multi-task, and make the performance up. See
2159 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
2161 ret = start_ordered_ops(inode, start, end);
2165 btrfs_inode_lock(inode, BTRFS_ILOCK_MMAP);
2167 atomic_inc(&root->log_batch);
2170 * Always check for the full sync flag while holding the inode's lock,
2171 * to avoid races with other tasks. The flag must be either set all the
2172 * time during logging or always off all the time while logging.
2174 full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2175 &BTRFS_I(inode)->runtime_flags);
2178 * Before we acquired the inode's lock and the mmap lock, someone may
2179 * have dirtied more pages in the target range. We need to make sure
2180 * that writeback for any such pages does not start while we are logging
2181 * the inode, because if it does, any of the following might happen when
2182 * we are not doing a full inode sync:
2184 * 1) We log an extent after its writeback finishes but before its
2185 * checksums are added to the csum tree, leading to -EIO errors
2186 * when attempting to read the extent after a log replay.
2188 * 2) We can end up logging an extent before its writeback finishes.
2189 * Therefore after the log replay we will have a file extent item
2190 * pointing to an unwritten extent (and no data checksums as well).
2192 * So trigger writeback for any eventual new dirty pages and then we
2193 * wait for all ordered extents to complete below.
2195 ret = start_ordered_ops(inode, start, end);
2197 btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP);
2202 * We have to do this here to avoid the priority inversion of waiting on
2203 * IO of a lower priority task while holding a transaction open.
2205 * For a full fsync we wait for the ordered extents to complete while
2206 * for a fast fsync we wait just for writeback to complete, and then
2207 * attach the ordered extents to the transaction so that a transaction
2208 * commit waits for their completion, to avoid data loss if we fsync,
2209 * the current transaction commits before the ordered extents complete
2210 * and a power failure happens right after that.
2212 * For zoned filesystem, if a write IO uses a ZONE_APPEND command, the
2213 * logical address recorded in the ordered extent may change. We need
2214 * to wait for the IO to stabilize the logical address.
2216 if (full_sync || btrfs_is_zoned(fs_info)) {
2217 ret = btrfs_wait_ordered_range(inode, start, len);
2220 * Get our ordered extents as soon as possible to avoid doing
2221 * checksum lookups in the csum tree, and use instead the
2222 * checksums attached to the ordered extents.
2224 btrfs_get_ordered_extents_for_logging(BTRFS_I(inode),
2225 &ctx.ordered_extents);
2226 ret = filemap_fdatawait_range(inode->i_mapping, start, end);
2230 goto out_release_extents;
2232 atomic_inc(&root->log_batch);
2235 if (skip_inode_logging(&ctx)) {
2237 * We've had everything committed since the last time we were
2238 * modified so clear this flag in case it was set for whatever
2239 * reason, it's no longer relevant.
2241 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2242 &BTRFS_I(inode)->runtime_flags);
2244 * An ordered extent might have started before and completed
2245 * already with io errors, in which case the inode was not
2246 * updated and we end up here. So check the inode's mapping
2247 * for any errors that might have happened since we last
2248 * checked called fsync.
2250 ret = filemap_check_wb_err(inode->i_mapping, file->f_wb_err);
2251 goto out_release_extents;
2255 * We use start here because we will need to wait on the IO to complete
2256 * in btrfs_sync_log, which could require joining a transaction (for
2257 * example checking cross references in the nocow path). If we use join
2258 * here we could get into a situation where we're waiting on IO to
2259 * happen that is blocked on a transaction trying to commit. With start
2260 * we inc the extwriter counter, so we wait for all extwriters to exit
2261 * before we start blocking joiners. This comment is to keep somebody
2262 * from thinking they are super smart and changing this to
2263 * btrfs_join_transaction *cough*Josef*cough*.
2265 trans = btrfs_start_transaction(root, 0);
2266 if (IS_ERR(trans)) {
2267 ret = PTR_ERR(trans);
2268 goto out_release_extents;
2270 trans->in_fsync = true;
2272 ret = btrfs_log_dentry_safe(trans, dentry, &ctx);
2273 btrfs_release_log_ctx_extents(&ctx);
2275 /* Fallthrough and commit/free transaction. */
2279 /* we've logged all the items and now have a consistent
2280 * version of the file in the log. It is possible that
2281 * someone will come in and modify the file, but that's
2282 * fine because the log is consistent on disk, and we
2283 * have references to all of the file's extents
2285 * It is possible that someone will come in and log the
2286 * file again, but that will end up using the synchronization
2287 * inside btrfs_sync_log to keep things safe.
2289 btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP);
2291 if (ret != BTRFS_NO_LOG_SYNC) {
2293 ret = btrfs_sync_log(trans, root, &ctx);
2295 ret = btrfs_end_transaction(trans);
2300 ret = btrfs_wait_ordered_range(inode, start, len);
2302 btrfs_end_transaction(trans);
2306 ret = btrfs_commit_transaction(trans);
2308 ret = btrfs_end_transaction(trans);
2311 ASSERT(list_empty(&ctx.list));
2312 err = file_check_and_advance_wb_err(file);
2315 return ret > 0 ? -EIO : ret;
2317 out_release_extents:
2318 btrfs_release_log_ctx_extents(&ctx);
2319 btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP);
2323 static const struct vm_operations_struct btrfs_file_vm_ops = {
2324 .fault = filemap_fault,
2325 .map_pages = filemap_map_pages,
2326 .page_mkwrite = btrfs_page_mkwrite,
2329 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
2331 struct address_space *mapping = filp->f_mapping;
2333 if (!mapping->a_ops->readpage)
2336 file_accessed(filp);
2337 vma->vm_ops = &btrfs_file_vm_ops;
2342 static int hole_mergeable(struct btrfs_inode *inode, struct extent_buffer *leaf,
2343 int slot, u64 start, u64 end)
2345 struct btrfs_file_extent_item *fi;
2346 struct btrfs_key key;
2348 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
2351 btrfs_item_key_to_cpu(leaf, &key, slot);
2352 if (key.objectid != btrfs_ino(inode) ||
2353 key.type != BTRFS_EXTENT_DATA_KEY)
2356 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2358 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2361 if (btrfs_file_extent_disk_bytenr(leaf, fi))
2364 if (key.offset == end)
2366 if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
2371 static int fill_holes(struct btrfs_trans_handle *trans,
2372 struct btrfs_inode *inode,
2373 struct btrfs_path *path, u64 offset, u64 end)
2375 struct btrfs_fs_info *fs_info = trans->fs_info;
2376 struct btrfs_root *root = inode->root;
2377 struct extent_buffer *leaf;
2378 struct btrfs_file_extent_item *fi;
2379 struct extent_map *hole_em;
2380 struct extent_map_tree *em_tree = &inode->extent_tree;
2381 struct btrfs_key key;
2384 if (btrfs_fs_incompat(fs_info, NO_HOLES))
2387 key.objectid = btrfs_ino(inode);
2388 key.type = BTRFS_EXTENT_DATA_KEY;
2389 key.offset = offset;
2391 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2394 * We should have dropped this offset, so if we find it then
2395 * something has gone horribly wrong.
2402 leaf = path->nodes[0];
2403 if (hole_mergeable(inode, leaf, path->slots[0] - 1, offset, end)) {
2407 fi = btrfs_item_ptr(leaf, path->slots[0],
2408 struct btrfs_file_extent_item);
2409 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2411 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2412 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2413 btrfs_set_file_extent_offset(leaf, fi, 0);
2414 btrfs_mark_buffer_dirty(leaf);
2418 if (hole_mergeable(inode, leaf, path->slots[0], offset, end)) {
2421 key.offset = offset;
2422 btrfs_set_item_key_safe(fs_info, path, &key);
2423 fi = btrfs_item_ptr(leaf, path->slots[0],
2424 struct btrfs_file_extent_item);
2425 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2427 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2428 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2429 btrfs_set_file_extent_offset(leaf, fi, 0);
2430 btrfs_mark_buffer_dirty(leaf);
2433 btrfs_release_path(path);
2435 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode),
2436 offset, 0, 0, end - offset, 0, end - offset, 0, 0, 0);
2441 btrfs_release_path(path);
2443 hole_em = alloc_extent_map();
2445 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2446 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
2448 hole_em->start = offset;
2449 hole_em->len = end - offset;
2450 hole_em->ram_bytes = hole_em->len;
2451 hole_em->orig_start = offset;
2453 hole_em->block_start = EXTENT_MAP_HOLE;
2454 hole_em->block_len = 0;
2455 hole_em->orig_block_len = 0;
2456 hole_em->compress_type = BTRFS_COMPRESS_NONE;
2457 hole_em->generation = trans->transid;
2460 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2461 write_lock(&em_tree->lock);
2462 ret = add_extent_mapping(em_tree, hole_em, 1);
2463 write_unlock(&em_tree->lock);
2464 } while (ret == -EEXIST);
2465 free_extent_map(hole_em);
2467 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2468 &inode->runtime_flags);
2475 * Find a hole extent on given inode and change start/len to the end of hole
2476 * extent.(hole/vacuum extent whose em->start <= start &&
2477 * em->start + em->len > start)
2478 * When a hole extent is found, return 1 and modify start/len.
2480 static int find_first_non_hole(struct btrfs_inode *inode, u64 *start, u64 *len)
2482 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2483 struct extent_map *em;
2486 em = btrfs_get_extent(inode, NULL, 0,
2487 round_down(*start, fs_info->sectorsize),
2488 round_up(*len, fs_info->sectorsize));
2492 /* Hole or vacuum extent(only exists in no-hole mode) */
2493 if (em->block_start == EXTENT_MAP_HOLE) {
2495 *len = em->start + em->len > *start + *len ?
2496 0 : *start + *len - em->start - em->len;
2497 *start = em->start + em->len;
2499 free_extent_map(em);
2503 static int btrfs_punch_hole_lock_range(struct inode *inode,
2504 const u64 lockstart,
2506 struct extent_state **cached_state)
2509 * For subpage case, if the range is not at page boundary, we could
2510 * have pages at the leading/tailing part of the range.
2511 * This could lead to dead loop since filemap_range_has_page()
2512 * will always return true.
2513 * So here we need to do extra page alignment for
2514 * filemap_range_has_page().
2516 const u64 page_lockstart = round_up(lockstart, PAGE_SIZE);
2517 const u64 page_lockend = round_down(lockend + 1, PAGE_SIZE) - 1;
2520 struct btrfs_ordered_extent *ordered;
2523 truncate_pagecache_range(inode, lockstart, lockend);
2525 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2527 ordered = btrfs_lookup_first_ordered_extent(BTRFS_I(inode),
2531 * We need to make sure we have no ordered extents in this range
2532 * and nobody raced in and read a page in this range, if we did
2533 * we need to try again.
2536 (ordered->file_offset + ordered->num_bytes <= lockstart ||
2537 ordered->file_offset > lockend)) &&
2538 !filemap_range_has_page(inode->i_mapping,
2539 page_lockstart, page_lockend)) {
2541 btrfs_put_ordered_extent(ordered);
2545 btrfs_put_ordered_extent(ordered);
2546 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
2547 lockend, cached_state);
2548 ret = btrfs_wait_ordered_range(inode, lockstart,
2549 lockend - lockstart + 1);
2556 static int btrfs_insert_replace_extent(struct btrfs_trans_handle *trans,
2557 struct btrfs_inode *inode,
2558 struct btrfs_path *path,
2559 struct btrfs_replace_extent_info *extent_info,
2560 const u64 replace_len,
2561 const u64 bytes_to_drop)
2563 struct btrfs_fs_info *fs_info = trans->fs_info;
2564 struct btrfs_root *root = inode->root;
2565 struct btrfs_file_extent_item *extent;
2566 struct extent_buffer *leaf;
2567 struct btrfs_key key;
2569 struct btrfs_ref ref = { 0 };
2572 if (replace_len == 0)
2575 if (extent_info->disk_offset == 0 &&
2576 btrfs_fs_incompat(fs_info, NO_HOLES)) {
2577 btrfs_update_inode_bytes(inode, 0, bytes_to_drop);
2581 key.objectid = btrfs_ino(inode);
2582 key.type = BTRFS_EXTENT_DATA_KEY;
2583 key.offset = extent_info->file_offset;
2584 ret = btrfs_insert_empty_item(trans, root, path, &key,
2585 sizeof(struct btrfs_file_extent_item));
2588 leaf = path->nodes[0];
2589 slot = path->slots[0];
2590 write_extent_buffer(leaf, extent_info->extent_buf,
2591 btrfs_item_ptr_offset(leaf, slot),
2592 sizeof(struct btrfs_file_extent_item));
2593 extent = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2594 ASSERT(btrfs_file_extent_type(leaf, extent) != BTRFS_FILE_EXTENT_INLINE);
2595 btrfs_set_file_extent_offset(leaf, extent, extent_info->data_offset);
2596 btrfs_set_file_extent_num_bytes(leaf, extent, replace_len);
2597 if (extent_info->is_new_extent)
2598 btrfs_set_file_extent_generation(leaf, extent, trans->transid);
2599 btrfs_mark_buffer_dirty(leaf);
2600 btrfs_release_path(path);
2602 ret = btrfs_inode_set_file_extent_range(inode, extent_info->file_offset,
2607 /* If it's a hole, nothing more needs to be done. */
2608 if (extent_info->disk_offset == 0) {
2609 btrfs_update_inode_bytes(inode, 0, bytes_to_drop);
2613 btrfs_update_inode_bytes(inode, replace_len, bytes_to_drop);
2615 if (extent_info->is_new_extent && extent_info->insertions == 0) {
2616 key.objectid = extent_info->disk_offset;
2617 key.type = BTRFS_EXTENT_ITEM_KEY;
2618 key.offset = extent_info->disk_len;
2619 ret = btrfs_alloc_reserved_file_extent(trans, root,
2621 extent_info->file_offset,
2622 extent_info->qgroup_reserved,
2627 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF,
2628 extent_info->disk_offset,
2629 extent_info->disk_len, 0);
2630 ref_offset = extent_info->file_offset - extent_info->data_offset;
2631 btrfs_init_data_ref(&ref, root->root_key.objectid,
2632 btrfs_ino(inode), ref_offset, 0, false);
2633 ret = btrfs_inc_extent_ref(trans, &ref);
2636 extent_info->insertions++;
2642 * The respective range must have been previously locked, as well as the inode.
2643 * The end offset is inclusive (last byte of the range).
2644 * @extent_info is NULL for fallocate's hole punching and non-NULL when replacing
2645 * the file range with an extent.
2646 * When not punching a hole, we don't want to end up in a state where we dropped
2647 * extents without inserting a new one, so we must abort the transaction to avoid
2650 int btrfs_replace_file_extents(struct btrfs_inode *inode,
2651 struct btrfs_path *path, const u64 start,
2653 struct btrfs_replace_extent_info *extent_info,
2654 struct btrfs_trans_handle **trans_out)
2656 struct btrfs_drop_extents_args drop_args = { 0 };
2657 struct btrfs_root *root = inode->root;
2658 struct btrfs_fs_info *fs_info = root->fs_info;
2659 u64 min_size = btrfs_calc_insert_metadata_size(fs_info, 1);
2660 u64 ino_size = round_up(inode->vfs_inode.i_size, fs_info->sectorsize);
2661 struct btrfs_trans_handle *trans = NULL;
2662 struct btrfs_block_rsv *rsv;
2663 unsigned int rsv_count;
2665 u64 len = end - start;
2671 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
2676 rsv->size = btrfs_calc_insert_metadata_size(fs_info, 1);
2680 * 1 - update the inode
2681 * 1 - removing the extents in the range
2682 * 1 - adding the hole extent if no_holes isn't set or if we are
2683 * replacing the range with a new extent
2685 if (!btrfs_fs_incompat(fs_info, NO_HOLES) || extent_info)
2690 trans = btrfs_start_transaction(root, rsv_count);
2691 if (IS_ERR(trans)) {
2692 ret = PTR_ERR(trans);
2697 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
2700 trans->block_rsv = rsv;
2703 drop_args.path = path;
2704 drop_args.end = end + 1;
2705 drop_args.drop_cache = true;
2706 while (cur_offset < end) {
2707 drop_args.start = cur_offset;
2708 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2709 /* If we are punching a hole decrement the inode's byte count */
2711 btrfs_update_inode_bytes(inode, 0,
2712 drop_args.bytes_found);
2713 if (ret != -ENOSPC) {
2715 * The only time we don't want to abort is if we are
2716 * attempting to clone a partial inline extent, in which
2717 * case we'll get EOPNOTSUPP. However if we aren't
2718 * clone we need to abort no matter what, because if we
2719 * got EOPNOTSUPP via prealloc then we messed up and
2723 (ret != -EOPNOTSUPP ||
2724 (extent_info && extent_info->is_new_extent)))
2725 btrfs_abort_transaction(trans, ret);
2729 trans->block_rsv = &fs_info->trans_block_rsv;
2731 if (!extent_info && cur_offset < drop_args.drop_end &&
2732 cur_offset < ino_size) {
2733 ret = fill_holes(trans, inode, path, cur_offset,
2734 drop_args.drop_end);
2737 * If we failed then we didn't insert our hole
2738 * entries for the area we dropped, so now the
2739 * fs is corrupted, so we must abort the
2742 btrfs_abort_transaction(trans, ret);
2745 } else if (!extent_info && cur_offset < drop_args.drop_end) {
2747 * We are past the i_size here, but since we didn't
2748 * insert holes we need to clear the mapped area so we
2749 * know to not set disk_i_size in this area until a new
2750 * file extent is inserted here.
2752 ret = btrfs_inode_clear_file_extent_range(inode,
2754 drop_args.drop_end - cur_offset);
2757 * We couldn't clear our area, so we could
2758 * presumably adjust up and corrupt the fs, so
2761 btrfs_abort_transaction(trans, ret);
2767 drop_args.drop_end > extent_info->file_offset) {
2768 u64 replace_len = drop_args.drop_end -
2769 extent_info->file_offset;
2771 ret = btrfs_insert_replace_extent(trans, inode, path,
2772 extent_info, replace_len,
2773 drop_args.bytes_found);
2775 btrfs_abort_transaction(trans, ret);
2778 extent_info->data_len -= replace_len;
2779 extent_info->data_offset += replace_len;
2780 extent_info->file_offset += replace_len;
2783 ret = btrfs_update_inode(trans, root, inode);
2787 btrfs_end_transaction(trans);
2788 btrfs_btree_balance_dirty(fs_info);
2790 trans = btrfs_start_transaction(root, rsv_count);
2791 if (IS_ERR(trans)) {
2792 ret = PTR_ERR(trans);
2797 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
2798 rsv, min_size, false);
2799 BUG_ON(ret); /* shouldn't happen */
2800 trans->block_rsv = rsv;
2802 cur_offset = drop_args.drop_end;
2803 len = end - cur_offset;
2804 if (!extent_info && len) {
2805 ret = find_first_non_hole(inode, &cur_offset, &len);
2806 if (unlikely(ret < 0))
2816 * If we were cloning, force the next fsync to be a full one since we
2817 * we replaced (or just dropped in the case of cloning holes when
2818 * NO_HOLES is enabled) file extent items and did not setup new extent
2819 * maps for the replacement extents (or holes).
2821 if (extent_info && !extent_info->is_new_extent)
2822 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
2827 trans->block_rsv = &fs_info->trans_block_rsv;
2829 * If we are using the NO_HOLES feature we might have had already an
2830 * hole that overlaps a part of the region [lockstart, lockend] and
2831 * ends at (or beyond) lockend. Since we have no file extent items to
2832 * represent holes, drop_end can be less than lockend and so we must
2833 * make sure we have an extent map representing the existing hole (the
2834 * call to __btrfs_drop_extents() might have dropped the existing extent
2835 * map representing the existing hole), otherwise the fast fsync path
2836 * will not record the existence of the hole region
2837 * [existing_hole_start, lockend].
2839 if (drop_args.drop_end <= end)
2840 drop_args.drop_end = end + 1;
2842 * Don't insert file hole extent item if it's for a range beyond eof
2843 * (because it's useless) or if it represents a 0 bytes range (when
2844 * cur_offset == drop_end).
2846 if (!extent_info && cur_offset < ino_size &&
2847 cur_offset < drop_args.drop_end) {
2848 ret = fill_holes(trans, inode, path, cur_offset,
2849 drop_args.drop_end);
2851 /* Same comment as above. */
2852 btrfs_abort_transaction(trans, ret);
2855 } else if (!extent_info && cur_offset < drop_args.drop_end) {
2856 /* See the comment in the loop above for the reasoning here. */
2857 ret = btrfs_inode_clear_file_extent_range(inode, cur_offset,
2858 drop_args.drop_end - cur_offset);
2860 btrfs_abort_transaction(trans, ret);
2866 ret = btrfs_insert_replace_extent(trans, inode, path,
2867 extent_info, extent_info->data_len,
2868 drop_args.bytes_found);
2870 btrfs_abort_transaction(trans, ret);
2879 trans->block_rsv = &fs_info->trans_block_rsv;
2881 btrfs_end_transaction(trans);
2885 btrfs_free_block_rsv(fs_info, rsv);
2890 static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
2892 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2893 struct btrfs_root *root = BTRFS_I(inode)->root;
2894 struct extent_state *cached_state = NULL;
2895 struct btrfs_path *path;
2896 struct btrfs_trans_handle *trans = NULL;
2901 u64 orig_start = offset;
2905 bool truncated_block = false;
2906 bool updated_inode = false;
2908 ret = btrfs_wait_ordered_range(inode, offset, len);
2912 btrfs_inode_lock(inode, BTRFS_ILOCK_MMAP);
2913 ino_size = round_up(inode->i_size, fs_info->sectorsize);
2914 ret = find_first_non_hole(BTRFS_I(inode), &offset, &len);
2916 goto out_only_mutex;
2918 /* Already in a large hole */
2920 goto out_only_mutex;
2923 lockstart = round_up(offset, btrfs_inode_sectorsize(BTRFS_I(inode)));
2924 lockend = round_down(offset + len,
2925 btrfs_inode_sectorsize(BTRFS_I(inode))) - 1;
2926 same_block = (BTRFS_BYTES_TO_BLKS(fs_info, offset))
2927 == (BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1));
2929 * We needn't truncate any block which is beyond the end of the file
2930 * because we are sure there is no data there.
2933 * Only do this if we are in the same block and we aren't doing the
2936 if (same_block && len < fs_info->sectorsize) {
2937 if (offset < ino_size) {
2938 truncated_block = true;
2939 ret = btrfs_truncate_block(BTRFS_I(inode), offset, len,
2944 goto out_only_mutex;
2947 /* zero back part of the first block */
2948 if (offset < ino_size) {
2949 truncated_block = true;
2950 ret = btrfs_truncate_block(BTRFS_I(inode), offset, 0, 0);
2952 btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP);
2957 /* Check the aligned pages after the first unaligned page,
2958 * if offset != orig_start, which means the first unaligned page
2959 * including several following pages are already in holes,
2960 * the extra check can be skipped */
2961 if (offset == orig_start) {
2962 /* after truncate page, check hole again */
2963 len = offset + len - lockstart;
2965 ret = find_first_non_hole(BTRFS_I(inode), &offset, &len);
2967 goto out_only_mutex;
2970 goto out_only_mutex;
2975 /* Check the tail unaligned part is in a hole */
2976 tail_start = lockend + 1;
2977 tail_len = offset + len - tail_start;
2979 ret = find_first_non_hole(BTRFS_I(inode), &tail_start, &tail_len);
2980 if (unlikely(ret < 0))
2981 goto out_only_mutex;
2983 /* zero the front end of the last page */
2984 if (tail_start + tail_len < ino_size) {
2985 truncated_block = true;
2986 ret = btrfs_truncate_block(BTRFS_I(inode),
2987 tail_start + tail_len,
2990 goto out_only_mutex;
2995 if (lockend < lockstart) {
2997 goto out_only_mutex;
3000 ret = btrfs_punch_hole_lock_range(inode, lockstart, lockend,
3003 goto out_only_mutex;
3005 path = btrfs_alloc_path();
3011 ret = btrfs_replace_file_extents(BTRFS_I(inode), path, lockstart,
3012 lockend, NULL, &trans);
3013 btrfs_free_path(path);
3017 ASSERT(trans != NULL);
3018 inode_inc_iversion(inode);
3019 inode->i_mtime = inode->i_ctime = current_time(inode);
3020 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
3021 updated_inode = true;
3022 btrfs_end_transaction(trans);
3023 btrfs_btree_balance_dirty(fs_info);
3025 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
3028 if (!updated_inode && truncated_block && !ret) {
3030 * If we only end up zeroing part of a page, we still need to
3031 * update the inode item, so that all the time fields are
3032 * updated as well as the necessary btrfs inode in memory fields
3033 * for detecting, at fsync time, if the inode isn't yet in the
3034 * log tree or it's there but not up to date.
3036 struct timespec64 now = current_time(inode);
3038 inode_inc_iversion(inode);
3039 inode->i_mtime = now;
3040 inode->i_ctime = now;
3041 trans = btrfs_start_transaction(root, 1);
3042 if (IS_ERR(trans)) {
3043 ret = PTR_ERR(trans);
3047 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
3048 ret2 = btrfs_end_transaction(trans);
3053 btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP);
3057 /* Helper structure to record which range is already reserved */
3058 struct falloc_range {
3059 struct list_head list;
3065 * Helper function to add falloc range
3067 * Caller should have locked the larger range of extent containing
3070 static int add_falloc_range(struct list_head *head, u64 start, u64 len)
3072 struct falloc_range *range = NULL;
3074 if (!list_empty(head)) {
3076 * As fallocate iterates by bytenr order, we only need to check
3079 range = list_last_entry(head, struct falloc_range, list);
3080 if (range->start + range->len == start) {
3086 range = kmalloc(sizeof(*range), GFP_KERNEL);
3089 range->start = start;
3091 list_add_tail(&range->list, head);
3095 static int btrfs_fallocate_update_isize(struct inode *inode,
3099 struct btrfs_trans_handle *trans;
3100 struct btrfs_root *root = BTRFS_I(inode)->root;
3104 if (mode & FALLOC_FL_KEEP_SIZE || end <= i_size_read(inode))
3107 trans = btrfs_start_transaction(root, 1);
3109 return PTR_ERR(trans);
3111 inode->i_ctime = current_time(inode);
3112 i_size_write(inode, end);
3113 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
3114 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
3115 ret2 = btrfs_end_transaction(trans);
3117 return ret ? ret : ret2;
3121 RANGE_BOUNDARY_WRITTEN_EXTENT,
3122 RANGE_BOUNDARY_PREALLOC_EXTENT,
3123 RANGE_BOUNDARY_HOLE,
3126 static int btrfs_zero_range_check_range_boundary(struct btrfs_inode *inode,
3129 const u64 sectorsize = btrfs_inode_sectorsize(inode);
3130 struct extent_map *em;
3133 offset = round_down(offset, sectorsize);
3134 em = btrfs_get_extent(inode, NULL, 0, offset, sectorsize);
3138 if (em->block_start == EXTENT_MAP_HOLE)
3139 ret = RANGE_BOUNDARY_HOLE;
3140 else if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
3141 ret = RANGE_BOUNDARY_PREALLOC_EXTENT;
3143 ret = RANGE_BOUNDARY_WRITTEN_EXTENT;
3145 free_extent_map(em);
3149 static int btrfs_zero_range(struct inode *inode,
3154 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3155 struct extent_map *em;
3156 struct extent_changeset *data_reserved = NULL;
3159 const u64 sectorsize = btrfs_inode_sectorsize(BTRFS_I(inode));
3160 u64 alloc_start = round_down(offset, sectorsize);
3161 u64 alloc_end = round_up(offset + len, sectorsize);
3162 u64 bytes_to_reserve = 0;
3163 bool space_reserved = false;
3165 inode_dio_wait(inode);
3167 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, alloc_start,
3168 alloc_end - alloc_start);
3175 * Avoid hole punching and extent allocation for some cases. More cases
3176 * could be considered, but these are unlikely common and we keep things
3177 * as simple as possible for now. Also, intentionally, if the target
3178 * range contains one or more prealloc extents together with regular
3179 * extents and holes, we drop all the existing extents and allocate a
3180 * new prealloc extent, so that we get a larger contiguous disk extent.
3182 if (em->start <= alloc_start &&
3183 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3184 const u64 em_end = em->start + em->len;
3186 if (em_end >= offset + len) {
3188 * The whole range is already a prealloc extent,
3189 * do nothing except updating the inode's i_size if
3192 free_extent_map(em);
3193 ret = btrfs_fallocate_update_isize(inode, offset + len,
3198 * Part of the range is already a prealloc extent, so operate
3199 * only on the remaining part of the range.
3201 alloc_start = em_end;
3202 ASSERT(IS_ALIGNED(alloc_start, sectorsize));
3203 len = offset + len - alloc_start;
3204 offset = alloc_start;
3205 alloc_hint = em->block_start + em->len;
3207 free_extent_map(em);
3209 if (BTRFS_BYTES_TO_BLKS(fs_info, offset) ==
3210 BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1)) {
3211 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, alloc_start,
3218 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3219 free_extent_map(em);
3220 ret = btrfs_fallocate_update_isize(inode, offset + len,
3224 if (len < sectorsize && em->block_start != EXTENT_MAP_HOLE) {
3225 free_extent_map(em);
3226 ret = btrfs_truncate_block(BTRFS_I(inode), offset, len,
3229 ret = btrfs_fallocate_update_isize(inode,
3234 free_extent_map(em);
3235 alloc_start = round_down(offset, sectorsize);
3236 alloc_end = alloc_start + sectorsize;
3240 alloc_start = round_up(offset, sectorsize);
3241 alloc_end = round_down(offset + len, sectorsize);
3244 * For unaligned ranges, check the pages at the boundaries, they might
3245 * map to an extent, in which case we need to partially zero them, or
3246 * they might map to a hole, in which case we need our allocation range
3249 if (!IS_ALIGNED(offset, sectorsize)) {
3250 ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode),
3254 if (ret == RANGE_BOUNDARY_HOLE) {
3255 alloc_start = round_down(offset, sectorsize);
3257 } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) {
3258 ret = btrfs_truncate_block(BTRFS_I(inode), offset, 0, 0);
3266 if (!IS_ALIGNED(offset + len, sectorsize)) {
3267 ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode),
3271 if (ret == RANGE_BOUNDARY_HOLE) {
3272 alloc_end = round_up(offset + len, sectorsize);
3274 } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) {
3275 ret = btrfs_truncate_block(BTRFS_I(inode), offset + len,
3285 if (alloc_start < alloc_end) {
3286 struct extent_state *cached_state = NULL;
3287 const u64 lockstart = alloc_start;
3288 const u64 lockend = alloc_end - 1;
3290 bytes_to_reserve = alloc_end - alloc_start;
3291 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode),
3295 space_reserved = true;
3296 ret = btrfs_punch_hole_lock_range(inode, lockstart, lockend,
3300 ret = btrfs_qgroup_reserve_data(BTRFS_I(inode), &data_reserved,
3301 alloc_start, bytes_to_reserve);
3303 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
3304 lockend, &cached_state);
3307 ret = btrfs_prealloc_file_range(inode, mode, alloc_start,
3308 alloc_end - alloc_start,
3310 offset + len, &alloc_hint);
3311 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
3312 lockend, &cached_state);
3313 /* btrfs_prealloc_file_range releases reserved space on error */
3315 space_reserved = false;
3319 ret = btrfs_fallocate_update_isize(inode, offset + len, mode);
3321 if (ret && space_reserved)
3322 btrfs_free_reserved_data_space(BTRFS_I(inode), data_reserved,
3323 alloc_start, bytes_to_reserve);
3324 extent_changeset_free(data_reserved);
3329 static long btrfs_fallocate(struct file *file, int mode,
3330 loff_t offset, loff_t len)
3332 struct inode *inode = file_inode(file);
3333 struct extent_state *cached_state = NULL;
3334 struct extent_changeset *data_reserved = NULL;
3335 struct falloc_range *range;
3336 struct falloc_range *tmp;
3337 struct list_head reserve_list;
3345 struct extent_map *em;
3346 int blocksize = btrfs_inode_sectorsize(BTRFS_I(inode));
3349 /* Do not allow fallocate in ZONED mode */
3350 if (btrfs_is_zoned(btrfs_sb(inode->i_sb)))
3353 alloc_start = round_down(offset, blocksize);
3354 alloc_end = round_up(offset + len, blocksize);
3355 cur_offset = alloc_start;
3357 /* Make sure we aren't being give some crap mode */
3358 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
3359 FALLOC_FL_ZERO_RANGE))
3362 if (mode & FALLOC_FL_PUNCH_HOLE)
3363 return btrfs_punch_hole(inode, offset, len);
3366 * Only trigger disk allocation, don't trigger qgroup reserve
3368 * For qgroup space, it will be checked later.
3370 if (!(mode & FALLOC_FL_ZERO_RANGE)) {
3371 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode),
3372 alloc_end - alloc_start);
3377 btrfs_inode_lock(inode, BTRFS_ILOCK_MMAP);
3379 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) {
3380 ret = inode_newsize_ok(inode, offset + len);
3386 * TODO: Move these two operations after we have checked
3387 * accurate reserved space, or fallocate can still fail but
3388 * with page truncated or size expanded.
3390 * But that's a minor problem and won't do much harm BTW.
3392 if (alloc_start > inode->i_size) {
3393 ret = btrfs_cont_expand(BTRFS_I(inode), i_size_read(inode),
3397 } else if (offset + len > inode->i_size) {
3399 * If we are fallocating from the end of the file onward we
3400 * need to zero out the end of the block if i_size lands in the
3401 * middle of a block.
3403 ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0);
3409 * wait for ordered IO before we have any locks. We'll loop again
3410 * below with the locks held.
3412 ret = btrfs_wait_ordered_range(inode, alloc_start,
3413 alloc_end - alloc_start);
3417 if (mode & FALLOC_FL_ZERO_RANGE) {
3418 ret = btrfs_zero_range(inode, offset, len, mode);
3419 btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP);
3423 locked_end = alloc_end - 1;
3425 struct btrfs_ordered_extent *ordered;
3427 /* the extent lock is ordered inside the running
3430 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
3431 locked_end, &cached_state);
3432 ordered = btrfs_lookup_first_ordered_extent(BTRFS_I(inode),
3436 ordered->file_offset + ordered->num_bytes > alloc_start &&
3437 ordered->file_offset < alloc_end) {
3438 btrfs_put_ordered_extent(ordered);
3439 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
3440 alloc_start, locked_end,
3443 * we can't wait on the range with the transaction
3444 * running or with the extent lock held
3446 ret = btrfs_wait_ordered_range(inode, alloc_start,
3447 alloc_end - alloc_start);
3452 btrfs_put_ordered_extent(ordered);
3457 /* First, check if we exceed the qgroup limit */
3458 INIT_LIST_HEAD(&reserve_list);
3459 while (cur_offset < alloc_end) {
3460 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
3461 alloc_end - cur_offset);
3466 last_byte = min(extent_map_end(em), alloc_end);
3467 actual_end = min_t(u64, extent_map_end(em), offset + len);
3468 last_byte = ALIGN(last_byte, blocksize);
3469 if (em->block_start == EXTENT_MAP_HOLE ||
3470 (cur_offset >= inode->i_size &&
3471 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
3472 ret = add_falloc_range(&reserve_list, cur_offset,
3473 last_byte - cur_offset);
3475 free_extent_map(em);
3478 ret = btrfs_qgroup_reserve_data(BTRFS_I(inode),
3479 &data_reserved, cur_offset,
3480 last_byte - cur_offset);
3482 cur_offset = last_byte;
3483 free_extent_map(em);
3488 * Do not need to reserve unwritten extent for this
3489 * range, free reserved data space first, otherwise
3490 * it'll result in false ENOSPC error.
3492 btrfs_free_reserved_data_space(BTRFS_I(inode),
3493 data_reserved, cur_offset,
3494 last_byte - cur_offset);
3496 free_extent_map(em);
3497 cur_offset = last_byte;
3501 * If ret is still 0, means we're OK to fallocate.
3502 * Or just cleanup the list and exit.
3504 list_for_each_entry_safe(range, tmp, &reserve_list, list) {
3506 ret = btrfs_prealloc_file_range(inode, mode,
3508 range->len, i_blocksize(inode),
3509 offset + len, &alloc_hint);
3511 btrfs_free_reserved_data_space(BTRFS_I(inode),
3512 data_reserved, range->start,
3514 list_del(&range->list);
3521 * We didn't need to allocate any more space, but we still extended the
3522 * size of the file so we need to update i_size and the inode item.
3524 ret = btrfs_fallocate_update_isize(inode, actual_end, mode);
3526 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
3529 btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP);
3530 /* Let go of our reservation. */
3531 if (ret != 0 && !(mode & FALLOC_FL_ZERO_RANGE))
3532 btrfs_free_reserved_data_space(BTRFS_I(inode), data_reserved,
3533 cur_offset, alloc_end - cur_offset);
3534 extent_changeset_free(data_reserved);
3538 static loff_t find_desired_extent(struct btrfs_inode *inode, loff_t offset,
3541 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3542 struct extent_map *em = NULL;
3543 struct extent_state *cached_state = NULL;
3544 loff_t i_size = inode->vfs_inode.i_size;
3551 if (i_size == 0 || offset >= i_size)
3555 * offset can be negative, in this case we start finding DATA/HOLE from
3556 * the very start of the file.
3558 start = max_t(loff_t, 0, offset);
3560 lockstart = round_down(start, fs_info->sectorsize);
3561 lockend = round_up(i_size, fs_info->sectorsize);
3562 if (lockend <= lockstart)
3563 lockend = lockstart + fs_info->sectorsize;
3565 len = lockend - lockstart + 1;
3567 lock_extent_bits(&inode->io_tree, lockstart, lockend, &cached_state);
3569 while (start < i_size) {
3570 em = btrfs_get_extent_fiemap(inode, start, len);
3577 if (whence == SEEK_HOLE &&
3578 (em->block_start == EXTENT_MAP_HOLE ||
3579 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
3581 else if (whence == SEEK_DATA &&
3582 (em->block_start != EXTENT_MAP_HOLE &&
3583 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
3586 start = em->start + em->len;
3587 free_extent_map(em);
3591 free_extent_map(em);
3592 unlock_extent_cached(&inode->io_tree, lockstart, lockend,
3597 if (whence == SEEK_DATA && start >= i_size)
3600 offset = min_t(loff_t, start, i_size);
3606 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
3608 struct inode *inode = file->f_mapping->host;
3612 return generic_file_llseek(file, offset, whence);
3615 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
3616 offset = find_desired_extent(BTRFS_I(inode), offset, whence);
3617 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
3624 return vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
3627 static int btrfs_file_open(struct inode *inode, struct file *filp)
3631 filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC;
3633 ret = fsverity_file_open(inode, filp);
3636 return generic_file_open(inode, filp);
3639 static int check_direct_read(struct btrfs_fs_info *fs_info,
3640 const struct iov_iter *iter, loff_t offset)
3645 ret = check_direct_IO(fs_info, iter, offset);
3649 if (!iter_is_iovec(iter))
3652 for (seg = 0; seg < iter->nr_segs; seg++)
3653 for (i = seg + 1; i < iter->nr_segs; i++)
3654 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
3659 static ssize_t btrfs_direct_read(struct kiocb *iocb, struct iov_iter *to)
3661 struct inode *inode = file_inode(iocb->ki_filp);
3664 if (fsverity_active(inode))
3667 if (check_direct_read(btrfs_sb(inode->i_sb), to, iocb->ki_pos))
3670 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
3671 ret = iomap_dio_rw(iocb, to, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
3673 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
3677 static ssize_t btrfs_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
3681 if (iocb->ki_flags & IOCB_DIRECT) {
3682 ret = btrfs_direct_read(iocb, to);
3683 if (ret < 0 || !iov_iter_count(to) ||
3684 iocb->ki_pos >= i_size_read(file_inode(iocb->ki_filp)))
3688 return filemap_read(iocb, to, ret);
3691 const struct file_operations btrfs_file_operations = {
3692 .llseek = btrfs_file_llseek,
3693 .read_iter = btrfs_file_read_iter,
3694 .splice_read = generic_file_splice_read,
3695 .write_iter = btrfs_file_write_iter,
3696 .splice_write = iter_file_splice_write,
3697 .mmap = btrfs_file_mmap,
3698 .open = btrfs_file_open,
3699 .release = btrfs_release_file,
3700 .fsync = btrfs_sync_file,
3701 .fallocate = btrfs_fallocate,
3702 .unlocked_ioctl = btrfs_ioctl,
3703 #ifdef CONFIG_COMPAT
3704 .compat_ioctl = btrfs_compat_ioctl,
3706 .remap_file_range = btrfs_remap_file_range,
3709 void __cold btrfs_auto_defrag_exit(void)
3711 kmem_cache_destroy(btrfs_inode_defrag_cachep);
3714 int __init btrfs_auto_defrag_init(void)
3716 btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
3717 sizeof(struct inode_defrag), 0,
3720 if (!btrfs_inode_defrag_cachep)
3726 int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end)
3731 * So with compression we will find and lock a dirty page and clear the
3732 * first one as dirty, setup an async extent, and immediately return
3733 * with the entire range locked but with nobody actually marked with
3734 * writeback. So we can't just filemap_write_and_wait_range() and
3735 * expect it to work since it will just kick off a thread to do the
3736 * actual work. So we need to call filemap_fdatawrite_range _again_
3737 * since it will wait on the page lock, which won't be unlocked until
3738 * after the pages have been marked as writeback and so we're good to go
3739 * from there. We have to do this otherwise we'll miss the ordered
3740 * extents and that results in badness. Please Josef, do not think you
3741 * know better and pull this out at some point in the future, it is
3742 * right and you are wrong.
3744 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
3745 if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
3746 &BTRFS_I(inode)->runtime_flags))
3747 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);