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>
21 #include "transaction.h"
22 #include "btrfs_inode.h"
23 #include "print-tree.h"
28 #include "compression.h"
29 #include "delalloc-space.h"
32 static struct kmem_cache *btrfs_inode_defrag_cachep;
34 * when auto defrag is enabled we
35 * queue up these defrag structs to remember which
36 * inodes need defragging passes
39 struct rb_node rb_node;
43 * transid where the defrag was added, we search for
44 * extents newer than this
51 /* last offset we were able to defrag */
54 /* if we've wrapped around back to zero once already */
58 static int __compare_inode_defrag(struct inode_defrag *defrag1,
59 struct inode_defrag *defrag2)
61 if (defrag1->root > defrag2->root)
63 else if (defrag1->root < defrag2->root)
65 else if (defrag1->ino > defrag2->ino)
67 else if (defrag1->ino < defrag2->ino)
73 /* pop a record for an inode into the defrag tree. The lock
74 * must be held already
76 * If you're inserting a record for an older transid than an
77 * existing record, the transid already in the tree is lowered
79 * If an existing record is found the defrag item you
82 static int __btrfs_add_inode_defrag(struct btrfs_inode *inode,
83 struct inode_defrag *defrag)
85 struct btrfs_fs_info *fs_info = inode->root->fs_info;
86 struct inode_defrag *entry;
88 struct rb_node *parent = NULL;
91 p = &fs_info->defrag_inodes.rb_node;
94 entry = rb_entry(parent, struct inode_defrag, rb_node);
96 ret = __compare_inode_defrag(defrag, entry);
100 p = &parent->rb_right;
102 /* if we're reinserting an entry for
103 * an old defrag run, make sure to
104 * lower the transid of our existing record
106 if (defrag->transid < entry->transid)
107 entry->transid = defrag->transid;
108 if (defrag->last_offset > entry->last_offset)
109 entry->last_offset = defrag->last_offset;
113 set_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags);
114 rb_link_node(&defrag->rb_node, parent, p);
115 rb_insert_color(&defrag->rb_node, &fs_info->defrag_inodes);
119 static inline int __need_auto_defrag(struct btrfs_fs_info *fs_info)
121 if (!btrfs_test_opt(fs_info, AUTO_DEFRAG))
124 if (btrfs_fs_closing(fs_info))
131 * insert a defrag record for this inode if auto defrag is
134 int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans,
135 struct btrfs_inode *inode)
137 struct btrfs_root *root = inode->root;
138 struct btrfs_fs_info *fs_info = root->fs_info;
139 struct inode_defrag *defrag;
143 if (!__need_auto_defrag(fs_info))
146 if (test_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags))
150 transid = trans->transid;
152 transid = inode->root->last_trans;
154 defrag = kmem_cache_zalloc(btrfs_inode_defrag_cachep, GFP_NOFS);
158 defrag->ino = btrfs_ino(inode);
159 defrag->transid = transid;
160 defrag->root = root->root_key.objectid;
162 spin_lock(&fs_info->defrag_inodes_lock);
163 if (!test_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags)) {
165 * If we set IN_DEFRAG flag and evict the inode from memory,
166 * and then re-read this inode, this new inode doesn't have
167 * IN_DEFRAG flag. At the case, we may find the existed defrag.
169 ret = __btrfs_add_inode_defrag(inode, defrag);
171 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
173 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
175 spin_unlock(&fs_info->defrag_inodes_lock);
180 * Requeue the defrag object. If there is a defrag object that points to
181 * the same inode in the tree, we will merge them together (by
182 * __btrfs_add_inode_defrag()) and free the one that we want to requeue.
184 static void btrfs_requeue_inode_defrag(struct btrfs_inode *inode,
185 struct inode_defrag *defrag)
187 struct btrfs_fs_info *fs_info = inode->root->fs_info;
190 if (!__need_auto_defrag(fs_info))
194 * Here we don't check the IN_DEFRAG flag, because we need merge
197 spin_lock(&fs_info->defrag_inodes_lock);
198 ret = __btrfs_add_inode_defrag(inode, defrag);
199 spin_unlock(&fs_info->defrag_inodes_lock);
204 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
208 * pick the defragable inode that we want, if it doesn't exist, we will get
211 static struct inode_defrag *
212 btrfs_pick_defrag_inode(struct btrfs_fs_info *fs_info, u64 root, u64 ino)
214 struct inode_defrag *entry = NULL;
215 struct inode_defrag tmp;
217 struct rb_node *parent = NULL;
223 spin_lock(&fs_info->defrag_inodes_lock);
224 p = fs_info->defrag_inodes.rb_node;
227 entry = rb_entry(parent, struct inode_defrag, rb_node);
229 ret = __compare_inode_defrag(&tmp, entry);
233 p = parent->rb_right;
238 if (parent && __compare_inode_defrag(&tmp, entry) > 0) {
239 parent = rb_next(parent);
241 entry = rb_entry(parent, struct inode_defrag, rb_node);
247 rb_erase(parent, &fs_info->defrag_inodes);
248 spin_unlock(&fs_info->defrag_inodes_lock);
252 void btrfs_cleanup_defrag_inodes(struct btrfs_fs_info *fs_info)
254 struct inode_defrag *defrag;
255 struct rb_node *node;
257 spin_lock(&fs_info->defrag_inodes_lock);
258 node = rb_first(&fs_info->defrag_inodes);
260 rb_erase(node, &fs_info->defrag_inodes);
261 defrag = rb_entry(node, struct inode_defrag, rb_node);
262 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
264 cond_resched_lock(&fs_info->defrag_inodes_lock);
266 node = rb_first(&fs_info->defrag_inodes);
268 spin_unlock(&fs_info->defrag_inodes_lock);
271 #define BTRFS_DEFRAG_BATCH 1024
273 static int __btrfs_run_defrag_inode(struct btrfs_fs_info *fs_info,
274 struct inode_defrag *defrag)
276 struct btrfs_root *inode_root;
278 struct btrfs_ioctl_defrag_range_args range;
283 inode_root = btrfs_get_fs_root(fs_info, defrag->root, true);
284 if (IS_ERR(inode_root)) {
285 ret = PTR_ERR(inode_root);
289 inode = btrfs_iget(fs_info->sb, defrag->ino, inode_root);
290 btrfs_put_root(inode_root);
292 ret = PTR_ERR(inode);
296 /* do a chunk of defrag */
297 clear_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
298 memset(&range, 0, sizeof(range));
300 range.start = defrag->last_offset;
302 sb_start_write(fs_info->sb);
303 num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid,
305 sb_end_write(fs_info->sb);
307 * if we filled the whole defrag batch, there
308 * must be more work to do. Queue this defrag
311 if (num_defrag == BTRFS_DEFRAG_BATCH) {
312 defrag->last_offset = range.start;
313 btrfs_requeue_inode_defrag(BTRFS_I(inode), defrag);
314 } else if (defrag->last_offset && !defrag->cycled) {
316 * we didn't fill our defrag batch, but
317 * we didn't start at zero. Make sure we loop
318 * around to the start of the file.
320 defrag->last_offset = 0;
322 btrfs_requeue_inode_defrag(BTRFS_I(inode), defrag);
324 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
330 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
335 * run through the list of inodes in the FS that need
338 int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info)
340 struct inode_defrag *defrag;
342 u64 root_objectid = 0;
344 atomic_inc(&fs_info->defrag_running);
346 /* Pause the auto defragger. */
347 if (test_bit(BTRFS_FS_STATE_REMOUNTING,
351 if (!__need_auto_defrag(fs_info))
354 /* find an inode to defrag */
355 defrag = btrfs_pick_defrag_inode(fs_info, root_objectid,
358 if (root_objectid || first_ino) {
367 first_ino = defrag->ino + 1;
368 root_objectid = defrag->root;
370 __btrfs_run_defrag_inode(fs_info, defrag);
372 atomic_dec(&fs_info->defrag_running);
375 * during unmount, we use the transaction_wait queue to
376 * wait for the defragger to stop
378 wake_up(&fs_info->transaction_wait);
382 /* simple helper to fault in pages and copy. This should go away
383 * and be replaced with calls into generic code.
385 static noinline int btrfs_copy_from_user(loff_t pos, size_t write_bytes,
386 struct page **prepared_pages,
390 size_t total_copied = 0;
392 int offset = offset_in_page(pos);
394 while (write_bytes > 0) {
395 size_t count = min_t(size_t,
396 PAGE_SIZE - offset, write_bytes);
397 struct page *page = prepared_pages[pg];
399 * Copy data from userspace to the current page
401 copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
403 /* Flush processor's dcache for this page */
404 flush_dcache_page(page);
407 * if we get a partial write, we can end up with
408 * partially up to date pages. These add
409 * a lot of complexity, so make sure they don't
410 * happen by forcing this copy to be retried.
412 * The rest of the btrfs_file_write code will fall
413 * back to page at a time copies after we return 0.
415 if (!PageUptodate(page) && copied < count)
418 iov_iter_advance(i, copied);
419 write_bytes -= copied;
420 total_copied += copied;
422 /* Return to btrfs_file_write_iter to fault page */
423 if (unlikely(copied == 0))
426 if (copied < PAGE_SIZE - offset) {
437 * unlocks pages after btrfs_file_write is done with them
439 static void btrfs_drop_pages(struct page **pages, size_t num_pages)
442 for (i = 0; i < num_pages; i++) {
443 /* page checked is some magic around finding pages that
444 * have been modified without going through btrfs_set_page_dirty
445 * clear it here. There should be no need to mark the pages
446 * accessed as prepare_pages should have marked them accessed
447 * in prepare_pages via find_or_create_page()
449 ClearPageChecked(pages[i]);
450 unlock_page(pages[i]);
456 * After btrfs_copy_from_user(), update the following things for delalloc:
457 * - Mark newly dirtied pages as DELALLOC in the io tree.
458 * Used to advise which range is to be written back.
459 * - Mark modified pages as Uptodate/Dirty and not needing COW fixup
460 * - Update inode size for past EOF write
462 int btrfs_dirty_pages(struct btrfs_inode *inode, struct page **pages,
463 size_t num_pages, loff_t pos, size_t write_bytes,
464 struct extent_state **cached, bool noreserve)
466 struct btrfs_fs_info *fs_info = inode->root->fs_info;
471 u64 end_of_last_block;
472 u64 end_pos = pos + write_bytes;
473 loff_t isize = i_size_read(&inode->vfs_inode);
474 unsigned int extra_bits = 0;
476 if (write_bytes == 0)
480 extra_bits |= EXTENT_NORESERVE;
482 start_pos = round_down(pos, fs_info->sectorsize);
483 num_bytes = round_up(write_bytes + pos - start_pos,
484 fs_info->sectorsize);
486 end_of_last_block = start_pos + num_bytes - 1;
489 * The pages may have already been dirty, clear out old accounting so
490 * we can set things up properly
492 clear_extent_bit(&inode->io_tree, start_pos, end_of_last_block,
493 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
496 err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
501 for (i = 0; i < num_pages; i++) {
502 struct page *p = pages[i];
509 * we've only changed i_size in ram, and we haven't updated
510 * the disk i_size. There is no need to log the inode
514 i_size_write(&inode->vfs_inode, end_pos);
519 * this drops all the extents in the cache that intersect the range
520 * [start, end]. Existing extents are split as required.
522 void btrfs_drop_extent_cache(struct btrfs_inode *inode, u64 start, u64 end,
525 struct extent_map *em;
526 struct extent_map *split = NULL;
527 struct extent_map *split2 = NULL;
528 struct extent_map_tree *em_tree = &inode->extent_tree;
529 u64 len = end - start + 1;
537 WARN_ON(end < start);
538 if (end == (u64)-1) {
547 split = alloc_extent_map();
549 split2 = alloc_extent_map();
550 if (!split || !split2)
553 write_lock(&em_tree->lock);
554 em = lookup_extent_mapping(em_tree, start, len);
556 write_unlock(&em_tree->lock);
560 gen = em->generation;
561 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
562 if (testend && em->start + em->len >= start + len) {
564 write_unlock(&em_tree->lock);
567 start = em->start + em->len;
569 len = start + len - (em->start + em->len);
571 write_unlock(&em_tree->lock);
574 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
575 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
576 clear_bit(EXTENT_FLAG_LOGGING, &flags);
577 modified = !list_empty(&em->list);
581 if (em->start < start) {
582 split->start = em->start;
583 split->len = start - em->start;
585 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
586 split->orig_start = em->orig_start;
587 split->block_start = em->block_start;
590 split->block_len = em->block_len;
592 split->block_len = split->len;
593 split->orig_block_len = max(split->block_len,
595 split->ram_bytes = em->ram_bytes;
597 split->orig_start = split->start;
598 split->block_len = 0;
599 split->block_start = em->block_start;
600 split->orig_block_len = 0;
601 split->ram_bytes = split->len;
604 split->generation = gen;
605 split->flags = flags;
606 split->compress_type = em->compress_type;
607 replace_extent_mapping(em_tree, em, split, modified);
608 free_extent_map(split);
612 if (testend && em->start + em->len > start + len) {
613 u64 diff = start + len - em->start;
615 split->start = start + len;
616 split->len = em->start + em->len - (start + len);
617 split->flags = flags;
618 split->compress_type = em->compress_type;
619 split->generation = gen;
621 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
622 split->orig_block_len = max(em->block_len,
625 split->ram_bytes = em->ram_bytes;
627 split->block_len = em->block_len;
628 split->block_start = em->block_start;
629 split->orig_start = em->orig_start;
631 split->block_len = split->len;
632 split->block_start = em->block_start
634 split->orig_start = em->orig_start;
637 split->ram_bytes = split->len;
638 split->orig_start = split->start;
639 split->block_len = 0;
640 split->block_start = em->block_start;
641 split->orig_block_len = 0;
644 if (extent_map_in_tree(em)) {
645 replace_extent_mapping(em_tree, em, split,
648 ret = add_extent_mapping(em_tree, split,
650 ASSERT(ret == 0); /* Logic error */
652 free_extent_map(split);
656 if (extent_map_in_tree(em))
657 remove_extent_mapping(em_tree, em);
658 write_unlock(&em_tree->lock);
662 /* once for the tree*/
666 free_extent_map(split);
668 free_extent_map(split2);
672 * this is very complex, but the basic idea is to drop all extents
673 * in the range start - end. hint_block is filled in with a block number
674 * that would be a good hint to the block allocator for this file.
676 * If an extent intersects the range but is not entirely inside the range
677 * it is either truncated or split. Anything entirely inside the range
678 * is deleted from the tree.
680 * Note: the VFS' inode number of bytes is not updated, it's up to the caller
681 * to deal with that. We set the field 'bytes_found' of the arguments structure
682 * with the number of allocated bytes found in the target range, so that the
683 * caller can update the inode's number of bytes in an atomic way when
684 * replacing extents in a range to avoid races with stat(2).
686 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
687 struct btrfs_root *root, struct btrfs_inode *inode,
688 struct btrfs_drop_extents_args *args)
690 struct btrfs_fs_info *fs_info = root->fs_info;
691 struct extent_buffer *leaf;
692 struct btrfs_file_extent_item *fi;
693 struct btrfs_ref ref = { 0 };
694 struct btrfs_key key;
695 struct btrfs_key new_key;
696 u64 ino = btrfs_ino(inode);
697 u64 search_start = args->start;
700 u64 extent_offset = 0;
702 u64 last_end = args->start;
708 int modify_tree = -1;
711 int leafs_visited = 0;
712 struct btrfs_path *path = args->path;
714 args->bytes_found = 0;
715 args->extent_inserted = false;
717 /* Must always have a path if ->replace_extent is true */
718 ASSERT(!(args->replace_extent && !args->path));
721 path = btrfs_alloc_path();
728 if (args->drop_cache)
729 btrfs_drop_extent_cache(inode, args->start, args->end - 1, 0);
731 if (args->start >= inode->disk_i_size && !args->replace_extent)
734 update_refs = (test_bit(BTRFS_ROOT_SHAREABLE, &root->state) ||
735 root == fs_info->tree_root);
738 ret = btrfs_lookup_file_extent(trans, root, path, ino,
739 search_start, modify_tree);
742 if (ret > 0 && path->slots[0] > 0 && search_start == args->start) {
743 leaf = path->nodes[0];
744 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
745 if (key.objectid == ino &&
746 key.type == BTRFS_EXTENT_DATA_KEY)
752 leaf = path->nodes[0];
753 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
755 ret = btrfs_next_leaf(root, path);
763 leaf = path->nodes[0];
767 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
769 if (key.objectid > ino)
771 if (WARN_ON_ONCE(key.objectid < ino) ||
772 key.type < BTRFS_EXTENT_DATA_KEY) {
777 if (key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= args->end)
780 fi = btrfs_item_ptr(leaf, path->slots[0],
781 struct btrfs_file_extent_item);
782 extent_type = btrfs_file_extent_type(leaf, fi);
784 if (extent_type == BTRFS_FILE_EXTENT_REG ||
785 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
786 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
787 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
788 extent_offset = btrfs_file_extent_offset(leaf, fi);
789 extent_end = key.offset +
790 btrfs_file_extent_num_bytes(leaf, fi);
791 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
792 extent_end = key.offset +
793 btrfs_file_extent_ram_bytes(leaf, fi);
800 * Don't skip extent items representing 0 byte lengths. They
801 * used to be created (bug) if while punching holes we hit
802 * -ENOSPC condition. So if we find one here, just ensure we
803 * delete it, otherwise we would insert a new file extent item
804 * with the same key (offset) as that 0 bytes length file
805 * extent item in the call to setup_items_for_insert() later
808 if (extent_end == key.offset && extent_end >= search_start) {
809 last_end = extent_end;
810 goto delete_extent_item;
813 if (extent_end <= search_start) {
819 search_start = max(key.offset, args->start);
820 if (recow || !modify_tree) {
822 btrfs_release_path(path);
827 * | - range to drop - |
828 * | -------- extent -------- |
830 if (args->start > key.offset && args->end < extent_end) {
832 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
837 memcpy(&new_key, &key, sizeof(new_key));
838 new_key.offset = args->start;
839 ret = btrfs_duplicate_item(trans, root, path,
841 if (ret == -EAGAIN) {
842 btrfs_release_path(path);
848 leaf = path->nodes[0];
849 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
850 struct btrfs_file_extent_item);
851 btrfs_set_file_extent_num_bytes(leaf, fi,
852 args->start - key.offset);
854 fi = btrfs_item_ptr(leaf, path->slots[0],
855 struct btrfs_file_extent_item);
857 extent_offset += args->start - key.offset;
858 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
859 btrfs_set_file_extent_num_bytes(leaf, fi,
860 extent_end - args->start);
861 btrfs_mark_buffer_dirty(leaf);
863 if (update_refs && disk_bytenr > 0) {
864 btrfs_init_generic_ref(&ref,
865 BTRFS_ADD_DELAYED_REF,
866 disk_bytenr, num_bytes, 0);
867 btrfs_init_data_ref(&ref,
868 root->root_key.objectid,
870 args->start - extent_offset);
871 ret = btrfs_inc_extent_ref(trans, &ref);
872 BUG_ON(ret); /* -ENOMEM */
874 key.offset = args->start;
877 * From here on out we will have actually dropped something, so
878 * last_end can be updated.
880 last_end = extent_end;
883 * | ---- range to drop ----- |
884 * | -------- extent -------- |
886 if (args->start <= key.offset && args->end < extent_end) {
887 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
892 memcpy(&new_key, &key, sizeof(new_key));
893 new_key.offset = args->end;
894 btrfs_set_item_key_safe(fs_info, path, &new_key);
896 extent_offset += args->end - key.offset;
897 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
898 btrfs_set_file_extent_num_bytes(leaf, fi,
899 extent_end - args->end);
900 btrfs_mark_buffer_dirty(leaf);
901 if (update_refs && disk_bytenr > 0)
902 args->bytes_found += args->end - key.offset;
906 search_start = extent_end;
908 * | ---- range to drop ----- |
909 * | -------- extent -------- |
911 if (args->start > key.offset && args->end >= extent_end) {
913 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
918 btrfs_set_file_extent_num_bytes(leaf, fi,
919 args->start - key.offset);
920 btrfs_mark_buffer_dirty(leaf);
921 if (update_refs && disk_bytenr > 0)
922 args->bytes_found += extent_end - args->start;
923 if (args->end == extent_end)
931 * | ---- range to drop ----- |
932 * | ------ extent ------ |
934 if (args->start <= key.offset && args->end >= extent_end) {
937 del_slot = path->slots[0];
940 BUG_ON(del_slot + del_nr != path->slots[0]);
945 extent_type == BTRFS_FILE_EXTENT_INLINE) {
946 args->bytes_found += extent_end - key.offset;
947 extent_end = ALIGN(extent_end,
948 fs_info->sectorsize);
949 } else if (update_refs && disk_bytenr > 0) {
950 btrfs_init_generic_ref(&ref,
951 BTRFS_DROP_DELAYED_REF,
952 disk_bytenr, num_bytes, 0);
953 btrfs_init_data_ref(&ref,
954 root->root_key.objectid,
956 key.offset - extent_offset);
957 ret = btrfs_free_extent(trans, &ref);
958 BUG_ON(ret); /* -ENOMEM */
959 args->bytes_found += extent_end - key.offset;
962 if (args->end == extent_end)
965 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
970 ret = btrfs_del_items(trans, root, path, del_slot,
973 btrfs_abort_transaction(trans, ret);
980 btrfs_release_path(path);
987 if (!ret && del_nr > 0) {
989 * Set path->slots[0] to first slot, so that after the delete
990 * if items are move off from our leaf to its immediate left or
991 * right neighbor leafs, we end up with a correct and adjusted
992 * path->slots[0] for our insertion (if args->replace_extent).
994 path->slots[0] = del_slot;
995 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
997 btrfs_abort_transaction(trans, ret);
1000 leaf = path->nodes[0];
1002 * If btrfs_del_items() was called, it might have deleted a leaf, in
1003 * which case it unlocked our path, so check path->locks[0] matches a
1006 if (!ret && args->replace_extent && leafs_visited == 1 &&
1007 path->locks[0] == BTRFS_WRITE_LOCK &&
1008 btrfs_leaf_free_space(leaf) >=
1009 sizeof(struct btrfs_item) + args->extent_item_size) {
1012 key.type = BTRFS_EXTENT_DATA_KEY;
1013 key.offset = args->start;
1014 if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) {
1015 struct btrfs_key slot_key;
1017 btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]);
1018 if (btrfs_comp_cpu_keys(&key, &slot_key) > 0)
1021 setup_items_for_insert(root, path, &key,
1022 &args->extent_item_size, 1);
1023 args->extent_inserted = true;
1027 btrfs_free_path(path);
1028 else if (!args->extent_inserted)
1029 btrfs_release_path(path);
1031 args->drop_end = found ? min(args->end, last_end) : args->end;
1036 static int extent_mergeable(struct extent_buffer *leaf, int slot,
1037 u64 objectid, u64 bytenr, u64 orig_offset,
1038 u64 *start, u64 *end)
1040 struct btrfs_file_extent_item *fi;
1041 struct btrfs_key key;
1044 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
1047 btrfs_item_key_to_cpu(leaf, &key, slot);
1048 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
1051 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1052 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
1053 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
1054 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
1055 btrfs_file_extent_compression(leaf, fi) ||
1056 btrfs_file_extent_encryption(leaf, fi) ||
1057 btrfs_file_extent_other_encoding(leaf, fi))
1060 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1061 if ((*start && *start != key.offset) || (*end && *end != extent_end))
1064 *start = key.offset;
1070 * Mark extent in the range start - end as written.
1072 * This changes extent type from 'pre-allocated' to 'regular'. If only
1073 * part of extent is marked as written, the extent will be split into
1076 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
1077 struct btrfs_inode *inode, u64 start, u64 end)
1079 struct btrfs_fs_info *fs_info = trans->fs_info;
1080 struct btrfs_root *root = inode->root;
1081 struct extent_buffer *leaf;
1082 struct btrfs_path *path;
1083 struct btrfs_file_extent_item *fi;
1084 struct btrfs_ref ref = { 0 };
1085 struct btrfs_key key;
1086 struct btrfs_key new_key;
1098 u64 ino = btrfs_ino(inode);
1100 path = btrfs_alloc_path();
1107 key.type = BTRFS_EXTENT_DATA_KEY;
1110 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1113 if (ret > 0 && path->slots[0] > 0)
1116 leaf = path->nodes[0];
1117 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1118 if (key.objectid != ino ||
1119 key.type != BTRFS_EXTENT_DATA_KEY) {
1121 btrfs_abort_transaction(trans, ret);
1124 fi = btrfs_item_ptr(leaf, path->slots[0],
1125 struct btrfs_file_extent_item);
1126 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_PREALLOC) {
1128 btrfs_abort_transaction(trans, ret);
1131 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1132 if (key.offset > start || extent_end < end) {
1134 btrfs_abort_transaction(trans, ret);
1138 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1139 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1140 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
1141 memcpy(&new_key, &key, sizeof(new_key));
1143 if (start == key.offset && end < extent_end) {
1146 if (extent_mergeable(leaf, path->slots[0] - 1,
1147 ino, bytenr, orig_offset,
1148 &other_start, &other_end)) {
1149 new_key.offset = end;
1150 btrfs_set_item_key_safe(fs_info, path, &new_key);
1151 fi = btrfs_item_ptr(leaf, path->slots[0],
1152 struct btrfs_file_extent_item);
1153 btrfs_set_file_extent_generation(leaf, fi,
1155 btrfs_set_file_extent_num_bytes(leaf, fi,
1157 btrfs_set_file_extent_offset(leaf, fi,
1159 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1160 struct btrfs_file_extent_item);
1161 btrfs_set_file_extent_generation(leaf, fi,
1163 btrfs_set_file_extent_num_bytes(leaf, fi,
1165 btrfs_mark_buffer_dirty(leaf);
1170 if (start > key.offset && end == extent_end) {
1173 if (extent_mergeable(leaf, path->slots[0] + 1,
1174 ino, bytenr, orig_offset,
1175 &other_start, &other_end)) {
1176 fi = btrfs_item_ptr(leaf, path->slots[0],
1177 struct btrfs_file_extent_item);
1178 btrfs_set_file_extent_num_bytes(leaf, fi,
1179 start - key.offset);
1180 btrfs_set_file_extent_generation(leaf, fi,
1183 new_key.offset = start;
1184 btrfs_set_item_key_safe(fs_info, path, &new_key);
1186 fi = btrfs_item_ptr(leaf, path->slots[0],
1187 struct btrfs_file_extent_item);
1188 btrfs_set_file_extent_generation(leaf, fi,
1190 btrfs_set_file_extent_num_bytes(leaf, fi,
1192 btrfs_set_file_extent_offset(leaf, fi,
1193 start - orig_offset);
1194 btrfs_mark_buffer_dirty(leaf);
1199 while (start > key.offset || end < extent_end) {
1200 if (key.offset == start)
1203 new_key.offset = split;
1204 ret = btrfs_duplicate_item(trans, root, path, &new_key);
1205 if (ret == -EAGAIN) {
1206 btrfs_release_path(path);
1210 btrfs_abort_transaction(trans, ret);
1214 leaf = path->nodes[0];
1215 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1216 struct btrfs_file_extent_item);
1217 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1218 btrfs_set_file_extent_num_bytes(leaf, fi,
1219 split - key.offset);
1221 fi = btrfs_item_ptr(leaf, path->slots[0],
1222 struct btrfs_file_extent_item);
1224 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1225 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
1226 btrfs_set_file_extent_num_bytes(leaf, fi,
1227 extent_end - split);
1228 btrfs_mark_buffer_dirty(leaf);
1230 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, bytenr,
1232 btrfs_init_data_ref(&ref, root->root_key.objectid, ino,
1234 ret = btrfs_inc_extent_ref(trans, &ref);
1236 btrfs_abort_transaction(trans, ret);
1240 if (split == start) {
1243 if (start != key.offset) {
1245 btrfs_abort_transaction(trans, ret);
1256 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
1258 btrfs_init_data_ref(&ref, root->root_key.objectid, ino, orig_offset);
1259 if (extent_mergeable(leaf, path->slots[0] + 1,
1260 ino, bytenr, orig_offset,
1261 &other_start, &other_end)) {
1263 btrfs_release_path(path);
1266 extent_end = other_end;
1267 del_slot = path->slots[0] + 1;
1269 ret = btrfs_free_extent(trans, &ref);
1271 btrfs_abort_transaction(trans, ret);
1277 if (extent_mergeable(leaf, path->slots[0] - 1,
1278 ino, bytenr, orig_offset,
1279 &other_start, &other_end)) {
1281 btrfs_release_path(path);
1284 key.offset = other_start;
1285 del_slot = path->slots[0];
1287 ret = btrfs_free_extent(trans, &ref);
1289 btrfs_abort_transaction(trans, ret);
1294 fi = btrfs_item_ptr(leaf, path->slots[0],
1295 struct btrfs_file_extent_item);
1296 btrfs_set_file_extent_type(leaf, fi,
1297 BTRFS_FILE_EXTENT_REG);
1298 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1299 btrfs_mark_buffer_dirty(leaf);
1301 fi = btrfs_item_ptr(leaf, del_slot - 1,
1302 struct btrfs_file_extent_item);
1303 btrfs_set_file_extent_type(leaf, fi,
1304 BTRFS_FILE_EXTENT_REG);
1305 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1306 btrfs_set_file_extent_num_bytes(leaf, fi,
1307 extent_end - key.offset);
1308 btrfs_mark_buffer_dirty(leaf);
1310 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1312 btrfs_abort_transaction(trans, ret);
1317 btrfs_free_path(path);
1322 * on error we return an unlocked page and the error value
1323 * on success we return a locked page and 0
1325 static int prepare_uptodate_page(struct inode *inode,
1326 struct page *page, u64 pos,
1327 bool force_uptodate)
1331 if (((pos & (PAGE_SIZE - 1)) || force_uptodate) &&
1332 !PageUptodate(page)) {
1333 ret = btrfs_readpage(NULL, page);
1337 if (!PageUptodate(page)) {
1341 if (page->mapping != inode->i_mapping) {
1350 * this just gets pages into the page cache and locks them down.
1352 static noinline int prepare_pages(struct inode *inode, struct page **pages,
1353 size_t num_pages, loff_t pos,
1354 size_t write_bytes, bool force_uptodate)
1357 unsigned long index = pos >> PAGE_SHIFT;
1358 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1362 for (i = 0; i < num_pages; i++) {
1364 pages[i] = find_or_create_page(inode->i_mapping, index + i,
1365 mask | __GFP_WRITE);
1373 err = prepare_uptodate_page(inode, pages[i], pos,
1375 if (!err && i == num_pages - 1)
1376 err = prepare_uptodate_page(inode, pages[i],
1377 pos + write_bytes, false);
1380 if (err == -EAGAIN) {
1387 wait_on_page_writeback(pages[i]);
1392 while (faili >= 0) {
1393 unlock_page(pages[faili]);
1394 put_page(pages[faili]);
1402 * This function locks the extent and properly waits for data=ordered extents
1403 * to finish before allowing the pages to be modified if need.
1406 * 1 - the extent is locked
1407 * 0 - the extent is not locked, and everything is OK
1408 * -EAGAIN - need re-prepare the pages
1409 * the other < 0 number - Something wrong happens
1412 lock_and_cleanup_extent_if_need(struct btrfs_inode *inode, struct page **pages,
1413 size_t num_pages, loff_t pos,
1415 u64 *lockstart, u64 *lockend,
1416 struct extent_state **cached_state)
1418 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1424 start_pos = round_down(pos, fs_info->sectorsize);
1425 last_pos = round_up(pos + write_bytes, fs_info->sectorsize) - 1;
1427 if (start_pos < inode->vfs_inode.i_size) {
1428 struct btrfs_ordered_extent *ordered;
1430 lock_extent_bits(&inode->io_tree, start_pos, last_pos,
1432 ordered = btrfs_lookup_ordered_range(inode, start_pos,
1433 last_pos - start_pos + 1);
1435 ordered->file_offset + ordered->num_bytes > start_pos &&
1436 ordered->file_offset <= last_pos) {
1437 unlock_extent_cached(&inode->io_tree, start_pos,
1438 last_pos, cached_state);
1439 for (i = 0; i < num_pages; i++) {
1440 unlock_page(pages[i]);
1443 btrfs_start_ordered_extent(ordered, 1);
1444 btrfs_put_ordered_extent(ordered);
1448 btrfs_put_ordered_extent(ordered);
1450 *lockstart = start_pos;
1451 *lockend = last_pos;
1456 * It's possible the pages are dirty right now, but we don't want
1457 * to clean them yet because copy_from_user may catch a page fault
1458 * and we might have to fall back to one page at a time. If that
1459 * happens, we'll unlock these pages and we'd have a window where
1460 * reclaim could sneak in and drop the once-dirty page on the floor
1461 * without writing it.
1463 * We have the pages locked and the extent range locked, so there's
1464 * no way someone can start IO on any dirty pages in this range.
1466 * We'll call btrfs_dirty_pages() later on, and that will flip around
1467 * delalloc bits and dirty the pages as required.
1469 for (i = 0; i < num_pages; i++) {
1470 set_page_extent_mapped(pages[i]);
1471 WARN_ON(!PageLocked(pages[i]));
1477 static int check_can_nocow(struct btrfs_inode *inode, loff_t pos,
1478 size_t *write_bytes, bool nowait)
1480 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1481 struct btrfs_root *root = inode->root;
1482 u64 lockstart, lockend;
1486 if (!(inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
1489 if (!nowait && !btrfs_drew_try_write_lock(&root->snapshot_lock))
1492 lockstart = round_down(pos, fs_info->sectorsize);
1493 lockend = round_up(pos + *write_bytes,
1494 fs_info->sectorsize) - 1;
1495 num_bytes = lockend - lockstart + 1;
1498 struct btrfs_ordered_extent *ordered;
1500 if (!try_lock_extent(&inode->io_tree, lockstart, lockend))
1503 ordered = btrfs_lookup_ordered_range(inode, lockstart,
1506 btrfs_put_ordered_extent(ordered);
1511 btrfs_lock_and_flush_ordered_range(inode, lockstart,
1515 ret = can_nocow_extent(&inode->vfs_inode, lockstart, &num_bytes,
1516 NULL, NULL, NULL, false);
1520 btrfs_drew_write_unlock(&root->snapshot_lock);
1522 *write_bytes = min_t(size_t, *write_bytes ,
1523 num_bytes - pos + lockstart);
1526 unlock_extent(&inode->io_tree, lockstart, lockend);
1531 static int check_nocow_nolock(struct btrfs_inode *inode, loff_t pos,
1532 size_t *write_bytes)
1534 return check_can_nocow(inode, pos, write_bytes, true);
1538 * Check if we can do nocow write into the range [@pos, @pos + @write_bytes)
1541 * @write_bytes: The length to write, will be updated to the nocow writeable
1544 * This function will flush ordered extents in the range to ensure proper
1548 * >0 and update @write_bytes if we can do nocow write
1549 * 0 if we can't do nocow write
1550 * -EAGAIN if we can't get the needed lock or there are ordered extents
1551 * for * (nowait == true) case
1552 * <0 if other error happened
1554 * NOTE: Callers need to release the lock by btrfs_check_nocow_unlock().
1556 int btrfs_check_nocow_lock(struct btrfs_inode *inode, loff_t pos,
1557 size_t *write_bytes)
1559 return check_can_nocow(inode, pos, write_bytes, false);
1562 void btrfs_check_nocow_unlock(struct btrfs_inode *inode)
1564 btrfs_drew_write_unlock(&inode->root->snapshot_lock);
1567 static void update_time_for_write(struct inode *inode)
1569 struct timespec64 now;
1571 if (IS_NOCMTIME(inode))
1574 now = current_time(inode);
1575 if (!timespec64_equal(&inode->i_mtime, &now))
1576 inode->i_mtime = now;
1578 if (!timespec64_equal(&inode->i_ctime, &now))
1579 inode->i_ctime = now;
1581 if (IS_I_VERSION(inode))
1582 inode_inc_iversion(inode);
1585 static int btrfs_write_check(struct kiocb *iocb, struct iov_iter *from,
1588 struct file *file = iocb->ki_filp;
1589 struct inode *inode = file_inode(file);
1590 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1591 loff_t pos = iocb->ki_pos;
1596 if (iocb->ki_flags & IOCB_NOWAIT) {
1597 size_t nocow_bytes = count;
1599 /* We will allocate space in case nodatacow is not set, so bail */
1600 if (check_nocow_nolock(BTRFS_I(inode), pos, &nocow_bytes) <= 0)
1603 * There are holes in the range or parts of the range that must
1604 * be COWed (shared extents, RO block groups, etc), so just bail
1607 if (nocow_bytes < count)
1611 current->backing_dev_info = inode_to_bdi(inode);
1612 ret = file_remove_privs(file);
1617 * We reserve space for updating the inode when we reserve space for the
1618 * extent we are going to write, so we will enospc out there. We don't
1619 * need to start yet another transaction to update the inode as we will
1620 * update the inode when we finish writing whatever data we write.
1622 update_time_for_write(inode);
1624 start_pos = round_down(pos, fs_info->sectorsize);
1625 oldsize = i_size_read(inode);
1626 if (start_pos > oldsize) {
1627 /* Expand hole size to cover write data, preventing empty gap */
1628 loff_t end_pos = round_up(pos + count, fs_info->sectorsize);
1630 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, end_pos);
1632 current->backing_dev_info = NULL;
1640 static noinline ssize_t btrfs_buffered_write(struct kiocb *iocb,
1643 struct file *file = iocb->ki_filp;
1645 struct inode *inode = file_inode(file);
1646 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1647 struct page **pages = NULL;
1648 struct extent_changeset *data_reserved = NULL;
1649 u64 release_bytes = 0;
1652 size_t num_written = 0;
1655 bool only_release_metadata = false;
1656 bool force_page_uptodate = false;
1657 loff_t old_isize = i_size_read(inode);
1658 unsigned int ilock_flags = 0;
1660 if (iocb->ki_flags & IOCB_NOWAIT)
1661 ilock_flags |= BTRFS_ILOCK_TRY;
1663 ret = btrfs_inode_lock(inode, ilock_flags);
1667 ret = generic_write_checks(iocb, i);
1671 ret = btrfs_write_check(iocb, i, ret);
1676 nrptrs = min(DIV_ROUND_UP(iov_iter_count(i), PAGE_SIZE),
1677 PAGE_SIZE / (sizeof(struct page *)));
1678 nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1679 nrptrs = max(nrptrs, 8);
1680 pages = kmalloc_array(nrptrs, sizeof(struct page *), GFP_KERNEL);
1686 while (iov_iter_count(i) > 0) {
1687 struct extent_state *cached_state = NULL;
1688 size_t offset = offset_in_page(pos);
1689 size_t sector_offset;
1690 size_t write_bytes = min(iov_iter_count(i),
1691 nrptrs * (size_t)PAGE_SIZE -
1694 size_t reserve_bytes;
1697 size_t dirty_sectors;
1702 * Fault pages before locking them in prepare_pages
1703 * to avoid recursive lock
1705 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1710 only_release_metadata = false;
1711 sector_offset = pos & (fs_info->sectorsize - 1);
1713 extent_changeset_release(data_reserved);
1714 ret = btrfs_check_data_free_space(BTRFS_I(inode),
1715 &data_reserved, pos,
1719 * If we don't have to COW at the offset, reserve
1720 * metadata only. write_bytes may get smaller than
1723 if (btrfs_check_nocow_lock(BTRFS_I(inode), pos,
1725 only_release_metadata = true;
1730 num_pages = DIV_ROUND_UP(write_bytes + offset, PAGE_SIZE);
1731 WARN_ON(num_pages > nrptrs);
1732 reserve_bytes = round_up(write_bytes + sector_offset,
1733 fs_info->sectorsize);
1734 WARN_ON(reserve_bytes == 0);
1735 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
1738 if (!only_release_metadata)
1739 btrfs_free_reserved_data_space(BTRFS_I(inode),
1743 btrfs_check_nocow_unlock(BTRFS_I(inode));
1747 release_bytes = reserve_bytes;
1750 * This is going to setup the pages array with the number of
1751 * pages we want, so we don't really need to worry about the
1752 * contents of pages from loop to loop
1754 ret = prepare_pages(inode, pages, num_pages,
1756 force_page_uptodate);
1758 btrfs_delalloc_release_extents(BTRFS_I(inode),
1763 extents_locked = lock_and_cleanup_extent_if_need(
1764 BTRFS_I(inode), pages,
1765 num_pages, pos, write_bytes, &lockstart,
1766 &lockend, &cached_state);
1767 if (extents_locked < 0) {
1768 if (extents_locked == -EAGAIN)
1770 btrfs_delalloc_release_extents(BTRFS_I(inode),
1772 ret = extents_locked;
1776 copied = btrfs_copy_from_user(pos, write_bytes, pages, i);
1778 num_sectors = BTRFS_BYTES_TO_BLKS(fs_info, reserve_bytes);
1779 dirty_sectors = round_up(copied + sector_offset,
1780 fs_info->sectorsize);
1781 dirty_sectors = BTRFS_BYTES_TO_BLKS(fs_info, dirty_sectors);
1784 * if we have trouble faulting in the pages, fall
1785 * back to one page at a time
1787 if (copied < write_bytes)
1791 force_page_uptodate = true;
1795 force_page_uptodate = false;
1796 dirty_pages = DIV_ROUND_UP(copied + offset,
1800 if (num_sectors > dirty_sectors) {
1801 /* release everything except the sectors we dirtied */
1802 release_bytes -= dirty_sectors << fs_info->sectorsize_bits;
1803 if (only_release_metadata) {
1804 btrfs_delalloc_release_metadata(BTRFS_I(inode),
1805 release_bytes, true);
1809 __pos = round_down(pos,
1810 fs_info->sectorsize) +
1811 (dirty_pages << PAGE_SHIFT);
1812 btrfs_delalloc_release_space(BTRFS_I(inode),
1813 data_reserved, __pos,
1814 release_bytes, true);
1818 release_bytes = round_up(copied + sector_offset,
1819 fs_info->sectorsize);
1821 ret = btrfs_dirty_pages(BTRFS_I(inode), pages,
1822 dirty_pages, pos, copied,
1823 &cached_state, only_release_metadata);
1826 * If we have not locked the extent range, because the range's
1827 * start offset is >= i_size, we might still have a non-NULL
1828 * cached extent state, acquired while marking the extent range
1829 * as delalloc through btrfs_dirty_pages(). Therefore free any
1830 * possible cached extent state to avoid a memory leak.
1833 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1834 lockstart, lockend, &cached_state);
1836 free_extent_state(cached_state);
1838 btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes);
1840 btrfs_drop_pages(pages, num_pages);
1845 if (only_release_metadata)
1846 btrfs_check_nocow_unlock(BTRFS_I(inode));
1848 btrfs_drop_pages(pages, num_pages);
1852 balance_dirty_pages_ratelimited(inode->i_mapping);
1855 num_written += copied;
1860 if (release_bytes) {
1861 if (only_release_metadata) {
1862 btrfs_check_nocow_unlock(BTRFS_I(inode));
1863 btrfs_delalloc_release_metadata(BTRFS_I(inode),
1864 release_bytes, true);
1866 btrfs_delalloc_release_space(BTRFS_I(inode),
1868 round_down(pos, fs_info->sectorsize),
1869 release_bytes, true);
1873 extent_changeset_free(data_reserved);
1874 if (num_written > 0) {
1875 pagecache_isize_extended(inode, old_isize, iocb->ki_pos);
1876 iocb->ki_pos += num_written;
1879 btrfs_inode_unlock(inode, ilock_flags);
1880 return num_written ? num_written : ret;
1883 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
1884 const struct iov_iter *iter, loff_t offset)
1886 const u32 blocksize_mask = fs_info->sectorsize - 1;
1888 if (offset & blocksize_mask)
1891 if (iov_iter_alignment(iter) & blocksize_mask)
1897 static ssize_t btrfs_direct_write(struct kiocb *iocb, struct iov_iter *from)
1899 struct file *file = iocb->ki_filp;
1900 struct inode *inode = file_inode(file);
1901 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1903 ssize_t written = 0;
1904 ssize_t written_buffered;
1907 unsigned int ilock_flags = 0;
1908 struct iomap_dio *dio = NULL;
1910 if (iocb->ki_flags & IOCB_NOWAIT)
1911 ilock_flags |= BTRFS_ILOCK_TRY;
1913 /* If the write DIO is within EOF, use a shared lock */
1914 if (iocb->ki_pos + iov_iter_count(from) <= i_size_read(inode))
1915 ilock_flags |= BTRFS_ILOCK_SHARED;
1918 err = btrfs_inode_lock(inode, ilock_flags);
1922 err = generic_write_checks(iocb, from);
1924 btrfs_inode_unlock(inode, ilock_flags);
1928 err = btrfs_write_check(iocb, from, err);
1930 btrfs_inode_unlock(inode, ilock_flags);
1936 * Re-check since file size may have changed just before taking the
1937 * lock or pos may have changed because of O_APPEND in generic_write_check()
1939 if ((ilock_flags & BTRFS_ILOCK_SHARED) &&
1940 pos + iov_iter_count(from) > i_size_read(inode)) {
1941 btrfs_inode_unlock(inode, ilock_flags);
1942 ilock_flags &= ~BTRFS_ILOCK_SHARED;
1946 if (check_direct_IO(fs_info, from, pos)) {
1947 btrfs_inode_unlock(inode, ilock_flags);
1951 dio = __iomap_dio_rw(iocb, from, &btrfs_dio_iomap_ops,
1952 &btrfs_dio_ops, is_sync_kiocb(iocb));
1954 btrfs_inode_unlock(inode, ilock_flags);
1956 if (IS_ERR_OR_NULL(dio)) {
1957 err = PTR_ERR_OR_ZERO(dio);
1958 if (err < 0 && err != -ENOTBLK)
1961 written = iomap_dio_complete(dio);
1964 if (written < 0 || !iov_iter_count(from)) {
1971 written_buffered = btrfs_buffered_write(iocb, from);
1972 if (written_buffered < 0) {
1973 err = written_buffered;
1977 * Ensure all data is persisted. We want the next direct IO read to be
1978 * able to read what was just written.
1980 endbyte = pos + written_buffered - 1;
1981 err = btrfs_fdatawrite_range(inode, pos, endbyte);
1984 err = filemap_fdatawait_range(inode->i_mapping, pos, endbyte);
1987 written += written_buffered;
1988 iocb->ki_pos = pos + written_buffered;
1989 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_SHIFT,
1990 endbyte >> PAGE_SHIFT);
1992 return written ? written : err;
1995 static ssize_t btrfs_file_write_iter(struct kiocb *iocb,
1996 struct iov_iter *from)
1998 struct file *file = iocb->ki_filp;
1999 struct btrfs_inode *inode = BTRFS_I(file_inode(file));
2000 ssize_t num_written = 0;
2001 const bool sync = iocb->ki_flags & IOCB_DSYNC;
2004 * If the fs flips readonly due to some impossible error, although we
2005 * have opened a file as writable, we have to stop this write operation
2006 * to ensure consistency.
2008 if (test_bit(BTRFS_FS_STATE_ERROR, &inode->root->fs_info->fs_state))
2011 if (!(iocb->ki_flags & IOCB_DIRECT) &&
2012 (iocb->ki_flags & IOCB_NOWAIT))
2016 atomic_inc(&inode->sync_writers);
2018 if (iocb->ki_flags & IOCB_DIRECT)
2019 num_written = btrfs_direct_write(iocb, from);
2021 num_written = btrfs_buffered_write(iocb, from);
2024 * We also have to set last_sub_trans to the current log transid,
2025 * otherwise subsequent syncs to a file that's been synced in this
2026 * transaction will appear to have already occurred.
2028 spin_lock(&inode->lock);
2029 inode->last_sub_trans = inode->root->log_transid;
2030 spin_unlock(&inode->lock);
2031 if (num_written > 0)
2032 num_written = generic_write_sync(iocb, num_written);
2035 atomic_dec(&inode->sync_writers);
2037 current->backing_dev_info = NULL;
2041 int btrfs_release_file(struct inode *inode, struct file *filp)
2043 struct btrfs_file_private *private = filp->private_data;
2045 if (private && private->filldir_buf)
2046 kfree(private->filldir_buf);
2048 filp->private_data = NULL;
2051 * Set by setattr when we are about to truncate a file from a non-zero
2052 * size to a zero size. This tries to flush down new bytes that may
2053 * have been written if the application were using truncate to replace
2056 if (test_and_clear_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
2057 &BTRFS_I(inode)->runtime_flags))
2058 filemap_flush(inode->i_mapping);
2062 static int start_ordered_ops(struct inode *inode, loff_t start, loff_t end)
2065 struct blk_plug plug;
2068 * This is only called in fsync, which would do synchronous writes, so
2069 * a plug can merge adjacent IOs as much as possible. Esp. in case of
2070 * multiple disks using raid profile, a large IO can be split to
2071 * several segments of stripe length (currently 64K).
2073 blk_start_plug(&plug);
2074 atomic_inc(&BTRFS_I(inode)->sync_writers);
2075 ret = btrfs_fdatawrite_range(inode, start, end);
2076 atomic_dec(&BTRFS_I(inode)->sync_writers);
2077 blk_finish_plug(&plug);
2083 * fsync call for both files and directories. This logs the inode into
2084 * the tree log instead of forcing full commits whenever possible.
2086 * It needs to call filemap_fdatawait so that all ordered extent updates are
2087 * in the metadata btree are up to date for copying to the log.
2089 * It drops the inode mutex before doing the tree log commit. This is an
2090 * important optimization for directories because holding the mutex prevents
2091 * new operations on the dir while we write to disk.
2093 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
2095 struct dentry *dentry = file_dentry(file);
2096 struct inode *inode = d_inode(dentry);
2097 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2098 struct btrfs_root *root = BTRFS_I(inode)->root;
2099 struct btrfs_trans_handle *trans;
2100 struct btrfs_log_ctx ctx;
2105 trace_btrfs_sync_file(file, datasync);
2107 btrfs_init_log_ctx(&ctx, inode);
2110 * Always set the range to a full range, otherwise we can get into
2111 * several problems, from missing file extent items to represent holes
2112 * when not using the NO_HOLES feature, to log tree corruption due to
2113 * races between hole detection during logging and completion of ordered
2114 * extents outside the range, to missing checksums due to ordered extents
2115 * for which we flushed only a subset of their pages.
2119 len = (u64)LLONG_MAX + 1;
2122 * We write the dirty pages in the range and wait until they complete
2123 * out of the ->i_mutex. If so, we can flush the dirty pages by
2124 * multi-task, and make the performance up. See
2125 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
2127 ret = start_ordered_ops(inode, start, end);
2133 atomic_inc(&root->log_batch);
2136 * Always check for the full sync flag while holding the inode's lock,
2137 * to avoid races with other tasks. The flag must be either set all the
2138 * time during logging or always off all the time while logging.
2140 full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2141 &BTRFS_I(inode)->runtime_flags);
2144 * Before we acquired the inode's lock, someone may have dirtied more
2145 * pages in the target range. We need to make sure that writeback for
2146 * any such pages does not start while we are logging the inode, because
2147 * if it does, any of the following might happen when we are not doing a
2150 * 1) We log an extent after its writeback finishes but before its
2151 * checksums are added to the csum tree, leading to -EIO errors
2152 * when attempting to read the extent after a log replay.
2154 * 2) We can end up logging an extent before its writeback finishes.
2155 * Therefore after the log replay we will have a file extent item
2156 * pointing to an unwritten extent (and no data checksums as well).
2158 * So trigger writeback for any eventual new dirty pages and then we
2159 * wait for all ordered extents to complete below.
2161 ret = start_ordered_ops(inode, start, end);
2163 inode_unlock(inode);
2168 * We have to do this here to avoid the priority inversion of waiting on
2169 * IO of a lower priority task while holding a transaction open.
2171 * For a full fsync we wait for the ordered extents to complete while
2172 * for a fast fsync we wait just for writeback to complete, and then
2173 * attach the ordered extents to the transaction so that a transaction
2174 * commit waits for their completion, to avoid data loss if we fsync,
2175 * the current transaction commits before the ordered extents complete
2176 * and a power failure happens right after that.
2179 ret = btrfs_wait_ordered_range(inode, start, len);
2182 * Get our ordered extents as soon as possible to avoid doing
2183 * checksum lookups in the csum tree, and use instead the
2184 * checksums attached to the ordered extents.
2186 btrfs_get_ordered_extents_for_logging(BTRFS_I(inode),
2187 &ctx.ordered_extents);
2188 ret = filemap_fdatawait_range(inode->i_mapping, start, end);
2192 goto out_release_extents;
2194 atomic_inc(&root->log_batch);
2197 * If we are doing a fast fsync we can not bail out if the inode's
2198 * last_trans is <= then the last committed transaction, because we only
2199 * update the last_trans of the inode during ordered extent completion,
2200 * and for a fast fsync we don't wait for that, we only wait for the
2201 * writeback to complete.
2204 if (btrfs_inode_in_log(BTRFS_I(inode), fs_info->generation) ||
2205 (BTRFS_I(inode)->last_trans <= fs_info->last_trans_committed &&
2206 (full_sync || list_empty(&ctx.ordered_extents)))) {
2208 * We've had everything committed since the last time we were
2209 * modified so clear this flag in case it was set for whatever
2210 * reason, it's no longer relevant.
2212 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2213 &BTRFS_I(inode)->runtime_flags);
2215 * An ordered extent might have started before and completed
2216 * already with io errors, in which case the inode was not
2217 * updated and we end up here. So check the inode's mapping
2218 * for any errors that might have happened since we last
2219 * checked called fsync.
2221 ret = filemap_check_wb_err(inode->i_mapping, file->f_wb_err);
2222 goto out_release_extents;
2226 * We use start here because we will need to wait on the IO to complete
2227 * in btrfs_sync_log, which could require joining a transaction (for
2228 * example checking cross references in the nocow path). If we use join
2229 * here we could get into a situation where we're waiting on IO to
2230 * happen that is blocked on a transaction trying to commit. With start
2231 * we inc the extwriter counter, so we wait for all extwriters to exit
2232 * before we start blocking joiners. This comment is to keep somebody
2233 * from thinking they are super smart and changing this to
2234 * btrfs_join_transaction *cough*Josef*cough*.
2236 trans = btrfs_start_transaction(root, 0);
2237 if (IS_ERR(trans)) {
2238 ret = PTR_ERR(trans);
2239 goto out_release_extents;
2242 ret = btrfs_log_dentry_safe(trans, dentry, &ctx);
2243 btrfs_release_log_ctx_extents(&ctx);
2245 /* Fallthrough and commit/free transaction. */
2249 /* we've logged all the items and now have a consistent
2250 * version of the file in the log. It is possible that
2251 * someone will come in and modify the file, but that's
2252 * fine because the log is consistent on disk, and we
2253 * have references to all of the file's extents
2255 * It is possible that someone will come in and log the
2256 * file again, but that will end up using the synchronization
2257 * inside btrfs_sync_log to keep things safe.
2259 inode_unlock(inode);
2261 if (ret != BTRFS_NO_LOG_SYNC) {
2263 ret = btrfs_sync_log(trans, root, &ctx);
2265 ret = btrfs_end_transaction(trans);
2270 ret = btrfs_wait_ordered_range(inode, start, len);
2272 btrfs_end_transaction(trans);
2276 ret = btrfs_commit_transaction(trans);
2278 ret = btrfs_end_transaction(trans);
2281 ASSERT(list_empty(&ctx.list));
2282 err = file_check_and_advance_wb_err(file);
2285 return ret > 0 ? -EIO : ret;
2287 out_release_extents:
2288 btrfs_release_log_ctx_extents(&ctx);
2289 inode_unlock(inode);
2293 static const struct vm_operations_struct btrfs_file_vm_ops = {
2294 .fault = filemap_fault,
2295 .map_pages = filemap_map_pages,
2296 .page_mkwrite = btrfs_page_mkwrite,
2299 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
2301 struct address_space *mapping = filp->f_mapping;
2303 if (!mapping->a_ops->readpage)
2306 file_accessed(filp);
2307 vma->vm_ops = &btrfs_file_vm_ops;
2312 static int hole_mergeable(struct btrfs_inode *inode, struct extent_buffer *leaf,
2313 int slot, u64 start, u64 end)
2315 struct btrfs_file_extent_item *fi;
2316 struct btrfs_key key;
2318 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
2321 btrfs_item_key_to_cpu(leaf, &key, slot);
2322 if (key.objectid != btrfs_ino(inode) ||
2323 key.type != BTRFS_EXTENT_DATA_KEY)
2326 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2328 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2331 if (btrfs_file_extent_disk_bytenr(leaf, fi))
2334 if (key.offset == end)
2336 if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
2341 static int fill_holes(struct btrfs_trans_handle *trans,
2342 struct btrfs_inode *inode,
2343 struct btrfs_path *path, u64 offset, u64 end)
2345 struct btrfs_fs_info *fs_info = trans->fs_info;
2346 struct btrfs_root *root = inode->root;
2347 struct extent_buffer *leaf;
2348 struct btrfs_file_extent_item *fi;
2349 struct extent_map *hole_em;
2350 struct extent_map_tree *em_tree = &inode->extent_tree;
2351 struct btrfs_key key;
2354 if (btrfs_fs_incompat(fs_info, NO_HOLES))
2357 key.objectid = btrfs_ino(inode);
2358 key.type = BTRFS_EXTENT_DATA_KEY;
2359 key.offset = offset;
2361 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2364 * We should have dropped this offset, so if we find it then
2365 * something has gone horribly wrong.
2372 leaf = path->nodes[0];
2373 if (hole_mergeable(inode, leaf, path->slots[0] - 1, offset, end)) {
2377 fi = btrfs_item_ptr(leaf, path->slots[0],
2378 struct btrfs_file_extent_item);
2379 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2381 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2382 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2383 btrfs_set_file_extent_offset(leaf, fi, 0);
2384 btrfs_mark_buffer_dirty(leaf);
2388 if (hole_mergeable(inode, leaf, path->slots[0], offset, end)) {
2391 key.offset = offset;
2392 btrfs_set_item_key_safe(fs_info, path, &key);
2393 fi = btrfs_item_ptr(leaf, path->slots[0],
2394 struct btrfs_file_extent_item);
2395 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2397 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2398 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2399 btrfs_set_file_extent_offset(leaf, fi, 0);
2400 btrfs_mark_buffer_dirty(leaf);
2403 btrfs_release_path(path);
2405 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode),
2406 offset, 0, 0, end - offset, 0, end - offset, 0, 0, 0);
2411 btrfs_release_path(path);
2413 hole_em = alloc_extent_map();
2415 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2416 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
2418 hole_em->start = offset;
2419 hole_em->len = end - offset;
2420 hole_em->ram_bytes = hole_em->len;
2421 hole_em->orig_start = offset;
2423 hole_em->block_start = EXTENT_MAP_HOLE;
2424 hole_em->block_len = 0;
2425 hole_em->orig_block_len = 0;
2426 hole_em->compress_type = BTRFS_COMPRESS_NONE;
2427 hole_em->generation = trans->transid;
2430 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2431 write_lock(&em_tree->lock);
2432 ret = add_extent_mapping(em_tree, hole_em, 1);
2433 write_unlock(&em_tree->lock);
2434 } while (ret == -EEXIST);
2435 free_extent_map(hole_em);
2437 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2438 &inode->runtime_flags);
2445 * Find a hole extent on given inode and change start/len to the end of hole
2446 * extent.(hole/vacuum extent whose em->start <= start &&
2447 * em->start + em->len > start)
2448 * When a hole extent is found, return 1 and modify start/len.
2450 static int find_first_non_hole(struct btrfs_inode *inode, u64 *start, u64 *len)
2452 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2453 struct extent_map *em;
2456 em = btrfs_get_extent(inode, NULL, 0,
2457 round_down(*start, fs_info->sectorsize),
2458 round_up(*len, fs_info->sectorsize));
2462 /* Hole or vacuum extent(only exists in no-hole mode) */
2463 if (em->block_start == EXTENT_MAP_HOLE) {
2465 *len = em->start + em->len > *start + *len ?
2466 0 : *start + *len - em->start - em->len;
2467 *start = em->start + em->len;
2469 free_extent_map(em);
2473 static int btrfs_punch_hole_lock_range(struct inode *inode,
2474 const u64 lockstart,
2476 struct extent_state **cached_state)
2479 struct btrfs_ordered_extent *ordered;
2482 truncate_pagecache_range(inode, lockstart, lockend);
2484 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2486 ordered = btrfs_lookup_first_ordered_extent(BTRFS_I(inode),
2490 * We need to make sure we have no ordered extents in this range
2491 * and nobody raced in and read a page in this range, if we did
2492 * we need to try again.
2495 (ordered->file_offset + ordered->num_bytes <= lockstart ||
2496 ordered->file_offset > lockend)) &&
2497 !filemap_range_has_page(inode->i_mapping,
2498 lockstart, lockend)) {
2500 btrfs_put_ordered_extent(ordered);
2504 btrfs_put_ordered_extent(ordered);
2505 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
2506 lockend, cached_state);
2507 ret = btrfs_wait_ordered_range(inode, lockstart,
2508 lockend - lockstart + 1);
2515 static int btrfs_insert_replace_extent(struct btrfs_trans_handle *trans,
2516 struct btrfs_inode *inode,
2517 struct btrfs_path *path,
2518 struct btrfs_replace_extent_info *extent_info,
2519 const u64 replace_len,
2520 const u64 bytes_to_drop)
2522 struct btrfs_fs_info *fs_info = trans->fs_info;
2523 struct btrfs_root *root = inode->root;
2524 struct btrfs_file_extent_item *extent;
2525 struct extent_buffer *leaf;
2526 struct btrfs_key key;
2528 struct btrfs_ref ref = { 0 };
2531 if (replace_len == 0)
2534 if (extent_info->disk_offset == 0 &&
2535 btrfs_fs_incompat(fs_info, NO_HOLES)) {
2536 btrfs_update_inode_bytes(inode, 0, bytes_to_drop);
2540 key.objectid = btrfs_ino(inode);
2541 key.type = BTRFS_EXTENT_DATA_KEY;
2542 key.offset = extent_info->file_offset;
2543 ret = btrfs_insert_empty_item(trans, root, path, &key,
2544 sizeof(struct btrfs_file_extent_item));
2547 leaf = path->nodes[0];
2548 slot = path->slots[0];
2549 write_extent_buffer(leaf, extent_info->extent_buf,
2550 btrfs_item_ptr_offset(leaf, slot),
2551 sizeof(struct btrfs_file_extent_item));
2552 extent = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2553 ASSERT(btrfs_file_extent_type(leaf, extent) != BTRFS_FILE_EXTENT_INLINE);
2554 btrfs_set_file_extent_offset(leaf, extent, extent_info->data_offset);
2555 btrfs_set_file_extent_num_bytes(leaf, extent, replace_len);
2556 if (extent_info->is_new_extent)
2557 btrfs_set_file_extent_generation(leaf, extent, trans->transid);
2558 btrfs_mark_buffer_dirty(leaf);
2559 btrfs_release_path(path);
2561 ret = btrfs_inode_set_file_extent_range(inode, extent_info->file_offset,
2566 /* If it's a hole, nothing more needs to be done. */
2567 if (extent_info->disk_offset == 0) {
2568 btrfs_update_inode_bytes(inode, 0, bytes_to_drop);
2572 btrfs_update_inode_bytes(inode, replace_len, bytes_to_drop);
2574 if (extent_info->is_new_extent && extent_info->insertions == 0) {
2575 key.objectid = extent_info->disk_offset;
2576 key.type = BTRFS_EXTENT_ITEM_KEY;
2577 key.offset = extent_info->disk_len;
2578 ret = btrfs_alloc_reserved_file_extent(trans, root,
2580 extent_info->file_offset,
2581 extent_info->qgroup_reserved,
2586 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF,
2587 extent_info->disk_offset,
2588 extent_info->disk_len, 0);
2589 ref_offset = extent_info->file_offset - extent_info->data_offset;
2590 btrfs_init_data_ref(&ref, root->root_key.objectid,
2591 btrfs_ino(inode), ref_offset);
2592 ret = btrfs_inc_extent_ref(trans, &ref);
2595 extent_info->insertions++;
2601 * The respective range must have been previously locked, as well as the inode.
2602 * The end offset is inclusive (last byte of the range).
2603 * @extent_info is NULL for fallocate's hole punching and non-NULL when replacing
2604 * the file range with an extent.
2605 * When not punching a hole, we don't want to end up in a state where we dropped
2606 * extents without inserting a new one, so we must abort the transaction to avoid
2609 int btrfs_replace_file_extents(struct inode *inode, struct btrfs_path *path,
2610 const u64 start, const u64 end,
2611 struct btrfs_replace_extent_info *extent_info,
2612 struct btrfs_trans_handle **trans_out)
2614 struct btrfs_drop_extents_args drop_args = { 0 };
2615 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2616 u64 min_size = btrfs_calc_insert_metadata_size(fs_info, 1);
2617 u64 ino_size = round_up(inode->i_size, fs_info->sectorsize);
2618 struct btrfs_root *root = BTRFS_I(inode)->root;
2619 struct btrfs_trans_handle *trans = NULL;
2620 struct btrfs_block_rsv *rsv;
2621 unsigned int rsv_count;
2623 u64 len = end - start;
2629 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
2634 rsv->size = btrfs_calc_insert_metadata_size(fs_info, 1);
2638 * 1 - update the inode
2639 * 1 - removing the extents in the range
2640 * 1 - adding the hole extent if no_holes isn't set or if we are
2641 * replacing the range with a new extent
2643 if (!btrfs_fs_incompat(fs_info, NO_HOLES) || extent_info)
2648 trans = btrfs_start_transaction(root, rsv_count);
2649 if (IS_ERR(trans)) {
2650 ret = PTR_ERR(trans);
2655 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
2658 trans->block_rsv = rsv;
2661 drop_args.path = path;
2662 drop_args.end = end + 1;
2663 drop_args.drop_cache = true;
2664 while (cur_offset < end) {
2665 drop_args.start = cur_offset;
2666 ret = btrfs_drop_extents(trans, root, BTRFS_I(inode), &drop_args);
2667 /* If we are punching a hole decrement the inode's byte count */
2669 btrfs_update_inode_bytes(BTRFS_I(inode), 0,
2670 drop_args.bytes_found);
2671 if (ret != -ENOSPC) {
2673 * When cloning we want to avoid transaction aborts when
2674 * nothing was done and we are attempting to clone parts
2675 * of inline extents, in such cases -EOPNOTSUPP is
2676 * returned by __btrfs_drop_extents() without having
2677 * changed anything in the file.
2679 if (extent_info && !extent_info->is_new_extent &&
2680 ret && ret != -EOPNOTSUPP)
2681 btrfs_abort_transaction(trans, ret);
2685 trans->block_rsv = &fs_info->trans_block_rsv;
2687 if (!extent_info && cur_offset < drop_args.drop_end &&
2688 cur_offset < ino_size) {
2689 ret = fill_holes(trans, BTRFS_I(inode), path,
2690 cur_offset, drop_args.drop_end);
2693 * If we failed then we didn't insert our hole
2694 * entries for the area we dropped, so now the
2695 * fs is corrupted, so we must abort the
2698 btrfs_abort_transaction(trans, ret);
2701 } else if (!extent_info && cur_offset < drop_args.drop_end) {
2703 * We are past the i_size here, but since we didn't
2704 * insert holes we need to clear the mapped area so we
2705 * know to not set disk_i_size in this area until a new
2706 * file extent is inserted here.
2708 ret = btrfs_inode_clear_file_extent_range(BTRFS_I(inode),
2710 drop_args.drop_end - cur_offset);
2713 * We couldn't clear our area, so we could
2714 * presumably adjust up and corrupt the fs, so
2717 btrfs_abort_transaction(trans, ret);
2723 drop_args.drop_end > extent_info->file_offset) {
2724 u64 replace_len = drop_args.drop_end -
2725 extent_info->file_offset;
2727 ret = btrfs_insert_replace_extent(trans, BTRFS_I(inode),
2728 path, extent_info, replace_len,
2729 drop_args.bytes_found);
2731 btrfs_abort_transaction(trans, ret);
2734 extent_info->data_len -= replace_len;
2735 extent_info->data_offset += replace_len;
2736 extent_info->file_offset += replace_len;
2739 cur_offset = drop_args.drop_end;
2741 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
2745 btrfs_end_transaction(trans);
2746 btrfs_btree_balance_dirty(fs_info);
2748 trans = btrfs_start_transaction(root, rsv_count);
2749 if (IS_ERR(trans)) {
2750 ret = PTR_ERR(trans);
2755 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
2756 rsv, min_size, false);
2757 BUG_ON(ret); /* shouldn't happen */
2758 trans->block_rsv = rsv;
2761 ret = find_first_non_hole(BTRFS_I(inode), &cur_offset,
2763 if (unlikely(ret < 0))
2773 * If we were cloning, force the next fsync to be a full one since we
2774 * we replaced (or just dropped in the case of cloning holes when
2775 * NO_HOLES is enabled) extents and extent maps.
2776 * This is for the sake of simplicity, and cloning into files larger
2777 * than 16Mb would force the full fsync any way (when
2778 * try_release_extent_mapping() is invoked during page cache truncation.
2780 if (extent_info && !extent_info->is_new_extent)
2781 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2782 &BTRFS_I(inode)->runtime_flags);
2787 trans->block_rsv = &fs_info->trans_block_rsv;
2789 * If we are using the NO_HOLES feature we might have had already an
2790 * hole that overlaps a part of the region [lockstart, lockend] and
2791 * ends at (or beyond) lockend. Since we have no file extent items to
2792 * represent holes, drop_end can be less than lockend and so we must
2793 * make sure we have an extent map representing the existing hole (the
2794 * call to __btrfs_drop_extents() might have dropped the existing extent
2795 * map representing the existing hole), otherwise the fast fsync path
2796 * will not record the existence of the hole region
2797 * [existing_hole_start, lockend].
2799 if (drop_args.drop_end <= end)
2800 drop_args.drop_end = end + 1;
2802 * Don't insert file hole extent item if it's for a range beyond eof
2803 * (because it's useless) or if it represents a 0 bytes range (when
2804 * cur_offset == drop_end).
2806 if (!extent_info && cur_offset < ino_size &&
2807 cur_offset < drop_args.drop_end) {
2808 ret = fill_holes(trans, BTRFS_I(inode), path,
2809 cur_offset, drop_args.drop_end);
2811 /* Same comment as above. */
2812 btrfs_abort_transaction(trans, ret);
2815 } else if (!extent_info && cur_offset < drop_args.drop_end) {
2816 /* See the comment in the loop above for the reasoning here. */
2817 ret = btrfs_inode_clear_file_extent_range(BTRFS_I(inode),
2818 cur_offset, drop_args.drop_end - cur_offset);
2820 btrfs_abort_transaction(trans, ret);
2826 ret = btrfs_insert_replace_extent(trans, BTRFS_I(inode), path,
2827 extent_info, extent_info->data_len,
2828 drop_args.bytes_found);
2830 btrfs_abort_transaction(trans, ret);
2839 trans->block_rsv = &fs_info->trans_block_rsv;
2841 btrfs_end_transaction(trans);
2845 btrfs_free_block_rsv(fs_info, rsv);
2850 static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
2852 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2853 struct btrfs_root *root = BTRFS_I(inode)->root;
2854 struct extent_state *cached_state = NULL;
2855 struct btrfs_path *path;
2856 struct btrfs_trans_handle *trans = NULL;
2861 u64 orig_start = offset;
2865 bool truncated_block = false;
2866 bool updated_inode = false;
2868 ret = btrfs_wait_ordered_range(inode, offset, len);
2873 ino_size = round_up(inode->i_size, fs_info->sectorsize);
2874 ret = find_first_non_hole(BTRFS_I(inode), &offset, &len);
2876 goto out_only_mutex;
2878 /* Already in a large hole */
2880 goto out_only_mutex;
2883 lockstart = round_up(offset, btrfs_inode_sectorsize(BTRFS_I(inode)));
2884 lockend = round_down(offset + len,
2885 btrfs_inode_sectorsize(BTRFS_I(inode))) - 1;
2886 same_block = (BTRFS_BYTES_TO_BLKS(fs_info, offset))
2887 == (BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1));
2889 * We needn't truncate any block which is beyond the end of the file
2890 * because we are sure there is no data there.
2893 * Only do this if we are in the same block and we aren't doing the
2896 if (same_block && len < fs_info->sectorsize) {
2897 if (offset < ino_size) {
2898 truncated_block = true;
2899 ret = btrfs_truncate_block(BTRFS_I(inode), offset, len,
2904 goto out_only_mutex;
2907 /* zero back part of the first block */
2908 if (offset < ino_size) {
2909 truncated_block = true;
2910 ret = btrfs_truncate_block(BTRFS_I(inode), offset, 0, 0);
2912 inode_unlock(inode);
2917 /* Check the aligned pages after the first unaligned page,
2918 * if offset != orig_start, which means the first unaligned page
2919 * including several following pages are already in holes,
2920 * the extra check can be skipped */
2921 if (offset == orig_start) {
2922 /* after truncate page, check hole again */
2923 len = offset + len - lockstart;
2925 ret = find_first_non_hole(BTRFS_I(inode), &offset, &len);
2927 goto out_only_mutex;
2930 goto out_only_mutex;
2935 /* Check the tail unaligned part is in a hole */
2936 tail_start = lockend + 1;
2937 tail_len = offset + len - tail_start;
2939 ret = find_first_non_hole(BTRFS_I(inode), &tail_start, &tail_len);
2940 if (unlikely(ret < 0))
2941 goto out_only_mutex;
2943 /* zero the front end of the last page */
2944 if (tail_start + tail_len < ino_size) {
2945 truncated_block = true;
2946 ret = btrfs_truncate_block(BTRFS_I(inode),
2947 tail_start + tail_len,
2950 goto out_only_mutex;
2955 if (lockend < lockstart) {
2957 goto out_only_mutex;
2960 ret = btrfs_punch_hole_lock_range(inode, lockstart, lockend,
2963 goto out_only_mutex;
2965 path = btrfs_alloc_path();
2971 ret = btrfs_replace_file_extents(inode, path, lockstart, lockend, NULL,
2973 btrfs_free_path(path);
2977 ASSERT(trans != NULL);
2978 inode_inc_iversion(inode);
2979 inode->i_mtime = inode->i_ctime = current_time(inode);
2980 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
2981 updated_inode = true;
2982 btrfs_end_transaction(trans);
2983 btrfs_btree_balance_dirty(fs_info);
2985 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2988 if (!updated_inode && truncated_block && !ret) {
2990 * If we only end up zeroing part of a page, we still need to
2991 * update the inode item, so that all the time fields are
2992 * updated as well as the necessary btrfs inode in memory fields
2993 * for detecting, at fsync time, if the inode isn't yet in the
2994 * log tree or it's there but not up to date.
2996 struct timespec64 now = current_time(inode);
2998 inode_inc_iversion(inode);
2999 inode->i_mtime = now;
3000 inode->i_ctime = now;
3001 trans = btrfs_start_transaction(root, 1);
3002 if (IS_ERR(trans)) {
3003 ret = PTR_ERR(trans);
3007 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
3008 ret2 = btrfs_end_transaction(trans);
3013 inode_unlock(inode);
3017 /* Helper structure to record which range is already reserved */
3018 struct falloc_range {
3019 struct list_head list;
3025 * Helper function to add falloc range
3027 * Caller should have locked the larger range of extent containing
3030 static int add_falloc_range(struct list_head *head, u64 start, u64 len)
3032 struct falloc_range *prev = NULL;
3033 struct falloc_range *range = NULL;
3035 if (list_empty(head))
3039 * As fallocate iterate by bytenr order, we only need to check
3042 prev = list_entry(head->prev, struct falloc_range, list);
3043 if (prev->start + prev->len == start) {
3048 range = kmalloc(sizeof(*range), GFP_KERNEL);
3051 range->start = start;
3053 list_add_tail(&range->list, head);
3057 static int btrfs_fallocate_update_isize(struct inode *inode,
3061 struct btrfs_trans_handle *trans;
3062 struct btrfs_root *root = BTRFS_I(inode)->root;
3066 if (mode & FALLOC_FL_KEEP_SIZE || end <= i_size_read(inode))
3069 trans = btrfs_start_transaction(root, 1);
3071 return PTR_ERR(trans);
3073 inode->i_ctime = current_time(inode);
3074 i_size_write(inode, end);
3075 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
3076 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
3077 ret2 = btrfs_end_transaction(trans);
3079 return ret ? ret : ret2;
3083 RANGE_BOUNDARY_WRITTEN_EXTENT,
3084 RANGE_BOUNDARY_PREALLOC_EXTENT,
3085 RANGE_BOUNDARY_HOLE,
3088 static int btrfs_zero_range_check_range_boundary(struct btrfs_inode *inode,
3091 const u64 sectorsize = btrfs_inode_sectorsize(inode);
3092 struct extent_map *em;
3095 offset = round_down(offset, sectorsize);
3096 em = btrfs_get_extent(inode, NULL, 0, offset, sectorsize);
3100 if (em->block_start == EXTENT_MAP_HOLE)
3101 ret = RANGE_BOUNDARY_HOLE;
3102 else if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
3103 ret = RANGE_BOUNDARY_PREALLOC_EXTENT;
3105 ret = RANGE_BOUNDARY_WRITTEN_EXTENT;
3107 free_extent_map(em);
3111 static int btrfs_zero_range(struct inode *inode,
3116 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3117 struct extent_map *em;
3118 struct extent_changeset *data_reserved = NULL;
3121 const u64 sectorsize = btrfs_inode_sectorsize(BTRFS_I(inode));
3122 u64 alloc_start = round_down(offset, sectorsize);
3123 u64 alloc_end = round_up(offset + len, sectorsize);
3124 u64 bytes_to_reserve = 0;
3125 bool space_reserved = false;
3127 inode_dio_wait(inode);
3129 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, alloc_start,
3130 alloc_end - alloc_start);
3137 * Avoid hole punching and extent allocation for some cases. More cases
3138 * could be considered, but these are unlikely common and we keep things
3139 * as simple as possible for now. Also, intentionally, if the target
3140 * range contains one or more prealloc extents together with regular
3141 * extents and holes, we drop all the existing extents and allocate a
3142 * new prealloc extent, so that we get a larger contiguous disk extent.
3144 if (em->start <= alloc_start &&
3145 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3146 const u64 em_end = em->start + em->len;
3148 if (em_end >= offset + len) {
3150 * The whole range is already a prealloc extent,
3151 * do nothing except updating the inode's i_size if
3154 free_extent_map(em);
3155 ret = btrfs_fallocate_update_isize(inode, offset + len,
3160 * Part of the range is already a prealloc extent, so operate
3161 * only on the remaining part of the range.
3163 alloc_start = em_end;
3164 ASSERT(IS_ALIGNED(alloc_start, sectorsize));
3165 len = offset + len - alloc_start;
3166 offset = alloc_start;
3167 alloc_hint = em->block_start + em->len;
3169 free_extent_map(em);
3171 if (BTRFS_BYTES_TO_BLKS(fs_info, offset) ==
3172 BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1)) {
3173 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, alloc_start,
3180 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3181 free_extent_map(em);
3182 ret = btrfs_fallocate_update_isize(inode, offset + len,
3186 if (len < sectorsize && em->block_start != EXTENT_MAP_HOLE) {
3187 free_extent_map(em);
3188 ret = btrfs_truncate_block(BTRFS_I(inode), offset, len,
3191 ret = btrfs_fallocate_update_isize(inode,
3196 free_extent_map(em);
3197 alloc_start = round_down(offset, sectorsize);
3198 alloc_end = alloc_start + sectorsize;
3202 alloc_start = round_up(offset, sectorsize);
3203 alloc_end = round_down(offset + len, sectorsize);
3206 * For unaligned ranges, check the pages at the boundaries, they might
3207 * map to an extent, in which case we need to partially zero them, or
3208 * they might map to a hole, in which case we need our allocation range
3211 if (!IS_ALIGNED(offset, sectorsize)) {
3212 ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode),
3216 if (ret == RANGE_BOUNDARY_HOLE) {
3217 alloc_start = round_down(offset, sectorsize);
3219 } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) {
3220 ret = btrfs_truncate_block(BTRFS_I(inode), offset, 0, 0);
3228 if (!IS_ALIGNED(offset + len, sectorsize)) {
3229 ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode),
3233 if (ret == RANGE_BOUNDARY_HOLE) {
3234 alloc_end = round_up(offset + len, sectorsize);
3236 } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) {
3237 ret = btrfs_truncate_block(BTRFS_I(inode), offset + len,
3247 if (alloc_start < alloc_end) {
3248 struct extent_state *cached_state = NULL;
3249 const u64 lockstart = alloc_start;
3250 const u64 lockend = alloc_end - 1;
3252 bytes_to_reserve = alloc_end - alloc_start;
3253 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode),
3257 space_reserved = true;
3258 ret = btrfs_punch_hole_lock_range(inode, lockstart, lockend,
3262 ret = btrfs_qgroup_reserve_data(BTRFS_I(inode), &data_reserved,
3263 alloc_start, bytes_to_reserve);
3266 ret = btrfs_prealloc_file_range(inode, mode, alloc_start,
3267 alloc_end - alloc_start,
3269 offset + len, &alloc_hint);
3270 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
3271 lockend, &cached_state);
3272 /* btrfs_prealloc_file_range releases reserved space on error */
3274 space_reserved = false;
3278 ret = btrfs_fallocate_update_isize(inode, offset + len, mode);
3280 if (ret && space_reserved)
3281 btrfs_free_reserved_data_space(BTRFS_I(inode), data_reserved,
3282 alloc_start, bytes_to_reserve);
3283 extent_changeset_free(data_reserved);
3288 static long btrfs_fallocate(struct file *file, int mode,
3289 loff_t offset, loff_t len)
3291 struct inode *inode = file_inode(file);
3292 struct extent_state *cached_state = NULL;
3293 struct extent_changeset *data_reserved = NULL;
3294 struct falloc_range *range;
3295 struct falloc_range *tmp;
3296 struct list_head reserve_list;
3304 struct extent_map *em;
3305 int blocksize = btrfs_inode_sectorsize(BTRFS_I(inode));
3308 /* Do not allow fallocate in ZONED mode */
3309 if (btrfs_is_zoned(btrfs_sb(inode->i_sb)))
3312 alloc_start = round_down(offset, blocksize);
3313 alloc_end = round_up(offset + len, blocksize);
3314 cur_offset = alloc_start;
3316 /* Make sure we aren't being give some crap mode */
3317 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
3318 FALLOC_FL_ZERO_RANGE))
3321 if (mode & FALLOC_FL_PUNCH_HOLE)
3322 return btrfs_punch_hole(inode, offset, len);
3325 * Only trigger disk allocation, don't trigger qgroup reserve
3327 * For qgroup space, it will be checked later.
3329 if (!(mode & FALLOC_FL_ZERO_RANGE)) {
3330 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode),
3331 alloc_end - alloc_start);
3336 btrfs_inode_lock(inode, 0);
3338 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) {
3339 ret = inode_newsize_ok(inode, offset + len);
3345 * TODO: Move these two operations after we have checked
3346 * accurate reserved space, or fallocate can still fail but
3347 * with page truncated or size expanded.
3349 * But that's a minor problem and won't do much harm BTW.
3351 if (alloc_start > inode->i_size) {
3352 ret = btrfs_cont_expand(BTRFS_I(inode), i_size_read(inode),
3356 } else if (offset + len > inode->i_size) {
3358 * If we are fallocating from the end of the file onward we
3359 * need to zero out the end of the block if i_size lands in the
3360 * middle of a block.
3362 ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0);
3368 * wait for ordered IO before we have any locks. We'll loop again
3369 * below with the locks held.
3371 ret = btrfs_wait_ordered_range(inode, alloc_start,
3372 alloc_end - alloc_start);
3376 if (mode & FALLOC_FL_ZERO_RANGE) {
3377 ret = btrfs_zero_range(inode, offset, len, mode);
3378 inode_unlock(inode);
3382 locked_end = alloc_end - 1;
3384 struct btrfs_ordered_extent *ordered;
3386 /* the extent lock is ordered inside the running
3389 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
3390 locked_end, &cached_state);
3391 ordered = btrfs_lookup_first_ordered_extent(BTRFS_I(inode),
3395 ordered->file_offset + ordered->num_bytes > alloc_start &&
3396 ordered->file_offset < alloc_end) {
3397 btrfs_put_ordered_extent(ordered);
3398 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
3399 alloc_start, locked_end,
3402 * we can't wait on the range with the transaction
3403 * running or with the extent lock held
3405 ret = btrfs_wait_ordered_range(inode, alloc_start,
3406 alloc_end - alloc_start);
3411 btrfs_put_ordered_extent(ordered);
3416 /* First, check if we exceed the qgroup limit */
3417 INIT_LIST_HEAD(&reserve_list);
3418 while (cur_offset < alloc_end) {
3419 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
3420 alloc_end - cur_offset);
3425 last_byte = min(extent_map_end(em), alloc_end);
3426 actual_end = min_t(u64, extent_map_end(em), offset + len);
3427 last_byte = ALIGN(last_byte, blocksize);
3428 if (em->block_start == EXTENT_MAP_HOLE ||
3429 (cur_offset >= inode->i_size &&
3430 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
3431 ret = add_falloc_range(&reserve_list, cur_offset,
3432 last_byte - cur_offset);
3434 free_extent_map(em);
3437 ret = btrfs_qgroup_reserve_data(BTRFS_I(inode),
3438 &data_reserved, cur_offset,
3439 last_byte - cur_offset);
3441 cur_offset = last_byte;
3442 free_extent_map(em);
3447 * Do not need to reserve unwritten extent for this
3448 * range, free reserved data space first, otherwise
3449 * it'll result in false ENOSPC error.
3451 btrfs_free_reserved_data_space(BTRFS_I(inode),
3452 data_reserved, cur_offset,
3453 last_byte - cur_offset);
3455 free_extent_map(em);
3456 cur_offset = last_byte;
3460 * If ret is still 0, means we're OK to fallocate.
3461 * Or just cleanup the list and exit.
3463 list_for_each_entry_safe(range, tmp, &reserve_list, list) {
3465 ret = btrfs_prealloc_file_range(inode, mode,
3467 range->len, i_blocksize(inode),
3468 offset + len, &alloc_hint);
3470 btrfs_free_reserved_data_space(BTRFS_I(inode),
3471 data_reserved, range->start,
3473 list_del(&range->list);
3480 * We didn't need to allocate any more space, but we still extended the
3481 * size of the file so we need to update i_size and the inode item.
3483 ret = btrfs_fallocate_update_isize(inode, actual_end, mode);
3485 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
3488 inode_unlock(inode);
3489 /* Let go of our reservation. */
3490 if (ret != 0 && !(mode & FALLOC_FL_ZERO_RANGE))
3491 btrfs_free_reserved_data_space(BTRFS_I(inode), data_reserved,
3492 cur_offset, alloc_end - cur_offset);
3493 extent_changeset_free(data_reserved);
3497 static loff_t find_desired_extent(struct inode *inode, loff_t offset,
3500 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3501 struct extent_map *em = NULL;
3502 struct extent_state *cached_state = NULL;
3503 loff_t i_size = inode->i_size;
3510 if (i_size == 0 || offset >= i_size)
3514 * offset can be negative, in this case we start finding DATA/HOLE from
3515 * the very start of the file.
3517 start = max_t(loff_t, 0, offset);
3519 lockstart = round_down(start, fs_info->sectorsize);
3520 lockend = round_up(i_size, fs_info->sectorsize);
3521 if (lockend <= lockstart)
3522 lockend = lockstart + fs_info->sectorsize;
3524 len = lockend - lockstart + 1;
3526 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
3529 while (start < i_size) {
3530 em = btrfs_get_extent_fiemap(BTRFS_I(inode), start, len);
3537 if (whence == SEEK_HOLE &&
3538 (em->block_start == EXTENT_MAP_HOLE ||
3539 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
3541 else if (whence == SEEK_DATA &&
3542 (em->block_start != EXTENT_MAP_HOLE &&
3543 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
3546 start = em->start + em->len;
3547 free_extent_map(em);
3551 free_extent_map(em);
3552 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
3557 if (whence == SEEK_DATA && start >= i_size)
3560 offset = min_t(loff_t, start, i_size);
3566 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
3568 struct inode *inode = file->f_mapping->host;
3572 return generic_file_llseek(file, offset, whence);
3575 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
3576 offset = find_desired_extent(inode, offset, whence);
3577 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
3584 return vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
3587 static int btrfs_file_open(struct inode *inode, struct file *filp)
3589 filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC;
3590 return generic_file_open(inode, filp);
3593 static int check_direct_read(struct btrfs_fs_info *fs_info,
3594 const struct iov_iter *iter, loff_t offset)
3599 ret = check_direct_IO(fs_info, iter, offset);
3603 if (!iter_is_iovec(iter))
3606 for (seg = 0; seg < iter->nr_segs; seg++)
3607 for (i = seg + 1; i < iter->nr_segs; i++)
3608 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
3613 static ssize_t btrfs_direct_read(struct kiocb *iocb, struct iov_iter *to)
3615 struct inode *inode = file_inode(iocb->ki_filp);
3618 if (check_direct_read(btrfs_sb(inode->i_sb), to, iocb->ki_pos))
3621 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
3622 ret = iomap_dio_rw(iocb, to, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
3623 is_sync_kiocb(iocb));
3624 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
3628 static ssize_t btrfs_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
3632 if (iocb->ki_flags & IOCB_DIRECT) {
3633 ret = btrfs_direct_read(iocb, to);
3634 if (ret < 0 || !iov_iter_count(to) ||
3635 iocb->ki_pos >= i_size_read(file_inode(iocb->ki_filp)))
3639 return generic_file_buffered_read(iocb, to, ret);
3642 const struct file_operations btrfs_file_operations = {
3643 .llseek = btrfs_file_llseek,
3644 .read_iter = btrfs_file_read_iter,
3645 .splice_read = generic_file_splice_read,
3646 .write_iter = btrfs_file_write_iter,
3647 .splice_write = iter_file_splice_write,
3648 .mmap = btrfs_file_mmap,
3649 .open = btrfs_file_open,
3650 .release = btrfs_release_file,
3651 .fsync = btrfs_sync_file,
3652 .fallocate = btrfs_fallocate,
3653 .unlocked_ioctl = btrfs_ioctl,
3654 #ifdef CONFIG_COMPAT
3655 .compat_ioctl = btrfs_compat_ioctl,
3657 .remap_file_range = btrfs_remap_file_range,
3660 void __cold btrfs_auto_defrag_exit(void)
3662 kmem_cache_destroy(btrfs_inode_defrag_cachep);
3665 int __init btrfs_auto_defrag_init(void)
3667 btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
3668 sizeof(struct inode_defrag), 0,
3671 if (!btrfs_inode_defrag_cachep)
3677 int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end)
3682 * So with compression we will find and lock a dirty page and clear the
3683 * first one as dirty, setup an async extent, and immediately return
3684 * with the entire range locked but with nobody actually marked with
3685 * writeback. So we can't just filemap_write_and_wait_range() and
3686 * expect it to work since it will just kick off a thread to do the
3687 * actual work. So we need to call filemap_fdatawrite_range _again_
3688 * since it will wait on the page lock, which won't be unlocked until
3689 * after the pages have been marked as writeback and so we're good to go
3690 * from there. We have to do this otherwise we'll miss the ordered
3691 * extents and that results in badness. Please Josef, do not think you
3692 * know better and pull this out at some point in the future, it is
3693 * right and you are wrong.
3695 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
3696 if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
3697 &BTRFS_I(inode)->runtime_flags))
3698 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);