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"
30 static struct kmem_cache *btrfs_inode_defrag_cachep;
32 * when auto defrag is enabled we
33 * queue up these defrag structs to remember which
34 * inodes need defragging passes
37 struct rb_node rb_node;
41 * transid where the defrag was added, we search for
42 * extents newer than this
49 /* last offset we were able to defrag */
52 /* if we've wrapped around back to zero once already */
56 static int __compare_inode_defrag(struct inode_defrag *defrag1,
57 struct inode_defrag *defrag2)
59 if (defrag1->root > defrag2->root)
61 else if (defrag1->root < defrag2->root)
63 else if (defrag1->ino > defrag2->ino)
65 else if (defrag1->ino < defrag2->ino)
71 /* pop a record for an inode into the defrag tree. The lock
72 * must be held already
74 * If you're inserting a record for an older transid than an
75 * existing record, the transid already in the tree is lowered
77 * If an existing record is found the defrag item you
80 static int __btrfs_add_inode_defrag(struct btrfs_inode *inode,
81 struct inode_defrag *defrag)
83 struct btrfs_fs_info *fs_info = inode->root->fs_info;
84 struct inode_defrag *entry;
86 struct rb_node *parent = NULL;
89 p = &fs_info->defrag_inodes.rb_node;
92 entry = rb_entry(parent, struct inode_defrag, rb_node);
94 ret = __compare_inode_defrag(defrag, entry);
98 p = &parent->rb_right;
100 /* if we're reinserting an entry for
101 * an old defrag run, make sure to
102 * lower the transid of our existing record
104 if (defrag->transid < entry->transid)
105 entry->transid = defrag->transid;
106 if (defrag->last_offset > entry->last_offset)
107 entry->last_offset = defrag->last_offset;
111 set_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags);
112 rb_link_node(&defrag->rb_node, parent, p);
113 rb_insert_color(&defrag->rb_node, &fs_info->defrag_inodes);
117 static inline int __need_auto_defrag(struct btrfs_fs_info *fs_info)
119 if (!btrfs_test_opt(fs_info, AUTO_DEFRAG))
122 if (btrfs_fs_closing(fs_info))
129 * insert a defrag record for this inode if auto defrag is
132 int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans,
133 struct btrfs_inode *inode)
135 struct btrfs_root *root = inode->root;
136 struct btrfs_fs_info *fs_info = root->fs_info;
137 struct inode_defrag *defrag;
141 if (!__need_auto_defrag(fs_info))
144 if (test_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags))
148 transid = trans->transid;
150 transid = inode->root->last_trans;
152 defrag = kmem_cache_zalloc(btrfs_inode_defrag_cachep, GFP_NOFS);
156 defrag->ino = btrfs_ino(inode);
157 defrag->transid = transid;
158 defrag->root = root->root_key.objectid;
160 spin_lock(&fs_info->defrag_inodes_lock);
161 if (!test_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags)) {
163 * If we set IN_DEFRAG flag and evict the inode from memory,
164 * and then re-read this inode, this new inode doesn't have
165 * IN_DEFRAG flag. At the case, we may find the existed defrag.
167 ret = __btrfs_add_inode_defrag(inode, defrag);
169 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
171 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
173 spin_unlock(&fs_info->defrag_inodes_lock);
178 * Requeue the defrag object. If there is a defrag object that points to
179 * the same inode in the tree, we will merge them together (by
180 * __btrfs_add_inode_defrag()) and free the one that we want to requeue.
182 static void btrfs_requeue_inode_defrag(struct btrfs_inode *inode,
183 struct inode_defrag *defrag)
185 struct btrfs_fs_info *fs_info = inode->root->fs_info;
188 if (!__need_auto_defrag(fs_info))
192 * Here we don't check the IN_DEFRAG flag, because we need merge
195 spin_lock(&fs_info->defrag_inodes_lock);
196 ret = __btrfs_add_inode_defrag(inode, defrag);
197 spin_unlock(&fs_info->defrag_inodes_lock);
202 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
206 * pick the defragable inode that we want, if it doesn't exist, we will get
209 static struct inode_defrag *
210 btrfs_pick_defrag_inode(struct btrfs_fs_info *fs_info, u64 root, u64 ino)
212 struct inode_defrag *entry = NULL;
213 struct inode_defrag tmp;
215 struct rb_node *parent = NULL;
221 spin_lock(&fs_info->defrag_inodes_lock);
222 p = fs_info->defrag_inodes.rb_node;
225 entry = rb_entry(parent, struct inode_defrag, rb_node);
227 ret = __compare_inode_defrag(&tmp, entry);
231 p = parent->rb_right;
236 if (parent && __compare_inode_defrag(&tmp, entry) > 0) {
237 parent = rb_next(parent);
239 entry = rb_entry(parent, struct inode_defrag, rb_node);
245 rb_erase(parent, &fs_info->defrag_inodes);
246 spin_unlock(&fs_info->defrag_inodes_lock);
250 void btrfs_cleanup_defrag_inodes(struct btrfs_fs_info *fs_info)
252 struct inode_defrag *defrag;
253 struct rb_node *node;
255 spin_lock(&fs_info->defrag_inodes_lock);
256 node = rb_first(&fs_info->defrag_inodes);
258 rb_erase(node, &fs_info->defrag_inodes);
259 defrag = rb_entry(node, struct inode_defrag, rb_node);
260 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
262 cond_resched_lock(&fs_info->defrag_inodes_lock);
264 node = rb_first(&fs_info->defrag_inodes);
266 spin_unlock(&fs_info->defrag_inodes_lock);
269 #define BTRFS_DEFRAG_BATCH 1024
271 static int __btrfs_run_defrag_inode(struct btrfs_fs_info *fs_info,
272 struct inode_defrag *defrag)
274 struct btrfs_root *inode_root;
276 struct btrfs_key key;
277 struct btrfs_ioctl_defrag_range_args range;
283 key.objectid = defrag->root;
284 key.type = BTRFS_ROOT_ITEM_KEY;
285 key.offset = (u64)-1;
287 index = srcu_read_lock(&fs_info->subvol_srcu);
289 inode_root = btrfs_read_fs_root_no_name(fs_info, &key);
290 if (IS_ERR(inode_root)) {
291 ret = PTR_ERR(inode_root);
295 key.objectid = defrag->ino;
296 key.type = BTRFS_INODE_ITEM_KEY;
298 inode = btrfs_iget(fs_info->sb, &key, inode_root, NULL);
300 ret = PTR_ERR(inode);
303 srcu_read_unlock(&fs_info->subvol_srcu, index);
305 /* do a chunk of defrag */
306 clear_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
307 memset(&range, 0, sizeof(range));
309 range.start = defrag->last_offset;
311 sb_start_write(fs_info->sb);
312 num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid,
314 sb_end_write(fs_info->sb);
316 * if we filled the whole defrag batch, there
317 * must be more work to do. Queue this defrag
320 if (num_defrag == BTRFS_DEFRAG_BATCH) {
321 defrag->last_offset = range.start;
322 btrfs_requeue_inode_defrag(BTRFS_I(inode), defrag);
323 } else if (defrag->last_offset && !defrag->cycled) {
325 * we didn't fill our defrag batch, but
326 * we didn't start at zero. Make sure we loop
327 * around to the start of the file.
329 defrag->last_offset = 0;
331 btrfs_requeue_inode_defrag(BTRFS_I(inode), defrag);
333 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
339 srcu_read_unlock(&fs_info->subvol_srcu, index);
340 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
345 * run through the list of inodes in the FS that need
348 int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info)
350 struct inode_defrag *defrag;
352 u64 root_objectid = 0;
354 atomic_inc(&fs_info->defrag_running);
356 /* Pause the auto defragger. */
357 if (test_bit(BTRFS_FS_STATE_REMOUNTING,
361 if (!__need_auto_defrag(fs_info))
364 /* find an inode to defrag */
365 defrag = btrfs_pick_defrag_inode(fs_info, root_objectid,
368 if (root_objectid || first_ino) {
377 first_ino = defrag->ino + 1;
378 root_objectid = defrag->root;
380 __btrfs_run_defrag_inode(fs_info, defrag);
382 atomic_dec(&fs_info->defrag_running);
385 * during unmount, we use the transaction_wait queue to
386 * wait for the defragger to stop
388 wake_up(&fs_info->transaction_wait);
392 /* simple helper to fault in pages and copy. This should go away
393 * and be replaced with calls into generic code.
395 static noinline int btrfs_copy_from_user(loff_t pos, size_t write_bytes,
396 struct page **prepared_pages,
400 size_t total_copied = 0;
402 int offset = offset_in_page(pos);
404 while (write_bytes > 0) {
405 size_t count = min_t(size_t,
406 PAGE_SIZE - offset, write_bytes);
407 struct page *page = prepared_pages[pg];
409 * Copy data from userspace to the current page
411 copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
413 /* Flush processor's dcache for this page */
414 flush_dcache_page(page);
417 * if we get a partial write, we can end up with
418 * partially up to date pages. These add
419 * a lot of complexity, so make sure they don't
420 * happen by forcing this copy to be retried.
422 * The rest of the btrfs_file_write code will fall
423 * back to page at a time copies after we return 0.
425 if (!PageUptodate(page) && copied < count)
428 iov_iter_advance(i, copied);
429 write_bytes -= copied;
430 total_copied += copied;
432 /* Return to btrfs_file_write_iter to fault page */
433 if (unlikely(copied == 0))
436 if (copied < PAGE_SIZE - offset) {
447 * unlocks pages after btrfs_file_write is done with them
449 static void btrfs_drop_pages(struct page **pages, size_t num_pages)
452 for (i = 0; i < num_pages; i++) {
453 /* page checked is some magic around finding pages that
454 * have been modified without going through btrfs_set_page_dirty
455 * clear it here. There should be no need to mark the pages
456 * accessed as prepare_pages should have marked them accessed
457 * in prepare_pages via find_or_create_page()
459 ClearPageChecked(pages[i]);
460 unlock_page(pages[i]);
465 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
468 struct extent_state **cached_state)
470 u64 search_start = start;
471 const u64 end = start + len - 1;
473 while (search_start < end) {
474 const u64 search_len = end - search_start + 1;
475 struct extent_map *em;
479 em = btrfs_get_extent(inode, NULL, 0, search_start,
484 if (em->block_start != EXTENT_MAP_HOLE)
488 if (em->start < search_start)
489 em_len -= search_start - em->start;
490 if (em_len > search_len)
493 ret = set_extent_bit(&inode->io_tree, search_start,
494 search_start + em_len - 1,
496 NULL, cached_state, GFP_NOFS);
498 search_start = extent_map_end(em);
507 * after copy_from_user, pages need to be dirtied and we need to make
508 * sure holes are created between the current EOF and the start of
509 * any next extents (if required).
511 * this also makes the decision about creating an inline extent vs
512 * doing real data extents, marking pages dirty and delalloc as required.
514 int btrfs_dirty_pages(struct inode *inode, struct page **pages,
515 size_t num_pages, loff_t pos, size_t write_bytes,
516 struct extent_state **cached)
518 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
523 u64 end_of_last_block;
524 u64 end_pos = pos + write_bytes;
525 loff_t isize = i_size_read(inode);
526 unsigned int extra_bits = 0;
528 start_pos = pos & ~((u64) fs_info->sectorsize - 1);
529 num_bytes = round_up(write_bytes + pos - start_pos,
530 fs_info->sectorsize);
532 end_of_last_block = start_pos + num_bytes - 1;
535 * The pages may have already been dirty, clear out old accounting so
536 * we can set things up properly
538 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos, end_of_last_block,
539 EXTENT_DIRTY | EXTENT_DELALLOC |
540 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0, cached);
542 if (!btrfs_is_free_space_inode(BTRFS_I(inode))) {
543 if (start_pos >= isize &&
544 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)) {
546 * There can't be any extents following eof in this case
547 * so just set the delalloc new bit for the range
550 extra_bits |= EXTENT_DELALLOC_NEW;
552 err = btrfs_find_new_delalloc_bytes(BTRFS_I(inode),
560 err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
561 extra_bits, cached, 0);
565 for (i = 0; i < num_pages; i++) {
566 struct page *p = pages[i];
573 * we've only changed i_size in ram, and we haven't updated
574 * the disk i_size. There is no need to log the inode
578 i_size_write(inode, end_pos);
583 * this drops all the extents in the cache that intersect the range
584 * [start, end]. Existing extents are split as required.
586 void btrfs_drop_extent_cache(struct btrfs_inode *inode, u64 start, u64 end,
589 struct extent_map *em;
590 struct extent_map *split = NULL;
591 struct extent_map *split2 = NULL;
592 struct extent_map_tree *em_tree = &inode->extent_tree;
593 u64 len = end - start + 1;
601 WARN_ON(end < start);
602 if (end == (u64)-1) {
611 split = alloc_extent_map();
613 split2 = alloc_extent_map();
614 if (!split || !split2)
617 write_lock(&em_tree->lock);
618 em = lookup_extent_mapping(em_tree, start, len);
620 write_unlock(&em_tree->lock);
624 gen = em->generation;
625 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
626 if (testend && em->start + em->len >= start + len) {
628 write_unlock(&em_tree->lock);
631 start = em->start + em->len;
633 len = start + len - (em->start + em->len);
635 write_unlock(&em_tree->lock);
638 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
639 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
640 clear_bit(EXTENT_FLAG_LOGGING, &flags);
641 modified = !list_empty(&em->list);
645 if (em->start < start) {
646 split->start = em->start;
647 split->len = start - em->start;
649 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
650 split->orig_start = em->orig_start;
651 split->block_start = em->block_start;
654 split->block_len = em->block_len;
656 split->block_len = split->len;
657 split->orig_block_len = max(split->block_len,
659 split->ram_bytes = em->ram_bytes;
661 split->orig_start = split->start;
662 split->block_len = 0;
663 split->block_start = em->block_start;
664 split->orig_block_len = 0;
665 split->ram_bytes = split->len;
668 split->generation = gen;
669 split->bdev = em->bdev;
670 split->flags = flags;
671 split->compress_type = em->compress_type;
672 replace_extent_mapping(em_tree, em, split, modified);
673 free_extent_map(split);
677 if (testend && em->start + em->len > start + len) {
678 u64 diff = start + len - em->start;
680 split->start = start + len;
681 split->len = em->start + em->len - (start + len);
682 split->bdev = em->bdev;
683 split->flags = flags;
684 split->compress_type = em->compress_type;
685 split->generation = gen;
687 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
688 split->orig_block_len = max(em->block_len,
691 split->ram_bytes = em->ram_bytes;
693 split->block_len = em->block_len;
694 split->block_start = em->block_start;
695 split->orig_start = em->orig_start;
697 split->block_len = split->len;
698 split->block_start = em->block_start
700 split->orig_start = em->orig_start;
703 split->ram_bytes = split->len;
704 split->orig_start = split->start;
705 split->block_len = 0;
706 split->block_start = em->block_start;
707 split->orig_block_len = 0;
710 if (extent_map_in_tree(em)) {
711 replace_extent_mapping(em_tree, em, split,
714 ret = add_extent_mapping(em_tree, split,
716 ASSERT(ret == 0); /* Logic error */
718 free_extent_map(split);
722 if (extent_map_in_tree(em))
723 remove_extent_mapping(em_tree, em);
724 write_unlock(&em_tree->lock);
728 /* once for the tree*/
732 free_extent_map(split);
734 free_extent_map(split2);
738 * this is very complex, but the basic idea is to drop all extents
739 * in the range start - end. hint_block is filled in with a block number
740 * that would be a good hint to the block allocator for this file.
742 * If an extent intersects the range but is not entirely inside the range
743 * it is either truncated or split. Anything entirely inside the range
744 * is deleted from the tree.
746 int __btrfs_drop_extents(struct btrfs_trans_handle *trans,
747 struct btrfs_root *root, struct inode *inode,
748 struct btrfs_path *path, u64 start, u64 end,
749 u64 *drop_end, int drop_cache,
751 u32 extent_item_size,
754 struct btrfs_fs_info *fs_info = root->fs_info;
755 struct extent_buffer *leaf;
756 struct btrfs_file_extent_item *fi;
757 struct btrfs_ref ref = { 0 };
758 struct btrfs_key key;
759 struct btrfs_key new_key;
760 u64 ino = btrfs_ino(BTRFS_I(inode));
761 u64 search_start = start;
764 u64 extent_offset = 0;
766 u64 last_end = start;
772 int modify_tree = -1;
775 int leafs_visited = 0;
778 btrfs_drop_extent_cache(BTRFS_I(inode), start, end - 1, 0);
780 if (start >= BTRFS_I(inode)->disk_i_size && !replace_extent)
783 update_refs = (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
784 root == fs_info->tree_root);
787 ret = btrfs_lookup_file_extent(trans, root, path, ino,
788 search_start, modify_tree);
791 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
792 leaf = path->nodes[0];
793 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
794 if (key.objectid == ino &&
795 key.type == BTRFS_EXTENT_DATA_KEY)
801 leaf = path->nodes[0];
802 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
804 ret = btrfs_next_leaf(root, path);
812 leaf = path->nodes[0];
816 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
818 if (key.objectid > ino)
820 if (WARN_ON_ONCE(key.objectid < ino) ||
821 key.type < BTRFS_EXTENT_DATA_KEY) {
826 if (key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
829 fi = btrfs_item_ptr(leaf, path->slots[0],
830 struct btrfs_file_extent_item);
831 extent_type = btrfs_file_extent_type(leaf, fi);
833 if (extent_type == BTRFS_FILE_EXTENT_REG ||
834 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
835 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
836 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
837 extent_offset = btrfs_file_extent_offset(leaf, fi);
838 extent_end = key.offset +
839 btrfs_file_extent_num_bytes(leaf, fi);
840 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
841 extent_end = key.offset +
842 btrfs_file_extent_ram_bytes(leaf, fi);
849 * Don't skip extent items representing 0 byte lengths. They
850 * used to be created (bug) if while punching holes we hit
851 * -ENOSPC condition. So if we find one here, just ensure we
852 * delete it, otherwise we would insert a new file extent item
853 * with the same key (offset) as that 0 bytes length file
854 * extent item in the call to setup_items_for_insert() later
857 if (extent_end == key.offset && extent_end >= search_start) {
858 last_end = extent_end;
859 goto delete_extent_item;
862 if (extent_end <= search_start) {
868 search_start = max(key.offset, start);
869 if (recow || !modify_tree) {
871 btrfs_release_path(path);
876 * | - range to drop - |
877 * | -------- extent -------- |
879 if (start > key.offset && end < extent_end) {
881 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
886 memcpy(&new_key, &key, sizeof(new_key));
887 new_key.offset = start;
888 ret = btrfs_duplicate_item(trans, root, path,
890 if (ret == -EAGAIN) {
891 btrfs_release_path(path);
897 leaf = path->nodes[0];
898 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
899 struct btrfs_file_extent_item);
900 btrfs_set_file_extent_num_bytes(leaf, fi,
903 fi = btrfs_item_ptr(leaf, path->slots[0],
904 struct btrfs_file_extent_item);
906 extent_offset += start - key.offset;
907 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
908 btrfs_set_file_extent_num_bytes(leaf, fi,
910 btrfs_mark_buffer_dirty(leaf);
912 if (update_refs && disk_bytenr > 0) {
913 btrfs_init_generic_ref(&ref,
914 BTRFS_ADD_DELAYED_REF,
915 disk_bytenr, num_bytes, 0);
916 btrfs_init_data_ref(&ref,
917 root->root_key.objectid,
919 start - extent_offset);
920 ret = btrfs_inc_extent_ref(trans, &ref);
921 BUG_ON(ret); /* -ENOMEM */
926 * From here on out we will have actually dropped something, so
927 * last_end can be updated.
929 last_end = extent_end;
932 * | ---- range to drop ----- |
933 * | -------- extent -------- |
935 if (start <= key.offset && end < extent_end) {
936 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
941 memcpy(&new_key, &key, sizeof(new_key));
942 new_key.offset = end;
943 btrfs_set_item_key_safe(fs_info, path, &new_key);
945 extent_offset += end - key.offset;
946 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
947 btrfs_set_file_extent_num_bytes(leaf, fi,
949 btrfs_mark_buffer_dirty(leaf);
950 if (update_refs && disk_bytenr > 0)
951 inode_sub_bytes(inode, end - key.offset);
955 search_start = extent_end;
957 * | ---- range to drop ----- |
958 * | -------- extent -------- |
960 if (start > key.offset && end >= extent_end) {
962 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
967 btrfs_set_file_extent_num_bytes(leaf, fi,
969 btrfs_mark_buffer_dirty(leaf);
970 if (update_refs && disk_bytenr > 0)
971 inode_sub_bytes(inode, extent_end - start);
972 if (end == extent_end)
980 * | ---- range to drop ----- |
981 * | ------ extent ------ |
983 if (start <= key.offset && end >= extent_end) {
986 del_slot = path->slots[0];
989 BUG_ON(del_slot + del_nr != path->slots[0]);
994 extent_type == BTRFS_FILE_EXTENT_INLINE) {
995 inode_sub_bytes(inode,
996 extent_end - key.offset);
997 extent_end = ALIGN(extent_end,
998 fs_info->sectorsize);
999 } else if (update_refs && disk_bytenr > 0) {
1000 btrfs_init_generic_ref(&ref,
1001 BTRFS_DROP_DELAYED_REF,
1002 disk_bytenr, num_bytes, 0);
1003 btrfs_init_data_ref(&ref,
1004 root->root_key.objectid,
1006 key.offset - extent_offset);
1007 ret = btrfs_free_extent(trans, &ref);
1008 BUG_ON(ret); /* -ENOMEM */
1009 inode_sub_bytes(inode,
1010 extent_end - key.offset);
1013 if (end == extent_end)
1016 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
1021 ret = btrfs_del_items(trans, root, path, del_slot,
1024 btrfs_abort_transaction(trans, ret);
1031 btrfs_release_path(path);
1038 if (!ret && del_nr > 0) {
1040 * Set path->slots[0] to first slot, so that after the delete
1041 * if items are move off from our leaf to its immediate left or
1042 * right neighbor leafs, we end up with a correct and adjusted
1043 * path->slots[0] for our insertion (if replace_extent != 0).
1045 path->slots[0] = del_slot;
1046 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1048 btrfs_abort_transaction(trans, ret);
1051 leaf = path->nodes[0];
1053 * If btrfs_del_items() was called, it might have deleted a leaf, in
1054 * which case it unlocked our path, so check path->locks[0] matches a
1057 if (!ret && replace_extent && leafs_visited == 1 &&
1058 (path->locks[0] == BTRFS_WRITE_LOCK_BLOCKING ||
1059 path->locks[0] == BTRFS_WRITE_LOCK) &&
1060 btrfs_leaf_free_space(leaf) >=
1061 sizeof(struct btrfs_item) + extent_item_size) {
1064 key.type = BTRFS_EXTENT_DATA_KEY;
1066 if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) {
1067 struct btrfs_key slot_key;
1069 btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]);
1070 if (btrfs_comp_cpu_keys(&key, &slot_key) > 0)
1073 setup_items_for_insert(root, path, &key,
1076 sizeof(struct btrfs_item) +
1077 extent_item_size, 1);
1081 if (!replace_extent || !(*key_inserted))
1082 btrfs_release_path(path);
1084 *drop_end = found ? min(end, last_end) : end;
1088 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
1089 struct btrfs_root *root, struct inode *inode, u64 start,
1090 u64 end, int drop_cache)
1092 struct btrfs_path *path;
1095 path = btrfs_alloc_path();
1098 ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL,
1099 drop_cache, 0, 0, NULL);
1100 btrfs_free_path(path);
1104 static int extent_mergeable(struct extent_buffer *leaf, int slot,
1105 u64 objectid, u64 bytenr, u64 orig_offset,
1106 u64 *start, u64 *end)
1108 struct btrfs_file_extent_item *fi;
1109 struct btrfs_key key;
1112 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
1115 btrfs_item_key_to_cpu(leaf, &key, slot);
1116 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
1119 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1120 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
1121 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
1122 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
1123 btrfs_file_extent_compression(leaf, fi) ||
1124 btrfs_file_extent_encryption(leaf, fi) ||
1125 btrfs_file_extent_other_encoding(leaf, fi))
1128 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1129 if ((*start && *start != key.offset) || (*end && *end != extent_end))
1132 *start = key.offset;
1138 * Mark extent in the range start - end as written.
1140 * This changes extent type from 'pre-allocated' to 'regular'. If only
1141 * part of extent is marked as written, the extent will be split into
1144 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
1145 struct btrfs_inode *inode, u64 start, u64 end)
1147 struct btrfs_fs_info *fs_info = trans->fs_info;
1148 struct btrfs_root *root = inode->root;
1149 struct extent_buffer *leaf;
1150 struct btrfs_path *path;
1151 struct btrfs_file_extent_item *fi;
1152 struct btrfs_ref ref = { 0 };
1153 struct btrfs_key key;
1154 struct btrfs_key new_key;
1166 u64 ino = btrfs_ino(inode);
1168 path = btrfs_alloc_path();
1175 key.type = BTRFS_EXTENT_DATA_KEY;
1178 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1181 if (ret > 0 && path->slots[0] > 0)
1184 leaf = path->nodes[0];
1185 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1186 if (key.objectid != ino ||
1187 key.type != BTRFS_EXTENT_DATA_KEY) {
1189 btrfs_abort_transaction(trans, ret);
1192 fi = btrfs_item_ptr(leaf, path->slots[0],
1193 struct btrfs_file_extent_item);
1194 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_PREALLOC) {
1196 btrfs_abort_transaction(trans, ret);
1199 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1200 if (key.offset > start || extent_end < end) {
1202 btrfs_abort_transaction(trans, ret);
1206 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1207 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1208 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
1209 memcpy(&new_key, &key, sizeof(new_key));
1211 if (start == key.offset && end < extent_end) {
1214 if (extent_mergeable(leaf, path->slots[0] - 1,
1215 ino, bytenr, orig_offset,
1216 &other_start, &other_end)) {
1217 new_key.offset = end;
1218 btrfs_set_item_key_safe(fs_info, path, &new_key);
1219 fi = btrfs_item_ptr(leaf, path->slots[0],
1220 struct btrfs_file_extent_item);
1221 btrfs_set_file_extent_generation(leaf, fi,
1223 btrfs_set_file_extent_num_bytes(leaf, fi,
1225 btrfs_set_file_extent_offset(leaf, fi,
1227 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1228 struct btrfs_file_extent_item);
1229 btrfs_set_file_extent_generation(leaf, fi,
1231 btrfs_set_file_extent_num_bytes(leaf, fi,
1233 btrfs_mark_buffer_dirty(leaf);
1238 if (start > key.offset && end == extent_end) {
1241 if (extent_mergeable(leaf, path->slots[0] + 1,
1242 ino, bytenr, orig_offset,
1243 &other_start, &other_end)) {
1244 fi = btrfs_item_ptr(leaf, path->slots[0],
1245 struct btrfs_file_extent_item);
1246 btrfs_set_file_extent_num_bytes(leaf, fi,
1247 start - key.offset);
1248 btrfs_set_file_extent_generation(leaf, fi,
1251 new_key.offset = start;
1252 btrfs_set_item_key_safe(fs_info, path, &new_key);
1254 fi = btrfs_item_ptr(leaf, path->slots[0],
1255 struct btrfs_file_extent_item);
1256 btrfs_set_file_extent_generation(leaf, fi,
1258 btrfs_set_file_extent_num_bytes(leaf, fi,
1260 btrfs_set_file_extent_offset(leaf, fi,
1261 start - orig_offset);
1262 btrfs_mark_buffer_dirty(leaf);
1267 while (start > key.offset || end < extent_end) {
1268 if (key.offset == start)
1271 new_key.offset = split;
1272 ret = btrfs_duplicate_item(trans, root, path, &new_key);
1273 if (ret == -EAGAIN) {
1274 btrfs_release_path(path);
1278 btrfs_abort_transaction(trans, ret);
1282 leaf = path->nodes[0];
1283 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1284 struct btrfs_file_extent_item);
1285 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1286 btrfs_set_file_extent_num_bytes(leaf, fi,
1287 split - key.offset);
1289 fi = btrfs_item_ptr(leaf, path->slots[0],
1290 struct btrfs_file_extent_item);
1292 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1293 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
1294 btrfs_set_file_extent_num_bytes(leaf, fi,
1295 extent_end - split);
1296 btrfs_mark_buffer_dirty(leaf);
1298 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, bytenr,
1300 btrfs_init_data_ref(&ref, root->root_key.objectid, ino,
1302 ret = btrfs_inc_extent_ref(trans, &ref);
1304 btrfs_abort_transaction(trans, ret);
1308 if (split == start) {
1311 if (start != key.offset) {
1313 btrfs_abort_transaction(trans, ret);
1324 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
1326 btrfs_init_data_ref(&ref, root->root_key.objectid, ino, orig_offset);
1327 if (extent_mergeable(leaf, path->slots[0] + 1,
1328 ino, bytenr, orig_offset,
1329 &other_start, &other_end)) {
1331 btrfs_release_path(path);
1334 extent_end = other_end;
1335 del_slot = path->slots[0] + 1;
1337 ret = btrfs_free_extent(trans, &ref);
1339 btrfs_abort_transaction(trans, ret);
1345 if (extent_mergeable(leaf, path->slots[0] - 1,
1346 ino, bytenr, orig_offset,
1347 &other_start, &other_end)) {
1349 btrfs_release_path(path);
1352 key.offset = other_start;
1353 del_slot = path->slots[0];
1355 ret = btrfs_free_extent(trans, &ref);
1357 btrfs_abort_transaction(trans, ret);
1362 fi = btrfs_item_ptr(leaf, path->slots[0],
1363 struct btrfs_file_extent_item);
1364 btrfs_set_file_extent_type(leaf, fi,
1365 BTRFS_FILE_EXTENT_REG);
1366 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1367 btrfs_mark_buffer_dirty(leaf);
1369 fi = btrfs_item_ptr(leaf, del_slot - 1,
1370 struct btrfs_file_extent_item);
1371 btrfs_set_file_extent_type(leaf, fi,
1372 BTRFS_FILE_EXTENT_REG);
1373 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1374 btrfs_set_file_extent_num_bytes(leaf, fi,
1375 extent_end - key.offset);
1376 btrfs_mark_buffer_dirty(leaf);
1378 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1380 btrfs_abort_transaction(trans, ret);
1385 btrfs_free_path(path);
1390 * on error we return an unlocked page and the error value
1391 * on success we return a locked page and 0
1393 static int prepare_uptodate_page(struct inode *inode,
1394 struct page *page, u64 pos,
1395 bool force_uptodate)
1399 if (((pos & (PAGE_SIZE - 1)) || force_uptodate) &&
1400 !PageUptodate(page)) {
1401 ret = btrfs_readpage(NULL, page);
1405 if (!PageUptodate(page)) {
1409 if (page->mapping != inode->i_mapping) {
1418 * this just gets pages into the page cache and locks them down.
1420 static noinline int prepare_pages(struct inode *inode, struct page **pages,
1421 size_t num_pages, loff_t pos,
1422 size_t write_bytes, bool force_uptodate)
1425 unsigned long index = pos >> PAGE_SHIFT;
1426 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1430 for (i = 0; i < num_pages; i++) {
1432 pages[i] = find_or_create_page(inode->i_mapping, index + i,
1433 mask | __GFP_WRITE);
1441 err = prepare_uptodate_page(inode, pages[i], pos,
1443 if (!err && i == num_pages - 1)
1444 err = prepare_uptodate_page(inode, pages[i],
1445 pos + write_bytes, false);
1448 if (err == -EAGAIN) {
1455 wait_on_page_writeback(pages[i]);
1460 while (faili >= 0) {
1461 unlock_page(pages[faili]);
1462 put_page(pages[faili]);
1470 * This function locks the extent and properly waits for data=ordered extents
1471 * to finish before allowing the pages to be modified if need.
1474 * 1 - the extent is locked
1475 * 0 - the extent is not locked, and everything is OK
1476 * -EAGAIN - need re-prepare the pages
1477 * the other < 0 number - Something wrong happens
1480 lock_and_cleanup_extent_if_need(struct btrfs_inode *inode, struct page **pages,
1481 size_t num_pages, loff_t pos,
1483 u64 *lockstart, u64 *lockend,
1484 struct extent_state **cached_state)
1486 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1492 start_pos = round_down(pos, fs_info->sectorsize);
1493 last_pos = start_pos
1494 + round_up(pos + write_bytes - start_pos,
1495 fs_info->sectorsize) - 1;
1497 if (start_pos < inode->vfs_inode.i_size) {
1498 struct btrfs_ordered_extent *ordered;
1500 lock_extent_bits(&inode->io_tree, start_pos, last_pos,
1502 ordered = btrfs_lookup_ordered_range(inode, start_pos,
1503 last_pos - start_pos + 1);
1505 ordered->file_offset + ordered->len > start_pos &&
1506 ordered->file_offset <= last_pos) {
1507 unlock_extent_cached(&inode->io_tree, start_pos,
1508 last_pos, cached_state);
1509 for (i = 0; i < num_pages; i++) {
1510 unlock_page(pages[i]);
1513 btrfs_start_ordered_extent(&inode->vfs_inode,
1515 btrfs_put_ordered_extent(ordered);
1519 btrfs_put_ordered_extent(ordered);
1521 *lockstart = start_pos;
1522 *lockend = last_pos;
1527 * It's possible the pages are dirty right now, but we don't want
1528 * to clean them yet because copy_from_user may catch a page fault
1529 * and we might have to fall back to one page at a time. If that
1530 * happens, we'll unlock these pages and we'd have a window where
1531 * reclaim could sneak in and drop the once-dirty page on the floor
1532 * without writing it.
1534 * We have the pages locked and the extent range locked, so there's
1535 * no way someone can start IO on any dirty pages in this range.
1537 * We'll call btrfs_dirty_pages() later on, and that will flip around
1538 * delalloc bits and dirty the pages as required.
1540 for (i = 0; i < num_pages; i++) {
1541 set_page_extent_mapped(pages[i]);
1542 WARN_ON(!PageLocked(pages[i]));
1548 static noinline int check_can_nocow(struct btrfs_inode *inode, loff_t pos,
1549 size_t *write_bytes)
1551 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1552 struct btrfs_root *root = inode->root;
1553 struct btrfs_ordered_extent *ordered;
1554 u64 lockstart, lockend;
1558 ret = btrfs_start_write_no_snapshotting(root);
1562 lockstart = round_down(pos, fs_info->sectorsize);
1563 lockend = round_up(pos + *write_bytes,
1564 fs_info->sectorsize) - 1;
1567 lock_extent(&inode->io_tree, lockstart, lockend);
1568 ordered = btrfs_lookup_ordered_range(inode, lockstart,
1569 lockend - lockstart + 1);
1573 unlock_extent(&inode->io_tree, lockstart, lockend);
1574 btrfs_start_ordered_extent(&inode->vfs_inode, ordered, 1);
1575 btrfs_put_ordered_extent(ordered);
1578 num_bytes = lockend - lockstart + 1;
1579 ret = can_nocow_extent(&inode->vfs_inode, lockstart, &num_bytes,
1583 btrfs_end_write_no_snapshotting(root);
1585 *write_bytes = min_t(size_t, *write_bytes ,
1586 num_bytes - pos + lockstart);
1589 unlock_extent(&inode->io_tree, lockstart, lockend);
1594 static noinline ssize_t btrfs_buffered_write(struct kiocb *iocb,
1597 struct file *file = iocb->ki_filp;
1598 loff_t pos = iocb->ki_pos;
1599 struct inode *inode = file_inode(file);
1600 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1601 struct btrfs_root *root = BTRFS_I(inode)->root;
1602 struct page **pages = NULL;
1603 struct extent_state *cached_state = NULL;
1604 struct extent_changeset *data_reserved = NULL;
1605 u64 release_bytes = 0;
1608 size_t num_written = 0;
1611 bool only_release_metadata = false;
1612 bool force_page_uptodate = false;
1614 nrptrs = min(DIV_ROUND_UP(iov_iter_count(i), PAGE_SIZE),
1615 PAGE_SIZE / (sizeof(struct page *)));
1616 nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1617 nrptrs = max(nrptrs, 8);
1618 pages = kmalloc_array(nrptrs, sizeof(struct page *), GFP_KERNEL);
1622 while (iov_iter_count(i) > 0) {
1623 size_t offset = offset_in_page(pos);
1624 size_t sector_offset;
1625 size_t write_bytes = min(iov_iter_count(i),
1626 nrptrs * (size_t)PAGE_SIZE -
1628 size_t num_pages = DIV_ROUND_UP(write_bytes + offset,
1630 size_t reserve_bytes;
1633 size_t dirty_sectors;
1637 WARN_ON(num_pages > nrptrs);
1640 * Fault pages before locking them in prepare_pages
1641 * to avoid recursive lock
1643 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1648 sector_offset = pos & (fs_info->sectorsize - 1);
1649 reserve_bytes = round_up(write_bytes + sector_offset,
1650 fs_info->sectorsize);
1652 extent_changeset_release(data_reserved);
1653 ret = btrfs_check_data_free_space(inode, &data_reserved, pos,
1656 if ((BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW |
1657 BTRFS_INODE_PREALLOC)) &&
1658 check_can_nocow(BTRFS_I(inode), pos,
1659 &write_bytes) > 0) {
1661 * For nodata cow case, no need to reserve
1664 only_release_metadata = true;
1666 * our prealloc extent may be smaller than
1667 * write_bytes, so scale down.
1669 num_pages = DIV_ROUND_UP(write_bytes + offset,
1671 reserve_bytes = round_up(write_bytes +
1673 fs_info->sectorsize);
1679 WARN_ON(reserve_bytes == 0);
1680 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
1683 if (!only_release_metadata)
1684 btrfs_free_reserved_data_space(inode,
1688 btrfs_end_write_no_snapshotting(root);
1692 release_bytes = reserve_bytes;
1695 * This is going to setup the pages array with the number of
1696 * pages we want, so we don't really need to worry about the
1697 * contents of pages from loop to loop
1699 ret = prepare_pages(inode, pages, num_pages,
1701 force_page_uptodate);
1703 btrfs_delalloc_release_extents(BTRFS_I(inode),
1704 reserve_bytes, true);
1708 extents_locked = lock_and_cleanup_extent_if_need(
1709 BTRFS_I(inode), pages,
1710 num_pages, pos, write_bytes, &lockstart,
1711 &lockend, &cached_state);
1712 if (extents_locked < 0) {
1713 if (extents_locked == -EAGAIN)
1715 btrfs_delalloc_release_extents(BTRFS_I(inode),
1716 reserve_bytes, true);
1717 ret = extents_locked;
1721 copied = btrfs_copy_from_user(pos, write_bytes, pages, i);
1723 num_sectors = BTRFS_BYTES_TO_BLKS(fs_info, reserve_bytes);
1724 dirty_sectors = round_up(copied + sector_offset,
1725 fs_info->sectorsize);
1726 dirty_sectors = BTRFS_BYTES_TO_BLKS(fs_info, dirty_sectors);
1729 * if we have trouble faulting in the pages, fall
1730 * back to one page at a time
1732 if (copied < write_bytes)
1736 force_page_uptodate = true;
1740 force_page_uptodate = false;
1741 dirty_pages = DIV_ROUND_UP(copied + offset,
1745 if (num_sectors > dirty_sectors) {
1746 /* release everything except the sectors we dirtied */
1747 release_bytes -= dirty_sectors <<
1748 fs_info->sb->s_blocksize_bits;
1749 if (only_release_metadata) {
1750 btrfs_delalloc_release_metadata(BTRFS_I(inode),
1751 release_bytes, true);
1755 __pos = round_down(pos,
1756 fs_info->sectorsize) +
1757 (dirty_pages << PAGE_SHIFT);
1758 btrfs_delalloc_release_space(inode,
1759 data_reserved, __pos,
1760 release_bytes, true);
1764 release_bytes = round_up(copied + sector_offset,
1765 fs_info->sectorsize);
1768 ret = btrfs_dirty_pages(inode, pages, dirty_pages,
1769 pos, copied, &cached_state);
1771 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1772 lockstart, lockend, &cached_state);
1773 btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes,
1776 btrfs_drop_pages(pages, num_pages);
1781 if (only_release_metadata)
1782 btrfs_end_write_no_snapshotting(root);
1784 if (only_release_metadata && copied > 0) {
1785 lockstart = round_down(pos,
1786 fs_info->sectorsize);
1787 lockend = round_up(pos + copied,
1788 fs_info->sectorsize) - 1;
1790 set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
1791 lockend, EXTENT_NORESERVE, NULL,
1793 only_release_metadata = false;
1796 btrfs_drop_pages(pages, num_pages);
1800 balance_dirty_pages_ratelimited(inode->i_mapping);
1801 if (dirty_pages < (fs_info->nodesize >> PAGE_SHIFT) + 1)
1802 btrfs_btree_balance_dirty(fs_info);
1805 num_written += copied;
1810 if (release_bytes) {
1811 if (only_release_metadata) {
1812 btrfs_end_write_no_snapshotting(root);
1813 btrfs_delalloc_release_metadata(BTRFS_I(inode),
1814 release_bytes, true);
1816 btrfs_delalloc_release_space(inode, data_reserved,
1817 round_down(pos, fs_info->sectorsize),
1818 release_bytes, true);
1822 extent_changeset_free(data_reserved);
1823 return num_written ? num_written : ret;
1826 static ssize_t __btrfs_direct_write(struct kiocb *iocb, struct iov_iter *from)
1828 struct file *file = iocb->ki_filp;
1829 struct inode *inode = file_inode(file);
1832 ssize_t written_buffered;
1836 written = generic_file_direct_write(iocb, from);
1838 if (written < 0 || !iov_iter_count(from))
1842 written_buffered = btrfs_buffered_write(iocb, from);
1843 if (written_buffered < 0) {
1844 err = written_buffered;
1848 * Ensure all data is persisted. We want the next direct IO read to be
1849 * able to read what was just written.
1851 endbyte = pos + written_buffered - 1;
1852 err = btrfs_fdatawrite_range(inode, pos, endbyte);
1855 err = filemap_fdatawait_range(inode->i_mapping, pos, endbyte);
1858 written += written_buffered;
1859 iocb->ki_pos = pos + written_buffered;
1860 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_SHIFT,
1861 endbyte >> PAGE_SHIFT);
1863 return written ? written : err;
1866 static void update_time_for_write(struct inode *inode)
1868 struct timespec64 now;
1870 if (IS_NOCMTIME(inode))
1873 now = current_time(inode);
1874 if (!timespec64_equal(&inode->i_mtime, &now))
1875 inode->i_mtime = now;
1877 if (!timespec64_equal(&inode->i_ctime, &now))
1878 inode->i_ctime = now;
1880 if (IS_I_VERSION(inode))
1881 inode_inc_iversion(inode);
1884 static ssize_t btrfs_file_write_iter(struct kiocb *iocb,
1885 struct iov_iter *from)
1887 struct file *file = iocb->ki_filp;
1888 struct inode *inode = file_inode(file);
1889 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1890 struct btrfs_root *root = BTRFS_I(inode)->root;
1893 ssize_t num_written = 0;
1894 bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host);
1897 size_t count = iov_iter_count(from);
1901 if (!(iocb->ki_flags & IOCB_DIRECT) &&
1902 (iocb->ki_flags & IOCB_NOWAIT))
1905 if (!inode_trylock(inode)) {
1906 if (iocb->ki_flags & IOCB_NOWAIT)
1911 err = generic_write_checks(iocb, from);
1913 inode_unlock(inode);
1918 if (iocb->ki_flags & IOCB_NOWAIT) {
1920 * We will allocate space in case nodatacow is not set,
1923 if (!(BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW |
1924 BTRFS_INODE_PREALLOC)) ||
1925 check_can_nocow(BTRFS_I(inode), pos, &count) <= 0) {
1926 inode_unlock(inode);
1931 current->backing_dev_info = inode_to_bdi(inode);
1932 err = file_remove_privs(file);
1934 inode_unlock(inode);
1939 * If BTRFS flips readonly due to some impossible error
1940 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1941 * although we have opened a file as writable, we have
1942 * to stop this write operation to ensure FS consistency.
1944 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
1945 inode_unlock(inode);
1951 * We reserve space for updating the inode when we reserve space for the
1952 * extent we are going to write, so we will enospc out there. We don't
1953 * need to start yet another transaction to update the inode as we will
1954 * update the inode when we finish writing whatever data we write.
1956 update_time_for_write(inode);
1958 start_pos = round_down(pos, fs_info->sectorsize);
1959 oldsize = i_size_read(inode);
1960 if (start_pos > oldsize) {
1961 /* Expand hole size to cover write data, preventing empty gap */
1962 end_pos = round_up(pos + count,
1963 fs_info->sectorsize);
1964 err = btrfs_cont_expand(inode, oldsize, end_pos);
1966 inode_unlock(inode);
1969 if (start_pos > round_up(oldsize, fs_info->sectorsize))
1974 atomic_inc(&BTRFS_I(inode)->sync_writers);
1976 if (iocb->ki_flags & IOCB_DIRECT) {
1977 num_written = __btrfs_direct_write(iocb, from);
1979 num_written = btrfs_buffered_write(iocb, from);
1980 if (num_written > 0)
1981 iocb->ki_pos = pos + num_written;
1983 pagecache_isize_extended(inode, oldsize,
1984 i_size_read(inode));
1987 inode_unlock(inode);
1990 * We also have to set last_sub_trans to the current log transid,
1991 * otherwise subsequent syncs to a file that's been synced in this
1992 * transaction will appear to have already occurred.
1994 spin_lock(&BTRFS_I(inode)->lock);
1995 BTRFS_I(inode)->last_sub_trans = root->log_transid;
1996 spin_unlock(&BTRFS_I(inode)->lock);
1997 if (num_written > 0)
1998 num_written = generic_write_sync(iocb, num_written);
2001 atomic_dec(&BTRFS_I(inode)->sync_writers);
2003 current->backing_dev_info = NULL;
2004 return num_written ? num_written : err;
2007 int btrfs_release_file(struct inode *inode, struct file *filp)
2009 struct btrfs_file_private *private = filp->private_data;
2011 if (private && private->filldir_buf)
2012 kfree(private->filldir_buf);
2014 filp->private_data = NULL;
2017 * ordered_data_close is set by setattr when we are about to truncate
2018 * a file from a non-zero size to a zero size. This tries to
2019 * flush down new bytes that may have been written if the
2020 * application were using truncate to replace a file in place.
2022 if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
2023 &BTRFS_I(inode)->runtime_flags))
2024 filemap_flush(inode->i_mapping);
2028 static int start_ordered_ops(struct inode *inode, loff_t start, loff_t end)
2031 struct blk_plug plug;
2034 * This is only called in fsync, which would do synchronous writes, so
2035 * a plug can merge adjacent IOs as much as possible. Esp. in case of
2036 * multiple disks using raid profile, a large IO can be split to
2037 * several segments of stripe length (currently 64K).
2039 blk_start_plug(&plug);
2040 atomic_inc(&BTRFS_I(inode)->sync_writers);
2041 ret = btrfs_fdatawrite_range(inode, start, end);
2042 atomic_dec(&BTRFS_I(inode)->sync_writers);
2043 blk_finish_plug(&plug);
2049 * fsync call for both files and directories. This logs the inode into
2050 * the tree log instead of forcing full commits whenever possible.
2052 * It needs to call filemap_fdatawait so that all ordered extent updates are
2053 * in the metadata btree are up to date for copying to the log.
2055 * It drops the inode mutex before doing the tree log commit. This is an
2056 * important optimization for directories because holding the mutex prevents
2057 * new operations on the dir while we write to disk.
2059 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
2061 struct dentry *dentry = file_dentry(file);
2062 struct inode *inode = d_inode(dentry);
2063 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2064 struct btrfs_root *root = BTRFS_I(inode)->root;
2065 struct btrfs_trans_handle *trans;
2066 struct btrfs_log_ctx ctx;
2071 * The range length can be represented by u64, we have to do the typecasts
2072 * to avoid signed overflow if it's [0, LLONG_MAX] eg. from fsync()
2074 len = (u64)end - (u64)start + 1;
2075 trace_btrfs_sync_file(file, datasync);
2077 btrfs_init_log_ctx(&ctx, inode);
2080 * We write the dirty pages in the range and wait until they complete
2081 * out of the ->i_mutex. If so, we can flush the dirty pages by
2082 * multi-task, and make the performance up. See
2083 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
2085 ret = start_ordered_ops(inode, start, end);
2092 * We take the dio_sem here because the tree log stuff can race with
2093 * lockless dio writes and get an extent map logged for an extent we
2094 * never waited on. We need it this high up for lockdep reasons.
2096 down_write(&BTRFS_I(inode)->dio_sem);
2098 atomic_inc(&root->log_batch);
2101 * Before we acquired the inode's lock, someone may have dirtied more
2102 * pages in the target range. We need to make sure that writeback for
2103 * any such pages does not start while we are logging the inode, because
2104 * if it does, any of the following might happen when we are not doing a
2107 * 1) We log an extent after its writeback finishes but before its
2108 * checksums are added to the csum tree, leading to -EIO errors
2109 * when attempting to read the extent after a log replay.
2111 * 2) We can end up logging an extent before its writeback finishes.
2112 * Therefore after the log replay we will have a file extent item
2113 * pointing to an unwritten extent (and no data checksums as well).
2115 * So trigger writeback for any eventual new dirty pages and then we
2116 * wait for all ordered extents to complete below.
2118 ret = start_ordered_ops(inode, start, end);
2120 inode_unlock(inode);
2125 * We have to do this here to avoid the priority inversion of waiting on
2126 * IO of a lower priority task while holding a transaction open.
2128 ret = btrfs_wait_ordered_range(inode, start, len);
2130 up_write(&BTRFS_I(inode)->dio_sem);
2131 inode_unlock(inode);
2134 atomic_inc(&root->log_batch);
2137 if (btrfs_inode_in_log(BTRFS_I(inode), fs_info->generation) ||
2138 BTRFS_I(inode)->last_trans <= fs_info->last_trans_committed) {
2140 * We've had everything committed since the last time we were
2141 * modified so clear this flag in case it was set for whatever
2142 * reason, it's no longer relevant.
2144 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2145 &BTRFS_I(inode)->runtime_flags);
2147 * An ordered extent might have started before and completed
2148 * already with io errors, in which case the inode was not
2149 * updated and we end up here. So check the inode's mapping
2150 * for any errors that might have happened since we last
2151 * checked called fsync.
2153 ret = filemap_check_wb_err(inode->i_mapping, file->f_wb_err);
2154 up_write(&BTRFS_I(inode)->dio_sem);
2155 inode_unlock(inode);
2160 * We use start here because we will need to wait on the IO to complete
2161 * in btrfs_sync_log, which could require joining a transaction (for
2162 * example checking cross references in the nocow path). If we use join
2163 * here we could get into a situation where we're waiting on IO to
2164 * happen that is blocked on a transaction trying to commit. With start
2165 * we inc the extwriter counter, so we wait for all extwriters to exit
2166 * before we start blocking joiners. This comment is to keep somebody
2167 * from thinking they are super smart and changing this to
2168 * btrfs_join_transaction *cough*Josef*cough*.
2170 trans = btrfs_start_transaction(root, 0);
2171 if (IS_ERR(trans)) {
2172 ret = PTR_ERR(trans);
2173 up_write(&BTRFS_I(inode)->dio_sem);
2174 inode_unlock(inode);
2178 ret = btrfs_log_dentry_safe(trans, dentry, start, end, &ctx);
2180 /* Fallthrough and commit/free transaction. */
2184 /* we've logged all the items and now have a consistent
2185 * version of the file in the log. It is possible that
2186 * someone will come in and modify the file, but that's
2187 * fine because the log is consistent on disk, and we
2188 * have references to all of the file's extents
2190 * It is possible that someone will come in and log the
2191 * file again, but that will end up using the synchronization
2192 * inside btrfs_sync_log to keep things safe.
2194 up_write(&BTRFS_I(inode)->dio_sem);
2195 inode_unlock(inode);
2197 if (ret != BTRFS_NO_LOG_SYNC) {
2199 ret = btrfs_sync_log(trans, root, &ctx);
2201 ret = btrfs_end_transaction(trans);
2205 ret = btrfs_commit_transaction(trans);
2207 ret = btrfs_end_transaction(trans);
2210 ASSERT(list_empty(&ctx.list));
2211 err = file_check_and_advance_wb_err(file);
2214 return ret > 0 ? -EIO : ret;
2217 static const struct vm_operations_struct btrfs_file_vm_ops = {
2218 .fault = filemap_fault,
2219 .map_pages = filemap_map_pages,
2220 .page_mkwrite = btrfs_page_mkwrite,
2223 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
2225 struct address_space *mapping = filp->f_mapping;
2227 if (!mapping->a_ops->readpage)
2230 file_accessed(filp);
2231 vma->vm_ops = &btrfs_file_vm_ops;
2236 static int hole_mergeable(struct btrfs_inode *inode, struct extent_buffer *leaf,
2237 int slot, u64 start, u64 end)
2239 struct btrfs_file_extent_item *fi;
2240 struct btrfs_key key;
2242 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
2245 btrfs_item_key_to_cpu(leaf, &key, slot);
2246 if (key.objectid != btrfs_ino(inode) ||
2247 key.type != BTRFS_EXTENT_DATA_KEY)
2250 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2252 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2255 if (btrfs_file_extent_disk_bytenr(leaf, fi))
2258 if (key.offset == end)
2260 if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
2265 static int fill_holes(struct btrfs_trans_handle *trans,
2266 struct btrfs_inode *inode,
2267 struct btrfs_path *path, u64 offset, u64 end)
2269 struct btrfs_fs_info *fs_info = trans->fs_info;
2270 struct btrfs_root *root = inode->root;
2271 struct extent_buffer *leaf;
2272 struct btrfs_file_extent_item *fi;
2273 struct extent_map *hole_em;
2274 struct extent_map_tree *em_tree = &inode->extent_tree;
2275 struct btrfs_key key;
2278 if (btrfs_fs_incompat(fs_info, NO_HOLES))
2281 key.objectid = btrfs_ino(inode);
2282 key.type = BTRFS_EXTENT_DATA_KEY;
2283 key.offset = offset;
2285 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2288 * We should have dropped this offset, so if we find it then
2289 * something has gone horribly wrong.
2296 leaf = path->nodes[0];
2297 if (hole_mergeable(inode, leaf, path->slots[0] - 1, offset, end)) {
2301 fi = btrfs_item_ptr(leaf, path->slots[0],
2302 struct btrfs_file_extent_item);
2303 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2305 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2306 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2307 btrfs_set_file_extent_offset(leaf, fi, 0);
2308 btrfs_mark_buffer_dirty(leaf);
2312 if (hole_mergeable(inode, leaf, path->slots[0], offset, end)) {
2315 key.offset = offset;
2316 btrfs_set_item_key_safe(fs_info, path, &key);
2317 fi = btrfs_item_ptr(leaf, path->slots[0],
2318 struct btrfs_file_extent_item);
2319 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2321 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2322 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2323 btrfs_set_file_extent_offset(leaf, fi, 0);
2324 btrfs_mark_buffer_dirty(leaf);
2327 btrfs_release_path(path);
2329 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode),
2330 offset, 0, 0, end - offset, 0, end - offset, 0, 0, 0);
2335 btrfs_release_path(path);
2337 hole_em = alloc_extent_map();
2339 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2340 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
2342 hole_em->start = offset;
2343 hole_em->len = end - offset;
2344 hole_em->ram_bytes = hole_em->len;
2345 hole_em->orig_start = offset;
2347 hole_em->block_start = EXTENT_MAP_HOLE;
2348 hole_em->block_len = 0;
2349 hole_em->orig_block_len = 0;
2350 hole_em->bdev = fs_info->fs_devices->latest_bdev;
2351 hole_em->compress_type = BTRFS_COMPRESS_NONE;
2352 hole_em->generation = trans->transid;
2355 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2356 write_lock(&em_tree->lock);
2357 ret = add_extent_mapping(em_tree, hole_em, 1);
2358 write_unlock(&em_tree->lock);
2359 } while (ret == -EEXIST);
2360 free_extent_map(hole_em);
2362 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2363 &inode->runtime_flags);
2370 * Find a hole extent on given inode and change start/len to the end of hole
2371 * extent.(hole/vacuum extent whose em->start <= start &&
2372 * em->start + em->len > start)
2373 * When a hole extent is found, return 1 and modify start/len.
2375 static int find_first_non_hole(struct inode *inode, u64 *start, u64 *len)
2377 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2378 struct extent_map *em;
2381 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0,
2382 round_down(*start, fs_info->sectorsize),
2383 round_up(*len, fs_info->sectorsize), 0);
2387 /* Hole or vacuum extent(only exists in no-hole mode) */
2388 if (em->block_start == EXTENT_MAP_HOLE) {
2390 *len = em->start + em->len > *start + *len ?
2391 0 : *start + *len - em->start - em->len;
2392 *start = em->start + em->len;
2394 free_extent_map(em);
2398 static int btrfs_punch_hole_lock_range(struct inode *inode,
2399 const u64 lockstart,
2401 struct extent_state **cached_state)
2404 struct btrfs_ordered_extent *ordered;
2407 truncate_pagecache_range(inode, lockstart, lockend);
2409 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2411 ordered = btrfs_lookup_first_ordered_extent(inode, lockend);
2414 * We need to make sure we have no ordered extents in this range
2415 * and nobody raced in and read a page in this range, if we did
2416 * we need to try again.
2419 (ordered->file_offset + ordered->len <= lockstart ||
2420 ordered->file_offset > lockend)) &&
2421 !filemap_range_has_page(inode->i_mapping,
2422 lockstart, lockend)) {
2424 btrfs_put_ordered_extent(ordered);
2428 btrfs_put_ordered_extent(ordered);
2429 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
2430 lockend, cached_state);
2431 ret = btrfs_wait_ordered_range(inode, lockstart,
2432 lockend - lockstart + 1);
2439 static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
2441 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2442 struct btrfs_root *root = BTRFS_I(inode)->root;
2443 struct extent_state *cached_state = NULL;
2444 struct btrfs_path *path;
2445 struct btrfs_block_rsv *rsv;
2446 struct btrfs_trans_handle *trans;
2451 u64 orig_start = offset;
2453 u64 min_size = btrfs_calc_trans_metadata_size(fs_info, 1);
2457 unsigned int rsv_count;
2459 bool no_holes = btrfs_fs_incompat(fs_info, NO_HOLES);
2461 bool truncated_block = false;
2462 bool updated_inode = false;
2464 ret = btrfs_wait_ordered_range(inode, offset, len);
2469 ino_size = round_up(inode->i_size, fs_info->sectorsize);
2470 ret = find_first_non_hole(inode, &offset, &len);
2472 goto out_only_mutex;
2474 /* Already in a large hole */
2476 goto out_only_mutex;
2479 lockstart = round_up(offset, btrfs_inode_sectorsize(inode));
2480 lockend = round_down(offset + len,
2481 btrfs_inode_sectorsize(inode)) - 1;
2482 same_block = (BTRFS_BYTES_TO_BLKS(fs_info, offset))
2483 == (BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1));
2485 * We needn't truncate any block which is beyond the end of the file
2486 * because we are sure there is no data there.
2489 * Only do this if we are in the same block and we aren't doing the
2492 if (same_block && len < fs_info->sectorsize) {
2493 if (offset < ino_size) {
2494 truncated_block = true;
2495 ret = btrfs_truncate_block(inode, offset, len, 0);
2499 goto out_only_mutex;
2502 /* zero back part of the first block */
2503 if (offset < ino_size) {
2504 truncated_block = true;
2505 ret = btrfs_truncate_block(inode, offset, 0, 0);
2507 inode_unlock(inode);
2512 /* Check the aligned pages after the first unaligned page,
2513 * if offset != orig_start, which means the first unaligned page
2514 * including several following pages are already in holes,
2515 * the extra check can be skipped */
2516 if (offset == orig_start) {
2517 /* after truncate page, check hole again */
2518 len = offset + len - lockstart;
2520 ret = find_first_non_hole(inode, &offset, &len);
2522 goto out_only_mutex;
2525 goto out_only_mutex;
2530 /* Check the tail unaligned part is in a hole */
2531 tail_start = lockend + 1;
2532 tail_len = offset + len - tail_start;
2534 ret = find_first_non_hole(inode, &tail_start, &tail_len);
2535 if (unlikely(ret < 0))
2536 goto out_only_mutex;
2538 /* zero the front end of the last page */
2539 if (tail_start + tail_len < ino_size) {
2540 truncated_block = true;
2541 ret = btrfs_truncate_block(inode,
2542 tail_start + tail_len,
2545 goto out_only_mutex;
2550 if (lockend < lockstart) {
2552 goto out_only_mutex;
2555 ret = btrfs_punch_hole_lock_range(inode, lockstart, lockend,
2558 inode_unlock(inode);
2559 goto out_only_mutex;
2562 path = btrfs_alloc_path();
2568 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
2573 rsv->size = btrfs_calc_trans_metadata_size(fs_info, 1);
2577 * 1 - update the inode
2578 * 1 - removing the extents in the range
2579 * 1 - adding the hole extent if no_holes isn't set
2581 rsv_count = no_holes ? 2 : 3;
2582 trans = btrfs_start_transaction(root, rsv_count);
2583 if (IS_ERR(trans)) {
2584 err = PTR_ERR(trans);
2588 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
2591 trans->block_rsv = rsv;
2593 cur_offset = lockstart;
2594 len = lockend - cur_offset;
2595 while (cur_offset < lockend) {
2596 ret = __btrfs_drop_extents(trans, root, inode, path,
2597 cur_offset, lockend + 1,
2598 &drop_end, 1, 0, 0, NULL);
2602 trans->block_rsv = &fs_info->trans_block_rsv;
2604 if (cur_offset < drop_end && cur_offset < ino_size) {
2605 ret = fill_holes(trans, BTRFS_I(inode), path,
2606 cur_offset, drop_end);
2609 * If we failed then we didn't insert our hole
2610 * entries for the area we dropped, so now the
2611 * fs is corrupted, so we must abort the
2614 btrfs_abort_transaction(trans, ret);
2620 cur_offset = drop_end;
2622 ret = btrfs_update_inode(trans, root, inode);
2628 btrfs_end_transaction(trans);
2629 btrfs_btree_balance_dirty(fs_info);
2631 trans = btrfs_start_transaction(root, rsv_count);
2632 if (IS_ERR(trans)) {
2633 ret = PTR_ERR(trans);
2638 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
2639 rsv, min_size, false);
2640 BUG_ON(ret); /* shouldn't happen */
2641 trans->block_rsv = rsv;
2643 ret = find_first_non_hole(inode, &cur_offset, &len);
2644 if (unlikely(ret < 0))
2657 trans->block_rsv = &fs_info->trans_block_rsv;
2659 * If we are using the NO_HOLES feature we might have had already an
2660 * hole that overlaps a part of the region [lockstart, lockend] and
2661 * ends at (or beyond) lockend. Since we have no file extent items to
2662 * represent holes, drop_end can be less than lockend and so we must
2663 * make sure we have an extent map representing the existing hole (the
2664 * call to __btrfs_drop_extents() might have dropped the existing extent
2665 * map representing the existing hole), otherwise the fast fsync path
2666 * will not record the existence of the hole region
2667 * [existing_hole_start, lockend].
2669 if (drop_end <= lockend)
2670 drop_end = lockend + 1;
2672 * Don't insert file hole extent item if it's for a range beyond eof
2673 * (because it's useless) or if it represents a 0 bytes range (when
2674 * cur_offset == drop_end).
2676 if (cur_offset < ino_size && cur_offset < drop_end) {
2677 ret = fill_holes(trans, BTRFS_I(inode), path,
2678 cur_offset, drop_end);
2680 /* Same comment as above. */
2681 btrfs_abort_transaction(trans, ret);
2691 inode_inc_iversion(inode);
2692 inode->i_mtime = inode->i_ctime = current_time(inode);
2694 trans->block_rsv = &fs_info->trans_block_rsv;
2695 ret = btrfs_update_inode(trans, root, inode);
2696 updated_inode = true;
2697 btrfs_end_transaction(trans);
2698 btrfs_btree_balance_dirty(fs_info);
2700 btrfs_free_path(path);
2701 btrfs_free_block_rsv(fs_info, rsv);
2703 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2706 if (!updated_inode && truncated_block && !ret && !err) {
2708 * If we only end up zeroing part of a page, we still need to
2709 * update the inode item, so that all the time fields are
2710 * updated as well as the necessary btrfs inode in memory fields
2711 * for detecting, at fsync time, if the inode isn't yet in the
2712 * log tree or it's there but not up to date.
2714 trans = btrfs_start_transaction(root, 1);
2715 if (IS_ERR(trans)) {
2716 err = PTR_ERR(trans);
2718 err = btrfs_update_inode(trans, root, inode);
2719 ret = btrfs_end_transaction(trans);
2722 inode_unlock(inode);
2728 /* Helper structure to record which range is already reserved */
2729 struct falloc_range {
2730 struct list_head list;
2736 * Helper function to add falloc range
2738 * Caller should have locked the larger range of extent containing
2741 static int add_falloc_range(struct list_head *head, u64 start, u64 len)
2743 struct falloc_range *prev = NULL;
2744 struct falloc_range *range = NULL;
2746 if (list_empty(head))
2750 * As fallocate iterate by bytenr order, we only need to check
2753 prev = list_entry(head->prev, struct falloc_range, list);
2754 if (prev->start + prev->len == start) {
2759 range = kmalloc(sizeof(*range), GFP_KERNEL);
2762 range->start = start;
2764 list_add_tail(&range->list, head);
2768 static int btrfs_fallocate_update_isize(struct inode *inode,
2772 struct btrfs_trans_handle *trans;
2773 struct btrfs_root *root = BTRFS_I(inode)->root;
2777 if (mode & FALLOC_FL_KEEP_SIZE || end <= i_size_read(inode))
2780 trans = btrfs_start_transaction(root, 1);
2782 return PTR_ERR(trans);
2784 inode->i_ctime = current_time(inode);
2785 i_size_write(inode, end);
2786 btrfs_ordered_update_i_size(inode, end, NULL);
2787 ret = btrfs_update_inode(trans, root, inode);
2788 ret2 = btrfs_end_transaction(trans);
2790 return ret ? ret : ret2;
2794 RANGE_BOUNDARY_WRITTEN_EXTENT = 0,
2795 RANGE_BOUNDARY_PREALLOC_EXTENT = 1,
2796 RANGE_BOUNDARY_HOLE = 2,
2799 static int btrfs_zero_range_check_range_boundary(struct inode *inode,
2802 const u64 sectorsize = btrfs_inode_sectorsize(inode);
2803 struct extent_map *em;
2806 offset = round_down(offset, sectorsize);
2807 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, offset, sectorsize, 0);
2811 if (em->block_start == EXTENT_MAP_HOLE)
2812 ret = RANGE_BOUNDARY_HOLE;
2813 else if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2814 ret = RANGE_BOUNDARY_PREALLOC_EXTENT;
2816 ret = RANGE_BOUNDARY_WRITTEN_EXTENT;
2818 free_extent_map(em);
2822 static int btrfs_zero_range(struct inode *inode,
2827 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2828 struct extent_map *em;
2829 struct extent_changeset *data_reserved = NULL;
2832 const u64 sectorsize = btrfs_inode_sectorsize(inode);
2833 u64 alloc_start = round_down(offset, sectorsize);
2834 u64 alloc_end = round_up(offset + len, sectorsize);
2835 u64 bytes_to_reserve = 0;
2836 bool space_reserved = false;
2838 inode_dio_wait(inode);
2840 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0,
2841 alloc_start, alloc_end - alloc_start, 0);
2848 * Avoid hole punching and extent allocation for some cases. More cases
2849 * could be considered, but these are unlikely common and we keep things
2850 * as simple as possible for now. Also, intentionally, if the target
2851 * range contains one or more prealloc extents together with regular
2852 * extents and holes, we drop all the existing extents and allocate a
2853 * new prealloc extent, so that we get a larger contiguous disk extent.
2855 if (em->start <= alloc_start &&
2856 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
2857 const u64 em_end = em->start + em->len;
2859 if (em_end >= offset + len) {
2861 * The whole range is already a prealloc extent,
2862 * do nothing except updating the inode's i_size if
2865 free_extent_map(em);
2866 ret = btrfs_fallocate_update_isize(inode, offset + len,
2871 * Part of the range is already a prealloc extent, so operate
2872 * only on the remaining part of the range.
2874 alloc_start = em_end;
2875 ASSERT(IS_ALIGNED(alloc_start, sectorsize));
2876 len = offset + len - alloc_start;
2877 offset = alloc_start;
2878 alloc_hint = em->block_start + em->len;
2880 free_extent_map(em);
2882 if (BTRFS_BYTES_TO_BLKS(fs_info, offset) ==
2883 BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1)) {
2884 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0,
2885 alloc_start, sectorsize, 0);
2891 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
2892 free_extent_map(em);
2893 ret = btrfs_fallocate_update_isize(inode, offset + len,
2897 if (len < sectorsize && em->block_start != EXTENT_MAP_HOLE) {
2898 free_extent_map(em);
2899 ret = btrfs_truncate_block(inode, offset, len, 0);
2901 ret = btrfs_fallocate_update_isize(inode,
2906 free_extent_map(em);
2907 alloc_start = round_down(offset, sectorsize);
2908 alloc_end = alloc_start + sectorsize;
2912 alloc_start = round_up(offset, sectorsize);
2913 alloc_end = round_down(offset + len, sectorsize);
2916 * For unaligned ranges, check the pages at the boundaries, they might
2917 * map to an extent, in which case we need to partially zero them, or
2918 * they might map to a hole, in which case we need our allocation range
2921 if (!IS_ALIGNED(offset, sectorsize)) {
2922 ret = btrfs_zero_range_check_range_boundary(inode, offset);
2925 if (ret == RANGE_BOUNDARY_HOLE) {
2926 alloc_start = round_down(offset, sectorsize);
2928 } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) {
2929 ret = btrfs_truncate_block(inode, offset, 0, 0);
2937 if (!IS_ALIGNED(offset + len, sectorsize)) {
2938 ret = btrfs_zero_range_check_range_boundary(inode,
2942 if (ret == RANGE_BOUNDARY_HOLE) {
2943 alloc_end = round_up(offset + len, sectorsize);
2945 } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) {
2946 ret = btrfs_truncate_block(inode, offset + len, 0, 1);
2955 if (alloc_start < alloc_end) {
2956 struct extent_state *cached_state = NULL;
2957 const u64 lockstart = alloc_start;
2958 const u64 lockend = alloc_end - 1;
2960 bytes_to_reserve = alloc_end - alloc_start;
2961 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode),
2965 space_reserved = true;
2966 ret = btrfs_qgroup_reserve_data(inode, &data_reserved,
2967 alloc_start, bytes_to_reserve);
2970 ret = btrfs_punch_hole_lock_range(inode, lockstart, lockend,
2974 ret = btrfs_prealloc_file_range(inode, mode, alloc_start,
2975 alloc_end - alloc_start,
2977 offset + len, &alloc_hint);
2978 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
2979 lockend, &cached_state);
2980 /* btrfs_prealloc_file_range releases reserved space on error */
2982 space_reserved = false;
2986 ret = btrfs_fallocate_update_isize(inode, offset + len, mode);
2988 if (ret && space_reserved)
2989 btrfs_free_reserved_data_space(inode, data_reserved,
2990 alloc_start, bytes_to_reserve);
2991 extent_changeset_free(data_reserved);
2996 static long btrfs_fallocate(struct file *file, int mode,
2997 loff_t offset, loff_t len)
2999 struct inode *inode = file_inode(file);
3000 struct extent_state *cached_state = NULL;
3001 struct extent_changeset *data_reserved = NULL;
3002 struct falloc_range *range;
3003 struct falloc_range *tmp;
3004 struct list_head reserve_list;
3012 struct extent_map *em;
3013 int blocksize = btrfs_inode_sectorsize(inode);
3016 alloc_start = round_down(offset, blocksize);
3017 alloc_end = round_up(offset + len, blocksize);
3018 cur_offset = alloc_start;
3020 /* Make sure we aren't being give some crap mode */
3021 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
3022 FALLOC_FL_ZERO_RANGE))
3025 if (mode & FALLOC_FL_PUNCH_HOLE)
3026 return btrfs_punch_hole(inode, offset, len);
3029 * Only trigger disk allocation, don't trigger qgroup reserve
3031 * For qgroup space, it will be checked later.
3033 if (!(mode & FALLOC_FL_ZERO_RANGE)) {
3034 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode),
3035 alloc_end - alloc_start);
3042 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) {
3043 ret = inode_newsize_ok(inode, offset + len);
3049 * TODO: Move these two operations after we have checked
3050 * accurate reserved space, or fallocate can still fail but
3051 * with page truncated or size expanded.
3053 * But that's a minor problem and won't do much harm BTW.
3055 if (alloc_start > inode->i_size) {
3056 ret = btrfs_cont_expand(inode, i_size_read(inode),
3060 } else if (offset + len > inode->i_size) {
3062 * If we are fallocating from the end of the file onward we
3063 * need to zero out the end of the block if i_size lands in the
3064 * middle of a block.
3066 ret = btrfs_truncate_block(inode, inode->i_size, 0, 0);
3072 * wait for ordered IO before we have any locks. We'll loop again
3073 * below with the locks held.
3075 ret = btrfs_wait_ordered_range(inode, alloc_start,
3076 alloc_end - alloc_start);
3080 if (mode & FALLOC_FL_ZERO_RANGE) {
3081 ret = btrfs_zero_range(inode, offset, len, mode);
3082 inode_unlock(inode);
3086 locked_end = alloc_end - 1;
3088 struct btrfs_ordered_extent *ordered;
3090 /* the extent lock is ordered inside the running
3093 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
3094 locked_end, &cached_state);
3095 ordered = btrfs_lookup_first_ordered_extent(inode, locked_end);
3098 ordered->file_offset + ordered->len > alloc_start &&
3099 ordered->file_offset < alloc_end) {
3100 btrfs_put_ordered_extent(ordered);
3101 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
3102 alloc_start, locked_end,
3105 * we can't wait on the range with the transaction
3106 * running or with the extent lock held
3108 ret = btrfs_wait_ordered_range(inode, alloc_start,
3109 alloc_end - alloc_start);
3114 btrfs_put_ordered_extent(ordered);
3119 /* First, check if we exceed the qgroup limit */
3120 INIT_LIST_HEAD(&reserve_list);
3121 while (cur_offset < alloc_end) {
3122 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
3123 alloc_end - cur_offset, 0);
3128 last_byte = min(extent_map_end(em), alloc_end);
3129 actual_end = min_t(u64, extent_map_end(em), offset + len);
3130 last_byte = ALIGN(last_byte, blocksize);
3131 if (em->block_start == EXTENT_MAP_HOLE ||
3132 (cur_offset >= inode->i_size &&
3133 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
3134 ret = add_falloc_range(&reserve_list, cur_offset,
3135 last_byte - cur_offset);
3137 free_extent_map(em);
3140 ret = btrfs_qgroup_reserve_data(inode, &data_reserved,
3141 cur_offset, last_byte - cur_offset);
3143 cur_offset = last_byte;
3144 free_extent_map(em);
3149 * Do not need to reserve unwritten extent for this
3150 * range, free reserved data space first, otherwise
3151 * it'll result in false ENOSPC error.
3153 btrfs_free_reserved_data_space(inode, data_reserved,
3154 cur_offset, last_byte - cur_offset);
3156 free_extent_map(em);
3157 cur_offset = last_byte;
3161 * If ret is still 0, means we're OK to fallocate.
3162 * Or just cleanup the list and exit.
3164 list_for_each_entry_safe(range, tmp, &reserve_list, list) {
3166 ret = btrfs_prealloc_file_range(inode, mode,
3168 range->len, i_blocksize(inode),
3169 offset + len, &alloc_hint);
3171 btrfs_free_reserved_data_space(inode,
3172 data_reserved, range->start,
3174 list_del(&range->list);
3181 * We didn't need to allocate any more space, but we still extended the
3182 * size of the file so we need to update i_size and the inode item.
3184 ret = btrfs_fallocate_update_isize(inode, actual_end, mode);
3186 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
3189 inode_unlock(inode);
3190 /* Let go of our reservation. */
3191 if (ret != 0 && !(mode & FALLOC_FL_ZERO_RANGE))
3192 btrfs_free_reserved_data_space(inode, data_reserved,
3193 cur_offset, alloc_end - cur_offset);
3194 extent_changeset_free(data_reserved);
3198 static int find_desired_extent(struct inode *inode, loff_t *offset, int whence)
3200 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3201 struct extent_map *em = NULL;
3202 struct extent_state *cached_state = NULL;
3209 if (inode->i_size == 0)
3213 * *offset can be negative, in this case we start finding DATA/HOLE from
3214 * the very start of the file.
3216 start = max_t(loff_t, 0, *offset);
3218 lockstart = round_down(start, fs_info->sectorsize);
3219 lockend = round_up(i_size_read(inode),
3220 fs_info->sectorsize);
3221 if (lockend <= lockstart)
3222 lockend = lockstart + fs_info->sectorsize;
3224 len = lockend - lockstart + 1;
3226 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
3229 while (start < inode->i_size) {
3230 em = btrfs_get_extent_fiemap(BTRFS_I(inode), start, len);
3237 if (whence == SEEK_HOLE &&
3238 (em->block_start == EXTENT_MAP_HOLE ||
3239 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
3241 else if (whence == SEEK_DATA &&
3242 (em->block_start != EXTENT_MAP_HOLE &&
3243 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
3246 start = em->start + em->len;
3247 free_extent_map(em);
3251 free_extent_map(em);
3253 if (whence == SEEK_DATA && start >= inode->i_size)
3256 *offset = min_t(loff_t, start, inode->i_size);
3258 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
3263 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
3265 struct inode *inode = file->f_mapping->host;
3272 offset = generic_file_llseek(file, offset, whence);
3276 if (offset >= i_size_read(inode)) {
3277 inode_unlock(inode);
3281 ret = find_desired_extent(inode, &offset, whence);
3283 inode_unlock(inode);
3288 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
3290 inode_unlock(inode);
3294 static int btrfs_file_open(struct inode *inode, struct file *filp)
3296 filp->f_mode |= FMODE_NOWAIT;
3297 return generic_file_open(inode, filp);
3300 const struct file_operations btrfs_file_operations = {
3301 .llseek = btrfs_file_llseek,
3302 .read_iter = generic_file_read_iter,
3303 .splice_read = generic_file_splice_read,
3304 .write_iter = btrfs_file_write_iter,
3305 .mmap = btrfs_file_mmap,
3306 .open = btrfs_file_open,
3307 .release = btrfs_release_file,
3308 .fsync = btrfs_sync_file,
3309 .fallocate = btrfs_fallocate,
3310 .unlocked_ioctl = btrfs_ioctl,
3311 #ifdef CONFIG_COMPAT
3312 .compat_ioctl = btrfs_compat_ioctl,
3314 .remap_file_range = btrfs_remap_file_range,
3317 void __cold btrfs_auto_defrag_exit(void)
3319 kmem_cache_destroy(btrfs_inode_defrag_cachep);
3322 int __init btrfs_auto_defrag_init(void)
3324 btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
3325 sizeof(struct inode_defrag), 0,
3328 if (!btrfs_inode_defrag_cachep)
3334 int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end)
3339 * So with compression we will find and lock a dirty page and clear the
3340 * first one as dirty, setup an async extent, and immediately return
3341 * with the entire range locked but with nobody actually marked with
3342 * writeback. So we can't just filemap_write_and_wait_range() and
3343 * expect it to work since it will just kick off a thread to do the
3344 * actual work. So we need to call filemap_fdatawrite_range _again_
3345 * since it will wait on the page lock, which won't be unlocked until
3346 * after the pages have been marked as writeback and so we're good to go
3347 * from there. We have to do this otherwise we'll miss the ordered
3348 * extents and that results in badness. Please Josef, do not think you
3349 * know better and pull this out at some point in the future, it is
3350 * right and you are wrong.
3352 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
3353 if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
3354 &BTRFS_I(inode)->runtime_flags))
3355 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);