2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 #include <linux/pagemap.h>
21 #include <linux/highmem.h>
22 #include <linux/time.h>
23 #include <linux/init.h>
24 #include <linux/string.h>
25 #include <linux/backing-dev.h>
26 #include <linux/mpage.h>
27 #include <linux/falloc.h>
28 #include <linux/swap.h>
29 #include <linux/writeback.h>
30 #include <linux/compat.h>
31 #include <linux/slab.h>
32 #include <linux/btrfs.h>
33 #include <linux/uio.h>
36 #include "transaction.h"
37 #include "btrfs_inode.h"
38 #include "print-tree.h"
43 #include "compression.h"
45 static struct kmem_cache *btrfs_inode_defrag_cachep;
47 * when auto defrag is enabled we
48 * queue up these defrag structs to remember which
49 * inodes need defragging passes
52 struct rb_node rb_node;
56 * transid where the defrag was added, we search for
57 * extents newer than this
64 /* last offset we were able to defrag */
67 /* if we've wrapped around back to zero once already */
71 static int __compare_inode_defrag(struct inode_defrag *defrag1,
72 struct inode_defrag *defrag2)
74 if (defrag1->root > defrag2->root)
76 else if (defrag1->root < defrag2->root)
78 else if (defrag1->ino > defrag2->ino)
80 else if (defrag1->ino < defrag2->ino)
86 /* pop a record for an inode into the defrag tree. The lock
87 * must be held already
89 * If you're inserting a record for an older transid than an
90 * existing record, the transid already in the tree is lowered
92 * If an existing record is found the defrag item you
95 static int __btrfs_add_inode_defrag(struct btrfs_inode *inode,
96 struct inode_defrag *defrag)
98 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
99 struct inode_defrag *entry;
101 struct rb_node *parent = NULL;
104 p = &fs_info->defrag_inodes.rb_node;
107 entry = rb_entry(parent, struct inode_defrag, rb_node);
109 ret = __compare_inode_defrag(defrag, entry);
111 p = &parent->rb_left;
113 p = &parent->rb_right;
115 /* if we're reinserting an entry for
116 * an old defrag run, make sure to
117 * lower the transid of our existing record
119 if (defrag->transid < entry->transid)
120 entry->transid = defrag->transid;
121 if (defrag->last_offset > entry->last_offset)
122 entry->last_offset = defrag->last_offset;
126 set_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags);
127 rb_link_node(&defrag->rb_node, parent, p);
128 rb_insert_color(&defrag->rb_node, &fs_info->defrag_inodes);
132 static inline int __need_auto_defrag(struct btrfs_fs_info *fs_info)
134 if (!btrfs_test_opt(fs_info, AUTO_DEFRAG))
137 if (btrfs_fs_closing(fs_info))
144 * insert a defrag record for this inode if auto defrag is
147 int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans,
148 struct btrfs_inode *inode)
150 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
151 struct btrfs_root *root = inode->root;
152 struct inode_defrag *defrag;
156 if (!__need_auto_defrag(fs_info))
159 if (test_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags))
163 transid = trans->transid;
165 transid = inode->root->last_trans;
167 defrag = kmem_cache_zalloc(btrfs_inode_defrag_cachep, GFP_NOFS);
171 defrag->ino = btrfs_ino(inode);
172 defrag->transid = transid;
173 defrag->root = root->root_key.objectid;
175 spin_lock(&fs_info->defrag_inodes_lock);
176 if (!test_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags)) {
178 * If we set IN_DEFRAG flag and evict the inode from memory,
179 * and then re-read this inode, this new inode doesn't have
180 * IN_DEFRAG flag. At the case, we may find the existed defrag.
182 ret = __btrfs_add_inode_defrag(inode, defrag);
184 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
186 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
188 spin_unlock(&fs_info->defrag_inodes_lock);
193 * Requeue the defrag object. If there is a defrag object that points to
194 * the same inode in the tree, we will merge them together (by
195 * __btrfs_add_inode_defrag()) and free the one that we want to requeue.
197 static void btrfs_requeue_inode_defrag(struct btrfs_inode *inode,
198 struct inode_defrag *defrag)
200 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
203 if (!__need_auto_defrag(fs_info))
207 * Here we don't check the IN_DEFRAG flag, because we need merge
210 spin_lock(&fs_info->defrag_inodes_lock);
211 ret = __btrfs_add_inode_defrag(inode, defrag);
212 spin_unlock(&fs_info->defrag_inodes_lock);
217 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
221 * pick the defragable inode that we want, if it doesn't exist, we will get
224 static struct inode_defrag *
225 btrfs_pick_defrag_inode(struct btrfs_fs_info *fs_info, u64 root, u64 ino)
227 struct inode_defrag *entry = NULL;
228 struct inode_defrag tmp;
230 struct rb_node *parent = NULL;
236 spin_lock(&fs_info->defrag_inodes_lock);
237 p = fs_info->defrag_inodes.rb_node;
240 entry = rb_entry(parent, struct inode_defrag, rb_node);
242 ret = __compare_inode_defrag(&tmp, entry);
246 p = parent->rb_right;
251 if (parent && __compare_inode_defrag(&tmp, entry) > 0) {
252 parent = rb_next(parent);
254 entry = rb_entry(parent, struct inode_defrag, rb_node);
260 rb_erase(parent, &fs_info->defrag_inodes);
261 spin_unlock(&fs_info->defrag_inodes_lock);
265 void btrfs_cleanup_defrag_inodes(struct btrfs_fs_info *fs_info)
267 struct inode_defrag *defrag;
268 struct rb_node *node;
270 spin_lock(&fs_info->defrag_inodes_lock);
271 node = rb_first(&fs_info->defrag_inodes);
273 rb_erase(node, &fs_info->defrag_inodes);
274 defrag = rb_entry(node, struct inode_defrag, rb_node);
275 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
277 cond_resched_lock(&fs_info->defrag_inodes_lock);
279 node = rb_first(&fs_info->defrag_inodes);
281 spin_unlock(&fs_info->defrag_inodes_lock);
284 #define BTRFS_DEFRAG_BATCH 1024
286 static int __btrfs_run_defrag_inode(struct btrfs_fs_info *fs_info,
287 struct inode_defrag *defrag)
289 struct btrfs_root *inode_root;
291 struct btrfs_key key;
292 struct btrfs_ioctl_defrag_range_args range;
298 key.objectid = defrag->root;
299 key.type = BTRFS_ROOT_ITEM_KEY;
300 key.offset = (u64)-1;
302 index = srcu_read_lock(&fs_info->subvol_srcu);
304 inode_root = btrfs_read_fs_root_no_name(fs_info, &key);
305 if (IS_ERR(inode_root)) {
306 ret = PTR_ERR(inode_root);
310 key.objectid = defrag->ino;
311 key.type = BTRFS_INODE_ITEM_KEY;
313 inode = btrfs_iget(fs_info->sb, &key, inode_root, NULL);
315 ret = PTR_ERR(inode);
318 srcu_read_unlock(&fs_info->subvol_srcu, index);
320 /* do a chunk of defrag */
321 clear_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
322 memset(&range, 0, sizeof(range));
324 range.start = defrag->last_offset;
326 sb_start_write(fs_info->sb);
327 num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid,
329 sb_end_write(fs_info->sb);
331 * if we filled the whole defrag batch, there
332 * must be more work to do. Queue this defrag
335 if (num_defrag == BTRFS_DEFRAG_BATCH) {
336 defrag->last_offset = range.start;
337 btrfs_requeue_inode_defrag(BTRFS_I(inode), defrag);
338 } else if (defrag->last_offset && !defrag->cycled) {
340 * we didn't fill our defrag batch, but
341 * we didn't start at zero. Make sure we loop
342 * around to the start of the file.
344 defrag->last_offset = 0;
346 btrfs_requeue_inode_defrag(BTRFS_I(inode), defrag);
348 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
354 srcu_read_unlock(&fs_info->subvol_srcu, index);
355 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
360 * run through the list of inodes in the FS that need
363 int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info)
365 struct inode_defrag *defrag;
367 u64 root_objectid = 0;
369 atomic_inc(&fs_info->defrag_running);
371 /* Pause the auto defragger. */
372 if (test_bit(BTRFS_FS_STATE_REMOUNTING,
376 if (!__need_auto_defrag(fs_info))
379 /* find an inode to defrag */
380 defrag = btrfs_pick_defrag_inode(fs_info, root_objectid,
383 if (root_objectid || first_ino) {
392 first_ino = defrag->ino + 1;
393 root_objectid = defrag->root;
395 __btrfs_run_defrag_inode(fs_info, defrag);
397 atomic_dec(&fs_info->defrag_running);
400 * during unmount, we use the transaction_wait queue to
401 * wait for the defragger to stop
403 wake_up(&fs_info->transaction_wait);
407 /* simple helper to fault in pages and copy. This should go away
408 * and be replaced with calls into generic code.
410 static noinline int btrfs_copy_from_user(loff_t pos, size_t write_bytes,
411 struct page **prepared_pages,
415 size_t total_copied = 0;
417 int offset = pos & (PAGE_SIZE - 1);
419 while (write_bytes > 0) {
420 size_t count = min_t(size_t,
421 PAGE_SIZE - offset, write_bytes);
422 struct page *page = prepared_pages[pg];
424 * Copy data from userspace to the current page
426 copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
428 /* Flush processor's dcache for this page */
429 flush_dcache_page(page);
432 * if we get a partial write, we can end up with
433 * partially up to date pages. These add
434 * a lot of complexity, so make sure they don't
435 * happen by forcing this copy to be retried.
437 * The rest of the btrfs_file_write code will fall
438 * back to page at a time copies after we return 0.
440 if (!PageUptodate(page) && copied < count)
443 iov_iter_advance(i, copied);
444 write_bytes -= copied;
445 total_copied += copied;
447 /* Return to btrfs_file_write_iter to fault page */
448 if (unlikely(copied == 0))
451 if (copied < PAGE_SIZE - offset) {
462 * unlocks pages after btrfs_file_write is done with them
464 static void btrfs_drop_pages(struct page **pages, size_t num_pages)
467 for (i = 0; i < num_pages; i++) {
468 /* page checked is some magic around finding pages that
469 * have been modified without going through btrfs_set_page_dirty
470 * clear it here. There should be no need to mark the pages
471 * accessed as prepare_pages should have marked them accessed
472 * in prepare_pages via find_or_create_page()
474 ClearPageChecked(pages[i]);
475 unlock_page(pages[i]);
481 * after copy_from_user, pages need to be dirtied and we need to make
482 * sure holes are created between the current EOF and the start of
483 * any next extents (if required).
485 * this also makes the decision about creating an inline extent vs
486 * doing real data extents, marking pages dirty and delalloc as required.
488 int btrfs_dirty_pages(struct inode *inode, struct page **pages,
489 size_t num_pages, loff_t pos, size_t write_bytes,
490 struct extent_state **cached)
492 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
497 u64 end_of_last_block;
498 u64 end_pos = pos + write_bytes;
499 loff_t isize = i_size_read(inode);
501 start_pos = pos & ~((u64) fs_info->sectorsize - 1);
502 num_bytes = round_up(write_bytes + pos - start_pos,
503 fs_info->sectorsize);
505 end_of_last_block = start_pos + num_bytes - 1;
506 err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
511 for (i = 0; i < num_pages; i++) {
512 struct page *p = pages[i];
519 * we've only changed i_size in ram, and we haven't updated
520 * the disk i_size. There is no need to log the inode
524 i_size_write(inode, end_pos);
529 * this drops all the extents in the cache that intersect the range
530 * [start, end]. Existing extents are split as required.
532 void btrfs_drop_extent_cache(struct btrfs_inode *inode, u64 start, u64 end,
535 struct extent_map *em;
536 struct extent_map *split = NULL;
537 struct extent_map *split2 = NULL;
538 struct extent_map_tree *em_tree = &inode->extent_tree;
539 u64 len = end - start + 1;
547 WARN_ON(end < start);
548 if (end == (u64)-1) {
557 split = alloc_extent_map();
559 split2 = alloc_extent_map();
560 if (!split || !split2)
563 write_lock(&em_tree->lock);
564 em = lookup_extent_mapping(em_tree, start, len);
566 write_unlock(&em_tree->lock);
570 gen = em->generation;
571 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
572 if (testend && em->start + em->len >= start + len) {
574 write_unlock(&em_tree->lock);
577 start = em->start + em->len;
579 len = start + len - (em->start + em->len);
581 write_unlock(&em_tree->lock);
584 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
585 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
586 clear_bit(EXTENT_FLAG_LOGGING, &flags);
587 modified = !list_empty(&em->list);
591 if (em->start < start) {
592 split->start = em->start;
593 split->len = start - em->start;
595 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
596 split->orig_start = em->orig_start;
597 split->block_start = em->block_start;
600 split->block_len = em->block_len;
602 split->block_len = split->len;
603 split->orig_block_len = max(split->block_len,
605 split->ram_bytes = em->ram_bytes;
607 split->orig_start = split->start;
608 split->block_len = 0;
609 split->block_start = em->block_start;
610 split->orig_block_len = 0;
611 split->ram_bytes = split->len;
614 split->generation = gen;
615 split->bdev = em->bdev;
616 split->flags = flags;
617 split->compress_type = em->compress_type;
618 replace_extent_mapping(em_tree, em, split, modified);
619 free_extent_map(split);
623 if (testend && em->start + em->len > start + len) {
624 u64 diff = start + len - em->start;
626 split->start = start + len;
627 split->len = em->start + em->len - (start + len);
628 split->bdev = em->bdev;
629 split->flags = flags;
630 split->compress_type = em->compress_type;
631 split->generation = gen;
633 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
634 split->orig_block_len = max(em->block_len,
637 split->ram_bytes = em->ram_bytes;
639 split->block_len = em->block_len;
640 split->block_start = em->block_start;
641 split->orig_start = em->orig_start;
643 split->block_len = split->len;
644 split->block_start = em->block_start
646 split->orig_start = em->orig_start;
649 split->ram_bytes = split->len;
650 split->orig_start = split->start;
651 split->block_len = 0;
652 split->block_start = em->block_start;
653 split->orig_block_len = 0;
656 if (extent_map_in_tree(em)) {
657 replace_extent_mapping(em_tree, em, split,
660 ret = add_extent_mapping(em_tree, split,
662 ASSERT(ret == 0); /* Logic error */
664 free_extent_map(split);
668 if (extent_map_in_tree(em))
669 remove_extent_mapping(em_tree, em);
670 write_unlock(&em_tree->lock);
674 /* once for the tree*/
678 free_extent_map(split);
680 free_extent_map(split2);
684 * this is very complex, but the basic idea is to drop all extents
685 * in the range start - end. hint_block is filled in with a block number
686 * that would be a good hint to the block allocator for this file.
688 * If an extent intersects the range but is not entirely inside the range
689 * it is either truncated or split. Anything entirely inside the range
690 * is deleted from the tree.
692 int __btrfs_drop_extents(struct btrfs_trans_handle *trans,
693 struct btrfs_root *root, struct inode *inode,
694 struct btrfs_path *path, u64 start, u64 end,
695 u64 *drop_end, int drop_cache,
697 u32 extent_item_size,
700 struct btrfs_fs_info *fs_info = root->fs_info;
701 struct extent_buffer *leaf;
702 struct btrfs_file_extent_item *fi;
703 struct btrfs_key key;
704 struct btrfs_key new_key;
705 u64 ino = btrfs_ino(BTRFS_I(inode));
706 u64 search_start = start;
709 u64 extent_offset = 0;
711 u64 last_end = start;
717 int modify_tree = -1;
720 int leafs_visited = 0;
723 btrfs_drop_extent_cache(BTRFS_I(inode), start, end - 1, 0);
725 if (start >= BTRFS_I(inode)->disk_i_size && !replace_extent)
728 update_refs = (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
729 root == fs_info->tree_root);
732 ret = btrfs_lookup_file_extent(trans, root, path, ino,
733 search_start, modify_tree);
736 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
737 leaf = path->nodes[0];
738 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
739 if (key.objectid == ino &&
740 key.type == BTRFS_EXTENT_DATA_KEY)
746 leaf = path->nodes[0];
747 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
749 ret = btrfs_next_leaf(root, path);
757 leaf = path->nodes[0];
761 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
763 if (key.objectid > ino)
765 if (WARN_ON_ONCE(key.objectid < ino) ||
766 key.type < BTRFS_EXTENT_DATA_KEY) {
771 if (key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
774 fi = btrfs_item_ptr(leaf, path->slots[0],
775 struct btrfs_file_extent_item);
776 extent_type = btrfs_file_extent_type(leaf, fi);
778 if (extent_type == BTRFS_FILE_EXTENT_REG ||
779 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
780 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
781 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
782 extent_offset = btrfs_file_extent_offset(leaf, fi);
783 extent_end = key.offset +
784 btrfs_file_extent_num_bytes(leaf, fi);
785 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
786 extent_end = key.offset +
787 btrfs_file_extent_inline_len(leaf,
795 * Don't skip extent items representing 0 byte lengths. They
796 * used to be created (bug) if while punching holes we hit
797 * -ENOSPC condition. So if we find one here, just ensure we
798 * delete it, otherwise we would insert a new file extent item
799 * with the same key (offset) as that 0 bytes length file
800 * extent item in the call to setup_items_for_insert() later
803 if (extent_end == key.offset && extent_end >= search_start) {
804 last_end = extent_end;
805 goto delete_extent_item;
808 if (extent_end <= search_start) {
814 search_start = max(key.offset, start);
815 if (recow || !modify_tree) {
817 btrfs_release_path(path);
822 * | - range to drop - |
823 * | -------- extent -------- |
825 if (start > key.offset && end < extent_end) {
827 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
832 memcpy(&new_key, &key, sizeof(new_key));
833 new_key.offset = start;
834 ret = btrfs_duplicate_item(trans, root, path,
836 if (ret == -EAGAIN) {
837 btrfs_release_path(path);
843 leaf = path->nodes[0];
844 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
845 struct btrfs_file_extent_item);
846 btrfs_set_file_extent_num_bytes(leaf, fi,
849 fi = btrfs_item_ptr(leaf, path->slots[0],
850 struct btrfs_file_extent_item);
852 extent_offset += start - key.offset;
853 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
854 btrfs_set_file_extent_num_bytes(leaf, fi,
856 btrfs_mark_buffer_dirty(leaf);
858 if (update_refs && disk_bytenr > 0) {
859 ret = btrfs_inc_extent_ref(trans, fs_info,
860 disk_bytenr, num_bytes, 0,
861 root->root_key.objectid,
863 start - extent_offset);
864 BUG_ON(ret); /* -ENOMEM */
869 * From here on out we will have actually dropped something, so
870 * last_end can be updated.
872 last_end = extent_end;
875 * | ---- range to drop ----- |
876 * | -------- extent -------- |
878 if (start <= key.offset && end < extent_end) {
879 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
884 memcpy(&new_key, &key, sizeof(new_key));
885 new_key.offset = end;
886 btrfs_set_item_key_safe(fs_info, path, &new_key);
888 extent_offset += end - key.offset;
889 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
890 btrfs_set_file_extent_num_bytes(leaf, fi,
892 btrfs_mark_buffer_dirty(leaf);
893 if (update_refs && disk_bytenr > 0)
894 inode_sub_bytes(inode, end - key.offset);
898 search_start = extent_end;
900 * | ---- range to drop ----- |
901 * | -------- extent -------- |
903 if (start > key.offset && end >= extent_end) {
905 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
910 btrfs_set_file_extent_num_bytes(leaf, fi,
912 btrfs_mark_buffer_dirty(leaf);
913 if (update_refs && disk_bytenr > 0)
914 inode_sub_bytes(inode, extent_end - start);
915 if (end == extent_end)
923 * | ---- range to drop ----- |
924 * | ------ extent ------ |
926 if (start <= key.offset && end >= extent_end) {
929 del_slot = path->slots[0];
932 BUG_ON(del_slot + del_nr != path->slots[0]);
937 extent_type == BTRFS_FILE_EXTENT_INLINE) {
938 inode_sub_bytes(inode,
939 extent_end - key.offset);
940 extent_end = ALIGN(extent_end,
941 fs_info->sectorsize);
942 } else if (update_refs && disk_bytenr > 0) {
943 ret = btrfs_free_extent(trans, fs_info,
944 disk_bytenr, num_bytes, 0,
945 root->root_key.objectid,
946 key.objectid, key.offset -
948 BUG_ON(ret); /* -ENOMEM */
949 inode_sub_bytes(inode,
950 extent_end - key.offset);
953 if (end == extent_end)
956 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
961 ret = btrfs_del_items(trans, root, path, del_slot,
964 btrfs_abort_transaction(trans, ret);
971 btrfs_release_path(path);
978 if (!ret && del_nr > 0) {
980 * Set path->slots[0] to first slot, so that after the delete
981 * if items are move off from our leaf to its immediate left or
982 * right neighbor leafs, we end up with a correct and adjusted
983 * path->slots[0] for our insertion (if replace_extent != 0).
985 path->slots[0] = del_slot;
986 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
988 btrfs_abort_transaction(trans, ret);
991 leaf = path->nodes[0];
993 * If btrfs_del_items() was called, it might have deleted a leaf, in
994 * which case it unlocked our path, so check path->locks[0] matches a
997 if (!ret && replace_extent && leafs_visited == 1 &&
998 (path->locks[0] == BTRFS_WRITE_LOCK_BLOCKING ||
999 path->locks[0] == BTRFS_WRITE_LOCK) &&
1000 btrfs_leaf_free_space(fs_info, leaf) >=
1001 sizeof(struct btrfs_item) + extent_item_size) {
1004 key.type = BTRFS_EXTENT_DATA_KEY;
1006 if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) {
1007 struct btrfs_key slot_key;
1009 btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]);
1010 if (btrfs_comp_cpu_keys(&key, &slot_key) > 0)
1013 setup_items_for_insert(root, path, &key,
1016 sizeof(struct btrfs_item) +
1017 extent_item_size, 1);
1021 if (!replace_extent || !(*key_inserted))
1022 btrfs_release_path(path);
1024 *drop_end = found ? min(end, last_end) : end;
1028 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
1029 struct btrfs_root *root, struct inode *inode, u64 start,
1030 u64 end, int drop_cache)
1032 struct btrfs_path *path;
1035 path = btrfs_alloc_path();
1038 ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL,
1039 drop_cache, 0, 0, NULL);
1040 btrfs_free_path(path);
1044 static int extent_mergeable(struct extent_buffer *leaf, int slot,
1045 u64 objectid, u64 bytenr, u64 orig_offset,
1046 u64 *start, u64 *end)
1048 struct btrfs_file_extent_item *fi;
1049 struct btrfs_key key;
1052 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
1055 btrfs_item_key_to_cpu(leaf, &key, slot);
1056 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
1059 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1060 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
1061 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
1062 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
1063 btrfs_file_extent_compression(leaf, fi) ||
1064 btrfs_file_extent_encryption(leaf, fi) ||
1065 btrfs_file_extent_other_encoding(leaf, fi))
1068 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1069 if ((*start && *start != key.offset) || (*end && *end != extent_end))
1072 *start = key.offset;
1078 * Mark extent in the range start - end as written.
1080 * This changes extent type from 'pre-allocated' to 'regular'. If only
1081 * part of extent is marked as written, the extent will be split into
1084 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
1085 struct btrfs_inode *inode, u64 start, u64 end)
1087 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1088 struct btrfs_root *root = inode->root;
1089 struct extent_buffer *leaf;
1090 struct btrfs_path *path;
1091 struct btrfs_file_extent_item *fi;
1092 struct btrfs_key key;
1093 struct btrfs_key new_key;
1105 u64 ino = btrfs_ino(inode);
1107 path = btrfs_alloc_path();
1114 key.type = BTRFS_EXTENT_DATA_KEY;
1117 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1120 if (ret > 0 && path->slots[0] > 0)
1123 leaf = path->nodes[0];
1124 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1125 if (key.objectid != ino ||
1126 key.type != BTRFS_EXTENT_DATA_KEY) {
1128 btrfs_abort_transaction(trans, ret);
1131 fi = btrfs_item_ptr(leaf, path->slots[0],
1132 struct btrfs_file_extent_item);
1133 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_PREALLOC) {
1135 btrfs_abort_transaction(trans, ret);
1138 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1139 if (key.offset > start || extent_end < end) {
1141 btrfs_abort_transaction(trans, ret);
1145 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1146 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1147 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
1148 memcpy(&new_key, &key, sizeof(new_key));
1150 if (start == key.offset && end < extent_end) {
1153 if (extent_mergeable(leaf, path->slots[0] - 1,
1154 ino, bytenr, orig_offset,
1155 &other_start, &other_end)) {
1156 new_key.offset = end;
1157 btrfs_set_item_key_safe(fs_info, path, &new_key);
1158 fi = btrfs_item_ptr(leaf, path->slots[0],
1159 struct btrfs_file_extent_item);
1160 btrfs_set_file_extent_generation(leaf, fi,
1162 btrfs_set_file_extent_num_bytes(leaf, fi,
1164 btrfs_set_file_extent_offset(leaf, fi,
1166 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1167 struct btrfs_file_extent_item);
1168 btrfs_set_file_extent_generation(leaf, fi,
1170 btrfs_set_file_extent_num_bytes(leaf, fi,
1172 btrfs_mark_buffer_dirty(leaf);
1177 if (start > key.offset && end == extent_end) {
1180 if (extent_mergeable(leaf, path->slots[0] + 1,
1181 ino, bytenr, orig_offset,
1182 &other_start, &other_end)) {
1183 fi = btrfs_item_ptr(leaf, path->slots[0],
1184 struct btrfs_file_extent_item);
1185 btrfs_set_file_extent_num_bytes(leaf, fi,
1186 start - key.offset);
1187 btrfs_set_file_extent_generation(leaf, fi,
1190 new_key.offset = start;
1191 btrfs_set_item_key_safe(fs_info, path, &new_key);
1193 fi = btrfs_item_ptr(leaf, path->slots[0],
1194 struct btrfs_file_extent_item);
1195 btrfs_set_file_extent_generation(leaf, fi,
1197 btrfs_set_file_extent_num_bytes(leaf, fi,
1199 btrfs_set_file_extent_offset(leaf, fi,
1200 start - orig_offset);
1201 btrfs_mark_buffer_dirty(leaf);
1206 while (start > key.offset || end < extent_end) {
1207 if (key.offset == start)
1210 new_key.offset = split;
1211 ret = btrfs_duplicate_item(trans, root, path, &new_key);
1212 if (ret == -EAGAIN) {
1213 btrfs_release_path(path);
1217 btrfs_abort_transaction(trans, ret);
1221 leaf = path->nodes[0];
1222 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1223 struct btrfs_file_extent_item);
1224 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1225 btrfs_set_file_extent_num_bytes(leaf, fi,
1226 split - key.offset);
1228 fi = btrfs_item_ptr(leaf, path->slots[0],
1229 struct btrfs_file_extent_item);
1231 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1232 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
1233 btrfs_set_file_extent_num_bytes(leaf, fi,
1234 extent_end - split);
1235 btrfs_mark_buffer_dirty(leaf);
1237 ret = btrfs_inc_extent_ref(trans, fs_info, bytenr, num_bytes,
1238 0, root->root_key.objectid,
1241 btrfs_abort_transaction(trans, ret);
1245 if (split == start) {
1248 if (start != key.offset) {
1250 btrfs_abort_transaction(trans, ret);
1261 if (extent_mergeable(leaf, path->slots[0] + 1,
1262 ino, bytenr, orig_offset,
1263 &other_start, &other_end)) {
1265 btrfs_release_path(path);
1268 extent_end = other_end;
1269 del_slot = path->slots[0] + 1;
1271 ret = btrfs_free_extent(trans, fs_info, bytenr, num_bytes,
1272 0, root->root_key.objectid,
1275 btrfs_abort_transaction(trans, ret);
1281 if (extent_mergeable(leaf, path->slots[0] - 1,
1282 ino, bytenr, orig_offset,
1283 &other_start, &other_end)) {
1285 btrfs_release_path(path);
1288 key.offset = other_start;
1289 del_slot = path->slots[0];
1291 ret = btrfs_free_extent(trans, fs_info, bytenr, num_bytes,
1292 0, root->root_key.objectid,
1295 btrfs_abort_transaction(trans, ret);
1300 fi = btrfs_item_ptr(leaf, path->slots[0],
1301 struct btrfs_file_extent_item);
1302 btrfs_set_file_extent_type(leaf, fi,
1303 BTRFS_FILE_EXTENT_REG);
1304 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1305 btrfs_mark_buffer_dirty(leaf);
1307 fi = btrfs_item_ptr(leaf, del_slot - 1,
1308 struct btrfs_file_extent_item);
1309 btrfs_set_file_extent_type(leaf, fi,
1310 BTRFS_FILE_EXTENT_REG);
1311 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1312 btrfs_set_file_extent_num_bytes(leaf, fi,
1313 extent_end - key.offset);
1314 btrfs_mark_buffer_dirty(leaf);
1316 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1318 btrfs_abort_transaction(trans, ret);
1323 btrfs_free_path(path);
1328 * on error we return an unlocked page and the error value
1329 * on success we return a locked page and 0
1331 static int prepare_uptodate_page(struct inode *inode,
1332 struct page *page, u64 pos,
1333 bool force_uptodate)
1337 if (((pos & (PAGE_SIZE - 1)) || force_uptodate) &&
1338 !PageUptodate(page)) {
1339 ret = btrfs_readpage(NULL, page);
1343 if (!PageUptodate(page)) {
1347 if (page->mapping != inode->i_mapping) {
1356 * this just gets pages into the page cache and locks them down.
1358 static noinline int prepare_pages(struct inode *inode, struct page **pages,
1359 size_t num_pages, loff_t pos,
1360 size_t write_bytes, bool force_uptodate)
1363 unsigned long index = pos >> PAGE_SHIFT;
1364 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1368 for (i = 0; i < num_pages; i++) {
1370 pages[i] = find_or_create_page(inode->i_mapping, index + i,
1371 mask | __GFP_WRITE);
1379 err = prepare_uptodate_page(inode, pages[i], pos,
1381 if (!err && i == num_pages - 1)
1382 err = prepare_uptodate_page(inode, pages[i],
1383 pos + write_bytes, false);
1386 if (err == -EAGAIN) {
1393 wait_on_page_writeback(pages[i]);
1398 while (faili >= 0) {
1399 unlock_page(pages[faili]);
1400 put_page(pages[faili]);
1408 * This function locks the extent and properly waits for data=ordered extents
1409 * to finish before allowing the pages to be modified if need.
1412 * 1 - the extent is locked
1413 * 0 - the extent is not locked, and everything is OK
1414 * -EAGAIN - need re-prepare the pages
1415 * the other < 0 number - Something wrong happens
1418 lock_and_cleanup_extent_if_need(struct btrfs_inode *inode, struct page **pages,
1419 size_t num_pages, loff_t pos,
1421 u64 *lockstart, u64 *lockend,
1422 struct extent_state **cached_state)
1424 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1430 start_pos = round_down(pos, fs_info->sectorsize);
1431 last_pos = start_pos
1432 + round_up(pos + write_bytes - start_pos,
1433 fs_info->sectorsize) - 1;
1435 if (start_pos < inode->vfs_inode.i_size) {
1436 struct btrfs_ordered_extent *ordered;
1437 lock_extent_bits(&inode->io_tree, start_pos, last_pos,
1439 ordered = btrfs_lookup_ordered_range(inode, start_pos,
1440 last_pos - start_pos + 1);
1442 ordered->file_offset + ordered->len > start_pos &&
1443 ordered->file_offset <= last_pos) {
1444 unlock_extent_cached(&inode->io_tree, start_pos,
1445 last_pos, cached_state, GFP_NOFS);
1446 for (i = 0; i < num_pages; i++) {
1447 unlock_page(pages[i]);
1450 btrfs_start_ordered_extent(&inode->vfs_inode,
1452 btrfs_put_ordered_extent(ordered);
1456 btrfs_put_ordered_extent(ordered);
1458 clear_extent_bit(&inode->io_tree, start_pos,
1459 last_pos, EXTENT_DIRTY | EXTENT_DELALLOC |
1460 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
1461 0, 0, cached_state, GFP_NOFS);
1462 *lockstart = start_pos;
1463 *lockend = last_pos;
1467 for (i = 0; i < num_pages; i++) {
1468 if (clear_page_dirty_for_io(pages[i]))
1469 account_page_redirty(pages[i]);
1470 set_page_extent_mapped(pages[i]);
1471 WARN_ON(!PageLocked(pages[i]));
1477 static noinline int check_can_nocow(struct btrfs_inode *inode, loff_t pos,
1478 size_t *write_bytes)
1480 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1481 struct btrfs_root *root = inode->root;
1482 struct btrfs_ordered_extent *ordered;
1483 u64 lockstart, lockend;
1487 ret = btrfs_start_write_no_snapshoting(root);
1491 lockstart = round_down(pos, fs_info->sectorsize);
1492 lockend = round_up(pos + *write_bytes,
1493 fs_info->sectorsize) - 1;
1496 lock_extent(&inode->io_tree, lockstart, lockend);
1497 ordered = btrfs_lookup_ordered_range(inode, lockstart,
1498 lockend - lockstart + 1);
1502 unlock_extent(&inode->io_tree, lockstart, lockend);
1503 btrfs_start_ordered_extent(&inode->vfs_inode, ordered, 1);
1504 btrfs_put_ordered_extent(ordered);
1507 num_bytes = lockend - lockstart + 1;
1508 ret = can_nocow_extent(&inode->vfs_inode, lockstart, &num_bytes,
1512 btrfs_end_write_no_snapshoting(root);
1514 *write_bytes = min_t(size_t, *write_bytes ,
1515 num_bytes - pos + lockstart);
1518 unlock_extent(&inode->io_tree, lockstart, lockend);
1523 static noinline ssize_t __btrfs_buffered_write(struct file *file,
1527 struct inode *inode = file_inode(file);
1528 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1529 struct btrfs_root *root = BTRFS_I(inode)->root;
1530 struct page **pages = NULL;
1531 struct extent_state *cached_state = NULL;
1532 u64 release_bytes = 0;
1535 size_t num_written = 0;
1538 bool only_release_metadata = false;
1539 bool force_page_uptodate = false;
1542 nrptrs = min(DIV_ROUND_UP(iov_iter_count(i), PAGE_SIZE),
1543 PAGE_SIZE / (sizeof(struct page *)));
1544 nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1545 nrptrs = max(nrptrs, 8);
1546 pages = kmalloc_array(nrptrs, sizeof(struct page *), GFP_KERNEL);
1550 while (iov_iter_count(i) > 0) {
1551 size_t offset = pos & (PAGE_SIZE - 1);
1552 size_t sector_offset;
1553 size_t write_bytes = min(iov_iter_count(i),
1554 nrptrs * (size_t)PAGE_SIZE -
1556 size_t num_pages = DIV_ROUND_UP(write_bytes + offset,
1558 size_t reserve_bytes;
1561 size_t dirty_sectors;
1564 WARN_ON(num_pages > nrptrs);
1567 * Fault pages before locking them in prepare_pages
1568 * to avoid recursive lock
1570 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1575 sector_offset = pos & (fs_info->sectorsize - 1);
1576 reserve_bytes = round_up(write_bytes + sector_offset,
1577 fs_info->sectorsize);
1579 ret = btrfs_check_data_free_space(inode, pos, write_bytes);
1581 if ((BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW |
1582 BTRFS_INODE_PREALLOC)) &&
1583 check_can_nocow(BTRFS_I(inode), pos,
1584 &write_bytes) > 0) {
1586 * For nodata cow case, no need to reserve
1589 only_release_metadata = true;
1591 * our prealloc extent may be smaller than
1592 * write_bytes, so scale down.
1594 num_pages = DIV_ROUND_UP(write_bytes + offset,
1596 reserve_bytes = round_up(write_bytes +
1598 fs_info->sectorsize);
1604 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
1607 if (!only_release_metadata)
1608 btrfs_free_reserved_data_space(inode, pos,
1611 btrfs_end_write_no_snapshoting(root);
1615 release_bytes = reserve_bytes;
1616 need_unlock = false;
1619 * This is going to setup the pages array with the number of
1620 * pages we want, so we don't really need to worry about the
1621 * contents of pages from loop to loop
1623 ret = prepare_pages(inode, pages, num_pages,
1625 force_page_uptodate);
1629 ret = lock_and_cleanup_extent_if_need(BTRFS_I(inode), pages,
1630 num_pages, pos, write_bytes, &lockstart,
1631 &lockend, &cached_state);
1636 } else if (ret > 0) {
1641 copied = btrfs_copy_from_user(pos, write_bytes, pages, i);
1643 num_sectors = BTRFS_BYTES_TO_BLKS(fs_info, reserve_bytes);
1644 dirty_sectors = round_up(copied + sector_offset,
1645 fs_info->sectorsize);
1646 dirty_sectors = BTRFS_BYTES_TO_BLKS(fs_info, dirty_sectors);
1649 * if we have trouble faulting in the pages, fall
1650 * back to one page at a time
1652 if (copied < write_bytes)
1656 force_page_uptodate = true;
1660 force_page_uptodate = false;
1661 dirty_pages = DIV_ROUND_UP(copied + offset,
1666 * If we had a short copy we need to release the excess delaloc
1667 * bytes we reserved. We need to increment outstanding_extents
1668 * because btrfs_delalloc_release_space and
1669 * btrfs_delalloc_release_metadata will decrement it, but
1670 * we still have an outstanding extent for the chunk we actually
1673 if (num_sectors > dirty_sectors) {
1674 /* release everything except the sectors we dirtied */
1675 release_bytes -= dirty_sectors <<
1676 fs_info->sb->s_blocksize_bits;
1678 spin_lock(&BTRFS_I(inode)->lock);
1679 BTRFS_I(inode)->outstanding_extents++;
1680 spin_unlock(&BTRFS_I(inode)->lock);
1682 if (only_release_metadata) {
1683 btrfs_delalloc_release_metadata(BTRFS_I(inode),
1688 __pos = round_down(pos,
1689 fs_info->sectorsize) +
1690 (dirty_pages << PAGE_SHIFT);
1691 btrfs_delalloc_release_space(inode, __pos,
1696 release_bytes = round_up(copied + sector_offset,
1697 fs_info->sectorsize);
1700 ret = btrfs_dirty_pages(inode, pages, dirty_pages,
1703 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1704 lockstart, lockend, &cached_state,
1707 btrfs_drop_pages(pages, num_pages);
1712 if (only_release_metadata)
1713 btrfs_end_write_no_snapshoting(root);
1715 if (only_release_metadata && copied > 0) {
1716 lockstart = round_down(pos,
1717 fs_info->sectorsize);
1718 lockend = round_up(pos + copied,
1719 fs_info->sectorsize) - 1;
1721 set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
1722 lockend, EXTENT_NORESERVE, NULL,
1724 only_release_metadata = false;
1727 btrfs_drop_pages(pages, num_pages);
1731 balance_dirty_pages_ratelimited(inode->i_mapping);
1732 if (dirty_pages < (fs_info->nodesize >> PAGE_SHIFT) + 1)
1733 btrfs_btree_balance_dirty(fs_info);
1736 num_written += copied;
1741 if (release_bytes) {
1742 if (only_release_metadata) {
1743 btrfs_end_write_no_snapshoting(root);
1744 btrfs_delalloc_release_metadata(BTRFS_I(inode),
1747 btrfs_delalloc_release_space(inode,
1748 round_down(pos, fs_info->sectorsize),
1753 return num_written ? num_written : ret;
1756 static ssize_t __btrfs_direct_write(struct kiocb *iocb, struct iov_iter *from)
1758 struct file *file = iocb->ki_filp;
1759 struct inode *inode = file_inode(file);
1760 loff_t pos = iocb->ki_pos;
1762 ssize_t written_buffered;
1766 written = generic_file_direct_write(iocb, from);
1768 if (written < 0 || !iov_iter_count(from))
1772 written_buffered = __btrfs_buffered_write(file, from, pos);
1773 if (written_buffered < 0) {
1774 err = written_buffered;
1778 * Ensure all data is persisted. We want the next direct IO read to be
1779 * able to read what was just written.
1781 endbyte = pos + written_buffered - 1;
1782 err = btrfs_fdatawrite_range(inode, pos, endbyte);
1785 err = filemap_fdatawait_range(inode->i_mapping, pos, endbyte);
1788 written += written_buffered;
1789 iocb->ki_pos = pos + written_buffered;
1790 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_SHIFT,
1791 endbyte >> PAGE_SHIFT);
1793 return written ? written : err;
1796 static void update_time_for_write(struct inode *inode)
1798 struct timespec now;
1800 if (IS_NOCMTIME(inode))
1803 now = current_time(inode);
1804 if (!timespec_equal(&inode->i_mtime, &now))
1805 inode->i_mtime = now;
1807 if (!timespec_equal(&inode->i_ctime, &now))
1808 inode->i_ctime = now;
1810 if (IS_I_VERSION(inode))
1811 inode_inc_iversion(inode);
1814 static ssize_t btrfs_file_write_iter(struct kiocb *iocb,
1815 struct iov_iter *from)
1817 struct file *file = iocb->ki_filp;
1818 struct inode *inode = file_inode(file);
1819 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1820 struct btrfs_root *root = BTRFS_I(inode)->root;
1823 ssize_t num_written = 0;
1824 bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host);
1832 err = generic_write_checks(iocb, from);
1834 inode_unlock(inode);
1838 current->backing_dev_info = inode_to_bdi(inode);
1839 err = file_remove_privs(file);
1841 inode_unlock(inode);
1846 * If BTRFS flips readonly due to some impossible error
1847 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1848 * although we have opened a file as writable, we have
1849 * to stop this write operation to ensure FS consistency.
1851 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
1852 inode_unlock(inode);
1858 * We reserve space for updating the inode when we reserve space for the
1859 * extent we are going to write, so we will enospc out there. We don't
1860 * need to start yet another transaction to update the inode as we will
1861 * update the inode when we finish writing whatever data we write.
1863 update_time_for_write(inode);
1866 count = iov_iter_count(from);
1867 start_pos = round_down(pos, fs_info->sectorsize);
1868 oldsize = i_size_read(inode);
1869 if (start_pos > oldsize) {
1870 /* Expand hole size to cover write data, preventing empty gap */
1871 end_pos = round_up(pos + count,
1872 fs_info->sectorsize);
1873 err = btrfs_cont_expand(inode, oldsize, end_pos);
1875 inode_unlock(inode);
1878 if (start_pos > round_up(oldsize, fs_info->sectorsize))
1883 atomic_inc(&BTRFS_I(inode)->sync_writers);
1885 if (iocb->ki_flags & IOCB_DIRECT) {
1886 num_written = __btrfs_direct_write(iocb, from);
1888 num_written = __btrfs_buffered_write(file, from, pos);
1889 if (num_written > 0)
1890 iocb->ki_pos = pos + num_written;
1892 pagecache_isize_extended(inode, oldsize,
1893 i_size_read(inode));
1896 inode_unlock(inode);
1899 * We also have to set last_sub_trans to the current log transid,
1900 * otherwise subsequent syncs to a file that's been synced in this
1901 * transaction will appear to have already occurred.
1903 spin_lock(&BTRFS_I(inode)->lock);
1904 BTRFS_I(inode)->last_sub_trans = root->log_transid;
1905 spin_unlock(&BTRFS_I(inode)->lock);
1906 if (num_written > 0)
1907 num_written = generic_write_sync(iocb, num_written);
1910 atomic_dec(&BTRFS_I(inode)->sync_writers);
1912 current->backing_dev_info = NULL;
1913 return num_written ? num_written : err;
1916 int btrfs_release_file(struct inode *inode, struct file *filp)
1918 if (filp->private_data)
1919 btrfs_ioctl_trans_end(filp);
1921 * ordered_data_close is set by settattr when we are about to truncate
1922 * a file from a non-zero size to a zero size. This tries to
1923 * flush down new bytes that may have been written if the
1924 * application were using truncate to replace a file in place.
1926 if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
1927 &BTRFS_I(inode)->runtime_flags))
1928 filemap_flush(inode->i_mapping);
1932 static int start_ordered_ops(struct inode *inode, loff_t start, loff_t end)
1936 atomic_inc(&BTRFS_I(inode)->sync_writers);
1937 ret = btrfs_fdatawrite_range(inode, start, end);
1938 atomic_dec(&BTRFS_I(inode)->sync_writers);
1944 * fsync call for both files and directories. This logs the inode into
1945 * the tree log instead of forcing full commits whenever possible.
1947 * It needs to call filemap_fdatawait so that all ordered extent updates are
1948 * in the metadata btree are up to date for copying to the log.
1950 * It drops the inode mutex before doing the tree log commit. This is an
1951 * important optimization for directories because holding the mutex prevents
1952 * new operations on the dir while we write to disk.
1954 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1956 struct dentry *dentry = file_dentry(file);
1957 struct inode *inode = d_inode(dentry);
1958 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1959 struct btrfs_root *root = BTRFS_I(inode)->root;
1960 struct btrfs_trans_handle *trans;
1961 struct btrfs_log_ctx ctx;
1967 * The range length can be represented by u64, we have to do the typecasts
1968 * to avoid signed overflow if it's [0, LLONG_MAX] eg. from fsync()
1970 len = (u64)end - (u64)start + 1;
1971 trace_btrfs_sync_file(file, datasync);
1974 * We write the dirty pages in the range and wait until they complete
1975 * out of the ->i_mutex. If so, we can flush the dirty pages by
1976 * multi-task, and make the performance up. See
1977 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1979 ret = start_ordered_ops(inode, start, end);
1984 atomic_inc(&root->log_batch);
1985 full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1986 &BTRFS_I(inode)->runtime_flags);
1988 * We might have have had more pages made dirty after calling
1989 * start_ordered_ops and before acquiring the inode's i_mutex.
1993 * For a full sync, we need to make sure any ordered operations
1994 * start and finish before we start logging the inode, so that
1995 * all extents are persisted and the respective file extent
1996 * items are in the fs/subvol btree.
1998 ret = btrfs_wait_ordered_range(inode, start, len);
2001 * Start any new ordered operations before starting to log the
2002 * inode. We will wait for them to finish in btrfs_sync_log().
2004 * Right before acquiring the inode's mutex, we might have new
2005 * writes dirtying pages, which won't immediately start the
2006 * respective ordered operations - that is done through the
2007 * fill_delalloc callbacks invoked from the writepage and
2008 * writepages address space operations. So make sure we start
2009 * all ordered operations before starting to log our inode. Not
2010 * doing this means that while logging the inode, writeback
2011 * could start and invoke writepage/writepages, which would call
2012 * the fill_delalloc callbacks (cow_file_range,
2013 * submit_compressed_extents). These callbacks add first an
2014 * extent map to the modified list of extents and then create
2015 * the respective ordered operation, which means in
2016 * tree-log.c:btrfs_log_inode() we might capture all existing
2017 * ordered operations (with btrfs_get_logged_extents()) before
2018 * the fill_delalloc callback adds its ordered operation, and by
2019 * the time we visit the modified list of extent maps (with
2020 * btrfs_log_changed_extents()), we see and process the extent
2021 * map they created. We then use the extent map to construct a
2022 * file extent item for logging without waiting for the
2023 * respective ordered operation to finish - this file extent
2024 * item points to a disk location that might not have yet been
2025 * written to, containing random data - so after a crash a log
2026 * replay will make our inode have file extent items that point
2027 * to disk locations containing invalid data, as we returned
2028 * success to userspace without waiting for the respective
2029 * ordered operation to finish, because it wasn't captured by
2030 * btrfs_get_logged_extents().
2032 ret = start_ordered_ops(inode, start, end);
2035 inode_unlock(inode);
2038 atomic_inc(&root->log_batch);
2041 * If the last transaction that changed this file was before the current
2042 * transaction and we have the full sync flag set in our inode, we can
2043 * bail out now without any syncing.
2045 * Note that we can't bail out if the full sync flag isn't set. This is
2046 * because when the full sync flag is set we start all ordered extents
2047 * and wait for them to fully complete - when they complete they update
2048 * the inode's last_trans field through:
2050 * btrfs_finish_ordered_io() ->
2051 * btrfs_update_inode_fallback() ->
2052 * btrfs_update_inode() ->
2053 * btrfs_set_inode_last_trans()
2055 * So we are sure that last_trans is up to date and can do this check to
2056 * bail out safely. For the fast path, when the full sync flag is not
2057 * set in our inode, we can not do it because we start only our ordered
2058 * extents and don't wait for them to complete (that is when
2059 * btrfs_finish_ordered_io runs), so here at this point their last_trans
2060 * value might be less than or equals to fs_info->last_trans_committed,
2061 * and setting a speculative last_trans for an inode when a buffered
2062 * write is made (such as fs_info->generation + 1 for example) would not
2063 * be reliable since after setting the value and before fsync is called
2064 * any number of transactions can start and commit (transaction kthread
2065 * commits the current transaction periodically), and a transaction
2066 * commit does not start nor waits for ordered extents to complete.
2069 if (btrfs_inode_in_log(BTRFS_I(inode), fs_info->generation) ||
2070 (full_sync && BTRFS_I(inode)->last_trans <=
2071 fs_info->last_trans_committed) ||
2072 (!btrfs_have_ordered_extents_in_range(inode, start, len) &&
2073 BTRFS_I(inode)->last_trans
2074 <= fs_info->last_trans_committed)) {
2076 * We've had everything committed since the last time we were
2077 * modified so clear this flag in case it was set for whatever
2078 * reason, it's no longer relevant.
2080 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2081 &BTRFS_I(inode)->runtime_flags);
2083 * An ordered extent might have started before and completed
2084 * already with io errors, in which case the inode was not
2085 * updated and we end up here. So check the inode's mapping
2086 * flags for any errors that might have happened while doing
2087 * writeback of file data.
2089 ret = filemap_check_errors(inode->i_mapping);
2090 inode_unlock(inode);
2095 * ok we haven't committed the transaction yet, lets do a commit
2097 if (file->private_data)
2098 btrfs_ioctl_trans_end(file);
2101 * We use start here because we will need to wait on the IO to complete
2102 * in btrfs_sync_log, which could require joining a transaction (for
2103 * example checking cross references in the nocow path). If we use join
2104 * here we could get into a situation where we're waiting on IO to
2105 * happen that is blocked on a transaction trying to commit. With start
2106 * we inc the extwriter counter, so we wait for all extwriters to exit
2107 * before we start blocking join'ers. This comment is to keep somebody
2108 * from thinking they are super smart and changing this to
2109 * btrfs_join_transaction *cough*Josef*cough*.
2111 trans = btrfs_start_transaction(root, 0);
2112 if (IS_ERR(trans)) {
2113 ret = PTR_ERR(trans);
2114 inode_unlock(inode);
2119 btrfs_init_log_ctx(&ctx, inode);
2121 ret = btrfs_log_dentry_safe(trans, root, dentry, start, end, &ctx);
2123 /* Fallthrough and commit/free transaction. */
2127 /* we've logged all the items and now have a consistent
2128 * version of the file in the log. It is possible that
2129 * someone will come in and modify the file, but that's
2130 * fine because the log is consistent on disk, and we
2131 * have references to all of the file's extents
2133 * It is possible that someone will come in and log the
2134 * file again, but that will end up using the synchronization
2135 * inside btrfs_sync_log to keep things safe.
2137 inode_unlock(inode);
2140 * If any of the ordered extents had an error, just return it to user
2141 * space, so that the application knows some writes didn't succeed and
2142 * can take proper action (retry for e.g.). Blindly committing the
2143 * transaction in this case, would fool userspace that everything was
2144 * successful. And we also want to make sure our log doesn't contain
2145 * file extent items pointing to extents that weren't fully written to -
2146 * just like in the non fast fsync path, where we check for the ordered
2147 * operation's error flag before writing to the log tree and return -EIO
2148 * if any of them had this flag set (btrfs_wait_ordered_range) -
2149 * therefore we need to check for errors in the ordered operations,
2150 * which are indicated by ctx.io_err.
2153 btrfs_end_transaction(trans);
2158 if (ret != BTRFS_NO_LOG_SYNC) {
2160 ret = btrfs_sync_log(trans, root, &ctx);
2162 ret = btrfs_end_transaction(trans);
2167 ret = btrfs_wait_ordered_range(inode, start, len);
2169 btrfs_end_transaction(trans);
2173 ret = btrfs_commit_transaction(trans);
2175 ret = btrfs_end_transaction(trans);
2178 return ret > 0 ? -EIO : ret;
2181 static const struct vm_operations_struct btrfs_file_vm_ops = {
2182 .fault = filemap_fault,
2183 .map_pages = filemap_map_pages,
2184 .page_mkwrite = btrfs_page_mkwrite,
2187 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
2189 struct address_space *mapping = filp->f_mapping;
2191 if (!mapping->a_ops->readpage)
2194 file_accessed(filp);
2195 vma->vm_ops = &btrfs_file_vm_ops;
2200 static int hole_mergeable(struct btrfs_inode *inode, struct extent_buffer *leaf,
2201 int slot, u64 start, u64 end)
2203 struct btrfs_file_extent_item *fi;
2204 struct btrfs_key key;
2206 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
2209 btrfs_item_key_to_cpu(leaf, &key, slot);
2210 if (key.objectid != btrfs_ino(inode) ||
2211 key.type != BTRFS_EXTENT_DATA_KEY)
2214 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2216 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2219 if (btrfs_file_extent_disk_bytenr(leaf, fi))
2222 if (key.offset == end)
2224 if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
2229 static int fill_holes(struct btrfs_trans_handle *trans,
2230 struct btrfs_inode *inode,
2231 struct btrfs_path *path, u64 offset, u64 end)
2233 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
2234 struct btrfs_root *root = inode->root;
2235 struct extent_buffer *leaf;
2236 struct btrfs_file_extent_item *fi;
2237 struct extent_map *hole_em;
2238 struct extent_map_tree *em_tree = &inode->extent_tree;
2239 struct btrfs_key key;
2242 if (btrfs_fs_incompat(fs_info, NO_HOLES))
2245 key.objectid = btrfs_ino(inode);
2246 key.type = BTRFS_EXTENT_DATA_KEY;
2247 key.offset = offset;
2249 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2252 * We should have dropped this offset, so if we find it then
2253 * something has gone horribly wrong.
2260 leaf = path->nodes[0];
2261 if (hole_mergeable(inode, leaf, path->slots[0] - 1, offset, end)) {
2265 fi = btrfs_item_ptr(leaf, path->slots[0],
2266 struct btrfs_file_extent_item);
2267 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2269 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2270 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2271 btrfs_set_file_extent_offset(leaf, fi, 0);
2272 btrfs_mark_buffer_dirty(leaf);
2276 if (hole_mergeable(inode, leaf, path->slots[0], offset, end)) {
2279 key.offset = offset;
2280 btrfs_set_item_key_safe(fs_info, path, &key);
2281 fi = btrfs_item_ptr(leaf, path->slots[0],
2282 struct btrfs_file_extent_item);
2283 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2285 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2286 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2287 btrfs_set_file_extent_offset(leaf, fi, 0);
2288 btrfs_mark_buffer_dirty(leaf);
2291 btrfs_release_path(path);
2293 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode),
2294 offset, 0, 0, end - offset, 0, end - offset, 0, 0, 0);
2299 btrfs_release_path(path);
2301 hole_em = alloc_extent_map();
2303 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2304 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
2306 hole_em->start = offset;
2307 hole_em->len = end - offset;
2308 hole_em->ram_bytes = hole_em->len;
2309 hole_em->orig_start = offset;
2311 hole_em->block_start = EXTENT_MAP_HOLE;
2312 hole_em->block_len = 0;
2313 hole_em->orig_block_len = 0;
2314 hole_em->bdev = fs_info->fs_devices->latest_bdev;
2315 hole_em->compress_type = BTRFS_COMPRESS_NONE;
2316 hole_em->generation = trans->transid;
2319 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2320 write_lock(&em_tree->lock);
2321 ret = add_extent_mapping(em_tree, hole_em, 1);
2322 write_unlock(&em_tree->lock);
2323 } while (ret == -EEXIST);
2324 free_extent_map(hole_em);
2326 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2327 &inode->runtime_flags);
2334 * Find a hole extent on given inode and change start/len to the end of hole
2335 * extent.(hole/vacuum extent whose em->start <= start &&
2336 * em->start + em->len > start)
2337 * When a hole extent is found, return 1 and modify start/len.
2339 static int find_first_non_hole(struct inode *inode, u64 *start, u64 *len)
2341 struct extent_map *em;
2344 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, *start, *len, 0);
2345 if (IS_ERR_OR_NULL(em)) {
2353 /* Hole or vacuum extent(only exists in no-hole mode) */
2354 if (em->block_start == EXTENT_MAP_HOLE) {
2356 *len = em->start + em->len > *start + *len ?
2357 0 : *start + *len - em->start - em->len;
2358 *start = em->start + em->len;
2360 free_extent_map(em);
2364 static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
2366 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2367 struct btrfs_root *root = BTRFS_I(inode)->root;
2368 struct extent_state *cached_state = NULL;
2369 struct btrfs_path *path;
2370 struct btrfs_block_rsv *rsv;
2371 struct btrfs_trans_handle *trans;
2376 u64 orig_start = offset;
2378 u64 min_size = btrfs_calc_trans_metadata_size(fs_info, 1);
2382 unsigned int rsv_count;
2384 bool no_holes = btrfs_fs_incompat(fs_info, NO_HOLES);
2386 bool truncated_block = false;
2387 bool updated_inode = false;
2389 ret = btrfs_wait_ordered_range(inode, offset, len);
2394 ino_size = round_up(inode->i_size, fs_info->sectorsize);
2395 ret = find_first_non_hole(inode, &offset, &len);
2397 goto out_only_mutex;
2399 /* Already in a large hole */
2401 goto out_only_mutex;
2404 lockstart = round_up(offset, btrfs_inode_sectorsize(inode));
2405 lockend = round_down(offset + len,
2406 btrfs_inode_sectorsize(inode)) - 1;
2407 same_block = (BTRFS_BYTES_TO_BLKS(fs_info, offset))
2408 == (BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1));
2410 * We needn't truncate any block which is beyond the end of the file
2411 * because we are sure there is no data there.
2414 * Only do this if we are in the same block and we aren't doing the
2417 if (same_block && len < fs_info->sectorsize) {
2418 if (offset < ino_size) {
2419 truncated_block = true;
2420 ret = btrfs_truncate_block(inode, offset, len, 0);
2424 goto out_only_mutex;
2427 /* zero back part of the first block */
2428 if (offset < ino_size) {
2429 truncated_block = true;
2430 ret = btrfs_truncate_block(inode, offset, 0, 0);
2432 inode_unlock(inode);
2437 /* Check the aligned pages after the first unaligned page,
2438 * if offset != orig_start, which means the first unaligned page
2439 * including several following pages are already in holes,
2440 * the extra check can be skipped */
2441 if (offset == orig_start) {
2442 /* after truncate page, check hole again */
2443 len = offset + len - lockstart;
2445 ret = find_first_non_hole(inode, &offset, &len);
2447 goto out_only_mutex;
2450 goto out_only_mutex;
2455 /* Check the tail unaligned part is in a hole */
2456 tail_start = lockend + 1;
2457 tail_len = offset + len - tail_start;
2459 ret = find_first_non_hole(inode, &tail_start, &tail_len);
2460 if (unlikely(ret < 0))
2461 goto out_only_mutex;
2463 /* zero the front end of the last page */
2464 if (tail_start + tail_len < ino_size) {
2465 truncated_block = true;
2466 ret = btrfs_truncate_block(inode,
2467 tail_start + tail_len,
2470 goto out_only_mutex;
2475 if (lockend < lockstart) {
2477 goto out_only_mutex;
2481 struct btrfs_ordered_extent *ordered;
2483 truncate_pagecache_range(inode, lockstart, lockend);
2485 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2487 ordered = btrfs_lookup_first_ordered_extent(inode, lockend);
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->len <= lockstart ||
2496 ordered->file_offset > lockend)) &&
2497 !btrfs_page_exists_in_range(inode, lockstart, lockend)) {
2499 btrfs_put_ordered_extent(ordered);
2503 btrfs_put_ordered_extent(ordered);
2504 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
2505 lockend, &cached_state, GFP_NOFS);
2506 ret = btrfs_wait_ordered_range(inode, lockstart,
2507 lockend - lockstart + 1);
2509 inode_unlock(inode);
2514 path = btrfs_alloc_path();
2520 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
2525 rsv->size = btrfs_calc_trans_metadata_size(fs_info, 1);
2529 * 1 - update the inode
2530 * 1 - removing the extents in the range
2531 * 1 - adding the hole extent if no_holes isn't set
2533 rsv_count = no_holes ? 2 : 3;
2534 trans = btrfs_start_transaction(root, rsv_count);
2535 if (IS_ERR(trans)) {
2536 err = PTR_ERR(trans);
2540 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
2543 trans->block_rsv = rsv;
2545 cur_offset = lockstart;
2546 len = lockend - cur_offset;
2547 while (cur_offset < lockend) {
2548 ret = __btrfs_drop_extents(trans, root, inode, path,
2549 cur_offset, lockend + 1,
2550 &drop_end, 1, 0, 0, NULL);
2554 trans->block_rsv = &fs_info->trans_block_rsv;
2556 if (cur_offset < drop_end && cur_offset < ino_size) {
2557 ret = fill_holes(trans, BTRFS_I(inode), path,
2558 cur_offset, drop_end);
2561 * If we failed then we didn't insert our hole
2562 * entries for the area we dropped, so now the
2563 * fs is corrupted, so we must abort the
2566 btrfs_abort_transaction(trans, ret);
2572 cur_offset = drop_end;
2574 ret = btrfs_update_inode(trans, root, inode);
2580 btrfs_end_transaction(trans);
2581 btrfs_btree_balance_dirty(fs_info);
2583 trans = btrfs_start_transaction(root, rsv_count);
2584 if (IS_ERR(trans)) {
2585 ret = PTR_ERR(trans);
2590 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
2592 BUG_ON(ret); /* shouldn't happen */
2593 trans->block_rsv = rsv;
2595 ret = find_first_non_hole(inode, &cur_offset, &len);
2596 if (unlikely(ret < 0))
2609 trans->block_rsv = &fs_info->trans_block_rsv;
2611 * If we are using the NO_HOLES feature we might have had already an
2612 * hole that overlaps a part of the region [lockstart, lockend] and
2613 * ends at (or beyond) lockend. Since we have no file extent items to
2614 * represent holes, drop_end can be less than lockend and so we must
2615 * make sure we have an extent map representing the existing hole (the
2616 * call to __btrfs_drop_extents() might have dropped the existing extent
2617 * map representing the existing hole), otherwise the fast fsync path
2618 * will not record the existence of the hole region
2619 * [existing_hole_start, lockend].
2621 if (drop_end <= lockend)
2622 drop_end = lockend + 1;
2624 * Don't insert file hole extent item if it's for a range beyond eof
2625 * (because it's useless) or if it represents a 0 bytes range (when
2626 * cur_offset == drop_end).
2628 if (cur_offset < ino_size && cur_offset < drop_end) {
2629 ret = fill_holes(trans, BTRFS_I(inode), path,
2630 cur_offset, drop_end);
2632 /* Same comment as above. */
2633 btrfs_abort_transaction(trans, ret);
2643 inode_inc_iversion(inode);
2644 inode->i_mtime = inode->i_ctime = current_time(inode);
2646 trans->block_rsv = &fs_info->trans_block_rsv;
2647 ret = btrfs_update_inode(trans, root, inode);
2648 updated_inode = true;
2649 btrfs_end_transaction(trans);
2650 btrfs_btree_balance_dirty(fs_info);
2652 btrfs_free_path(path);
2653 btrfs_free_block_rsv(fs_info, rsv);
2655 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2656 &cached_state, GFP_NOFS);
2658 if (!updated_inode && truncated_block && !ret && !err) {
2660 * If we only end up zeroing part of a page, we still need to
2661 * update the inode item, so that all the time fields are
2662 * updated as well as the necessary btrfs inode in memory fields
2663 * for detecting, at fsync time, if the inode isn't yet in the
2664 * log tree or it's there but not up to date.
2666 trans = btrfs_start_transaction(root, 1);
2667 if (IS_ERR(trans)) {
2668 err = PTR_ERR(trans);
2670 err = btrfs_update_inode(trans, root, inode);
2671 ret = btrfs_end_transaction(trans);
2674 inode_unlock(inode);
2680 /* Helper structure to record which range is already reserved */
2681 struct falloc_range {
2682 struct list_head list;
2688 * Helper function to add falloc range
2690 * Caller should have locked the larger range of extent containing
2693 static int add_falloc_range(struct list_head *head, u64 start, u64 len)
2695 struct falloc_range *prev = NULL;
2696 struct falloc_range *range = NULL;
2698 if (list_empty(head))
2702 * As fallocate iterate by bytenr order, we only need to check
2705 prev = list_entry(head->prev, struct falloc_range, list);
2706 if (prev->start + prev->len == start) {
2711 range = kmalloc(sizeof(*range), GFP_KERNEL);
2714 range->start = start;
2716 list_add_tail(&range->list, head);
2720 static long btrfs_fallocate(struct file *file, int mode,
2721 loff_t offset, loff_t len)
2723 struct inode *inode = file_inode(file);
2724 struct extent_state *cached_state = NULL;
2725 struct falloc_range *range;
2726 struct falloc_range *tmp;
2727 struct list_head reserve_list;
2735 struct extent_map *em;
2736 int blocksize = btrfs_inode_sectorsize(inode);
2739 alloc_start = round_down(offset, blocksize);
2740 alloc_end = round_up(offset + len, blocksize);
2741 cur_offset = alloc_start;
2743 /* Make sure we aren't being give some crap mode */
2744 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2747 if (mode & FALLOC_FL_PUNCH_HOLE)
2748 return btrfs_punch_hole(inode, offset, len);
2751 * Only trigger disk allocation, don't trigger qgroup reserve
2753 * For qgroup space, it will be checked later.
2755 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode),
2756 alloc_end - alloc_start);
2762 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) {
2763 ret = inode_newsize_ok(inode, offset + len);
2769 * TODO: Move these two operations after we have checked
2770 * accurate reserved space, or fallocate can still fail but
2771 * with page truncated or size expanded.
2773 * But that's a minor problem and won't do much harm BTW.
2775 if (alloc_start > inode->i_size) {
2776 ret = btrfs_cont_expand(inode, i_size_read(inode),
2780 } else if (offset + len > inode->i_size) {
2782 * If we are fallocating from the end of the file onward we
2783 * need to zero out the end of the block if i_size lands in the
2784 * middle of a block.
2786 ret = btrfs_truncate_block(inode, inode->i_size, 0, 0);
2792 * wait for ordered IO before we have any locks. We'll loop again
2793 * below with the locks held.
2795 ret = btrfs_wait_ordered_range(inode, alloc_start,
2796 alloc_end - alloc_start);
2800 locked_end = alloc_end - 1;
2802 struct btrfs_ordered_extent *ordered;
2804 /* the extent lock is ordered inside the running
2807 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
2808 locked_end, &cached_state);
2809 ordered = btrfs_lookup_first_ordered_extent(inode,
2812 ordered->file_offset + ordered->len > alloc_start &&
2813 ordered->file_offset < alloc_end) {
2814 btrfs_put_ordered_extent(ordered);
2815 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
2816 alloc_start, locked_end,
2817 &cached_state, GFP_KERNEL);
2819 * we can't wait on the range with the transaction
2820 * running or with the extent lock held
2822 ret = btrfs_wait_ordered_range(inode, alloc_start,
2823 alloc_end - alloc_start);
2828 btrfs_put_ordered_extent(ordered);
2833 /* First, check if we exceed the qgroup limit */
2834 INIT_LIST_HEAD(&reserve_list);
2836 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
2837 alloc_end - cur_offset, 0);
2838 if (IS_ERR_OR_NULL(em)) {
2845 last_byte = min(extent_map_end(em), alloc_end);
2846 actual_end = min_t(u64, extent_map_end(em), offset + len);
2847 last_byte = ALIGN(last_byte, blocksize);
2848 if (em->block_start == EXTENT_MAP_HOLE ||
2849 (cur_offset >= inode->i_size &&
2850 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
2851 ret = add_falloc_range(&reserve_list, cur_offset,
2852 last_byte - cur_offset);
2854 free_extent_map(em);
2857 ret = btrfs_qgroup_reserve_data(inode, cur_offset,
2858 last_byte - cur_offset);
2863 * Do not need to reserve unwritten extent for this
2864 * range, free reserved data space first, otherwise
2865 * it'll result in false ENOSPC error.
2867 btrfs_free_reserved_data_space(inode, cur_offset,
2868 last_byte - cur_offset);
2870 free_extent_map(em);
2871 cur_offset = last_byte;
2872 if (cur_offset >= alloc_end)
2877 * If ret is still 0, means we're OK to fallocate.
2878 * Or just cleanup the list and exit.
2880 list_for_each_entry_safe(range, tmp, &reserve_list, list) {
2882 ret = btrfs_prealloc_file_range(inode, mode,
2884 range->len, i_blocksize(inode),
2885 offset + len, &alloc_hint);
2887 btrfs_free_reserved_data_space(inode, range->start,
2889 list_del(&range->list);
2895 if (actual_end > inode->i_size &&
2896 !(mode & FALLOC_FL_KEEP_SIZE)) {
2897 struct btrfs_trans_handle *trans;
2898 struct btrfs_root *root = BTRFS_I(inode)->root;
2901 * We didn't need to allocate any more space, but we
2902 * still extended the size of the file so we need to
2903 * update i_size and the inode item.
2905 trans = btrfs_start_transaction(root, 1);
2906 if (IS_ERR(trans)) {
2907 ret = PTR_ERR(trans);
2909 inode->i_ctime = current_time(inode);
2910 i_size_write(inode, actual_end);
2911 btrfs_ordered_update_i_size(inode, actual_end, NULL);
2912 ret = btrfs_update_inode(trans, root, inode);
2914 btrfs_end_transaction(trans);
2916 ret = btrfs_end_transaction(trans);
2920 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
2921 &cached_state, GFP_KERNEL);
2923 inode_unlock(inode);
2924 /* Let go of our reservation. */
2926 btrfs_free_reserved_data_space(inode, alloc_start,
2927 alloc_end - cur_offset);
2931 static int find_desired_extent(struct inode *inode, loff_t *offset, int whence)
2933 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2934 struct extent_map *em = NULL;
2935 struct extent_state *cached_state = NULL;
2942 if (inode->i_size == 0)
2946 * *offset can be negative, in this case we start finding DATA/HOLE from
2947 * the very start of the file.
2949 start = max_t(loff_t, 0, *offset);
2951 lockstart = round_down(start, fs_info->sectorsize);
2952 lockend = round_up(i_size_read(inode),
2953 fs_info->sectorsize);
2954 if (lockend <= lockstart)
2955 lockend = lockstart + fs_info->sectorsize;
2957 len = lockend - lockstart + 1;
2959 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2962 while (start < inode->i_size) {
2963 em = btrfs_get_extent_fiemap(BTRFS_I(inode), NULL, 0,
2971 if (whence == SEEK_HOLE &&
2972 (em->block_start == EXTENT_MAP_HOLE ||
2973 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2975 else if (whence == SEEK_DATA &&
2976 (em->block_start != EXTENT_MAP_HOLE &&
2977 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2980 start = em->start + em->len;
2981 free_extent_map(em);
2985 free_extent_map(em);
2987 if (whence == SEEK_DATA && start >= inode->i_size)
2990 *offset = min_t(loff_t, start, inode->i_size);
2992 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2993 &cached_state, GFP_NOFS);
2997 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
2999 struct inode *inode = file->f_mapping->host;
3006 offset = generic_file_llseek(file, offset, whence);
3010 if (offset >= i_size_read(inode)) {
3011 inode_unlock(inode);
3015 ret = find_desired_extent(inode, &offset, whence);
3017 inode_unlock(inode);
3022 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
3024 inode_unlock(inode);
3028 const struct file_operations btrfs_file_operations = {
3029 .llseek = btrfs_file_llseek,
3030 .read_iter = generic_file_read_iter,
3031 .splice_read = generic_file_splice_read,
3032 .write_iter = btrfs_file_write_iter,
3033 .mmap = btrfs_file_mmap,
3034 .open = generic_file_open,
3035 .release = btrfs_release_file,
3036 .fsync = btrfs_sync_file,
3037 .fallocate = btrfs_fallocate,
3038 .unlocked_ioctl = btrfs_ioctl,
3039 #ifdef CONFIG_COMPAT
3040 .compat_ioctl = btrfs_compat_ioctl,
3042 .clone_file_range = btrfs_clone_file_range,
3043 .dedupe_file_range = btrfs_dedupe_file_range,
3046 void btrfs_auto_defrag_exit(void)
3048 kmem_cache_destroy(btrfs_inode_defrag_cachep);
3051 int btrfs_auto_defrag_init(void)
3053 btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
3054 sizeof(struct inode_defrag), 0,
3057 if (!btrfs_inode_defrag_cachep)
3063 int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end)
3068 * So with compression we will find and lock a dirty page and clear the
3069 * first one as dirty, setup an async extent, and immediately return
3070 * with the entire range locked but with nobody actually marked with
3071 * writeback. So we can't just filemap_write_and_wait_range() and
3072 * expect it to work since it will just kick off a thread to do the
3073 * actual work. So we need to call filemap_fdatawrite_range _again_
3074 * since it will wait on the page lock, which won't be unlocked until
3075 * after the pages have been marked as writeback and so we're good to go
3076 * from there. We have to do this otherwise we'll miss the ordered
3077 * extents and that results in badness. Please Josef, do not think you
3078 * know better and pull this out at some point in the future, it is
3079 * right and you are wrong.
3081 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
3082 if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
3083 &BTRFS_I(inode)->runtime_flags))
3084 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);