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/aio.h>
28 #include <linux/falloc.h>
29 #include <linux/swap.h>
30 #include <linux/writeback.h>
31 #include <linux/statfs.h>
32 #include <linux/compat.h>
33 #include <linux/slab.h>
34 #include <linux/btrfs.h>
37 #include "transaction.h"
38 #include "btrfs_inode.h"
39 #include "print-tree.h"
44 static struct kmem_cache *btrfs_inode_defrag_cachep;
46 * when auto defrag is enabled we
47 * queue up these defrag structs to remember which
48 * inodes need defragging passes
51 struct rb_node rb_node;
55 * transid where the defrag was added, we search for
56 * extents newer than this
63 /* last offset we were able to defrag */
66 /* if we've wrapped around back to zero once already */
70 static int __compare_inode_defrag(struct inode_defrag *defrag1,
71 struct inode_defrag *defrag2)
73 if (defrag1->root > defrag2->root)
75 else if (defrag1->root < defrag2->root)
77 else if (defrag1->ino > defrag2->ino)
79 else if (defrag1->ino < defrag2->ino)
85 /* pop a record for an inode into the defrag tree. The lock
86 * must be held already
88 * If you're inserting a record for an older transid than an
89 * existing record, the transid already in the tree is lowered
91 * If an existing record is found the defrag item you
94 static int __btrfs_add_inode_defrag(struct inode *inode,
95 struct inode_defrag *defrag)
97 struct btrfs_root *root = BTRFS_I(inode)->root;
98 struct inode_defrag *entry;
100 struct rb_node *parent = NULL;
103 p = &root->fs_info->defrag_inodes.rb_node;
106 entry = rb_entry(parent, struct inode_defrag, rb_node);
108 ret = __compare_inode_defrag(defrag, entry);
110 p = &parent->rb_left;
112 p = &parent->rb_right;
114 /* if we're reinserting an entry for
115 * an old defrag run, make sure to
116 * lower the transid of our existing record
118 if (defrag->transid < entry->transid)
119 entry->transid = defrag->transid;
120 if (defrag->last_offset > entry->last_offset)
121 entry->last_offset = defrag->last_offset;
125 set_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
126 rb_link_node(&defrag->rb_node, parent, p);
127 rb_insert_color(&defrag->rb_node, &root->fs_info->defrag_inodes);
131 static inline int __need_auto_defrag(struct btrfs_root *root)
133 if (!btrfs_test_opt(root, AUTO_DEFRAG))
136 if (btrfs_fs_closing(root->fs_info))
143 * insert a defrag record for this inode if auto defrag is
146 int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans,
149 struct btrfs_root *root = BTRFS_I(inode)->root;
150 struct inode_defrag *defrag;
154 if (!__need_auto_defrag(root))
157 if (test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags))
161 transid = trans->transid;
163 transid = BTRFS_I(inode)->root->last_trans;
165 defrag = kmem_cache_zalloc(btrfs_inode_defrag_cachep, GFP_NOFS);
169 defrag->ino = btrfs_ino(inode);
170 defrag->transid = transid;
171 defrag->root = root->root_key.objectid;
173 spin_lock(&root->fs_info->defrag_inodes_lock);
174 if (!test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags)) {
176 * If we set IN_DEFRAG flag and evict the inode from memory,
177 * and then re-read this inode, this new inode doesn't have
178 * IN_DEFRAG flag. At the case, we may find the existed defrag.
180 ret = __btrfs_add_inode_defrag(inode, defrag);
182 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
184 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
186 spin_unlock(&root->fs_info->defrag_inodes_lock);
191 * Requeue the defrag object. If there is a defrag object that points to
192 * the same inode in the tree, we will merge them together (by
193 * __btrfs_add_inode_defrag()) and free the one that we want to requeue.
195 static void btrfs_requeue_inode_defrag(struct inode *inode,
196 struct inode_defrag *defrag)
198 struct btrfs_root *root = BTRFS_I(inode)->root;
201 if (!__need_auto_defrag(root))
205 * Here we don't check the IN_DEFRAG flag, because we need merge
208 spin_lock(&root->fs_info->defrag_inodes_lock);
209 ret = __btrfs_add_inode_defrag(inode, defrag);
210 spin_unlock(&root->fs_info->defrag_inodes_lock);
215 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
219 * pick the defragable inode that we want, if it doesn't exist, we will get
222 static struct inode_defrag *
223 btrfs_pick_defrag_inode(struct btrfs_fs_info *fs_info, u64 root, u64 ino)
225 struct inode_defrag *entry = NULL;
226 struct inode_defrag tmp;
228 struct rb_node *parent = NULL;
234 spin_lock(&fs_info->defrag_inodes_lock);
235 p = fs_info->defrag_inodes.rb_node;
238 entry = rb_entry(parent, struct inode_defrag, rb_node);
240 ret = __compare_inode_defrag(&tmp, entry);
244 p = parent->rb_right;
249 if (parent && __compare_inode_defrag(&tmp, entry) > 0) {
250 parent = rb_next(parent);
252 entry = rb_entry(parent, struct inode_defrag, rb_node);
258 rb_erase(parent, &fs_info->defrag_inodes);
259 spin_unlock(&fs_info->defrag_inodes_lock);
263 void btrfs_cleanup_defrag_inodes(struct btrfs_fs_info *fs_info)
265 struct inode_defrag *defrag;
266 struct rb_node *node;
268 spin_lock(&fs_info->defrag_inodes_lock);
269 node = rb_first(&fs_info->defrag_inodes);
271 rb_erase(node, &fs_info->defrag_inodes);
272 defrag = rb_entry(node, struct inode_defrag, rb_node);
273 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
275 if (need_resched()) {
276 spin_unlock(&fs_info->defrag_inodes_lock);
278 spin_lock(&fs_info->defrag_inodes_lock);
281 node = rb_first(&fs_info->defrag_inodes);
283 spin_unlock(&fs_info->defrag_inodes_lock);
286 #define BTRFS_DEFRAG_BATCH 1024
288 static int __btrfs_run_defrag_inode(struct btrfs_fs_info *fs_info,
289 struct inode_defrag *defrag)
291 struct btrfs_root *inode_root;
293 struct btrfs_key key;
294 struct btrfs_ioctl_defrag_range_args range;
300 key.objectid = defrag->root;
301 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
302 key.offset = (u64)-1;
304 index = srcu_read_lock(&fs_info->subvol_srcu);
306 inode_root = btrfs_read_fs_root_no_name(fs_info, &key);
307 if (IS_ERR(inode_root)) {
308 ret = PTR_ERR(inode_root);
312 key.objectid = defrag->ino;
313 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
315 inode = btrfs_iget(fs_info->sb, &key, inode_root, NULL);
317 ret = PTR_ERR(inode);
320 srcu_read_unlock(&fs_info->subvol_srcu, index);
322 /* do a chunk of defrag */
323 clear_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
324 memset(&range, 0, sizeof(range));
326 range.start = defrag->last_offset;
328 sb_start_write(fs_info->sb);
329 num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid,
331 sb_end_write(fs_info->sb);
333 * if we filled the whole defrag batch, there
334 * must be more work to do. Queue this defrag
337 if (num_defrag == BTRFS_DEFRAG_BATCH) {
338 defrag->last_offset = range.start;
339 btrfs_requeue_inode_defrag(inode, defrag);
340 } else if (defrag->last_offset && !defrag->cycled) {
342 * we didn't fill our defrag batch, but
343 * we didn't start at zero. Make sure we loop
344 * around to the start of the file.
346 defrag->last_offset = 0;
348 btrfs_requeue_inode_defrag(inode, defrag);
350 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
356 srcu_read_unlock(&fs_info->subvol_srcu, index);
357 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
362 * run through the list of inodes in the FS that need
365 int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info)
367 struct inode_defrag *defrag;
369 u64 root_objectid = 0;
371 atomic_inc(&fs_info->defrag_running);
373 /* Pause the auto defragger. */
374 if (test_bit(BTRFS_FS_STATE_REMOUNTING,
378 if (!__need_auto_defrag(fs_info->tree_root))
381 /* find an inode to defrag */
382 defrag = btrfs_pick_defrag_inode(fs_info, root_objectid,
385 if (root_objectid || first_ino) {
394 first_ino = defrag->ino + 1;
395 root_objectid = defrag->root;
397 __btrfs_run_defrag_inode(fs_info, defrag);
399 atomic_dec(&fs_info->defrag_running);
402 * during unmount, we use the transaction_wait queue to
403 * wait for the defragger to stop
405 wake_up(&fs_info->transaction_wait);
409 /* simple helper to fault in pages and copy. This should go away
410 * and be replaced with calls into generic code.
412 static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
414 struct page **prepared_pages,
418 size_t total_copied = 0;
420 int offset = pos & (PAGE_CACHE_SIZE - 1);
422 while (write_bytes > 0) {
423 size_t count = min_t(size_t,
424 PAGE_CACHE_SIZE - offset, write_bytes);
425 struct page *page = prepared_pages[pg];
427 * Copy data from userspace to the current page
429 copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
431 /* Flush processor's dcache for this page */
432 flush_dcache_page(page);
435 * if we get a partial write, we can end up with
436 * partially up to date pages. These add
437 * a lot of complexity, so make sure they don't
438 * happen by forcing this copy to be retried.
440 * The rest of the btrfs_file_write code will fall
441 * back to page at a time copies after we return 0.
443 if (!PageUptodate(page) && copied < count)
446 iov_iter_advance(i, copied);
447 write_bytes -= copied;
448 total_copied += copied;
450 /* Return to btrfs_file_aio_write to fault page */
451 if (unlikely(copied == 0))
454 if (unlikely(copied < PAGE_CACHE_SIZE - offset)) {
465 * unlocks pages after btrfs_file_write is done with them
467 static void btrfs_drop_pages(struct page **pages, size_t num_pages)
470 for (i = 0; i < num_pages; i++) {
471 /* page checked is some magic around finding pages that
472 * have been modified without going through btrfs_set_page_dirty
475 ClearPageChecked(pages[i]);
476 unlock_page(pages[i]);
477 mark_page_accessed(pages[i]);
478 page_cache_release(pages[i]);
483 * after copy_from_user, pages need to be dirtied and we need to make
484 * sure holes are created between the current EOF and the start of
485 * any next extents (if required).
487 * this also makes the decision about creating an inline extent vs
488 * doing real data extents, marking pages dirty and delalloc as required.
490 int btrfs_dirty_pages(struct btrfs_root *root, struct inode *inode,
491 struct page **pages, size_t num_pages,
492 loff_t pos, size_t write_bytes,
493 struct extent_state **cached)
499 u64 end_of_last_block;
500 u64 end_pos = pos + write_bytes;
501 loff_t isize = i_size_read(inode);
503 start_pos = pos & ~((u64)root->sectorsize - 1);
504 num_bytes = ALIGN(write_bytes + pos - start_pos, root->sectorsize);
506 end_of_last_block = start_pos + num_bytes - 1;
507 err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
512 for (i = 0; i < num_pages; i++) {
513 struct page *p = pages[i];
520 * we've only changed i_size in ram, and we haven't updated
521 * the disk i_size. There is no need to log the inode
525 i_size_write(inode, end_pos);
530 * this drops all the extents in the cache that intersect the range
531 * [start, end]. Existing extents are split as required.
533 void btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
536 struct extent_map *em;
537 struct extent_map *split = NULL;
538 struct extent_map *split2 = NULL;
539 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
540 u64 len = end - start + 1;
548 WARN_ON(end < start);
549 if (end == (u64)-1) {
558 split = alloc_extent_map();
560 split2 = alloc_extent_map();
561 if (!split || !split2)
564 write_lock(&em_tree->lock);
565 em = lookup_extent_mapping(em_tree, start, len);
567 write_unlock(&em_tree->lock);
571 gen = em->generation;
572 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
573 if (testend && em->start + em->len >= start + len) {
575 write_unlock(&em_tree->lock);
578 start = em->start + em->len;
580 len = start + len - (em->start + em->len);
582 write_unlock(&em_tree->lock);
585 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
586 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
587 clear_bit(EXTENT_FLAG_LOGGING, &flags);
588 modified = !list_empty(&em->list);
592 if (em->start < start) {
593 split->start = em->start;
594 split->len = start - em->start;
596 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
597 split->orig_start = em->orig_start;
598 split->block_start = em->block_start;
601 split->block_len = em->block_len;
603 split->block_len = split->len;
604 split->orig_block_len = max(split->block_len,
606 split->ram_bytes = em->ram_bytes;
608 split->orig_start = split->start;
609 split->block_len = 0;
610 split->block_start = em->block_start;
611 split->orig_block_len = 0;
612 split->ram_bytes = split->len;
615 split->generation = gen;
616 split->bdev = em->bdev;
617 split->flags = flags;
618 split->compress_type = em->compress_type;
619 replace_extent_mapping(em_tree, em, split, modified);
620 free_extent_map(split);
624 if (testend && em->start + em->len > start + len) {
625 u64 diff = start + len - em->start;
627 split->start = start + len;
628 split->len = em->start + em->len - (start + len);
629 split->bdev = em->bdev;
630 split->flags = flags;
631 split->compress_type = em->compress_type;
632 split->generation = gen;
634 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
635 split->orig_block_len = max(em->block_len,
638 split->ram_bytes = em->ram_bytes;
640 split->block_len = em->block_len;
641 split->block_start = em->block_start;
642 split->orig_start = em->orig_start;
644 split->block_len = split->len;
645 split->block_start = em->block_start
647 split->orig_start = em->orig_start;
650 split->ram_bytes = split->len;
651 split->orig_start = split->start;
652 split->block_len = 0;
653 split->block_start = em->block_start;
654 split->orig_block_len = 0;
657 if (extent_map_in_tree(em)) {
658 replace_extent_mapping(em_tree, em, split,
661 ret = add_extent_mapping(em_tree, split,
663 ASSERT(ret == 0); /* Logic error */
665 free_extent_map(split);
669 if (extent_map_in_tree(em))
670 remove_extent_mapping(em_tree, em);
671 write_unlock(&em_tree->lock);
675 /* once for the tree*/
679 free_extent_map(split);
681 free_extent_map(split2);
685 * this is very complex, but the basic idea is to drop all extents
686 * in the range start - end. hint_block is filled in with a block number
687 * that would be a good hint to the block allocator for this file.
689 * If an extent intersects the range but is not entirely inside the range
690 * it is either truncated or split. Anything entirely inside the range
691 * is deleted from the tree.
693 int __btrfs_drop_extents(struct btrfs_trans_handle *trans,
694 struct btrfs_root *root, struct inode *inode,
695 struct btrfs_path *path, u64 start, u64 end,
696 u64 *drop_end, int drop_cache,
698 u32 extent_item_size,
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(inode);
706 u64 search_start = start;
709 u64 extent_offset = 0;
716 int modify_tree = -1;
719 int leafs_visited = 0;
722 btrfs_drop_extent_cache(inode, start, end - 1, 0);
724 if (start >= BTRFS_I(inode)->disk_i_size && !replace_extent)
727 update_refs = (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
728 root == root->fs_info->tree_root);
731 ret = btrfs_lookup_file_extent(trans, root, path, ino,
732 search_start, modify_tree);
735 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
736 leaf = path->nodes[0];
737 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
738 if (key.objectid == ino &&
739 key.type == BTRFS_EXTENT_DATA_KEY)
745 leaf = path->nodes[0];
746 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
748 ret = btrfs_next_leaf(root, path);
756 leaf = path->nodes[0];
760 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
761 if (key.objectid > ino ||
762 key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
765 fi = btrfs_item_ptr(leaf, path->slots[0],
766 struct btrfs_file_extent_item);
767 extent_type = btrfs_file_extent_type(leaf, fi);
769 if (extent_type == BTRFS_FILE_EXTENT_REG ||
770 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
771 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
772 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
773 extent_offset = btrfs_file_extent_offset(leaf, fi);
774 extent_end = key.offset +
775 btrfs_file_extent_num_bytes(leaf, fi);
776 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
777 extent_end = key.offset +
778 btrfs_file_extent_inline_len(leaf,
782 extent_end = search_start;
786 * Don't skip extent items representing 0 byte lengths. They
787 * used to be created (bug) if while punching holes we hit
788 * -ENOSPC condition. So if we find one here, just ensure we
789 * delete it, otherwise we would insert a new file extent item
790 * with the same key (offset) as that 0 bytes length file
791 * extent item in the call to setup_items_for_insert() later
794 if (extent_end == key.offset && extent_end >= search_start)
795 goto delete_extent_item;
797 if (extent_end <= search_start) {
803 search_start = max(key.offset, start);
804 if (recow || !modify_tree) {
806 btrfs_release_path(path);
811 * | - range to drop - |
812 * | -------- extent -------- |
814 if (start > key.offset && end < extent_end) {
816 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
821 memcpy(&new_key, &key, sizeof(new_key));
822 new_key.offset = start;
823 ret = btrfs_duplicate_item(trans, root, path,
825 if (ret == -EAGAIN) {
826 btrfs_release_path(path);
832 leaf = path->nodes[0];
833 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
834 struct btrfs_file_extent_item);
835 btrfs_set_file_extent_num_bytes(leaf, fi,
838 fi = btrfs_item_ptr(leaf, path->slots[0],
839 struct btrfs_file_extent_item);
841 extent_offset += start - key.offset;
842 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
843 btrfs_set_file_extent_num_bytes(leaf, fi,
845 btrfs_mark_buffer_dirty(leaf);
847 if (update_refs && disk_bytenr > 0) {
848 ret = btrfs_inc_extent_ref(trans, root,
849 disk_bytenr, num_bytes, 0,
850 root->root_key.objectid,
852 start - extent_offset, 0);
853 BUG_ON(ret); /* -ENOMEM */
858 * | ---- range to drop ----- |
859 * | -------- extent -------- |
861 if (start <= key.offset && end < extent_end) {
862 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
867 memcpy(&new_key, &key, sizeof(new_key));
868 new_key.offset = end;
869 btrfs_set_item_key_safe(root, path, &new_key);
871 extent_offset += end - key.offset;
872 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
873 btrfs_set_file_extent_num_bytes(leaf, fi,
875 btrfs_mark_buffer_dirty(leaf);
876 if (update_refs && disk_bytenr > 0)
877 inode_sub_bytes(inode, end - key.offset);
881 search_start = extent_end;
883 * | ---- range to drop ----- |
884 * | -------- extent -------- |
886 if (start > key.offset && end >= extent_end) {
888 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
893 btrfs_set_file_extent_num_bytes(leaf, fi,
895 btrfs_mark_buffer_dirty(leaf);
896 if (update_refs && disk_bytenr > 0)
897 inode_sub_bytes(inode, extent_end - start);
898 if (end == extent_end)
906 * | ---- range to drop ----- |
907 * | ------ extent ------ |
909 if (start <= key.offset && end >= extent_end) {
912 del_slot = path->slots[0];
915 BUG_ON(del_slot + del_nr != path->slots[0]);
920 extent_type == BTRFS_FILE_EXTENT_INLINE) {
921 inode_sub_bytes(inode,
922 extent_end - key.offset);
923 extent_end = ALIGN(extent_end,
925 } else if (update_refs && disk_bytenr > 0) {
926 ret = btrfs_free_extent(trans, root,
927 disk_bytenr, num_bytes, 0,
928 root->root_key.objectid,
929 key.objectid, key.offset -
931 BUG_ON(ret); /* -ENOMEM */
932 inode_sub_bytes(inode,
933 extent_end - key.offset);
936 if (end == extent_end)
939 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
944 ret = btrfs_del_items(trans, root, path, del_slot,
947 btrfs_abort_transaction(trans, root, ret);
954 btrfs_release_path(path);
961 if (!ret && del_nr > 0) {
963 * Set path->slots[0] to first slot, so that after the delete
964 * if items are move off from our leaf to its immediate left or
965 * right neighbor leafs, we end up with a correct and adjusted
966 * path->slots[0] for our insertion (if replace_extent != 0).
968 path->slots[0] = del_slot;
969 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
971 btrfs_abort_transaction(trans, root, ret);
974 leaf = path->nodes[0];
976 * If btrfs_del_items() was called, it might have deleted a leaf, in
977 * which case it unlocked our path, so check path->locks[0] matches a
980 if (!ret && replace_extent && leafs_visited == 1 &&
981 (path->locks[0] == BTRFS_WRITE_LOCK_BLOCKING ||
982 path->locks[0] == BTRFS_WRITE_LOCK) &&
983 btrfs_leaf_free_space(root, leaf) >=
984 sizeof(struct btrfs_item) + extent_item_size) {
987 key.type = BTRFS_EXTENT_DATA_KEY;
989 if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) {
990 struct btrfs_key slot_key;
992 btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]);
993 if (btrfs_comp_cpu_keys(&key, &slot_key) > 0)
996 setup_items_for_insert(root, path, &key,
999 sizeof(struct btrfs_item) +
1000 extent_item_size, 1);
1004 if (!replace_extent || !(*key_inserted))
1005 btrfs_release_path(path);
1007 *drop_end = found ? min(end, extent_end) : end;
1011 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
1012 struct btrfs_root *root, struct inode *inode, u64 start,
1013 u64 end, int drop_cache)
1015 struct btrfs_path *path;
1018 path = btrfs_alloc_path();
1021 ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL,
1022 drop_cache, 0, 0, NULL);
1023 btrfs_free_path(path);
1027 static int extent_mergeable(struct extent_buffer *leaf, int slot,
1028 u64 objectid, u64 bytenr, u64 orig_offset,
1029 u64 *start, u64 *end)
1031 struct btrfs_file_extent_item *fi;
1032 struct btrfs_key key;
1035 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
1038 btrfs_item_key_to_cpu(leaf, &key, slot);
1039 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
1042 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1043 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
1044 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
1045 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
1046 btrfs_file_extent_compression(leaf, fi) ||
1047 btrfs_file_extent_encryption(leaf, fi) ||
1048 btrfs_file_extent_other_encoding(leaf, fi))
1051 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1052 if ((*start && *start != key.offset) || (*end && *end != extent_end))
1055 *start = key.offset;
1061 * Mark extent in the range start - end as written.
1063 * This changes extent type from 'pre-allocated' to 'regular'. If only
1064 * part of extent is marked as written, the extent will be split into
1067 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
1068 struct inode *inode, u64 start, u64 end)
1070 struct btrfs_root *root = BTRFS_I(inode)->root;
1071 struct extent_buffer *leaf;
1072 struct btrfs_path *path;
1073 struct btrfs_file_extent_item *fi;
1074 struct btrfs_key key;
1075 struct btrfs_key new_key;
1087 u64 ino = btrfs_ino(inode);
1089 path = btrfs_alloc_path();
1096 key.type = BTRFS_EXTENT_DATA_KEY;
1099 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1102 if (ret > 0 && path->slots[0] > 0)
1105 leaf = path->nodes[0];
1106 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1107 BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY);
1108 fi = btrfs_item_ptr(leaf, path->slots[0],
1109 struct btrfs_file_extent_item);
1110 BUG_ON(btrfs_file_extent_type(leaf, fi) !=
1111 BTRFS_FILE_EXTENT_PREALLOC);
1112 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1113 BUG_ON(key.offset > start || extent_end < end);
1115 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1116 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1117 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
1118 memcpy(&new_key, &key, sizeof(new_key));
1120 if (start == key.offset && end < extent_end) {
1123 if (extent_mergeable(leaf, path->slots[0] - 1,
1124 ino, bytenr, orig_offset,
1125 &other_start, &other_end)) {
1126 new_key.offset = end;
1127 btrfs_set_item_key_safe(root, path, &new_key);
1128 fi = btrfs_item_ptr(leaf, path->slots[0],
1129 struct btrfs_file_extent_item);
1130 btrfs_set_file_extent_generation(leaf, fi,
1132 btrfs_set_file_extent_num_bytes(leaf, fi,
1134 btrfs_set_file_extent_offset(leaf, fi,
1136 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1137 struct btrfs_file_extent_item);
1138 btrfs_set_file_extent_generation(leaf, fi,
1140 btrfs_set_file_extent_num_bytes(leaf, fi,
1142 btrfs_mark_buffer_dirty(leaf);
1147 if (start > key.offset && end == extent_end) {
1150 if (extent_mergeable(leaf, path->slots[0] + 1,
1151 ino, bytenr, orig_offset,
1152 &other_start, &other_end)) {
1153 fi = btrfs_item_ptr(leaf, path->slots[0],
1154 struct btrfs_file_extent_item);
1155 btrfs_set_file_extent_num_bytes(leaf, fi,
1156 start - key.offset);
1157 btrfs_set_file_extent_generation(leaf, fi,
1160 new_key.offset = start;
1161 btrfs_set_item_key_safe(root, path, &new_key);
1163 fi = btrfs_item_ptr(leaf, path->slots[0],
1164 struct btrfs_file_extent_item);
1165 btrfs_set_file_extent_generation(leaf, fi,
1167 btrfs_set_file_extent_num_bytes(leaf, fi,
1169 btrfs_set_file_extent_offset(leaf, fi,
1170 start - orig_offset);
1171 btrfs_mark_buffer_dirty(leaf);
1176 while (start > key.offset || end < extent_end) {
1177 if (key.offset == start)
1180 new_key.offset = split;
1181 ret = btrfs_duplicate_item(trans, root, path, &new_key);
1182 if (ret == -EAGAIN) {
1183 btrfs_release_path(path);
1187 btrfs_abort_transaction(trans, root, ret);
1191 leaf = path->nodes[0];
1192 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1193 struct btrfs_file_extent_item);
1194 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1195 btrfs_set_file_extent_num_bytes(leaf, fi,
1196 split - key.offset);
1198 fi = btrfs_item_ptr(leaf, path->slots[0],
1199 struct btrfs_file_extent_item);
1201 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1202 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
1203 btrfs_set_file_extent_num_bytes(leaf, fi,
1204 extent_end - split);
1205 btrfs_mark_buffer_dirty(leaf);
1207 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
1208 root->root_key.objectid,
1209 ino, orig_offset, 0);
1210 BUG_ON(ret); /* -ENOMEM */
1212 if (split == start) {
1215 BUG_ON(start != key.offset);
1224 if (extent_mergeable(leaf, path->slots[0] + 1,
1225 ino, bytenr, orig_offset,
1226 &other_start, &other_end)) {
1228 btrfs_release_path(path);
1231 extent_end = other_end;
1232 del_slot = path->slots[0] + 1;
1234 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1235 0, root->root_key.objectid,
1236 ino, orig_offset, 0);
1237 BUG_ON(ret); /* -ENOMEM */
1241 if (extent_mergeable(leaf, path->slots[0] - 1,
1242 ino, bytenr, orig_offset,
1243 &other_start, &other_end)) {
1245 btrfs_release_path(path);
1248 key.offset = other_start;
1249 del_slot = path->slots[0];
1251 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1252 0, root->root_key.objectid,
1253 ino, orig_offset, 0);
1254 BUG_ON(ret); /* -ENOMEM */
1257 fi = btrfs_item_ptr(leaf, path->slots[0],
1258 struct btrfs_file_extent_item);
1259 btrfs_set_file_extent_type(leaf, fi,
1260 BTRFS_FILE_EXTENT_REG);
1261 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1262 btrfs_mark_buffer_dirty(leaf);
1264 fi = btrfs_item_ptr(leaf, del_slot - 1,
1265 struct btrfs_file_extent_item);
1266 btrfs_set_file_extent_type(leaf, fi,
1267 BTRFS_FILE_EXTENT_REG);
1268 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1269 btrfs_set_file_extent_num_bytes(leaf, fi,
1270 extent_end - key.offset);
1271 btrfs_mark_buffer_dirty(leaf);
1273 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1275 btrfs_abort_transaction(trans, root, ret);
1280 btrfs_free_path(path);
1285 * on error we return an unlocked page and the error value
1286 * on success we return a locked page and 0
1288 static int prepare_uptodate_page(struct page *page, u64 pos,
1289 bool force_uptodate)
1293 if (((pos & (PAGE_CACHE_SIZE - 1)) || force_uptodate) &&
1294 !PageUptodate(page)) {
1295 ret = btrfs_readpage(NULL, page);
1299 if (!PageUptodate(page)) {
1308 * this just gets pages into the page cache and locks them down.
1310 static noinline int prepare_pages(struct inode *inode, struct page **pages,
1311 size_t num_pages, loff_t pos,
1312 size_t write_bytes, bool force_uptodate)
1315 unsigned long index = pos >> PAGE_CACHE_SHIFT;
1316 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1320 for (i = 0; i < num_pages; i++) {
1321 pages[i] = find_or_create_page(inode->i_mapping, index + i,
1322 mask | __GFP_WRITE);
1330 err = prepare_uptodate_page(pages[i], pos,
1332 if (i == num_pages - 1)
1333 err = prepare_uptodate_page(pages[i],
1334 pos + write_bytes, false);
1336 page_cache_release(pages[i]);
1340 wait_on_page_writeback(pages[i]);
1345 while (faili >= 0) {
1346 unlock_page(pages[faili]);
1347 page_cache_release(pages[faili]);
1355 * This function locks the extent and properly waits for data=ordered extents
1356 * to finish before allowing the pages to be modified if need.
1359 * 1 - the extent is locked
1360 * 0 - the extent is not locked, and everything is OK
1361 * -EAGAIN - need re-prepare the pages
1362 * the other < 0 number - Something wrong happens
1365 lock_and_cleanup_extent_if_need(struct inode *inode, struct page **pages,
1366 size_t num_pages, loff_t pos,
1367 u64 *lockstart, u64 *lockend,
1368 struct extent_state **cached_state)
1375 start_pos = pos & ~((u64)PAGE_CACHE_SIZE - 1);
1376 last_pos = start_pos + ((u64)num_pages << PAGE_CACHE_SHIFT) - 1;
1378 if (start_pos < inode->i_size) {
1379 struct btrfs_ordered_extent *ordered;
1380 lock_extent_bits(&BTRFS_I(inode)->io_tree,
1381 start_pos, last_pos, 0, cached_state);
1382 ordered = btrfs_lookup_ordered_range(inode, start_pos,
1383 last_pos - start_pos + 1);
1385 ordered->file_offset + ordered->len > start_pos &&
1386 ordered->file_offset <= last_pos) {
1387 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1388 start_pos, last_pos,
1389 cached_state, GFP_NOFS);
1390 for (i = 0; i < num_pages; i++) {
1391 unlock_page(pages[i]);
1392 page_cache_release(pages[i]);
1394 btrfs_start_ordered_extent(inode, ordered, 1);
1395 btrfs_put_ordered_extent(ordered);
1399 btrfs_put_ordered_extent(ordered);
1401 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
1402 last_pos, EXTENT_DIRTY | EXTENT_DELALLOC |
1403 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
1404 0, 0, cached_state, GFP_NOFS);
1405 *lockstart = start_pos;
1406 *lockend = last_pos;
1410 for (i = 0; i < num_pages; i++) {
1411 if (clear_page_dirty_for_io(pages[i]))
1412 account_page_redirty(pages[i]);
1413 set_page_extent_mapped(pages[i]);
1414 WARN_ON(!PageLocked(pages[i]));
1420 static noinline int check_can_nocow(struct inode *inode, loff_t pos,
1421 size_t *write_bytes)
1423 struct btrfs_root *root = BTRFS_I(inode)->root;
1424 struct btrfs_ordered_extent *ordered;
1425 u64 lockstart, lockend;
1429 ret = btrfs_start_nocow_write(root);
1433 lockstart = round_down(pos, root->sectorsize);
1434 lockend = round_up(pos + *write_bytes, root->sectorsize) - 1;
1437 lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1438 ordered = btrfs_lookup_ordered_range(inode, lockstart,
1439 lockend - lockstart + 1);
1443 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1444 btrfs_start_ordered_extent(inode, ordered, 1);
1445 btrfs_put_ordered_extent(ordered);
1448 num_bytes = lockend - lockstart + 1;
1449 ret = can_nocow_extent(inode, lockstart, &num_bytes, NULL, NULL, NULL);
1452 btrfs_end_nocow_write(root);
1454 *write_bytes = min_t(size_t, *write_bytes ,
1455 num_bytes - pos + lockstart);
1458 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1463 static noinline ssize_t __btrfs_buffered_write(struct file *file,
1467 struct inode *inode = file_inode(file);
1468 struct btrfs_root *root = BTRFS_I(inode)->root;
1469 struct page **pages = NULL;
1470 struct extent_state *cached_state = NULL;
1471 u64 release_bytes = 0;
1474 unsigned long first_index;
1475 size_t num_written = 0;
1478 bool only_release_metadata = false;
1479 bool force_page_uptodate = false;
1482 nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) /
1483 PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
1484 (sizeof(struct page *)));
1485 nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1486 nrptrs = max(nrptrs, 8);
1487 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
1491 first_index = pos >> PAGE_CACHE_SHIFT;
1493 while (iov_iter_count(i) > 0) {
1494 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1495 size_t write_bytes = min(iov_iter_count(i),
1496 nrptrs * (size_t)PAGE_CACHE_SIZE -
1498 size_t num_pages = (write_bytes + offset +
1499 PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1500 size_t reserve_bytes;
1504 WARN_ON(num_pages > nrptrs);
1507 * Fault pages before locking them in prepare_pages
1508 * to avoid recursive lock
1510 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1515 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1516 ret = btrfs_check_data_free_space(inode, reserve_bytes);
1517 if (ret == -ENOSPC &&
1518 (BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW |
1519 BTRFS_INODE_PREALLOC))) {
1520 ret = check_can_nocow(inode, pos, &write_bytes);
1522 only_release_metadata = true;
1524 * our prealloc extent may be smaller than
1525 * write_bytes, so scale down.
1527 num_pages = (write_bytes + offset +
1528 PAGE_CACHE_SIZE - 1) >>
1530 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1540 ret = btrfs_delalloc_reserve_metadata(inode, reserve_bytes);
1542 if (!only_release_metadata)
1543 btrfs_free_reserved_data_space(inode,
1546 btrfs_end_nocow_write(root);
1550 release_bytes = reserve_bytes;
1551 need_unlock = false;
1554 * This is going to setup the pages array with the number of
1555 * pages we want, so we don't really need to worry about the
1556 * contents of pages from loop to loop
1558 ret = prepare_pages(inode, pages, num_pages,
1560 force_page_uptodate);
1564 ret = lock_and_cleanup_extent_if_need(inode, pages, num_pages,
1565 pos, &lockstart, &lockend,
1571 } else if (ret > 0) {
1576 copied = btrfs_copy_from_user(pos, num_pages,
1577 write_bytes, pages, i);
1580 * if we have trouble faulting in the pages, fall
1581 * back to one page at a time
1583 if (copied < write_bytes)
1587 force_page_uptodate = true;
1590 force_page_uptodate = false;
1591 dirty_pages = (copied + offset +
1592 PAGE_CACHE_SIZE - 1) >>
1597 * If we had a short copy we need to release the excess delaloc
1598 * bytes we reserved. We need to increment outstanding_extents
1599 * because btrfs_delalloc_release_space will decrement it, but
1600 * we still have an outstanding extent for the chunk we actually
1603 if (num_pages > dirty_pages) {
1604 release_bytes = (num_pages - dirty_pages) <<
1607 spin_lock(&BTRFS_I(inode)->lock);
1608 BTRFS_I(inode)->outstanding_extents++;
1609 spin_unlock(&BTRFS_I(inode)->lock);
1611 if (only_release_metadata)
1612 btrfs_delalloc_release_metadata(inode,
1615 btrfs_delalloc_release_space(inode,
1619 release_bytes = dirty_pages << PAGE_CACHE_SHIFT;
1622 ret = btrfs_dirty_pages(root, inode, pages,
1623 dirty_pages, pos, copied,
1626 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1627 lockstart, lockend, &cached_state,
1630 btrfs_drop_pages(pages, num_pages);
1635 if (only_release_metadata)
1636 btrfs_end_nocow_write(root);
1638 if (only_release_metadata && copied > 0) {
1639 u64 lockstart = round_down(pos, root->sectorsize);
1640 u64 lockend = lockstart +
1641 (dirty_pages << PAGE_CACHE_SHIFT) - 1;
1643 set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
1644 lockend, EXTENT_NORESERVE, NULL,
1646 only_release_metadata = false;
1649 btrfs_drop_pages(pages, num_pages);
1653 balance_dirty_pages_ratelimited(inode->i_mapping);
1654 if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1655 btrfs_btree_balance_dirty(root);
1658 num_written += copied;
1663 if (release_bytes) {
1664 if (only_release_metadata) {
1665 btrfs_end_nocow_write(root);
1666 btrfs_delalloc_release_metadata(inode, release_bytes);
1668 btrfs_delalloc_release_space(inode, release_bytes);
1672 return num_written ? num_written : ret;
1675 static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1676 const struct iovec *iov,
1677 unsigned long nr_segs, loff_t pos,
1678 size_t count, size_t ocount)
1680 struct file *file = iocb->ki_filp;
1683 ssize_t written_buffered;
1687 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
1690 if (written < 0 || written == count)
1695 iov_iter_init(&i, iov, nr_segs, count, written);
1696 written_buffered = __btrfs_buffered_write(file, &i, pos);
1697 if (written_buffered < 0) {
1698 err = written_buffered;
1701 endbyte = pos + written_buffered - 1;
1702 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
1705 written += written_buffered;
1706 iocb->ki_pos = pos + written_buffered;
1707 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
1708 endbyte >> PAGE_CACHE_SHIFT);
1710 return written ? written : err;
1713 static void update_time_for_write(struct inode *inode)
1715 struct timespec now;
1717 if (IS_NOCMTIME(inode))
1720 now = current_fs_time(inode->i_sb);
1721 if (!timespec_equal(&inode->i_mtime, &now))
1722 inode->i_mtime = now;
1724 if (!timespec_equal(&inode->i_ctime, &now))
1725 inode->i_ctime = now;
1727 if (IS_I_VERSION(inode))
1728 inode_inc_iversion(inode);
1731 static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
1732 const struct iovec *iov,
1733 unsigned long nr_segs, loff_t pos)
1735 struct file *file = iocb->ki_filp;
1736 struct inode *inode = file_inode(file);
1737 struct btrfs_root *root = BTRFS_I(inode)->root;
1740 ssize_t num_written = 0;
1742 size_t count, ocount;
1743 bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host);
1745 mutex_lock(&inode->i_mutex);
1747 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
1749 mutex_unlock(&inode->i_mutex);
1754 current->backing_dev_info = inode->i_mapping->backing_dev_info;
1755 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1757 mutex_unlock(&inode->i_mutex);
1762 mutex_unlock(&inode->i_mutex);
1766 err = file_remove_suid(file);
1768 mutex_unlock(&inode->i_mutex);
1773 * If BTRFS flips readonly due to some impossible error
1774 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1775 * although we have opened a file as writable, we have
1776 * to stop this write operation to ensure FS consistency.
1778 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
1779 mutex_unlock(&inode->i_mutex);
1785 * We reserve space for updating the inode when we reserve space for the
1786 * extent we are going to write, so we will enospc out there. We don't
1787 * need to start yet another transaction to update the inode as we will
1788 * update the inode when we finish writing whatever data we write.
1790 update_time_for_write(inode);
1792 start_pos = round_down(pos, root->sectorsize);
1793 if (start_pos > i_size_read(inode)) {
1794 /* Expand hole size to cover write data, preventing empty gap */
1795 end_pos = round_up(pos + count, root->sectorsize);
1796 err = btrfs_cont_expand(inode, i_size_read(inode), end_pos);
1798 mutex_unlock(&inode->i_mutex);
1804 atomic_inc(&BTRFS_I(inode)->sync_writers);
1806 if (unlikely(file->f_flags & O_DIRECT)) {
1807 num_written = __btrfs_direct_write(iocb, iov, nr_segs,
1808 pos, count, ocount);
1812 iov_iter_init(&i, iov, nr_segs, count, num_written);
1814 num_written = __btrfs_buffered_write(file, &i, pos);
1815 if (num_written > 0)
1816 iocb->ki_pos = pos + num_written;
1819 mutex_unlock(&inode->i_mutex);
1822 * we want to make sure fsync finds this change
1823 * but we haven't joined a transaction running right now.
1825 * Later on, someone is sure to update the inode and get the
1826 * real transid recorded.
1828 * We set last_trans now to the fs_info generation + 1,
1829 * this will either be one more than the running transaction
1830 * or the generation used for the next transaction if there isn't
1831 * one running right now.
1833 * We also have to set last_sub_trans to the current log transid,
1834 * otherwise subsequent syncs to a file that's been synced in this
1835 * transaction will appear to have already occured.
1837 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1838 BTRFS_I(inode)->last_sub_trans = root->log_transid;
1839 if (num_written > 0) {
1840 err = generic_write_sync(file, pos, num_written);
1846 atomic_dec(&BTRFS_I(inode)->sync_writers);
1848 current->backing_dev_info = NULL;
1849 return num_written ? num_written : err;
1852 int btrfs_release_file(struct inode *inode, struct file *filp)
1855 * ordered_data_close is set by settattr when we are about to truncate
1856 * a file from a non-zero size to a zero size. This tries to
1857 * flush down new bytes that may have been written if the
1858 * application were using truncate to replace a file in place.
1860 if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
1861 &BTRFS_I(inode)->runtime_flags)) {
1862 struct btrfs_trans_handle *trans;
1863 struct btrfs_root *root = BTRFS_I(inode)->root;
1866 * We need to block on a committing transaction to keep us from
1867 * throwing a ordered operation on to the list and causing
1868 * something like sync to deadlock trying to flush out this
1871 trans = btrfs_start_transaction(root, 0);
1873 return PTR_ERR(trans);
1874 btrfs_add_ordered_operation(trans, BTRFS_I(inode)->root, inode);
1875 btrfs_end_transaction(trans, root);
1876 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1877 filemap_flush(inode->i_mapping);
1879 if (filp->private_data)
1880 btrfs_ioctl_trans_end(filp);
1885 * fsync call for both files and directories. This logs the inode into
1886 * the tree log instead of forcing full commits whenever possible.
1888 * It needs to call filemap_fdatawait so that all ordered extent updates are
1889 * in the metadata btree are up to date for copying to the log.
1891 * It drops the inode mutex before doing the tree log commit. This is an
1892 * important optimization for directories because holding the mutex prevents
1893 * new operations on the dir while we write to disk.
1895 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1897 struct dentry *dentry = file->f_path.dentry;
1898 struct inode *inode = dentry->d_inode;
1899 struct btrfs_root *root = BTRFS_I(inode)->root;
1900 struct btrfs_trans_handle *trans;
1901 struct btrfs_log_ctx ctx;
1905 trace_btrfs_sync_file(file, datasync);
1908 * We write the dirty pages in the range and wait until they complete
1909 * out of the ->i_mutex. If so, we can flush the dirty pages by
1910 * multi-task, and make the performance up. See
1911 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1913 atomic_inc(&BTRFS_I(inode)->sync_writers);
1914 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
1915 if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1916 &BTRFS_I(inode)->runtime_flags))
1917 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
1918 atomic_dec(&BTRFS_I(inode)->sync_writers);
1922 mutex_lock(&inode->i_mutex);
1925 * We flush the dirty pages again to avoid some dirty pages in the
1928 atomic_inc(&root->log_batch);
1929 full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1930 &BTRFS_I(inode)->runtime_flags);
1932 ret = btrfs_wait_ordered_range(inode, start, end - start + 1);
1934 mutex_unlock(&inode->i_mutex);
1938 atomic_inc(&root->log_batch);
1941 * check the transaction that last modified this inode
1942 * and see if its already been committed
1944 if (!BTRFS_I(inode)->last_trans) {
1945 mutex_unlock(&inode->i_mutex);
1950 * if the last transaction that changed this file was before
1951 * the current transaction, we can bail out now without any
1955 if (btrfs_inode_in_log(inode, root->fs_info->generation) ||
1956 BTRFS_I(inode)->last_trans <=
1957 root->fs_info->last_trans_committed) {
1958 BTRFS_I(inode)->last_trans = 0;
1961 * We'v had everything committed since the last time we were
1962 * modified so clear this flag in case it was set for whatever
1963 * reason, it's no longer relevant.
1965 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1966 &BTRFS_I(inode)->runtime_flags);
1967 mutex_unlock(&inode->i_mutex);
1972 * ok we haven't committed the transaction yet, lets do a commit
1974 if (file->private_data)
1975 btrfs_ioctl_trans_end(file);
1978 * We use start here because we will need to wait on the IO to complete
1979 * in btrfs_sync_log, which could require joining a transaction (for
1980 * example checking cross references in the nocow path). If we use join
1981 * here we could get into a situation where we're waiting on IO to
1982 * happen that is blocked on a transaction trying to commit. With start
1983 * we inc the extwriter counter, so we wait for all extwriters to exit
1984 * before we start blocking join'ers. This comment is to keep somebody
1985 * from thinking they are super smart and changing this to
1986 * btrfs_join_transaction *cough*Josef*cough*.
1988 trans = btrfs_start_transaction(root, 0);
1989 if (IS_ERR(trans)) {
1990 ret = PTR_ERR(trans);
1991 mutex_unlock(&inode->i_mutex);
1996 btrfs_init_log_ctx(&ctx);
1998 ret = btrfs_log_dentry_safe(trans, root, dentry, &ctx);
2000 /* Fallthrough and commit/free transaction. */
2004 /* we've logged all the items and now have a consistent
2005 * version of the file in the log. It is possible that
2006 * someone will come in and modify the file, but that's
2007 * fine because the log is consistent on disk, and we
2008 * have references to all of the file's extents
2010 * It is possible that someone will come in and log the
2011 * file again, but that will end up using the synchronization
2012 * inside btrfs_sync_log to keep things safe.
2014 mutex_unlock(&inode->i_mutex);
2016 if (ret != BTRFS_NO_LOG_SYNC) {
2018 ret = btrfs_sync_log(trans, root, &ctx);
2020 ret = btrfs_end_transaction(trans, root);
2025 ret = btrfs_wait_ordered_range(inode, start,
2030 ret = btrfs_commit_transaction(trans, root);
2032 ret = btrfs_end_transaction(trans, root);
2035 return ret > 0 ? -EIO : ret;
2038 static const struct vm_operations_struct btrfs_file_vm_ops = {
2039 .fault = filemap_fault,
2040 .map_pages = filemap_map_pages,
2041 .page_mkwrite = btrfs_page_mkwrite,
2042 .remap_pages = generic_file_remap_pages,
2045 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
2047 struct address_space *mapping = filp->f_mapping;
2049 if (!mapping->a_ops->readpage)
2052 file_accessed(filp);
2053 vma->vm_ops = &btrfs_file_vm_ops;
2058 static int hole_mergeable(struct inode *inode, struct extent_buffer *leaf,
2059 int slot, u64 start, u64 end)
2061 struct btrfs_file_extent_item *fi;
2062 struct btrfs_key key;
2064 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
2067 btrfs_item_key_to_cpu(leaf, &key, slot);
2068 if (key.objectid != btrfs_ino(inode) ||
2069 key.type != BTRFS_EXTENT_DATA_KEY)
2072 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2074 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2077 if (btrfs_file_extent_disk_bytenr(leaf, fi))
2080 if (key.offset == end)
2082 if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
2087 static int fill_holes(struct btrfs_trans_handle *trans, struct inode *inode,
2088 struct btrfs_path *path, u64 offset, u64 end)
2090 struct btrfs_root *root = BTRFS_I(inode)->root;
2091 struct extent_buffer *leaf;
2092 struct btrfs_file_extent_item *fi;
2093 struct extent_map *hole_em;
2094 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2095 struct btrfs_key key;
2098 if (btrfs_fs_incompat(root->fs_info, NO_HOLES))
2101 key.objectid = btrfs_ino(inode);
2102 key.type = BTRFS_EXTENT_DATA_KEY;
2103 key.offset = offset;
2105 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2110 leaf = path->nodes[0];
2111 if (hole_mergeable(inode, leaf, path->slots[0]-1, offset, end)) {
2115 fi = btrfs_item_ptr(leaf, path->slots[0],
2116 struct btrfs_file_extent_item);
2117 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2119 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2120 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2121 btrfs_set_file_extent_offset(leaf, fi, 0);
2122 btrfs_mark_buffer_dirty(leaf);
2126 if (hole_mergeable(inode, leaf, path->slots[0]+1, offset, end)) {
2130 key.offset = offset;
2131 btrfs_set_item_key_safe(root, path, &key);
2132 fi = btrfs_item_ptr(leaf, path->slots[0],
2133 struct btrfs_file_extent_item);
2134 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2136 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2137 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2138 btrfs_set_file_extent_offset(leaf, fi, 0);
2139 btrfs_mark_buffer_dirty(leaf);
2142 btrfs_release_path(path);
2144 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
2145 0, 0, end - offset, 0, end - offset,
2151 btrfs_release_path(path);
2153 hole_em = alloc_extent_map();
2155 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2156 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2157 &BTRFS_I(inode)->runtime_flags);
2159 hole_em->start = offset;
2160 hole_em->len = end - offset;
2161 hole_em->ram_bytes = hole_em->len;
2162 hole_em->orig_start = offset;
2164 hole_em->block_start = EXTENT_MAP_HOLE;
2165 hole_em->block_len = 0;
2166 hole_em->orig_block_len = 0;
2167 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
2168 hole_em->compress_type = BTRFS_COMPRESS_NONE;
2169 hole_em->generation = trans->transid;
2172 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2173 write_lock(&em_tree->lock);
2174 ret = add_extent_mapping(em_tree, hole_em, 1);
2175 write_unlock(&em_tree->lock);
2176 } while (ret == -EEXIST);
2177 free_extent_map(hole_em);
2179 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2180 &BTRFS_I(inode)->runtime_flags);
2186 static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
2188 struct btrfs_root *root = BTRFS_I(inode)->root;
2189 struct extent_state *cached_state = NULL;
2190 struct btrfs_path *path;
2191 struct btrfs_block_rsv *rsv;
2192 struct btrfs_trans_handle *trans;
2193 u64 lockstart = round_up(offset, BTRFS_I(inode)->root->sectorsize);
2194 u64 lockend = round_down(offset + len,
2195 BTRFS_I(inode)->root->sectorsize) - 1;
2196 u64 cur_offset = lockstart;
2197 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
2202 bool same_page = ((offset >> PAGE_CACHE_SHIFT) ==
2203 ((offset + len - 1) >> PAGE_CACHE_SHIFT));
2204 bool no_holes = btrfs_fs_incompat(root->fs_info, NO_HOLES);
2207 ret = btrfs_wait_ordered_range(inode, offset, len);
2211 mutex_lock(&inode->i_mutex);
2212 ino_size = round_up(inode->i_size, PAGE_CACHE_SIZE);
2214 * We needn't truncate any page which is beyond the end of the file
2215 * because we are sure there is no data there.
2218 * Only do this if we are in the same page and we aren't doing the
2221 if (same_page && len < PAGE_CACHE_SIZE) {
2222 if (offset < ino_size)
2223 ret = btrfs_truncate_page(inode, offset, len, 0);
2224 mutex_unlock(&inode->i_mutex);
2228 /* zero back part of the first page */
2229 if (offset < ino_size) {
2230 ret = btrfs_truncate_page(inode, offset, 0, 0);
2232 mutex_unlock(&inode->i_mutex);
2237 /* zero the front end of the last page */
2238 if (offset + len < ino_size) {
2239 ret = btrfs_truncate_page(inode, offset + len, 0, 1);
2241 mutex_unlock(&inode->i_mutex);
2246 if (lockend < lockstart) {
2247 mutex_unlock(&inode->i_mutex);
2252 struct btrfs_ordered_extent *ordered;
2254 truncate_pagecache_range(inode, lockstart, lockend);
2256 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2258 ordered = btrfs_lookup_first_ordered_extent(inode, lockend);
2261 * We need to make sure we have no ordered extents in this range
2262 * and nobody raced in and read a page in this range, if we did
2263 * we need to try again.
2266 (ordered->file_offset + ordered->len <= lockstart ||
2267 ordered->file_offset > lockend)) &&
2268 !test_range_bit(&BTRFS_I(inode)->io_tree, lockstart,
2269 lockend, EXTENT_UPTODATE, 0,
2272 btrfs_put_ordered_extent(ordered);
2276 btrfs_put_ordered_extent(ordered);
2277 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
2278 lockend, &cached_state, GFP_NOFS);
2279 ret = btrfs_wait_ordered_range(inode, lockstart,
2280 lockend - lockstart + 1);
2282 mutex_unlock(&inode->i_mutex);
2287 path = btrfs_alloc_path();
2293 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2298 rsv->size = btrfs_calc_trunc_metadata_size(root, 1);
2302 * 1 - update the inode
2303 * 1 - removing the extents in the range
2304 * 1 - adding the hole extent if no_holes isn't set
2306 rsv_count = no_holes ? 2 : 3;
2307 trans = btrfs_start_transaction(root, rsv_count);
2308 if (IS_ERR(trans)) {
2309 err = PTR_ERR(trans);
2313 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
2316 trans->block_rsv = rsv;
2318 while (cur_offset < lockend) {
2319 ret = __btrfs_drop_extents(trans, root, inode, path,
2320 cur_offset, lockend + 1,
2321 &drop_end, 1, 0, 0, NULL);
2325 trans->block_rsv = &root->fs_info->trans_block_rsv;
2327 if (cur_offset < ino_size) {
2328 ret = fill_holes(trans, inode, path, cur_offset,
2336 cur_offset = drop_end;
2338 ret = btrfs_update_inode(trans, root, inode);
2344 btrfs_end_transaction(trans, root);
2345 btrfs_btree_balance_dirty(root);
2347 trans = btrfs_start_transaction(root, rsv_count);
2348 if (IS_ERR(trans)) {
2349 ret = PTR_ERR(trans);
2354 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
2356 BUG_ON(ret); /* shouldn't happen */
2357 trans->block_rsv = rsv;
2365 trans->block_rsv = &root->fs_info->trans_block_rsv;
2367 * Don't insert file hole extent item if it's for a range beyond eof
2368 * (because it's useless) or if it represents a 0 bytes range (when
2369 * cur_offset == drop_end).
2371 if (cur_offset < ino_size && cur_offset < drop_end) {
2372 ret = fill_holes(trans, inode, path, cur_offset, drop_end);
2383 inode_inc_iversion(inode);
2384 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2386 trans->block_rsv = &root->fs_info->trans_block_rsv;
2387 ret = btrfs_update_inode(trans, root, inode);
2388 btrfs_end_transaction(trans, root);
2389 btrfs_btree_balance_dirty(root);
2391 btrfs_free_path(path);
2392 btrfs_free_block_rsv(root, rsv);
2394 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2395 &cached_state, GFP_NOFS);
2396 mutex_unlock(&inode->i_mutex);
2402 static long btrfs_fallocate(struct file *file, int mode,
2403 loff_t offset, loff_t len)
2405 struct inode *inode = file_inode(file);
2406 struct extent_state *cached_state = NULL;
2407 struct btrfs_root *root = BTRFS_I(inode)->root;
2414 struct extent_map *em;
2415 int blocksize = BTRFS_I(inode)->root->sectorsize;
2418 alloc_start = round_down(offset, blocksize);
2419 alloc_end = round_up(offset + len, blocksize);
2421 /* Make sure we aren't being give some crap mode */
2422 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2425 if (mode & FALLOC_FL_PUNCH_HOLE)
2426 return btrfs_punch_hole(inode, offset, len);
2429 * Make sure we have enough space before we do the
2432 ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
2435 if (root->fs_info->quota_enabled) {
2436 ret = btrfs_qgroup_reserve(root, alloc_end - alloc_start);
2438 goto out_reserve_fail;
2441 mutex_lock(&inode->i_mutex);
2442 ret = inode_newsize_ok(inode, alloc_end);
2446 if (alloc_start > inode->i_size) {
2447 ret = btrfs_cont_expand(inode, i_size_read(inode),
2453 * If we are fallocating from the end of the file onward we
2454 * need to zero out the end of the page if i_size lands in the
2457 ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
2463 * wait for ordered IO before we have any locks. We'll loop again
2464 * below with the locks held.
2466 ret = btrfs_wait_ordered_range(inode, alloc_start,
2467 alloc_end - alloc_start);
2471 locked_end = alloc_end - 1;
2473 struct btrfs_ordered_extent *ordered;
2475 /* the extent lock is ordered inside the running
2478 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
2479 locked_end, 0, &cached_state);
2480 ordered = btrfs_lookup_first_ordered_extent(inode,
2483 ordered->file_offset + ordered->len > alloc_start &&
2484 ordered->file_offset < alloc_end) {
2485 btrfs_put_ordered_extent(ordered);
2486 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
2487 alloc_start, locked_end,
2488 &cached_state, GFP_NOFS);
2490 * we can't wait on the range with the transaction
2491 * running or with the extent lock held
2493 ret = btrfs_wait_ordered_range(inode, alloc_start,
2494 alloc_end - alloc_start);
2499 btrfs_put_ordered_extent(ordered);
2504 cur_offset = alloc_start;
2508 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2509 alloc_end - cur_offset, 0);
2510 if (IS_ERR_OR_NULL(em)) {
2517 last_byte = min(extent_map_end(em), alloc_end);
2518 actual_end = min_t(u64, extent_map_end(em), offset + len);
2519 last_byte = ALIGN(last_byte, blocksize);
2521 if (em->block_start == EXTENT_MAP_HOLE ||
2522 (cur_offset >= inode->i_size &&
2523 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
2524 ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
2525 last_byte - cur_offset,
2526 1 << inode->i_blkbits,
2531 free_extent_map(em);
2534 } else if (actual_end > inode->i_size &&
2535 !(mode & FALLOC_FL_KEEP_SIZE)) {
2537 * We didn't need to allocate any more space, but we
2538 * still extended the size of the file so we need to
2541 inode->i_ctime = CURRENT_TIME;
2542 i_size_write(inode, actual_end);
2543 btrfs_ordered_update_i_size(inode, actual_end, NULL);
2545 free_extent_map(em);
2547 cur_offset = last_byte;
2548 if (cur_offset >= alloc_end) {
2553 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
2554 &cached_state, GFP_NOFS);
2556 mutex_unlock(&inode->i_mutex);
2557 if (root->fs_info->quota_enabled)
2558 btrfs_qgroup_free(root, alloc_end - alloc_start);
2560 /* Let go of our reservation. */
2561 btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
2565 static int find_desired_extent(struct inode *inode, loff_t *offset, int whence)
2567 struct btrfs_root *root = BTRFS_I(inode)->root;
2568 struct extent_map *em = NULL;
2569 struct extent_state *cached_state = NULL;
2570 u64 lockstart = *offset;
2571 u64 lockend = i_size_read(inode);
2572 u64 start = *offset;
2573 u64 len = i_size_read(inode);
2576 lockend = max_t(u64, root->sectorsize, lockend);
2577 if (lockend <= lockstart)
2578 lockend = lockstart + root->sectorsize;
2581 len = lockend - lockstart + 1;
2583 len = max_t(u64, len, root->sectorsize);
2584 if (inode->i_size == 0)
2587 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, 0,
2590 while (start < inode->i_size) {
2591 em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0);
2598 if (whence == SEEK_HOLE &&
2599 (em->block_start == EXTENT_MAP_HOLE ||
2600 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2602 else if (whence == SEEK_DATA &&
2603 (em->block_start != EXTENT_MAP_HOLE &&
2604 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2607 start = em->start + em->len;
2608 free_extent_map(em);
2612 free_extent_map(em);
2614 if (whence == SEEK_DATA && start >= inode->i_size)
2617 *offset = min_t(loff_t, start, inode->i_size);
2619 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2620 &cached_state, GFP_NOFS);
2624 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
2626 struct inode *inode = file->f_mapping->host;
2629 mutex_lock(&inode->i_mutex);
2633 offset = generic_file_llseek(file, offset, whence);
2637 if (offset >= i_size_read(inode)) {
2638 mutex_unlock(&inode->i_mutex);
2642 ret = find_desired_extent(inode, &offset, whence);
2644 mutex_unlock(&inode->i_mutex);
2649 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2651 mutex_unlock(&inode->i_mutex);
2655 const struct file_operations btrfs_file_operations = {
2656 .llseek = btrfs_file_llseek,
2657 .read = do_sync_read,
2658 .write = do_sync_write,
2659 .aio_read = generic_file_aio_read,
2660 .splice_read = generic_file_splice_read,
2661 .aio_write = btrfs_file_aio_write,
2662 .mmap = btrfs_file_mmap,
2663 .open = generic_file_open,
2664 .release = btrfs_release_file,
2665 .fsync = btrfs_sync_file,
2666 .fallocate = btrfs_fallocate,
2667 .unlocked_ioctl = btrfs_ioctl,
2668 #ifdef CONFIG_COMPAT
2669 .compat_ioctl = btrfs_ioctl,
2673 void btrfs_auto_defrag_exit(void)
2675 if (btrfs_inode_defrag_cachep)
2676 kmem_cache_destroy(btrfs_inode_defrag_cachep);
2679 int btrfs_auto_defrag_init(void)
2681 btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
2682 sizeof(struct inode_defrag), 0,
2683 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2685 if (!btrfs_inode_defrag_cachep)