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/statfs.h>
31 #include <linux/compat.h>
32 #include <linux/slab.h>
35 #include "transaction.h"
36 #include "btrfs_inode.h"
38 #include "print-tree.h"
44 * when auto defrag is enabled we
45 * queue up these defrag structs to remember which
46 * inodes need defragging passes
49 struct rb_node rb_node;
53 * transid where the defrag was added, we search for
54 * extents newer than this
61 /* last offset we were able to defrag */
64 /* if we've wrapped around back to zero once already */
68 /* pop a record for an inode into the defrag tree. The lock
69 * must be held already
71 * If you're inserting a record for an older transid than an
72 * existing record, the transid already in the tree is lowered
74 * If an existing record is found the defrag item you
77 static void __btrfs_add_inode_defrag(struct inode *inode,
78 struct inode_defrag *defrag)
80 struct btrfs_root *root = BTRFS_I(inode)->root;
81 struct inode_defrag *entry;
83 struct rb_node *parent = NULL;
85 p = &root->fs_info->defrag_inodes.rb_node;
88 entry = rb_entry(parent, struct inode_defrag, rb_node);
90 if (defrag->ino < entry->ino)
92 else if (defrag->ino > entry->ino)
93 p = &parent->rb_right;
95 /* if we're reinserting an entry for
96 * an old defrag run, make sure to
97 * lower the transid of our existing record
99 if (defrag->transid < entry->transid)
100 entry->transid = defrag->transid;
101 if (defrag->last_offset > entry->last_offset)
102 entry->last_offset = defrag->last_offset;
106 BTRFS_I(inode)->in_defrag = 1;
107 rb_link_node(&defrag->rb_node, parent, p);
108 rb_insert_color(&defrag->rb_node, &root->fs_info->defrag_inodes);
118 * insert a defrag record for this inode if auto defrag is
121 int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans,
124 struct btrfs_root *root = BTRFS_I(inode)->root;
125 struct inode_defrag *defrag;
128 if (!btrfs_test_opt(root, AUTO_DEFRAG))
131 if (btrfs_fs_closing(root->fs_info))
134 if (BTRFS_I(inode)->in_defrag)
138 transid = trans->transid;
140 transid = BTRFS_I(inode)->root->last_trans;
142 defrag = kzalloc(sizeof(*defrag), GFP_NOFS);
146 defrag->ino = btrfs_ino(inode);
147 defrag->transid = transid;
148 defrag->root = root->root_key.objectid;
150 spin_lock(&root->fs_info->defrag_inodes_lock);
151 if (!BTRFS_I(inode)->in_defrag)
152 __btrfs_add_inode_defrag(inode, defrag);
155 spin_unlock(&root->fs_info->defrag_inodes_lock);
160 * must be called with the defrag_inodes lock held
162 struct inode_defrag *btrfs_find_defrag_inode(struct btrfs_fs_info *info, u64 ino,
163 struct rb_node **next)
165 struct inode_defrag *entry = NULL;
167 struct rb_node *parent = NULL;
169 p = info->defrag_inodes.rb_node;
172 entry = rb_entry(parent, struct inode_defrag, rb_node);
174 if (ino < entry->ino)
176 else if (ino > entry->ino)
177 p = parent->rb_right;
183 while (parent && ino > entry->ino) {
184 parent = rb_next(parent);
185 entry = rb_entry(parent, struct inode_defrag, rb_node);
193 * run through the list of inodes in the FS that need
196 int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info)
198 struct inode_defrag *defrag;
199 struct btrfs_root *inode_root;
202 struct btrfs_key key;
203 struct btrfs_ioctl_defrag_range_args range;
206 int defrag_batch = 1024;
208 memset(&range, 0, sizeof(range));
211 atomic_inc(&fs_info->defrag_running);
212 spin_lock(&fs_info->defrag_inodes_lock);
216 /* find an inode to defrag */
217 defrag = btrfs_find_defrag_inode(fs_info, first_ino, &n);
220 defrag = rb_entry(n, struct inode_defrag, rb_node);
221 else if (first_ino) {
229 /* remove it from the rbtree */
230 first_ino = defrag->ino + 1;
231 rb_erase(&defrag->rb_node, &fs_info->defrag_inodes);
233 if (btrfs_fs_closing(fs_info))
236 spin_unlock(&fs_info->defrag_inodes_lock);
239 key.objectid = defrag->root;
240 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
241 key.offset = (u64)-1;
242 inode_root = btrfs_read_fs_root_no_name(fs_info, &key);
243 if (IS_ERR(inode_root))
246 key.objectid = defrag->ino;
247 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
250 inode = btrfs_iget(fs_info->sb, &key, inode_root, NULL);
254 /* do a chunk of defrag */
255 BTRFS_I(inode)->in_defrag = 0;
256 range.start = defrag->last_offset;
257 num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid,
260 * if we filled the whole defrag batch, there
261 * must be more work to do. Queue this defrag
264 if (num_defrag == defrag_batch) {
265 defrag->last_offset = range.start;
266 __btrfs_add_inode_defrag(inode, defrag);
268 * we don't want to kfree defrag, we added it back to
272 } else if (defrag->last_offset && !defrag->cycled) {
274 * we didn't fill our defrag batch, but
275 * we didn't start at zero. Make sure we loop
276 * around to the start of the file.
278 defrag->last_offset = 0;
280 __btrfs_add_inode_defrag(inode, defrag);
286 spin_lock(&fs_info->defrag_inodes_lock);
290 spin_unlock(&fs_info->defrag_inodes_lock);
292 atomic_dec(&fs_info->defrag_running);
295 * during unmount, we use the transaction_wait queue to
296 * wait for the defragger to stop
298 wake_up(&fs_info->transaction_wait);
302 /* simple helper to fault in pages and copy. This should go away
303 * and be replaced with calls into generic code.
305 static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
307 struct page **prepared_pages,
311 size_t total_copied = 0;
313 int offset = pos & (PAGE_CACHE_SIZE - 1);
315 while (write_bytes > 0) {
316 size_t count = min_t(size_t,
317 PAGE_CACHE_SIZE - offset, write_bytes);
318 struct page *page = prepared_pages[pg];
320 * Copy data from userspace to the current page
322 * Disable pagefault to avoid recursive lock since
323 * the pages are already locked
326 copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
329 /* Flush processor's dcache for this page */
330 flush_dcache_page(page);
333 * if we get a partial write, we can end up with
334 * partially up to date pages. These add
335 * a lot of complexity, so make sure they don't
336 * happen by forcing this copy to be retried.
338 * The rest of the btrfs_file_write code will fall
339 * back to page at a time copies after we return 0.
341 if (!PageUptodate(page) && copied < count)
344 iov_iter_advance(i, copied);
345 write_bytes -= copied;
346 total_copied += copied;
348 /* Return to btrfs_file_aio_write to fault page */
349 if (unlikely(copied == 0))
352 if (unlikely(copied < PAGE_CACHE_SIZE - offset)) {
363 * unlocks pages after btrfs_file_write is done with them
365 void btrfs_drop_pages(struct page **pages, size_t num_pages)
368 for (i = 0; i < num_pages; i++) {
369 /* page checked is some magic around finding pages that
370 * have been modified without going through btrfs_set_page_dirty
373 ClearPageChecked(pages[i]);
374 unlock_page(pages[i]);
375 mark_page_accessed(pages[i]);
376 page_cache_release(pages[i]);
381 * after copy_from_user, pages need to be dirtied and we need to make
382 * sure holes are created between the current EOF and the start of
383 * any next extents (if required).
385 * this also makes the decision about creating an inline extent vs
386 * doing real data extents, marking pages dirty and delalloc as required.
388 int btrfs_dirty_pages(struct btrfs_root *root, struct inode *inode,
389 struct page **pages, size_t num_pages,
390 loff_t pos, size_t write_bytes,
391 struct extent_state **cached)
397 u64 end_of_last_block;
398 u64 end_pos = pos + write_bytes;
399 loff_t isize = i_size_read(inode);
401 start_pos = pos & ~((u64)root->sectorsize - 1);
402 num_bytes = (write_bytes + pos - start_pos +
403 root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
405 end_of_last_block = start_pos + num_bytes - 1;
406 err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
411 for (i = 0; i < num_pages; i++) {
412 struct page *p = pages[i];
419 * we've only changed i_size in ram, and we haven't updated
420 * the disk i_size. There is no need to log the inode
424 i_size_write(inode, end_pos);
429 * this drops all the extents in the cache that intersect the range
430 * [start, end]. Existing extents are split as required.
432 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
435 struct extent_map *em;
436 struct extent_map *split = NULL;
437 struct extent_map *split2 = NULL;
438 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
439 u64 len = end - start + 1;
445 WARN_ON(end < start);
446 if (end == (u64)-1) {
452 split = alloc_extent_map();
454 split2 = alloc_extent_map();
455 BUG_ON(!split || !split2);
457 write_lock(&em_tree->lock);
458 em = lookup_extent_mapping(em_tree, start, len);
460 write_unlock(&em_tree->lock);
464 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
465 if (testend && em->start + em->len >= start + len) {
467 write_unlock(&em_tree->lock);
470 start = em->start + em->len;
472 len = start + len - (em->start + em->len);
474 write_unlock(&em_tree->lock);
477 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
478 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
479 remove_extent_mapping(em_tree, em);
481 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
483 split->start = em->start;
484 split->len = start - em->start;
485 split->orig_start = em->orig_start;
486 split->block_start = em->block_start;
489 split->block_len = em->block_len;
491 split->block_len = split->len;
493 split->bdev = em->bdev;
494 split->flags = flags;
495 split->compress_type = em->compress_type;
496 ret = add_extent_mapping(em_tree, split);
498 free_extent_map(split);
502 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
503 testend && em->start + em->len > start + len) {
504 u64 diff = start + len - em->start;
506 split->start = start + len;
507 split->len = em->start + em->len - (start + len);
508 split->bdev = em->bdev;
509 split->flags = flags;
510 split->compress_type = em->compress_type;
513 split->block_len = em->block_len;
514 split->block_start = em->block_start;
515 split->orig_start = em->orig_start;
517 split->block_len = split->len;
518 split->block_start = em->block_start + diff;
519 split->orig_start = split->start;
522 ret = add_extent_mapping(em_tree, split);
524 free_extent_map(split);
527 write_unlock(&em_tree->lock);
531 /* once for the tree*/
535 free_extent_map(split);
537 free_extent_map(split2);
542 * this is very complex, but the basic idea is to drop all extents
543 * in the range start - end. hint_block is filled in with a block number
544 * that would be a good hint to the block allocator for this file.
546 * If an extent intersects the range but is not entirely inside the range
547 * it is either truncated or split. Anything entirely inside the range
548 * is deleted from the tree.
550 int btrfs_drop_extents(struct btrfs_trans_handle *trans, struct inode *inode,
551 u64 start, u64 end, u64 *hint_byte, int drop_cache)
553 struct btrfs_root *root = BTRFS_I(inode)->root;
554 struct extent_buffer *leaf;
555 struct btrfs_file_extent_item *fi;
556 struct btrfs_path *path;
557 struct btrfs_key key;
558 struct btrfs_key new_key;
559 u64 ino = btrfs_ino(inode);
560 u64 search_start = start;
563 u64 extent_offset = 0;
572 btrfs_drop_extent_cache(inode, start, end - 1, 0);
574 path = btrfs_alloc_path();
580 ret = btrfs_lookup_file_extent(trans, root, path, ino,
584 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
585 leaf = path->nodes[0];
586 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
587 if (key.objectid == ino &&
588 key.type == BTRFS_EXTENT_DATA_KEY)
593 leaf = path->nodes[0];
594 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
596 ret = btrfs_next_leaf(root, path);
603 leaf = path->nodes[0];
607 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
609 if (key.objectid > ino)
611 if (WARN_ON_ONCE(key.objectid < ino) ||
612 key.type < BTRFS_EXTENT_DATA_KEY) {
616 if (key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
619 fi = btrfs_item_ptr(leaf, path->slots[0],
620 struct btrfs_file_extent_item);
621 extent_type = btrfs_file_extent_type(leaf, fi);
623 if (extent_type == BTRFS_FILE_EXTENT_REG ||
624 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
625 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
626 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
627 extent_offset = btrfs_file_extent_offset(leaf, fi);
628 extent_end = key.offset +
629 btrfs_file_extent_num_bytes(leaf, fi);
630 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
631 extent_end = key.offset +
632 btrfs_file_extent_inline_len(leaf,
639 if (extent_end <= search_start) {
644 search_start = max(key.offset, start);
646 btrfs_release_path(path);
651 * | - range to drop - |
652 * | -------- extent -------- |
654 if (start > key.offset && end < extent_end) {
656 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
658 memcpy(&new_key, &key, sizeof(new_key));
659 new_key.offset = start;
660 ret = btrfs_duplicate_item(trans, root, path,
662 if (ret == -EAGAIN) {
663 btrfs_release_path(path);
669 leaf = path->nodes[0];
670 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
671 struct btrfs_file_extent_item);
672 btrfs_set_file_extent_num_bytes(leaf, fi,
675 fi = btrfs_item_ptr(leaf, path->slots[0],
676 struct btrfs_file_extent_item);
678 extent_offset += start - key.offset;
679 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
680 btrfs_set_file_extent_num_bytes(leaf, fi,
682 btrfs_mark_buffer_dirty(leaf);
684 if (disk_bytenr > 0) {
685 ret = btrfs_inc_extent_ref(trans, root,
686 disk_bytenr, num_bytes, 0,
687 root->root_key.objectid,
689 start - extent_offset);
691 *hint_byte = disk_bytenr;
696 * | ---- range to drop ----- |
697 * | -------- extent -------- |
699 if (start <= key.offset && end < extent_end) {
700 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
702 memcpy(&new_key, &key, sizeof(new_key));
703 new_key.offset = end;
704 btrfs_set_item_key_safe(trans, root, path, &new_key);
706 extent_offset += end - key.offset;
707 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
708 btrfs_set_file_extent_num_bytes(leaf, fi,
710 btrfs_mark_buffer_dirty(leaf);
711 if (disk_bytenr > 0) {
712 inode_sub_bytes(inode, end - key.offset);
713 *hint_byte = disk_bytenr;
718 search_start = extent_end;
720 * | ---- range to drop ----- |
721 * | -------- extent -------- |
723 if (start > key.offset && end >= extent_end) {
725 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
727 btrfs_set_file_extent_num_bytes(leaf, fi,
729 btrfs_mark_buffer_dirty(leaf);
730 if (disk_bytenr > 0) {
731 inode_sub_bytes(inode, extent_end - start);
732 *hint_byte = disk_bytenr;
734 if (end == extent_end)
742 * | ---- range to drop ----- |
743 * | ------ extent ------ |
745 if (start <= key.offset && end >= extent_end) {
747 del_slot = path->slots[0];
750 BUG_ON(del_slot + del_nr != path->slots[0]);
754 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
755 inode_sub_bytes(inode,
756 extent_end - key.offset);
757 extent_end = ALIGN(extent_end,
759 } else if (disk_bytenr > 0) {
760 ret = btrfs_free_extent(trans, root,
761 disk_bytenr, num_bytes, 0,
762 root->root_key.objectid,
763 key.objectid, key.offset -
766 inode_sub_bytes(inode,
767 extent_end - key.offset);
768 *hint_byte = disk_bytenr;
771 if (end == extent_end)
774 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
779 ret = btrfs_del_items(trans, root, path, del_slot,
786 btrfs_release_path(path);
794 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
798 btrfs_free_path(path);
802 static int extent_mergeable(struct extent_buffer *leaf, int slot,
803 u64 objectid, u64 bytenr, u64 orig_offset,
804 u64 *start, u64 *end)
806 struct btrfs_file_extent_item *fi;
807 struct btrfs_key key;
810 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
813 btrfs_item_key_to_cpu(leaf, &key, slot);
814 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
817 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
818 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
819 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
820 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
821 btrfs_file_extent_compression(leaf, fi) ||
822 btrfs_file_extent_encryption(leaf, fi) ||
823 btrfs_file_extent_other_encoding(leaf, fi))
826 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
827 if ((*start && *start != key.offset) || (*end && *end != extent_end))
836 * Mark extent in the range start - end as written.
838 * This changes extent type from 'pre-allocated' to 'regular'. If only
839 * part of extent is marked as written, the extent will be split into
842 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
843 struct inode *inode, u64 start, u64 end)
845 struct btrfs_root *root = BTRFS_I(inode)->root;
846 struct extent_buffer *leaf;
847 struct btrfs_path *path;
848 struct btrfs_file_extent_item *fi;
849 struct btrfs_key key;
850 struct btrfs_key new_key;
862 u64 ino = btrfs_ino(inode);
864 btrfs_drop_extent_cache(inode, start, end - 1, 0);
866 path = btrfs_alloc_path();
873 key.type = BTRFS_EXTENT_DATA_KEY;
876 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
879 if (ret > 0 && path->slots[0] > 0)
882 leaf = path->nodes[0];
883 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
884 BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY);
885 fi = btrfs_item_ptr(leaf, path->slots[0],
886 struct btrfs_file_extent_item);
887 BUG_ON(btrfs_file_extent_type(leaf, fi) !=
888 BTRFS_FILE_EXTENT_PREALLOC);
889 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
890 BUG_ON(key.offset > start || extent_end < end);
892 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
893 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
894 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
895 memcpy(&new_key, &key, sizeof(new_key));
897 if (start == key.offset && end < extent_end) {
900 if (extent_mergeable(leaf, path->slots[0] - 1,
901 ino, bytenr, orig_offset,
902 &other_start, &other_end)) {
903 new_key.offset = end;
904 btrfs_set_item_key_safe(trans, root, path, &new_key);
905 fi = btrfs_item_ptr(leaf, path->slots[0],
906 struct btrfs_file_extent_item);
907 btrfs_set_file_extent_num_bytes(leaf, fi,
909 btrfs_set_file_extent_offset(leaf, fi,
911 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
912 struct btrfs_file_extent_item);
913 btrfs_set_file_extent_num_bytes(leaf, fi,
915 btrfs_mark_buffer_dirty(leaf);
920 if (start > key.offset && end == extent_end) {
923 if (extent_mergeable(leaf, path->slots[0] + 1,
924 ino, bytenr, orig_offset,
925 &other_start, &other_end)) {
926 fi = btrfs_item_ptr(leaf, path->slots[0],
927 struct btrfs_file_extent_item);
928 btrfs_set_file_extent_num_bytes(leaf, fi,
931 new_key.offset = start;
932 btrfs_set_item_key_safe(trans, root, path, &new_key);
934 fi = btrfs_item_ptr(leaf, path->slots[0],
935 struct btrfs_file_extent_item);
936 btrfs_set_file_extent_num_bytes(leaf, fi,
938 btrfs_set_file_extent_offset(leaf, fi,
939 start - orig_offset);
940 btrfs_mark_buffer_dirty(leaf);
945 while (start > key.offset || end < extent_end) {
946 if (key.offset == start)
949 new_key.offset = split;
950 ret = btrfs_duplicate_item(trans, root, path, &new_key);
951 if (ret == -EAGAIN) {
952 btrfs_release_path(path);
957 leaf = path->nodes[0];
958 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
959 struct btrfs_file_extent_item);
960 btrfs_set_file_extent_num_bytes(leaf, fi,
963 fi = btrfs_item_ptr(leaf, path->slots[0],
964 struct btrfs_file_extent_item);
966 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
967 btrfs_set_file_extent_num_bytes(leaf, fi,
969 btrfs_mark_buffer_dirty(leaf);
971 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
972 root->root_key.objectid,
976 if (split == start) {
979 BUG_ON(start != key.offset);
988 if (extent_mergeable(leaf, path->slots[0] + 1,
989 ino, bytenr, orig_offset,
990 &other_start, &other_end)) {
992 btrfs_release_path(path);
995 extent_end = other_end;
996 del_slot = path->slots[0] + 1;
998 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
999 0, root->root_key.objectid,
1005 if (extent_mergeable(leaf, path->slots[0] - 1,
1006 ino, bytenr, orig_offset,
1007 &other_start, &other_end)) {
1009 btrfs_release_path(path);
1012 key.offset = other_start;
1013 del_slot = path->slots[0];
1015 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1016 0, root->root_key.objectid,
1021 fi = btrfs_item_ptr(leaf, path->slots[0],
1022 struct btrfs_file_extent_item);
1023 btrfs_set_file_extent_type(leaf, fi,
1024 BTRFS_FILE_EXTENT_REG);
1025 btrfs_mark_buffer_dirty(leaf);
1027 fi = btrfs_item_ptr(leaf, del_slot - 1,
1028 struct btrfs_file_extent_item);
1029 btrfs_set_file_extent_type(leaf, fi,
1030 BTRFS_FILE_EXTENT_REG);
1031 btrfs_set_file_extent_num_bytes(leaf, fi,
1032 extent_end - key.offset);
1033 btrfs_mark_buffer_dirty(leaf);
1035 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1039 btrfs_free_path(path);
1044 * on error we return an unlocked page and the error value
1045 * on success we return a locked page and 0
1047 static int prepare_uptodate_page(struct page *page, u64 pos,
1048 bool force_uptodate)
1052 if (((pos & (PAGE_CACHE_SIZE - 1)) || force_uptodate) &&
1053 !PageUptodate(page)) {
1054 ret = btrfs_readpage(NULL, page);
1058 if (!PageUptodate(page)) {
1067 * this gets pages into the page cache and locks them down, it also properly
1068 * waits for data=ordered extents to finish before allowing the pages to be
1071 static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
1072 struct page **pages, size_t num_pages,
1073 loff_t pos, unsigned long first_index,
1074 size_t write_bytes, bool force_uptodate)
1076 struct extent_state *cached_state = NULL;
1078 unsigned long index = pos >> PAGE_CACHE_SHIFT;
1079 struct inode *inode = fdentry(file)->d_inode;
1080 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1086 start_pos = pos & ~((u64)root->sectorsize - 1);
1087 last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
1090 for (i = 0; i < num_pages; i++) {
1091 pages[i] = find_or_create_page(inode->i_mapping, index + i,
1100 err = prepare_uptodate_page(pages[i], pos,
1102 if (i == num_pages - 1)
1103 err = prepare_uptodate_page(pages[i],
1104 pos + write_bytes, false);
1106 page_cache_release(pages[i]);
1110 wait_on_page_writeback(pages[i]);
1113 if (start_pos < inode->i_size) {
1114 struct btrfs_ordered_extent *ordered;
1115 lock_extent_bits(&BTRFS_I(inode)->io_tree,
1116 start_pos, last_pos - 1, 0, &cached_state,
1118 ordered = btrfs_lookup_first_ordered_extent(inode,
1121 ordered->file_offset + ordered->len > start_pos &&
1122 ordered->file_offset < last_pos) {
1123 btrfs_put_ordered_extent(ordered);
1124 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1125 start_pos, last_pos - 1,
1126 &cached_state, GFP_NOFS);
1127 for (i = 0; i < num_pages; i++) {
1128 unlock_page(pages[i]);
1129 page_cache_release(pages[i]);
1131 btrfs_wait_ordered_range(inode, start_pos,
1132 last_pos - start_pos);
1136 btrfs_put_ordered_extent(ordered);
1138 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
1139 last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1140 EXTENT_DO_ACCOUNTING, 0, 0, &cached_state,
1142 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1143 start_pos, last_pos - 1, &cached_state,
1146 for (i = 0; i < num_pages; i++) {
1147 clear_page_dirty_for_io(pages[i]);
1148 set_page_extent_mapped(pages[i]);
1149 WARN_ON(!PageLocked(pages[i]));
1153 while (faili >= 0) {
1154 unlock_page(pages[faili]);
1155 page_cache_release(pages[faili]);
1162 static noinline ssize_t __btrfs_buffered_write(struct file *file,
1166 struct inode *inode = fdentry(file)->d_inode;
1167 struct btrfs_root *root = BTRFS_I(inode)->root;
1168 struct page **pages = NULL;
1169 unsigned long first_index;
1170 size_t num_written = 0;
1173 bool force_page_uptodate = false;
1175 nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) /
1176 PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
1177 (sizeof(struct page *)));
1178 nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1179 nrptrs = max(nrptrs, 8);
1180 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
1184 first_index = pos >> PAGE_CACHE_SHIFT;
1186 while (iov_iter_count(i) > 0) {
1187 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1188 size_t write_bytes = min(iov_iter_count(i),
1189 nrptrs * (size_t)PAGE_CACHE_SIZE -
1191 size_t num_pages = (write_bytes + offset +
1192 PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1196 WARN_ON(num_pages > nrptrs);
1199 * Fault pages before locking them in prepare_pages
1200 * to avoid recursive lock
1202 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1207 ret = btrfs_delalloc_reserve_space(inode,
1208 num_pages << PAGE_CACHE_SHIFT);
1213 * This is going to setup the pages array with the number of
1214 * pages we want, so we don't really need to worry about the
1215 * contents of pages from loop to loop
1217 ret = prepare_pages(root, file, pages, num_pages,
1218 pos, first_index, write_bytes,
1219 force_page_uptodate);
1221 btrfs_delalloc_release_space(inode,
1222 num_pages << PAGE_CACHE_SHIFT);
1226 copied = btrfs_copy_from_user(pos, num_pages,
1227 write_bytes, pages, i);
1230 * if we have trouble faulting in the pages, fall
1231 * back to one page at a time
1233 if (copied < write_bytes)
1237 force_page_uptodate = true;
1240 force_page_uptodate = false;
1241 dirty_pages = (copied + offset +
1242 PAGE_CACHE_SIZE - 1) >>
1247 * If we had a short copy we need to release the excess delaloc
1248 * bytes we reserved. We need to increment outstanding_extents
1249 * because btrfs_delalloc_release_space will decrement it, but
1250 * we still have an outstanding extent for the chunk we actually
1253 if (num_pages > dirty_pages) {
1255 spin_lock(&BTRFS_I(inode)->lock);
1256 BTRFS_I(inode)->outstanding_extents++;
1257 spin_unlock(&BTRFS_I(inode)->lock);
1259 btrfs_delalloc_release_space(inode,
1260 (num_pages - dirty_pages) <<
1265 ret = btrfs_dirty_pages(root, inode, pages,
1266 dirty_pages, pos, copied,
1269 btrfs_delalloc_release_space(inode,
1270 dirty_pages << PAGE_CACHE_SHIFT);
1271 btrfs_drop_pages(pages, num_pages);
1276 btrfs_drop_pages(pages, num_pages);
1280 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
1282 if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1283 btrfs_btree_balance_dirty(root, 1);
1284 btrfs_throttle(root);
1287 num_written += copied;
1292 return num_written ? num_written : ret;
1295 static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1296 const struct iovec *iov,
1297 unsigned long nr_segs, loff_t pos,
1298 loff_t *ppos, size_t count, size_t ocount)
1300 struct file *file = iocb->ki_filp;
1301 struct inode *inode = fdentry(file)->d_inode;
1304 ssize_t written_buffered;
1308 written = generic_file_direct_write(iocb, iov, &nr_segs, pos, ppos,
1312 * the generic O_DIRECT will update in-memory i_size after the
1313 * DIOs are done. But our endio handlers that update the on
1314 * disk i_size never update past the in memory i_size. So we
1315 * need one more update here to catch any additions to the
1318 if (inode->i_size != BTRFS_I(inode)->disk_i_size) {
1319 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
1320 mark_inode_dirty(inode);
1323 if (written < 0 || written == count)
1328 iov_iter_init(&i, iov, nr_segs, count, written);
1329 written_buffered = __btrfs_buffered_write(file, &i, pos);
1330 if (written_buffered < 0) {
1331 err = written_buffered;
1334 endbyte = pos + written_buffered - 1;
1335 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
1338 written += written_buffered;
1339 *ppos = pos + written_buffered;
1340 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
1341 endbyte >> PAGE_CACHE_SHIFT);
1343 return written ? written : err;
1346 static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
1347 const struct iovec *iov,
1348 unsigned long nr_segs, loff_t pos)
1350 struct file *file = iocb->ki_filp;
1351 struct inode *inode = fdentry(file)->d_inode;
1352 struct btrfs_root *root = BTRFS_I(inode)->root;
1353 loff_t *ppos = &iocb->ki_pos;
1355 ssize_t num_written = 0;
1357 size_t count, ocount;
1359 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1361 mutex_lock(&inode->i_mutex);
1363 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
1365 mutex_unlock(&inode->i_mutex);
1370 current->backing_dev_info = inode->i_mapping->backing_dev_info;
1371 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1373 mutex_unlock(&inode->i_mutex);
1378 mutex_unlock(&inode->i_mutex);
1382 err = file_remove_suid(file);
1384 mutex_unlock(&inode->i_mutex);
1389 * If BTRFS flips readonly due to some impossible error
1390 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1391 * although we have opened a file as writable, we have
1392 * to stop this write operation to ensure FS consistency.
1394 if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
1395 mutex_unlock(&inode->i_mutex);
1400 err = btrfs_update_time(file);
1402 mutex_unlock(&inode->i_mutex);
1405 BTRFS_I(inode)->sequence++;
1407 start_pos = round_down(pos, root->sectorsize);
1408 if (start_pos > i_size_read(inode)) {
1409 err = btrfs_cont_expand(inode, i_size_read(inode), start_pos);
1411 mutex_unlock(&inode->i_mutex);
1416 if (unlikely(file->f_flags & O_DIRECT)) {
1417 num_written = __btrfs_direct_write(iocb, iov, nr_segs,
1418 pos, ppos, count, ocount);
1422 iov_iter_init(&i, iov, nr_segs, count, num_written);
1424 num_written = __btrfs_buffered_write(file, &i, pos);
1425 if (num_written > 0)
1426 *ppos = pos + num_written;
1429 mutex_unlock(&inode->i_mutex);
1432 * we want to make sure fsync finds this change
1433 * but we haven't joined a transaction running right now.
1435 * Later on, someone is sure to update the inode and get the
1436 * real transid recorded.
1438 * We set last_trans now to the fs_info generation + 1,
1439 * this will either be one more than the running transaction
1440 * or the generation used for the next transaction if there isn't
1441 * one running right now.
1443 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1444 if (num_written > 0 || num_written == -EIOCBQUEUED) {
1445 err = generic_write_sync(file, pos, num_written);
1446 if (err < 0 && num_written > 0)
1450 current->backing_dev_info = NULL;
1451 return num_written ? num_written : err;
1454 int btrfs_release_file(struct inode *inode, struct file *filp)
1457 * ordered_data_close is set by settattr when we are about to truncate
1458 * a file from a non-zero size to a zero size. This tries to
1459 * flush down new bytes that may have been written if the
1460 * application were using truncate to replace a file in place.
1462 if (BTRFS_I(inode)->ordered_data_close) {
1463 BTRFS_I(inode)->ordered_data_close = 0;
1464 btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode);
1465 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1466 filemap_flush(inode->i_mapping);
1468 if (filp->private_data)
1469 btrfs_ioctl_trans_end(filp);
1474 * fsync call for both files and directories. This logs the inode into
1475 * the tree log instead of forcing full commits whenever possible.
1477 * It needs to call filemap_fdatawait so that all ordered extent updates are
1478 * in the metadata btree are up to date for copying to the log.
1480 * It drops the inode mutex before doing the tree log commit. This is an
1481 * important optimization for directories because holding the mutex prevents
1482 * new operations on the dir while we write to disk.
1484 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1486 struct dentry *dentry = file->f_path.dentry;
1487 struct inode *inode = dentry->d_inode;
1488 struct btrfs_root *root = BTRFS_I(inode)->root;
1490 struct btrfs_trans_handle *trans;
1492 trace_btrfs_sync_file(file, datasync);
1494 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
1497 mutex_lock(&inode->i_mutex);
1499 /* we wait first, since the writeback may change the inode */
1501 btrfs_wait_ordered_range(inode, 0, (u64)-1);
1505 * check the transaction that last modified this inode
1506 * and see if its already been committed
1508 if (!BTRFS_I(inode)->last_trans) {
1509 mutex_unlock(&inode->i_mutex);
1514 * if the last transaction that changed this file was before
1515 * the current transaction, we can bail out now without any
1519 if (BTRFS_I(inode)->last_trans <=
1520 root->fs_info->last_trans_committed) {
1521 BTRFS_I(inode)->last_trans = 0;
1522 mutex_unlock(&inode->i_mutex);
1527 * ok we haven't committed the transaction yet, lets do a commit
1529 if (file->private_data)
1530 btrfs_ioctl_trans_end(file);
1532 trans = btrfs_start_transaction(root, 0);
1533 if (IS_ERR(trans)) {
1534 ret = PTR_ERR(trans);
1535 mutex_unlock(&inode->i_mutex);
1539 ret = btrfs_log_dentry_safe(trans, root, dentry);
1541 mutex_unlock(&inode->i_mutex);
1545 /* we've logged all the items and now have a consistent
1546 * version of the file in the log. It is possible that
1547 * someone will come in and modify the file, but that's
1548 * fine because the log is consistent on disk, and we
1549 * have references to all of the file's extents
1551 * It is possible that someone will come in and log the
1552 * file again, but that will end up using the synchronization
1553 * inside btrfs_sync_log to keep things safe.
1555 mutex_unlock(&inode->i_mutex);
1557 if (ret != BTRFS_NO_LOG_SYNC) {
1559 ret = btrfs_commit_transaction(trans, root);
1561 ret = btrfs_sync_log(trans, root);
1563 ret = btrfs_end_transaction(trans, root);
1565 ret = btrfs_commit_transaction(trans, root);
1568 ret = btrfs_end_transaction(trans, root);
1571 return ret > 0 ? -EIO : ret;
1574 static const struct vm_operations_struct btrfs_file_vm_ops = {
1575 .fault = filemap_fault,
1576 .page_mkwrite = btrfs_page_mkwrite,
1579 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
1581 struct address_space *mapping = filp->f_mapping;
1583 if (!mapping->a_ops->readpage)
1586 file_accessed(filp);
1587 vma->vm_ops = &btrfs_file_vm_ops;
1588 vma->vm_flags |= VM_CAN_NONLINEAR;
1593 static long btrfs_fallocate(struct file *file, int mode,
1594 loff_t offset, loff_t len)
1596 struct inode *inode = file->f_path.dentry->d_inode;
1597 struct extent_state *cached_state = NULL;
1604 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
1605 struct extent_map *em;
1608 alloc_start = offset & ~mask;
1609 alloc_end = (offset + len + mask) & ~mask;
1611 /* We only support the FALLOC_FL_KEEP_SIZE mode */
1612 if (mode & ~FALLOC_FL_KEEP_SIZE)
1616 * wait for ordered IO before we have any locks. We'll loop again
1617 * below with the locks held.
1619 btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
1621 mutex_lock(&inode->i_mutex);
1622 ret = inode_newsize_ok(inode, alloc_end);
1626 if (alloc_start > inode->i_size) {
1627 ret = btrfs_cont_expand(inode, i_size_read(inode),
1633 locked_end = alloc_end - 1;
1635 struct btrfs_ordered_extent *ordered;
1637 /* the extent lock is ordered inside the running
1640 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
1641 locked_end, 0, &cached_state, GFP_NOFS);
1642 ordered = btrfs_lookup_first_ordered_extent(inode,
1645 ordered->file_offset + ordered->len > alloc_start &&
1646 ordered->file_offset < alloc_end) {
1647 btrfs_put_ordered_extent(ordered);
1648 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1649 alloc_start, locked_end,
1650 &cached_state, GFP_NOFS);
1652 * we can't wait on the range with the transaction
1653 * running or with the extent lock held
1655 btrfs_wait_ordered_range(inode, alloc_start,
1656 alloc_end - alloc_start);
1659 btrfs_put_ordered_extent(ordered);
1664 cur_offset = alloc_start;
1668 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
1669 alloc_end - cur_offset, 0);
1670 BUG_ON(IS_ERR_OR_NULL(em));
1671 last_byte = min(extent_map_end(em), alloc_end);
1672 actual_end = min_t(u64, extent_map_end(em), offset + len);
1673 last_byte = (last_byte + mask) & ~mask;
1675 if (em->block_start == EXTENT_MAP_HOLE ||
1676 (cur_offset >= inode->i_size &&
1677 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
1680 * Make sure we have enough space before we do the
1683 ret = btrfs_check_data_free_space(inode, last_byte -
1686 free_extent_map(em);
1690 ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
1691 last_byte - cur_offset,
1692 1 << inode->i_blkbits,
1696 /* Let go of our reservation. */
1697 btrfs_free_reserved_data_space(inode, last_byte -
1700 free_extent_map(em);
1703 } else if (actual_end > inode->i_size &&
1704 !(mode & FALLOC_FL_KEEP_SIZE)) {
1706 * We didn't need to allocate any more space, but we
1707 * still extended the size of the file so we need to
1710 inode->i_ctime = CURRENT_TIME;
1711 i_size_write(inode, actual_end);
1712 btrfs_ordered_update_i_size(inode, actual_end, NULL);
1714 free_extent_map(em);
1716 cur_offset = last_byte;
1717 if (cur_offset >= alloc_end) {
1722 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
1723 &cached_state, GFP_NOFS);
1725 mutex_unlock(&inode->i_mutex);
1729 static int find_desired_extent(struct inode *inode, loff_t *offset, int origin)
1731 struct btrfs_root *root = BTRFS_I(inode)->root;
1732 struct extent_map *em;
1733 struct extent_state *cached_state = NULL;
1734 u64 lockstart = *offset;
1735 u64 lockend = i_size_read(inode);
1736 u64 start = *offset;
1737 u64 orig_start = *offset;
1738 u64 len = i_size_read(inode);
1742 lockend = max_t(u64, root->sectorsize, lockend);
1743 if (lockend <= lockstart)
1744 lockend = lockstart + root->sectorsize;
1746 len = lockend - lockstart + 1;
1748 len = max_t(u64, len, root->sectorsize);
1749 if (inode->i_size == 0)
1752 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, 0,
1753 &cached_state, GFP_NOFS);
1756 * Delalloc is such a pain. If we have a hole and we have pending
1757 * delalloc for a portion of the hole we will get back a hole that
1758 * exists for the entire range since it hasn't been actually written
1759 * yet. So to take care of this case we need to look for an extent just
1760 * before the position we want in case there is outstanding delalloc
1763 if (origin == SEEK_HOLE && start != 0) {
1764 if (start <= root->sectorsize)
1765 em = btrfs_get_extent_fiemap(inode, NULL, 0, 0,
1766 root->sectorsize, 0);
1768 em = btrfs_get_extent_fiemap(inode, NULL, 0,
1769 start - root->sectorsize,
1770 root->sectorsize, 0);
1775 last_end = em->start + em->len;
1776 if (em->block_start == EXTENT_MAP_DELALLOC)
1777 last_end = min_t(u64, last_end, inode->i_size);
1778 free_extent_map(em);
1782 em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0);
1788 if (em->block_start == EXTENT_MAP_HOLE) {
1789 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
1790 if (last_end <= orig_start) {
1791 free_extent_map(em);
1797 if (origin == SEEK_HOLE) {
1799 free_extent_map(em);
1803 if (origin == SEEK_DATA) {
1804 if (em->block_start == EXTENT_MAP_DELALLOC) {
1805 if (start >= inode->i_size) {
1806 free_extent_map(em);
1813 free_extent_map(em);
1818 start = em->start + em->len;
1819 last_end = em->start + em->len;
1821 if (em->block_start == EXTENT_MAP_DELALLOC)
1822 last_end = min_t(u64, last_end, inode->i_size);
1824 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
1825 free_extent_map(em);
1829 free_extent_map(em);
1833 *offset = min(*offset, inode->i_size);
1835 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
1836 &cached_state, GFP_NOFS);
1840 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int origin)
1842 struct inode *inode = file->f_mapping->host;
1845 mutex_lock(&inode->i_mutex);
1849 offset = generic_file_llseek(file, offset, origin);
1853 if (offset >= i_size_read(inode)) {
1854 mutex_unlock(&inode->i_mutex);
1858 ret = find_desired_extent(inode, &offset, origin);
1860 mutex_unlock(&inode->i_mutex);
1865 if (offset < 0 && !(file->f_mode & FMODE_UNSIGNED_OFFSET)) {
1869 if (offset > inode->i_sb->s_maxbytes) {
1874 /* Special lock needed here? */
1875 if (offset != file->f_pos) {
1876 file->f_pos = offset;
1877 file->f_version = 0;
1880 mutex_unlock(&inode->i_mutex);
1884 const struct file_operations btrfs_file_operations = {
1885 .llseek = btrfs_file_llseek,
1886 .read = do_sync_read,
1887 .write = do_sync_write,
1888 .aio_read = generic_file_aio_read,
1889 .splice_read = generic_file_splice_read,
1890 .aio_write = btrfs_file_aio_write,
1891 .mmap = btrfs_file_mmap,
1892 .open = generic_file_open,
1893 .release = btrfs_release_file,
1894 .fsync = btrfs_sync_file,
1895 .fallocate = btrfs_fallocate,
1896 .unlocked_ioctl = btrfs_ioctl,
1897 #ifdef CONFIG_COMPAT
1898 .compat_ioctl = btrfs_compat_ioctl,