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 int __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;
129 if (!btrfs_test_opt(root, AUTO_DEFRAG))
132 if (btrfs_fs_closing(root->fs_info))
135 if (BTRFS_I(inode)->in_defrag)
139 transid = trans->transid;
141 transid = BTRFS_I(inode)->root->last_trans;
143 defrag = kzalloc(sizeof(*defrag), GFP_NOFS);
147 defrag->ino = btrfs_ino(inode);
148 defrag->transid = transid;
149 defrag->root = root->root_key.objectid;
151 spin_lock(&root->fs_info->defrag_inodes_lock);
152 if (!BTRFS_I(inode)->in_defrag)
153 ret = __btrfs_add_inode_defrag(inode, defrag);
154 spin_unlock(&root->fs_info->defrag_inodes_lock);
159 * must be called with the defrag_inodes lock held
161 struct inode_defrag *btrfs_find_defrag_inode(struct btrfs_fs_info *info, u64 ino,
162 struct rb_node **next)
164 struct inode_defrag *entry = NULL;
166 struct rb_node *parent = NULL;
168 p = info->defrag_inodes.rb_node;
171 entry = rb_entry(parent, struct inode_defrag, rb_node);
173 if (ino < entry->ino)
175 else if (ino > entry->ino)
176 p = parent->rb_right;
182 while (parent && ino > entry->ino) {
183 parent = rb_next(parent);
184 entry = rb_entry(parent, struct inode_defrag, rb_node);
192 * run through the list of inodes in the FS that need
195 int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info)
197 struct inode_defrag *defrag;
198 struct btrfs_root *inode_root;
201 struct btrfs_key key;
202 struct btrfs_ioctl_defrag_range_args range;
205 int defrag_batch = 1024;
207 memset(&range, 0, sizeof(range));
210 atomic_inc(&fs_info->defrag_running);
211 spin_lock(&fs_info->defrag_inodes_lock);
215 /* find an inode to defrag */
216 defrag = btrfs_find_defrag_inode(fs_info, first_ino, &n);
219 defrag = rb_entry(n, struct inode_defrag, rb_node);
220 else if (first_ino) {
228 /* remove it from the rbtree */
229 first_ino = defrag->ino + 1;
230 rb_erase(&defrag->rb_node, &fs_info->defrag_inodes);
232 if (btrfs_fs_closing(fs_info))
235 spin_unlock(&fs_info->defrag_inodes_lock);
238 key.objectid = defrag->root;
239 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
240 key.offset = (u64)-1;
241 inode_root = btrfs_read_fs_root_no_name(fs_info, &key);
242 if (IS_ERR(inode_root))
245 key.objectid = defrag->ino;
246 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
249 inode = btrfs_iget(fs_info->sb, &key, inode_root, NULL);
253 /* do a chunk of defrag */
254 BTRFS_I(inode)->in_defrag = 0;
255 range.start = defrag->last_offset;
256 num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid,
259 * if we filled the whole defrag batch, there
260 * must be more work to do. Queue this defrag
263 if (num_defrag == defrag_batch) {
264 defrag->last_offset = range.start;
265 __btrfs_add_inode_defrag(inode, defrag);
267 * we don't want to kfree defrag, we added it back to
271 } else if (defrag->last_offset && !defrag->cycled) {
273 * we didn't fill our defrag batch, but
274 * we didn't start at zero. Make sure we loop
275 * around to the start of the file.
277 defrag->last_offset = 0;
279 __btrfs_add_inode_defrag(inode, defrag);
285 spin_lock(&fs_info->defrag_inodes_lock);
289 spin_unlock(&fs_info->defrag_inodes_lock);
291 atomic_dec(&fs_info->defrag_running);
294 * during unmount, we use the transaction_wait queue to
295 * wait for the defragger to stop
297 wake_up(&fs_info->transaction_wait);
301 /* simple helper to fault in pages and copy. This should go away
302 * and be replaced with calls into generic code.
304 static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
306 struct page **prepared_pages,
310 size_t total_copied = 0;
312 int offset = pos & (PAGE_CACHE_SIZE - 1);
314 while (write_bytes > 0) {
315 size_t count = min_t(size_t,
316 PAGE_CACHE_SIZE - offset, write_bytes);
317 struct page *page = prepared_pages[pg];
319 * Copy data from userspace to the current page
321 * Disable pagefault to avoid recursive lock since
322 * the pages are already locked
325 copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
328 /* Flush processor's dcache for this page */
329 flush_dcache_page(page);
332 * if we get a partial write, we can end up with
333 * partially up to date pages. These add
334 * a lot of complexity, so make sure they don't
335 * happen by forcing this copy to be retried.
337 * The rest of the btrfs_file_write code will fall
338 * back to page at a time copies after we return 0.
340 if (!PageUptodate(page) && copied < count)
343 iov_iter_advance(i, copied);
344 write_bytes -= copied;
345 total_copied += copied;
347 /* Return to btrfs_file_aio_write to fault page */
348 if (unlikely(copied == 0))
351 if (unlikely(copied < PAGE_CACHE_SIZE - offset)) {
362 * unlocks pages after btrfs_file_write is done with them
364 void btrfs_drop_pages(struct page **pages, size_t num_pages)
367 for (i = 0; i < num_pages; i++) {
368 /* page checked is some magic around finding pages that
369 * have been modified without going through btrfs_set_page_dirty
372 ClearPageChecked(pages[i]);
373 unlock_page(pages[i]);
374 mark_page_accessed(pages[i]);
375 page_cache_release(pages[i]);
380 * after copy_from_user, pages need to be dirtied and we need to make
381 * sure holes are created between the current EOF and the start of
382 * any next extents (if required).
384 * this also makes the decision about creating an inline extent vs
385 * doing real data extents, marking pages dirty and delalloc as required.
387 int btrfs_dirty_pages(struct btrfs_root *root, struct inode *inode,
388 struct page **pages, size_t num_pages,
389 loff_t pos, size_t write_bytes,
390 struct extent_state **cached)
396 u64 end_of_last_block;
397 u64 end_pos = pos + write_bytes;
398 loff_t isize = i_size_read(inode);
400 start_pos = pos & ~((u64)root->sectorsize - 1);
401 num_bytes = (write_bytes + pos - start_pos +
402 root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
404 end_of_last_block = start_pos + num_bytes - 1;
405 err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
410 for (i = 0; i < num_pages; i++) {
411 struct page *p = pages[i];
418 * we've only changed i_size in ram, and we haven't updated
419 * the disk i_size. There is no need to log the inode
423 i_size_write(inode, end_pos);
428 * this drops all the extents in the cache that intersect the range
429 * [start, end]. Existing extents are split as required.
431 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
434 struct extent_map *em;
435 struct extent_map *split = NULL;
436 struct extent_map *split2 = NULL;
437 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
438 u64 len = end - start + 1;
444 WARN_ON(end < start);
445 if (end == (u64)-1) {
451 split = alloc_extent_map();
453 split2 = alloc_extent_map();
454 BUG_ON(!split || !split2);
456 write_lock(&em_tree->lock);
457 em = lookup_extent_mapping(em_tree, start, len);
459 write_unlock(&em_tree->lock);
463 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
464 if (testend && em->start + em->len >= start + len) {
466 write_unlock(&em_tree->lock);
469 start = em->start + em->len;
471 len = start + len - (em->start + em->len);
473 write_unlock(&em_tree->lock);
476 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
477 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
478 remove_extent_mapping(em_tree, em);
480 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
482 split->start = em->start;
483 split->len = start - em->start;
484 split->orig_start = em->orig_start;
485 split->block_start = em->block_start;
488 split->block_len = em->block_len;
490 split->block_len = split->len;
492 split->bdev = em->bdev;
493 split->flags = flags;
494 split->compress_type = em->compress_type;
495 ret = add_extent_mapping(em_tree, split);
497 free_extent_map(split);
501 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
502 testend && em->start + em->len > start + len) {
503 u64 diff = start + len - em->start;
505 split->start = start + len;
506 split->len = em->start + em->len - (start + len);
507 split->bdev = em->bdev;
508 split->flags = flags;
509 split->compress_type = em->compress_type;
512 split->block_len = em->block_len;
513 split->block_start = em->block_start;
514 split->orig_start = em->orig_start;
516 split->block_len = split->len;
517 split->block_start = em->block_start + diff;
518 split->orig_start = split->start;
521 ret = add_extent_mapping(em_tree, split);
523 free_extent_map(split);
526 write_unlock(&em_tree->lock);
530 /* once for the tree*/
534 free_extent_map(split);
536 free_extent_map(split2);
541 * this is very complex, but the basic idea is to drop all extents
542 * in the range start - end. hint_block is filled in with a block number
543 * that would be a good hint to the block allocator for this file.
545 * If an extent intersects the range but is not entirely inside the range
546 * it is either truncated or split. Anything entirely inside the range
547 * is deleted from the tree.
549 int btrfs_drop_extents(struct btrfs_trans_handle *trans, struct inode *inode,
550 u64 start, u64 end, u64 *hint_byte, int drop_cache)
552 struct btrfs_root *root = BTRFS_I(inode)->root;
553 struct extent_buffer *leaf;
554 struct btrfs_file_extent_item *fi;
555 struct btrfs_path *path;
556 struct btrfs_key key;
557 struct btrfs_key new_key;
558 u64 ino = btrfs_ino(inode);
559 u64 search_start = start;
562 u64 extent_offset = 0;
571 btrfs_drop_extent_cache(inode, start, end - 1, 0);
573 path = btrfs_alloc_path();
579 ret = btrfs_lookup_file_extent(trans, root, path, ino,
583 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
584 leaf = path->nodes[0];
585 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
586 if (key.objectid == ino &&
587 key.type == BTRFS_EXTENT_DATA_KEY)
592 leaf = path->nodes[0];
593 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
595 ret = btrfs_next_leaf(root, path);
602 leaf = path->nodes[0];
606 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
607 if (key.objectid > ino ||
608 key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
611 fi = btrfs_item_ptr(leaf, path->slots[0],
612 struct btrfs_file_extent_item);
613 extent_type = btrfs_file_extent_type(leaf, fi);
615 if (extent_type == BTRFS_FILE_EXTENT_REG ||
616 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
617 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
618 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
619 extent_offset = btrfs_file_extent_offset(leaf, fi);
620 extent_end = key.offset +
621 btrfs_file_extent_num_bytes(leaf, fi);
622 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
623 extent_end = key.offset +
624 btrfs_file_extent_inline_len(leaf, fi);
627 extent_end = search_start;
630 if (extent_end <= search_start) {
635 search_start = max(key.offset, start);
637 btrfs_release_path(path);
642 * | - range to drop - |
643 * | -------- extent -------- |
645 if (start > key.offset && end < extent_end) {
647 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
649 memcpy(&new_key, &key, sizeof(new_key));
650 new_key.offset = start;
651 ret = btrfs_duplicate_item(trans, root, path,
653 if (ret == -EAGAIN) {
654 btrfs_release_path(path);
660 leaf = path->nodes[0];
661 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
662 struct btrfs_file_extent_item);
663 btrfs_set_file_extent_num_bytes(leaf, fi,
666 fi = btrfs_item_ptr(leaf, path->slots[0],
667 struct btrfs_file_extent_item);
669 extent_offset += start - key.offset;
670 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
671 btrfs_set_file_extent_num_bytes(leaf, fi,
673 btrfs_mark_buffer_dirty(leaf);
675 if (disk_bytenr > 0) {
676 ret = btrfs_inc_extent_ref(trans, root,
677 disk_bytenr, num_bytes, 0,
678 root->root_key.objectid,
680 start - extent_offset);
682 *hint_byte = disk_bytenr;
687 * | ---- range to drop ----- |
688 * | -------- extent -------- |
690 if (start <= key.offset && end < extent_end) {
691 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
693 memcpy(&new_key, &key, sizeof(new_key));
694 new_key.offset = end;
695 btrfs_set_item_key_safe(trans, root, path, &new_key);
697 extent_offset += end - key.offset;
698 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
699 btrfs_set_file_extent_num_bytes(leaf, fi,
701 btrfs_mark_buffer_dirty(leaf);
702 if (disk_bytenr > 0) {
703 inode_sub_bytes(inode, end - key.offset);
704 *hint_byte = disk_bytenr;
709 search_start = extent_end;
711 * | ---- range to drop ----- |
712 * | -------- extent -------- |
714 if (start > key.offset && end >= extent_end) {
716 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
718 btrfs_set_file_extent_num_bytes(leaf, fi,
720 btrfs_mark_buffer_dirty(leaf);
721 if (disk_bytenr > 0) {
722 inode_sub_bytes(inode, extent_end - start);
723 *hint_byte = disk_bytenr;
725 if (end == extent_end)
733 * | ---- range to drop ----- |
734 * | ------ extent ------ |
736 if (start <= key.offset && end >= extent_end) {
738 del_slot = path->slots[0];
741 BUG_ON(del_slot + del_nr != path->slots[0]);
745 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
746 inode_sub_bytes(inode,
747 extent_end - key.offset);
748 extent_end = ALIGN(extent_end,
750 } else if (disk_bytenr > 0) {
751 ret = btrfs_free_extent(trans, root,
752 disk_bytenr, num_bytes, 0,
753 root->root_key.objectid,
754 key.objectid, key.offset -
757 inode_sub_bytes(inode,
758 extent_end - key.offset);
759 *hint_byte = disk_bytenr;
762 if (end == extent_end)
765 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
770 ret = btrfs_del_items(trans, root, path, del_slot,
777 btrfs_release_path(path);
785 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
789 btrfs_free_path(path);
793 static int extent_mergeable(struct extent_buffer *leaf, int slot,
794 u64 objectid, u64 bytenr, u64 orig_offset,
795 u64 *start, u64 *end)
797 struct btrfs_file_extent_item *fi;
798 struct btrfs_key key;
801 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
804 btrfs_item_key_to_cpu(leaf, &key, slot);
805 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
808 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
809 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
810 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
811 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
812 btrfs_file_extent_compression(leaf, fi) ||
813 btrfs_file_extent_encryption(leaf, fi) ||
814 btrfs_file_extent_other_encoding(leaf, fi))
817 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
818 if ((*start && *start != key.offset) || (*end && *end != extent_end))
827 * Mark extent in the range start - end as written.
829 * This changes extent type from 'pre-allocated' to 'regular'. If only
830 * part of extent is marked as written, the extent will be split into
833 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
834 struct inode *inode, u64 start, u64 end)
836 struct btrfs_root *root = BTRFS_I(inode)->root;
837 struct extent_buffer *leaf;
838 struct btrfs_path *path;
839 struct btrfs_file_extent_item *fi;
840 struct btrfs_key key;
841 struct btrfs_key new_key;
853 u64 ino = btrfs_ino(inode);
855 btrfs_drop_extent_cache(inode, start, end - 1, 0);
857 path = btrfs_alloc_path();
864 key.type = BTRFS_EXTENT_DATA_KEY;
867 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
870 if (ret > 0 && path->slots[0] > 0)
873 leaf = path->nodes[0];
874 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
875 BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY);
876 fi = btrfs_item_ptr(leaf, path->slots[0],
877 struct btrfs_file_extent_item);
878 BUG_ON(btrfs_file_extent_type(leaf, fi) !=
879 BTRFS_FILE_EXTENT_PREALLOC);
880 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
881 BUG_ON(key.offset > start || extent_end < end);
883 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
884 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
885 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
886 memcpy(&new_key, &key, sizeof(new_key));
888 if (start == key.offset && end < extent_end) {
891 if (extent_mergeable(leaf, path->slots[0] - 1,
892 ino, bytenr, orig_offset,
893 &other_start, &other_end)) {
894 new_key.offset = end;
895 btrfs_set_item_key_safe(trans, root, path, &new_key);
896 fi = btrfs_item_ptr(leaf, path->slots[0],
897 struct btrfs_file_extent_item);
898 btrfs_set_file_extent_num_bytes(leaf, fi,
900 btrfs_set_file_extent_offset(leaf, fi,
902 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
903 struct btrfs_file_extent_item);
904 btrfs_set_file_extent_num_bytes(leaf, fi,
906 btrfs_mark_buffer_dirty(leaf);
911 if (start > key.offset && end == extent_end) {
914 if (extent_mergeable(leaf, path->slots[0] + 1,
915 ino, bytenr, orig_offset,
916 &other_start, &other_end)) {
917 fi = btrfs_item_ptr(leaf, path->slots[0],
918 struct btrfs_file_extent_item);
919 btrfs_set_file_extent_num_bytes(leaf, fi,
922 new_key.offset = start;
923 btrfs_set_item_key_safe(trans, root, path, &new_key);
925 fi = btrfs_item_ptr(leaf, path->slots[0],
926 struct btrfs_file_extent_item);
927 btrfs_set_file_extent_num_bytes(leaf, fi,
929 btrfs_set_file_extent_offset(leaf, fi,
930 start - orig_offset);
931 btrfs_mark_buffer_dirty(leaf);
936 while (start > key.offset || end < extent_end) {
937 if (key.offset == start)
940 new_key.offset = split;
941 ret = btrfs_duplicate_item(trans, root, path, &new_key);
942 if (ret == -EAGAIN) {
943 btrfs_release_path(path);
948 leaf = path->nodes[0];
949 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
950 struct btrfs_file_extent_item);
951 btrfs_set_file_extent_num_bytes(leaf, fi,
954 fi = btrfs_item_ptr(leaf, path->slots[0],
955 struct btrfs_file_extent_item);
957 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
958 btrfs_set_file_extent_num_bytes(leaf, fi,
960 btrfs_mark_buffer_dirty(leaf);
962 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
963 root->root_key.objectid,
967 if (split == start) {
970 BUG_ON(start != key.offset);
979 if (extent_mergeable(leaf, path->slots[0] + 1,
980 ino, bytenr, orig_offset,
981 &other_start, &other_end)) {
983 btrfs_release_path(path);
986 extent_end = other_end;
987 del_slot = path->slots[0] + 1;
989 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
990 0, root->root_key.objectid,
996 if (extent_mergeable(leaf, path->slots[0] - 1,
997 ino, bytenr, orig_offset,
998 &other_start, &other_end)) {
1000 btrfs_release_path(path);
1003 key.offset = other_start;
1004 del_slot = path->slots[0];
1006 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1007 0, root->root_key.objectid,
1012 fi = btrfs_item_ptr(leaf, path->slots[0],
1013 struct btrfs_file_extent_item);
1014 btrfs_set_file_extent_type(leaf, fi,
1015 BTRFS_FILE_EXTENT_REG);
1016 btrfs_mark_buffer_dirty(leaf);
1018 fi = btrfs_item_ptr(leaf, del_slot - 1,
1019 struct btrfs_file_extent_item);
1020 btrfs_set_file_extent_type(leaf, fi,
1021 BTRFS_FILE_EXTENT_REG);
1022 btrfs_set_file_extent_num_bytes(leaf, fi,
1023 extent_end - key.offset);
1024 btrfs_mark_buffer_dirty(leaf);
1026 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1030 btrfs_free_path(path);
1035 * on error we return an unlocked page and the error value
1036 * on success we return a locked page and 0
1038 static int prepare_uptodate_page(struct page *page, u64 pos)
1042 if ((pos & (PAGE_CACHE_SIZE - 1)) && !PageUptodate(page)) {
1043 ret = btrfs_readpage(NULL, page);
1047 if (!PageUptodate(page)) {
1056 * this gets pages into the page cache and locks them down, it also properly
1057 * waits for data=ordered extents to finish before allowing the pages to be
1060 static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
1061 struct page **pages, size_t num_pages,
1062 loff_t pos, unsigned long first_index,
1063 unsigned long last_index, size_t write_bytes)
1065 struct extent_state *cached_state = NULL;
1067 unsigned long index = pos >> PAGE_CACHE_SHIFT;
1068 struct inode *inode = fdentry(file)->d_inode;
1074 start_pos = pos & ~((u64)root->sectorsize - 1);
1075 last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
1077 if (start_pos > inode->i_size) {
1078 err = btrfs_cont_expand(inode, i_size_read(inode), start_pos);
1084 for (i = 0; i < num_pages; i++) {
1085 pages[i] = find_or_create_page(inode->i_mapping, index + i,
1094 err = prepare_uptodate_page(pages[i], pos);
1095 if (i == num_pages - 1)
1096 err = prepare_uptodate_page(pages[i],
1099 page_cache_release(pages[i]);
1103 wait_on_page_writeback(pages[i]);
1106 if (start_pos < inode->i_size) {
1107 struct btrfs_ordered_extent *ordered;
1108 lock_extent_bits(&BTRFS_I(inode)->io_tree,
1109 start_pos, last_pos - 1, 0, &cached_state,
1111 ordered = btrfs_lookup_first_ordered_extent(inode,
1114 ordered->file_offset + ordered->len > start_pos &&
1115 ordered->file_offset < last_pos) {
1116 btrfs_put_ordered_extent(ordered);
1117 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1118 start_pos, last_pos - 1,
1119 &cached_state, GFP_NOFS);
1120 for (i = 0; i < num_pages; i++) {
1121 unlock_page(pages[i]);
1122 page_cache_release(pages[i]);
1124 btrfs_wait_ordered_range(inode, start_pos,
1125 last_pos - start_pos);
1129 btrfs_put_ordered_extent(ordered);
1131 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
1132 last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1133 EXTENT_DO_ACCOUNTING, 0, 0, &cached_state,
1135 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1136 start_pos, last_pos - 1, &cached_state,
1139 for (i = 0; i < num_pages; i++) {
1140 clear_page_dirty_for_io(pages[i]);
1141 set_page_extent_mapped(pages[i]);
1142 WARN_ON(!PageLocked(pages[i]));
1146 while (faili >= 0) {
1147 unlock_page(pages[faili]);
1148 page_cache_release(pages[faili]);
1155 static noinline ssize_t __btrfs_buffered_write(struct file *file,
1159 struct inode *inode = fdentry(file)->d_inode;
1160 struct btrfs_root *root = BTRFS_I(inode)->root;
1161 struct page **pages = NULL;
1162 unsigned long first_index;
1163 unsigned long last_index;
1164 size_t num_written = 0;
1168 nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) /
1169 PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
1170 (sizeof(struct page *)));
1171 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
1175 first_index = pos >> PAGE_CACHE_SHIFT;
1176 last_index = (pos + iov_iter_count(i)) >> PAGE_CACHE_SHIFT;
1178 while (iov_iter_count(i) > 0) {
1179 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1180 size_t write_bytes = min(iov_iter_count(i),
1181 nrptrs * (size_t)PAGE_CACHE_SIZE -
1183 size_t num_pages = (write_bytes + offset +
1184 PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1188 WARN_ON(num_pages > nrptrs);
1191 * Fault pages before locking them in prepare_pages
1192 * to avoid recursive lock
1194 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1199 ret = btrfs_delalloc_reserve_space(inode,
1200 num_pages << PAGE_CACHE_SHIFT);
1205 * This is going to setup the pages array with the number of
1206 * pages we want, so we don't really need to worry about the
1207 * contents of pages from loop to loop
1209 ret = prepare_pages(root, file, pages, num_pages,
1210 pos, first_index, last_index,
1213 btrfs_delalloc_release_space(inode,
1214 num_pages << PAGE_CACHE_SHIFT);
1218 copied = btrfs_copy_from_user(pos, num_pages,
1219 write_bytes, pages, i);
1222 * if we have trouble faulting in the pages, fall
1223 * back to one page at a time
1225 if (copied < write_bytes)
1231 dirty_pages = (copied + offset +
1232 PAGE_CACHE_SIZE - 1) >>
1236 * If we had a short copy we need to release the excess delaloc
1237 * bytes we reserved. We need to increment outstanding_extents
1238 * because btrfs_delalloc_release_space will decrement it, but
1239 * we still have an outstanding extent for the chunk we actually
1242 if (num_pages > dirty_pages) {
1244 spin_lock(&BTRFS_I(inode)->lock);
1245 BTRFS_I(inode)->outstanding_extents++;
1246 spin_unlock(&BTRFS_I(inode)->lock);
1248 btrfs_delalloc_release_space(inode,
1249 (num_pages - dirty_pages) <<
1254 ret = btrfs_dirty_pages(root, inode, pages,
1255 dirty_pages, pos, copied,
1258 btrfs_delalloc_release_space(inode,
1259 dirty_pages << PAGE_CACHE_SHIFT);
1260 btrfs_drop_pages(pages, num_pages);
1265 btrfs_drop_pages(pages, num_pages);
1269 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
1271 if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1272 btrfs_btree_balance_dirty(root, 1);
1273 btrfs_throttle(root);
1276 num_written += copied;
1281 return num_written ? num_written : ret;
1284 static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1285 const struct iovec *iov,
1286 unsigned long nr_segs, loff_t pos,
1287 loff_t *ppos, size_t count, size_t ocount)
1289 struct file *file = iocb->ki_filp;
1290 struct inode *inode = fdentry(file)->d_inode;
1293 ssize_t written_buffered;
1297 written = generic_file_direct_write(iocb, iov, &nr_segs, pos, ppos,
1301 * the generic O_DIRECT will update in-memory i_size after the
1302 * DIOs are done. But our endio handlers that update the on
1303 * disk i_size never update past the in memory i_size. So we
1304 * need one more update here to catch any additions to the
1307 if (inode->i_size != BTRFS_I(inode)->disk_i_size) {
1308 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
1309 mark_inode_dirty(inode);
1312 if (written < 0 || written == count)
1317 iov_iter_init(&i, iov, nr_segs, count, written);
1318 written_buffered = __btrfs_buffered_write(file, &i, pos);
1319 if (written_buffered < 0) {
1320 err = written_buffered;
1323 endbyte = pos + written_buffered - 1;
1324 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
1327 written += written_buffered;
1328 *ppos = pos + written_buffered;
1329 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
1330 endbyte >> PAGE_CACHE_SHIFT);
1332 return written ? written : err;
1335 static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
1336 const struct iovec *iov,
1337 unsigned long nr_segs, loff_t pos)
1339 struct file *file = iocb->ki_filp;
1340 struct inode *inode = fdentry(file)->d_inode;
1341 struct btrfs_root *root = BTRFS_I(inode)->root;
1342 loff_t *ppos = &iocb->ki_pos;
1343 ssize_t num_written = 0;
1345 size_t count, ocount;
1347 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1349 mutex_lock(&inode->i_mutex);
1351 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
1353 mutex_unlock(&inode->i_mutex);
1358 current->backing_dev_info = inode->i_mapping->backing_dev_info;
1359 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1361 mutex_unlock(&inode->i_mutex);
1366 mutex_unlock(&inode->i_mutex);
1370 err = file_remove_suid(file);
1372 mutex_unlock(&inode->i_mutex);
1377 * If BTRFS flips readonly due to some impossible error
1378 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1379 * although we have opened a file as writable, we have
1380 * to stop this write operation to ensure FS consistency.
1382 if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
1383 mutex_unlock(&inode->i_mutex);
1388 file_update_time(file);
1389 BTRFS_I(inode)->sequence++;
1391 if (unlikely(file->f_flags & O_DIRECT)) {
1392 num_written = __btrfs_direct_write(iocb, iov, nr_segs,
1393 pos, ppos, count, ocount);
1397 iov_iter_init(&i, iov, nr_segs, count, num_written);
1399 num_written = __btrfs_buffered_write(file, &i, pos);
1400 if (num_written > 0)
1401 *ppos = pos + num_written;
1404 mutex_unlock(&inode->i_mutex);
1407 * we want to make sure fsync finds this change
1408 * but we haven't joined a transaction running right now.
1410 * Later on, someone is sure to update the inode and get the
1411 * real transid recorded.
1413 * We set last_trans now to the fs_info generation + 1,
1414 * this will either be one more than the running transaction
1415 * or the generation used for the next transaction if there isn't
1416 * one running right now.
1418 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1419 if (num_written > 0 || num_written == -EIOCBQUEUED) {
1420 err = generic_write_sync(file, pos, num_written);
1421 if (err < 0 && num_written > 0)
1425 current->backing_dev_info = NULL;
1426 return num_written ? num_written : err;
1429 int btrfs_release_file(struct inode *inode, struct file *filp)
1432 * ordered_data_close is set by settattr when we are about to truncate
1433 * a file from a non-zero size to a zero size. This tries to
1434 * flush down new bytes that may have been written if the
1435 * application were using truncate to replace a file in place.
1437 if (BTRFS_I(inode)->ordered_data_close) {
1438 BTRFS_I(inode)->ordered_data_close = 0;
1439 btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode);
1440 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1441 filemap_flush(inode->i_mapping);
1443 if (filp->private_data)
1444 btrfs_ioctl_trans_end(filp);
1449 * fsync call for both files and directories. This logs the inode into
1450 * the tree log instead of forcing full commits whenever possible.
1452 * It needs to call filemap_fdatawait so that all ordered extent updates are
1453 * in the metadata btree are up to date for copying to the log.
1455 * It drops the inode mutex before doing the tree log commit. This is an
1456 * important optimization for directories because holding the mutex prevents
1457 * new operations on the dir while we write to disk.
1459 int btrfs_sync_file(struct file *file, int datasync)
1461 struct dentry *dentry = file->f_path.dentry;
1462 struct inode *inode = dentry->d_inode;
1463 struct btrfs_root *root = BTRFS_I(inode)->root;
1465 struct btrfs_trans_handle *trans;
1467 trace_btrfs_sync_file(file, datasync);
1469 /* we wait first, since the writeback may change the inode */
1471 /* the VFS called filemap_fdatawrite for us */
1472 btrfs_wait_ordered_range(inode, 0, (u64)-1);
1476 * check the transaction that last modified this inode
1477 * and see if its already been committed
1479 if (!BTRFS_I(inode)->last_trans)
1483 * if the last transaction that changed this file was before
1484 * the current transaction, we can bail out now without any
1488 if (BTRFS_I(inode)->last_trans <=
1489 root->fs_info->last_trans_committed) {
1490 BTRFS_I(inode)->last_trans = 0;
1495 * ok we haven't committed the transaction yet, lets do a commit
1497 if (file->private_data)
1498 btrfs_ioctl_trans_end(file);
1500 trans = btrfs_start_transaction(root, 0);
1501 if (IS_ERR(trans)) {
1502 ret = PTR_ERR(trans);
1506 ret = btrfs_log_dentry_safe(trans, root, dentry);
1510 /* we've logged all the items and now have a consistent
1511 * version of the file in the log. It is possible that
1512 * someone will come in and modify the file, but that's
1513 * fine because the log is consistent on disk, and we
1514 * have references to all of the file's extents
1516 * It is possible that someone will come in and log the
1517 * file again, but that will end up using the synchronization
1518 * inside btrfs_sync_log to keep things safe.
1520 mutex_unlock(&dentry->d_inode->i_mutex);
1522 if (ret != BTRFS_NO_LOG_SYNC) {
1524 ret = btrfs_commit_transaction(trans, root);
1526 ret = btrfs_sync_log(trans, root);
1528 ret = btrfs_end_transaction(trans, root);
1530 ret = btrfs_commit_transaction(trans, root);
1533 ret = btrfs_end_transaction(trans, root);
1535 mutex_lock(&dentry->d_inode->i_mutex);
1537 return ret > 0 ? -EIO : ret;
1540 static const struct vm_operations_struct btrfs_file_vm_ops = {
1541 .fault = filemap_fault,
1542 .page_mkwrite = btrfs_page_mkwrite,
1545 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
1547 struct address_space *mapping = filp->f_mapping;
1549 if (!mapping->a_ops->readpage)
1552 file_accessed(filp);
1553 vma->vm_ops = &btrfs_file_vm_ops;
1554 vma->vm_flags |= VM_CAN_NONLINEAR;
1559 static long btrfs_fallocate(struct file *file, int mode,
1560 loff_t offset, loff_t len)
1562 struct inode *inode = file->f_path.dentry->d_inode;
1563 struct extent_state *cached_state = NULL;
1570 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
1571 struct extent_map *em;
1574 alloc_start = offset & ~mask;
1575 alloc_end = (offset + len + mask) & ~mask;
1577 /* We only support the FALLOC_FL_KEEP_SIZE mode */
1578 if (mode & ~FALLOC_FL_KEEP_SIZE)
1582 * wait for ordered IO before we have any locks. We'll loop again
1583 * below with the locks held.
1585 btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
1587 mutex_lock(&inode->i_mutex);
1588 ret = inode_newsize_ok(inode, alloc_end);
1592 if (alloc_start > inode->i_size) {
1593 ret = btrfs_cont_expand(inode, i_size_read(inode),
1599 ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
1603 locked_end = alloc_end - 1;
1605 struct btrfs_ordered_extent *ordered;
1607 /* the extent lock is ordered inside the running
1610 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
1611 locked_end, 0, &cached_state, GFP_NOFS);
1612 ordered = btrfs_lookup_first_ordered_extent(inode,
1615 ordered->file_offset + ordered->len > alloc_start &&
1616 ordered->file_offset < alloc_end) {
1617 btrfs_put_ordered_extent(ordered);
1618 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1619 alloc_start, locked_end,
1620 &cached_state, GFP_NOFS);
1622 * we can't wait on the range with the transaction
1623 * running or with the extent lock held
1625 btrfs_wait_ordered_range(inode, alloc_start,
1626 alloc_end - alloc_start);
1629 btrfs_put_ordered_extent(ordered);
1634 cur_offset = alloc_start;
1636 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
1637 alloc_end - cur_offset, 0);
1638 BUG_ON(IS_ERR_OR_NULL(em));
1639 last_byte = min(extent_map_end(em), alloc_end);
1640 last_byte = (last_byte + mask) & ~mask;
1641 if (em->block_start == EXTENT_MAP_HOLE ||
1642 (cur_offset >= inode->i_size &&
1643 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
1644 ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
1645 last_byte - cur_offset,
1646 1 << inode->i_blkbits,
1650 free_extent_map(em);
1654 free_extent_map(em);
1656 cur_offset = last_byte;
1657 if (cur_offset >= alloc_end) {
1662 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
1663 &cached_state, GFP_NOFS);
1665 btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
1667 mutex_unlock(&inode->i_mutex);
1671 const struct file_operations btrfs_file_operations = {
1672 .llseek = generic_file_llseek,
1673 .read = do_sync_read,
1674 .write = do_sync_write,
1675 .aio_read = generic_file_aio_read,
1676 .splice_read = generic_file_splice_read,
1677 .aio_write = btrfs_file_aio_write,
1678 .mmap = btrfs_file_mmap,
1679 .open = generic_file_open,
1680 .release = btrfs_release_file,
1681 .fsync = btrfs_sync_file,
1682 .fallocate = btrfs_fallocate,
1683 .unlocked_ioctl = btrfs_ioctl,
1684 #ifdef CONFIG_COMPAT
1685 .compat_ioctl = btrfs_ioctl,