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 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 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;
298 key.objectid = defrag->root;
299 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
300 key.offset = (u64)-1;
301 inode_root = btrfs_read_fs_root_no_name(fs_info, &key);
302 if (IS_ERR(inode_root)) {
303 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
304 return PTR_ERR(inode_root);
307 key.objectid = defrag->ino;
308 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
310 inode = btrfs_iget(fs_info->sb, &key, inode_root, NULL);
312 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
313 return PTR_ERR(inode);
316 /* do a chunk of defrag */
317 clear_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
318 memset(&range, 0, sizeof(range));
320 range.start = defrag->last_offset;
322 sb_start_write(fs_info->sb);
323 num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid,
325 sb_end_write(fs_info->sb);
327 * if we filled the whole defrag batch, there
328 * must be more work to do. Queue this defrag
331 if (num_defrag == BTRFS_DEFRAG_BATCH) {
332 defrag->last_offset = range.start;
333 btrfs_requeue_inode_defrag(inode, defrag);
334 } else if (defrag->last_offset && !defrag->cycled) {
336 * we didn't fill our defrag batch, but
337 * we didn't start at zero. Make sure we loop
338 * around to the start of the file.
340 defrag->last_offset = 0;
342 btrfs_requeue_inode_defrag(inode, defrag);
344 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
352 * run through the list of inodes in the FS that need
355 int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info)
357 struct inode_defrag *defrag;
359 u64 root_objectid = 0;
361 atomic_inc(&fs_info->defrag_running);
363 if (!__need_auto_defrag(fs_info->tree_root))
366 /* find an inode to defrag */
367 defrag = btrfs_pick_defrag_inode(fs_info, root_objectid,
370 if (root_objectid || first_ino) {
379 first_ino = defrag->ino + 1;
380 root_objectid = defrag->root;
382 __btrfs_run_defrag_inode(fs_info, defrag);
384 atomic_dec(&fs_info->defrag_running);
387 * during unmount, we use the transaction_wait queue to
388 * wait for the defragger to stop
390 wake_up(&fs_info->transaction_wait);
394 /* simple helper to fault in pages and copy. This should go away
395 * and be replaced with calls into generic code.
397 static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
399 struct page **prepared_pages,
403 size_t total_copied = 0;
405 int offset = pos & (PAGE_CACHE_SIZE - 1);
407 while (write_bytes > 0) {
408 size_t count = min_t(size_t,
409 PAGE_CACHE_SIZE - offset, write_bytes);
410 struct page *page = prepared_pages[pg];
412 * Copy data from userspace to the current page
414 * Disable pagefault to avoid recursive lock since
415 * the pages are already locked
418 copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
421 /* Flush processor's dcache for this page */
422 flush_dcache_page(page);
425 * if we get a partial write, we can end up with
426 * partially up to date pages. These add
427 * a lot of complexity, so make sure they don't
428 * happen by forcing this copy to be retried.
430 * The rest of the btrfs_file_write code will fall
431 * back to page at a time copies after we return 0.
433 if (!PageUptodate(page) && copied < count)
436 iov_iter_advance(i, copied);
437 write_bytes -= copied;
438 total_copied += copied;
440 /* Return to btrfs_file_aio_write to fault page */
441 if (unlikely(copied == 0))
444 if (unlikely(copied < PAGE_CACHE_SIZE - offset)) {
455 * unlocks pages after btrfs_file_write is done with them
457 void btrfs_drop_pages(struct page **pages, size_t num_pages)
460 for (i = 0; i < num_pages; i++) {
461 /* page checked is some magic around finding pages that
462 * have been modified without going through btrfs_set_page_dirty
465 ClearPageChecked(pages[i]);
466 unlock_page(pages[i]);
467 mark_page_accessed(pages[i]);
468 page_cache_release(pages[i]);
473 * after copy_from_user, pages need to be dirtied and we need to make
474 * sure holes are created between the current EOF and the start of
475 * any next extents (if required).
477 * this also makes the decision about creating an inline extent vs
478 * doing real data extents, marking pages dirty and delalloc as required.
480 int btrfs_dirty_pages(struct btrfs_root *root, struct inode *inode,
481 struct page **pages, size_t num_pages,
482 loff_t pos, size_t write_bytes,
483 struct extent_state **cached)
489 u64 end_of_last_block;
490 u64 end_pos = pos + write_bytes;
491 loff_t isize = i_size_read(inode);
493 start_pos = pos & ~((u64)root->sectorsize - 1);
494 num_bytes = (write_bytes + pos - start_pos +
495 root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
497 end_of_last_block = start_pos + num_bytes - 1;
498 err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
503 for (i = 0; i < num_pages; i++) {
504 struct page *p = pages[i];
511 * we've only changed i_size in ram, and we haven't updated
512 * the disk i_size. There is no need to log the inode
516 i_size_write(inode, end_pos);
521 * this drops all the extents in the cache that intersect the range
522 * [start, end]. Existing extents are split as required.
524 void btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
527 struct extent_map *em;
528 struct extent_map *split = NULL;
529 struct extent_map *split2 = NULL;
530 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
531 u64 len = end - start + 1;
538 WARN_ON(end < start);
539 if (end == (u64)-1) {
547 split = alloc_extent_map();
549 split2 = alloc_extent_map();
550 if (!split || !split2)
553 write_lock(&em_tree->lock);
554 em = lookup_extent_mapping(em_tree, start, len);
556 write_unlock(&em_tree->lock);
560 gen = em->generation;
561 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
562 if (testend && em->start + em->len >= start + len) {
564 write_unlock(&em_tree->lock);
567 start = em->start + em->len;
569 len = start + len - (em->start + em->len);
571 write_unlock(&em_tree->lock);
574 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
575 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
576 remove_extent_mapping(em_tree, em);
580 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
582 split->start = em->start;
583 split->len = start - em->start;
584 split->orig_start = em->orig_start;
585 split->block_start = em->block_start;
588 split->block_len = em->block_len;
590 split->block_len = split->len;
591 split->orig_block_len = max(split->block_len,
593 split->generation = gen;
594 split->bdev = em->bdev;
595 split->flags = flags;
596 split->compress_type = em->compress_type;
597 ret = add_extent_mapping(em_tree, split);
598 BUG_ON(ret); /* Logic error */
599 list_move(&split->list, &em_tree->modified_extents);
600 free_extent_map(split);
604 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
605 testend && em->start + em->len > start + len) {
606 u64 diff = start + len - em->start;
608 split->start = start + len;
609 split->len = em->start + em->len - (start + len);
610 split->bdev = em->bdev;
611 split->flags = flags;
612 split->compress_type = em->compress_type;
613 split->generation = gen;
614 split->orig_block_len = max(em->block_len,
618 split->block_len = em->block_len;
619 split->block_start = em->block_start;
620 split->orig_start = em->orig_start;
622 split->block_len = split->len;
623 split->block_start = em->block_start + diff;
624 split->orig_start = em->orig_start;
627 ret = add_extent_mapping(em_tree, split);
628 BUG_ON(ret); /* Logic error */
629 list_move(&split->list, &em_tree->modified_extents);
630 free_extent_map(split);
634 write_unlock(&em_tree->lock);
638 /* once for the tree*/
642 free_extent_map(split);
644 free_extent_map(split2);
648 * this is very complex, but the basic idea is to drop all extents
649 * in the range start - end. hint_block is filled in with a block number
650 * that would be a good hint to the block allocator for this file.
652 * If an extent intersects the range but is not entirely inside the range
653 * it is either truncated or split. Anything entirely inside the range
654 * is deleted from the tree.
656 int __btrfs_drop_extents(struct btrfs_trans_handle *trans,
657 struct btrfs_root *root, struct inode *inode,
658 struct btrfs_path *path, u64 start, u64 end,
659 u64 *drop_end, int drop_cache)
661 struct extent_buffer *leaf;
662 struct btrfs_file_extent_item *fi;
663 struct btrfs_key key;
664 struct btrfs_key new_key;
665 u64 ino = btrfs_ino(inode);
666 u64 search_start = start;
669 u64 extent_offset = 0;
676 int modify_tree = -1;
677 int update_refs = (root->ref_cows || root == root->fs_info->tree_root);
681 btrfs_drop_extent_cache(inode, start, end - 1, 0);
683 if (start >= BTRFS_I(inode)->disk_i_size)
688 ret = btrfs_lookup_file_extent(trans, root, path, ino,
689 search_start, modify_tree);
692 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
693 leaf = path->nodes[0];
694 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
695 if (key.objectid == ino &&
696 key.type == BTRFS_EXTENT_DATA_KEY)
701 leaf = path->nodes[0];
702 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
704 ret = btrfs_next_leaf(root, path);
711 leaf = path->nodes[0];
715 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
716 if (key.objectid > ino ||
717 key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
720 fi = btrfs_item_ptr(leaf, path->slots[0],
721 struct btrfs_file_extent_item);
722 extent_type = btrfs_file_extent_type(leaf, fi);
724 if (extent_type == BTRFS_FILE_EXTENT_REG ||
725 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
726 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
727 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
728 extent_offset = btrfs_file_extent_offset(leaf, fi);
729 extent_end = key.offset +
730 btrfs_file_extent_num_bytes(leaf, fi);
731 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
732 extent_end = key.offset +
733 btrfs_file_extent_inline_len(leaf, fi);
736 extent_end = search_start;
739 if (extent_end <= search_start) {
745 search_start = max(key.offset, start);
746 if (recow || !modify_tree) {
748 btrfs_release_path(path);
753 * | - range to drop - |
754 * | -------- extent -------- |
756 if (start > key.offset && end < extent_end) {
758 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
760 memcpy(&new_key, &key, sizeof(new_key));
761 new_key.offset = start;
762 ret = btrfs_duplicate_item(trans, root, path,
764 if (ret == -EAGAIN) {
765 btrfs_release_path(path);
771 leaf = path->nodes[0];
772 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
773 struct btrfs_file_extent_item);
774 btrfs_set_file_extent_num_bytes(leaf, fi,
777 fi = btrfs_item_ptr(leaf, path->slots[0],
778 struct btrfs_file_extent_item);
780 extent_offset += start - key.offset;
781 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
782 btrfs_set_file_extent_num_bytes(leaf, fi,
784 btrfs_mark_buffer_dirty(leaf);
786 if (update_refs && disk_bytenr > 0) {
787 ret = btrfs_inc_extent_ref(trans, root,
788 disk_bytenr, num_bytes, 0,
789 root->root_key.objectid,
791 start - extent_offset, 0);
792 BUG_ON(ret); /* -ENOMEM */
797 * | ---- range to drop ----- |
798 * | -------- extent -------- |
800 if (start <= key.offset && end < extent_end) {
801 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
803 memcpy(&new_key, &key, sizeof(new_key));
804 new_key.offset = end;
805 btrfs_set_item_key_safe(trans, root, path, &new_key);
807 extent_offset += end - key.offset;
808 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
809 btrfs_set_file_extent_num_bytes(leaf, fi,
811 btrfs_mark_buffer_dirty(leaf);
812 if (update_refs && disk_bytenr > 0)
813 inode_sub_bytes(inode, end - key.offset);
817 search_start = extent_end;
819 * | ---- range to drop ----- |
820 * | -------- extent -------- |
822 if (start > key.offset && end >= extent_end) {
824 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
826 btrfs_set_file_extent_num_bytes(leaf, fi,
828 btrfs_mark_buffer_dirty(leaf);
829 if (update_refs && disk_bytenr > 0)
830 inode_sub_bytes(inode, extent_end - start);
831 if (end == extent_end)
839 * | ---- range to drop ----- |
840 * | ------ extent ------ |
842 if (start <= key.offset && end >= extent_end) {
844 del_slot = path->slots[0];
847 BUG_ON(del_slot + del_nr != path->slots[0]);
852 extent_type == BTRFS_FILE_EXTENT_INLINE) {
853 inode_sub_bytes(inode,
854 extent_end - key.offset);
855 extent_end = ALIGN(extent_end,
857 } else if (update_refs && disk_bytenr > 0) {
858 ret = btrfs_free_extent(trans, root,
859 disk_bytenr, num_bytes, 0,
860 root->root_key.objectid,
861 key.objectid, key.offset -
863 BUG_ON(ret); /* -ENOMEM */
864 inode_sub_bytes(inode,
865 extent_end - key.offset);
868 if (end == extent_end)
871 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
876 ret = btrfs_del_items(trans, root, path, del_slot,
879 btrfs_abort_transaction(trans, root, ret);
886 btrfs_release_path(path);
893 if (!ret && del_nr > 0) {
894 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
896 btrfs_abort_transaction(trans, root, ret);
900 *drop_end = found ? min(end, extent_end) : end;
901 btrfs_release_path(path);
905 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
906 struct btrfs_root *root, struct inode *inode, u64 start,
907 u64 end, int drop_cache)
909 struct btrfs_path *path;
912 path = btrfs_alloc_path();
915 ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL,
917 btrfs_free_path(path);
921 static int extent_mergeable(struct extent_buffer *leaf, int slot,
922 u64 objectid, u64 bytenr, u64 orig_offset,
923 u64 *start, u64 *end)
925 struct btrfs_file_extent_item *fi;
926 struct btrfs_key key;
929 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
932 btrfs_item_key_to_cpu(leaf, &key, slot);
933 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
936 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
937 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
938 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
939 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
940 btrfs_file_extent_compression(leaf, fi) ||
941 btrfs_file_extent_encryption(leaf, fi) ||
942 btrfs_file_extent_other_encoding(leaf, fi))
945 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
946 if ((*start && *start != key.offset) || (*end && *end != extent_end))
955 * Mark extent in the range start - end as written.
957 * This changes extent type from 'pre-allocated' to 'regular'. If only
958 * part of extent is marked as written, the extent will be split into
961 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
962 struct inode *inode, u64 start, u64 end)
964 struct btrfs_root *root = BTRFS_I(inode)->root;
965 struct extent_buffer *leaf;
966 struct btrfs_path *path;
967 struct btrfs_file_extent_item *fi;
968 struct btrfs_key key;
969 struct btrfs_key new_key;
981 u64 ino = btrfs_ino(inode);
983 path = btrfs_alloc_path();
990 key.type = BTRFS_EXTENT_DATA_KEY;
993 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
996 if (ret > 0 && path->slots[0] > 0)
999 leaf = path->nodes[0];
1000 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1001 BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY);
1002 fi = btrfs_item_ptr(leaf, path->slots[0],
1003 struct btrfs_file_extent_item);
1004 BUG_ON(btrfs_file_extent_type(leaf, fi) !=
1005 BTRFS_FILE_EXTENT_PREALLOC);
1006 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1007 BUG_ON(key.offset > start || extent_end < end);
1009 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1010 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1011 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
1012 memcpy(&new_key, &key, sizeof(new_key));
1014 if (start == key.offset && end < extent_end) {
1017 if (extent_mergeable(leaf, path->slots[0] - 1,
1018 ino, bytenr, orig_offset,
1019 &other_start, &other_end)) {
1020 new_key.offset = end;
1021 btrfs_set_item_key_safe(trans, root, path, &new_key);
1022 fi = btrfs_item_ptr(leaf, path->slots[0],
1023 struct btrfs_file_extent_item);
1024 btrfs_set_file_extent_generation(leaf, fi,
1026 btrfs_set_file_extent_num_bytes(leaf, fi,
1028 btrfs_set_file_extent_offset(leaf, fi,
1030 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1031 struct btrfs_file_extent_item);
1032 btrfs_set_file_extent_generation(leaf, fi,
1034 btrfs_set_file_extent_num_bytes(leaf, fi,
1036 btrfs_mark_buffer_dirty(leaf);
1041 if (start > key.offset && end == extent_end) {
1044 if (extent_mergeable(leaf, path->slots[0] + 1,
1045 ino, bytenr, orig_offset,
1046 &other_start, &other_end)) {
1047 fi = btrfs_item_ptr(leaf, path->slots[0],
1048 struct btrfs_file_extent_item);
1049 btrfs_set_file_extent_num_bytes(leaf, fi,
1050 start - key.offset);
1051 btrfs_set_file_extent_generation(leaf, fi,
1054 new_key.offset = start;
1055 btrfs_set_item_key_safe(trans, root, path, &new_key);
1057 fi = btrfs_item_ptr(leaf, path->slots[0],
1058 struct btrfs_file_extent_item);
1059 btrfs_set_file_extent_generation(leaf, fi,
1061 btrfs_set_file_extent_num_bytes(leaf, fi,
1063 btrfs_set_file_extent_offset(leaf, fi,
1064 start - orig_offset);
1065 btrfs_mark_buffer_dirty(leaf);
1070 while (start > key.offset || end < extent_end) {
1071 if (key.offset == start)
1074 new_key.offset = split;
1075 ret = btrfs_duplicate_item(trans, root, path, &new_key);
1076 if (ret == -EAGAIN) {
1077 btrfs_release_path(path);
1081 btrfs_abort_transaction(trans, root, ret);
1085 leaf = path->nodes[0];
1086 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1087 struct btrfs_file_extent_item);
1088 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1089 btrfs_set_file_extent_num_bytes(leaf, fi,
1090 split - key.offset);
1092 fi = btrfs_item_ptr(leaf, path->slots[0],
1093 struct btrfs_file_extent_item);
1095 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1096 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
1097 btrfs_set_file_extent_num_bytes(leaf, fi,
1098 extent_end - split);
1099 btrfs_mark_buffer_dirty(leaf);
1101 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
1102 root->root_key.objectid,
1103 ino, orig_offset, 0);
1104 BUG_ON(ret); /* -ENOMEM */
1106 if (split == start) {
1109 BUG_ON(start != key.offset);
1118 if (extent_mergeable(leaf, path->slots[0] + 1,
1119 ino, bytenr, orig_offset,
1120 &other_start, &other_end)) {
1122 btrfs_release_path(path);
1125 extent_end = other_end;
1126 del_slot = path->slots[0] + 1;
1128 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1129 0, root->root_key.objectid,
1130 ino, orig_offset, 0);
1131 BUG_ON(ret); /* -ENOMEM */
1135 if (extent_mergeable(leaf, path->slots[0] - 1,
1136 ino, bytenr, orig_offset,
1137 &other_start, &other_end)) {
1139 btrfs_release_path(path);
1142 key.offset = other_start;
1143 del_slot = path->slots[0];
1145 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1146 0, root->root_key.objectid,
1147 ino, orig_offset, 0);
1148 BUG_ON(ret); /* -ENOMEM */
1151 fi = btrfs_item_ptr(leaf, path->slots[0],
1152 struct btrfs_file_extent_item);
1153 btrfs_set_file_extent_type(leaf, fi,
1154 BTRFS_FILE_EXTENT_REG);
1155 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1156 btrfs_mark_buffer_dirty(leaf);
1158 fi = btrfs_item_ptr(leaf, del_slot - 1,
1159 struct btrfs_file_extent_item);
1160 btrfs_set_file_extent_type(leaf, fi,
1161 BTRFS_FILE_EXTENT_REG);
1162 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1163 btrfs_set_file_extent_num_bytes(leaf, fi,
1164 extent_end - key.offset);
1165 btrfs_mark_buffer_dirty(leaf);
1167 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1169 btrfs_abort_transaction(trans, root, ret);
1174 btrfs_free_path(path);
1179 * on error we return an unlocked page and the error value
1180 * on success we return a locked page and 0
1182 static int prepare_uptodate_page(struct page *page, u64 pos,
1183 bool force_uptodate)
1187 if (((pos & (PAGE_CACHE_SIZE - 1)) || force_uptodate) &&
1188 !PageUptodate(page)) {
1189 ret = btrfs_readpage(NULL, page);
1193 if (!PageUptodate(page)) {
1202 * this gets pages into the page cache and locks them down, it also properly
1203 * waits for data=ordered extents to finish before allowing the pages to be
1206 static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
1207 struct page **pages, size_t num_pages,
1208 loff_t pos, unsigned long first_index,
1209 size_t write_bytes, bool force_uptodate)
1211 struct extent_state *cached_state = NULL;
1213 unsigned long index = pos >> PAGE_CACHE_SHIFT;
1214 struct inode *inode = fdentry(file)->d_inode;
1215 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1221 start_pos = pos & ~((u64)root->sectorsize - 1);
1222 last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
1225 for (i = 0; i < num_pages; i++) {
1226 pages[i] = find_or_create_page(inode->i_mapping, index + i,
1227 mask | __GFP_WRITE);
1235 err = prepare_uptodate_page(pages[i], pos,
1237 if (i == num_pages - 1)
1238 err = prepare_uptodate_page(pages[i],
1239 pos + write_bytes, false);
1241 page_cache_release(pages[i]);
1245 wait_on_page_writeback(pages[i]);
1248 if (start_pos < inode->i_size) {
1249 struct btrfs_ordered_extent *ordered;
1250 lock_extent_bits(&BTRFS_I(inode)->io_tree,
1251 start_pos, last_pos - 1, 0, &cached_state);
1252 ordered = btrfs_lookup_first_ordered_extent(inode,
1255 ordered->file_offset + ordered->len > start_pos &&
1256 ordered->file_offset < last_pos) {
1257 btrfs_put_ordered_extent(ordered);
1258 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1259 start_pos, last_pos - 1,
1260 &cached_state, GFP_NOFS);
1261 for (i = 0; i < num_pages; i++) {
1262 unlock_page(pages[i]);
1263 page_cache_release(pages[i]);
1265 btrfs_wait_ordered_range(inode, start_pos,
1266 last_pos - start_pos);
1270 btrfs_put_ordered_extent(ordered);
1272 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
1273 last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1274 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
1275 0, 0, &cached_state, GFP_NOFS);
1276 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1277 start_pos, last_pos - 1, &cached_state,
1280 for (i = 0; i < num_pages; i++) {
1281 if (clear_page_dirty_for_io(pages[i]))
1282 account_page_redirty(pages[i]);
1283 set_page_extent_mapped(pages[i]);
1284 WARN_ON(!PageLocked(pages[i]));
1288 while (faili >= 0) {
1289 unlock_page(pages[faili]);
1290 page_cache_release(pages[faili]);
1297 static noinline ssize_t __btrfs_buffered_write(struct file *file,
1301 struct inode *inode = fdentry(file)->d_inode;
1302 struct btrfs_root *root = BTRFS_I(inode)->root;
1303 struct page **pages = NULL;
1304 unsigned long first_index;
1305 size_t num_written = 0;
1308 bool force_page_uptodate = false;
1310 nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) /
1311 PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
1312 (sizeof(struct page *)));
1313 nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1314 nrptrs = max(nrptrs, 8);
1315 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
1319 first_index = pos >> PAGE_CACHE_SHIFT;
1321 while (iov_iter_count(i) > 0) {
1322 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1323 size_t write_bytes = min(iov_iter_count(i),
1324 nrptrs * (size_t)PAGE_CACHE_SIZE -
1326 size_t num_pages = (write_bytes + offset +
1327 PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1331 WARN_ON(num_pages > nrptrs);
1334 * Fault pages before locking them in prepare_pages
1335 * to avoid recursive lock
1337 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1342 ret = btrfs_delalloc_reserve_space(inode,
1343 num_pages << PAGE_CACHE_SHIFT);
1348 * This is going to setup the pages array with the number of
1349 * pages we want, so we don't really need to worry about the
1350 * contents of pages from loop to loop
1352 ret = prepare_pages(root, file, pages, num_pages,
1353 pos, first_index, write_bytes,
1354 force_page_uptodate);
1356 btrfs_delalloc_release_space(inode,
1357 num_pages << PAGE_CACHE_SHIFT);
1361 copied = btrfs_copy_from_user(pos, num_pages,
1362 write_bytes, pages, i);
1365 * if we have trouble faulting in the pages, fall
1366 * back to one page at a time
1368 if (copied < write_bytes)
1372 force_page_uptodate = true;
1375 force_page_uptodate = false;
1376 dirty_pages = (copied + offset +
1377 PAGE_CACHE_SIZE - 1) >>
1382 * If we had a short copy we need to release the excess delaloc
1383 * bytes we reserved. We need to increment outstanding_extents
1384 * because btrfs_delalloc_release_space will decrement it, but
1385 * we still have an outstanding extent for the chunk we actually
1388 if (num_pages > dirty_pages) {
1390 spin_lock(&BTRFS_I(inode)->lock);
1391 BTRFS_I(inode)->outstanding_extents++;
1392 spin_unlock(&BTRFS_I(inode)->lock);
1394 btrfs_delalloc_release_space(inode,
1395 (num_pages - dirty_pages) <<
1400 ret = btrfs_dirty_pages(root, inode, pages,
1401 dirty_pages, pos, copied,
1404 btrfs_delalloc_release_space(inode,
1405 dirty_pages << PAGE_CACHE_SHIFT);
1406 btrfs_drop_pages(pages, num_pages);
1411 btrfs_drop_pages(pages, num_pages);
1415 balance_dirty_pages_ratelimited(inode->i_mapping);
1416 if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1417 btrfs_btree_balance_dirty(root);
1420 num_written += copied;
1425 return num_written ? num_written : ret;
1428 static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1429 const struct iovec *iov,
1430 unsigned long nr_segs, loff_t pos,
1431 loff_t *ppos, size_t count, size_t ocount)
1433 struct file *file = iocb->ki_filp;
1436 ssize_t written_buffered;
1440 written = generic_file_direct_write(iocb, iov, &nr_segs, pos, ppos,
1443 if (written < 0 || written == count)
1448 iov_iter_init(&i, iov, nr_segs, count, written);
1449 written_buffered = __btrfs_buffered_write(file, &i, pos);
1450 if (written_buffered < 0) {
1451 err = written_buffered;
1454 endbyte = pos + written_buffered - 1;
1455 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
1458 written += written_buffered;
1459 *ppos = pos + written_buffered;
1460 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
1461 endbyte >> PAGE_CACHE_SHIFT);
1463 return written ? written : err;
1466 static void update_time_for_write(struct inode *inode)
1468 struct timespec now;
1470 if (IS_NOCMTIME(inode))
1473 now = current_fs_time(inode->i_sb);
1474 if (!timespec_equal(&inode->i_mtime, &now))
1475 inode->i_mtime = now;
1477 if (!timespec_equal(&inode->i_ctime, &now))
1478 inode->i_ctime = now;
1480 if (IS_I_VERSION(inode))
1481 inode_inc_iversion(inode);
1484 static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
1485 const struct iovec *iov,
1486 unsigned long nr_segs, loff_t pos)
1488 struct file *file = iocb->ki_filp;
1489 struct inode *inode = fdentry(file)->d_inode;
1490 struct btrfs_root *root = BTRFS_I(inode)->root;
1491 loff_t *ppos = &iocb->ki_pos;
1493 ssize_t num_written = 0;
1495 size_t count, ocount;
1496 bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host);
1498 sb_start_write(inode->i_sb);
1500 mutex_lock(&inode->i_mutex);
1502 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
1504 mutex_unlock(&inode->i_mutex);
1509 current->backing_dev_info = inode->i_mapping->backing_dev_info;
1510 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1512 mutex_unlock(&inode->i_mutex);
1517 mutex_unlock(&inode->i_mutex);
1521 err = file_remove_suid(file);
1523 mutex_unlock(&inode->i_mutex);
1528 * If BTRFS flips readonly due to some impossible error
1529 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1530 * although we have opened a file as writable, we have
1531 * to stop this write operation to ensure FS consistency.
1533 if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
1534 mutex_unlock(&inode->i_mutex);
1540 * We reserve space for updating the inode when we reserve space for the
1541 * extent we are going to write, so we will enospc out there. We don't
1542 * need to start yet another transaction to update the inode as we will
1543 * update the inode when we finish writing whatever data we write.
1545 update_time_for_write(inode);
1547 start_pos = round_down(pos, root->sectorsize);
1548 if (start_pos > i_size_read(inode)) {
1549 err = btrfs_cont_expand(inode, i_size_read(inode), start_pos);
1551 mutex_unlock(&inode->i_mutex);
1557 atomic_inc(&BTRFS_I(inode)->sync_writers);
1559 if (unlikely(file->f_flags & O_DIRECT)) {
1560 num_written = __btrfs_direct_write(iocb, iov, nr_segs,
1561 pos, ppos, count, ocount);
1565 iov_iter_init(&i, iov, nr_segs, count, num_written);
1567 num_written = __btrfs_buffered_write(file, &i, pos);
1568 if (num_written > 0)
1569 *ppos = pos + num_written;
1572 mutex_unlock(&inode->i_mutex);
1575 * we want to make sure fsync finds this change
1576 * but we haven't joined a transaction running right now.
1578 * Later on, someone is sure to update the inode and get the
1579 * real transid recorded.
1581 * We set last_trans now to the fs_info generation + 1,
1582 * this will either be one more than the running transaction
1583 * or the generation used for the next transaction if there isn't
1584 * one running right now.
1586 * We also have to set last_sub_trans to the current log transid,
1587 * otherwise subsequent syncs to a file that's been synced in this
1588 * transaction will appear to have already occured.
1590 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1591 BTRFS_I(inode)->last_sub_trans = root->log_transid;
1592 if (num_written > 0 || num_written == -EIOCBQUEUED) {
1593 err = generic_write_sync(file, pos, num_written);
1594 if (err < 0 && num_written > 0)
1599 atomic_dec(&BTRFS_I(inode)->sync_writers);
1600 sb_end_write(inode->i_sb);
1601 current->backing_dev_info = NULL;
1602 return num_written ? num_written : err;
1605 int btrfs_release_file(struct inode *inode, struct file *filp)
1608 * ordered_data_close is set by settattr when we are about to truncate
1609 * a file from a non-zero size to a zero size. This tries to
1610 * flush down new bytes that may have been written if the
1611 * application were using truncate to replace a file in place.
1613 if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
1614 &BTRFS_I(inode)->runtime_flags)) {
1615 btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode);
1616 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1617 filemap_flush(inode->i_mapping);
1619 if (filp->private_data)
1620 btrfs_ioctl_trans_end(filp);
1625 * fsync call for both files and directories. This logs the inode into
1626 * the tree log instead of forcing full commits whenever possible.
1628 * It needs to call filemap_fdatawait so that all ordered extent updates are
1629 * in the metadata btree are up to date for copying to the log.
1631 * It drops the inode mutex before doing the tree log commit. This is an
1632 * important optimization for directories because holding the mutex prevents
1633 * new operations on the dir while we write to disk.
1635 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1637 struct dentry *dentry = file->f_path.dentry;
1638 struct inode *inode = dentry->d_inode;
1639 struct btrfs_root *root = BTRFS_I(inode)->root;
1641 struct btrfs_trans_handle *trans;
1643 trace_btrfs_sync_file(file, datasync);
1646 * We write the dirty pages in the range and wait until they complete
1647 * out of the ->i_mutex. If so, we can flush the dirty pages by
1648 * multi-task, and make the performance up.
1650 atomic_inc(&BTRFS_I(inode)->sync_writers);
1651 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
1652 atomic_dec(&BTRFS_I(inode)->sync_writers);
1656 mutex_lock(&inode->i_mutex);
1659 * We flush the dirty pages again to avoid some dirty pages in the
1662 atomic_inc(&root->log_batch);
1663 btrfs_wait_ordered_range(inode, start, end - start + 1);
1664 atomic_inc(&root->log_batch);
1667 * check the transaction that last modified this inode
1668 * and see if its already been committed
1670 if (!BTRFS_I(inode)->last_trans) {
1671 mutex_unlock(&inode->i_mutex);
1676 * if the last transaction that changed this file was before
1677 * the current transaction, we can bail out now without any
1681 if (btrfs_inode_in_log(inode, root->fs_info->generation) ||
1682 BTRFS_I(inode)->last_trans <=
1683 root->fs_info->last_trans_committed) {
1684 BTRFS_I(inode)->last_trans = 0;
1687 * We'v had everything committed since the last time we were
1688 * modified so clear this flag in case it was set for whatever
1689 * reason, it's no longer relevant.
1691 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1692 &BTRFS_I(inode)->runtime_flags);
1693 mutex_unlock(&inode->i_mutex);
1698 * ok we haven't committed the transaction yet, lets do a commit
1700 if (file->private_data)
1701 btrfs_ioctl_trans_end(file);
1703 trans = btrfs_start_transaction(root, 0);
1704 if (IS_ERR(trans)) {
1705 ret = PTR_ERR(trans);
1706 mutex_unlock(&inode->i_mutex);
1710 ret = btrfs_log_dentry_safe(trans, root, dentry);
1712 mutex_unlock(&inode->i_mutex);
1716 /* we've logged all the items and now have a consistent
1717 * version of the file in the log. It is possible that
1718 * someone will come in and modify the file, but that's
1719 * fine because the log is consistent on disk, and we
1720 * have references to all of the file's extents
1722 * It is possible that someone will come in and log the
1723 * file again, but that will end up using the synchronization
1724 * inside btrfs_sync_log to keep things safe.
1726 mutex_unlock(&inode->i_mutex);
1728 if (ret != BTRFS_NO_LOG_SYNC) {
1730 ret = btrfs_commit_transaction(trans, root);
1732 ret = btrfs_sync_log(trans, root);
1734 ret = btrfs_end_transaction(trans, root);
1736 ret = btrfs_commit_transaction(trans, root);
1739 ret = btrfs_end_transaction(trans, root);
1742 return ret > 0 ? -EIO : ret;
1745 static const struct vm_operations_struct btrfs_file_vm_ops = {
1746 .fault = filemap_fault,
1747 .page_mkwrite = btrfs_page_mkwrite,
1748 .remap_pages = generic_file_remap_pages,
1751 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
1753 struct address_space *mapping = filp->f_mapping;
1755 if (!mapping->a_ops->readpage)
1758 file_accessed(filp);
1759 vma->vm_ops = &btrfs_file_vm_ops;
1764 static int hole_mergeable(struct inode *inode, struct extent_buffer *leaf,
1765 int slot, u64 start, u64 end)
1767 struct btrfs_file_extent_item *fi;
1768 struct btrfs_key key;
1770 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
1773 btrfs_item_key_to_cpu(leaf, &key, slot);
1774 if (key.objectid != btrfs_ino(inode) ||
1775 key.type != BTRFS_EXTENT_DATA_KEY)
1778 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1780 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
1783 if (btrfs_file_extent_disk_bytenr(leaf, fi))
1786 if (key.offset == end)
1788 if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
1793 static int fill_holes(struct btrfs_trans_handle *trans, struct inode *inode,
1794 struct btrfs_path *path, u64 offset, u64 end)
1796 struct btrfs_root *root = BTRFS_I(inode)->root;
1797 struct extent_buffer *leaf;
1798 struct btrfs_file_extent_item *fi;
1799 struct extent_map *hole_em;
1800 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1801 struct btrfs_key key;
1804 key.objectid = btrfs_ino(inode);
1805 key.type = BTRFS_EXTENT_DATA_KEY;
1806 key.offset = offset;
1809 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1814 leaf = path->nodes[0];
1815 if (hole_mergeable(inode, leaf, path->slots[0]-1, offset, end)) {
1819 fi = btrfs_item_ptr(leaf, path->slots[0],
1820 struct btrfs_file_extent_item);
1821 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
1823 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1824 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
1825 btrfs_set_file_extent_offset(leaf, fi, 0);
1826 btrfs_mark_buffer_dirty(leaf);
1830 if (hole_mergeable(inode, leaf, path->slots[0]+1, offset, end)) {
1834 key.offset = offset;
1835 btrfs_set_item_key_safe(trans, root, path, &key);
1836 fi = btrfs_item_ptr(leaf, path->slots[0],
1837 struct btrfs_file_extent_item);
1838 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
1840 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1841 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
1842 btrfs_set_file_extent_offset(leaf, fi, 0);
1843 btrfs_mark_buffer_dirty(leaf);
1846 btrfs_release_path(path);
1848 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
1849 0, 0, end - offset, 0, end - offset,
1855 btrfs_release_path(path);
1857 hole_em = alloc_extent_map();
1859 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
1860 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1861 &BTRFS_I(inode)->runtime_flags);
1863 hole_em->start = offset;
1864 hole_em->len = end - offset;
1865 hole_em->orig_start = offset;
1867 hole_em->block_start = EXTENT_MAP_HOLE;
1868 hole_em->block_len = 0;
1869 hole_em->orig_block_len = 0;
1870 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
1871 hole_em->compress_type = BTRFS_COMPRESS_NONE;
1872 hole_em->generation = trans->transid;
1875 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
1876 write_lock(&em_tree->lock);
1877 ret = add_extent_mapping(em_tree, hole_em);
1879 list_move(&hole_em->list,
1880 &em_tree->modified_extents);
1881 write_unlock(&em_tree->lock);
1882 } while (ret == -EEXIST);
1883 free_extent_map(hole_em);
1885 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1886 &BTRFS_I(inode)->runtime_flags);
1892 static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
1894 struct btrfs_root *root = BTRFS_I(inode)->root;
1895 struct extent_state *cached_state = NULL;
1896 struct btrfs_path *path;
1897 struct btrfs_block_rsv *rsv;
1898 struct btrfs_trans_handle *trans;
1899 u64 lockstart = round_up(offset, BTRFS_I(inode)->root->sectorsize);
1900 u64 lockend = round_down(offset + len,
1901 BTRFS_I(inode)->root->sectorsize) - 1;
1902 u64 cur_offset = lockstart;
1903 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
1907 bool same_page = ((offset >> PAGE_CACHE_SHIFT) ==
1908 ((offset + len - 1) >> PAGE_CACHE_SHIFT));
1910 btrfs_wait_ordered_range(inode, offset, len);
1912 mutex_lock(&inode->i_mutex);
1914 * We needn't truncate any page which is beyond the end of the file
1915 * because we are sure there is no data there.
1918 * Only do this if we are in the same page and we aren't doing the
1921 if (same_page && len < PAGE_CACHE_SIZE) {
1922 if (offset < round_up(inode->i_size, PAGE_CACHE_SIZE))
1923 ret = btrfs_truncate_page(inode, offset, len, 0);
1924 mutex_unlock(&inode->i_mutex);
1928 /* zero back part of the first page */
1929 if (offset < round_up(inode->i_size, PAGE_CACHE_SIZE)) {
1930 ret = btrfs_truncate_page(inode, offset, 0, 0);
1932 mutex_unlock(&inode->i_mutex);
1937 /* zero the front end of the last page */
1938 if (offset + len < round_up(inode->i_size, PAGE_CACHE_SIZE)) {
1939 ret = btrfs_truncate_page(inode, offset + len, 0, 1);
1941 mutex_unlock(&inode->i_mutex);
1946 if (lockend < lockstart) {
1947 mutex_unlock(&inode->i_mutex);
1952 struct btrfs_ordered_extent *ordered;
1954 truncate_pagecache_range(inode, lockstart, lockend);
1956 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
1958 ordered = btrfs_lookup_first_ordered_extent(inode, lockend);
1961 * We need to make sure we have no ordered extents in this range
1962 * and nobody raced in and read a page in this range, if we did
1963 * we need to try again.
1966 (ordered->file_offset + ordered->len < lockstart ||
1967 ordered->file_offset > lockend)) &&
1968 !test_range_bit(&BTRFS_I(inode)->io_tree, lockstart,
1969 lockend, EXTENT_UPTODATE, 0,
1972 btrfs_put_ordered_extent(ordered);
1976 btrfs_put_ordered_extent(ordered);
1977 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
1978 lockend, &cached_state, GFP_NOFS);
1979 btrfs_wait_ordered_range(inode, lockstart,
1980 lockend - lockstart + 1);
1983 path = btrfs_alloc_path();
1989 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
1994 rsv->size = btrfs_calc_trunc_metadata_size(root, 1);
1998 * 1 - update the inode
1999 * 1 - removing the extents in the range
2000 * 1 - adding the hole extent
2002 trans = btrfs_start_transaction(root, 3);
2003 if (IS_ERR(trans)) {
2004 err = PTR_ERR(trans);
2008 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
2011 trans->block_rsv = rsv;
2013 while (cur_offset < lockend) {
2014 ret = __btrfs_drop_extents(trans, root, inode, path,
2015 cur_offset, lockend + 1,
2020 trans->block_rsv = &root->fs_info->trans_block_rsv;
2022 ret = fill_holes(trans, inode, path, cur_offset, drop_end);
2028 cur_offset = drop_end;
2030 ret = btrfs_update_inode(trans, root, inode);
2036 btrfs_end_transaction(trans, root);
2037 btrfs_btree_balance_dirty(root);
2039 trans = btrfs_start_transaction(root, 3);
2040 if (IS_ERR(trans)) {
2041 ret = PTR_ERR(trans);
2046 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
2048 BUG_ON(ret); /* shouldn't happen */
2049 trans->block_rsv = rsv;
2057 trans->block_rsv = &root->fs_info->trans_block_rsv;
2058 ret = fill_holes(trans, inode, path, cur_offset, drop_end);
2068 inode_inc_iversion(inode);
2069 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2071 trans->block_rsv = &root->fs_info->trans_block_rsv;
2072 ret = btrfs_update_inode(trans, root, inode);
2073 btrfs_end_transaction(trans, root);
2074 btrfs_btree_balance_dirty(root);
2076 btrfs_free_path(path);
2077 btrfs_free_block_rsv(root, rsv);
2079 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2080 &cached_state, GFP_NOFS);
2081 mutex_unlock(&inode->i_mutex);
2087 static long btrfs_fallocate(struct file *file, int mode,
2088 loff_t offset, loff_t len)
2090 struct inode *inode = file->f_path.dentry->d_inode;
2091 struct extent_state *cached_state = NULL;
2098 struct extent_map *em;
2099 int blocksize = BTRFS_I(inode)->root->sectorsize;
2102 alloc_start = round_down(offset, blocksize);
2103 alloc_end = round_up(offset + len, blocksize);
2105 /* Make sure we aren't being give some crap mode */
2106 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2109 if (mode & FALLOC_FL_PUNCH_HOLE)
2110 return btrfs_punch_hole(inode, offset, len);
2113 * Make sure we have enough space before we do the
2116 ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
2121 * wait for ordered IO before we have any locks. We'll loop again
2122 * below with the locks held.
2124 btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
2126 mutex_lock(&inode->i_mutex);
2127 ret = inode_newsize_ok(inode, alloc_end);
2131 if (alloc_start > inode->i_size) {
2132 ret = btrfs_cont_expand(inode, i_size_read(inode),
2138 locked_end = alloc_end - 1;
2140 struct btrfs_ordered_extent *ordered;
2142 /* the extent lock is ordered inside the running
2145 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
2146 locked_end, 0, &cached_state);
2147 ordered = btrfs_lookup_first_ordered_extent(inode,
2150 ordered->file_offset + ordered->len > alloc_start &&
2151 ordered->file_offset < alloc_end) {
2152 btrfs_put_ordered_extent(ordered);
2153 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
2154 alloc_start, locked_end,
2155 &cached_state, GFP_NOFS);
2157 * we can't wait on the range with the transaction
2158 * running or with the extent lock held
2160 btrfs_wait_ordered_range(inode, alloc_start,
2161 alloc_end - alloc_start);
2164 btrfs_put_ordered_extent(ordered);
2169 cur_offset = alloc_start;
2173 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2174 alloc_end - cur_offset, 0);
2175 if (IS_ERR_OR_NULL(em)) {
2182 last_byte = min(extent_map_end(em), alloc_end);
2183 actual_end = min_t(u64, extent_map_end(em), offset + len);
2184 last_byte = ALIGN(last_byte, blocksize);
2186 if (em->block_start == EXTENT_MAP_HOLE ||
2187 (cur_offset >= inode->i_size &&
2188 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
2189 ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
2190 last_byte - cur_offset,
2191 1 << inode->i_blkbits,
2196 free_extent_map(em);
2199 } else if (actual_end > inode->i_size &&
2200 !(mode & FALLOC_FL_KEEP_SIZE)) {
2202 * We didn't need to allocate any more space, but we
2203 * still extended the size of the file so we need to
2206 inode->i_ctime = CURRENT_TIME;
2207 i_size_write(inode, actual_end);
2208 btrfs_ordered_update_i_size(inode, actual_end, NULL);
2210 free_extent_map(em);
2212 cur_offset = last_byte;
2213 if (cur_offset >= alloc_end) {
2218 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
2219 &cached_state, GFP_NOFS);
2221 mutex_unlock(&inode->i_mutex);
2222 /* Let go of our reservation. */
2223 btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
2227 static int find_desired_extent(struct inode *inode, loff_t *offset, int whence)
2229 struct btrfs_root *root = BTRFS_I(inode)->root;
2230 struct extent_map *em;
2231 struct extent_state *cached_state = NULL;
2232 u64 lockstart = *offset;
2233 u64 lockend = i_size_read(inode);
2234 u64 start = *offset;
2235 u64 orig_start = *offset;
2236 u64 len = i_size_read(inode);
2240 lockend = max_t(u64, root->sectorsize, lockend);
2241 if (lockend <= lockstart)
2242 lockend = lockstart + root->sectorsize;
2244 len = lockend - lockstart + 1;
2246 len = max_t(u64, len, root->sectorsize);
2247 if (inode->i_size == 0)
2250 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, 0,
2254 * Delalloc is such a pain. If we have a hole and we have pending
2255 * delalloc for a portion of the hole we will get back a hole that
2256 * exists for the entire range since it hasn't been actually written
2257 * yet. So to take care of this case we need to look for an extent just
2258 * before the position we want in case there is outstanding delalloc
2261 if (whence == SEEK_HOLE && start != 0) {
2262 if (start <= root->sectorsize)
2263 em = btrfs_get_extent_fiemap(inode, NULL, 0, 0,
2264 root->sectorsize, 0);
2266 em = btrfs_get_extent_fiemap(inode, NULL, 0,
2267 start - root->sectorsize,
2268 root->sectorsize, 0);
2273 last_end = em->start + em->len;
2274 if (em->block_start == EXTENT_MAP_DELALLOC)
2275 last_end = min_t(u64, last_end, inode->i_size);
2276 free_extent_map(em);
2280 em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0);
2286 if (em->block_start == EXTENT_MAP_HOLE) {
2287 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2288 if (last_end <= orig_start) {
2289 free_extent_map(em);
2295 if (whence == SEEK_HOLE) {
2297 free_extent_map(em);
2301 if (whence == SEEK_DATA) {
2302 if (em->block_start == EXTENT_MAP_DELALLOC) {
2303 if (start >= inode->i_size) {
2304 free_extent_map(em);
2311 free_extent_map(em);
2316 start = em->start + em->len;
2317 last_end = em->start + em->len;
2319 if (em->block_start == EXTENT_MAP_DELALLOC)
2320 last_end = min_t(u64, last_end, inode->i_size);
2322 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2323 free_extent_map(em);
2327 free_extent_map(em);
2331 *offset = min(*offset, inode->i_size);
2333 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2334 &cached_state, GFP_NOFS);
2338 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
2340 struct inode *inode = file->f_mapping->host;
2343 mutex_lock(&inode->i_mutex);
2347 offset = generic_file_llseek(file, offset, whence);
2351 if (offset >= i_size_read(inode)) {
2352 mutex_unlock(&inode->i_mutex);
2356 ret = find_desired_extent(inode, &offset, whence);
2358 mutex_unlock(&inode->i_mutex);
2363 if (offset < 0 && !(file->f_mode & FMODE_UNSIGNED_OFFSET)) {
2367 if (offset > inode->i_sb->s_maxbytes) {
2372 /* Special lock needed here? */
2373 if (offset != file->f_pos) {
2374 file->f_pos = offset;
2375 file->f_version = 0;
2378 mutex_unlock(&inode->i_mutex);
2382 const struct file_operations btrfs_file_operations = {
2383 .llseek = btrfs_file_llseek,
2384 .read = do_sync_read,
2385 .write = do_sync_write,
2386 .aio_read = generic_file_aio_read,
2387 .splice_read = generic_file_splice_read,
2388 .aio_write = btrfs_file_aio_write,
2389 .mmap = btrfs_file_mmap,
2390 .open = generic_file_open,
2391 .release = btrfs_release_file,
2392 .fsync = btrfs_sync_file,
2393 .fallocate = btrfs_fallocate,
2394 .unlocked_ioctl = btrfs_ioctl,
2395 #ifdef CONFIG_COMPAT
2396 .compat_ioctl = btrfs_ioctl,
2400 void btrfs_auto_defrag_exit(void)
2402 if (btrfs_inode_defrag_cachep)
2403 kmem_cache_destroy(btrfs_inode_defrag_cachep);
2406 int btrfs_auto_defrag_init(void)
2408 btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
2409 sizeof(struct inode_defrag), 0,
2410 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2412 if (!btrfs_inode_defrag_cachep)