2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 #include <linux/pagemap.h>
21 #include <linux/highmem.h>
22 #include <linux/time.h>
23 #include <linux/init.h>
24 #include <linux/string.h>
25 #include <linux/backing-dev.h>
26 #include <linux/mpage.h>
27 #include <linux/aio.h>
28 #include <linux/falloc.h>
29 #include <linux/swap.h>
30 #include <linux/writeback.h>
31 #include <linux/statfs.h>
32 #include <linux/compat.h>
33 #include <linux/slab.h>
34 #include <linux/btrfs.h>
37 #include "transaction.h"
38 #include "btrfs_inode.h"
39 #include "print-tree.h"
44 static struct kmem_cache *btrfs_inode_defrag_cachep;
46 * when auto defrag is enabled we
47 * queue up these defrag structs to remember which
48 * inodes need defragging passes
51 struct rb_node rb_node;
55 * transid where the defrag was added, we search for
56 * extents newer than this
63 /* last offset we were able to defrag */
66 /* if we've wrapped around back to zero once already */
70 static int __compare_inode_defrag(struct inode_defrag *defrag1,
71 struct inode_defrag *defrag2)
73 if (defrag1->root > defrag2->root)
75 else if (defrag1->root < defrag2->root)
77 else if (defrag1->ino > defrag2->ino)
79 else if (defrag1->ino < defrag2->ino)
85 /* pop a record for an inode into the defrag tree. The lock
86 * must be held already
88 * If you're inserting a record for an older transid than an
89 * existing record, the transid already in the tree is lowered
91 * If an existing record is found the defrag item you
94 static int __btrfs_add_inode_defrag(struct inode *inode,
95 struct inode_defrag *defrag)
97 struct btrfs_root *root = BTRFS_I(inode)->root;
98 struct inode_defrag *entry;
100 struct rb_node *parent = NULL;
103 p = &root->fs_info->defrag_inodes.rb_node;
106 entry = rb_entry(parent, struct inode_defrag, rb_node);
108 ret = __compare_inode_defrag(defrag, entry);
110 p = &parent->rb_left;
112 p = &parent->rb_right;
114 /* if we're reinserting an entry for
115 * an old defrag run, make sure to
116 * lower the transid of our existing record
118 if (defrag->transid < entry->transid)
119 entry->transid = defrag->transid;
120 if (defrag->last_offset > entry->last_offset)
121 entry->last_offset = defrag->last_offset;
125 set_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
126 rb_link_node(&defrag->rb_node, parent, p);
127 rb_insert_color(&defrag->rb_node, &root->fs_info->defrag_inodes);
131 static inline int __need_auto_defrag(struct btrfs_root *root)
133 if (!btrfs_test_opt(root, AUTO_DEFRAG))
136 if (btrfs_fs_closing(root->fs_info))
143 * insert a defrag record for this inode if auto defrag is
146 int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans,
149 struct btrfs_root *root = BTRFS_I(inode)->root;
150 struct inode_defrag *defrag;
154 if (!__need_auto_defrag(root))
157 if (test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags))
161 transid = trans->transid;
163 transid = BTRFS_I(inode)->root->last_trans;
165 defrag = kmem_cache_zalloc(btrfs_inode_defrag_cachep, GFP_NOFS);
169 defrag->ino = btrfs_ino(inode);
170 defrag->transid = transid;
171 defrag->root = root->root_key.objectid;
173 spin_lock(&root->fs_info->defrag_inodes_lock);
174 if (!test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags)) {
176 * If we set IN_DEFRAG flag and evict the inode from memory,
177 * and then re-read this inode, this new inode doesn't have
178 * IN_DEFRAG flag. At the case, we may find the existed defrag.
180 ret = __btrfs_add_inode_defrag(inode, defrag);
182 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
184 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
186 spin_unlock(&root->fs_info->defrag_inodes_lock);
191 * Requeue the defrag object. If there is a defrag object that points to
192 * the same inode in the tree, we will merge them together (by
193 * __btrfs_add_inode_defrag()) and free the one that we want to requeue.
195 static void btrfs_requeue_inode_defrag(struct inode *inode,
196 struct inode_defrag *defrag)
198 struct btrfs_root *root = BTRFS_I(inode)->root;
201 if (!__need_auto_defrag(root))
205 * Here we don't check the IN_DEFRAG flag, because we need merge
208 spin_lock(&root->fs_info->defrag_inodes_lock);
209 ret = __btrfs_add_inode_defrag(inode, defrag);
210 spin_unlock(&root->fs_info->defrag_inodes_lock);
215 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
219 * pick the defragable inode that we want, if it doesn't exist, we will get
222 static struct inode_defrag *
223 btrfs_pick_defrag_inode(struct btrfs_fs_info *fs_info, u64 root, u64 ino)
225 struct inode_defrag *entry = NULL;
226 struct inode_defrag tmp;
228 struct rb_node *parent = NULL;
234 spin_lock(&fs_info->defrag_inodes_lock);
235 p = fs_info->defrag_inodes.rb_node;
238 entry = rb_entry(parent, struct inode_defrag, rb_node);
240 ret = __compare_inode_defrag(&tmp, entry);
244 p = parent->rb_right;
249 if (parent && __compare_inode_defrag(&tmp, entry) > 0) {
250 parent = rb_next(parent);
252 entry = rb_entry(parent, struct inode_defrag, rb_node);
258 rb_erase(parent, &fs_info->defrag_inodes);
259 spin_unlock(&fs_info->defrag_inodes_lock);
263 void btrfs_cleanup_defrag_inodes(struct btrfs_fs_info *fs_info)
265 struct inode_defrag *defrag;
266 struct rb_node *node;
268 spin_lock(&fs_info->defrag_inodes_lock);
269 node = rb_first(&fs_info->defrag_inodes);
271 rb_erase(node, &fs_info->defrag_inodes);
272 defrag = rb_entry(node, struct inode_defrag, rb_node);
273 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
275 if (need_resched()) {
276 spin_unlock(&fs_info->defrag_inodes_lock);
278 spin_lock(&fs_info->defrag_inodes_lock);
281 node = rb_first(&fs_info->defrag_inodes);
283 spin_unlock(&fs_info->defrag_inodes_lock);
286 #define BTRFS_DEFRAG_BATCH 1024
288 static int __btrfs_run_defrag_inode(struct btrfs_fs_info *fs_info,
289 struct inode_defrag *defrag)
291 struct btrfs_root *inode_root;
293 struct btrfs_key key;
294 struct btrfs_ioctl_defrag_range_args range;
300 key.objectid = defrag->root;
301 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
302 key.offset = (u64)-1;
304 index = srcu_read_lock(&fs_info->subvol_srcu);
306 inode_root = btrfs_read_fs_root_no_name(fs_info, &key);
307 if (IS_ERR(inode_root)) {
308 ret = PTR_ERR(inode_root);
312 key.objectid = defrag->ino;
313 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
315 inode = btrfs_iget(fs_info->sb, &key, inode_root, NULL);
317 ret = PTR_ERR(inode);
320 srcu_read_unlock(&fs_info->subvol_srcu, index);
322 /* do a chunk of defrag */
323 clear_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
324 memset(&range, 0, sizeof(range));
326 range.start = defrag->last_offset;
328 sb_start_write(fs_info->sb);
329 num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid,
331 sb_end_write(fs_info->sb);
333 * if we filled the whole defrag batch, there
334 * must be more work to do. Queue this defrag
337 if (num_defrag == BTRFS_DEFRAG_BATCH) {
338 defrag->last_offset = range.start;
339 btrfs_requeue_inode_defrag(inode, defrag);
340 } else if (defrag->last_offset && !defrag->cycled) {
342 * we didn't fill our defrag batch, but
343 * we didn't start at zero. Make sure we loop
344 * around to the start of the file.
346 defrag->last_offset = 0;
348 btrfs_requeue_inode_defrag(inode, defrag);
350 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
356 srcu_read_unlock(&fs_info->subvol_srcu, index);
357 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
362 * run through the list of inodes in the FS that need
365 int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info)
367 struct inode_defrag *defrag;
369 u64 root_objectid = 0;
371 atomic_inc(&fs_info->defrag_running);
373 /* Pause the auto defragger. */
374 if (test_bit(BTRFS_FS_STATE_REMOUNTING,
378 if (!__need_auto_defrag(fs_info->tree_root))
381 /* find an inode to defrag */
382 defrag = btrfs_pick_defrag_inode(fs_info, root_objectid,
385 if (root_objectid || first_ino) {
394 first_ino = defrag->ino + 1;
395 root_objectid = defrag->root;
397 __btrfs_run_defrag_inode(fs_info, defrag);
399 atomic_dec(&fs_info->defrag_running);
402 * during unmount, we use the transaction_wait queue to
403 * wait for the defragger to stop
405 wake_up(&fs_info->transaction_wait);
409 /* simple helper to fault in pages and copy. This should go away
410 * and be replaced with calls into generic code.
412 static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
414 struct page **prepared_pages,
418 size_t total_copied = 0;
420 int offset = pos & (PAGE_CACHE_SIZE - 1);
422 while (write_bytes > 0) {
423 size_t count = min_t(size_t,
424 PAGE_CACHE_SIZE - offset, write_bytes);
425 struct page *page = prepared_pages[pg];
427 * Copy data from userspace to the current page
429 * Disable pagefault to avoid recursive lock since
430 * the pages are already locked
433 copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
436 /* Flush processor's dcache for this page */
437 flush_dcache_page(page);
440 * if we get a partial write, we can end up with
441 * partially up to date pages. These add
442 * a lot of complexity, so make sure they don't
443 * happen by forcing this copy to be retried.
445 * The rest of the btrfs_file_write code will fall
446 * back to page at a time copies after we return 0.
448 if (!PageUptodate(page) && copied < count)
451 iov_iter_advance(i, copied);
452 write_bytes -= copied;
453 total_copied += copied;
455 /* Return to btrfs_file_aio_write to fault page */
456 if (unlikely(copied == 0))
459 if (unlikely(copied < PAGE_CACHE_SIZE - offset)) {
470 * unlocks pages after btrfs_file_write is done with them
472 static void btrfs_drop_pages(struct page **pages, size_t num_pages)
475 for (i = 0; i < num_pages; i++) {
476 /* page checked is some magic around finding pages that
477 * have been modified without going through btrfs_set_page_dirty
480 ClearPageChecked(pages[i]);
481 unlock_page(pages[i]);
482 mark_page_accessed(pages[i]);
483 page_cache_release(pages[i]);
488 * after copy_from_user, pages need to be dirtied and we need to make
489 * sure holes are created between the current EOF and the start of
490 * any next extents (if required).
492 * this also makes the decision about creating an inline extent vs
493 * doing real data extents, marking pages dirty and delalloc as required.
495 int btrfs_dirty_pages(struct btrfs_root *root, struct inode *inode,
496 struct page **pages, size_t num_pages,
497 loff_t pos, size_t write_bytes,
498 struct extent_state **cached)
504 u64 end_of_last_block;
505 u64 end_pos = pos + write_bytes;
506 loff_t isize = i_size_read(inode);
508 start_pos = pos & ~((u64)root->sectorsize - 1);
509 num_bytes = ALIGN(write_bytes + pos - start_pos, root->sectorsize);
511 end_of_last_block = start_pos + num_bytes - 1;
512 err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
517 for (i = 0; i < num_pages; i++) {
518 struct page *p = pages[i];
525 * we've only changed i_size in ram, and we haven't updated
526 * the disk i_size. There is no need to log the inode
530 i_size_write(inode, end_pos);
535 * this drops all the extents in the cache that intersect the range
536 * [start, end]. Existing extents are split as required.
538 void btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
541 struct extent_map *em;
542 struct extent_map *split = NULL;
543 struct extent_map *split2 = NULL;
544 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
545 u64 len = end - start + 1;
553 WARN_ON(end < start);
554 if (end == (u64)-1) {
563 split = alloc_extent_map();
565 split2 = alloc_extent_map();
566 if (!split || !split2)
569 write_lock(&em_tree->lock);
570 em = lookup_extent_mapping(em_tree, start, len);
572 write_unlock(&em_tree->lock);
576 gen = em->generation;
577 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
578 if (testend && em->start + em->len >= start + len) {
580 write_unlock(&em_tree->lock);
583 start = em->start + em->len;
585 len = start + len - (em->start + em->len);
587 write_unlock(&em_tree->lock);
590 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
591 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
592 clear_bit(EXTENT_FLAG_LOGGING, &flags);
593 modified = !list_empty(&em->list);
597 if (em->start < start) {
598 split->start = em->start;
599 split->len = start - em->start;
601 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
602 split->orig_start = em->orig_start;
603 split->block_start = em->block_start;
606 split->block_len = em->block_len;
608 split->block_len = split->len;
609 split->orig_block_len = max(split->block_len,
611 split->ram_bytes = em->ram_bytes;
613 split->orig_start = split->start;
614 split->block_len = 0;
615 split->block_start = em->block_start;
616 split->orig_block_len = 0;
617 split->ram_bytes = split->len;
620 split->generation = gen;
621 split->bdev = em->bdev;
622 split->flags = flags;
623 split->compress_type = em->compress_type;
624 replace_extent_mapping(em_tree, em, split, modified);
625 free_extent_map(split);
629 if (testend && em->start + em->len > start + len) {
630 u64 diff = start + len - em->start;
632 split->start = start + len;
633 split->len = em->start + em->len - (start + len);
634 split->bdev = em->bdev;
635 split->flags = flags;
636 split->compress_type = em->compress_type;
637 split->generation = gen;
639 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
640 split->orig_block_len = max(em->block_len,
643 split->ram_bytes = em->ram_bytes;
645 split->block_len = em->block_len;
646 split->block_start = em->block_start;
647 split->orig_start = em->orig_start;
649 split->block_len = split->len;
650 split->block_start = em->block_start
652 split->orig_start = em->orig_start;
655 split->ram_bytes = split->len;
656 split->orig_start = split->start;
657 split->block_len = 0;
658 split->block_start = em->block_start;
659 split->orig_block_len = 0;
662 if (extent_map_in_tree(em)) {
663 replace_extent_mapping(em_tree, em, split,
666 ret = add_extent_mapping(em_tree, split,
668 ASSERT(ret == 0); /* Logic error */
670 free_extent_map(split);
674 if (extent_map_in_tree(em))
675 remove_extent_mapping(em_tree, em);
676 write_unlock(&em_tree->lock);
680 /* once for the tree*/
684 free_extent_map(split);
686 free_extent_map(split2);
690 * this is very complex, but the basic idea is to drop all extents
691 * in the range start - end. hint_block is filled in with a block number
692 * that would be a good hint to the block allocator for this file.
694 * If an extent intersects the range but is not entirely inside the range
695 * it is either truncated or split. Anything entirely inside the range
696 * is deleted from the tree.
698 int __btrfs_drop_extents(struct btrfs_trans_handle *trans,
699 struct btrfs_root *root, struct inode *inode,
700 struct btrfs_path *path, u64 start, u64 end,
701 u64 *drop_end, int drop_cache,
703 u32 extent_item_size,
706 struct extent_buffer *leaf;
707 struct btrfs_file_extent_item *fi;
708 struct btrfs_key key;
709 struct btrfs_key new_key;
710 u64 ino = btrfs_ino(inode);
711 u64 search_start = start;
714 u64 extent_offset = 0;
721 int modify_tree = -1;
722 int update_refs = (root->ref_cows || root == root->fs_info->tree_root);
724 int leafs_visited = 0;
727 btrfs_drop_extent_cache(inode, start, end - 1, 0);
729 if (start >= BTRFS_I(inode)->disk_i_size && !replace_extent)
734 ret = btrfs_lookup_file_extent(trans, root, path, ino,
735 search_start, modify_tree);
738 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
739 leaf = path->nodes[0];
740 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
741 if (key.objectid == ino &&
742 key.type == BTRFS_EXTENT_DATA_KEY)
748 leaf = path->nodes[0];
749 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
751 ret = btrfs_next_leaf(root, path);
759 leaf = path->nodes[0];
763 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
764 if (key.objectid > ino ||
765 key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
768 fi = btrfs_item_ptr(leaf, path->slots[0],
769 struct btrfs_file_extent_item);
770 extent_type = btrfs_file_extent_type(leaf, fi);
772 if (extent_type == BTRFS_FILE_EXTENT_REG ||
773 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
774 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
775 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
776 extent_offset = btrfs_file_extent_offset(leaf, fi);
777 extent_end = key.offset +
778 btrfs_file_extent_num_bytes(leaf, fi);
779 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
780 extent_end = key.offset +
781 btrfs_file_extent_inline_len(leaf,
785 extent_end = search_start;
788 if (extent_end <= search_start) {
794 search_start = max(key.offset, start);
795 if (recow || !modify_tree) {
797 btrfs_release_path(path);
802 * | - range to drop - |
803 * | -------- extent -------- |
805 if (start > key.offset && end < extent_end) {
807 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
809 memcpy(&new_key, &key, sizeof(new_key));
810 new_key.offset = start;
811 ret = btrfs_duplicate_item(trans, root, path,
813 if (ret == -EAGAIN) {
814 btrfs_release_path(path);
820 leaf = path->nodes[0];
821 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
822 struct btrfs_file_extent_item);
823 btrfs_set_file_extent_num_bytes(leaf, fi,
826 fi = btrfs_item_ptr(leaf, path->slots[0],
827 struct btrfs_file_extent_item);
829 extent_offset += start - key.offset;
830 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
831 btrfs_set_file_extent_num_bytes(leaf, fi,
833 btrfs_mark_buffer_dirty(leaf);
835 if (update_refs && disk_bytenr > 0) {
836 ret = btrfs_inc_extent_ref(trans, root,
837 disk_bytenr, num_bytes, 0,
838 root->root_key.objectid,
840 start - extent_offset, 0);
841 BUG_ON(ret); /* -ENOMEM */
846 * | ---- range to drop ----- |
847 * | -------- extent -------- |
849 if (start <= key.offset && end < extent_end) {
850 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
852 memcpy(&new_key, &key, sizeof(new_key));
853 new_key.offset = end;
854 btrfs_set_item_key_safe(root, path, &new_key);
856 extent_offset += end - key.offset;
857 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
858 btrfs_set_file_extent_num_bytes(leaf, fi,
860 btrfs_mark_buffer_dirty(leaf);
861 if (update_refs && disk_bytenr > 0)
862 inode_sub_bytes(inode, end - key.offset);
866 search_start = extent_end;
868 * | ---- range to drop ----- |
869 * | -------- extent -------- |
871 if (start > key.offset && end >= extent_end) {
873 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
875 btrfs_set_file_extent_num_bytes(leaf, fi,
877 btrfs_mark_buffer_dirty(leaf);
878 if (update_refs && disk_bytenr > 0)
879 inode_sub_bytes(inode, extent_end - start);
880 if (end == extent_end)
888 * | ---- range to drop ----- |
889 * | ------ extent ------ |
891 if (start <= key.offset && end >= extent_end) {
893 del_slot = path->slots[0];
896 BUG_ON(del_slot + del_nr != path->slots[0]);
901 extent_type == BTRFS_FILE_EXTENT_INLINE) {
902 inode_sub_bytes(inode,
903 extent_end - key.offset);
904 extent_end = ALIGN(extent_end,
906 } else if (update_refs && disk_bytenr > 0) {
907 ret = btrfs_free_extent(trans, root,
908 disk_bytenr, num_bytes, 0,
909 root->root_key.objectid,
910 key.objectid, key.offset -
912 BUG_ON(ret); /* -ENOMEM */
913 inode_sub_bytes(inode,
914 extent_end - key.offset);
917 if (end == extent_end)
920 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
925 ret = btrfs_del_items(trans, root, path, del_slot,
928 btrfs_abort_transaction(trans, root, ret);
935 btrfs_release_path(path);
942 if (!ret && del_nr > 0) {
944 * Set path->slots[0] to first slot, so that after the delete
945 * if items are move off from our leaf to its immediate left or
946 * right neighbor leafs, we end up with a correct and adjusted
947 * path->slots[0] for our insertion (if replace_extent != 0).
949 path->slots[0] = del_slot;
950 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
952 btrfs_abort_transaction(trans, root, ret);
955 leaf = path->nodes[0];
957 * If btrfs_del_items() was called, it might have deleted a leaf, in
958 * which case it unlocked our path, so check path->locks[0] matches a
961 if (!ret && replace_extent && leafs_visited == 1 &&
962 (path->locks[0] == BTRFS_WRITE_LOCK_BLOCKING ||
963 path->locks[0] == BTRFS_WRITE_LOCK) &&
964 btrfs_leaf_free_space(root, leaf) >=
965 sizeof(struct btrfs_item) + extent_item_size) {
968 key.type = BTRFS_EXTENT_DATA_KEY;
970 if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) {
971 struct btrfs_key slot_key;
973 btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]);
974 if (btrfs_comp_cpu_keys(&key, &slot_key) > 0)
977 setup_items_for_insert(root, path, &key,
980 sizeof(struct btrfs_item) +
981 extent_item_size, 1);
985 if (!replace_extent || !(*key_inserted))
986 btrfs_release_path(path);
988 *drop_end = found ? min(end, extent_end) : end;
992 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
993 struct btrfs_root *root, struct inode *inode, u64 start,
994 u64 end, int drop_cache)
996 struct btrfs_path *path;
999 path = btrfs_alloc_path();
1002 ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL,
1003 drop_cache, 0, 0, NULL);
1004 btrfs_free_path(path);
1008 static int extent_mergeable(struct extent_buffer *leaf, int slot,
1009 u64 objectid, u64 bytenr, u64 orig_offset,
1010 u64 *start, u64 *end)
1012 struct btrfs_file_extent_item *fi;
1013 struct btrfs_key key;
1016 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
1019 btrfs_item_key_to_cpu(leaf, &key, slot);
1020 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
1023 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1024 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
1025 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
1026 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
1027 btrfs_file_extent_compression(leaf, fi) ||
1028 btrfs_file_extent_encryption(leaf, fi) ||
1029 btrfs_file_extent_other_encoding(leaf, fi))
1032 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1033 if ((*start && *start != key.offset) || (*end && *end != extent_end))
1036 *start = key.offset;
1042 * Mark extent in the range start - end as written.
1044 * This changes extent type from 'pre-allocated' to 'regular'. If only
1045 * part of extent is marked as written, the extent will be split into
1048 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
1049 struct inode *inode, u64 start, u64 end)
1051 struct btrfs_root *root = BTRFS_I(inode)->root;
1052 struct extent_buffer *leaf;
1053 struct btrfs_path *path;
1054 struct btrfs_file_extent_item *fi;
1055 struct btrfs_key key;
1056 struct btrfs_key new_key;
1068 u64 ino = btrfs_ino(inode);
1070 path = btrfs_alloc_path();
1077 key.type = BTRFS_EXTENT_DATA_KEY;
1080 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1083 if (ret > 0 && path->slots[0] > 0)
1086 leaf = path->nodes[0];
1087 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1088 BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY);
1089 fi = btrfs_item_ptr(leaf, path->slots[0],
1090 struct btrfs_file_extent_item);
1091 BUG_ON(btrfs_file_extent_type(leaf, fi) !=
1092 BTRFS_FILE_EXTENT_PREALLOC);
1093 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1094 BUG_ON(key.offset > start || extent_end < end);
1096 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1097 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1098 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
1099 memcpy(&new_key, &key, sizeof(new_key));
1101 if (start == key.offset && end < extent_end) {
1104 if (extent_mergeable(leaf, path->slots[0] - 1,
1105 ino, bytenr, orig_offset,
1106 &other_start, &other_end)) {
1107 new_key.offset = end;
1108 btrfs_set_item_key_safe(root, path, &new_key);
1109 fi = btrfs_item_ptr(leaf, path->slots[0],
1110 struct btrfs_file_extent_item);
1111 btrfs_set_file_extent_generation(leaf, fi,
1113 btrfs_set_file_extent_num_bytes(leaf, fi,
1115 btrfs_set_file_extent_offset(leaf, fi,
1117 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1118 struct btrfs_file_extent_item);
1119 btrfs_set_file_extent_generation(leaf, fi,
1121 btrfs_set_file_extent_num_bytes(leaf, fi,
1123 btrfs_mark_buffer_dirty(leaf);
1128 if (start > key.offset && end == extent_end) {
1131 if (extent_mergeable(leaf, path->slots[0] + 1,
1132 ino, bytenr, orig_offset,
1133 &other_start, &other_end)) {
1134 fi = btrfs_item_ptr(leaf, path->slots[0],
1135 struct btrfs_file_extent_item);
1136 btrfs_set_file_extent_num_bytes(leaf, fi,
1137 start - key.offset);
1138 btrfs_set_file_extent_generation(leaf, fi,
1141 new_key.offset = start;
1142 btrfs_set_item_key_safe(root, path, &new_key);
1144 fi = btrfs_item_ptr(leaf, path->slots[0],
1145 struct btrfs_file_extent_item);
1146 btrfs_set_file_extent_generation(leaf, fi,
1148 btrfs_set_file_extent_num_bytes(leaf, fi,
1150 btrfs_set_file_extent_offset(leaf, fi,
1151 start - orig_offset);
1152 btrfs_mark_buffer_dirty(leaf);
1157 while (start > key.offset || end < extent_end) {
1158 if (key.offset == start)
1161 new_key.offset = split;
1162 ret = btrfs_duplicate_item(trans, root, path, &new_key);
1163 if (ret == -EAGAIN) {
1164 btrfs_release_path(path);
1168 btrfs_abort_transaction(trans, root, ret);
1172 leaf = path->nodes[0];
1173 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1174 struct btrfs_file_extent_item);
1175 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1176 btrfs_set_file_extent_num_bytes(leaf, fi,
1177 split - key.offset);
1179 fi = btrfs_item_ptr(leaf, path->slots[0],
1180 struct btrfs_file_extent_item);
1182 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1183 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
1184 btrfs_set_file_extent_num_bytes(leaf, fi,
1185 extent_end - split);
1186 btrfs_mark_buffer_dirty(leaf);
1188 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
1189 root->root_key.objectid,
1190 ino, orig_offset, 0);
1191 BUG_ON(ret); /* -ENOMEM */
1193 if (split == start) {
1196 BUG_ON(start != key.offset);
1205 if (extent_mergeable(leaf, path->slots[0] + 1,
1206 ino, bytenr, orig_offset,
1207 &other_start, &other_end)) {
1209 btrfs_release_path(path);
1212 extent_end = other_end;
1213 del_slot = path->slots[0] + 1;
1215 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1216 0, root->root_key.objectid,
1217 ino, orig_offset, 0);
1218 BUG_ON(ret); /* -ENOMEM */
1222 if (extent_mergeable(leaf, path->slots[0] - 1,
1223 ino, bytenr, orig_offset,
1224 &other_start, &other_end)) {
1226 btrfs_release_path(path);
1229 key.offset = other_start;
1230 del_slot = path->slots[0];
1232 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1233 0, root->root_key.objectid,
1234 ino, orig_offset, 0);
1235 BUG_ON(ret); /* -ENOMEM */
1238 fi = btrfs_item_ptr(leaf, path->slots[0],
1239 struct btrfs_file_extent_item);
1240 btrfs_set_file_extent_type(leaf, fi,
1241 BTRFS_FILE_EXTENT_REG);
1242 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1243 btrfs_mark_buffer_dirty(leaf);
1245 fi = btrfs_item_ptr(leaf, del_slot - 1,
1246 struct btrfs_file_extent_item);
1247 btrfs_set_file_extent_type(leaf, fi,
1248 BTRFS_FILE_EXTENT_REG);
1249 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1250 btrfs_set_file_extent_num_bytes(leaf, fi,
1251 extent_end - key.offset);
1252 btrfs_mark_buffer_dirty(leaf);
1254 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1256 btrfs_abort_transaction(trans, root, ret);
1261 btrfs_free_path(path);
1266 * on error we return an unlocked page and the error value
1267 * on success we return a locked page and 0
1269 static int prepare_uptodate_page(struct page *page, u64 pos,
1270 bool force_uptodate)
1274 if (((pos & (PAGE_CACHE_SIZE - 1)) || force_uptodate) &&
1275 !PageUptodate(page)) {
1276 ret = btrfs_readpage(NULL, page);
1280 if (!PageUptodate(page)) {
1289 * this just gets pages into the page cache and locks them down.
1291 static noinline int prepare_pages(struct inode *inode, struct page **pages,
1292 size_t num_pages, loff_t pos,
1293 size_t write_bytes, bool force_uptodate)
1296 unsigned long index = pos >> PAGE_CACHE_SHIFT;
1297 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1301 for (i = 0; i < num_pages; i++) {
1302 pages[i] = find_or_create_page(inode->i_mapping, index + i,
1303 mask | __GFP_WRITE);
1311 err = prepare_uptodate_page(pages[i], pos,
1313 if (i == num_pages - 1)
1314 err = prepare_uptodate_page(pages[i],
1315 pos + write_bytes, false);
1317 page_cache_release(pages[i]);
1321 wait_on_page_writeback(pages[i]);
1326 while (faili >= 0) {
1327 unlock_page(pages[faili]);
1328 page_cache_release(pages[faili]);
1336 * This function locks the extent and properly waits for data=ordered extents
1337 * to finish before allowing the pages to be modified if need.
1340 * 1 - the extent is locked
1341 * 0 - the extent is not locked, and everything is OK
1342 * -EAGAIN - need re-prepare the pages
1343 * the other < 0 number - Something wrong happens
1346 lock_and_cleanup_extent_if_need(struct inode *inode, struct page **pages,
1347 size_t num_pages, loff_t pos,
1348 u64 *lockstart, u64 *lockend,
1349 struct extent_state **cached_state)
1356 start_pos = pos & ~((u64)PAGE_CACHE_SIZE - 1);
1357 last_pos = start_pos + ((u64)num_pages << PAGE_CACHE_SHIFT) - 1;
1359 if (start_pos < inode->i_size) {
1360 struct btrfs_ordered_extent *ordered;
1361 lock_extent_bits(&BTRFS_I(inode)->io_tree,
1362 start_pos, last_pos, 0, cached_state);
1363 ordered = btrfs_lookup_first_ordered_extent(inode, last_pos);
1365 ordered->file_offset + ordered->len > start_pos &&
1366 ordered->file_offset <= last_pos) {
1367 btrfs_put_ordered_extent(ordered);
1368 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1369 start_pos, last_pos,
1370 cached_state, GFP_NOFS);
1371 for (i = 0; i < num_pages; i++) {
1372 unlock_page(pages[i]);
1373 page_cache_release(pages[i]);
1375 ret = btrfs_wait_ordered_range(inode, start_pos,
1376 last_pos - start_pos + 1);
1383 btrfs_put_ordered_extent(ordered);
1385 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
1386 last_pos, EXTENT_DIRTY | EXTENT_DELALLOC |
1387 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
1388 0, 0, cached_state, GFP_NOFS);
1389 *lockstart = start_pos;
1390 *lockend = last_pos;
1394 for (i = 0; i < num_pages; i++) {
1395 if (clear_page_dirty_for_io(pages[i]))
1396 account_page_redirty(pages[i]);
1397 set_page_extent_mapped(pages[i]);
1398 WARN_ON(!PageLocked(pages[i]));
1404 static noinline int check_can_nocow(struct inode *inode, loff_t pos,
1405 size_t *write_bytes)
1407 struct btrfs_root *root = BTRFS_I(inode)->root;
1408 struct btrfs_ordered_extent *ordered;
1409 u64 lockstart, lockend;
1413 lockstart = round_down(pos, root->sectorsize);
1414 lockend = round_up(pos + *write_bytes, root->sectorsize) - 1;
1417 lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1418 ordered = btrfs_lookup_ordered_range(inode, lockstart,
1419 lockend - lockstart + 1);
1423 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1424 btrfs_start_ordered_extent(inode, ordered, 1);
1425 btrfs_put_ordered_extent(ordered);
1428 num_bytes = lockend - lockstart + 1;
1429 ret = can_nocow_extent(inode, lockstart, &num_bytes, NULL, NULL, NULL);
1433 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
1434 EXTENT_DIRTY | EXTENT_DELALLOC |
1435 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1437 *write_bytes = min_t(size_t, *write_bytes ,
1438 num_bytes - pos + lockstart);
1441 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1446 static noinline ssize_t __btrfs_buffered_write(struct file *file,
1450 struct inode *inode = file_inode(file);
1451 struct btrfs_root *root = BTRFS_I(inode)->root;
1452 struct page **pages = NULL;
1453 struct extent_state *cached_state = NULL;
1454 u64 release_bytes = 0;
1457 unsigned long first_index;
1458 size_t num_written = 0;
1461 bool only_release_metadata = false;
1462 bool force_page_uptodate = false;
1465 nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) /
1466 PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
1467 (sizeof(struct page *)));
1468 nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1469 nrptrs = max(nrptrs, 8);
1470 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
1474 first_index = pos >> PAGE_CACHE_SHIFT;
1476 while (iov_iter_count(i) > 0) {
1477 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1478 size_t write_bytes = min(iov_iter_count(i),
1479 nrptrs * (size_t)PAGE_CACHE_SIZE -
1481 size_t num_pages = (write_bytes + offset +
1482 PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1483 size_t reserve_bytes;
1487 WARN_ON(num_pages > nrptrs);
1490 * Fault pages before locking them in prepare_pages
1491 * to avoid recursive lock
1493 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1498 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1499 ret = btrfs_check_data_free_space(inode, reserve_bytes);
1500 if (ret == -ENOSPC &&
1501 (BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW |
1502 BTRFS_INODE_PREALLOC))) {
1503 ret = check_can_nocow(inode, pos, &write_bytes);
1505 only_release_metadata = true;
1507 * our prealloc extent may be smaller than
1508 * write_bytes, so scale down.
1510 num_pages = (write_bytes + offset +
1511 PAGE_CACHE_SIZE - 1) >>
1513 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1523 ret = btrfs_delalloc_reserve_metadata(inode, reserve_bytes);
1525 if (!only_release_metadata)
1526 btrfs_free_reserved_data_space(inode,
1531 release_bytes = reserve_bytes;
1532 need_unlock = false;
1535 * This is going to setup the pages array with the number of
1536 * pages we want, so we don't really need to worry about the
1537 * contents of pages from loop to loop
1539 ret = prepare_pages(inode, pages, num_pages,
1541 force_page_uptodate);
1545 ret = lock_and_cleanup_extent_if_need(inode, pages, num_pages,
1546 pos, &lockstart, &lockend,
1552 } else if (ret > 0) {
1557 copied = btrfs_copy_from_user(pos, num_pages,
1558 write_bytes, pages, i);
1561 * if we have trouble faulting in the pages, fall
1562 * back to one page at a time
1564 if (copied < write_bytes)
1568 force_page_uptodate = true;
1571 force_page_uptodate = false;
1572 dirty_pages = (copied + offset +
1573 PAGE_CACHE_SIZE - 1) >>
1578 * If we had a short copy we need to release the excess delaloc
1579 * bytes we reserved. We need to increment outstanding_extents
1580 * because btrfs_delalloc_release_space will decrement it, but
1581 * we still have an outstanding extent for the chunk we actually
1584 if (num_pages > dirty_pages) {
1585 release_bytes = (num_pages - dirty_pages) <<
1588 spin_lock(&BTRFS_I(inode)->lock);
1589 BTRFS_I(inode)->outstanding_extents++;
1590 spin_unlock(&BTRFS_I(inode)->lock);
1592 if (only_release_metadata)
1593 btrfs_delalloc_release_metadata(inode,
1596 btrfs_delalloc_release_space(inode,
1600 release_bytes = dirty_pages << PAGE_CACHE_SHIFT;
1603 ret = btrfs_dirty_pages(root, inode, pages,
1604 dirty_pages, pos, copied,
1607 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1608 lockstart, lockend, &cached_state,
1611 btrfs_drop_pages(pages, num_pages);
1616 if (only_release_metadata && copied > 0) {
1617 u64 lockstart = round_down(pos, root->sectorsize);
1618 u64 lockend = lockstart +
1619 (dirty_pages << PAGE_CACHE_SHIFT) - 1;
1621 set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
1622 lockend, EXTENT_NORESERVE, NULL,
1624 only_release_metadata = false;
1627 btrfs_drop_pages(pages, num_pages);
1631 balance_dirty_pages_ratelimited(inode->i_mapping);
1632 if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1633 btrfs_btree_balance_dirty(root);
1636 num_written += copied;
1641 if (release_bytes) {
1642 if (only_release_metadata)
1643 btrfs_delalloc_release_metadata(inode, release_bytes);
1645 btrfs_delalloc_release_space(inode, release_bytes);
1648 return num_written ? num_written : ret;
1651 static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1652 const struct iovec *iov,
1653 unsigned long nr_segs, loff_t pos,
1654 loff_t *ppos, size_t count, size_t ocount)
1656 struct file *file = iocb->ki_filp;
1659 ssize_t written_buffered;
1663 written = generic_file_direct_write(iocb, iov, &nr_segs, pos, ppos,
1666 if (written < 0 || written == count)
1671 iov_iter_init(&i, iov, nr_segs, count, written);
1672 written_buffered = __btrfs_buffered_write(file, &i, pos);
1673 if (written_buffered < 0) {
1674 err = written_buffered;
1677 endbyte = pos + written_buffered - 1;
1678 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
1681 written += written_buffered;
1682 *ppos = pos + written_buffered;
1683 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
1684 endbyte >> PAGE_CACHE_SHIFT);
1686 return written ? written : err;
1689 static void update_time_for_write(struct inode *inode)
1691 struct timespec now;
1693 if (IS_NOCMTIME(inode))
1696 now = current_fs_time(inode->i_sb);
1697 if (!timespec_equal(&inode->i_mtime, &now))
1698 inode->i_mtime = now;
1700 if (!timespec_equal(&inode->i_ctime, &now))
1701 inode->i_ctime = now;
1703 if (IS_I_VERSION(inode))
1704 inode_inc_iversion(inode);
1707 static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
1708 const struct iovec *iov,
1709 unsigned long nr_segs, loff_t pos)
1711 struct file *file = iocb->ki_filp;
1712 struct inode *inode = file_inode(file);
1713 struct btrfs_root *root = BTRFS_I(inode)->root;
1714 loff_t *ppos = &iocb->ki_pos;
1716 ssize_t num_written = 0;
1718 size_t count, ocount;
1719 bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host);
1721 mutex_lock(&inode->i_mutex);
1723 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
1725 mutex_unlock(&inode->i_mutex);
1730 current->backing_dev_info = inode->i_mapping->backing_dev_info;
1731 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1733 mutex_unlock(&inode->i_mutex);
1738 mutex_unlock(&inode->i_mutex);
1742 err = file_remove_suid(file);
1744 mutex_unlock(&inode->i_mutex);
1749 * If BTRFS flips readonly due to some impossible error
1750 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1751 * although we have opened a file as writable, we have
1752 * to stop this write operation to ensure FS consistency.
1754 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
1755 mutex_unlock(&inode->i_mutex);
1761 * We reserve space for updating the inode when we reserve space for the
1762 * extent we are going to write, so we will enospc out there. We don't
1763 * need to start yet another transaction to update the inode as we will
1764 * update the inode when we finish writing whatever data we write.
1766 update_time_for_write(inode);
1768 start_pos = round_down(pos, root->sectorsize);
1769 if (start_pos > i_size_read(inode)) {
1770 err = btrfs_cont_expand(inode, i_size_read(inode), start_pos);
1772 mutex_unlock(&inode->i_mutex);
1778 atomic_inc(&BTRFS_I(inode)->sync_writers);
1780 if (unlikely(file->f_flags & O_DIRECT)) {
1781 num_written = __btrfs_direct_write(iocb, iov, nr_segs,
1782 pos, ppos, count, ocount);
1786 iov_iter_init(&i, iov, nr_segs, count, num_written);
1788 num_written = __btrfs_buffered_write(file, &i, pos);
1789 if (num_written > 0)
1790 *ppos = pos + num_written;
1793 mutex_unlock(&inode->i_mutex);
1796 * we want to make sure fsync finds this change
1797 * but we haven't joined a transaction running right now.
1799 * Later on, someone is sure to update the inode and get the
1800 * real transid recorded.
1802 * We set last_trans now to the fs_info generation + 1,
1803 * this will either be one more than the running transaction
1804 * or the generation used for the next transaction if there isn't
1805 * one running right now.
1807 * We also have to set last_sub_trans to the current log transid,
1808 * otherwise subsequent syncs to a file that's been synced in this
1809 * transaction will appear to have already occured.
1811 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1812 BTRFS_I(inode)->last_sub_trans = root->log_transid;
1813 if (num_written > 0) {
1814 err = generic_write_sync(file, pos, num_written);
1815 if (err < 0 && num_written > 0)
1820 atomic_dec(&BTRFS_I(inode)->sync_writers);
1822 current->backing_dev_info = NULL;
1823 return num_written ? num_written : err;
1826 int btrfs_release_file(struct inode *inode, struct file *filp)
1829 * ordered_data_close is set by settattr when we are about to truncate
1830 * a file from a non-zero size to a zero size. This tries to
1831 * flush down new bytes that may have been written if the
1832 * application were using truncate to replace a file in place.
1834 if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
1835 &BTRFS_I(inode)->runtime_flags)) {
1836 struct btrfs_trans_handle *trans;
1837 struct btrfs_root *root = BTRFS_I(inode)->root;
1840 * We need to block on a committing transaction to keep us from
1841 * throwing a ordered operation on to the list and causing
1842 * something like sync to deadlock trying to flush out this
1845 trans = btrfs_start_transaction(root, 0);
1847 return PTR_ERR(trans);
1848 btrfs_add_ordered_operation(trans, BTRFS_I(inode)->root, inode);
1849 btrfs_end_transaction(trans, root);
1850 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1851 filemap_flush(inode->i_mapping);
1853 if (filp->private_data)
1854 btrfs_ioctl_trans_end(filp);
1859 * fsync call for both files and directories. This logs the inode into
1860 * the tree log instead of forcing full commits whenever possible.
1862 * It needs to call filemap_fdatawait so that all ordered extent updates are
1863 * in the metadata btree are up to date for copying to the log.
1865 * It drops the inode mutex before doing the tree log commit. This is an
1866 * important optimization for directories because holding the mutex prevents
1867 * new operations on the dir while we write to disk.
1869 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1871 struct dentry *dentry = file->f_path.dentry;
1872 struct inode *inode = dentry->d_inode;
1873 struct btrfs_root *root = BTRFS_I(inode)->root;
1874 struct btrfs_trans_handle *trans;
1875 struct btrfs_log_ctx ctx;
1879 trace_btrfs_sync_file(file, datasync);
1882 * We write the dirty pages in the range and wait until they complete
1883 * out of the ->i_mutex. If so, we can flush the dirty pages by
1884 * multi-task, and make the performance up. See
1885 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1887 atomic_inc(&BTRFS_I(inode)->sync_writers);
1888 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
1889 if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1890 &BTRFS_I(inode)->runtime_flags))
1891 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
1892 atomic_dec(&BTRFS_I(inode)->sync_writers);
1896 mutex_lock(&inode->i_mutex);
1899 * We flush the dirty pages again to avoid some dirty pages in the
1902 atomic_inc(&root->log_batch);
1903 full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1904 &BTRFS_I(inode)->runtime_flags);
1906 ret = btrfs_wait_ordered_range(inode, start, end - start + 1);
1908 mutex_unlock(&inode->i_mutex);
1912 atomic_inc(&root->log_batch);
1915 * check the transaction that last modified this inode
1916 * and see if its already been committed
1918 if (!BTRFS_I(inode)->last_trans) {
1919 mutex_unlock(&inode->i_mutex);
1924 * if the last transaction that changed this file was before
1925 * the current transaction, we can bail out now without any
1929 if (btrfs_inode_in_log(inode, root->fs_info->generation) ||
1930 BTRFS_I(inode)->last_trans <=
1931 root->fs_info->last_trans_committed) {
1932 BTRFS_I(inode)->last_trans = 0;
1935 * We'v had everything committed since the last time we were
1936 * modified so clear this flag in case it was set for whatever
1937 * reason, it's no longer relevant.
1939 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1940 &BTRFS_I(inode)->runtime_flags);
1941 mutex_unlock(&inode->i_mutex);
1946 * ok we haven't committed the transaction yet, lets do a commit
1948 if (file->private_data)
1949 btrfs_ioctl_trans_end(file);
1952 * We use start here because we will need to wait on the IO to complete
1953 * in btrfs_sync_log, which could require joining a transaction (for
1954 * example checking cross references in the nocow path). If we use join
1955 * here we could get into a situation where we're waiting on IO to
1956 * happen that is blocked on a transaction trying to commit. With start
1957 * we inc the extwriter counter, so we wait for all extwriters to exit
1958 * before we start blocking join'ers. This comment is to keep somebody
1959 * from thinking they are super smart and changing this to
1960 * btrfs_join_transaction *cough*Josef*cough*.
1962 trans = btrfs_start_transaction(root, 0);
1963 if (IS_ERR(trans)) {
1964 ret = PTR_ERR(trans);
1965 mutex_unlock(&inode->i_mutex);
1970 btrfs_init_log_ctx(&ctx);
1972 ret = btrfs_log_dentry_safe(trans, root, dentry, &ctx);
1974 /* Fallthrough and commit/free transaction. */
1978 /* we've logged all the items and now have a consistent
1979 * version of the file in the log. It is possible that
1980 * someone will come in and modify the file, but that's
1981 * fine because the log is consistent on disk, and we
1982 * have references to all of the file's extents
1984 * It is possible that someone will come in and log the
1985 * file again, but that will end up using the synchronization
1986 * inside btrfs_sync_log to keep things safe.
1988 mutex_unlock(&inode->i_mutex);
1990 if (ret != BTRFS_NO_LOG_SYNC) {
1992 ret = btrfs_sync_log(trans, root, &ctx);
1994 ret = btrfs_end_transaction(trans, root);
1999 ret = btrfs_wait_ordered_range(inode, start,
2004 ret = btrfs_commit_transaction(trans, root);
2006 ret = btrfs_end_transaction(trans, root);
2009 return ret > 0 ? -EIO : ret;
2012 static const struct vm_operations_struct btrfs_file_vm_ops = {
2013 .fault = filemap_fault,
2014 .page_mkwrite = btrfs_page_mkwrite,
2015 .remap_pages = generic_file_remap_pages,
2018 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
2020 struct address_space *mapping = filp->f_mapping;
2022 if (!mapping->a_ops->readpage)
2025 file_accessed(filp);
2026 vma->vm_ops = &btrfs_file_vm_ops;
2031 static int hole_mergeable(struct inode *inode, struct extent_buffer *leaf,
2032 int slot, u64 start, u64 end)
2034 struct btrfs_file_extent_item *fi;
2035 struct btrfs_key key;
2037 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
2040 btrfs_item_key_to_cpu(leaf, &key, slot);
2041 if (key.objectid != btrfs_ino(inode) ||
2042 key.type != BTRFS_EXTENT_DATA_KEY)
2045 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2047 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2050 if (btrfs_file_extent_disk_bytenr(leaf, fi))
2053 if (key.offset == end)
2055 if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
2060 static int fill_holes(struct btrfs_trans_handle *trans, struct inode *inode,
2061 struct btrfs_path *path, u64 offset, u64 end)
2063 struct btrfs_root *root = BTRFS_I(inode)->root;
2064 struct extent_buffer *leaf;
2065 struct btrfs_file_extent_item *fi;
2066 struct extent_map *hole_em;
2067 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2068 struct btrfs_key key;
2071 if (btrfs_fs_incompat(root->fs_info, NO_HOLES))
2074 key.objectid = btrfs_ino(inode);
2075 key.type = BTRFS_EXTENT_DATA_KEY;
2076 key.offset = offset;
2078 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2083 leaf = path->nodes[0];
2084 if (hole_mergeable(inode, leaf, path->slots[0]-1, offset, end)) {
2088 fi = btrfs_item_ptr(leaf, path->slots[0],
2089 struct btrfs_file_extent_item);
2090 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2092 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2093 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2094 btrfs_set_file_extent_offset(leaf, fi, 0);
2095 btrfs_mark_buffer_dirty(leaf);
2099 if (hole_mergeable(inode, leaf, path->slots[0]+1, offset, end)) {
2103 key.offset = offset;
2104 btrfs_set_item_key_safe(root, path, &key);
2105 fi = btrfs_item_ptr(leaf, path->slots[0],
2106 struct btrfs_file_extent_item);
2107 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2109 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2110 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2111 btrfs_set_file_extent_offset(leaf, fi, 0);
2112 btrfs_mark_buffer_dirty(leaf);
2115 btrfs_release_path(path);
2117 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
2118 0, 0, end - offset, 0, end - offset,
2124 btrfs_release_path(path);
2126 hole_em = alloc_extent_map();
2128 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2129 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2130 &BTRFS_I(inode)->runtime_flags);
2132 hole_em->start = offset;
2133 hole_em->len = end - offset;
2134 hole_em->ram_bytes = hole_em->len;
2135 hole_em->orig_start = offset;
2137 hole_em->block_start = EXTENT_MAP_HOLE;
2138 hole_em->block_len = 0;
2139 hole_em->orig_block_len = 0;
2140 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
2141 hole_em->compress_type = BTRFS_COMPRESS_NONE;
2142 hole_em->generation = trans->transid;
2145 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2146 write_lock(&em_tree->lock);
2147 ret = add_extent_mapping(em_tree, hole_em, 1);
2148 write_unlock(&em_tree->lock);
2149 } while (ret == -EEXIST);
2150 free_extent_map(hole_em);
2152 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2153 &BTRFS_I(inode)->runtime_flags);
2159 static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
2161 struct btrfs_root *root = BTRFS_I(inode)->root;
2162 struct extent_state *cached_state = NULL;
2163 struct btrfs_path *path;
2164 struct btrfs_block_rsv *rsv;
2165 struct btrfs_trans_handle *trans;
2166 u64 lockstart = round_up(offset, BTRFS_I(inode)->root->sectorsize);
2167 u64 lockend = round_down(offset + len,
2168 BTRFS_I(inode)->root->sectorsize) - 1;
2169 u64 cur_offset = lockstart;
2170 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
2175 bool same_page = ((offset >> PAGE_CACHE_SHIFT) ==
2176 ((offset + len - 1) >> PAGE_CACHE_SHIFT));
2177 bool no_holes = btrfs_fs_incompat(root->fs_info, NO_HOLES);
2178 u64 ino_size = round_up(inode->i_size, PAGE_CACHE_SIZE);
2180 ret = btrfs_wait_ordered_range(inode, offset, len);
2184 mutex_lock(&inode->i_mutex);
2186 * We needn't truncate any page which is beyond the end of the file
2187 * because we are sure there is no data there.
2190 * Only do this if we are in the same page and we aren't doing the
2193 if (same_page && len < PAGE_CACHE_SIZE) {
2194 if (offset < ino_size)
2195 ret = btrfs_truncate_page(inode, offset, len, 0);
2196 mutex_unlock(&inode->i_mutex);
2200 /* zero back part of the first page */
2201 if (offset < ino_size) {
2202 ret = btrfs_truncate_page(inode, offset, 0, 0);
2204 mutex_unlock(&inode->i_mutex);
2209 /* zero the front end of the last page */
2210 if (offset + len < ino_size) {
2211 ret = btrfs_truncate_page(inode, offset + len, 0, 1);
2213 mutex_unlock(&inode->i_mutex);
2218 if (lockend < lockstart) {
2219 mutex_unlock(&inode->i_mutex);
2224 struct btrfs_ordered_extent *ordered;
2226 truncate_pagecache_range(inode, lockstart, lockend);
2228 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2230 ordered = btrfs_lookup_first_ordered_extent(inode, lockend);
2233 * We need to make sure we have no ordered extents in this range
2234 * and nobody raced in and read a page in this range, if we did
2235 * we need to try again.
2238 (ordered->file_offset + ordered->len <= lockstart ||
2239 ordered->file_offset > lockend)) &&
2240 !test_range_bit(&BTRFS_I(inode)->io_tree, lockstart,
2241 lockend, EXTENT_UPTODATE, 0,
2244 btrfs_put_ordered_extent(ordered);
2248 btrfs_put_ordered_extent(ordered);
2249 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
2250 lockend, &cached_state, GFP_NOFS);
2251 ret = btrfs_wait_ordered_range(inode, lockstart,
2252 lockend - lockstart + 1);
2254 mutex_unlock(&inode->i_mutex);
2259 path = btrfs_alloc_path();
2265 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2270 rsv->size = btrfs_calc_trunc_metadata_size(root, 1);
2274 * 1 - update the inode
2275 * 1 - removing the extents in the range
2276 * 1 - adding the hole extent if no_holes isn't set
2278 rsv_count = no_holes ? 2 : 3;
2279 trans = btrfs_start_transaction(root, rsv_count);
2280 if (IS_ERR(trans)) {
2281 err = PTR_ERR(trans);
2285 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
2288 trans->block_rsv = rsv;
2290 while (cur_offset < lockend) {
2291 ret = __btrfs_drop_extents(trans, root, inode, path,
2292 cur_offset, lockend + 1,
2293 &drop_end, 1, 0, 0, NULL);
2297 trans->block_rsv = &root->fs_info->trans_block_rsv;
2299 if (cur_offset < ino_size) {
2300 ret = fill_holes(trans, inode, path, cur_offset,
2308 cur_offset = drop_end;
2310 ret = btrfs_update_inode(trans, root, inode);
2316 btrfs_end_transaction(trans, root);
2317 btrfs_btree_balance_dirty(root);
2319 trans = btrfs_start_transaction(root, rsv_count);
2320 if (IS_ERR(trans)) {
2321 ret = PTR_ERR(trans);
2326 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
2328 BUG_ON(ret); /* shouldn't happen */
2329 trans->block_rsv = rsv;
2337 trans->block_rsv = &root->fs_info->trans_block_rsv;
2338 if (cur_offset < ino_size) {
2339 ret = fill_holes(trans, inode, path, cur_offset, drop_end);
2350 inode_inc_iversion(inode);
2351 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2353 trans->block_rsv = &root->fs_info->trans_block_rsv;
2354 ret = btrfs_update_inode(trans, root, inode);
2355 btrfs_end_transaction(trans, root);
2356 btrfs_btree_balance_dirty(root);
2358 btrfs_free_path(path);
2359 btrfs_free_block_rsv(root, rsv);
2361 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2362 &cached_state, GFP_NOFS);
2363 mutex_unlock(&inode->i_mutex);
2369 static long btrfs_fallocate(struct file *file, int mode,
2370 loff_t offset, loff_t len)
2372 struct inode *inode = file_inode(file);
2373 struct extent_state *cached_state = NULL;
2374 struct btrfs_root *root = BTRFS_I(inode)->root;
2381 struct extent_map *em;
2382 int blocksize = BTRFS_I(inode)->root->sectorsize;
2385 alloc_start = round_down(offset, blocksize);
2386 alloc_end = round_up(offset + len, blocksize);
2388 /* Make sure we aren't being give some crap mode */
2389 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2392 if (mode & FALLOC_FL_PUNCH_HOLE)
2393 return btrfs_punch_hole(inode, offset, len);
2396 * Make sure we have enough space before we do the
2399 ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
2402 if (root->fs_info->quota_enabled) {
2403 ret = btrfs_qgroup_reserve(root, alloc_end - alloc_start);
2405 goto out_reserve_fail;
2408 mutex_lock(&inode->i_mutex);
2409 ret = inode_newsize_ok(inode, alloc_end);
2413 if (alloc_start > inode->i_size) {
2414 ret = btrfs_cont_expand(inode, i_size_read(inode),
2420 * If we are fallocating from the end of the file onward we
2421 * need to zero out the end of the page if i_size lands in the
2424 ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
2430 * wait for ordered IO before we have any locks. We'll loop again
2431 * below with the locks held.
2433 ret = btrfs_wait_ordered_range(inode, alloc_start,
2434 alloc_end - alloc_start);
2438 locked_end = alloc_end - 1;
2440 struct btrfs_ordered_extent *ordered;
2442 /* the extent lock is ordered inside the running
2445 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
2446 locked_end, 0, &cached_state);
2447 ordered = btrfs_lookup_first_ordered_extent(inode,
2450 ordered->file_offset + ordered->len > alloc_start &&
2451 ordered->file_offset < alloc_end) {
2452 btrfs_put_ordered_extent(ordered);
2453 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
2454 alloc_start, locked_end,
2455 &cached_state, GFP_NOFS);
2457 * we can't wait on the range with the transaction
2458 * running or with the extent lock held
2460 ret = btrfs_wait_ordered_range(inode, alloc_start,
2461 alloc_end - alloc_start);
2466 btrfs_put_ordered_extent(ordered);
2471 cur_offset = alloc_start;
2475 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2476 alloc_end - cur_offset, 0);
2477 if (IS_ERR_OR_NULL(em)) {
2484 last_byte = min(extent_map_end(em), alloc_end);
2485 actual_end = min_t(u64, extent_map_end(em), offset + len);
2486 last_byte = ALIGN(last_byte, blocksize);
2488 if (em->block_start == EXTENT_MAP_HOLE ||
2489 (cur_offset >= inode->i_size &&
2490 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
2491 ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
2492 last_byte - cur_offset,
2493 1 << inode->i_blkbits,
2498 free_extent_map(em);
2501 } else if (actual_end > inode->i_size &&
2502 !(mode & FALLOC_FL_KEEP_SIZE)) {
2504 * We didn't need to allocate any more space, but we
2505 * still extended the size of the file so we need to
2508 inode->i_ctime = CURRENT_TIME;
2509 i_size_write(inode, actual_end);
2510 btrfs_ordered_update_i_size(inode, actual_end, NULL);
2512 free_extent_map(em);
2514 cur_offset = last_byte;
2515 if (cur_offset >= alloc_end) {
2520 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
2521 &cached_state, GFP_NOFS);
2523 mutex_unlock(&inode->i_mutex);
2524 if (root->fs_info->quota_enabled)
2525 btrfs_qgroup_free(root, alloc_end - alloc_start);
2527 /* Let go of our reservation. */
2528 btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
2532 static int find_desired_extent(struct inode *inode, loff_t *offset, int whence)
2534 struct btrfs_root *root = BTRFS_I(inode)->root;
2535 struct extent_map *em = NULL;
2536 struct extent_state *cached_state = NULL;
2537 u64 lockstart = *offset;
2538 u64 lockend = i_size_read(inode);
2539 u64 start = *offset;
2540 u64 len = i_size_read(inode);
2543 lockend = max_t(u64, root->sectorsize, lockend);
2544 if (lockend <= lockstart)
2545 lockend = lockstart + root->sectorsize;
2548 len = lockend - lockstart + 1;
2550 len = max_t(u64, len, root->sectorsize);
2551 if (inode->i_size == 0)
2554 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, 0,
2557 while (start < inode->i_size) {
2558 em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0);
2565 if (whence == SEEK_HOLE &&
2566 (em->block_start == EXTENT_MAP_HOLE ||
2567 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2569 else if (whence == SEEK_DATA &&
2570 (em->block_start != EXTENT_MAP_HOLE &&
2571 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2574 start = em->start + em->len;
2575 free_extent_map(em);
2579 free_extent_map(em);
2581 if (whence == SEEK_DATA && start >= inode->i_size)
2584 *offset = min_t(loff_t, start, inode->i_size);
2586 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2587 &cached_state, GFP_NOFS);
2591 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
2593 struct inode *inode = file->f_mapping->host;
2596 mutex_lock(&inode->i_mutex);
2600 offset = generic_file_llseek(file, offset, whence);
2604 if (offset >= i_size_read(inode)) {
2605 mutex_unlock(&inode->i_mutex);
2609 ret = find_desired_extent(inode, &offset, whence);
2611 mutex_unlock(&inode->i_mutex);
2616 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2618 mutex_unlock(&inode->i_mutex);
2622 const struct file_operations btrfs_file_operations = {
2623 .llseek = btrfs_file_llseek,
2624 .read = do_sync_read,
2625 .write = do_sync_write,
2626 .aio_read = generic_file_aio_read,
2627 .splice_read = generic_file_splice_read,
2628 .aio_write = btrfs_file_aio_write,
2629 .mmap = btrfs_file_mmap,
2630 .open = generic_file_open,
2631 .release = btrfs_release_file,
2632 .fsync = btrfs_sync_file,
2633 .fallocate = btrfs_fallocate,
2634 .unlocked_ioctl = btrfs_ioctl,
2635 #ifdef CONFIG_COMPAT
2636 .compat_ioctl = btrfs_ioctl,
2640 void btrfs_auto_defrag_exit(void)
2642 if (btrfs_inode_defrag_cachep)
2643 kmem_cache_destroy(btrfs_inode_defrag_cachep);
2646 int btrfs_auto_defrag_init(void)
2648 btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
2649 sizeof(struct inode_defrag), 0,
2650 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2652 if (!btrfs_inode_defrag_cachep)