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 ret = btrfs_start_nocow_write(root);
1417 lockstart = round_down(pos, root->sectorsize);
1418 lockend = round_up(pos + *write_bytes, root->sectorsize) - 1;
1421 lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1422 ordered = btrfs_lookup_ordered_range(inode, lockstart,
1423 lockend - lockstart + 1);
1427 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1428 btrfs_start_ordered_extent(inode, ordered, 1);
1429 btrfs_put_ordered_extent(ordered);
1432 num_bytes = lockend - lockstart + 1;
1433 ret = can_nocow_extent(inode, lockstart, &num_bytes, NULL, NULL, NULL);
1436 btrfs_end_nocow_write(root);
1438 *write_bytes = min_t(size_t, *write_bytes ,
1439 num_bytes - pos + lockstart);
1442 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1447 static noinline ssize_t __btrfs_buffered_write(struct file *file,
1451 struct inode *inode = file_inode(file);
1452 struct btrfs_root *root = BTRFS_I(inode)->root;
1453 struct page **pages = NULL;
1454 struct extent_state *cached_state = NULL;
1455 u64 release_bytes = 0;
1458 unsigned long first_index;
1459 size_t num_written = 0;
1462 bool only_release_metadata = false;
1463 bool force_page_uptodate = false;
1466 nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) /
1467 PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
1468 (sizeof(struct page *)));
1469 nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1470 nrptrs = max(nrptrs, 8);
1471 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
1475 first_index = pos >> PAGE_CACHE_SHIFT;
1477 while (iov_iter_count(i) > 0) {
1478 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1479 size_t write_bytes = min(iov_iter_count(i),
1480 nrptrs * (size_t)PAGE_CACHE_SIZE -
1482 size_t num_pages = (write_bytes + offset +
1483 PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1484 size_t reserve_bytes;
1488 WARN_ON(num_pages > nrptrs);
1491 * Fault pages before locking them in prepare_pages
1492 * to avoid recursive lock
1494 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1499 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1500 ret = btrfs_check_data_free_space(inode, reserve_bytes);
1501 if (ret == -ENOSPC &&
1502 (BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW |
1503 BTRFS_INODE_PREALLOC))) {
1504 ret = check_can_nocow(inode, pos, &write_bytes);
1506 only_release_metadata = true;
1508 * our prealloc extent may be smaller than
1509 * write_bytes, so scale down.
1511 num_pages = (write_bytes + offset +
1512 PAGE_CACHE_SIZE - 1) >>
1514 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1524 ret = btrfs_delalloc_reserve_metadata(inode, reserve_bytes);
1526 if (!only_release_metadata)
1527 btrfs_free_reserved_data_space(inode,
1530 btrfs_end_nocow_write(root);
1534 release_bytes = reserve_bytes;
1535 need_unlock = false;
1538 * This is going to setup the pages array with the number of
1539 * pages we want, so we don't really need to worry about the
1540 * contents of pages from loop to loop
1542 ret = prepare_pages(inode, pages, num_pages,
1544 force_page_uptodate);
1548 ret = lock_and_cleanup_extent_if_need(inode, pages, num_pages,
1549 pos, &lockstart, &lockend,
1555 } else if (ret > 0) {
1560 copied = btrfs_copy_from_user(pos, num_pages,
1561 write_bytes, pages, i);
1564 * if we have trouble faulting in the pages, fall
1565 * back to one page at a time
1567 if (copied < write_bytes)
1571 force_page_uptodate = true;
1574 force_page_uptodate = false;
1575 dirty_pages = (copied + offset +
1576 PAGE_CACHE_SIZE - 1) >>
1581 * If we had a short copy we need to release the excess delaloc
1582 * bytes we reserved. We need to increment outstanding_extents
1583 * because btrfs_delalloc_release_space will decrement it, but
1584 * we still have an outstanding extent for the chunk we actually
1587 if (num_pages > dirty_pages) {
1588 release_bytes = (num_pages - dirty_pages) <<
1591 spin_lock(&BTRFS_I(inode)->lock);
1592 BTRFS_I(inode)->outstanding_extents++;
1593 spin_unlock(&BTRFS_I(inode)->lock);
1595 if (only_release_metadata)
1596 btrfs_delalloc_release_metadata(inode,
1599 btrfs_delalloc_release_space(inode,
1603 release_bytes = dirty_pages << PAGE_CACHE_SHIFT;
1606 ret = btrfs_dirty_pages(root, inode, pages,
1607 dirty_pages, pos, copied,
1610 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1611 lockstart, lockend, &cached_state,
1614 btrfs_drop_pages(pages, num_pages);
1619 if (only_release_metadata)
1620 btrfs_end_nocow_write(root);
1622 if (only_release_metadata && copied > 0) {
1623 u64 lockstart = round_down(pos, root->sectorsize);
1624 u64 lockend = lockstart +
1625 (dirty_pages << PAGE_CACHE_SHIFT) - 1;
1627 set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
1628 lockend, EXTENT_NORESERVE, NULL,
1630 only_release_metadata = false;
1633 btrfs_drop_pages(pages, num_pages);
1637 balance_dirty_pages_ratelimited(inode->i_mapping);
1638 if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1639 btrfs_btree_balance_dirty(root);
1642 num_written += copied;
1647 if (release_bytes) {
1648 if (only_release_metadata) {
1649 btrfs_end_nocow_write(root);
1650 btrfs_delalloc_release_metadata(inode, release_bytes);
1652 btrfs_delalloc_release_space(inode, release_bytes);
1656 return num_written ? num_written : ret;
1659 static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1660 const struct iovec *iov,
1661 unsigned long nr_segs, loff_t pos,
1662 loff_t *ppos, size_t count, size_t ocount)
1664 struct file *file = iocb->ki_filp;
1667 ssize_t written_buffered;
1671 written = generic_file_direct_write(iocb, iov, &nr_segs, pos, ppos,
1674 if (written < 0 || written == count)
1679 iov_iter_init(&i, iov, nr_segs, count, written);
1680 written_buffered = __btrfs_buffered_write(file, &i, pos);
1681 if (written_buffered < 0) {
1682 err = written_buffered;
1685 endbyte = pos + written_buffered - 1;
1686 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
1689 written += written_buffered;
1690 *ppos = pos + written_buffered;
1691 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
1692 endbyte >> PAGE_CACHE_SHIFT);
1694 return written ? written : err;
1697 static void update_time_for_write(struct inode *inode)
1699 struct timespec now;
1701 if (IS_NOCMTIME(inode))
1704 now = current_fs_time(inode->i_sb);
1705 if (!timespec_equal(&inode->i_mtime, &now))
1706 inode->i_mtime = now;
1708 if (!timespec_equal(&inode->i_ctime, &now))
1709 inode->i_ctime = now;
1711 if (IS_I_VERSION(inode))
1712 inode_inc_iversion(inode);
1715 static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
1716 const struct iovec *iov,
1717 unsigned long nr_segs, loff_t pos)
1719 struct file *file = iocb->ki_filp;
1720 struct inode *inode = file_inode(file);
1721 struct btrfs_root *root = BTRFS_I(inode)->root;
1722 loff_t *ppos = &iocb->ki_pos;
1724 ssize_t num_written = 0;
1726 size_t count, ocount;
1727 bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host);
1729 mutex_lock(&inode->i_mutex);
1731 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
1733 mutex_unlock(&inode->i_mutex);
1738 current->backing_dev_info = inode->i_mapping->backing_dev_info;
1739 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1741 mutex_unlock(&inode->i_mutex);
1746 mutex_unlock(&inode->i_mutex);
1750 err = file_remove_suid(file);
1752 mutex_unlock(&inode->i_mutex);
1757 * If BTRFS flips readonly due to some impossible error
1758 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1759 * although we have opened a file as writable, we have
1760 * to stop this write operation to ensure FS consistency.
1762 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
1763 mutex_unlock(&inode->i_mutex);
1769 * We reserve space for updating the inode when we reserve space for the
1770 * extent we are going to write, so we will enospc out there. We don't
1771 * need to start yet another transaction to update the inode as we will
1772 * update the inode when we finish writing whatever data we write.
1774 update_time_for_write(inode);
1776 start_pos = round_down(pos, root->sectorsize);
1777 if (start_pos > i_size_read(inode)) {
1778 err = btrfs_cont_expand(inode, i_size_read(inode), start_pos);
1780 mutex_unlock(&inode->i_mutex);
1786 atomic_inc(&BTRFS_I(inode)->sync_writers);
1788 if (unlikely(file->f_flags & O_DIRECT)) {
1789 num_written = __btrfs_direct_write(iocb, iov, nr_segs,
1790 pos, ppos, count, ocount);
1794 iov_iter_init(&i, iov, nr_segs, count, num_written);
1796 num_written = __btrfs_buffered_write(file, &i, pos);
1797 if (num_written > 0)
1798 *ppos = pos + num_written;
1801 mutex_unlock(&inode->i_mutex);
1804 * we want to make sure fsync finds this change
1805 * but we haven't joined a transaction running right now.
1807 * Later on, someone is sure to update the inode and get the
1808 * real transid recorded.
1810 * We set last_trans now to the fs_info generation + 1,
1811 * this will either be one more than the running transaction
1812 * or the generation used for the next transaction if there isn't
1813 * one running right now.
1815 * We also have to set last_sub_trans to the current log transid,
1816 * otherwise subsequent syncs to a file that's been synced in this
1817 * transaction will appear to have already occured.
1819 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1820 BTRFS_I(inode)->last_sub_trans = root->log_transid;
1821 if (num_written > 0) {
1822 err = generic_write_sync(file, pos, num_written);
1823 if (err < 0 && num_written > 0)
1828 atomic_dec(&BTRFS_I(inode)->sync_writers);
1830 current->backing_dev_info = NULL;
1831 return num_written ? num_written : err;
1834 int btrfs_release_file(struct inode *inode, struct file *filp)
1837 * ordered_data_close is set by settattr when we are about to truncate
1838 * a file from a non-zero size to a zero size. This tries to
1839 * flush down new bytes that may have been written if the
1840 * application were using truncate to replace a file in place.
1842 if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
1843 &BTRFS_I(inode)->runtime_flags)) {
1844 struct btrfs_trans_handle *trans;
1845 struct btrfs_root *root = BTRFS_I(inode)->root;
1848 * We need to block on a committing transaction to keep us from
1849 * throwing a ordered operation on to the list and causing
1850 * something like sync to deadlock trying to flush out this
1853 trans = btrfs_start_transaction(root, 0);
1855 return PTR_ERR(trans);
1856 btrfs_add_ordered_operation(trans, BTRFS_I(inode)->root, inode);
1857 btrfs_end_transaction(trans, root);
1858 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1859 filemap_flush(inode->i_mapping);
1861 if (filp->private_data)
1862 btrfs_ioctl_trans_end(filp);
1867 * fsync call for both files and directories. This logs the inode into
1868 * the tree log instead of forcing full commits whenever possible.
1870 * It needs to call filemap_fdatawait so that all ordered extent updates are
1871 * in the metadata btree are up to date for copying to the log.
1873 * It drops the inode mutex before doing the tree log commit. This is an
1874 * important optimization for directories because holding the mutex prevents
1875 * new operations on the dir while we write to disk.
1877 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1879 struct dentry *dentry = file->f_path.dentry;
1880 struct inode *inode = dentry->d_inode;
1881 struct btrfs_root *root = BTRFS_I(inode)->root;
1882 struct btrfs_trans_handle *trans;
1883 struct btrfs_log_ctx ctx;
1887 trace_btrfs_sync_file(file, datasync);
1890 * We write the dirty pages in the range and wait until they complete
1891 * out of the ->i_mutex. If so, we can flush the dirty pages by
1892 * multi-task, and make the performance up. See
1893 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1895 atomic_inc(&BTRFS_I(inode)->sync_writers);
1896 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
1897 if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1898 &BTRFS_I(inode)->runtime_flags))
1899 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
1900 atomic_dec(&BTRFS_I(inode)->sync_writers);
1904 mutex_lock(&inode->i_mutex);
1907 * We flush the dirty pages again to avoid some dirty pages in the
1910 atomic_inc(&root->log_batch);
1911 full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1912 &BTRFS_I(inode)->runtime_flags);
1914 ret = btrfs_wait_ordered_range(inode, start, end - start + 1);
1916 mutex_unlock(&inode->i_mutex);
1920 atomic_inc(&root->log_batch);
1923 * check the transaction that last modified this inode
1924 * and see if its already been committed
1926 if (!BTRFS_I(inode)->last_trans) {
1927 mutex_unlock(&inode->i_mutex);
1932 * if the last transaction that changed this file was before
1933 * the current transaction, we can bail out now without any
1937 if (btrfs_inode_in_log(inode, root->fs_info->generation) ||
1938 BTRFS_I(inode)->last_trans <=
1939 root->fs_info->last_trans_committed) {
1940 BTRFS_I(inode)->last_trans = 0;
1943 * We'v had everything committed since the last time we were
1944 * modified so clear this flag in case it was set for whatever
1945 * reason, it's no longer relevant.
1947 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1948 &BTRFS_I(inode)->runtime_flags);
1949 mutex_unlock(&inode->i_mutex);
1954 * ok we haven't committed the transaction yet, lets do a commit
1956 if (file->private_data)
1957 btrfs_ioctl_trans_end(file);
1960 * We use start here because we will need to wait on the IO to complete
1961 * in btrfs_sync_log, which could require joining a transaction (for
1962 * example checking cross references in the nocow path). If we use join
1963 * here we could get into a situation where we're waiting on IO to
1964 * happen that is blocked on a transaction trying to commit. With start
1965 * we inc the extwriter counter, so we wait for all extwriters to exit
1966 * before we start blocking join'ers. This comment is to keep somebody
1967 * from thinking they are super smart and changing this to
1968 * btrfs_join_transaction *cough*Josef*cough*.
1970 trans = btrfs_start_transaction(root, 0);
1971 if (IS_ERR(trans)) {
1972 ret = PTR_ERR(trans);
1973 mutex_unlock(&inode->i_mutex);
1978 btrfs_init_log_ctx(&ctx);
1980 ret = btrfs_log_dentry_safe(trans, root, dentry, &ctx);
1982 /* Fallthrough and commit/free transaction. */
1986 /* we've logged all the items and now have a consistent
1987 * version of the file in the log. It is possible that
1988 * someone will come in and modify the file, but that's
1989 * fine because the log is consistent on disk, and we
1990 * have references to all of the file's extents
1992 * It is possible that someone will come in and log the
1993 * file again, but that will end up using the synchronization
1994 * inside btrfs_sync_log to keep things safe.
1996 mutex_unlock(&inode->i_mutex);
1998 if (ret != BTRFS_NO_LOG_SYNC) {
2000 ret = btrfs_sync_log(trans, root, &ctx);
2002 ret = btrfs_end_transaction(trans, root);
2007 ret = btrfs_wait_ordered_range(inode, start,
2012 ret = btrfs_commit_transaction(trans, root);
2014 ret = btrfs_end_transaction(trans, root);
2017 return ret > 0 ? -EIO : ret;
2020 static const struct vm_operations_struct btrfs_file_vm_ops = {
2021 .fault = filemap_fault,
2022 .page_mkwrite = btrfs_page_mkwrite,
2023 .remap_pages = generic_file_remap_pages,
2026 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
2028 struct address_space *mapping = filp->f_mapping;
2030 if (!mapping->a_ops->readpage)
2033 file_accessed(filp);
2034 vma->vm_ops = &btrfs_file_vm_ops;
2039 static int hole_mergeable(struct inode *inode, struct extent_buffer *leaf,
2040 int slot, u64 start, u64 end)
2042 struct btrfs_file_extent_item *fi;
2043 struct btrfs_key key;
2045 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
2048 btrfs_item_key_to_cpu(leaf, &key, slot);
2049 if (key.objectid != btrfs_ino(inode) ||
2050 key.type != BTRFS_EXTENT_DATA_KEY)
2053 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2055 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2058 if (btrfs_file_extent_disk_bytenr(leaf, fi))
2061 if (key.offset == end)
2063 if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
2068 static int fill_holes(struct btrfs_trans_handle *trans, struct inode *inode,
2069 struct btrfs_path *path, u64 offset, u64 end)
2071 struct btrfs_root *root = BTRFS_I(inode)->root;
2072 struct extent_buffer *leaf;
2073 struct btrfs_file_extent_item *fi;
2074 struct extent_map *hole_em;
2075 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2076 struct btrfs_key key;
2079 if (btrfs_fs_incompat(root->fs_info, NO_HOLES))
2082 key.objectid = btrfs_ino(inode);
2083 key.type = BTRFS_EXTENT_DATA_KEY;
2084 key.offset = offset;
2086 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2091 leaf = path->nodes[0];
2092 if (hole_mergeable(inode, leaf, path->slots[0]-1, offset, end)) {
2096 fi = btrfs_item_ptr(leaf, path->slots[0],
2097 struct btrfs_file_extent_item);
2098 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2100 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2101 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2102 btrfs_set_file_extent_offset(leaf, fi, 0);
2103 btrfs_mark_buffer_dirty(leaf);
2107 if (hole_mergeable(inode, leaf, path->slots[0]+1, offset, end)) {
2111 key.offset = offset;
2112 btrfs_set_item_key_safe(root, path, &key);
2113 fi = btrfs_item_ptr(leaf, path->slots[0],
2114 struct btrfs_file_extent_item);
2115 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2117 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2118 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2119 btrfs_set_file_extent_offset(leaf, fi, 0);
2120 btrfs_mark_buffer_dirty(leaf);
2123 btrfs_release_path(path);
2125 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
2126 0, 0, end - offset, 0, end - offset,
2132 btrfs_release_path(path);
2134 hole_em = alloc_extent_map();
2136 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2137 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2138 &BTRFS_I(inode)->runtime_flags);
2140 hole_em->start = offset;
2141 hole_em->len = end - offset;
2142 hole_em->ram_bytes = hole_em->len;
2143 hole_em->orig_start = offset;
2145 hole_em->block_start = EXTENT_MAP_HOLE;
2146 hole_em->block_len = 0;
2147 hole_em->orig_block_len = 0;
2148 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
2149 hole_em->compress_type = BTRFS_COMPRESS_NONE;
2150 hole_em->generation = trans->transid;
2153 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2154 write_lock(&em_tree->lock);
2155 ret = add_extent_mapping(em_tree, hole_em, 1);
2156 write_unlock(&em_tree->lock);
2157 } while (ret == -EEXIST);
2158 free_extent_map(hole_em);
2160 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2161 &BTRFS_I(inode)->runtime_flags);
2167 static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
2169 struct btrfs_root *root = BTRFS_I(inode)->root;
2170 struct extent_state *cached_state = NULL;
2171 struct btrfs_path *path;
2172 struct btrfs_block_rsv *rsv;
2173 struct btrfs_trans_handle *trans;
2174 u64 lockstart = round_up(offset, BTRFS_I(inode)->root->sectorsize);
2175 u64 lockend = round_down(offset + len,
2176 BTRFS_I(inode)->root->sectorsize) - 1;
2177 u64 cur_offset = lockstart;
2178 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
2183 bool same_page = ((offset >> PAGE_CACHE_SHIFT) ==
2184 ((offset + len - 1) >> PAGE_CACHE_SHIFT));
2185 bool no_holes = btrfs_fs_incompat(root->fs_info, NO_HOLES);
2186 u64 ino_size = round_up(inode->i_size, PAGE_CACHE_SIZE);
2188 ret = btrfs_wait_ordered_range(inode, offset, len);
2192 mutex_lock(&inode->i_mutex);
2194 * We needn't truncate any page which is beyond the end of the file
2195 * because we are sure there is no data there.
2198 * Only do this if we are in the same page and we aren't doing the
2201 if (same_page && len < PAGE_CACHE_SIZE) {
2202 if (offset < ino_size)
2203 ret = btrfs_truncate_page(inode, offset, len, 0);
2204 mutex_unlock(&inode->i_mutex);
2208 /* zero back part of the first page */
2209 if (offset < ino_size) {
2210 ret = btrfs_truncate_page(inode, offset, 0, 0);
2212 mutex_unlock(&inode->i_mutex);
2217 /* zero the front end of the last page */
2218 if (offset + len < ino_size) {
2219 ret = btrfs_truncate_page(inode, offset + len, 0, 1);
2221 mutex_unlock(&inode->i_mutex);
2226 if (lockend < lockstart) {
2227 mutex_unlock(&inode->i_mutex);
2232 struct btrfs_ordered_extent *ordered;
2234 truncate_pagecache_range(inode, lockstart, lockend);
2236 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2238 ordered = btrfs_lookup_first_ordered_extent(inode, lockend);
2241 * We need to make sure we have no ordered extents in this range
2242 * and nobody raced in and read a page in this range, if we did
2243 * we need to try again.
2246 (ordered->file_offset + ordered->len <= lockstart ||
2247 ordered->file_offset > lockend)) &&
2248 !test_range_bit(&BTRFS_I(inode)->io_tree, lockstart,
2249 lockend, EXTENT_UPTODATE, 0,
2252 btrfs_put_ordered_extent(ordered);
2256 btrfs_put_ordered_extent(ordered);
2257 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
2258 lockend, &cached_state, GFP_NOFS);
2259 ret = btrfs_wait_ordered_range(inode, lockstart,
2260 lockend - lockstart + 1);
2262 mutex_unlock(&inode->i_mutex);
2267 path = btrfs_alloc_path();
2273 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2278 rsv->size = btrfs_calc_trunc_metadata_size(root, 1);
2282 * 1 - update the inode
2283 * 1 - removing the extents in the range
2284 * 1 - adding the hole extent if no_holes isn't set
2286 rsv_count = no_holes ? 2 : 3;
2287 trans = btrfs_start_transaction(root, rsv_count);
2288 if (IS_ERR(trans)) {
2289 err = PTR_ERR(trans);
2293 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
2296 trans->block_rsv = rsv;
2298 while (cur_offset < lockend) {
2299 ret = __btrfs_drop_extents(trans, root, inode, path,
2300 cur_offset, lockend + 1,
2301 &drop_end, 1, 0, 0, NULL);
2305 trans->block_rsv = &root->fs_info->trans_block_rsv;
2307 if (cur_offset < ino_size) {
2308 ret = fill_holes(trans, inode, path, cur_offset,
2316 cur_offset = drop_end;
2318 ret = btrfs_update_inode(trans, root, inode);
2324 btrfs_end_transaction(trans, root);
2325 btrfs_btree_balance_dirty(root);
2327 trans = btrfs_start_transaction(root, rsv_count);
2328 if (IS_ERR(trans)) {
2329 ret = PTR_ERR(trans);
2334 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
2336 BUG_ON(ret); /* shouldn't happen */
2337 trans->block_rsv = rsv;
2345 trans->block_rsv = &root->fs_info->trans_block_rsv;
2346 if (cur_offset < ino_size) {
2347 ret = fill_holes(trans, inode, path, cur_offset, drop_end);
2358 inode_inc_iversion(inode);
2359 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2361 trans->block_rsv = &root->fs_info->trans_block_rsv;
2362 ret = btrfs_update_inode(trans, root, inode);
2363 btrfs_end_transaction(trans, root);
2364 btrfs_btree_balance_dirty(root);
2366 btrfs_free_path(path);
2367 btrfs_free_block_rsv(root, rsv);
2369 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2370 &cached_state, GFP_NOFS);
2371 mutex_unlock(&inode->i_mutex);
2377 static long btrfs_fallocate(struct file *file, int mode,
2378 loff_t offset, loff_t len)
2380 struct inode *inode = file_inode(file);
2381 struct extent_state *cached_state = NULL;
2382 struct btrfs_root *root = BTRFS_I(inode)->root;
2389 struct extent_map *em;
2390 int blocksize = BTRFS_I(inode)->root->sectorsize;
2393 alloc_start = round_down(offset, blocksize);
2394 alloc_end = round_up(offset + len, blocksize);
2396 /* Make sure we aren't being give some crap mode */
2397 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2400 if (mode & FALLOC_FL_PUNCH_HOLE)
2401 return btrfs_punch_hole(inode, offset, len);
2404 * Make sure we have enough space before we do the
2407 ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
2410 if (root->fs_info->quota_enabled) {
2411 ret = btrfs_qgroup_reserve(root, alloc_end - alloc_start);
2413 goto out_reserve_fail;
2416 mutex_lock(&inode->i_mutex);
2417 ret = inode_newsize_ok(inode, alloc_end);
2421 if (alloc_start > inode->i_size) {
2422 ret = btrfs_cont_expand(inode, i_size_read(inode),
2428 * If we are fallocating from the end of the file onward we
2429 * need to zero out the end of the page if i_size lands in the
2432 ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
2438 * wait for ordered IO before we have any locks. We'll loop again
2439 * below with the locks held.
2441 ret = btrfs_wait_ordered_range(inode, alloc_start,
2442 alloc_end - alloc_start);
2446 locked_end = alloc_end - 1;
2448 struct btrfs_ordered_extent *ordered;
2450 /* the extent lock is ordered inside the running
2453 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
2454 locked_end, 0, &cached_state);
2455 ordered = btrfs_lookup_first_ordered_extent(inode,
2458 ordered->file_offset + ordered->len > alloc_start &&
2459 ordered->file_offset < alloc_end) {
2460 btrfs_put_ordered_extent(ordered);
2461 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
2462 alloc_start, locked_end,
2463 &cached_state, GFP_NOFS);
2465 * we can't wait on the range with the transaction
2466 * running or with the extent lock held
2468 ret = btrfs_wait_ordered_range(inode, alloc_start,
2469 alloc_end - alloc_start);
2474 btrfs_put_ordered_extent(ordered);
2479 cur_offset = alloc_start;
2483 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2484 alloc_end - cur_offset, 0);
2485 if (IS_ERR_OR_NULL(em)) {
2492 last_byte = min(extent_map_end(em), alloc_end);
2493 actual_end = min_t(u64, extent_map_end(em), offset + len);
2494 last_byte = ALIGN(last_byte, blocksize);
2496 if (em->block_start == EXTENT_MAP_HOLE ||
2497 (cur_offset >= inode->i_size &&
2498 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
2499 ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
2500 last_byte - cur_offset,
2501 1 << inode->i_blkbits,
2506 free_extent_map(em);
2509 } else if (actual_end > inode->i_size &&
2510 !(mode & FALLOC_FL_KEEP_SIZE)) {
2512 * We didn't need to allocate any more space, but we
2513 * still extended the size of the file so we need to
2516 inode->i_ctime = CURRENT_TIME;
2517 i_size_write(inode, actual_end);
2518 btrfs_ordered_update_i_size(inode, actual_end, NULL);
2520 free_extent_map(em);
2522 cur_offset = last_byte;
2523 if (cur_offset >= alloc_end) {
2528 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
2529 &cached_state, GFP_NOFS);
2531 mutex_unlock(&inode->i_mutex);
2532 if (root->fs_info->quota_enabled)
2533 btrfs_qgroup_free(root, alloc_end - alloc_start);
2535 /* Let go of our reservation. */
2536 btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
2540 static int find_desired_extent(struct inode *inode, loff_t *offset, int whence)
2542 struct btrfs_root *root = BTRFS_I(inode)->root;
2543 struct extent_map *em = NULL;
2544 struct extent_state *cached_state = NULL;
2545 u64 lockstart = *offset;
2546 u64 lockend = i_size_read(inode);
2547 u64 start = *offset;
2548 u64 len = i_size_read(inode);
2551 lockend = max_t(u64, root->sectorsize, lockend);
2552 if (lockend <= lockstart)
2553 lockend = lockstart + root->sectorsize;
2556 len = lockend - lockstart + 1;
2558 len = max_t(u64, len, root->sectorsize);
2559 if (inode->i_size == 0)
2562 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, 0,
2565 while (start < inode->i_size) {
2566 em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0);
2573 if (whence == SEEK_HOLE &&
2574 (em->block_start == EXTENT_MAP_HOLE ||
2575 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2577 else if (whence == SEEK_DATA &&
2578 (em->block_start != EXTENT_MAP_HOLE &&
2579 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2582 start = em->start + em->len;
2583 free_extent_map(em);
2587 free_extent_map(em);
2589 if (whence == SEEK_DATA && start >= inode->i_size)
2592 *offset = min_t(loff_t, start, inode->i_size);
2594 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2595 &cached_state, GFP_NOFS);
2599 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
2601 struct inode *inode = file->f_mapping->host;
2604 mutex_lock(&inode->i_mutex);
2608 offset = generic_file_llseek(file, offset, whence);
2612 if (offset >= i_size_read(inode)) {
2613 mutex_unlock(&inode->i_mutex);
2617 ret = find_desired_extent(inode, &offset, whence);
2619 mutex_unlock(&inode->i_mutex);
2624 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2626 mutex_unlock(&inode->i_mutex);
2630 const struct file_operations btrfs_file_operations = {
2631 .llseek = btrfs_file_llseek,
2632 .read = do_sync_read,
2633 .write = do_sync_write,
2634 .aio_read = generic_file_aio_read,
2635 .splice_read = generic_file_splice_read,
2636 .aio_write = btrfs_file_aio_write,
2637 .mmap = btrfs_file_mmap,
2638 .open = generic_file_open,
2639 .release = btrfs_release_file,
2640 .fsync = btrfs_sync_file,
2641 .fallocate = btrfs_fallocate,
2642 .unlocked_ioctl = btrfs_ioctl,
2643 #ifdef CONFIG_COMPAT
2644 .compat_ioctl = btrfs_ioctl,
2648 void btrfs_auto_defrag_exit(void)
2650 if (btrfs_inode_defrag_cachep)
2651 kmem_cache_destroy(btrfs_inode_defrag_cachep);
2654 int btrfs_auto_defrag_init(void)
2656 btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
2657 sizeof(struct inode_defrag), 0,
2658 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2660 if (!btrfs_inode_defrag_cachep)