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 copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
431 /* Flush processor's dcache for this page */
432 flush_dcache_page(page);
435 * if we get a partial write, we can end up with
436 * partially up to date pages. These add
437 * a lot of complexity, so make sure they don't
438 * happen by forcing this copy to be retried.
440 * The rest of the btrfs_file_write code will fall
441 * back to page at a time copies after we return 0.
443 if (!PageUptodate(page) && copied < count)
446 iov_iter_advance(i, copied);
447 write_bytes -= copied;
448 total_copied += copied;
450 /* Return to btrfs_file_aio_write to fault page */
451 if (unlikely(copied == 0))
454 if (unlikely(copied < PAGE_CACHE_SIZE - offset)) {
465 * unlocks pages after btrfs_file_write is done with them
467 static void btrfs_drop_pages(struct page **pages, size_t num_pages)
470 for (i = 0; i < num_pages; i++) {
471 /* page checked is some magic around finding pages that
472 * have been modified without going through btrfs_set_page_dirty
475 ClearPageChecked(pages[i]);
476 unlock_page(pages[i]);
477 mark_page_accessed(pages[i]);
478 page_cache_release(pages[i]);
483 * after copy_from_user, pages need to be dirtied and we need to make
484 * sure holes are created between the current EOF and the start of
485 * any next extents (if required).
487 * this also makes the decision about creating an inline extent vs
488 * doing real data extents, marking pages dirty and delalloc as required.
490 int btrfs_dirty_pages(struct btrfs_root *root, struct inode *inode,
491 struct page **pages, size_t num_pages,
492 loff_t pos, size_t write_bytes,
493 struct extent_state **cached)
499 u64 end_of_last_block;
500 u64 end_pos = pos + write_bytes;
501 loff_t isize = i_size_read(inode);
503 start_pos = pos & ~((u64)root->sectorsize - 1);
504 num_bytes = ALIGN(write_bytes + pos - start_pos, root->sectorsize);
506 end_of_last_block = start_pos + num_bytes - 1;
507 err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
512 for (i = 0; i < num_pages; i++) {
513 struct page *p = pages[i];
520 * we've only changed i_size in ram, and we haven't updated
521 * the disk i_size. There is no need to log the inode
525 i_size_write(inode, end_pos);
530 * this drops all the extents in the cache that intersect the range
531 * [start, end]. Existing extents are split as required.
533 void btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
536 struct extent_map *em;
537 struct extent_map *split = NULL;
538 struct extent_map *split2 = NULL;
539 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
540 u64 len = end - start + 1;
548 WARN_ON(end < start);
549 if (end == (u64)-1) {
558 split = alloc_extent_map();
560 split2 = alloc_extent_map();
561 if (!split || !split2)
564 write_lock(&em_tree->lock);
565 em = lookup_extent_mapping(em_tree, start, len);
567 write_unlock(&em_tree->lock);
571 gen = em->generation;
572 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
573 if (testend && em->start + em->len >= start + len) {
575 write_unlock(&em_tree->lock);
578 start = em->start + em->len;
580 len = start + len - (em->start + em->len);
582 write_unlock(&em_tree->lock);
585 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
586 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
587 clear_bit(EXTENT_FLAG_LOGGING, &flags);
588 modified = !list_empty(&em->list);
589 remove_extent_mapping(em_tree, em);
593 if (em->start < start) {
594 split->start = em->start;
595 split->len = start - em->start;
597 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
598 split->orig_start = em->orig_start;
599 split->block_start = em->block_start;
602 split->block_len = em->block_len;
604 split->block_len = split->len;
605 split->orig_block_len = max(split->block_len,
607 split->ram_bytes = em->ram_bytes;
609 split->orig_start = split->start;
610 split->block_len = 0;
611 split->block_start = em->block_start;
612 split->orig_block_len = 0;
613 split->ram_bytes = split->len;
616 split->generation = gen;
617 split->bdev = em->bdev;
618 split->flags = flags;
619 split->compress_type = em->compress_type;
620 ret = add_extent_mapping(em_tree, split, modified);
621 BUG_ON(ret); /* Logic error */
622 free_extent_map(split);
626 if (testend && em->start + em->len > start + len) {
627 u64 diff = start + len - em->start;
629 split->start = start + len;
630 split->len = em->start + em->len - (start + len);
631 split->bdev = em->bdev;
632 split->flags = flags;
633 split->compress_type = em->compress_type;
634 split->generation = gen;
636 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
637 split->orig_block_len = max(em->block_len,
640 split->ram_bytes = em->ram_bytes;
642 split->block_len = em->block_len;
643 split->block_start = em->block_start;
644 split->orig_start = em->orig_start;
646 split->block_len = split->len;
647 split->block_start = em->block_start
649 split->orig_start = em->orig_start;
652 split->ram_bytes = split->len;
653 split->orig_start = split->start;
654 split->block_len = 0;
655 split->block_start = em->block_start;
656 split->orig_block_len = 0;
659 ret = add_extent_mapping(em_tree, split, modified);
660 BUG_ON(ret); /* Logic error */
661 free_extent_map(split);
665 write_unlock(&em_tree->lock);
669 /* once for the tree*/
673 free_extent_map(split);
675 free_extent_map(split2);
679 * this is very complex, but the basic idea is to drop all extents
680 * in the range start - end. hint_block is filled in with a block number
681 * that would be a good hint to the block allocator for this file.
683 * If an extent intersects the range but is not entirely inside the range
684 * it is either truncated or split. Anything entirely inside the range
685 * is deleted from the tree.
687 int __btrfs_drop_extents(struct btrfs_trans_handle *trans,
688 struct btrfs_root *root, struct inode *inode,
689 struct btrfs_path *path, u64 start, u64 end,
690 u64 *drop_end, int drop_cache,
692 u32 extent_item_size,
695 struct extent_buffer *leaf;
696 struct btrfs_file_extent_item *fi;
697 struct btrfs_key key;
698 struct btrfs_key new_key;
699 u64 ino = btrfs_ino(inode);
700 u64 search_start = start;
703 u64 extent_offset = 0;
710 int modify_tree = -1;
711 int update_refs = (root->ref_cows || root == root->fs_info->tree_root);
713 int leafs_visited = 0;
716 btrfs_drop_extent_cache(inode, start, end - 1, 0);
718 if (start >= BTRFS_I(inode)->disk_i_size)
723 ret = btrfs_lookup_file_extent(trans, root, path, ino,
724 search_start, modify_tree);
727 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
728 leaf = path->nodes[0];
729 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
730 if (key.objectid == ino &&
731 key.type == BTRFS_EXTENT_DATA_KEY)
737 leaf = path->nodes[0];
738 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
740 ret = btrfs_next_leaf(root, path);
748 leaf = path->nodes[0];
752 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
753 if (key.objectid > ino ||
754 key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
757 fi = btrfs_item_ptr(leaf, path->slots[0],
758 struct btrfs_file_extent_item);
759 extent_type = btrfs_file_extent_type(leaf, fi);
761 if (extent_type == BTRFS_FILE_EXTENT_REG ||
762 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
763 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
764 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
765 extent_offset = btrfs_file_extent_offset(leaf, fi);
766 extent_end = key.offset +
767 btrfs_file_extent_num_bytes(leaf, fi);
768 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
769 extent_end = key.offset +
770 btrfs_file_extent_inline_len(leaf,
774 extent_end = search_start;
777 if (extent_end <= search_start) {
783 search_start = max(key.offset, start);
784 if (recow || !modify_tree) {
786 btrfs_release_path(path);
791 * | - range to drop - |
792 * | -------- extent -------- |
794 if (start > key.offset && end < extent_end) {
796 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
798 memcpy(&new_key, &key, sizeof(new_key));
799 new_key.offset = start;
800 ret = btrfs_duplicate_item(trans, root, path,
802 if (ret == -EAGAIN) {
803 btrfs_release_path(path);
809 leaf = path->nodes[0];
810 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
811 struct btrfs_file_extent_item);
812 btrfs_set_file_extent_num_bytes(leaf, fi,
815 fi = btrfs_item_ptr(leaf, path->slots[0],
816 struct btrfs_file_extent_item);
818 extent_offset += start - key.offset;
819 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
820 btrfs_set_file_extent_num_bytes(leaf, fi,
822 btrfs_mark_buffer_dirty(leaf);
824 if (update_refs && disk_bytenr > 0) {
825 ret = btrfs_inc_extent_ref(trans, root,
826 disk_bytenr, num_bytes, 0,
827 root->root_key.objectid,
829 start - extent_offset, 0);
830 BUG_ON(ret); /* -ENOMEM */
835 * | ---- range to drop ----- |
836 * | -------- extent -------- |
838 if (start <= key.offset && end < extent_end) {
839 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
841 memcpy(&new_key, &key, sizeof(new_key));
842 new_key.offset = end;
843 btrfs_set_item_key_safe(root, path, &new_key);
845 extent_offset += end - key.offset;
846 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
847 btrfs_set_file_extent_num_bytes(leaf, fi,
849 btrfs_mark_buffer_dirty(leaf);
850 if (update_refs && disk_bytenr > 0)
851 inode_sub_bytes(inode, end - key.offset);
855 search_start = extent_end;
857 * | ---- range to drop ----- |
858 * | -------- extent -------- |
860 if (start > key.offset && end >= extent_end) {
862 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
864 btrfs_set_file_extent_num_bytes(leaf, fi,
866 btrfs_mark_buffer_dirty(leaf);
867 if (update_refs && disk_bytenr > 0)
868 inode_sub_bytes(inode, extent_end - start);
869 if (end == extent_end)
877 * | ---- range to drop ----- |
878 * | ------ extent ------ |
880 if (start <= key.offset && end >= extent_end) {
882 del_slot = path->slots[0];
885 BUG_ON(del_slot + del_nr != path->slots[0]);
890 extent_type == BTRFS_FILE_EXTENT_INLINE) {
891 inode_sub_bytes(inode,
892 extent_end - key.offset);
893 extent_end = ALIGN(extent_end,
895 } else if (update_refs && disk_bytenr > 0) {
896 ret = btrfs_free_extent(trans, root,
897 disk_bytenr, num_bytes, 0,
898 root->root_key.objectid,
899 key.objectid, key.offset -
901 BUG_ON(ret); /* -ENOMEM */
902 inode_sub_bytes(inode,
903 extent_end - key.offset);
906 if (end == extent_end)
909 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
914 ret = btrfs_del_items(trans, root, path, del_slot,
917 btrfs_abort_transaction(trans, root, ret);
924 btrfs_release_path(path);
931 if (!ret && del_nr > 0) {
933 * Set path->slots[0] to first slot, so that after the delete
934 * if items are move off from our leaf to its immediate left or
935 * right neighbor leafs, we end up with a correct and adjusted
936 * path->slots[0] for our insertion.
938 path->slots[0] = del_slot;
939 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
941 btrfs_abort_transaction(trans, root, ret);
943 leaf = path->nodes[0];
945 * leaf eb has flag EXTENT_BUFFER_STALE if it was deleted (that
946 * is, its contents got pushed to its neighbors), in which case
947 * it means path->locks[0] == 0
949 if (!ret && replace_extent && leafs_visited == 1 &&
951 btrfs_leaf_free_space(root, leaf) >=
952 sizeof(struct btrfs_item) + extent_item_size) {
955 key.type = BTRFS_EXTENT_DATA_KEY;
957 setup_items_for_insert(root, path, &key,
960 sizeof(struct btrfs_item) +
961 extent_item_size, 1);
966 if (!replace_extent || !(*key_inserted))
967 btrfs_release_path(path);
969 *drop_end = found ? min(end, extent_end) : end;
973 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
974 struct btrfs_root *root, struct inode *inode, u64 start,
975 u64 end, int drop_cache)
977 struct btrfs_path *path;
980 path = btrfs_alloc_path();
983 ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL,
984 drop_cache, 0, 0, NULL);
985 btrfs_free_path(path);
989 static int extent_mergeable(struct extent_buffer *leaf, int slot,
990 u64 objectid, u64 bytenr, u64 orig_offset,
991 u64 *start, u64 *end)
993 struct btrfs_file_extent_item *fi;
994 struct btrfs_key key;
997 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
1000 btrfs_item_key_to_cpu(leaf, &key, slot);
1001 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
1004 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1005 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
1006 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
1007 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
1008 btrfs_file_extent_compression(leaf, fi) ||
1009 btrfs_file_extent_encryption(leaf, fi) ||
1010 btrfs_file_extent_other_encoding(leaf, fi))
1013 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1014 if ((*start && *start != key.offset) || (*end && *end != extent_end))
1017 *start = key.offset;
1023 * Mark extent in the range start - end as written.
1025 * This changes extent type from 'pre-allocated' to 'regular'. If only
1026 * part of extent is marked as written, the extent will be split into
1029 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
1030 struct inode *inode, u64 start, u64 end)
1032 struct btrfs_root *root = BTRFS_I(inode)->root;
1033 struct extent_buffer *leaf;
1034 struct btrfs_path *path;
1035 struct btrfs_file_extent_item *fi;
1036 struct btrfs_key key;
1037 struct btrfs_key new_key;
1049 u64 ino = btrfs_ino(inode);
1051 path = btrfs_alloc_path();
1058 key.type = BTRFS_EXTENT_DATA_KEY;
1061 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1064 if (ret > 0 && path->slots[0] > 0)
1067 leaf = path->nodes[0];
1068 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1069 BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY);
1070 fi = btrfs_item_ptr(leaf, path->slots[0],
1071 struct btrfs_file_extent_item);
1072 BUG_ON(btrfs_file_extent_type(leaf, fi) !=
1073 BTRFS_FILE_EXTENT_PREALLOC);
1074 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1075 BUG_ON(key.offset > start || extent_end < end);
1077 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1078 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1079 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
1080 memcpy(&new_key, &key, sizeof(new_key));
1082 if (start == key.offset && end < extent_end) {
1085 if (extent_mergeable(leaf, path->slots[0] - 1,
1086 ino, bytenr, orig_offset,
1087 &other_start, &other_end)) {
1088 new_key.offset = end;
1089 btrfs_set_item_key_safe(root, path, &new_key);
1090 fi = btrfs_item_ptr(leaf, path->slots[0],
1091 struct btrfs_file_extent_item);
1092 btrfs_set_file_extent_generation(leaf, fi,
1094 btrfs_set_file_extent_num_bytes(leaf, fi,
1096 btrfs_set_file_extent_offset(leaf, fi,
1098 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1099 struct btrfs_file_extent_item);
1100 btrfs_set_file_extent_generation(leaf, fi,
1102 btrfs_set_file_extent_num_bytes(leaf, fi,
1104 btrfs_mark_buffer_dirty(leaf);
1109 if (start > key.offset && end == extent_end) {
1112 if (extent_mergeable(leaf, path->slots[0] + 1,
1113 ino, bytenr, orig_offset,
1114 &other_start, &other_end)) {
1115 fi = btrfs_item_ptr(leaf, path->slots[0],
1116 struct btrfs_file_extent_item);
1117 btrfs_set_file_extent_num_bytes(leaf, fi,
1118 start - key.offset);
1119 btrfs_set_file_extent_generation(leaf, fi,
1122 new_key.offset = start;
1123 btrfs_set_item_key_safe(root, path, &new_key);
1125 fi = btrfs_item_ptr(leaf, path->slots[0],
1126 struct btrfs_file_extent_item);
1127 btrfs_set_file_extent_generation(leaf, fi,
1129 btrfs_set_file_extent_num_bytes(leaf, fi,
1131 btrfs_set_file_extent_offset(leaf, fi,
1132 start - orig_offset);
1133 btrfs_mark_buffer_dirty(leaf);
1138 while (start > key.offset || end < extent_end) {
1139 if (key.offset == start)
1142 new_key.offset = split;
1143 ret = btrfs_duplicate_item(trans, root, path, &new_key);
1144 if (ret == -EAGAIN) {
1145 btrfs_release_path(path);
1149 btrfs_abort_transaction(trans, root, ret);
1153 leaf = path->nodes[0];
1154 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1155 struct btrfs_file_extent_item);
1156 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1157 btrfs_set_file_extent_num_bytes(leaf, fi,
1158 split - key.offset);
1160 fi = btrfs_item_ptr(leaf, path->slots[0],
1161 struct btrfs_file_extent_item);
1163 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1164 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
1165 btrfs_set_file_extent_num_bytes(leaf, fi,
1166 extent_end - split);
1167 btrfs_mark_buffer_dirty(leaf);
1169 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
1170 root->root_key.objectid,
1171 ino, orig_offset, 0);
1172 BUG_ON(ret); /* -ENOMEM */
1174 if (split == start) {
1177 BUG_ON(start != key.offset);
1186 if (extent_mergeable(leaf, path->slots[0] + 1,
1187 ino, bytenr, orig_offset,
1188 &other_start, &other_end)) {
1190 btrfs_release_path(path);
1193 extent_end = other_end;
1194 del_slot = path->slots[0] + 1;
1196 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1197 0, root->root_key.objectid,
1198 ino, orig_offset, 0);
1199 BUG_ON(ret); /* -ENOMEM */
1203 if (extent_mergeable(leaf, path->slots[0] - 1,
1204 ino, bytenr, orig_offset,
1205 &other_start, &other_end)) {
1207 btrfs_release_path(path);
1210 key.offset = other_start;
1211 del_slot = path->slots[0];
1213 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1214 0, root->root_key.objectid,
1215 ino, orig_offset, 0);
1216 BUG_ON(ret); /* -ENOMEM */
1219 fi = btrfs_item_ptr(leaf, path->slots[0],
1220 struct btrfs_file_extent_item);
1221 btrfs_set_file_extent_type(leaf, fi,
1222 BTRFS_FILE_EXTENT_REG);
1223 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1224 btrfs_mark_buffer_dirty(leaf);
1226 fi = btrfs_item_ptr(leaf, del_slot - 1,
1227 struct btrfs_file_extent_item);
1228 btrfs_set_file_extent_type(leaf, fi,
1229 BTRFS_FILE_EXTENT_REG);
1230 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1231 btrfs_set_file_extent_num_bytes(leaf, fi,
1232 extent_end - key.offset);
1233 btrfs_mark_buffer_dirty(leaf);
1235 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1237 btrfs_abort_transaction(trans, root, ret);
1242 btrfs_free_path(path);
1247 * on error we return an unlocked page and the error value
1248 * on success we return a locked page and 0
1250 static int prepare_uptodate_page(struct page *page, u64 pos,
1251 bool force_uptodate)
1255 if (((pos & (PAGE_CACHE_SIZE - 1)) || force_uptodate) &&
1256 !PageUptodate(page)) {
1257 ret = btrfs_readpage(NULL, page);
1261 if (!PageUptodate(page)) {
1270 * this just gets pages into the page cache and locks them down.
1272 static noinline int prepare_pages(struct inode *inode, struct page **pages,
1273 size_t num_pages, loff_t pos,
1274 size_t write_bytes, bool force_uptodate)
1277 unsigned long index = pos >> PAGE_CACHE_SHIFT;
1278 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1282 for (i = 0; i < num_pages; i++) {
1283 pages[i] = find_or_create_page(inode->i_mapping, index + i,
1284 mask | __GFP_WRITE);
1292 err = prepare_uptodate_page(pages[i], pos,
1294 if (i == num_pages - 1)
1295 err = prepare_uptodate_page(pages[i],
1296 pos + write_bytes, false);
1298 page_cache_release(pages[i]);
1302 wait_on_page_writeback(pages[i]);
1307 while (faili >= 0) {
1308 unlock_page(pages[faili]);
1309 page_cache_release(pages[faili]);
1317 * This function locks the extent and properly waits for data=ordered extents
1318 * to finish before allowing the pages to be modified if need.
1321 * 1 - the extent is locked
1322 * 0 - the extent is not locked, and everything is OK
1323 * -EAGAIN - need re-prepare the pages
1324 * the other < 0 number - Something wrong happens
1327 lock_and_cleanup_extent_if_need(struct inode *inode, struct page **pages,
1328 size_t num_pages, loff_t pos,
1329 u64 *lockstart, u64 *lockend,
1330 struct extent_state **cached_state)
1337 start_pos = pos & ~((u64)PAGE_CACHE_SIZE - 1);
1338 last_pos = start_pos + ((u64)num_pages << PAGE_CACHE_SHIFT) - 1;
1340 if (start_pos < inode->i_size) {
1341 struct btrfs_ordered_extent *ordered;
1342 lock_extent_bits(&BTRFS_I(inode)->io_tree,
1343 start_pos, last_pos, 0, cached_state);
1344 ordered = btrfs_lookup_first_ordered_extent(inode, last_pos);
1346 ordered->file_offset + ordered->len > start_pos &&
1347 ordered->file_offset <= last_pos) {
1348 btrfs_put_ordered_extent(ordered);
1349 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1350 start_pos, last_pos,
1351 cached_state, GFP_NOFS);
1352 for (i = 0; i < num_pages; i++) {
1353 unlock_page(pages[i]);
1354 page_cache_release(pages[i]);
1356 ret = btrfs_wait_ordered_range(inode, start_pos,
1357 last_pos - start_pos + 1);
1364 btrfs_put_ordered_extent(ordered);
1366 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
1367 last_pos, EXTENT_DIRTY | EXTENT_DELALLOC |
1368 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
1369 0, 0, cached_state, GFP_NOFS);
1370 *lockstart = start_pos;
1371 *lockend = last_pos;
1375 for (i = 0; i < num_pages; i++) {
1376 if (clear_page_dirty_for_io(pages[i]))
1377 account_page_redirty(pages[i]);
1378 set_page_extent_mapped(pages[i]);
1379 WARN_ON(!PageLocked(pages[i]));
1385 static noinline int check_can_nocow(struct inode *inode, loff_t pos,
1386 size_t *write_bytes)
1388 struct btrfs_root *root = BTRFS_I(inode)->root;
1389 struct btrfs_ordered_extent *ordered;
1390 u64 lockstart, lockend;
1394 lockstart = round_down(pos, root->sectorsize);
1395 lockend = lockstart + round_up(*write_bytes, root->sectorsize) - 1;
1398 lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1399 ordered = btrfs_lookup_ordered_range(inode, lockstart,
1400 lockend - lockstart + 1);
1404 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1405 btrfs_start_ordered_extent(inode, ordered, 1);
1406 btrfs_put_ordered_extent(ordered);
1409 num_bytes = lockend - lockstart + 1;
1410 ret = can_nocow_extent(inode, lockstart, &num_bytes, NULL, NULL, NULL);
1414 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
1415 EXTENT_DIRTY | EXTENT_DELALLOC |
1416 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1418 *write_bytes = min_t(size_t, *write_bytes, num_bytes);
1421 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1426 static noinline ssize_t __btrfs_buffered_write(struct file *file,
1430 struct inode *inode = file_inode(file);
1431 struct btrfs_root *root = BTRFS_I(inode)->root;
1432 struct page **pages = NULL;
1433 struct extent_state *cached_state = NULL;
1434 u64 release_bytes = 0;
1437 unsigned long first_index;
1438 size_t num_written = 0;
1441 bool only_release_metadata = false;
1442 bool force_page_uptodate = false;
1445 nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) /
1446 PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
1447 (sizeof(struct page *)));
1448 nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1449 nrptrs = max(nrptrs, 8);
1450 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
1454 first_index = pos >> PAGE_CACHE_SHIFT;
1456 while (iov_iter_count(i) > 0) {
1457 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1458 size_t write_bytes = min(iov_iter_count(i),
1459 nrptrs * (size_t)PAGE_CACHE_SIZE -
1461 size_t num_pages = (write_bytes + offset +
1462 PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1463 size_t reserve_bytes;
1467 WARN_ON(num_pages > nrptrs);
1470 * Fault pages before locking them in prepare_pages
1471 * to avoid recursive lock
1473 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1478 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1479 ret = btrfs_check_data_free_space(inode, reserve_bytes);
1480 if (ret == -ENOSPC &&
1481 (BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW |
1482 BTRFS_INODE_PREALLOC))) {
1483 ret = check_can_nocow(inode, pos, &write_bytes);
1485 only_release_metadata = true;
1487 * our prealloc extent may be smaller than
1488 * write_bytes, so scale down.
1490 num_pages = (write_bytes + offset +
1491 PAGE_CACHE_SIZE - 1) >>
1493 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1503 ret = btrfs_delalloc_reserve_metadata(inode, reserve_bytes);
1505 if (!only_release_metadata)
1506 btrfs_free_reserved_data_space(inode,
1511 release_bytes = reserve_bytes;
1512 need_unlock = false;
1515 * This is going to setup the pages array with the number of
1516 * pages we want, so we don't really need to worry about the
1517 * contents of pages from loop to loop
1519 ret = prepare_pages(inode, pages, num_pages,
1521 force_page_uptodate);
1525 ret = lock_and_cleanup_extent_if_need(inode, pages, num_pages,
1526 pos, &lockstart, &lockend,
1532 } else if (ret > 0) {
1537 copied = btrfs_copy_from_user(pos, num_pages,
1538 write_bytes, pages, i);
1541 * if we have trouble faulting in the pages, fall
1542 * back to one page at a time
1544 if (copied < write_bytes)
1548 force_page_uptodate = true;
1551 force_page_uptodate = false;
1552 dirty_pages = (copied + offset +
1553 PAGE_CACHE_SIZE - 1) >>
1558 * If we had a short copy we need to release the excess delaloc
1559 * bytes we reserved. We need to increment outstanding_extents
1560 * because btrfs_delalloc_release_space will decrement it, but
1561 * we still have an outstanding extent for the chunk we actually
1564 if (num_pages > dirty_pages) {
1565 release_bytes = (num_pages - dirty_pages) <<
1568 spin_lock(&BTRFS_I(inode)->lock);
1569 BTRFS_I(inode)->outstanding_extents++;
1570 spin_unlock(&BTRFS_I(inode)->lock);
1572 if (only_release_metadata)
1573 btrfs_delalloc_release_metadata(inode,
1576 btrfs_delalloc_release_space(inode,
1580 release_bytes = dirty_pages << PAGE_CACHE_SHIFT;
1583 ret = btrfs_dirty_pages(root, inode, pages,
1584 dirty_pages, pos, copied,
1587 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1588 lockstart, lockend, &cached_state,
1591 btrfs_drop_pages(pages, num_pages);
1596 if (only_release_metadata && copied > 0) {
1597 u64 lockstart = round_down(pos, root->sectorsize);
1598 u64 lockend = lockstart +
1599 (dirty_pages << PAGE_CACHE_SHIFT) - 1;
1601 set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
1602 lockend, EXTENT_NORESERVE, NULL,
1604 only_release_metadata = false;
1607 btrfs_drop_pages(pages, num_pages);
1611 balance_dirty_pages_ratelimited(inode->i_mapping);
1612 if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1613 btrfs_btree_balance_dirty(root);
1616 num_written += copied;
1621 if (release_bytes) {
1622 if (only_release_metadata)
1623 btrfs_delalloc_release_metadata(inode, release_bytes);
1625 btrfs_delalloc_release_space(inode, release_bytes);
1628 return num_written ? num_written : ret;
1631 static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1632 const struct iovec *iov,
1633 unsigned long nr_segs, loff_t pos,
1634 size_t count, size_t ocount)
1636 struct file *file = iocb->ki_filp;
1639 ssize_t written_buffered;
1643 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
1646 if (written < 0 || written == count)
1651 iov_iter_init(&i, iov, nr_segs, count, written);
1652 written_buffered = __btrfs_buffered_write(file, &i, pos);
1653 if (written_buffered < 0) {
1654 err = written_buffered;
1657 endbyte = pos + written_buffered - 1;
1658 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
1661 written += written_buffered;
1662 iocb->ki_pos = pos + written_buffered;
1663 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
1664 endbyte >> PAGE_CACHE_SHIFT);
1666 return written ? written : err;
1669 static void update_time_for_write(struct inode *inode)
1671 struct timespec now;
1673 if (IS_NOCMTIME(inode))
1676 now = current_fs_time(inode->i_sb);
1677 if (!timespec_equal(&inode->i_mtime, &now))
1678 inode->i_mtime = now;
1680 if (!timespec_equal(&inode->i_ctime, &now))
1681 inode->i_ctime = now;
1683 if (IS_I_VERSION(inode))
1684 inode_inc_iversion(inode);
1687 static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
1688 const struct iovec *iov,
1689 unsigned long nr_segs, loff_t pos)
1691 struct file *file = iocb->ki_filp;
1692 struct inode *inode = file_inode(file);
1693 struct btrfs_root *root = BTRFS_I(inode)->root;
1695 ssize_t num_written = 0;
1697 size_t count, ocount;
1698 bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host);
1700 mutex_lock(&inode->i_mutex);
1702 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
1704 mutex_unlock(&inode->i_mutex);
1709 current->backing_dev_info = inode->i_mapping->backing_dev_info;
1710 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1712 mutex_unlock(&inode->i_mutex);
1717 mutex_unlock(&inode->i_mutex);
1721 err = file_remove_suid(file);
1723 mutex_unlock(&inode->i_mutex);
1728 * If BTRFS flips readonly due to some impossible error
1729 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1730 * although we have opened a file as writable, we have
1731 * to stop this write operation to ensure FS consistency.
1733 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
1734 mutex_unlock(&inode->i_mutex);
1740 * We reserve space for updating the inode when we reserve space for the
1741 * extent we are going to write, so we will enospc out there. We don't
1742 * need to start yet another transaction to update the inode as we will
1743 * update the inode when we finish writing whatever data we write.
1745 update_time_for_write(inode);
1747 start_pos = round_down(pos, root->sectorsize);
1748 if (start_pos > i_size_read(inode)) {
1749 err = btrfs_cont_expand(inode, i_size_read(inode), start_pos);
1751 mutex_unlock(&inode->i_mutex);
1757 atomic_inc(&BTRFS_I(inode)->sync_writers);
1759 if (unlikely(file->f_flags & O_DIRECT)) {
1760 num_written = __btrfs_direct_write(iocb, iov, nr_segs,
1761 pos, count, ocount);
1765 iov_iter_init(&i, iov, nr_segs, count, num_written);
1767 num_written = __btrfs_buffered_write(file, &i, pos);
1768 if (num_written > 0)
1769 iocb->ki_pos = pos + num_written;
1772 mutex_unlock(&inode->i_mutex);
1775 * we want to make sure fsync finds this change
1776 * but we haven't joined a transaction running right now.
1778 * Later on, someone is sure to update the inode and get the
1779 * real transid recorded.
1781 * We set last_trans now to the fs_info generation + 1,
1782 * this will either be one more than the running transaction
1783 * or the generation used for the next transaction if there isn't
1784 * one running right now.
1786 * We also have to set last_sub_trans to the current log transid,
1787 * otherwise subsequent syncs to a file that's been synced in this
1788 * transaction will appear to have already occured.
1790 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1791 BTRFS_I(inode)->last_sub_trans = root->log_transid;
1792 if (num_written > 0) {
1793 err = generic_write_sync(file, pos, num_written);
1794 if (err < 0 && num_written > 0)
1799 atomic_dec(&BTRFS_I(inode)->sync_writers);
1801 current->backing_dev_info = NULL;
1802 return num_written ? num_written : err;
1805 int btrfs_release_file(struct inode *inode, struct file *filp)
1808 * ordered_data_close is set by settattr when we are about to truncate
1809 * a file from a non-zero size to a zero size. This tries to
1810 * flush down new bytes that may have been written if the
1811 * application were using truncate to replace a file in place.
1813 if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
1814 &BTRFS_I(inode)->runtime_flags)) {
1815 struct btrfs_trans_handle *trans;
1816 struct btrfs_root *root = BTRFS_I(inode)->root;
1819 * We need to block on a committing transaction to keep us from
1820 * throwing a ordered operation on to the list and causing
1821 * something like sync to deadlock trying to flush out this
1824 trans = btrfs_start_transaction(root, 0);
1826 return PTR_ERR(trans);
1827 btrfs_add_ordered_operation(trans, BTRFS_I(inode)->root, inode);
1828 btrfs_end_transaction(trans, root);
1829 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1830 filemap_flush(inode->i_mapping);
1832 if (filp->private_data)
1833 btrfs_ioctl_trans_end(filp);
1838 * fsync call for both files and directories. This logs the inode into
1839 * the tree log instead of forcing full commits whenever possible.
1841 * It needs to call filemap_fdatawait so that all ordered extent updates are
1842 * in the metadata btree are up to date for copying to the log.
1844 * It drops the inode mutex before doing the tree log commit. This is an
1845 * important optimization for directories because holding the mutex prevents
1846 * new operations on the dir while we write to disk.
1848 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1850 struct dentry *dentry = file->f_path.dentry;
1851 struct inode *inode = dentry->d_inode;
1852 struct btrfs_root *root = BTRFS_I(inode)->root;
1854 struct btrfs_trans_handle *trans;
1857 trace_btrfs_sync_file(file, datasync);
1860 * We write the dirty pages in the range and wait until they complete
1861 * out of the ->i_mutex. If so, we can flush the dirty pages by
1862 * multi-task, and make the performance up. See
1863 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1865 atomic_inc(&BTRFS_I(inode)->sync_writers);
1866 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
1867 if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1868 &BTRFS_I(inode)->runtime_flags))
1869 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
1870 atomic_dec(&BTRFS_I(inode)->sync_writers);
1874 mutex_lock(&inode->i_mutex);
1877 * We flush the dirty pages again to avoid some dirty pages in the
1880 atomic_inc(&root->log_batch);
1881 full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1882 &BTRFS_I(inode)->runtime_flags);
1884 ret = btrfs_wait_ordered_range(inode, start, end - start + 1);
1886 mutex_unlock(&inode->i_mutex);
1890 atomic_inc(&root->log_batch);
1893 * check the transaction that last modified this inode
1894 * and see if its already been committed
1896 if (!BTRFS_I(inode)->last_trans) {
1897 mutex_unlock(&inode->i_mutex);
1902 * if the last transaction that changed this file was before
1903 * the current transaction, we can bail out now without any
1907 if (btrfs_inode_in_log(inode, root->fs_info->generation) ||
1908 BTRFS_I(inode)->last_trans <=
1909 root->fs_info->last_trans_committed) {
1910 BTRFS_I(inode)->last_trans = 0;
1913 * We'v had everything committed since the last time we were
1914 * modified so clear this flag in case it was set for whatever
1915 * reason, it's no longer relevant.
1917 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1918 &BTRFS_I(inode)->runtime_flags);
1919 mutex_unlock(&inode->i_mutex);
1924 * ok we haven't committed the transaction yet, lets do a commit
1926 if (file->private_data)
1927 btrfs_ioctl_trans_end(file);
1930 * We use start here because we will need to wait on the IO to complete
1931 * in btrfs_sync_log, which could require joining a transaction (for
1932 * example checking cross references in the nocow path). If we use join
1933 * here we could get into a situation where we're waiting on IO to
1934 * happen that is blocked on a transaction trying to commit. With start
1935 * we inc the extwriter counter, so we wait for all extwriters to exit
1936 * before we start blocking join'ers. This comment is to keep somebody
1937 * from thinking they are super smart and changing this to
1938 * btrfs_join_transaction *cough*Josef*cough*.
1940 trans = btrfs_start_transaction(root, 0);
1941 if (IS_ERR(trans)) {
1942 ret = PTR_ERR(trans);
1943 mutex_unlock(&inode->i_mutex);
1948 ret = btrfs_log_dentry_safe(trans, root, dentry);
1950 /* Fallthrough and commit/free transaction. */
1954 /* we've logged all the items and now have a consistent
1955 * version of the file in the log. It is possible that
1956 * someone will come in and modify the file, but that's
1957 * fine because the log is consistent on disk, and we
1958 * have references to all of the file's extents
1960 * It is possible that someone will come in and log the
1961 * file again, but that will end up using the synchronization
1962 * inside btrfs_sync_log to keep things safe.
1964 mutex_unlock(&inode->i_mutex);
1966 if (ret != BTRFS_NO_LOG_SYNC) {
1968 ret = btrfs_sync_log(trans, root);
1970 ret = btrfs_end_transaction(trans, root);
1975 ret = btrfs_wait_ordered_range(inode, start,
1980 ret = btrfs_commit_transaction(trans, root);
1982 ret = btrfs_end_transaction(trans, root);
1985 return ret > 0 ? -EIO : ret;
1988 static const struct vm_operations_struct btrfs_file_vm_ops = {
1989 .fault = filemap_fault,
1990 .page_mkwrite = btrfs_page_mkwrite,
1991 .remap_pages = generic_file_remap_pages,
1994 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
1996 struct address_space *mapping = filp->f_mapping;
1998 if (!mapping->a_ops->readpage)
2001 file_accessed(filp);
2002 vma->vm_ops = &btrfs_file_vm_ops;
2007 static int hole_mergeable(struct inode *inode, struct extent_buffer *leaf,
2008 int slot, u64 start, u64 end)
2010 struct btrfs_file_extent_item *fi;
2011 struct btrfs_key key;
2013 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
2016 btrfs_item_key_to_cpu(leaf, &key, slot);
2017 if (key.objectid != btrfs_ino(inode) ||
2018 key.type != BTRFS_EXTENT_DATA_KEY)
2021 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2023 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2026 if (btrfs_file_extent_disk_bytenr(leaf, fi))
2029 if (key.offset == end)
2031 if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
2036 static int fill_holes(struct btrfs_trans_handle *trans, struct inode *inode,
2037 struct btrfs_path *path, u64 offset, u64 end)
2039 struct btrfs_root *root = BTRFS_I(inode)->root;
2040 struct extent_buffer *leaf;
2041 struct btrfs_file_extent_item *fi;
2042 struct extent_map *hole_em;
2043 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2044 struct btrfs_key key;
2047 if (btrfs_fs_incompat(root->fs_info, NO_HOLES))
2050 key.objectid = btrfs_ino(inode);
2051 key.type = BTRFS_EXTENT_DATA_KEY;
2052 key.offset = offset;
2054 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2059 leaf = path->nodes[0];
2060 if (hole_mergeable(inode, leaf, path->slots[0]-1, offset, end)) {
2064 fi = btrfs_item_ptr(leaf, path->slots[0],
2065 struct btrfs_file_extent_item);
2066 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2068 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2069 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2070 btrfs_set_file_extent_offset(leaf, fi, 0);
2071 btrfs_mark_buffer_dirty(leaf);
2075 if (hole_mergeable(inode, leaf, path->slots[0]+1, offset, end)) {
2079 key.offset = offset;
2080 btrfs_set_item_key_safe(root, path, &key);
2081 fi = btrfs_item_ptr(leaf, path->slots[0],
2082 struct btrfs_file_extent_item);
2083 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2085 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2086 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2087 btrfs_set_file_extent_offset(leaf, fi, 0);
2088 btrfs_mark_buffer_dirty(leaf);
2091 btrfs_release_path(path);
2093 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
2094 0, 0, end - offset, 0, end - offset,
2100 btrfs_release_path(path);
2102 hole_em = alloc_extent_map();
2104 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2105 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2106 &BTRFS_I(inode)->runtime_flags);
2108 hole_em->start = offset;
2109 hole_em->len = end - offset;
2110 hole_em->ram_bytes = hole_em->len;
2111 hole_em->orig_start = offset;
2113 hole_em->block_start = EXTENT_MAP_HOLE;
2114 hole_em->block_len = 0;
2115 hole_em->orig_block_len = 0;
2116 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
2117 hole_em->compress_type = BTRFS_COMPRESS_NONE;
2118 hole_em->generation = trans->transid;
2121 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2122 write_lock(&em_tree->lock);
2123 ret = add_extent_mapping(em_tree, hole_em, 1);
2124 write_unlock(&em_tree->lock);
2125 } while (ret == -EEXIST);
2126 free_extent_map(hole_em);
2128 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2129 &BTRFS_I(inode)->runtime_flags);
2135 static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
2137 struct btrfs_root *root = BTRFS_I(inode)->root;
2138 struct extent_state *cached_state = NULL;
2139 struct btrfs_path *path;
2140 struct btrfs_block_rsv *rsv;
2141 struct btrfs_trans_handle *trans;
2142 u64 lockstart = round_up(offset, BTRFS_I(inode)->root->sectorsize);
2143 u64 lockend = round_down(offset + len,
2144 BTRFS_I(inode)->root->sectorsize) - 1;
2145 u64 cur_offset = lockstart;
2146 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
2151 bool same_page = ((offset >> PAGE_CACHE_SHIFT) ==
2152 ((offset + len - 1) >> PAGE_CACHE_SHIFT));
2153 bool no_holes = btrfs_fs_incompat(root->fs_info, NO_HOLES);
2155 ret = btrfs_wait_ordered_range(inode, offset, len);
2159 mutex_lock(&inode->i_mutex);
2161 * We needn't truncate any page which is beyond the end of the file
2162 * because we are sure there is no data there.
2165 * Only do this if we are in the same page and we aren't doing the
2168 if (same_page && len < PAGE_CACHE_SIZE) {
2169 if (offset < round_up(inode->i_size, PAGE_CACHE_SIZE))
2170 ret = btrfs_truncate_page(inode, offset, len, 0);
2171 mutex_unlock(&inode->i_mutex);
2175 /* zero back part of the first page */
2176 if (offset < round_up(inode->i_size, PAGE_CACHE_SIZE)) {
2177 ret = btrfs_truncate_page(inode, offset, 0, 0);
2179 mutex_unlock(&inode->i_mutex);
2184 /* zero the front end of the last page */
2185 if (offset + len < round_up(inode->i_size, PAGE_CACHE_SIZE)) {
2186 ret = btrfs_truncate_page(inode, offset + len, 0, 1);
2188 mutex_unlock(&inode->i_mutex);
2193 if (lockend < lockstart) {
2194 mutex_unlock(&inode->i_mutex);
2199 struct btrfs_ordered_extent *ordered;
2201 truncate_pagecache_range(inode, lockstart, lockend);
2203 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2205 ordered = btrfs_lookup_first_ordered_extent(inode, lockend);
2208 * We need to make sure we have no ordered extents in this range
2209 * and nobody raced in and read a page in this range, if we did
2210 * we need to try again.
2213 (ordered->file_offset + ordered->len <= lockstart ||
2214 ordered->file_offset > lockend)) &&
2215 !test_range_bit(&BTRFS_I(inode)->io_tree, lockstart,
2216 lockend, EXTENT_UPTODATE, 0,
2219 btrfs_put_ordered_extent(ordered);
2223 btrfs_put_ordered_extent(ordered);
2224 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
2225 lockend, &cached_state, GFP_NOFS);
2226 ret = btrfs_wait_ordered_range(inode, lockstart,
2227 lockend - lockstart + 1);
2229 mutex_unlock(&inode->i_mutex);
2234 path = btrfs_alloc_path();
2240 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2245 rsv->size = btrfs_calc_trunc_metadata_size(root, 1);
2249 * 1 - update the inode
2250 * 1 - removing the extents in the range
2251 * 1 - adding the hole extent if no_holes isn't set
2253 rsv_count = no_holes ? 2 : 3;
2254 trans = btrfs_start_transaction(root, rsv_count);
2255 if (IS_ERR(trans)) {
2256 err = PTR_ERR(trans);
2260 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
2263 trans->block_rsv = rsv;
2265 while (cur_offset < lockend) {
2266 ret = __btrfs_drop_extents(trans, root, inode, path,
2267 cur_offset, lockend + 1,
2268 &drop_end, 1, 0, 0, NULL);
2272 trans->block_rsv = &root->fs_info->trans_block_rsv;
2274 ret = fill_holes(trans, inode, path, cur_offset, drop_end);
2280 cur_offset = drop_end;
2282 ret = btrfs_update_inode(trans, root, inode);
2288 btrfs_end_transaction(trans, root);
2289 btrfs_btree_balance_dirty(root);
2291 trans = btrfs_start_transaction(root, rsv_count);
2292 if (IS_ERR(trans)) {
2293 ret = PTR_ERR(trans);
2298 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
2300 BUG_ON(ret); /* shouldn't happen */
2301 trans->block_rsv = rsv;
2309 trans->block_rsv = &root->fs_info->trans_block_rsv;
2310 ret = fill_holes(trans, inode, path, cur_offset, drop_end);
2320 inode_inc_iversion(inode);
2321 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2323 trans->block_rsv = &root->fs_info->trans_block_rsv;
2324 ret = btrfs_update_inode(trans, root, inode);
2325 btrfs_end_transaction(trans, root);
2326 btrfs_btree_balance_dirty(root);
2328 btrfs_free_path(path);
2329 btrfs_free_block_rsv(root, rsv);
2331 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2332 &cached_state, GFP_NOFS);
2333 mutex_unlock(&inode->i_mutex);
2339 static long btrfs_fallocate(struct file *file, int mode,
2340 loff_t offset, loff_t len)
2342 struct inode *inode = file_inode(file);
2343 struct extent_state *cached_state = NULL;
2344 struct btrfs_root *root = BTRFS_I(inode)->root;
2351 struct extent_map *em;
2352 int blocksize = BTRFS_I(inode)->root->sectorsize;
2355 alloc_start = round_down(offset, blocksize);
2356 alloc_end = round_up(offset + len, blocksize);
2358 /* Make sure we aren't being give some crap mode */
2359 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2362 if (mode & FALLOC_FL_PUNCH_HOLE)
2363 return btrfs_punch_hole(inode, offset, len);
2366 * Make sure we have enough space before we do the
2369 ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
2372 if (root->fs_info->quota_enabled) {
2373 ret = btrfs_qgroup_reserve(root, alloc_end - alloc_start);
2375 goto out_reserve_fail;
2378 mutex_lock(&inode->i_mutex);
2379 ret = inode_newsize_ok(inode, alloc_end);
2383 if (alloc_start > inode->i_size) {
2384 ret = btrfs_cont_expand(inode, i_size_read(inode),
2390 * If we are fallocating from the end of the file onward we
2391 * need to zero out the end of the page if i_size lands in the
2394 ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
2400 * wait for ordered IO before we have any locks. We'll loop again
2401 * below with the locks held.
2403 ret = btrfs_wait_ordered_range(inode, alloc_start,
2404 alloc_end - alloc_start);
2408 locked_end = alloc_end - 1;
2410 struct btrfs_ordered_extent *ordered;
2412 /* the extent lock is ordered inside the running
2415 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
2416 locked_end, 0, &cached_state);
2417 ordered = btrfs_lookup_first_ordered_extent(inode,
2420 ordered->file_offset + ordered->len > alloc_start &&
2421 ordered->file_offset < alloc_end) {
2422 btrfs_put_ordered_extent(ordered);
2423 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
2424 alloc_start, locked_end,
2425 &cached_state, GFP_NOFS);
2427 * we can't wait on the range with the transaction
2428 * running or with the extent lock held
2430 ret = btrfs_wait_ordered_range(inode, alloc_start,
2431 alloc_end - alloc_start);
2436 btrfs_put_ordered_extent(ordered);
2441 cur_offset = alloc_start;
2445 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2446 alloc_end - cur_offset, 0);
2447 if (IS_ERR_OR_NULL(em)) {
2454 last_byte = min(extent_map_end(em), alloc_end);
2455 actual_end = min_t(u64, extent_map_end(em), offset + len);
2456 last_byte = ALIGN(last_byte, blocksize);
2458 if (em->block_start == EXTENT_MAP_HOLE ||
2459 (cur_offset >= inode->i_size &&
2460 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
2461 ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
2462 last_byte - cur_offset,
2463 1 << inode->i_blkbits,
2468 free_extent_map(em);
2471 } else if (actual_end > inode->i_size &&
2472 !(mode & FALLOC_FL_KEEP_SIZE)) {
2474 * We didn't need to allocate any more space, but we
2475 * still extended the size of the file so we need to
2478 inode->i_ctime = CURRENT_TIME;
2479 i_size_write(inode, actual_end);
2480 btrfs_ordered_update_i_size(inode, actual_end, NULL);
2482 free_extent_map(em);
2484 cur_offset = last_byte;
2485 if (cur_offset >= alloc_end) {
2490 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
2491 &cached_state, GFP_NOFS);
2493 mutex_unlock(&inode->i_mutex);
2494 if (root->fs_info->quota_enabled)
2495 btrfs_qgroup_free(root, alloc_end - alloc_start);
2497 /* Let go of our reservation. */
2498 btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
2502 static int find_desired_extent(struct inode *inode, loff_t *offset, int whence)
2504 struct btrfs_root *root = BTRFS_I(inode)->root;
2505 struct extent_map *em = NULL;
2506 struct extent_state *cached_state = NULL;
2507 u64 lockstart = *offset;
2508 u64 lockend = i_size_read(inode);
2509 u64 start = *offset;
2510 u64 len = i_size_read(inode);
2513 lockend = max_t(u64, root->sectorsize, lockend);
2514 if (lockend <= lockstart)
2515 lockend = lockstart + root->sectorsize;
2518 len = lockend - lockstart + 1;
2520 len = max_t(u64, len, root->sectorsize);
2521 if (inode->i_size == 0)
2524 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, 0,
2527 while (start < inode->i_size) {
2528 em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0);
2535 if (whence == SEEK_HOLE &&
2536 (em->block_start == EXTENT_MAP_HOLE ||
2537 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2539 else if (whence == SEEK_DATA &&
2540 (em->block_start != EXTENT_MAP_HOLE &&
2541 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2544 start = em->start + em->len;
2545 free_extent_map(em);
2549 free_extent_map(em);
2551 if (whence == SEEK_DATA && start >= inode->i_size)
2554 *offset = min_t(loff_t, start, inode->i_size);
2556 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2557 &cached_state, GFP_NOFS);
2561 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
2563 struct inode *inode = file->f_mapping->host;
2566 mutex_lock(&inode->i_mutex);
2570 offset = generic_file_llseek(file, offset, whence);
2574 if (offset >= i_size_read(inode)) {
2575 mutex_unlock(&inode->i_mutex);
2579 ret = find_desired_extent(inode, &offset, whence);
2581 mutex_unlock(&inode->i_mutex);
2586 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2588 mutex_unlock(&inode->i_mutex);
2592 const struct file_operations btrfs_file_operations = {
2593 .llseek = btrfs_file_llseek,
2594 .read = do_sync_read,
2595 .write = do_sync_write,
2596 .aio_read = generic_file_aio_read,
2597 .splice_read = generic_file_splice_read,
2598 .aio_write = btrfs_file_aio_write,
2599 .mmap = btrfs_file_mmap,
2600 .open = generic_file_open,
2601 .release = btrfs_release_file,
2602 .fsync = btrfs_sync_file,
2603 .fallocate = btrfs_fallocate,
2604 .unlocked_ioctl = btrfs_ioctl,
2605 #ifdef CONFIG_COMPAT
2606 .compat_ioctl = btrfs_ioctl,
2610 void btrfs_auto_defrag_exit(void)
2612 if (btrfs_inode_defrag_cachep)
2613 kmem_cache_destroy(btrfs_inode_defrag_cachep);
2616 int btrfs_auto_defrag_init(void)
2618 btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
2619 sizeof(struct inode_defrag), 0,
2620 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2622 if (!btrfs_inode_defrag_cachep)