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);
594 remove_extent_mapping(em_tree, em);
598 if (em->start < start) {
599 split->start = em->start;
600 split->len = start - em->start;
602 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
603 split->orig_start = em->orig_start;
604 split->block_start = em->block_start;
607 split->block_len = em->block_len;
609 split->block_len = split->len;
610 split->orig_block_len = max(split->block_len,
612 split->ram_bytes = em->ram_bytes;
614 split->orig_start = split->start;
615 split->block_len = 0;
616 split->block_start = em->block_start;
617 split->orig_block_len = 0;
618 split->ram_bytes = split->len;
621 split->generation = gen;
622 split->bdev = em->bdev;
623 split->flags = flags;
624 split->compress_type = em->compress_type;
625 ret = add_extent_mapping(em_tree, split, modified);
626 BUG_ON(ret); /* Logic error */
627 free_extent_map(split);
631 if (testend && em->start + em->len > start + len) {
632 u64 diff = start + len - em->start;
634 split->start = start + len;
635 split->len = em->start + em->len - (start + len);
636 split->bdev = em->bdev;
637 split->flags = flags;
638 split->compress_type = em->compress_type;
639 split->generation = gen;
641 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
642 split->orig_block_len = max(em->block_len,
645 split->ram_bytes = em->ram_bytes;
647 split->block_len = em->block_len;
648 split->block_start = em->block_start;
649 split->orig_start = em->orig_start;
651 split->block_len = split->len;
652 split->block_start = em->block_start
654 split->orig_start = em->orig_start;
657 split->ram_bytes = split->len;
658 split->orig_start = split->start;
659 split->block_len = 0;
660 split->block_start = em->block_start;
661 split->orig_block_len = 0;
664 ret = add_extent_mapping(em_tree, split, modified);
665 BUG_ON(ret); /* Logic error */
666 free_extent_map(split);
670 write_unlock(&em_tree->lock);
674 /* once for the tree*/
678 free_extent_map(split);
680 free_extent_map(split2);
684 * this is very complex, but the basic idea is to drop all extents
685 * in the range start - end. hint_block is filled in with a block number
686 * that would be a good hint to the block allocator for this file.
688 * If an extent intersects the range but is not entirely inside the range
689 * it is either truncated or split. Anything entirely inside the range
690 * is deleted from the tree.
692 int __btrfs_drop_extents(struct btrfs_trans_handle *trans,
693 struct btrfs_root *root, struct inode *inode,
694 struct btrfs_path *path, u64 start, u64 end,
695 u64 *drop_end, int drop_cache,
697 u32 extent_item_size,
700 struct extent_buffer *leaf;
701 struct btrfs_file_extent_item *fi;
702 struct btrfs_key key;
703 struct btrfs_key new_key;
704 u64 ino = btrfs_ino(inode);
705 u64 search_start = start;
708 u64 extent_offset = 0;
715 int modify_tree = -1;
716 int update_refs = (root->ref_cows || root == root->fs_info->tree_root);
718 int leafs_visited = 0;
721 btrfs_drop_extent_cache(inode, start, end - 1, 0);
723 if (start >= BTRFS_I(inode)->disk_i_size)
728 ret = btrfs_lookup_file_extent(trans, root, path, ino,
729 search_start, modify_tree);
732 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
733 leaf = path->nodes[0];
734 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
735 if (key.objectid == ino &&
736 key.type == BTRFS_EXTENT_DATA_KEY)
742 leaf = path->nodes[0];
743 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
745 ret = btrfs_next_leaf(root, path);
753 leaf = path->nodes[0];
757 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
758 if (key.objectid > ino ||
759 key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
762 fi = btrfs_item_ptr(leaf, path->slots[0],
763 struct btrfs_file_extent_item);
764 extent_type = btrfs_file_extent_type(leaf, fi);
766 if (extent_type == BTRFS_FILE_EXTENT_REG ||
767 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
768 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
769 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
770 extent_offset = btrfs_file_extent_offset(leaf, fi);
771 extent_end = key.offset +
772 btrfs_file_extent_num_bytes(leaf, fi);
773 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
774 extent_end = key.offset +
775 btrfs_file_extent_inline_len(leaf,
779 extent_end = search_start;
782 if (extent_end <= search_start) {
788 search_start = max(key.offset, start);
789 if (recow || !modify_tree) {
791 btrfs_release_path(path);
796 * | - range to drop - |
797 * | -------- extent -------- |
799 if (start > key.offset && end < extent_end) {
801 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
803 memcpy(&new_key, &key, sizeof(new_key));
804 new_key.offset = start;
805 ret = btrfs_duplicate_item(trans, root, path,
807 if (ret == -EAGAIN) {
808 btrfs_release_path(path);
814 leaf = path->nodes[0];
815 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
816 struct btrfs_file_extent_item);
817 btrfs_set_file_extent_num_bytes(leaf, fi,
820 fi = btrfs_item_ptr(leaf, path->slots[0],
821 struct btrfs_file_extent_item);
823 extent_offset += start - key.offset;
824 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
825 btrfs_set_file_extent_num_bytes(leaf, fi,
827 btrfs_mark_buffer_dirty(leaf);
829 if (update_refs && disk_bytenr > 0) {
830 ret = btrfs_inc_extent_ref(trans, root,
831 disk_bytenr, num_bytes, 0,
832 root->root_key.objectid,
834 start - extent_offset, 0);
835 BUG_ON(ret); /* -ENOMEM */
840 * | ---- range to drop ----- |
841 * | -------- extent -------- |
843 if (start <= key.offset && end < extent_end) {
844 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
846 memcpy(&new_key, &key, sizeof(new_key));
847 new_key.offset = end;
848 btrfs_set_item_key_safe(root, path, &new_key);
850 extent_offset += end - key.offset;
851 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
852 btrfs_set_file_extent_num_bytes(leaf, fi,
854 btrfs_mark_buffer_dirty(leaf);
855 if (update_refs && disk_bytenr > 0)
856 inode_sub_bytes(inode, end - key.offset);
860 search_start = extent_end;
862 * | ---- range to drop ----- |
863 * | -------- extent -------- |
865 if (start > key.offset && end >= extent_end) {
867 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
869 btrfs_set_file_extent_num_bytes(leaf, fi,
871 btrfs_mark_buffer_dirty(leaf);
872 if (update_refs && disk_bytenr > 0)
873 inode_sub_bytes(inode, extent_end - start);
874 if (end == extent_end)
882 * | ---- range to drop ----- |
883 * | ------ extent ------ |
885 if (start <= key.offset && end >= extent_end) {
887 del_slot = path->slots[0];
890 BUG_ON(del_slot + del_nr != path->slots[0]);
895 extent_type == BTRFS_FILE_EXTENT_INLINE) {
896 inode_sub_bytes(inode,
897 extent_end - key.offset);
898 extent_end = ALIGN(extent_end,
900 } else if (update_refs && disk_bytenr > 0) {
901 ret = btrfs_free_extent(trans, root,
902 disk_bytenr, num_bytes, 0,
903 root->root_key.objectid,
904 key.objectid, key.offset -
906 BUG_ON(ret); /* -ENOMEM */
907 inode_sub_bytes(inode,
908 extent_end - key.offset);
911 if (end == extent_end)
914 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
919 ret = btrfs_del_items(trans, root, path, del_slot,
922 btrfs_abort_transaction(trans, root, ret);
929 btrfs_release_path(path);
936 if (!ret && del_nr > 0) {
938 * Set path->slots[0] to first slot, so that after the delete
939 * if items are move off from our leaf to its immediate left or
940 * right neighbor leafs, we end up with a correct and adjusted
941 * path->slots[0] for our insertion.
943 path->slots[0] = del_slot;
944 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
946 btrfs_abort_transaction(trans, root, ret);
948 leaf = path->nodes[0];
950 * leaf eb has flag EXTENT_BUFFER_STALE if it was deleted (that
951 * is, its contents got pushed to its neighbors), in which case
952 * it means path->locks[0] == 0
954 if (!ret && replace_extent && leafs_visited == 1 &&
956 btrfs_leaf_free_space(root, leaf) >=
957 sizeof(struct btrfs_item) + extent_item_size) {
960 key.type = BTRFS_EXTENT_DATA_KEY;
962 setup_items_for_insert(root, path, &key,
965 sizeof(struct btrfs_item) +
966 extent_item_size, 1);
971 if (!replace_extent || !(*key_inserted))
972 btrfs_release_path(path);
974 *drop_end = found ? min(end, extent_end) : end;
978 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
979 struct btrfs_root *root, struct inode *inode, u64 start,
980 u64 end, int drop_cache)
982 struct btrfs_path *path;
985 path = btrfs_alloc_path();
988 ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL,
989 drop_cache, 0, 0, NULL);
990 btrfs_free_path(path);
994 static int extent_mergeable(struct extent_buffer *leaf, int slot,
995 u64 objectid, u64 bytenr, u64 orig_offset,
996 u64 *start, u64 *end)
998 struct btrfs_file_extent_item *fi;
999 struct btrfs_key key;
1002 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
1005 btrfs_item_key_to_cpu(leaf, &key, slot);
1006 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
1009 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1010 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
1011 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
1012 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
1013 btrfs_file_extent_compression(leaf, fi) ||
1014 btrfs_file_extent_encryption(leaf, fi) ||
1015 btrfs_file_extent_other_encoding(leaf, fi))
1018 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1019 if ((*start && *start != key.offset) || (*end && *end != extent_end))
1022 *start = key.offset;
1028 * Mark extent in the range start - end as written.
1030 * This changes extent type from 'pre-allocated' to 'regular'. If only
1031 * part of extent is marked as written, the extent will be split into
1034 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
1035 struct inode *inode, u64 start, u64 end)
1037 struct btrfs_root *root = BTRFS_I(inode)->root;
1038 struct extent_buffer *leaf;
1039 struct btrfs_path *path;
1040 struct btrfs_file_extent_item *fi;
1041 struct btrfs_key key;
1042 struct btrfs_key new_key;
1054 u64 ino = btrfs_ino(inode);
1056 path = btrfs_alloc_path();
1063 key.type = BTRFS_EXTENT_DATA_KEY;
1066 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1069 if (ret > 0 && path->slots[0] > 0)
1072 leaf = path->nodes[0];
1073 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1074 BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY);
1075 fi = btrfs_item_ptr(leaf, path->slots[0],
1076 struct btrfs_file_extent_item);
1077 BUG_ON(btrfs_file_extent_type(leaf, fi) !=
1078 BTRFS_FILE_EXTENT_PREALLOC);
1079 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1080 BUG_ON(key.offset > start || extent_end < end);
1082 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1083 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1084 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
1085 memcpy(&new_key, &key, sizeof(new_key));
1087 if (start == key.offset && end < extent_end) {
1090 if (extent_mergeable(leaf, path->slots[0] - 1,
1091 ino, bytenr, orig_offset,
1092 &other_start, &other_end)) {
1093 new_key.offset = end;
1094 btrfs_set_item_key_safe(root, path, &new_key);
1095 fi = btrfs_item_ptr(leaf, path->slots[0],
1096 struct btrfs_file_extent_item);
1097 btrfs_set_file_extent_generation(leaf, fi,
1099 btrfs_set_file_extent_num_bytes(leaf, fi,
1101 btrfs_set_file_extent_offset(leaf, fi,
1103 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1104 struct btrfs_file_extent_item);
1105 btrfs_set_file_extent_generation(leaf, fi,
1107 btrfs_set_file_extent_num_bytes(leaf, fi,
1109 btrfs_mark_buffer_dirty(leaf);
1114 if (start > key.offset && end == extent_end) {
1117 if (extent_mergeable(leaf, path->slots[0] + 1,
1118 ino, bytenr, orig_offset,
1119 &other_start, &other_end)) {
1120 fi = btrfs_item_ptr(leaf, path->slots[0],
1121 struct btrfs_file_extent_item);
1122 btrfs_set_file_extent_num_bytes(leaf, fi,
1123 start - key.offset);
1124 btrfs_set_file_extent_generation(leaf, fi,
1127 new_key.offset = start;
1128 btrfs_set_item_key_safe(root, path, &new_key);
1130 fi = btrfs_item_ptr(leaf, path->slots[0],
1131 struct btrfs_file_extent_item);
1132 btrfs_set_file_extent_generation(leaf, fi,
1134 btrfs_set_file_extent_num_bytes(leaf, fi,
1136 btrfs_set_file_extent_offset(leaf, fi,
1137 start - orig_offset);
1138 btrfs_mark_buffer_dirty(leaf);
1143 while (start > key.offset || end < extent_end) {
1144 if (key.offset == start)
1147 new_key.offset = split;
1148 ret = btrfs_duplicate_item(trans, root, path, &new_key);
1149 if (ret == -EAGAIN) {
1150 btrfs_release_path(path);
1154 btrfs_abort_transaction(trans, root, ret);
1158 leaf = path->nodes[0];
1159 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1160 struct btrfs_file_extent_item);
1161 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1162 btrfs_set_file_extent_num_bytes(leaf, fi,
1163 split - key.offset);
1165 fi = btrfs_item_ptr(leaf, path->slots[0],
1166 struct btrfs_file_extent_item);
1168 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1169 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
1170 btrfs_set_file_extent_num_bytes(leaf, fi,
1171 extent_end - split);
1172 btrfs_mark_buffer_dirty(leaf);
1174 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
1175 root->root_key.objectid,
1176 ino, orig_offset, 0);
1177 BUG_ON(ret); /* -ENOMEM */
1179 if (split == start) {
1182 BUG_ON(start != key.offset);
1191 if (extent_mergeable(leaf, path->slots[0] + 1,
1192 ino, bytenr, orig_offset,
1193 &other_start, &other_end)) {
1195 btrfs_release_path(path);
1198 extent_end = other_end;
1199 del_slot = path->slots[0] + 1;
1201 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1202 0, root->root_key.objectid,
1203 ino, orig_offset, 0);
1204 BUG_ON(ret); /* -ENOMEM */
1208 if (extent_mergeable(leaf, path->slots[0] - 1,
1209 ino, bytenr, orig_offset,
1210 &other_start, &other_end)) {
1212 btrfs_release_path(path);
1215 key.offset = other_start;
1216 del_slot = path->slots[0];
1218 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1219 0, root->root_key.objectid,
1220 ino, orig_offset, 0);
1221 BUG_ON(ret); /* -ENOMEM */
1224 fi = btrfs_item_ptr(leaf, path->slots[0],
1225 struct btrfs_file_extent_item);
1226 btrfs_set_file_extent_type(leaf, fi,
1227 BTRFS_FILE_EXTENT_REG);
1228 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1229 btrfs_mark_buffer_dirty(leaf);
1231 fi = btrfs_item_ptr(leaf, del_slot - 1,
1232 struct btrfs_file_extent_item);
1233 btrfs_set_file_extent_type(leaf, fi,
1234 BTRFS_FILE_EXTENT_REG);
1235 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1236 btrfs_set_file_extent_num_bytes(leaf, fi,
1237 extent_end - key.offset);
1238 btrfs_mark_buffer_dirty(leaf);
1240 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1242 btrfs_abort_transaction(trans, root, ret);
1247 btrfs_free_path(path);
1252 * on error we return an unlocked page and the error value
1253 * on success we return a locked page and 0
1255 static int prepare_uptodate_page(struct page *page, u64 pos,
1256 bool force_uptodate)
1260 if (((pos & (PAGE_CACHE_SIZE - 1)) || force_uptodate) &&
1261 !PageUptodate(page)) {
1262 ret = btrfs_readpage(NULL, page);
1266 if (!PageUptodate(page)) {
1275 * this just gets pages into the page cache and locks them down.
1277 static noinline int prepare_pages(struct inode *inode, struct page **pages,
1278 size_t num_pages, loff_t pos,
1279 size_t write_bytes, bool force_uptodate)
1282 unsigned long index = pos >> PAGE_CACHE_SHIFT;
1283 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1287 for (i = 0; i < num_pages; i++) {
1288 pages[i] = find_or_create_page(inode->i_mapping, index + i,
1289 mask | __GFP_WRITE);
1297 err = prepare_uptodate_page(pages[i], pos,
1299 if (i == num_pages - 1)
1300 err = prepare_uptodate_page(pages[i],
1301 pos + write_bytes, false);
1303 page_cache_release(pages[i]);
1307 wait_on_page_writeback(pages[i]);
1312 while (faili >= 0) {
1313 unlock_page(pages[faili]);
1314 page_cache_release(pages[faili]);
1322 * This function locks the extent and properly waits for data=ordered extents
1323 * to finish before allowing the pages to be modified if need.
1326 * 1 - the extent is locked
1327 * 0 - the extent is not locked, and everything is OK
1328 * -EAGAIN - need re-prepare the pages
1329 * the other < 0 number - Something wrong happens
1332 lock_and_cleanup_extent_if_need(struct inode *inode, struct page **pages,
1333 size_t num_pages, loff_t pos,
1334 u64 *lockstart, u64 *lockend,
1335 struct extent_state **cached_state)
1342 start_pos = pos & ~((u64)PAGE_CACHE_SIZE - 1);
1343 last_pos = start_pos + ((u64)num_pages << PAGE_CACHE_SHIFT) - 1;
1345 if (start_pos < inode->i_size) {
1346 struct btrfs_ordered_extent *ordered;
1347 lock_extent_bits(&BTRFS_I(inode)->io_tree,
1348 start_pos, last_pos, 0, cached_state);
1349 ordered = btrfs_lookup_first_ordered_extent(inode, last_pos);
1351 ordered->file_offset + ordered->len > start_pos &&
1352 ordered->file_offset <= last_pos) {
1353 btrfs_put_ordered_extent(ordered);
1354 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1355 start_pos, last_pos,
1356 cached_state, GFP_NOFS);
1357 for (i = 0; i < num_pages; i++) {
1358 unlock_page(pages[i]);
1359 page_cache_release(pages[i]);
1361 ret = btrfs_wait_ordered_range(inode, start_pos,
1362 last_pos - start_pos + 1);
1369 btrfs_put_ordered_extent(ordered);
1371 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
1372 last_pos, EXTENT_DIRTY | EXTENT_DELALLOC |
1373 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
1374 0, 0, cached_state, GFP_NOFS);
1375 *lockstart = start_pos;
1376 *lockend = last_pos;
1380 for (i = 0; i < num_pages; i++) {
1381 if (clear_page_dirty_for_io(pages[i]))
1382 account_page_redirty(pages[i]);
1383 set_page_extent_mapped(pages[i]);
1384 WARN_ON(!PageLocked(pages[i]));
1390 static noinline int check_can_nocow(struct inode *inode, loff_t pos,
1391 size_t *write_bytes)
1393 struct btrfs_root *root = BTRFS_I(inode)->root;
1394 struct btrfs_ordered_extent *ordered;
1395 u64 lockstart, lockend;
1399 lockstart = round_down(pos, root->sectorsize);
1400 lockend = lockstart + round_up(*write_bytes, root->sectorsize) - 1;
1403 lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1404 ordered = btrfs_lookup_ordered_range(inode, lockstart,
1405 lockend - lockstart + 1);
1409 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1410 btrfs_start_ordered_extent(inode, ordered, 1);
1411 btrfs_put_ordered_extent(ordered);
1414 num_bytes = lockend - lockstart + 1;
1415 ret = can_nocow_extent(inode, lockstart, &num_bytes, NULL, NULL, NULL);
1419 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
1420 EXTENT_DIRTY | EXTENT_DELALLOC |
1421 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1423 *write_bytes = min_t(size_t, *write_bytes, num_bytes);
1426 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1431 static noinline ssize_t __btrfs_buffered_write(struct file *file,
1435 struct inode *inode = file_inode(file);
1436 struct btrfs_root *root = BTRFS_I(inode)->root;
1437 struct page **pages = NULL;
1438 struct extent_state *cached_state = NULL;
1439 u64 release_bytes = 0;
1442 unsigned long first_index;
1443 size_t num_written = 0;
1446 bool only_release_metadata = false;
1447 bool force_page_uptodate = false;
1450 nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) /
1451 PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
1452 (sizeof(struct page *)));
1453 nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1454 nrptrs = max(nrptrs, 8);
1455 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
1459 first_index = pos >> PAGE_CACHE_SHIFT;
1461 while (iov_iter_count(i) > 0) {
1462 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1463 size_t write_bytes = min(iov_iter_count(i),
1464 nrptrs * (size_t)PAGE_CACHE_SIZE -
1466 size_t num_pages = (write_bytes + offset +
1467 PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1468 size_t reserve_bytes;
1472 WARN_ON(num_pages > nrptrs);
1475 * Fault pages before locking them in prepare_pages
1476 * to avoid recursive lock
1478 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1483 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1484 ret = btrfs_check_data_free_space(inode, reserve_bytes);
1485 if (ret == -ENOSPC &&
1486 (BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW |
1487 BTRFS_INODE_PREALLOC))) {
1488 ret = check_can_nocow(inode, pos, &write_bytes);
1490 only_release_metadata = true;
1492 * our prealloc extent may be smaller than
1493 * write_bytes, so scale down.
1495 num_pages = (write_bytes + offset +
1496 PAGE_CACHE_SIZE - 1) >>
1498 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1508 ret = btrfs_delalloc_reserve_metadata(inode, reserve_bytes);
1510 if (!only_release_metadata)
1511 btrfs_free_reserved_data_space(inode,
1516 release_bytes = reserve_bytes;
1517 need_unlock = false;
1520 * This is going to setup the pages array with the number of
1521 * pages we want, so we don't really need to worry about the
1522 * contents of pages from loop to loop
1524 ret = prepare_pages(inode, pages, num_pages,
1526 force_page_uptodate);
1530 ret = lock_and_cleanup_extent_if_need(inode, pages, num_pages,
1531 pos, &lockstart, &lockend,
1537 } else if (ret > 0) {
1542 copied = btrfs_copy_from_user(pos, num_pages,
1543 write_bytes, pages, i);
1546 * if we have trouble faulting in the pages, fall
1547 * back to one page at a time
1549 if (copied < write_bytes)
1553 force_page_uptodate = true;
1556 force_page_uptodate = false;
1557 dirty_pages = (copied + offset +
1558 PAGE_CACHE_SIZE - 1) >>
1563 * If we had a short copy we need to release the excess delaloc
1564 * bytes we reserved. We need to increment outstanding_extents
1565 * because btrfs_delalloc_release_space will decrement it, but
1566 * we still have an outstanding extent for the chunk we actually
1569 if (num_pages > dirty_pages) {
1570 release_bytes = (num_pages - dirty_pages) <<
1573 spin_lock(&BTRFS_I(inode)->lock);
1574 BTRFS_I(inode)->outstanding_extents++;
1575 spin_unlock(&BTRFS_I(inode)->lock);
1577 if (only_release_metadata)
1578 btrfs_delalloc_release_metadata(inode,
1581 btrfs_delalloc_release_space(inode,
1585 release_bytes = dirty_pages << PAGE_CACHE_SHIFT;
1588 ret = btrfs_dirty_pages(root, inode, pages,
1589 dirty_pages, pos, copied,
1592 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1593 lockstart, lockend, &cached_state,
1596 btrfs_drop_pages(pages, num_pages);
1601 if (only_release_metadata && copied > 0) {
1602 u64 lockstart = round_down(pos, root->sectorsize);
1603 u64 lockend = lockstart +
1604 (dirty_pages << PAGE_CACHE_SHIFT) - 1;
1606 set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
1607 lockend, EXTENT_NORESERVE, NULL,
1609 only_release_metadata = false;
1612 btrfs_drop_pages(pages, num_pages);
1616 balance_dirty_pages_ratelimited(inode->i_mapping);
1617 if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1618 btrfs_btree_balance_dirty(root);
1621 num_written += copied;
1626 if (release_bytes) {
1627 if (only_release_metadata)
1628 btrfs_delalloc_release_metadata(inode, release_bytes);
1630 btrfs_delalloc_release_space(inode, release_bytes);
1633 return num_written ? num_written : ret;
1636 static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1637 const struct iovec *iov,
1638 unsigned long nr_segs, loff_t pos,
1639 loff_t *ppos, size_t count, size_t ocount)
1641 struct file *file = iocb->ki_filp;
1644 ssize_t written_buffered;
1648 written = generic_file_direct_write(iocb, iov, &nr_segs, pos, ppos,
1651 if (written < 0 || written == count)
1656 iov_iter_init(&i, iov, nr_segs, count, written);
1657 written_buffered = __btrfs_buffered_write(file, &i, pos);
1658 if (written_buffered < 0) {
1659 err = written_buffered;
1662 endbyte = pos + written_buffered - 1;
1663 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
1666 written += written_buffered;
1667 *ppos = pos + written_buffered;
1668 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
1669 endbyte >> PAGE_CACHE_SHIFT);
1671 return written ? written : err;
1674 static void update_time_for_write(struct inode *inode)
1676 struct timespec now;
1678 if (IS_NOCMTIME(inode))
1681 now = current_fs_time(inode->i_sb);
1682 if (!timespec_equal(&inode->i_mtime, &now))
1683 inode->i_mtime = now;
1685 if (!timespec_equal(&inode->i_ctime, &now))
1686 inode->i_ctime = now;
1688 if (IS_I_VERSION(inode))
1689 inode_inc_iversion(inode);
1692 static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
1693 const struct iovec *iov,
1694 unsigned long nr_segs, loff_t pos)
1696 struct file *file = iocb->ki_filp;
1697 struct inode *inode = file_inode(file);
1698 struct btrfs_root *root = BTRFS_I(inode)->root;
1699 loff_t *ppos = &iocb->ki_pos;
1701 ssize_t num_written = 0;
1703 size_t count, ocount;
1704 bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host);
1706 mutex_lock(&inode->i_mutex);
1708 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
1710 mutex_unlock(&inode->i_mutex);
1715 current->backing_dev_info = inode->i_mapping->backing_dev_info;
1716 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1718 mutex_unlock(&inode->i_mutex);
1723 mutex_unlock(&inode->i_mutex);
1727 err = file_remove_suid(file);
1729 mutex_unlock(&inode->i_mutex);
1734 * If BTRFS flips readonly due to some impossible error
1735 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1736 * although we have opened a file as writable, we have
1737 * to stop this write operation to ensure FS consistency.
1739 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
1740 mutex_unlock(&inode->i_mutex);
1746 * We reserve space for updating the inode when we reserve space for the
1747 * extent we are going to write, so we will enospc out there. We don't
1748 * need to start yet another transaction to update the inode as we will
1749 * update the inode when we finish writing whatever data we write.
1751 update_time_for_write(inode);
1753 start_pos = round_down(pos, root->sectorsize);
1754 if (start_pos > i_size_read(inode)) {
1755 err = btrfs_cont_expand(inode, i_size_read(inode), start_pos);
1757 mutex_unlock(&inode->i_mutex);
1763 atomic_inc(&BTRFS_I(inode)->sync_writers);
1765 if (unlikely(file->f_flags & O_DIRECT)) {
1766 num_written = __btrfs_direct_write(iocb, iov, nr_segs,
1767 pos, ppos, count, ocount);
1771 iov_iter_init(&i, iov, nr_segs, count, num_written);
1773 num_written = __btrfs_buffered_write(file, &i, pos);
1774 if (num_written > 0)
1775 *ppos = pos + num_written;
1778 mutex_unlock(&inode->i_mutex);
1781 * we want to make sure fsync finds this change
1782 * but we haven't joined a transaction running right now.
1784 * Later on, someone is sure to update the inode and get the
1785 * real transid recorded.
1787 * We set last_trans now to the fs_info generation + 1,
1788 * this will either be one more than the running transaction
1789 * or the generation used for the next transaction if there isn't
1790 * one running right now.
1792 * We also have to set last_sub_trans to the current log transid,
1793 * otherwise subsequent syncs to a file that's been synced in this
1794 * transaction will appear to have already occured.
1796 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1797 BTRFS_I(inode)->last_sub_trans = root->log_transid;
1798 if (num_written > 0) {
1799 err = generic_write_sync(file, pos, num_written);
1805 atomic_dec(&BTRFS_I(inode)->sync_writers);
1807 current->backing_dev_info = NULL;
1808 return num_written ? num_written : err;
1811 int btrfs_release_file(struct inode *inode, struct file *filp)
1814 * ordered_data_close is set by settattr when we are about to truncate
1815 * a file from a non-zero size to a zero size. This tries to
1816 * flush down new bytes that may have been written if the
1817 * application were using truncate to replace a file in place.
1819 if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
1820 &BTRFS_I(inode)->runtime_flags)) {
1821 struct btrfs_trans_handle *trans;
1822 struct btrfs_root *root = BTRFS_I(inode)->root;
1825 * We need to block on a committing transaction to keep us from
1826 * throwing a ordered operation on to the list and causing
1827 * something like sync to deadlock trying to flush out this
1830 trans = btrfs_start_transaction(root, 0);
1832 return PTR_ERR(trans);
1833 btrfs_add_ordered_operation(trans, BTRFS_I(inode)->root, inode);
1834 btrfs_end_transaction(trans, root);
1835 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1836 filemap_flush(inode->i_mapping);
1838 if (filp->private_data)
1839 btrfs_ioctl_trans_end(filp);
1844 * fsync call for both files and directories. This logs the inode into
1845 * the tree log instead of forcing full commits whenever possible.
1847 * It needs to call filemap_fdatawait so that all ordered extent updates are
1848 * in the metadata btree are up to date for copying to the log.
1850 * It drops the inode mutex before doing the tree log commit. This is an
1851 * important optimization for directories because holding the mutex prevents
1852 * new operations on the dir while we write to disk.
1854 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1856 struct dentry *dentry = file->f_path.dentry;
1857 struct inode *inode = dentry->d_inode;
1858 struct btrfs_root *root = BTRFS_I(inode)->root;
1860 struct btrfs_trans_handle *trans;
1863 trace_btrfs_sync_file(file, datasync);
1866 * We write the dirty pages in the range and wait until they complete
1867 * out of the ->i_mutex. If so, we can flush the dirty pages by
1868 * multi-task, and make the performance up. See
1869 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1871 atomic_inc(&BTRFS_I(inode)->sync_writers);
1872 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
1873 if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1874 &BTRFS_I(inode)->runtime_flags))
1875 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
1876 atomic_dec(&BTRFS_I(inode)->sync_writers);
1880 mutex_lock(&inode->i_mutex);
1883 * We flush the dirty pages again to avoid some dirty pages in the
1886 atomic_inc(&root->log_batch);
1887 full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1888 &BTRFS_I(inode)->runtime_flags);
1890 ret = btrfs_wait_ordered_range(inode, start, end - start + 1);
1892 mutex_unlock(&inode->i_mutex);
1896 atomic_inc(&root->log_batch);
1899 * check the transaction that last modified this inode
1900 * and see if its already been committed
1902 if (!BTRFS_I(inode)->last_trans) {
1903 mutex_unlock(&inode->i_mutex);
1908 * if the last transaction that changed this file was before
1909 * the current transaction, we can bail out now without any
1913 if (btrfs_inode_in_log(inode, root->fs_info->generation) ||
1914 BTRFS_I(inode)->last_trans <=
1915 root->fs_info->last_trans_committed) {
1916 BTRFS_I(inode)->last_trans = 0;
1919 * We'v had everything committed since the last time we were
1920 * modified so clear this flag in case it was set for whatever
1921 * reason, it's no longer relevant.
1923 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1924 &BTRFS_I(inode)->runtime_flags);
1925 mutex_unlock(&inode->i_mutex);
1930 * ok we haven't committed the transaction yet, lets do a commit
1932 if (file->private_data)
1933 btrfs_ioctl_trans_end(file);
1936 * We use start here because we will need to wait on the IO to complete
1937 * in btrfs_sync_log, which could require joining a transaction (for
1938 * example checking cross references in the nocow path). If we use join
1939 * here we could get into a situation where we're waiting on IO to
1940 * happen that is blocked on a transaction trying to commit. With start
1941 * we inc the extwriter counter, so we wait for all extwriters to exit
1942 * before we start blocking join'ers. This comment is to keep somebody
1943 * from thinking they are super smart and changing this to
1944 * btrfs_join_transaction *cough*Josef*cough*.
1946 trans = btrfs_start_transaction(root, 0);
1947 if (IS_ERR(trans)) {
1948 ret = PTR_ERR(trans);
1949 mutex_unlock(&inode->i_mutex);
1954 ret = btrfs_log_dentry_safe(trans, root, dentry);
1956 /* Fallthrough and commit/free transaction. */
1960 /* we've logged all the items and now have a consistent
1961 * version of the file in the log. It is possible that
1962 * someone will come in and modify the file, but that's
1963 * fine because the log is consistent on disk, and we
1964 * have references to all of the file's extents
1966 * It is possible that someone will come in and log the
1967 * file again, but that will end up using the synchronization
1968 * inside btrfs_sync_log to keep things safe.
1970 mutex_unlock(&inode->i_mutex);
1972 if (ret != BTRFS_NO_LOG_SYNC) {
1974 ret = btrfs_sync_log(trans, root);
1976 ret = btrfs_end_transaction(trans, root);
1981 ret = btrfs_wait_ordered_range(inode, start,
1986 ret = btrfs_commit_transaction(trans, root);
1988 ret = btrfs_end_transaction(trans, root);
1991 return ret > 0 ? -EIO : ret;
1994 static const struct vm_operations_struct btrfs_file_vm_ops = {
1995 .fault = filemap_fault,
1996 .page_mkwrite = btrfs_page_mkwrite,
1997 .remap_pages = generic_file_remap_pages,
2000 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
2002 struct address_space *mapping = filp->f_mapping;
2004 if (!mapping->a_ops->readpage)
2007 file_accessed(filp);
2008 vma->vm_ops = &btrfs_file_vm_ops;
2013 static int hole_mergeable(struct inode *inode, struct extent_buffer *leaf,
2014 int slot, u64 start, u64 end)
2016 struct btrfs_file_extent_item *fi;
2017 struct btrfs_key key;
2019 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
2022 btrfs_item_key_to_cpu(leaf, &key, slot);
2023 if (key.objectid != btrfs_ino(inode) ||
2024 key.type != BTRFS_EXTENT_DATA_KEY)
2027 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2029 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2032 if (btrfs_file_extent_disk_bytenr(leaf, fi))
2035 if (key.offset == end)
2037 if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
2042 static int fill_holes(struct btrfs_trans_handle *trans, struct inode *inode,
2043 struct btrfs_path *path, u64 offset, u64 end)
2045 struct btrfs_root *root = BTRFS_I(inode)->root;
2046 struct extent_buffer *leaf;
2047 struct btrfs_file_extent_item *fi;
2048 struct extent_map *hole_em;
2049 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2050 struct btrfs_key key;
2053 if (btrfs_fs_incompat(root->fs_info, NO_HOLES))
2056 key.objectid = btrfs_ino(inode);
2057 key.type = BTRFS_EXTENT_DATA_KEY;
2058 key.offset = offset;
2060 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2065 leaf = path->nodes[0];
2066 if (hole_mergeable(inode, leaf, path->slots[0]-1, offset, end)) {
2070 fi = btrfs_item_ptr(leaf, path->slots[0],
2071 struct btrfs_file_extent_item);
2072 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2074 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2075 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2076 btrfs_set_file_extent_offset(leaf, fi, 0);
2077 btrfs_mark_buffer_dirty(leaf);
2081 if (hole_mergeable(inode, leaf, path->slots[0]+1, offset, end)) {
2085 key.offset = offset;
2086 btrfs_set_item_key_safe(root, path, &key);
2087 fi = btrfs_item_ptr(leaf, path->slots[0],
2088 struct btrfs_file_extent_item);
2089 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2091 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2092 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2093 btrfs_set_file_extent_offset(leaf, fi, 0);
2094 btrfs_mark_buffer_dirty(leaf);
2097 btrfs_release_path(path);
2099 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
2100 0, 0, end - offset, 0, end - offset,
2106 btrfs_release_path(path);
2108 hole_em = alloc_extent_map();
2110 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2111 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2112 &BTRFS_I(inode)->runtime_flags);
2114 hole_em->start = offset;
2115 hole_em->len = end - offset;
2116 hole_em->ram_bytes = hole_em->len;
2117 hole_em->orig_start = offset;
2119 hole_em->block_start = EXTENT_MAP_HOLE;
2120 hole_em->block_len = 0;
2121 hole_em->orig_block_len = 0;
2122 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
2123 hole_em->compress_type = BTRFS_COMPRESS_NONE;
2124 hole_em->generation = trans->transid;
2127 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2128 write_lock(&em_tree->lock);
2129 ret = add_extent_mapping(em_tree, hole_em, 1);
2130 write_unlock(&em_tree->lock);
2131 } while (ret == -EEXIST);
2132 free_extent_map(hole_em);
2134 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2135 &BTRFS_I(inode)->runtime_flags);
2141 static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
2143 struct btrfs_root *root = BTRFS_I(inode)->root;
2144 struct extent_state *cached_state = NULL;
2145 struct btrfs_path *path;
2146 struct btrfs_block_rsv *rsv;
2147 struct btrfs_trans_handle *trans;
2148 u64 lockstart = round_up(offset, BTRFS_I(inode)->root->sectorsize);
2149 u64 lockend = round_down(offset + len,
2150 BTRFS_I(inode)->root->sectorsize) - 1;
2151 u64 cur_offset = lockstart;
2152 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
2157 bool same_page = ((offset >> PAGE_CACHE_SHIFT) ==
2158 ((offset + len - 1) >> PAGE_CACHE_SHIFT));
2159 bool no_holes = btrfs_fs_incompat(root->fs_info, NO_HOLES);
2161 ret = btrfs_wait_ordered_range(inode, offset, len);
2165 mutex_lock(&inode->i_mutex);
2167 * We needn't truncate any page which is beyond the end of the file
2168 * because we are sure there is no data there.
2171 * Only do this if we are in the same page and we aren't doing the
2174 if (same_page && len < PAGE_CACHE_SIZE) {
2175 if (offset < round_up(inode->i_size, PAGE_CACHE_SIZE))
2176 ret = btrfs_truncate_page(inode, offset, len, 0);
2177 mutex_unlock(&inode->i_mutex);
2181 /* zero back part of the first page */
2182 if (offset < round_up(inode->i_size, PAGE_CACHE_SIZE)) {
2183 ret = btrfs_truncate_page(inode, offset, 0, 0);
2185 mutex_unlock(&inode->i_mutex);
2190 /* zero the front end of the last page */
2191 if (offset + len < round_up(inode->i_size, PAGE_CACHE_SIZE)) {
2192 ret = btrfs_truncate_page(inode, offset + len, 0, 1);
2194 mutex_unlock(&inode->i_mutex);
2199 if (lockend < lockstart) {
2200 mutex_unlock(&inode->i_mutex);
2205 struct btrfs_ordered_extent *ordered;
2207 truncate_pagecache_range(inode, lockstart, lockend);
2209 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2211 ordered = btrfs_lookup_first_ordered_extent(inode, lockend);
2214 * We need to make sure we have no ordered extents in this range
2215 * and nobody raced in and read a page in this range, if we did
2216 * we need to try again.
2219 (ordered->file_offset + ordered->len <= lockstart ||
2220 ordered->file_offset > lockend)) &&
2221 !test_range_bit(&BTRFS_I(inode)->io_tree, lockstart,
2222 lockend, EXTENT_UPTODATE, 0,
2225 btrfs_put_ordered_extent(ordered);
2229 btrfs_put_ordered_extent(ordered);
2230 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
2231 lockend, &cached_state, GFP_NOFS);
2232 ret = btrfs_wait_ordered_range(inode, lockstart,
2233 lockend - lockstart + 1);
2235 mutex_unlock(&inode->i_mutex);
2240 path = btrfs_alloc_path();
2246 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2251 rsv->size = btrfs_calc_trunc_metadata_size(root, 1);
2255 * 1 - update the inode
2256 * 1 - removing the extents in the range
2257 * 1 - adding the hole extent if no_holes isn't set
2259 rsv_count = no_holes ? 2 : 3;
2260 trans = btrfs_start_transaction(root, rsv_count);
2261 if (IS_ERR(trans)) {
2262 err = PTR_ERR(trans);
2266 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
2269 trans->block_rsv = rsv;
2271 while (cur_offset < lockend) {
2272 ret = __btrfs_drop_extents(trans, root, inode, path,
2273 cur_offset, lockend + 1,
2274 &drop_end, 1, 0, 0, NULL);
2278 trans->block_rsv = &root->fs_info->trans_block_rsv;
2280 ret = fill_holes(trans, inode, path, cur_offset, drop_end);
2286 cur_offset = drop_end;
2288 ret = btrfs_update_inode(trans, root, inode);
2294 btrfs_end_transaction(trans, root);
2295 btrfs_btree_balance_dirty(root);
2297 trans = btrfs_start_transaction(root, rsv_count);
2298 if (IS_ERR(trans)) {
2299 ret = PTR_ERR(trans);
2304 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
2306 BUG_ON(ret); /* shouldn't happen */
2307 trans->block_rsv = rsv;
2315 trans->block_rsv = &root->fs_info->trans_block_rsv;
2316 ret = fill_holes(trans, inode, path, cur_offset, drop_end);
2326 inode_inc_iversion(inode);
2327 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2329 trans->block_rsv = &root->fs_info->trans_block_rsv;
2330 ret = btrfs_update_inode(trans, root, inode);
2331 btrfs_end_transaction(trans, root);
2332 btrfs_btree_balance_dirty(root);
2334 btrfs_free_path(path);
2335 btrfs_free_block_rsv(root, rsv);
2337 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2338 &cached_state, GFP_NOFS);
2339 mutex_unlock(&inode->i_mutex);
2345 static long btrfs_fallocate(struct file *file, int mode,
2346 loff_t offset, loff_t len)
2348 struct inode *inode = file_inode(file);
2349 struct extent_state *cached_state = NULL;
2350 struct btrfs_root *root = BTRFS_I(inode)->root;
2357 struct extent_map *em;
2358 int blocksize = BTRFS_I(inode)->root->sectorsize;
2361 alloc_start = round_down(offset, blocksize);
2362 alloc_end = round_up(offset + len, blocksize);
2364 /* Make sure we aren't being give some crap mode */
2365 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2368 if (mode & FALLOC_FL_PUNCH_HOLE)
2369 return btrfs_punch_hole(inode, offset, len);
2372 * Make sure we have enough space before we do the
2375 ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
2378 if (root->fs_info->quota_enabled) {
2379 ret = btrfs_qgroup_reserve(root, alloc_end - alloc_start);
2381 goto out_reserve_fail;
2384 mutex_lock(&inode->i_mutex);
2385 ret = inode_newsize_ok(inode, alloc_end);
2389 if (alloc_start > inode->i_size) {
2390 ret = btrfs_cont_expand(inode, i_size_read(inode),
2396 * If we are fallocating from the end of the file onward we
2397 * need to zero out the end of the page if i_size lands in the
2400 ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
2406 * wait for ordered IO before we have any locks. We'll loop again
2407 * below with the locks held.
2409 ret = btrfs_wait_ordered_range(inode, alloc_start,
2410 alloc_end - alloc_start);
2414 locked_end = alloc_end - 1;
2416 struct btrfs_ordered_extent *ordered;
2418 /* the extent lock is ordered inside the running
2421 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
2422 locked_end, 0, &cached_state);
2423 ordered = btrfs_lookup_first_ordered_extent(inode,
2426 ordered->file_offset + ordered->len > alloc_start &&
2427 ordered->file_offset < alloc_end) {
2428 btrfs_put_ordered_extent(ordered);
2429 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
2430 alloc_start, locked_end,
2431 &cached_state, GFP_NOFS);
2433 * we can't wait on the range with the transaction
2434 * running or with the extent lock held
2436 ret = btrfs_wait_ordered_range(inode, alloc_start,
2437 alloc_end - alloc_start);
2442 btrfs_put_ordered_extent(ordered);
2447 cur_offset = alloc_start;
2451 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2452 alloc_end - cur_offset, 0);
2453 if (IS_ERR_OR_NULL(em)) {
2460 last_byte = min(extent_map_end(em), alloc_end);
2461 actual_end = min_t(u64, extent_map_end(em), offset + len);
2462 last_byte = ALIGN(last_byte, blocksize);
2464 if (em->block_start == EXTENT_MAP_HOLE ||
2465 (cur_offset >= inode->i_size &&
2466 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
2467 ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
2468 last_byte - cur_offset,
2469 1 << inode->i_blkbits,
2474 free_extent_map(em);
2477 } else if (actual_end > inode->i_size &&
2478 !(mode & FALLOC_FL_KEEP_SIZE)) {
2480 * We didn't need to allocate any more space, but we
2481 * still extended the size of the file so we need to
2484 inode->i_ctime = CURRENT_TIME;
2485 i_size_write(inode, actual_end);
2486 btrfs_ordered_update_i_size(inode, actual_end, NULL);
2488 free_extent_map(em);
2490 cur_offset = last_byte;
2491 if (cur_offset >= alloc_end) {
2496 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
2497 &cached_state, GFP_NOFS);
2499 mutex_unlock(&inode->i_mutex);
2500 if (root->fs_info->quota_enabled)
2501 btrfs_qgroup_free(root, alloc_end - alloc_start);
2503 /* Let go of our reservation. */
2504 btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
2508 static int find_desired_extent(struct inode *inode, loff_t *offset, int whence)
2510 struct btrfs_root *root = BTRFS_I(inode)->root;
2511 struct extent_map *em = NULL;
2512 struct extent_state *cached_state = NULL;
2513 u64 lockstart = *offset;
2514 u64 lockend = i_size_read(inode);
2515 u64 start = *offset;
2516 u64 len = i_size_read(inode);
2519 lockend = max_t(u64, root->sectorsize, lockend);
2520 if (lockend <= lockstart)
2521 lockend = lockstart + root->sectorsize;
2524 len = lockend - lockstart + 1;
2526 len = max_t(u64, len, root->sectorsize);
2527 if (inode->i_size == 0)
2530 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, 0,
2533 while (start < inode->i_size) {
2534 em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0);
2541 if (whence == SEEK_HOLE &&
2542 (em->block_start == EXTENT_MAP_HOLE ||
2543 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2545 else if (whence == SEEK_DATA &&
2546 (em->block_start != EXTENT_MAP_HOLE &&
2547 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2550 start = em->start + em->len;
2551 free_extent_map(em);
2555 free_extent_map(em);
2557 if (whence == SEEK_DATA && start >= inode->i_size)
2560 *offset = min_t(loff_t, start, inode->i_size);
2562 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2563 &cached_state, GFP_NOFS);
2567 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
2569 struct inode *inode = file->f_mapping->host;
2572 mutex_lock(&inode->i_mutex);
2576 offset = generic_file_llseek(file, offset, whence);
2580 if (offset >= i_size_read(inode)) {
2581 mutex_unlock(&inode->i_mutex);
2585 ret = find_desired_extent(inode, &offset, whence);
2587 mutex_unlock(&inode->i_mutex);
2592 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2594 mutex_unlock(&inode->i_mutex);
2598 const struct file_operations btrfs_file_operations = {
2599 .llseek = btrfs_file_llseek,
2600 .read = do_sync_read,
2601 .write = do_sync_write,
2602 .aio_read = generic_file_aio_read,
2603 .splice_read = generic_file_splice_read,
2604 .aio_write = btrfs_file_aio_write,
2605 .mmap = btrfs_file_mmap,
2606 .open = generic_file_open,
2607 .release = btrfs_release_file,
2608 .fsync = btrfs_sync_file,
2609 .fallocate = btrfs_fallocate,
2610 .unlocked_ioctl = btrfs_ioctl,
2611 #ifdef CONFIG_COMPAT
2612 .compat_ioctl = btrfs_ioctl,
2616 void btrfs_auto_defrag_exit(void)
2618 if (btrfs_inode_defrag_cachep)
2619 kmem_cache_destroy(btrfs_inode_defrag_cachep);
2622 int btrfs_auto_defrag_init(void)
2624 btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
2625 sizeof(struct inode_defrag), 0,
2626 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2628 if (!btrfs_inode_defrag_cachep)