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 && !replace_extent)
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 (if replace_extent != 0).
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);
949 leaf = path->nodes[0];
951 * If btrfs_del_items() was called, it might have deleted a leaf, in
952 * which case it unlocked our path, so check path->locks[0] matches a
955 if (!ret && replace_extent && leafs_visited == 1 &&
956 (path->locks[0] == BTRFS_WRITE_LOCK_BLOCKING ||
957 path->locks[0] == BTRFS_WRITE_LOCK) &&
958 btrfs_leaf_free_space(root, leaf) >=
959 sizeof(struct btrfs_item) + extent_item_size) {
962 key.type = BTRFS_EXTENT_DATA_KEY;
964 if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) {
965 struct btrfs_key slot_key;
967 btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]);
968 if (btrfs_comp_cpu_keys(&key, &slot_key) > 0)
971 setup_items_for_insert(root, path, &key,
974 sizeof(struct btrfs_item) +
975 extent_item_size, 1);
979 if (!replace_extent || !(*key_inserted))
980 btrfs_release_path(path);
982 *drop_end = found ? min(end, extent_end) : end;
986 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
987 struct btrfs_root *root, struct inode *inode, u64 start,
988 u64 end, int drop_cache)
990 struct btrfs_path *path;
993 path = btrfs_alloc_path();
996 ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL,
997 drop_cache, 0, 0, NULL);
998 btrfs_free_path(path);
1002 static int extent_mergeable(struct extent_buffer *leaf, int slot,
1003 u64 objectid, u64 bytenr, u64 orig_offset,
1004 u64 *start, u64 *end)
1006 struct btrfs_file_extent_item *fi;
1007 struct btrfs_key key;
1010 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
1013 btrfs_item_key_to_cpu(leaf, &key, slot);
1014 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
1017 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1018 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
1019 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
1020 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
1021 btrfs_file_extent_compression(leaf, fi) ||
1022 btrfs_file_extent_encryption(leaf, fi) ||
1023 btrfs_file_extent_other_encoding(leaf, fi))
1026 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1027 if ((*start && *start != key.offset) || (*end && *end != extent_end))
1030 *start = key.offset;
1036 * Mark extent in the range start - end as written.
1038 * This changes extent type from 'pre-allocated' to 'regular'. If only
1039 * part of extent is marked as written, the extent will be split into
1042 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
1043 struct inode *inode, u64 start, u64 end)
1045 struct btrfs_root *root = BTRFS_I(inode)->root;
1046 struct extent_buffer *leaf;
1047 struct btrfs_path *path;
1048 struct btrfs_file_extent_item *fi;
1049 struct btrfs_key key;
1050 struct btrfs_key new_key;
1062 u64 ino = btrfs_ino(inode);
1064 path = btrfs_alloc_path();
1071 key.type = BTRFS_EXTENT_DATA_KEY;
1074 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1077 if (ret > 0 && path->slots[0] > 0)
1080 leaf = path->nodes[0];
1081 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1082 BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY);
1083 fi = btrfs_item_ptr(leaf, path->slots[0],
1084 struct btrfs_file_extent_item);
1085 BUG_ON(btrfs_file_extent_type(leaf, fi) !=
1086 BTRFS_FILE_EXTENT_PREALLOC);
1087 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1088 BUG_ON(key.offset > start || extent_end < end);
1090 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1091 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1092 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
1093 memcpy(&new_key, &key, sizeof(new_key));
1095 if (start == key.offset && end < extent_end) {
1098 if (extent_mergeable(leaf, path->slots[0] - 1,
1099 ino, bytenr, orig_offset,
1100 &other_start, &other_end)) {
1101 new_key.offset = end;
1102 btrfs_set_item_key_safe(root, path, &new_key);
1103 fi = btrfs_item_ptr(leaf, path->slots[0],
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_set_file_extent_offset(leaf, fi,
1111 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1112 struct btrfs_file_extent_item);
1113 btrfs_set_file_extent_generation(leaf, fi,
1115 btrfs_set_file_extent_num_bytes(leaf, fi,
1117 btrfs_mark_buffer_dirty(leaf);
1122 if (start > key.offset && end == extent_end) {
1125 if (extent_mergeable(leaf, path->slots[0] + 1,
1126 ino, bytenr, orig_offset,
1127 &other_start, &other_end)) {
1128 fi = btrfs_item_ptr(leaf, path->slots[0],
1129 struct btrfs_file_extent_item);
1130 btrfs_set_file_extent_num_bytes(leaf, fi,
1131 start - key.offset);
1132 btrfs_set_file_extent_generation(leaf, fi,
1135 new_key.offset = start;
1136 btrfs_set_item_key_safe(root, path, &new_key);
1138 fi = btrfs_item_ptr(leaf, path->slots[0],
1139 struct btrfs_file_extent_item);
1140 btrfs_set_file_extent_generation(leaf, fi,
1142 btrfs_set_file_extent_num_bytes(leaf, fi,
1144 btrfs_set_file_extent_offset(leaf, fi,
1145 start - orig_offset);
1146 btrfs_mark_buffer_dirty(leaf);
1151 while (start > key.offset || end < extent_end) {
1152 if (key.offset == start)
1155 new_key.offset = split;
1156 ret = btrfs_duplicate_item(trans, root, path, &new_key);
1157 if (ret == -EAGAIN) {
1158 btrfs_release_path(path);
1162 btrfs_abort_transaction(trans, root, ret);
1166 leaf = path->nodes[0];
1167 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1168 struct btrfs_file_extent_item);
1169 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1170 btrfs_set_file_extent_num_bytes(leaf, fi,
1171 split - key.offset);
1173 fi = btrfs_item_ptr(leaf, path->slots[0],
1174 struct btrfs_file_extent_item);
1176 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1177 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
1178 btrfs_set_file_extent_num_bytes(leaf, fi,
1179 extent_end - split);
1180 btrfs_mark_buffer_dirty(leaf);
1182 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
1183 root->root_key.objectid,
1184 ino, orig_offset, 0);
1185 BUG_ON(ret); /* -ENOMEM */
1187 if (split == start) {
1190 BUG_ON(start != key.offset);
1199 if (extent_mergeable(leaf, path->slots[0] + 1,
1200 ino, bytenr, orig_offset,
1201 &other_start, &other_end)) {
1203 btrfs_release_path(path);
1206 extent_end = other_end;
1207 del_slot = path->slots[0] + 1;
1209 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1210 0, root->root_key.objectid,
1211 ino, orig_offset, 0);
1212 BUG_ON(ret); /* -ENOMEM */
1216 if (extent_mergeable(leaf, path->slots[0] - 1,
1217 ino, bytenr, orig_offset,
1218 &other_start, &other_end)) {
1220 btrfs_release_path(path);
1223 key.offset = other_start;
1224 del_slot = path->slots[0];
1226 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1227 0, root->root_key.objectid,
1228 ino, orig_offset, 0);
1229 BUG_ON(ret); /* -ENOMEM */
1232 fi = btrfs_item_ptr(leaf, path->slots[0],
1233 struct btrfs_file_extent_item);
1234 btrfs_set_file_extent_type(leaf, fi,
1235 BTRFS_FILE_EXTENT_REG);
1236 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1237 btrfs_mark_buffer_dirty(leaf);
1239 fi = btrfs_item_ptr(leaf, del_slot - 1,
1240 struct btrfs_file_extent_item);
1241 btrfs_set_file_extent_type(leaf, fi,
1242 BTRFS_FILE_EXTENT_REG);
1243 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1244 btrfs_set_file_extent_num_bytes(leaf, fi,
1245 extent_end - key.offset);
1246 btrfs_mark_buffer_dirty(leaf);
1248 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1250 btrfs_abort_transaction(trans, root, ret);
1255 btrfs_free_path(path);
1260 * on error we return an unlocked page and the error value
1261 * on success we return a locked page and 0
1263 static int prepare_uptodate_page(struct page *page, u64 pos,
1264 bool force_uptodate)
1268 if (((pos & (PAGE_CACHE_SIZE - 1)) || force_uptodate) &&
1269 !PageUptodate(page)) {
1270 ret = btrfs_readpage(NULL, page);
1274 if (!PageUptodate(page)) {
1283 * this just gets pages into the page cache and locks them down.
1285 static noinline int prepare_pages(struct inode *inode, struct page **pages,
1286 size_t num_pages, loff_t pos,
1287 size_t write_bytes, bool force_uptodate)
1290 unsigned long index = pos >> PAGE_CACHE_SHIFT;
1291 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1295 for (i = 0; i < num_pages; i++) {
1296 pages[i] = find_or_create_page(inode->i_mapping, index + i,
1297 mask | __GFP_WRITE);
1305 err = prepare_uptodate_page(pages[i], pos,
1307 if (i == num_pages - 1)
1308 err = prepare_uptodate_page(pages[i],
1309 pos + write_bytes, false);
1311 page_cache_release(pages[i]);
1315 wait_on_page_writeback(pages[i]);
1320 while (faili >= 0) {
1321 unlock_page(pages[faili]);
1322 page_cache_release(pages[faili]);
1330 * This function locks the extent and properly waits for data=ordered extents
1331 * to finish before allowing the pages to be modified if need.
1334 * 1 - the extent is locked
1335 * 0 - the extent is not locked, and everything is OK
1336 * -EAGAIN - need re-prepare the pages
1337 * the other < 0 number - Something wrong happens
1340 lock_and_cleanup_extent_if_need(struct inode *inode, struct page **pages,
1341 size_t num_pages, loff_t pos,
1342 u64 *lockstart, u64 *lockend,
1343 struct extent_state **cached_state)
1350 start_pos = pos & ~((u64)PAGE_CACHE_SIZE - 1);
1351 last_pos = start_pos + ((u64)num_pages << PAGE_CACHE_SHIFT) - 1;
1353 if (start_pos < inode->i_size) {
1354 struct btrfs_ordered_extent *ordered;
1355 lock_extent_bits(&BTRFS_I(inode)->io_tree,
1356 start_pos, last_pos, 0, cached_state);
1357 ordered = btrfs_lookup_first_ordered_extent(inode, last_pos);
1359 ordered->file_offset + ordered->len > start_pos &&
1360 ordered->file_offset <= last_pos) {
1361 btrfs_put_ordered_extent(ordered);
1362 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1363 start_pos, last_pos,
1364 cached_state, GFP_NOFS);
1365 for (i = 0; i < num_pages; i++) {
1366 unlock_page(pages[i]);
1367 page_cache_release(pages[i]);
1369 ret = btrfs_wait_ordered_range(inode, start_pos,
1370 last_pos - start_pos + 1);
1377 btrfs_put_ordered_extent(ordered);
1379 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
1380 last_pos, EXTENT_DIRTY | EXTENT_DELALLOC |
1381 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
1382 0, 0, cached_state, GFP_NOFS);
1383 *lockstart = start_pos;
1384 *lockend = last_pos;
1388 for (i = 0; i < num_pages; i++) {
1389 if (clear_page_dirty_for_io(pages[i]))
1390 account_page_redirty(pages[i]);
1391 set_page_extent_mapped(pages[i]);
1392 WARN_ON(!PageLocked(pages[i]));
1398 static noinline int check_can_nocow(struct inode *inode, loff_t pos,
1399 size_t *write_bytes)
1401 struct btrfs_root *root = BTRFS_I(inode)->root;
1402 struct btrfs_ordered_extent *ordered;
1403 u64 lockstart, lockend;
1407 lockstart = round_down(pos, root->sectorsize);
1408 lockend = lockstart + round_up(*write_bytes, root->sectorsize) - 1;
1411 lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1412 ordered = btrfs_lookup_ordered_range(inode, lockstart,
1413 lockend - lockstart + 1);
1417 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1418 btrfs_start_ordered_extent(inode, ordered, 1);
1419 btrfs_put_ordered_extent(ordered);
1422 num_bytes = lockend - lockstart + 1;
1423 ret = can_nocow_extent(inode, lockstart, &num_bytes, NULL, NULL, NULL);
1427 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
1428 EXTENT_DIRTY | EXTENT_DELALLOC |
1429 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1431 *write_bytes = min_t(size_t, *write_bytes, num_bytes);
1434 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1439 static noinline ssize_t __btrfs_buffered_write(struct file *file,
1443 struct inode *inode = file_inode(file);
1444 struct btrfs_root *root = BTRFS_I(inode)->root;
1445 struct page **pages = NULL;
1446 struct extent_state *cached_state = NULL;
1447 u64 release_bytes = 0;
1450 unsigned long first_index;
1451 size_t num_written = 0;
1454 bool only_release_metadata = false;
1455 bool force_page_uptodate = false;
1458 nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) /
1459 PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
1460 (sizeof(struct page *)));
1461 nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1462 nrptrs = max(nrptrs, 8);
1463 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
1467 first_index = pos >> PAGE_CACHE_SHIFT;
1469 while (iov_iter_count(i) > 0) {
1470 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1471 size_t write_bytes = min(iov_iter_count(i),
1472 nrptrs * (size_t)PAGE_CACHE_SIZE -
1474 size_t num_pages = (write_bytes + offset +
1475 PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1476 size_t reserve_bytes;
1480 WARN_ON(num_pages > nrptrs);
1483 * Fault pages before locking them in prepare_pages
1484 * to avoid recursive lock
1486 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1491 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1492 ret = btrfs_check_data_free_space(inode, reserve_bytes);
1493 if (ret == -ENOSPC &&
1494 (BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW |
1495 BTRFS_INODE_PREALLOC))) {
1496 ret = check_can_nocow(inode, pos, &write_bytes);
1498 only_release_metadata = true;
1500 * our prealloc extent may be smaller than
1501 * write_bytes, so scale down.
1503 num_pages = (write_bytes + offset +
1504 PAGE_CACHE_SIZE - 1) >>
1506 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1516 ret = btrfs_delalloc_reserve_metadata(inode, reserve_bytes);
1518 if (!only_release_metadata)
1519 btrfs_free_reserved_data_space(inode,
1524 release_bytes = reserve_bytes;
1525 need_unlock = false;
1528 * This is going to setup the pages array with the number of
1529 * pages we want, so we don't really need to worry about the
1530 * contents of pages from loop to loop
1532 ret = prepare_pages(inode, pages, num_pages,
1534 force_page_uptodate);
1538 ret = lock_and_cleanup_extent_if_need(inode, pages, num_pages,
1539 pos, &lockstart, &lockend,
1545 } else if (ret > 0) {
1550 copied = btrfs_copy_from_user(pos, num_pages,
1551 write_bytes, pages, i);
1554 * if we have trouble faulting in the pages, fall
1555 * back to one page at a time
1557 if (copied < write_bytes)
1561 force_page_uptodate = true;
1564 force_page_uptodate = false;
1565 dirty_pages = (copied + offset +
1566 PAGE_CACHE_SIZE - 1) >>
1571 * If we had a short copy we need to release the excess delaloc
1572 * bytes we reserved. We need to increment outstanding_extents
1573 * because btrfs_delalloc_release_space will decrement it, but
1574 * we still have an outstanding extent for the chunk we actually
1577 if (num_pages > dirty_pages) {
1578 release_bytes = (num_pages - dirty_pages) <<
1581 spin_lock(&BTRFS_I(inode)->lock);
1582 BTRFS_I(inode)->outstanding_extents++;
1583 spin_unlock(&BTRFS_I(inode)->lock);
1585 if (only_release_metadata)
1586 btrfs_delalloc_release_metadata(inode,
1589 btrfs_delalloc_release_space(inode,
1593 release_bytes = dirty_pages << PAGE_CACHE_SHIFT;
1596 ret = btrfs_dirty_pages(root, inode, pages,
1597 dirty_pages, pos, copied,
1600 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1601 lockstart, lockend, &cached_state,
1604 btrfs_drop_pages(pages, num_pages);
1609 if (only_release_metadata && copied > 0) {
1610 u64 lockstart = round_down(pos, root->sectorsize);
1611 u64 lockend = lockstart +
1612 (dirty_pages << PAGE_CACHE_SHIFT) - 1;
1614 set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
1615 lockend, EXTENT_NORESERVE, NULL,
1617 only_release_metadata = false;
1620 btrfs_drop_pages(pages, num_pages);
1624 balance_dirty_pages_ratelimited(inode->i_mapping);
1625 if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1626 btrfs_btree_balance_dirty(root);
1629 num_written += copied;
1634 if (release_bytes) {
1635 if (only_release_metadata)
1636 btrfs_delalloc_release_metadata(inode, release_bytes);
1638 btrfs_delalloc_release_space(inode, release_bytes);
1641 return num_written ? num_written : ret;
1644 static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1645 const struct iovec *iov,
1646 unsigned long nr_segs, loff_t pos,
1647 loff_t *ppos, size_t count, size_t ocount)
1649 struct file *file = iocb->ki_filp;
1652 ssize_t written_buffered;
1656 written = generic_file_direct_write(iocb, iov, &nr_segs, pos, ppos,
1659 if (written < 0 || written == count)
1664 iov_iter_init(&i, iov, nr_segs, count, written);
1665 written_buffered = __btrfs_buffered_write(file, &i, pos);
1666 if (written_buffered < 0) {
1667 err = written_buffered;
1670 endbyte = pos + written_buffered - 1;
1671 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
1674 written += written_buffered;
1675 *ppos = pos + written_buffered;
1676 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
1677 endbyte >> PAGE_CACHE_SHIFT);
1679 return written ? written : err;
1682 static void update_time_for_write(struct inode *inode)
1684 struct timespec now;
1686 if (IS_NOCMTIME(inode))
1689 now = current_fs_time(inode->i_sb);
1690 if (!timespec_equal(&inode->i_mtime, &now))
1691 inode->i_mtime = now;
1693 if (!timespec_equal(&inode->i_ctime, &now))
1694 inode->i_ctime = now;
1696 if (IS_I_VERSION(inode))
1697 inode_inc_iversion(inode);
1700 static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
1701 const struct iovec *iov,
1702 unsigned long nr_segs, loff_t pos)
1704 struct file *file = iocb->ki_filp;
1705 struct inode *inode = file_inode(file);
1706 struct btrfs_root *root = BTRFS_I(inode)->root;
1707 loff_t *ppos = &iocb->ki_pos;
1709 ssize_t num_written = 0;
1711 size_t count, ocount;
1712 bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host);
1714 mutex_lock(&inode->i_mutex);
1716 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
1718 mutex_unlock(&inode->i_mutex);
1723 current->backing_dev_info = inode->i_mapping->backing_dev_info;
1724 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1726 mutex_unlock(&inode->i_mutex);
1731 mutex_unlock(&inode->i_mutex);
1735 err = file_remove_suid(file);
1737 mutex_unlock(&inode->i_mutex);
1742 * If BTRFS flips readonly due to some impossible error
1743 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1744 * although we have opened a file as writable, we have
1745 * to stop this write operation to ensure FS consistency.
1747 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
1748 mutex_unlock(&inode->i_mutex);
1754 * We reserve space for updating the inode when we reserve space for the
1755 * extent we are going to write, so we will enospc out there. We don't
1756 * need to start yet another transaction to update the inode as we will
1757 * update the inode when we finish writing whatever data we write.
1759 update_time_for_write(inode);
1761 start_pos = round_down(pos, root->sectorsize);
1762 if (start_pos > i_size_read(inode)) {
1763 err = btrfs_cont_expand(inode, i_size_read(inode), start_pos);
1765 mutex_unlock(&inode->i_mutex);
1771 atomic_inc(&BTRFS_I(inode)->sync_writers);
1773 if (unlikely(file->f_flags & O_DIRECT)) {
1774 num_written = __btrfs_direct_write(iocb, iov, nr_segs,
1775 pos, ppos, count, ocount);
1779 iov_iter_init(&i, iov, nr_segs, count, num_written);
1781 num_written = __btrfs_buffered_write(file, &i, pos);
1782 if (num_written > 0)
1783 *ppos = pos + num_written;
1786 mutex_unlock(&inode->i_mutex);
1789 * we want to make sure fsync finds this change
1790 * but we haven't joined a transaction running right now.
1792 * Later on, someone is sure to update the inode and get the
1793 * real transid recorded.
1795 * We set last_trans now to the fs_info generation + 1,
1796 * this will either be one more than the running transaction
1797 * or the generation used for the next transaction if there isn't
1798 * one running right now.
1800 * We also have to set last_sub_trans to the current log transid,
1801 * otherwise subsequent syncs to a file that's been synced in this
1802 * transaction will appear to have already occured.
1804 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1805 BTRFS_I(inode)->last_sub_trans = root->log_transid;
1806 if (num_written > 0) {
1807 err = generic_write_sync(file, pos, num_written);
1808 if (err < 0 && num_written > 0)
1813 atomic_dec(&BTRFS_I(inode)->sync_writers);
1815 current->backing_dev_info = NULL;
1816 return num_written ? num_written : err;
1819 int btrfs_release_file(struct inode *inode, struct file *filp)
1822 * ordered_data_close is set by settattr when we are about to truncate
1823 * a file from a non-zero size to a zero size. This tries to
1824 * flush down new bytes that may have been written if the
1825 * application were using truncate to replace a file in place.
1827 if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
1828 &BTRFS_I(inode)->runtime_flags)) {
1829 struct btrfs_trans_handle *trans;
1830 struct btrfs_root *root = BTRFS_I(inode)->root;
1833 * We need to block on a committing transaction to keep us from
1834 * throwing a ordered operation on to the list and causing
1835 * something like sync to deadlock trying to flush out this
1838 trans = btrfs_start_transaction(root, 0);
1840 return PTR_ERR(trans);
1841 btrfs_add_ordered_operation(trans, BTRFS_I(inode)->root, inode);
1842 btrfs_end_transaction(trans, root);
1843 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1844 filemap_flush(inode->i_mapping);
1846 if (filp->private_data)
1847 btrfs_ioctl_trans_end(filp);
1852 * fsync call for both files and directories. This logs the inode into
1853 * the tree log instead of forcing full commits whenever possible.
1855 * It needs to call filemap_fdatawait so that all ordered extent updates are
1856 * in the metadata btree are up to date for copying to the log.
1858 * It drops the inode mutex before doing the tree log commit. This is an
1859 * important optimization for directories because holding the mutex prevents
1860 * new operations on the dir while we write to disk.
1862 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1864 struct dentry *dentry = file->f_path.dentry;
1865 struct inode *inode = dentry->d_inode;
1866 struct btrfs_root *root = BTRFS_I(inode)->root;
1868 struct btrfs_trans_handle *trans;
1871 trace_btrfs_sync_file(file, datasync);
1874 * We write the dirty pages in the range and wait until they complete
1875 * out of the ->i_mutex. If so, we can flush the dirty pages by
1876 * multi-task, and make the performance up. See
1877 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1879 atomic_inc(&BTRFS_I(inode)->sync_writers);
1880 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
1881 if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1882 &BTRFS_I(inode)->runtime_flags))
1883 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
1884 atomic_dec(&BTRFS_I(inode)->sync_writers);
1888 mutex_lock(&inode->i_mutex);
1891 * We flush the dirty pages again to avoid some dirty pages in the
1894 atomic_inc(&root->log_batch);
1895 full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1896 &BTRFS_I(inode)->runtime_flags);
1898 ret = btrfs_wait_ordered_range(inode, start, end - start + 1);
1900 mutex_unlock(&inode->i_mutex);
1904 atomic_inc(&root->log_batch);
1907 * check the transaction that last modified this inode
1908 * and see if its already been committed
1910 if (!BTRFS_I(inode)->last_trans) {
1911 mutex_unlock(&inode->i_mutex);
1916 * if the last transaction that changed this file was before
1917 * the current transaction, we can bail out now without any
1921 if (btrfs_inode_in_log(inode, root->fs_info->generation) ||
1922 BTRFS_I(inode)->last_trans <=
1923 root->fs_info->last_trans_committed) {
1924 BTRFS_I(inode)->last_trans = 0;
1927 * We'v had everything committed since the last time we were
1928 * modified so clear this flag in case it was set for whatever
1929 * reason, it's no longer relevant.
1931 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1932 &BTRFS_I(inode)->runtime_flags);
1933 mutex_unlock(&inode->i_mutex);
1938 * ok we haven't committed the transaction yet, lets do a commit
1940 if (file->private_data)
1941 btrfs_ioctl_trans_end(file);
1944 * We use start here because we will need to wait on the IO to complete
1945 * in btrfs_sync_log, which could require joining a transaction (for
1946 * example checking cross references in the nocow path). If we use join
1947 * here we could get into a situation where we're waiting on IO to
1948 * happen that is blocked on a transaction trying to commit. With start
1949 * we inc the extwriter counter, so we wait for all extwriters to exit
1950 * before we start blocking join'ers. This comment is to keep somebody
1951 * from thinking they are super smart and changing this to
1952 * btrfs_join_transaction *cough*Josef*cough*.
1954 trans = btrfs_start_transaction(root, 0);
1955 if (IS_ERR(trans)) {
1956 ret = PTR_ERR(trans);
1957 mutex_unlock(&inode->i_mutex);
1962 ret = btrfs_log_dentry_safe(trans, root, dentry);
1964 /* Fallthrough and commit/free transaction. */
1968 /* we've logged all the items and now have a consistent
1969 * version of the file in the log. It is possible that
1970 * someone will come in and modify the file, but that's
1971 * fine because the log is consistent on disk, and we
1972 * have references to all of the file's extents
1974 * It is possible that someone will come in and log the
1975 * file again, but that will end up using the synchronization
1976 * inside btrfs_sync_log to keep things safe.
1978 mutex_unlock(&inode->i_mutex);
1980 if (ret != BTRFS_NO_LOG_SYNC) {
1982 ret = btrfs_sync_log(trans, root);
1984 ret = btrfs_end_transaction(trans, root);
1989 ret = btrfs_wait_ordered_range(inode, start,
1994 ret = btrfs_commit_transaction(trans, root);
1996 ret = btrfs_end_transaction(trans, root);
1999 return ret > 0 ? -EIO : ret;
2002 static const struct vm_operations_struct btrfs_file_vm_ops = {
2003 .fault = filemap_fault,
2004 .page_mkwrite = btrfs_page_mkwrite,
2005 .remap_pages = generic_file_remap_pages,
2008 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
2010 struct address_space *mapping = filp->f_mapping;
2012 if (!mapping->a_ops->readpage)
2015 file_accessed(filp);
2016 vma->vm_ops = &btrfs_file_vm_ops;
2021 static int hole_mergeable(struct inode *inode, struct extent_buffer *leaf,
2022 int slot, u64 start, u64 end)
2024 struct btrfs_file_extent_item *fi;
2025 struct btrfs_key key;
2027 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
2030 btrfs_item_key_to_cpu(leaf, &key, slot);
2031 if (key.objectid != btrfs_ino(inode) ||
2032 key.type != BTRFS_EXTENT_DATA_KEY)
2035 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2037 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2040 if (btrfs_file_extent_disk_bytenr(leaf, fi))
2043 if (key.offset == end)
2045 if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
2050 static int fill_holes(struct btrfs_trans_handle *trans, struct inode *inode,
2051 struct btrfs_path *path, u64 offset, u64 end)
2053 struct btrfs_root *root = BTRFS_I(inode)->root;
2054 struct extent_buffer *leaf;
2055 struct btrfs_file_extent_item *fi;
2056 struct extent_map *hole_em;
2057 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2058 struct btrfs_key key;
2061 if (btrfs_fs_incompat(root->fs_info, NO_HOLES))
2064 key.objectid = btrfs_ino(inode);
2065 key.type = BTRFS_EXTENT_DATA_KEY;
2066 key.offset = offset;
2068 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2073 leaf = path->nodes[0];
2074 if (hole_mergeable(inode, leaf, path->slots[0]-1, offset, end)) {
2078 fi = btrfs_item_ptr(leaf, path->slots[0],
2079 struct btrfs_file_extent_item);
2080 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2082 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2083 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2084 btrfs_set_file_extent_offset(leaf, fi, 0);
2085 btrfs_mark_buffer_dirty(leaf);
2089 if (hole_mergeable(inode, leaf, path->slots[0]+1, offset, end)) {
2093 key.offset = offset;
2094 btrfs_set_item_key_safe(root, path, &key);
2095 fi = btrfs_item_ptr(leaf, path->slots[0],
2096 struct btrfs_file_extent_item);
2097 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2099 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2100 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2101 btrfs_set_file_extent_offset(leaf, fi, 0);
2102 btrfs_mark_buffer_dirty(leaf);
2105 btrfs_release_path(path);
2107 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
2108 0, 0, end - offset, 0, end - offset,
2114 btrfs_release_path(path);
2116 hole_em = alloc_extent_map();
2118 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2119 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2120 &BTRFS_I(inode)->runtime_flags);
2122 hole_em->start = offset;
2123 hole_em->len = end - offset;
2124 hole_em->ram_bytes = hole_em->len;
2125 hole_em->orig_start = offset;
2127 hole_em->block_start = EXTENT_MAP_HOLE;
2128 hole_em->block_len = 0;
2129 hole_em->orig_block_len = 0;
2130 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
2131 hole_em->compress_type = BTRFS_COMPRESS_NONE;
2132 hole_em->generation = trans->transid;
2135 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2136 write_lock(&em_tree->lock);
2137 ret = add_extent_mapping(em_tree, hole_em, 1);
2138 write_unlock(&em_tree->lock);
2139 } while (ret == -EEXIST);
2140 free_extent_map(hole_em);
2142 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2143 &BTRFS_I(inode)->runtime_flags);
2149 static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
2151 struct btrfs_root *root = BTRFS_I(inode)->root;
2152 struct extent_state *cached_state = NULL;
2153 struct btrfs_path *path;
2154 struct btrfs_block_rsv *rsv;
2155 struct btrfs_trans_handle *trans;
2156 u64 lockstart = round_up(offset, BTRFS_I(inode)->root->sectorsize);
2157 u64 lockend = round_down(offset + len,
2158 BTRFS_I(inode)->root->sectorsize) - 1;
2159 u64 cur_offset = lockstart;
2160 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
2165 bool same_page = ((offset >> PAGE_CACHE_SHIFT) ==
2166 ((offset + len - 1) >> PAGE_CACHE_SHIFT));
2167 bool no_holes = btrfs_fs_incompat(root->fs_info, NO_HOLES);
2169 ret = btrfs_wait_ordered_range(inode, offset, len);
2173 mutex_lock(&inode->i_mutex);
2175 * We needn't truncate any page which is beyond the end of the file
2176 * because we are sure there is no data there.
2179 * Only do this if we are in the same page and we aren't doing the
2182 if (same_page && len < PAGE_CACHE_SIZE) {
2183 if (offset < round_up(inode->i_size, PAGE_CACHE_SIZE))
2184 ret = btrfs_truncate_page(inode, offset, len, 0);
2185 mutex_unlock(&inode->i_mutex);
2189 /* zero back part of the first page */
2190 if (offset < round_up(inode->i_size, PAGE_CACHE_SIZE)) {
2191 ret = btrfs_truncate_page(inode, offset, 0, 0);
2193 mutex_unlock(&inode->i_mutex);
2198 /* zero the front end of the last page */
2199 if (offset + len < round_up(inode->i_size, PAGE_CACHE_SIZE)) {
2200 ret = btrfs_truncate_page(inode, offset + len, 0, 1);
2202 mutex_unlock(&inode->i_mutex);
2207 if (lockend < lockstart) {
2208 mutex_unlock(&inode->i_mutex);
2213 struct btrfs_ordered_extent *ordered;
2215 truncate_pagecache_range(inode, lockstart, lockend);
2217 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2219 ordered = btrfs_lookup_first_ordered_extent(inode, lockend);
2222 * We need to make sure we have no ordered extents in this range
2223 * and nobody raced in and read a page in this range, if we did
2224 * we need to try again.
2227 (ordered->file_offset + ordered->len <= lockstart ||
2228 ordered->file_offset > lockend)) &&
2229 !test_range_bit(&BTRFS_I(inode)->io_tree, lockstart,
2230 lockend, EXTENT_UPTODATE, 0,
2233 btrfs_put_ordered_extent(ordered);
2237 btrfs_put_ordered_extent(ordered);
2238 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
2239 lockend, &cached_state, GFP_NOFS);
2240 ret = btrfs_wait_ordered_range(inode, lockstart,
2241 lockend - lockstart + 1);
2243 mutex_unlock(&inode->i_mutex);
2248 path = btrfs_alloc_path();
2254 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2259 rsv->size = btrfs_calc_trunc_metadata_size(root, 1);
2263 * 1 - update the inode
2264 * 1 - removing the extents in the range
2265 * 1 - adding the hole extent if no_holes isn't set
2267 rsv_count = no_holes ? 2 : 3;
2268 trans = btrfs_start_transaction(root, rsv_count);
2269 if (IS_ERR(trans)) {
2270 err = PTR_ERR(trans);
2274 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
2277 trans->block_rsv = rsv;
2279 while (cur_offset < lockend) {
2280 ret = __btrfs_drop_extents(trans, root, inode, path,
2281 cur_offset, lockend + 1,
2282 &drop_end, 1, 0, 0, NULL);
2286 trans->block_rsv = &root->fs_info->trans_block_rsv;
2288 ret = fill_holes(trans, inode, path, cur_offset, drop_end);
2294 cur_offset = drop_end;
2296 ret = btrfs_update_inode(trans, root, inode);
2302 btrfs_end_transaction(trans, root);
2303 btrfs_btree_balance_dirty(root);
2305 trans = btrfs_start_transaction(root, rsv_count);
2306 if (IS_ERR(trans)) {
2307 ret = PTR_ERR(trans);
2312 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
2314 BUG_ON(ret); /* shouldn't happen */
2315 trans->block_rsv = rsv;
2323 trans->block_rsv = &root->fs_info->trans_block_rsv;
2324 ret = fill_holes(trans, inode, path, cur_offset, drop_end);
2334 inode_inc_iversion(inode);
2335 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2337 trans->block_rsv = &root->fs_info->trans_block_rsv;
2338 ret = btrfs_update_inode(trans, root, inode);
2339 btrfs_end_transaction(trans, root);
2340 btrfs_btree_balance_dirty(root);
2342 btrfs_free_path(path);
2343 btrfs_free_block_rsv(root, rsv);
2345 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2346 &cached_state, GFP_NOFS);
2347 mutex_unlock(&inode->i_mutex);
2353 static long btrfs_fallocate(struct file *file, int mode,
2354 loff_t offset, loff_t len)
2356 struct inode *inode = file_inode(file);
2357 struct extent_state *cached_state = NULL;
2358 struct btrfs_root *root = BTRFS_I(inode)->root;
2365 struct extent_map *em;
2366 int blocksize = BTRFS_I(inode)->root->sectorsize;
2369 alloc_start = round_down(offset, blocksize);
2370 alloc_end = round_up(offset + len, blocksize);
2372 /* Make sure we aren't being give some crap mode */
2373 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2376 if (mode & FALLOC_FL_PUNCH_HOLE)
2377 return btrfs_punch_hole(inode, offset, len);
2380 * Make sure we have enough space before we do the
2383 ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
2386 if (root->fs_info->quota_enabled) {
2387 ret = btrfs_qgroup_reserve(root, alloc_end - alloc_start);
2389 goto out_reserve_fail;
2392 mutex_lock(&inode->i_mutex);
2393 ret = inode_newsize_ok(inode, alloc_end);
2397 if (alloc_start > inode->i_size) {
2398 ret = btrfs_cont_expand(inode, i_size_read(inode),
2404 * If we are fallocating from the end of the file onward we
2405 * need to zero out the end of the page if i_size lands in the
2408 ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
2414 * wait for ordered IO before we have any locks. We'll loop again
2415 * below with the locks held.
2417 ret = btrfs_wait_ordered_range(inode, alloc_start,
2418 alloc_end - alloc_start);
2422 locked_end = alloc_end - 1;
2424 struct btrfs_ordered_extent *ordered;
2426 /* the extent lock is ordered inside the running
2429 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
2430 locked_end, 0, &cached_state);
2431 ordered = btrfs_lookup_first_ordered_extent(inode,
2434 ordered->file_offset + ordered->len > alloc_start &&
2435 ordered->file_offset < alloc_end) {
2436 btrfs_put_ordered_extent(ordered);
2437 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
2438 alloc_start, locked_end,
2439 &cached_state, GFP_NOFS);
2441 * we can't wait on the range with the transaction
2442 * running or with the extent lock held
2444 ret = btrfs_wait_ordered_range(inode, alloc_start,
2445 alloc_end - alloc_start);
2450 btrfs_put_ordered_extent(ordered);
2455 cur_offset = alloc_start;
2459 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2460 alloc_end - cur_offset, 0);
2461 if (IS_ERR_OR_NULL(em)) {
2468 last_byte = min(extent_map_end(em), alloc_end);
2469 actual_end = min_t(u64, extent_map_end(em), offset + len);
2470 last_byte = ALIGN(last_byte, blocksize);
2472 if (em->block_start == EXTENT_MAP_HOLE ||
2473 (cur_offset >= inode->i_size &&
2474 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
2475 ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
2476 last_byte - cur_offset,
2477 1 << inode->i_blkbits,
2482 free_extent_map(em);
2485 } else if (actual_end > inode->i_size &&
2486 !(mode & FALLOC_FL_KEEP_SIZE)) {
2488 * We didn't need to allocate any more space, but we
2489 * still extended the size of the file so we need to
2492 inode->i_ctime = CURRENT_TIME;
2493 i_size_write(inode, actual_end);
2494 btrfs_ordered_update_i_size(inode, actual_end, NULL);
2496 free_extent_map(em);
2498 cur_offset = last_byte;
2499 if (cur_offset >= alloc_end) {
2504 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
2505 &cached_state, GFP_NOFS);
2507 mutex_unlock(&inode->i_mutex);
2508 if (root->fs_info->quota_enabled)
2509 btrfs_qgroup_free(root, alloc_end - alloc_start);
2511 /* Let go of our reservation. */
2512 btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
2516 static int find_desired_extent(struct inode *inode, loff_t *offset, int whence)
2518 struct btrfs_root *root = BTRFS_I(inode)->root;
2519 struct extent_map *em = NULL;
2520 struct extent_state *cached_state = NULL;
2521 u64 lockstart = *offset;
2522 u64 lockend = i_size_read(inode);
2523 u64 start = *offset;
2524 u64 len = i_size_read(inode);
2527 lockend = max_t(u64, root->sectorsize, lockend);
2528 if (lockend <= lockstart)
2529 lockend = lockstart + root->sectorsize;
2532 len = lockend - lockstart + 1;
2534 len = max_t(u64, len, root->sectorsize);
2535 if (inode->i_size == 0)
2538 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, 0,
2541 while (start < inode->i_size) {
2542 em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0);
2549 if (whence == SEEK_HOLE &&
2550 (em->block_start == EXTENT_MAP_HOLE ||
2551 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2553 else if (whence == SEEK_DATA &&
2554 (em->block_start != EXTENT_MAP_HOLE &&
2555 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2558 start = em->start + em->len;
2559 free_extent_map(em);
2563 free_extent_map(em);
2565 if (whence == SEEK_DATA && start >= inode->i_size)
2568 *offset = min_t(loff_t, start, inode->i_size);
2570 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2571 &cached_state, GFP_NOFS);
2575 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
2577 struct inode *inode = file->f_mapping->host;
2580 mutex_lock(&inode->i_mutex);
2584 offset = generic_file_llseek(file, offset, whence);
2588 if (offset >= i_size_read(inode)) {
2589 mutex_unlock(&inode->i_mutex);
2593 ret = find_desired_extent(inode, &offset, whence);
2595 mutex_unlock(&inode->i_mutex);
2600 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2602 mutex_unlock(&inode->i_mutex);
2606 const struct file_operations btrfs_file_operations = {
2607 .llseek = btrfs_file_llseek,
2608 .read = do_sync_read,
2609 .write = do_sync_write,
2610 .aio_read = generic_file_aio_read,
2611 .splice_read = generic_file_splice_read,
2612 .aio_write = btrfs_file_aio_write,
2613 .mmap = btrfs_file_mmap,
2614 .open = generic_file_open,
2615 .release = btrfs_release_file,
2616 .fsync = btrfs_sync_file,
2617 .fallocate = btrfs_fallocate,
2618 .unlocked_ioctl = btrfs_ioctl,
2619 #ifdef CONFIG_COMPAT
2620 .compat_ioctl = btrfs_ioctl,
2624 void btrfs_auto_defrag_exit(void)
2626 if (btrfs_inode_defrag_cachep)
2627 kmem_cache_destroy(btrfs_inode_defrag_cachep);
2630 int btrfs_auto_defrag_init(void)
2632 btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
2633 sizeof(struct inode_defrag), 0,
2634 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2636 if (!btrfs_inode_defrag_cachep)