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/falloc.h>
28 #include <linux/swap.h>
29 #include <linux/writeback.h>
30 #include <linux/statfs.h>
31 #include <linux/compat.h>
32 #include <linux/slab.h>
33 #include <linux/btrfs.h>
36 #include "transaction.h"
37 #include "btrfs_inode.h"
38 #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->block_start < EXTENT_MAP_LAST_BYTE &&
600 split->start = em->start;
601 split->len = start - em->start;
602 split->orig_start = em->orig_start;
603 split->block_start = em->block_start;
606 split->block_len = em->block_len;
608 split->block_len = split->len;
609 split->ram_bytes = em->ram_bytes;
610 split->orig_block_len = max(split->block_len,
612 split->generation = gen;
613 split->bdev = em->bdev;
614 split->flags = flags;
615 split->compress_type = em->compress_type;
616 ret = add_extent_mapping(em_tree, split, modified);
617 BUG_ON(ret); /* Logic error */
618 free_extent_map(split);
622 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
623 testend && em->start + em->len > start + len) {
624 u64 diff = start + len - em->start;
626 split->start = start + len;
627 split->len = em->start + em->len - (start + len);
628 split->bdev = em->bdev;
629 split->flags = flags;
630 split->compress_type = em->compress_type;
631 split->generation = gen;
632 split->orig_block_len = max(em->block_len,
634 split->ram_bytes = em->ram_bytes;
637 split->block_len = em->block_len;
638 split->block_start = em->block_start;
639 split->orig_start = em->orig_start;
641 split->block_len = split->len;
642 split->block_start = em->block_start + diff;
643 split->orig_start = em->orig_start;
646 ret = add_extent_mapping(em_tree, split, modified);
647 BUG_ON(ret); /* Logic error */
648 free_extent_map(split);
652 write_unlock(&em_tree->lock);
656 /* once for the tree*/
660 free_extent_map(split);
662 free_extent_map(split2);
666 * this is very complex, but the basic idea is to drop all extents
667 * in the range start - end. hint_block is filled in with a block number
668 * that would be a good hint to the block allocator for this file.
670 * If an extent intersects the range but is not entirely inside the range
671 * it is either truncated or split. Anything entirely inside the range
672 * is deleted from the tree.
674 int __btrfs_drop_extents(struct btrfs_trans_handle *trans,
675 struct btrfs_root *root, struct inode *inode,
676 struct btrfs_path *path, u64 start, u64 end,
677 u64 *drop_end, int drop_cache)
679 struct extent_buffer *leaf;
680 struct btrfs_file_extent_item *fi;
681 struct btrfs_key key;
682 struct btrfs_key new_key;
683 u64 ino = btrfs_ino(inode);
684 u64 search_start = start;
687 u64 extent_offset = 0;
694 int modify_tree = -1;
695 int update_refs = (root->ref_cows || root == root->fs_info->tree_root);
699 btrfs_drop_extent_cache(inode, start, end - 1, 0);
701 if (start >= BTRFS_I(inode)->disk_i_size)
706 ret = btrfs_lookup_file_extent(trans, root, path, ino,
707 search_start, modify_tree);
710 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
711 leaf = path->nodes[0];
712 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
713 if (key.objectid == ino &&
714 key.type == BTRFS_EXTENT_DATA_KEY)
719 leaf = path->nodes[0];
720 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
722 ret = btrfs_next_leaf(root, path);
729 leaf = path->nodes[0];
733 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
734 if (key.objectid > ino ||
735 key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
738 fi = btrfs_item_ptr(leaf, path->slots[0],
739 struct btrfs_file_extent_item);
740 extent_type = btrfs_file_extent_type(leaf, fi);
742 if (extent_type == BTRFS_FILE_EXTENT_REG ||
743 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
744 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
745 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
746 extent_offset = btrfs_file_extent_offset(leaf, fi);
747 extent_end = key.offset +
748 btrfs_file_extent_num_bytes(leaf, fi);
749 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
750 extent_end = key.offset +
751 btrfs_file_extent_inline_len(leaf, fi);
754 extent_end = search_start;
757 if (extent_end <= search_start) {
763 search_start = max(key.offset, start);
764 if (recow || !modify_tree) {
766 btrfs_release_path(path);
771 * | - range to drop - |
772 * | -------- extent -------- |
774 if (start > key.offset && end < extent_end) {
776 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
778 memcpy(&new_key, &key, sizeof(new_key));
779 new_key.offset = start;
780 ret = btrfs_duplicate_item(trans, root, path,
782 if (ret == -EAGAIN) {
783 btrfs_release_path(path);
789 leaf = path->nodes[0];
790 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
791 struct btrfs_file_extent_item);
792 btrfs_set_file_extent_num_bytes(leaf, fi,
795 fi = btrfs_item_ptr(leaf, path->slots[0],
796 struct btrfs_file_extent_item);
798 extent_offset += start - key.offset;
799 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
800 btrfs_set_file_extent_num_bytes(leaf, fi,
802 btrfs_mark_buffer_dirty(leaf);
804 if (update_refs && disk_bytenr > 0) {
805 ret = btrfs_inc_extent_ref(trans, root,
806 disk_bytenr, num_bytes, 0,
807 root->root_key.objectid,
809 start - extent_offset, 0);
810 BUG_ON(ret); /* -ENOMEM */
815 * | ---- range to drop ----- |
816 * | -------- extent -------- |
818 if (start <= key.offset && end < extent_end) {
819 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
821 memcpy(&new_key, &key, sizeof(new_key));
822 new_key.offset = end;
823 btrfs_set_item_key_safe(root, path, &new_key);
825 extent_offset += end - key.offset;
826 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
827 btrfs_set_file_extent_num_bytes(leaf, fi,
829 btrfs_mark_buffer_dirty(leaf);
830 if (update_refs && disk_bytenr > 0)
831 inode_sub_bytes(inode, end - key.offset);
835 search_start = extent_end;
837 * | ---- range to drop ----- |
838 * | -------- extent -------- |
840 if (start > key.offset && end >= extent_end) {
842 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
844 btrfs_set_file_extent_num_bytes(leaf, fi,
846 btrfs_mark_buffer_dirty(leaf);
847 if (update_refs && disk_bytenr > 0)
848 inode_sub_bytes(inode, extent_end - start);
849 if (end == extent_end)
857 * | ---- range to drop ----- |
858 * | ------ extent ------ |
860 if (start <= key.offset && end >= extent_end) {
862 del_slot = path->slots[0];
865 BUG_ON(del_slot + del_nr != path->slots[0]);
870 extent_type == BTRFS_FILE_EXTENT_INLINE) {
871 inode_sub_bytes(inode,
872 extent_end - key.offset);
873 extent_end = ALIGN(extent_end,
875 } else if (update_refs && disk_bytenr > 0) {
876 ret = btrfs_free_extent(trans, root,
877 disk_bytenr, num_bytes, 0,
878 root->root_key.objectid,
879 key.objectid, key.offset -
881 BUG_ON(ret); /* -ENOMEM */
882 inode_sub_bytes(inode,
883 extent_end - key.offset);
886 if (end == extent_end)
889 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
894 ret = btrfs_del_items(trans, root, path, del_slot,
897 btrfs_abort_transaction(trans, root, ret);
904 btrfs_release_path(path);
911 if (!ret && del_nr > 0) {
912 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
914 btrfs_abort_transaction(trans, root, ret);
918 *drop_end = found ? min(end, extent_end) : end;
919 btrfs_release_path(path);
923 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
924 struct btrfs_root *root, struct inode *inode, u64 start,
925 u64 end, int drop_cache)
927 struct btrfs_path *path;
930 path = btrfs_alloc_path();
933 ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL,
935 btrfs_free_path(path);
939 static int extent_mergeable(struct extent_buffer *leaf, int slot,
940 u64 objectid, u64 bytenr, u64 orig_offset,
941 u64 *start, u64 *end)
943 struct btrfs_file_extent_item *fi;
944 struct btrfs_key key;
947 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
950 btrfs_item_key_to_cpu(leaf, &key, slot);
951 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
954 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
955 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
956 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
957 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
958 btrfs_file_extent_compression(leaf, fi) ||
959 btrfs_file_extent_encryption(leaf, fi) ||
960 btrfs_file_extent_other_encoding(leaf, fi))
963 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
964 if ((*start && *start != key.offset) || (*end && *end != extent_end))
973 * Mark extent in the range start - end as written.
975 * This changes extent type from 'pre-allocated' to 'regular'. If only
976 * part of extent is marked as written, the extent will be split into
979 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
980 struct inode *inode, u64 start, u64 end)
982 struct btrfs_root *root = BTRFS_I(inode)->root;
983 struct extent_buffer *leaf;
984 struct btrfs_path *path;
985 struct btrfs_file_extent_item *fi;
986 struct btrfs_key key;
987 struct btrfs_key new_key;
999 u64 ino = btrfs_ino(inode);
1001 path = btrfs_alloc_path();
1008 key.type = BTRFS_EXTENT_DATA_KEY;
1011 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1014 if (ret > 0 && path->slots[0] > 0)
1017 leaf = path->nodes[0];
1018 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1019 BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY);
1020 fi = btrfs_item_ptr(leaf, path->slots[0],
1021 struct btrfs_file_extent_item);
1022 BUG_ON(btrfs_file_extent_type(leaf, fi) !=
1023 BTRFS_FILE_EXTENT_PREALLOC);
1024 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1025 BUG_ON(key.offset > start || extent_end < end);
1027 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1028 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1029 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
1030 memcpy(&new_key, &key, sizeof(new_key));
1032 if (start == key.offset && end < extent_end) {
1035 if (extent_mergeable(leaf, path->slots[0] - 1,
1036 ino, bytenr, orig_offset,
1037 &other_start, &other_end)) {
1038 new_key.offset = end;
1039 btrfs_set_item_key_safe(root, path, &new_key);
1040 fi = btrfs_item_ptr(leaf, path->slots[0],
1041 struct btrfs_file_extent_item);
1042 btrfs_set_file_extent_generation(leaf, fi,
1044 btrfs_set_file_extent_num_bytes(leaf, fi,
1046 btrfs_set_file_extent_offset(leaf, fi,
1048 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1049 struct btrfs_file_extent_item);
1050 btrfs_set_file_extent_generation(leaf, fi,
1052 btrfs_set_file_extent_num_bytes(leaf, fi,
1054 btrfs_mark_buffer_dirty(leaf);
1059 if (start > key.offset && end == extent_end) {
1062 if (extent_mergeable(leaf, path->slots[0] + 1,
1063 ino, bytenr, orig_offset,
1064 &other_start, &other_end)) {
1065 fi = btrfs_item_ptr(leaf, path->slots[0],
1066 struct btrfs_file_extent_item);
1067 btrfs_set_file_extent_num_bytes(leaf, fi,
1068 start - key.offset);
1069 btrfs_set_file_extent_generation(leaf, fi,
1072 new_key.offset = start;
1073 btrfs_set_item_key_safe(root, path, &new_key);
1075 fi = btrfs_item_ptr(leaf, path->slots[0],
1076 struct btrfs_file_extent_item);
1077 btrfs_set_file_extent_generation(leaf, fi,
1079 btrfs_set_file_extent_num_bytes(leaf, fi,
1081 btrfs_set_file_extent_offset(leaf, fi,
1082 start - orig_offset);
1083 btrfs_mark_buffer_dirty(leaf);
1088 while (start > key.offset || end < extent_end) {
1089 if (key.offset == start)
1092 new_key.offset = split;
1093 ret = btrfs_duplicate_item(trans, root, path, &new_key);
1094 if (ret == -EAGAIN) {
1095 btrfs_release_path(path);
1099 btrfs_abort_transaction(trans, root, ret);
1103 leaf = path->nodes[0];
1104 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1105 struct btrfs_file_extent_item);
1106 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1107 btrfs_set_file_extent_num_bytes(leaf, fi,
1108 split - key.offset);
1110 fi = btrfs_item_ptr(leaf, path->slots[0],
1111 struct btrfs_file_extent_item);
1113 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1114 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
1115 btrfs_set_file_extent_num_bytes(leaf, fi,
1116 extent_end - split);
1117 btrfs_mark_buffer_dirty(leaf);
1119 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
1120 root->root_key.objectid,
1121 ino, orig_offset, 0);
1122 BUG_ON(ret); /* -ENOMEM */
1124 if (split == start) {
1127 BUG_ON(start != key.offset);
1136 if (extent_mergeable(leaf, path->slots[0] + 1,
1137 ino, bytenr, orig_offset,
1138 &other_start, &other_end)) {
1140 btrfs_release_path(path);
1143 extent_end = other_end;
1144 del_slot = path->slots[0] + 1;
1146 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1147 0, root->root_key.objectid,
1148 ino, orig_offset, 0);
1149 BUG_ON(ret); /* -ENOMEM */
1153 if (extent_mergeable(leaf, path->slots[0] - 1,
1154 ino, bytenr, orig_offset,
1155 &other_start, &other_end)) {
1157 btrfs_release_path(path);
1160 key.offset = other_start;
1161 del_slot = path->slots[0];
1163 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1164 0, root->root_key.objectid,
1165 ino, orig_offset, 0);
1166 BUG_ON(ret); /* -ENOMEM */
1169 fi = btrfs_item_ptr(leaf, path->slots[0],
1170 struct btrfs_file_extent_item);
1171 btrfs_set_file_extent_type(leaf, fi,
1172 BTRFS_FILE_EXTENT_REG);
1173 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1174 btrfs_mark_buffer_dirty(leaf);
1176 fi = btrfs_item_ptr(leaf, del_slot - 1,
1177 struct btrfs_file_extent_item);
1178 btrfs_set_file_extent_type(leaf, fi,
1179 BTRFS_FILE_EXTENT_REG);
1180 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1181 btrfs_set_file_extent_num_bytes(leaf, fi,
1182 extent_end - key.offset);
1183 btrfs_mark_buffer_dirty(leaf);
1185 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1187 btrfs_abort_transaction(trans, root, ret);
1192 btrfs_free_path(path);
1197 * on error we return an unlocked page and the error value
1198 * on success we return a locked page and 0
1200 static int prepare_uptodate_page(struct page *page, u64 pos,
1201 bool force_uptodate)
1205 if (((pos & (PAGE_CACHE_SIZE - 1)) || force_uptodate) &&
1206 !PageUptodate(page)) {
1207 ret = btrfs_readpage(NULL, page);
1211 if (!PageUptodate(page)) {
1220 * this gets pages into the page cache and locks them down, it also properly
1221 * waits for data=ordered extents to finish before allowing the pages to be
1224 static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
1225 struct page **pages, size_t num_pages,
1226 loff_t pos, unsigned long first_index,
1227 size_t write_bytes, bool force_uptodate)
1229 struct extent_state *cached_state = NULL;
1231 unsigned long index = pos >> PAGE_CACHE_SHIFT;
1232 struct inode *inode = file_inode(file);
1233 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1239 start_pos = pos & ~((u64)root->sectorsize - 1);
1240 last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
1243 for (i = 0; i < num_pages; i++) {
1244 pages[i] = find_or_create_page(inode->i_mapping, index + i,
1245 mask | __GFP_WRITE);
1253 err = prepare_uptodate_page(pages[i], pos,
1255 if (i == num_pages - 1)
1256 err = prepare_uptodate_page(pages[i],
1257 pos + write_bytes, false);
1259 page_cache_release(pages[i]);
1263 wait_on_page_writeback(pages[i]);
1266 if (start_pos < inode->i_size) {
1267 struct btrfs_ordered_extent *ordered;
1268 lock_extent_bits(&BTRFS_I(inode)->io_tree,
1269 start_pos, last_pos - 1, 0, &cached_state);
1270 ordered = btrfs_lookup_first_ordered_extent(inode,
1273 ordered->file_offset + ordered->len > start_pos &&
1274 ordered->file_offset < last_pos) {
1275 btrfs_put_ordered_extent(ordered);
1276 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1277 start_pos, last_pos - 1,
1278 &cached_state, GFP_NOFS);
1279 for (i = 0; i < num_pages; i++) {
1280 unlock_page(pages[i]);
1281 page_cache_release(pages[i]);
1283 btrfs_wait_ordered_range(inode, start_pos,
1284 last_pos - start_pos);
1288 btrfs_put_ordered_extent(ordered);
1290 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
1291 last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1292 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
1293 0, 0, &cached_state, GFP_NOFS);
1294 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1295 start_pos, last_pos - 1, &cached_state,
1298 for (i = 0; i < num_pages; i++) {
1299 if (clear_page_dirty_for_io(pages[i]))
1300 account_page_redirty(pages[i]);
1301 set_page_extent_mapped(pages[i]);
1302 WARN_ON(!PageLocked(pages[i]));
1306 while (faili >= 0) {
1307 unlock_page(pages[faili]);
1308 page_cache_release(pages[faili]);
1315 static noinline int check_can_nocow(struct inode *inode, loff_t pos,
1316 size_t *write_bytes)
1318 struct btrfs_trans_handle *trans;
1319 struct btrfs_root *root = BTRFS_I(inode)->root;
1320 struct btrfs_ordered_extent *ordered;
1321 u64 lockstart, lockend;
1325 lockstart = round_down(pos, root->sectorsize);
1326 lockend = lockstart + round_up(*write_bytes, root->sectorsize) - 1;
1329 lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1330 ordered = btrfs_lookup_ordered_range(inode, lockstart,
1331 lockend - lockstart + 1);
1335 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1336 btrfs_start_ordered_extent(inode, ordered, 1);
1337 btrfs_put_ordered_extent(ordered);
1340 trans = btrfs_join_transaction(root);
1341 if (IS_ERR(trans)) {
1342 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1343 return PTR_ERR(trans);
1346 num_bytes = lockend - lockstart + 1;
1347 ret = can_nocow_extent(trans, inode, lockstart, &num_bytes, NULL, NULL,
1349 btrfs_end_transaction(trans, root);
1353 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
1354 EXTENT_DIRTY | EXTENT_DELALLOC |
1355 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1357 *write_bytes = min_t(size_t, *write_bytes, num_bytes);
1360 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1365 static noinline ssize_t __btrfs_buffered_write(struct file *file,
1369 struct inode *inode = file_inode(file);
1370 struct btrfs_root *root = BTRFS_I(inode)->root;
1371 struct page **pages = NULL;
1372 u64 release_bytes = 0;
1373 unsigned long first_index;
1374 size_t num_written = 0;
1377 bool only_release_metadata = false;
1378 bool force_page_uptodate = false;
1380 nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) /
1381 PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
1382 (sizeof(struct page *)));
1383 nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1384 nrptrs = max(nrptrs, 8);
1385 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
1389 first_index = pos >> PAGE_CACHE_SHIFT;
1391 while (iov_iter_count(i) > 0) {
1392 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1393 size_t write_bytes = min(iov_iter_count(i),
1394 nrptrs * (size_t)PAGE_CACHE_SIZE -
1396 size_t num_pages = (write_bytes + offset +
1397 PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1398 size_t reserve_bytes;
1402 WARN_ON(num_pages > nrptrs);
1405 * Fault pages before locking them in prepare_pages
1406 * to avoid recursive lock
1408 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1413 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1414 ret = btrfs_check_data_free_space(inode, reserve_bytes);
1415 if (ret == -ENOSPC &&
1416 (BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW |
1417 BTRFS_INODE_PREALLOC))) {
1418 ret = check_can_nocow(inode, pos, &write_bytes);
1420 only_release_metadata = true;
1422 * our prealloc extent may be smaller than
1423 * write_bytes, so scale down.
1425 num_pages = (write_bytes + offset +
1426 PAGE_CACHE_SIZE - 1) >>
1428 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1438 ret = btrfs_delalloc_reserve_metadata(inode, reserve_bytes);
1440 if (!only_release_metadata)
1441 btrfs_free_reserved_data_space(inode,
1446 release_bytes = reserve_bytes;
1449 * This is going to setup the pages array with the number of
1450 * pages we want, so we don't really need to worry about the
1451 * contents of pages from loop to loop
1453 ret = prepare_pages(root, file, pages, num_pages,
1454 pos, first_index, write_bytes,
1455 force_page_uptodate);
1459 copied = btrfs_copy_from_user(pos, num_pages,
1460 write_bytes, pages, i);
1463 * if we have trouble faulting in the pages, fall
1464 * back to one page at a time
1466 if (copied < write_bytes)
1470 force_page_uptodate = true;
1473 force_page_uptodate = false;
1474 dirty_pages = (copied + offset +
1475 PAGE_CACHE_SIZE - 1) >>
1480 * If we had a short copy we need to release the excess delaloc
1481 * bytes we reserved. We need to increment outstanding_extents
1482 * because btrfs_delalloc_release_space will decrement it, but
1483 * we still have an outstanding extent for the chunk we actually
1486 if (num_pages > dirty_pages) {
1487 release_bytes = (num_pages - dirty_pages) <<
1490 spin_lock(&BTRFS_I(inode)->lock);
1491 BTRFS_I(inode)->outstanding_extents++;
1492 spin_unlock(&BTRFS_I(inode)->lock);
1494 if (only_release_metadata)
1495 btrfs_delalloc_release_metadata(inode,
1498 btrfs_delalloc_release_space(inode,
1502 release_bytes = dirty_pages << PAGE_CACHE_SHIFT;
1504 ret = btrfs_dirty_pages(root, inode, pages,
1505 dirty_pages, pos, copied,
1508 btrfs_drop_pages(pages, num_pages);
1514 btrfs_drop_pages(pages, num_pages);
1516 if (only_release_metadata && copied > 0) {
1517 u64 lockstart = round_down(pos, root->sectorsize);
1518 u64 lockend = lockstart +
1519 (dirty_pages << PAGE_CACHE_SHIFT) - 1;
1521 set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
1522 lockend, EXTENT_NORESERVE, NULL,
1524 only_release_metadata = false;
1529 balance_dirty_pages_ratelimited(inode->i_mapping);
1530 if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1531 btrfs_btree_balance_dirty(root);
1534 num_written += copied;
1539 if (release_bytes) {
1540 if (only_release_metadata)
1541 btrfs_delalloc_release_metadata(inode, release_bytes);
1543 btrfs_delalloc_release_space(inode, release_bytes);
1546 return num_written ? num_written : ret;
1549 static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1550 const struct iovec *iov,
1551 unsigned long nr_segs, loff_t pos,
1552 loff_t *ppos, size_t count, size_t ocount)
1554 struct file *file = iocb->ki_filp;
1557 ssize_t written_buffered;
1561 written = generic_file_direct_write(iocb, iov, &nr_segs, pos, ppos,
1564 if (written < 0 || written == count)
1569 iov_iter_init(&i, iov, nr_segs, count, written);
1570 written_buffered = __btrfs_buffered_write(file, &i, pos);
1571 if (written_buffered < 0) {
1572 err = written_buffered;
1575 endbyte = pos + written_buffered - 1;
1576 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
1579 written += written_buffered;
1580 *ppos = pos + written_buffered;
1581 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
1582 endbyte >> PAGE_CACHE_SHIFT);
1584 return written ? written : err;
1587 static void update_time_for_write(struct inode *inode)
1589 struct timespec now;
1591 if (IS_NOCMTIME(inode))
1594 now = current_fs_time(inode->i_sb);
1595 if (!timespec_equal(&inode->i_mtime, &now))
1596 inode->i_mtime = now;
1598 if (!timespec_equal(&inode->i_ctime, &now))
1599 inode->i_ctime = now;
1601 if (IS_I_VERSION(inode))
1602 inode_inc_iversion(inode);
1605 static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
1606 const struct iovec *iov,
1607 unsigned long nr_segs, loff_t pos)
1609 struct file *file = iocb->ki_filp;
1610 struct inode *inode = file_inode(file);
1611 struct btrfs_root *root = BTRFS_I(inode)->root;
1612 loff_t *ppos = &iocb->ki_pos;
1614 ssize_t num_written = 0;
1616 size_t count, ocount;
1617 bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host);
1619 sb_start_write(inode->i_sb);
1621 mutex_lock(&inode->i_mutex);
1623 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
1625 mutex_unlock(&inode->i_mutex);
1630 current->backing_dev_info = inode->i_mapping->backing_dev_info;
1631 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1633 mutex_unlock(&inode->i_mutex);
1638 mutex_unlock(&inode->i_mutex);
1642 err = file_remove_suid(file);
1644 mutex_unlock(&inode->i_mutex);
1649 * If BTRFS flips readonly due to some impossible error
1650 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1651 * although we have opened a file as writable, we have
1652 * to stop this write operation to ensure FS consistency.
1654 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
1655 mutex_unlock(&inode->i_mutex);
1661 * We reserve space for updating the inode when we reserve space for the
1662 * extent we are going to write, so we will enospc out there. We don't
1663 * need to start yet another transaction to update the inode as we will
1664 * update the inode when we finish writing whatever data we write.
1666 update_time_for_write(inode);
1668 start_pos = round_down(pos, root->sectorsize);
1669 if (start_pos > i_size_read(inode)) {
1670 err = btrfs_cont_expand(inode, i_size_read(inode), start_pos);
1672 mutex_unlock(&inode->i_mutex);
1678 atomic_inc(&BTRFS_I(inode)->sync_writers);
1680 if (unlikely(file->f_flags & O_DIRECT)) {
1681 num_written = __btrfs_direct_write(iocb, iov, nr_segs,
1682 pos, ppos, count, ocount);
1686 iov_iter_init(&i, iov, nr_segs, count, num_written);
1688 num_written = __btrfs_buffered_write(file, &i, pos);
1689 if (num_written > 0)
1690 *ppos = pos + num_written;
1693 mutex_unlock(&inode->i_mutex);
1696 * we want to make sure fsync finds this change
1697 * but we haven't joined a transaction running right now.
1699 * Later on, someone is sure to update the inode and get the
1700 * real transid recorded.
1702 * We set last_trans now to the fs_info generation + 1,
1703 * this will either be one more than the running transaction
1704 * or the generation used for the next transaction if there isn't
1705 * one running right now.
1707 * We also have to set last_sub_trans to the current log transid,
1708 * otherwise subsequent syncs to a file that's been synced in this
1709 * transaction will appear to have already occured.
1711 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1712 BTRFS_I(inode)->last_sub_trans = root->log_transid;
1713 if (num_written > 0 || num_written == -EIOCBQUEUED) {
1714 err = generic_write_sync(file, pos, num_written);
1715 if (err < 0 && num_written > 0)
1720 atomic_dec(&BTRFS_I(inode)->sync_writers);
1722 sb_end_write(inode->i_sb);
1723 current->backing_dev_info = NULL;
1724 return num_written ? num_written : err;
1727 int btrfs_release_file(struct inode *inode, struct file *filp)
1730 * ordered_data_close is set by settattr when we are about to truncate
1731 * a file from a non-zero size to a zero size. This tries to
1732 * flush down new bytes that may have been written if the
1733 * application were using truncate to replace a file in place.
1735 if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
1736 &BTRFS_I(inode)->runtime_flags)) {
1737 struct btrfs_trans_handle *trans;
1738 struct btrfs_root *root = BTRFS_I(inode)->root;
1741 * We need to block on a committing transaction to keep us from
1742 * throwing a ordered operation on to the list and causing
1743 * something like sync to deadlock trying to flush out this
1746 trans = btrfs_start_transaction(root, 0);
1748 return PTR_ERR(trans);
1749 btrfs_add_ordered_operation(trans, BTRFS_I(inode)->root, inode);
1750 btrfs_end_transaction(trans, root);
1751 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1752 filemap_flush(inode->i_mapping);
1754 if (filp->private_data)
1755 btrfs_ioctl_trans_end(filp);
1760 * fsync call for both files and directories. This logs the inode into
1761 * the tree log instead of forcing full commits whenever possible.
1763 * It needs to call filemap_fdatawait so that all ordered extent updates are
1764 * in the metadata btree are up to date for copying to the log.
1766 * It drops the inode mutex before doing the tree log commit. This is an
1767 * important optimization for directories because holding the mutex prevents
1768 * new operations on the dir while we write to disk.
1770 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1772 struct dentry *dentry = file->f_path.dentry;
1773 struct inode *inode = dentry->d_inode;
1774 struct btrfs_root *root = BTRFS_I(inode)->root;
1776 struct btrfs_trans_handle *trans;
1779 trace_btrfs_sync_file(file, datasync);
1782 * We write the dirty pages in the range and wait until they complete
1783 * out of the ->i_mutex. If so, we can flush the dirty pages by
1784 * multi-task, and make the performance up. See
1785 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1787 atomic_inc(&BTRFS_I(inode)->sync_writers);
1788 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
1789 if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1790 &BTRFS_I(inode)->runtime_flags))
1791 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
1792 atomic_dec(&BTRFS_I(inode)->sync_writers);
1796 mutex_lock(&inode->i_mutex);
1799 * We flush the dirty pages again to avoid some dirty pages in the
1802 atomic_inc(&root->log_batch);
1803 full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1804 &BTRFS_I(inode)->runtime_flags);
1806 btrfs_wait_ordered_range(inode, start, end - start + 1);
1807 atomic_inc(&root->log_batch);
1810 * check the transaction that last modified this inode
1811 * and see if its already been committed
1813 if (!BTRFS_I(inode)->last_trans) {
1814 mutex_unlock(&inode->i_mutex);
1819 * if the last transaction that changed this file was before
1820 * the current transaction, we can bail out now without any
1824 if (btrfs_inode_in_log(inode, root->fs_info->generation) ||
1825 BTRFS_I(inode)->last_trans <=
1826 root->fs_info->last_trans_committed) {
1827 BTRFS_I(inode)->last_trans = 0;
1830 * We'v had everything committed since the last time we were
1831 * modified so clear this flag in case it was set for whatever
1832 * reason, it's no longer relevant.
1834 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1835 &BTRFS_I(inode)->runtime_flags);
1836 mutex_unlock(&inode->i_mutex);
1841 * ok we haven't committed the transaction yet, lets do a commit
1843 if (file->private_data)
1844 btrfs_ioctl_trans_end(file);
1846 trans = btrfs_start_transaction(root, 0);
1847 if (IS_ERR(trans)) {
1848 ret = PTR_ERR(trans);
1849 mutex_unlock(&inode->i_mutex);
1853 ret = btrfs_log_dentry_safe(trans, root, dentry);
1855 mutex_unlock(&inode->i_mutex);
1859 /* we've logged all the items and now have a consistent
1860 * version of the file in the log. It is possible that
1861 * someone will come in and modify the file, but that's
1862 * fine because the log is consistent on disk, and we
1863 * have references to all of the file's extents
1865 * It is possible that someone will come in and log the
1866 * file again, but that will end up using the synchronization
1867 * inside btrfs_sync_log to keep things safe.
1869 mutex_unlock(&inode->i_mutex);
1871 if (ret != BTRFS_NO_LOG_SYNC) {
1874 * If we didn't already wait for ordered extents we need
1878 btrfs_wait_ordered_range(inode, start,
1880 ret = btrfs_commit_transaction(trans, root);
1882 ret = btrfs_sync_log(trans, root);
1884 ret = btrfs_end_transaction(trans, root);
1887 btrfs_wait_ordered_range(inode, start,
1890 ret = btrfs_commit_transaction(trans, root);
1894 ret = btrfs_end_transaction(trans, root);
1897 return ret > 0 ? -EIO : ret;
1900 static const struct vm_operations_struct btrfs_file_vm_ops = {
1901 .fault = filemap_fault,
1902 .page_mkwrite = btrfs_page_mkwrite,
1903 .remap_pages = generic_file_remap_pages,
1906 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
1908 struct address_space *mapping = filp->f_mapping;
1910 if (!mapping->a_ops->readpage)
1913 file_accessed(filp);
1914 vma->vm_ops = &btrfs_file_vm_ops;
1919 static int hole_mergeable(struct inode *inode, struct extent_buffer *leaf,
1920 int slot, u64 start, u64 end)
1922 struct btrfs_file_extent_item *fi;
1923 struct btrfs_key key;
1925 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
1928 btrfs_item_key_to_cpu(leaf, &key, slot);
1929 if (key.objectid != btrfs_ino(inode) ||
1930 key.type != BTRFS_EXTENT_DATA_KEY)
1933 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1935 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
1938 if (btrfs_file_extent_disk_bytenr(leaf, fi))
1941 if (key.offset == end)
1943 if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
1948 static int fill_holes(struct btrfs_trans_handle *trans, struct inode *inode,
1949 struct btrfs_path *path, u64 offset, u64 end)
1951 struct btrfs_root *root = BTRFS_I(inode)->root;
1952 struct extent_buffer *leaf;
1953 struct btrfs_file_extent_item *fi;
1954 struct extent_map *hole_em;
1955 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1956 struct btrfs_key key;
1959 key.objectid = btrfs_ino(inode);
1960 key.type = BTRFS_EXTENT_DATA_KEY;
1961 key.offset = offset;
1964 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1969 leaf = path->nodes[0];
1970 if (hole_mergeable(inode, leaf, path->slots[0]-1, offset, end)) {
1974 fi = btrfs_item_ptr(leaf, path->slots[0],
1975 struct btrfs_file_extent_item);
1976 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
1978 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1979 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
1980 btrfs_set_file_extent_offset(leaf, fi, 0);
1981 btrfs_mark_buffer_dirty(leaf);
1985 if (hole_mergeable(inode, leaf, path->slots[0]+1, offset, end)) {
1989 key.offset = offset;
1990 btrfs_set_item_key_safe(root, path, &key);
1991 fi = btrfs_item_ptr(leaf, path->slots[0],
1992 struct btrfs_file_extent_item);
1993 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
1995 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1996 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
1997 btrfs_set_file_extent_offset(leaf, fi, 0);
1998 btrfs_mark_buffer_dirty(leaf);
2001 btrfs_release_path(path);
2003 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
2004 0, 0, end - offset, 0, end - offset,
2010 btrfs_release_path(path);
2012 hole_em = alloc_extent_map();
2014 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2015 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2016 &BTRFS_I(inode)->runtime_flags);
2018 hole_em->start = offset;
2019 hole_em->len = end - offset;
2020 hole_em->ram_bytes = hole_em->len;
2021 hole_em->orig_start = offset;
2023 hole_em->block_start = EXTENT_MAP_HOLE;
2024 hole_em->block_len = 0;
2025 hole_em->orig_block_len = 0;
2026 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
2027 hole_em->compress_type = BTRFS_COMPRESS_NONE;
2028 hole_em->generation = trans->transid;
2031 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2032 write_lock(&em_tree->lock);
2033 ret = add_extent_mapping(em_tree, hole_em, 1);
2034 write_unlock(&em_tree->lock);
2035 } while (ret == -EEXIST);
2036 free_extent_map(hole_em);
2038 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2039 &BTRFS_I(inode)->runtime_flags);
2045 static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
2047 struct btrfs_root *root = BTRFS_I(inode)->root;
2048 struct extent_state *cached_state = NULL;
2049 struct btrfs_path *path;
2050 struct btrfs_block_rsv *rsv;
2051 struct btrfs_trans_handle *trans;
2052 u64 lockstart = round_up(offset, BTRFS_I(inode)->root->sectorsize);
2053 u64 lockend = round_down(offset + len,
2054 BTRFS_I(inode)->root->sectorsize) - 1;
2055 u64 cur_offset = lockstart;
2056 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
2060 bool same_page = ((offset >> PAGE_CACHE_SHIFT) ==
2061 ((offset + len - 1) >> PAGE_CACHE_SHIFT));
2063 btrfs_wait_ordered_range(inode, offset, len);
2065 mutex_lock(&inode->i_mutex);
2067 * We needn't truncate any page which is beyond the end of the file
2068 * because we are sure there is no data there.
2071 * Only do this if we are in the same page and we aren't doing the
2074 if (same_page && len < PAGE_CACHE_SIZE) {
2075 if (offset < round_up(inode->i_size, PAGE_CACHE_SIZE))
2076 ret = btrfs_truncate_page(inode, offset, len, 0);
2077 mutex_unlock(&inode->i_mutex);
2081 /* zero back part of the first page */
2082 if (offset < round_up(inode->i_size, PAGE_CACHE_SIZE)) {
2083 ret = btrfs_truncate_page(inode, offset, 0, 0);
2085 mutex_unlock(&inode->i_mutex);
2090 /* zero the front end of the last page */
2091 if (offset + len < round_up(inode->i_size, PAGE_CACHE_SIZE)) {
2092 ret = btrfs_truncate_page(inode, offset + len, 0, 1);
2094 mutex_unlock(&inode->i_mutex);
2099 if (lockend < lockstart) {
2100 mutex_unlock(&inode->i_mutex);
2105 struct btrfs_ordered_extent *ordered;
2107 truncate_pagecache_range(inode, lockstart, lockend);
2109 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2111 ordered = btrfs_lookup_first_ordered_extent(inode, lockend);
2114 * We need to make sure we have no ordered extents in this range
2115 * and nobody raced in and read a page in this range, if we did
2116 * we need to try again.
2119 (ordered->file_offset + ordered->len < lockstart ||
2120 ordered->file_offset > lockend)) &&
2121 !test_range_bit(&BTRFS_I(inode)->io_tree, lockstart,
2122 lockend, EXTENT_UPTODATE, 0,
2125 btrfs_put_ordered_extent(ordered);
2129 btrfs_put_ordered_extent(ordered);
2130 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
2131 lockend, &cached_state, GFP_NOFS);
2132 btrfs_wait_ordered_range(inode, lockstart,
2133 lockend - lockstart + 1);
2136 path = btrfs_alloc_path();
2142 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2147 rsv->size = btrfs_calc_trunc_metadata_size(root, 1);
2151 * 1 - update the inode
2152 * 1 - removing the extents in the range
2153 * 1 - adding the hole extent
2155 trans = btrfs_start_transaction(root, 3);
2156 if (IS_ERR(trans)) {
2157 err = PTR_ERR(trans);
2161 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
2164 trans->block_rsv = rsv;
2166 while (cur_offset < lockend) {
2167 ret = __btrfs_drop_extents(trans, root, inode, path,
2168 cur_offset, lockend + 1,
2173 trans->block_rsv = &root->fs_info->trans_block_rsv;
2175 ret = fill_holes(trans, inode, path, cur_offset, drop_end);
2181 cur_offset = drop_end;
2183 ret = btrfs_update_inode(trans, root, inode);
2189 btrfs_end_transaction(trans, root);
2190 btrfs_btree_balance_dirty(root);
2192 trans = btrfs_start_transaction(root, 3);
2193 if (IS_ERR(trans)) {
2194 ret = PTR_ERR(trans);
2199 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
2201 BUG_ON(ret); /* shouldn't happen */
2202 trans->block_rsv = rsv;
2210 trans->block_rsv = &root->fs_info->trans_block_rsv;
2211 ret = fill_holes(trans, inode, path, cur_offset, drop_end);
2221 inode_inc_iversion(inode);
2222 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2224 trans->block_rsv = &root->fs_info->trans_block_rsv;
2225 ret = btrfs_update_inode(trans, root, inode);
2226 btrfs_end_transaction(trans, root);
2227 btrfs_btree_balance_dirty(root);
2229 btrfs_free_path(path);
2230 btrfs_free_block_rsv(root, rsv);
2232 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2233 &cached_state, GFP_NOFS);
2234 mutex_unlock(&inode->i_mutex);
2240 static long btrfs_fallocate(struct file *file, int mode,
2241 loff_t offset, loff_t len)
2243 struct inode *inode = file_inode(file);
2244 struct extent_state *cached_state = NULL;
2245 struct btrfs_root *root = BTRFS_I(inode)->root;
2252 struct extent_map *em;
2253 int blocksize = BTRFS_I(inode)->root->sectorsize;
2256 alloc_start = round_down(offset, blocksize);
2257 alloc_end = round_up(offset + len, blocksize);
2259 /* Make sure we aren't being give some crap mode */
2260 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2263 if (mode & FALLOC_FL_PUNCH_HOLE)
2264 return btrfs_punch_hole(inode, offset, len);
2267 * Make sure we have enough space before we do the
2270 ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
2273 if (root->fs_info->quota_enabled) {
2274 ret = btrfs_qgroup_reserve(root, alloc_end - alloc_start);
2276 goto out_reserve_fail;
2279 mutex_lock(&inode->i_mutex);
2280 ret = inode_newsize_ok(inode, alloc_end);
2284 if (alloc_start > inode->i_size) {
2285 ret = btrfs_cont_expand(inode, i_size_read(inode),
2291 * If we are fallocating from the end of the file onward we
2292 * need to zero out the end of the page if i_size lands in the
2295 ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
2301 * wait for ordered IO before we have any locks. We'll loop again
2302 * below with the locks held.
2304 btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
2306 locked_end = alloc_end - 1;
2308 struct btrfs_ordered_extent *ordered;
2310 /* the extent lock is ordered inside the running
2313 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
2314 locked_end, 0, &cached_state);
2315 ordered = btrfs_lookup_first_ordered_extent(inode,
2318 ordered->file_offset + ordered->len > alloc_start &&
2319 ordered->file_offset < alloc_end) {
2320 btrfs_put_ordered_extent(ordered);
2321 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
2322 alloc_start, locked_end,
2323 &cached_state, GFP_NOFS);
2325 * we can't wait on the range with the transaction
2326 * running or with the extent lock held
2328 btrfs_wait_ordered_range(inode, alloc_start,
2329 alloc_end - alloc_start);
2332 btrfs_put_ordered_extent(ordered);
2337 cur_offset = alloc_start;
2341 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2342 alloc_end - cur_offset, 0);
2343 if (IS_ERR_OR_NULL(em)) {
2350 last_byte = min(extent_map_end(em), alloc_end);
2351 actual_end = min_t(u64, extent_map_end(em), offset + len);
2352 last_byte = ALIGN(last_byte, blocksize);
2354 if (em->block_start == EXTENT_MAP_HOLE ||
2355 (cur_offset >= inode->i_size &&
2356 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
2357 ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
2358 last_byte - cur_offset,
2359 1 << inode->i_blkbits,
2364 free_extent_map(em);
2367 } else if (actual_end > inode->i_size &&
2368 !(mode & FALLOC_FL_KEEP_SIZE)) {
2370 * We didn't need to allocate any more space, but we
2371 * still extended the size of the file so we need to
2374 inode->i_ctime = CURRENT_TIME;
2375 i_size_write(inode, actual_end);
2376 btrfs_ordered_update_i_size(inode, actual_end, NULL);
2378 free_extent_map(em);
2380 cur_offset = last_byte;
2381 if (cur_offset >= alloc_end) {
2386 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
2387 &cached_state, GFP_NOFS);
2389 mutex_unlock(&inode->i_mutex);
2390 if (root->fs_info->quota_enabled)
2391 btrfs_qgroup_free(root, alloc_end - alloc_start);
2393 /* Let go of our reservation. */
2394 btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
2398 static int find_desired_extent(struct inode *inode, loff_t *offset, int whence)
2400 struct btrfs_root *root = BTRFS_I(inode)->root;
2401 struct extent_map *em;
2402 struct extent_state *cached_state = NULL;
2403 u64 lockstart = *offset;
2404 u64 lockend = i_size_read(inode);
2405 u64 start = *offset;
2406 u64 orig_start = *offset;
2407 u64 len = i_size_read(inode);
2411 lockend = max_t(u64, root->sectorsize, lockend);
2412 if (lockend <= lockstart)
2413 lockend = lockstart + root->sectorsize;
2416 len = lockend - lockstart + 1;
2418 len = max_t(u64, len, root->sectorsize);
2419 if (inode->i_size == 0)
2422 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, 0,
2426 * Delalloc is such a pain. If we have a hole and we have pending
2427 * delalloc for a portion of the hole we will get back a hole that
2428 * exists for the entire range since it hasn't been actually written
2429 * yet. So to take care of this case we need to look for an extent just
2430 * before the position we want in case there is outstanding delalloc
2433 if (whence == SEEK_HOLE && start != 0) {
2434 if (start <= root->sectorsize)
2435 em = btrfs_get_extent_fiemap(inode, NULL, 0, 0,
2436 root->sectorsize, 0);
2438 em = btrfs_get_extent_fiemap(inode, NULL, 0,
2439 start - root->sectorsize,
2440 root->sectorsize, 0);
2445 last_end = em->start + em->len;
2446 if (em->block_start == EXTENT_MAP_DELALLOC)
2447 last_end = min_t(u64, last_end, inode->i_size);
2448 free_extent_map(em);
2452 em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0);
2458 if (em->block_start == EXTENT_MAP_HOLE) {
2459 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2460 if (last_end <= orig_start) {
2461 free_extent_map(em);
2467 if (whence == SEEK_HOLE) {
2469 free_extent_map(em);
2473 if (whence == SEEK_DATA) {
2474 if (em->block_start == EXTENT_MAP_DELALLOC) {
2475 if (start >= inode->i_size) {
2476 free_extent_map(em);
2482 if (!test_bit(EXTENT_FLAG_PREALLOC,
2485 free_extent_map(em);
2491 start = em->start + em->len;
2492 last_end = em->start + em->len;
2494 if (em->block_start == EXTENT_MAP_DELALLOC)
2495 last_end = min_t(u64, last_end, inode->i_size);
2497 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2498 free_extent_map(em);
2502 free_extent_map(em);
2506 *offset = min(*offset, inode->i_size);
2508 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2509 &cached_state, GFP_NOFS);
2513 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
2515 struct inode *inode = file->f_mapping->host;
2518 mutex_lock(&inode->i_mutex);
2522 offset = generic_file_llseek(file, offset, whence);
2526 if (offset >= i_size_read(inode)) {
2527 mutex_unlock(&inode->i_mutex);
2531 ret = find_desired_extent(inode, &offset, whence);
2533 mutex_unlock(&inode->i_mutex);
2538 if (offset < 0 && !(file->f_mode & FMODE_UNSIGNED_OFFSET)) {
2542 if (offset > inode->i_sb->s_maxbytes) {
2547 /* Special lock needed here? */
2548 if (offset != file->f_pos) {
2549 file->f_pos = offset;
2550 file->f_version = 0;
2553 mutex_unlock(&inode->i_mutex);
2557 const struct file_operations btrfs_file_operations = {
2558 .llseek = btrfs_file_llseek,
2559 .read = do_sync_read,
2560 .write = do_sync_write,
2561 .aio_read = generic_file_aio_read,
2562 .splice_read = generic_file_splice_read,
2563 .aio_write = btrfs_file_aio_write,
2564 .mmap = btrfs_file_mmap,
2565 .open = generic_file_open,
2566 .release = btrfs_release_file,
2567 .fsync = btrfs_sync_file,
2568 .fallocate = btrfs_fallocate,
2569 .unlocked_ioctl = btrfs_ioctl,
2570 #ifdef CONFIG_COMPAT
2571 .compat_ioctl = btrfs_ioctl,
2575 void btrfs_auto_defrag_exit(void)
2577 if (btrfs_inode_defrag_cachep)
2578 kmem_cache_destroy(btrfs_inode_defrag_cachep);
2581 int btrfs_auto_defrag_init(void)
2583 btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
2584 sizeof(struct inode_defrag), 0,
2585 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2587 if (!btrfs_inode_defrag_cachep)