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"
45 static struct kmem_cache *btrfs_inode_defrag_cachep;
47 * when auto defrag is enabled we
48 * queue up these defrag structs to remember which
49 * inodes need defragging passes
52 struct rb_node rb_node;
56 * transid where the defrag was added, we search for
57 * extents newer than this
64 /* last offset we were able to defrag */
67 /* if we've wrapped around back to zero once already */
71 static int __compare_inode_defrag(struct inode_defrag *defrag1,
72 struct inode_defrag *defrag2)
74 if (defrag1->root > defrag2->root)
76 else if (defrag1->root < defrag2->root)
78 else if (defrag1->ino > defrag2->ino)
80 else if (defrag1->ino < defrag2->ino)
86 /* pop a record for an inode into the defrag tree. The lock
87 * must be held already
89 * If you're inserting a record for an older transid than an
90 * existing record, the transid already in the tree is lowered
92 * If an existing record is found the defrag item you
95 static int __btrfs_add_inode_defrag(struct inode *inode,
96 struct inode_defrag *defrag)
98 struct btrfs_root *root = BTRFS_I(inode)->root;
99 struct inode_defrag *entry;
101 struct rb_node *parent = NULL;
104 p = &root->fs_info->defrag_inodes.rb_node;
107 entry = rb_entry(parent, struct inode_defrag, rb_node);
109 ret = __compare_inode_defrag(defrag, entry);
111 p = &parent->rb_left;
113 p = &parent->rb_right;
115 /* if we're reinserting an entry for
116 * an old defrag run, make sure to
117 * lower the transid of our existing record
119 if (defrag->transid < entry->transid)
120 entry->transid = defrag->transid;
121 if (defrag->last_offset > entry->last_offset)
122 entry->last_offset = defrag->last_offset;
126 set_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
127 rb_link_node(&defrag->rb_node, parent, p);
128 rb_insert_color(&defrag->rb_node, &root->fs_info->defrag_inodes);
132 static inline int __need_auto_defrag(struct btrfs_root *root)
134 if (!btrfs_test_opt(root, AUTO_DEFRAG))
137 if (btrfs_fs_closing(root->fs_info))
144 * insert a defrag record for this inode if auto defrag is
147 int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans,
150 struct btrfs_root *root = BTRFS_I(inode)->root;
151 struct inode_defrag *defrag;
155 if (!__need_auto_defrag(root))
158 if (test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags))
162 transid = trans->transid;
164 transid = BTRFS_I(inode)->root->last_trans;
166 defrag = kmem_cache_zalloc(btrfs_inode_defrag_cachep, GFP_NOFS);
170 defrag->ino = btrfs_ino(inode);
171 defrag->transid = transid;
172 defrag->root = root->root_key.objectid;
174 spin_lock(&root->fs_info->defrag_inodes_lock);
175 if (!test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags)) {
177 * If we set IN_DEFRAG flag and evict the inode from memory,
178 * and then re-read this inode, this new inode doesn't have
179 * IN_DEFRAG flag. At the case, we may find the existed defrag.
181 ret = __btrfs_add_inode_defrag(inode, defrag);
183 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
185 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
187 spin_unlock(&root->fs_info->defrag_inodes_lock);
192 * Requeue the defrag object. If there is a defrag object that points to
193 * the same inode in the tree, we will merge them together (by
194 * __btrfs_add_inode_defrag()) and free the one that we want to requeue.
196 static void btrfs_requeue_inode_defrag(struct inode *inode,
197 struct inode_defrag *defrag)
199 struct btrfs_root *root = BTRFS_I(inode)->root;
202 if (!__need_auto_defrag(root))
206 * Here we don't check the IN_DEFRAG flag, because we need merge
209 spin_lock(&root->fs_info->defrag_inodes_lock);
210 ret = __btrfs_add_inode_defrag(inode, defrag);
211 spin_unlock(&root->fs_info->defrag_inodes_lock);
216 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
220 * pick the defragable inode that we want, if it doesn't exist, we will get
223 static struct inode_defrag *
224 btrfs_pick_defrag_inode(struct btrfs_fs_info *fs_info, u64 root, u64 ino)
226 struct inode_defrag *entry = NULL;
227 struct inode_defrag tmp;
229 struct rb_node *parent = NULL;
235 spin_lock(&fs_info->defrag_inodes_lock);
236 p = fs_info->defrag_inodes.rb_node;
239 entry = rb_entry(parent, struct inode_defrag, rb_node);
241 ret = __compare_inode_defrag(&tmp, entry);
245 p = parent->rb_right;
250 if (parent && __compare_inode_defrag(&tmp, entry) > 0) {
251 parent = rb_next(parent);
253 entry = rb_entry(parent, struct inode_defrag, rb_node);
259 rb_erase(parent, &fs_info->defrag_inodes);
260 spin_unlock(&fs_info->defrag_inodes_lock);
264 void btrfs_cleanup_defrag_inodes(struct btrfs_fs_info *fs_info)
266 struct inode_defrag *defrag;
267 struct rb_node *node;
269 spin_lock(&fs_info->defrag_inodes_lock);
270 node = rb_first(&fs_info->defrag_inodes);
272 rb_erase(node, &fs_info->defrag_inodes);
273 defrag = rb_entry(node, struct inode_defrag, rb_node);
274 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
276 if (need_resched()) {
277 spin_unlock(&fs_info->defrag_inodes_lock);
279 spin_lock(&fs_info->defrag_inodes_lock);
282 node = rb_first(&fs_info->defrag_inodes);
284 spin_unlock(&fs_info->defrag_inodes_lock);
287 #define BTRFS_DEFRAG_BATCH 1024
289 static int __btrfs_run_defrag_inode(struct btrfs_fs_info *fs_info,
290 struct inode_defrag *defrag)
292 struct btrfs_root *inode_root;
294 struct btrfs_key key;
295 struct btrfs_ioctl_defrag_range_args range;
301 key.objectid = defrag->root;
302 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
303 key.offset = (u64)-1;
305 index = srcu_read_lock(&fs_info->subvol_srcu);
307 inode_root = btrfs_read_fs_root_no_name(fs_info, &key);
308 if (IS_ERR(inode_root)) {
309 ret = PTR_ERR(inode_root);
313 key.objectid = defrag->ino;
314 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
316 inode = btrfs_iget(fs_info->sb, &key, inode_root, NULL);
318 ret = PTR_ERR(inode);
321 srcu_read_unlock(&fs_info->subvol_srcu, index);
323 /* do a chunk of defrag */
324 clear_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
325 memset(&range, 0, sizeof(range));
327 range.start = defrag->last_offset;
329 sb_start_write(fs_info->sb);
330 num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid,
332 sb_end_write(fs_info->sb);
334 * if we filled the whole defrag batch, there
335 * must be more work to do. Queue this defrag
338 if (num_defrag == BTRFS_DEFRAG_BATCH) {
339 defrag->last_offset = range.start;
340 btrfs_requeue_inode_defrag(inode, defrag);
341 } else if (defrag->last_offset && !defrag->cycled) {
343 * we didn't fill our defrag batch, but
344 * we didn't start at zero. Make sure we loop
345 * around to the start of the file.
347 defrag->last_offset = 0;
349 btrfs_requeue_inode_defrag(inode, defrag);
351 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
357 srcu_read_unlock(&fs_info->subvol_srcu, index);
358 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
363 * run through the list of inodes in the FS that need
366 int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info)
368 struct inode_defrag *defrag;
370 u64 root_objectid = 0;
372 atomic_inc(&fs_info->defrag_running);
374 /* Pause the auto defragger. */
375 if (test_bit(BTRFS_FS_STATE_REMOUNTING,
379 if (!__need_auto_defrag(fs_info->tree_root))
382 /* find an inode to defrag */
383 defrag = btrfs_pick_defrag_inode(fs_info, root_objectid,
386 if (root_objectid || first_ino) {
395 first_ino = defrag->ino + 1;
396 root_objectid = defrag->root;
398 __btrfs_run_defrag_inode(fs_info, defrag);
400 atomic_dec(&fs_info->defrag_running);
403 * during unmount, we use the transaction_wait queue to
404 * wait for the defragger to stop
406 wake_up(&fs_info->transaction_wait);
410 /* simple helper to fault in pages and copy. This should go away
411 * and be replaced with calls into generic code.
413 static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
415 struct page **prepared_pages,
419 size_t total_copied = 0;
421 int offset = pos & (PAGE_CACHE_SIZE - 1);
423 while (write_bytes > 0) {
424 size_t count = min_t(size_t,
425 PAGE_CACHE_SIZE - offset, write_bytes);
426 struct page *page = prepared_pages[pg];
428 * Copy data from userspace to the current page
430 * Disable pagefault to avoid recursive lock since
431 * the pages are already locked
434 copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
437 /* Flush processor's dcache for this page */
438 flush_dcache_page(page);
441 * if we get a partial write, we can end up with
442 * partially up to date pages. These add
443 * a lot of complexity, so make sure they don't
444 * happen by forcing this copy to be retried.
446 * The rest of the btrfs_file_write code will fall
447 * back to page at a time copies after we return 0.
449 if (!PageUptodate(page) && copied < count)
452 iov_iter_advance(i, copied);
453 write_bytes -= copied;
454 total_copied += copied;
456 /* Return to btrfs_file_write_iter to fault page */
457 if (unlikely(copied == 0))
460 if (unlikely(copied < PAGE_CACHE_SIZE - offset)) {
471 * unlocks pages after btrfs_file_write is done with them
473 static void btrfs_drop_pages(struct page **pages, size_t num_pages)
476 for (i = 0; i < num_pages; i++) {
477 /* page checked is some magic around finding pages that
478 * have been modified without going through btrfs_set_page_dirty
481 ClearPageChecked(pages[i]);
482 unlock_page(pages[i]);
483 mark_page_accessed(pages[i]);
484 page_cache_release(pages[i]);
489 * after copy_from_user, pages need to be dirtied and we need to make
490 * sure holes are created between the current EOF and the start of
491 * any next extents (if required).
493 * this also makes the decision about creating an inline extent vs
494 * doing real data extents, marking pages dirty and delalloc as required.
496 int btrfs_dirty_pages(struct btrfs_root *root, struct inode *inode,
497 struct page **pages, size_t num_pages,
498 loff_t pos, size_t write_bytes,
499 struct extent_state **cached)
505 u64 end_of_last_block;
506 u64 end_pos = pos + write_bytes;
507 loff_t isize = i_size_read(inode);
509 start_pos = pos & ~((u64)root->sectorsize - 1);
510 num_bytes = ALIGN(write_bytes + pos - start_pos, root->sectorsize);
512 end_of_last_block = start_pos + num_bytes - 1;
513 err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
518 for (i = 0; i < num_pages; i++) {
519 struct page *p = pages[i];
526 * we've only changed i_size in ram, and we haven't updated
527 * the disk i_size. There is no need to log the inode
531 i_size_write(inode, end_pos);
536 * this drops all the extents in the cache that intersect the range
537 * [start, end]. Existing extents are split as required.
539 void btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
542 struct extent_map *em;
543 struct extent_map *split = NULL;
544 struct extent_map *split2 = NULL;
545 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
546 u64 len = end - start + 1;
554 WARN_ON(end < start);
555 if (end == (u64)-1) {
564 split = alloc_extent_map();
566 split2 = alloc_extent_map();
567 if (!split || !split2)
570 write_lock(&em_tree->lock);
571 em = lookup_extent_mapping(em_tree, start, len);
573 write_unlock(&em_tree->lock);
577 gen = em->generation;
578 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
579 if (testend && em->start + em->len >= start + len) {
581 write_unlock(&em_tree->lock);
584 start = em->start + em->len;
586 len = start + len - (em->start + em->len);
588 write_unlock(&em_tree->lock);
591 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
592 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
593 clear_bit(EXTENT_FLAG_LOGGING, &flags);
594 modified = !list_empty(&em->list);
595 remove_extent_mapping(em_tree, em);
599 if (em->start < start) {
600 split->start = em->start;
601 split->len = start - em->start;
603 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
604 split->orig_start = em->orig_start;
605 split->block_start = em->block_start;
608 split->block_len = em->block_len;
610 split->block_len = split->len;
611 split->orig_block_len = max(split->block_len,
613 split->ram_bytes = em->ram_bytes;
615 split->orig_start = split->start;
616 split->block_len = 0;
617 split->block_start = em->block_start;
618 split->orig_block_len = 0;
619 split->ram_bytes = split->len;
622 split->generation = gen;
623 split->bdev = em->bdev;
624 split->flags = flags;
625 split->compress_type = em->compress_type;
626 ret = add_extent_mapping(em_tree, split, modified);
627 BUG_ON(ret); /* Logic error */
628 free_extent_map(split);
632 if (testend && em->start + em->len > start + len) {
633 u64 diff = start + len - em->start;
635 split->start = start + len;
636 split->len = em->start + em->len - (start + len);
637 split->bdev = em->bdev;
638 split->flags = flags;
639 split->compress_type = em->compress_type;
640 split->generation = gen;
642 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
643 split->orig_block_len = max(em->block_len,
646 split->ram_bytes = em->ram_bytes;
648 split->block_len = em->block_len;
649 split->block_start = em->block_start;
650 split->orig_start = em->orig_start;
652 split->block_len = split->len;
653 split->block_start = em->block_start
655 split->orig_start = em->orig_start;
658 split->ram_bytes = split->len;
659 split->orig_start = split->start;
660 split->block_len = 0;
661 split->block_start = em->block_start;
662 split->orig_block_len = 0;
665 ret = add_extent_mapping(em_tree, split, modified);
666 BUG_ON(ret); /* Logic error */
667 free_extent_map(split);
671 write_unlock(&em_tree->lock);
675 /* once for the tree*/
679 free_extent_map(split);
681 free_extent_map(split2);
685 * this is very complex, but the basic idea is to drop all extents
686 * in the range start - end. hint_block is filled in with a block number
687 * that would be a good hint to the block allocator for this file.
689 * If an extent intersects the range but is not entirely inside the range
690 * it is either truncated or split. Anything entirely inside the range
691 * is deleted from the tree.
693 int __btrfs_drop_extents(struct btrfs_trans_handle *trans,
694 struct btrfs_root *root, struct inode *inode,
695 struct btrfs_path *path, u64 start, u64 end,
696 u64 *drop_end, int drop_cache)
698 struct extent_buffer *leaf;
699 struct btrfs_file_extent_item *fi;
700 struct btrfs_key key;
701 struct btrfs_key new_key;
702 u64 ino = btrfs_ino(inode);
703 u64 search_start = start;
706 u64 extent_offset = 0;
713 int modify_tree = -1;
714 int update_refs = (root->ref_cows || root == root->fs_info->tree_root);
718 btrfs_drop_extent_cache(inode, start, end - 1, 0);
720 if (start >= BTRFS_I(inode)->disk_i_size)
725 ret = btrfs_lookup_file_extent(trans, root, path, ino,
726 search_start, modify_tree);
729 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
730 leaf = path->nodes[0];
731 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
732 if (key.objectid == ino &&
733 key.type == BTRFS_EXTENT_DATA_KEY)
738 leaf = path->nodes[0];
739 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
741 ret = btrfs_next_leaf(root, path);
748 leaf = path->nodes[0];
752 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
753 if (key.objectid > ino ||
754 key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
757 fi = btrfs_item_ptr(leaf, path->slots[0],
758 struct btrfs_file_extent_item);
759 extent_type = btrfs_file_extent_type(leaf, fi);
761 if (extent_type == BTRFS_FILE_EXTENT_REG ||
762 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
763 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
764 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
765 extent_offset = btrfs_file_extent_offset(leaf, fi);
766 extent_end = key.offset +
767 btrfs_file_extent_num_bytes(leaf, fi);
768 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
769 extent_end = key.offset +
770 btrfs_file_extent_inline_len(leaf, fi);
773 extent_end = search_start;
776 if (extent_end <= search_start) {
782 search_start = max(key.offset, start);
783 if (recow || !modify_tree) {
785 btrfs_release_path(path);
790 * | - range to drop - |
791 * | -------- extent -------- |
793 if (start > key.offset && end < extent_end) {
795 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
797 memcpy(&new_key, &key, sizeof(new_key));
798 new_key.offset = start;
799 ret = btrfs_duplicate_item(trans, root, path,
801 if (ret == -EAGAIN) {
802 btrfs_release_path(path);
808 leaf = path->nodes[0];
809 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
810 struct btrfs_file_extent_item);
811 btrfs_set_file_extent_num_bytes(leaf, fi,
814 fi = btrfs_item_ptr(leaf, path->slots[0],
815 struct btrfs_file_extent_item);
817 extent_offset += start - key.offset;
818 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
819 btrfs_set_file_extent_num_bytes(leaf, fi,
821 btrfs_mark_buffer_dirty(leaf);
823 if (update_refs && disk_bytenr > 0) {
824 ret = btrfs_inc_extent_ref(trans, root,
825 disk_bytenr, num_bytes, 0,
826 root->root_key.objectid,
828 start - extent_offset, 0);
829 BUG_ON(ret); /* -ENOMEM */
834 * | ---- range to drop ----- |
835 * | -------- extent -------- |
837 if (start <= key.offset && end < extent_end) {
838 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
840 memcpy(&new_key, &key, sizeof(new_key));
841 new_key.offset = end;
842 btrfs_set_item_key_safe(root, path, &new_key);
844 extent_offset += end - key.offset;
845 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
846 btrfs_set_file_extent_num_bytes(leaf, fi,
848 btrfs_mark_buffer_dirty(leaf);
849 if (update_refs && disk_bytenr > 0)
850 inode_sub_bytes(inode, end - key.offset);
854 search_start = extent_end;
856 * | ---- range to drop ----- |
857 * | -------- extent -------- |
859 if (start > key.offset && end >= extent_end) {
861 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
863 btrfs_set_file_extent_num_bytes(leaf, fi,
865 btrfs_mark_buffer_dirty(leaf);
866 if (update_refs && disk_bytenr > 0)
867 inode_sub_bytes(inode, extent_end - start);
868 if (end == extent_end)
876 * | ---- range to drop ----- |
877 * | ------ extent ------ |
879 if (start <= key.offset && end >= extent_end) {
881 del_slot = path->slots[0];
884 BUG_ON(del_slot + del_nr != path->slots[0]);
889 extent_type == BTRFS_FILE_EXTENT_INLINE) {
890 inode_sub_bytes(inode,
891 extent_end - key.offset);
892 extent_end = ALIGN(extent_end,
894 } else if (update_refs && disk_bytenr > 0) {
895 ret = btrfs_free_extent(trans, root,
896 disk_bytenr, num_bytes, 0,
897 root->root_key.objectid,
898 key.objectid, key.offset -
900 BUG_ON(ret); /* -ENOMEM */
901 inode_sub_bytes(inode,
902 extent_end - key.offset);
905 if (end == extent_end)
908 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
913 ret = btrfs_del_items(trans, root, path, del_slot,
916 btrfs_abort_transaction(trans, root, ret);
923 btrfs_release_path(path);
930 if (!ret && del_nr > 0) {
931 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
933 btrfs_abort_transaction(trans, root, ret);
937 *drop_end = found ? min(end, extent_end) : end;
938 btrfs_release_path(path);
942 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
943 struct btrfs_root *root, struct inode *inode, u64 start,
944 u64 end, int drop_cache)
946 struct btrfs_path *path;
949 path = btrfs_alloc_path();
952 ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL,
954 btrfs_free_path(path);
958 static int extent_mergeable(struct extent_buffer *leaf, int slot,
959 u64 objectid, u64 bytenr, u64 orig_offset,
960 u64 *start, u64 *end)
962 struct btrfs_file_extent_item *fi;
963 struct btrfs_key key;
966 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
969 btrfs_item_key_to_cpu(leaf, &key, slot);
970 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
973 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
974 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
975 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
976 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
977 btrfs_file_extent_compression(leaf, fi) ||
978 btrfs_file_extent_encryption(leaf, fi) ||
979 btrfs_file_extent_other_encoding(leaf, fi))
982 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
983 if ((*start && *start != key.offset) || (*end && *end != extent_end))
992 * Mark extent in the range start - end as written.
994 * This changes extent type from 'pre-allocated' to 'regular'. If only
995 * part of extent is marked as written, the extent will be split into
998 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
999 struct inode *inode, u64 start, u64 end)
1001 struct btrfs_root *root = BTRFS_I(inode)->root;
1002 struct extent_buffer *leaf;
1003 struct btrfs_path *path;
1004 struct btrfs_file_extent_item *fi;
1005 struct btrfs_key key;
1006 struct btrfs_key new_key;
1018 u64 ino = btrfs_ino(inode);
1020 path = btrfs_alloc_path();
1027 key.type = BTRFS_EXTENT_DATA_KEY;
1030 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1033 if (ret > 0 && path->slots[0] > 0)
1036 leaf = path->nodes[0];
1037 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1038 BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY);
1039 fi = btrfs_item_ptr(leaf, path->slots[0],
1040 struct btrfs_file_extent_item);
1041 BUG_ON(btrfs_file_extent_type(leaf, fi) !=
1042 BTRFS_FILE_EXTENT_PREALLOC);
1043 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1044 BUG_ON(key.offset > start || extent_end < end);
1046 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1047 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1048 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
1049 memcpy(&new_key, &key, sizeof(new_key));
1051 if (start == key.offset && end < extent_end) {
1054 if (extent_mergeable(leaf, path->slots[0] - 1,
1055 ino, bytenr, orig_offset,
1056 &other_start, &other_end)) {
1057 new_key.offset = end;
1058 btrfs_set_item_key_safe(root, path, &new_key);
1059 fi = btrfs_item_ptr(leaf, path->slots[0],
1060 struct btrfs_file_extent_item);
1061 btrfs_set_file_extent_generation(leaf, fi,
1063 btrfs_set_file_extent_num_bytes(leaf, fi,
1065 btrfs_set_file_extent_offset(leaf, fi,
1067 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1068 struct btrfs_file_extent_item);
1069 btrfs_set_file_extent_generation(leaf, fi,
1071 btrfs_set_file_extent_num_bytes(leaf, fi,
1073 btrfs_mark_buffer_dirty(leaf);
1078 if (start > key.offset && end == extent_end) {
1081 if (extent_mergeable(leaf, path->slots[0] + 1,
1082 ino, bytenr, orig_offset,
1083 &other_start, &other_end)) {
1084 fi = btrfs_item_ptr(leaf, path->slots[0],
1085 struct btrfs_file_extent_item);
1086 btrfs_set_file_extent_num_bytes(leaf, fi,
1087 start - key.offset);
1088 btrfs_set_file_extent_generation(leaf, fi,
1091 new_key.offset = start;
1092 btrfs_set_item_key_safe(root, path, &new_key);
1094 fi = btrfs_item_ptr(leaf, path->slots[0],
1095 struct btrfs_file_extent_item);
1096 btrfs_set_file_extent_generation(leaf, fi,
1098 btrfs_set_file_extent_num_bytes(leaf, fi,
1100 btrfs_set_file_extent_offset(leaf, fi,
1101 start - orig_offset);
1102 btrfs_mark_buffer_dirty(leaf);
1107 while (start > key.offset || end < extent_end) {
1108 if (key.offset == start)
1111 new_key.offset = split;
1112 ret = btrfs_duplicate_item(trans, root, path, &new_key);
1113 if (ret == -EAGAIN) {
1114 btrfs_release_path(path);
1118 btrfs_abort_transaction(trans, root, ret);
1122 leaf = path->nodes[0];
1123 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1124 struct btrfs_file_extent_item);
1125 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1126 btrfs_set_file_extent_num_bytes(leaf, fi,
1127 split - key.offset);
1129 fi = btrfs_item_ptr(leaf, path->slots[0],
1130 struct btrfs_file_extent_item);
1132 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1133 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
1134 btrfs_set_file_extent_num_bytes(leaf, fi,
1135 extent_end - split);
1136 btrfs_mark_buffer_dirty(leaf);
1138 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
1139 root->root_key.objectid,
1140 ino, orig_offset, 0);
1141 BUG_ON(ret); /* -ENOMEM */
1143 if (split == start) {
1146 BUG_ON(start != key.offset);
1155 if (extent_mergeable(leaf, path->slots[0] + 1,
1156 ino, bytenr, orig_offset,
1157 &other_start, &other_end)) {
1159 btrfs_release_path(path);
1162 extent_end = other_end;
1163 del_slot = path->slots[0] + 1;
1165 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1166 0, root->root_key.objectid,
1167 ino, orig_offset, 0);
1168 BUG_ON(ret); /* -ENOMEM */
1172 if (extent_mergeable(leaf, path->slots[0] - 1,
1173 ino, bytenr, orig_offset,
1174 &other_start, &other_end)) {
1176 btrfs_release_path(path);
1179 key.offset = other_start;
1180 del_slot = path->slots[0];
1182 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1183 0, root->root_key.objectid,
1184 ino, orig_offset, 0);
1185 BUG_ON(ret); /* -ENOMEM */
1188 fi = btrfs_item_ptr(leaf, path->slots[0],
1189 struct btrfs_file_extent_item);
1190 btrfs_set_file_extent_type(leaf, fi,
1191 BTRFS_FILE_EXTENT_REG);
1192 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1193 btrfs_mark_buffer_dirty(leaf);
1195 fi = btrfs_item_ptr(leaf, del_slot - 1,
1196 struct btrfs_file_extent_item);
1197 btrfs_set_file_extent_type(leaf, fi,
1198 BTRFS_FILE_EXTENT_REG);
1199 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1200 btrfs_set_file_extent_num_bytes(leaf, fi,
1201 extent_end - key.offset);
1202 btrfs_mark_buffer_dirty(leaf);
1204 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1206 btrfs_abort_transaction(trans, root, ret);
1211 btrfs_free_path(path);
1216 * on error we return an unlocked page and the error value
1217 * on success we return a locked page and 0
1219 static int prepare_uptodate_page(struct page *page, u64 pos,
1220 bool force_uptodate)
1224 if (((pos & (PAGE_CACHE_SIZE - 1)) || force_uptodate) &&
1225 !PageUptodate(page)) {
1226 ret = btrfs_readpage(NULL, page);
1230 if (!PageUptodate(page)) {
1239 * this gets pages into the page cache and locks them down, it also properly
1240 * waits for data=ordered extents to finish before allowing the pages to be
1243 static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
1244 struct page **pages, size_t num_pages,
1245 loff_t pos, unsigned long first_index,
1246 size_t write_bytes, bool force_uptodate)
1248 struct extent_state *cached_state = NULL;
1250 unsigned long index = pos >> PAGE_CACHE_SHIFT;
1251 struct inode *inode = file_inode(file);
1252 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1258 start_pos = pos & ~((u64)root->sectorsize - 1);
1259 last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
1262 for (i = 0; i < num_pages; i++) {
1263 pages[i] = find_or_create_page(inode->i_mapping, index + i,
1264 mask | __GFP_WRITE);
1272 err = prepare_uptodate_page(pages[i], pos,
1274 if (i == num_pages - 1)
1275 err = prepare_uptodate_page(pages[i],
1276 pos + write_bytes, false);
1278 page_cache_release(pages[i]);
1282 wait_on_page_writeback(pages[i]);
1285 if (start_pos < inode->i_size) {
1286 struct btrfs_ordered_extent *ordered;
1287 lock_extent_bits(&BTRFS_I(inode)->io_tree,
1288 start_pos, last_pos - 1, 0, &cached_state);
1289 ordered = btrfs_lookup_first_ordered_extent(inode,
1292 ordered->file_offset + ordered->len > start_pos &&
1293 ordered->file_offset < last_pos) {
1294 btrfs_put_ordered_extent(ordered);
1295 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1296 start_pos, last_pos - 1,
1297 &cached_state, GFP_NOFS);
1298 for (i = 0; i < num_pages; i++) {
1299 unlock_page(pages[i]);
1300 page_cache_release(pages[i]);
1302 btrfs_wait_ordered_range(inode, start_pos,
1303 last_pos - start_pos);
1307 btrfs_put_ordered_extent(ordered);
1309 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
1310 last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1311 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
1312 0, 0, &cached_state, GFP_NOFS);
1313 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1314 start_pos, last_pos - 1, &cached_state,
1317 for (i = 0; i < num_pages; i++) {
1318 if (clear_page_dirty_for_io(pages[i]))
1319 account_page_redirty(pages[i]);
1320 set_page_extent_mapped(pages[i]);
1321 WARN_ON(!PageLocked(pages[i]));
1325 while (faili >= 0) {
1326 unlock_page(pages[faili]);
1327 page_cache_release(pages[faili]);
1334 static noinline int check_can_nocow(struct inode *inode, loff_t pos,
1335 size_t *write_bytes)
1337 struct btrfs_root *root = BTRFS_I(inode)->root;
1338 struct btrfs_ordered_extent *ordered;
1339 u64 lockstart, lockend;
1343 lockstart = round_down(pos, root->sectorsize);
1344 lockend = lockstart + round_up(*write_bytes, root->sectorsize) - 1;
1347 lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1348 ordered = btrfs_lookup_ordered_range(inode, lockstart,
1349 lockend - lockstart + 1);
1353 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1354 btrfs_start_ordered_extent(inode, ordered, 1);
1355 btrfs_put_ordered_extent(ordered);
1358 num_bytes = lockend - lockstart + 1;
1359 ret = can_nocow_extent(inode, lockstart, &num_bytes, NULL, NULL, NULL);
1363 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
1364 EXTENT_DIRTY | EXTENT_DELALLOC |
1365 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1367 *write_bytes = min_t(size_t, *write_bytes, num_bytes);
1370 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1375 static noinline ssize_t __btrfs_buffered_write(struct file *file,
1379 struct inode *inode = file_inode(file);
1380 struct btrfs_root *root = BTRFS_I(inode)->root;
1381 struct page **pages = NULL;
1382 u64 release_bytes = 0;
1383 unsigned long first_index;
1384 size_t num_written = 0;
1387 bool only_release_metadata = false;
1388 bool force_page_uptodate = false;
1390 nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) /
1391 PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
1392 (sizeof(struct page *)));
1393 nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1394 nrptrs = max(nrptrs, 8);
1395 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
1399 first_index = pos >> PAGE_CACHE_SHIFT;
1401 while (iov_iter_count(i) > 0) {
1402 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1403 size_t write_bytes = min(iov_iter_count(i),
1404 nrptrs * (size_t)PAGE_CACHE_SIZE -
1406 size_t num_pages = (write_bytes + offset +
1407 PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1408 size_t reserve_bytes;
1412 WARN_ON(num_pages > nrptrs);
1415 * Fault pages before locking them in prepare_pages
1416 * to avoid recursive lock
1418 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1423 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1424 ret = btrfs_check_data_free_space(inode, reserve_bytes);
1425 if (ret == -ENOSPC &&
1426 (BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW |
1427 BTRFS_INODE_PREALLOC))) {
1428 ret = check_can_nocow(inode, pos, &write_bytes);
1430 only_release_metadata = true;
1432 * our prealloc extent may be smaller than
1433 * write_bytes, so scale down.
1435 num_pages = (write_bytes + offset +
1436 PAGE_CACHE_SIZE - 1) >>
1438 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1448 ret = btrfs_delalloc_reserve_metadata(inode, reserve_bytes);
1450 if (!only_release_metadata)
1451 btrfs_free_reserved_data_space(inode,
1456 release_bytes = reserve_bytes;
1459 * This is going to setup the pages array with the number of
1460 * pages we want, so we don't really need to worry about the
1461 * contents of pages from loop to loop
1463 ret = prepare_pages(root, file, pages, num_pages,
1464 pos, first_index, write_bytes,
1465 force_page_uptodate);
1469 copied = btrfs_copy_from_user(pos, num_pages,
1470 write_bytes, pages, i);
1473 * if we have trouble faulting in the pages, fall
1474 * back to one page at a time
1476 if (copied < write_bytes)
1480 force_page_uptodate = true;
1483 force_page_uptodate = false;
1484 dirty_pages = (copied + offset +
1485 PAGE_CACHE_SIZE - 1) >>
1490 * If we had a short copy we need to release the excess delaloc
1491 * bytes we reserved. We need to increment outstanding_extents
1492 * because btrfs_delalloc_release_space will decrement it, but
1493 * we still have an outstanding extent for the chunk we actually
1496 if (num_pages > dirty_pages) {
1497 release_bytes = (num_pages - dirty_pages) <<
1500 spin_lock(&BTRFS_I(inode)->lock);
1501 BTRFS_I(inode)->outstanding_extents++;
1502 spin_unlock(&BTRFS_I(inode)->lock);
1504 if (only_release_metadata)
1505 btrfs_delalloc_release_metadata(inode,
1508 btrfs_delalloc_release_space(inode,
1512 release_bytes = dirty_pages << PAGE_CACHE_SHIFT;
1514 ret = btrfs_dirty_pages(root, inode, pages,
1515 dirty_pages, pos, copied,
1518 btrfs_drop_pages(pages, num_pages);
1524 btrfs_drop_pages(pages, num_pages);
1526 if (only_release_metadata && copied > 0) {
1527 u64 lockstart = round_down(pos, root->sectorsize);
1528 u64 lockend = lockstart +
1529 (dirty_pages << PAGE_CACHE_SHIFT) - 1;
1531 set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
1532 lockend, EXTENT_NORESERVE, NULL,
1534 only_release_metadata = false;
1539 balance_dirty_pages_ratelimited(inode->i_mapping);
1540 if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1541 btrfs_btree_balance_dirty(root);
1544 num_written += copied;
1549 if (release_bytes) {
1550 if (only_release_metadata)
1551 btrfs_delalloc_release_metadata(inode, release_bytes);
1553 btrfs_delalloc_release_space(inode, release_bytes);
1556 return num_written ? num_written : ret;
1559 static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1560 struct iov_iter *iter, loff_t pos,
1561 loff_t *ppos, size_t count)
1563 struct file *file = iocb->ki_filp;
1565 ssize_t written_buffered;
1569 written = generic_file_direct_write_iter(iocb, iter, pos, ppos, count);
1571 if (written < 0 || written == count)
1576 written_buffered = __btrfs_buffered_write(file, iter, pos);
1577 if (written_buffered < 0) {
1578 err = written_buffered;
1581 endbyte = pos + written_buffered - 1;
1582 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
1585 written += written_buffered;
1586 *ppos = pos + written_buffered;
1587 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
1588 endbyte >> PAGE_CACHE_SHIFT);
1590 return written ? written : err;
1593 static void update_time_for_write(struct inode *inode)
1595 struct timespec now;
1597 if (IS_NOCMTIME(inode))
1600 now = current_fs_time(inode->i_sb);
1601 if (!timespec_equal(&inode->i_mtime, &now))
1602 inode->i_mtime = now;
1604 if (!timespec_equal(&inode->i_ctime, &now))
1605 inode->i_ctime = now;
1607 if (IS_I_VERSION(inode))
1608 inode_inc_iversion(inode);
1611 static ssize_t btrfs_file_write_iter(struct kiocb *iocb,
1612 struct iov_iter *iter, loff_t pos)
1614 struct file *file = iocb->ki_filp;
1615 struct inode *inode = file_inode(file);
1616 struct btrfs_root *root = BTRFS_I(inode)->root;
1617 loff_t *ppos = &iocb->ki_pos;
1619 ssize_t num_written = 0;
1622 bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host);
1624 mutex_lock(&inode->i_mutex);
1626 count = iov_iter_count(iter);
1628 current->backing_dev_info = inode->i_mapping->backing_dev_info;
1629 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1631 mutex_unlock(&inode->i_mutex);
1636 mutex_unlock(&inode->i_mutex);
1640 err = file_remove_suid(file);
1642 mutex_unlock(&inode->i_mutex);
1647 * If BTRFS flips readonly due to some impossible error
1648 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1649 * although we have opened a file as writable, we have
1650 * to stop this write operation to ensure FS consistency.
1652 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
1653 mutex_unlock(&inode->i_mutex);
1659 * We reserve space for updating the inode when we reserve space for the
1660 * extent we are going to write, so we will enospc out there. We don't
1661 * need to start yet another transaction to update the inode as we will
1662 * update the inode when we finish writing whatever data we write.
1664 update_time_for_write(inode);
1666 start_pos = round_down(pos, root->sectorsize);
1667 if (start_pos > i_size_read(inode)) {
1668 err = btrfs_cont_expand(inode, i_size_read(inode), start_pos);
1670 mutex_unlock(&inode->i_mutex);
1676 atomic_inc(&BTRFS_I(inode)->sync_writers);
1678 if (unlikely(file->f_flags & O_DIRECT)) {
1679 num_written = __btrfs_direct_write(iocb, iter, pos, ppos,
1682 num_written = __btrfs_buffered_write(file, iter, pos);
1683 if (num_written > 0)
1684 *ppos = pos + num_written;
1687 mutex_unlock(&inode->i_mutex);
1690 * we want to make sure fsync finds this change
1691 * but we haven't joined a transaction running right now.
1693 * Later on, someone is sure to update the inode and get the
1694 * real transid recorded.
1696 * We set last_trans now to the fs_info generation + 1,
1697 * this will either be one more than the running transaction
1698 * or the generation used for the next transaction if there isn't
1699 * one running right now.
1701 * We also have to set last_sub_trans to the current log transid,
1702 * otherwise subsequent syncs to a file that's been synced in this
1703 * transaction will appear to have already occured.
1705 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1706 BTRFS_I(inode)->last_sub_trans = root->log_transid;
1707 if (num_written > 0) {
1708 err = generic_write_sync(file, pos, num_written);
1709 if (err < 0 && num_written > 0)
1714 atomic_dec(&BTRFS_I(inode)->sync_writers);
1716 current->backing_dev_info = NULL;
1717 return num_written ? num_written : err;
1720 int btrfs_release_file(struct inode *inode, struct file *filp)
1723 * ordered_data_close is set by settattr when we are about to truncate
1724 * a file from a non-zero size to a zero size. This tries to
1725 * flush down new bytes that may have been written if the
1726 * application were using truncate to replace a file in place.
1728 if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
1729 &BTRFS_I(inode)->runtime_flags)) {
1730 struct btrfs_trans_handle *trans;
1731 struct btrfs_root *root = BTRFS_I(inode)->root;
1734 * We need to block on a committing transaction to keep us from
1735 * throwing a ordered operation on to the list and causing
1736 * something like sync to deadlock trying to flush out this
1739 trans = btrfs_start_transaction(root, 0);
1741 return PTR_ERR(trans);
1742 btrfs_add_ordered_operation(trans, BTRFS_I(inode)->root, inode);
1743 btrfs_end_transaction(trans, root);
1744 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1745 filemap_flush(inode->i_mapping);
1747 if (filp->private_data)
1748 btrfs_ioctl_trans_end(filp);
1753 * fsync call for both files and directories. This logs the inode into
1754 * the tree log instead of forcing full commits whenever possible.
1756 * It needs to call filemap_fdatawait so that all ordered extent updates are
1757 * in the metadata btree are up to date for copying to the log.
1759 * It drops the inode mutex before doing the tree log commit. This is an
1760 * important optimization for directories because holding the mutex prevents
1761 * new operations on the dir while we write to disk.
1763 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1765 struct dentry *dentry = file->f_path.dentry;
1766 struct inode *inode = dentry->d_inode;
1767 struct btrfs_root *root = BTRFS_I(inode)->root;
1769 struct btrfs_trans_handle *trans;
1772 trace_btrfs_sync_file(file, datasync);
1775 * We write the dirty pages in the range and wait until they complete
1776 * out of the ->i_mutex. If so, we can flush the dirty pages by
1777 * multi-task, and make the performance up. See
1778 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1780 atomic_inc(&BTRFS_I(inode)->sync_writers);
1781 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
1782 if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1783 &BTRFS_I(inode)->runtime_flags))
1784 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
1785 atomic_dec(&BTRFS_I(inode)->sync_writers);
1789 mutex_lock(&inode->i_mutex);
1792 * We flush the dirty pages again to avoid some dirty pages in the
1795 atomic_inc(&root->log_batch);
1796 full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1797 &BTRFS_I(inode)->runtime_flags);
1799 btrfs_wait_ordered_range(inode, start, end - start + 1);
1800 atomic_inc(&root->log_batch);
1803 * check the transaction that last modified this inode
1804 * and see if its already been committed
1806 if (!BTRFS_I(inode)->last_trans) {
1807 mutex_unlock(&inode->i_mutex);
1812 * if the last transaction that changed this file was before
1813 * the current transaction, we can bail out now without any
1817 if (btrfs_inode_in_log(inode, root->fs_info->generation) ||
1818 BTRFS_I(inode)->last_trans <=
1819 root->fs_info->last_trans_committed) {
1820 BTRFS_I(inode)->last_trans = 0;
1823 * We'v had everything committed since the last time we were
1824 * modified so clear this flag in case it was set for whatever
1825 * reason, it's no longer relevant.
1827 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1828 &BTRFS_I(inode)->runtime_flags);
1829 mutex_unlock(&inode->i_mutex);
1834 * ok we haven't committed the transaction yet, lets do a commit
1836 if (file->private_data)
1837 btrfs_ioctl_trans_end(file);
1839 trans = btrfs_start_transaction(root, 0);
1840 if (IS_ERR(trans)) {
1841 ret = PTR_ERR(trans);
1842 mutex_unlock(&inode->i_mutex);
1846 ret = btrfs_log_dentry_safe(trans, root, dentry);
1848 /* Fallthrough and commit/free transaction. */
1852 /* we've logged all the items and now have a consistent
1853 * version of the file in the log. It is possible that
1854 * someone will come in and modify the file, but that's
1855 * fine because the log is consistent on disk, and we
1856 * have references to all of the file's extents
1858 * It is possible that someone will come in and log the
1859 * file again, but that will end up using the synchronization
1860 * inside btrfs_sync_log to keep things safe.
1862 mutex_unlock(&inode->i_mutex);
1864 if (ret != BTRFS_NO_LOG_SYNC) {
1867 * If we didn't already wait for ordered extents we need
1871 btrfs_wait_ordered_range(inode, start,
1873 ret = btrfs_commit_transaction(trans, root);
1875 ret = btrfs_sync_log(trans, root);
1877 ret = btrfs_end_transaction(trans, root);
1880 btrfs_wait_ordered_range(inode, start,
1883 ret = btrfs_commit_transaction(trans, root);
1887 ret = btrfs_end_transaction(trans, root);
1890 return ret > 0 ? -EIO : ret;
1893 static const struct vm_operations_struct btrfs_file_vm_ops = {
1894 .fault = filemap_fault,
1895 .page_mkwrite = btrfs_page_mkwrite,
1896 .remap_pages = generic_file_remap_pages,
1899 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
1901 struct address_space *mapping = filp->f_mapping;
1903 if (!mapping->a_ops->readpage)
1906 file_accessed(filp);
1907 vma->vm_ops = &btrfs_file_vm_ops;
1912 static int hole_mergeable(struct inode *inode, struct extent_buffer *leaf,
1913 int slot, u64 start, u64 end)
1915 struct btrfs_file_extent_item *fi;
1916 struct btrfs_key key;
1918 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
1921 btrfs_item_key_to_cpu(leaf, &key, slot);
1922 if (key.objectid != btrfs_ino(inode) ||
1923 key.type != BTRFS_EXTENT_DATA_KEY)
1926 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1928 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
1931 if (btrfs_file_extent_disk_bytenr(leaf, fi))
1934 if (key.offset == end)
1936 if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
1941 static int fill_holes(struct btrfs_trans_handle *trans, struct inode *inode,
1942 struct btrfs_path *path, u64 offset, u64 end)
1944 struct btrfs_root *root = BTRFS_I(inode)->root;
1945 struct extent_buffer *leaf;
1946 struct btrfs_file_extent_item *fi;
1947 struct extent_map *hole_em;
1948 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1949 struct btrfs_key key;
1952 key.objectid = btrfs_ino(inode);
1953 key.type = BTRFS_EXTENT_DATA_KEY;
1954 key.offset = offset;
1957 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1962 leaf = path->nodes[0];
1963 if (hole_mergeable(inode, leaf, path->slots[0]-1, offset, end)) {
1967 fi = btrfs_item_ptr(leaf, path->slots[0],
1968 struct btrfs_file_extent_item);
1969 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
1971 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1972 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
1973 btrfs_set_file_extent_offset(leaf, fi, 0);
1974 btrfs_mark_buffer_dirty(leaf);
1978 if (hole_mergeable(inode, leaf, path->slots[0]+1, offset, end)) {
1982 key.offset = offset;
1983 btrfs_set_item_key_safe(root, path, &key);
1984 fi = btrfs_item_ptr(leaf, path->slots[0],
1985 struct btrfs_file_extent_item);
1986 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
1988 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1989 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
1990 btrfs_set_file_extent_offset(leaf, fi, 0);
1991 btrfs_mark_buffer_dirty(leaf);
1994 btrfs_release_path(path);
1996 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
1997 0, 0, end - offset, 0, end - offset,
2003 btrfs_release_path(path);
2005 hole_em = alloc_extent_map();
2007 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2008 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2009 &BTRFS_I(inode)->runtime_flags);
2011 hole_em->start = offset;
2012 hole_em->len = end - offset;
2013 hole_em->ram_bytes = hole_em->len;
2014 hole_em->orig_start = offset;
2016 hole_em->block_start = EXTENT_MAP_HOLE;
2017 hole_em->block_len = 0;
2018 hole_em->orig_block_len = 0;
2019 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
2020 hole_em->compress_type = BTRFS_COMPRESS_NONE;
2021 hole_em->generation = trans->transid;
2024 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2025 write_lock(&em_tree->lock);
2026 ret = add_extent_mapping(em_tree, hole_em, 1);
2027 write_unlock(&em_tree->lock);
2028 } while (ret == -EEXIST);
2029 free_extent_map(hole_em);
2031 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2032 &BTRFS_I(inode)->runtime_flags);
2038 static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
2040 struct btrfs_root *root = BTRFS_I(inode)->root;
2041 struct extent_state *cached_state = NULL;
2042 struct btrfs_path *path;
2043 struct btrfs_block_rsv *rsv;
2044 struct btrfs_trans_handle *trans;
2045 u64 lockstart = round_up(offset, BTRFS_I(inode)->root->sectorsize);
2046 u64 lockend = round_down(offset + len,
2047 BTRFS_I(inode)->root->sectorsize) - 1;
2048 u64 cur_offset = lockstart;
2049 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
2053 bool same_page = ((offset >> PAGE_CACHE_SHIFT) ==
2054 ((offset + len - 1) >> PAGE_CACHE_SHIFT));
2056 btrfs_wait_ordered_range(inode, offset, len);
2058 mutex_lock(&inode->i_mutex);
2060 * We needn't truncate any page which is beyond the end of the file
2061 * because we are sure there is no data there.
2064 * Only do this if we are in the same page and we aren't doing the
2067 if (same_page && len < PAGE_CACHE_SIZE) {
2068 if (offset < round_up(inode->i_size, PAGE_CACHE_SIZE))
2069 ret = btrfs_truncate_page(inode, offset, len, 0);
2070 mutex_unlock(&inode->i_mutex);
2074 /* zero back part of the first page */
2075 if (offset < round_up(inode->i_size, PAGE_CACHE_SIZE)) {
2076 ret = btrfs_truncate_page(inode, offset, 0, 0);
2078 mutex_unlock(&inode->i_mutex);
2083 /* zero the front end of the last page */
2084 if (offset + len < round_up(inode->i_size, PAGE_CACHE_SIZE)) {
2085 ret = btrfs_truncate_page(inode, offset + len, 0, 1);
2087 mutex_unlock(&inode->i_mutex);
2092 if (lockend < lockstart) {
2093 mutex_unlock(&inode->i_mutex);
2098 struct btrfs_ordered_extent *ordered;
2100 truncate_pagecache_range(inode, lockstart, lockend);
2102 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2104 ordered = btrfs_lookup_first_ordered_extent(inode, lockend);
2107 * We need to make sure we have no ordered extents in this range
2108 * and nobody raced in and read a page in this range, if we did
2109 * we need to try again.
2112 (ordered->file_offset + ordered->len < lockstart ||
2113 ordered->file_offset > lockend)) &&
2114 !test_range_bit(&BTRFS_I(inode)->io_tree, lockstart,
2115 lockend, EXTENT_UPTODATE, 0,
2118 btrfs_put_ordered_extent(ordered);
2122 btrfs_put_ordered_extent(ordered);
2123 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
2124 lockend, &cached_state, GFP_NOFS);
2125 btrfs_wait_ordered_range(inode, lockstart,
2126 lockend - lockstart + 1);
2129 path = btrfs_alloc_path();
2135 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2140 rsv->size = btrfs_calc_trunc_metadata_size(root, 1);
2144 * 1 - update the inode
2145 * 1 - removing the extents in the range
2146 * 1 - adding the hole extent
2148 trans = btrfs_start_transaction(root, 3);
2149 if (IS_ERR(trans)) {
2150 err = PTR_ERR(trans);
2154 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
2157 trans->block_rsv = rsv;
2159 while (cur_offset < lockend) {
2160 ret = __btrfs_drop_extents(trans, root, inode, path,
2161 cur_offset, lockend + 1,
2166 trans->block_rsv = &root->fs_info->trans_block_rsv;
2168 ret = fill_holes(trans, inode, path, cur_offset, drop_end);
2174 cur_offset = drop_end;
2176 ret = btrfs_update_inode(trans, root, inode);
2182 btrfs_end_transaction(trans, root);
2183 btrfs_btree_balance_dirty(root);
2185 trans = btrfs_start_transaction(root, 3);
2186 if (IS_ERR(trans)) {
2187 ret = PTR_ERR(trans);
2192 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
2194 BUG_ON(ret); /* shouldn't happen */
2195 trans->block_rsv = rsv;
2203 trans->block_rsv = &root->fs_info->trans_block_rsv;
2204 ret = fill_holes(trans, inode, path, cur_offset, drop_end);
2214 inode_inc_iversion(inode);
2215 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2217 trans->block_rsv = &root->fs_info->trans_block_rsv;
2218 ret = btrfs_update_inode(trans, root, inode);
2219 btrfs_end_transaction(trans, root);
2220 btrfs_btree_balance_dirty(root);
2222 btrfs_free_path(path);
2223 btrfs_free_block_rsv(root, rsv);
2225 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2226 &cached_state, GFP_NOFS);
2227 mutex_unlock(&inode->i_mutex);
2233 static long btrfs_fallocate(struct file *file, int mode,
2234 loff_t offset, loff_t len)
2236 struct inode *inode = file_inode(file);
2237 struct extent_state *cached_state = NULL;
2238 struct btrfs_root *root = BTRFS_I(inode)->root;
2245 struct extent_map *em;
2246 int blocksize = BTRFS_I(inode)->root->sectorsize;
2249 alloc_start = round_down(offset, blocksize);
2250 alloc_end = round_up(offset + len, blocksize);
2252 /* Make sure we aren't being give some crap mode */
2253 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2256 if (mode & FALLOC_FL_PUNCH_HOLE)
2257 return btrfs_punch_hole(inode, offset, len);
2260 * Make sure we have enough space before we do the
2263 ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
2266 if (root->fs_info->quota_enabled) {
2267 ret = btrfs_qgroup_reserve(root, alloc_end - alloc_start);
2269 goto out_reserve_fail;
2272 mutex_lock(&inode->i_mutex);
2273 ret = inode_newsize_ok(inode, alloc_end);
2277 if (alloc_start > inode->i_size) {
2278 ret = btrfs_cont_expand(inode, i_size_read(inode),
2284 * If we are fallocating from the end of the file onward we
2285 * need to zero out the end of the page if i_size lands in the
2288 ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
2294 * wait for ordered IO before we have any locks. We'll loop again
2295 * below with the locks held.
2297 btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
2299 locked_end = alloc_end - 1;
2301 struct btrfs_ordered_extent *ordered;
2303 /* the extent lock is ordered inside the running
2306 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
2307 locked_end, 0, &cached_state);
2308 ordered = btrfs_lookup_first_ordered_extent(inode,
2311 ordered->file_offset + ordered->len > alloc_start &&
2312 ordered->file_offset < alloc_end) {
2313 btrfs_put_ordered_extent(ordered);
2314 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
2315 alloc_start, locked_end,
2316 &cached_state, GFP_NOFS);
2318 * we can't wait on the range with the transaction
2319 * running or with the extent lock held
2321 btrfs_wait_ordered_range(inode, alloc_start,
2322 alloc_end - alloc_start);
2325 btrfs_put_ordered_extent(ordered);
2330 cur_offset = alloc_start;
2334 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2335 alloc_end - cur_offset, 0);
2336 if (IS_ERR_OR_NULL(em)) {
2343 last_byte = min(extent_map_end(em), alloc_end);
2344 actual_end = min_t(u64, extent_map_end(em), offset + len);
2345 last_byte = ALIGN(last_byte, blocksize);
2347 if (em->block_start == EXTENT_MAP_HOLE ||
2348 (cur_offset >= inode->i_size &&
2349 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
2350 ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
2351 last_byte - cur_offset,
2352 1 << inode->i_blkbits,
2357 free_extent_map(em);
2360 } else if (actual_end > inode->i_size &&
2361 !(mode & FALLOC_FL_KEEP_SIZE)) {
2363 * We didn't need to allocate any more space, but we
2364 * still extended the size of the file so we need to
2367 inode->i_ctime = CURRENT_TIME;
2368 i_size_write(inode, actual_end);
2369 btrfs_ordered_update_i_size(inode, actual_end, NULL);
2371 free_extent_map(em);
2373 cur_offset = last_byte;
2374 if (cur_offset >= alloc_end) {
2379 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
2380 &cached_state, GFP_NOFS);
2382 mutex_unlock(&inode->i_mutex);
2383 if (root->fs_info->quota_enabled)
2384 btrfs_qgroup_free(root, alloc_end - alloc_start);
2386 /* Let go of our reservation. */
2387 btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
2391 static int find_desired_extent(struct inode *inode, loff_t *offset, int whence)
2393 struct btrfs_root *root = BTRFS_I(inode)->root;
2394 struct extent_map *em;
2395 struct extent_state *cached_state = NULL;
2396 u64 lockstart = *offset;
2397 u64 lockend = i_size_read(inode);
2398 u64 start = *offset;
2399 u64 orig_start = *offset;
2400 u64 len = i_size_read(inode);
2404 lockend = max_t(u64, root->sectorsize, lockend);
2405 if (lockend <= lockstart)
2406 lockend = lockstart + root->sectorsize;
2409 len = lockend - lockstart + 1;
2411 len = max_t(u64, len, root->sectorsize);
2412 if (inode->i_size == 0)
2415 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, 0,
2419 * Delalloc is such a pain. If we have a hole and we have pending
2420 * delalloc for a portion of the hole we will get back a hole that
2421 * exists for the entire range since it hasn't been actually written
2422 * yet. So to take care of this case we need to look for an extent just
2423 * before the position we want in case there is outstanding delalloc
2426 if (whence == SEEK_HOLE && start != 0) {
2427 if (start <= root->sectorsize)
2428 em = btrfs_get_extent_fiemap(inode, NULL, 0, 0,
2429 root->sectorsize, 0);
2431 em = btrfs_get_extent_fiemap(inode, NULL, 0,
2432 start - root->sectorsize,
2433 root->sectorsize, 0);
2438 last_end = em->start + em->len;
2439 if (em->block_start == EXTENT_MAP_DELALLOC)
2440 last_end = min_t(u64, last_end, inode->i_size);
2441 free_extent_map(em);
2445 em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0);
2451 if (em->block_start == EXTENT_MAP_HOLE) {
2452 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2453 if (last_end <= orig_start) {
2454 free_extent_map(em);
2460 if (whence == SEEK_HOLE) {
2462 free_extent_map(em);
2466 if (whence == SEEK_DATA) {
2467 if (em->block_start == EXTENT_MAP_DELALLOC) {
2468 if (start >= inode->i_size) {
2469 free_extent_map(em);
2475 if (!test_bit(EXTENT_FLAG_PREALLOC,
2478 free_extent_map(em);
2484 start = em->start + em->len;
2485 last_end = em->start + em->len;
2487 if (em->block_start == EXTENT_MAP_DELALLOC)
2488 last_end = min_t(u64, last_end, inode->i_size);
2490 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2491 free_extent_map(em);
2495 free_extent_map(em);
2499 *offset = min(*offset, inode->i_size);
2501 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2502 &cached_state, GFP_NOFS);
2506 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
2508 struct inode *inode = file->f_mapping->host;
2511 mutex_lock(&inode->i_mutex);
2515 offset = generic_file_llseek(file, offset, whence);
2519 if (offset >= i_size_read(inode)) {
2520 mutex_unlock(&inode->i_mutex);
2524 ret = find_desired_extent(inode, &offset, whence);
2526 mutex_unlock(&inode->i_mutex);
2531 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2533 mutex_unlock(&inode->i_mutex);
2537 const struct file_operations btrfs_file_operations = {
2538 .llseek = btrfs_file_llseek,
2539 .read = do_sync_read,
2540 .write = do_sync_write,
2541 .splice_read = generic_file_splice_read,
2542 .read_iter = generic_file_read_iter,
2543 .write_iter = btrfs_file_write_iter,
2544 .mmap = btrfs_file_mmap,
2545 .open = generic_file_open,
2546 .release = btrfs_release_file,
2547 .fsync = btrfs_sync_file,
2548 .fallocate = btrfs_fallocate,
2549 .unlocked_ioctl = btrfs_ioctl,
2550 #ifdef CONFIG_COMPAT
2551 .compat_ioctl = btrfs_ioctl,
2555 void btrfs_auto_defrag_exit(void)
2557 if (btrfs_inode_defrag_cachep)
2558 kmem_cache_destroy(btrfs_inode_defrag_cachep);
2561 int btrfs_auto_defrag_init(void)
2563 btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
2564 sizeof(struct inode_defrag), 0,
2565 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2567 if (!btrfs_inode_defrag_cachep)