2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 #include <linux/pagemap.h>
21 #include <linux/highmem.h>
22 #include <linux/time.h>
23 #include <linux/init.h>
24 #include <linux/string.h>
25 #include <linux/backing-dev.h>
26 #include <linux/mpage.h>
27 #include <linux/aio.h>
28 #include <linux/falloc.h>
29 #include <linux/swap.h>
30 #include <linux/writeback.h>
31 #include <linux/statfs.h>
32 #include <linux/compat.h>
33 #include <linux/slab.h>
34 #include <linux/btrfs.h>
37 #include "transaction.h"
38 #include "btrfs_inode.h"
39 #include "print-tree.h"
44 static struct kmem_cache *btrfs_inode_defrag_cachep;
46 * when auto defrag is enabled we
47 * queue up these defrag structs to remember which
48 * inodes need defragging passes
51 struct rb_node rb_node;
55 * transid where the defrag was added, we search for
56 * extents newer than this
63 /* last offset we were able to defrag */
66 /* if we've wrapped around back to zero once already */
70 static int __compare_inode_defrag(struct inode_defrag *defrag1,
71 struct inode_defrag *defrag2)
73 if (defrag1->root > defrag2->root)
75 else if (defrag1->root < defrag2->root)
77 else if (defrag1->ino > defrag2->ino)
79 else if (defrag1->ino < defrag2->ino)
85 /* pop a record for an inode into the defrag tree. The lock
86 * must be held already
88 * If you're inserting a record for an older transid than an
89 * existing record, the transid already in the tree is lowered
91 * If an existing record is found the defrag item you
94 static int __btrfs_add_inode_defrag(struct inode *inode,
95 struct inode_defrag *defrag)
97 struct btrfs_root *root = BTRFS_I(inode)->root;
98 struct inode_defrag *entry;
100 struct rb_node *parent = NULL;
103 p = &root->fs_info->defrag_inodes.rb_node;
106 entry = rb_entry(parent, struct inode_defrag, rb_node);
108 ret = __compare_inode_defrag(defrag, entry);
110 p = &parent->rb_left;
112 p = &parent->rb_right;
114 /* if we're reinserting an entry for
115 * an old defrag run, make sure to
116 * lower the transid of our existing record
118 if (defrag->transid < entry->transid)
119 entry->transid = defrag->transid;
120 if (defrag->last_offset > entry->last_offset)
121 entry->last_offset = defrag->last_offset;
125 set_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
126 rb_link_node(&defrag->rb_node, parent, p);
127 rb_insert_color(&defrag->rb_node, &root->fs_info->defrag_inodes);
131 static inline int __need_auto_defrag(struct btrfs_root *root)
133 if (!btrfs_test_opt(root, AUTO_DEFRAG))
136 if (btrfs_fs_closing(root->fs_info))
143 * insert a defrag record for this inode if auto defrag is
146 int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans,
149 struct btrfs_root *root = BTRFS_I(inode)->root;
150 struct inode_defrag *defrag;
154 if (!__need_auto_defrag(root))
157 if (test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags))
161 transid = trans->transid;
163 transid = BTRFS_I(inode)->root->last_trans;
165 defrag = kmem_cache_zalloc(btrfs_inode_defrag_cachep, GFP_NOFS);
169 defrag->ino = btrfs_ino(inode);
170 defrag->transid = transid;
171 defrag->root = root->root_key.objectid;
173 spin_lock(&root->fs_info->defrag_inodes_lock);
174 if (!test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags)) {
176 * If we set IN_DEFRAG flag and evict the inode from memory,
177 * and then re-read this inode, this new inode doesn't have
178 * IN_DEFRAG flag. At the case, we may find the existed defrag.
180 ret = __btrfs_add_inode_defrag(inode, defrag);
182 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
184 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
186 spin_unlock(&root->fs_info->defrag_inodes_lock);
191 * Requeue the defrag object. If there is a defrag object that points to
192 * the same inode in the tree, we will merge them together (by
193 * __btrfs_add_inode_defrag()) and free the one that we want to requeue.
195 static void btrfs_requeue_inode_defrag(struct inode *inode,
196 struct inode_defrag *defrag)
198 struct btrfs_root *root = BTRFS_I(inode)->root;
201 if (!__need_auto_defrag(root))
205 * Here we don't check the IN_DEFRAG flag, because we need merge
208 spin_lock(&root->fs_info->defrag_inodes_lock);
209 ret = __btrfs_add_inode_defrag(inode, defrag);
210 spin_unlock(&root->fs_info->defrag_inodes_lock);
215 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
219 * pick the defragable inode that we want, if it doesn't exist, we will get
222 static struct inode_defrag *
223 btrfs_pick_defrag_inode(struct btrfs_fs_info *fs_info, u64 root, u64 ino)
225 struct inode_defrag *entry = NULL;
226 struct inode_defrag tmp;
228 struct rb_node *parent = NULL;
234 spin_lock(&fs_info->defrag_inodes_lock);
235 p = fs_info->defrag_inodes.rb_node;
238 entry = rb_entry(parent, struct inode_defrag, rb_node);
240 ret = __compare_inode_defrag(&tmp, entry);
244 p = parent->rb_right;
249 if (parent && __compare_inode_defrag(&tmp, entry) > 0) {
250 parent = rb_next(parent);
252 entry = rb_entry(parent, struct inode_defrag, rb_node);
258 rb_erase(parent, &fs_info->defrag_inodes);
259 spin_unlock(&fs_info->defrag_inodes_lock);
263 void btrfs_cleanup_defrag_inodes(struct btrfs_fs_info *fs_info)
265 struct inode_defrag *defrag;
266 struct rb_node *node;
268 spin_lock(&fs_info->defrag_inodes_lock);
269 node = rb_first(&fs_info->defrag_inodes);
271 rb_erase(node, &fs_info->defrag_inodes);
272 defrag = rb_entry(node, struct inode_defrag, rb_node);
273 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
275 if (need_resched()) {
276 spin_unlock(&fs_info->defrag_inodes_lock);
278 spin_lock(&fs_info->defrag_inodes_lock);
281 node = rb_first(&fs_info->defrag_inodes);
283 spin_unlock(&fs_info->defrag_inodes_lock);
286 #define BTRFS_DEFRAG_BATCH 1024
288 static int __btrfs_run_defrag_inode(struct btrfs_fs_info *fs_info,
289 struct inode_defrag *defrag)
291 struct btrfs_root *inode_root;
293 struct btrfs_key key;
294 struct btrfs_ioctl_defrag_range_args range;
300 key.objectid = defrag->root;
301 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
302 key.offset = (u64)-1;
304 index = srcu_read_lock(&fs_info->subvol_srcu);
306 inode_root = btrfs_read_fs_root_no_name(fs_info, &key);
307 if (IS_ERR(inode_root)) {
308 ret = PTR_ERR(inode_root);
312 key.objectid = defrag->ino;
313 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
315 inode = btrfs_iget(fs_info->sb, &key, inode_root, NULL);
317 ret = PTR_ERR(inode);
320 srcu_read_unlock(&fs_info->subvol_srcu, index);
322 /* do a chunk of defrag */
323 clear_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
324 memset(&range, 0, sizeof(range));
326 range.start = defrag->last_offset;
328 sb_start_write(fs_info->sb);
329 num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid,
331 sb_end_write(fs_info->sb);
333 * if we filled the whole defrag batch, there
334 * must be more work to do. Queue this defrag
337 if (num_defrag == BTRFS_DEFRAG_BATCH) {
338 defrag->last_offset = range.start;
339 btrfs_requeue_inode_defrag(inode, defrag);
340 } else if (defrag->last_offset && !defrag->cycled) {
342 * we didn't fill our defrag batch, but
343 * we didn't start at zero. Make sure we loop
344 * around to the start of the file.
346 defrag->last_offset = 0;
348 btrfs_requeue_inode_defrag(inode, defrag);
350 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
356 srcu_read_unlock(&fs_info->subvol_srcu, index);
357 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
362 * run through the list of inodes in the FS that need
365 int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info)
367 struct inode_defrag *defrag;
369 u64 root_objectid = 0;
371 atomic_inc(&fs_info->defrag_running);
373 /* Pause the auto defragger. */
374 if (test_bit(BTRFS_FS_STATE_REMOUNTING,
378 if (!__need_auto_defrag(fs_info->tree_root))
381 /* find an inode to defrag */
382 defrag = btrfs_pick_defrag_inode(fs_info, root_objectid,
385 if (root_objectid || first_ino) {
394 first_ino = defrag->ino + 1;
395 root_objectid = defrag->root;
397 __btrfs_run_defrag_inode(fs_info, defrag);
399 atomic_dec(&fs_info->defrag_running);
402 * during unmount, we use the transaction_wait queue to
403 * wait for the defragger to stop
405 wake_up(&fs_info->transaction_wait);
409 /* simple helper to fault in pages and copy. This should go away
410 * and be replaced with calls into generic code.
412 static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
414 struct page **prepared_pages,
418 size_t total_copied = 0;
420 int offset = pos & (PAGE_CACHE_SIZE - 1);
422 while (write_bytes > 0) {
423 size_t count = min_t(size_t,
424 PAGE_CACHE_SIZE - offset, write_bytes);
425 struct page *page = prepared_pages[pg];
427 * Copy data from userspace to the current page
429 * Disable pagefault to avoid recursive lock since
430 * the pages are already locked
433 copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
436 /* Flush processor's dcache for this page */
437 flush_dcache_page(page);
440 * if we get a partial write, we can end up with
441 * partially up to date pages. These add
442 * a lot of complexity, so make sure they don't
443 * happen by forcing this copy to be retried.
445 * The rest of the btrfs_file_write code will fall
446 * back to page at a time copies after we return 0.
448 if (!PageUptodate(page) && copied < count)
451 iov_iter_advance(i, copied);
452 write_bytes -= copied;
453 total_copied += copied;
455 /* Return to btrfs_file_aio_write to fault page */
456 if (unlikely(copied == 0))
459 if (unlikely(copied < PAGE_CACHE_SIZE - offset)) {
470 * unlocks pages after btrfs_file_write is done with them
472 static void btrfs_drop_pages(struct page **pages, size_t num_pages)
475 for (i = 0; i < num_pages; i++) {
476 /* page checked is some magic around finding pages that
477 * have been modified without going through btrfs_set_page_dirty
480 ClearPageChecked(pages[i]);
481 unlock_page(pages[i]);
482 mark_page_accessed(pages[i]);
483 page_cache_release(pages[i]);
488 * after copy_from_user, pages need to be dirtied and we need to make
489 * sure holes are created between the current EOF and the start of
490 * any next extents (if required).
492 * this also makes the decision about creating an inline extent vs
493 * doing real data extents, marking pages dirty and delalloc as required.
495 int btrfs_dirty_pages(struct btrfs_root *root, struct inode *inode,
496 struct page **pages, size_t num_pages,
497 loff_t pos, size_t write_bytes,
498 struct extent_state **cached)
504 u64 end_of_last_block;
505 u64 end_pos = pos + write_bytes;
506 loff_t isize = i_size_read(inode);
508 start_pos = pos & ~((u64)root->sectorsize - 1);
509 num_bytes = ALIGN(write_bytes + pos - start_pos, root->sectorsize);
511 end_of_last_block = start_pos + num_bytes - 1;
512 err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
517 for (i = 0; i < num_pages; i++) {
518 struct page *p = pages[i];
525 * we've only changed i_size in ram, and we haven't updated
526 * the disk i_size. There is no need to log the inode
530 i_size_write(inode, end_pos);
535 * this drops all the extents in the cache that intersect the range
536 * [start, end]. Existing extents are split as required.
538 void btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
541 struct extent_map *em;
542 struct extent_map *split = NULL;
543 struct extent_map *split2 = NULL;
544 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
545 u64 len = end - start + 1;
553 WARN_ON(end < start);
554 if (end == (u64)-1) {
563 split = alloc_extent_map();
565 split2 = alloc_extent_map();
566 if (!split || !split2)
569 write_lock(&em_tree->lock);
570 em = lookup_extent_mapping(em_tree, start, len);
572 write_unlock(&em_tree->lock);
576 gen = em->generation;
577 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
578 if (testend && em->start + em->len >= start + len) {
580 write_unlock(&em_tree->lock);
583 start = em->start + em->len;
585 len = start + len - (em->start + em->len);
587 write_unlock(&em_tree->lock);
590 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
591 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
592 clear_bit(EXTENT_FLAG_LOGGING, &flags);
593 modified = !list_empty(&em->list);
594 remove_extent_mapping(em_tree, em);
598 if (em->start < start) {
599 split->start = em->start;
600 split->len = start - em->start;
602 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
603 split->orig_start = em->orig_start;
604 split->block_start = em->block_start;
607 split->block_len = em->block_len;
609 split->block_len = split->len;
610 split->orig_block_len = max(split->block_len,
612 split->ram_bytes = em->ram_bytes;
614 split->orig_start = split->start;
615 split->block_len = 0;
616 split->block_start = em->block_start;
617 split->orig_block_len = 0;
618 split->ram_bytes = split->len;
621 split->generation = gen;
622 split->bdev = em->bdev;
623 split->flags = flags;
624 split->compress_type = em->compress_type;
625 ret = add_extent_mapping(em_tree, split, modified);
626 BUG_ON(ret); /* Logic error */
627 free_extent_map(split);
631 if (testend && em->start + em->len > start + len) {
632 u64 diff = start + len - em->start;
634 split->start = start + len;
635 split->len = em->start + em->len - (start + len);
636 split->bdev = em->bdev;
637 split->flags = flags;
638 split->compress_type = em->compress_type;
639 split->generation = gen;
641 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
642 split->orig_block_len = max(em->block_len,
645 split->ram_bytes = em->ram_bytes;
647 split->block_len = em->block_len;
648 split->block_start = em->block_start;
649 split->orig_start = em->orig_start;
651 split->block_len = split->len;
652 split->block_start = em->block_start
654 split->orig_start = em->orig_start;
657 split->ram_bytes = split->len;
658 split->orig_start = split->start;
659 split->block_len = 0;
660 split->block_start = em->block_start;
661 split->orig_block_len = 0;
664 ret = add_extent_mapping(em_tree, split, modified);
665 BUG_ON(ret); /* Logic error */
666 free_extent_map(split);
670 write_unlock(&em_tree->lock);
674 /* once for the tree*/
678 free_extent_map(split);
680 free_extent_map(split2);
684 * this is very complex, but the basic idea is to drop all extents
685 * in the range start - end. hint_block is filled in with a block number
686 * that would be a good hint to the block allocator for this file.
688 * If an extent intersects the range but is not entirely inside the range
689 * it is either truncated or split. Anything entirely inside the range
690 * is deleted from the tree.
692 int __btrfs_drop_extents(struct btrfs_trans_handle *trans,
693 struct btrfs_root *root, struct inode *inode,
694 struct btrfs_path *path, u64 start, u64 end,
695 u64 *drop_end, int drop_cache)
697 struct extent_buffer *leaf;
698 struct btrfs_file_extent_item *fi;
699 struct btrfs_key key;
700 struct btrfs_key new_key;
701 u64 ino = btrfs_ino(inode);
702 u64 search_start = start;
705 u64 extent_offset = 0;
712 int modify_tree = -1;
713 int update_refs = (root->ref_cows || root == root->fs_info->tree_root);
717 btrfs_drop_extent_cache(inode, start, end - 1, 0);
719 if (start >= BTRFS_I(inode)->disk_i_size)
724 ret = btrfs_lookup_file_extent(trans, root, path, ino,
725 search_start, modify_tree);
728 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
729 leaf = path->nodes[0];
730 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
731 if (key.objectid == ino &&
732 key.type == BTRFS_EXTENT_DATA_KEY)
737 leaf = path->nodes[0];
738 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
740 ret = btrfs_next_leaf(root, path);
747 leaf = path->nodes[0];
751 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
752 if (key.objectid > ino ||
753 key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
756 fi = btrfs_item_ptr(leaf, path->slots[0],
757 struct btrfs_file_extent_item);
758 extent_type = btrfs_file_extent_type(leaf, fi);
760 if (extent_type == BTRFS_FILE_EXTENT_REG ||
761 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
762 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
763 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
764 extent_offset = btrfs_file_extent_offset(leaf, fi);
765 extent_end = key.offset +
766 btrfs_file_extent_num_bytes(leaf, fi);
767 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
768 extent_end = key.offset +
769 btrfs_file_extent_inline_len(leaf, fi);
772 extent_end = search_start;
775 if (extent_end <= search_start) {
781 search_start = max(key.offset, start);
782 if (recow || !modify_tree) {
784 btrfs_release_path(path);
789 * | - range to drop - |
790 * | -------- extent -------- |
792 if (start > key.offset && end < extent_end) {
794 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
796 memcpy(&new_key, &key, sizeof(new_key));
797 new_key.offset = start;
798 ret = btrfs_duplicate_item(trans, root, path,
800 if (ret == -EAGAIN) {
801 btrfs_release_path(path);
807 leaf = path->nodes[0];
808 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
809 struct btrfs_file_extent_item);
810 btrfs_set_file_extent_num_bytes(leaf, fi,
813 fi = btrfs_item_ptr(leaf, path->slots[0],
814 struct btrfs_file_extent_item);
816 extent_offset += start - key.offset;
817 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
818 btrfs_set_file_extent_num_bytes(leaf, fi,
820 btrfs_mark_buffer_dirty(leaf);
822 if (update_refs && disk_bytenr > 0) {
823 ret = btrfs_inc_extent_ref(trans, root,
824 disk_bytenr, num_bytes, 0,
825 root->root_key.objectid,
827 start - extent_offset, 0);
828 BUG_ON(ret); /* -ENOMEM */
833 * | ---- range to drop ----- |
834 * | -------- extent -------- |
836 if (start <= key.offset && end < extent_end) {
837 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
839 memcpy(&new_key, &key, sizeof(new_key));
840 new_key.offset = end;
841 btrfs_set_item_key_safe(root, path, &new_key);
843 extent_offset += end - key.offset;
844 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
845 btrfs_set_file_extent_num_bytes(leaf, fi,
847 btrfs_mark_buffer_dirty(leaf);
848 if (update_refs && disk_bytenr > 0)
849 inode_sub_bytes(inode, end - key.offset);
853 search_start = extent_end;
855 * | ---- range to drop ----- |
856 * | -------- extent -------- |
858 if (start > key.offset && end >= extent_end) {
860 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
862 btrfs_set_file_extent_num_bytes(leaf, fi,
864 btrfs_mark_buffer_dirty(leaf);
865 if (update_refs && disk_bytenr > 0)
866 inode_sub_bytes(inode, extent_end - start);
867 if (end == extent_end)
875 * | ---- range to drop ----- |
876 * | ------ extent ------ |
878 if (start <= key.offset && end >= extent_end) {
880 del_slot = path->slots[0];
883 BUG_ON(del_slot + del_nr != path->slots[0]);
888 extent_type == BTRFS_FILE_EXTENT_INLINE) {
889 inode_sub_bytes(inode,
890 extent_end - key.offset);
891 extent_end = ALIGN(extent_end,
893 } else if (update_refs && disk_bytenr > 0) {
894 ret = btrfs_free_extent(trans, root,
895 disk_bytenr, num_bytes, 0,
896 root->root_key.objectid,
897 key.objectid, key.offset -
899 BUG_ON(ret); /* -ENOMEM */
900 inode_sub_bytes(inode,
901 extent_end - key.offset);
904 if (end == extent_end)
907 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
912 ret = btrfs_del_items(trans, root, path, del_slot,
915 btrfs_abort_transaction(trans, root, ret);
922 btrfs_release_path(path);
929 if (!ret && del_nr > 0) {
930 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
932 btrfs_abort_transaction(trans, root, ret);
936 *drop_end = found ? min(end, extent_end) : end;
937 btrfs_release_path(path);
941 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
942 struct btrfs_root *root, struct inode *inode, u64 start,
943 u64 end, int drop_cache)
945 struct btrfs_path *path;
948 path = btrfs_alloc_path();
951 ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL,
953 btrfs_free_path(path);
957 static int extent_mergeable(struct extent_buffer *leaf, int slot,
958 u64 objectid, u64 bytenr, u64 orig_offset,
959 u64 *start, u64 *end)
961 struct btrfs_file_extent_item *fi;
962 struct btrfs_key key;
965 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
968 btrfs_item_key_to_cpu(leaf, &key, slot);
969 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
972 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
973 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
974 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
975 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
976 btrfs_file_extent_compression(leaf, fi) ||
977 btrfs_file_extent_encryption(leaf, fi) ||
978 btrfs_file_extent_other_encoding(leaf, fi))
981 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
982 if ((*start && *start != key.offset) || (*end && *end != extent_end))
991 * Mark extent in the range start - end as written.
993 * This changes extent type from 'pre-allocated' to 'regular'. If only
994 * part of extent is marked as written, the extent will be split into
997 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
998 struct inode *inode, u64 start, u64 end)
1000 struct btrfs_root *root = BTRFS_I(inode)->root;
1001 struct extent_buffer *leaf;
1002 struct btrfs_path *path;
1003 struct btrfs_file_extent_item *fi;
1004 struct btrfs_key key;
1005 struct btrfs_key new_key;
1017 u64 ino = btrfs_ino(inode);
1019 path = btrfs_alloc_path();
1026 key.type = BTRFS_EXTENT_DATA_KEY;
1029 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1032 if (ret > 0 && path->slots[0] > 0)
1035 leaf = path->nodes[0];
1036 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1037 BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY);
1038 fi = btrfs_item_ptr(leaf, path->slots[0],
1039 struct btrfs_file_extent_item);
1040 BUG_ON(btrfs_file_extent_type(leaf, fi) !=
1041 BTRFS_FILE_EXTENT_PREALLOC);
1042 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1043 BUG_ON(key.offset > start || extent_end < end);
1045 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1046 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1047 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
1048 memcpy(&new_key, &key, sizeof(new_key));
1050 if (start == key.offset && end < extent_end) {
1053 if (extent_mergeable(leaf, path->slots[0] - 1,
1054 ino, bytenr, orig_offset,
1055 &other_start, &other_end)) {
1056 new_key.offset = end;
1057 btrfs_set_item_key_safe(root, path, &new_key);
1058 fi = btrfs_item_ptr(leaf, path->slots[0],
1059 struct btrfs_file_extent_item);
1060 btrfs_set_file_extent_generation(leaf, fi,
1062 btrfs_set_file_extent_num_bytes(leaf, fi,
1064 btrfs_set_file_extent_offset(leaf, fi,
1066 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1067 struct btrfs_file_extent_item);
1068 btrfs_set_file_extent_generation(leaf, fi,
1070 btrfs_set_file_extent_num_bytes(leaf, fi,
1072 btrfs_mark_buffer_dirty(leaf);
1077 if (start > key.offset && end == extent_end) {
1080 if (extent_mergeable(leaf, path->slots[0] + 1,
1081 ino, bytenr, orig_offset,
1082 &other_start, &other_end)) {
1083 fi = btrfs_item_ptr(leaf, path->slots[0],
1084 struct btrfs_file_extent_item);
1085 btrfs_set_file_extent_num_bytes(leaf, fi,
1086 start - key.offset);
1087 btrfs_set_file_extent_generation(leaf, fi,
1090 new_key.offset = start;
1091 btrfs_set_item_key_safe(root, path, &new_key);
1093 fi = btrfs_item_ptr(leaf, path->slots[0],
1094 struct btrfs_file_extent_item);
1095 btrfs_set_file_extent_generation(leaf, fi,
1097 btrfs_set_file_extent_num_bytes(leaf, fi,
1099 btrfs_set_file_extent_offset(leaf, fi,
1100 start - orig_offset);
1101 btrfs_mark_buffer_dirty(leaf);
1106 while (start > key.offset || end < extent_end) {
1107 if (key.offset == start)
1110 new_key.offset = split;
1111 ret = btrfs_duplicate_item(trans, root, path, &new_key);
1112 if (ret == -EAGAIN) {
1113 btrfs_release_path(path);
1117 btrfs_abort_transaction(trans, root, ret);
1121 leaf = path->nodes[0];
1122 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1123 struct btrfs_file_extent_item);
1124 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1125 btrfs_set_file_extent_num_bytes(leaf, fi,
1126 split - key.offset);
1128 fi = btrfs_item_ptr(leaf, path->slots[0],
1129 struct btrfs_file_extent_item);
1131 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1132 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
1133 btrfs_set_file_extent_num_bytes(leaf, fi,
1134 extent_end - split);
1135 btrfs_mark_buffer_dirty(leaf);
1137 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
1138 root->root_key.objectid,
1139 ino, orig_offset, 0);
1140 BUG_ON(ret); /* -ENOMEM */
1142 if (split == start) {
1145 BUG_ON(start != key.offset);
1154 if (extent_mergeable(leaf, path->slots[0] + 1,
1155 ino, bytenr, orig_offset,
1156 &other_start, &other_end)) {
1158 btrfs_release_path(path);
1161 extent_end = other_end;
1162 del_slot = path->slots[0] + 1;
1164 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1165 0, root->root_key.objectid,
1166 ino, orig_offset, 0);
1167 BUG_ON(ret); /* -ENOMEM */
1171 if (extent_mergeable(leaf, path->slots[0] - 1,
1172 ino, bytenr, orig_offset,
1173 &other_start, &other_end)) {
1175 btrfs_release_path(path);
1178 key.offset = other_start;
1179 del_slot = path->slots[0];
1181 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1182 0, root->root_key.objectid,
1183 ino, orig_offset, 0);
1184 BUG_ON(ret); /* -ENOMEM */
1187 fi = btrfs_item_ptr(leaf, path->slots[0],
1188 struct btrfs_file_extent_item);
1189 btrfs_set_file_extent_type(leaf, fi,
1190 BTRFS_FILE_EXTENT_REG);
1191 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1192 btrfs_mark_buffer_dirty(leaf);
1194 fi = btrfs_item_ptr(leaf, del_slot - 1,
1195 struct btrfs_file_extent_item);
1196 btrfs_set_file_extent_type(leaf, fi,
1197 BTRFS_FILE_EXTENT_REG);
1198 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1199 btrfs_set_file_extent_num_bytes(leaf, fi,
1200 extent_end - key.offset);
1201 btrfs_mark_buffer_dirty(leaf);
1203 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1205 btrfs_abort_transaction(trans, root, ret);
1210 btrfs_free_path(path);
1215 * on error we return an unlocked page and the error value
1216 * on success we return a locked page and 0
1218 static int prepare_uptodate_page(struct page *page, u64 pos,
1219 bool force_uptodate)
1223 if (((pos & (PAGE_CACHE_SIZE - 1)) || force_uptodate) &&
1224 !PageUptodate(page)) {
1225 ret = btrfs_readpage(NULL, page);
1229 if (!PageUptodate(page)) {
1238 * this just gets pages into the page cache and locks them down.
1240 static noinline int prepare_pages(struct inode *inode, struct page **pages,
1241 size_t num_pages, loff_t pos,
1242 size_t write_bytes, bool force_uptodate)
1245 unsigned long index = pos >> PAGE_CACHE_SHIFT;
1246 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1250 for (i = 0; i < num_pages; i++) {
1251 pages[i] = find_or_create_page(inode->i_mapping, index + i,
1252 mask | __GFP_WRITE);
1260 err = prepare_uptodate_page(pages[i], pos,
1262 if (i == num_pages - 1)
1263 err = prepare_uptodate_page(pages[i],
1264 pos + write_bytes, false);
1266 page_cache_release(pages[i]);
1270 wait_on_page_writeback(pages[i]);
1275 while (faili >= 0) {
1276 unlock_page(pages[faili]);
1277 page_cache_release(pages[faili]);
1285 * This function locks the extent and properly waits for data=ordered extents
1286 * to finish before allowing the pages to be modified if need.
1289 * 1 - the extent is locked
1290 * 0 - the extent is not locked, and everything is OK
1291 * -EAGAIN - need re-prepare the pages
1292 * the other < 0 number - Something wrong happens
1295 lock_and_cleanup_extent_if_need(struct inode *inode, struct page **pages,
1296 size_t num_pages, loff_t pos,
1297 u64 *lockstart, u64 *lockend,
1298 struct extent_state **cached_state)
1305 start_pos = pos & ~((u64)PAGE_CACHE_SIZE - 1);
1306 last_pos = start_pos + ((u64)num_pages << PAGE_CACHE_SHIFT) - 1;
1308 if (start_pos < inode->i_size) {
1309 struct btrfs_ordered_extent *ordered;
1310 lock_extent_bits(&BTRFS_I(inode)->io_tree,
1311 start_pos, last_pos, 0, cached_state);
1312 ordered = btrfs_lookup_first_ordered_extent(inode, last_pos);
1314 ordered->file_offset + ordered->len > start_pos &&
1315 ordered->file_offset <= last_pos) {
1316 btrfs_put_ordered_extent(ordered);
1317 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1318 start_pos, last_pos,
1319 cached_state, GFP_NOFS);
1320 for (i = 0; i < num_pages; i++) {
1321 unlock_page(pages[i]);
1322 page_cache_release(pages[i]);
1324 ret = btrfs_wait_ordered_range(inode, start_pos,
1325 last_pos - start_pos + 1);
1332 btrfs_put_ordered_extent(ordered);
1334 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
1335 last_pos, EXTENT_DIRTY | EXTENT_DELALLOC |
1336 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
1337 0, 0, cached_state, GFP_NOFS);
1338 *lockstart = start_pos;
1339 *lockend = last_pos;
1343 for (i = 0; i < num_pages; i++) {
1344 if (clear_page_dirty_for_io(pages[i]))
1345 account_page_redirty(pages[i]);
1346 set_page_extent_mapped(pages[i]);
1347 WARN_ON(!PageLocked(pages[i]));
1353 static noinline int check_can_nocow(struct inode *inode, loff_t pos,
1354 size_t *write_bytes)
1356 struct btrfs_root *root = BTRFS_I(inode)->root;
1357 struct btrfs_ordered_extent *ordered;
1358 u64 lockstart, lockend;
1362 lockstart = round_down(pos, root->sectorsize);
1363 lockend = lockstart + round_up(*write_bytes, root->sectorsize) - 1;
1366 lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1367 ordered = btrfs_lookup_ordered_range(inode, lockstart,
1368 lockend - lockstart + 1);
1372 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1373 btrfs_start_ordered_extent(inode, ordered, 1);
1374 btrfs_put_ordered_extent(ordered);
1377 num_bytes = lockend - lockstart + 1;
1378 ret = can_nocow_extent(inode, lockstart, &num_bytes, NULL, NULL, NULL);
1382 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
1383 EXTENT_DIRTY | EXTENT_DELALLOC |
1384 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1386 *write_bytes = min_t(size_t, *write_bytes, num_bytes);
1389 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1394 static noinline ssize_t __btrfs_buffered_write(struct file *file,
1398 struct inode *inode = file_inode(file);
1399 struct btrfs_root *root = BTRFS_I(inode)->root;
1400 struct page **pages = NULL;
1401 struct extent_state *cached_state = NULL;
1402 u64 release_bytes = 0;
1405 unsigned long first_index;
1406 size_t num_written = 0;
1409 bool only_release_metadata = false;
1410 bool force_page_uptodate = false;
1413 nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) /
1414 PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
1415 (sizeof(struct page *)));
1416 nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1417 nrptrs = max(nrptrs, 8);
1418 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
1422 first_index = pos >> PAGE_CACHE_SHIFT;
1424 while (iov_iter_count(i) > 0) {
1425 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1426 size_t write_bytes = min(iov_iter_count(i),
1427 nrptrs * (size_t)PAGE_CACHE_SIZE -
1429 size_t num_pages = (write_bytes + offset +
1430 PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1431 size_t reserve_bytes;
1435 WARN_ON(num_pages > nrptrs);
1438 * Fault pages before locking them in prepare_pages
1439 * to avoid recursive lock
1441 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1446 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1447 ret = btrfs_check_data_free_space(inode, reserve_bytes);
1448 if (ret == -ENOSPC &&
1449 (BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW |
1450 BTRFS_INODE_PREALLOC))) {
1451 ret = check_can_nocow(inode, pos, &write_bytes);
1453 only_release_metadata = true;
1455 * our prealloc extent may be smaller than
1456 * write_bytes, so scale down.
1458 num_pages = (write_bytes + offset +
1459 PAGE_CACHE_SIZE - 1) >>
1461 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1471 ret = btrfs_delalloc_reserve_metadata(inode, reserve_bytes);
1473 if (!only_release_metadata)
1474 btrfs_free_reserved_data_space(inode,
1479 release_bytes = reserve_bytes;
1480 need_unlock = false;
1483 * This is going to setup the pages array with the number of
1484 * pages we want, so we don't really need to worry about the
1485 * contents of pages from loop to loop
1487 ret = prepare_pages(inode, pages, num_pages,
1489 force_page_uptodate);
1493 ret = lock_and_cleanup_extent_if_need(inode, pages, num_pages,
1494 pos, &lockstart, &lockend,
1500 } else if (ret > 0) {
1505 copied = btrfs_copy_from_user(pos, num_pages,
1506 write_bytes, pages, i);
1509 * if we have trouble faulting in the pages, fall
1510 * back to one page at a time
1512 if (copied < write_bytes)
1516 force_page_uptodate = true;
1519 force_page_uptodate = false;
1520 dirty_pages = (copied + offset +
1521 PAGE_CACHE_SIZE - 1) >>
1526 * If we had a short copy we need to release the excess delaloc
1527 * bytes we reserved. We need to increment outstanding_extents
1528 * because btrfs_delalloc_release_space will decrement it, but
1529 * we still have an outstanding extent for the chunk we actually
1532 if (num_pages > dirty_pages) {
1533 release_bytes = (num_pages - dirty_pages) <<
1536 spin_lock(&BTRFS_I(inode)->lock);
1537 BTRFS_I(inode)->outstanding_extents++;
1538 spin_unlock(&BTRFS_I(inode)->lock);
1540 if (only_release_metadata)
1541 btrfs_delalloc_release_metadata(inode,
1544 btrfs_delalloc_release_space(inode,
1548 release_bytes = dirty_pages << PAGE_CACHE_SHIFT;
1551 ret = btrfs_dirty_pages(root, inode, pages,
1552 dirty_pages, pos, copied,
1555 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1556 lockstart, lockend, &cached_state,
1558 btrfs_drop_pages(pages, num_pages);
1563 if (only_release_metadata && copied > 0) {
1564 u64 lockstart = round_down(pos, root->sectorsize);
1565 u64 lockend = lockstart +
1566 (dirty_pages << PAGE_CACHE_SHIFT) - 1;
1568 set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
1569 lockend, EXTENT_NORESERVE, NULL,
1571 only_release_metadata = false;
1576 balance_dirty_pages_ratelimited(inode->i_mapping);
1577 if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1578 btrfs_btree_balance_dirty(root);
1581 num_written += copied;
1586 if (release_bytes) {
1587 if (only_release_metadata)
1588 btrfs_delalloc_release_metadata(inode, release_bytes);
1590 btrfs_delalloc_release_space(inode, release_bytes);
1593 return num_written ? num_written : ret;
1596 static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1597 const struct iovec *iov,
1598 unsigned long nr_segs, loff_t pos,
1599 loff_t *ppos, size_t count, size_t ocount)
1601 struct file *file = iocb->ki_filp;
1604 ssize_t written_buffered;
1608 written = generic_file_direct_write(iocb, iov, &nr_segs, pos, ppos,
1611 if (written < 0 || written == count)
1616 iov_iter_init(&i, iov, nr_segs, count, written);
1617 written_buffered = __btrfs_buffered_write(file, &i, pos);
1618 if (written_buffered < 0) {
1619 err = written_buffered;
1622 endbyte = pos + written_buffered - 1;
1623 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
1626 written += written_buffered;
1627 *ppos = pos + written_buffered;
1628 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
1629 endbyte >> PAGE_CACHE_SHIFT);
1631 return written ? written : err;
1634 static void update_time_for_write(struct inode *inode)
1636 struct timespec now;
1638 if (IS_NOCMTIME(inode))
1641 now = current_fs_time(inode->i_sb);
1642 if (!timespec_equal(&inode->i_mtime, &now))
1643 inode->i_mtime = now;
1645 if (!timespec_equal(&inode->i_ctime, &now))
1646 inode->i_ctime = now;
1648 if (IS_I_VERSION(inode))
1649 inode_inc_iversion(inode);
1652 static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
1653 const struct iovec *iov,
1654 unsigned long nr_segs, loff_t pos)
1656 struct file *file = iocb->ki_filp;
1657 struct inode *inode = file_inode(file);
1658 struct btrfs_root *root = BTRFS_I(inode)->root;
1659 loff_t *ppos = &iocb->ki_pos;
1661 ssize_t num_written = 0;
1663 size_t count, ocount;
1664 bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host);
1666 mutex_lock(&inode->i_mutex);
1668 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
1670 mutex_unlock(&inode->i_mutex);
1675 current->backing_dev_info = inode->i_mapping->backing_dev_info;
1676 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1678 mutex_unlock(&inode->i_mutex);
1683 mutex_unlock(&inode->i_mutex);
1687 err = file_remove_suid(file);
1689 mutex_unlock(&inode->i_mutex);
1694 * If BTRFS flips readonly due to some impossible error
1695 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1696 * although we have opened a file as writable, we have
1697 * to stop this write operation to ensure FS consistency.
1699 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
1700 mutex_unlock(&inode->i_mutex);
1706 * We reserve space for updating the inode when we reserve space for the
1707 * extent we are going to write, so we will enospc out there. We don't
1708 * need to start yet another transaction to update the inode as we will
1709 * update the inode when we finish writing whatever data we write.
1711 update_time_for_write(inode);
1713 start_pos = round_down(pos, root->sectorsize);
1714 if (start_pos > i_size_read(inode)) {
1715 err = btrfs_cont_expand(inode, i_size_read(inode), start_pos);
1717 mutex_unlock(&inode->i_mutex);
1723 atomic_inc(&BTRFS_I(inode)->sync_writers);
1725 if (unlikely(file->f_flags & O_DIRECT)) {
1726 num_written = __btrfs_direct_write(iocb, iov, nr_segs,
1727 pos, ppos, count, ocount);
1731 iov_iter_init(&i, iov, nr_segs, count, num_written);
1733 num_written = __btrfs_buffered_write(file, &i, pos);
1734 if (num_written > 0)
1735 *ppos = pos + num_written;
1738 mutex_unlock(&inode->i_mutex);
1741 * we want to make sure fsync finds this change
1742 * but we haven't joined a transaction running right now.
1744 * Later on, someone is sure to update the inode and get the
1745 * real transid recorded.
1747 * We set last_trans now to the fs_info generation + 1,
1748 * this will either be one more than the running transaction
1749 * or the generation used for the next transaction if there isn't
1750 * one running right now.
1752 * We also have to set last_sub_trans to the current log transid,
1753 * otherwise subsequent syncs to a file that's been synced in this
1754 * transaction will appear to have already occured.
1756 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1757 BTRFS_I(inode)->last_sub_trans = root->log_transid;
1758 if (num_written > 0) {
1759 err = generic_write_sync(file, pos, num_written);
1760 if (err < 0 && num_written > 0)
1765 atomic_dec(&BTRFS_I(inode)->sync_writers);
1767 current->backing_dev_info = NULL;
1768 return num_written ? num_written : err;
1771 int btrfs_release_file(struct inode *inode, struct file *filp)
1774 * ordered_data_close is set by settattr when we are about to truncate
1775 * a file from a non-zero size to a zero size. This tries to
1776 * flush down new bytes that may have been written if the
1777 * application were using truncate to replace a file in place.
1779 if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
1780 &BTRFS_I(inode)->runtime_flags)) {
1781 struct btrfs_trans_handle *trans;
1782 struct btrfs_root *root = BTRFS_I(inode)->root;
1785 * We need to block on a committing transaction to keep us from
1786 * throwing a ordered operation on to the list and causing
1787 * something like sync to deadlock trying to flush out this
1790 trans = btrfs_start_transaction(root, 0);
1792 return PTR_ERR(trans);
1793 btrfs_add_ordered_operation(trans, BTRFS_I(inode)->root, inode);
1794 btrfs_end_transaction(trans, root);
1795 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1796 filemap_flush(inode->i_mapping);
1798 if (filp->private_data)
1799 btrfs_ioctl_trans_end(filp);
1804 * fsync call for both files and directories. This logs the inode into
1805 * the tree log instead of forcing full commits whenever possible.
1807 * It needs to call filemap_fdatawait so that all ordered extent updates are
1808 * in the metadata btree are up to date for copying to the log.
1810 * It drops the inode mutex before doing the tree log commit. This is an
1811 * important optimization for directories because holding the mutex prevents
1812 * new operations on the dir while we write to disk.
1814 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1816 struct dentry *dentry = file->f_path.dentry;
1817 struct inode *inode = dentry->d_inode;
1818 struct btrfs_root *root = BTRFS_I(inode)->root;
1820 struct btrfs_trans_handle *trans;
1823 trace_btrfs_sync_file(file, datasync);
1826 * We write the dirty pages in the range and wait until they complete
1827 * out of the ->i_mutex. If so, we can flush the dirty pages by
1828 * multi-task, and make the performance up. See
1829 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1831 atomic_inc(&BTRFS_I(inode)->sync_writers);
1832 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
1833 if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1834 &BTRFS_I(inode)->runtime_flags))
1835 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
1836 atomic_dec(&BTRFS_I(inode)->sync_writers);
1840 mutex_lock(&inode->i_mutex);
1843 * We flush the dirty pages again to avoid some dirty pages in the
1846 atomic_inc(&root->log_batch);
1847 full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1848 &BTRFS_I(inode)->runtime_flags);
1850 ret = btrfs_wait_ordered_range(inode, start, end - start + 1);
1852 mutex_unlock(&inode->i_mutex);
1856 atomic_inc(&root->log_batch);
1859 * check the transaction that last modified this inode
1860 * and see if its already been committed
1862 if (!BTRFS_I(inode)->last_trans) {
1863 mutex_unlock(&inode->i_mutex);
1868 * if the last transaction that changed this file was before
1869 * the current transaction, we can bail out now without any
1873 if (btrfs_inode_in_log(inode, root->fs_info->generation) ||
1874 BTRFS_I(inode)->last_trans <=
1875 root->fs_info->last_trans_committed) {
1876 BTRFS_I(inode)->last_trans = 0;
1879 * We'v had everything committed since the last time we were
1880 * modified so clear this flag in case it was set for whatever
1881 * reason, it's no longer relevant.
1883 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1884 &BTRFS_I(inode)->runtime_flags);
1885 mutex_unlock(&inode->i_mutex);
1890 * ok we haven't committed the transaction yet, lets do a commit
1892 if (file->private_data)
1893 btrfs_ioctl_trans_end(file);
1895 trans = btrfs_start_transaction(root, 0);
1896 if (IS_ERR(trans)) {
1897 ret = PTR_ERR(trans);
1898 mutex_unlock(&inode->i_mutex);
1902 ret = btrfs_log_dentry_safe(trans, root, dentry);
1904 /* Fallthrough and commit/free transaction. */
1908 /* we've logged all the items and now have a consistent
1909 * version of the file in the log. It is possible that
1910 * someone will come in and modify the file, but that's
1911 * fine because the log is consistent on disk, and we
1912 * have references to all of the file's extents
1914 * It is possible that someone will come in and log the
1915 * file again, but that will end up using the synchronization
1916 * inside btrfs_sync_log to keep things safe.
1918 mutex_unlock(&inode->i_mutex);
1920 if (ret != BTRFS_NO_LOG_SYNC) {
1922 ret = btrfs_sync_log(trans, root);
1924 ret = btrfs_end_transaction(trans, root);
1929 ret = btrfs_wait_ordered_range(inode, start,
1934 ret = btrfs_commit_transaction(trans, root);
1936 ret = btrfs_end_transaction(trans, root);
1939 return ret > 0 ? -EIO : ret;
1942 static const struct vm_operations_struct btrfs_file_vm_ops = {
1943 .fault = filemap_fault,
1944 .page_mkwrite = btrfs_page_mkwrite,
1945 .remap_pages = generic_file_remap_pages,
1948 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
1950 struct address_space *mapping = filp->f_mapping;
1952 if (!mapping->a_ops->readpage)
1955 file_accessed(filp);
1956 vma->vm_ops = &btrfs_file_vm_ops;
1961 static int hole_mergeable(struct inode *inode, struct extent_buffer *leaf,
1962 int slot, u64 start, u64 end)
1964 struct btrfs_file_extent_item *fi;
1965 struct btrfs_key key;
1967 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
1970 btrfs_item_key_to_cpu(leaf, &key, slot);
1971 if (key.objectid != btrfs_ino(inode) ||
1972 key.type != BTRFS_EXTENT_DATA_KEY)
1975 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1977 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
1980 if (btrfs_file_extent_disk_bytenr(leaf, fi))
1983 if (key.offset == end)
1985 if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
1990 static int fill_holes(struct btrfs_trans_handle *trans, struct inode *inode,
1991 struct btrfs_path *path, u64 offset, u64 end)
1993 struct btrfs_root *root = BTRFS_I(inode)->root;
1994 struct extent_buffer *leaf;
1995 struct btrfs_file_extent_item *fi;
1996 struct extent_map *hole_em;
1997 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1998 struct btrfs_key key;
2001 if (btrfs_fs_incompat(root->fs_info, NO_HOLES))
2004 key.objectid = btrfs_ino(inode);
2005 key.type = BTRFS_EXTENT_DATA_KEY;
2006 key.offset = offset;
2008 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2013 leaf = path->nodes[0];
2014 if (hole_mergeable(inode, leaf, path->slots[0]-1, offset, end)) {
2018 fi = btrfs_item_ptr(leaf, path->slots[0],
2019 struct btrfs_file_extent_item);
2020 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2022 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2023 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2024 btrfs_set_file_extent_offset(leaf, fi, 0);
2025 btrfs_mark_buffer_dirty(leaf);
2029 if (hole_mergeable(inode, leaf, path->slots[0]+1, offset, end)) {
2033 key.offset = offset;
2034 btrfs_set_item_key_safe(root, path, &key);
2035 fi = btrfs_item_ptr(leaf, path->slots[0],
2036 struct btrfs_file_extent_item);
2037 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2039 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2040 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2041 btrfs_set_file_extent_offset(leaf, fi, 0);
2042 btrfs_mark_buffer_dirty(leaf);
2045 btrfs_release_path(path);
2047 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
2048 0, 0, end - offset, 0, end - offset,
2054 btrfs_release_path(path);
2056 hole_em = alloc_extent_map();
2058 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2059 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2060 &BTRFS_I(inode)->runtime_flags);
2062 hole_em->start = offset;
2063 hole_em->len = end - offset;
2064 hole_em->ram_bytes = hole_em->len;
2065 hole_em->orig_start = offset;
2067 hole_em->block_start = EXTENT_MAP_HOLE;
2068 hole_em->block_len = 0;
2069 hole_em->orig_block_len = 0;
2070 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
2071 hole_em->compress_type = BTRFS_COMPRESS_NONE;
2072 hole_em->generation = trans->transid;
2075 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2076 write_lock(&em_tree->lock);
2077 ret = add_extent_mapping(em_tree, hole_em, 1);
2078 write_unlock(&em_tree->lock);
2079 } while (ret == -EEXIST);
2080 free_extent_map(hole_em);
2082 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2083 &BTRFS_I(inode)->runtime_flags);
2089 static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
2091 struct btrfs_root *root = BTRFS_I(inode)->root;
2092 struct extent_state *cached_state = NULL;
2093 struct btrfs_path *path;
2094 struct btrfs_block_rsv *rsv;
2095 struct btrfs_trans_handle *trans;
2096 u64 lockstart = round_up(offset, BTRFS_I(inode)->root->sectorsize);
2097 u64 lockend = round_down(offset + len,
2098 BTRFS_I(inode)->root->sectorsize) - 1;
2099 u64 cur_offset = lockstart;
2100 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
2105 bool same_page = ((offset >> PAGE_CACHE_SHIFT) ==
2106 ((offset + len - 1) >> PAGE_CACHE_SHIFT));
2107 bool no_holes = btrfs_fs_incompat(root->fs_info, NO_HOLES);
2109 ret = btrfs_wait_ordered_range(inode, offset, len);
2113 mutex_lock(&inode->i_mutex);
2115 * We needn't truncate any page which is beyond the end of the file
2116 * because we are sure there is no data there.
2119 * Only do this if we are in the same page and we aren't doing the
2122 if (same_page && len < PAGE_CACHE_SIZE) {
2123 if (offset < round_up(inode->i_size, PAGE_CACHE_SIZE))
2124 ret = btrfs_truncate_page(inode, offset, len, 0);
2125 mutex_unlock(&inode->i_mutex);
2129 /* zero back part of the first page */
2130 if (offset < round_up(inode->i_size, PAGE_CACHE_SIZE)) {
2131 ret = btrfs_truncate_page(inode, offset, 0, 0);
2133 mutex_unlock(&inode->i_mutex);
2138 /* zero the front end of the last page */
2139 if (offset + len < round_up(inode->i_size, PAGE_CACHE_SIZE)) {
2140 ret = btrfs_truncate_page(inode, offset + len, 0, 1);
2142 mutex_unlock(&inode->i_mutex);
2147 if (lockend < lockstart) {
2148 mutex_unlock(&inode->i_mutex);
2153 struct btrfs_ordered_extent *ordered;
2155 truncate_pagecache_range(inode, lockstart, lockend);
2157 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2159 ordered = btrfs_lookup_first_ordered_extent(inode, lockend);
2162 * We need to make sure we have no ordered extents in this range
2163 * and nobody raced in and read a page in this range, if we did
2164 * we need to try again.
2167 (ordered->file_offset + ordered->len <= lockstart ||
2168 ordered->file_offset > lockend)) &&
2169 !test_range_bit(&BTRFS_I(inode)->io_tree, lockstart,
2170 lockend, EXTENT_UPTODATE, 0,
2173 btrfs_put_ordered_extent(ordered);
2177 btrfs_put_ordered_extent(ordered);
2178 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
2179 lockend, &cached_state, GFP_NOFS);
2180 ret = btrfs_wait_ordered_range(inode, lockstart,
2181 lockend - lockstart + 1);
2183 mutex_unlock(&inode->i_mutex);
2188 path = btrfs_alloc_path();
2194 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2199 rsv->size = btrfs_calc_trunc_metadata_size(root, 1);
2203 * 1 - update the inode
2204 * 1 - removing the extents in the range
2205 * 1 - adding the hole extent if no_holes isn't set
2207 rsv_count = no_holes ? 2 : 3;
2208 trans = btrfs_start_transaction(root, rsv_count);
2209 if (IS_ERR(trans)) {
2210 err = PTR_ERR(trans);
2214 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
2217 trans->block_rsv = rsv;
2219 while (cur_offset < lockend) {
2220 ret = __btrfs_drop_extents(trans, root, inode, path,
2221 cur_offset, lockend + 1,
2226 trans->block_rsv = &root->fs_info->trans_block_rsv;
2228 ret = fill_holes(trans, inode, path, cur_offset, drop_end);
2234 cur_offset = drop_end;
2236 ret = btrfs_update_inode(trans, root, inode);
2242 btrfs_end_transaction(trans, root);
2243 btrfs_btree_balance_dirty(root);
2245 trans = btrfs_start_transaction(root, rsv_count);
2246 if (IS_ERR(trans)) {
2247 ret = PTR_ERR(trans);
2252 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
2254 BUG_ON(ret); /* shouldn't happen */
2255 trans->block_rsv = rsv;
2263 trans->block_rsv = &root->fs_info->trans_block_rsv;
2264 ret = fill_holes(trans, inode, path, cur_offset, drop_end);
2274 inode_inc_iversion(inode);
2275 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2277 trans->block_rsv = &root->fs_info->trans_block_rsv;
2278 ret = btrfs_update_inode(trans, root, inode);
2279 btrfs_end_transaction(trans, root);
2280 btrfs_btree_balance_dirty(root);
2282 btrfs_free_path(path);
2283 btrfs_free_block_rsv(root, rsv);
2285 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2286 &cached_state, GFP_NOFS);
2287 mutex_unlock(&inode->i_mutex);
2293 static long btrfs_fallocate(struct file *file, int mode,
2294 loff_t offset, loff_t len)
2296 struct inode *inode = file_inode(file);
2297 struct extent_state *cached_state = NULL;
2298 struct btrfs_root *root = BTRFS_I(inode)->root;
2305 struct extent_map *em;
2306 int blocksize = BTRFS_I(inode)->root->sectorsize;
2309 alloc_start = round_down(offset, blocksize);
2310 alloc_end = round_up(offset + len, blocksize);
2312 /* Make sure we aren't being give some crap mode */
2313 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2316 if (mode & FALLOC_FL_PUNCH_HOLE)
2317 return btrfs_punch_hole(inode, offset, len);
2320 * Make sure we have enough space before we do the
2323 ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
2326 if (root->fs_info->quota_enabled) {
2327 ret = btrfs_qgroup_reserve(root, alloc_end - alloc_start);
2329 goto out_reserve_fail;
2332 mutex_lock(&inode->i_mutex);
2333 ret = inode_newsize_ok(inode, alloc_end);
2337 if (alloc_start > inode->i_size) {
2338 ret = btrfs_cont_expand(inode, i_size_read(inode),
2344 * If we are fallocating from the end of the file onward we
2345 * need to zero out the end of the page if i_size lands in the
2348 ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
2354 * wait for ordered IO before we have any locks. We'll loop again
2355 * below with the locks held.
2357 ret = btrfs_wait_ordered_range(inode, alloc_start,
2358 alloc_end - alloc_start);
2362 locked_end = alloc_end - 1;
2364 struct btrfs_ordered_extent *ordered;
2366 /* the extent lock is ordered inside the running
2369 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
2370 locked_end, 0, &cached_state);
2371 ordered = btrfs_lookup_first_ordered_extent(inode,
2374 ordered->file_offset + ordered->len > alloc_start &&
2375 ordered->file_offset < alloc_end) {
2376 btrfs_put_ordered_extent(ordered);
2377 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
2378 alloc_start, locked_end,
2379 &cached_state, GFP_NOFS);
2381 * we can't wait on the range with the transaction
2382 * running or with the extent lock held
2384 ret = btrfs_wait_ordered_range(inode, alloc_start,
2385 alloc_end - alloc_start);
2390 btrfs_put_ordered_extent(ordered);
2395 cur_offset = alloc_start;
2399 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2400 alloc_end - cur_offset, 0);
2401 if (IS_ERR_OR_NULL(em)) {
2408 last_byte = min(extent_map_end(em), alloc_end);
2409 actual_end = min_t(u64, extent_map_end(em), offset + len);
2410 last_byte = ALIGN(last_byte, blocksize);
2412 if (em->block_start == EXTENT_MAP_HOLE ||
2413 (cur_offset >= inode->i_size &&
2414 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
2415 ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
2416 last_byte - cur_offset,
2417 1 << inode->i_blkbits,
2422 free_extent_map(em);
2425 } else if (actual_end > inode->i_size &&
2426 !(mode & FALLOC_FL_KEEP_SIZE)) {
2428 * We didn't need to allocate any more space, but we
2429 * still extended the size of the file so we need to
2432 inode->i_ctime = CURRENT_TIME;
2433 i_size_write(inode, actual_end);
2434 btrfs_ordered_update_i_size(inode, actual_end, NULL);
2436 free_extent_map(em);
2438 cur_offset = last_byte;
2439 if (cur_offset >= alloc_end) {
2444 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
2445 &cached_state, GFP_NOFS);
2447 mutex_unlock(&inode->i_mutex);
2448 if (root->fs_info->quota_enabled)
2449 btrfs_qgroup_free(root, alloc_end - alloc_start);
2451 /* Let go of our reservation. */
2452 btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
2456 static int find_desired_extent(struct inode *inode, loff_t *offset, int whence)
2458 struct btrfs_root *root = BTRFS_I(inode)->root;
2459 struct extent_map *em = NULL;
2460 struct extent_state *cached_state = NULL;
2461 u64 lockstart = *offset;
2462 u64 lockend = i_size_read(inode);
2463 u64 start = *offset;
2464 u64 len = i_size_read(inode);
2467 lockend = max_t(u64, root->sectorsize, lockend);
2468 if (lockend <= lockstart)
2469 lockend = lockstart + root->sectorsize;
2472 len = lockend - lockstart + 1;
2474 len = max_t(u64, len, root->sectorsize);
2475 if (inode->i_size == 0)
2478 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, 0,
2481 while (start < inode->i_size) {
2482 em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0);
2489 if (whence == SEEK_HOLE &&
2490 (em->block_start == EXTENT_MAP_HOLE ||
2491 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2493 else if (whence == SEEK_DATA &&
2494 (em->block_start != EXTENT_MAP_HOLE &&
2495 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2498 start = em->start + em->len;
2499 free_extent_map(em);
2503 free_extent_map(em);
2505 if (whence == SEEK_DATA && start >= inode->i_size)
2508 *offset = min_t(loff_t, start, inode->i_size);
2510 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2511 &cached_state, GFP_NOFS);
2515 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
2517 struct inode *inode = file->f_mapping->host;
2520 mutex_lock(&inode->i_mutex);
2524 offset = generic_file_llseek(file, offset, whence);
2528 if (offset >= i_size_read(inode)) {
2529 mutex_unlock(&inode->i_mutex);
2533 ret = find_desired_extent(inode, &offset, whence);
2535 mutex_unlock(&inode->i_mutex);
2540 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2542 mutex_unlock(&inode->i_mutex);
2546 const struct file_operations btrfs_file_operations = {
2547 .llseek = btrfs_file_llseek,
2548 .read = do_sync_read,
2549 .write = do_sync_write,
2550 .aio_read = generic_file_aio_read,
2551 .splice_read = generic_file_splice_read,
2552 .aio_write = btrfs_file_aio_write,
2553 .mmap = btrfs_file_mmap,
2554 .open = generic_file_open,
2555 .release = btrfs_release_file,
2556 .fsync = btrfs_sync_file,
2557 .fallocate = btrfs_fallocate,
2558 .unlocked_ioctl = btrfs_ioctl,
2559 #ifdef CONFIG_COMPAT
2560 .compat_ioctl = btrfs_ioctl,
2564 void btrfs_auto_defrag_exit(void)
2566 if (btrfs_inode_defrag_cachep)
2567 kmem_cache_destroy(btrfs_inode_defrag_cachep);
2570 int btrfs_auto_defrag_init(void)
2572 btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
2573 sizeof(struct inode_defrag), 0,
2574 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2576 if (!btrfs_inode_defrag_cachep)