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 copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
431 /* Flush processor's dcache for this page */
432 flush_dcache_page(page);
435 * if we get a partial write, we can end up with
436 * partially up to date pages. These add
437 * a lot of complexity, so make sure they don't
438 * happen by forcing this copy to be retried.
440 * The rest of the btrfs_file_write code will fall
441 * back to page at a time copies after we return 0.
443 if (!PageUptodate(page) && copied < count)
446 iov_iter_advance(i, copied);
447 write_bytes -= copied;
448 total_copied += copied;
450 /* Return to btrfs_file_aio_write to fault page */
451 if (unlikely(copied == 0))
454 if (unlikely(copied < PAGE_CACHE_SIZE - offset)) {
465 * unlocks pages after btrfs_file_write is done with them
467 static void btrfs_drop_pages(struct page **pages, size_t num_pages)
470 for (i = 0; i < num_pages; i++) {
471 /* page checked is some magic around finding pages that
472 * have been modified without going through btrfs_set_page_dirty
473 * clear it here. There should be no need to mark the pages
474 * accessed as prepare_pages should have marked them accessed
475 * in prepare_pages via find_or_create_page()
477 ClearPageChecked(pages[i]);
478 unlock_page(pages[i]);
479 page_cache_release(pages[i]);
484 * after copy_from_user, pages need to be dirtied and we need to make
485 * sure holes are created between the current EOF and the start of
486 * any next extents (if required).
488 * this also makes the decision about creating an inline extent vs
489 * doing real data extents, marking pages dirty and delalloc as required.
491 int btrfs_dirty_pages(struct btrfs_root *root, struct inode *inode,
492 struct page **pages, size_t num_pages,
493 loff_t pos, size_t write_bytes,
494 struct extent_state **cached)
500 u64 end_of_last_block;
501 u64 end_pos = pos + write_bytes;
502 loff_t isize = i_size_read(inode);
504 start_pos = pos & ~((u64)root->sectorsize - 1);
505 num_bytes = ALIGN(write_bytes + pos - start_pos, root->sectorsize);
507 end_of_last_block = start_pos + num_bytes - 1;
508 err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
513 for (i = 0; i < num_pages; i++) {
514 struct page *p = pages[i];
521 * we've only changed i_size in ram, and we haven't updated
522 * the disk i_size. There is no need to log the inode
526 i_size_write(inode, end_pos);
531 * this drops all the extents in the cache that intersect the range
532 * [start, end]. Existing extents are split as required.
534 void btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
537 struct extent_map *em;
538 struct extent_map *split = NULL;
539 struct extent_map *split2 = NULL;
540 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
541 u64 len = end - start + 1;
549 WARN_ON(end < start);
550 if (end == (u64)-1) {
559 split = alloc_extent_map();
561 split2 = alloc_extent_map();
562 if (!split || !split2)
565 write_lock(&em_tree->lock);
566 em = lookup_extent_mapping(em_tree, start, len);
568 write_unlock(&em_tree->lock);
572 gen = em->generation;
573 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
574 if (testend && em->start + em->len >= start + len) {
576 write_unlock(&em_tree->lock);
579 start = em->start + em->len;
581 len = start + len - (em->start + em->len);
583 write_unlock(&em_tree->lock);
586 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
587 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
588 clear_bit(EXTENT_FLAG_LOGGING, &flags);
589 modified = !list_empty(&em->list);
593 if (em->start < start) {
594 split->start = em->start;
595 split->len = start - em->start;
597 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
598 split->orig_start = em->orig_start;
599 split->block_start = em->block_start;
602 split->block_len = em->block_len;
604 split->block_len = split->len;
605 split->orig_block_len = max(split->block_len,
607 split->ram_bytes = em->ram_bytes;
609 split->orig_start = split->start;
610 split->block_len = 0;
611 split->block_start = em->block_start;
612 split->orig_block_len = 0;
613 split->ram_bytes = split->len;
616 split->generation = gen;
617 split->bdev = em->bdev;
618 split->flags = flags;
619 split->compress_type = em->compress_type;
620 replace_extent_mapping(em_tree, em, split, modified);
621 free_extent_map(split);
625 if (testend && em->start + em->len > start + len) {
626 u64 diff = start + len - em->start;
628 split->start = start + len;
629 split->len = em->start + em->len - (start + len);
630 split->bdev = em->bdev;
631 split->flags = flags;
632 split->compress_type = em->compress_type;
633 split->generation = gen;
635 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
636 split->orig_block_len = max(em->block_len,
639 split->ram_bytes = em->ram_bytes;
641 split->block_len = em->block_len;
642 split->block_start = em->block_start;
643 split->orig_start = em->orig_start;
645 split->block_len = split->len;
646 split->block_start = em->block_start
648 split->orig_start = em->orig_start;
651 split->ram_bytes = split->len;
652 split->orig_start = split->start;
653 split->block_len = 0;
654 split->block_start = em->block_start;
655 split->orig_block_len = 0;
658 if (extent_map_in_tree(em)) {
659 replace_extent_mapping(em_tree, em, split,
662 ret = add_extent_mapping(em_tree, split,
664 ASSERT(ret == 0); /* Logic error */
666 free_extent_map(split);
670 if (extent_map_in_tree(em))
671 remove_extent_mapping(em_tree, em);
672 write_unlock(&em_tree->lock);
676 /* once for the tree*/
680 free_extent_map(split);
682 free_extent_map(split2);
686 * this is very complex, but the basic idea is to drop all extents
687 * in the range start - end. hint_block is filled in with a block number
688 * that would be a good hint to the block allocator for this file.
690 * If an extent intersects the range but is not entirely inside the range
691 * it is either truncated or split. Anything entirely inside the range
692 * is deleted from the tree.
694 int __btrfs_drop_extents(struct btrfs_trans_handle *trans,
695 struct btrfs_root *root, struct inode *inode,
696 struct btrfs_path *path, u64 start, u64 end,
697 u64 *drop_end, int drop_cache,
699 u32 extent_item_size,
702 struct extent_buffer *leaf;
703 struct btrfs_file_extent_item *fi;
704 struct btrfs_key key;
705 struct btrfs_key new_key;
706 u64 ino = btrfs_ino(inode);
707 u64 search_start = start;
710 u64 extent_offset = 0;
717 int modify_tree = -1;
718 int update_refs = (root->ref_cows || root == root->fs_info->tree_root);
720 int leafs_visited = 0;
723 btrfs_drop_extent_cache(inode, start, end - 1, 0);
725 if (start >= BTRFS_I(inode)->disk_i_size && !replace_extent)
730 ret = btrfs_lookup_file_extent(trans, root, path, ino,
731 search_start, modify_tree);
734 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
735 leaf = path->nodes[0];
736 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
737 if (key.objectid == ino &&
738 key.type == BTRFS_EXTENT_DATA_KEY)
744 leaf = path->nodes[0];
745 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
747 ret = btrfs_next_leaf(root, path);
755 leaf = path->nodes[0];
759 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
760 if (key.objectid > ino ||
761 key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
764 fi = btrfs_item_ptr(leaf, path->slots[0],
765 struct btrfs_file_extent_item);
766 extent_type = btrfs_file_extent_type(leaf, fi);
768 if (extent_type == BTRFS_FILE_EXTENT_REG ||
769 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
770 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
771 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
772 extent_offset = btrfs_file_extent_offset(leaf, fi);
773 extent_end = key.offset +
774 btrfs_file_extent_num_bytes(leaf, fi);
775 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
776 extent_end = key.offset +
777 btrfs_file_extent_inline_len(leaf,
781 extent_end = search_start;
784 if (extent_end <= search_start) {
790 search_start = max(key.offset, start);
791 if (recow || !modify_tree) {
793 btrfs_release_path(path);
798 * | - range to drop - |
799 * | -------- extent -------- |
801 if (start > key.offset && end < extent_end) {
803 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
808 memcpy(&new_key, &key, sizeof(new_key));
809 new_key.offset = start;
810 ret = btrfs_duplicate_item(trans, root, path,
812 if (ret == -EAGAIN) {
813 btrfs_release_path(path);
819 leaf = path->nodes[0];
820 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
821 struct btrfs_file_extent_item);
822 btrfs_set_file_extent_num_bytes(leaf, fi,
825 fi = btrfs_item_ptr(leaf, path->slots[0],
826 struct btrfs_file_extent_item);
828 extent_offset += start - key.offset;
829 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
830 btrfs_set_file_extent_num_bytes(leaf, fi,
832 btrfs_mark_buffer_dirty(leaf);
834 if (update_refs && disk_bytenr > 0) {
835 ret = btrfs_inc_extent_ref(trans, root,
836 disk_bytenr, num_bytes, 0,
837 root->root_key.objectid,
839 start - extent_offset, 0);
840 BUG_ON(ret); /* -ENOMEM */
845 * | ---- range to drop ----- |
846 * | -------- extent -------- |
848 if (start <= key.offset && end < extent_end) {
849 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
854 memcpy(&new_key, &key, sizeof(new_key));
855 new_key.offset = end;
856 btrfs_set_item_key_safe(root, path, &new_key);
858 extent_offset += end - key.offset;
859 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
860 btrfs_set_file_extent_num_bytes(leaf, fi,
862 btrfs_mark_buffer_dirty(leaf);
863 if (update_refs && disk_bytenr > 0)
864 inode_sub_bytes(inode, end - key.offset);
868 search_start = extent_end;
870 * | ---- range to drop ----- |
871 * | -------- extent -------- |
873 if (start > key.offset && end >= extent_end) {
875 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
880 btrfs_set_file_extent_num_bytes(leaf, fi,
882 btrfs_mark_buffer_dirty(leaf);
883 if (update_refs && disk_bytenr > 0)
884 inode_sub_bytes(inode, extent_end - start);
885 if (end == extent_end)
893 * | ---- range to drop ----- |
894 * | ------ extent ------ |
896 if (start <= key.offset && end >= extent_end) {
898 del_slot = path->slots[0];
901 BUG_ON(del_slot + del_nr != path->slots[0]);
906 extent_type == BTRFS_FILE_EXTENT_INLINE) {
907 inode_sub_bytes(inode,
908 extent_end - key.offset);
909 extent_end = ALIGN(extent_end,
911 } else if (update_refs && disk_bytenr > 0) {
912 ret = btrfs_free_extent(trans, root,
913 disk_bytenr, num_bytes, 0,
914 root->root_key.objectid,
915 key.objectid, key.offset -
917 BUG_ON(ret); /* -ENOMEM */
918 inode_sub_bytes(inode,
919 extent_end - key.offset);
922 if (end == extent_end)
925 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
930 ret = btrfs_del_items(trans, root, path, del_slot,
933 btrfs_abort_transaction(trans, root, ret);
940 btrfs_release_path(path);
947 if (!ret && del_nr > 0) {
949 * Set path->slots[0] to first slot, so that after the delete
950 * if items are move off from our leaf to its immediate left or
951 * right neighbor leafs, we end up with a correct and adjusted
952 * path->slots[0] for our insertion (if replace_extent != 0).
954 path->slots[0] = del_slot;
955 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
957 btrfs_abort_transaction(trans, root, ret);
960 leaf = path->nodes[0];
962 * If btrfs_del_items() was called, it might have deleted a leaf, in
963 * which case it unlocked our path, so check path->locks[0] matches a
966 if (!ret && replace_extent && leafs_visited == 1 &&
967 (path->locks[0] == BTRFS_WRITE_LOCK_BLOCKING ||
968 path->locks[0] == BTRFS_WRITE_LOCK) &&
969 btrfs_leaf_free_space(root, leaf) >=
970 sizeof(struct btrfs_item) + extent_item_size) {
973 key.type = BTRFS_EXTENT_DATA_KEY;
975 if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) {
976 struct btrfs_key slot_key;
978 btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]);
979 if (btrfs_comp_cpu_keys(&key, &slot_key) > 0)
982 setup_items_for_insert(root, path, &key,
985 sizeof(struct btrfs_item) +
986 extent_item_size, 1);
990 if (!replace_extent || !(*key_inserted))
991 btrfs_release_path(path);
993 *drop_end = found ? min(end, extent_end) : end;
997 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
998 struct btrfs_root *root, struct inode *inode, u64 start,
999 u64 end, int drop_cache)
1001 struct btrfs_path *path;
1004 path = btrfs_alloc_path();
1007 ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL,
1008 drop_cache, 0, 0, NULL);
1009 btrfs_free_path(path);
1013 static int extent_mergeable(struct extent_buffer *leaf, int slot,
1014 u64 objectid, u64 bytenr, u64 orig_offset,
1015 u64 *start, u64 *end)
1017 struct btrfs_file_extent_item *fi;
1018 struct btrfs_key key;
1021 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
1024 btrfs_item_key_to_cpu(leaf, &key, slot);
1025 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
1028 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1029 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
1030 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
1031 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
1032 btrfs_file_extent_compression(leaf, fi) ||
1033 btrfs_file_extent_encryption(leaf, fi) ||
1034 btrfs_file_extent_other_encoding(leaf, fi))
1037 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1038 if ((*start && *start != key.offset) || (*end && *end != extent_end))
1041 *start = key.offset;
1047 * Mark extent in the range start - end as written.
1049 * This changes extent type from 'pre-allocated' to 'regular'. If only
1050 * part of extent is marked as written, the extent will be split into
1053 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
1054 struct inode *inode, u64 start, u64 end)
1056 struct btrfs_root *root = BTRFS_I(inode)->root;
1057 struct extent_buffer *leaf;
1058 struct btrfs_path *path;
1059 struct btrfs_file_extent_item *fi;
1060 struct btrfs_key key;
1061 struct btrfs_key new_key;
1073 u64 ino = btrfs_ino(inode);
1075 path = btrfs_alloc_path();
1082 key.type = BTRFS_EXTENT_DATA_KEY;
1085 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1088 if (ret > 0 && path->slots[0] > 0)
1091 leaf = path->nodes[0];
1092 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1093 BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY);
1094 fi = btrfs_item_ptr(leaf, path->slots[0],
1095 struct btrfs_file_extent_item);
1096 BUG_ON(btrfs_file_extent_type(leaf, fi) !=
1097 BTRFS_FILE_EXTENT_PREALLOC);
1098 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1099 BUG_ON(key.offset > start || extent_end < end);
1101 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1102 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1103 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
1104 memcpy(&new_key, &key, sizeof(new_key));
1106 if (start == key.offset && end < extent_end) {
1109 if (extent_mergeable(leaf, path->slots[0] - 1,
1110 ino, bytenr, orig_offset,
1111 &other_start, &other_end)) {
1112 new_key.offset = end;
1113 btrfs_set_item_key_safe(root, path, &new_key);
1114 fi = btrfs_item_ptr(leaf, path->slots[0],
1115 struct btrfs_file_extent_item);
1116 btrfs_set_file_extent_generation(leaf, fi,
1118 btrfs_set_file_extent_num_bytes(leaf, fi,
1120 btrfs_set_file_extent_offset(leaf, fi,
1122 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1123 struct btrfs_file_extent_item);
1124 btrfs_set_file_extent_generation(leaf, fi,
1126 btrfs_set_file_extent_num_bytes(leaf, fi,
1128 btrfs_mark_buffer_dirty(leaf);
1133 if (start > key.offset && end == extent_end) {
1136 if (extent_mergeable(leaf, path->slots[0] + 1,
1137 ino, bytenr, orig_offset,
1138 &other_start, &other_end)) {
1139 fi = btrfs_item_ptr(leaf, path->slots[0],
1140 struct btrfs_file_extent_item);
1141 btrfs_set_file_extent_num_bytes(leaf, fi,
1142 start - key.offset);
1143 btrfs_set_file_extent_generation(leaf, fi,
1146 new_key.offset = start;
1147 btrfs_set_item_key_safe(root, path, &new_key);
1149 fi = btrfs_item_ptr(leaf, path->slots[0],
1150 struct btrfs_file_extent_item);
1151 btrfs_set_file_extent_generation(leaf, fi,
1153 btrfs_set_file_extent_num_bytes(leaf, fi,
1155 btrfs_set_file_extent_offset(leaf, fi,
1156 start - orig_offset);
1157 btrfs_mark_buffer_dirty(leaf);
1162 while (start > key.offset || end < extent_end) {
1163 if (key.offset == start)
1166 new_key.offset = split;
1167 ret = btrfs_duplicate_item(trans, root, path, &new_key);
1168 if (ret == -EAGAIN) {
1169 btrfs_release_path(path);
1173 btrfs_abort_transaction(trans, root, ret);
1177 leaf = path->nodes[0];
1178 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1179 struct btrfs_file_extent_item);
1180 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1181 btrfs_set_file_extent_num_bytes(leaf, fi,
1182 split - key.offset);
1184 fi = btrfs_item_ptr(leaf, path->slots[0],
1185 struct btrfs_file_extent_item);
1187 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1188 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
1189 btrfs_set_file_extent_num_bytes(leaf, fi,
1190 extent_end - split);
1191 btrfs_mark_buffer_dirty(leaf);
1193 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
1194 root->root_key.objectid,
1195 ino, orig_offset, 0);
1196 BUG_ON(ret); /* -ENOMEM */
1198 if (split == start) {
1201 BUG_ON(start != key.offset);
1210 if (extent_mergeable(leaf, path->slots[0] + 1,
1211 ino, bytenr, orig_offset,
1212 &other_start, &other_end)) {
1214 btrfs_release_path(path);
1217 extent_end = other_end;
1218 del_slot = path->slots[0] + 1;
1220 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1221 0, root->root_key.objectid,
1222 ino, orig_offset, 0);
1223 BUG_ON(ret); /* -ENOMEM */
1227 if (extent_mergeable(leaf, path->slots[0] - 1,
1228 ino, bytenr, orig_offset,
1229 &other_start, &other_end)) {
1231 btrfs_release_path(path);
1234 key.offset = other_start;
1235 del_slot = path->slots[0];
1237 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1238 0, root->root_key.objectid,
1239 ino, orig_offset, 0);
1240 BUG_ON(ret); /* -ENOMEM */
1243 fi = btrfs_item_ptr(leaf, path->slots[0],
1244 struct btrfs_file_extent_item);
1245 btrfs_set_file_extent_type(leaf, fi,
1246 BTRFS_FILE_EXTENT_REG);
1247 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1248 btrfs_mark_buffer_dirty(leaf);
1250 fi = btrfs_item_ptr(leaf, del_slot - 1,
1251 struct btrfs_file_extent_item);
1252 btrfs_set_file_extent_type(leaf, fi,
1253 BTRFS_FILE_EXTENT_REG);
1254 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1255 btrfs_set_file_extent_num_bytes(leaf, fi,
1256 extent_end - key.offset);
1257 btrfs_mark_buffer_dirty(leaf);
1259 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1261 btrfs_abort_transaction(trans, root, ret);
1266 btrfs_free_path(path);
1271 * on error we return an unlocked page and the error value
1272 * on success we return a locked page and 0
1274 static int prepare_uptodate_page(struct page *page, u64 pos,
1275 bool force_uptodate)
1279 if (((pos & (PAGE_CACHE_SIZE - 1)) || force_uptodate) &&
1280 !PageUptodate(page)) {
1281 ret = btrfs_readpage(NULL, page);
1285 if (!PageUptodate(page)) {
1294 * this just gets pages into the page cache and locks them down.
1296 static noinline int prepare_pages(struct inode *inode, struct page **pages,
1297 size_t num_pages, loff_t pos,
1298 size_t write_bytes, bool force_uptodate)
1301 unsigned long index = pos >> PAGE_CACHE_SHIFT;
1302 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1306 for (i = 0; i < num_pages; i++) {
1307 pages[i] = find_or_create_page(inode->i_mapping, index + i,
1308 mask | __GFP_WRITE);
1316 err = prepare_uptodate_page(pages[i], pos,
1318 if (i == num_pages - 1)
1319 err = prepare_uptodate_page(pages[i],
1320 pos + write_bytes, false);
1322 page_cache_release(pages[i]);
1326 wait_on_page_writeback(pages[i]);
1331 while (faili >= 0) {
1332 unlock_page(pages[faili]);
1333 page_cache_release(pages[faili]);
1341 * This function locks the extent and properly waits for data=ordered extents
1342 * to finish before allowing the pages to be modified if need.
1345 * 1 - the extent is locked
1346 * 0 - the extent is not locked, and everything is OK
1347 * -EAGAIN - need re-prepare the pages
1348 * the other < 0 number - Something wrong happens
1351 lock_and_cleanup_extent_if_need(struct inode *inode, struct page **pages,
1352 size_t num_pages, loff_t pos,
1353 u64 *lockstart, u64 *lockend,
1354 struct extent_state **cached_state)
1361 start_pos = pos & ~((u64)PAGE_CACHE_SIZE - 1);
1362 last_pos = start_pos + ((u64)num_pages << PAGE_CACHE_SHIFT) - 1;
1364 if (start_pos < inode->i_size) {
1365 struct btrfs_ordered_extent *ordered;
1366 lock_extent_bits(&BTRFS_I(inode)->io_tree,
1367 start_pos, last_pos, 0, cached_state);
1368 ordered = btrfs_lookup_ordered_range(inode, start_pos,
1369 last_pos - start_pos + 1);
1371 ordered->file_offset + ordered->len > start_pos &&
1372 ordered->file_offset <= last_pos) {
1373 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1374 start_pos, last_pos,
1375 cached_state, GFP_NOFS);
1376 for (i = 0; i < num_pages; i++) {
1377 unlock_page(pages[i]);
1378 page_cache_release(pages[i]);
1380 btrfs_start_ordered_extent(inode, ordered, 1);
1381 btrfs_put_ordered_extent(ordered);
1385 btrfs_put_ordered_extent(ordered);
1387 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
1388 last_pos, EXTENT_DIRTY | EXTENT_DELALLOC |
1389 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
1390 0, 0, cached_state, GFP_NOFS);
1391 *lockstart = start_pos;
1392 *lockend = last_pos;
1396 for (i = 0; i < num_pages; i++) {
1397 if (clear_page_dirty_for_io(pages[i]))
1398 account_page_redirty(pages[i]);
1399 set_page_extent_mapped(pages[i]);
1400 WARN_ON(!PageLocked(pages[i]));
1406 static noinline int check_can_nocow(struct inode *inode, loff_t pos,
1407 size_t *write_bytes)
1409 struct btrfs_root *root = BTRFS_I(inode)->root;
1410 struct btrfs_ordered_extent *ordered;
1411 u64 lockstart, lockend;
1415 ret = btrfs_start_nocow_write(root);
1419 lockstart = round_down(pos, root->sectorsize);
1420 lockend = round_up(pos + *write_bytes, root->sectorsize) - 1;
1423 lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1424 ordered = btrfs_lookup_ordered_range(inode, lockstart,
1425 lockend - lockstart + 1);
1429 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1430 btrfs_start_ordered_extent(inode, ordered, 1);
1431 btrfs_put_ordered_extent(ordered);
1434 num_bytes = lockend - lockstart + 1;
1435 ret = can_nocow_extent(inode, lockstart, &num_bytes, NULL, NULL, NULL);
1438 btrfs_end_nocow_write(root);
1440 *write_bytes = min_t(size_t, *write_bytes ,
1441 num_bytes - pos + lockstart);
1444 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1449 static noinline ssize_t __btrfs_buffered_write(struct file *file,
1453 struct inode *inode = file_inode(file);
1454 struct btrfs_root *root = BTRFS_I(inode)->root;
1455 struct page **pages = NULL;
1456 struct extent_state *cached_state = NULL;
1457 u64 release_bytes = 0;
1460 unsigned long first_index;
1461 size_t num_written = 0;
1464 bool only_release_metadata = false;
1465 bool force_page_uptodate = false;
1468 nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) /
1469 PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
1470 (sizeof(struct page *)));
1471 nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1472 nrptrs = max(nrptrs, 8);
1473 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
1477 first_index = pos >> PAGE_CACHE_SHIFT;
1479 while (iov_iter_count(i) > 0) {
1480 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1481 size_t write_bytes = min(iov_iter_count(i),
1482 nrptrs * (size_t)PAGE_CACHE_SIZE -
1484 size_t num_pages = (write_bytes + offset +
1485 PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1486 size_t reserve_bytes;
1490 WARN_ON(num_pages > nrptrs);
1493 * Fault pages before locking them in prepare_pages
1494 * to avoid recursive lock
1496 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1501 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1502 ret = btrfs_check_data_free_space(inode, reserve_bytes);
1503 if (ret == -ENOSPC &&
1504 (BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW |
1505 BTRFS_INODE_PREALLOC))) {
1506 ret = check_can_nocow(inode, pos, &write_bytes);
1508 only_release_metadata = true;
1510 * our prealloc extent may be smaller than
1511 * write_bytes, so scale down.
1513 num_pages = (write_bytes + offset +
1514 PAGE_CACHE_SIZE - 1) >>
1516 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1526 ret = btrfs_delalloc_reserve_metadata(inode, reserve_bytes);
1528 if (!only_release_metadata)
1529 btrfs_free_reserved_data_space(inode,
1532 btrfs_end_nocow_write(root);
1536 release_bytes = reserve_bytes;
1537 need_unlock = false;
1540 * This is going to setup the pages array with the number of
1541 * pages we want, so we don't really need to worry about the
1542 * contents of pages from loop to loop
1544 ret = prepare_pages(inode, pages, num_pages,
1546 force_page_uptodate);
1550 ret = lock_and_cleanup_extent_if_need(inode, pages, num_pages,
1551 pos, &lockstart, &lockend,
1557 } else if (ret > 0) {
1562 copied = btrfs_copy_from_user(pos, num_pages,
1563 write_bytes, pages, i);
1566 * if we have trouble faulting in the pages, fall
1567 * back to one page at a time
1569 if (copied < write_bytes)
1573 force_page_uptodate = true;
1576 force_page_uptodate = false;
1577 dirty_pages = (copied + offset +
1578 PAGE_CACHE_SIZE - 1) >>
1583 * If we had a short copy we need to release the excess delaloc
1584 * bytes we reserved. We need to increment outstanding_extents
1585 * because btrfs_delalloc_release_space will decrement it, but
1586 * we still have an outstanding extent for the chunk we actually
1589 if (num_pages > dirty_pages) {
1590 release_bytes = (num_pages - dirty_pages) <<
1593 spin_lock(&BTRFS_I(inode)->lock);
1594 BTRFS_I(inode)->outstanding_extents++;
1595 spin_unlock(&BTRFS_I(inode)->lock);
1597 if (only_release_metadata)
1598 btrfs_delalloc_release_metadata(inode,
1601 btrfs_delalloc_release_space(inode,
1605 release_bytes = dirty_pages << PAGE_CACHE_SHIFT;
1608 ret = btrfs_dirty_pages(root, inode, pages,
1609 dirty_pages, pos, copied,
1612 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1613 lockstart, lockend, &cached_state,
1616 btrfs_drop_pages(pages, num_pages);
1621 if (only_release_metadata)
1622 btrfs_end_nocow_write(root);
1624 if (only_release_metadata && copied > 0) {
1625 u64 lockstart = round_down(pos, root->sectorsize);
1626 u64 lockend = lockstart +
1627 (dirty_pages << PAGE_CACHE_SHIFT) - 1;
1629 set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
1630 lockend, EXTENT_NORESERVE, NULL,
1632 only_release_metadata = false;
1635 btrfs_drop_pages(pages, num_pages);
1639 balance_dirty_pages_ratelimited(inode->i_mapping);
1640 if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1641 btrfs_btree_balance_dirty(root);
1644 num_written += copied;
1649 if (release_bytes) {
1650 if (only_release_metadata) {
1651 btrfs_end_nocow_write(root);
1652 btrfs_delalloc_release_metadata(inode, release_bytes);
1654 btrfs_delalloc_release_space(inode, release_bytes);
1658 return num_written ? num_written : ret;
1661 static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1662 const struct iovec *iov,
1663 unsigned long nr_segs, loff_t pos,
1664 size_t count, size_t ocount)
1666 struct file *file = iocb->ki_filp;
1669 ssize_t written_buffered;
1673 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
1676 if (written < 0 || written == count)
1681 iov_iter_init(&i, iov, nr_segs, count, written);
1682 written_buffered = __btrfs_buffered_write(file, &i, pos);
1683 if (written_buffered < 0) {
1684 err = written_buffered;
1687 endbyte = pos + written_buffered - 1;
1688 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
1691 written += written_buffered;
1692 iocb->ki_pos = pos + written_buffered;
1693 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
1694 endbyte >> PAGE_CACHE_SHIFT);
1696 return written ? written : err;
1699 static void update_time_for_write(struct inode *inode)
1701 struct timespec now;
1703 if (IS_NOCMTIME(inode))
1706 now = current_fs_time(inode->i_sb);
1707 if (!timespec_equal(&inode->i_mtime, &now))
1708 inode->i_mtime = now;
1710 if (!timespec_equal(&inode->i_ctime, &now))
1711 inode->i_ctime = now;
1713 if (IS_I_VERSION(inode))
1714 inode_inc_iversion(inode);
1717 static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
1718 const struct iovec *iov,
1719 unsigned long nr_segs, loff_t pos)
1721 struct file *file = iocb->ki_filp;
1722 struct inode *inode = file_inode(file);
1723 struct btrfs_root *root = BTRFS_I(inode)->root;
1726 ssize_t num_written = 0;
1728 size_t count, ocount;
1729 bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host);
1731 mutex_lock(&inode->i_mutex);
1733 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
1735 mutex_unlock(&inode->i_mutex);
1740 current->backing_dev_info = inode->i_mapping->backing_dev_info;
1741 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1743 mutex_unlock(&inode->i_mutex);
1748 mutex_unlock(&inode->i_mutex);
1752 err = file_remove_suid(file);
1754 mutex_unlock(&inode->i_mutex);
1759 * If BTRFS flips readonly due to some impossible error
1760 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1761 * although we have opened a file as writable, we have
1762 * to stop this write operation to ensure FS consistency.
1764 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
1765 mutex_unlock(&inode->i_mutex);
1771 * We reserve space for updating the inode when we reserve space for the
1772 * extent we are going to write, so we will enospc out there. We don't
1773 * need to start yet another transaction to update the inode as we will
1774 * update the inode when we finish writing whatever data we write.
1776 update_time_for_write(inode);
1778 start_pos = round_down(pos, root->sectorsize);
1779 if (start_pos > i_size_read(inode)) {
1780 /* Expand hole size to cover write data, preventing empty gap */
1781 end_pos = round_up(pos + count, root->sectorsize);
1782 err = btrfs_cont_expand(inode, i_size_read(inode), end_pos);
1784 mutex_unlock(&inode->i_mutex);
1790 atomic_inc(&BTRFS_I(inode)->sync_writers);
1792 if (unlikely(file->f_flags & O_DIRECT)) {
1793 num_written = __btrfs_direct_write(iocb, iov, nr_segs,
1794 pos, count, ocount);
1798 iov_iter_init(&i, iov, nr_segs, count, num_written);
1800 num_written = __btrfs_buffered_write(file, &i, pos);
1801 if (num_written > 0)
1802 iocb->ki_pos = pos + num_written;
1805 mutex_unlock(&inode->i_mutex);
1808 * we want to make sure fsync finds this change
1809 * but we haven't joined a transaction running right now.
1811 * Later on, someone is sure to update the inode and get the
1812 * real transid recorded.
1814 * We set last_trans now to the fs_info generation + 1,
1815 * this will either be one more than the running transaction
1816 * or the generation used for the next transaction if there isn't
1817 * one running right now.
1819 * We also have to set last_sub_trans to the current log transid,
1820 * otherwise subsequent syncs to a file that's been synced in this
1821 * transaction will appear to have already occured.
1823 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1824 BTRFS_I(inode)->last_sub_trans = root->log_transid;
1825 if (num_written > 0) {
1826 err = generic_write_sync(file, pos, num_written);
1832 atomic_dec(&BTRFS_I(inode)->sync_writers);
1834 current->backing_dev_info = NULL;
1835 return num_written ? num_written : err;
1838 int btrfs_release_file(struct inode *inode, struct file *filp)
1841 * ordered_data_close is set by settattr when we are about to truncate
1842 * a file from a non-zero size to a zero size. This tries to
1843 * flush down new bytes that may have been written if the
1844 * application were using truncate to replace a file in place.
1846 if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
1847 &BTRFS_I(inode)->runtime_flags)) {
1848 struct btrfs_trans_handle *trans;
1849 struct btrfs_root *root = BTRFS_I(inode)->root;
1852 * We need to block on a committing transaction to keep us from
1853 * throwing a ordered operation on to the list and causing
1854 * something like sync to deadlock trying to flush out this
1857 trans = btrfs_start_transaction(root, 0);
1859 return PTR_ERR(trans);
1860 btrfs_add_ordered_operation(trans, BTRFS_I(inode)->root, inode);
1861 btrfs_end_transaction(trans, root);
1862 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1863 filemap_flush(inode->i_mapping);
1865 if (filp->private_data)
1866 btrfs_ioctl_trans_end(filp);
1871 * fsync call for both files and directories. This logs the inode into
1872 * the tree log instead of forcing full commits whenever possible.
1874 * It needs to call filemap_fdatawait so that all ordered extent updates are
1875 * in the metadata btree are up to date for copying to the log.
1877 * It drops the inode mutex before doing the tree log commit. This is an
1878 * important optimization for directories because holding the mutex prevents
1879 * new operations on the dir while we write to disk.
1881 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1883 struct dentry *dentry = file->f_path.dentry;
1884 struct inode *inode = dentry->d_inode;
1885 struct btrfs_root *root = BTRFS_I(inode)->root;
1886 struct btrfs_trans_handle *trans;
1887 struct btrfs_log_ctx ctx;
1891 trace_btrfs_sync_file(file, datasync);
1894 * We write the dirty pages in the range and wait until they complete
1895 * out of the ->i_mutex. If so, we can flush the dirty pages by
1896 * multi-task, and make the performance up. See
1897 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1899 atomic_inc(&BTRFS_I(inode)->sync_writers);
1900 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
1901 if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1902 &BTRFS_I(inode)->runtime_flags))
1903 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
1904 atomic_dec(&BTRFS_I(inode)->sync_writers);
1908 mutex_lock(&inode->i_mutex);
1911 * We flush the dirty pages again to avoid some dirty pages in the
1914 atomic_inc(&root->log_batch);
1915 full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1916 &BTRFS_I(inode)->runtime_flags);
1918 ret = btrfs_wait_ordered_range(inode, start, end - start + 1);
1920 mutex_unlock(&inode->i_mutex);
1924 atomic_inc(&root->log_batch);
1927 * check the transaction that last modified this inode
1928 * and see if its already been committed
1930 if (!BTRFS_I(inode)->last_trans) {
1931 mutex_unlock(&inode->i_mutex);
1936 * if the last transaction that changed this file was before
1937 * the current transaction, we can bail out now without any
1941 if (btrfs_inode_in_log(inode, root->fs_info->generation) ||
1942 BTRFS_I(inode)->last_trans <=
1943 root->fs_info->last_trans_committed) {
1944 BTRFS_I(inode)->last_trans = 0;
1947 * We'v had everything committed since the last time we were
1948 * modified so clear this flag in case it was set for whatever
1949 * reason, it's no longer relevant.
1951 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1952 &BTRFS_I(inode)->runtime_flags);
1953 mutex_unlock(&inode->i_mutex);
1958 * ok we haven't committed the transaction yet, lets do a commit
1960 if (file->private_data)
1961 btrfs_ioctl_trans_end(file);
1964 * We use start here because we will need to wait on the IO to complete
1965 * in btrfs_sync_log, which could require joining a transaction (for
1966 * example checking cross references in the nocow path). If we use join
1967 * here we could get into a situation where we're waiting on IO to
1968 * happen that is blocked on a transaction trying to commit. With start
1969 * we inc the extwriter counter, so we wait for all extwriters to exit
1970 * before we start blocking join'ers. This comment is to keep somebody
1971 * from thinking they are super smart and changing this to
1972 * btrfs_join_transaction *cough*Josef*cough*.
1974 trans = btrfs_start_transaction(root, 0);
1975 if (IS_ERR(trans)) {
1976 ret = PTR_ERR(trans);
1977 mutex_unlock(&inode->i_mutex);
1982 btrfs_init_log_ctx(&ctx);
1984 ret = btrfs_log_dentry_safe(trans, root, dentry, &ctx);
1986 /* Fallthrough and commit/free transaction. */
1990 /* we've logged all the items and now have a consistent
1991 * version of the file in the log. It is possible that
1992 * someone will come in and modify the file, but that's
1993 * fine because the log is consistent on disk, and we
1994 * have references to all of the file's extents
1996 * It is possible that someone will come in and log the
1997 * file again, but that will end up using the synchronization
1998 * inside btrfs_sync_log to keep things safe.
2000 mutex_unlock(&inode->i_mutex);
2002 if (ret != BTRFS_NO_LOG_SYNC) {
2004 ret = btrfs_sync_log(trans, root, &ctx);
2006 ret = btrfs_end_transaction(trans, root);
2011 ret = btrfs_wait_ordered_range(inode, start,
2016 ret = btrfs_commit_transaction(trans, root);
2018 ret = btrfs_end_transaction(trans, root);
2021 return ret > 0 ? -EIO : ret;
2024 static const struct vm_operations_struct btrfs_file_vm_ops = {
2025 .fault = filemap_fault,
2026 .map_pages = filemap_map_pages,
2027 .page_mkwrite = btrfs_page_mkwrite,
2028 .remap_pages = generic_file_remap_pages,
2031 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
2033 struct address_space *mapping = filp->f_mapping;
2035 if (!mapping->a_ops->readpage)
2038 file_accessed(filp);
2039 vma->vm_ops = &btrfs_file_vm_ops;
2044 static int hole_mergeable(struct inode *inode, struct extent_buffer *leaf,
2045 int slot, u64 start, u64 end)
2047 struct btrfs_file_extent_item *fi;
2048 struct btrfs_key key;
2050 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
2053 btrfs_item_key_to_cpu(leaf, &key, slot);
2054 if (key.objectid != btrfs_ino(inode) ||
2055 key.type != BTRFS_EXTENT_DATA_KEY)
2058 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2060 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2063 if (btrfs_file_extent_disk_bytenr(leaf, fi))
2066 if (key.offset == end)
2068 if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
2073 static int fill_holes(struct btrfs_trans_handle *trans, struct inode *inode,
2074 struct btrfs_path *path, u64 offset, u64 end)
2076 struct btrfs_root *root = BTRFS_I(inode)->root;
2077 struct extent_buffer *leaf;
2078 struct btrfs_file_extent_item *fi;
2079 struct extent_map *hole_em;
2080 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2081 struct btrfs_key key;
2084 if (btrfs_fs_incompat(root->fs_info, NO_HOLES))
2087 key.objectid = btrfs_ino(inode);
2088 key.type = BTRFS_EXTENT_DATA_KEY;
2089 key.offset = offset;
2091 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2096 leaf = path->nodes[0];
2097 if (hole_mergeable(inode, leaf, path->slots[0]-1, offset, end)) {
2101 fi = btrfs_item_ptr(leaf, path->slots[0],
2102 struct btrfs_file_extent_item);
2103 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2105 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2106 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2107 btrfs_set_file_extent_offset(leaf, fi, 0);
2108 btrfs_mark_buffer_dirty(leaf);
2112 if (hole_mergeable(inode, leaf, path->slots[0]+1, offset, end)) {
2116 key.offset = offset;
2117 btrfs_set_item_key_safe(root, path, &key);
2118 fi = btrfs_item_ptr(leaf, path->slots[0],
2119 struct btrfs_file_extent_item);
2120 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2122 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2123 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2124 btrfs_set_file_extent_offset(leaf, fi, 0);
2125 btrfs_mark_buffer_dirty(leaf);
2128 btrfs_release_path(path);
2130 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
2131 0, 0, end - offset, 0, end - offset,
2137 btrfs_release_path(path);
2139 hole_em = alloc_extent_map();
2141 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2142 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2143 &BTRFS_I(inode)->runtime_flags);
2145 hole_em->start = offset;
2146 hole_em->len = end - offset;
2147 hole_em->ram_bytes = hole_em->len;
2148 hole_em->orig_start = offset;
2150 hole_em->block_start = EXTENT_MAP_HOLE;
2151 hole_em->block_len = 0;
2152 hole_em->orig_block_len = 0;
2153 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
2154 hole_em->compress_type = BTRFS_COMPRESS_NONE;
2155 hole_em->generation = trans->transid;
2158 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2159 write_lock(&em_tree->lock);
2160 ret = add_extent_mapping(em_tree, hole_em, 1);
2161 write_unlock(&em_tree->lock);
2162 } while (ret == -EEXIST);
2163 free_extent_map(hole_em);
2165 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2166 &BTRFS_I(inode)->runtime_flags);
2172 static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
2174 struct btrfs_root *root = BTRFS_I(inode)->root;
2175 struct extent_state *cached_state = NULL;
2176 struct btrfs_path *path;
2177 struct btrfs_block_rsv *rsv;
2178 struct btrfs_trans_handle *trans;
2179 u64 lockstart = round_up(offset, BTRFS_I(inode)->root->sectorsize);
2180 u64 lockend = round_down(offset + len,
2181 BTRFS_I(inode)->root->sectorsize) - 1;
2182 u64 cur_offset = lockstart;
2183 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
2188 bool same_page = ((offset >> PAGE_CACHE_SHIFT) ==
2189 ((offset + len - 1) >> PAGE_CACHE_SHIFT));
2190 bool no_holes = btrfs_fs_incompat(root->fs_info, NO_HOLES);
2191 u64 ino_size = round_up(inode->i_size, PAGE_CACHE_SIZE);
2193 ret = btrfs_wait_ordered_range(inode, offset, len);
2197 mutex_lock(&inode->i_mutex);
2199 * We needn't truncate any page which is beyond the end of the file
2200 * because we are sure there is no data there.
2203 * Only do this if we are in the same page and we aren't doing the
2206 if (same_page && len < PAGE_CACHE_SIZE) {
2207 if (offset < ino_size)
2208 ret = btrfs_truncate_page(inode, offset, len, 0);
2209 mutex_unlock(&inode->i_mutex);
2213 /* zero back part of the first page */
2214 if (offset < ino_size) {
2215 ret = btrfs_truncate_page(inode, offset, 0, 0);
2217 mutex_unlock(&inode->i_mutex);
2222 /* zero the front end of the last page */
2223 if (offset + len < ino_size) {
2224 ret = btrfs_truncate_page(inode, offset + len, 0, 1);
2226 mutex_unlock(&inode->i_mutex);
2231 if (lockend < lockstart) {
2232 mutex_unlock(&inode->i_mutex);
2237 struct btrfs_ordered_extent *ordered;
2239 truncate_pagecache_range(inode, lockstart, lockend);
2241 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2243 ordered = btrfs_lookup_first_ordered_extent(inode, lockend);
2246 * We need to make sure we have no ordered extents in this range
2247 * and nobody raced in and read a page in this range, if we did
2248 * we need to try again.
2251 (ordered->file_offset + ordered->len <= lockstart ||
2252 ordered->file_offset > lockend)) &&
2253 !test_range_bit(&BTRFS_I(inode)->io_tree, lockstart,
2254 lockend, EXTENT_UPTODATE, 0,
2257 btrfs_put_ordered_extent(ordered);
2261 btrfs_put_ordered_extent(ordered);
2262 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
2263 lockend, &cached_state, GFP_NOFS);
2264 ret = btrfs_wait_ordered_range(inode, lockstart,
2265 lockend - lockstart + 1);
2267 mutex_unlock(&inode->i_mutex);
2272 path = btrfs_alloc_path();
2278 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2283 rsv->size = btrfs_calc_trunc_metadata_size(root, 1);
2287 * 1 - update the inode
2288 * 1 - removing the extents in the range
2289 * 1 - adding the hole extent if no_holes isn't set
2291 rsv_count = no_holes ? 2 : 3;
2292 trans = btrfs_start_transaction(root, rsv_count);
2293 if (IS_ERR(trans)) {
2294 err = PTR_ERR(trans);
2298 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
2301 trans->block_rsv = rsv;
2303 while (cur_offset < lockend) {
2304 ret = __btrfs_drop_extents(trans, root, inode, path,
2305 cur_offset, lockend + 1,
2306 &drop_end, 1, 0, 0, NULL);
2310 trans->block_rsv = &root->fs_info->trans_block_rsv;
2312 if (cur_offset < ino_size) {
2313 ret = fill_holes(trans, inode, path, cur_offset,
2321 cur_offset = drop_end;
2323 ret = btrfs_update_inode(trans, root, inode);
2329 btrfs_end_transaction(trans, root);
2330 btrfs_btree_balance_dirty(root);
2332 trans = btrfs_start_transaction(root, rsv_count);
2333 if (IS_ERR(trans)) {
2334 ret = PTR_ERR(trans);
2339 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
2341 BUG_ON(ret); /* shouldn't happen */
2342 trans->block_rsv = rsv;
2350 trans->block_rsv = &root->fs_info->trans_block_rsv;
2351 if (cur_offset < ino_size) {
2352 ret = fill_holes(trans, inode, path, cur_offset, drop_end);
2363 inode_inc_iversion(inode);
2364 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2366 trans->block_rsv = &root->fs_info->trans_block_rsv;
2367 ret = btrfs_update_inode(trans, root, inode);
2368 btrfs_end_transaction(trans, root);
2369 btrfs_btree_balance_dirty(root);
2371 btrfs_free_path(path);
2372 btrfs_free_block_rsv(root, rsv);
2374 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2375 &cached_state, GFP_NOFS);
2376 mutex_unlock(&inode->i_mutex);
2382 static long btrfs_fallocate(struct file *file, int mode,
2383 loff_t offset, loff_t len)
2385 struct inode *inode = file_inode(file);
2386 struct extent_state *cached_state = NULL;
2387 struct btrfs_root *root = BTRFS_I(inode)->root;
2394 struct extent_map *em;
2395 int blocksize = BTRFS_I(inode)->root->sectorsize;
2398 alloc_start = round_down(offset, blocksize);
2399 alloc_end = round_up(offset + len, blocksize);
2401 /* Make sure we aren't being give some crap mode */
2402 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2405 if (mode & FALLOC_FL_PUNCH_HOLE)
2406 return btrfs_punch_hole(inode, offset, len);
2409 * Make sure we have enough space before we do the
2412 ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
2415 if (root->fs_info->quota_enabled) {
2416 ret = btrfs_qgroup_reserve(root, alloc_end - alloc_start);
2418 goto out_reserve_fail;
2421 mutex_lock(&inode->i_mutex);
2422 ret = inode_newsize_ok(inode, alloc_end);
2426 if (alloc_start > inode->i_size) {
2427 ret = btrfs_cont_expand(inode, i_size_read(inode),
2433 * If we are fallocating from the end of the file onward we
2434 * need to zero out the end of the page if i_size lands in the
2437 ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
2443 * wait for ordered IO before we have any locks. We'll loop again
2444 * below with the locks held.
2446 ret = btrfs_wait_ordered_range(inode, alloc_start,
2447 alloc_end - alloc_start);
2451 locked_end = alloc_end - 1;
2453 struct btrfs_ordered_extent *ordered;
2455 /* the extent lock is ordered inside the running
2458 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
2459 locked_end, 0, &cached_state);
2460 ordered = btrfs_lookup_first_ordered_extent(inode,
2463 ordered->file_offset + ordered->len > alloc_start &&
2464 ordered->file_offset < alloc_end) {
2465 btrfs_put_ordered_extent(ordered);
2466 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
2467 alloc_start, locked_end,
2468 &cached_state, GFP_NOFS);
2470 * we can't wait on the range with the transaction
2471 * running or with the extent lock held
2473 ret = btrfs_wait_ordered_range(inode, alloc_start,
2474 alloc_end - alloc_start);
2479 btrfs_put_ordered_extent(ordered);
2484 cur_offset = alloc_start;
2488 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2489 alloc_end - cur_offset, 0);
2490 if (IS_ERR_OR_NULL(em)) {
2497 last_byte = min(extent_map_end(em), alloc_end);
2498 actual_end = min_t(u64, extent_map_end(em), offset + len);
2499 last_byte = ALIGN(last_byte, blocksize);
2501 if (em->block_start == EXTENT_MAP_HOLE ||
2502 (cur_offset >= inode->i_size &&
2503 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
2504 ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
2505 last_byte - cur_offset,
2506 1 << inode->i_blkbits,
2511 free_extent_map(em);
2514 } else if (actual_end > inode->i_size &&
2515 !(mode & FALLOC_FL_KEEP_SIZE)) {
2517 * We didn't need to allocate any more space, but we
2518 * still extended the size of the file so we need to
2521 inode->i_ctime = CURRENT_TIME;
2522 i_size_write(inode, actual_end);
2523 btrfs_ordered_update_i_size(inode, actual_end, NULL);
2525 free_extent_map(em);
2527 cur_offset = last_byte;
2528 if (cur_offset >= alloc_end) {
2533 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
2534 &cached_state, GFP_NOFS);
2536 mutex_unlock(&inode->i_mutex);
2537 if (root->fs_info->quota_enabled)
2538 btrfs_qgroup_free(root, alloc_end - alloc_start);
2540 /* Let go of our reservation. */
2541 btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
2545 static int find_desired_extent(struct inode *inode, loff_t *offset, int whence)
2547 struct btrfs_root *root = BTRFS_I(inode)->root;
2548 struct extent_map *em = NULL;
2549 struct extent_state *cached_state = NULL;
2550 u64 lockstart = *offset;
2551 u64 lockend = i_size_read(inode);
2552 u64 start = *offset;
2553 u64 len = i_size_read(inode);
2556 lockend = max_t(u64, root->sectorsize, lockend);
2557 if (lockend <= lockstart)
2558 lockend = lockstart + root->sectorsize;
2561 len = lockend - lockstart + 1;
2563 len = max_t(u64, len, root->sectorsize);
2564 if (inode->i_size == 0)
2567 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, 0,
2570 while (start < inode->i_size) {
2571 em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0);
2578 if (whence == SEEK_HOLE &&
2579 (em->block_start == EXTENT_MAP_HOLE ||
2580 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2582 else if (whence == SEEK_DATA &&
2583 (em->block_start != EXTENT_MAP_HOLE &&
2584 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2587 start = em->start + em->len;
2588 free_extent_map(em);
2592 free_extent_map(em);
2594 if (whence == SEEK_DATA && start >= inode->i_size)
2597 *offset = min_t(loff_t, start, inode->i_size);
2599 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2600 &cached_state, GFP_NOFS);
2604 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
2606 struct inode *inode = file->f_mapping->host;
2609 mutex_lock(&inode->i_mutex);
2613 offset = generic_file_llseek(file, offset, whence);
2617 if (offset >= i_size_read(inode)) {
2618 mutex_unlock(&inode->i_mutex);
2622 ret = find_desired_extent(inode, &offset, whence);
2624 mutex_unlock(&inode->i_mutex);
2629 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2631 mutex_unlock(&inode->i_mutex);
2635 const struct file_operations btrfs_file_operations = {
2636 .llseek = btrfs_file_llseek,
2637 .read = do_sync_read,
2638 .write = do_sync_write,
2639 .aio_read = generic_file_aio_read,
2640 .splice_read = generic_file_splice_read,
2641 .aio_write = btrfs_file_aio_write,
2642 .mmap = btrfs_file_mmap,
2643 .open = generic_file_open,
2644 .release = btrfs_release_file,
2645 .fsync = btrfs_sync_file,
2646 .fallocate = btrfs_fallocate,
2647 .unlocked_ioctl = btrfs_ioctl,
2648 #ifdef CONFIG_COMPAT
2649 .compat_ioctl = btrfs_ioctl,
2653 void btrfs_auto_defrag_exit(void)
2655 if (btrfs_inode_defrag_cachep)
2656 kmem_cache_destroy(btrfs_inode_defrag_cachep);
2659 int btrfs_auto_defrag_init(void)
2661 btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
2662 sizeof(struct inode_defrag), 0,
2663 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2665 if (!btrfs_inode_defrag_cachep)