2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 #include <linux/pagemap.h>
21 #include <linux/highmem.h>
22 #include <linux/time.h>
23 #include <linux/init.h>
24 #include <linux/string.h>
25 #include <linux/backing-dev.h>
26 #include <linux/mpage.h>
27 #include <linux/falloc.h>
28 #include <linux/swap.h>
29 #include <linux/writeback.h>
30 #include <linux/statfs.h>
31 #include <linux/compat.h>
32 #include <linux/slab.h>
33 #include <linux/btrfs.h>
34 #include <linux/uio.h>
37 #include "transaction.h"
38 #include "btrfs_inode.h"
39 #include "print-tree.h"
45 static struct kmem_cache *btrfs_inode_defrag_cachep;
47 * when auto defrag is enabled we
48 * queue up these defrag structs to remember which
49 * inodes need defragging passes
52 struct rb_node rb_node;
56 * transid where the defrag was added, we search for
57 * extents newer than this
64 /* last offset we were able to defrag */
67 /* if we've wrapped around back to zero once already */
71 static int __compare_inode_defrag(struct inode_defrag *defrag1,
72 struct inode_defrag *defrag2)
74 if (defrag1->root > defrag2->root)
76 else if (defrag1->root < defrag2->root)
78 else if (defrag1->ino > defrag2->ino)
80 else if (defrag1->ino < defrag2->ino)
86 /* pop a record for an inode into the defrag tree. The lock
87 * must be held already
89 * If you're inserting a record for an older transid than an
90 * existing record, the transid already in the tree is lowered
92 * If an existing record is found the defrag item you
95 static int __btrfs_add_inode_defrag(struct inode *inode,
96 struct inode_defrag *defrag)
98 struct btrfs_root *root = BTRFS_I(inode)->root;
99 struct inode_defrag *entry;
101 struct rb_node *parent = NULL;
104 p = &root->fs_info->defrag_inodes.rb_node;
107 entry = rb_entry(parent, struct inode_defrag, rb_node);
109 ret = __compare_inode_defrag(defrag, entry);
111 p = &parent->rb_left;
113 p = &parent->rb_right;
115 /* if we're reinserting an entry for
116 * an old defrag run, make sure to
117 * lower the transid of our existing record
119 if (defrag->transid < entry->transid)
120 entry->transid = defrag->transid;
121 if (defrag->last_offset > entry->last_offset)
122 entry->last_offset = defrag->last_offset;
126 set_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
127 rb_link_node(&defrag->rb_node, parent, p);
128 rb_insert_color(&defrag->rb_node, &root->fs_info->defrag_inodes);
132 static inline int __need_auto_defrag(struct btrfs_root *root)
134 if (!btrfs_test_opt(root, AUTO_DEFRAG))
137 if (btrfs_fs_closing(root->fs_info))
144 * insert a defrag record for this inode if auto defrag is
147 int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans,
150 struct btrfs_root *root = BTRFS_I(inode)->root;
151 struct inode_defrag *defrag;
155 if (!__need_auto_defrag(root))
158 if (test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags))
162 transid = trans->transid;
164 transid = BTRFS_I(inode)->root->last_trans;
166 defrag = kmem_cache_zalloc(btrfs_inode_defrag_cachep, GFP_NOFS);
170 defrag->ino = btrfs_ino(inode);
171 defrag->transid = transid;
172 defrag->root = root->root_key.objectid;
174 spin_lock(&root->fs_info->defrag_inodes_lock);
175 if (!test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags)) {
177 * If we set IN_DEFRAG flag and evict the inode from memory,
178 * and then re-read this inode, this new inode doesn't have
179 * IN_DEFRAG flag. At the case, we may find the existed defrag.
181 ret = __btrfs_add_inode_defrag(inode, defrag);
183 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
185 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
187 spin_unlock(&root->fs_info->defrag_inodes_lock);
192 * Requeue the defrag object. If there is a defrag object that points to
193 * the same inode in the tree, we will merge them together (by
194 * __btrfs_add_inode_defrag()) and free the one that we want to requeue.
196 static void btrfs_requeue_inode_defrag(struct inode *inode,
197 struct inode_defrag *defrag)
199 struct btrfs_root *root = BTRFS_I(inode)->root;
202 if (!__need_auto_defrag(root))
206 * Here we don't check the IN_DEFRAG flag, because we need merge
209 spin_lock(&root->fs_info->defrag_inodes_lock);
210 ret = __btrfs_add_inode_defrag(inode, defrag);
211 spin_unlock(&root->fs_info->defrag_inodes_lock);
216 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
220 * pick the defragable inode that we want, if it doesn't exist, we will get
223 static struct inode_defrag *
224 btrfs_pick_defrag_inode(struct btrfs_fs_info *fs_info, u64 root, u64 ino)
226 struct inode_defrag *entry = NULL;
227 struct inode_defrag tmp;
229 struct rb_node *parent = NULL;
235 spin_lock(&fs_info->defrag_inodes_lock);
236 p = fs_info->defrag_inodes.rb_node;
239 entry = rb_entry(parent, struct inode_defrag, rb_node);
241 ret = __compare_inode_defrag(&tmp, entry);
245 p = parent->rb_right;
250 if (parent && __compare_inode_defrag(&tmp, entry) > 0) {
251 parent = rb_next(parent);
253 entry = rb_entry(parent, struct inode_defrag, rb_node);
259 rb_erase(parent, &fs_info->defrag_inodes);
260 spin_unlock(&fs_info->defrag_inodes_lock);
264 void btrfs_cleanup_defrag_inodes(struct btrfs_fs_info *fs_info)
266 struct inode_defrag *defrag;
267 struct rb_node *node;
269 spin_lock(&fs_info->defrag_inodes_lock);
270 node = rb_first(&fs_info->defrag_inodes);
272 rb_erase(node, &fs_info->defrag_inodes);
273 defrag = rb_entry(node, struct inode_defrag, rb_node);
274 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
276 cond_resched_lock(&fs_info->defrag_inodes_lock);
278 node = rb_first(&fs_info->defrag_inodes);
280 spin_unlock(&fs_info->defrag_inodes_lock);
283 #define BTRFS_DEFRAG_BATCH 1024
285 static int __btrfs_run_defrag_inode(struct btrfs_fs_info *fs_info,
286 struct inode_defrag *defrag)
288 struct btrfs_root *inode_root;
290 struct btrfs_key key;
291 struct btrfs_ioctl_defrag_range_args range;
297 key.objectid = defrag->root;
298 key.type = BTRFS_ROOT_ITEM_KEY;
299 key.offset = (u64)-1;
301 index = srcu_read_lock(&fs_info->subvol_srcu);
303 inode_root = btrfs_read_fs_root_no_name(fs_info, &key);
304 if (IS_ERR(inode_root)) {
305 ret = PTR_ERR(inode_root);
309 key.objectid = defrag->ino;
310 key.type = BTRFS_INODE_ITEM_KEY;
312 inode = btrfs_iget(fs_info->sb, &key, inode_root, NULL);
314 ret = PTR_ERR(inode);
317 srcu_read_unlock(&fs_info->subvol_srcu, index);
319 /* do a chunk of defrag */
320 clear_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
321 memset(&range, 0, sizeof(range));
323 range.start = defrag->last_offset;
325 sb_start_write(fs_info->sb);
326 num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid,
328 sb_end_write(fs_info->sb);
330 * if we filled the whole defrag batch, there
331 * must be more work to do. Queue this defrag
334 if (num_defrag == BTRFS_DEFRAG_BATCH) {
335 defrag->last_offset = range.start;
336 btrfs_requeue_inode_defrag(inode, defrag);
337 } else if (defrag->last_offset && !defrag->cycled) {
339 * we didn't fill our defrag batch, but
340 * we didn't start at zero. Make sure we loop
341 * around to the start of the file.
343 defrag->last_offset = 0;
345 btrfs_requeue_inode_defrag(inode, defrag);
347 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
353 srcu_read_unlock(&fs_info->subvol_srcu, index);
354 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
359 * run through the list of inodes in the FS that need
362 int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info)
364 struct inode_defrag *defrag;
366 u64 root_objectid = 0;
368 atomic_inc(&fs_info->defrag_running);
370 /* Pause the auto defragger. */
371 if (test_bit(BTRFS_FS_STATE_REMOUNTING,
375 if (!__need_auto_defrag(fs_info->tree_root))
378 /* find an inode to defrag */
379 defrag = btrfs_pick_defrag_inode(fs_info, root_objectid,
382 if (root_objectid || first_ino) {
391 first_ino = defrag->ino + 1;
392 root_objectid = defrag->root;
394 __btrfs_run_defrag_inode(fs_info, defrag);
396 atomic_dec(&fs_info->defrag_running);
399 * during unmount, we use the transaction_wait queue to
400 * wait for the defragger to stop
402 wake_up(&fs_info->transaction_wait);
406 /* simple helper to fault in pages and copy. This should go away
407 * and be replaced with calls into generic code.
409 static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
411 struct page **prepared_pages,
415 size_t total_copied = 0;
417 int offset = pos & (PAGE_CACHE_SIZE - 1);
419 while (write_bytes > 0) {
420 size_t count = min_t(size_t,
421 PAGE_CACHE_SIZE - offset, write_bytes);
422 struct page *page = prepared_pages[pg];
424 * Copy data from userspace to the current page
426 copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
428 /* Flush processor's dcache for this page */
429 flush_dcache_page(page);
432 * if we get a partial write, we can end up with
433 * partially up to date pages. These add
434 * a lot of complexity, so make sure they don't
435 * happen by forcing this copy to be retried.
437 * The rest of the btrfs_file_write code will fall
438 * back to page at a time copies after we return 0.
440 if (!PageUptodate(page) && copied < count)
443 iov_iter_advance(i, copied);
444 write_bytes -= copied;
445 total_copied += copied;
447 /* Return to btrfs_file_write_iter to fault page */
448 if (unlikely(copied == 0))
451 if (copied < PAGE_CACHE_SIZE - offset) {
462 * unlocks pages after btrfs_file_write is done with them
464 static void btrfs_drop_pages(struct page **pages, size_t num_pages)
467 for (i = 0; i < num_pages; i++) {
468 /* page checked is some magic around finding pages that
469 * have been modified without going through btrfs_set_page_dirty
470 * clear it here. There should be no need to mark the pages
471 * accessed as prepare_pages should have marked them accessed
472 * in prepare_pages via find_or_create_page()
474 ClearPageChecked(pages[i]);
475 unlock_page(pages[i]);
476 page_cache_release(pages[i]);
481 * after copy_from_user, pages need to be dirtied and we need to make
482 * sure holes are created between the current EOF and the start of
483 * any next extents (if required).
485 * this also makes the decision about creating an inline extent vs
486 * doing real data extents, marking pages dirty and delalloc as required.
488 int btrfs_dirty_pages(struct btrfs_root *root, struct inode *inode,
489 struct page **pages, size_t num_pages,
490 loff_t pos, size_t write_bytes,
491 struct extent_state **cached)
497 u64 end_of_last_block;
498 u64 end_pos = pos + write_bytes;
499 loff_t isize = i_size_read(inode);
501 start_pos = pos & ~((u64)root->sectorsize - 1);
502 num_bytes = ALIGN(write_bytes + pos - start_pos, root->sectorsize);
504 end_of_last_block = start_pos + num_bytes - 1;
505 err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
510 for (i = 0; i < num_pages; i++) {
511 struct page *p = pages[i];
518 * we've only changed i_size in ram, and we haven't updated
519 * the disk i_size. There is no need to log the inode
523 i_size_write(inode, end_pos);
528 * this drops all the extents in the cache that intersect the range
529 * [start, end]. Existing extents are split as required.
531 void btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
534 struct extent_map *em;
535 struct extent_map *split = NULL;
536 struct extent_map *split2 = NULL;
537 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
538 u64 len = end - start + 1;
546 WARN_ON(end < start);
547 if (end == (u64)-1) {
556 split = alloc_extent_map();
558 split2 = alloc_extent_map();
559 if (!split || !split2)
562 write_lock(&em_tree->lock);
563 em = lookup_extent_mapping(em_tree, start, len);
565 write_unlock(&em_tree->lock);
569 gen = em->generation;
570 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
571 if (testend && em->start + em->len >= start + len) {
573 write_unlock(&em_tree->lock);
576 start = em->start + em->len;
578 len = start + len - (em->start + em->len);
580 write_unlock(&em_tree->lock);
583 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
584 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
585 clear_bit(EXTENT_FLAG_LOGGING, &flags);
586 modified = !list_empty(&em->list);
590 if (em->start < start) {
591 split->start = em->start;
592 split->len = start - em->start;
594 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
595 split->orig_start = em->orig_start;
596 split->block_start = em->block_start;
599 split->block_len = em->block_len;
601 split->block_len = split->len;
602 split->orig_block_len = max(split->block_len,
604 split->ram_bytes = em->ram_bytes;
606 split->orig_start = split->start;
607 split->block_len = 0;
608 split->block_start = em->block_start;
609 split->orig_block_len = 0;
610 split->ram_bytes = split->len;
613 split->generation = gen;
614 split->bdev = em->bdev;
615 split->flags = flags;
616 split->compress_type = em->compress_type;
617 replace_extent_mapping(em_tree, em, split, modified);
618 free_extent_map(split);
622 if (testend && em->start + em->len > start + len) {
623 u64 diff = start + len - em->start;
625 split->start = start + len;
626 split->len = em->start + em->len - (start + len);
627 split->bdev = em->bdev;
628 split->flags = flags;
629 split->compress_type = em->compress_type;
630 split->generation = gen;
632 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
633 split->orig_block_len = max(em->block_len,
636 split->ram_bytes = em->ram_bytes;
638 split->block_len = em->block_len;
639 split->block_start = em->block_start;
640 split->orig_start = em->orig_start;
642 split->block_len = split->len;
643 split->block_start = em->block_start
645 split->orig_start = em->orig_start;
648 split->ram_bytes = split->len;
649 split->orig_start = split->start;
650 split->block_len = 0;
651 split->block_start = em->block_start;
652 split->orig_block_len = 0;
655 if (extent_map_in_tree(em)) {
656 replace_extent_mapping(em_tree, em, split,
659 ret = add_extent_mapping(em_tree, split,
661 ASSERT(ret == 0); /* Logic error */
663 free_extent_map(split);
667 if (extent_map_in_tree(em))
668 remove_extent_mapping(em_tree, em);
669 write_unlock(&em_tree->lock);
673 /* once for the tree*/
677 free_extent_map(split);
679 free_extent_map(split2);
683 * this is very complex, but the basic idea is to drop all extents
684 * in the range start - end. hint_block is filled in with a block number
685 * that would be a good hint to the block allocator for this file.
687 * If an extent intersects the range but is not entirely inside the range
688 * it is either truncated or split. Anything entirely inside the range
689 * is deleted from the tree.
691 int __btrfs_drop_extents(struct btrfs_trans_handle *trans,
692 struct btrfs_root *root, struct inode *inode,
693 struct btrfs_path *path, u64 start, u64 end,
694 u64 *drop_end, int drop_cache,
696 u32 extent_item_size,
699 struct extent_buffer *leaf;
700 struct btrfs_file_extent_item *fi;
701 struct btrfs_key key;
702 struct btrfs_key new_key;
703 u64 ino = btrfs_ino(inode);
704 u64 search_start = start;
707 u64 extent_offset = 0;
714 int modify_tree = -1;
717 int leafs_visited = 0;
720 btrfs_drop_extent_cache(inode, start, end - 1, 0);
722 if (start >= BTRFS_I(inode)->disk_i_size && !replace_extent)
725 update_refs = (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
726 root == root->fs_info->tree_root);
729 ret = btrfs_lookup_file_extent(trans, root, path, ino,
730 search_start, modify_tree);
733 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
734 leaf = path->nodes[0];
735 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
736 if (key.objectid == ino &&
737 key.type == BTRFS_EXTENT_DATA_KEY)
743 leaf = path->nodes[0];
744 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
746 ret = btrfs_next_leaf(root, path);
754 leaf = path->nodes[0];
758 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
759 if (key.objectid > ino ||
760 key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
763 fi = btrfs_item_ptr(leaf, path->slots[0],
764 struct btrfs_file_extent_item);
765 extent_type = btrfs_file_extent_type(leaf, fi);
767 if (extent_type == BTRFS_FILE_EXTENT_REG ||
768 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
769 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
770 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
771 extent_offset = btrfs_file_extent_offset(leaf, fi);
772 extent_end = key.offset +
773 btrfs_file_extent_num_bytes(leaf, fi);
774 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
775 extent_end = key.offset +
776 btrfs_file_extent_inline_len(leaf,
780 extent_end = search_start;
784 * Don't skip extent items representing 0 byte lengths. They
785 * used to be created (bug) if while punching holes we hit
786 * -ENOSPC condition. So if we find one here, just ensure we
787 * delete it, otherwise we would insert a new file extent item
788 * with the same key (offset) as that 0 bytes length file
789 * extent item in the call to setup_items_for_insert() later
792 if (extent_end == key.offset && extent_end >= search_start)
793 goto delete_extent_item;
795 if (extent_end <= search_start) {
801 search_start = max(key.offset, start);
802 if (recow || !modify_tree) {
804 btrfs_release_path(path);
809 * | - range to drop - |
810 * | -------- extent -------- |
812 if (start > key.offset && end < extent_end) {
814 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
819 memcpy(&new_key, &key, sizeof(new_key));
820 new_key.offset = start;
821 ret = btrfs_duplicate_item(trans, root, path,
823 if (ret == -EAGAIN) {
824 btrfs_release_path(path);
830 leaf = path->nodes[0];
831 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
832 struct btrfs_file_extent_item);
833 btrfs_set_file_extent_num_bytes(leaf, fi,
836 fi = btrfs_item_ptr(leaf, path->slots[0],
837 struct btrfs_file_extent_item);
839 extent_offset += start - key.offset;
840 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
841 btrfs_set_file_extent_num_bytes(leaf, fi,
843 btrfs_mark_buffer_dirty(leaf);
845 if (update_refs && disk_bytenr > 0) {
846 ret = btrfs_inc_extent_ref(trans, root,
847 disk_bytenr, num_bytes, 0,
848 root->root_key.objectid,
850 start - extent_offset);
851 BUG_ON(ret); /* -ENOMEM */
856 * | ---- range to drop ----- |
857 * | -------- extent -------- |
859 if (start <= key.offset && end < extent_end) {
860 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
865 memcpy(&new_key, &key, sizeof(new_key));
866 new_key.offset = end;
867 btrfs_set_item_key_safe(root->fs_info, path, &new_key);
869 extent_offset += end - key.offset;
870 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
871 btrfs_set_file_extent_num_bytes(leaf, fi,
873 btrfs_mark_buffer_dirty(leaf);
874 if (update_refs && disk_bytenr > 0)
875 inode_sub_bytes(inode, end - key.offset);
879 search_start = extent_end;
881 * | ---- range to drop ----- |
882 * | -------- extent -------- |
884 if (start > key.offset && end >= extent_end) {
886 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
891 btrfs_set_file_extent_num_bytes(leaf, fi,
893 btrfs_mark_buffer_dirty(leaf);
894 if (update_refs && disk_bytenr > 0)
895 inode_sub_bytes(inode, extent_end - start);
896 if (end == extent_end)
904 * | ---- range to drop ----- |
905 * | ------ extent ------ |
907 if (start <= key.offset && end >= extent_end) {
910 del_slot = path->slots[0];
913 BUG_ON(del_slot + del_nr != path->slots[0]);
918 extent_type == BTRFS_FILE_EXTENT_INLINE) {
919 inode_sub_bytes(inode,
920 extent_end - key.offset);
921 extent_end = ALIGN(extent_end,
923 } else if (update_refs && disk_bytenr > 0) {
924 ret = btrfs_free_extent(trans, root,
925 disk_bytenr, num_bytes, 0,
926 root->root_key.objectid,
927 key.objectid, key.offset -
929 BUG_ON(ret); /* -ENOMEM */
930 inode_sub_bytes(inode,
931 extent_end - key.offset);
934 if (end == extent_end)
937 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
942 ret = btrfs_del_items(trans, root, path, del_slot,
945 btrfs_abort_transaction(trans, root, ret);
952 btrfs_release_path(path);
959 if (!ret && del_nr > 0) {
961 * Set path->slots[0] to first slot, so that after the delete
962 * if items are move off from our leaf to its immediate left or
963 * right neighbor leafs, we end up with a correct and adjusted
964 * path->slots[0] for our insertion (if replace_extent != 0).
966 path->slots[0] = del_slot;
967 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
969 btrfs_abort_transaction(trans, root, ret);
972 leaf = path->nodes[0];
974 * If btrfs_del_items() was called, it might have deleted a leaf, in
975 * which case it unlocked our path, so check path->locks[0] matches a
978 if (!ret && replace_extent && leafs_visited == 1 &&
979 (path->locks[0] == BTRFS_WRITE_LOCK_BLOCKING ||
980 path->locks[0] == BTRFS_WRITE_LOCK) &&
981 btrfs_leaf_free_space(root, leaf) >=
982 sizeof(struct btrfs_item) + extent_item_size) {
985 key.type = BTRFS_EXTENT_DATA_KEY;
987 if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) {
988 struct btrfs_key slot_key;
990 btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]);
991 if (btrfs_comp_cpu_keys(&key, &slot_key) > 0)
994 setup_items_for_insert(root, path, &key,
997 sizeof(struct btrfs_item) +
998 extent_item_size, 1);
1002 if (!replace_extent || !(*key_inserted))
1003 btrfs_release_path(path);
1005 *drop_end = found ? min(end, extent_end) : end;
1009 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
1010 struct btrfs_root *root, struct inode *inode, u64 start,
1011 u64 end, int drop_cache)
1013 struct btrfs_path *path;
1016 path = btrfs_alloc_path();
1019 ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL,
1020 drop_cache, 0, 0, NULL);
1021 btrfs_free_path(path);
1025 static int extent_mergeable(struct extent_buffer *leaf, int slot,
1026 u64 objectid, u64 bytenr, u64 orig_offset,
1027 u64 *start, u64 *end)
1029 struct btrfs_file_extent_item *fi;
1030 struct btrfs_key key;
1033 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
1036 btrfs_item_key_to_cpu(leaf, &key, slot);
1037 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
1040 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1041 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
1042 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
1043 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
1044 btrfs_file_extent_compression(leaf, fi) ||
1045 btrfs_file_extent_encryption(leaf, fi) ||
1046 btrfs_file_extent_other_encoding(leaf, fi))
1049 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1050 if ((*start && *start != key.offset) || (*end && *end != extent_end))
1053 *start = key.offset;
1059 * Mark extent in the range start - end as written.
1061 * This changes extent type from 'pre-allocated' to 'regular'. If only
1062 * part of extent is marked as written, the extent will be split into
1065 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
1066 struct inode *inode, u64 start, u64 end)
1068 struct btrfs_root *root = BTRFS_I(inode)->root;
1069 struct extent_buffer *leaf;
1070 struct btrfs_path *path;
1071 struct btrfs_file_extent_item *fi;
1072 struct btrfs_key key;
1073 struct btrfs_key new_key;
1085 u64 ino = btrfs_ino(inode);
1087 path = btrfs_alloc_path();
1094 key.type = BTRFS_EXTENT_DATA_KEY;
1097 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1100 if (ret > 0 && path->slots[0] > 0)
1103 leaf = path->nodes[0];
1104 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1105 BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY);
1106 fi = btrfs_item_ptr(leaf, path->slots[0],
1107 struct btrfs_file_extent_item);
1108 BUG_ON(btrfs_file_extent_type(leaf, fi) !=
1109 BTRFS_FILE_EXTENT_PREALLOC);
1110 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1111 BUG_ON(key.offset > start || extent_end < end);
1113 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1114 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1115 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
1116 memcpy(&new_key, &key, sizeof(new_key));
1118 if (start == key.offset && end < extent_end) {
1121 if (extent_mergeable(leaf, path->slots[0] - 1,
1122 ino, bytenr, orig_offset,
1123 &other_start, &other_end)) {
1124 new_key.offset = end;
1125 btrfs_set_item_key_safe(root->fs_info, path, &new_key);
1126 fi = btrfs_item_ptr(leaf, path->slots[0],
1127 struct btrfs_file_extent_item);
1128 btrfs_set_file_extent_generation(leaf, fi,
1130 btrfs_set_file_extent_num_bytes(leaf, fi,
1132 btrfs_set_file_extent_offset(leaf, fi,
1134 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1135 struct btrfs_file_extent_item);
1136 btrfs_set_file_extent_generation(leaf, fi,
1138 btrfs_set_file_extent_num_bytes(leaf, fi,
1140 btrfs_mark_buffer_dirty(leaf);
1145 if (start > key.offset && end == extent_end) {
1148 if (extent_mergeable(leaf, path->slots[0] + 1,
1149 ino, bytenr, orig_offset,
1150 &other_start, &other_end)) {
1151 fi = btrfs_item_ptr(leaf, path->slots[0],
1152 struct btrfs_file_extent_item);
1153 btrfs_set_file_extent_num_bytes(leaf, fi,
1154 start - key.offset);
1155 btrfs_set_file_extent_generation(leaf, fi,
1158 new_key.offset = start;
1159 btrfs_set_item_key_safe(root->fs_info, path, &new_key);
1161 fi = btrfs_item_ptr(leaf, path->slots[0],
1162 struct btrfs_file_extent_item);
1163 btrfs_set_file_extent_generation(leaf, fi,
1165 btrfs_set_file_extent_num_bytes(leaf, fi,
1167 btrfs_set_file_extent_offset(leaf, fi,
1168 start - orig_offset);
1169 btrfs_mark_buffer_dirty(leaf);
1174 while (start > key.offset || end < extent_end) {
1175 if (key.offset == start)
1178 new_key.offset = split;
1179 ret = btrfs_duplicate_item(trans, root, path, &new_key);
1180 if (ret == -EAGAIN) {
1181 btrfs_release_path(path);
1185 btrfs_abort_transaction(trans, root, ret);
1189 leaf = path->nodes[0];
1190 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1191 struct btrfs_file_extent_item);
1192 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1193 btrfs_set_file_extent_num_bytes(leaf, fi,
1194 split - key.offset);
1196 fi = btrfs_item_ptr(leaf, path->slots[0],
1197 struct btrfs_file_extent_item);
1199 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1200 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
1201 btrfs_set_file_extent_num_bytes(leaf, fi,
1202 extent_end - split);
1203 btrfs_mark_buffer_dirty(leaf);
1205 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
1206 root->root_key.objectid,
1208 BUG_ON(ret); /* -ENOMEM */
1210 if (split == start) {
1213 BUG_ON(start != key.offset);
1222 if (extent_mergeable(leaf, path->slots[0] + 1,
1223 ino, bytenr, orig_offset,
1224 &other_start, &other_end)) {
1226 btrfs_release_path(path);
1229 extent_end = other_end;
1230 del_slot = path->slots[0] + 1;
1232 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1233 0, root->root_key.objectid,
1235 BUG_ON(ret); /* -ENOMEM */
1239 if (extent_mergeable(leaf, path->slots[0] - 1,
1240 ino, bytenr, orig_offset,
1241 &other_start, &other_end)) {
1243 btrfs_release_path(path);
1246 key.offset = other_start;
1247 del_slot = path->slots[0];
1249 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1250 0, root->root_key.objectid,
1252 BUG_ON(ret); /* -ENOMEM */
1255 fi = btrfs_item_ptr(leaf, path->slots[0],
1256 struct btrfs_file_extent_item);
1257 btrfs_set_file_extent_type(leaf, fi,
1258 BTRFS_FILE_EXTENT_REG);
1259 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1260 btrfs_mark_buffer_dirty(leaf);
1262 fi = btrfs_item_ptr(leaf, del_slot - 1,
1263 struct btrfs_file_extent_item);
1264 btrfs_set_file_extent_type(leaf, fi,
1265 BTRFS_FILE_EXTENT_REG);
1266 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1267 btrfs_set_file_extent_num_bytes(leaf, fi,
1268 extent_end - key.offset);
1269 btrfs_mark_buffer_dirty(leaf);
1271 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1273 btrfs_abort_transaction(trans, root, ret);
1278 btrfs_free_path(path);
1283 * on error we return an unlocked page and the error value
1284 * on success we return a locked page and 0
1286 static int prepare_uptodate_page(struct page *page, u64 pos,
1287 bool force_uptodate)
1291 if (((pos & (PAGE_CACHE_SIZE - 1)) || force_uptodate) &&
1292 !PageUptodate(page)) {
1293 ret = btrfs_readpage(NULL, page);
1297 if (!PageUptodate(page)) {
1306 * this just gets pages into the page cache and locks them down.
1308 static noinline int prepare_pages(struct inode *inode, struct page **pages,
1309 size_t num_pages, loff_t pos,
1310 size_t write_bytes, bool force_uptodate)
1313 unsigned long index = pos >> PAGE_CACHE_SHIFT;
1314 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1318 for (i = 0; i < num_pages; i++) {
1319 pages[i] = find_or_create_page(inode->i_mapping, index + i,
1320 mask | __GFP_WRITE);
1328 err = prepare_uptodate_page(pages[i], pos,
1330 if (i == num_pages - 1)
1331 err = prepare_uptodate_page(pages[i],
1332 pos + write_bytes, false);
1334 page_cache_release(pages[i]);
1338 wait_on_page_writeback(pages[i]);
1343 while (faili >= 0) {
1344 unlock_page(pages[faili]);
1345 page_cache_release(pages[faili]);
1353 * This function locks the extent and properly waits for data=ordered extents
1354 * to finish before allowing the pages to be modified if need.
1357 * 1 - the extent is locked
1358 * 0 - the extent is not locked, and everything is OK
1359 * -EAGAIN - need re-prepare the pages
1360 * the other < 0 number - Something wrong happens
1363 lock_and_cleanup_extent_if_need(struct inode *inode, struct page **pages,
1364 size_t num_pages, loff_t pos,
1365 u64 *lockstart, u64 *lockend,
1366 struct extent_state **cached_state)
1373 start_pos = pos & ~((u64)PAGE_CACHE_SIZE - 1);
1374 last_pos = start_pos + ((u64)num_pages << PAGE_CACHE_SHIFT) - 1;
1376 if (start_pos < inode->i_size) {
1377 struct btrfs_ordered_extent *ordered;
1378 lock_extent_bits(&BTRFS_I(inode)->io_tree,
1379 start_pos, last_pos, 0, cached_state);
1380 ordered = btrfs_lookup_ordered_range(inode, start_pos,
1381 last_pos - start_pos + 1);
1383 ordered->file_offset + ordered->len > start_pos &&
1384 ordered->file_offset <= last_pos) {
1385 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1386 start_pos, last_pos,
1387 cached_state, GFP_NOFS);
1388 for (i = 0; i < num_pages; i++) {
1389 unlock_page(pages[i]);
1390 page_cache_release(pages[i]);
1392 btrfs_start_ordered_extent(inode, ordered, 1);
1393 btrfs_put_ordered_extent(ordered);
1397 btrfs_put_ordered_extent(ordered);
1399 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
1400 last_pos, EXTENT_DIRTY | EXTENT_DELALLOC |
1401 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
1402 0, 0, cached_state, GFP_NOFS);
1403 *lockstart = start_pos;
1404 *lockend = last_pos;
1408 for (i = 0; i < num_pages; i++) {
1409 if (clear_page_dirty_for_io(pages[i]))
1410 account_page_redirty(pages[i]);
1411 set_page_extent_mapped(pages[i]);
1412 WARN_ON(!PageLocked(pages[i]));
1418 static noinline int check_can_nocow(struct inode *inode, loff_t pos,
1419 size_t *write_bytes)
1421 struct btrfs_root *root = BTRFS_I(inode)->root;
1422 struct btrfs_ordered_extent *ordered;
1423 u64 lockstart, lockend;
1427 ret = btrfs_start_write_no_snapshoting(root);
1431 lockstart = round_down(pos, root->sectorsize);
1432 lockend = round_up(pos + *write_bytes, root->sectorsize) - 1;
1435 lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1436 ordered = btrfs_lookup_ordered_range(inode, lockstart,
1437 lockend - lockstart + 1);
1441 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1442 btrfs_start_ordered_extent(inode, ordered, 1);
1443 btrfs_put_ordered_extent(ordered);
1446 num_bytes = lockend - lockstart + 1;
1447 ret = can_nocow_extent(inode, lockstart, &num_bytes, NULL, NULL, NULL);
1450 btrfs_end_write_no_snapshoting(root);
1452 *write_bytes = min_t(size_t, *write_bytes ,
1453 num_bytes - pos + lockstart);
1456 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1461 static noinline ssize_t __btrfs_buffered_write(struct file *file,
1465 struct inode *inode = file_inode(file);
1466 struct btrfs_root *root = BTRFS_I(inode)->root;
1467 struct page **pages = NULL;
1468 struct extent_state *cached_state = NULL;
1469 u64 release_bytes = 0;
1472 size_t num_written = 0;
1475 bool only_release_metadata = false;
1476 bool force_page_uptodate = false;
1479 nrptrs = min(DIV_ROUND_UP(iov_iter_count(i), PAGE_CACHE_SIZE),
1480 PAGE_CACHE_SIZE / (sizeof(struct page *)));
1481 nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1482 nrptrs = max(nrptrs, 8);
1483 pages = kmalloc_array(nrptrs, sizeof(struct page *), GFP_KERNEL);
1487 while (iov_iter_count(i) > 0) {
1488 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1489 size_t write_bytes = min(iov_iter_count(i),
1490 nrptrs * (size_t)PAGE_CACHE_SIZE -
1492 size_t num_pages = DIV_ROUND_UP(write_bytes + offset,
1494 size_t reserve_bytes;
1498 WARN_ON(num_pages > nrptrs);
1501 * Fault pages before locking them in prepare_pages
1502 * to avoid recursive lock
1504 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1509 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1511 if (BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW |
1512 BTRFS_INODE_PREALLOC)) {
1513 ret = check_can_nocow(inode, pos, &write_bytes);
1518 * For nodata cow case, no need to reserve
1521 only_release_metadata = true;
1523 * our prealloc extent may be smaller than
1524 * write_bytes, so scale down.
1526 num_pages = DIV_ROUND_UP(write_bytes + offset,
1528 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1529 goto reserve_metadata;
1532 ret = btrfs_check_data_free_space(inode, pos, write_bytes);
1537 ret = btrfs_delalloc_reserve_metadata(inode, reserve_bytes);
1539 if (!only_release_metadata)
1540 btrfs_free_reserved_data_space(inode, pos,
1543 btrfs_end_write_no_snapshoting(root);
1547 release_bytes = reserve_bytes;
1548 need_unlock = false;
1551 * This is going to setup the pages array with the number of
1552 * pages we want, so we don't really need to worry about the
1553 * contents of pages from loop to loop
1555 ret = prepare_pages(inode, pages, num_pages,
1557 force_page_uptodate);
1561 ret = lock_and_cleanup_extent_if_need(inode, pages, num_pages,
1562 pos, &lockstart, &lockend,
1568 } else if (ret > 0) {
1573 copied = btrfs_copy_from_user(pos, num_pages,
1574 write_bytes, pages, i);
1577 * if we have trouble faulting in the pages, fall
1578 * back to one page at a time
1580 if (copied < write_bytes)
1584 force_page_uptodate = true;
1587 force_page_uptodate = false;
1588 dirty_pages = DIV_ROUND_UP(copied + offset,
1593 * If we had a short copy we need to release the excess delaloc
1594 * bytes we reserved. We need to increment outstanding_extents
1595 * because btrfs_delalloc_release_space will decrement it, but
1596 * we still have an outstanding extent for the chunk we actually
1599 if (num_pages > dirty_pages) {
1600 release_bytes = (num_pages - dirty_pages) <<
1603 spin_lock(&BTRFS_I(inode)->lock);
1604 BTRFS_I(inode)->outstanding_extents++;
1605 spin_unlock(&BTRFS_I(inode)->lock);
1607 if (only_release_metadata) {
1608 btrfs_delalloc_release_metadata(inode,
1613 __pos = round_down(pos, root->sectorsize) +
1614 (dirty_pages << PAGE_CACHE_SHIFT);
1615 btrfs_delalloc_release_space(inode, __pos,
1620 release_bytes = dirty_pages << PAGE_CACHE_SHIFT;
1623 ret = btrfs_dirty_pages(root, inode, pages,
1624 dirty_pages, pos, copied,
1627 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1628 lockstart, lockend, &cached_state,
1631 btrfs_drop_pages(pages, num_pages);
1636 if (only_release_metadata)
1637 btrfs_end_write_no_snapshoting(root);
1639 if (only_release_metadata && copied > 0) {
1640 lockstart = round_down(pos, root->sectorsize);
1641 lockend = lockstart +
1642 (dirty_pages << PAGE_CACHE_SHIFT) - 1;
1644 set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
1645 lockend, EXTENT_NORESERVE, NULL,
1647 only_release_metadata = false;
1650 btrfs_drop_pages(pages, num_pages);
1654 balance_dirty_pages_ratelimited(inode->i_mapping);
1655 if (dirty_pages < (root->nodesize >> PAGE_CACHE_SHIFT) + 1)
1656 btrfs_btree_balance_dirty(root);
1659 num_written += copied;
1664 if (release_bytes) {
1665 if (only_release_metadata) {
1666 btrfs_end_write_no_snapshoting(root);
1667 btrfs_delalloc_release_metadata(inode, release_bytes);
1669 btrfs_delalloc_release_space(inode, pos, release_bytes);
1673 return num_written ? num_written : ret;
1676 static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1677 struct iov_iter *from,
1680 struct file *file = iocb->ki_filp;
1681 struct inode *inode = file_inode(file);
1683 ssize_t written_buffered;
1687 written = generic_file_direct_write(iocb, from, pos);
1689 if (written < 0 || !iov_iter_count(from))
1693 written_buffered = __btrfs_buffered_write(file, from, pos);
1694 if (written_buffered < 0) {
1695 err = written_buffered;
1699 * Ensure all data is persisted. We want the next direct IO read to be
1700 * able to read what was just written.
1702 endbyte = pos + written_buffered - 1;
1703 err = btrfs_fdatawrite_range(inode, pos, endbyte);
1706 err = filemap_fdatawait_range(inode->i_mapping, pos, endbyte);
1709 written += written_buffered;
1710 iocb->ki_pos = pos + written_buffered;
1711 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
1712 endbyte >> PAGE_CACHE_SHIFT);
1714 return written ? written : err;
1717 static void update_time_for_write(struct inode *inode)
1719 struct timespec now;
1721 if (IS_NOCMTIME(inode))
1724 now = current_fs_time(inode->i_sb);
1725 if (!timespec_equal(&inode->i_mtime, &now))
1726 inode->i_mtime = now;
1728 if (!timespec_equal(&inode->i_ctime, &now))
1729 inode->i_ctime = now;
1731 if (IS_I_VERSION(inode))
1732 inode_inc_iversion(inode);
1735 static ssize_t btrfs_file_write_iter(struct kiocb *iocb,
1736 struct iov_iter *from)
1738 struct file *file = iocb->ki_filp;
1739 struct inode *inode = file_inode(file);
1740 struct btrfs_root *root = BTRFS_I(inode)->root;
1743 ssize_t num_written = 0;
1744 bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host);
1749 mutex_lock(&inode->i_mutex);
1750 err = generic_write_checks(iocb, from);
1752 mutex_unlock(&inode->i_mutex);
1756 current->backing_dev_info = inode_to_bdi(inode);
1757 err = file_remove_privs(file);
1759 mutex_unlock(&inode->i_mutex);
1764 * If BTRFS flips readonly due to some impossible error
1765 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1766 * although we have opened a file as writable, we have
1767 * to stop this write operation to ensure FS consistency.
1769 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
1770 mutex_unlock(&inode->i_mutex);
1776 * We reserve space for updating the inode when we reserve space for the
1777 * extent we are going to write, so we will enospc out there. We don't
1778 * need to start yet another transaction to update the inode as we will
1779 * update the inode when we finish writing whatever data we write.
1781 update_time_for_write(inode);
1784 count = iov_iter_count(from);
1785 start_pos = round_down(pos, root->sectorsize);
1786 if (start_pos > i_size_read(inode)) {
1787 /* Expand hole size to cover write data, preventing empty gap */
1788 end_pos = round_up(pos + count, root->sectorsize);
1789 err = btrfs_cont_expand(inode, i_size_read(inode), end_pos);
1791 mutex_unlock(&inode->i_mutex);
1797 atomic_inc(&BTRFS_I(inode)->sync_writers);
1799 if (iocb->ki_flags & IOCB_DIRECT) {
1800 num_written = __btrfs_direct_write(iocb, from, pos);
1802 num_written = __btrfs_buffered_write(file, from, pos);
1803 if (num_written > 0)
1804 iocb->ki_pos = pos + num_written;
1807 mutex_unlock(&inode->i_mutex);
1810 * We also have to set last_sub_trans to the current log transid,
1811 * otherwise subsequent syncs to a file that's been synced in this
1812 * transaction will appear to have already occured.
1814 spin_lock(&BTRFS_I(inode)->lock);
1815 BTRFS_I(inode)->last_sub_trans = root->log_transid;
1816 spin_unlock(&BTRFS_I(inode)->lock);
1817 if (num_written > 0) {
1818 err = generic_write_sync(file, pos, num_written);
1824 atomic_dec(&BTRFS_I(inode)->sync_writers);
1826 current->backing_dev_info = NULL;
1827 return num_written ? num_written : err;
1830 int btrfs_release_file(struct inode *inode, struct file *filp)
1832 if (filp->private_data)
1833 btrfs_ioctl_trans_end(filp);
1835 * ordered_data_close is set by settattr when we are about to truncate
1836 * a file from a non-zero size to a zero size. This tries to
1837 * flush down new bytes that may have been written if the
1838 * application were using truncate to replace a file in place.
1840 if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
1841 &BTRFS_I(inode)->runtime_flags))
1842 filemap_flush(inode->i_mapping);
1846 static int start_ordered_ops(struct inode *inode, loff_t start, loff_t end)
1850 atomic_inc(&BTRFS_I(inode)->sync_writers);
1851 ret = btrfs_fdatawrite_range(inode, start, end);
1852 atomic_dec(&BTRFS_I(inode)->sync_writers);
1858 * fsync call for both files and directories. This logs the inode into
1859 * the tree log instead of forcing full commits whenever possible.
1861 * It needs to call filemap_fdatawait so that all ordered extent updates are
1862 * in the metadata btree are up to date for copying to the log.
1864 * It drops the inode mutex before doing the tree log commit. This is an
1865 * important optimization for directories because holding the mutex prevents
1866 * new operations on the dir while we write to disk.
1868 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1870 struct dentry *dentry = file->f_path.dentry;
1871 struct inode *inode = d_inode(dentry);
1872 struct btrfs_root *root = BTRFS_I(inode)->root;
1873 struct btrfs_trans_handle *trans;
1874 struct btrfs_log_ctx ctx;
1877 const u64 len = end - start + 1;
1879 trace_btrfs_sync_file(file, datasync);
1882 * We write the dirty pages in the range and wait until they complete
1883 * out of the ->i_mutex. If so, we can flush the dirty pages by
1884 * multi-task, and make the performance up. See
1885 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1887 ret = start_ordered_ops(inode, start, end);
1891 mutex_lock(&inode->i_mutex);
1892 atomic_inc(&root->log_batch);
1893 full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1894 &BTRFS_I(inode)->runtime_flags);
1896 * We might have have had more pages made dirty after calling
1897 * start_ordered_ops and before acquiring the inode's i_mutex.
1901 * For a full sync, we need to make sure any ordered operations
1902 * start and finish before we start logging the inode, so that
1903 * all extents are persisted and the respective file extent
1904 * items are in the fs/subvol btree.
1906 ret = btrfs_wait_ordered_range(inode, start, len);
1909 * Start any new ordered operations before starting to log the
1910 * inode. We will wait for them to finish in btrfs_sync_log().
1912 * Right before acquiring the inode's mutex, we might have new
1913 * writes dirtying pages, which won't immediately start the
1914 * respective ordered operations - that is done through the
1915 * fill_delalloc callbacks invoked from the writepage and
1916 * writepages address space operations. So make sure we start
1917 * all ordered operations before starting to log our inode. Not
1918 * doing this means that while logging the inode, writeback
1919 * could start and invoke writepage/writepages, which would call
1920 * the fill_delalloc callbacks (cow_file_range,
1921 * submit_compressed_extents). These callbacks add first an
1922 * extent map to the modified list of extents and then create
1923 * the respective ordered operation, which means in
1924 * tree-log.c:btrfs_log_inode() we might capture all existing
1925 * ordered operations (with btrfs_get_logged_extents()) before
1926 * the fill_delalloc callback adds its ordered operation, and by
1927 * the time we visit the modified list of extent maps (with
1928 * btrfs_log_changed_extents()), we see and process the extent
1929 * map they created. We then use the extent map to construct a
1930 * file extent item for logging without waiting for the
1931 * respective ordered operation to finish - this file extent
1932 * item points to a disk location that might not have yet been
1933 * written to, containing random data - so after a crash a log
1934 * replay will make our inode have file extent items that point
1935 * to disk locations containing invalid data, as we returned
1936 * success to userspace without waiting for the respective
1937 * ordered operation to finish, because it wasn't captured by
1938 * btrfs_get_logged_extents().
1940 ret = start_ordered_ops(inode, start, end);
1943 mutex_unlock(&inode->i_mutex);
1946 atomic_inc(&root->log_batch);
1949 * If the last transaction that changed this file was before the current
1950 * transaction and we have the full sync flag set in our inode, we can
1951 * bail out now without any syncing.
1953 * Note that we can't bail out if the full sync flag isn't set. This is
1954 * because when the full sync flag is set we start all ordered extents
1955 * and wait for them to fully complete - when they complete they update
1956 * the inode's last_trans field through:
1958 * btrfs_finish_ordered_io() ->
1959 * btrfs_update_inode_fallback() ->
1960 * btrfs_update_inode() ->
1961 * btrfs_set_inode_last_trans()
1963 * So we are sure that last_trans is up to date and can do this check to
1964 * bail out safely. For the fast path, when the full sync flag is not
1965 * set in our inode, we can not do it because we start only our ordered
1966 * extents and don't wait for them to complete (that is when
1967 * btrfs_finish_ordered_io runs), so here at this point their last_trans
1968 * value might be less than or equals to fs_info->last_trans_committed,
1969 * and setting a speculative last_trans for an inode when a buffered
1970 * write is made (such as fs_info->generation + 1 for example) would not
1971 * be reliable since after setting the value and before fsync is called
1972 * any number of transactions can start and commit (transaction kthread
1973 * commits the current transaction periodically), and a transaction
1974 * commit does not start nor waits for ordered extents to complete.
1977 if (btrfs_inode_in_log(inode, root->fs_info->generation) ||
1978 (BTRFS_I(inode)->last_trans <=
1979 root->fs_info->last_trans_committed &&
1981 !btrfs_have_ordered_extents_in_range(inode, start, len)))) {
1983 * We'v had everything committed since the last time we were
1984 * modified so clear this flag in case it was set for whatever
1985 * reason, it's no longer relevant.
1987 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1988 &BTRFS_I(inode)->runtime_flags);
1989 mutex_unlock(&inode->i_mutex);
1994 * ok we haven't committed the transaction yet, lets do a commit
1996 if (file->private_data)
1997 btrfs_ioctl_trans_end(file);
2000 * We use start here because we will need to wait on the IO to complete
2001 * in btrfs_sync_log, which could require joining a transaction (for
2002 * example checking cross references in the nocow path). If we use join
2003 * here we could get into a situation where we're waiting on IO to
2004 * happen that is blocked on a transaction trying to commit. With start
2005 * we inc the extwriter counter, so we wait for all extwriters to exit
2006 * before we start blocking join'ers. This comment is to keep somebody
2007 * from thinking they are super smart and changing this to
2008 * btrfs_join_transaction *cough*Josef*cough*.
2010 trans = btrfs_start_transaction(root, 0);
2011 if (IS_ERR(trans)) {
2012 ret = PTR_ERR(trans);
2013 mutex_unlock(&inode->i_mutex);
2018 btrfs_init_log_ctx(&ctx);
2020 ret = btrfs_log_dentry_safe(trans, root, dentry, start, end, &ctx);
2022 /* Fallthrough and commit/free transaction. */
2026 /* we've logged all the items and now have a consistent
2027 * version of the file in the log. It is possible that
2028 * someone will come in and modify the file, but that's
2029 * fine because the log is consistent on disk, and we
2030 * have references to all of the file's extents
2032 * It is possible that someone will come in and log the
2033 * file again, but that will end up using the synchronization
2034 * inside btrfs_sync_log to keep things safe.
2036 mutex_unlock(&inode->i_mutex);
2039 * If any of the ordered extents had an error, just return it to user
2040 * space, so that the application knows some writes didn't succeed and
2041 * can take proper action (retry for e.g.). Blindly committing the
2042 * transaction in this case, would fool userspace that everything was
2043 * successful. And we also want to make sure our log doesn't contain
2044 * file extent items pointing to extents that weren't fully written to -
2045 * just like in the non fast fsync path, where we check for the ordered
2046 * operation's error flag before writing to the log tree and return -EIO
2047 * if any of them had this flag set (btrfs_wait_ordered_range) -
2048 * therefore we need to check for errors in the ordered operations,
2049 * which are indicated by ctx.io_err.
2052 btrfs_end_transaction(trans, root);
2057 if (ret != BTRFS_NO_LOG_SYNC) {
2059 ret = btrfs_sync_log(trans, root, &ctx);
2061 ret = btrfs_end_transaction(trans, root);
2066 ret = btrfs_wait_ordered_range(inode, start,
2069 btrfs_end_transaction(trans, root);
2073 ret = btrfs_commit_transaction(trans, root);
2075 ret = btrfs_end_transaction(trans, root);
2078 return ret > 0 ? -EIO : ret;
2081 static const struct vm_operations_struct btrfs_file_vm_ops = {
2082 .fault = filemap_fault,
2083 .map_pages = filemap_map_pages,
2084 .page_mkwrite = btrfs_page_mkwrite,
2087 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
2089 struct address_space *mapping = filp->f_mapping;
2091 if (!mapping->a_ops->readpage)
2094 file_accessed(filp);
2095 vma->vm_ops = &btrfs_file_vm_ops;
2100 static int hole_mergeable(struct inode *inode, struct extent_buffer *leaf,
2101 int slot, u64 start, u64 end)
2103 struct btrfs_file_extent_item *fi;
2104 struct btrfs_key key;
2106 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
2109 btrfs_item_key_to_cpu(leaf, &key, slot);
2110 if (key.objectid != btrfs_ino(inode) ||
2111 key.type != BTRFS_EXTENT_DATA_KEY)
2114 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2116 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2119 if (btrfs_file_extent_disk_bytenr(leaf, fi))
2122 if (key.offset == end)
2124 if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
2129 static int fill_holes(struct btrfs_trans_handle *trans, struct inode *inode,
2130 struct btrfs_path *path, u64 offset, u64 end)
2132 struct btrfs_root *root = BTRFS_I(inode)->root;
2133 struct extent_buffer *leaf;
2134 struct btrfs_file_extent_item *fi;
2135 struct extent_map *hole_em;
2136 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2137 struct btrfs_key key;
2140 if (btrfs_fs_incompat(root->fs_info, NO_HOLES))
2143 key.objectid = btrfs_ino(inode);
2144 key.type = BTRFS_EXTENT_DATA_KEY;
2145 key.offset = offset;
2147 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2152 leaf = path->nodes[0];
2153 if (hole_mergeable(inode, leaf, path->slots[0]-1, offset, end)) {
2157 fi = btrfs_item_ptr(leaf, path->slots[0],
2158 struct btrfs_file_extent_item);
2159 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2161 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2162 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2163 btrfs_set_file_extent_offset(leaf, fi, 0);
2164 btrfs_mark_buffer_dirty(leaf);
2168 if (hole_mergeable(inode, leaf, path->slots[0], offset, end)) {
2171 key.offset = offset;
2172 btrfs_set_item_key_safe(root->fs_info, path, &key);
2173 fi = btrfs_item_ptr(leaf, path->slots[0],
2174 struct btrfs_file_extent_item);
2175 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2177 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2178 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2179 btrfs_set_file_extent_offset(leaf, fi, 0);
2180 btrfs_mark_buffer_dirty(leaf);
2183 btrfs_release_path(path);
2185 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
2186 0, 0, end - offset, 0, end - offset,
2192 btrfs_release_path(path);
2194 hole_em = alloc_extent_map();
2196 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2197 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2198 &BTRFS_I(inode)->runtime_flags);
2200 hole_em->start = offset;
2201 hole_em->len = end - offset;
2202 hole_em->ram_bytes = hole_em->len;
2203 hole_em->orig_start = offset;
2205 hole_em->block_start = EXTENT_MAP_HOLE;
2206 hole_em->block_len = 0;
2207 hole_em->orig_block_len = 0;
2208 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
2209 hole_em->compress_type = BTRFS_COMPRESS_NONE;
2210 hole_em->generation = trans->transid;
2213 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2214 write_lock(&em_tree->lock);
2215 ret = add_extent_mapping(em_tree, hole_em, 1);
2216 write_unlock(&em_tree->lock);
2217 } while (ret == -EEXIST);
2218 free_extent_map(hole_em);
2220 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2221 &BTRFS_I(inode)->runtime_flags);
2228 * Find a hole extent on given inode and change start/len to the end of hole
2229 * extent.(hole/vacuum extent whose em->start <= start &&
2230 * em->start + em->len > start)
2231 * When a hole extent is found, return 1 and modify start/len.
2233 static int find_first_non_hole(struct inode *inode, u64 *start, u64 *len)
2235 struct extent_map *em;
2238 em = btrfs_get_extent(inode, NULL, 0, *start, *len, 0);
2239 if (IS_ERR_OR_NULL(em)) {
2247 /* Hole or vacuum extent(only exists in no-hole mode) */
2248 if (em->block_start == EXTENT_MAP_HOLE) {
2250 *len = em->start + em->len > *start + *len ?
2251 0 : *start + *len - em->start - em->len;
2252 *start = em->start + em->len;
2254 free_extent_map(em);
2258 static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
2260 struct btrfs_root *root = BTRFS_I(inode)->root;
2261 struct extent_state *cached_state = NULL;
2262 struct btrfs_path *path;
2263 struct btrfs_block_rsv *rsv;
2264 struct btrfs_trans_handle *trans;
2269 u64 orig_start = offset;
2271 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
2275 unsigned int rsv_count;
2277 bool no_holes = btrfs_fs_incompat(root->fs_info, NO_HOLES);
2279 bool truncated_page = false;
2280 bool updated_inode = false;
2282 ret = btrfs_wait_ordered_range(inode, offset, len);
2286 mutex_lock(&inode->i_mutex);
2287 ino_size = round_up(inode->i_size, PAGE_CACHE_SIZE);
2288 ret = find_first_non_hole(inode, &offset, &len);
2290 goto out_only_mutex;
2292 /* Already in a large hole */
2294 goto out_only_mutex;
2297 lockstart = round_up(offset, BTRFS_I(inode)->root->sectorsize);
2298 lockend = round_down(offset + len,
2299 BTRFS_I(inode)->root->sectorsize) - 1;
2300 same_page = ((offset >> PAGE_CACHE_SHIFT) ==
2301 ((offset + len - 1) >> PAGE_CACHE_SHIFT));
2304 * We needn't truncate any page which is beyond the end of the file
2305 * because we are sure there is no data there.
2308 * Only do this if we are in the same page and we aren't doing the
2311 if (same_page && len < PAGE_CACHE_SIZE) {
2312 if (offset < ino_size) {
2313 truncated_page = true;
2314 ret = btrfs_truncate_page(inode, offset, len, 0);
2318 goto out_only_mutex;
2321 /* zero back part of the first page */
2322 if (offset < ino_size) {
2323 truncated_page = true;
2324 ret = btrfs_truncate_page(inode, offset, 0, 0);
2326 mutex_unlock(&inode->i_mutex);
2331 /* Check the aligned pages after the first unaligned page,
2332 * if offset != orig_start, which means the first unaligned page
2333 * including serveral following pages are already in holes,
2334 * the extra check can be skipped */
2335 if (offset == orig_start) {
2336 /* after truncate page, check hole again */
2337 len = offset + len - lockstart;
2339 ret = find_first_non_hole(inode, &offset, &len);
2341 goto out_only_mutex;
2344 goto out_only_mutex;
2349 /* Check the tail unaligned part is in a hole */
2350 tail_start = lockend + 1;
2351 tail_len = offset + len - tail_start;
2353 ret = find_first_non_hole(inode, &tail_start, &tail_len);
2354 if (unlikely(ret < 0))
2355 goto out_only_mutex;
2357 /* zero the front end of the last page */
2358 if (tail_start + tail_len < ino_size) {
2359 truncated_page = true;
2360 ret = btrfs_truncate_page(inode,
2361 tail_start + tail_len, 0, 1);
2363 goto out_only_mutex;
2368 if (lockend < lockstart) {
2370 goto out_only_mutex;
2374 struct btrfs_ordered_extent *ordered;
2376 truncate_pagecache_range(inode, lockstart, lockend);
2378 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2380 ordered = btrfs_lookup_first_ordered_extent(inode, lockend);
2383 * We need to make sure we have no ordered extents in this range
2384 * and nobody raced in and read a page in this range, if we did
2385 * we need to try again.
2388 (ordered->file_offset + ordered->len <= lockstart ||
2389 ordered->file_offset > lockend)) &&
2390 !btrfs_page_exists_in_range(inode, lockstart, lockend)) {
2392 btrfs_put_ordered_extent(ordered);
2396 btrfs_put_ordered_extent(ordered);
2397 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
2398 lockend, &cached_state, GFP_NOFS);
2399 ret = btrfs_wait_ordered_range(inode, lockstart,
2400 lockend - lockstart + 1);
2402 mutex_unlock(&inode->i_mutex);
2407 path = btrfs_alloc_path();
2413 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2418 rsv->size = btrfs_calc_trunc_metadata_size(root, 1);
2422 * 1 - update the inode
2423 * 1 - removing the extents in the range
2424 * 1 - adding the hole extent if no_holes isn't set
2426 rsv_count = no_holes ? 2 : 3;
2427 trans = btrfs_start_transaction(root, rsv_count);
2428 if (IS_ERR(trans)) {
2429 err = PTR_ERR(trans);
2433 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
2436 trans->block_rsv = rsv;
2438 cur_offset = lockstart;
2439 len = lockend - cur_offset;
2440 while (cur_offset < lockend) {
2441 ret = __btrfs_drop_extents(trans, root, inode, path,
2442 cur_offset, lockend + 1,
2443 &drop_end, 1, 0, 0, NULL);
2447 trans->block_rsv = &root->fs_info->trans_block_rsv;
2449 if (cur_offset < ino_size) {
2450 ret = fill_holes(trans, inode, path, cur_offset,
2458 cur_offset = drop_end;
2460 ret = btrfs_update_inode(trans, root, inode);
2466 btrfs_end_transaction(trans, root);
2467 btrfs_btree_balance_dirty(root);
2469 trans = btrfs_start_transaction(root, rsv_count);
2470 if (IS_ERR(trans)) {
2471 ret = PTR_ERR(trans);
2476 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
2478 BUG_ON(ret); /* shouldn't happen */
2479 trans->block_rsv = rsv;
2481 ret = find_first_non_hole(inode, &cur_offset, &len);
2482 if (unlikely(ret < 0))
2495 trans->block_rsv = &root->fs_info->trans_block_rsv;
2497 * If we are using the NO_HOLES feature we might have had already an
2498 * hole that overlaps a part of the region [lockstart, lockend] and
2499 * ends at (or beyond) lockend. Since we have no file extent items to
2500 * represent holes, drop_end can be less than lockend and so we must
2501 * make sure we have an extent map representing the existing hole (the
2502 * call to __btrfs_drop_extents() might have dropped the existing extent
2503 * map representing the existing hole), otherwise the fast fsync path
2504 * will not record the existence of the hole region
2505 * [existing_hole_start, lockend].
2507 if (drop_end <= lockend)
2508 drop_end = lockend + 1;
2510 * Don't insert file hole extent item if it's for a range beyond eof
2511 * (because it's useless) or if it represents a 0 bytes range (when
2512 * cur_offset == drop_end).
2514 if (cur_offset < ino_size && cur_offset < drop_end) {
2515 ret = fill_holes(trans, inode, path, cur_offset, drop_end);
2526 inode_inc_iversion(inode);
2527 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2529 trans->block_rsv = &root->fs_info->trans_block_rsv;
2530 ret = btrfs_update_inode(trans, root, inode);
2531 updated_inode = true;
2532 btrfs_end_transaction(trans, root);
2533 btrfs_btree_balance_dirty(root);
2535 btrfs_free_path(path);
2536 btrfs_free_block_rsv(root, rsv);
2538 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2539 &cached_state, GFP_NOFS);
2541 if (!updated_inode && truncated_page && !ret && !err) {
2543 * If we only end up zeroing part of a page, we still need to
2544 * update the inode item, so that all the time fields are
2545 * updated as well as the necessary btrfs inode in memory fields
2546 * for detecting, at fsync time, if the inode isn't yet in the
2547 * log tree or it's there but not up to date.
2549 trans = btrfs_start_transaction(root, 1);
2550 if (IS_ERR(trans)) {
2551 err = PTR_ERR(trans);
2553 err = btrfs_update_inode(trans, root, inode);
2554 ret = btrfs_end_transaction(trans, root);
2557 mutex_unlock(&inode->i_mutex);
2563 /* Helper structure to record which range is already reserved */
2564 struct falloc_range {
2565 struct list_head list;
2571 * Helper function to add falloc range
2573 * Caller should have locked the larger range of extent containing
2576 static int add_falloc_range(struct list_head *head, u64 start, u64 len)
2578 struct falloc_range *prev = NULL;
2579 struct falloc_range *range = NULL;
2581 if (list_empty(head))
2585 * As fallocate iterate by bytenr order, we only need to check
2588 prev = list_entry(head->prev, struct falloc_range, list);
2589 if (prev->start + prev->len == start) {
2594 range = kmalloc(sizeof(*range), GFP_NOFS);
2597 range->start = start;
2599 list_add_tail(&range->list, head);
2603 static long btrfs_fallocate(struct file *file, int mode,
2604 loff_t offset, loff_t len)
2606 struct inode *inode = file_inode(file);
2607 struct extent_state *cached_state = NULL;
2608 struct falloc_range *range;
2609 struct falloc_range *tmp;
2610 struct list_head reserve_list;
2618 struct extent_map *em;
2619 int blocksize = BTRFS_I(inode)->root->sectorsize;
2622 alloc_start = round_down(offset, blocksize);
2623 alloc_end = round_up(offset + len, blocksize);
2625 /* Make sure we aren't being give some crap mode */
2626 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2629 if (mode & FALLOC_FL_PUNCH_HOLE)
2630 return btrfs_punch_hole(inode, offset, len);
2633 * Only trigger disk allocation, don't trigger qgroup reserve
2635 * For qgroup space, it will be checked later.
2637 ret = btrfs_alloc_data_chunk_ondemand(inode, alloc_end - alloc_start);
2641 mutex_lock(&inode->i_mutex);
2642 ret = inode_newsize_ok(inode, alloc_end);
2647 * TODO: Move these two operations after we have checked
2648 * accurate reserved space, or fallocate can still fail but
2649 * with page truncated or size expanded.
2651 * But that's a minor problem and won't do much harm BTW.
2653 if (alloc_start > inode->i_size) {
2654 ret = btrfs_cont_expand(inode, i_size_read(inode),
2658 } else if (offset + len > inode->i_size) {
2660 * If we are fallocating from the end of the file onward we
2661 * need to zero out the end of the page if i_size lands in the
2664 ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
2670 * wait for ordered IO before we have any locks. We'll loop again
2671 * below with the locks held.
2673 ret = btrfs_wait_ordered_range(inode, alloc_start,
2674 alloc_end - alloc_start);
2678 locked_end = alloc_end - 1;
2680 struct btrfs_ordered_extent *ordered;
2682 /* the extent lock is ordered inside the running
2685 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
2686 locked_end, 0, &cached_state);
2687 ordered = btrfs_lookup_first_ordered_extent(inode,
2690 ordered->file_offset + ordered->len > alloc_start &&
2691 ordered->file_offset < alloc_end) {
2692 btrfs_put_ordered_extent(ordered);
2693 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
2694 alloc_start, locked_end,
2695 &cached_state, GFP_NOFS);
2697 * we can't wait on the range with the transaction
2698 * running or with the extent lock held
2700 ret = btrfs_wait_ordered_range(inode, alloc_start,
2701 alloc_end - alloc_start);
2706 btrfs_put_ordered_extent(ordered);
2711 /* First, check if we exceed the qgroup limit */
2712 INIT_LIST_HEAD(&reserve_list);
2713 cur_offset = alloc_start;
2715 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2716 alloc_end - cur_offset, 0);
2717 if (IS_ERR_OR_NULL(em)) {
2724 last_byte = min(extent_map_end(em), alloc_end);
2725 actual_end = min_t(u64, extent_map_end(em), offset + len);
2726 last_byte = ALIGN(last_byte, blocksize);
2727 if (em->block_start == EXTENT_MAP_HOLE ||
2728 (cur_offset >= inode->i_size &&
2729 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
2730 ret = add_falloc_range(&reserve_list, cur_offset,
2731 last_byte - cur_offset);
2733 free_extent_map(em);
2736 ret = btrfs_qgroup_reserve_data(inode, cur_offset,
2737 last_byte - cur_offset);
2741 free_extent_map(em);
2742 cur_offset = last_byte;
2743 if (cur_offset >= alloc_end)
2748 * If ret is still 0, means we're OK to fallocate.
2749 * Or just cleanup the list and exit.
2751 list_for_each_entry_safe(range, tmp, &reserve_list, list) {
2753 ret = btrfs_prealloc_file_range(inode, mode,
2755 range->len, 1 << inode->i_blkbits,
2756 offset + len, &alloc_hint);
2757 list_del(&range->list);
2763 if (actual_end > inode->i_size &&
2764 !(mode & FALLOC_FL_KEEP_SIZE)) {
2765 struct btrfs_trans_handle *trans;
2766 struct btrfs_root *root = BTRFS_I(inode)->root;
2769 * We didn't need to allocate any more space, but we
2770 * still extended the size of the file so we need to
2771 * update i_size and the inode item.
2773 trans = btrfs_start_transaction(root, 1);
2774 if (IS_ERR(trans)) {
2775 ret = PTR_ERR(trans);
2777 inode->i_ctime = CURRENT_TIME;
2778 i_size_write(inode, actual_end);
2779 btrfs_ordered_update_i_size(inode, actual_end, NULL);
2780 ret = btrfs_update_inode(trans, root, inode);
2782 btrfs_end_transaction(trans, root);
2784 ret = btrfs_end_transaction(trans, root);
2788 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
2789 &cached_state, GFP_NOFS);
2792 * As we waited the extent range, the data_rsv_map must be empty
2793 * in the range, as written data range will be released from it.
2794 * And for prealloacted extent, it will also be released when
2795 * its metadata is written.
2796 * So this is completely used as cleanup.
2798 btrfs_qgroup_free_data(inode, alloc_start, alloc_end - alloc_start);
2799 mutex_unlock(&inode->i_mutex);
2800 /* Let go of our reservation. */
2801 btrfs_free_reserved_data_space(inode, alloc_start,
2802 alloc_end - alloc_start);
2806 static int find_desired_extent(struct inode *inode, loff_t *offset, int whence)
2808 struct btrfs_root *root = BTRFS_I(inode)->root;
2809 struct extent_map *em = NULL;
2810 struct extent_state *cached_state = NULL;
2817 if (inode->i_size == 0)
2821 * *offset can be negative, in this case we start finding DATA/HOLE from
2822 * the very start of the file.
2824 start = max_t(loff_t, 0, *offset);
2826 lockstart = round_down(start, root->sectorsize);
2827 lockend = round_up(i_size_read(inode), root->sectorsize);
2828 if (lockend <= lockstart)
2829 lockend = lockstart + root->sectorsize;
2831 len = lockend - lockstart + 1;
2833 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, 0,
2836 while (start < inode->i_size) {
2837 em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0);
2844 if (whence == SEEK_HOLE &&
2845 (em->block_start == EXTENT_MAP_HOLE ||
2846 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2848 else if (whence == SEEK_DATA &&
2849 (em->block_start != EXTENT_MAP_HOLE &&
2850 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2853 start = em->start + em->len;
2854 free_extent_map(em);
2858 free_extent_map(em);
2860 if (whence == SEEK_DATA && start >= inode->i_size)
2863 *offset = min_t(loff_t, start, inode->i_size);
2865 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2866 &cached_state, GFP_NOFS);
2870 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
2872 struct inode *inode = file->f_mapping->host;
2875 mutex_lock(&inode->i_mutex);
2879 offset = generic_file_llseek(file, offset, whence);
2883 if (offset >= i_size_read(inode)) {
2884 mutex_unlock(&inode->i_mutex);
2888 ret = find_desired_extent(inode, &offset, whence);
2890 mutex_unlock(&inode->i_mutex);
2895 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2897 mutex_unlock(&inode->i_mutex);
2901 const struct file_operations btrfs_file_operations = {
2902 .llseek = btrfs_file_llseek,
2903 .read_iter = generic_file_read_iter,
2904 .splice_read = generic_file_splice_read,
2905 .write_iter = btrfs_file_write_iter,
2906 .mmap = btrfs_file_mmap,
2907 .open = generic_file_open,
2908 .release = btrfs_release_file,
2909 .fsync = btrfs_sync_file,
2910 .fallocate = btrfs_fallocate,
2911 .unlocked_ioctl = btrfs_ioctl,
2912 #ifdef CONFIG_COMPAT
2913 .compat_ioctl = btrfs_ioctl,
2917 void btrfs_auto_defrag_exit(void)
2919 if (btrfs_inode_defrag_cachep)
2920 kmem_cache_destroy(btrfs_inode_defrag_cachep);
2923 int btrfs_auto_defrag_init(void)
2925 btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
2926 sizeof(struct inode_defrag), 0,
2927 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2929 if (!btrfs_inode_defrag_cachep)
2935 int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end)
2940 * So with compression we will find and lock a dirty page and clear the
2941 * first one as dirty, setup an async extent, and immediately return
2942 * with the entire range locked but with nobody actually marked with
2943 * writeback. So we can't just filemap_write_and_wait_range() and
2944 * expect it to work since it will just kick off a thread to do the
2945 * actual work. So we need to call filemap_fdatawrite_range _again_
2946 * since it will wait on the page lock, which won't be unlocked until
2947 * after the pages have been marked as writeback and so we're good to go
2948 * from there. We have to do this otherwise we'll miss the ordered
2949 * extents and that results in badness. Please Josef, do not think you
2950 * know better and pull this out at some point in the future, it is
2951 * right and you are wrong.
2953 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
2954 if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
2955 &BTRFS_I(inode)->runtime_flags))
2956 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);