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, size_t write_bytes,
410 struct page **prepared_pages,
414 size_t total_copied = 0;
416 int offset = pos & (PAGE_CACHE_SIZE - 1);
418 while (write_bytes > 0) {
419 size_t count = min_t(size_t,
420 PAGE_CACHE_SIZE - offset, write_bytes);
421 struct page *page = prepared_pages[pg];
423 * Copy data from userspace to the current page
425 copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
427 /* Flush processor's dcache for this page */
428 flush_dcache_page(page);
431 * if we get a partial write, we can end up with
432 * partially up to date pages. These add
433 * a lot of complexity, so make sure they don't
434 * happen by forcing this copy to be retried.
436 * The rest of the btrfs_file_write code will fall
437 * back to page at a time copies after we return 0.
439 if (!PageUptodate(page) && copied < count)
442 iov_iter_advance(i, copied);
443 write_bytes -= copied;
444 total_copied += copied;
446 /* Return to btrfs_file_write_iter to fault page */
447 if (unlikely(copied == 0))
450 if (copied < PAGE_CACHE_SIZE - offset) {
461 * unlocks pages after btrfs_file_write is done with them
463 static void btrfs_drop_pages(struct page **pages, size_t num_pages)
466 for (i = 0; i < num_pages; i++) {
467 /* page checked is some magic around finding pages that
468 * have been modified without going through btrfs_set_page_dirty
469 * clear it here. There should be no need to mark the pages
470 * accessed as prepare_pages should have marked them accessed
471 * in prepare_pages via find_or_create_page()
473 ClearPageChecked(pages[i]);
474 unlock_page(pages[i]);
475 page_cache_release(pages[i]);
480 * after copy_from_user, pages need to be dirtied and we need to make
481 * sure holes are created between the current EOF and the start of
482 * any next extents (if required).
484 * this also makes the decision about creating an inline extent vs
485 * doing real data extents, marking pages dirty and delalloc as required.
487 int btrfs_dirty_pages(struct btrfs_root *root, struct inode *inode,
488 struct page **pages, size_t num_pages,
489 loff_t pos, size_t write_bytes,
490 struct extent_state **cached)
496 u64 end_of_last_block;
497 u64 end_pos = pos + write_bytes;
498 loff_t isize = i_size_read(inode);
500 start_pos = pos & ~((u64)root->sectorsize - 1);
501 num_bytes = round_up(write_bytes + pos - start_pos, root->sectorsize);
503 end_of_last_block = start_pos + num_bytes - 1;
504 err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
509 for (i = 0; i < num_pages; i++) {
510 struct page *p = pages[i];
517 * we've only changed i_size in ram, and we haven't updated
518 * the disk i_size. There is no need to log the inode
522 i_size_write(inode, end_pos);
527 * this drops all the extents in the cache that intersect the range
528 * [start, end]. Existing extents are split as required.
530 void btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
533 struct extent_map *em;
534 struct extent_map *split = NULL;
535 struct extent_map *split2 = NULL;
536 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
537 u64 len = end - start + 1;
545 WARN_ON(end < start);
546 if (end == (u64)-1) {
555 split = alloc_extent_map();
557 split2 = alloc_extent_map();
558 if (!split || !split2)
561 write_lock(&em_tree->lock);
562 em = lookup_extent_mapping(em_tree, start, len);
564 write_unlock(&em_tree->lock);
568 gen = em->generation;
569 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
570 if (testend && em->start + em->len >= start + len) {
572 write_unlock(&em_tree->lock);
575 start = em->start + em->len;
577 len = start + len - (em->start + em->len);
579 write_unlock(&em_tree->lock);
582 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
583 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
584 clear_bit(EXTENT_FLAG_LOGGING, &flags);
585 modified = !list_empty(&em->list);
589 if (em->start < start) {
590 split->start = em->start;
591 split->len = start - em->start;
593 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
594 split->orig_start = em->orig_start;
595 split->block_start = em->block_start;
598 split->block_len = em->block_len;
600 split->block_len = split->len;
601 split->orig_block_len = max(split->block_len,
603 split->ram_bytes = em->ram_bytes;
605 split->orig_start = split->start;
606 split->block_len = 0;
607 split->block_start = em->block_start;
608 split->orig_block_len = 0;
609 split->ram_bytes = split->len;
612 split->generation = gen;
613 split->bdev = em->bdev;
614 split->flags = flags;
615 split->compress_type = em->compress_type;
616 replace_extent_mapping(em_tree, em, split, modified);
617 free_extent_map(split);
621 if (testend && em->start + em->len > start + len) {
622 u64 diff = start + len - em->start;
624 split->start = start + len;
625 split->len = em->start + em->len - (start + len);
626 split->bdev = em->bdev;
627 split->flags = flags;
628 split->compress_type = em->compress_type;
629 split->generation = gen;
631 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
632 split->orig_block_len = max(em->block_len,
635 split->ram_bytes = em->ram_bytes;
637 split->block_len = em->block_len;
638 split->block_start = em->block_start;
639 split->orig_start = em->orig_start;
641 split->block_len = split->len;
642 split->block_start = em->block_start
644 split->orig_start = em->orig_start;
647 split->ram_bytes = split->len;
648 split->orig_start = split->start;
649 split->block_len = 0;
650 split->block_start = em->block_start;
651 split->orig_block_len = 0;
654 if (extent_map_in_tree(em)) {
655 replace_extent_mapping(em_tree, em, split,
658 ret = add_extent_mapping(em_tree, split,
660 ASSERT(ret == 0); /* Logic error */
662 free_extent_map(split);
666 if (extent_map_in_tree(em))
667 remove_extent_mapping(em_tree, em);
668 write_unlock(&em_tree->lock);
672 /* once for the tree*/
676 free_extent_map(split);
678 free_extent_map(split2);
682 * this is very complex, but the basic idea is to drop all extents
683 * in the range start - end. hint_block is filled in with a block number
684 * that would be a good hint to the block allocator for this file.
686 * If an extent intersects the range but is not entirely inside the range
687 * it is either truncated or split. Anything entirely inside the range
688 * is deleted from the tree.
690 int __btrfs_drop_extents(struct btrfs_trans_handle *trans,
691 struct btrfs_root *root, struct inode *inode,
692 struct btrfs_path *path, u64 start, u64 end,
693 u64 *drop_end, int drop_cache,
695 u32 extent_item_size,
698 struct extent_buffer *leaf;
699 struct btrfs_file_extent_item *fi;
700 struct btrfs_key key;
701 struct btrfs_key new_key;
702 u64 ino = btrfs_ino(inode);
703 u64 search_start = start;
706 u64 extent_offset = 0;
713 int modify_tree = -1;
716 int leafs_visited = 0;
719 btrfs_drop_extent_cache(inode, start, end - 1, 0);
721 if (start >= BTRFS_I(inode)->disk_i_size && !replace_extent)
724 update_refs = (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
725 root == root->fs_info->tree_root);
728 ret = btrfs_lookup_file_extent(trans, root, path, ino,
729 search_start, modify_tree);
732 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
733 leaf = path->nodes[0];
734 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
735 if (key.objectid == ino &&
736 key.type == BTRFS_EXTENT_DATA_KEY)
742 leaf = path->nodes[0];
743 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
745 ret = btrfs_next_leaf(root, path);
753 leaf = path->nodes[0];
757 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
759 if (key.objectid > ino)
761 if (WARN_ON_ONCE(key.objectid < ino) ||
762 key.type < BTRFS_EXTENT_DATA_KEY) {
767 if (key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
770 fi = btrfs_item_ptr(leaf, path->slots[0],
771 struct btrfs_file_extent_item);
772 extent_type = btrfs_file_extent_type(leaf, fi);
774 if (extent_type == BTRFS_FILE_EXTENT_REG ||
775 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
776 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
777 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
778 extent_offset = btrfs_file_extent_offset(leaf, fi);
779 extent_end = key.offset +
780 btrfs_file_extent_num_bytes(leaf, fi);
781 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
782 extent_end = key.offset +
783 btrfs_file_extent_inline_len(leaf,
791 * Don't skip extent items representing 0 byte lengths. They
792 * used to be created (bug) if while punching holes we hit
793 * -ENOSPC condition. So if we find one here, just ensure we
794 * delete it, otherwise we would insert a new file extent item
795 * with the same key (offset) as that 0 bytes length file
796 * extent item in the call to setup_items_for_insert() later
799 if (extent_end == key.offset && extent_end >= search_start)
800 goto delete_extent_item;
802 if (extent_end <= search_start) {
808 search_start = max(key.offset, start);
809 if (recow || !modify_tree) {
811 btrfs_release_path(path);
816 * | - range to drop - |
817 * | -------- extent -------- |
819 if (start > key.offset && end < extent_end) {
821 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
826 memcpy(&new_key, &key, sizeof(new_key));
827 new_key.offset = start;
828 ret = btrfs_duplicate_item(trans, root, path,
830 if (ret == -EAGAIN) {
831 btrfs_release_path(path);
837 leaf = path->nodes[0];
838 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
839 struct btrfs_file_extent_item);
840 btrfs_set_file_extent_num_bytes(leaf, fi,
843 fi = btrfs_item_ptr(leaf, path->slots[0],
844 struct btrfs_file_extent_item);
846 extent_offset += start - key.offset;
847 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
848 btrfs_set_file_extent_num_bytes(leaf, fi,
850 btrfs_mark_buffer_dirty(leaf);
852 if (update_refs && disk_bytenr > 0) {
853 ret = btrfs_inc_extent_ref(trans, root,
854 disk_bytenr, num_bytes, 0,
855 root->root_key.objectid,
857 start - extent_offset);
858 BUG_ON(ret); /* -ENOMEM */
863 * | ---- range to drop ----- |
864 * | -------- extent -------- |
866 if (start <= key.offset && end < extent_end) {
867 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
872 memcpy(&new_key, &key, sizeof(new_key));
873 new_key.offset = end;
874 btrfs_set_item_key_safe(root->fs_info, path, &new_key);
876 extent_offset += end - key.offset;
877 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
878 btrfs_set_file_extent_num_bytes(leaf, fi,
880 btrfs_mark_buffer_dirty(leaf);
881 if (update_refs && disk_bytenr > 0)
882 inode_sub_bytes(inode, end - key.offset);
886 search_start = extent_end;
888 * | ---- range to drop ----- |
889 * | -------- extent -------- |
891 if (start > key.offset && end >= extent_end) {
893 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
898 btrfs_set_file_extent_num_bytes(leaf, fi,
900 btrfs_mark_buffer_dirty(leaf);
901 if (update_refs && disk_bytenr > 0)
902 inode_sub_bytes(inode, extent_end - start);
903 if (end == extent_end)
911 * | ---- range to drop ----- |
912 * | ------ extent ------ |
914 if (start <= key.offset && end >= extent_end) {
917 del_slot = path->slots[0];
920 BUG_ON(del_slot + del_nr != path->slots[0]);
925 extent_type == BTRFS_FILE_EXTENT_INLINE) {
926 inode_sub_bytes(inode,
927 extent_end - key.offset);
928 extent_end = ALIGN(extent_end,
930 } else if (update_refs && disk_bytenr > 0) {
931 ret = btrfs_free_extent(trans, root,
932 disk_bytenr, num_bytes, 0,
933 root->root_key.objectid,
934 key.objectid, key.offset -
936 BUG_ON(ret); /* -ENOMEM */
937 inode_sub_bytes(inode,
938 extent_end - key.offset);
941 if (end == extent_end)
944 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
949 ret = btrfs_del_items(trans, root, path, del_slot,
952 btrfs_abort_transaction(trans, root, ret);
959 btrfs_release_path(path);
966 if (!ret && del_nr > 0) {
968 * Set path->slots[0] to first slot, so that after the delete
969 * if items are move off from our leaf to its immediate left or
970 * right neighbor leafs, we end up with a correct and adjusted
971 * path->slots[0] for our insertion (if replace_extent != 0).
973 path->slots[0] = del_slot;
974 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
976 btrfs_abort_transaction(trans, root, ret);
979 leaf = path->nodes[0];
981 * If btrfs_del_items() was called, it might have deleted a leaf, in
982 * which case it unlocked our path, so check path->locks[0] matches a
985 if (!ret && replace_extent && leafs_visited == 1 &&
986 (path->locks[0] == BTRFS_WRITE_LOCK_BLOCKING ||
987 path->locks[0] == BTRFS_WRITE_LOCK) &&
988 btrfs_leaf_free_space(root, leaf) >=
989 sizeof(struct btrfs_item) + extent_item_size) {
992 key.type = BTRFS_EXTENT_DATA_KEY;
994 if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) {
995 struct btrfs_key slot_key;
997 btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]);
998 if (btrfs_comp_cpu_keys(&key, &slot_key) > 0)
1001 setup_items_for_insert(root, path, &key,
1004 sizeof(struct btrfs_item) +
1005 extent_item_size, 1);
1009 if (!replace_extent || !(*key_inserted))
1010 btrfs_release_path(path);
1012 *drop_end = found ? min(end, extent_end) : end;
1016 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
1017 struct btrfs_root *root, struct inode *inode, u64 start,
1018 u64 end, int drop_cache)
1020 struct btrfs_path *path;
1023 path = btrfs_alloc_path();
1026 ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL,
1027 drop_cache, 0, 0, NULL);
1028 btrfs_free_path(path);
1032 static int extent_mergeable(struct extent_buffer *leaf, int slot,
1033 u64 objectid, u64 bytenr, u64 orig_offset,
1034 u64 *start, u64 *end)
1036 struct btrfs_file_extent_item *fi;
1037 struct btrfs_key key;
1040 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
1043 btrfs_item_key_to_cpu(leaf, &key, slot);
1044 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
1047 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1048 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
1049 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
1050 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
1051 btrfs_file_extent_compression(leaf, fi) ||
1052 btrfs_file_extent_encryption(leaf, fi) ||
1053 btrfs_file_extent_other_encoding(leaf, fi))
1056 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1057 if ((*start && *start != key.offset) || (*end && *end != extent_end))
1060 *start = key.offset;
1066 * Mark extent in the range start - end as written.
1068 * This changes extent type from 'pre-allocated' to 'regular'. If only
1069 * part of extent is marked as written, the extent will be split into
1072 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
1073 struct inode *inode, u64 start, u64 end)
1075 struct btrfs_root *root = BTRFS_I(inode)->root;
1076 struct extent_buffer *leaf;
1077 struct btrfs_path *path;
1078 struct btrfs_file_extent_item *fi;
1079 struct btrfs_key key;
1080 struct btrfs_key new_key;
1092 u64 ino = btrfs_ino(inode);
1094 path = btrfs_alloc_path();
1101 key.type = BTRFS_EXTENT_DATA_KEY;
1104 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1107 if (ret > 0 && path->slots[0] > 0)
1110 leaf = path->nodes[0];
1111 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1112 BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY);
1113 fi = btrfs_item_ptr(leaf, path->slots[0],
1114 struct btrfs_file_extent_item);
1115 BUG_ON(btrfs_file_extent_type(leaf, fi) !=
1116 BTRFS_FILE_EXTENT_PREALLOC);
1117 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1118 BUG_ON(key.offset > start || extent_end < end);
1120 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1121 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1122 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
1123 memcpy(&new_key, &key, sizeof(new_key));
1125 if (start == key.offset && end < extent_end) {
1128 if (extent_mergeable(leaf, path->slots[0] - 1,
1129 ino, bytenr, orig_offset,
1130 &other_start, &other_end)) {
1131 new_key.offset = end;
1132 btrfs_set_item_key_safe(root->fs_info, path, &new_key);
1133 fi = btrfs_item_ptr(leaf, path->slots[0],
1134 struct btrfs_file_extent_item);
1135 btrfs_set_file_extent_generation(leaf, fi,
1137 btrfs_set_file_extent_num_bytes(leaf, fi,
1139 btrfs_set_file_extent_offset(leaf, fi,
1141 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1142 struct btrfs_file_extent_item);
1143 btrfs_set_file_extent_generation(leaf, fi,
1145 btrfs_set_file_extent_num_bytes(leaf, fi,
1147 btrfs_mark_buffer_dirty(leaf);
1152 if (start > key.offset && end == extent_end) {
1155 if (extent_mergeable(leaf, path->slots[0] + 1,
1156 ino, bytenr, orig_offset,
1157 &other_start, &other_end)) {
1158 fi = btrfs_item_ptr(leaf, path->slots[0],
1159 struct btrfs_file_extent_item);
1160 btrfs_set_file_extent_num_bytes(leaf, fi,
1161 start - key.offset);
1162 btrfs_set_file_extent_generation(leaf, fi,
1165 new_key.offset = start;
1166 btrfs_set_item_key_safe(root->fs_info, path, &new_key);
1168 fi = btrfs_item_ptr(leaf, path->slots[0],
1169 struct btrfs_file_extent_item);
1170 btrfs_set_file_extent_generation(leaf, fi,
1172 btrfs_set_file_extent_num_bytes(leaf, fi,
1174 btrfs_set_file_extent_offset(leaf, fi,
1175 start - orig_offset);
1176 btrfs_mark_buffer_dirty(leaf);
1181 while (start > key.offset || end < extent_end) {
1182 if (key.offset == start)
1185 new_key.offset = split;
1186 ret = btrfs_duplicate_item(trans, root, path, &new_key);
1187 if (ret == -EAGAIN) {
1188 btrfs_release_path(path);
1192 btrfs_abort_transaction(trans, root, ret);
1196 leaf = path->nodes[0];
1197 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1198 struct btrfs_file_extent_item);
1199 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1200 btrfs_set_file_extent_num_bytes(leaf, fi,
1201 split - key.offset);
1203 fi = btrfs_item_ptr(leaf, path->slots[0],
1204 struct btrfs_file_extent_item);
1206 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1207 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
1208 btrfs_set_file_extent_num_bytes(leaf, fi,
1209 extent_end - split);
1210 btrfs_mark_buffer_dirty(leaf);
1212 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
1213 root->root_key.objectid,
1215 BUG_ON(ret); /* -ENOMEM */
1217 if (split == start) {
1220 BUG_ON(start != key.offset);
1229 if (extent_mergeable(leaf, path->slots[0] + 1,
1230 ino, bytenr, orig_offset,
1231 &other_start, &other_end)) {
1233 btrfs_release_path(path);
1236 extent_end = other_end;
1237 del_slot = path->slots[0] + 1;
1239 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1240 0, root->root_key.objectid,
1242 BUG_ON(ret); /* -ENOMEM */
1246 if (extent_mergeable(leaf, path->slots[0] - 1,
1247 ino, bytenr, orig_offset,
1248 &other_start, &other_end)) {
1250 btrfs_release_path(path);
1253 key.offset = other_start;
1254 del_slot = path->slots[0];
1256 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1257 0, root->root_key.objectid,
1259 BUG_ON(ret); /* -ENOMEM */
1262 fi = btrfs_item_ptr(leaf, path->slots[0],
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_mark_buffer_dirty(leaf);
1269 fi = btrfs_item_ptr(leaf, del_slot - 1,
1270 struct btrfs_file_extent_item);
1271 btrfs_set_file_extent_type(leaf, fi,
1272 BTRFS_FILE_EXTENT_REG);
1273 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1274 btrfs_set_file_extent_num_bytes(leaf, fi,
1275 extent_end - key.offset);
1276 btrfs_mark_buffer_dirty(leaf);
1278 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1280 btrfs_abort_transaction(trans, root, ret);
1285 btrfs_free_path(path);
1290 * on error we return an unlocked page and the error value
1291 * on success we return a locked page and 0
1293 static int prepare_uptodate_page(struct inode *inode,
1294 struct page *page, u64 pos,
1295 bool force_uptodate)
1299 if (((pos & (PAGE_CACHE_SIZE - 1)) || force_uptodate) &&
1300 !PageUptodate(page)) {
1301 ret = btrfs_readpage(NULL, page);
1305 if (!PageUptodate(page)) {
1309 if (page->mapping != inode->i_mapping) {
1318 * this just gets pages into the page cache and locks them down.
1320 static noinline int prepare_pages(struct inode *inode, struct page **pages,
1321 size_t num_pages, loff_t pos,
1322 size_t write_bytes, bool force_uptodate)
1325 unsigned long index = pos >> PAGE_CACHE_SHIFT;
1326 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1330 for (i = 0; i < num_pages; i++) {
1332 pages[i] = find_or_create_page(inode->i_mapping, index + i,
1333 mask | __GFP_WRITE);
1341 err = prepare_uptodate_page(inode, pages[i], pos,
1343 if (!err && i == num_pages - 1)
1344 err = prepare_uptodate_page(inode, pages[i],
1345 pos + write_bytes, false);
1347 page_cache_release(pages[i]);
1348 if (err == -EAGAIN) {
1355 wait_on_page_writeback(pages[i]);
1360 while (faili >= 0) {
1361 unlock_page(pages[faili]);
1362 page_cache_release(pages[faili]);
1370 * This function locks the extent and properly waits for data=ordered extents
1371 * to finish before allowing the pages to be modified if need.
1374 * 1 - the extent is locked
1375 * 0 - the extent is not locked, and everything is OK
1376 * -EAGAIN - need re-prepare the pages
1377 * the other < 0 number - Something wrong happens
1380 lock_and_cleanup_extent_if_need(struct inode *inode, struct page **pages,
1381 size_t num_pages, loff_t pos,
1383 u64 *lockstart, u64 *lockend,
1384 struct extent_state **cached_state)
1386 struct btrfs_root *root = BTRFS_I(inode)->root;
1392 start_pos = round_down(pos, root->sectorsize);
1393 last_pos = start_pos
1394 + round_up(pos + write_bytes - start_pos, root->sectorsize) - 1;
1396 if (start_pos < inode->i_size) {
1397 struct btrfs_ordered_extent *ordered;
1398 lock_extent_bits(&BTRFS_I(inode)->io_tree,
1399 start_pos, last_pos, cached_state);
1400 ordered = btrfs_lookup_ordered_range(inode, start_pos,
1401 last_pos - start_pos + 1);
1403 ordered->file_offset + ordered->len > start_pos &&
1404 ordered->file_offset <= last_pos) {
1405 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1406 start_pos, last_pos,
1407 cached_state, GFP_NOFS);
1408 for (i = 0; i < num_pages; i++) {
1409 unlock_page(pages[i]);
1410 page_cache_release(pages[i]);
1412 btrfs_start_ordered_extent(inode, ordered, 1);
1413 btrfs_put_ordered_extent(ordered);
1417 btrfs_put_ordered_extent(ordered);
1419 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
1420 last_pos, EXTENT_DIRTY | EXTENT_DELALLOC |
1421 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
1422 0, 0, cached_state, GFP_NOFS);
1423 *lockstart = start_pos;
1424 *lockend = last_pos;
1428 for (i = 0; i < num_pages; i++) {
1429 if (clear_page_dirty_for_io(pages[i]))
1430 account_page_redirty(pages[i]);
1431 set_page_extent_mapped(pages[i]);
1432 WARN_ON(!PageLocked(pages[i]));
1438 static noinline int check_can_nocow(struct inode *inode, loff_t pos,
1439 size_t *write_bytes)
1441 struct btrfs_root *root = BTRFS_I(inode)->root;
1442 struct btrfs_ordered_extent *ordered;
1443 u64 lockstart, lockend;
1447 ret = btrfs_start_write_no_snapshoting(root);
1451 lockstart = round_down(pos, root->sectorsize);
1452 lockend = round_up(pos + *write_bytes, root->sectorsize) - 1;
1455 lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1456 ordered = btrfs_lookup_ordered_range(inode, lockstart,
1457 lockend - lockstart + 1);
1461 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1462 btrfs_start_ordered_extent(inode, ordered, 1);
1463 btrfs_put_ordered_extent(ordered);
1466 num_bytes = lockend - lockstart + 1;
1467 ret = can_nocow_extent(inode, lockstart, &num_bytes, NULL, NULL, NULL);
1470 btrfs_end_write_no_snapshoting(root);
1472 *write_bytes = min_t(size_t, *write_bytes ,
1473 num_bytes - pos + lockstart);
1476 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1481 static noinline ssize_t __btrfs_buffered_write(struct file *file,
1485 struct inode *inode = file_inode(file);
1486 struct btrfs_root *root = BTRFS_I(inode)->root;
1487 struct page **pages = NULL;
1488 struct extent_state *cached_state = NULL;
1489 u64 release_bytes = 0;
1492 size_t num_written = 0;
1495 bool only_release_metadata = false;
1496 bool force_page_uptodate = false;
1499 nrptrs = min(DIV_ROUND_UP(iov_iter_count(i), PAGE_CACHE_SIZE),
1500 PAGE_CACHE_SIZE / (sizeof(struct page *)));
1501 nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1502 nrptrs = max(nrptrs, 8);
1503 pages = kmalloc_array(nrptrs, sizeof(struct page *), GFP_KERNEL);
1507 while (iov_iter_count(i) > 0) {
1508 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1509 size_t sector_offset;
1510 size_t write_bytes = min(iov_iter_count(i),
1511 nrptrs * (size_t)PAGE_CACHE_SIZE -
1513 size_t num_pages = DIV_ROUND_UP(write_bytes + offset,
1515 size_t reserve_bytes;
1518 size_t dirty_sectors;
1521 WARN_ON(num_pages > nrptrs);
1524 * Fault pages before locking them in prepare_pages
1525 * to avoid recursive lock
1527 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1532 sector_offset = pos & (root->sectorsize - 1);
1533 reserve_bytes = round_up(write_bytes + sector_offset,
1536 if ((BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW |
1537 BTRFS_INODE_PREALLOC)) &&
1538 check_can_nocow(inode, pos, &write_bytes) > 0) {
1540 * For nodata cow case, no need to reserve
1543 only_release_metadata = true;
1545 * our prealloc extent may be smaller than
1546 * write_bytes, so scale down.
1548 num_pages = DIV_ROUND_UP(write_bytes + offset,
1550 reserve_bytes = round_up(write_bytes + sector_offset,
1552 goto reserve_metadata;
1555 ret = btrfs_check_data_free_space(inode, pos, write_bytes);
1560 ret = btrfs_delalloc_reserve_metadata(inode, reserve_bytes);
1562 if (!only_release_metadata)
1563 btrfs_free_reserved_data_space(inode, pos,
1566 btrfs_end_write_no_snapshoting(root);
1570 release_bytes = reserve_bytes;
1571 need_unlock = false;
1574 * This is going to setup the pages array with the number of
1575 * pages we want, so we don't really need to worry about the
1576 * contents of pages from loop to loop
1578 ret = prepare_pages(inode, pages, num_pages,
1580 force_page_uptodate);
1584 ret = lock_and_cleanup_extent_if_need(inode, pages, num_pages,
1585 pos, write_bytes, &lockstart,
1586 &lockend, &cached_state);
1591 } else if (ret > 0) {
1596 copied = btrfs_copy_from_user(pos, write_bytes, pages, i);
1599 * if we have trouble faulting in the pages, fall
1600 * back to one page at a time
1602 if (copied < write_bytes)
1606 force_page_uptodate = true;
1609 force_page_uptodate = false;
1610 dirty_pages = DIV_ROUND_UP(copied + offset,
1615 * If we had a short copy we need to release the excess delaloc
1616 * bytes we reserved. We need to increment outstanding_extents
1617 * because btrfs_delalloc_release_space will decrement it, but
1618 * we still have an outstanding extent for the chunk we actually
1621 num_sectors = BTRFS_BYTES_TO_BLKS(root->fs_info,
1623 dirty_sectors = round_up(copied + sector_offset,
1625 dirty_sectors = BTRFS_BYTES_TO_BLKS(root->fs_info,
1628 if (num_sectors > dirty_sectors) {
1629 release_bytes = (write_bytes - copied)
1630 & ~((u64)root->sectorsize - 1);
1632 spin_lock(&BTRFS_I(inode)->lock);
1633 BTRFS_I(inode)->outstanding_extents++;
1634 spin_unlock(&BTRFS_I(inode)->lock);
1636 if (only_release_metadata) {
1637 btrfs_delalloc_release_metadata(inode,
1642 __pos = round_down(pos, root->sectorsize) +
1643 (dirty_pages << PAGE_CACHE_SHIFT);
1644 btrfs_delalloc_release_space(inode, __pos,
1649 release_bytes = round_up(copied + sector_offset,
1653 ret = btrfs_dirty_pages(root, inode, pages,
1654 dirty_pages, pos, copied,
1657 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1658 lockstart, lockend, &cached_state,
1661 btrfs_drop_pages(pages, num_pages);
1666 if (only_release_metadata)
1667 btrfs_end_write_no_snapshoting(root);
1669 if (only_release_metadata && copied > 0) {
1670 lockstart = round_down(pos, root->sectorsize);
1671 lockend = round_up(pos + copied, root->sectorsize) - 1;
1673 set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
1674 lockend, EXTENT_NORESERVE, NULL,
1676 only_release_metadata = false;
1679 btrfs_drop_pages(pages, num_pages);
1683 balance_dirty_pages_ratelimited(inode->i_mapping);
1684 if (dirty_pages < (root->nodesize >> PAGE_CACHE_SHIFT) + 1)
1685 btrfs_btree_balance_dirty(root);
1688 num_written += copied;
1693 if (release_bytes) {
1694 if (only_release_metadata) {
1695 btrfs_end_write_no_snapshoting(root);
1696 btrfs_delalloc_release_metadata(inode, release_bytes);
1698 btrfs_delalloc_release_space(inode, pos, release_bytes);
1702 return num_written ? num_written : ret;
1705 static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1706 struct iov_iter *from,
1709 struct file *file = iocb->ki_filp;
1710 struct inode *inode = file_inode(file);
1712 ssize_t written_buffered;
1716 written = generic_file_direct_write(iocb, from, pos);
1718 if (written < 0 || !iov_iter_count(from))
1722 written_buffered = __btrfs_buffered_write(file, from, pos);
1723 if (written_buffered < 0) {
1724 err = written_buffered;
1728 * Ensure all data is persisted. We want the next direct IO read to be
1729 * able to read what was just written.
1731 endbyte = pos + written_buffered - 1;
1732 err = btrfs_fdatawrite_range(inode, pos, endbyte);
1735 err = filemap_fdatawait_range(inode->i_mapping, pos, endbyte);
1738 written += written_buffered;
1739 iocb->ki_pos = pos + written_buffered;
1740 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
1741 endbyte >> PAGE_CACHE_SHIFT);
1743 return written ? written : err;
1746 static void update_time_for_write(struct inode *inode)
1748 struct timespec now;
1750 if (IS_NOCMTIME(inode))
1753 now = current_fs_time(inode->i_sb);
1754 if (!timespec_equal(&inode->i_mtime, &now))
1755 inode->i_mtime = now;
1757 if (!timespec_equal(&inode->i_ctime, &now))
1758 inode->i_ctime = now;
1760 if (IS_I_VERSION(inode))
1761 inode_inc_iversion(inode);
1764 static ssize_t btrfs_file_write_iter(struct kiocb *iocb,
1765 struct iov_iter *from)
1767 struct file *file = iocb->ki_filp;
1768 struct inode *inode = file_inode(file);
1769 struct btrfs_root *root = BTRFS_I(inode)->root;
1772 ssize_t num_written = 0;
1773 bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host);
1781 err = generic_write_checks(iocb, from);
1783 inode_unlock(inode);
1787 current->backing_dev_info = inode_to_bdi(inode);
1788 err = file_remove_privs(file);
1790 inode_unlock(inode);
1795 * If BTRFS flips readonly due to some impossible error
1796 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1797 * although we have opened a file as writable, we have
1798 * to stop this write operation to ensure FS consistency.
1800 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
1801 inode_unlock(inode);
1807 * We reserve space for updating the inode when we reserve space for the
1808 * extent we are going to write, so we will enospc out there. We don't
1809 * need to start yet another transaction to update the inode as we will
1810 * update the inode when we finish writing whatever data we write.
1812 update_time_for_write(inode);
1815 count = iov_iter_count(from);
1816 start_pos = round_down(pos, root->sectorsize);
1817 oldsize = i_size_read(inode);
1818 if (start_pos > oldsize) {
1819 /* Expand hole size to cover write data, preventing empty gap */
1820 end_pos = round_up(pos + count, root->sectorsize);
1821 err = btrfs_cont_expand(inode, oldsize, end_pos);
1823 inode_unlock(inode);
1826 if (start_pos > round_up(oldsize, root->sectorsize))
1831 atomic_inc(&BTRFS_I(inode)->sync_writers);
1833 if (iocb->ki_flags & IOCB_DIRECT) {
1834 num_written = __btrfs_direct_write(iocb, from, pos);
1836 num_written = __btrfs_buffered_write(file, from, pos);
1837 if (num_written > 0)
1838 iocb->ki_pos = pos + num_written;
1840 pagecache_isize_extended(inode, oldsize,
1841 i_size_read(inode));
1844 inode_unlock(inode);
1847 * We also have to set last_sub_trans to the current log transid,
1848 * otherwise subsequent syncs to a file that's been synced in this
1849 * transaction will appear to have already occured.
1851 spin_lock(&BTRFS_I(inode)->lock);
1852 BTRFS_I(inode)->last_sub_trans = root->log_transid;
1853 spin_unlock(&BTRFS_I(inode)->lock);
1854 if (num_written > 0) {
1855 err = generic_write_sync(file, pos, num_written);
1861 atomic_dec(&BTRFS_I(inode)->sync_writers);
1863 current->backing_dev_info = NULL;
1864 return num_written ? num_written : err;
1867 int btrfs_release_file(struct inode *inode, struct file *filp)
1869 if (filp->private_data)
1870 btrfs_ioctl_trans_end(filp);
1872 * ordered_data_close is set by settattr when we are about to truncate
1873 * a file from a non-zero size to a zero size. This tries to
1874 * flush down new bytes that may have been written if the
1875 * application were using truncate to replace a file in place.
1877 if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
1878 &BTRFS_I(inode)->runtime_flags))
1879 filemap_flush(inode->i_mapping);
1883 static int start_ordered_ops(struct inode *inode, loff_t start, loff_t end)
1887 atomic_inc(&BTRFS_I(inode)->sync_writers);
1888 ret = btrfs_fdatawrite_range(inode, start, end);
1889 atomic_dec(&BTRFS_I(inode)->sync_writers);
1895 * fsync call for both files and directories. This logs the inode into
1896 * the tree log instead of forcing full commits whenever possible.
1898 * It needs to call filemap_fdatawait so that all ordered extent updates are
1899 * in the metadata btree are up to date for copying to the log.
1901 * It drops the inode mutex before doing the tree log commit. This is an
1902 * important optimization for directories because holding the mutex prevents
1903 * new operations on the dir while we write to disk.
1905 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1907 struct dentry *dentry = file->f_path.dentry;
1908 struct inode *inode = d_inode(dentry);
1909 struct btrfs_root *root = BTRFS_I(inode)->root;
1910 struct btrfs_trans_handle *trans;
1911 struct btrfs_log_ctx ctx;
1917 * The range length can be represented by u64, we have to do the typecasts
1918 * to avoid signed overflow if it's [0, LLONG_MAX] eg. from fsync()
1920 len = (u64)end - (u64)start + 1;
1921 trace_btrfs_sync_file(file, datasync);
1924 * We write the dirty pages in the range and wait until they complete
1925 * out of the ->i_mutex. If so, we can flush the dirty pages by
1926 * multi-task, and make the performance up. See
1927 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1929 ret = start_ordered_ops(inode, start, end);
1934 atomic_inc(&root->log_batch);
1935 full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1936 &BTRFS_I(inode)->runtime_flags);
1938 * We might have have had more pages made dirty after calling
1939 * start_ordered_ops and before acquiring the inode's i_mutex.
1943 * For a full sync, we need to make sure any ordered operations
1944 * start and finish before we start logging the inode, so that
1945 * all extents are persisted and the respective file extent
1946 * items are in the fs/subvol btree.
1948 ret = btrfs_wait_ordered_range(inode, start, len);
1951 * Start any new ordered operations before starting to log the
1952 * inode. We will wait for them to finish in btrfs_sync_log().
1954 * Right before acquiring the inode's mutex, we might have new
1955 * writes dirtying pages, which won't immediately start the
1956 * respective ordered operations - that is done through the
1957 * fill_delalloc callbacks invoked from the writepage and
1958 * writepages address space operations. So make sure we start
1959 * all ordered operations before starting to log our inode. Not
1960 * doing this means that while logging the inode, writeback
1961 * could start and invoke writepage/writepages, which would call
1962 * the fill_delalloc callbacks (cow_file_range,
1963 * submit_compressed_extents). These callbacks add first an
1964 * extent map to the modified list of extents and then create
1965 * the respective ordered operation, which means in
1966 * tree-log.c:btrfs_log_inode() we might capture all existing
1967 * ordered operations (with btrfs_get_logged_extents()) before
1968 * the fill_delalloc callback adds its ordered operation, and by
1969 * the time we visit the modified list of extent maps (with
1970 * btrfs_log_changed_extents()), we see and process the extent
1971 * map they created. We then use the extent map to construct a
1972 * file extent item for logging without waiting for the
1973 * respective ordered operation to finish - this file extent
1974 * item points to a disk location that might not have yet been
1975 * written to, containing random data - so after a crash a log
1976 * replay will make our inode have file extent items that point
1977 * to disk locations containing invalid data, as we returned
1978 * success to userspace without waiting for the respective
1979 * ordered operation to finish, because it wasn't captured by
1980 * btrfs_get_logged_extents().
1982 ret = start_ordered_ops(inode, start, end);
1985 inode_unlock(inode);
1988 atomic_inc(&root->log_batch);
1991 * If the last transaction that changed this file was before the current
1992 * transaction and we have the full sync flag set in our inode, we can
1993 * bail out now without any syncing.
1995 * Note that we can't bail out if the full sync flag isn't set. This is
1996 * because when the full sync flag is set we start all ordered extents
1997 * and wait for them to fully complete - when they complete they update
1998 * the inode's last_trans field through:
2000 * btrfs_finish_ordered_io() ->
2001 * btrfs_update_inode_fallback() ->
2002 * btrfs_update_inode() ->
2003 * btrfs_set_inode_last_trans()
2005 * So we are sure that last_trans is up to date and can do this check to
2006 * bail out safely. For the fast path, when the full sync flag is not
2007 * set in our inode, we can not do it because we start only our ordered
2008 * extents and don't wait for them to complete (that is when
2009 * btrfs_finish_ordered_io runs), so here at this point their last_trans
2010 * value might be less than or equals to fs_info->last_trans_committed,
2011 * and setting a speculative last_trans for an inode when a buffered
2012 * write is made (such as fs_info->generation + 1 for example) would not
2013 * be reliable since after setting the value and before fsync is called
2014 * any number of transactions can start and commit (transaction kthread
2015 * commits the current transaction periodically), and a transaction
2016 * commit does not start nor waits for ordered extents to complete.
2019 if (btrfs_inode_in_log(inode, root->fs_info->generation) ||
2020 (BTRFS_I(inode)->last_trans <=
2021 root->fs_info->last_trans_committed &&
2023 !btrfs_have_ordered_extents_in_range(inode, start, len)))) {
2025 * We'v had everything committed since the last time we were
2026 * modified so clear this flag in case it was set for whatever
2027 * reason, it's no longer relevant.
2029 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2030 &BTRFS_I(inode)->runtime_flags);
2031 inode_unlock(inode);
2036 * ok we haven't committed the transaction yet, lets do a commit
2038 if (file->private_data)
2039 btrfs_ioctl_trans_end(file);
2042 * We use start here because we will need to wait on the IO to complete
2043 * in btrfs_sync_log, which could require joining a transaction (for
2044 * example checking cross references in the nocow path). If we use join
2045 * here we could get into a situation where we're waiting on IO to
2046 * happen that is blocked on a transaction trying to commit. With start
2047 * we inc the extwriter counter, so we wait for all extwriters to exit
2048 * before we start blocking join'ers. This comment is to keep somebody
2049 * from thinking they are super smart and changing this to
2050 * btrfs_join_transaction *cough*Josef*cough*.
2052 trans = btrfs_start_transaction(root, 0);
2053 if (IS_ERR(trans)) {
2054 ret = PTR_ERR(trans);
2055 inode_unlock(inode);
2060 btrfs_init_log_ctx(&ctx);
2062 ret = btrfs_log_dentry_safe(trans, root, dentry, start, end, &ctx);
2064 /* Fallthrough and commit/free transaction. */
2068 /* we've logged all the items and now have a consistent
2069 * version of the file in the log. It is possible that
2070 * someone will come in and modify the file, but that's
2071 * fine because the log is consistent on disk, and we
2072 * have references to all of the file's extents
2074 * It is possible that someone will come in and log the
2075 * file again, but that will end up using the synchronization
2076 * inside btrfs_sync_log to keep things safe.
2078 inode_unlock(inode);
2081 * If any of the ordered extents had an error, just return it to user
2082 * space, so that the application knows some writes didn't succeed and
2083 * can take proper action (retry for e.g.). Blindly committing the
2084 * transaction in this case, would fool userspace that everything was
2085 * successful. And we also want to make sure our log doesn't contain
2086 * file extent items pointing to extents that weren't fully written to -
2087 * just like in the non fast fsync path, where we check for the ordered
2088 * operation's error flag before writing to the log tree and return -EIO
2089 * if any of them had this flag set (btrfs_wait_ordered_range) -
2090 * therefore we need to check for errors in the ordered operations,
2091 * which are indicated by ctx.io_err.
2094 btrfs_end_transaction(trans, root);
2099 if (ret != BTRFS_NO_LOG_SYNC) {
2101 ret = btrfs_sync_log(trans, root, &ctx);
2103 ret = btrfs_end_transaction(trans, root);
2108 ret = btrfs_wait_ordered_range(inode, start, len);
2110 btrfs_end_transaction(trans, root);
2114 ret = btrfs_commit_transaction(trans, root);
2116 ret = btrfs_end_transaction(trans, root);
2119 return ret > 0 ? -EIO : ret;
2122 static const struct vm_operations_struct btrfs_file_vm_ops = {
2123 .fault = filemap_fault,
2124 .map_pages = filemap_map_pages,
2125 .page_mkwrite = btrfs_page_mkwrite,
2128 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
2130 struct address_space *mapping = filp->f_mapping;
2132 if (!mapping->a_ops->readpage)
2135 file_accessed(filp);
2136 vma->vm_ops = &btrfs_file_vm_ops;
2141 static int hole_mergeable(struct inode *inode, struct extent_buffer *leaf,
2142 int slot, u64 start, u64 end)
2144 struct btrfs_file_extent_item *fi;
2145 struct btrfs_key key;
2147 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
2150 btrfs_item_key_to_cpu(leaf, &key, slot);
2151 if (key.objectid != btrfs_ino(inode) ||
2152 key.type != BTRFS_EXTENT_DATA_KEY)
2155 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2157 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2160 if (btrfs_file_extent_disk_bytenr(leaf, fi))
2163 if (key.offset == end)
2165 if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
2170 static int fill_holes(struct btrfs_trans_handle *trans, struct inode *inode,
2171 struct btrfs_path *path, u64 offset, u64 end)
2173 struct btrfs_root *root = BTRFS_I(inode)->root;
2174 struct extent_buffer *leaf;
2175 struct btrfs_file_extent_item *fi;
2176 struct extent_map *hole_em;
2177 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2178 struct btrfs_key key;
2181 if (btrfs_fs_incompat(root->fs_info, NO_HOLES))
2184 key.objectid = btrfs_ino(inode);
2185 key.type = BTRFS_EXTENT_DATA_KEY;
2186 key.offset = offset;
2188 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2193 leaf = path->nodes[0];
2194 if (hole_mergeable(inode, leaf, path->slots[0]-1, offset, end)) {
2198 fi = btrfs_item_ptr(leaf, path->slots[0],
2199 struct btrfs_file_extent_item);
2200 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2202 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2203 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2204 btrfs_set_file_extent_offset(leaf, fi, 0);
2205 btrfs_mark_buffer_dirty(leaf);
2209 if (hole_mergeable(inode, leaf, path->slots[0], offset, end)) {
2212 key.offset = offset;
2213 btrfs_set_item_key_safe(root->fs_info, path, &key);
2214 fi = btrfs_item_ptr(leaf, path->slots[0],
2215 struct btrfs_file_extent_item);
2216 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2218 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2219 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2220 btrfs_set_file_extent_offset(leaf, fi, 0);
2221 btrfs_mark_buffer_dirty(leaf);
2224 btrfs_release_path(path);
2226 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
2227 0, 0, end - offset, 0, end - offset,
2233 btrfs_release_path(path);
2235 hole_em = alloc_extent_map();
2237 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2238 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2239 &BTRFS_I(inode)->runtime_flags);
2241 hole_em->start = offset;
2242 hole_em->len = end - offset;
2243 hole_em->ram_bytes = hole_em->len;
2244 hole_em->orig_start = offset;
2246 hole_em->block_start = EXTENT_MAP_HOLE;
2247 hole_em->block_len = 0;
2248 hole_em->orig_block_len = 0;
2249 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
2250 hole_em->compress_type = BTRFS_COMPRESS_NONE;
2251 hole_em->generation = trans->transid;
2254 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2255 write_lock(&em_tree->lock);
2256 ret = add_extent_mapping(em_tree, hole_em, 1);
2257 write_unlock(&em_tree->lock);
2258 } while (ret == -EEXIST);
2259 free_extent_map(hole_em);
2261 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2262 &BTRFS_I(inode)->runtime_flags);
2269 * Find a hole extent on given inode and change start/len to the end of hole
2270 * extent.(hole/vacuum extent whose em->start <= start &&
2271 * em->start + em->len > start)
2272 * When a hole extent is found, return 1 and modify start/len.
2274 static int find_first_non_hole(struct inode *inode, u64 *start, u64 *len)
2276 struct extent_map *em;
2279 em = btrfs_get_extent(inode, NULL, 0, *start, *len, 0);
2280 if (IS_ERR_OR_NULL(em)) {
2288 /* Hole or vacuum extent(only exists in no-hole mode) */
2289 if (em->block_start == EXTENT_MAP_HOLE) {
2291 *len = em->start + em->len > *start + *len ?
2292 0 : *start + *len - em->start - em->len;
2293 *start = em->start + em->len;
2295 free_extent_map(em);
2299 static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
2301 struct btrfs_root *root = BTRFS_I(inode)->root;
2302 struct extent_state *cached_state = NULL;
2303 struct btrfs_path *path;
2304 struct btrfs_block_rsv *rsv;
2305 struct btrfs_trans_handle *trans;
2310 u64 orig_start = offset;
2312 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
2316 unsigned int rsv_count;
2318 bool no_holes = btrfs_fs_incompat(root->fs_info, NO_HOLES);
2320 bool truncated_block = false;
2321 bool updated_inode = false;
2323 ret = btrfs_wait_ordered_range(inode, offset, len);
2328 ino_size = round_up(inode->i_size, root->sectorsize);
2329 ret = find_first_non_hole(inode, &offset, &len);
2331 goto out_only_mutex;
2333 /* Already in a large hole */
2335 goto out_only_mutex;
2338 lockstart = round_up(offset, BTRFS_I(inode)->root->sectorsize);
2339 lockend = round_down(offset + len,
2340 BTRFS_I(inode)->root->sectorsize) - 1;
2341 same_block = (BTRFS_BYTES_TO_BLKS(root->fs_info, offset))
2342 == (BTRFS_BYTES_TO_BLKS(root->fs_info, offset + len - 1));
2344 * We needn't truncate any block which is beyond the end of the file
2345 * because we are sure there is no data there.
2348 * Only do this if we are in the same block and we aren't doing the
2351 if (same_block && len < root->sectorsize) {
2352 if (offset < ino_size) {
2353 truncated_block = true;
2354 ret = btrfs_truncate_block(inode, offset, len, 0);
2358 goto out_only_mutex;
2361 /* zero back part of the first block */
2362 if (offset < ino_size) {
2363 truncated_block = true;
2364 ret = btrfs_truncate_block(inode, offset, 0, 0);
2366 inode_unlock(inode);
2371 /* Check the aligned pages after the first unaligned page,
2372 * if offset != orig_start, which means the first unaligned page
2373 * including serveral following pages are already in holes,
2374 * the extra check can be skipped */
2375 if (offset == orig_start) {
2376 /* after truncate page, check hole again */
2377 len = offset + len - lockstart;
2379 ret = find_first_non_hole(inode, &offset, &len);
2381 goto out_only_mutex;
2384 goto out_only_mutex;
2389 /* Check the tail unaligned part is in a hole */
2390 tail_start = lockend + 1;
2391 tail_len = offset + len - tail_start;
2393 ret = find_first_non_hole(inode, &tail_start, &tail_len);
2394 if (unlikely(ret < 0))
2395 goto out_only_mutex;
2397 /* zero the front end of the last page */
2398 if (tail_start + tail_len < ino_size) {
2399 truncated_block = true;
2400 ret = btrfs_truncate_block(inode,
2401 tail_start + tail_len,
2404 goto out_only_mutex;
2409 if (lockend < lockstart) {
2411 goto out_only_mutex;
2415 struct btrfs_ordered_extent *ordered;
2417 truncate_pagecache_range(inode, lockstart, lockend);
2419 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2421 ordered = btrfs_lookup_first_ordered_extent(inode, lockend);
2424 * We need to make sure we have no ordered extents in this range
2425 * and nobody raced in and read a page in this range, if we did
2426 * we need to try again.
2429 (ordered->file_offset + ordered->len <= lockstart ||
2430 ordered->file_offset > lockend)) &&
2431 !btrfs_page_exists_in_range(inode, lockstart, lockend)) {
2433 btrfs_put_ordered_extent(ordered);
2437 btrfs_put_ordered_extent(ordered);
2438 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
2439 lockend, &cached_state, GFP_NOFS);
2440 ret = btrfs_wait_ordered_range(inode, lockstart,
2441 lockend - lockstart + 1);
2443 inode_unlock(inode);
2448 path = btrfs_alloc_path();
2454 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2459 rsv->size = btrfs_calc_trunc_metadata_size(root, 1);
2463 * 1 - update the inode
2464 * 1 - removing the extents in the range
2465 * 1 - adding the hole extent if no_holes isn't set
2467 rsv_count = no_holes ? 2 : 3;
2468 trans = btrfs_start_transaction(root, rsv_count);
2469 if (IS_ERR(trans)) {
2470 err = PTR_ERR(trans);
2474 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
2477 trans->block_rsv = rsv;
2479 cur_offset = lockstart;
2480 len = lockend - cur_offset;
2481 while (cur_offset < lockend) {
2482 ret = __btrfs_drop_extents(trans, root, inode, path,
2483 cur_offset, lockend + 1,
2484 &drop_end, 1, 0, 0, NULL);
2488 trans->block_rsv = &root->fs_info->trans_block_rsv;
2490 if (cur_offset < ino_size) {
2491 ret = fill_holes(trans, inode, path, cur_offset,
2499 cur_offset = drop_end;
2501 ret = btrfs_update_inode(trans, root, inode);
2507 btrfs_end_transaction(trans, root);
2508 btrfs_btree_balance_dirty(root);
2510 trans = btrfs_start_transaction(root, rsv_count);
2511 if (IS_ERR(trans)) {
2512 ret = PTR_ERR(trans);
2517 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
2519 BUG_ON(ret); /* shouldn't happen */
2520 trans->block_rsv = rsv;
2522 ret = find_first_non_hole(inode, &cur_offset, &len);
2523 if (unlikely(ret < 0))
2536 trans->block_rsv = &root->fs_info->trans_block_rsv;
2538 * If we are using the NO_HOLES feature we might have had already an
2539 * hole that overlaps a part of the region [lockstart, lockend] and
2540 * ends at (or beyond) lockend. Since we have no file extent items to
2541 * represent holes, drop_end can be less than lockend and so we must
2542 * make sure we have an extent map representing the existing hole (the
2543 * call to __btrfs_drop_extents() might have dropped the existing extent
2544 * map representing the existing hole), otherwise the fast fsync path
2545 * will not record the existence of the hole region
2546 * [existing_hole_start, lockend].
2548 if (drop_end <= lockend)
2549 drop_end = lockend + 1;
2551 * Don't insert file hole extent item if it's for a range beyond eof
2552 * (because it's useless) or if it represents a 0 bytes range (when
2553 * cur_offset == drop_end).
2555 if (cur_offset < ino_size && cur_offset < drop_end) {
2556 ret = fill_holes(trans, inode, path, cur_offset, drop_end);
2567 inode_inc_iversion(inode);
2568 inode->i_mtime = inode->i_ctime = current_fs_time(inode->i_sb);
2570 trans->block_rsv = &root->fs_info->trans_block_rsv;
2571 ret = btrfs_update_inode(trans, root, inode);
2572 updated_inode = true;
2573 btrfs_end_transaction(trans, root);
2574 btrfs_btree_balance_dirty(root);
2576 btrfs_free_path(path);
2577 btrfs_free_block_rsv(root, rsv);
2579 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2580 &cached_state, GFP_NOFS);
2582 if (!updated_inode && truncated_block && !ret && !err) {
2584 * If we only end up zeroing part of a page, we still need to
2585 * update the inode item, so that all the time fields are
2586 * updated as well as the necessary btrfs inode in memory fields
2587 * for detecting, at fsync time, if the inode isn't yet in the
2588 * log tree or it's there but not up to date.
2590 trans = btrfs_start_transaction(root, 1);
2591 if (IS_ERR(trans)) {
2592 err = PTR_ERR(trans);
2594 err = btrfs_update_inode(trans, root, inode);
2595 ret = btrfs_end_transaction(trans, root);
2598 inode_unlock(inode);
2604 /* Helper structure to record which range is already reserved */
2605 struct falloc_range {
2606 struct list_head list;
2612 * Helper function to add falloc range
2614 * Caller should have locked the larger range of extent containing
2617 static int add_falloc_range(struct list_head *head, u64 start, u64 len)
2619 struct falloc_range *prev = NULL;
2620 struct falloc_range *range = NULL;
2622 if (list_empty(head))
2626 * As fallocate iterate by bytenr order, we only need to check
2629 prev = list_entry(head->prev, struct falloc_range, list);
2630 if (prev->start + prev->len == start) {
2635 range = kmalloc(sizeof(*range), GFP_KERNEL);
2638 range->start = start;
2640 list_add_tail(&range->list, head);
2644 static long btrfs_fallocate(struct file *file, int mode,
2645 loff_t offset, loff_t len)
2647 struct inode *inode = file_inode(file);
2648 struct extent_state *cached_state = NULL;
2649 struct falloc_range *range;
2650 struct falloc_range *tmp;
2651 struct list_head reserve_list;
2659 struct extent_map *em;
2660 int blocksize = BTRFS_I(inode)->root->sectorsize;
2663 alloc_start = round_down(offset, blocksize);
2664 alloc_end = round_up(offset + len, blocksize);
2666 /* Make sure we aren't being give some crap mode */
2667 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2670 if (mode & FALLOC_FL_PUNCH_HOLE)
2671 return btrfs_punch_hole(inode, offset, len);
2674 * Only trigger disk allocation, don't trigger qgroup reserve
2676 * For qgroup space, it will be checked later.
2678 ret = btrfs_alloc_data_chunk_ondemand(inode, alloc_end - alloc_start);
2683 ret = inode_newsize_ok(inode, alloc_end);
2688 * TODO: Move these two operations after we have checked
2689 * accurate reserved space, or fallocate can still fail but
2690 * with page truncated or size expanded.
2692 * But that's a minor problem and won't do much harm BTW.
2694 if (alloc_start > inode->i_size) {
2695 ret = btrfs_cont_expand(inode, i_size_read(inode),
2699 } else if (offset + len > inode->i_size) {
2701 * If we are fallocating from the end of the file onward we
2702 * need to zero out the end of the block if i_size lands in the
2703 * middle of a block.
2705 ret = btrfs_truncate_block(inode, inode->i_size, 0, 0);
2711 * wait for ordered IO before we have any locks. We'll loop again
2712 * below with the locks held.
2714 ret = btrfs_wait_ordered_range(inode, alloc_start,
2715 alloc_end - alloc_start);
2719 locked_end = alloc_end - 1;
2721 struct btrfs_ordered_extent *ordered;
2723 /* the extent lock is ordered inside the running
2726 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
2727 locked_end, &cached_state);
2728 ordered = btrfs_lookup_first_ordered_extent(inode,
2731 ordered->file_offset + ordered->len > alloc_start &&
2732 ordered->file_offset < alloc_end) {
2733 btrfs_put_ordered_extent(ordered);
2734 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
2735 alloc_start, locked_end,
2736 &cached_state, GFP_KERNEL);
2738 * we can't wait on the range with the transaction
2739 * running or with the extent lock held
2741 ret = btrfs_wait_ordered_range(inode, alloc_start,
2742 alloc_end - alloc_start);
2747 btrfs_put_ordered_extent(ordered);
2752 /* First, check if we exceed the qgroup limit */
2753 INIT_LIST_HEAD(&reserve_list);
2754 cur_offset = alloc_start;
2756 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2757 alloc_end - cur_offset, 0);
2758 if (IS_ERR_OR_NULL(em)) {
2765 last_byte = min(extent_map_end(em), alloc_end);
2766 actual_end = min_t(u64, extent_map_end(em), offset + len);
2767 last_byte = ALIGN(last_byte, blocksize);
2768 if (em->block_start == EXTENT_MAP_HOLE ||
2769 (cur_offset >= inode->i_size &&
2770 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
2771 ret = add_falloc_range(&reserve_list, cur_offset,
2772 last_byte - cur_offset);
2774 free_extent_map(em);
2777 ret = btrfs_qgroup_reserve_data(inode, cur_offset,
2778 last_byte - cur_offset);
2782 free_extent_map(em);
2783 cur_offset = last_byte;
2784 if (cur_offset >= alloc_end)
2789 * If ret is still 0, means we're OK to fallocate.
2790 * Or just cleanup the list and exit.
2792 list_for_each_entry_safe(range, tmp, &reserve_list, list) {
2794 ret = btrfs_prealloc_file_range(inode, mode,
2796 range->len, 1 << inode->i_blkbits,
2797 offset + len, &alloc_hint);
2798 list_del(&range->list);
2804 if (actual_end > inode->i_size &&
2805 !(mode & FALLOC_FL_KEEP_SIZE)) {
2806 struct btrfs_trans_handle *trans;
2807 struct btrfs_root *root = BTRFS_I(inode)->root;
2810 * We didn't need to allocate any more space, but we
2811 * still extended the size of the file so we need to
2812 * update i_size and the inode item.
2814 trans = btrfs_start_transaction(root, 1);
2815 if (IS_ERR(trans)) {
2816 ret = PTR_ERR(trans);
2818 inode->i_ctime = current_fs_time(inode->i_sb);
2819 i_size_write(inode, actual_end);
2820 btrfs_ordered_update_i_size(inode, actual_end, NULL);
2821 ret = btrfs_update_inode(trans, root, inode);
2823 btrfs_end_transaction(trans, root);
2825 ret = btrfs_end_transaction(trans, root);
2829 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
2830 &cached_state, GFP_KERNEL);
2833 * As we waited the extent range, the data_rsv_map must be empty
2834 * in the range, as written data range will be released from it.
2835 * And for prealloacted extent, it will also be released when
2836 * its metadata is written.
2837 * So this is completely used as cleanup.
2839 btrfs_qgroup_free_data(inode, alloc_start, alloc_end - alloc_start);
2840 inode_unlock(inode);
2841 /* Let go of our reservation. */
2842 btrfs_free_reserved_data_space(inode, alloc_start,
2843 alloc_end - alloc_start);
2847 static int find_desired_extent(struct inode *inode, loff_t *offset, int whence)
2849 struct btrfs_root *root = BTRFS_I(inode)->root;
2850 struct extent_map *em = NULL;
2851 struct extent_state *cached_state = NULL;
2858 if (inode->i_size == 0)
2862 * *offset can be negative, in this case we start finding DATA/HOLE from
2863 * the very start of the file.
2865 start = max_t(loff_t, 0, *offset);
2867 lockstart = round_down(start, root->sectorsize);
2868 lockend = round_up(i_size_read(inode), root->sectorsize);
2869 if (lockend <= lockstart)
2870 lockend = lockstart + root->sectorsize;
2872 len = lockend - lockstart + 1;
2874 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2877 while (start < inode->i_size) {
2878 em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0);
2885 if (whence == SEEK_HOLE &&
2886 (em->block_start == EXTENT_MAP_HOLE ||
2887 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2889 else if (whence == SEEK_DATA &&
2890 (em->block_start != EXTENT_MAP_HOLE &&
2891 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2894 start = em->start + em->len;
2895 free_extent_map(em);
2899 free_extent_map(em);
2901 if (whence == SEEK_DATA && start >= inode->i_size)
2904 *offset = min_t(loff_t, start, inode->i_size);
2906 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2907 &cached_state, GFP_NOFS);
2911 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
2913 struct inode *inode = file->f_mapping->host;
2920 offset = generic_file_llseek(file, offset, whence);
2924 if (offset >= i_size_read(inode)) {
2925 inode_unlock(inode);
2929 ret = find_desired_extent(inode, &offset, whence);
2931 inode_unlock(inode);
2936 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2938 inode_unlock(inode);
2942 const struct file_operations btrfs_file_operations = {
2943 .llseek = btrfs_file_llseek,
2944 .read_iter = generic_file_read_iter,
2945 .splice_read = generic_file_splice_read,
2946 .write_iter = btrfs_file_write_iter,
2947 .mmap = btrfs_file_mmap,
2948 .open = generic_file_open,
2949 .release = btrfs_release_file,
2950 .fsync = btrfs_sync_file,
2951 .fallocate = btrfs_fallocate,
2952 .unlocked_ioctl = btrfs_ioctl,
2953 #ifdef CONFIG_COMPAT
2954 .compat_ioctl = btrfs_ioctl,
2956 .copy_file_range = btrfs_copy_file_range,
2957 .clone_file_range = btrfs_clone_file_range,
2958 .dedupe_file_range = btrfs_dedupe_file_range,
2961 void btrfs_auto_defrag_exit(void)
2963 kmem_cache_destroy(btrfs_inode_defrag_cachep);
2966 int btrfs_auto_defrag_init(void)
2968 btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
2969 sizeof(struct inode_defrag), 0,
2970 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2972 if (!btrfs_inode_defrag_cachep)
2978 int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end)
2983 * So with compression we will find and lock a dirty page and clear the
2984 * first one as dirty, setup an async extent, and immediately return
2985 * with the entire range locked but with nobody actually marked with
2986 * writeback. So we can't just filemap_write_and_wait_range() and
2987 * expect it to work since it will just kick off a thread to do the
2988 * actual work. So we need to call filemap_fdatawrite_range _again_
2989 * since it will wait on the page lock, which won't be unlocked until
2990 * after the pages have been marked as writeback and so we're good to go
2991 * from there. We have to do this otherwise we'll miss the ordered
2992 * extents and that results in badness. Please Josef, do not think you
2993 * know better and pull this out at some point in the future, it is
2994 * right and you are wrong.
2996 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
2997 if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
2998 &BTRFS_I(inode)->runtime_flags))
2999 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);