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]);
760 if (key.objectid > ino)
762 if (WARN_ON_ONCE(key.objectid < ino) ||
763 key.type < BTRFS_EXTENT_DATA_KEY) {
768 if (key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
771 fi = btrfs_item_ptr(leaf, path->slots[0],
772 struct btrfs_file_extent_item);
773 extent_type = btrfs_file_extent_type(leaf, fi);
775 if (extent_type == BTRFS_FILE_EXTENT_REG ||
776 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
777 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
778 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
779 extent_offset = btrfs_file_extent_offset(leaf, fi);
780 extent_end = key.offset +
781 btrfs_file_extent_num_bytes(leaf, fi);
782 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
783 extent_end = key.offset +
784 btrfs_file_extent_inline_len(leaf,
792 * Don't skip extent items representing 0 byte lengths. They
793 * used to be created (bug) if while punching holes we hit
794 * -ENOSPC condition. So if we find one here, just ensure we
795 * delete it, otherwise we would insert a new file extent item
796 * with the same key (offset) as that 0 bytes length file
797 * extent item in the call to setup_items_for_insert() later
800 if (extent_end == key.offset && extent_end >= search_start)
801 goto delete_extent_item;
803 if (extent_end <= search_start) {
809 search_start = max(key.offset, start);
810 if (recow || !modify_tree) {
812 btrfs_release_path(path);
817 * | - range to drop - |
818 * | -------- extent -------- |
820 if (start > key.offset && end < extent_end) {
822 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
827 memcpy(&new_key, &key, sizeof(new_key));
828 new_key.offset = start;
829 ret = btrfs_duplicate_item(trans, root, path,
831 if (ret == -EAGAIN) {
832 btrfs_release_path(path);
838 leaf = path->nodes[0];
839 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
840 struct btrfs_file_extent_item);
841 btrfs_set_file_extent_num_bytes(leaf, fi,
844 fi = btrfs_item_ptr(leaf, path->slots[0],
845 struct btrfs_file_extent_item);
847 extent_offset += start - key.offset;
848 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
849 btrfs_set_file_extent_num_bytes(leaf, fi,
851 btrfs_mark_buffer_dirty(leaf);
853 if (update_refs && disk_bytenr > 0) {
854 ret = btrfs_inc_extent_ref(trans, root,
855 disk_bytenr, num_bytes, 0,
856 root->root_key.objectid,
858 start - extent_offset);
859 BUG_ON(ret); /* -ENOMEM */
864 * | ---- range to drop ----- |
865 * | -------- extent -------- |
867 if (start <= key.offset && end < extent_end) {
868 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
873 memcpy(&new_key, &key, sizeof(new_key));
874 new_key.offset = end;
875 btrfs_set_item_key_safe(root->fs_info, path, &new_key);
877 extent_offset += end - key.offset;
878 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
879 btrfs_set_file_extent_num_bytes(leaf, fi,
881 btrfs_mark_buffer_dirty(leaf);
882 if (update_refs && disk_bytenr > 0)
883 inode_sub_bytes(inode, end - key.offset);
887 search_start = extent_end;
889 * | ---- range to drop ----- |
890 * | -------- extent -------- |
892 if (start > key.offset && end >= extent_end) {
894 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
899 btrfs_set_file_extent_num_bytes(leaf, fi,
901 btrfs_mark_buffer_dirty(leaf);
902 if (update_refs && disk_bytenr > 0)
903 inode_sub_bytes(inode, extent_end - start);
904 if (end == extent_end)
912 * | ---- range to drop ----- |
913 * | ------ extent ------ |
915 if (start <= key.offset && end >= extent_end) {
918 del_slot = path->slots[0];
921 BUG_ON(del_slot + del_nr != path->slots[0]);
926 extent_type == BTRFS_FILE_EXTENT_INLINE) {
927 inode_sub_bytes(inode,
928 extent_end - key.offset);
929 extent_end = ALIGN(extent_end,
931 } else if (update_refs && disk_bytenr > 0) {
932 ret = btrfs_free_extent(trans, root,
933 disk_bytenr, num_bytes, 0,
934 root->root_key.objectid,
935 key.objectid, key.offset -
937 BUG_ON(ret); /* -ENOMEM */
938 inode_sub_bytes(inode,
939 extent_end - key.offset);
942 if (end == extent_end)
945 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
950 ret = btrfs_del_items(trans, root, path, del_slot,
953 btrfs_abort_transaction(trans, root, ret);
960 btrfs_release_path(path);
967 if (!ret && del_nr > 0) {
969 * Set path->slots[0] to first slot, so that after the delete
970 * if items are move off from our leaf to its immediate left or
971 * right neighbor leafs, we end up with a correct and adjusted
972 * path->slots[0] for our insertion (if replace_extent != 0).
974 path->slots[0] = del_slot;
975 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
977 btrfs_abort_transaction(trans, root, ret);
980 leaf = path->nodes[0];
982 * If btrfs_del_items() was called, it might have deleted a leaf, in
983 * which case it unlocked our path, so check path->locks[0] matches a
986 if (!ret && replace_extent && leafs_visited == 1 &&
987 (path->locks[0] == BTRFS_WRITE_LOCK_BLOCKING ||
988 path->locks[0] == BTRFS_WRITE_LOCK) &&
989 btrfs_leaf_free_space(root, leaf) >=
990 sizeof(struct btrfs_item) + extent_item_size) {
993 key.type = BTRFS_EXTENT_DATA_KEY;
995 if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) {
996 struct btrfs_key slot_key;
998 btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]);
999 if (btrfs_comp_cpu_keys(&key, &slot_key) > 0)
1002 setup_items_for_insert(root, path, &key,
1005 sizeof(struct btrfs_item) +
1006 extent_item_size, 1);
1010 if (!replace_extent || !(*key_inserted))
1011 btrfs_release_path(path);
1013 *drop_end = found ? min(end, extent_end) : end;
1017 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
1018 struct btrfs_root *root, struct inode *inode, u64 start,
1019 u64 end, int drop_cache)
1021 struct btrfs_path *path;
1024 path = btrfs_alloc_path();
1027 ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL,
1028 drop_cache, 0, 0, NULL);
1029 btrfs_free_path(path);
1033 static int extent_mergeable(struct extent_buffer *leaf, int slot,
1034 u64 objectid, u64 bytenr, u64 orig_offset,
1035 u64 *start, u64 *end)
1037 struct btrfs_file_extent_item *fi;
1038 struct btrfs_key key;
1041 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
1044 btrfs_item_key_to_cpu(leaf, &key, slot);
1045 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
1048 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1049 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
1050 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
1051 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
1052 btrfs_file_extent_compression(leaf, fi) ||
1053 btrfs_file_extent_encryption(leaf, fi) ||
1054 btrfs_file_extent_other_encoding(leaf, fi))
1057 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1058 if ((*start && *start != key.offset) || (*end && *end != extent_end))
1061 *start = key.offset;
1067 * Mark extent in the range start - end as written.
1069 * This changes extent type from 'pre-allocated' to 'regular'. If only
1070 * part of extent is marked as written, the extent will be split into
1073 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
1074 struct inode *inode, u64 start, u64 end)
1076 struct btrfs_root *root = BTRFS_I(inode)->root;
1077 struct extent_buffer *leaf;
1078 struct btrfs_path *path;
1079 struct btrfs_file_extent_item *fi;
1080 struct btrfs_key key;
1081 struct btrfs_key new_key;
1093 u64 ino = btrfs_ino(inode);
1095 path = btrfs_alloc_path();
1102 key.type = BTRFS_EXTENT_DATA_KEY;
1105 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1108 if (ret > 0 && path->slots[0] > 0)
1111 leaf = path->nodes[0];
1112 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1113 BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY);
1114 fi = btrfs_item_ptr(leaf, path->slots[0],
1115 struct btrfs_file_extent_item);
1116 BUG_ON(btrfs_file_extent_type(leaf, fi) !=
1117 BTRFS_FILE_EXTENT_PREALLOC);
1118 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1119 BUG_ON(key.offset > start || extent_end < end);
1121 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1122 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1123 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
1124 memcpy(&new_key, &key, sizeof(new_key));
1126 if (start == key.offset && end < extent_end) {
1129 if (extent_mergeable(leaf, path->slots[0] - 1,
1130 ino, bytenr, orig_offset,
1131 &other_start, &other_end)) {
1132 new_key.offset = end;
1133 btrfs_set_item_key_safe(root->fs_info, path, &new_key);
1134 fi = btrfs_item_ptr(leaf, path->slots[0],
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_set_file_extent_offset(leaf, fi,
1142 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1143 struct btrfs_file_extent_item);
1144 btrfs_set_file_extent_generation(leaf, fi,
1146 btrfs_set_file_extent_num_bytes(leaf, fi,
1148 btrfs_mark_buffer_dirty(leaf);
1153 if (start > key.offset && end == extent_end) {
1156 if (extent_mergeable(leaf, path->slots[0] + 1,
1157 ino, bytenr, orig_offset,
1158 &other_start, &other_end)) {
1159 fi = btrfs_item_ptr(leaf, path->slots[0],
1160 struct btrfs_file_extent_item);
1161 btrfs_set_file_extent_num_bytes(leaf, fi,
1162 start - key.offset);
1163 btrfs_set_file_extent_generation(leaf, fi,
1166 new_key.offset = start;
1167 btrfs_set_item_key_safe(root->fs_info, path, &new_key);
1169 fi = btrfs_item_ptr(leaf, path->slots[0],
1170 struct btrfs_file_extent_item);
1171 btrfs_set_file_extent_generation(leaf, fi,
1173 btrfs_set_file_extent_num_bytes(leaf, fi,
1175 btrfs_set_file_extent_offset(leaf, fi,
1176 start - orig_offset);
1177 btrfs_mark_buffer_dirty(leaf);
1182 while (start > key.offset || end < extent_end) {
1183 if (key.offset == start)
1186 new_key.offset = split;
1187 ret = btrfs_duplicate_item(trans, root, path, &new_key);
1188 if (ret == -EAGAIN) {
1189 btrfs_release_path(path);
1193 btrfs_abort_transaction(trans, root, ret);
1197 leaf = path->nodes[0];
1198 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1199 struct btrfs_file_extent_item);
1200 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1201 btrfs_set_file_extent_num_bytes(leaf, fi,
1202 split - key.offset);
1204 fi = btrfs_item_ptr(leaf, path->slots[0],
1205 struct btrfs_file_extent_item);
1207 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1208 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
1209 btrfs_set_file_extent_num_bytes(leaf, fi,
1210 extent_end - split);
1211 btrfs_mark_buffer_dirty(leaf);
1213 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
1214 root->root_key.objectid,
1216 BUG_ON(ret); /* -ENOMEM */
1218 if (split == start) {
1221 BUG_ON(start != key.offset);
1230 if (extent_mergeable(leaf, path->slots[0] + 1,
1231 ino, bytenr, orig_offset,
1232 &other_start, &other_end)) {
1234 btrfs_release_path(path);
1237 extent_end = other_end;
1238 del_slot = path->slots[0] + 1;
1240 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1241 0, root->root_key.objectid,
1243 BUG_ON(ret); /* -ENOMEM */
1247 if (extent_mergeable(leaf, path->slots[0] - 1,
1248 ino, bytenr, orig_offset,
1249 &other_start, &other_end)) {
1251 btrfs_release_path(path);
1254 key.offset = other_start;
1255 del_slot = path->slots[0];
1257 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1258 0, root->root_key.objectid,
1260 BUG_ON(ret); /* -ENOMEM */
1263 fi = btrfs_item_ptr(leaf, path->slots[0],
1264 struct btrfs_file_extent_item);
1265 btrfs_set_file_extent_type(leaf, fi,
1266 BTRFS_FILE_EXTENT_REG);
1267 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1268 btrfs_mark_buffer_dirty(leaf);
1270 fi = btrfs_item_ptr(leaf, del_slot - 1,
1271 struct btrfs_file_extent_item);
1272 btrfs_set_file_extent_type(leaf, fi,
1273 BTRFS_FILE_EXTENT_REG);
1274 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1275 btrfs_set_file_extent_num_bytes(leaf, fi,
1276 extent_end - key.offset);
1277 btrfs_mark_buffer_dirty(leaf);
1279 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1281 btrfs_abort_transaction(trans, root, ret);
1286 btrfs_free_path(path);
1291 * on error we return an unlocked page and the error value
1292 * on success we return a locked page and 0
1294 static int prepare_uptodate_page(struct inode *inode,
1295 struct page *page, u64 pos,
1296 bool force_uptodate)
1300 if (((pos & (PAGE_CACHE_SIZE - 1)) || force_uptodate) &&
1301 !PageUptodate(page)) {
1302 ret = btrfs_readpage(NULL, page);
1306 if (!PageUptodate(page)) {
1310 if (page->mapping != inode->i_mapping) {
1319 * this just gets pages into the page cache and locks them down.
1321 static noinline int prepare_pages(struct inode *inode, struct page **pages,
1322 size_t num_pages, loff_t pos,
1323 size_t write_bytes, bool force_uptodate)
1326 unsigned long index = pos >> PAGE_CACHE_SHIFT;
1327 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1331 for (i = 0; i < num_pages; i++) {
1333 pages[i] = find_or_create_page(inode->i_mapping, index + i,
1334 mask | __GFP_WRITE);
1342 err = prepare_uptodate_page(inode, pages[i], pos,
1344 if (!err && i == num_pages - 1)
1345 err = prepare_uptodate_page(inode, pages[i],
1346 pos + write_bytes, false);
1348 page_cache_release(pages[i]);
1349 if (err == -EAGAIN) {
1356 wait_on_page_writeback(pages[i]);
1361 while (faili >= 0) {
1362 unlock_page(pages[faili]);
1363 page_cache_release(pages[faili]);
1371 * This function locks the extent and properly waits for data=ordered extents
1372 * to finish before allowing the pages to be modified if need.
1375 * 1 - the extent is locked
1376 * 0 - the extent is not locked, and everything is OK
1377 * -EAGAIN - need re-prepare the pages
1378 * the other < 0 number - Something wrong happens
1381 lock_and_cleanup_extent_if_need(struct inode *inode, struct page **pages,
1382 size_t num_pages, loff_t pos,
1383 u64 *lockstart, u64 *lockend,
1384 struct extent_state **cached_state)
1391 start_pos = pos & ~((u64)PAGE_CACHE_SIZE - 1);
1392 last_pos = start_pos + ((u64)num_pages << PAGE_CACHE_SHIFT) - 1;
1394 if (start_pos < inode->i_size) {
1395 struct btrfs_ordered_extent *ordered;
1396 lock_extent_bits(&BTRFS_I(inode)->io_tree,
1397 start_pos, last_pos, 0, cached_state);
1398 ordered = btrfs_lookup_ordered_range(inode, start_pos,
1399 last_pos - start_pos + 1);
1401 ordered->file_offset + ordered->len > start_pos &&
1402 ordered->file_offset <= last_pos) {
1403 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1404 start_pos, last_pos,
1405 cached_state, GFP_NOFS);
1406 for (i = 0; i < num_pages; i++) {
1407 unlock_page(pages[i]);
1408 page_cache_release(pages[i]);
1410 btrfs_start_ordered_extent(inode, ordered, 1);
1411 btrfs_put_ordered_extent(ordered);
1415 btrfs_put_ordered_extent(ordered);
1417 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
1418 last_pos, EXTENT_DIRTY | EXTENT_DELALLOC |
1419 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
1420 0, 0, cached_state, GFP_NOFS);
1421 *lockstart = start_pos;
1422 *lockend = last_pos;
1426 for (i = 0; i < num_pages; i++) {
1427 if (clear_page_dirty_for_io(pages[i]))
1428 account_page_redirty(pages[i]);
1429 set_page_extent_mapped(pages[i]);
1430 WARN_ON(!PageLocked(pages[i]));
1436 static noinline int check_can_nocow(struct inode *inode, loff_t pos,
1437 size_t *write_bytes)
1439 struct btrfs_root *root = BTRFS_I(inode)->root;
1440 struct btrfs_ordered_extent *ordered;
1441 u64 lockstart, lockend;
1445 ret = btrfs_start_write_no_snapshoting(root);
1449 lockstart = round_down(pos, root->sectorsize);
1450 lockend = round_up(pos + *write_bytes, root->sectorsize) - 1;
1453 lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1454 ordered = btrfs_lookup_ordered_range(inode, lockstart,
1455 lockend - lockstart + 1);
1459 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1460 btrfs_start_ordered_extent(inode, ordered, 1);
1461 btrfs_put_ordered_extent(ordered);
1464 num_bytes = lockend - lockstart + 1;
1465 ret = can_nocow_extent(inode, lockstart, &num_bytes, NULL, NULL, NULL);
1468 btrfs_end_write_no_snapshoting(root);
1470 *write_bytes = min_t(size_t, *write_bytes ,
1471 num_bytes - pos + lockstart);
1474 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1479 static noinline ssize_t __btrfs_buffered_write(struct file *file,
1483 struct inode *inode = file_inode(file);
1484 struct btrfs_root *root = BTRFS_I(inode)->root;
1485 struct page **pages = NULL;
1486 struct extent_state *cached_state = NULL;
1487 u64 release_bytes = 0;
1490 size_t num_written = 0;
1493 bool only_release_metadata = false;
1494 bool force_page_uptodate = false;
1497 nrptrs = min(DIV_ROUND_UP(iov_iter_count(i), PAGE_CACHE_SIZE),
1498 PAGE_CACHE_SIZE / (sizeof(struct page *)));
1499 nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1500 nrptrs = max(nrptrs, 8);
1501 pages = kmalloc_array(nrptrs, sizeof(struct page *), GFP_KERNEL);
1505 while (iov_iter_count(i) > 0) {
1506 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1507 size_t write_bytes = min(iov_iter_count(i),
1508 nrptrs * (size_t)PAGE_CACHE_SIZE -
1510 size_t num_pages = DIV_ROUND_UP(write_bytes + offset,
1512 size_t reserve_bytes;
1516 WARN_ON(num_pages > nrptrs);
1519 * Fault pages before locking them in prepare_pages
1520 * to avoid recursive lock
1522 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1527 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1528 only_release_metadata = false;
1530 if ((BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW |
1531 BTRFS_INODE_PREALLOC)) &&
1532 check_can_nocow(inode, pos, &write_bytes) > 0) {
1534 * For nodata cow case, no need to reserve
1537 only_release_metadata = true;
1539 * our prealloc extent may be smaller than
1540 * write_bytes, so scale down.
1542 num_pages = DIV_ROUND_UP(write_bytes + offset,
1544 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1545 goto reserve_metadata;
1548 ret = btrfs_check_data_free_space(inode, pos, write_bytes);
1553 ret = btrfs_delalloc_reserve_metadata(inode, reserve_bytes);
1555 if (!only_release_metadata)
1556 btrfs_free_reserved_data_space(inode, pos,
1559 btrfs_end_write_no_snapshoting(root);
1563 release_bytes = reserve_bytes;
1564 need_unlock = false;
1567 * This is going to setup the pages array with the number of
1568 * pages we want, so we don't really need to worry about the
1569 * contents of pages from loop to loop
1571 ret = prepare_pages(inode, pages, num_pages,
1573 force_page_uptodate);
1577 ret = lock_and_cleanup_extent_if_need(inode, pages, num_pages,
1578 pos, &lockstart, &lockend,
1584 } else if (ret > 0) {
1589 copied = btrfs_copy_from_user(pos, num_pages,
1590 write_bytes, pages, i);
1593 * if we have trouble faulting in the pages, fall
1594 * back to one page at a time
1596 if (copied < write_bytes)
1600 force_page_uptodate = true;
1603 force_page_uptodate = false;
1604 dirty_pages = DIV_ROUND_UP(copied + offset,
1609 * If we had a short copy we need to release the excess delaloc
1610 * bytes we reserved. We need to increment outstanding_extents
1611 * because btrfs_delalloc_release_space will decrement it, but
1612 * we still have an outstanding extent for the chunk we actually
1615 if (num_pages > dirty_pages) {
1616 release_bytes = (num_pages - dirty_pages) <<
1619 spin_lock(&BTRFS_I(inode)->lock);
1620 BTRFS_I(inode)->outstanding_extents++;
1621 spin_unlock(&BTRFS_I(inode)->lock);
1623 if (only_release_metadata) {
1624 btrfs_delalloc_release_metadata(inode,
1629 __pos = round_down(pos, root->sectorsize) +
1630 (dirty_pages << PAGE_CACHE_SHIFT);
1631 btrfs_delalloc_release_space(inode, __pos,
1636 release_bytes = dirty_pages << PAGE_CACHE_SHIFT;
1639 ret = btrfs_dirty_pages(root, inode, pages,
1640 dirty_pages, pos, copied,
1643 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1644 lockstart, lockend, &cached_state,
1647 btrfs_drop_pages(pages, num_pages);
1652 if (only_release_metadata)
1653 btrfs_end_write_no_snapshoting(root);
1655 if (only_release_metadata && copied > 0) {
1656 lockstart = round_down(pos, root->sectorsize);
1657 lockend = lockstart +
1658 (dirty_pages << PAGE_CACHE_SHIFT) - 1;
1660 set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
1661 lockend, EXTENT_NORESERVE, NULL,
1665 btrfs_drop_pages(pages, num_pages);
1669 balance_dirty_pages_ratelimited(inode->i_mapping);
1670 if (dirty_pages < (root->nodesize >> PAGE_CACHE_SHIFT) + 1)
1671 btrfs_btree_balance_dirty(root);
1674 num_written += copied;
1679 if (release_bytes) {
1680 if (only_release_metadata) {
1681 btrfs_end_write_no_snapshoting(root);
1682 btrfs_delalloc_release_metadata(inode, release_bytes);
1684 btrfs_delalloc_release_space(inode, pos, release_bytes);
1688 return num_written ? num_written : ret;
1691 static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1692 struct iov_iter *from,
1695 struct file *file = iocb->ki_filp;
1696 struct inode *inode = file_inode(file);
1698 ssize_t written_buffered;
1702 written = generic_file_direct_write(iocb, from, pos);
1704 if (written < 0 || !iov_iter_count(from))
1708 written_buffered = __btrfs_buffered_write(file, from, pos);
1709 if (written_buffered < 0) {
1710 err = written_buffered;
1714 * Ensure all data is persisted. We want the next direct IO read to be
1715 * able to read what was just written.
1717 endbyte = pos + written_buffered - 1;
1718 err = btrfs_fdatawrite_range(inode, pos, endbyte);
1721 err = filemap_fdatawait_range(inode->i_mapping, pos, endbyte);
1724 written += written_buffered;
1725 iocb->ki_pos = pos + written_buffered;
1726 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
1727 endbyte >> PAGE_CACHE_SHIFT);
1729 return written ? written : err;
1732 static void update_time_for_write(struct inode *inode)
1734 struct timespec now;
1736 if (IS_NOCMTIME(inode))
1739 now = current_fs_time(inode->i_sb);
1740 if (!timespec_equal(&inode->i_mtime, &now))
1741 inode->i_mtime = now;
1743 if (!timespec_equal(&inode->i_ctime, &now))
1744 inode->i_ctime = now;
1746 if (IS_I_VERSION(inode))
1747 inode_inc_iversion(inode);
1750 static ssize_t btrfs_file_write_iter(struct kiocb *iocb,
1751 struct iov_iter *from)
1753 struct file *file = iocb->ki_filp;
1754 struct inode *inode = file_inode(file);
1755 struct btrfs_root *root = BTRFS_I(inode)->root;
1758 ssize_t num_written = 0;
1759 bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host);
1764 mutex_lock(&inode->i_mutex);
1765 err = generic_write_checks(iocb, from);
1767 mutex_unlock(&inode->i_mutex);
1771 current->backing_dev_info = inode_to_bdi(inode);
1772 err = file_remove_privs(file);
1774 mutex_unlock(&inode->i_mutex);
1779 * If BTRFS flips readonly due to some impossible error
1780 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1781 * although we have opened a file as writable, we have
1782 * to stop this write operation to ensure FS consistency.
1784 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
1785 mutex_unlock(&inode->i_mutex);
1791 * We reserve space for updating the inode when we reserve space for the
1792 * extent we are going to write, so we will enospc out there. We don't
1793 * need to start yet another transaction to update the inode as we will
1794 * update the inode when we finish writing whatever data we write.
1796 update_time_for_write(inode);
1799 count = iov_iter_count(from);
1800 start_pos = round_down(pos, root->sectorsize);
1801 if (start_pos > i_size_read(inode)) {
1802 /* Expand hole size to cover write data, preventing empty gap */
1803 end_pos = round_up(pos + count, root->sectorsize);
1804 err = btrfs_cont_expand(inode, i_size_read(inode), end_pos);
1806 mutex_unlock(&inode->i_mutex);
1812 atomic_inc(&BTRFS_I(inode)->sync_writers);
1814 if (iocb->ki_flags & IOCB_DIRECT) {
1815 num_written = __btrfs_direct_write(iocb, from, pos);
1817 num_written = __btrfs_buffered_write(file, from, pos);
1818 if (num_written > 0)
1819 iocb->ki_pos = pos + num_written;
1822 mutex_unlock(&inode->i_mutex);
1825 * We also have to set last_sub_trans to the current log transid,
1826 * otherwise subsequent syncs to a file that's been synced in this
1827 * transaction will appear to have already occured.
1829 spin_lock(&BTRFS_I(inode)->lock);
1830 BTRFS_I(inode)->last_sub_trans = root->log_transid;
1831 spin_unlock(&BTRFS_I(inode)->lock);
1832 if (num_written > 0) {
1833 err = generic_write_sync(file, pos, num_written);
1839 atomic_dec(&BTRFS_I(inode)->sync_writers);
1841 current->backing_dev_info = NULL;
1842 return num_written ? num_written : err;
1845 int btrfs_release_file(struct inode *inode, struct file *filp)
1847 if (filp->private_data)
1848 btrfs_ioctl_trans_end(filp);
1850 * ordered_data_close is set by settattr when we are about to truncate
1851 * a file from a non-zero size to a zero size. This tries to
1852 * flush down new bytes that may have been written if the
1853 * application were using truncate to replace a file in place.
1855 if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
1856 &BTRFS_I(inode)->runtime_flags))
1857 filemap_flush(inode->i_mapping);
1861 static int start_ordered_ops(struct inode *inode, loff_t start, loff_t end)
1864 struct blk_plug plug;
1867 * This is only called in fsync, which would do synchronous writes, so
1868 * a plug can merge adjacent IOs as much as possible. Esp. in case of
1869 * multiple disks using raid profile, a large IO can be split to
1870 * several segments of stripe length (currently 64K).
1872 blk_start_plug(&plug);
1873 atomic_inc(&BTRFS_I(inode)->sync_writers);
1874 ret = btrfs_fdatawrite_range(inode, start, end);
1875 atomic_dec(&BTRFS_I(inode)->sync_writers);
1876 blk_finish_plug(&plug);
1882 * fsync call for both files and directories. This logs the inode into
1883 * the tree log instead of forcing full commits whenever possible.
1885 * It needs to call filemap_fdatawait so that all ordered extent updates are
1886 * in the metadata btree are up to date for copying to the log.
1888 * It drops the inode mutex before doing the tree log commit. This is an
1889 * important optimization for directories because holding the mutex prevents
1890 * new operations on the dir while we write to disk.
1892 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1894 struct dentry *dentry = file_dentry(file);
1895 struct inode *inode = d_inode(dentry);
1896 struct btrfs_root *root = BTRFS_I(inode)->root;
1897 struct btrfs_trans_handle *trans;
1898 struct btrfs_log_ctx ctx;
1904 * If the inode needs a full sync, make sure we use a full range to
1905 * avoid log tree corruption, due to hole detection racing with ordered
1906 * extent completion for adjacent ranges, and assertion failures during
1909 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1910 &BTRFS_I(inode)->runtime_flags)) {
1916 * The range length can be represented by u64, we have to do the typecasts
1917 * to avoid signed overflow if it's [0, LLONG_MAX] eg. from fsync()
1919 len = (u64)end - (u64)start + 1;
1920 trace_btrfs_sync_file(file, datasync);
1923 * We write the dirty pages in the range and wait until they complete
1924 * out of the ->i_mutex. If so, we can flush the dirty pages by
1925 * multi-task, and make the performance up. See
1926 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1928 ret = start_ordered_ops(inode, start, end);
1932 mutex_lock(&inode->i_mutex);
1933 atomic_inc(&root->log_batch);
1934 full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1935 &BTRFS_I(inode)->runtime_flags);
1937 * We might have have had more pages made dirty after calling
1938 * start_ordered_ops and before acquiring the inode's i_mutex.
1942 * For a full sync, we need to make sure any ordered operations
1943 * start and finish before we start logging the inode, so that
1944 * all extents are persisted and the respective file extent
1945 * items are in the fs/subvol btree.
1947 ret = btrfs_wait_ordered_range(inode, start, len);
1950 * Start any new ordered operations before starting to log the
1951 * inode. We will wait for them to finish in btrfs_sync_log().
1953 * Right before acquiring the inode's mutex, we might have new
1954 * writes dirtying pages, which won't immediately start the
1955 * respective ordered operations - that is done through the
1956 * fill_delalloc callbacks invoked from the writepage and
1957 * writepages address space operations. So make sure we start
1958 * all ordered operations before starting to log our inode. Not
1959 * doing this means that while logging the inode, writeback
1960 * could start and invoke writepage/writepages, which would call
1961 * the fill_delalloc callbacks (cow_file_range,
1962 * submit_compressed_extents). These callbacks add first an
1963 * extent map to the modified list of extents and then create
1964 * the respective ordered operation, which means in
1965 * tree-log.c:btrfs_log_inode() we might capture all existing
1966 * ordered operations (with btrfs_get_logged_extents()) before
1967 * the fill_delalloc callback adds its ordered operation, and by
1968 * the time we visit the modified list of extent maps (with
1969 * btrfs_log_changed_extents()), we see and process the extent
1970 * map they created. We then use the extent map to construct a
1971 * file extent item for logging without waiting for the
1972 * respective ordered operation to finish - this file extent
1973 * item points to a disk location that might not have yet been
1974 * written to, containing random data - so after a crash a log
1975 * replay will make our inode have file extent items that point
1976 * to disk locations containing invalid data, as we returned
1977 * success to userspace without waiting for the respective
1978 * ordered operation to finish, because it wasn't captured by
1979 * btrfs_get_logged_extents().
1981 ret = start_ordered_ops(inode, start, end);
1984 mutex_unlock(&inode->i_mutex);
1987 atomic_inc(&root->log_batch);
1990 * If the last transaction that changed this file was before the current
1991 * transaction and we have the full sync flag set in our inode, we can
1992 * bail out now without any syncing.
1994 * Note that we can't bail out if the full sync flag isn't set. This is
1995 * because when the full sync flag is set we start all ordered extents
1996 * and wait for them to fully complete - when they complete they update
1997 * the inode's last_trans field through:
1999 * btrfs_finish_ordered_io() ->
2000 * btrfs_update_inode_fallback() ->
2001 * btrfs_update_inode() ->
2002 * btrfs_set_inode_last_trans()
2004 * So we are sure that last_trans is up to date and can do this check to
2005 * bail out safely. For the fast path, when the full sync flag is not
2006 * set in our inode, we can not do it because we start only our ordered
2007 * extents and don't wait for them to complete (that is when
2008 * btrfs_finish_ordered_io runs), so here at this point their last_trans
2009 * value might be less than or equals to fs_info->last_trans_committed,
2010 * and setting a speculative last_trans for an inode when a buffered
2011 * write is made (such as fs_info->generation + 1 for example) would not
2012 * be reliable since after setting the value and before fsync is called
2013 * any number of transactions can start and commit (transaction kthread
2014 * commits the current transaction periodically), and a transaction
2015 * commit does not start nor waits for ordered extents to complete.
2018 if (btrfs_inode_in_log(inode, root->fs_info->generation) ||
2019 (BTRFS_I(inode)->last_trans <=
2020 root->fs_info->last_trans_committed &&
2022 !btrfs_have_ordered_extents_in_range(inode, start, len)))) {
2024 * We'v had everything committed since the last time we were
2025 * modified so clear this flag in case it was set for whatever
2026 * reason, it's no longer relevant.
2028 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2029 &BTRFS_I(inode)->runtime_flags);
2030 mutex_unlock(&inode->i_mutex);
2035 * ok we haven't committed the transaction yet, lets do a commit
2037 if (file->private_data)
2038 btrfs_ioctl_trans_end(file);
2041 * We use start here because we will need to wait on the IO to complete
2042 * in btrfs_sync_log, which could require joining a transaction (for
2043 * example checking cross references in the nocow path). If we use join
2044 * here we could get into a situation where we're waiting on IO to
2045 * happen that is blocked on a transaction trying to commit. With start
2046 * we inc the extwriter counter, so we wait for all extwriters to exit
2047 * before we start blocking join'ers. This comment is to keep somebody
2048 * from thinking they are super smart and changing this to
2049 * btrfs_join_transaction *cough*Josef*cough*.
2051 trans = btrfs_start_transaction(root, 0);
2052 if (IS_ERR(trans)) {
2053 ret = PTR_ERR(trans);
2054 mutex_unlock(&inode->i_mutex);
2059 btrfs_init_log_ctx(&ctx);
2061 ret = btrfs_log_dentry_safe(trans, root, dentry, start, end, &ctx);
2063 /* Fallthrough and commit/free transaction. */
2067 /* we've logged all the items and now have a consistent
2068 * version of the file in the log. It is possible that
2069 * someone will come in and modify the file, but that's
2070 * fine because the log is consistent on disk, and we
2071 * have references to all of the file's extents
2073 * It is possible that someone will come in and log the
2074 * file again, but that will end up using the synchronization
2075 * inside btrfs_sync_log to keep things safe.
2077 mutex_unlock(&inode->i_mutex);
2080 * If any of the ordered extents had an error, just return it to user
2081 * space, so that the application knows some writes didn't succeed and
2082 * can take proper action (retry for e.g.). Blindly committing the
2083 * transaction in this case, would fool userspace that everything was
2084 * successful. And we also want to make sure our log doesn't contain
2085 * file extent items pointing to extents that weren't fully written to -
2086 * just like in the non fast fsync path, where we check for the ordered
2087 * operation's error flag before writing to the log tree and return -EIO
2088 * if any of them had this flag set (btrfs_wait_ordered_range) -
2089 * therefore we need to check for errors in the ordered operations,
2090 * which are indicated by ctx.io_err.
2093 btrfs_end_transaction(trans, root);
2098 if (ret != BTRFS_NO_LOG_SYNC) {
2100 ret = btrfs_sync_log(trans, root, &ctx);
2102 ret = btrfs_end_transaction(trans, root);
2107 ret = btrfs_wait_ordered_range(inode, start, len);
2109 btrfs_end_transaction(trans, root);
2113 ret = btrfs_commit_transaction(trans, root);
2115 ret = btrfs_end_transaction(trans, root);
2118 return ret > 0 ? -EIO : ret;
2121 static const struct vm_operations_struct btrfs_file_vm_ops = {
2122 .fault = filemap_fault,
2123 .map_pages = filemap_map_pages,
2124 .page_mkwrite = btrfs_page_mkwrite,
2127 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
2129 struct address_space *mapping = filp->f_mapping;
2131 if (!mapping->a_ops->readpage)
2134 file_accessed(filp);
2135 vma->vm_ops = &btrfs_file_vm_ops;
2140 static int hole_mergeable(struct inode *inode, struct extent_buffer *leaf,
2141 int slot, u64 start, u64 end)
2143 struct btrfs_file_extent_item *fi;
2144 struct btrfs_key key;
2146 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
2149 btrfs_item_key_to_cpu(leaf, &key, slot);
2150 if (key.objectid != btrfs_ino(inode) ||
2151 key.type != BTRFS_EXTENT_DATA_KEY)
2154 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2156 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2159 if (btrfs_file_extent_disk_bytenr(leaf, fi))
2162 if (key.offset == end)
2164 if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
2169 static int fill_holes(struct btrfs_trans_handle *trans, struct inode *inode,
2170 struct btrfs_path *path, u64 offset, u64 end)
2172 struct btrfs_root *root = BTRFS_I(inode)->root;
2173 struct extent_buffer *leaf;
2174 struct btrfs_file_extent_item *fi;
2175 struct extent_map *hole_em;
2176 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2177 struct btrfs_key key;
2180 if (btrfs_fs_incompat(root->fs_info, NO_HOLES))
2183 key.objectid = btrfs_ino(inode);
2184 key.type = BTRFS_EXTENT_DATA_KEY;
2185 key.offset = offset;
2187 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2192 leaf = path->nodes[0];
2193 if (hole_mergeable(inode, leaf, path->slots[0]-1, offset, end)) {
2197 fi = btrfs_item_ptr(leaf, path->slots[0],
2198 struct btrfs_file_extent_item);
2199 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2201 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2202 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2203 btrfs_set_file_extent_offset(leaf, fi, 0);
2204 btrfs_mark_buffer_dirty(leaf);
2208 if (hole_mergeable(inode, leaf, path->slots[0], offset, end)) {
2211 key.offset = offset;
2212 btrfs_set_item_key_safe(root->fs_info, path, &key);
2213 fi = btrfs_item_ptr(leaf, path->slots[0],
2214 struct btrfs_file_extent_item);
2215 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2217 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2218 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2219 btrfs_set_file_extent_offset(leaf, fi, 0);
2220 btrfs_mark_buffer_dirty(leaf);
2223 btrfs_release_path(path);
2225 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
2226 0, 0, end - offset, 0, end - offset,
2232 btrfs_release_path(path);
2234 hole_em = alloc_extent_map();
2236 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2237 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2238 &BTRFS_I(inode)->runtime_flags);
2240 hole_em->start = offset;
2241 hole_em->len = end - offset;
2242 hole_em->ram_bytes = hole_em->len;
2243 hole_em->orig_start = offset;
2245 hole_em->block_start = EXTENT_MAP_HOLE;
2246 hole_em->block_len = 0;
2247 hole_em->orig_block_len = 0;
2248 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
2249 hole_em->compress_type = BTRFS_COMPRESS_NONE;
2250 hole_em->generation = trans->transid;
2253 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2254 write_lock(&em_tree->lock);
2255 ret = add_extent_mapping(em_tree, hole_em, 1);
2256 write_unlock(&em_tree->lock);
2257 } while (ret == -EEXIST);
2258 free_extent_map(hole_em);
2260 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2261 &BTRFS_I(inode)->runtime_flags);
2268 * Find a hole extent on given inode and change start/len to the end of hole
2269 * extent.(hole/vacuum extent whose em->start <= start &&
2270 * em->start + em->len > start)
2271 * When a hole extent is found, return 1 and modify start/len.
2273 static int find_first_non_hole(struct inode *inode, u64 *start, u64 *len)
2275 struct extent_map *em;
2278 em = btrfs_get_extent(inode, NULL, 0, *start, *len, 0);
2279 if (IS_ERR_OR_NULL(em)) {
2287 /* Hole or vacuum extent(only exists in no-hole mode) */
2288 if (em->block_start == EXTENT_MAP_HOLE) {
2290 *len = em->start + em->len > *start + *len ?
2291 0 : *start + *len - em->start - em->len;
2292 *start = em->start + em->len;
2294 free_extent_map(em);
2298 static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
2300 struct btrfs_root *root = BTRFS_I(inode)->root;
2301 struct extent_state *cached_state = NULL;
2302 struct btrfs_path *path;
2303 struct btrfs_block_rsv *rsv;
2304 struct btrfs_trans_handle *trans;
2309 u64 orig_start = offset;
2311 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
2315 unsigned int rsv_count;
2317 bool no_holes = btrfs_fs_incompat(root->fs_info, NO_HOLES);
2319 bool truncated_page = false;
2320 bool updated_inode = false;
2322 ret = btrfs_wait_ordered_range(inode, offset, len);
2326 mutex_lock(&inode->i_mutex);
2327 ino_size = round_up(inode->i_size, PAGE_CACHE_SIZE);
2328 ret = find_first_non_hole(inode, &offset, &len);
2330 goto out_only_mutex;
2332 /* Already in a large hole */
2334 goto out_only_mutex;
2337 lockstart = round_up(offset, BTRFS_I(inode)->root->sectorsize);
2338 lockend = round_down(offset + len,
2339 BTRFS_I(inode)->root->sectorsize) - 1;
2340 same_page = ((offset >> PAGE_CACHE_SHIFT) ==
2341 ((offset + len - 1) >> PAGE_CACHE_SHIFT));
2344 * We needn't truncate any page 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 page and we aren't doing the
2351 if (same_page && len < PAGE_CACHE_SIZE) {
2352 if (offset < ino_size) {
2353 truncated_page = true;
2354 ret = btrfs_truncate_page(inode, offset, len, 0);
2358 goto out_only_mutex;
2361 /* zero back part of the first page */
2362 if (offset < ino_size) {
2363 truncated_page = true;
2364 ret = btrfs_truncate_page(inode, offset, 0, 0);
2366 mutex_unlock(&inode->i_mutex);
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_page = true;
2400 ret = btrfs_truncate_page(inode,
2401 tail_start + tail_len, 0, 1);
2403 goto out_only_mutex;
2408 if (lockend < lockstart) {
2410 goto out_only_mutex;
2414 struct btrfs_ordered_extent *ordered;
2416 truncate_pagecache_range(inode, lockstart, lockend);
2418 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2420 ordered = btrfs_lookup_first_ordered_extent(inode, lockend);
2423 * We need to make sure we have no ordered extents in this range
2424 * and nobody raced in and read a page in this range, if we did
2425 * we need to try again.
2428 (ordered->file_offset + ordered->len <= lockstart ||
2429 ordered->file_offset > lockend)) &&
2430 !btrfs_page_exists_in_range(inode, lockstart, lockend)) {
2432 btrfs_put_ordered_extent(ordered);
2436 btrfs_put_ordered_extent(ordered);
2437 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
2438 lockend, &cached_state, GFP_NOFS);
2439 ret = btrfs_wait_ordered_range(inode, lockstart,
2440 lockend - lockstart + 1);
2442 mutex_unlock(&inode->i_mutex);
2447 path = btrfs_alloc_path();
2453 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2458 rsv->size = btrfs_calc_trunc_metadata_size(root, 1);
2462 * 1 - update the inode
2463 * 1 - removing the extents in the range
2464 * 1 - adding the hole extent if no_holes isn't set
2466 rsv_count = no_holes ? 2 : 3;
2467 trans = btrfs_start_transaction(root, rsv_count);
2468 if (IS_ERR(trans)) {
2469 err = PTR_ERR(trans);
2473 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
2476 trans->block_rsv = rsv;
2478 cur_offset = lockstart;
2479 len = lockend - cur_offset;
2480 while (cur_offset < lockend) {
2481 ret = __btrfs_drop_extents(trans, root, inode, path,
2482 cur_offset, lockend + 1,
2483 &drop_end, 1, 0, 0, NULL);
2487 trans->block_rsv = &root->fs_info->trans_block_rsv;
2489 if (cur_offset < ino_size) {
2490 ret = fill_holes(trans, inode, path, cur_offset,
2498 cur_offset = drop_end;
2500 ret = btrfs_update_inode(trans, root, inode);
2506 btrfs_end_transaction(trans, root);
2507 btrfs_btree_balance_dirty(root);
2509 trans = btrfs_start_transaction(root, rsv_count);
2510 if (IS_ERR(trans)) {
2511 ret = PTR_ERR(trans);
2516 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
2518 BUG_ON(ret); /* shouldn't happen */
2519 trans->block_rsv = rsv;
2521 ret = find_first_non_hole(inode, &cur_offset, &len);
2522 if (unlikely(ret < 0))
2535 trans->block_rsv = &root->fs_info->trans_block_rsv;
2537 * If we are using the NO_HOLES feature we might have had already an
2538 * hole that overlaps a part of the region [lockstart, lockend] and
2539 * ends at (or beyond) lockend. Since we have no file extent items to
2540 * represent holes, drop_end can be less than lockend and so we must
2541 * make sure we have an extent map representing the existing hole (the
2542 * call to __btrfs_drop_extents() might have dropped the existing extent
2543 * map representing the existing hole), otherwise the fast fsync path
2544 * will not record the existence of the hole region
2545 * [existing_hole_start, lockend].
2547 if (drop_end <= lockend)
2548 drop_end = lockend + 1;
2550 * Don't insert file hole extent item if it's for a range beyond eof
2551 * (because it's useless) or if it represents a 0 bytes range (when
2552 * cur_offset == drop_end).
2554 if (cur_offset < ino_size && cur_offset < drop_end) {
2555 ret = fill_holes(trans, inode, path, cur_offset, drop_end);
2566 inode_inc_iversion(inode);
2567 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2569 trans->block_rsv = &root->fs_info->trans_block_rsv;
2570 ret = btrfs_update_inode(trans, root, inode);
2571 updated_inode = true;
2572 btrfs_end_transaction(trans, root);
2573 btrfs_btree_balance_dirty(root);
2575 btrfs_free_path(path);
2576 btrfs_free_block_rsv(root, rsv);
2578 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2579 &cached_state, GFP_NOFS);
2581 if (!updated_inode && truncated_page && !ret && !err) {
2583 * If we only end up zeroing part of a page, we still need to
2584 * update the inode item, so that all the time fields are
2585 * updated as well as the necessary btrfs inode in memory fields
2586 * for detecting, at fsync time, if the inode isn't yet in the
2587 * log tree or it's there but not up to date.
2589 trans = btrfs_start_transaction(root, 1);
2590 if (IS_ERR(trans)) {
2591 err = PTR_ERR(trans);
2593 err = btrfs_update_inode(trans, root, inode);
2594 ret = btrfs_end_transaction(trans, root);
2597 mutex_unlock(&inode->i_mutex);
2603 /* Helper structure to record which range is already reserved */
2604 struct falloc_range {
2605 struct list_head list;
2611 * Helper function to add falloc range
2613 * Caller should have locked the larger range of extent containing
2616 static int add_falloc_range(struct list_head *head, u64 start, u64 len)
2618 struct falloc_range *prev = NULL;
2619 struct falloc_range *range = NULL;
2621 if (list_empty(head))
2625 * As fallocate iterate by bytenr order, we only need to check
2628 prev = list_entry(head->prev, struct falloc_range, list);
2629 if (prev->start + prev->len == start) {
2634 range = kmalloc(sizeof(*range), GFP_NOFS);
2637 range->start = start;
2639 list_add_tail(&range->list, head);
2643 static long btrfs_fallocate(struct file *file, int mode,
2644 loff_t offset, loff_t len)
2646 struct inode *inode = file_inode(file);
2647 struct extent_state *cached_state = NULL;
2648 struct falloc_range *range;
2649 struct falloc_range *tmp;
2650 struct list_head reserve_list;
2658 struct extent_map *em;
2659 int blocksize = BTRFS_I(inode)->root->sectorsize;
2662 alloc_start = round_down(offset, blocksize);
2663 alloc_end = round_up(offset + len, blocksize);
2665 /* Make sure we aren't being give some crap mode */
2666 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2669 if (mode & FALLOC_FL_PUNCH_HOLE)
2670 return btrfs_punch_hole(inode, offset, len);
2673 * Only trigger disk allocation, don't trigger qgroup reserve
2675 * For qgroup space, it will be checked later.
2677 ret = btrfs_alloc_data_chunk_ondemand(inode, alloc_end - alloc_start);
2681 mutex_lock(&inode->i_mutex);
2682 ret = inode_newsize_ok(inode, alloc_end);
2687 * TODO: Move these two operations after we have checked
2688 * accurate reserved space, or fallocate can still fail but
2689 * with page truncated or size expanded.
2691 * But that's a minor problem and won't do much harm BTW.
2693 if (alloc_start > inode->i_size) {
2694 ret = btrfs_cont_expand(inode, i_size_read(inode),
2698 } else if (offset + len > inode->i_size) {
2700 * If we are fallocating from the end of the file onward we
2701 * need to zero out the end of the page if i_size lands in the
2704 ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
2710 * wait for ordered IO before we have any locks. We'll loop again
2711 * below with the locks held.
2713 ret = btrfs_wait_ordered_range(inode, alloc_start,
2714 alloc_end - alloc_start);
2718 locked_end = alloc_end - 1;
2720 struct btrfs_ordered_extent *ordered;
2722 /* the extent lock is ordered inside the running
2725 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
2726 locked_end, 0, &cached_state);
2727 ordered = btrfs_lookup_first_ordered_extent(inode,
2730 ordered->file_offset + ordered->len > alloc_start &&
2731 ordered->file_offset < alloc_end) {
2732 btrfs_put_ordered_extent(ordered);
2733 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
2734 alloc_start, locked_end,
2735 &cached_state, GFP_NOFS);
2737 * we can't wait on the range with the transaction
2738 * running or with the extent lock held
2740 ret = btrfs_wait_ordered_range(inode, alloc_start,
2741 alloc_end - alloc_start);
2746 btrfs_put_ordered_extent(ordered);
2751 /* First, check if we exceed the qgroup limit */
2752 INIT_LIST_HEAD(&reserve_list);
2753 cur_offset = alloc_start;
2755 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2756 alloc_end - cur_offset, 0);
2757 if (IS_ERR_OR_NULL(em)) {
2764 last_byte = min(extent_map_end(em), alloc_end);
2765 actual_end = min_t(u64, extent_map_end(em), offset + len);
2766 last_byte = ALIGN(last_byte, blocksize);
2767 if (em->block_start == EXTENT_MAP_HOLE ||
2768 (cur_offset >= inode->i_size &&
2769 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
2770 ret = add_falloc_range(&reserve_list, cur_offset,
2771 last_byte - cur_offset);
2773 free_extent_map(em);
2776 ret = btrfs_qgroup_reserve_data(inode, cur_offset,
2777 last_byte - cur_offset);
2781 free_extent_map(em);
2782 cur_offset = last_byte;
2783 if (cur_offset >= alloc_end)
2788 * If ret is still 0, means we're OK to fallocate.
2789 * Or just cleanup the list and exit.
2791 list_for_each_entry_safe(range, tmp, &reserve_list, list) {
2793 ret = btrfs_prealloc_file_range(inode, mode,
2795 range->len, i_blocksize(inode),
2796 offset + len, &alloc_hint);
2797 list_del(&range->list);
2803 if (actual_end > inode->i_size &&
2804 !(mode & FALLOC_FL_KEEP_SIZE)) {
2805 struct btrfs_trans_handle *trans;
2806 struct btrfs_root *root = BTRFS_I(inode)->root;
2809 * We didn't need to allocate any more space, but we
2810 * still extended the size of the file so we need to
2811 * update i_size and the inode item.
2813 trans = btrfs_start_transaction(root, 1);
2814 if (IS_ERR(trans)) {
2815 ret = PTR_ERR(trans);
2817 inode->i_ctime = CURRENT_TIME;
2818 i_size_write(inode, actual_end);
2819 btrfs_ordered_update_i_size(inode, actual_end, NULL);
2820 ret = btrfs_update_inode(trans, root, inode);
2822 btrfs_end_transaction(trans, root);
2824 ret = btrfs_end_transaction(trans, root);
2828 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
2829 &cached_state, GFP_NOFS);
2832 * As we waited the extent range, the data_rsv_map must be empty
2833 * in the range, as written data range will be released from it.
2834 * And for prealloacted extent, it will also be released when
2835 * its metadata is written.
2836 * So this is completely used as cleanup.
2838 btrfs_qgroup_free_data(inode, alloc_start, alloc_end - alloc_start);
2839 mutex_unlock(&inode->i_mutex);
2840 /* Let go of our reservation. */
2841 btrfs_free_reserved_data_space(inode, alloc_start,
2842 alloc_end - alloc_start);
2846 static int find_desired_extent(struct inode *inode, loff_t *offset, int whence)
2848 struct btrfs_root *root = BTRFS_I(inode)->root;
2849 struct extent_map *em = NULL;
2850 struct extent_state *cached_state = NULL;
2857 if (inode->i_size == 0)
2861 * *offset can be negative, in this case we start finding DATA/HOLE from
2862 * the very start of the file.
2864 start = max_t(loff_t, 0, *offset);
2866 lockstart = round_down(start, root->sectorsize);
2867 lockend = round_up(i_size_read(inode), root->sectorsize);
2868 if (lockend <= lockstart)
2869 lockend = lockstart + root->sectorsize;
2871 len = lockend - lockstart + 1;
2873 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, 0,
2876 while (start < inode->i_size) {
2877 em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0);
2884 if (whence == SEEK_HOLE &&
2885 (em->block_start == EXTENT_MAP_HOLE ||
2886 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2888 else if (whence == SEEK_DATA &&
2889 (em->block_start != EXTENT_MAP_HOLE &&
2890 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2893 start = em->start + em->len;
2894 free_extent_map(em);
2898 free_extent_map(em);
2900 if (whence == SEEK_DATA && start >= inode->i_size)
2903 *offset = min_t(loff_t, start, inode->i_size);
2905 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2906 &cached_state, GFP_NOFS);
2910 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
2912 struct inode *inode = file->f_mapping->host;
2915 mutex_lock(&inode->i_mutex);
2919 offset = generic_file_llseek(file, offset, whence);
2923 if (offset >= i_size_read(inode)) {
2924 mutex_unlock(&inode->i_mutex);
2928 ret = find_desired_extent(inode, &offset, whence);
2930 mutex_unlock(&inode->i_mutex);
2935 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2937 mutex_unlock(&inode->i_mutex);
2941 const struct file_operations btrfs_file_operations = {
2942 .llseek = btrfs_file_llseek,
2943 .read_iter = generic_file_read_iter,
2944 .splice_read = generic_file_splice_read,
2945 .write_iter = btrfs_file_write_iter,
2946 .mmap = btrfs_file_mmap,
2947 .open = generic_file_open,
2948 .release = btrfs_release_file,
2949 .fsync = btrfs_sync_file,
2950 .fallocate = btrfs_fallocate,
2951 .unlocked_ioctl = btrfs_ioctl,
2952 #ifdef CONFIG_COMPAT
2953 .compat_ioctl = btrfs_ioctl,
2957 void btrfs_auto_defrag_exit(void)
2959 if (btrfs_inode_defrag_cachep)
2960 kmem_cache_destroy(btrfs_inode_defrag_cachep);
2963 int btrfs_auto_defrag_init(void)
2965 btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
2966 sizeof(struct inode_defrag), 0,
2967 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2969 if (!btrfs_inode_defrag_cachep)
2975 int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end)
2980 * So with compression we will find and lock a dirty page and clear the
2981 * first one as dirty, setup an async extent, and immediately return
2982 * with the entire range locked but with nobody actually marked with
2983 * writeback. So we can't just filemap_write_and_wait_range() and
2984 * expect it to work since it will just kick off a thread to do the
2985 * actual work. So we need to call filemap_fdatawrite_range _again_
2986 * since it will wait on the page lock, which won't be unlocked until
2987 * after the pages have been marked as writeback and so we're good to go
2988 * from there. We have to do this otherwise we'll miss the ordered
2989 * extents and that results in badness. Please Josef, do not think you
2990 * know better and pull this out at some point in the future, it is
2991 * right and you are wrong.
2993 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
2994 if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
2995 &BTRFS_I(inode)->runtime_flags))
2996 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);