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"
44 #include "compression.h"
46 static struct kmem_cache *btrfs_inode_defrag_cachep;
48 * when auto defrag is enabled we
49 * queue up these defrag structs to remember which
50 * inodes need defragging passes
53 struct rb_node rb_node;
57 * transid where the defrag was added, we search for
58 * extents newer than this
65 /* last offset we were able to defrag */
68 /* if we've wrapped around back to zero once already */
72 static int __compare_inode_defrag(struct inode_defrag *defrag1,
73 struct inode_defrag *defrag2)
75 if (defrag1->root > defrag2->root)
77 else if (defrag1->root < defrag2->root)
79 else if (defrag1->ino > defrag2->ino)
81 else if (defrag1->ino < defrag2->ino)
87 /* pop a record for an inode into the defrag tree. The lock
88 * must be held already
90 * If you're inserting a record for an older transid than an
91 * existing record, the transid already in the tree is lowered
93 * If an existing record is found the defrag item you
96 static int __btrfs_add_inode_defrag(struct inode *inode,
97 struct inode_defrag *defrag)
99 struct btrfs_root *root = BTRFS_I(inode)->root;
100 struct inode_defrag *entry;
102 struct rb_node *parent = NULL;
105 p = &root->fs_info->defrag_inodes.rb_node;
108 entry = rb_entry(parent, struct inode_defrag, rb_node);
110 ret = __compare_inode_defrag(defrag, entry);
112 p = &parent->rb_left;
114 p = &parent->rb_right;
116 /* if we're reinserting an entry for
117 * an old defrag run, make sure to
118 * lower the transid of our existing record
120 if (defrag->transid < entry->transid)
121 entry->transid = defrag->transid;
122 if (defrag->last_offset > entry->last_offset)
123 entry->last_offset = defrag->last_offset;
127 set_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
128 rb_link_node(&defrag->rb_node, parent, p);
129 rb_insert_color(&defrag->rb_node, &root->fs_info->defrag_inodes);
133 static inline int __need_auto_defrag(struct btrfs_root *root)
135 if (!btrfs_test_opt(root->fs_info, AUTO_DEFRAG))
138 if (btrfs_fs_closing(root->fs_info))
145 * insert a defrag record for this inode if auto defrag is
148 int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans,
151 struct btrfs_root *root = BTRFS_I(inode)->root;
152 struct inode_defrag *defrag;
156 if (!__need_auto_defrag(root))
159 if (test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags))
163 transid = trans->transid;
165 transid = BTRFS_I(inode)->root->last_trans;
167 defrag = kmem_cache_zalloc(btrfs_inode_defrag_cachep, GFP_NOFS);
171 defrag->ino = btrfs_ino(inode);
172 defrag->transid = transid;
173 defrag->root = root->root_key.objectid;
175 spin_lock(&root->fs_info->defrag_inodes_lock);
176 if (!test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags)) {
178 * If we set IN_DEFRAG flag and evict the inode from memory,
179 * and then re-read this inode, this new inode doesn't have
180 * IN_DEFRAG flag. At the case, we may find the existed defrag.
182 ret = __btrfs_add_inode_defrag(inode, defrag);
184 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
186 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
188 spin_unlock(&root->fs_info->defrag_inodes_lock);
193 * Requeue the defrag object. If there is a defrag object that points to
194 * the same inode in the tree, we will merge them together (by
195 * __btrfs_add_inode_defrag()) and free the one that we want to requeue.
197 static void btrfs_requeue_inode_defrag(struct inode *inode,
198 struct inode_defrag *defrag)
200 struct btrfs_root *root = BTRFS_I(inode)->root;
203 if (!__need_auto_defrag(root))
207 * Here we don't check the IN_DEFRAG flag, because we need merge
210 spin_lock(&root->fs_info->defrag_inodes_lock);
211 ret = __btrfs_add_inode_defrag(inode, defrag);
212 spin_unlock(&root->fs_info->defrag_inodes_lock);
217 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
221 * pick the defragable inode that we want, if it doesn't exist, we will get
224 static struct inode_defrag *
225 btrfs_pick_defrag_inode(struct btrfs_fs_info *fs_info, u64 root, u64 ino)
227 struct inode_defrag *entry = NULL;
228 struct inode_defrag tmp;
230 struct rb_node *parent = NULL;
236 spin_lock(&fs_info->defrag_inodes_lock);
237 p = fs_info->defrag_inodes.rb_node;
240 entry = rb_entry(parent, struct inode_defrag, rb_node);
242 ret = __compare_inode_defrag(&tmp, entry);
246 p = parent->rb_right;
251 if (parent && __compare_inode_defrag(&tmp, entry) > 0) {
252 parent = rb_next(parent);
254 entry = rb_entry(parent, struct inode_defrag, rb_node);
260 rb_erase(parent, &fs_info->defrag_inodes);
261 spin_unlock(&fs_info->defrag_inodes_lock);
265 void btrfs_cleanup_defrag_inodes(struct btrfs_fs_info *fs_info)
267 struct inode_defrag *defrag;
268 struct rb_node *node;
270 spin_lock(&fs_info->defrag_inodes_lock);
271 node = rb_first(&fs_info->defrag_inodes);
273 rb_erase(node, &fs_info->defrag_inodes);
274 defrag = rb_entry(node, struct inode_defrag, rb_node);
275 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
277 cond_resched_lock(&fs_info->defrag_inodes_lock);
279 node = rb_first(&fs_info->defrag_inodes);
281 spin_unlock(&fs_info->defrag_inodes_lock);
284 #define BTRFS_DEFRAG_BATCH 1024
286 static int __btrfs_run_defrag_inode(struct btrfs_fs_info *fs_info,
287 struct inode_defrag *defrag)
289 struct btrfs_root *inode_root;
291 struct btrfs_key key;
292 struct btrfs_ioctl_defrag_range_args range;
298 key.objectid = defrag->root;
299 key.type = BTRFS_ROOT_ITEM_KEY;
300 key.offset = (u64)-1;
302 index = srcu_read_lock(&fs_info->subvol_srcu);
304 inode_root = btrfs_read_fs_root_no_name(fs_info, &key);
305 if (IS_ERR(inode_root)) {
306 ret = PTR_ERR(inode_root);
310 key.objectid = defrag->ino;
311 key.type = BTRFS_INODE_ITEM_KEY;
313 inode = btrfs_iget(fs_info->sb, &key, inode_root, NULL);
315 ret = PTR_ERR(inode);
318 srcu_read_unlock(&fs_info->subvol_srcu, index);
320 /* do a chunk of defrag */
321 clear_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
322 memset(&range, 0, sizeof(range));
324 range.start = defrag->last_offset;
326 sb_start_write(fs_info->sb);
327 num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid,
329 sb_end_write(fs_info->sb);
331 * if we filled the whole defrag batch, there
332 * must be more work to do. Queue this defrag
335 if (num_defrag == BTRFS_DEFRAG_BATCH) {
336 defrag->last_offset = range.start;
337 btrfs_requeue_inode_defrag(inode, defrag);
338 } else if (defrag->last_offset && !defrag->cycled) {
340 * we didn't fill our defrag batch, but
341 * we didn't start at zero. Make sure we loop
342 * around to the start of the file.
344 defrag->last_offset = 0;
346 btrfs_requeue_inode_defrag(inode, defrag);
348 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
354 srcu_read_unlock(&fs_info->subvol_srcu, index);
355 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
360 * run through the list of inodes in the FS that need
363 int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info)
365 struct inode_defrag *defrag;
367 u64 root_objectid = 0;
369 atomic_inc(&fs_info->defrag_running);
371 /* Pause the auto defragger. */
372 if (test_bit(BTRFS_FS_STATE_REMOUNTING,
376 if (!__need_auto_defrag(fs_info->tree_root))
379 /* find an inode to defrag */
380 defrag = btrfs_pick_defrag_inode(fs_info, root_objectid,
383 if (root_objectid || first_ino) {
392 first_ino = defrag->ino + 1;
393 root_objectid = defrag->root;
395 __btrfs_run_defrag_inode(fs_info, defrag);
397 atomic_dec(&fs_info->defrag_running);
400 * during unmount, we use the transaction_wait queue to
401 * wait for the defragger to stop
403 wake_up(&fs_info->transaction_wait);
407 /* simple helper to fault in pages and copy. This should go away
408 * and be replaced with calls into generic code.
410 static noinline int btrfs_copy_from_user(loff_t pos, size_t write_bytes,
411 struct page **prepared_pages,
415 size_t total_copied = 0;
417 int offset = pos & (PAGE_SIZE - 1);
419 while (write_bytes > 0) {
420 size_t count = min_t(size_t,
421 PAGE_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_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]);
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 = round_up(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, 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, 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, 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, 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_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_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);
1349 if (err == -EAGAIN) {
1356 wait_on_page_writeback(pages[i]);
1361 while (faili >= 0) {
1362 unlock_page(pages[faili]);
1363 put_page(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,
1384 u64 *lockstart, u64 *lockend,
1385 struct extent_state **cached_state)
1387 struct btrfs_root *root = BTRFS_I(inode)->root;
1393 start_pos = round_down(pos, root->sectorsize);
1394 last_pos = start_pos
1395 + round_up(pos + write_bytes - start_pos, root->sectorsize) - 1;
1397 if (start_pos < inode->i_size) {
1398 struct btrfs_ordered_extent *ordered;
1399 lock_extent_bits(&BTRFS_I(inode)->io_tree,
1400 start_pos, last_pos, cached_state);
1401 ordered = btrfs_lookup_ordered_range(inode, start_pos,
1402 last_pos - start_pos + 1);
1404 ordered->file_offset + ordered->len > start_pos &&
1405 ordered->file_offset <= last_pos) {
1406 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1407 start_pos, last_pos,
1408 cached_state, GFP_NOFS);
1409 for (i = 0; i < num_pages; i++) {
1410 unlock_page(pages[i]);
1413 btrfs_start_ordered_extent(inode, ordered, 1);
1414 btrfs_put_ordered_extent(ordered);
1418 btrfs_put_ordered_extent(ordered);
1420 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
1421 last_pos, EXTENT_DIRTY | EXTENT_DELALLOC |
1422 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
1423 0, 0, cached_state, GFP_NOFS);
1424 *lockstart = start_pos;
1425 *lockend = last_pos;
1429 for (i = 0; i < num_pages; i++) {
1430 if (clear_page_dirty_for_io(pages[i]))
1431 account_page_redirty(pages[i]);
1432 set_page_extent_mapped(pages[i]);
1433 WARN_ON(!PageLocked(pages[i]));
1439 static noinline int check_can_nocow(struct inode *inode, loff_t pos,
1440 size_t *write_bytes)
1442 struct btrfs_root *root = BTRFS_I(inode)->root;
1443 struct btrfs_ordered_extent *ordered;
1444 u64 lockstart, lockend;
1448 ret = btrfs_start_write_no_snapshoting(root);
1452 lockstart = round_down(pos, root->sectorsize);
1453 lockend = round_up(pos + *write_bytes, root->sectorsize) - 1;
1456 lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1457 ordered = btrfs_lookup_ordered_range(inode, lockstart,
1458 lockend - lockstart + 1);
1462 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1463 btrfs_start_ordered_extent(inode, ordered, 1);
1464 btrfs_put_ordered_extent(ordered);
1467 num_bytes = lockend - lockstart + 1;
1468 ret = can_nocow_extent(inode, lockstart, &num_bytes, NULL, NULL, NULL);
1471 btrfs_end_write_no_snapshoting(root);
1473 *write_bytes = min_t(size_t, *write_bytes ,
1474 num_bytes - pos + lockstart);
1477 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1482 static noinline ssize_t __btrfs_buffered_write(struct file *file,
1486 struct inode *inode = file_inode(file);
1487 struct btrfs_root *root = BTRFS_I(inode)->root;
1488 struct page **pages = NULL;
1489 struct extent_state *cached_state = NULL;
1490 u64 release_bytes = 0;
1493 size_t num_written = 0;
1496 bool only_release_metadata = false;
1497 bool force_page_uptodate = false;
1500 nrptrs = min(DIV_ROUND_UP(iov_iter_count(i), PAGE_SIZE),
1501 PAGE_SIZE / (sizeof(struct page *)));
1502 nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1503 nrptrs = max(nrptrs, 8);
1504 pages = kmalloc_array(nrptrs, sizeof(struct page *), GFP_KERNEL);
1508 while (iov_iter_count(i) > 0) {
1509 size_t offset = pos & (PAGE_SIZE - 1);
1510 size_t sector_offset;
1511 size_t write_bytes = min(iov_iter_count(i),
1512 nrptrs * (size_t)PAGE_SIZE -
1514 size_t num_pages = DIV_ROUND_UP(write_bytes + offset,
1516 size_t reserve_bytes;
1519 size_t dirty_sectors;
1522 WARN_ON(num_pages > nrptrs);
1525 * Fault pages before locking them in prepare_pages
1526 * to avoid recursive lock
1528 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1533 sector_offset = pos & (root->sectorsize - 1);
1534 reserve_bytes = round_up(write_bytes + sector_offset,
1537 ret = btrfs_check_data_free_space(inode, pos, write_bytes);
1539 if ((BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW |
1540 BTRFS_INODE_PREALLOC)) &&
1541 check_can_nocow(inode, pos, &write_bytes) > 0) {
1543 * For nodata cow case, no need to reserve
1546 only_release_metadata = true;
1548 * our prealloc extent may be smaller than
1549 * write_bytes, so scale down.
1551 num_pages = DIV_ROUND_UP(write_bytes + offset,
1553 reserve_bytes = round_up(write_bytes +
1561 ret = btrfs_delalloc_reserve_metadata(inode, reserve_bytes);
1563 if (!only_release_metadata)
1564 btrfs_free_reserved_data_space(inode, pos,
1567 btrfs_end_write_no_snapshoting(root);
1571 release_bytes = reserve_bytes;
1572 need_unlock = false;
1575 * This is going to setup the pages array with the number of
1576 * pages we want, so we don't really need to worry about the
1577 * contents of pages from loop to loop
1579 ret = prepare_pages(inode, pages, num_pages,
1581 force_page_uptodate);
1585 ret = lock_and_cleanup_extent_if_need(inode, pages, num_pages,
1586 pos, write_bytes, &lockstart,
1587 &lockend, &cached_state);
1592 } else if (ret > 0) {
1597 copied = btrfs_copy_from_user(pos, write_bytes, pages, i);
1599 num_sectors = BTRFS_BYTES_TO_BLKS(root->fs_info,
1601 dirty_sectors = round_up(copied + sector_offset,
1603 dirty_sectors = BTRFS_BYTES_TO_BLKS(root->fs_info,
1607 * if we have trouble faulting in the pages, fall
1608 * back to one page at a time
1610 if (copied < write_bytes)
1614 force_page_uptodate = true;
1618 force_page_uptodate = false;
1619 dirty_pages = DIV_ROUND_UP(copied + offset,
1624 * If we had a short copy we need to release the excess delaloc
1625 * bytes we reserved. We need to increment outstanding_extents
1626 * because btrfs_delalloc_release_space and
1627 * btrfs_delalloc_release_metadata will decrement it, but
1628 * we still have an outstanding extent for the chunk we actually
1631 if (num_sectors > dirty_sectors) {
1633 /* release everything except the sectors we dirtied */
1634 release_bytes -= dirty_sectors <<
1635 root->fs_info->sb->s_blocksize_bits;
1638 spin_lock(&BTRFS_I(inode)->lock);
1639 BTRFS_I(inode)->outstanding_extents++;
1640 spin_unlock(&BTRFS_I(inode)->lock);
1642 if (only_release_metadata) {
1643 btrfs_delalloc_release_metadata(inode,
1648 __pos = round_down(pos, root->sectorsize) +
1649 (dirty_pages << PAGE_SHIFT);
1650 btrfs_delalloc_release_space(inode, __pos,
1655 release_bytes = round_up(copied + sector_offset,
1659 ret = btrfs_dirty_pages(root, inode, pages,
1660 dirty_pages, pos, copied,
1663 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1664 lockstart, lockend, &cached_state,
1667 btrfs_drop_pages(pages, num_pages);
1672 if (only_release_metadata)
1673 btrfs_end_write_no_snapshoting(root);
1675 if (only_release_metadata && copied > 0) {
1676 lockstart = round_down(pos, root->sectorsize);
1677 lockend = round_up(pos + copied, root->sectorsize) - 1;
1679 set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
1680 lockend, EXTENT_NORESERVE, NULL,
1682 only_release_metadata = false;
1685 btrfs_drop_pages(pages, num_pages);
1689 balance_dirty_pages_ratelimited(inode->i_mapping);
1690 if (dirty_pages < (root->nodesize >> PAGE_SHIFT) + 1)
1691 btrfs_btree_balance_dirty(root);
1694 num_written += copied;
1699 if (release_bytes) {
1700 if (only_release_metadata) {
1701 btrfs_end_write_no_snapshoting(root);
1702 btrfs_delalloc_release_metadata(inode, release_bytes);
1704 btrfs_delalloc_release_space(inode,
1705 round_down(pos, root->sectorsize),
1710 return num_written ? num_written : ret;
1713 static ssize_t __btrfs_direct_write(struct kiocb *iocb, struct iov_iter *from)
1715 struct file *file = iocb->ki_filp;
1716 struct inode *inode = file_inode(file);
1717 loff_t pos = iocb->ki_pos;
1719 ssize_t written_buffered;
1723 written = generic_file_direct_write(iocb, from);
1725 if (written < 0 || !iov_iter_count(from))
1729 written_buffered = __btrfs_buffered_write(file, from, pos);
1730 if (written_buffered < 0) {
1731 err = written_buffered;
1735 * Ensure all data is persisted. We want the next direct IO read to be
1736 * able to read what was just written.
1738 endbyte = pos + written_buffered - 1;
1739 err = btrfs_fdatawrite_range(inode, pos, endbyte);
1742 err = filemap_fdatawait_range(inode->i_mapping, pos, endbyte);
1745 written += written_buffered;
1746 iocb->ki_pos = pos + written_buffered;
1747 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_SHIFT,
1748 endbyte >> PAGE_SHIFT);
1750 return written ? written : err;
1753 static void update_time_for_write(struct inode *inode)
1755 struct timespec now;
1757 if (IS_NOCMTIME(inode))
1760 now = current_fs_time(inode->i_sb);
1761 if (!timespec_equal(&inode->i_mtime, &now))
1762 inode->i_mtime = now;
1764 if (!timespec_equal(&inode->i_ctime, &now))
1765 inode->i_ctime = now;
1767 if (IS_I_VERSION(inode))
1768 inode_inc_iversion(inode);
1771 static ssize_t btrfs_file_write_iter(struct kiocb *iocb,
1772 struct iov_iter *from)
1774 struct file *file = iocb->ki_filp;
1775 struct inode *inode = file_inode(file);
1776 struct btrfs_root *root = BTRFS_I(inode)->root;
1779 ssize_t num_written = 0;
1780 bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host);
1788 err = generic_write_checks(iocb, from);
1790 inode_unlock(inode);
1794 current->backing_dev_info = inode_to_bdi(inode);
1795 err = file_remove_privs(file);
1797 inode_unlock(inode);
1802 * If BTRFS flips readonly due to some impossible error
1803 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1804 * although we have opened a file as writable, we have
1805 * to stop this write operation to ensure FS consistency.
1807 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
1808 inode_unlock(inode);
1814 * We reserve space for updating the inode when we reserve space for the
1815 * extent we are going to write, so we will enospc out there. We don't
1816 * need to start yet another transaction to update the inode as we will
1817 * update the inode when we finish writing whatever data we write.
1819 update_time_for_write(inode);
1822 count = iov_iter_count(from);
1823 start_pos = round_down(pos, root->sectorsize);
1824 oldsize = i_size_read(inode);
1825 if (start_pos > oldsize) {
1826 /* Expand hole size to cover write data, preventing empty gap */
1827 end_pos = round_up(pos + count, root->sectorsize);
1828 err = btrfs_cont_expand(inode, oldsize, end_pos);
1830 inode_unlock(inode);
1833 if (start_pos > round_up(oldsize, root->sectorsize))
1838 atomic_inc(&BTRFS_I(inode)->sync_writers);
1840 if (iocb->ki_flags & IOCB_DIRECT) {
1841 num_written = __btrfs_direct_write(iocb, from);
1843 num_written = __btrfs_buffered_write(file, from, pos);
1844 if (num_written > 0)
1845 iocb->ki_pos = pos + num_written;
1847 pagecache_isize_extended(inode, oldsize,
1848 i_size_read(inode));
1851 inode_unlock(inode);
1854 * We also have to set last_sub_trans to the current log transid,
1855 * otherwise subsequent syncs to a file that's been synced in this
1856 * transaction will appear to have already occurred.
1858 spin_lock(&BTRFS_I(inode)->lock);
1859 BTRFS_I(inode)->last_sub_trans = root->log_transid;
1860 spin_unlock(&BTRFS_I(inode)->lock);
1861 if (num_written > 0)
1862 num_written = generic_write_sync(iocb, num_written);
1865 atomic_dec(&BTRFS_I(inode)->sync_writers);
1867 current->backing_dev_info = NULL;
1868 return num_written ? num_written : err;
1871 int btrfs_release_file(struct inode *inode, struct file *filp)
1873 if (filp->private_data)
1874 btrfs_ioctl_trans_end(filp);
1876 * ordered_data_close is set by settattr when we are about to truncate
1877 * a file from a non-zero size to a zero size. This tries to
1878 * flush down new bytes that may have been written if the
1879 * application were using truncate to replace a file in place.
1881 if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
1882 &BTRFS_I(inode)->runtime_flags))
1883 filemap_flush(inode->i_mapping);
1887 static int start_ordered_ops(struct inode *inode, loff_t start, loff_t end)
1891 atomic_inc(&BTRFS_I(inode)->sync_writers);
1892 ret = btrfs_fdatawrite_range(inode, start, end);
1893 atomic_dec(&BTRFS_I(inode)->sync_writers);
1899 * fsync call for both files and directories. This logs the inode into
1900 * the tree log instead of forcing full commits whenever possible.
1902 * It needs to call filemap_fdatawait so that all ordered extent updates are
1903 * in the metadata btree are up to date for copying to the log.
1905 * It drops the inode mutex before doing the tree log commit. This is an
1906 * important optimization for directories because holding the mutex prevents
1907 * new operations on the dir while we write to disk.
1909 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1911 struct dentry *dentry = file_dentry(file);
1912 struct inode *inode = d_inode(dentry);
1913 struct btrfs_root *root = BTRFS_I(inode)->root;
1914 struct btrfs_trans_handle *trans;
1915 struct btrfs_log_ctx ctx;
1921 * The range length can be represented by u64, we have to do the typecasts
1922 * to avoid signed overflow if it's [0, LLONG_MAX] eg. from fsync()
1924 len = (u64)end - (u64)start + 1;
1925 trace_btrfs_sync_file(file, datasync);
1928 * We write the dirty pages in the range and wait until they complete
1929 * out of the ->i_mutex. If so, we can flush the dirty pages by
1930 * multi-task, and make the performance up. See
1931 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1933 ret = start_ordered_ops(inode, start, end);
1938 atomic_inc(&root->log_batch);
1939 full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1940 &BTRFS_I(inode)->runtime_flags);
1942 * We might have have had more pages made dirty after calling
1943 * start_ordered_ops and before acquiring the inode's i_mutex.
1947 * For a full sync, we need to make sure any ordered operations
1948 * start and finish before we start logging the inode, so that
1949 * all extents are persisted and the respective file extent
1950 * items are in the fs/subvol btree.
1952 ret = btrfs_wait_ordered_range(inode, start, len);
1955 * Start any new ordered operations before starting to log the
1956 * inode. We will wait for them to finish in btrfs_sync_log().
1958 * Right before acquiring the inode's mutex, we might have new
1959 * writes dirtying pages, which won't immediately start the
1960 * respective ordered operations - that is done through the
1961 * fill_delalloc callbacks invoked from the writepage and
1962 * writepages address space operations. So make sure we start
1963 * all ordered operations before starting to log our inode. Not
1964 * doing this means that while logging the inode, writeback
1965 * could start and invoke writepage/writepages, which would call
1966 * the fill_delalloc callbacks (cow_file_range,
1967 * submit_compressed_extents). These callbacks add first an
1968 * extent map to the modified list of extents and then create
1969 * the respective ordered operation, which means in
1970 * tree-log.c:btrfs_log_inode() we might capture all existing
1971 * ordered operations (with btrfs_get_logged_extents()) before
1972 * the fill_delalloc callback adds its ordered operation, and by
1973 * the time we visit the modified list of extent maps (with
1974 * btrfs_log_changed_extents()), we see and process the extent
1975 * map they created. We then use the extent map to construct a
1976 * file extent item for logging without waiting for the
1977 * respective ordered operation to finish - this file extent
1978 * item points to a disk location that might not have yet been
1979 * written to, containing random data - so after a crash a log
1980 * replay will make our inode have file extent items that point
1981 * to disk locations containing invalid data, as we returned
1982 * success to userspace without waiting for the respective
1983 * ordered operation to finish, because it wasn't captured by
1984 * btrfs_get_logged_extents().
1986 ret = start_ordered_ops(inode, start, end);
1989 inode_unlock(inode);
1992 atomic_inc(&root->log_batch);
1995 * If the last transaction that changed this file was before the current
1996 * transaction and we have the full sync flag set in our inode, we can
1997 * bail out now without any syncing.
1999 * Note that we can't bail out if the full sync flag isn't set. This is
2000 * because when the full sync flag is set we start all ordered extents
2001 * and wait for them to fully complete - when they complete they update
2002 * the inode's last_trans field through:
2004 * btrfs_finish_ordered_io() ->
2005 * btrfs_update_inode_fallback() ->
2006 * btrfs_update_inode() ->
2007 * btrfs_set_inode_last_trans()
2009 * So we are sure that last_trans is up to date and can do this check to
2010 * bail out safely. For the fast path, when the full sync flag is not
2011 * set in our inode, we can not do it because we start only our ordered
2012 * extents and don't wait for them to complete (that is when
2013 * btrfs_finish_ordered_io runs), so here at this point their last_trans
2014 * value might be less than or equals to fs_info->last_trans_committed,
2015 * and setting a speculative last_trans for an inode when a buffered
2016 * write is made (such as fs_info->generation + 1 for example) would not
2017 * be reliable since after setting the value and before fsync is called
2018 * any number of transactions can start and commit (transaction kthread
2019 * commits the current transaction periodically), and a transaction
2020 * commit does not start nor waits for ordered extents to complete.
2023 if (btrfs_inode_in_log(inode, root->fs_info->generation) ||
2024 (full_sync && BTRFS_I(inode)->last_trans <=
2025 root->fs_info->last_trans_committed) ||
2026 (!btrfs_have_ordered_extents_in_range(inode, start, len) &&
2027 BTRFS_I(inode)->last_trans
2028 <= root->fs_info->last_trans_committed)) {
2030 * We've had everything committed since the last time we were
2031 * modified so clear this flag in case it was set for whatever
2032 * reason, it's no longer relevant.
2034 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2035 &BTRFS_I(inode)->runtime_flags);
2036 inode_unlock(inode);
2041 * ok we haven't committed the transaction yet, lets do a commit
2043 if (file->private_data)
2044 btrfs_ioctl_trans_end(file);
2047 * We use start here because we will need to wait on the IO to complete
2048 * in btrfs_sync_log, which could require joining a transaction (for
2049 * example checking cross references in the nocow path). If we use join
2050 * here we could get into a situation where we're waiting on IO to
2051 * happen that is blocked on a transaction trying to commit. With start
2052 * we inc the extwriter counter, so we wait for all extwriters to exit
2053 * before we start blocking join'ers. This comment is to keep somebody
2054 * from thinking they are super smart and changing this to
2055 * btrfs_join_transaction *cough*Josef*cough*.
2057 trans = btrfs_start_transaction(root, 0);
2058 if (IS_ERR(trans)) {
2059 ret = PTR_ERR(trans);
2060 inode_unlock(inode);
2065 btrfs_init_log_ctx(&ctx);
2067 ret = btrfs_log_dentry_safe(trans, root, dentry, start, end, &ctx);
2069 /* Fallthrough and commit/free transaction. */
2073 /* we've logged all the items and now have a consistent
2074 * version of the file in the log. It is possible that
2075 * someone will come in and modify the file, but that's
2076 * fine because the log is consistent on disk, and we
2077 * have references to all of the file's extents
2079 * It is possible that someone will come in and log the
2080 * file again, but that will end up using the synchronization
2081 * inside btrfs_sync_log to keep things safe.
2083 inode_unlock(inode);
2086 * If any of the ordered extents had an error, just return it to user
2087 * space, so that the application knows some writes didn't succeed and
2088 * can take proper action (retry for e.g.). Blindly committing the
2089 * transaction in this case, would fool userspace that everything was
2090 * successful. And we also want to make sure our log doesn't contain
2091 * file extent items pointing to extents that weren't fully written to -
2092 * just like in the non fast fsync path, where we check for the ordered
2093 * operation's error flag before writing to the log tree and return -EIO
2094 * if any of them had this flag set (btrfs_wait_ordered_range) -
2095 * therefore we need to check for errors in the ordered operations,
2096 * which are indicated by ctx.io_err.
2099 btrfs_end_transaction(trans, root);
2104 if (ret != BTRFS_NO_LOG_SYNC) {
2106 ret = btrfs_sync_log(trans, root, &ctx);
2108 ret = btrfs_end_transaction(trans, root);
2113 ret = btrfs_wait_ordered_range(inode, start, len);
2115 btrfs_end_transaction(trans, root);
2119 ret = btrfs_commit_transaction(trans, root);
2121 ret = btrfs_end_transaction(trans, root);
2124 return ret > 0 ? -EIO : ret;
2127 static const struct vm_operations_struct btrfs_file_vm_ops = {
2128 .fault = filemap_fault,
2129 .map_pages = filemap_map_pages,
2130 .page_mkwrite = btrfs_page_mkwrite,
2133 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
2135 struct address_space *mapping = filp->f_mapping;
2137 if (!mapping->a_ops->readpage)
2140 file_accessed(filp);
2141 vma->vm_ops = &btrfs_file_vm_ops;
2146 static int hole_mergeable(struct inode *inode, struct extent_buffer *leaf,
2147 int slot, u64 start, u64 end)
2149 struct btrfs_file_extent_item *fi;
2150 struct btrfs_key key;
2152 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
2155 btrfs_item_key_to_cpu(leaf, &key, slot);
2156 if (key.objectid != btrfs_ino(inode) ||
2157 key.type != BTRFS_EXTENT_DATA_KEY)
2160 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2162 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2165 if (btrfs_file_extent_disk_bytenr(leaf, fi))
2168 if (key.offset == end)
2170 if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
2175 static int fill_holes(struct btrfs_trans_handle *trans, struct inode *inode,
2176 struct btrfs_path *path, u64 offset, u64 end)
2178 struct btrfs_root *root = BTRFS_I(inode)->root;
2179 struct extent_buffer *leaf;
2180 struct btrfs_file_extent_item *fi;
2181 struct extent_map *hole_em;
2182 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2183 struct btrfs_key key;
2186 if (btrfs_fs_incompat(root->fs_info, NO_HOLES))
2189 key.objectid = btrfs_ino(inode);
2190 key.type = BTRFS_EXTENT_DATA_KEY;
2191 key.offset = offset;
2193 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2198 leaf = path->nodes[0];
2199 if (hole_mergeable(inode, leaf, path->slots[0]-1, offset, end)) {
2203 fi = btrfs_item_ptr(leaf, path->slots[0],
2204 struct btrfs_file_extent_item);
2205 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2207 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2208 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2209 btrfs_set_file_extent_offset(leaf, fi, 0);
2210 btrfs_mark_buffer_dirty(leaf);
2214 if (hole_mergeable(inode, leaf, path->slots[0], offset, end)) {
2217 key.offset = offset;
2218 btrfs_set_item_key_safe(root->fs_info, path, &key);
2219 fi = btrfs_item_ptr(leaf, path->slots[0],
2220 struct btrfs_file_extent_item);
2221 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2223 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2224 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2225 btrfs_set_file_extent_offset(leaf, fi, 0);
2226 btrfs_mark_buffer_dirty(leaf);
2229 btrfs_release_path(path);
2231 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
2232 0, 0, end - offset, 0, end - offset,
2238 btrfs_release_path(path);
2240 hole_em = alloc_extent_map();
2242 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2243 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2244 &BTRFS_I(inode)->runtime_flags);
2246 hole_em->start = offset;
2247 hole_em->len = end - offset;
2248 hole_em->ram_bytes = hole_em->len;
2249 hole_em->orig_start = offset;
2251 hole_em->block_start = EXTENT_MAP_HOLE;
2252 hole_em->block_len = 0;
2253 hole_em->orig_block_len = 0;
2254 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
2255 hole_em->compress_type = BTRFS_COMPRESS_NONE;
2256 hole_em->generation = trans->transid;
2259 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2260 write_lock(&em_tree->lock);
2261 ret = add_extent_mapping(em_tree, hole_em, 1);
2262 write_unlock(&em_tree->lock);
2263 } while (ret == -EEXIST);
2264 free_extent_map(hole_em);
2266 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2267 &BTRFS_I(inode)->runtime_flags);
2274 * Find a hole extent on given inode and change start/len to the end of hole
2275 * extent.(hole/vacuum extent whose em->start <= start &&
2276 * em->start + em->len > start)
2277 * When a hole extent is found, return 1 and modify start/len.
2279 static int find_first_non_hole(struct inode *inode, u64 *start, u64 *len)
2281 struct extent_map *em;
2284 em = btrfs_get_extent(inode, NULL, 0, *start, *len, 0);
2285 if (IS_ERR_OR_NULL(em)) {
2293 /* Hole or vacuum extent(only exists in no-hole mode) */
2294 if (em->block_start == EXTENT_MAP_HOLE) {
2296 *len = em->start + em->len > *start + *len ?
2297 0 : *start + *len - em->start - em->len;
2298 *start = em->start + em->len;
2300 free_extent_map(em);
2304 static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
2306 struct btrfs_root *root = BTRFS_I(inode)->root;
2307 struct extent_state *cached_state = NULL;
2308 struct btrfs_path *path;
2309 struct btrfs_block_rsv *rsv;
2310 struct btrfs_trans_handle *trans;
2315 u64 orig_start = offset;
2317 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
2321 unsigned int rsv_count;
2323 bool no_holes = btrfs_fs_incompat(root->fs_info, NO_HOLES);
2325 bool truncated_block = false;
2326 bool updated_inode = false;
2328 ret = btrfs_wait_ordered_range(inode, offset, len);
2333 ino_size = round_up(inode->i_size, root->sectorsize);
2334 ret = find_first_non_hole(inode, &offset, &len);
2336 goto out_only_mutex;
2338 /* Already in a large hole */
2340 goto out_only_mutex;
2343 lockstart = round_up(offset, BTRFS_I(inode)->root->sectorsize);
2344 lockend = round_down(offset + len,
2345 BTRFS_I(inode)->root->sectorsize) - 1;
2346 same_block = (BTRFS_BYTES_TO_BLKS(root->fs_info, offset))
2347 == (BTRFS_BYTES_TO_BLKS(root->fs_info, offset + len - 1));
2349 * We needn't truncate any block which is beyond the end of the file
2350 * because we are sure there is no data there.
2353 * Only do this if we are in the same block and we aren't doing the
2356 if (same_block && len < root->sectorsize) {
2357 if (offset < ino_size) {
2358 truncated_block = true;
2359 ret = btrfs_truncate_block(inode, offset, len, 0);
2363 goto out_only_mutex;
2366 /* zero back part of the first block */
2367 if (offset < ino_size) {
2368 truncated_block = true;
2369 ret = btrfs_truncate_block(inode, offset, 0, 0);
2371 inode_unlock(inode);
2376 /* Check the aligned pages after the first unaligned page,
2377 * if offset != orig_start, which means the first unaligned page
2378 * including several following pages are already in holes,
2379 * the extra check can be skipped */
2380 if (offset == orig_start) {
2381 /* after truncate page, check hole again */
2382 len = offset + len - lockstart;
2384 ret = find_first_non_hole(inode, &offset, &len);
2386 goto out_only_mutex;
2389 goto out_only_mutex;
2394 /* Check the tail unaligned part is in a hole */
2395 tail_start = lockend + 1;
2396 tail_len = offset + len - tail_start;
2398 ret = find_first_non_hole(inode, &tail_start, &tail_len);
2399 if (unlikely(ret < 0))
2400 goto out_only_mutex;
2402 /* zero the front end of the last page */
2403 if (tail_start + tail_len < ino_size) {
2404 truncated_block = true;
2405 ret = btrfs_truncate_block(inode,
2406 tail_start + tail_len,
2409 goto out_only_mutex;
2414 if (lockend < lockstart) {
2416 goto out_only_mutex;
2420 struct btrfs_ordered_extent *ordered;
2422 truncate_pagecache_range(inode, lockstart, lockend);
2424 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2426 ordered = btrfs_lookup_first_ordered_extent(inode, lockend);
2429 * We need to make sure we have no ordered extents in this range
2430 * and nobody raced in and read a page in this range, if we did
2431 * we need to try again.
2434 (ordered->file_offset + ordered->len <= lockstart ||
2435 ordered->file_offset > lockend)) &&
2436 !btrfs_page_exists_in_range(inode, lockstart, lockend)) {
2438 btrfs_put_ordered_extent(ordered);
2442 btrfs_put_ordered_extent(ordered);
2443 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
2444 lockend, &cached_state, GFP_NOFS);
2445 ret = btrfs_wait_ordered_range(inode, lockstart,
2446 lockend - lockstart + 1);
2448 inode_unlock(inode);
2453 path = btrfs_alloc_path();
2459 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2464 rsv->size = btrfs_calc_trunc_metadata_size(root, 1);
2468 * 1 - update the inode
2469 * 1 - removing the extents in the range
2470 * 1 - adding the hole extent if no_holes isn't set
2472 rsv_count = no_holes ? 2 : 3;
2473 trans = btrfs_start_transaction(root, rsv_count);
2474 if (IS_ERR(trans)) {
2475 err = PTR_ERR(trans);
2479 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
2482 trans->block_rsv = rsv;
2484 cur_offset = lockstart;
2485 len = lockend - cur_offset;
2486 while (cur_offset < lockend) {
2487 ret = __btrfs_drop_extents(trans, root, inode, path,
2488 cur_offset, lockend + 1,
2489 &drop_end, 1, 0, 0, NULL);
2493 trans->block_rsv = &root->fs_info->trans_block_rsv;
2495 if (cur_offset < ino_size) {
2496 ret = fill_holes(trans, inode, path, cur_offset,
2504 cur_offset = drop_end;
2506 ret = btrfs_update_inode(trans, root, inode);
2512 btrfs_end_transaction(trans, root);
2513 btrfs_btree_balance_dirty(root);
2515 trans = btrfs_start_transaction(root, rsv_count);
2516 if (IS_ERR(trans)) {
2517 ret = PTR_ERR(trans);
2522 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
2524 BUG_ON(ret); /* shouldn't happen */
2525 trans->block_rsv = rsv;
2527 ret = find_first_non_hole(inode, &cur_offset, &len);
2528 if (unlikely(ret < 0))
2541 trans->block_rsv = &root->fs_info->trans_block_rsv;
2543 * If we are using the NO_HOLES feature we might have had already an
2544 * hole that overlaps a part of the region [lockstart, lockend] and
2545 * ends at (or beyond) lockend. Since we have no file extent items to
2546 * represent holes, drop_end can be less than lockend and so we must
2547 * make sure we have an extent map representing the existing hole (the
2548 * call to __btrfs_drop_extents() might have dropped the existing extent
2549 * map representing the existing hole), otherwise the fast fsync path
2550 * will not record the existence of the hole region
2551 * [existing_hole_start, lockend].
2553 if (drop_end <= lockend)
2554 drop_end = lockend + 1;
2556 * Don't insert file hole extent item if it's for a range beyond eof
2557 * (because it's useless) or if it represents a 0 bytes range (when
2558 * cur_offset == drop_end).
2560 if (cur_offset < ino_size && cur_offset < drop_end) {
2561 ret = fill_holes(trans, inode, path, cur_offset, drop_end);
2572 inode_inc_iversion(inode);
2573 inode->i_mtime = inode->i_ctime = current_fs_time(inode->i_sb);
2575 trans->block_rsv = &root->fs_info->trans_block_rsv;
2576 ret = btrfs_update_inode(trans, root, inode);
2577 updated_inode = true;
2578 btrfs_end_transaction(trans, root);
2579 btrfs_btree_balance_dirty(root);
2581 btrfs_free_path(path);
2582 btrfs_free_block_rsv(root, rsv);
2584 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2585 &cached_state, GFP_NOFS);
2587 if (!updated_inode && truncated_block && !ret && !err) {
2589 * If we only end up zeroing part of a page, we still need to
2590 * update the inode item, so that all the time fields are
2591 * updated as well as the necessary btrfs inode in memory fields
2592 * for detecting, at fsync time, if the inode isn't yet in the
2593 * log tree or it's there but not up to date.
2595 trans = btrfs_start_transaction(root, 1);
2596 if (IS_ERR(trans)) {
2597 err = PTR_ERR(trans);
2599 err = btrfs_update_inode(trans, root, inode);
2600 ret = btrfs_end_transaction(trans, root);
2603 inode_unlock(inode);
2609 /* Helper structure to record which range is already reserved */
2610 struct falloc_range {
2611 struct list_head list;
2617 * Helper function to add falloc range
2619 * Caller should have locked the larger range of extent containing
2622 static int add_falloc_range(struct list_head *head, u64 start, u64 len)
2624 struct falloc_range *prev = NULL;
2625 struct falloc_range *range = NULL;
2627 if (list_empty(head))
2631 * As fallocate iterate by bytenr order, we only need to check
2634 prev = list_entry(head->prev, struct falloc_range, list);
2635 if (prev->start + prev->len == start) {
2640 range = kmalloc(sizeof(*range), GFP_KERNEL);
2643 range->start = start;
2645 list_add_tail(&range->list, head);
2649 static long btrfs_fallocate(struct file *file, int mode,
2650 loff_t offset, loff_t len)
2652 struct inode *inode = file_inode(file);
2653 struct extent_state *cached_state = NULL;
2654 struct falloc_range *range;
2655 struct falloc_range *tmp;
2656 struct list_head reserve_list;
2664 struct extent_map *em;
2665 int blocksize = BTRFS_I(inode)->root->sectorsize;
2668 alloc_start = round_down(offset, blocksize);
2669 alloc_end = round_up(offset + len, blocksize);
2671 /* Make sure we aren't being give some crap mode */
2672 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2675 if (mode & FALLOC_FL_PUNCH_HOLE)
2676 return btrfs_punch_hole(inode, offset, len);
2679 * Only trigger disk allocation, don't trigger qgroup reserve
2681 * For qgroup space, it will be checked later.
2683 ret = btrfs_alloc_data_chunk_ondemand(inode, alloc_end - alloc_start);
2689 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) {
2690 ret = inode_newsize_ok(inode, offset + len);
2696 * TODO: Move these two operations after we have checked
2697 * accurate reserved space, or fallocate can still fail but
2698 * with page truncated or size expanded.
2700 * But that's a minor problem and won't do much harm BTW.
2702 if (alloc_start > inode->i_size) {
2703 ret = btrfs_cont_expand(inode, i_size_read(inode),
2707 } else if (offset + len > inode->i_size) {
2709 * If we are fallocating from the end of the file onward we
2710 * need to zero out the end of the block if i_size lands in the
2711 * middle of a block.
2713 ret = btrfs_truncate_block(inode, inode->i_size, 0, 0);
2719 * wait for ordered IO before we have any locks. We'll loop again
2720 * below with the locks held.
2722 ret = btrfs_wait_ordered_range(inode, alloc_start,
2723 alloc_end - alloc_start);
2727 locked_end = alloc_end - 1;
2729 struct btrfs_ordered_extent *ordered;
2731 /* the extent lock is ordered inside the running
2734 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
2735 locked_end, &cached_state);
2736 ordered = btrfs_lookup_first_ordered_extent(inode,
2739 ordered->file_offset + ordered->len > alloc_start &&
2740 ordered->file_offset < alloc_end) {
2741 btrfs_put_ordered_extent(ordered);
2742 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
2743 alloc_start, locked_end,
2744 &cached_state, GFP_KERNEL);
2746 * we can't wait on the range with the transaction
2747 * running or with the extent lock held
2749 ret = btrfs_wait_ordered_range(inode, alloc_start,
2750 alloc_end - alloc_start);
2755 btrfs_put_ordered_extent(ordered);
2760 /* First, check if we exceed the qgroup limit */
2761 INIT_LIST_HEAD(&reserve_list);
2762 cur_offset = alloc_start;
2764 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2765 alloc_end - cur_offset, 0);
2766 if (IS_ERR_OR_NULL(em)) {
2773 last_byte = min(extent_map_end(em), alloc_end);
2774 actual_end = min_t(u64, extent_map_end(em), offset + len);
2775 last_byte = ALIGN(last_byte, blocksize);
2776 if (em->block_start == EXTENT_MAP_HOLE ||
2777 (cur_offset >= inode->i_size &&
2778 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
2779 ret = add_falloc_range(&reserve_list, cur_offset,
2780 last_byte - cur_offset);
2782 free_extent_map(em);
2785 ret = btrfs_qgroup_reserve_data(inode, cur_offset,
2786 last_byte - cur_offset);
2790 free_extent_map(em);
2791 cur_offset = last_byte;
2792 if (cur_offset >= alloc_end)
2797 * If ret is still 0, means we're OK to fallocate.
2798 * Or just cleanup the list and exit.
2800 list_for_each_entry_safe(range, tmp, &reserve_list, list) {
2802 ret = btrfs_prealloc_file_range(inode, mode,
2804 range->len, 1 << inode->i_blkbits,
2805 offset + len, &alloc_hint);
2806 list_del(&range->list);
2812 if (actual_end > inode->i_size &&
2813 !(mode & FALLOC_FL_KEEP_SIZE)) {
2814 struct btrfs_trans_handle *trans;
2815 struct btrfs_root *root = BTRFS_I(inode)->root;
2818 * We didn't need to allocate any more space, but we
2819 * still extended the size of the file so we need to
2820 * update i_size and the inode item.
2822 trans = btrfs_start_transaction(root, 1);
2823 if (IS_ERR(trans)) {
2824 ret = PTR_ERR(trans);
2826 inode->i_ctime = current_fs_time(inode->i_sb);
2827 i_size_write(inode, actual_end);
2828 btrfs_ordered_update_i_size(inode, actual_end, NULL);
2829 ret = btrfs_update_inode(trans, root, inode);
2831 btrfs_end_transaction(trans, root);
2833 ret = btrfs_end_transaction(trans, root);
2837 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
2838 &cached_state, GFP_KERNEL);
2841 * As we waited the extent range, the data_rsv_map must be empty
2842 * in the range, as written data range will be released from it.
2843 * And for prealloacted extent, it will also be released when
2844 * its metadata is written.
2845 * So this is completely used as cleanup.
2847 btrfs_qgroup_free_data(inode, alloc_start, alloc_end - alloc_start);
2848 inode_unlock(inode);
2849 /* Let go of our reservation. */
2850 btrfs_free_reserved_data_space(inode, alloc_start,
2851 alloc_end - alloc_start);
2855 static int find_desired_extent(struct inode *inode, loff_t *offset, int whence)
2857 struct btrfs_root *root = BTRFS_I(inode)->root;
2858 struct extent_map *em = NULL;
2859 struct extent_state *cached_state = NULL;
2866 if (inode->i_size == 0)
2870 * *offset can be negative, in this case we start finding DATA/HOLE from
2871 * the very start of the file.
2873 start = max_t(loff_t, 0, *offset);
2875 lockstart = round_down(start, root->sectorsize);
2876 lockend = round_up(i_size_read(inode), root->sectorsize);
2877 if (lockend <= lockstart)
2878 lockend = lockstart + root->sectorsize;
2880 len = lockend - lockstart + 1;
2882 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2885 while (start < inode->i_size) {
2886 em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0);
2893 if (whence == SEEK_HOLE &&
2894 (em->block_start == EXTENT_MAP_HOLE ||
2895 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2897 else if (whence == SEEK_DATA &&
2898 (em->block_start != EXTENT_MAP_HOLE &&
2899 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2902 start = em->start + em->len;
2903 free_extent_map(em);
2907 free_extent_map(em);
2909 if (whence == SEEK_DATA && start >= inode->i_size)
2912 *offset = min_t(loff_t, start, inode->i_size);
2914 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2915 &cached_state, GFP_NOFS);
2919 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
2921 struct inode *inode = file->f_mapping->host;
2928 offset = generic_file_llseek(file, offset, whence);
2932 if (offset >= i_size_read(inode)) {
2933 inode_unlock(inode);
2937 ret = find_desired_extent(inode, &offset, whence);
2939 inode_unlock(inode);
2944 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2946 inode_unlock(inode);
2950 const struct file_operations btrfs_file_operations = {
2951 .llseek = btrfs_file_llseek,
2952 .read_iter = generic_file_read_iter,
2953 .splice_read = generic_file_splice_read,
2954 .write_iter = btrfs_file_write_iter,
2955 .mmap = btrfs_file_mmap,
2956 .open = generic_file_open,
2957 .release = btrfs_release_file,
2958 .fsync = btrfs_sync_file,
2959 .fallocate = btrfs_fallocate,
2960 .unlocked_ioctl = btrfs_ioctl,
2961 #ifdef CONFIG_COMPAT
2962 .compat_ioctl = btrfs_compat_ioctl,
2964 .copy_file_range = btrfs_copy_file_range,
2965 .clone_file_range = btrfs_clone_file_range,
2966 .dedupe_file_range = btrfs_dedupe_file_range,
2969 void btrfs_auto_defrag_exit(void)
2971 kmem_cache_destroy(btrfs_inode_defrag_cachep);
2974 int btrfs_auto_defrag_init(void)
2976 btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
2977 sizeof(struct inode_defrag), 0,
2980 if (!btrfs_inode_defrag_cachep)
2986 int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end)
2991 * So with compression we will find and lock a dirty page and clear the
2992 * first one as dirty, setup an async extent, and immediately return
2993 * with the entire range locked but with nobody actually marked with
2994 * writeback. So we can't just filemap_write_and_wait_range() and
2995 * expect it to work since it will just kick off a thread to do the
2996 * actual work. So we need to call filemap_fdatawrite_range _again_
2997 * since it will wait on the page lock, which won't be unlocked until
2998 * after the pages have been marked as writeback and so we're good to go
2999 * from there. We have to do this otherwise we'll miss the ordered
3000 * extents and that results in badness. Please Josef, do not think you
3001 * know better and pull this out at some point in the future, it is
3002 * right and you are wrong.
3004 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
3005 if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
3006 &BTRFS_I(inode)->runtime_flags))
3007 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);