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>
35 #include "transaction.h"
36 #include "btrfs_inode.h"
38 #include "print-tree.h"
44 * when auto defrag is enabled we
45 * queue up these defrag structs to remember which
46 * inodes need defragging passes
49 struct rb_node rb_node;
53 * transid where the defrag was added, we search for
54 * extents newer than this
61 /* last offset we were able to defrag */
64 /* if we've wrapped around back to zero once already */
68 static int __compare_inode_defrag(struct inode_defrag *defrag1,
69 struct inode_defrag *defrag2)
71 if (defrag1->root > defrag2->root)
73 else if (defrag1->root < defrag2->root)
75 else if (defrag1->ino > defrag2->ino)
77 else if (defrag1->ino < defrag2->ino)
83 /* pop a record for an inode into the defrag tree. The lock
84 * must be held already
86 * If you're inserting a record for an older transid than an
87 * existing record, the transid already in the tree is lowered
89 * If an existing record is found the defrag item you
92 static void __btrfs_add_inode_defrag(struct inode *inode,
93 struct inode_defrag *defrag)
95 struct btrfs_root *root = BTRFS_I(inode)->root;
96 struct inode_defrag *entry;
98 struct rb_node *parent = NULL;
101 p = &root->fs_info->defrag_inodes.rb_node;
104 entry = rb_entry(parent, struct inode_defrag, rb_node);
106 ret = __compare_inode_defrag(defrag, entry);
108 p = &parent->rb_left;
110 p = &parent->rb_right;
112 /* if we're reinserting an entry for
113 * an old defrag run, make sure to
114 * lower the transid of our existing record
116 if (defrag->transid < entry->transid)
117 entry->transid = defrag->transid;
118 if (defrag->last_offset > entry->last_offset)
119 entry->last_offset = defrag->last_offset;
123 set_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
124 rb_link_node(&defrag->rb_node, parent, p);
125 rb_insert_color(&defrag->rb_node, &root->fs_info->defrag_inodes);
135 * insert a defrag record for this inode if auto defrag is
138 int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans,
141 struct btrfs_root *root = BTRFS_I(inode)->root;
142 struct inode_defrag *defrag;
145 if (!btrfs_test_opt(root, AUTO_DEFRAG))
148 if (btrfs_fs_closing(root->fs_info))
151 if (test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags))
155 transid = trans->transid;
157 transid = BTRFS_I(inode)->root->last_trans;
159 defrag = kzalloc(sizeof(*defrag), GFP_NOFS);
163 defrag->ino = btrfs_ino(inode);
164 defrag->transid = transid;
165 defrag->root = root->root_key.objectid;
167 spin_lock(&root->fs_info->defrag_inodes_lock);
168 if (!test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags))
169 __btrfs_add_inode_defrag(inode, defrag);
172 spin_unlock(&root->fs_info->defrag_inodes_lock);
177 * must be called with the defrag_inodes lock held
179 struct inode_defrag *btrfs_find_defrag_inode(struct btrfs_fs_info *info,
181 struct rb_node **next)
183 struct inode_defrag *entry = NULL;
184 struct inode_defrag tmp;
186 struct rb_node *parent = NULL;
192 p = info->defrag_inodes.rb_node;
195 entry = rb_entry(parent, struct inode_defrag, rb_node);
197 ret = __compare_inode_defrag(&tmp, entry);
201 p = parent->rb_right;
207 while (parent && __compare_inode_defrag(&tmp, entry) > 0) {
208 parent = rb_next(parent);
209 entry = rb_entry(parent, struct inode_defrag, rb_node);
217 * run through the list of inodes in the FS that need
220 int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info)
222 struct inode_defrag *defrag;
223 struct btrfs_root *inode_root;
226 struct btrfs_key key;
227 struct btrfs_ioctl_defrag_range_args range;
229 u64 root_objectid = 0;
231 int defrag_batch = 1024;
233 memset(&range, 0, sizeof(range));
236 atomic_inc(&fs_info->defrag_running);
237 spin_lock(&fs_info->defrag_inodes_lock);
241 /* find an inode to defrag */
242 defrag = btrfs_find_defrag_inode(fs_info, root_objectid,
246 defrag = rb_entry(n, struct inode_defrag,
248 } else if (root_objectid || first_ino) {
257 /* remove it from the rbtree */
258 first_ino = defrag->ino + 1;
259 root_objectid = defrag->root;
260 rb_erase(&defrag->rb_node, &fs_info->defrag_inodes);
262 if (btrfs_fs_closing(fs_info))
265 spin_unlock(&fs_info->defrag_inodes_lock);
268 key.objectid = defrag->root;
269 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
270 key.offset = (u64)-1;
271 inode_root = btrfs_read_fs_root_no_name(fs_info, &key);
272 if (IS_ERR(inode_root))
275 key.objectid = defrag->ino;
276 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
279 inode = btrfs_iget(fs_info->sb, &key, inode_root, NULL);
283 /* do a chunk of defrag */
284 clear_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
285 range.start = defrag->last_offset;
286 num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid,
289 * if we filled the whole defrag batch, there
290 * must be more work to do. Queue this defrag
293 if (num_defrag == defrag_batch) {
294 defrag->last_offset = range.start;
295 __btrfs_add_inode_defrag(inode, defrag);
297 * we don't want to kfree defrag, we added it back to
301 } else if (defrag->last_offset && !defrag->cycled) {
303 * we didn't fill our defrag batch, but
304 * we didn't start at zero. Make sure we loop
305 * around to the start of the file.
307 defrag->last_offset = 0;
309 __btrfs_add_inode_defrag(inode, defrag);
315 spin_lock(&fs_info->defrag_inodes_lock);
319 spin_unlock(&fs_info->defrag_inodes_lock);
321 atomic_dec(&fs_info->defrag_running);
324 * during unmount, we use the transaction_wait queue to
325 * wait for the defragger to stop
327 wake_up(&fs_info->transaction_wait);
331 /* simple helper to fault in pages and copy. This should go away
332 * and be replaced with calls into generic code.
334 static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
336 struct page **prepared_pages,
340 size_t total_copied = 0;
342 int offset = pos & (PAGE_CACHE_SIZE - 1);
344 while (write_bytes > 0) {
345 size_t count = min_t(size_t,
346 PAGE_CACHE_SIZE - offset, write_bytes);
347 struct page *page = prepared_pages[pg];
349 * Copy data from userspace to the current page
351 * Disable pagefault to avoid recursive lock since
352 * the pages are already locked
355 copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
358 /* Flush processor's dcache for this page */
359 flush_dcache_page(page);
362 * if we get a partial write, we can end up with
363 * partially up to date pages. These add
364 * a lot of complexity, so make sure they don't
365 * happen by forcing this copy to be retried.
367 * The rest of the btrfs_file_write code will fall
368 * back to page at a time copies after we return 0.
370 if (!PageUptodate(page) && copied < count)
373 iov_iter_advance(i, copied);
374 write_bytes -= copied;
375 total_copied += copied;
377 /* Return to btrfs_file_aio_write to fault page */
378 if (unlikely(copied == 0))
381 if (unlikely(copied < PAGE_CACHE_SIZE - offset)) {
392 * unlocks pages after btrfs_file_write is done with them
394 void btrfs_drop_pages(struct page **pages, size_t num_pages)
397 for (i = 0; i < num_pages; i++) {
398 /* page checked is some magic around finding pages that
399 * have been modified without going through btrfs_set_page_dirty
402 ClearPageChecked(pages[i]);
403 unlock_page(pages[i]);
404 mark_page_accessed(pages[i]);
405 page_cache_release(pages[i]);
410 * after copy_from_user, pages need to be dirtied and we need to make
411 * sure holes are created between the current EOF and the start of
412 * any next extents (if required).
414 * this also makes the decision about creating an inline extent vs
415 * doing real data extents, marking pages dirty and delalloc as required.
417 int btrfs_dirty_pages(struct btrfs_root *root, struct inode *inode,
418 struct page **pages, size_t num_pages,
419 loff_t pos, size_t write_bytes,
420 struct extent_state **cached)
426 u64 end_of_last_block;
427 u64 end_pos = pos + write_bytes;
428 loff_t isize = i_size_read(inode);
430 start_pos = pos & ~((u64)root->sectorsize - 1);
431 num_bytes = (write_bytes + pos - start_pos +
432 root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
434 end_of_last_block = start_pos + num_bytes - 1;
435 err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
440 for (i = 0; i < num_pages; i++) {
441 struct page *p = pages[i];
448 * we've only changed i_size in ram, and we haven't updated
449 * the disk i_size. There is no need to log the inode
453 i_size_write(inode, end_pos);
458 * this drops all the extents in the cache that intersect the range
459 * [start, end]. Existing extents are split as required.
461 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
464 struct extent_map *em;
465 struct extent_map *split = NULL;
466 struct extent_map *split2 = NULL;
467 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
468 u64 len = end - start + 1;
475 WARN_ON(end < start);
476 if (end == (u64)-1) {
482 split = alloc_extent_map();
484 split2 = alloc_extent_map();
485 BUG_ON(!split || !split2); /* -ENOMEM */
487 write_lock(&em_tree->lock);
488 em = lookup_extent_mapping(em_tree, start, len);
490 write_unlock(&em_tree->lock);
494 gen = em->generation;
495 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
496 if (testend && em->start + em->len >= start + len) {
498 write_unlock(&em_tree->lock);
501 start = em->start + em->len;
503 len = start + len - (em->start + em->len);
505 write_unlock(&em_tree->lock);
508 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
509 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
510 remove_extent_mapping(em_tree, em);
512 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
514 split->start = em->start;
515 split->len = start - em->start;
516 split->orig_start = em->orig_start;
517 split->block_start = em->block_start;
520 split->block_len = em->block_len;
522 split->block_len = split->len;
523 split->generation = gen;
524 split->bdev = em->bdev;
525 split->flags = flags;
526 split->compress_type = em->compress_type;
527 ret = add_extent_mapping(em_tree, split);
528 BUG_ON(ret); /* Logic error */
529 list_move(&split->list, &em_tree->modified_extents);
530 free_extent_map(split);
534 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
535 testend && em->start + em->len > start + len) {
536 u64 diff = start + len - em->start;
538 split->start = start + len;
539 split->len = em->start + em->len - (start + len);
540 split->bdev = em->bdev;
541 split->flags = flags;
542 split->compress_type = em->compress_type;
543 split->generation = gen;
546 split->block_len = em->block_len;
547 split->block_start = em->block_start;
548 split->orig_start = em->orig_start;
550 split->block_len = split->len;
551 split->block_start = em->block_start + diff;
552 split->orig_start = split->start;
555 ret = add_extent_mapping(em_tree, split);
556 BUG_ON(ret); /* Logic error */
557 list_move(&split->list, &em_tree->modified_extents);
558 free_extent_map(split);
561 write_unlock(&em_tree->lock);
565 /* once for the tree*/
569 free_extent_map(split);
571 free_extent_map(split2);
576 * this is very complex, but the basic idea is to drop all extents
577 * in the range start - end. hint_block is filled in with a block number
578 * that would be a good hint to the block allocator for this file.
580 * If an extent intersects the range but is not entirely inside the range
581 * it is either truncated or split. Anything entirely inside the range
582 * is deleted from the tree.
584 int __btrfs_drop_extents(struct btrfs_trans_handle *trans,
585 struct btrfs_root *root, struct inode *inode,
586 struct btrfs_path *path, u64 start, u64 end,
587 u64 *hint_byte, int drop_cache)
589 struct extent_buffer *leaf;
590 struct btrfs_file_extent_item *fi;
591 struct btrfs_key key;
592 struct btrfs_key new_key;
593 u64 ino = btrfs_ino(inode);
594 u64 search_start = start;
597 u64 extent_offset = 0;
604 int modify_tree = -1;
605 int update_refs = (root->ref_cows || root == root->fs_info->tree_root);
608 btrfs_drop_extent_cache(inode, start, end - 1, 0);
610 if (start >= BTRFS_I(inode)->disk_i_size)
615 ret = btrfs_lookup_file_extent(trans, root, path, ino,
616 search_start, modify_tree);
619 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
620 leaf = path->nodes[0];
621 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
622 if (key.objectid == ino &&
623 key.type == BTRFS_EXTENT_DATA_KEY)
628 leaf = path->nodes[0];
629 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
631 ret = btrfs_next_leaf(root, path);
638 leaf = path->nodes[0];
642 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
643 if (key.objectid > ino ||
644 key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
647 fi = btrfs_item_ptr(leaf, path->slots[0],
648 struct btrfs_file_extent_item);
649 extent_type = btrfs_file_extent_type(leaf, fi);
651 if (extent_type == BTRFS_FILE_EXTENT_REG ||
652 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
653 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
654 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
655 extent_offset = btrfs_file_extent_offset(leaf, fi);
656 extent_end = key.offset +
657 btrfs_file_extent_num_bytes(leaf, fi);
658 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
659 extent_end = key.offset +
660 btrfs_file_extent_inline_len(leaf, fi);
663 extent_end = search_start;
666 if (extent_end <= search_start) {
671 search_start = max(key.offset, start);
672 if (recow || !modify_tree) {
674 btrfs_release_path(path);
679 * | - range to drop - |
680 * | -------- extent -------- |
682 if (start > key.offset && end < extent_end) {
684 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
686 memcpy(&new_key, &key, sizeof(new_key));
687 new_key.offset = start;
688 ret = btrfs_duplicate_item(trans, root, path,
690 if (ret == -EAGAIN) {
691 btrfs_release_path(path);
697 leaf = path->nodes[0];
698 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
699 struct btrfs_file_extent_item);
700 btrfs_set_file_extent_num_bytes(leaf, fi,
703 fi = btrfs_item_ptr(leaf, path->slots[0],
704 struct btrfs_file_extent_item);
706 extent_offset += start - key.offset;
707 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
708 btrfs_set_file_extent_num_bytes(leaf, fi,
710 btrfs_mark_buffer_dirty(leaf);
712 if (update_refs && disk_bytenr > 0) {
713 ret = btrfs_inc_extent_ref(trans, root,
714 disk_bytenr, num_bytes, 0,
715 root->root_key.objectid,
717 start - extent_offset, 0);
718 BUG_ON(ret); /* -ENOMEM */
719 *hint_byte = disk_bytenr;
724 * | ---- range to drop ----- |
725 * | -------- extent -------- |
727 if (start <= key.offset && end < extent_end) {
728 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
730 memcpy(&new_key, &key, sizeof(new_key));
731 new_key.offset = end;
732 btrfs_set_item_key_safe(trans, root, path, &new_key);
734 extent_offset += end - key.offset;
735 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
736 btrfs_set_file_extent_num_bytes(leaf, fi,
738 btrfs_mark_buffer_dirty(leaf);
739 if (update_refs && disk_bytenr > 0) {
740 inode_sub_bytes(inode, end - key.offset);
741 *hint_byte = disk_bytenr;
746 search_start = extent_end;
748 * | ---- range to drop ----- |
749 * | -------- extent -------- |
751 if (start > key.offset && end >= extent_end) {
753 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
755 btrfs_set_file_extent_num_bytes(leaf, fi,
757 btrfs_mark_buffer_dirty(leaf);
758 if (update_refs && disk_bytenr > 0) {
759 inode_sub_bytes(inode, extent_end - start);
760 *hint_byte = disk_bytenr;
762 if (end == extent_end)
770 * | ---- range to drop ----- |
771 * | ------ extent ------ |
773 if (start <= key.offset && end >= extent_end) {
775 del_slot = path->slots[0];
778 BUG_ON(del_slot + del_nr != path->slots[0]);
783 extent_type == BTRFS_FILE_EXTENT_INLINE) {
784 inode_sub_bytes(inode,
785 extent_end - key.offset);
786 extent_end = ALIGN(extent_end,
788 } else if (update_refs && disk_bytenr > 0) {
789 ret = btrfs_free_extent(trans, root,
790 disk_bytenr, num_bytes, 0,
791 root->root_key.objectid,
792 key.objectid, key.offset -
794 BUG_ON(ret); /* -ENOMEM */
795 inode_sub_bytes(inode,
796 extent_end - key.offset);
797 *hint_byte = disk_bytenr;
800 if (end == extent_end)
803 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
808 ret = btrfs_del_items(trans, root, path, del_slot,
811 btrfs_abort_transaction(trans, root, ret);
818 btrfs_release_path(path);
825 if (!ret && del_nr > 0) {
826 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
828 btrfs_abort_transaction(trans, root, ret);
831 btrfs_release_path(path);
835 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
836 struct btrfs_root *root, struct inode *inode, u64 start,
837 u64 end, u64 *hint_byte, int drop_cache)
839 struct btrfs_path *path;
842 path = btrfs_alloc_path();
845 ret = __btrfs_drop_extents(trans, root, inode, path, start, end,
846 hint_byte, drop_cache);
847 btrfs_free_path(path);
851 static int extent_mergeable(struct extent_buffer *leaf, int slot,
852 u64 objectid, u64 bytenr, u64 orig_offset,
853 u64 *start, u64 *end)
855 struct btrfs_file_extent_item *fi;
856 struct btrfs_key key;
859 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
862 btrfs_item_key_to_cpu(leaf, &key, slot);
863 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
866 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
867 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
868 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
869 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
870 btrfs_file_extent_compression(leaf, fi) ||
871 btrfs_file_extent_encryption(leaf, fi) ||
872 btrfs_file_extent_other_encoding(leaf, fi))
875 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
876 if ((*start && *start != key.offset) || (*end && *end != extent_end))
885 * Mark extent in the range start - end as written.
887 * This changes extent type from 'pre-allocated' to 'regular'. If only
888 * part of extent is marked as written, the extent will be split into
891 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
892 struct inode *inode, u64 start, u64 end)
894 struct btrfs_root *root = BTRFS_I(inode)->root;
895 struct extent_buffer *leaf;
896 struct btrfs_path *path;
897 struct btrfs_file_extent_item *fi;
898 struct btrfs_key key;
899 struct btrfs_key new_key;
911 u64 ino = btrfs_ino(inode);
913 path = btrfs_alloc_path();
920 key.type = BTRFS_EXTENT_DATA_KEY;
923 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
926 if (ret > 0 && path->slots[0] > 0)
929 leaf = path->nodes[0];
930 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
931 BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY);
932 fi = btrfs_item_ptr(leaf, path->slots[0],
933 struct btrfs_file_extent_item);
934 BUG_ON(btrfs_file_extent_type(leaf, fi) !=
935 BTRFS_FILE_EXTENT_PREALLOC);
936 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
937 BUG_ON(key.offset > start || extent_end < end);
939 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
940 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
941 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
942 memcpy(&new_key, &key, sizeof(new_key));
944 if (start == key.offset && end < extent_end) {
947 if (extent_mergeable(leaf, path->slots[0] - 1,
948 ino, bytenr, orig_offset,
949 &other_start, &other_end)) {
950 new_key.offset = end;
951 btrfs_set_item_key_safe(trans, root, path, &new_key);
952 fi = btrfs_item_ptr(leaf, path->slots[0],
953 struct btrfs_file_extent_item);
954 btrfs_set_file_extent_generation(leaf, fi,
956 btrfs_set_file_extent_num_bytes(leaf, fi,
958 btrfs_set_file_extent_offset(leaf, fi,
960 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
961 struct btrfs_file_extent_item);
962 btrfs_set_file_extent_generation(leaf, fi,
964 btrfs_set_file_extent_num_bytes(leaf, fi,
966 btrfs_mark_buffer_dirty(leaf);
971 if (start > key.offset && end == extent_end) {
974 if (extent_mergeable(leaf, path->slots[0] + 1,
975 ino, bytenr, orig_offset,
976 &other_start, &other_end)) {
977 fi = btrfs_item_ptr(leaf, path->slots[0],
978 struct btrfs_file_extent_item);
979 btrfs_set_file_extent_num_bytes(leaf, fi,
981 btrfs_set_file_extent_generation(leaf, fi,
984 new_key.offset = start;
985 btrfs_set_item_key_safe(trans, root, path, &new_key);
987 fi = btrfs_item_ptr(leaf, path->slots[0],
988 struct btrfs_file_extent_item);
989 btrfs_set_file_extent_generation(leaf, fi,
991 btrfs_set_file_extent_num_bytes(leaf, fi,
993 btrfs_set_file_extent_offset(leaf, fi,
994 start - orig_offset);
995 btrfs_mark_buffer_dirty(leaf);
1000 while (start > key.offset || end < extent_end) {
1001 if (key.offset == start)
1004 new_key.offset = split;
1005 ret = btrfs_duplicate_item(trans, root, path, &new_key);
1006 if (ret == -EAGAIN) {
1007 btrfs_release_path(path);
1011 btrfs_abort_transaction(trans, root, ret);
1015 leaf = path->nodes[0];
1016 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1017 struct btrfs_file_extent_item);
1018 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1019 btrfs_set_file_extent_num_bytes(leaf, fi,
1020 split - key.offset);
1022 fi = btrfs_item_ptr(leaf, path->slots[0],
1023 struct btrfs_file_extent_item);
1025 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1026 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
1027 btrfs_set_file_extent_num_bytes(leaf, fi,
1028 extent_end - split);
1029 btrfs_mark_buffer_dirty(leaf);
1031 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
1032 root->root_key.objectid,
1033 ino, orig_offset, 0);
1034 BUG_ON(ret); /* -ENOMEM */
1036 if (split == start) {
1039 BUG_ON(start != key.offset);
1048 if (extent_mergeable(leaf, path->slots[0] + 1,
1049 ino, bytenr, orig_offset,
1050 &other_start, &other_end)) {
1052 btrfs_release_path(path);
1055 extent_end = other_end;
1056 del_slot = path->slots[0] + 1;
1058 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1059 0, root->root_key.objectid,
1060 ino, orig_offset, 0);
1061 BUG_ON(ret); /* -ENOMEM */
1065 if (extent_mergeable(leaf, path->slots[0] - 1,
1066 ino, bytenr, orig_offset,
1067 &other_start, &other_end)) {
1069 btrfs_release_path(path);
1072 key.offset = other_start;
1073 del_slot = path->slots[0];
1075 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1076 0, root->root_key.objectid,
1077 ino, orig_offset, 0);
1078 BUG_ON(ret); /* -ENOMEM */
1081 fi = btrfs_item_ptr(leaf, path->slots[0],
1082 struct btrfs_file_extent_item);
1083 btrfs_set_file_extent_type(leaf, fi,
1084 BTRFS_FILE_EXTENT_REG);
1085 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1086 btrfs_mark_buffer_dirty(leaf);
1088 fi = btrfs_item_ptr(leaf, del_slot - 1,
1089 struct btrfs_file_extent_item);
1090 btrfs_set_file_extent_type(leaf, fi,
1091 BTRFS_FILE_EXTENT_REG);
1092 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1093 btrfs_set_file_extent_num_bytes(leaf, fi,
1094 extent_end - key.offset);
1095 btrfs_mark_buffer_dirty(leaf);
1097 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1099 btrfs_abort_transaction(trans, root, ret);
1104 btrfs_free_path(path);
1109 * on error we return an unlocked page and the error value
1110 * on success we return a locked page and 0
1112 static int prepare_uptodate_page(struct page *page, u64 pos,
1113 bool force_uptodate)
1117 if (((pos & (PAGE_CACHE_SIZE - 1)) || force_uptodate) &&
1118 !PageUptodate(page)) {
1119 ret = btrfs_readpage(NULL, page);
1123 if (!PageUptodate(page)) {
1132 * this gets pages into the page cache and locks them down, it also properly
1133 * waits for data=ordered extents to finish before allowing the pages to be
1136 static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
1137 struct page **pages, size_t num_pages,
1138 loff_t pos, unsigned long first_index,
1139 size_t write_bytes, bool force_uptodate)
1141 struct extent_state *cached_state = NULL;
1143 unsigned long index = pos >> PAGE_CACHE_SHIFT;
1144 struct inode *inode = fdentry(file)->d_inode;
1145 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1151 start_pos = pos & ~((u64)root->sectorsize - 1);
1152 last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
1155 for (i = 0; i < num_pages; i++) {
1156 pages[i] = find_or_create_page(inode->i_mapping, index + i,
1157 mask | __GFP_WRITE);
1165 err = prepare_uptodate_page(pages[i], pos,
1167 if (i == num_pages - 1)
1168 err = prepare_uptodate_page(pages[i],
1169 pos + write_bytes, false);
1171 page_cache_release(pages[i]);
1175 wait_on_page_writeback(pages[i]);
1178 if (start_pos < inode->i_size) {
1179 struct btrfs_ordered_extent *ordered;
1180 lock_extent_bits(&BTRFS_I(inode)->io_tree,
1181 start_pos, last_pos - 1, 0, &cached_state);
1182 ordered = btrfs_lookup_first_ordered_extent(inode,
1185 ordered->file_offset + ordered->len > start_pos &&
1186 ordered->file_offset < last_pos) {
1187 btrfs_put_ordered_extent(ordered);
1188 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1189 start_pos, last_pos - 1,
1190 &cached_state, GFP_NOFS);
1191 for (i = 0; i < num_pages; i++) {
1192 unlock_page(pages[i]);
1193 page_cache_release(pages[i]);
1195 btrfs_wait_ordered_range(inode, start_pos,
1196 last_pos - start_pos);
1200 btrfs_put_ordered_extent(ordered);
1202 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
1203 last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1204 EXTENT_DO_ACCOUNTING, 0, 0, &cached_state,
1206 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1207 start_pos, last_pos - 1, &cached_state,
1210 for (i = 0; i < num_pages; i++) {
1211 if (clear_page_dirty_for_io(pages[i]))
1212 account_page_redirty(pages[i]);
1213 set_page_extent_mapped(pages[i]);
1214 WARN_ON(!PageLocked(pages[i]));
1218 while (faili >= 0) {
1219 unlock_page(pages[faili]);
1220 page_cache_release(pages[faili]);
1227 static noinline ssize_t __btrfs_buffered_write(struct file *file,
1231 struct inode *inode = fdentry(file)->d_inode;
1232 struct btrfs_root *root = BTRFS_I(inode)->root;
1233 struct page **pages = NULL;
1234 unsigned long first_index;
1235 size_t num_written = 0;
1238 bool force_page_uptodate = false;
1240 nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) /
1241 PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
1242 (sizeof(struct page *)));
1243 nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1244 nrptrs = max(nrptrs, 8);
1245 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
1249 first_index = pos >> PAGE_CACHE_SHIFT;
1251 while (iov_iter_count(i) > 0) {
1252 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1253 size_t write_bytes = min(iov_iter_count(i),
1254 nrptrs * (size_t)PAGE_CACHE_SIZE -
1256 size_t num_pages = (write_bytes + offset +
1257 PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1261 WARN_ON(num_pages > nrptrs);
1264 * Fault pages before locking them in prepare_pages
1265 * to avoid recursive lock
1267 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1272 ret = btrfs_delalloc_reserve_space(inode,
1273 num_pages << PAGE_CACHE_SHIFT);
1278 * This is going to setup the pages array with the number of
1279 * pages we want, so we don't really need to worry about the
1280 * contents of pages from loop to loop
1282 ret = prepare_pages(root, file, pages, num_pages,
1283 pos, first_index, write_bytes,
1284 force_page_uptodate);
1286 btrfs_delalloc_release_space(inode,
1287 num_pages << PAGE_CACHE_SHIFT);
1291 copied = btrfs_copy_from_user(pos, num_pages,
1292 write_bytes, pages, i);
1295 * if we have trouble faulting in the pages, fall
1296 * back to one page at a time
1298 if (copied < write_bytes)
1302 force_page_uptodate = true;
1305 force_page_uptodate = false;
1306 dirty_pages = (copied + offset +
1307 PAGE_CACHE_SIZE - 1) >>
1312 * If we had a short copy we need to release the excess delaloc
1313 * bytes we reserved. We need to increment outstanding_extents
1314 * because btrfs_delalloc_release_space will decrement it, but
1315 * we still have an outstanding extent for the chunk we actually
1318 if (num_pages > dirty_pages) {
1320 spin_lock(&BTRFS_I(inode)->lock);
1321 BTRFS_I(inode)->outstanding_extents++;
1322 spin_unlock(&BTRFS_I(inode)->lock);
1324 btrfs_delalloc_release_space(inode,
1325 (num_pages - dirty_pages) <<
1330 ret = btrfs_dirty_pages(root, inode, pages,
1331 dirty_pages, pos, copied,
1334 btrfs_delalloc_release_space(inode,
1335 dirty_pages << PAGE_CACHE_SHIFT);
1336 btrfs_drop_pages(pages, num_pages);
1341 btrfs_drop_pages(pages, num_pages);
1345 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
1347 if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1348 btrfs_btree_balance_dirty(root, 1);
1351 num_written += copied;
1356 return num_written ? num_written : ret;
1359 static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1360 const struct iovec *iov,
1361 unsigned long nr_segs, loff_t pos,
1362 loff_t *ppos, size_t count, size_t ocount)
1364 struct file *file = iocb->ki_filp;
1367 ssize_t written_buffered;
1371 written = generic_file_direct_write(iocb, iov, &nr_segs, pos, ppos,
1374 if (written < 0 || written == count)
1379 iov_iter_init(&i, iov, nr_segs, count, written);
1380 written_buffered = __btrfs_buffered_write(file, &i, pos);
1381 if (written_buffered < 0) {
1382 err = written_buffered;
1385 endbyte = pos + written_buffered - 1;
1386 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
1389 written += written_buffered;
1390 *ppos = pos + written_buffered;
1391 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
1392 endbyte >> PAGE_CACHE_SHIFT);
1394 return written ? written : err;
1397 static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
1398 const struct iovec *iov,
1399 unsigned long nr_segs, loff_t pos)
1401 struct file *file = iocb->ki_filp;
1402 struct inode *inode = fdentry(file)->d_inode;
1403 struct btrfs_root *root = BTRFS_I(inode)->root;
1404 loff_t *ppos = &iocb->ki_pos;
1406 ssize_t num_written = 0;
1408 size_t count, ocount;
1410 sb_start_write(inode->i_sb);
1412 mutex_lock(&inode->i_mutex);
1414 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
1416 mutex_unlock(&inode->i_mutex);
1421 current->backing_dev_info = inode->i_mapping->backing_dev_info;
1422 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1424 mutex_unlock(&inode->i_mutex);
1429 mutex_unlock(&inode->i_mutex);
1433 err = file_remove_suid(file);
1435 mutex_unlock(&inode->i_mutex);
1440 * If BTRFS flips readonly due to some impossible error
1441 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1442 * although we have opened a file as writable, we have
1443 * to stop this write operation to ensure FS consistency.
1445 if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
1446 mutex_unlock(&inode->i_mutex);
1451 err = file_update_time(file);
1453 mutex_unlock(&inode->i_mutex);
1457 start_pos = round_down(pos, root->sectorsize);
1458 if (start_pos > i_size_read(inode)) {
1459 err = btrfs_cont_expand(inode, i_size_read(inode), start_pos);
1461 mutex_unlock(&inode->i_mutex);
1466 if (unlikely(file->f_flags & O_DIRECT)) {
1467 num_written = __btrfs_direct_write(iocb, iov, nr_segs,
1468 pos, ppos, count, ocount);
1472 iov_iter_init(&i, iov, nr_segs, count, num_written);
1474 num_written = __btrfs_buffered_write(file, &i, pos);
1475 if (num_written > 0)
1476 *ppos = pos + num_written;
1479 mutex_unlock(&inode->i_mutex);
1482 * we want to make sure fsync finds this change
1483 * but we haven't joined a transaction running right now.
1485 * Later on, someone is sure to update the inode and get the
1486 * real transid recorded.
1488 * We set last_trans now to the fs_info generation + 1,
1489 * this will either be one more than the running transaction
1490 * or the generation used for the next transaction if there isn't
1491 * one running right now.
1493 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1494 if (num_written > 0 || num_written == -EIOCBQUEUED) {
1495 err = generic_write_sync(file, pos, num_written);
1496 if (err < 0 && num_written > 0)
1500 sb_end_write(inode->i_sb);
1501 current->backing_dev_info = NULL;
1502 return num_written ? num_written : err;
1505 int btrfs_release_file(struct inode *inode, struct file *filp)
1508 * ordered_data_close is set by settattr when we are about to truncate
1509 * a file from a non-zero size to a zero size. This tries to
1510 * flush down new bytes that may have been written if the
1511 * application were using truncate to replace a file in place.
1513 if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
1514 &BTRFS_I(inode)->runtime_flags)) {
1515 btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode);
1516 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1517 filemap_flush(inode->i_mapping);
1519 if (filp->private_data)
1520 btrfs_ioctl_trans_end(filp);
1525 * fsync call for both files and directories. This logs the inode into
1526 * the tree log instead of forcing full commits whenever possible.
1528 * It needs to call filemap_fdatawait so that all ordered extent updates are
1529 * in the metadata btree are up to date for copying to the log.
1531 * It drops the inode mutex before doing the tree log commit. This is an
1532 * important optimization for directories because holding the mutex prevents
1533 * new operations on the dir while we write to disk.
1535 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1537 struct dentry *dentry = file->f_path.dentry;
1538 struct inode *inode = dentry->d_inode;
1539 struct btrfs_root *root = BTRFS_I(inode)->root;
1541 struct btrfs_trans_handle *trans;
1543 trace_btrfs_sync_file(file, datasync);
1545 mutex_lock(&inode->i_mutex);
1548 * we wait first, since the writeback may change the inode, also wait
1549 * ordered range does a filemape_write_and_wait_range which is why we
1550 * don't do it above like other file systems.
1553 btrfs_wait_ordered_range(inode, start, end);
1557 * check the transaction that last modified this inode
1558 * and see if its already been committed
1560 if (!BTRFS_I(inode)->last_trans) {
1561 mutex_unlock(&inode->i_mutex);
1566 * if the last transaction that changed this file was before
1567 * the current transaction, we can bail out now without any
1571 if (btrfs_inode_in_log(inode, root->fs_info->generation) ||
1572 BTRFS_I(inode)->last_trans <=
1573 root->fs_info->last_trans_committed) {
1574 BTRFS_I(inode)->last_trans = 0;
1577 * We'v had everything committed since the last time we were
1578 * modified so clear this flag in case it was set for whatever
1579 * reason, it's no longer relevant.
1581 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1582 &BTRFS_I(inode)->runtime_flags);
1583 mutex_unlock(&inode->i_mutex);
1588 * ok we haven't committed the transaction yet, lets do a commit
1590 if (file->private_data)
1591 btrfs_ioctl_trans_end(file);
1593 trans = btrfs_start_transaction(root, 0);
1594 if (IS_ERR(trans)) {
1595 ret = PTR_ERR(trans);
1596 mutex_unlock(&inode->i_mutex);
1600 ret = btrfs_log_dentry_safe(trans, root, dentry);
1602 mutex_unlock(&inode->i_mutex);
1606 /* we've logged all the items and now have a consistent
1607 * version of the file in the log. It is possible that
1608 * someone will come in and modify the file, but that's
1609 * fine because the log is consistent on disk, and we
1610 * have references to all of the file's extents
1612 * It is possible that someone will come in and log the
1613 * file again, but that will end up using the synchronization
1614 * inside btrfs_sync_log to keep things safe.
1616 mutex_unlock(&inode->i_mutex);
1618 if (ret != BTRFS_NO_LOG_SYNC) {
1620 ret = btrfs_commit_transaction(trans, root);
1622 ret = btrfs_sync_log(trans, root);
1624 ret = btrfs_end_transaction(trans, root);
1626 ret = btrfs_commit_transaction(trans, root);
1629 ret = btrfs_end_transaction(trans, root);
1632 return ret > 0 ? -EIO : ret;
1635 static const struct vm_operations_struct btrfs_file_vm_ops = {
1636 .fault = filemap_fault,
1637 .page_mkwrite = btrfs_page_mkwrite,
1640 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
1642 struct address_space *mapping = filp->f_mapping;
1644 if (!mapping->a_ops->readpage)
1647 file_accessed(filp);
1648 vma->vm_ops = &btrfs_file_vm_ops;
1649 vma->vm_flags |= VM_CAN_NONLINEAR;
1654 static long btrfs_fallocate(struct file *file, int mode,
1655 loff_t offset, loff_t len)
1657 struct inode *inode = file->f_path.dentry->d_inode;
1658 struct extent_state *cached_state = NULL;
1665 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
1666 struct extent_map *em;
1669 alloc_start = offset & ~mask;
1670 alloc_end = (offset + len + mask) & ~mask;
1672 /* We only support the FALLOC_FL_KEEP_SIZE mode */
1673 if (mode & ~FALLOC_FL_KEEP_SIZE)
1677 * Make sure we have enough space before we do the
1680 ret = btrfs_check_data_free_space(inode, len);
1685 * wait for ordered IO before we have any locks. We'll loop again
1686 * below with the locks held.
1688 btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
1690 mutex_lock(&inode->i_mutex);
1691 ret = inode_newsize_ok(inode, alloc_end);
1695 if (alloc_start > inode->i_size) {
1696 ret = btrfs_cont_expand(inode, i_size_read(inode),
1702 locked_end = alloc_end - 1;
1704 struct btrfs_ordered_extent *ordered;
1706 /* the extent lock is ordered inside the running
1709 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
1710 locked_end, 0, &cached_state);
1711 ordered = btrfs_lookup_first_ordered_extent(inode,
1714 ordered->file_offset + ordered->len > alloc_start &&
1715 ordered->file_offset < alloc_end) {
1716 btrfs_put_ordered_extent(ordered);
1717 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1718 alloc_start, locked_end,
1719 &cached_state, GFP_NOFS);
1721 * we can't wait on the range with the transaction
1722 * running or with the extent lock held
1724 btrfs_wait_ordered_range(inode, alloc_start,
1725 alloc_end - alloc_start);
1728 btrfs_put_ordered_extent(ordered);
1733 cur_offset = alloc_start;
1737 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
1738 alloc_end - cur_offset, 0);
1739 if (IS_ERR_OR_NULL(em)) {
1746 last_byte = min(extent_map_end(em), alloc_end);
1747 actual_end = min_t(u64, extent_map_end(em), offset + len);
1748 last_byte = (last_byte + mask) & ~mask;
1750 if (em->block_start == EXTENT_MAP_HOLE ||
1751 (cur_offset >= inode->i_size &&
1752 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
1753 ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
1754 last_byte - cur_offset,
1755 1 << inode->i_blkbits,
1760 free_extent_map(em);
1763 } else if (actual_end > inode->i_size &&
1764 !(mode & FALLOC_FL_KEEP_SIZE)) {
1766 * We didn't need to allocate any more space, but we
1767 * still extended the size of the file so we need to
1770 inode->i_ctime = CURRENT_TIME;
1771 i_size_write(inode, actual_end);
1772 btrfs_ordered_update_i_size(inode, actual_end, NULL);
1774 free_extent_map(em);
1776 cur_offset = last_byte;
1777 if (cur_offset >= alloc_end) {
1782 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
1783 &cached_state, GFP_NOFS);
1785 mutex_unlock(&inode->i_mutex);
1786 /* Let go of our reservation. */
1787 btrfs_free_reserved_data_space(inode, len);
1791 static int find_desired_extent(struct inode *inode, loff_t *offset, int origin)
1793 struct btrfs_root *root = BTRFS_I(inode)->root;
1794 struct extent_map *em;
1795 struct extent_state *cached_state = NULL;
1796 u64 lockstart = *offset;
1797 u64 lockend = i_size_read(inode);
1798 u64 start = *offset;
1799 u64 orig_start = *offset;
1800 u64 len = i_size_read(inode);
1804 lockend = max_t(u64, root->sectorsize, lockend);
1805 if (lockend <= lockstart)
1806 lockend = lockstart + root->sectorsize;
1808 len = lockend - lockstart + 1;
1810 len = max_t(u64, len, root->sectorsize);
1811 if (inode->i_size == 0)
1814 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, 0,
1818 * Delalloc is such a pain. If we have a hole and we have pending
1819 * delalloc for a portion of the hole we will get back a hole that
1820 * exists for the entire range since it hasn't been actually written
1821 * yet. So to take care of this case we need to look for an extent just
1822 * before the position we want in case there is outstanding delalloc
1825 if (origin == SEEK_HOLE && start != 0) {
1826 if (start <= root->sectorsize)
1827 em = btrfs_get_extent_fiemap(inode, NULL, 0, 0,
1828 root->sectorsize, 0);
1830 em = btrfs_get_extent_fiemap(inode, NULL, 0,
1831 start - root->sectorsize,
1832 root->sectorsize, 0);
1837 last_end = em->start + em->len;
1838 if (em->block_start == EXTENT_MAP_DELALLOC)
1839 last_end = min_t(u64, last_end, inode->i_size);
1840 free_extent_map(em);
1844 em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0);
1850 if (em->block_start == EXTENT_MAP_HOLE) {
1851 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
1852 if (last_end <= orig_start) {
1853 free_extent_map(em);
1859 if (origin == SEEK_HOLE) {
1861 free_extent_map(em);
1865 if (origin == SEEK_DATA) {
1866 if (em->block_start == EXTENT_MAP_DELALLOC) {
1867 if (start >= inode->i_size) {
1868 free_extent_map(em);
1875 free_extent_map(em);
1880 start = em->start + em->len;
1881 last_end = em->start + em->len;
1883 if (em->block_start == EXTENT_MAP_DELALLOC)
1884 last_end = min_t(u64, last_end, inode->i_size);
1886 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
1887 free_extent_map(em);
1891 free_extent_map(em);
1895 *offset = min(*offset, inode->i_size);
1897 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
1898 &cached_state, GFP_NOFS);
1902 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int origin)
1904 struct inode *inode = file->f_mapping->host;
1907 mutex_lock(&inode->i_mutex);
1911 offset = generic_file_llseek(file, offset, origin);
1915 if (offset >= i_size_read(inode)) {
1916 mutex_unlock(&inode->i_mutex);
1920 ret = find_desired_extent(inode, &offset, origin);
1922 mutex_unlock(&inode->i_mutex);
1927 if (offset < 0 && !(file->f_mode & FMODE_UNSIGNED_OFFSET)) {
1931 if (offset > inode->i_sb->s_maxbytes) {
1936 /* Special lock needed here? */
1937 if (offset != file->f_pos) {
1938 file->f_pos = offset;
1939 file->f_version = 0;
1942 mutex_unlock(&inode->i_mutex);
1946 const struct file_operations btrfs_file_operations = {
1947 .llseek = btrfs_file_llseek,
1948 .read = do_sync_read,
1949 .write = do_sync_write,
1950 .aio_read = generic_file_aio_read,
1951 .splice_read = generic_file_splice_read,
1952 .aio_write = btrfs_file_aio_write,
1953 .mmap = btrfs_file_mmap,
1954 .open = generic_file_open,
1955 .release = btrfs_release_file,
1956 .fsync = btrfs_sync_file,
1957 .fallocate = btrfs_fallocate,
1958 .unlocked_ioctl = btrfs_ioctl,
1959 #ifdef CONFIG_COMPAT
1960 .compat_ioctl = btrfs_ioctl,