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"
45 * when auto defrag is enabled we
46 * queue up these defrag structs to remember which
47 * inodes need defragging passes
50 struct rb_node rb_node;
54 * transid where the defrag was added, we search for
55 * extents newer than this
62 /* last offset we were able to defrag */
65 /* if we've wrapped around back to zero once already */
69 static int __compare_inode_defrag(struct inode_defrag *defrag1,
70 struct inode_defrag *defrag2)
72 if (defrag1->root > defrag2->root)
74 else if (defrag1->root < defrag2->root)
76 else if (defrag1->ino > defrag2->ino)
78 else if (defrag1->ino < defrag2->ino)
84 /* pop a record for an inode into the defrag tree. The lock
85 * must be held already
87 * If you're inserting a record for an older transid than an
88 * existing record, the transid already in the tree is lowered
90 * If an existing record is found the defrag item you
93 static void __btrfs_add_inode_defrag(struct inode *inode,
94 struct inode_defrag *defrag)
96 struct btrfs_root *root = BTRFS_I(inode)->root;
97 struct inode_defrag *entry;
99 struct rb_node *parent = NULL;
102 p = &root->fs_info->defrag_inodes.rb_node;
105 entry = rb_entry(parent, struct inode_defrag, rb_node);
107 ret = __compare_inode_defrag(defrag, entry);
109 p = &parent->rb_left;
111 p = &parent->rb_right;
113 /* if we're reinserting an entry for
114 * an old defrag run, make sure to
115 * lower the transid of our existing record
117 if (defrag->transid < entry->transid)
118 entry->transid = defrag->transid;
119 if (defrag->last_offset > entry->last_offset)
120 entry->last_offset = defrag->last_offset;
124 set_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
125 rb_link_node(&defrag->rb_node, parent, p);
126 rb_insert_color(&defrag->rb_node, &root->fs_info->defrag_inodes);
136 * insert a defrag record for this inode if auto defrag is
139 int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans,
142 struct btrfs_root *root = BTRFS_I(inode)->root;
143 struct inode_defrag *defrag;
146 if (!btrfs_test_opt(root, AUTO_DEFRAG))
149 if (btrfs_fs_closing(root->fs_info))
152 if (test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags))
156 transid = trans->transid;
158 transid = BTRFS_I(inode)->root->last_trans;
160 defrag = kzalloc(sizeof(*defrag), GFP_NOFS);
164 defrag->ino = btrfs_ino(inode);
165 defrag->transid = transid;
166 defrag->root = root->root_key.objectid;
168 spin_lock(&root->fs_info->defrag_inodes_lock);
169 if (!test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags))
170 __btrfs_add_inode_defrag(inode, defrag);
173 spin_unlock(&root->fs_info->defrag_inodes_lock);
178 * must be called with the defrag_inodes lock held
180 struct inode_defrag *btrfs_find_defrag_inode(struct btrfs_fs_info *info,
182 struct rb_node **next)
184 struct inode_defrag *entry = NULL;
185 struct inode_defrag tmp;
187 struct rb_node *parent = NULL;
193 p = info->defrag_inodes.rb_node;
196 entry = rb_entry(parent, struct inode_defrag, rb_node);
198 ret = __compare_inode_defrag(&tmp, entry);
202 p = parent->rb_right;
208 while (parent && __compare_inode_defrag(&tmp, entry) > 0) {
209 parent = rb_next(parent);
210 entry = rb_entry(parent, struct inode_defrag, rb_node);
218 * run through the list of inodes in the FS that need
221 int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info)
223 struct inode_defrag *defrag;
224 struct btrfs_root *inode_root;
227 struct btrfs_key key;
228 struct btrfs_ioctl_defrag_range_args range;
230 u64 root_objectid = 0;
232 int defrag_batch = 1024;
234 memset(&range, 0, sizeof(range));
237 atomic_inc(&fs_info->defrag_running);
238 spin_lock(&fs_info->defrag_inodes_lock);
242 /* find an inode to defrag */
243 defrag = btrfs_find_defrag_inode(fs_info, root_objectid,
247 defrag = rb_entry(n, struct inode_defrag,
249 } else if (root_objectid || first_ino) {
258 /* remove it from the rbtree */
259 first_ino = defrag->ino + 1;
260 root_objectid = defrag->root;
261 rb_erase(&defrag->rb_node, &fs_info->defrag_inodes);
263 if (btrfs_fs_closing(fs_info))
266 spin_unlock(&fs_info->defrag_inodes_lock);
269 key.objectid = defrag->root;
270 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
271 key.offset = (u64)-1;
272 inode_root = btrfs_read_fs_root_no_name(fs_info, &key);
273 if (IS_ERR(inode_root))
276 key.objectid = defrag->ino;
277 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
280 inode = btrfs_iget(fs_info->sb, &key, inode_root, NULL);
284 /* do a chunk of defrag */
285 clear_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
286 range.start = defrag->last_offset;
287 num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid,
290 * if we filled the whole defrag batch, there
291 * must be more work to do. Queue this defrag
294 if (num_defrag == defrag_batch) {
295 defrag->last_offset = range.start;
296 __btrfs_add_inode_defrag(inode, defrag);
298 * we don't want to kfree defrag, we added it back to
302 } else if (defrag->last_offset && !defrag->cycled) {
304 * we didn't fill our defrag batch, but
305 * we didn't start at zero. Make sure we loop
306 * around to the start of the file.
308 defrag->last_offset = 0;
310 __btrfs_add_inode_defrag(inode, defrag);
316 spin_lock(&fs_info->defrag_inodes_lock);
320 spin_unlock(&fs_info->defrag_inodes_lock);
322 atomic_dec(&fs_info->defrag_running);
325 * during unmount, we use the transaction_wait queue to
326 * wait for the defragger to stop
328 wake_up(&fs_info->transaction_wait);
332 /* simple helper to fault in pages and copy. This should go away
333 * and be replaced with calls into generic code.
335 static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
337 struct page **prepared_pages,
341 size_t total_copied = 0;
343 int offset = pos & (PAGE_CACHE_SIZE - 1);
345 while (write_bytes > 0) {
346 size_t count = min_t(size_t,
347 PAGE_CACHE_SIZE - offset, write_bytes);
348 struct page *page = prepared_pages[pg];
350 * Copy data from userspace to the current page
352 * Disable pagefault to avoid recursive lock since
353 * the pages are already locked
356 copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
359 /* Flush processor's dcache for this page */
360 flush_dcache_page(page);
363 * if we get a partial write, we can end up with
364 * partially up to date pages. These add
365 * a lot of complexity, so make sure they don't
366 * happen by forcing this copy to be retried.
368 * The rest of the btrfs_file_write code will fall
369 * back to page at a time copies after we return 0.
371 if (!PageUptodate(page) && copied < count)
374 iov_iter_advance(i, copied);
375 write_bytes -= copied;
376 total_copied += copied;
378 /* Return to btrfs_file_aio_write to fault page */
379 if (unlikely(copied == 0))
382 if (unlikely(copied < PAGE_CACHE_SIZE - offset)) {
393 * unlocks pages after btrfs_file_write is done with them
395 void btrfs_drop_pages(struct page **pages, size_t num_pages)
398 for (i = 0; i < num_pages; i++) {
399 /* page checked is some magic around finding pages that
400 * have been modified without going through btrfs_set_page_dirty
403 ClearPageChecked(pages[i]);
404 unlock_page(pages[i]);
405 mark_page_accessed(pages[i]);
406 page_cache_release(pages[i]);
411 * after copy_from_user, pages need to be dirtied and we need to make
412 * sure holes are created between the current EOF and the start of
413 * any next extents (if required).
415 * this also makes the decision about creating an inline extent vs
416 * doing real data extents, marking pages dirty and delalloc as required.
418 int btrfs_dirty_pages(struct btrfs_root *root, struct inode *inode,
419 struct page **pages, size_t num_pages,
420 loff_t pos, size_t write_bytes,
421 struct extent_state **cached)
427 u64 end_of_last_block;
428 u64 end_pos = pos + write_bytes;
429 loff_t isize = i_size_read(inode);
431 start_pos = pos & ~((u64)root->sectorsize - 1);
432 num_bytes = (write_bytes + pos - start_pos +
433 root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
435 end_of_last_block = start_pos + num_bytes - 1;
436 err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
441 for (i = 0; i < num_pages; i++) {
442 struct page *p = pages[i];
449 * we've only changed i_size in ram, and we haven't updated
450 * the disk i_size. There is no need to log the inode
454 i_size_write(inode, end_pos);
459 * this drops all the extents in the cache that intersect the range
460 * [start, end]. Existing extents are split as required.
462 void btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
465 struct extent_map *em;
466 struct extent_map *split = NULL;
467 struct extent_map *split2 = NULL;
468 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
469 u64 len = end - start + 1;
476 WARN_ON(end < start);
477 if (end == (u64)-1) {
485 split = alloc_extent_map();
487 split2 = alloc_extent_map();
488 if (!split || !split2)
491 write_lock(&em_tree->lock);
492 em = lookup_extent_mapping(em_tree, start, len);
494 write_unlock(&em_tree->lock);
498 gen = em->generation;
499 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
500 if (testend && em->start + em->len >= start + len) {
502 write_unlock(&em_tree->lock);
505 start = em->start + em->len;
507 len = start + len - (em->start + em->len);
509 write_unlock(&em_tree->lock);
512 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
513 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
514 remove_extent_mapping(em_tree, em);
518 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
520 split->start = em->start;
521 split->len = start - em->start;
522 split->orig_start = em->orig_start;
523 split->block_start = em->block_start;
526 split->block_len = em->block_len;
528 split->block_len = split->len;
529 split->generation = gen;
530 split->bdev = em->bdev;
531 split->flags = flags;
532 split->compress_type = em->compress_type;
533 ret = add_extent_mapping(em_tree, split);
534 BUG_ON(ret); /* Logic error */
535 list_move(&split->list, &em_tree->modified_extents);
536 free_extent_map(split);
540 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
541 testend && em->start + em->len > start + len) {
542 u64 diff = start + len - em->start;
544 split->start = start + len;
545 split->len = em->start + em->len - (start + len);
546 split->bdev = em->bdev;
547 split->flags = flags;
548 split->compress_type = em->compress_type;
549 split->generation = gen;
552 split->block_len = em->block_len;
553 split->block_start = em->block_start;
554 split->orig_start = em->orig_start;
556 split->block_len = split->len;
557 split->block_start = em->block_start + diff;
558 split->orig_start = split->start;
561 ret = add_extent_mapping(em_tree, split);
562 BUG_ON(ret); /* Logic error */
563 list_move(&split->list, &em_tree->modified_extents);
564 free_extent_map(split);
568 write_unlock(&em_tree->lock);
572 /* once for the tree*/
576 free_extent_map(split);
578 free_extent_map(split2);
582 * this is very complex, but the basic idea is to drop all extents
583 * in the range start - end. hint_block is filled in with a block number
584 * that would be a good hint to the block allocator for this file.
586 * If an extent intersects the range but is not entirely inside the range
587 * it is either truncated or split. Anything entirely inside the range
588 * is deleted from the tree.
590 int __btrfs_drop_extents(struct btrfs_trans_handle *trans,
591 struct btrfs_root *root, struct inode *inode,
592 struct btrfs_path *path, u64 start, u64 end,
593 u64 *drop_end, int drop_cache)
595 struct extent_buffer *leaf;
596 struct btrfs_file_extent_item *fi;
597 struct btrfs_key key;
598 struct btrfs_key new_key;
599 u64 ino = btrfs_ino(inode);
600 u64 search_start = start;
603 u64 extent_offset = 0;
610 int modify_tree = -1;
611 int update_refs = (root->ref_cows || root == root->fs_info->tree_root);
614 btrfs_drop_extent_cache(inode, start, end - 1, 0);
616 if (start >= BTRFS_I(inode)->disk_i_size)
621 ret = btrfs_lookup_file_extent(trans, root, path, ino,
622 search_start, modify_tree);
625 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
626 leaf = path->nodes[0];
627 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
628 if (key.objectid == ino &&
629 key.type == BTRFS_EXTENT_DATA_KEY)
634 leaf = path->nodes[0];
635 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
637 ret = btrfs_next_leaf(root, path);
644 leaf = path->nodes[0];
648 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
649 if (key.objectid > ino ||
650 key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
653 fi = btrfs_item_ptr(leaf, path->slots[0],
654 struct btrfs_file_extent_item);
655 extent_type = btrfs_file_extent_type(leaf, fi);
657 if (extent_type == BTRFS_FILE_EXTENT_REG ||
658 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
659 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
660 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
661 extent_offset = btrfs_file_extent_offset(leaf, fi);
662 extent_end = key.offset +
663 btrfs_file_extent_num_bytes(leaf, fi);
664 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
665 extent_end = key.offset +
666 btrfs_file_extent_inline_len(leaf, fi);
669 extent_end = search_start;
672 if (extent_end <= search_start) {
677 search_start = max(key.offset, start);
678 if (recow || !modify_tree) {
680 btrfs_release_path(path);
685 * | - range to drop - |
686 * | -------- extent -------- |
688 if (start > key.offset && end < extent_end) {
690 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
692 memcpy(&new_key, &key, sizeof(new_key));
693 new_key.offset = start;
694 ret = btrfs_duplicate_item(trans, root, path,
696 if (ret == -EAGAIN) {
697 btrfs_release_path(path);
703 leaf = path->nodes[0];
704 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
705 struct btrfs_file_extent_item);
706 btrfs_set_file_extent_num_bytes(leaf, fi,
709 fi = btrfs_item_ptr(leaf, path->slots[0],
710 struct btrfs_file_extent_item);
712 extent_offset += start - key.offset;
713 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
714 btrfs_set_file_extent_num_bytes(leaf, fi,
716 btrfs_mark_buffer_dirty(leaf);
718 if (update_refs && disk_bytenr > 0) {
719 ret = btrfs_inc_extent_ref(trans, root,
720 disk_bytenr, num_bytes, 0,
721 root->root_key.objectid,
723 start - extent_offset, 0);
724 BUG_ON(ret); /* -ENOMEM */
729 * | ---- range to drop ----- |
730 * | -------- extent -------- |
732 if (start <= key.offset && end < extent_end) {
733 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
735 memcpy(&new_key, &key, sizeof(new_key));
736 new_key.offset = end;
737 btrfs_set_item_key_safe(trans, root, path, &new_key);
739 extent_offset += end - key.offset;
740 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
741 btrfs_set_file_extent_num_bytes(leaf, fi,
743 btrfs_mark_buffer_dirty(leaf);
744 if (update_refs && disk_bytenr > 0)
745 inode_sub_bytes(inode, end - key.offset);
749 search_start = extent_end;
751 * | ---- range to drop ----- |
752 * | -------- extent -------- |
754 if (start > key.offset && end >= extent_end) {
756 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
758 btrfs_set_file_extent_num_bytes(leaf, fi,
760 btrfs_mark_buffer_dirty(leaf);
761 if (update_refs && disk_bytenr > 0)
762 inode_sub_bytes(inode, extent_end - start);
763 if (end == extent_end)
771 * | ---- range to drop ----- |
772 * | ------ extent ------ |
774 if (start <= key.offset && end >= extent_end) {
776 del_slot = path->slots[0];
779 BUG_ON(del_slot + del_nr != path->slots[0]);
784 extent_type == BTRFS_FILE_EXTENT_INLINE) {
785 inode_sub_bytes(inode,
786 extent_end - key.offset);
787 extent_end = ALIGN(extent_end,
789 } else if (update_refs && disk_bytenr > 0) {
790 ret = btrfs_free_extent(trans, root,
791 disk_bytenr, num_bytes, 0,
792 root->root_key.objectid,
793 key.objectid, key.offset -
795 BUG_ON(ret); /* -ENOMEM */
796 inode_sub_bytes(inode,
797 extent_end - key.offset);
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);
832 *drop_end = min(end, extent_end);
833 btrfs_release_path(path);
837 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
838 struct btrfs_root *root, struct inode *inode, u64 start,
839 u64 end, int drop_cache)
841 struct btrfs_path *path;
844 path = btrfs_alloc_path();
847 ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL,
849 btrfs_free_path(path);
853 static int extent_mergeable(struct extent_buffer *leaf, int slot,
854 u64 objectid, u64 bytenr, u64 orig_offset,
855 u64 *start, u64 *end)
857 struct btrfs_file_extent_item *fi;
858 struct btrfs_key key;
861 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
864 btrfs_item_key_to_cpu(leaf, &key, slot);
865 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
868 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
869 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
870 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
871 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
872 btrfs_file_extent_compression(leaf, fi) ||
873 btrfs_file_extent_encryption(leaf, fi) ||
874 btrfs_file_extent_other_encoding(leaf, fi))
877 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
878 if ((*start && *start != key.offset) || (*end && *end != extent_end))
887 * Mark extent in the range start - end as written.
889 * This changes extent type from 'pre-allocated' to 'regular'. If only
890 * part of extent is marked as written, the extent will be split into
893 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
894 struct inode *inode, u64 start, u64 end)
896 struct btrfs_root *root = BTRFS_I(inode)->root;
897 struct extent_buffer *leaf;
898 struct btrfs_path *path;
899 struct btrfs_file_extent_item *fi;
900 struct btrfs_key key;
901 struct btrfs_key new_key;
913 u64 ino = btrfs_ino(inode);
915 path = btrfs_alloc_path();
922 key.type = BTRFS_EXTENT_DATA_KEY;
925 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
928 if (ret > 0 && path->slots[0] > 0)
931 leaf = path->nodes[0];
932 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
933 BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY);
934 fi = btrfs_item_ptr(leaf, path->slots[0],
935 struct btrfs_file_extent_item);
936 BUG_ON(btrfs_file_extent_type(leaf, fi) !=
937 BTRFS_FILE_EXTENT_PREALLOC);
938 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
939 BUG_ON(key.offset > start || extent_end < end);
941 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
942 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
943 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
944 memcpy(&new_key, &key, sizeof(new_key));
946 if (start == key.offset && end < extent_end) {
949 if (extent_mergeable(leaf, path->slots[0] - 1,
950 ino, bytenr, orig_offset,
951 &other_start, &other_end)) {
952 new_key.offset = end;
953 btrfs_set_item_key_safe(trans, root, path, &new_key);
954 fi = btrfs_item_ptr(leaf, path->slots[0],
955 struct btrfs_file_extent_item);
956 btrfs_set_file_extent_generation(leaf, fi,
958 btrfs_set_file_extent_num_bytes(leaf, fi,
960 btrfs_set_file_extent_offset(leaf, fi,
962 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
963 struct btrfs_file_extent_item);
964 btrfs_set_file_extent_generation(leaf, fi,
966 btrfs_set_file_extent_num_bytes(leaf, fi,
968 btrfs_mark_buffer_dirty(leaf);
973 if (start > key.offset && end == extent_end) {
976 if (extent_mergeable(leaf, path->slots[0] + 1,
977 ino, bytenr, orig_offset,
978 &other_start, &other_end)) {
979 fi = btrfs_item_ptr(leaf, path->slots[0],
980 struct btrfs_file_extent_item);
981 btrfs_set_file_extent_num_bytes(leaf, fi,
983 btrfs_set_file_extent_generation(leaf, fi,
986 new_key.offset = start;
987 btrfs_set_item_key_safe(trans, root, path, &new_key);
989 fi = btrfs_item_ptr(leaf, path->slots[0],
990 struct btrfs_file_extent_item);
991 btrfs_set_file_extent_generation(leaf, fi,
993 btrfs_set_file_extent_num_bytes(leaf, fi,
995 btrfs_set_file_extent_offset(leaf, fi,
996 start - orig_offset);
997 btrfs_mark_buffer_dirty(leaf);
1002 while (start > key.offset || end < extent_end) {
1003 if (key.offset == start)
1006 new_key.offset = split;
1007 ret = btrfs_duplicate_item(trans, root, path, &new_key);
1008 if (ret == -EAGAIN) {
1009 btrfs_release_path(path);
1013 btrfs_abort_transaction(trans, root, ret);
1017 leaf = path->nodes[0];
1018 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1019 struct btrfs_file_extent_item);
1020 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1021 btrfs_set_file_extent_num_bytes(leaf, fi,
1022 split - key.offset);
1024 fi = btrfs_item_ptr(leaf, path->slots[0],
1025 struct btrfs_file_extent_item);
1027 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1028 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
1029 btrfs_set_file_extent_num_bytes(leaf, fi,
1030 extent_end - split);
1031 btrfs_mark_buffer_dirty(leaf);
1033 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
1034 root->root_key.objectid,
1035 ino, orig_offset, 0);
1036 BUG_ON(ret); /* -ENOMEM */
1038 if (split == start) {
1041 BUG_ON(start != key.offset);
1050 if (extent_mergeable(leaf, path->slots[0] + 1,
1051 ino, bytenr, orig_offset,
1052 &other_start, &other_end)) {
1054 btrfs_release_path(path);
1057 extent_end = other_end;
1058 del_slot = path->slots[0] + 1;
1060 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1061 0, root->root_key.objectid,
1062 ino, orig_offset, 0);
1063 BUG_ON(ret); /* -ENOMEM */
1067 if (extent_mergeable(leaf, path->slots[0] - 1,
1068 ino, bytenr, orig_offset,
1069 &other_start, &other_end)) {
1071 btrfs_release_path(path);
1074 key.offset = other_start;
1075 del_slot = path->slots[0];
1077 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1078 0, root->root_key.objectid,
1079 ino, orig_offset, 0);
1080 BUG_ON(ret); /* -ENOMEM */
1083 fi = btrfs_item_ptr(leaf, path->slots[0],
1084 struct btrfs_file_extent_item);
1085 btrfs_set_file_extent_type(leaf, fi,
1086 BTRFS_FILE_EXTENT_REG);
1087 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1088 btrfs_mark_buffer_dirty(leaf);
1090 fi = btrfs_item_ptr(leaf, del_slot - 1,
1091 struct btrfs_file_extent_item);
1092 btrfs_set_file_extent_type(leaf, fi,
1093 BTRFS_FILE_EXTENT_REG);
1094 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1095 btrfs_set_file_extent_num_bytes(leaf, fi,
1096 extent_end - key.offset);
1097 btrfs_mark_buffer_dirty(leaf);
1099 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1101 btrfs_abort_transaction(trans, root, ret);
1106 btrfs_free_path(path);
1111 * on error we return an unlocked page and the error value
1112 * on success we return a locked page and 0
1114 static int prepare_uptodate_page(struct page *page, u64 pos,
1115 bool force_uptodate)
1119 if (((pos & (PAGE_CACHE_SIZE - 1)) || force_uptodate) &&
1120 !PageUptodate(page)) {
1121 ret = btrfs_readpage(NULL, page);
1125 if (!PageUptodate(page)) {
1134 * this gets pages into the page cache and locks them down, it also properly
1135 * waits for data=ordered extents to finish before allowing the pages to be
1138 static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
1139 struct page **pages, size_t num_pages,
1140 loff_t pos, unsigned long first_index,
1141 size_t write_bytes, bool force_uptodate)
1143 struct extent_state *cached_state = NULL;
1145 unsigned long index = pos >> PAGE_CACHE_SHIFT;
1146 struct inode *inode = fdentry(file)->d_inode;
1147 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1153 start_pos = pos & ~((u64)root->sectorsize - 1);
1154 last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
1157 for (i = 0; i < num_pages; i++) {
1158 pages[i] = find_or_create_page(inode->i_mapping, index + i,
1159 mask | __GFP_WRITE);
1167 err = prepare_uptodate_page(pages[i], pos,
1169 if (i == num_pages - 1)
1170 err = prepare_uptodate_page(pages[i],
1171 pos + write_bytes, false);
1173 page_cache_release(pages[i]);
1177 wait_on_page_writeback(pages[i]);
1180 if (start_pos < inode->i_size) {
1181 struct btrfs_ordered_extent *ordered;
1182 lock_extent_bits(&BTRFS_I(inode)->io_tree,
1183 start_pos, last_pos - 1, 0, &cached_state);
1184 ordered = btrfs_lookup_first_ordered_extent(inode,
1187 ordered->file_offset + ordered->len > start_pos &&
1188 ordered->file_offset < last_pos) {
1189 btrfs_put_ordered_extent(ordered);
1190 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1191 start_pos, last_pos - 1,
1192 &cached_state, GFP_NOFS);
1193 for (i = 0; i < num_pages; i++) {
1194 unlock_page(pages[i]);
1195 page_cache_release(pages[i]);
1197 btrfs_wait_ordered_range(inode, start_pos,
1198 last_pos - start_pos);
1202 btrfs_put_ordered_extent(ordered);
1204 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
1205 last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1206 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
1207 0, 0, &cached_state, GFP_NOFS);
1208 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1209 start_pos, last_pos - 1, &cached_state,
1212 for (i = 0; i < num_pages; i++) {
1213 if (clear_page_dirty_for_io(pages[i]))
1214 account_page_redirty(pages[i]);
1215 set_page_extent_mapped(pages[i]);
1216 WARN_ON(!PageLocked(pages[i]));
1220 while (faili >= 0) {
1221 unlock_page(pages[faili]);
1222 page_cache_release(pages[faili]);
1229 static noinline ssize_t __btrfs_buffered_write(struct file *file,
1233 struct inode *inode = fdentry(file)->d_inode;
1234 struct btrfs_root *root = BTRFS_I(inode)->root;
1235 struct page **pages = NULL;
1236 unsigned long first_index;
1237 size_t num_written = 0;
1240 bool force_page_uptodate = false;
1242 nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) /
1243 PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
1244 (sizeof(struct page *)));
1245 nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1246 nrptrs = max(nrptrs, 8);
1247 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
1251 first_index = pos >> PAGE_CACHE_SHIFT;
1253 while (iov_iter_count(i) > 0) {
1254 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1255 size_t write_bytes = min(iov_iter_count(i),
1256 nrptrs * (size_t)PAGE_CACHE_SIZE -
1258 size_t num_pages = (write_bytes + offset +
1259 PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1263 WARN_ON(num_pages > nrptrs);
1266 * Fault pages before locking them in prepare_pages
1267 * to avoid recursive lock
1269 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1274 ret = btrfs_delalloc_reserve_space(inode,
1275 num_pages << PAGE_CACHE_SHIFT);
1280 * This is going to setup the pages array with the number of
1281 * pages we want, so we don't really need to worry about the
1282 * contents of pages from loop to loop
1284 ret = prepare_pages(root, file, pages, num_pages,
1285 pos, first_index, write_bytes,
1286 force_page_uptodate);
1288 btrfs_delalloc_release_space(inode,
1289 num_pages << PAGE_CACHE_SHIFT);
1293 copied = btrfs_copy_from_user(pos, num_pages,
1294 write_bytes, pages, i);
1297 * if we have trouble faulting in the pages, fall
1298 * back to one page at a time
1300 if (copied < write_bytes)
1304 force_page_uptodate = true;
1307 force_page_uptodate = false;
1308 dirty_pages = (copied + offset +
1309 PAGE_CACHE_SIZE - 1) >>
1314 * If we had a short copy we need to release the excess delaloc
1315 * bytes we reserved. We need to increment outstanding_extents
1316 * because btrfs_delalloc_release_space will decrement it, but
1317 * we still have an outstanding extent for the chunk we actually
1320 if (num_pages > dirty_pages) {
1322 spin_lock(&BTRFS_I(inode)->lock);
1323 BTRFS_I(inode)->outstanding_extents++;
1324 spin_unlock(&BTRFS_I(inode)->lock);
1326 btrfs_delalloc_release_space(inode,
1327 (num_pages - dirty_pages) <<
1332 ret = btrfs_dirty_pages(root, inode, pages,
1333 dirty_pages, pos, copied,
1336 btrfs_delalloc_release_space(inode,
1337 dirty_pages << PAGE_CACHE_SHIFT);
1338 btrfs_drop_pages(pages, num_pages);
1343 btrfs_drop_pages(pages, num_pages);
1347 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
1349 if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1350 btrfs_btree_balance_dirty(root, 1);
1353 num_written += copied;
1358 return num_written ? num_written : ret;
1361 static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1362 const struct iovec *iov,
1363 unsigned long nr_segs, loff_t pos,
1364 loff_t *ppos, size_t count, size_t ocount)
1366 struct file *file = iocb->ki_filp;
1369 ssize_t written_buffered;
1373 written = generic_file_direct_write(iocb, iov, &nr_segs, pos, ppos,
1376 if (written < 0 || written == count)
1381 iov_iter_init(&i, iov, nr_segs, count, written);
1382 written_buffered = __btrfs_buffered_write(file, &i, pos);
1383 if (written_buffered < 0) {
1384 err = written_buffered;
1387 endbyte = pos + written_buffered - 1;
1388 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
1391 written += written_buffered;
1392 *ppos = pos + written_buffered;
1393 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
1394 endbyte >> PAGE_CACHE_SHIFT);
1396 return written ? written : err;
1399 static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
1400 const struct iovec *iov,
1401 unsigned long nr_segs, loff_t pos)
1403 struct file *file = iocb->ki_filp;
1404 struct inode *inode = fdentry(file)->d_inode;
1405 struct btrfs_root *root = BTRFS_I(inode)->root;
1406 loff_t *ppos = &iocb->ki_pos;
1408 ssize_t num_written = 0;
1410 size_t count, ocount;
1412 sb_start_write(inode->i_sb);
1414 mutex_lock(&inode->i_mutex);
1416 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
1418 mutex_unlock(&inode->i_mutex);
1423 current->backing_dev_info = inode->i_mapping->backing_dev_info;
1424 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1426 mutex_unlock(&inode->i_mutex);
1431 mutex_unlock(&inode->i_mutex);
1435 err = file_remove_suid(file);
1437 mutex_unlock(&inode->i_mutex);
1442 * If BTRFS flips readonly due to some impossible error
1443 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1444 * although we have opened a file as writable, we have
1445 * to stop this write operation to ensure FS consistency.
1447 if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
1448 mutex_unlock(&inode->i_mutex);
1453 err = file_update_time(file);
1455 mutex_unlock(&inode->i_mutex);
1459 start_pos = round_down(pos, root->sectorsize);
1460 if (start_pos > i_size_read(inode)) {
1461 err = btrfs_cont_expand(inode, i_size_read(inode), start_pos);
1463 mutex_unlock(&inode->i_mutex);
1468 if (unlikely(file->f_flags & O_DIRECT)) {
1469 num_written = __btrfs_direct_write(iocb, iov, nr_segs,
1470 pos, ppos, count, ocount);
1474 iov_iter_init(&i, iov, nr_segs, count, num_written);
1476 num_written = __btrfs_buffered_write(file, &i, pos);
1477 if (num_written > 0)
1478 *ppos = pos + num_written;
1481 mutex_unlock(&inode->i_mutex);
1484 * we want to make sure fsync finds this change
1485 * but we haven't joined a transaction running right now.
1487 * Later on, someone is sure to update the inode and get the
1488 * real transid recorded.
1490 * We set last_trans now to the fs_info generation + 1,
1491 * this will either be one more than the running transaction
1492 * or the generation used for the next transaction if there isn't
1493 * one running right now.
1495 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1496 if (num_written > 0 || num_written == -EIOCBQUEUED) {
1497 err = generic_write_sync(file, pos, num_written);
1498 if (err < 0 && num_written > 0)
1502 sb_end_write(inode->i_sb);
1503 current->backing_dev_info = NULL;
1504 return num_written ? num_written : err;
1507 int btrfs_release_file(struct inode *inode, struct file *filp)
1510 * ordered_data_close is set by settattr when we are about to truncate
1511 * a file from a non-zero size to a zero size. This tries to
1512 * flush down new bytes that may have been written if the
1513 * application were using truncate to replace a file in place.
1515 if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
1516 &BTRFS_I(inode)->runtime_flags)) {
1517 btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode);
1518 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1519 filemap_flush(inode->i_mapping);
1521 if (filp->private_data)
1522 btrfs_ioctl_trans_end(filp);
1527 * fsync call for both files and directories. This logs the inode into
1528 * the tree log instead of forcing full commits whenever possible.
1530 * It needs to call filemap_fdatawait so that all ordered extent updates are
1531 * in the metadata btree are up to date for copying to the log.
1533 * It drops the inode mutex before doing the tree log commit. This is an
1534 * important optimization for directories because holding the mutex prevents
1535 * new operations on the dir while we write to disk.
1537 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1539 struct dentry *dentry = file->f_path.dentry;
1540 struct inode *inode = dentry->d_inode;
1541 struct btrfs_root *root = BTRFS_I(inode)->root;
1543 struct btrfs_trans_handle *trans;
1545 trace_btrfs_sync_file(file, datasync);
1548 * We write the dirty pages in the range and wait until they complete
1549 * out of the ->i_mutex. If so, we can flush the dirty pages by
1550 * multi-task, and make the performance up.
1552 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
1556 mutex_lock(&inode->i_mutex);
1559 * We flush the dirty pages again to avoid some dirty pages in the
1562 atomic_inc(&root->log_batch);
1563 btrfs_wait_ordered_range(inode, start, end);
1564 atomic_inc(&root->log_batch);
1567 * check the transaction that last modified this inode
1568 * and see if its already been committed
1570 if (!BTRFS_I(inode)->last_trans) {
1571 mutex_unlock(&inode->i_mutex);
1576 * if the last transaction that changed this file was before
1577 * the current transaction, we can bail out now without any
1581 if (btrfs_inode_in_log(inode, root->fs_info->generation) ||
1582 BTRFS_I(inode)->last_trans <=
1583 root->fs_info->last_trans_committed) {
1584 BTRFS_I(inode)->last_trans = 0;
1587 * We'v had everything committed since the last time we were
1588 * modified so clear this flag in case it was set for whatever
1589 * reason, it's no longer relevant.
1591 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1592 &BTRFS_I(inode)->runtime_flags);
1593 mutex_unlock(&inode->i_mutex);
1598 * ok we haven't committed the transaction yet, lets do a commit
1600 if (file->private_data)
1601 btrfs_ioctl_trans_end(file);
1603 trans = btrfs_start_transaction(root, 0);
1604 if (IS_ERR(trans)) {
1605 ret = PTR_ERR(trans);
1606 mutex_unlock(&inode->i_mutex);
1610 ret = btrfs_log_dentry_safe(trans, root, dentry);
1612 mutex_unlock(&inode->i_mutex);
1616 /* we've logged all the items and now have a consistent
1617 * version of the file in the log. It is possible that
1618 * someone will come in and modify the file, but that's
1619 * fine because the log is consistent on disk, and we
1620 * have references to all of the file's extents
1622 * It is possible that someone will come in and log the
1623 * file again, but that will end up using the synchronization
1624 * inside btrfs_sync_log to keep things safe.
1626 mutex_unlock(&inode->i_mutex);
1628 if (ret != BTRFS_NO_LOG_SYNC) {
1630 ret = btrfs_commit_transaction(trans, root);
1632 ret = btrfs_sync_log(trans, root);
1634 ret = btrfs_end_transaction(trans, root);
1636 ret = btrfs_commit_transaction(trans, root);
1639 ret = btrfs_end_transaction(trans, root);
1642 return ret > 0 ? -EIO : ret;
1645 static const struct vm_operations_struct btrfs_file_vm_ops = {
1646 .fault = filemap_fault,
1647 .page_mkwrite = btrfs_page_mkwrite,
1650 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
1652 struct address_space *mapping = filp->f_mapping;
1654 if (!mapping->a_ops->readpage)
1657 file_accessed(filp);
1658 vma->vm_ops = &btrfs_file_vm_ops;
1659 vma->vm_flags |= VM_CAN_NONLINEAR;
1664 static int hole_mergeable(struct inode *inode, struct extent_buffer *leaf,
1665 int slot, u64 start, u64 end)
1667 struct btrfs_file_extent_item *fi;
1668 struct btrfs_key key;
1670 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
1673 btrfs_item_key_to_cpu(leaf, &key, slot);
1674 if (key.objectid != btrfs_ino(inode) ||
1675 key.type != BTRFS_EXTENT_DATA_KEY)
1678 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1680 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
1683 if (btrfs_file_extent_disk_bytenr(leaf, fi))
1686 if (key.offset == end)
1688 if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
1693 static int fill_holes(struct btrfs_trans_handle *trans, struct inode *inode,
1694 struct btrfs_path *path, u64 offset, u64 end)
1696 struct btrfs_root *root = BTRFS_I(inode)->root;
1697 struct extent_buffer *leaf;
1698 struct btrfs_file_extent_item *fi;
1699 struct extent_map *hole_em;
1700 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1701 struct btrfs_key key;
1704 key.objectid = btrfs_ino(inode);
1705 key.type = BTRFS_EXTENT_DATA_KEY;
1706 key.offset = offset;
1709 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1714 leaf = path->nodes[0];
1715 if (hole_mergeable(inode, leaf, path->slots[0]-1, offset, end)) {
1719 fi = btrfs_item_ptr(leaf, path->slots[0],
1720 struct btrfs_file_extent_item);
1721 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
1723 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1724 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
1725 btrfs_set_file_extent_offset(leaf, fi, 0);
1726 btrfs_mark_buffer_dirty(leaf);
1730 if (hole_mergeable(inode, leaf, path->slots[0]+1, offset, end)) {
1734 key.offset = offset;
1735 btrfs_set_item_key_safe(trans, root, path, &key);
1736 fi = btrfs_item_ptr(leaf, path->slots[0],
1737 struct btrfs_file_extent_item);
1738 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
1740 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1741 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
1742 btrfs_set_file_extent_offset(leaf, fi, 0);
1743 btrfs_mark_buffer_dirty(leaf);
1746 btrfs_release_path(path);
1748 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
1749 0, 0, end - offset, 0, end - offset,
1755 btrfs_release_path(path);
1757 hole_em = alloc_extent_map();
1759 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
1760 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1761 &BTRFS_I(inode)->runtime_flags);
1763 hole_em->start = offset;
1764 hole_em->len = end - offset;
1765 hole_em->orig_start = offset;
1767 hole_em->block_start = EXTENT_MAP_HOLE;
1768 hole_em->block_len = 0;
1769 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
1770 hole_em->compress_type = BTRFS_COMPRESS_NONE;
1771 hole_em->generation = trans->transid;
1774 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
1775 write_lock(&em_tree->lock);
1776 ret = add_extent_mapping(em_tree, hole_em);
1778 list_move(&hole_em->list,
1779 &em_tree->modified_extents);
1780 write_unlock(&em_tree->lock);
1781 } while (ret == -EEXIST);
1782 free_extent_map(hole_em);
1784 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1785 &BTRFS_I(inode)->runtime_flags);
1791 static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
1793 struct btrfs_root *root = BTRFS_I(inode)->root;
1794 struct extent_state *cached_state = NULL;
1795 struct btrfs_path *path;
1796 struct btrfs_block_rsv *rsv;
1797 struct btrfs_trans_handle *trans;
1798 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
1799 u64 lockstart = (offset + mask) & ~mask;
1800 u64 lockend = ((offset + len) & ~mask) - 1;
1801 u64 cur_offset = lockstart;
1802 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
1807 bool same_page = (offset >> PAGE_CACHE_SHIFT) ==
1808 ((offset + len) >> PAGE_CACHE_SHIFT);
1810 btrfs_wait_ordered_range(inode, offset, len);
1812 mutex_lock(&inode->i_mutex);
1813 if (offset >= inode->i_size) {
1814 mutex_unlock(&inode->i_mutex);
1819 * Only do this if we are in the same page and we aren't doing the
1822 if (same_page && len < PAGE_CACHE_SIZE) {
1823 ret = btrfs_truncate_page(inode, offset, len, 0);
1824 mutex_unlock(&inode->i_mutex);
1828 /* zero back part of the first page */
1829 ret = btrfs_truncate_page(inode, offset, 0, 0);
1831 mutex_unlock(&inode->i_mutex);
1835 /* zero the front end of the last page */
1836 ret = btrfs_truncate_page(inode, offset + len, 0, 1);
1838 mutex_unlock(&inode->i_mutex);
1842 if (lockend < lockstart) {
1843 mutex_unlock(&inode->i_mutex);
1848 struct btrfs_ordered_extent *ordered;
1850 truncate_pagecache_range(inode, lockstart, lockend);
1852 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
1854 ordered = btrfs_lookup_first_ordered_extent(inode, lockend);
1857 * We need to make sure we have no ordered extents in this range
1858 * and nobody raced in and read a page in this range, if we did
1859 * we need to try again.
1862 (ordered->file_offset + ordered->len < lockstart ||
1863 ordered->file_offset > lockend)) &&
1864 !test_range_bit(&BTRFS_I(inode)->io_tree, lockstart,
1865 lockend, EXTENT_UPTODATE, 0,
1868 btrfs_put_ordered_extent(ordered);
1872 btrfs_put_ordered_extent(ordered);
1873 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
1874 lockend, &cached_state, GFP_NOFS);
1875 btrfs_wait_ordered_range(inode, lockstart,
1876 lockend - lockstart + 1);
1879 path = btrfs_alloc_path();
1885 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
1890 rsv->size = btrfs_calc_trunc_metadata_size(root, 1);
1894 * 1 - update the inode
1895 * 1 - removing the extents in the range
1896 * 1 - adding the hole extent
1898 trans = btrfs_start_transaction(root, 3);
1899 if (IS_ERR(trans)) {
1900 err = PTR_ERR(trans);
1904 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
1907 trans->block_rsv = rsv;
1909 while (cur_offset < lockend) {
1910 ret = __btrfs_drop_extents(trans, root, inode, path,
1911 cur_offset, lockend + 1,
1916 trans->block_rsv = &root->fs_info->trans_block_rsv;
1918 ret = fill_holes(trans, inode, path, cur_offset, drop_end);
1924 cur_offset = drop_end;
1926 ret = btrfs_update_inode(trans, root, inode);
1932 nr = trans->blocks_used;
1933 btrfs_end_transaction(trans, root);
1934 btrfs_btree_balance_dirty(root, nr);
1936 trans = btrfs_start_transaction(root, 3);
1937 if (IS_ERR(trans)) {
1938 ret = PTR_ERR(trans);
1943 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
1945 BUG_ON(ret); /* shouldn't happen */
1946 trans->block_rsv = rsv;
1954 trans->block_rsv = &root->fs_info->trans_block_rsv;
1955 ret = fill_holes(trans, inode, path, cur_offset, drop_end);
1965 trans->block_rsv = &root->fs_info->trans_block_rsv;
1966 ret = btrfs_update_inode(trans, root, inode);
1967 nr = trans->blocks_used;
1968 btrfs_end_transaction(trans, root);
1969 btrfs_btree_balance_dirty(root, nr);
1971 btrfs_free_path(path);
1972 btrfs_free_block_rsv(root, rsv);
1974 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
1975 &cached_state, GFP_NOFS);
1976 mutex_unlock(&inode->i_mutex);
1982 static long btrfs_fallocate(struct file *file, int mode,
1983 loff_t offset, loff_t len)
1985 struct inode *inode = file->f_path.dentry->d_inode;
1986 struct extent_state *cached_state = NULL;
1993 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
1994 struct extent_map *em;
1997 alloc_start = offset & ~mask;
1998 alloc_end = (offset + len + mask) & ~mask;
2000 /* Make sure we aren't being give some crap mode */
2001 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2004 if (mode & FALLOC_FL_PUNCH_HOLE)
2005 return btrfs_punch_hole(inode, offset, len);
2008 * Make sure we have enough space before we do the
2011 ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start + 1);
2016 * wait for ordered IO before we have any locks. We'll loop again
2017 * below with the locks held.
2019 btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
2021 mutex_lock(&inode->i_mutex);
2022 ret = inode_newsize_ok(inode, alloc_end);
2026 if (alloc_start > inode->i_size) {
2027 ret = btrfs_cont_expand(inode, i_size_read(inode),
2033 locked_end = alloc_end - 1;
2035 struct btrfs_ordered_extent *ordered;
2037 /* the extent lock is ordered inside the running
2040 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
2041 locked_end, 0, &cached_state);
2042 ordered = btrfs_lookup_first_ordered_extent(inode,
2045 ordered->file_offset + ordered->len > alloc_start &&
2046 ordered->file_offset < alloc_end) {
2047 btrfs_put_ordered_extent(ordered);
2048 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
2049 alloc_start, locked_end,
2050 &cached_state, GFP_NOFS);
2052 * we can't wait on the range with the transaction
2053 * running or with the extent lock held
2055 btrfs_wait_ordered_range(inode, alloc_start,
2056 alloc_end - alloc_start);
2059 btrfs_put_ordered_extent(ordered);
2064 cur_offset = alloc_start;
2068 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2069 alloc_end - cur_offset, 0);
2070 if (IS_ERR_OR_NULL(em)) {
2077 last_byte = min(extent_map_end(em), alloc_end);
2078 actual_end = min_t(u64, extent_map_end(em), offset + len);
2079 last_byte = (last_byte + mask) & ~mask;
2081 if (em->block_start == EXTENT_MAP_HOLE ||
2082 (cur_offset >= inode->i_size &&
2083 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
2084 ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
2085 last_byte - cur_offset,
2086 1 << inode->i_blkbits,
2091 free_extent_map(em);
2094 } else if (actual_end > inode->i_size &&
2095 !(mode & FALLOC_FL_KEEP_SIZE)) {
2097 * We didn't need to allocate any more space, but we
2098 * still extended the size of the file so we need to
2101 inode->i_ctime = CURRENT_TIME;
2102 i_size_write(inode, actual_end);
2103 btrfs_ordered_update_i_size(inode, actual_end, NULL);
2105 free_extent_map(em);
2107 cur_offset = last_byte;
2108 if (cur_offset >= alloc_end) {
2113 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
2114 &cached_state, GFP_NOFS);
2116 mutex_unlock(&inode->i_mutex);
2117 /* Let go of our reservation. */
2118 btrfs_free_reserved_data_space(inode, alloc_end - alloc_start + 1);
2122 static int find_desired_extent(struct inode *inode, loff_t *offset, int origin)
2124 struct btrfs_root *root = BTRFS_I(inode)->root;
2125 struct extent_map *em;
2126 struct extent_state *cached_state = NULL;
2127 u64 lockstart = *offset;
2128 u64 lockend = i_size_read(inode);
2129 u64 start = *offset;
2130 u64 orig_start = *offset;
2131 u64 len = i_size_read(inode);
2135 lockend = max_t(u64, root->sectorsize, lockend);
2136 if (lockend <= lockstart)
2137 lockend = lockstart + root->sectorsize;
2139 len = lockend - lockstart + 1;
2141 len = max_t(u64, len, root->sectorsize);
2142 if (inode->i_size == 0)
2145 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, 0,
2149 * Delalloc is such a pain. If we have a hole and we have pending
2150 * delalloc for a portion of the hole we will get back a hole that
2151 * exists for the entire range since it hasn't been actually written
2152 * yet. So to take care of this case we need to look for an extent just
2153 * before the position we want in case there is outstanding delalloc
2156 if (origin == SEEK_HOLE && start != 0) {
2157 if (start <= root->sectorsize)
2158 em = btrfs_get_extent_fiemap(inode, NULL, 0, 0,
2159 root->sectorsize, 0);
2161 em = btrfs_get_extent_fiemap(inode, NULL, 0,
2162 start - root->sectorsize,
2163 root->sectorsize, 0);
2168 last_end = em->start + em->len;
2169 if (em->block_start == EXTENT_MAP_DELALLOC)
2170 last_end = min_t(u64, last_end, inode->i_size);
2171 free_extent_map(em);
2175 em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0);
2181 if (em->block_start == EXTENT_MAP_HOLE) {
2182 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2183 if (last_end <= orig_start) {
2184 free_extent_map(em);
2190 if (origin == SEEK_HOLE) {
2192 free_extent_map(em);
2196 if (origin == SEEK_DATA) {
2197 if (em->block_start == EXTENT_MAP_DELALLOC) {
2198 if (start >= inode->i_size) {
2199 free_extent_map(em);
2206 free_extent_map(em);
2211 start = em->start + em->len;
2212 last_end = em->start + em->len;
2214 if (em->block_start == EXTENT_MAP_DELALLOC)
2215 last_end = min_t(u64, last_end, inode->i_size);
2217 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2218 free_extent_map(em);
2222 free_extent_map(em);
2226 *offset = min(*offset, inode->i_size);
2228 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2229 &cached_state, GFP_NOFS);
2233 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int origin)
2235 struct inode *inode = file->f_mapping->host;
2238 mutex_lock(&inode->i_mutex);
2242 offset = generic_file_llseek(file, offset, origin);
2246 if (offset >= i_size_read(inode)) {
2247 mutex_unlock(&inode->i_mutex);
2251 ret = find_desired_extent(inode, &offset, origin);
2253 mutex_unlock(&inode->i_mutex);
2258 if (offset < 0 && !(file->f_mode & FMODE_UNSIGNED_OFFSET)) {
2262 if (offset > inode->i_sb->s_maxbytes) {
2267 /* Special lock needed here? */
2268 if (offset != file->f_pos) {
2269 file->f_pos = offset;
2270 file->f_version = 0;
2273 mutex_unlock(&inode->i_mutex);
2277 const struct file_operations btrfs_file_operations = {
2278 .llseek = btrfs_file_llseek,
2279 .read = do_sync_read,
2280 .write = do_sync_write,
2281 .aio_read = generic_file_aio_read,
2282 .splice_read = generic_file_splice_read,
2283 .aio_write = btrfs_file_aio_write,
2284 .mmap = btrfs_file_mmap,
2285 .open = generic_file_open,
2286 .release = btrfs_release_file,
2287 .fsync = btrfs_sync_file,
2288 .fallocate = btrfs_fallocate,
2289 .unlocked_ioctl = btrfs_ioctl,
2290 #ifdef CONFIG_COMPAT
2291 .compat_ioctl = btrfs_ioctl,