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/smp_lock.h>
26 #include <linux/backing-dev.h>
27 #include <linux/mpage.h>
28 #include <linux/swap.h>
29 #include <linux/writeback.h>
30 #include <linux/statfs.h>
31 #include <linux/compat.h>
32 #include <linux/version.h>
35 #include "transaction.h"
36 #include "btrfs_inode.h"
38 #include "print-tree.h"
44 /* simple helper to fault in pages and copy. This should go away
45 * and be replaced with calls into generic code.
47 static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
49 struct page **prepared_pages,
50 const char __user *buf)
54 int offset = pos & (PAGE_CACHE_SIZE - 1);
56 for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
57 size_t count = min_t(size_t,
58 PAGE_CACHE_SIZE - offset, write_bytes);
59 struct page *page = prepared_pages[i];
60 fault_in_pages_readable(buf, count);
62 /* Copy data from userspace to the current page */
64 page_fault = __copy_from_user(page_address(page) + offset,
66 /* Flush processor's dcache for this page */
67 flush_dcache_page(page);
75 return page_fault ? -EFAULT : 0;
79 * unlocks pages after btrfs_file_write is done with them
81 static noinline void btrfs_drop_pages(struct page **pages, size_t num_pages)
84 for (i = 0; i < num_pages; i++) {
87 /* page checked is some magic around finding pages that
88 * have been modified without going through btrfs_set_page_dirty
91 ClearPageChecked(pages[i]);
92 unlock_page(pages[i]);
93 mark_page_accessed(pages[i]);
94 page_cache_release(pages[i]);
99 * after copy_from_user, pages need to be dirtied and we need to make
100 * sure holes are created between the current EOF and the start of
101 * any next extents (if required).
103 * this also makes the decision about creating an inline extent vs
104 * doing real data extents, marking pages dirty and delalloc as required.
106 static noinline int dirty_and_release_pages(struct btrfs_trans_handle *trans,
107 struct btrfs_root *root,
116 struct inode *inode = fdentry(file)->d_inode;
117 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
121 u64 end_of_last_block;
122 u64 end_pos = pos + write_bytes;
123 loff_t isize = i_size_read(inode);
125 start_pos = pos & ~((u64)root->sectorsize - 1);
126 num_bytes = (write_bytes + pos - start_pos +
127 root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
129 end_of_last_block = start_pos + num_bytes - 1;
131 lock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
132 trans = btrfs_join_transaction(root, 1);
137 btrfs_set_trans_block_group(trans, inode);
140 set_extent_uptodate(io_tree, start_pos, end_of_last_block, GFP_NOFS);
142 /* check for reserved extents on each page, we don't want
143 * to reset the delalloc bit on things that already have
146 btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block);
147 for (i = 0; i < num_pages; i++) {
148 struct page *p = pages[i];
153 if (end_pos > isize) {
154 i_size_write(inode, end_pos);
155 btrfs_update_inode(trans, root, inode);
157 err = btrfs_end_transaction(trans, root);
159 unlock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
164 * this drops all the extents in the cache that intersect the range
165 * [start, end]. Existing extents are split as required.
167 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
170 struct extent_map *em;
171 struct extent_map *split = NULL;
172 struct extent_map *split2 = NULL;
173 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
174 u64 len = end - start + 1;
180 WARN_ON(end < start);
181 if (end == (u64)-1) {
187 split = alloc_extent_map(GFP_NOFS);
189 split2 = alloc_extent_map(GFP_NOFS);
191 spin_lock(&em_tree->lock);
192 em = lookup_extent_mapping(em_tree, start, len);
194 spin_unlock(&em_tree->lock);
198 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
199 spin_unlock(&em_tree->lock);
200 if (em->start <= start &&
201 (!testend || em->start + em->len >= start + len)) {
205 if (start < em->start) {
206 len = em->start - start;
208 len = start + len - (em->start + em->len);
209 start = em->start + em->len;
214 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
215 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
216 remove_extent_mapping(em_tree, em);
218 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
220 split->start = em->start;
221 split->len = start - em->start;
222 split->orig_start = em->orig_start;
223 split->block_start = em->block_start;
226 split->block_len = em->block_len;
228 split->block_len = split->len;
230 split->bdev = em->bdev;
231 split->flags = flags;
232 ret = add_extent_mapping(em_tree, split);
234 free_extent_map(split);
238 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
239 testend && em->start + em->len > start + len) {
240 u64 diff = start + len - em->start;
242 split->start = start + len;
243 split->len = em->start + em->len - (start + len);
244 split->bdev = em->bdev;
245 split->flags = flags;
248 split->block_len = em->block_len;
249 split->block_start = em->block_start;
250 split->orig_start = em->orig_start;
252 split->block_len = split->len;
253 split->block_start = em->block_start + diff;
254 split->orig_start = split->start;
257 ret = add_extent_mapping(em_tree, split);
259 free_extent_map(split);
262 spin_unlock(&em_tree->lock);
266 /* once for the tree*/
270 free_extent_map(split);
272 free_extent_map(split2);
276 int btrfs_check_file(struct btrfs_root *root, struct inode *inode)
280 struct btrfs_path *path;
281 struct btrfs_key found_key;
282 struct extent_buffer *leaf;
283 struct btrfs_file_extent_item *extent;
292 path = btrfs_alloc_path();
293 ret = btrfs_lookup_file_extent(NULL, root, path, inode->i_ino,
296 nritems = btrfs_header_nritems(path->nodes[0]);
297 if (path->slots[0] >= nritems) {
298 ret = btrfs_next_leaf(root, path);
301 nritems = btrfs_header_nritems(path->nodes[0]);
303 slot = path->slots[0];
304 leaf = path->nodes[0];
305 btrfs_item_key_to_cpu(leaf, &found_key, slot);
306 if (found_key.objectid != inode->i_ino)
308 if (found_key.type != BTRFS_EXTENT_DATA_KEY)
311 if (found_key.offset < last_offset) {
313 btrfs_print_leaf(root, leaf);
314 printk(KERN_ERR "inode %lu found offset %llu "
315 "expected %llu\n", inode->i_ino,
316 (unsigned long long)found_key.offset,
317 (unsigned long long)last_offset);
321 extent = btrfs_item_ptr(leaf, slot,
322 struct btrfs_file_extent_item);
323 found_type = btrfs_file_extent_type(leaf, extent);
324 if (found_type == BTRFS_FILE_EXTENT_REG) {
325 extent_end = found_key.offset +
326 btrfs_file_extent_num_bytes(leaf, extent);
327 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
328 struct btrfs_item *item;
329 item = btrfs_item_nr(leaf, slot);
330 extent_end = found_key.offset +
331 btrfs_file_extent_inline_len(leaf, extent);
332 extent_end = (extent_end + root->sectorsize - 1) &
333 ~((u64)root->sectorsize - 1);
335 last_offset = extent_end;
338 if (0 && last_offset < inode->i_size) {
340 btrfs_print_leaf(root, leaf);
341 printk(KERN_ERR "inode %lu found offset %llu size %llu\n",
342 inode->i_ino, (unsigned long long)last_offset,
343 (unsigned long long)inode->i_size);
348 btrfs_free_path(path);
354 * this is very complex, but the basic idea is to drop all extents
355 * in the range start - end. hint_block is filled in with a block number
356 * that would be a good hint to the block allocator for this file.
358 * If an extent intersects the range but is not entirely inside the range
359 * it is either truncated or split. Anything entirely inside the range
360 * is deleted from the tree.
362 * inline_limit is used to tell this code which offsets in the file to keep
363 * if they contain inline extents.
365 noinline int btrfs_drop_extents(struct btrfs_trans_handle *trans,
366 struct btrfs_root *root, struct inode *inode,
367 u64 start, u64 end, u64 inline_limit, u64 *hint_byte)
370 u64 locked_end = end;
371 u64 search_start = start;
378 u16 other_encoding = 0;
381 struct extent_buffer *leaf;
382 struct btrfs_file_extent_item *extent;
383 struct btrfs_path *path;
384 struct btrfs_key key;
385 struct btrfs_file_extent_item old;
396 btrfs_drop_extent_cache(inode, start, end - 1, 0);
398 path = btrfs_alloc_path();
403 btrfs_release_path(root, path);
404 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
409 if (path->slots[0] == 0) {
426 leaf = path->nodes[0];
427 slot = path->slots[0];
429 btrfs_item_key_to_cpu(leaf, &key, slot);
430 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY &&
434 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
435 key.objectid != inode->i_ino) {
439 search_start = max(key.offset, start);
442 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
443 extent = btrfs_item_ptr(leaf, slot,
444 struct btrfs_file_extent_item);
445 found_type = btrfs_file_extent_type(leaf, extent);
446 compression = btrfs_file_extent_compression(leaf,
448 encryption = btrfs_file_extent_encryption(leaf,
450 other_encoding = btrfs_file_extent_other_encoding(leaf,
452 if (found_type == BTRFS_FILE_EXTENT_REG ||
453 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
455 btrfs_file_extent_disk_bytenr(leaf,
458 *hint_byte = extent_end;
460 extent_end = key.offset +
461 btrfs_file_extent_num_bytes(leaf, extent);
462 ram_bytes = btrfs_file_extent_ram_bytes(leaf,
465 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
467 extent_end = key.offset +
468 btrfs_file_extent_inline_len(leaf, extent);
471 extent_end = search_start;
474 /* we found nothing we can drop */
475 if ((!found_extent && !found_inline) ||
476 search_start >= extent_end) {
479 nritems = btrfs_header_nritems(leaf);
480 if (slot >= nritems - 1) {
481 nextret = btrfs_next_leaf(root, path);
491 if (end <= extent_end && start >= key.offset && found_inline)
492 *hint_byte = EXTENT_MAP_INLINE;
495 read_extent_buffer(leaf, &old, (unsigned long)extent,
497 root_gen = btrfs_header_generation(leaf);
498 root_owner = btrfs_header_owner(leaf);
499 leaf_start = leaf->start;
502 if (end < extent_end && end >= key.offset) {
504 if (found_inline && start <= key.offset)
508 if (bookend && found_extent) {
509 if (locked_end < extent_end) {
510 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
511 locked_end, extent_end - 1,
514 btrfs_release_path(root, path);
515 lock_extent(&BTRFS_I(inode)->io_tree,
516 locked_end, extent_end - 1,
518 locked_end = extent_end;
521 locked_end = extent_end;
523 orig_parent = path->nodes[0]->start;
524 disk_bytenr = le64_to_cpu(old.disk_bytenr);
525 if (disk_bytenr != 0) {
526 ret = btrfs_inc_extent_ref(trans, root,
528 le64_to_cpu(old.disk_num_bytes),
529 orig_parent, root->root_key.objectid,
530 trans->transid, inode->i_ino);
536 u64 mask = root->sectorsize - 1;
537 search_start = (extent_end + mask) & ~mask;
539 search_start = extent_end;
541 /* truncate existing extent */
542 if (start > key.offset) {
546 WARN_ON(start & (root->sectorsize - 1));
548 new_num = start - key.offset;
549 old_num = btrfs_file_extent_num_bytes(leaf,
552 btrfs_file_extent_disk_bytenr(leaf,
554 if (btrfs_file_extent_disk_bytenr(leaf,
556 inode_sub_bytes(inode, old_num -
559 if (!compression && !encryption) {
560 btrfs_set_file_extent_ram_bytes(leaf,
563 btrfs_set_file_extent_num_bytes(leaf,
565 btrfs_mark_buffer_dirty(leaf);
566 } else if (key.offset < inline_limit &&
567 (end > extent_end) &&
568 (inline_limit < extent_end)) {
570 new_size = btrfs_file_extent_calc_inline_size(
571 inline_limit - key.offset);
572 inode_sub_bytes(inode, extent_end -
574 btrfs_set_file_extent_ram_bytes(leaf, extent,
576 if (!compression && !encryption) {
577 btrfs_truncate_item(trans, root, path,
582 /* delete the entire extent */
585 inode_sub_bytes(inode, extent_end -
587 ret = btrfs_del_item(trans, root, path);
588 /* TODO update progress marker and return */
591 btrfs_release_path(root, path);
592 /* the extent will be freed later */
594 if (bookend && found_inline && start <= key.offset) {
596 new_size = btrfs_file_extent_calc_inline_size(
598 inode_sub_bytes(inode, end - key.offset);
599 btrfs_set_file_extent_ram_bytes(leaf, extent,
601 if (!compression && !encryption)
602 ret = btrfs_truncate_item(trans, root, path,
606 /* create bookend, splitting the extent in two */
607 if (bookend && found_extent) {
608 struct btrfs_key ins;
609 ins.objectid = inode->i_ino;
611 btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
613 btrfs_release_path(root, path);
614 ret = btrfs_insert_empty_item(trans, root, path, &ins,
618 leaf = path->nodes[0];
619 extent = btrfs_item_ptr(leaf, path->slots[0],
620 struct btrfs_file_extent_item);
621 write_extent_buffer(leaf, &old,
622 (unsigned long)extent, sizeof(old));
624 btrfs_set_file_extent_compression(leaf, extent,
626 btrfs_set_file_extent_encryption(leaf, extent,
628 btrfs_set_file_extent_other_encoding(leaf, extent,
630 btrfs_set_file_extent_offset(leaf, extent,
631 le64_to_cpu(old.offset) + end - key.offset);
632 WARN_ON(le64_to_cpu(old.num_bytes) <
634 btrfs_set_file_extent_num_bytes(leaf, extent,
638 * set the ram bytes to the size of the full extent
639 * before splitting. This is a worst case flag,
640 * but its the best we can do because we don't know
641 * how splitting affects compression
643 btrfs_set_file_extent_ram_bytes(leaf, extent,
645 btrfs_set_file_extent_type(leaf, extent, found_type);
647 btrfs_mark_buffer_dirty(path->nodes[0]);
649 if (disk_bytenr != 0) {
650 ret = btrfs_update_extent_ref(trans, root,
651 disk_bytenr, orig_parent,
653 root->root_key.objectid,
654 trans->transid, ins.objectid);
658 btrfs_release_path(root, path);
659 if (disk_bytenr != 0)
660 inode_add_bytes(inode, extent_end - end);
663 if (found_extent && !keep) {
664 u64 old_disk_bytenr = le64_to_cpu(old.disk_bytenr);
666 if (old_disk_bytenr != 0) {
667 inode_sub_bytes(inode,
668 le64_to_cpu(old.num_bytes));
669 ret = btrfs_free_extent(trans, root,
671 le64_to_cpu(old.disk_num_bytes),
672 leaf_start, root_owner,
673 root_gen, key.objectid, 0);
675 *hint_byte = old_disk_bytenr;
679 if (search_start >= end) {
685 btrfs_free_path(path);
686 if (locked_end > end) {
687 unlock_extent(&BTRFS_I(inode)->io_tree, end, locked_end - 1,
690 btrfs_check_file(root, inode);
694 static int extent_mergeable(struct extent_buffer *leaf, int slot,
695 u64 objectid, u64 bytenr, u64 *start, u64 *end)
697 struct btrfs_file_extent_item *fi;
698 struct btrfs_key key;
701 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
704 btrfs_item_key_to_cpu(leaf, &key, slot);
705 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
708 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
709 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
710 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
711 btrfs_file_extent_compression(leaf, fi) ||
712 btrfs_file_extent_encryption(leaf, fi) ||
713 btrfs_file_extent_other_encoding(leaf, fi))
716 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
717 if ((*start && *start != key.offset) || (*end && *end != extent_end))
726 * Mark extent in the range start - end as written.
728 * This changes extent type from 'pre-allocated' to 'regular'. If only
729 * part of extent is marked as written, the extent will be split into
732 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
733 struct btrfs_root *root,
734 struct inode *inode, u64 start, u64 end)
736 struct extent_buffer *leaf;
737 struct btrfs_path *path;
738 struct btrfs_file_extent_item *fi;
739 struct btrfs_key key;
747 u64 locked_end = end;
753 btrfs_drop_extent_cache(inode, start, end - 1, 0);
755 path = btrfs_alloc_path();
758 key.objectid = inode->i_ino;
759 key.type = BTRFS_EXTENT_DATA_KEY;
763 key.offset = split - 1;
765 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
766 if (ret > 0 && path->slots[0] > 0)
769 leaf = path->nodes[0];
770 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
771 BUG_ON(key.objectid != inode->i_ino ||
772 key.type != BTRFS_EXTENT_DATA_KEY);
773 fi = btrfs_item_ptr(leaf, path->slots[0],
774 struct btrfs_file_extent_item);
775 extent_type = btrfs_file_extent_type(leaf, fi);
776 BUG_ON(extent_type != BTRFS_FILE_EXTENT_PREALLOC);
777 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
778 BUG_ON(key.offset > start || extent_end < end);
780 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
781 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
782 extent_offset = btrfs_file_extent_offset(leaf, fi);
784 if (key.offset == start)
787 if (key.offset == start && extent_end == end) {
790 u64 leaf_owner = btrfs_header_owner(leaf);
791 u64 leaf_gen = btrfs_header_generation(leaf);
794 if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
795 bytenr, &other_start, &other_end)) {
796 extent_end = other_end;
797 del_slot = path->slots[0] + 1;
799 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
800 leaf->start, leaf_owner,
801 leaf_gen, inode->i_ino, 0);
806 if (extent_mergeable(leaf, path->slots[0] - 1, inode->i_ino,
807 bytenr, &other_start, &other_end)) {
808 key.offset = other_start;
809 del_slot = path->slots[0];
811 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
812 leaf->start, leaf_owner,
813 leaf_gen, inode->i_ino, 0);
818 btrfs_set_file_extent_type(leaf, fi,
819 BTRFS_FILE_EXTENT_REG);
823 fi = btrfs_item_ptr(leaf, del_slot - 1,
824 struct btrfs_file_extent_item);
825 btrfs_set_file_extent_type(leaf, fi, BTRFS_FILE_EXTENT_REG);
826 btrfs_set_file_extent_num_bytes(leaf, fi,
827 extent_end - key.offset);
828 btrfs_mark_buffer_dirty(leaf);
830 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
833 } else if (split == start) {
834 if (locked_end < extent_end) {
835 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
836 locked_end, extent_end - 1, GFP_NOFS);
838 btrfs_release_path(root, path);
839 lock_extent(&BTRFS_I(inode)->io_tree,
840 locked_end, extent_end - 1, GFP_NOFS);
841 locked_end = extent_end;
844 locked_end = extent_end;
846 btrfs_set_file_extent_num_bytes(leaf, fi, split - key.offset);
847 extent_offset += split - key.offset;
849 BUG_ON(key.offset != start);
850 btrfs_set_file_extent_offset(leaf, fi, extent_offset +
852 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - split);
854 btrfs_set_item_key_safe(trans, root, path, &key);
858 if (extent_end == end) {
860 extent_type = BTRFS_FILE_EXTENT_REG;
862 if (extent_end == end && split == start) {
865 if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
866 bytenr, &other_start, &other_end)) {
868 fi = btrfs_item_ptr(leaf, path->slots[0],
869 struct btrfs_file_extent_item);
871 btrfs_set_item_key_safe(trans, root, path, &key);
872 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
873 btrfs_set_file_extent_num_bytes(leaf, fi,
878 if (extent_end == end && split == end) {
881 if (extent_mergeable(leaf, path->slots[0] - 1 , inode->i_ino,
882 bytenr, &other_start, &other_end)) {
884 fi = btrfs_item_ptr(leaf, path->slots[0],
885 struct btrfs_file_extent_item);
886 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end -
892 btrfs_mark_buffer_dirty(leaf);
894 orig_parent = leaf->start;
895 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes,
896 orig_parent, root->root_key.objectid,
897 trans->transid, inode->i_ino);
899 btrfs_release_path(root, path);
902 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*fi));
905 leaf = path->nodes[0];
906 fi = btrfs_item_ptr(leaf, path->slots[0],
907 struct btrfs_file_extent_item);
908 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
909 btrfs_set_file_extent_type(leaf, fi, extent_type);
910 btrfs_set_file_extent_disk_bytenr(leaf, fi, bytenr);
911 btrfs_set_file_extent_disk_num_bytes(leaf, fi, num_bytes);
912 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
913 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - key.offset);
914 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
915 btrfs_set_file_extent_compression(leaf, fi, 0);
916 btrfs_set_file_extent_encryption(leaf, fi, 0);
917 btrfs_set_file_extent_other_encoding(leaf, fi, 0);
919 if (orig_parent != leaf->start) {
920 ret = btrfs_update_extent_ref(trans, root, bytenr,
921 orig_parent, leaf->start,
922 root->root_key.objectid,
923 trans->transid, inode->i_ino);
927 btrfs_mark_buffer_dirty(leaf);
928 btrfs_release_path(root, path);
929 if (split_end && split == start) {
933 if (locked_end > end) {
934 unlock_extent(&BTRFS_I(inode)->io_tree, end, locked_end - 1,
937 btrfs_free_path(path);
942 * this gets pages into the page cache and locks them down, it also properly
943 * waits for data=ordered extents to finish before allowing the pages to be
946 static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
947 struct page **pages, size_t num_pages,
948 loff_t pos, unsigned long first_index,
949 unsigned long last_index, size_t write_bytes)
952 unsigned long index = pos >> PAGE_CACHE_SHIFT;
953 struct inode *inode = fdentry(file)->d_inode;
958 start_pos = pos & ~((u64)root->sectorsize - 1);
959 last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
961 if (start_pos > inode->i_size) {
962 err = btrfs_cont_expand(inode, start_pos);
967 memset(pages, 0, num_pages * sizeof(struct page *));
969 for (i = 0; i < num_pages; i++) {
970 pages[i] = grab_cache_page(inode->i_mapping, index + i);
975 wait_on_page_writeback(pages[i]);
977 if (start_pos < inode->i_size) {
978 struct btrfs_ordered_extent *ordered;
979 lock_extent(&BTRFS_I(inode)->io_tree,
980 start_pos, last_pos - 1, GFP_NOFS);
981 ordered = btrfs_lookup_first_ordered_extent(inode,
984 ordered->file_offset + ordered->len > start_pos &&
985 ordered->file_offset < last_pos) {
986 btrfs_put_ordered_extent(ordered);
987 unlock_extent(&BTRFS_I(inode)->io_tree,
988 start_pos, last_pos - 1, GFP_NOFS);
989 for (i = 0; i < num_pages; i++) {
990 unlock_page(pages[i]);
991 page_cache_release(pages[i]);
993 btrfs_wait_ordered_range(inode, start_pos,
994 last_pos - start_pos);
998 btrfs_put_ordered_extent(ordered);
1000 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
1001 last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC,
1003 unlock_extent(&BTRFS_I(inode)->io_tree,
1004 start_pos, last_pos - 1, GFP_NOFS);
1006 for (i = 0; i < num_pages; i++) {
1007 clear_page_dirty_for_io(pages[i]);
1008 set_page_extent_mapped(pages[i]);
1009 WARN_ON(!PageLocked(pages[i]));
1014 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
1015 size_t count, loff_t *ppos)
1019 ssize_t num_written = 0;
1022 struct inode *inode = fdentry(file)->d_inode;
1023 struct btrfs_root *root = BTRFS_I(inode)->root;
1024 struct page **pages = NULL;
1026 struct page *pinned[2];
1027 unsigned long first_index;
1028 unsigned long last_index;
1031 will_write = ((file->f_flags & O_SYNC) || IS_SYNC(inode) ||
1032 (file->f_flags & O_DIRECT));
1034 nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
1035 PAGE_CACHE_SIZE / (sizeof(struct page *)));
1042 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1043 current->backing_dev_info = inode->i_mapping->backing_dev_info;
1044 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1050 err = file_remove_suid(file);
1053 file_update_time(file);
1055 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
1057 mutex_lock(&inode->i_mutex);
1058 BTRFS_I(inode)->sequence++;
1059 first_index = pos >> PAGE_CACHE_SHIFT;
1060 last_index = (pos + count) >> PAGE_CACHE_SHIFT;
1063 * there are lots of better ways to do this, but this code
1064 * makes sure the first and last page in the file range are
1065 * up to date and ready for cow
1067 if ((pos & (PAGE_CACHE_SIZE - 1))) {
1068 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
1069 if (!PageUptodate(pinned[0])) {
1070 ret = btrfs_readpage(NULL, pinned[0]);
1072 wait_on_page_locked(pinned[0]);
1074 unlock_page(pinned[0]);
1077 if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
1078 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
1079 if (!PageUptodate(pinned[1])) {
1080 ret = btrfs_readpage(NULL, pinned[1]);
1082 wait_on_page_locked(pinned[1]);
1084 unlock_page(pinned[1]);
1089 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1090 size_t write_bytes = min(count, nrptrs *
1091 (size_t)PAGE_CACHE_SIZE -
1093 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
1096 WARN_ON(num_pages > nrptrs);
1097 memset(pages, 0, sizeof(struct page *) * nrptrs);
1099 ret = btrfs_check_free_space(root, write_bytes, 0);
1103 ret = prepare_pages(root, file, pages, num_pages,
1104 pos, first_index, last_index,
1109 ret = btrfs_copy_from_user(pos, num_pages,
1110 write_bytes, pages, buf);
1112 btrfs_drop_pages(pages, num_pages);
1116 ret = dirty_and_release_pages(NULL, root, file, pages,
1117 num_pages, pos, write_bytes);
1118 btrfs_drop_pages(pages, num_pages);
1123 btrfs_fdatawrite_range(inode->i_mapping, pos,
1124 pos + write_bytes - 1,
1127 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
1130 (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1131 btrfs_btree_balance_dirty(root, 1);
1132 btrfs_throttle(root);
1136 count -= write_bytes;
1138 num_written += write_bytes;
1143 mutex_unlock(&inode->i_mutex);
1148 page_cache_release(pinned[0]);
1150 page_cache_release(pinned[1]);
1153 if (num_written > 0 && will_write) {
1154 struct btrfs_trans_handle *trans;
1156 err = btrfs_wait_ordered_range(inode, start_pos, num_written);
1160 if ((file->f_flags & O_SYNC) || IS_SYNC(inode)) {
1161 trans = btrfs_start_transaction(root, 1);
1162 ret = btrfs_log_dentry_safe(trans, root,
1165 btrfs_sync_log(trans, root);
1166 btrfs_end_transaction(trans, root);
1168 btrfs_commit_transaction(trans, root);
1171 if (file->f_flags & O_DIRECT) {
1172 invalidate_mapping_pages(inode->i_mapping,
1173 start_pos >> PAGE_CACHE_SHIFT,
1174 (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
1177 current->backing_dev_info = NULL;
1178 return num_written ? num_written : err;
1181 int btrfs_release_file(struct inode *inode, struct file *filp)
1183 if (filp->private_data)
1184 btrfs_ioctl_trans_end(filp);
1189 * fsync call for both files and directories. This logs the inode into
1190 * the tree log instead of forcing full commits whenever possible.
1192 * It needs to call filemap_fdatawait so that all ordered extent updates are
1193 * in the metadata btree are up to date for copying to the log.
1195 * It drops the inode mutex before doing the tree log commit. This is an
1196 * important optimization for directories because holding the mutex prevents
1197 * new operations on the dir while we write to disk.
1199 int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
1201 struct inode *inode = dentry->d_inode;
1202 struct btrfs_root *root = BTRFS_I(inode)->root;
1204 struct btrfs_trans_handle *trans;
1207 * check the transaction that last modified this inode
1208 * and see if its already been committed
1210 if (!BTRFS_I(inode)->last_trans)
1213 mutex_lock(&root->fs_info->trans_mutex);
1214 if (BTRFS_I(inode)->last_trans <=
1215 root->fs_info->last_trans_committed) {
1216 BTRFS_I(inode)->last_trans = 0;
1217 mutex_unlock(&root->fs_info->trans_mutex);
1220 mutex_unlock(&root->fs_info->trans_mutex);
1222 root->fs_info->tree_log_batch++;
1223 filemap_fdatawrite(inode->i_mapping);
1224 btrfs_wait_ordered_range(inode, 0, (u64)-1);
1225 root->fs_info->tree_log_batch++;
1228 * ok we haven't committed the transaction yet, lets do a commit
1230 if (file->private_data)
1231 btrfs_ioctl_trans_end(file);
1233 trans = btrfs_start_transaction(root, 1);
1239 ret = btrfs_log_dentry_safe(trans, root, file->f_dentry);
1243 /* we've logged all the items and now have a consistent
1244 * version of the file in the log. It is possible that
1245 * someone will come in and modify the file, but that's
1246 * fine because the log is consistent on disk, and we
1247 * have references to all of the file's extents
1249 * It is possible that someone will come in and log the
1250 * file again, but that will end up using the synchronization
1251 * inside btrfs_sync_log to keep things safe.
1253 mutex_unlock(&file->f_dentry->d_inode->i_mutex);
1256 ret = btrfs_commit_transaction(trans, root);
1258 btrfs_sync_log(trans, root);
1259 ret = btrfs_end_transaction(trans, root);
1261 mutex_lock(&file->f_dentry->d_inode->i_mutex);
1263 return ret > 0 ? EIO : ret;
1266 static struct vm_operations_struct btrfs_file_vm_ops = {
1267 .fault = filemap_fault,
1268 .page_mkwrite = btrfs_page_mkwrite,
1271 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
1273 vma->vm_ops = &btrfs_file_vm_ops;
1274 file_accessed(filp);
1278 struct file_operations btrfs_file_operations = {
1279 .llseek = generic_file_llseek,
1280 .read = do_sync_read,
1281 .aio_read = generic_file_aio_read,
1282 .splice_read = generic_file_splice_read,
1283 .write = btrfs_file_write,
1284 .mmap = btrfs_file_mmap,
1285 .open = generic_file_open,
1286 .release = btrfs_release_file,
1287 .fsync = btrfs_sync_file,
1288 .unlocked_ioctl = btrfs_ioctl,
1289 #ifdef CONFIG_COMPAT
1290 .compat_ioctl = btrfs_ioctl,