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 int noinline 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 void noinline 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 int noinline 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 if ((end_of_last_block & 4095) == 0) {
141 printk("strange end of last %Lu %zu %Lu\n", start_pos, write_bytes, end_of_last_block);
143 set_extent_uptodate(io_tree, start_pos, end_of_last_block, GFP_NOFS);
145 /* check for reserved extents on each page, we don't want
146 * to reset the delalloc bit on things that already have
149 btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block);
150 for (i = 0; i < num_pages; i++) {
151 struct page *p = pages[i];
156 if (end_pos > isize) {
157 i_size_write(inode, end_pos);
158 btrfs_update_inode(trans, root, inode);
160 err = btrfs_end_transaction(trans, root);
162 unlock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
167 * this drops all the extents in the cache that intersect the range
168 * [start, end]. Existing extents are split as required.
170 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
173 struct extent_map *em;
174 struct extent_map *split = NULL;
175 struct extent_map *split2 = NULL;
176 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
177 u64 len = end - start + 1;
183 WARN_ON(end < start);
184 if (end == (u64)-1) {
190 split = alloc_extent_map(GFP_NOFS);
192 split2 = alloc_extent_map(GFP_NOFS);
194 spin_lock(&em_tree->lock);
195 em = lookup_extent_mapping(em_tree, start, len);
197 spin_unlock(&em_tree->lock);
201 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
202 spin_unlock(&em_tree->lock);
203 if (em->start <= start &&
204 (!testend || em->start + em->len >= start + len)) {
208 if (start < em->start) {
209 len = em->start - start;
211 len = start + len - (em->start + em->len);
212 start = em->start + em->len;
217 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
218 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
219 remove_extent_mapping(em_tree, em);
221 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
223 split->start = em->start;
224 split->len = start - em->start;
225 split->orig_start = em->orig_start;
226 split->block_start = em->block_start;
229 split->block_len = em->block_len;
231 split->block_len = split->len;
233 split->bdev = em->bdev;
234 split->flags = flags;
235 ret = add_extent_mapping(em_tree, split);
237 free_extent_map(split);
241 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
242 testend && em->start + em->len > start + len) {
243 u64 diff = start + len - em->start;
245 split->start = start + len;
246 split->len = em->start + em->len - (start + len);
247 split->orig_start = em->orig_start;
248 split->bdev = em->bdev;
249 split->flags = flags;
252 split->block_len = em->block_len;
253 split->block_start = em->block_start;
255 split->block_len = split->len;
256 split->block_start = em->block_start + diff;
259 ret = add_extent_mapping(em_tree, split);
261 free_extent_map(split);
264 spin_unlock(&em_tree->lock);
268 /* once for the tree*/
272 free_extent_map(split);
274 free_extent_map(split2);
278 int btrfs_check_file(struct btrfs_root *root, struct inode *inode)
282 struct btrfs_path *path;
283 struct btrfs_key found_key;
284 struct extent_buffer *leaf;
285 struct btrfs_file_extent_item *extent;
294 path = btrfs_alloc_path();
295 ret = btrfs_lookup_file_extent(NULL, root, path, inode->i_ino,
298 nritems = btrfs_header_nritems(path->nodes[0]);
299 if (path->slots[0] >= nritems) {
300 ret = btrfs_next_leaf(root, path);
303 nritems = btrfs_header_nritems(path->nodes[0]);
305 slot = path->slots[0];
306 leaf = path->nodes[0];
307 btrfs_item_key_to_cpu(leaf, &found_key, slot);
308 if (found_key.objectid != inode->i_ino)
310 if (found_key.type != BTRFS_EXTENT_DATA_KEY)
313 if (found_key.offset < last_offset) {
315 btrfs_print_leaf(root, leaf);
316 printk("inode %lu found offset %Lu expected %Lu\n",
317 inode->i_ino, found_key.offset, 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("inode %lu found offset %Lu size %Lu\n", inode->i_ino,
342 last_offset, inode->i_size);
347 btrfs_free_path(path);
353 * this is very complex, but the basic idea is to drop all extents
354 * in the range start - end. hint_block is filled in with a block number
355 * that would be a good hint to the block allocator for this file.
357 * If an extent intersects the range but is not entirely inside the range
358 * it is either truncated or split. Anything entirely inside the range
359 * is deleted from the tree.
361 * inline_limit is used to tell this code which offsets in the file to keep
362 * if they contain inline extents.
364 int noinline btrfs_drop_extents(struct btrfs_trans_handle *trans,
365 struct btrfs_root *root, struct inode *inode,
366 u64 start, u64 end, u64 inline_limit, u64 *hint_byte)
369 u64 locked_end = end;
370 u64 search_start = start;
377 u16 other_encoding = 0;
380 struct extent_buffer *leaf;
381 struct btrfs_file_extent_item *extent;
382 struct btrfs_path *path;
383 struct btrfs_key key;
384 struct btrfs_file_extent_item old;
395 btrfs_drop_extent_cache(inode, start, end - 1, 0);
397 path = btrfs_alloc_path();
402 btrfs_release_path(root, path);
403 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
408 if (path->slots[0] == 0) {
425 leaf = path->nodes[0];
426 slot = path->slots[0];
428 btrfs_item_key_to_cpu(leaf, &key, slot);
429 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY &&
433 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
434 key.objectid != inode->i_ino) {
438 search_start = key.offset;
441 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
442 extent = btrfs_item_ptr(leaf, slot,
443 struct btrfs_file_extent_item);
444 found_type = btrfs_file_extent_type(leaf, extent);
445 compression = btrfs_file_extent_compression(leaf,
447 encryption = btrfs_file_extent_encryption(leaf,
449 other_encoding = btrfs_file_extent_other_encoding(leaf,
451 if (found_type == BTRFS_FILE_EXTENT_REG ||
452 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
454 btrfs_file_extent_disk_bytenr(leaf,
457 *hint_byte = extent_end;
459 extent_end = key.offset +
460 btrfs_file_extent_num_bytes(leaf, extent);
461 ram_bytes = btrfs_file_extent_ram_bytes(leaf,
464 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
466 extent_end = key.offset +
467 btrfs_file_extent_inline_len(leaf, extent);
470 extent_end = search_start;
473 /* we found nothing we can drop */
474 if ((!found_extent && !found_inline) ||
475 search_start >= extent_end) {
478 nritems = btrfs_header_nritems(leaf);
479 if (slot >= nritems - 1) {
480 nextret = btrfs_next_leaf(root, path);
490 if (end <= extent_end && start >= key.offset && found_inline)
491 *hint_byte = EXTENT_MAP_INLINE;
494 read_extent_buffer(leaf, &old, (unsigned long)extent,
496 root_gen = btrfs_header_generation(leaf);
497 root_owner = btrfs_header_owner(leaf);
498 leaf_start = leaf->start;
501 if (end < extent_end && end >= key.offset) {
503 if (found_inline && start <= key.offset)
507 if (bookend && found_extent) {
508 if (locked_end < extent_end) {
509 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
510 locked_end, extent_end - 1,
513 btrfs_release_path(root, path);
514 lock_extent(&BTRFS_I(inode)->io_tree,
515 locked_end, extent_end - 1,
517 locked_end = extent_end;
520 locked_end = extent_end;
522 orig_parent = path->nodes[0]->start;
523 disk_bytenr = le64_to_cpu(old.disk_bytenr);
524 if (disk_bytenr != 0) {
525 ret = btrfs_inc_extent_ref(trans, root,
527 le64_to_cpu(old.disk_num_bytes),
528 orig_parent, root->root_key.objectid,
529 trans->transid, inode->i_ino);
535 u64 mask = root->sectorsize - 1;
536 search_start = (extent_end + mask) & ~mask;
538 search_start = extent_end;
540 /* truncate existing extent */
541 if (start > key.offset) {
545 WARN_ON(start & (root->sectorsize - 1));
547 new_num = start - key.offset;
548 old_num = btrfs_file_extent_num_bytes(leaf,
551 btrfs_file_extent_disk_bytenr(leaf,
553 if (btrfs_file_extent_disk_bytenr(leaf,
555 inode_sub_bytes(inode, old_num -
558 if (!compression && !encryption) {
559 btrfs_set_file_extent_ram_bytes(leaf,
562 btrfs_set_file_extent_num_bytes(leaf,
564 btrfs_mark_buffer_dirty(leaf);
565 } else if (key.offset < inline_limit &&
566 (end > extent_end) &&
567 (inline_limit < extent_end)) {
569 new_size = btrfs_file_extent_calc_inline_size(
570 inline_limit - key.offset);
571 inode_sub_bytes(inode, extent_end -
573 btrfs_set_file_extent_ram_bytes(leaf, extent,
575 if (!compression && !encryption) {
576 btrfs_truncate_item(trans, root, path,
581 /* delete the entire extent */
584 inode_sub_bytes(inode, extent_end -
586 ret = btrfs_del_item(trans, root, path);
587 /* TODO update progress marker and return */
590 btrfs_release_path(root, path);
591 /* the extent will be freed later */
593 if (bookend && found_inline && start <= key.offset) {
595 new_size = btrfs_file_extent_calc_inline_size(
597 inode_sub_bytes(inode, end - key.offset);
598 btrfs_set_file_extent_ram_bytes(leaf, extent,
600 if (!compression && !encryption)
601 ret = btrfs_truncate_item(trans, root, path,
605 /* create bookend, splitting the extent in two */
606 if (bookend && found_extent) {
607 struct btrfs_key ins;
608 ins.objectid = inode->i_ino;
610 btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
612 btrfs_release_path(root, path);
613 ret = btrfs_insert_empty_item(trans, root, path, &ins,
617 leaf = path->nodes[0];
618 extent = btrfs_item_ptr(leaf, path->slots[0],
619 struct btrfs_file_extent_item);
620 write_extent_buffer(leaf, &old,
621 (unsigned long)extent, sizeof(old));
623 btrfs_set_file_extent_compression(leaf, extent,
625 btrfs_set_file_extent_encryption(leaf, extent,
627 btrfs_set_file_extent_other_encoding(leaf, extent,
629 btrfs_set_file_extent_offset(leaf, extent,
630 le64_to_cpu(old.offset) + end - key.offset);
631 WARN_ON(le64_to_cpu(old.num_bytes) <
633 btrfs_set_file_extent_num_bytes(leaf, extent,
637 * set the ram bytes to the size of the full extent
638 * before splitting. This is a worst case flag,
639 * but its the best we can do because we don't know
640 * how splitting affects compression
642 btrfs_set_file_extent_ram_bytes(leaf, extent,
644 btrfs_set_file_extent_type(leaf, extent, found_type);
646 btrfs_mark_buffer_dirty(path->nodes[0]);
648 if (disk_bytenr != 0) {
649 ret = btrfs_update_extent_ref(trans, root,
650 disk_bytenr, orig_parent,
652 root->root_key.objectid,
653 trans->transid, ins.objectid);
657 btrfs_release_path(root, path);
658 if (disk_bytenr != 0) {
659 inode_add_bytes(inode, extent_end - end);
663 if (found_extent && !keep) {
664 u64 disk_bytenr = le64_to_cpu(old.disk_bytenr);
666 if (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 = 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;
752 btrfs_drop_extent_cache(inode, start, end - 1, 0);
754 path = btrfs_alloc_path();
757 key.objectid = inode->i_ino;
758 key.type = BTRFS_EXTENT_DATA_KEY;
762 key.offset = split - 1;
764 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
765 if (ret > 0 && path->slots[0] > 0)
768 leaf = path->nodes[0];
769 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
770 BUG_ON(key.objectid != inode->i_ino ||
771 key.type != BTRFS_EXTENT_DATA_KEY);
772 fi = btrfs_item_ptr(leaf, path->slots[0],
773 struct btrfs_file_extent_item);
774 extent_type = btrfs_file_extent_type(leaf, fi);
775 BUG_ON(extent_type != BTRFS_FILE_EXTENT_PREALLOC);
776 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
777 BUG_ON(key.offset > start || extent_end < end);
779 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
780 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
781 extent_offset = btrfs_file_extent_offset(leaf, fi);
783 if (key.offset == start)
786 if (key.offset == start && extent_end == end) {
789 u64 leaf_owner = btrfs_header_owner(leaf);
790 u64 leaf_gen = btrfs_header_generation(leaf);
793 if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
794 bytenr, &other_start, &other_end)) {
795 extent_end = other_end;
796 del_slot = path->slots[0] + 1;
798 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
799 leaf->start, leaf_owner,
800 leaf_gen, inode->i_ino, 0);
805 if (extent_mergeable(leaf, path->slots[0] - 1, inode->i_ino,
806 bytenr, &other_start, &other_end)) {
807 key.offset = other_start;
808 del_slot = path->slots[0];
810 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
811 leaf->start, leaf_owner,
812 leaf_gen, inode->i_ino, 0);
817 btrfs_set_file_extent_type(leaf, fi,
818 BTRFS_FILE_EXTENT_REG);
822 fi = btrfs_item_ptr(leaf, del_slot - 1,
823 struct btrfs_file_extent_item);
824 btrfs_set_file_extent_type(leaf, fi, BTRFS_FILE_EXTENT_REG);
825 btrfs_set_file_extent_num_bytes(leaf, fi,
826 extent_end - key.offset);
827 btrfs_mark_buffer_dirty(leaf);
829 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
832 } else if (split == start) {
833 if (locked_end < extent_end) {
834 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
835 locked_end, extent_end - 1, GFP_NOFS);
837 btrfs_release_path(root, path);
838 lock_extent(&BTRFS_I(inode)->io_tree,
839 locked_end, extent_end - 1, GFP_NOFS);
840 locked_end = extent_end;
843 locked_end = extent_end;
845 btrfs_set_file_extent_num_bytes(leaf, fi, split - key.offset);
846 extent_offset += split - key.offset;
848 BUG_ON(key.offset != start);
849 btrfs_set_file_extent_offset(leaf, fi, extent_offset +
851 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - split);
853 btrfs_set_item_key_safe(trans, root, path, &key);
857 if (extent_end == end) {
859 extent_type = BTRFS_FILE_EXTENT_REG;
861 if (extent_end == end && split == start) {
864 if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
865 bytenr, &other_start, &other_end)) {
867 fi = btrfs_item_ptr(leaf, path->slots[0],
868 struct btrfs_file_extent_item);
870 btrfs_set_item_key_safe(trans, root, path, &key);
871 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
872 btrfs_set_file_extent_num_bytes(leaf, fi,
877 if (extent_end == end && split == end) {
880 if (extent_mergeable(leaf, path->slots[0] - 1 , inode->i_ino,
881 bytenr, &other_start, &other_end)) {
883 fi = btrfs_item_ptr(leaf, path->slots[0],
884 struct btrfs_file_extent_item);
885 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end -
891 btrfs_mark_buffer_dirty(leaf);
892 btrfs_release_path(root, path);
895 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*fi));
898 leaf = path->nodes[0];
899 fi = btrfs_item_ptr(leaf, path->slots[0],
900 struct btrfs_file_extent_item);
901 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
902 btrfs_set_file_extent_type(leaf, fi, extent_type);
903 btrfs_set_file_extent_disk_bytenr(leaf, fi, bytenr);
904 btrfs_set_file_extent_disk_num_bytes(leaf, fi, num_bytes);
905 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
906 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - key.offset);
907 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
908 btrfs_set_file_extent_compression(leaf, fi, 0);
909 btrfs_set_file_extent_encryption(leaf, fi, 0);
910 btrfs_set_file_extent_other_encoding(leaf, fi, 0);
912 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes,
913 leaf->start, root->root_key.objectid,
914 trans->transid, inode->i_ino);
917 btrfs_mark_buffer_dirty(leaf);
918 btrfs_release_path(root, path);
919 if (split_end && split == start) {
923 if (locked_end > end) {
924 unlock_extent(&BTRFS_I(inode)->io_tree, end, locked_end - 1,
927 btrfs_free_path(path);
932 * this gets pages into the page cache and locks them down, it also properly
933 * waits for data=ordered extents to finish before allowing the pages to be
936 static int noinline prepare_pages(struct btrfs_root *root, struct file *file,
937 struct page **pages, size_t num_pages,
938 loff_t pos, unsigned long first_index,
939 unsigned long last_index, size_t write_bytes)
942 unsigned long index = pos >> PAGE_CACHE_SHIFT;
943 struct inode *inode = fdentry(file)->d_inode;
948 start_pos = pos & ~((u64)root->sectorsize - 1);
949 last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
951 if (start_pos > inode->i_size) {
952 err = btrfs_cont_expand(inode, start_pos);
957 memset(pages, 0, num_pages * sizeof(struct page *));
959 for (i = 0; i < num_pages; i++) {
960 pages[i] = grab_cache_page(inode->i_mapping, index + i);
965 wait_on_page_writeback(pages[i]);
967 if (start_pos < inode->i_size) {
968 struct btrfs_ordered_extent *ordered;
969 lock_extent(&BTRFS_I(inode)->io_tree,
970 start_pos, last_pos - 1, GFP_NOFS);
971 ordered = btrfs_lookup_first_ordered_extent(inode, last_pos -1);
973 ordered->file_offset + ordered->len > start_pos &&
974 ordered->file_offset < last_pos) {
975 btrfs_put_ordered_extent(ordered);
976 unlock_extent(&BTRFS_I(inode)->io_tree,
977 start_pos, last_pos - 1, GFP_NOFS);
978 for (i = 0; i < num_pages; i++) {
979 unlock_page(pages[i]);
980 page_cache_release(pages[i]);
982 btrfs_wait_ordered_range(inode, start_pos,
983 last_pos - start_pos);
987 btrfs_put_ordered_extent(ordered);
989 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
990 last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC,
992 unlock_extent(&BTRFS_I(inode)->io_tree,
993 start_pos, last_pos - 1, GFP_NOFS);
995 for (i = 0; i < num_pages; i++) {
996 clear_page_dirty_for_io(pages[i]);
997 set_page_extent_mapped(pages[i]);
998 WARN_ON(!PageLocked(pages[i]));
1003 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
1004 size_t count, loff_t *ppos)
1008 ssize_t num_written = 0;
1011 struct inode *inode = fdentry(file)->d_inode;
1012 struct btrfs_root *root = BTRFS_I(inode)->root;
1013 struct page **pages = NULL;
1015 struct page *pinned[2];
1016 unsigned long first_index;
1017 unsigned long last_index;
1020 will_write = ((file->f_flags & O_SYNC) || IS_SYNC(inode) ||
1021 (file->f_flags & O_DIRECT));
1023 nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
1024 PAGE_CACHE_SIZE / (sizeof(struct page *)));
1031 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1032 current->backing_dev_info = inode->i_mapping->backing_dev_info;
1033 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1039 err = file_remove_suid(file);
1042 file_update_time(file);
1044 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
1046 mutex_lock(&inode->i_mutex);
1047 first_index = pos >> PAGE_CACHE_SHIFT;
1048 last_index = (pos + count) >> PAGE_CACHE_SHIFT;
1051 * if this is a nodatasum mount, force summing off for the inode
1052 * all the time. That way a later mount with summing on won't
1055 if (btrfs_test_opt(root, NODATASUM))
1056 btrfs_set_flag(inode, NODATASUM);
1059 * there are lots of better ways to do this, but this code
1060 * makes sure the first and last page in the file range are
1061 * up to date and ready for cow
1063 if ((pos & (PAGE_CACHE_SIZE - 1))) {
1064 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
1065 if (!PageUptodate(pinned[0])) {
1066 ret = btrfs_readpage(NULL, pinned[0]);
1068 wait_on_page_locked(pinned[0]);
1070 unlock_page(pinned[0]);
1073 if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
1074 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
1075 if (!PageUptodate(pinned[1])) {
1076 ret = btrfs_readpage(NULL, pinned[1]);
1078 wait_on_page_locked(pinned[1]);
1080 unlock_page(pinned[1]);
1085 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1086 size_t write_bytes = min(count, nrptrs *
1087 (size_t)PAGE_CACHE_SIZE -
1089 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
1092 WARN_ON(num_pages > nrptrs);
1093 memset(pages, 0, sizeof(pages));
1095 ret = btrfs_check_free_space(root, write_bytes, 0);
1099 ret = prepare_pages(root, file, pages, num_pages,
1100 pos, first_index, last_index,
1105 ret = btrfs_copy_from_user(pos, num_pages,
1106 write_bytes, pages, buf);
1108 btrfs_drop_pages(pages, num_pages);
1112 ret = dirty_and_release_pages(NULL, root, file, pages,
1113 num_pages, pos, write_bytes);
1114 btrfs_drop_pages(pages, num_pages);
1119 btrfs_fdatawrite_range(inode->i_mapping, pos,
1120 pos + write_bytes - 1,
1123 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
1126 (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1127 btrfs_btree_balance_dirty(root, 1);
1128 btrfs_throttle(root);
1132 count -= write_bytes;
1134 num_written += write_bytes;
1139 mutex_unlock(&inode->i_mutex);
1144 page_cache_release(pinned[0]);
1146 page_cache_release(pinned[1]);
1149 if (num_written > 0 && will_write) {
1150 struct btrfs_trans_handle *trans;
1152 err = btrfs_wait_ordered_range(inode, start_pos, num_written);
1156 if ((file->f_flags & O_SYNC) || IS_SYNC(inode)) {
1157 trans = btrfs_start_transaction(root, 1);
1158 ret = btrfs_log_dentry_safe(trans, root,
1161 btrfs_sync_log(trans, root);
1162 btrfs_end_transaction(trans, root);
1164 btrfs_commit_transaction(trans, root);
1167 if (file->f_flags & O_DIRECT) {
1168 invalidate_mapping_pages(inode->i_mapping,
1169 start_pos >> PAGE_CACHE_SHIFT,
1170 (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
1173 current->backing_dev_info = NULL;
1174 return num_written ? num_written : err;
1177 int btrfs_release_file(struct inode * inode, struct file * filp)
1179 if (filp->private_data)
1180 btrfs_ioctl_trans_end(filp);
1185 * fsync call for both files and directories. This logs the inode into
1186 * the tree log instead of forcing full commits whenever possible.
1188 * It needs to call filemap_fdatawait so that all ordered extent updates are
1189 * in the metadata btree are up to date for copying to the log.
1191 * It drops the inode mutex before doing the tree log commit. This is an
1192 * important optimization for directories because holding the mutex prevents
1193 * new operations on the dir while we write to disk.
1195 int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
1197 struct inode *inode = dentry->d_inode;
1198 struct btrfs_root *root = BTRFS_I(inode)->root;
1200 struct btrfs_trans_handle *trans;
1203 * check the transaction that last modified this inode
1204 * and see if its already been committed
1206 if (!BTRFS_I(inode)->last_trans)
1209 mutex_lock(&root->fs_info->trans_mutex);
1210 if (BTRFS_I(inode)->last_trans <=
1211 root->fs_info->last_trans_committed) {
1212 BTRFS_I(inode)->last_trans = 0;
1213 mutex_unlock(&root->fs_info->trans_mutex);
1216 mutex_unlock(&root->fs_info->trans_mutex);
1218 root->fs_info->tree_log_batch++;
1219 filemap_fdatawait(inode->i_mapping);
1220 root->fs_info->tree_log_batch++;
1223 * ok we haven't committed the transaction yet, lets do a commit
1225 if (file->private_data)
1226 btrfs_ioctl_trans_end(file);
1228 trans = btrfs_start_transaction(root, 1);
1234 ret = btrfs_log_dentry_safe(trans, root, file->f_dentry);
1239 /* we've logged all the items and now have a consistent
1240 * version of the file in the log. It is possible that
1241 * someone will come in and modify the file, but that's
1242 * fine because the log is consistent on disk, and we
1243 * have references to all of the file's extents
1245 * It is possible that someone will come in and log the
1246 * file again, but that will end up using the synchronization
1247 * inside btrfs_sync_log to keep things safe.
1249 mutex_unlock(&file->f_dentry->d_inode->i_mutex);
1252 ret = btrfs_commit_transaction(trans, root);
1254 btrfs_sync_log(trans, root);
1255 ret = btrfs_end_transaction(trans, root);
1257 mutex_lock(&file->f_dentry->d_inode->i_mutex);
1259 return ret > 0 ? EIO : ret;
1262 static struct vm_operations_struct btrfs_file_vm_ops = {
1263 .fault = filemap_fault,
1264 .page_mkwrite = btrfs_page_mkwrite,
1267 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
1269 vma->vm_ops = &btrfs_file_vm_ops;
1270 file_accessed(filp);
1274 struct file_operations btrfs_file_operations = {
1275 .llseek = generic_file_llseek,
1276 .read = do_sync_read,
1277 .aio_read = generic_file_aio_read,
1278 .splice_read = generic_file_splice_read,
1279 .write = btrfs_file_write,
1280 .mmap = btrfs_file_mmap,
1281 .open = generic_file_open,
1282 .release = btrfs_release_file,
1283 .fsync = btrfs_sync_file,
1284 .unlocked_ioctl = btrfs_ioctl,
1285 #ifdef CONFIG_COMPAT
1286 .compat_ioctl = btrfs_ioctl,