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/swap.h>
28 #include <linux/writeback.h>
29 #include <linux/statfs.h>
30 #include <linux/compat.h>
33 #include "transaction.h"
34 #include "btrfs_inode.h"
36 #include "print-tree.h"
42 /* simple helper to fault in pages and copy. This should go away
43 * and be replaced with calls into generic code.
45 static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
47 struct page **prepared_pages,
48 const char __user *buf)
52 int offset = pos & (PAGE_CACHE_SIZE - 1);
54 for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
55 size_t count = min_t(size_t,
56 PAGE_CACHE_SIZE - offset, write_bytes);
57 struct page *page = prepared_pages[i];
58 fault_in_pages_readable(buf, count);
60 /* Copy data from userspace to the current page */
62 page_fault = __copy_from_user(page_address(page) + offset,
64 /* Flush processor's dcache for this page */
65 flush_dcache_page(page);
73 return page_fault ? -EFAULT : 0;
77 * unlocks pages after btrfs_file_write is done with them
79 static noinline void btrfs_drop_pages(struct page **pages, size_t num_pages)
82 for (i = 0; i < num_pages; i++) {
85 /* page checked is some magic around finding pages that
86 * have been modified without going through btrfs_set_page_dirty
89 ClearPageChecked(pages[i]);
90 unlock_page(pages[i]);
91 mark_page_accessed(pages[i]);
92 page_cache_release(pages[i]);
97 * after copy_from_user, pages need to be dirtied and we need to make
98 * sure holes are created between the current EOF and the start of
99 * any next extents (if required).
101 * this also makes the decision about creating an inline extent vs
102 * doing real data extents, marking pages dirty and delalloc as required.
104 static noinline int dirty_and_release_pages(struct btrfs_trans_handle *trans,
105 struct btrfs_root *root,
114 struct inode *inode = fdentry(file)->d_inode;
117 u64 end_of_last_block;
118 u64 end_pos = pos + write_bytes;
119 loff_t isize = i_size_read(inode);
121 start_pos = pos & ~((u64)root->sectorsize - 1);
122 num_bytes = (write_bytes + pos - start_pos +
123 root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
125 end_of_last_block = start_pos + num_bytes - 1;
126 err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block);
130 for (i = 0; i < num_pages; i++) {
131 struct page *p = pages[i];
136 if (end_pos > isize) {
137 i_size_write(inode, end_pos);
138 /* we've only changed i_size in ram, and we haven't updated
139 * the disk i_size. There is no need to log the inode
147 * this drops all the extents in the cache that intersect the range
148 * [start, end]. Existing extents are split as required.
150 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
153 struct extent_map *em;
154 struct extent_map *split = NULL;
155 struct extent_map *split2 = NULL;
156 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
157 u64 len = end - start + 1;
163 WARN_ON(end < start);
164 if (end == (u64)-1) {
170 split = alloc_extent_map(GFP_NOFS);
172 split2 = alloc_extent_map(GFP_NOFS);
174 write_lock(&em_tree->lock);
175 em = lookup_extent_mapping(em_tree, start, len);
177 write_unlock(&em_tree->lock);
181 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
182 if (em->start <= start &&
183 (!testend || em->start + em->len >= start + len)) {
185 write_unlock(&em_tree->lock);
188 if (start < em->start) {
189 len = em->start - start;
191 len = start + len - (em->start + em->len);
192 start = em->start + em->len;
195 write_unlock(&em_tree->lock);
198 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
199 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
200 remove_extent_mapping(em_tree, em);
202 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
204 split->start = em->start;
205 split->len = start - em->start;
206 split->orig_start = em->orig_start;
207 split->block_start = em->block_start;
210 split->block_len = em->block_len;
212 split->block_len = split->len;
214 split->bdev = em->bdev;
215 split->flags = flags;
216 ret = add_extent_mapping(em_tree, split);
218 free_extent_map(split);
222 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
223 testend && em->start + em->len > start + len) {
224 u64 diff = start + len - em->start;
226 split->start = start + len;
227 split->len = em->start + em->len - (start + len);
228 split->bdev = em->bdev;
229 split->flags = flags;
232 split->block_len = em->block_len;
233 split->block_start = em->block_start;
234 split->orig_start = em->orig_start;
236 split->block_len = split->len;
237 split->block_start = em->block_start + diff;
238 split->orig_start = split->start;
241 ret = add_extent_mapping(em_tree, split);
243 free_extent_map(split);
246 write_unlock(&em_tree->lock);
250 /* once for the tree*/
254 free_extent_map(split);
256 free_extent_map(split2);
261 * this is very complex, but the basic idea is to drop all extents
262 * in the range start - end. hint_block is filled in with a block number
263 * that would be a good hint to the block allocator for this file.
265 * If an extent intersects the range but is not entirely inside the range
266 * it is either truncated or split. Anything entirely inside the range
267 * is deleted from the tree.
269 int btrfs_drop_extents(struct btrfs_trans_handle *trans, struct inode *inode,
270 u64 start, u64 end, u64 *hint_byte, int drop_cache)
272 struct btrfs_root *root = BTRFS_I(inode)->root;
273 struct extent_buffer *leaf;
274 struct btrfs_file_extent_item *fi;
275 struct btrfs_path *path;
276 struct btrfs_key key;
277 struct btrfs_key new_key;
278 u64 search_start = start;
281 u64 extent_offset = 0;
290 btrfs_drop_extent_cache(inode, start, end - 1, 0);
292 path = btrfs_alloc_path();
298 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
302 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
303 leaf = path->nodes[0];
304 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
305 if (key.objectid == inode->i_ino &&
306 key.type == BTRFS_EXTENT_DATA_KEY)
311 leaf = path->nodes[0];
312 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
314 ret = btrfs_next_leaf(root, path);
321 leaf = path->nodes[0];
325 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
326 if (key.objectid > inode->i_ino ||
327 key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
330 fi = btrfs_item_ptr(leaf, path->slots[0],
331 struct btrfs_file_extent_item);
332 extent_type = btrfs_file_extent_type(leaf, fi);
334 if (extent_type == BTRFS_FILE_EXTENT_REG ||
335 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
336 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
337 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
338 extent_offset = btrfs_file_extent_offset(leaf, fi);
339 extent_end = key.offset +
340 btrfs_file_extent_num_bytes(leaf, fi);
341 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
342 extent_end = key.offset +
343 btrfs_file_extent_inline_len(leaf, fi);
346 extent_end = search_start;
349 if (extent_end <= search_start) {
354 search_start = max(key.offset, start);
356 btrfs_release_path(root, path);
361 * | - range to drop - |
362 * | -------- extent -------- |
364 if (start > key.offset && end < extent_end) {
366 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
368 memcpy(&new_key, &key, sizeof(new_key));
369 new_key.offset = start;
370 ret = btrfs_duplicate_item(trans, root, path,
372 if (ret == -EAGAIN) {
373 btrfs_release_path(root, path);
379 leaf = path->nodes[0];
380 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
381 struct btrfs_file_extent_item);
382 btrfs_set_file_extent_num_bytes(leaf, fi,
385 fi = btrfs_item_ptr(leaf, path->slots[0],
386 struct btrfs_file_extent_item);
388 extent_offset += start - key.offset;
389 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
390 btrfs_set_file_extent_num_bytes(leaf, fi,
392 btrfs_mark_buffer_dirty(leaf);
394 if (disk_bytenr > 0) {
395 ret = btrfs_inc_extent_ref(trans, root,
396 disk_bytenr, num_bytes, 0,
397 root->root_key.objectid,
399 start - extent_offset);
401 *hint_byte = disk_bytenr;
406 * | ---- range to drop ----- |
407 * | -------- extent -------- |
409 if (start <= key.offset && end < extent_end) {
410 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
412 memcpy(&new_key, &key, sizeof(new_key));
413 new_key.offset = end;
414 btrfs_set_item_key_safe(trans, root, path, &new_key);
416 extent_offset += end - key.offset;
417 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
418 btrfs_set_file_extent_num_bytes(leaf, fi,
420 btrfs_mark_buffer_dirty(leaf);
421 if (disk_bytenr > 0) {
422 inode_sub_bytes(inode, end - key.offset);
423 *hint_byte = disk_bytenr;
428 search_start = extent_end;
430 * | ---- range to drop ----- |
431 * | -------- extent -------- |
433 if (start > key.offset && end >= extent_end) {
435 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
437 btrfs_set_file_extent_num_bytes(leaf, fi,
439 btrfs_mark_buffer_dirty(leaf);
440 if (disk_bytenr > 0) {
441 inode_sub_bytes(inode, extent_end - start);
442 *hint_byte = disk_bytenr;
444 if (end == extent_end)
452 * | ---- range to drop ----- |
453 * | ------ extent ------ |
455 if (start <= key.offset && end >= extent_end) {
457 del_slot = path->slots[0];
460 BUG_ON(del_slot + del_nr != path->slots[0]);
464 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
465 inode_sub_bytes(inode,
466 extent_end - key.offset);
467 extent_end = ALIGN(extent_end,
469 } else if (disk_bytenr > 0) {
470 ret = btrfs_free_extent(trans, root,
471 disk_bytenr, num_bytes, 0,
472 root->root_key.objectid,
473 key.objectid, key.offset -
476 inode_sub_bytes(inode,
477 extent_end - key.offset);
478 *hint_byte = disk_bytenr;
481 if (end == extent_end)
484 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
489 ret = btrfs_del_items(trans, root, path, del_slot,
496 btrfs_release_path(root, path);
504 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
508 btrfs_free_path(path);
512 static int extent_mergeable(struct extent_buffer *leaf, int slot,
513 u64 objectid, u64 bytenr, u64 *start, u64 *end)
515 struct btrfs_file_extent_item *fi;
516 struct btrfs_key key;
519 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
522 btrfs_item_key_to_cpu(leaf, &key, slot);
523 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
526 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
527 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
528 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
529 btrfs_file_extent_compression(leaf, fi) ||
530 btrfs_file_extent_encryption(leaf, fi) ||
531 btrfs_file_extent_other_encoding(leaf, fi))
534 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
535 if ((*start && *start != key.offset) || (*end && *end != extent_end))
544 * Mark extent in the range start - end as written.
546 * This changes extent type from 'pre-allocated' to 'regular'. If only
547 * part of extent is marked as written, the extent will be split into
550 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
551 struct inode *inode, u64 start, u64 end)
553 struct btrfs_root *root = BTRFS_I(inode)->root;
554 struct extent_buffer *leaf;
555 struct btrfs_path *path;
556 struct btrfs_file_extent_item *fi;
557 struct btrfs_key key;
558 struct btrfs_key new_key;
570 btrfs_drop_extent_cache(inode, start, end - 1, 0);
572 path = btrfs_alloc_path();
576 key.objectid = inode->i_ino;
577 key.type = BTRFS_EXTENT_DATA_KEY;
580 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
581 if (ret > 0 && path->slots[0] > 0)
584 leaf = path->nodes[0];
585 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
586 BUG_ON(key.objectid != inode->i_ino ||
587 key.type != BTRFS_EXTENT_DATA_KEY);
588 fi = btrfs_item_ptr(leaf, path->slots[0],
589 struct btrfs_file_extent_item);
590 BUG_ON(btrfs_file_extent_type(leaf, fi) !=
591 BTRFS_FILE_EXTENT_PREALLOC);
592 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
593 BUG_ON(key.offset > start || extent_end < end);
595 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
596 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
597 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
599 while (start > key.offset || end < extent_end) {
600 if (key.offset == start)
603 memcpy(&new_key, &key, sizeof(new_key));
604 new_key.offset = split;
605 ret = btrfs_duplicate_item(trans, root, path, &new_key);
606 if (ret == -EAGAIN) {
607 btrfs_release_path(root, path);
612 leaf = path->nodes[0];
613 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
614 struct btrfs_file_extent_item);
615 btrfs_set_file_extent_num_bytes(leaf, fi,
618 fi = btrfs_item_ptr(leaf, path->slots[0],
619 struct btrfs_file_extent_item);
621 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
622 btrfs_set_file_extent_num_bytes(leaf, fi,
624 btrfs_mark_buffer_dirty(leaf);
626 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
627 root->root_key.objectid,
628 inode->i_ino, orig_offset);
631 if (split == start) {
634 BUG_ON(start != key.offset);
640 fi = btrfs_item_ptr(leaf, path->slots[0],
641 struct btrfs_file_extent_item);
645 if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
646 bytenr, &other_start, &other_end)) {
647 extent_end = other_end;
648 del_slot = path->slots[0] + 1;
650 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
651 0, root->root_key.objectid,
652 inode->i_ino, orig_offset);
657 if (extent_mergeable(leaf, path->slots[0] - 1, inode->i_ino,
658 bytenr, &other_start, &other_end)) {
659 key.offset = other_start;
660 del_slot = path->slots[0];
662 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
663 0, root->root_key.objectid,
664 inode->i_ino, orig_offset);
668 btrfs_set_file_extent_type(leaf, fi,
669 BTRFS_FILE_EXTENT_REG);
670 btrfs_mark_buffer_dirty(leaf);
674 fi = btrfs_item_ptr(leaf, del_slot - 1,
675 struct btrfs_file_extent_item);
676 btrfs_set_file_extent_type(leaf, fi, BTRFS_FILE_EXTENT_REG);
677 btrfs_set_file_extent_num_bytes(leaf, fi,
678 extent_end - key.offset);
679 btrfs_mark_buffer_dirty(leaf);
681 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
684 btrfs_free_path(path);
689 * this gets pages into the page cache and locks them down, it also properly
690 * waits for data=ordered extents to finish before allowing the pages to be
693 static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
694 struct page **pages, size_t num_pages,
695 loff_t pos, unsigned long first_index,
696 unsigned long last_index, size_t write_bytes)
699 unsigned long index = pos >> PAGE_CACHE_SHIFT;
700 struct inode *inode = fdentry(file)->d_inode;
705 start_pos = pos & ~((u64)root->sectorsize - 1);
706 last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
708 if (start_pos > inode->i_size) {
709 err = btrfs_cont_expand(inode, start_pos);
714 memset(pages, 0, num_pages * sizeof(struct page *));
716 for (i = 0; i < num_pages; i++) {
717 pages[i] = grab_cache_page(inode->i_mapping, index + i);
722 wait_on_page_writeback(pages[i]);
724 if (start_pos < inode->i_size) {
725 struct btrfs_ordered_extent *ordered;
726 lock_extent(&BTRFS_I(inode)->io_tree,
727 start_pos, last_pos - 1, GFP_NOFS);
728 ordered = btrfs_lookup_first_ordered_extent(inode,
731 ordered->file_offset + ordered->len > start_pos &&
732 ordered->file_offset < last_pos) {
733 btrfs_put_ordered_extent(ordered);
734 unlock_extent(&BTRFS_I(inode)->io_tree,
735 start_pos, last_pos - 1, GFP_NOFS);
736 for (i = 0; i < num_pages; i++) {
737 unlock_page(pages[i]);
738 page_cache_release(pages[i]);
740 btrfs_wait_ordered_range(inode, start_pos,
741 last_pos - start_pos);
745 btrfs_put_ordered_extent(ordered);
747 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
748 last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
749 EXTENT_DO_ACCOUNTING,
751 unlock_extent(&BTRFS_I(inode)->io_tree,
752 start_pos, last_pos - 1, GFP_NOFS);
754 for (i = 0; i < num_pages; i++) {
755 clear_page_dirty_for_io(pages[i]);
756 set_page_extent_mapped(pages[i]);
757 WARN_ON(!PageLocked(pages[i]));
762 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
763 size_t count, loff_t *ppos)
767 ssize_t num_written = 0;
770 struct inode *inode = fdentry(file)->d_inode;
771 struct btrfs_root *root = BTRFS_I(inode)->root;
772 struct page **pages = NULL;
774 struct page *pinned[2];
775 unsigned long first_index;
776 unsigned long last_index;
779 will_write = ((file->f_flags & O_SYNC) || IS_SYNC(inode) ||
780 (file->f_flags & O_DIRECT));
782 nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
783 PAGE_CACHE_SIZE / (sizeof(struct page *)));
790 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
792 /* do the reserve before the mutex lock in case we have to do some
793 * flushing. We wouldn't deadlock, but this is more polite.
795 err = btrfs_reserve_metadata_for_delalloc(root, inode, 1);
799 mutex_lock(&inode->i_mutex);
801 current->backing_dev_info = inode->i_mapping->backing_dev_info;
802 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
809 err = file_remove_suid(file);
813 file_update_time(file);
815 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
817 /* generic_write_checks can change our pos */
820 BTRFS_I(inode)->sequence++;
821 first_index = pos >> PAGE_CACHE_SHIFT;
822 last_index = (pos + count) >> PAGE_CACHE_SHIFT;
825 * there are lots of better ways to do this, but this code
826 * makes sure the first and last page in the file range are
827 * up to date and ready for cow
829 if ((pos & (PAGE_CACHE_SIZE - 1))) {
830 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
831 if (!PageUptodate(pinned[0])) {
832 ret = btrfs_readpage(NULL, pinned[0]);
834 wait_on_page_locked(pinned[0]);
836 unlock_page(pinned[0]);
839 if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
840 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
841 if (!PageUptodate(pinned[1])) {
842 ret = btrfs_readpage(NULL, pinned[1]);
844 wait_on_page_locked(pinned[1]);
846 unlock_page(pinned[1]);
851 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
852 size_t write_bytes = min(count, nrptrs *
853 (size_t)PAGE_CACHE_SIZE -
855 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
858 WARN_ON(num_pages > nrptrs);
859 memset(pages, 0, sizeof(struct page *) * nrptrs);
861 ret = btrfs_check_data_free_space(root, inode, write_bytes);
865 ret = prepare_pages(root, file, pages, num_pages,
866 pos, first_index, last_index,
869 btrfs_free_reserved_data_space(root, inode,
874 ret = btrfs_copy_from_user(pos, num_pages,
875 write_bytes, pages, buf);
877 btrfs_free_reserved_data_space(root, inode,
879 btrfs_drop_pages(pages, num_pages);
883 ret = dirty_and_release_pages(NULL, root, file, pages,
884 num_pages, pos, write_bytes);
885 btrfs_drop_pages(pages, num_pages);
887 btrfs_free_reserved_data_space(root, inode,
893 filemap_fdatawrite_range(inode->i_mapping, pos,
894 pos + write_bytes - 1);
896 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
899 (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
900 btrfs_btree_balance_dirty(root, 1);
901 btrfs_throttle(root);
905 count -= write_bytes;
907 num_written += write_bytes;
912 mutex_unlock(&inode->i_mutex);
915 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
920 page_cache_release(pinned[0]);
922 page_cache_release(pinned[1]);
926 * we want to make sure fsync finds this change
927 * but we haven't joined a transaction running right now.
929 * Later on, someone is sure to update the inode and get the
930 * real transid recorded.
932 * We set last_trans now to the fs_info generation + 1,
933 * this will either be one more than the running transaction
934 * or the generation used for the next transaction if there isn't
935 * one running right now.
937 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
939 if (num_written > 0 && will_write) {
940 struct btrfs_trans_handle *trans;
942 err = btrfs_wait_ordered_range(inode, start_pos, num_written);
946 if ((file->f_flags & O_SYNC) || IS_SYNC(inode)) {
947 trans = btrfs_start_transaction(root, 1);
948 ret = btrfs_log_dentry_safe(trans, root,
951 ret = btrfs_sync_log(trans, root);
953 btrfs_end_transaction(trans, root);
955 btrfs_commit_transaction(trans, root);
956 } else if (ret != BTRFS_NO_LOG_SYNC) {
957 btrfs_commit_transaction(trans, root);
959 btrfs_end_transaction(trans, root);
962 if (file->f_flags & O_DIRECT) {
963 invalidate_mapping_pages(inode->i_mapping,
964 start_pos >> PAGE_CACHE_SHIFT,
965 (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
968 current->backing_dev_info = NULL;
969 return num_written ? num_written : err;
972 int btrfs_release_file(struct inode *inode, struct file *filp)
975 * ordered_data_close is set by settattr when we are about to truncate
976 * a file from a non-zero size to a zero size. This tries to
977 * flush down new bytes that may have been written if the
978 * application were using truncate to replace a file in place.
980 if (BTRFS_I(inode)->ordered_data_close) {
981 BTRFS_I(inode)->ordered_data_close = 0;
982 btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode);
983 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
984 filemap_flush(inode->i_mapping);
986 if (filp->private_data)
987 btrfs_ioctl_trans_end(filp);
992 * fsync call for both files and directories. This logs the inode into
993 * the tree log instead of forcing full commits whenever possible.
995 * It needs to call filemap_fdatawait so that all ordered extent updates are
996 * in the metadata btree are up to date for copying to the log.
998 * It drops the inode mutex before doing the tree log commit. This is an
999 * important optimization for directories because holding the mutex prevents
1000 * new operations on the dir while we write to disk.
1002 int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
1004 struct inode *inode = dentry->d_inode;
1005 struct btrfs_root *root = BTRFS_I(inode)->root;
1007 struct btrfs_trans_handle *trans;
1010 /* we wait first, since the writeback may change the inode */
1012 /* the VFS called filemap_fdatawrite for us */
1013 btrfs_wait_ordered_range(inode, 0, (u64)-1);
1017 * check the transaction that last modified this inode
1018 * and see if its already been committed
1020 if (!BTRFS_I(inode)->last_trans)
1024 * if the last transaction that changed this file was before
1025 * the current transaction, we can bail out now without any
1028 mutex_lock(&root->fs_info->trans_mutex);
1029 if (BTRFS_I(inode)->last_trans <=
1030 root->fs_info->last_trans_committed) {
1031 BTRFS_I(inode)->last_trans = 0;
1032 mutex_unlock(&root->fs_info->trans_mutex);
1035 mutex_unlock(&root->fs_info->trans_mutex);
1038 * ok we haven't committed the transaction yet, lets do a commit
1040 if (file && file->private_data)
1041 btrfs_ioctl_trans_end(file);
1043 trans = btrfs_start_transaction(root, 1);
1049 ret = btrfs_log_dentry_safe(trans, root, dentry);
1053 /* we've logged all the items and now have a consistent
1054 * version of the file in the log. It is possible that
1055 * someone will come in and modify the file, but that's
1056 * fine because the log is consistent on disk, and we
1057 * have references to all of the file's extents
1059 * It is possible that someone will come in and log the
1060 * file again, but that will end up using the synchronization
1061 * inside btrfs_sync_log to keep things safe.
1063 mutex_unlock(&dentry->d_inode->i_mutex);
1065 if (ret != BTRFS_NO_LOG_SYNC) {
1067 ret = btrfs_commit_transaction(trans, root);
1069 ret = btrfs_sync_log(trans, root);
1071 ret = btrfs_end_transaction(trans, root);
1073 ret = btrfs_commit_transaction(trans, root);
1076 ret = btrfs_end_transaction(trans, root);
1078 mutex_lock(&dentry->d_inode->i_mutex);
1080 return ret > 0 ? EIO : ret;
1083 static const struct vm_operations_struct btrfs_file_vm_ops = {
1084 .fault = filemap_fault,
1085 .page_mkwrite = btrfs_page_mkwrite,
1088 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
1090 vma->vm_ops = &btrfs_file_vm_ops;
1091 file_accessed(filp);
1095 const struct file_operations btrfs_file_operations = {
1096 .llseek = generic_file_llseek,
1097 .read = do_sync_read,
1098 .aio_read = generic_file_aio_read,
1099 .splice_read = generic_file_splice_read,
1100 .write = btrfs_file_write,
1101 .mmap = btrfs_file_mmap,
1102 .open = generic_file_open,
1103 .release = btrfs_release_file,
1104 .fsync = btrfs_sync_file,
1105 .unlocked_ioctl = btrfs_ioctl,
1106 #ifdef CONFIG_COMPAT
1107 .compat_ioctl = btrfs_ioctl,