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 (testend && em->start + em->len >= start + len) {
184 write_unlock(&em_tree->lock);
187 start = em->start + em->len;
189 len = start + len - (em->start + em->len);
191 write_unlock(&em_tree->lock);
194 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
195 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
196 remove_extent_mapping(em_tree, em);
198 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
200 split->start = em->start;
201 split->len = start - em->start;
202 split->orig_start = em->orig_start;
203 split->block_start = em->block_start;
206 split->block_len = em->block_len;
208 split->block_len = split->len;
210 split->bdev = em->bdev;
211 split->flags = flags;
212 ret = add_extent_mapping(em_tree, split);
214 free_extent_map(split);
218 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
219 testend && em->start + em->len > start + len) {
220 u64 diff = start + len - em->start;
222 split->start = start + len;
223 split->len = em->start + em->len - (start + len);
224 split->bdev = em->bdev;
225 split->flags = flags;
228 split->block_len = em->block_len;
229 split->block_start = em->block_start;
230 split->orig_start = em->orig_start;
232 split->block_len = split->len;
233 split->block_start = em->block_start + diff;
234 split->orig_start = split->start;
237 ret = add_extent_mapping(em_tree, split);
239 free_extent_map(split);
242 write_unlock(&em_tree->lock);
246 /* once for the tree*/
250 free_extent_map(split);
252 free_extent_map(split2);
257 * this is very complex, but the basic idea is to drop all extents
258 * in the range start - end. hint_block is filled in with a block number
259 * that would be a good hint to the block allocator for this file.
261 * If an extent intersects the range but is not entirely inside the range
262 * it is either truncated or split. Anything entirely inside the range
263 * is deleted from the tree.
265 int btrfs_drop_extents(struct btrfs_trans_handle *trans, struct inode *inode,
266 u64 start, u64 end, u64 *hint_byte, int drop_cache)
268 struct btrfs_root *root = BTRFS_I(inode)->root;
269 struct extent_buffer *leaf;
270 struct btrfs_file_extent_item *fi;
271 struct btrfs_path *path;
272 struct btrfs_key key;
273 struct btrfs_key new_key;
274 u64 search_start = start;
277 u64 extent_offset = 0;
286 btrfs_drop_extent_cache(inode, start, end - 1, 0);
288 path = btrfs_alloc_path();
294 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
298 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
299 leaf = path->nodes[0];
300 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
301 if (key.objectid == inode->i_ino &&
302 key.type == BTRFS_EXTENT_DATA_KEY)
307 leaf = path->nodes[0];
308 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
310 ret = btrfs_next_leaf(root, path);
317 leaf = path->nodes[0];
321 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
322 if (key.objectid > inode->i_ino ||
323 key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
326 fi = btrfs_item_ptr(leaf, path->slots[0],
327 struct btrfs_file_extent_item);
328 extent_type = btrfs_file_extent_type(leaf, fi);
330 if (extent_type == BTRFS_FILE_EXTENT_REG ||
331 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
332 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
333 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
334 extent_offset = btrfs_file_extent_offset(leaf, fi);
335 extent_end = key.offset +
336 btrfs_file_extent_num_bytes(leaf, fi);
337 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
338 extent_end = key.offset +
339 btrfs_file_extent_inline_len(leaf, fi);
342 extent_end = search_start;
345 if (extent_end <= search_start) {
350 search_start = max(key.offset, start);
352 btrfs_release_path(root, path);
357 * | - range to drop - |
358 * | -------- extent -------- |
360 if (start > key.offset && end < extent_end) {
362 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
364 memcpy(&new_key, &key, sizeof(new_key));
365 new_key.offset = start;
366 ret = btrfs_duplicate_item(trans, root, path,
368 if (ret == -EAGAIN) {
369 btrfs_release_path(root, path);
375 leaf = path->nodes[0];
376 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
377 struct btrfs_file_extent_item);
378 btrfs_set_file_extent_num_bytes(leaf, fi,
381 fi = btrfs_item_ptr(leaf, path->slots[0],
382 struct btrfs_file_extent_item);
384 extent_offset += start - key.offset;
385 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
386 btrfs_set_file_extent_num_bytes(leaf, fi,
388 btrfs_mark_buffer_dirty(leaf);
390 if (disk_bytenr > 0) {
391 ret = btrfs_inc_extent_ref(trans, root,
392 disk_bytenr, num_bytes, 0,
393 root->root_key.objectid,
395 start - extent_offset);
397 *hint_byte = disk_bytenr;
402 * | ---- range to drop ----- |
403 * | -------- extent -------- |
405 if (start <= key.offset && end < extent_end) {
406 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
408 memcpy(&new_key, &key, sizeof(new_key));
409 new_key.offset = end;
410 btrfs_set_item_key_safe(trans, root, path, &new_key);
412 extent_offset += end - key.offset;
413 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
414 btrfs_set_file_extent_num_bytes(leaf, fi,
416 btrfs_mark_buffer_dirty(leaf);
417 if (disk_bytenr > 0) {
418 inode_sub_bytes(inode, end - key.offset);
419 *hint_byte = disk_bytenr;
424 search_start = extent_end;
426 * | ---- range to drop ----- |
427 * | -------- extent -------- |
429 if (start > key.offset && end >= extent_end) {
431 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
433 btrfs_set_file_extent_num_bytes(leaf, fi,
435 btrfs_mark_buffer_dirty(leaf);
436 if (disk_bytenr > 0) {
437 inode_sub_bytes(inode, extent_end - start);
438 *hint_byte = disk_bytenr;
440 if (end == extent_end)
448 * | ---- range to drop ----- |
449 * | ------ extent ------ |
451 if (start <= key.offset && end >= extent_end) {
453 del_slot = path->slots[0];
456 BUG_ON(del_slot + del_nr != path->slots[0]);
460 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
461 inode_sub_bytes(inode,
462 extent_end - key.offset);
463 extent_end = ALIGN(extent_end,
465 } else if (disk_bytenr > 0) {
466 ret = btrfs_free_extent(trans, root,
467 disk_bytenr, num_bytes, 0,
468 root->root_key.objectid,
469 key.objectid, key.offset -
472 inode_sub_bytes(inode,
473 extent_end - key.offset);
474 *hint_byte = disk_bytenr;
477 if (end == extent_end)
480 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
485 ret = btrfs_del_items(trans, root, path, del_slot,
492 btrfs_release_path(root, path);
500 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
504 btrfs_free_path(path);
508 static int extent_mergeable(struct extent_buffer *leaf, int slot,
509 u64 objectid, u64 bytenr, u64 orig_offset,
510 u64 *start, u64 *end)
512 struct btrfs_file_extent_item *fi;
513 struct btrfs_key key;
516 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
519 btrfs_item_key_to_cpu(leaf, &key, slot);
520 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
523 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
524 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
525 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
526 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
527 btrfs_file_extent_compression(leaf, fi) ||
528 btrfs_file_extent_encryption(leaf, fi) ||
529 btrfs_file_extent_other_encoding(leaf, fi))
532 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
533 if ((*start && *start != key.offset) || (*end && *end != extent_end))
542 * Mark extent in the range start - end as written.
544 * This changes extent type from 'pre-allocated' to 'regular'. If only
545 * part of extent is marked as written, the extent will be split into
548 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
549 struct inode *inode, u64 start, u64 end)
551 struct btrfs_root *root = BTRFS_I(inode)->root;
552 struct extent_buffer *leaf;
553 struct btrfs_path *path;
554 struct btrfs_file_extent_item *fi;
555 struct btrfs_key key;
556 struct btrfs_key new_key;
569 btrfs_drop_extent_cache(inode, start, end - 1, 0);
571 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);
598 memcpy(&new_key, &key, sizeof(new_key));
600 if (start == key.offset && end < extent_end) {
603 if (extent_mergeable(leaf, path->slots[0] - 1,
604 inode->i_ino, bytenr, orig_offset,
605 &other_start, &other_end)) {
606 new_key.offset = end;
607 btrfs_set_item_key_safe(trans, root, path, &new_key);
608 fi = btrfs_item_ptr(leaf, path->slots[0],
609 struct btrfs_file_extent_item);
610 btrfs_set_file_extent_num_bytes(leaf, fi,
612 btrfs_set_file_extent_offset(leaf, fi,
614 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
615 struct btrfs_file_extent_item);
616 btrfs_set_file_extent_num_bytes(leaf, fi,
618 btrfs_mark_buffer_dirty(leaf);
623 if (start > key.offset && end == extent_end) {
626 if (extent_mergeable(leaf, path->slots[0] + 1,
627 inode->i_ino, bytenr, orig_offset,
628 &other_start, &other_end)) {
629 fi = btrfs_item_ptr(leaf, path->slots[0],
630 struct btrfs_file_extent_item);
631 btrfs_set_file_extent_num_bytes(leaf, fi,
634 new_key.offset = start;
635 btrfs_set_item_key_safe(trans, root, path, &new_key);
637 fi = btrfs_item_ptr(leaf, path->slots[0],
638 struct btrfs_file_extent_item);
639 btrfs_set_file_extent_num_bytes(leaf, fi,
641 btrfs_set_file_extent_offset(leaf, fi,
642 start - orig_offset);
643 btrfs_mark_buffer_dirty(leaf);
648 while (start > key.offset || end < extent_end) {
649 if (key.offset == start)
652 new_key.offset = split;
653 ret = btrfs_duplicate_item(trans, root, path, &new_key);
654 if (ret == -EAGAIN) {
655 btrfs_release_path(root, path);
660 leaf = path->nodes[0];
661 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
662 struct btrfs_file_extent_item);
663 btrfs_set_file_extent_num_bytes(leaf, fi,
666 fi = btrfs_item_ptr(leaf, path->slots[0],
667 struct btrfs_file_extent_item);
669 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
670 btrfs_set_file_extent_num_bytes(leaf, fi,
672 btrfs_mark_buffer_dirty(leaf);
674 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
675 root->root_key.objectid,
676 inode->i_ino, orig_offset);
679 if (split == start) {
682 BUG_ON(start != key.offset);
691 if (extent_mergeable(leaf, path->slots[0] + 1,
692 inode->i_ino, bytenr, orig_offset,
693 &other_start, &other_end)) {
695 btrfs_release_path(root, path);
698 extent_end = other_end;
699 del_slot = path->slots[0] + 1;
701 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
702 0, root->root_key.objectid,
703 inode->i_ino, orig_offset);
708 if (extent_mergeable(leaf, path->slots[0] - 1,
709 inode->i_ino, bytenr, orig_offset,
710 &other_start, &other_end)) {
712 btrfs_release_path(root, path);
715 key.offset = other_start;
716 del_slot = path->slots[0];
718 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
719 0, root->root_key.objectid,
720 inode->i_ino, orig_offset);
724 fi = btrfs_item_ptr(leaf, path->slots[0],
725 struct btrfs_file_extent_item);
726 btrfs_set_file_extent_type(leaf, fi,
727 BTRFS_FILE_EXTENT_REG);
728 btrfs_mark_buffer_dirty(leaf);
730 fi = btrfs_item_ptr(leaf, del_slot - 1,
731 struct btrfs_file_extent_item);
732 btrfs_set_file_extent_type(leaf, fi,
733 BTRFS_FILE_EXTENT_REG);
734 btrfs_set_file_extent_num_bytes(leaf, fi,
735 extent_end - key.offset);
736 btrfs_mark_buffer_dirty(leaf);
738 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
742 btrfs_free_path(path);
747 * this gets pages into the page cache and locks them down, it also properly
748 * waits for data=ordered extents to finish before allowing the pages to be
751 static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
752 struct page **pages, size_t num_pages,
753 loff_t pos, unsigned long first_index,
754 unsigned long last_index, size_t write_bytes)
757 unsigned long index = pos >> PAGE_CACHE_SHIFT;
758 struct inode *inode = fdentry(file)->d_inode;
763 start_pos = pos & ~((u64)root->sectorsize - 1);
764 last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
766 if (start_pos > inode->i_size) {
767 err = btrfs_cont_expand(inode, start_pos);
772 memset(pages, 0, num_pages * sizeof(struct page *));
774 for (i = 0; i < num_pages; i++) {
775 pages[i] = grab_cache_page(inode->i_mapping, index + i);
780 wait_on_page_writeback(pages[i]);
782 if (start_pos < inode->i_size) {
783 struct btrfs_ordered_extent *ordered;
784 lock_extent(&BTRFS_I(inode)->io_tree,
785 start_pos, last_pos - 1, GFP_NOFS);
786 ordered = btrfs_lookup_first_ordered_extent(inode,
789 ordered->file_offset + ordered->len > start_pos &&
790 ordered->file_offset < last_pos) {
791 btrfs_put_ordered_extent(ordered);
792 unlock_extent(&BTRFS_I(inode)->io_tree,
793 start_pos, last_pos - 1, GFP_NOFS);
794 for (i = 0; i < num_pages; i++) {
795 unlock_page(pages[i]);
796 page_cache_release(pages[i]);
798 btrfs_wait_ordered_range(inode, start_pos,
799 last_pos - start_pos);
803 btrfs_put_ordered_extent(ordered);
805 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
806 last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
807 EXTENT_DO_ACCOUNTING,
809 unlock_extent(&BTRFS_I(inode)->io_tree,
810 start_pos, last_pos - 1, GFP_NOFS);
812 for (i = 0; i < num_pages; i++) {
813 clear_page_dirty_for_io(pages[i]);
814 set_page_extent_mapped(pages[i]);
815 WARN_ON(!PageLocked(pages[i]));
820 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
821 size_t count, loff_t *ppos)
825 ssize_t num_written = 0;
828 struct inode *inode = fdentry(file)->d_inode;
829 struct btrfs_root *root = BTRFS_I(inode)->root;
830 struct page **pages = NULL;
832 struct page *pinned[2];
833 unsigned long first_index;
834 unsigned long last_index;
837 will_write = ((file->f_flags & O_DSYNC) || IS_SYNC(inode) ||
838 (file->f_flags & O_DIRECT));
840 nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
841 PAGE_CACHE_SIZE / (sizeof(struct page *)));
848 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
850 /* do the reserve before the mutex lock in case we have to do some
851 * flushing. We wouldn't deadlock, but this is more polite.
853 err = btrfs_reserve_metadata_for_delalloc(root, inode, 1);
857 mutex_lock(&inode->i_mutex);
859 current->backing_dev_info = inode->i_mapping->backing_dev_info;
860 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
867 err = file_remove_suid(file);
871 file_update_time(file);
873 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
875 /* generic_write_checks can change our pos */
878 BTRFS_I(inode)->sequence++;
879 first_index = pos >> PAGE_CACHE_SHIFT;
880 last_index = (pos + count) >> PAGE_CACHE_SHIFT;
883 * there are lots of better ways to do this, but this code
884 * makes sure the first and last page in the file range are
885 * up to date and ready for cow
887 if ((pos & (PAGE_CACHE_SIZE - 1))) {
888 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
889 if (!PageUptodate(pinned[0])) {
890 ret = btrfs_readpage(NULL, pinned[0]);
892 wait_on_page_locked(pinned[0]);
894 unlock_page(pinned[0]);
897 if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
898 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
899 if (!PageUptodate(pinned[1])) {
900 ret = btrfs_readpage(NULL, pinned[1]);
902 wait_on_page_locked(pinned[1]);
904 unlock_page(pinned[1]);
909 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
910 size_t write_bytes = min(count, nrptrs *
911 (size_t)PAGE_CACHE_SIZE -
913 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
916 WARN_ON(num_pages > nrptrs);
917 memset(pages, 0, sizeof(struct page *) * nrptrs);
919 ret = btrfs_check_data_free_space(root, inode, write_bytes);
923 ret = prepare_pages(root, file, pages, num_pages,
924 pos, first_index, last_index,
927 btrfs_free_reserved_data_space(root, inode,
932 ret = btrfs_copy_from_user(pos, num_pages,
933 write_bytes, pages, buf);
935 btrfs_free_reserved_data_space(root, inode,
937 btrfs_drop_pages(pages, num_pages);
941 ret = dirty_and_release_pages(NULL, root, file, pages,
942 num_pages, pos, write_bytes);
943 btrfs_drop_pages(pages, num_pages);
945 btrfs_free_reserved_data_space(root, inode,
951 filemap_fdatawrite_range(inode->i_mapping, pos,
952 pos + write_bytes - 1);
954 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
957 (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
958 btrfs_btree_balance_dirty(root, 1);
959 btrfs_throttle(root);
963 count -= write_bytes;
965 num_written += write_bytes;
970 mutex_unlock(&inode->i_mutex);
973 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
978 page_cache_release(pinned[0]);
980 page_cache_release(pinned[1]);
984 * we want to make sure fsync finds this change
985 * but we haven't joined a transaction running right now.
987 * Later on, someone is sure to update the inode and get the
988 * real transid recorded.
990 * We set last_trans now to the fs_info generation + 1,
991 * this will either be one more than the running transaction
992 * or the generation used for the next transaction if there isn't
993 * one running right now.
995 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
997 if (num_written > 0 && will_write) {
998 struct btrfs_trans_handle *trans;
1000 err = btrfs_wait_ordered_range(inode, start_pos, num_written);
1004 if ((file->f_flags & O_DSYNC) || IS_SYNC(inode)) {
1005 trans = btrfs_start_transaction(root, 1);
1006 ret = btrfs_log_dentry_safe(trans, root,
1009 ret = btrfs_sync_log(trans, root);
1011 btrfs_end_transaction(trans, root);
1013 btrfs_commit_transaction(trans, root);
1014 } else if (ret != BTRFS_NO_LOG_SYNC) {
1015 btrfs_commit_transaction(trans, root);
1017 btrfs_end_transaction(trans, root);
1020 if (file->f_flags & O_DIRECT) {
1021 invalidate_mapping_pages(inode->i_mapping,
1022 start_pos >> PAGE_CACHE_SHIFT,
1023 (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
1026 current->backing_dev_info = NULL;
1027 return num_written ? num_written : err;
1030 int btrfs_release_file(struct inode *inode, struct file *filp)
1033 * ordered_data_close is set by settattr when we are about to truncate
1034 * a file from a non-zero size to a zero size. This tries to
1035 * flush down new bytes that may have been written if the
1036 * application were using truncate to replace a file in place.
1038 if (BTRFS_I(inode)->ordered_data_close) {
1039 BTRFS_I(inode)->ordered_data_close = 0;
1040 btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode);
1041 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1042 filemap_flush(inode->i_mapping);
1044 if (filp->private_data)
1045 btrfs_ioctl_trans_end(filp);
1050 * fsync call for both files and directories. This logs the inode into
1051 * the tree log instead of forcing full commits whenever possible.
1053 * It needs to call filemap_fdatawait so that all ordered extent updates are
1054 * in the metadata btree are up to date for copying to the log.
1056 * It drops the inode mutex before doing the tree log commit. This is an
1057 * important optimization for directories because holding the mutex prevents
1058 * new operations on the dir while we write to disk.
1060 int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
1062 struct inode *inode = dentry->d_inode;
1063 struct btrfs_root *root = BTRFS_I(inode)->root;
1065 struct btrfs_trans_handle *trans;
1068 /* we wait first, since the writeback may change the inode */
1070 /* the VFS called filemap_fdatawrite for us */
1071 btrfs_wait_ordered_range(inode, 0, (u64)-1);
1075 * check the transaction that last modified this inode
1076 * and see if its already been committed
1078 if (!BTRFS_I(inode)->last_trans)
1082 * if the last transaction that changed this file was before
1083 * the current transaction, we can bail out now without any
1086 mutex_lock(&root->fs_info->trans_mutex);
1087 if (BTRFS_I(inode)->last_trans <=
1088 root->fs_info->last_trans_committed) {
1089 BTRFS_I(inode)->last_trans = 0;
1090 mutex_unlock(&root->fs_info->trans_mutex);
1093 mutex_unlock(&root->fs_info->trans_mutex);
1096 * ok we haven't committed the transaction yet, lets do a commit
1098 if (file && file->private_data)
1099 btrfs_ioctl_trans_end(file);
1101 trans = btrfs_start_transaction(root, 1);
1107 ret = btrfs_log_dentry_safe(trans, root, dentry);
1111 /* we've logged all the items and now have a consistent
1112 * version of the file in the log. It is possible that
1113 * someone will come in and modify the file, but that's
1114 * fine because the log is consistent on disk, and we
1115 * have references to all of the file's extents
1117 * It is possible that someone will come in and log the
1118 * file again, but that will end up using the synchronization
1119 * inside btrfs_sync_log to keep things safe.
1121 mutex_unlock(&dentry->d_inode->i_mutex);
1123 if (ret != BTRFS_NO_LOG_SYNC) {
1125 ret = btrfs_commit_transaction(trans, root);
1127 ret = btrfs_sync_log(trans, root);
1129 ret = btrfs_end_transaction(trans, root);
1131 ret = btrfs_commit_transaction(trans, root);
1134 ret = btrfs_end_transaction(trans, root);
1136 mutex_lock(&dentry->d_inode->i_mutex);
1138 return ret > 0 ? -EIO : ret;
1141 static const struct vm_operations_struct btrfs_file_vm_ops = {
1142 .fault = filemap_fault,
1143 .page_mkwrite = btrfs_page_mkwrite,
1146 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
1148 vma->vm_ops = &btrfs_file_vm_ops;
1149 file_accessed(filp);
1153 const struct file_operations btrfs_file_operations = {
1154 .llseek = generic_file_llseek,
1155 .read = do_sync_read,
1156 .aio_read = generic_file_aio_read,
1157 .splice_read = generic_file_splice_read,
1158 .write = btrfs_file_write,
1159 .mmap = btrfs_file_mmap,
1160 .open = generic_file_open,
1161 .release = btrfs_release_file,
1162 .fsync = btrfs_sync_file,
1163 .unlocked_ioctl = btrfs_ioctl,
1164 #ifdef CONFIG_COMPAT
1165 .compat_ioctl = btrfs_ioctl,