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 btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block);
127 for (i = 0; i < num_pages; i++) {
128 struct page *p = pages[i];
133 if (end_pos > isize) {
134 i_size_write(inode, end_pos);
135 /* we've only changed i_size in ram, and we haven't updated
136 * the disk i_size. There is no need to log the inode
144 * this drops all the extents in the cache that intersect the range
145 * [start, end]. Existing extents are split as required.
147 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
150 struct extent_map *em;
151 struct extent_map *split = NULL;
152 struct extent_map *split2 = NULL;
153 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
154 u64 len = end - start + 1;
160 WARN_ON(end < start);
161 if (end == (u64)-1) {
167 split = alloc_extent_map(GFP_NOFS);
169 split2 = alloc_extent_map(GFP_NOFS);
171 write_lock(&em_tree->lock);
172 em = lookup_extent_mapping(em_tree, start, len);
174 write_unlock(&em_tree->lock);
178 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
179 if (em->start <= start &&
180 (!testend || em->start + em->len >= start + len)) {
182 write_unlock(&em_tree->lock);
185 if (start < em->start) {
186 len = em->start - start;
188 len = start + len - (em->start + em->len);
189 start = em->start + em->len;
192 write_unlock(&em_tree->lock);
195 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
196 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
197 remove_extent_mapping(em_tree, em);
199 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
201 split->start = em->start;
202 split->len = start - em->start;
203 split->orig_start = em->orig_start;
204 split->block_start = em->block_start;
207 split->block_len = em->block_len;
209 split->block_len = split->len;
211 split->bdev = em->bdev;
212 split->flags = flags;
213 ret = add_extent_mapping(em_tree, split);
215 free_extent_map(split);
219 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
220 testend && em->start + em->len > start + len) {
221 u64 diff = start + len - em->start;
223 split->start = start + len;
224 split->len = em->start + em->len - (start + len);
225 split->bdev = em->bdev;
226 split->flags = flags;
229 split->block_len = em->block_len;
230 split->block_start = em->block_start;
231 split->orig_start = em->orig_start;
233 split->block_len = split->len;
234 split->block_start = em->block_start + diff;
235 split->orig_start = split->start;
238 ret = add_extent_mapping(em_tree, split);
240 free_extent_map(split);
243 write_unlock(&em_tree->lock);
247 /* once for the tree*/
251 free_extent_map(split);
253 free_extent_map(split2);
258 * this is very complex, but the basic idea is to drop all extents
259 * in the range start - end. hint_block is filled in with a block number
260 * that would be a good hint to the block allocator for this file.
262 * If an extent intersects the range but is not entirely inside the range
263 * it is either truncated or split. Anything entirely inside the range
264 * is deleted from the tree.
266 * inline_limit is used to tell this code which offsets in the file to keep
267 * if they contain inline extents.
269 noinline int btrfs_drop_extents(struct btrfs_trans_handle *trans,
270 struct btrfs_root *root, struct inode *inode,
271 u64 start, u64 end, u64 locked_end,
272 u64 inline_limit, u64 *hint_byte, int drop_cache)
275 u64 search_start = start;
278 u64 orig_locked_end = locked_end;
281 u16 other_encoding = 0;
282 struct extent_buffer *leaf;
283 struct btrfs_file_extent_item *extent;
284 struct btrfs_path *path;
285 struct btrfs_key key;
286 struct btrfs_file_extent_item old;
298 btrfs_drop_extent_cache(inode, start, end - 1, 0);
300 path = btrfs_alloc_path();
305 btrfs_release_path(root, path);
306 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
311 if (path->slots[0] == 0) {
325 leaf = path->nodes[0];
326 slot = path->slots[0];
328 btrfs_item_key_to_cpu(leaf, &key, slot);
329 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY &&
333 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
334 key.objectid != inode->i_ino) {
338 search_start = max(key.offset, start);
341 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
342 extent = btrfs_item_ptr(leaf, slot,
343 struct btrfs_file_extent_item);
344 found_type = btrfs_file_extent_type(leaf, extent);
345 compression = btrfs_file_extent_compression(leaf,
347 encryption = btrfs_file_extent_encryption(leaf,
349 other_encoding = btrfs_file_extent_other_encoding(leaf,
351 if (found_type == BTRFS_FILE_EXTENT_REG ||
352 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
354 btrfs_file_extent_disk_bytenr(leaf,
357 *hint_byte = extent_end;
359 extent_end = key.offset +
360 btrfs_file_extent_num_bytes(leaf, extent);
361 ram_bytes = btrfs_file_extent_ram_bytes(leaf,
364 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
366 extent_end = key.offset +
367 btrfs_file_extent_inline_len(leaf, extent);
370 extent_end = search_start;
373 /* we found nothing we can drop */
374 if ((!found_extent && !found_inline) ||
375 search_start >= extent_end) {
378 nritems = btrfs_header_nritems(leaf);
379 if (slot >= nritems - 1) {
380 nextret = btrfs_next_leaf(root, path);
390 if (end <= extent_end && start >= key.offset && found_inline)
391 *hint_byte = EXTENT_MAP_INLINE;
394 read_extent_buffer(leaf, &old, (unsigned long)extent,
398 if (end < extent_end && end >= key.offset) {
400 if (found_inline && start <= key.offset)
404 if (bookend && found_extent) {
405 if (locked_end < extent_end) {
406 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
407 locked_end, extent_end - 1,
410 btrfs_release_path(root, path);
411 lock_extent(&BTRFS_I(inode)->io_tree,
412 locked_end, extent_end - 1,
414 locked_end = extent_end;
417 locked_end = extent_end;
419 disk_bytenr = le64_to_cpu(old.disk_bytenr);
420 if (disk_bytenr != 0) {
421 ret = btrfs_inc_extent_ref(trans, root,
423 le64_to_cpu(old.disk_num_bytes), 0,
424 root->root_key.objectid,
425 key.objectid, key.offset -
426 le64_to_cpu(old.offset));
432 u64 mask = root->sectorsize - 1;
433 search_start = (extent_end + mask) & ~mask;
435 search_start = extent_end;
437 /* truncate existing extent */
438 if (start > key.offset) {
442 WARN_ON(start & (root->sectorsize - 1));
444 new_num = start - key.offset;
445 old_num = btrfs_file_extent_num_bytes(leaf,
448 btrfs_file_extent_disk_bytenr(leaf,
450 if (btrfs_file_extent_disk_bytenr(leaf,
452 inode_sub_bytes(inode, old_num -
455 btrfs_set_file_extent_num_bytes(leaf,
457 btrfs_mark_buffer_dirty(leaf);
458 } else if (key.offset < inline_limit &&
459 (end > extent_end) &&
460 (inline_limit < extent_end)) {
462 new_size = btrfs_file_extent_calc_inline_size(
463 inline_limit - key.offset);
464 inode_sub_bytes(inode, extent_end -
466 btrfs_set_file_extent_ram_bytes(leaf, extent,
468 if (!compression && !encryption) {
469 btrfs_truncate_item(trans, root, path,
474 /* delete the entire extent */
477 inode_sub_bytes(inode, extent_end -
479 ret = btrfs_del_item(trans, root, path);
480 /* TODO update progress marker and return */
483 btrfs_release_path(root, path);
484 /* the extent will be freed later */
486 if (bookend && found_inline && start <= key.offset) {
488 new_size = btrfs_file_extent_calc_inline_size(
490 inode_sub_bytes(inode, end - key.offset);
491 btrfs_set_file_extent_ram_bytes(leaf, extent,
493 if (!compression && !encryption)
494 ret = btrfs_truncate_item(trans, root, path,
498 /* create bookend, splitting the extent in two */
499 if (bookend && found_extent) {
500 struct btrfs_key ins;
501 ins.objectid = inode->i_ino;
503 btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
505 btrfs_release_path(root, path);
506 path->leave_spinning = 1;
507 ret = btrfs_insert_empty_item(trans, root, path, &ins,
511 leaf = path->nodes[0];
512 extent = btrfs_item_ptr(leaf, path->slots[0],
513 struct btrfs_file_extent_item);
514 write_extent_buffer(leaf, &old,
515 (unsigned long)extent, sizeof(old));
517 btrfs_set_file_extent_compression(leaf, extent,
519 btrfs_set_file_extent_encryption(leaf, extent,
521 btrfs_set_file_extent_other_encoding(leaf, extent,
523 btrfs_set_file_extent_offset(leaf, extent,
524 le64_to_cpu(old.offset) + end - key.offset);
525 WARN_ON(le64_to_cpu(old.num_bytes) <
527 btrfs_set_file_extent_num_bytes(leaf, extent,
531 * set the ram bytes to the size of the full extent
532 * before splitting. This is a worst case flag,
533 * but its the best we can do because we don't know
534 * how splitting affects compression
536 btrfs_set_file_extent_ram_bytes(leaf, extent,
538 btrfs_set_file_extent_type(leaf, extent, found_type);
540 btrfs_unlock_up_safe(path, 1);
541 btrfs_mark_buffer_dirty(path->nodes[0]);
542 btrfs_set_lock_blocking(path->nodes[0]);
544 path->leave_spinning = 0;
545 btrfs_release_path(root, path);
546 if (disk_bytenr != 0)
547 inode_add_bytes(inode, extent_end - end);
550 if (found_extent && !keep) {
551 u64 old_disk_bytenr = le64_to_cpu(old.disk_bytenr);
553 if (old_disk_bytenr != 0) {
554 inode_sub_bytes(inode,
555 le64_to_cpu(old.num_bytes));
556 ret = btrfs_free_extent(trans, root,
558 le64_to_cpu(old.disk_num_bytes),
559 0, root->root_key.objectid,
560 key.objectid, key.offset -
561 le64_to_cpu(old.offset));
563 *hint_byte = old_disk_bytenr;
567 if (search_start >= end) {
573 btrfs_free_path(path);
574 if (locked_end > orig_locked_end) {
575 unlock_extent(&BTRFS_I(inode)->io_tree, orig_locked_end,
576 locked_end - 1, GFP_NOFS);
581 static int extent_mergeable(struct extent_buffer *leaf, int slot,
582 u64 objectid, u64 bytenr, u64 *start, u64 *end)
584 struct btrfs_file_extent_item *fi;
585 struct btrfs_key key;
588 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
591 btrfs_item_key_to_cpu(leaf, &key, slot);
592 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
595 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
596 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
597 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
598 btrfs_file_extent_compression(leaf, fi) ||
599 btrfs_file_extent_encryption(leaf, fi) ||
600 btrfs_file_extent_other_encoding(leaf, fi))
603 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
604 if ((*start && *start != key.offset) || (*end && *end != extent_end))
613 * Mark extent in the range start - end as written.
615 * This changes extent type from 'pre-allocated' to 'regular'. If only
616 * part of extent is marked as written, the extent will be split into
619 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
620 struct btrfs_root *root,
621 struct inode *inode, u64 start, u64 end)
623 struct extent_buffer *leaf;
624 struct btrfs_path *path;
625 struct btrfs_file_extent_item *fi;
626 struct btrfs_key key;
634 u64 locked_end = end;
639 btrfs_drop_extent_cache(inode, start, end - 1, 0);
641 path = btrfs_alloc_path();
644 key.objectid = inode->i_ino;
645 key.type = BTRFS_EXTENT_DATA_KEY;
649 key.offset = split - 1;
651 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
652 if (ret > 0 && path->slots[0] > 0)
655 leaf = path->nodes[0];
656 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
657 BUG_ON(key.objectid != inode->i_ino ||
658 key.type != BTRFS_EXTENT_DATA_KEY);
659 fi = btrfs_item_ptr(leaf, path->slots[0],
660 struct btrfs_file_extent_item);
661 extent_type = btrfs_file_extent_type(leaf, fi);
662 BUG_ON(extent_type != BTRFS_FILE_EXTENT_PREALLOC);
663 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
664 BUG_ON(key.offset > start || extent_end < end);
666 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
667 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
668 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
670 if (key.offset == start)
673 if (key.offset == start && extent_end == end) {
678 if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
679 bytenr, &other_start, &other_end)) {
680 extent_end = other_end;
681 del_slot = path->slots[0] + 1;
683 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
684 0, root->root_key.objectid,
685 inode->i_ino, orig_offset);
690 if (extent_mergeable(leaf, path->slots[0] - 1, inode->i_ino,
691 bytenr, &other_start, &other_end)) {
692 key.offset = other_start;
693 del_slot = path->slots[0];
695 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
696 0, root->root_key.objectid,
697 inode->i_ino, orig_offset);
702 btrfs_set_file_extent_type(leaf, fi,
703 BTRFS_FILE_EXTENT_REG);
707 fi = btrfs_item_ptr(leaf, del_slot - 1,
708 struct btrfs_file_extent_item);
709 btrfs_set_file_extent_type(leaf, fi, BTRFS_FILE_EXTENT_REG);
710 btrfs_set_file_extent_num_bytes(leaf, fi,
711 extent_end - key.offset);
712 btrfs_mark_buffer_dirty(leaf);
714 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
717 } else if (split == start) {
718 if (locked_end < extent_end) {
719 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
720 locked_end, extent_end - 1, GFP_NOFS);
722 btrfs_release_path(root, path);
723 lock_extent(&BTRFS_I(inode)->io_tree,
724 locked_end, extent_end - 1, GFP_NOFS);
725 locked_end = extent_end;
728 locked_end = extent_end;
730 btrfs_set_file_extent_num_bytes(leaf, fi, split - key.offset);
732 BUG_ON(key.offset != start);
734 btrfs_set_file_extent_offset(leaf, fi, key.offset -
736 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - split);
737 btrfs_set_item_key_safe(trans, root, path, &key);
741 if (extent_end == end) {
743 extent_type = BTRFS_FILE_EXTENT_REG;
745 if (extent_end == end && split == start) {
748 if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
749 bytenr, &other_start, &other_end)) {
751 fi = btrfs_item_ptr(leaf, path->slots[0],
752 struct btrfs_file_extent_item);
754 btrfs_set_item_key_safe(trans, root, path, &key);
755 btrfs_set_file_extent_offset(leaf, fi, key.offset -
757 btrfs_set_file_extent_num_bytes(leaf, fi,
762 if (extent_end == end && split == end) {
765 if (extent_mergeable(leaf, path->slots[0] - 1 , inode->i_ino,
766 bytenr, &other_start, &other_end)) {
768 fi = btrfs_item_ptr(leaf, path->slots[0],
769 struct btrfs_file_extent_item);
770 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end -
776 btrfs_mark_buffer_dirty(leaf);
778 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
779 root->root_key.objectid,
780 inode->i_ino, orig_offset);
782 btrfs_release_path(root, path);
785 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*fi));
788 leaf = path->nodes[0];
789 fi = btrfs_item_ptr(leaf, path->slots[0],
790 struct btrfs_file_extent_item);
791 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
792 btrfs_set_file_extent_type(leaf, fi, extent_type);
793 btrfs_set_file_extent_disk_bytenr(leaf, fi, bytenr);
794 btrfs_set_file_extent_disk_num_bytes(leaf, fi, num_bytes);
795 btrfs_set_file_extent_offset(leaf, fi, key.offset - orig_offset);
796 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - key.offset);
797 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
798 btrfs_set_file_extent_compression(leaf, fi, 0);
799 btrfs_set_file_extent_encryption(leaf, fi, 0);
800 btrfs_set_file_extent_other_encoding(leaf, fi, 0);
802 btrfs_mark_buffer_dirty(leaf);
805 btrfs_release_path(root, path);
806 if (split_end && split == start) {
810 if (locked_end > end) {
811 unlock_extent(&BTRFS_I(inode)->io_tree, end, locked_end - 1,
814 btrfs_free_path(path);
819 * this gets pages into the page cache and locks them down, it also properly
820 * waits for data=ordered extents to finish before allowing the pages to be
823 static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
824 struct page **pages, size_t num_pages,
825 loff_t pos, unsigned long first_index,
826 unsigned long last_index, size_t write_bytes)
829 unsigned long index = pos >> PAGE_CACHE_SHIFT;
830 struct inode *inode = fdentry(file)->d_inode;
835 start_pos = pos & ~((u64)root->sectorsize - 1);
836 last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
838 if (start_pos > inode->i_size) {
839 err = btrfs_cont_expand(inode, start_pos);
844 memset(pages, 0, num_pages * sizeof(struct page *));
846 for (i = 0; i < num_pages; i++) {
847 pages[i] = grab_cache_page(inode->i_mapping, index + i);
852 wait_on_page_writeback(pages[i]);
854 if (start_pos < inode->i_size) {
855 struct btrfs_ordered_extent *ordered;
856 lock_extent(&BTRFS_I(inode)->io_tree,
857 start_pos, last_pos - 1, GFP_NOFS);
858 ordered = btrfs_lookup_first_ordered_extent(inode,
861 ordered->file_offset + ordered->len > start_pos &&
862 ordered->file_offset < last_pos) {
863 btrfs_put_ordered_extent(ordered);
864 unlock_extent(&BTRFS_I(inode)->io_tree,
865 start_pos, last_pos - 1, GFP_NOFS);
866 for (i = 0; i < num_pages; i++) {
867 unlock_page(pages[i]);
868 page_cache_release(pages[i]);
870 btrfs_wait_ordered_range(inode, start_pos,
871 last_pos - start_pos);
875 btrfs_put_ordered_extent(ordered);
877 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
878 last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC,
880 unlock_extent(&BTRFS_I(inode)->io_tree,
881 start_pos, last_pos - 1, GFP_NOFS);
883 for (i = 0; i < num_pages; i++) {
884 clear_page_dirty_for_io(pages[i]);
885 set_page_extent_mapped(pages[i]);
886 WARN_ON(!PageLocked(pages[i]));
891 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
892 size_t count, loff_t *ppos)
896 ssize_t num_written = 0;
899 struct inode *inode = fdentry(file)->d_inode;
900 struct btrfs_root *root = BTRFS_I(inode)->root;
901 struct page **pages = NULL;
903 struct page *pinned[2];
904 unsigned long first_index;
905 unsigned long last_index;
908 will_write = ((file->f_flags & O_SYNC) || IS_SYNC(inode) ||
909 (file->f_flags & O_DIRECT));
911 nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
912 PAGE_CACHE_SIZE / (sizeof(struct page *)));
919 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
920 current->backing_dev_info = inode->i_mapping->backing_dev_info;
921 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
927 err = file_remove_suid(file);
930 file_update_time(file);
932 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
934 mutex_lock(&inode->i_mutex);
935 BTRFS_I(inode)->sequence++;
936 first_index = pos >> PAGE_CACHE_SHIFT;
937 last_index = (pos + count) >> PAGE_CACHE_SHIFT;
940 * there are lots of better ways to do this, but this code
941 * makes sure the first and last page in the file range are
942 * up to date and ready for cow
944 if ((pos & (PAGE_CACHE_SIZE - 1))) {
945 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
946 if (!PageUptodate(pinned[0])) {
947 ret = btrfs_readpage(NULL, pinned[0]);
949 wait_on_page_locked(pinned[0]);
951 unlock_page(pinned[0]);
954 if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
955 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
956 if (!PageUptodate(pinned[1])) {
957 ret = btrfs_readpage(NULL, pinned[1]);
959 wait_on_page_locked(pinned[1]);
961 unlock_page(pinned[1]);
966 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
967 size_t write_bytes = min(count, nrptrs *
968 (size_t)PAGE_CACHE_SIZE -
970 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
973 WARN_ON(num_pages > nrptrs);
974 memset(pages, 0, sizeof(struct page *) * nrptrs);
976 ret = btrfs_check_data_free_space(root, inode, write_bytes);
980 ret = prepare_pages(root, file, pages, num_pages,
981 pos, first_index, last_index,
984 btrfs_free_reserved_data_space(root, inode,
989 ret = btrfs_copy_from_user(pos, num_pages,
990 write_bytes, pages, buf);
992 btrfs_free_reserved_data_space(root, inode,
994 btrfs_drop_pages(pages, num_pages);
998 ret = dirty_and_release_pages(NULL, root, file, pages,
999 num_pages, pos, write_bytes);
1000 btrfs_drop_pages(pages, num_pages);
1002 btrfs_free_reserved_data_space(root, inode,
1008 btrfs_fdatawrite_range(inode->i_mapping, pos,
1009 pos + write_bytes - 1,
1012 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
1015 (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1016 btrfs_btree_balance_dirty(root, 1);
1017 btrfs_throttle(root);
1021 count -= write_bytes;
1023 num_written += write_bytes;
1028 mutex_unlock(&inode->i_mutex);
1035 page_cache_release(pinned[0]);
1037 page_cache_release(pinned[1]);
1041 * we want to make sure fsync finds this change
1042 * but we haven't joined a transaction running right now.
1044 * Later on, someone is sure to update the inode and get the
1045 * real transid recorded.
1047 * We set last_trans now to the fs_info generation + 1,
1048 * this will either be one more than the running transaction
1049 * or the generation used for the next transaction if there isn't
1050 * one running right now.
1052 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1054 if (num_written > 0 && will_write) {
1055 struct btrfs_trans_handle *trans;
1057 err = btrfs_wait_ordered_range(inode, start_pos, num_written);
1061 if ((file->f_flags & O_SYNC) || IS_SYNC(inode)) {
1062 trans = btrfs_start_transaction(root, 1);
1063 ret = btrfs_log_dentry_safe(trans, root,
1066 ret = btrfs_sync_log(trans, root);
1068 btrfs_end_transaction(trans, root);
1070 btrfs_commit_transaction(trans, root);
1072 btrfs_commit_transaction(trans, root);
1075 if (file->f_flags & O_DIRECT) {
1076 invalidate_mapping_pages(inode->i_mapping,
1077 start_pos >> PAGE_CACHE_SHIFT,
1078 (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
1081 current->backing_dev_info = NULL;
1082 return num_written ? num_written : err;
1085 int btrfs_release_file(struct inode *inode, struct file *filp)
1088 * ordered_data_close is set by settattr when we are about to truncate
1089 * a file from a non-zero size to a zero size. This tries to
1090 * flush down new bytes that may have been written if the
1091 * application were using truncate to replace a file in place.
1093 if (BTRFS_I(inode)->ordered_data_close) {
1094 BTRFS_I(inode)->ordered_data_close = 0;
1095 btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode);
1096 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1097 filemap_flush(inode->i_mapping);
1099 if (filp->private_data)
1100 btrfs_ioctl_trans_end(filp);
1105 * fsync call for both files and directories. This logs the inode into
1106 * the tree log instead of forcing full commits whenever possible.
1108 * It needs to call filemap_fdatawait so that all ordered extent updates are
1109 * in the metadata btree are up to date for copying to the log.
1111 * It drops the inode mutex before doing the tree log commit. This is an
1112 * important optimization for directories because holding the mutex prevents
1113 * new operations on the dir while we write to disk.
1115 int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
1117 struct inode *inode = dentry->d_inode;
1118 struct btrfs_root *root = BTRFS_I(inode)->root;
1120 struct btrfs_trans_handle *trans;
1123 * check the transaction that last modified this inode
1124 * and see if its already been committed
1126 if (!BTRFS_I(inode)->last_trans)
1129 mutex_lock(&root->fs_info->trans_mutex);
1130 if (BTRFS_I(inode)->last_trans <=
1131 root->fs_info->last_trans_committed) {
1132 BTRFS_I(inode)->last_trans = 0;
1133 mutex_unlock(&root->fs_info->trans_mutex);
1136 mutex_unlock(&root->fs_info->trans_mutex);
1139 filemap_fdatawrite(inode->i_mapping);
1140 btrfs_wait_ordered_range(inode, 0, (u64)-1);
1143 if (datasync && !(inode->i_state & I_DIRTY_PAGES))
1146 * ok we haven't committed the transaction yet, lets do a commit
1148 if (file && file->private_data)
1149 btrfs_ioctl_trans_end(file);
1151 trans = btrfs_start_transaction(root, 1);
1157 ret = btrfs_log_dentry_safe(trans, root, dentry);
1161 /* we've logged all the items and now have a consistent
1162 * version of the file in the log. It is possible that
1163 * someone will come in and modify the file, but that's
1164 * fine because the log is consistent on disk, and we
1165 * have references to all of the file's extents
1167 * It is possible that someone will come in and log the
1168 * file again, but that will end up using the synchronization
1169 * inside btrfs_sync_log to keep things safe.
1171 mutex_unlock(&dentry->d_inode->i_mutex);
1174 ret = btrfs_commit_transaction(trans, root);
1176 ret = btrfs_sync_log(trans, root);
1178 ret = btrfs_end_transaction(trans, root);
1180 ret = btrfs_commit_transaction(trans, root);
1182 mutex_lock(&dentry->d_inode->i_mutex);
1184 return ret > 0 ? EIO : ret;
1187 static struct vm_operations_struct btrfs_file_vm_ops = {
1188 .fault = filemap_fault,
1189 .page_mkwrite = btrfs_page_mkwrite,
1192 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
1194 vma->vm_ops = &btrfs_file_vm_ops;
1195 file_accessed(filp);
1199 struct file_operations btrfs_file_operations = {
1200 .llseek = generic_file_llseek,
1201 .read = do_sync_read,
1202 .aio_read = generic_file_aio_read,
1203 .splice_read = generic_file_splice_read,
1204 .write = btrfs_file_write,
1205 .mmap = btrfs_file_mmap,
1206 .open = generic_file_open,
1207 .release = btrfs_release_file,
1208 .fsync = btrfs_sync_file,
1209 .unlocked_ioctl = btrfs_ioctl,
1210 #ifdef CONFIG_COMPAT
1211 .compat_ioctl = btrfs_ioctl,