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>
34 #include "transaction.h"
35 #include "btrfs_inode.h"
37 #include "print-tree.h"
43 /* simple helper to fault in pages and copy. This should go away
44 * and be replaced with calls into generic code.
46 static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
48 struct page **prepared_pages,
49 const char __user *buf)
53 int offset = pos & (PAGE_CACHE_SIZE - 1);
55 for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
56 size_t count = min_t(size_t,
57 PAGE_CACHE_SIZE - offset, write_bytes);
58 struct page *page = prepared_pages[i];
59 fault_in_pages_readable(buf, count);
61 /* Copy data from userspace to the current page */
63 page_fault = __copy_from_user(page_address(page) + offset,
65 /* Flush processor's dcache for this page */
66 flush_dcache_page(page);
74 return page_fault ? -EFAULT : 0;
78 * unlocks pages after btrfs_file_write is done with them
80 static noinline void btrfs_drop_pages(struct page **pages, size_t num_pages)
83 for (i = 0; i < num_pages; i++) {
86 /* page checked is some magic around finding pages that
87 * have been modified without going through btrfs_set_page_dirty
90 ClearPageChecked(pages[i]);
91 unlock_page(pages[i]);
92 mark_page_accessed(pages[i]);
93 page_cache_release(pages[i]);
98 * after copy_from_user, pages need to be dirtied and we need to make
99 * sure holes are created between the current EOF and the start of
100 * any next extents (if required).
102 * this also makes the decision about creating an inline extent vs
103 * doing real data extents, marking pages dirty and delalloc as required.
105 static noinline int dirty_and_release_pages(struct btrfs_trans_handle *trans,
106 struct btrfs_root *root,
115 struct inode *inode = fdentry(file)->d_inode;
118 u64 end_of_last_block;
119 u64 end_pos = pos + write_bytes;
120 loff_t isize = i_size_read(inode);
122 start_pos = pos & ~((u64)root->sectorsize - 1);
123 num_bytes = (write_bytes + pos - start_pos +
124 root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
126 end_of_last_block = start_pos + num_bytes - 1;
127 btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block);
128 for (i = 0; i < num_pages; i++) {
129 struct page *p = pages[i];
134 if (end_pos > isize) {
135 i_size_write(inode, end_pos);
136 /* we've only changed i_size in ram, and we haven't updated
137 * the disk i_size. There is no need to log the inode
145 * this drops all the extents in the cache that intersect the range
146 * [start, end]. Existing extents are split as required.
148 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
151 struct extent_map *em;
152 struct extent_map *split = NULL;
153 struct extent_map *split2 = NULL;
154 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
155 u64 len = end - start + 1;
161 WARN_ON(end < start);
162 if (end == (u64)-1) {
168 split = alloc_extent_map(GFP_NOFS);
170 split2 = alloc_extent_map(GFP_NOFS);
172 write_lock(&em_tree->lock);
173 em = lookup_extent_mapping(em_tree, start, len);
175 write_unlock(&em_tree->lock);
179 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
180 write_unlock(&em_tree->lock);
181 if (em->start <= start &&
182 (!testend || em->start + em->len >= start + len)) {
186 if (start < em->start) {
187 len = em->start - start;
189 len = start + len - (em->start + em->len);
190 start = em->start + em->len;
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)
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;
297 btrfs_drop_extent_cache(inode, start, end - 1, 0);
299 path = btrfs_alloc_path();
304 btrfs_release_path(root, path);
305 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
310 if (path->slots[0] == 0) {
324 leaf = path->nodes[0];
325 slot = path->slots[0];
327 btrfs_item_key_to_cpu(leaf, &key, slot);
328 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY &&
332 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
333 key.objectid != inode->i_ino) {
337 search_start = max(key.offset, start);
340 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
341 extent = btrfs_item_ptr(leaf, slot,
342 struct btrfs_file_extent_item);
343 found_type = btrfs_file_extent_type(leaf, extent);
344 compression = btrfs_file_extent_compression(leaf,
346 encryption = btrfs_file_extent_encryption(leaf,
348 other_encoding = btrfs_file_extent_other_encoding(leaf,
350 if (found_type == BTRFS_FILE_EXTENT_REG ||
351 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
353 btrfs_file_extent_disk_bytenr(leaf,
356 *hint_byte = extent_end;
358 extent_end = key.offset +
359 btrfs_file_extent_num_bytes(leaf, extent);
360 ram_bytes = btrfs_file_extent_ram_bytes(leaf,
363 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
365 extent_end = key.offset +
366 btrfs_file_extent_inline_len(leaf, extent);
369 extent_end = search_start;
372 /* we found nothing we can drop */
373 if ((!found_extent && !found_inline) ||
374 search_start >= extent_end) {
377 nritems = btrfs_header_nritems(leaf);
378 if (slot >= nritems - 1) {
379 nextret = btrfs_next_leaf(root, path);
389 if (end <= extent_end && start >= key.offset && found_inline)
390 *hint_byte = EXTENT_MAP_INLINE;
393 read_extent_buffer(leaf, &old, (unsigned long)extent,
397 if (end < extent_end && end >= key.offset) {
399 if (found_inline && start <= key.offset)
403 if (bookend && found_extent) {
404 if (locked_end < extent_end) {
405 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
406 locked_end, extent_end - 1,
409 btrfs_release_path(root, path);
410 lock_extent(&BTRFS_I(inode)->io_tree,
411 locked_end, extent_end - 1,
413 locked_end = extent_end;
416 locked_end = extent_end;
418 disk_bytenr = le64_to_cpu(old.disk_bytenr);
419 if (disk_bytenr != 0) {
420 ret = btrfs_inc_extent_ref(trans, root,
422 le64_to_cpu(old.disk_num_bytes), 0,
423 root->root_key.objectid,
424 key.objectid, key.offset -
425 le64_to_cpu(old.offset));
431 u64 mask = root->sectorsize - 1;
432 search_start = (extent_end + mask) & ~mask;
434 search_start = extent_end;
436 /* truncate existing extent */
437 if (start > key.offset) {
441 WARN_ON(start & (root->sectorsize - 1));
443 new_num = start - key.offset;
444 old_num = btrfs_file_extent_num_bytes(leaf,
447 btrfs_file_extent_disk_bytenr(leaf,
449 if (btrfs_file_extent_disk_bytenr(leaf,
451 inode_sub_bytes(inode, old_num -
454 btrfs_set_file_extent_num_bytes(leaf,
456 btrfs_mark_buffer_dirty(leaf);
457 } else if (key.offset < inline_limit &&
458 (end > extent_end) &&
459 (inline_limit < extent_end)) {
461 new_size = btrfs_file_extent_calc_inline_size(
462 inline_limit - key.offset);
463 inode_sub_bytes(inode, extent_end -
465 btrfs_set_file_extent_ram_bytes(leaf, extent,
467 if (!compression && !encryption) {
468 btrfs_truncate_item(trans, root, path,
473 /* delete the entire extent */
476 inode_sub_bytes(inode, extent_end -
478 ret = btrfs_del_item(trans, root, path);
479 /* TODO update progress marker and return */
482 btrfs_release_path(root, path);
483 /* the extent will be freed later */
485 if (bookend && found_inline && start <= key.offset) {
487 new_size = btrfs_file_extent_calc_inline_size(
489 inode_sub_bytes(inode, end - key.offset);
490 btrfs_set_file_extent_ram_bytes(leaf, extent,
492 if (!compression && !encryption)
493 ret = btrfs_truncate_item(trans, root, path,
497 /* create bookend, splitting the extent in two */
498 if (bookend && found_extent) {
499 struct btrfs_key ins;
500 ins.objectid = inode->i_ino;
502 btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
504 btrfs_release_path(root, path);
505 path->leave_spinning = 1;
506 ret = btrfs_insert_empty_item(trans, root, path, &ins,
510 leaf = path->nodes[0];
511 extent = btrfs_item_ptr(leaf, path->slots[0],
512 struct btrfs_file_extent_item);
513 write_extent_buffer(leaf, &old,
514 (unsigned long)extent, sizeof(old));
516 btrfs_set_file_extent_compression(leaf, extent,
518 btrfs_set_file_extent_encryption(leaf, extent,
520 btrfs_set_file_extent_other_encoding(leaf, extent,
522 btrfs_set_file_extent_offset(leaf, extent,
523 le64_to_cpu(old.offset) + end - key.offset);
524 WARN_ON(le64_to_cpu(old.num_bytes) <
526 btrfs_set_file_extent_num_bytes(leaf, extent,
530 * set the ram bytes to the size of the full extent
531 * before splitting. This is a worst case flag,
532 * but its the best we can do because we don't know
533 * how splitting affects compression
535 btrfs_set_file_extent_ram_bytes(leaf, extent,
537 btrfs_set_file_extent_type(leaf, extent, found_type);
539 btrfs_unlock_up_safe(path, 1);
540 btrfs_mark_buffer_dirty(path->nodes[0]);
541 btrfs_set_lock_blocking(path->nodes[0]);
543 path->leave_spinning = 0;
544 btrfs_release_path(root, path);
545 if (disk_bytenr != 0)
546 inode_add_bytes(inode, extent_end - end);
549 if (found_extent && !keep) {
550 u64 old_disk_bytenr = le64_to_cpu(old.disk_bytenr);
552 if (old_disk_bytenr != 0) {
553 inode_sub_bytes(inode,
554 le64_to_cpu(old.num_bytes));
555 ret = btrfs_free_extent(trans, root,
557 le64_to_cpu(old.disk_num_bytes),
558 0, root->root_key.objectid,
559 key.objectid, key.offset -
560 le64_to_cpu(old.offset));
562 *hint_byte = old_disk_bytenr;
566 if (search_start >= end) {
572 btrfs_free_path(path);
573 if (locked_end > orig_locked_end) {
574 unlock_extent(&BTRFS_I(inode)->io_tree, orig_locked_end,
575 locked_end - 1, GFP_NOFS);
580 static int extent_mergeable(struct extent_buffer *leaf, int slot,
581 u64 objectid, u64 bytenr, u64 *start, u64 *end)
583 struct btrfs_file_extent_item *fi;
584 struct btrfs_key key;
587 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
590 btrfs_item_key_to_cpu(leaf, &key, slot);
591 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
594 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
595 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
596 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
597 btrfs_file_extent_compression(leaf, fi) ||
598 btrfs_file_extent_encryption(leaf, fi) ||
599 btrfs_file_extent_other_encoding(leaf, fi))
602 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
603 if ((*start && *start != key.offset) || (*end && *end != extent_end))
612 * Mark extent in the range start - end as written.
614 * This changes extent type from 'pre-allocated' to 'regular'. If only
615 * part of extent is marked as written, the extent will be split into
618 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
619 struct btrfs_root *root,
620 struct inode *inode, u64 start, u64 end)
622 struct extent_buffer *leaf;
623 struct btrfs_path *path;
624 struct btrfs_file_extent_item *fi;
625 struct btrfs_key key;
633 u64 locked_end = end;
638 btrfs_drop_extent_cache(inode, start, end - 1, 0);
640 path = btrfs_alloc_path();
643 key.objectid = inode->i_ino;
644 key.type = BTRFS_EXTENT_DATA_KEY;
648 key.offset = split - 1;
650 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
651 if (ret > 0 && path->slots[0] > 0)
654 leaf = path->nodes[0];
655 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
656 BUG_ON(key.objectid != inode->i_ino ||
657 key.type != BTRFS_EXTENT_DATA_KEY);
658 fi = btrfs_item_ptr(leaf, path->slots[0],
659 struct btrfs_file_extent_item);
660 extent_type = btrfs_file_extent_type(leaf, fi);
661 BUG_ON(extent_type != BTRFS_FILE_EXTENT_PREALLOC);
662 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
663 BUG_ON(key.offset > start || extent_end < end);
665 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
666 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
667 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
669 if (key.offset == start)
672 if (key.offset == start && extent_end == end) {
677 if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
678 bytenr, &other_start, &other_end)) {
679 extent_end = other_end;
680 del_slot = path->slots[0] + 1;
682 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
683 0, root->root_key.objectid,
684 inode->i_ino, orig_offset);
689 if (extent_mergeable(leaf, path->slots[0] - 1, inode->i_ino,
690 bytenr, &other_start, &other_end)) {
691 key.offset = other_start;
692 del_slot = path->slots[0];
694 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
695 0, root->root_key.objectid,
696 inode->i_ino, orig_offset);
701 btrfs_set_file_extent_type(leaf, fi,
702 BTRFS_FILE_EXTENT_REG);
706 fi = btrfs_item_ptr(leaf, del_slot - 1,
707 struct btrfs_file_extent_item);
708 btrfs_set_file_extent_type(leaf, fi, BTRFS_FILE_EXTENT_REG);
709 btrfs_set_file_extent_num_bytes(leaf, fi,
710 extent_end - key.offset);
711 btrfs_mark_buffer_dirty(leaf);
713 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
716 } else if (split == start) {
717 if (locked_end < extent_end) {
718 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
719 locked_end, extent_end - 1, GFP_NOFS);
721 btrfs_release_path(root, path);
722 lock_extent(&BTRFS_I(inode)->io_tree,
723 locked_end, extent_end - 1, GFP_NOFS);
724 locked_end = extent_end;
727 locked_end = extent_end;
729 btrfs_set_file_extent_num_bytes(leaf, fi, split - key.offset);
731 BUG_ON(key.offset != start);
733 btrfs_set_file_extent_offset(leaf, fi, key.offset -
735 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - split);
736 btrfs_set_item_key_safe(trans, root, path, &key);
740 if (extent_end == end) {
742 extent_type = BTRFS_FILE_EXTENT_REG;
744 if (extent_end == end && split == start) {
747 if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
748 bytenr, &other_start, &other_end)) {
750 fi = btrfs_item_ptr(leaf, path->slots[0],
751 struct btrfs_file_extent_item);
753 btrfs_set_item_key_safe(trans, root, path, &key);
754 btrfs_set_file_extent_offset(leaf, fi, key.offset -
756 btrfs_set_file_extent_num_bytes(leaf, fi,
761 if (extent_end == end && split == end) {
764 if (extent_mergeable(leaf, path->slots[0] - 1 , inode->i_ino,
765 bytenr, &other_start, &other_end)) {
767 fi = btrfs_item_ptr(leaf, path->slots[0],
768 struct btrfs_file_extent_item);
769 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end -
775 btrfs_mark_buffer_dirty(leaf);
777 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
778 root->root_key.objectid,
779 inode->i_ino, orig_offset);
781 btrfs_release_path(root, path);
784 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*fi));
787 leaf = path->nodes[0];
788 fi = btrfs_item_ptr(leaf, path->slots[0],
789 struct btrfs_file_extent_item);
790 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
791 btrfs_set_file_extent_type(leaf, fi, extent_type);
792 btrfs_set_file_extent_disk_bytenr(leaf, fi, bytenr);
793 btrfs_set_file_extent_disk_num_bytes(leaf, fi, num_bytes);
794 btrfs_set_file_extent_offset(leaf, fi, key.offset - orig_offset);
795 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - key.offset);
796 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
797 btrfs_set_file_extent_compression(leaf, fi, 0);
798 btrfs_set_file_extent_encryption(leaf, fi, 0);
799 btrfs_set_file_extent_other_encoding(leaf, fi, 0);
801 btrfs_mark_buffer_dirty(leaf);
804 btrfs_release_path(root, path);
805 if (split_end && split == start) {
809 if (locked_end > end) {
810 unlock_extent(&BTRFS_I(inode)->io_tree, end, locked_end - 1,
813 btrfs_free_path(path);
818 * this gets pages into the page cache and locks them down, it also properly
819 * waits for data=ordered extents to finish before allowing the pages to be
822 static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
823 struct page **pages, size_t num_pages,
824 loff_t pos, unsigned long first_index,
825 unsigned long last_index, size_t write_bytes)
828 unsigned long index = pos >> PAGE_CACHE_SHIFT;
829 struct inode *inode = fdentry(file)->d_inode;
834 start_pos = pos & ~((u64)root->sectorsize - 1);
835 last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
837 if (start_pos > inode->i_size) {
838 err = btrfs_cont_expand(inode, start_pos);
843 memset(pages, 0, num_pages * sizeof(struct page *));
845 for (i = 0; i < num_pages; i++) {
846 pages[i] = grab_cache_page(inode->i_mapping, index + i);
851 wait_on_page_writeback(pages[i]);
853 if (start_pos < inode->i_size) {
854 struct btrfs_ordered_extent *ordered;
855 lock_extent(&BTRFS_I(inode)->io_tree,
856 start_pos, last_pos - 1, GFP_NOFS);
857 ordered = btrfs_lookup_first_ordered_extent(inode,
860 ordered->file_offset + ordered->len > start_pos &&
861 ordered->file_offset < last_pos) {
862 btrfs_put_ordered_extent(ordered);
863 unlock_extent(&BTRFS_I(inode)->io_tree,
864 start_pos, last_pos - 1, GFP_NOFS);
865 for (i = 0; i < num_pages; i++) {
866 unlock_page(pages[i]);
867 page_cache_release(pages[i]);
869 btrfs_wait_ordered_range(inode, start_pos,
870 last_pos - start_pos);
874 btrfs_put_ordered_extent(ordered);
876 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
877 last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC,
879 unlock_extent(&BTRFS_I(inode)->io_tree,
880 start_pos, last_pos - 1, GFP_NOFS);
882 for (i = 0; i < num_pages; i++) {
883 clear_page_dirty_for_io(pages[i]);
884 set_page_extent_mapped(pages[i]);
885 WARN_ON(!PageLocked(pages[i]));
890 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
891 size_t count, loff_t *ppos)
895 ssize_t num_written = 0;
898 struct inode *inode = fdentry(file)->d_inode;
899 struct btrfs_root *root = BTRFS_I(inode)->root;
900 struct page **pages = NULL;
902 struct page *pinned[2];
903 unsigned long first_index;
904 unsigned long last_index;
907 will_write = ((file->f_flags & O_SYNC) || IS_SYNC(inode) ||
908 (file->f_flags & O_DIRECT));
910 nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
911 PAGE_CACHE_SIZE / (sizeof(struct page *)));
918 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
919 current->backing_dev_info = inode->i_mapping->backing_dev_info;
920 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
926 err = file_remove_suid(file);
929 file_update_time(file);
931 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
933 mutex_lock(&inode->i_mutex);
934 BTRFS_I(inode)->sequence++;
935 first_index = pos >> PAGE_CACHE_SHIFT;
936 last_index = (pos + count) >> PAGE_CACHE_SHIFT;
939 * there are lots of better ways to do this, but this code
940 * makes sure the first and last page in the file range are
941 * up to date and ready for cow
943 if ((pos & (PAGE_CACHE_SIZE - 1))) {
944 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
945 if (!PageUptodate(pinned[0])) {
946 ret = btrfs_readpage(NULL, pinned[0]);
948 wait_on_page_locked(pinned[0]);
950 unlock_page(pinned[0]);
953 if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
954 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
955 if (!PageUptodate(pinned[1])) {
956 ret = btrfs_readpage(NULL, pinned[1]);
958 wait_on_page_locked(pinned[1]);
960 unlock_page(pinned[1]);
965 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
966 size_t write_bytes = min(count, nrptrs *
967 (size_t)PAGE_CACHE_SIZE -
969 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
972 WARN_ON(num_pages > nrptrs);
973 memset(pages, 0, sizeof(struct page *) * nrptrs);
975 ret = btrfs_check_data_free_space(root, inode, write_bytes);
979 ret = prepare_pages(root, file, pages, num_pages,
980 pos, first_index, last_index,
983 btrfs_free_reserved_data_space(root, inode,
988 ret = btrfs_copy_from_user(pos, num_pages,
989 write_bytes, pages, buf);
991 btrfs_free_reserved_data_space(root, inode,
993 btrfs_drop_pages(pages, num_pages);
997 ret = dirty_and_release_pages(NULL, root, file, pages,
998 num_pages, pos, write_bytes);
999 btrfs_drop_pages(pages, num_pages);
1001 btrfs_free_reserved_data_space(root, inode,
1007 btrfs_fdatawrite_range(inode->i_mapping, pos,
1008 pos + write_bytes - 1,
1011 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
1014 (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1015 btrfs_btree_balance_dirty(root, 1);
1016 btrfs_throttle(root);
1020 count -= write_bytes;
1022 num_written += write_bytes;
1027 mutex_unlock(&inode->i_mutex);
1034 page_cache_release(pinned[0]);
1036 page_cache_release(pinned[1]);
1040 * we want to make sure fsync finds this change
1041 * but we haven't joined a transaction running right now.
1043 * Later on, someone is sure to update the inode and get the
1044 * real transid recorded.
1046 * We set last_trans now to the fs_info generation + 1,
1047 * this will either be one more than the running transaction
1048 * or the generation used for the next transaction if there isn't
1049 * one running right now.
1051 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1053 if (num_written > 0 && will_write) {
1054 struct btrfs_trans_handle *trans;
1056 err = btrfs_wait_ordered_range(inode, start_pos, num_written);
1060 if ((file->f_flags & O_SYNC) || IS_SYNC(inode)) {
1061 trans = btrfs_start_transaction(root, 1);
1062 ret = btrfs_log_dentry_safe(trans, root,
1065 ret = btrfs_sync_log(trans, root);
1067 btrfs_end_transaction(trans, root);
1069 btrfs_commit_transaction(trans, root);
1071 btrfs_commit_transaction(trans, root);
1074 if (file->f_flags & O_DIRECT) {
1075 invalidate_mapping_pages(inode->i_mapping,
1076 start_pos >> PAGE_CACHE_SHIFT,
1077 (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
1080 current->backing_dev_info = NULL;
1081 return num_written ? num_written : err;
1084 int btrfs_release_file(struct inode *inode, struct file *filp)
1087 * ordered_data_close is set by settattr when we are about to truncate
1088 * a file from a non-zero size to a zero size. This tries to
1089 * flush down new bytes that may have been written if the
1090 * application were using truncate to replace a file in place.
1092 if (BTRFS_I(inode)->ordered_data_close) {
1093 BTRFS_I(inode)->ordered_data_close = 0;
1094 btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode);
1095 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1096 filemap_flush(inode->i_mapping);
1098 if (filp->private_data)
1099 btrfs_ioctl_trans_end(filp);
1104 * fsync call for both files and directories. This logs the inode into
1105 * the tree log instead of forcing full commits whenever possible.
1107 * It needs to call filemap_fdatawait so that all ordered extent updates are
1108 * in the metadata btree are up to date for copying to the log.
1110 * It drops the inode mutex before doing the tree log commit. This is an
1111 * important optimization for directories because holding the mutex prevents
1112 * new operations on the dir while we write to disk.
1114 int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
1116 struct inode *inode = dentry->d_inode;
1117 struct btrfs_root *root = BTRFS_I(inode)->root;
1119 struct btrfs_trans_handle *trans;
1122 * check the transaction that last modified this inode
1123 * and see if its already been committed
1125 if (!BTRFS_I(inode)->last_trans)
1128 mutex_lock(&root->fs_info->trans_mutex);
1129 if (BTRFS_I(inode)->last_trans <=
1130 root->fs_info->last_trans_committed) {
1131 BTRFS_I(inode)->last_trans = 0;
1132 mutex_unlock(&root->fs_info->trans_mutex);
1135 mutex_unlock(&root->fs_info->trans_mutex);
1138 filemap_fdatawrite(inode->i_mapping);
1139 btrfs_wait_ordered_range(inode, 0, (u64)-1);
1142 if (datasync && !(inode->i_state & I_DIRTY_PAGES))
1145 * ok we haven't committed the transaction yet, lets do a commit
1147 if (file && file->private_data)
1148 btrfs_ioctl_trans_end(file);
1150 trans = btrfs_start_transaction(root, 1);
1156 ret = btrfs_log_dentry_safe(trans, root, dentry);
1160 /* we've logged all the items and now have a consistent
1161 * version of the file in the log. It is possible that
1162 * someone will come in and modify the file, but that's
1163 * fine because the log is consistent on disk, and we
1164 * have references to all of the file's extents
1166 * It is possible that someone will come in and log the
1167 * file again, but that will end up using the synchronization
1168 * inside btrfs_sync_log to keep things safe.
1170 mutex_unlock(&dentry->d_inode->i_mutex);
1173 ret = btrfs_commit_transaction(trans, root);
1175 ret = btrfs_sync_log(trans, root);
1177 ret = btrfs_end_transaction(trans, root);
1179 ret = btrfs_commit_transaction(trans, root);
1181 mutex_lock(&dentry->d_inode->i_mutex);
1183 return ret > 0 ? EIO : ret;
1186 static struct vm_operations_struct btrfs_file_vm_ops = {
1187 .fault = filemap_fault,
1188 .page_mkwrite = btrfs_page_mkwrite,
1191 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
1193 vma->vm_ops = &btrfs_file_vm_ops;
1194 file_accessed(filp);
1198 struct file_operations btrfs_file_operations = {
1199 .llseek = generic_file_llseek,
1200 .read = do_sync_read,
1201 .aio_read = generic_file_aio_read,
1202 .splice_read = generic_file_splice_read,
1203 .write = btrfs_file_write,
1204 .mmap = btrfs_file_mmap,
1205 .open = generic_file_open,
1206 .release = btrfs_release_file,
1207 .fsync = btrfs_sync_file,
1208 .unlocked_ioctl = btrfs_ioctl,
1209 #ifdef CONFIG_COMPAT
1210 .compat_ioctl = btrfs_ioctl,