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 * inline_limit is used to tell this code which offsets in the file to keep
270 * if they contain inline extents.
272 noinline int btrfs_drop_extents(struct btrfs_trans_handle *trans,
273 struct btrfs_root *root, struct inode *inode,
274 u64 start, u64 end, u64 locked_end,
275 u64 inline_limit, u64 *hint_byte, int drop_cache)
278 u64 search_start = start;
281 u64 orig_locked_end = locked_end;
284 u16 other_encoding = 0;
285 struct extent_buffer *leaf;
286 struct btrfs_file_extent_item *extent;
287 struct btrfs_path *path;
288 struct btrfs_key key;
289 struct btrfs_file_extent_item old;
301 btrfs_drop_extent_cache(inode, start, end - 1, 0);
303 path = btrfs_alloc_path();
308 btrfs_release_path(root, path);
309 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
314 if (path->slots[0] == 0) {
328 leaf = path->nodes[0];
329 slot = path->slots[0];
331 btrfs_item_key_to_cpu(leaf, &key, slot);
332 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY &&
336 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
337 key.objectid != inode->i_ino) {
341 search_start = max(key.offset, start);
344 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
345 extent = btrfs_item_ptr(leaf, slot,
346 struct btrfs_file_extent_item);
347 found_type = btrfs_file_extent_type(leaf, extent);
348 compression = btrfs_file_extent_compression(leaf,
350 encryption = btrfs_file_extent_encryption(leaf,
352 other_encoding = btrfs_file_extent_other_encoding(leaf,
354 if (found_type == BTRFS_FILE_EXTENT_REG ||
355 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
357 btrfs_file_extent_disk_bytenr(leaf,
360 *hint_byte = extent_end;
362 extent_end = key.offset +
363 btrfs_file_extent_num_bytes(leaf, extent);
364 ram_bytes = btrfs_file_extent_ram_bytes(leaf,
367 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
369 extent_end = key.offset +
370 btrfs_file_extent_inline_len(leaf, extent);
373 extent_end = search_start;
376 /* we found nothing we can drop */
377 if ((!found_extent && !found_inline) ||
378 search_start >= extent_end) {
381 nritems = btrfs_header_nritems(leaf);
382 if (slot >= nritems - 1) {
383 nextret = btrfs_next_leaf(root, path);
393 if (end <= extent_end && start >= key.offset && found_inline)
394 *hint_byte = EXTENT_MAP_INLINE;
397 read_extent_buffer(leaf, &old, (unsigned long)extent,
401 if (end < extent_end && end >= key.offset) {
403 if (found_inline && start <= key.offset)
407 if (bookend && found_extent) {
408 if (locked_end < extent_end) {
409 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
410 locked_end, extent_end - 1,
413 btrfs_release_path(root, path);
414 lock_extent(&BTRFS_I(inode)->io_tree,
415 locked_end, extent_end - 1,
417 locked_end = extent_end;
420 locked_end = extent_end;
422 disk_bytenr = le64_to_cpu(old.disk_bytenr);
423 if (disk_bytenr != 0) {
424 ret = btrfs_inc_extent_ref(trans, root,
426 le64_to_cpu(old.disk_num_bytes), 0,
427 root->root_key.objectid,
428 key.objectid, key.offset -
429 le64_to_cpu(old.offset));
435 u64 mask = root->sectorsize - 1;
436 search_start = (extent_end + mask) & ~mask;
438 search_start = extent_end;
440 /* truncate existing extent */
441 if (start > key.offset) {
445 WARN_ON(start & (root->sectorsize - 1));
447 new_num = start - key.offset;
448 old_num = btrfs_file_extent_num_bytes(leaf,
451 btrfs_file_extent_disk_bytenr(leaf,
453 if (btrfs_file_extent_disk_bytenr(leaf,
455 inode_sub_bytes(inode, old_num -
458 btrfs_set_file_extent_num_bytes(leaf,
460 btrfs_mark_buffer_dirty(leaf);
461 } else if (key.offset < inline_limit &&
462 (end > extent_end) &&
463 (inline_limit < extent_end)) {
465 new_size = btrfs_file_extent_calc_inline_size(
466 inline_limit - key.offset);
467 inode_sub_bytes(inode, extent_end -
469 btrfs_set_file_extent_ram_bytes(leaf, extent,
471 if (!compression && !encryption) {
472 btrfs_truncate_item(trans, root, path,
477 /* delete the entire extent */
480 inode_sub_bytes(inode, extent_end -
482 ret = btrfs_del_item(trans, root, path);
483 /* TODO update progress marker and return */
486 btrfs_release_path(root, path);
487 /* the extent will be freed later */
489 if (bookend && found_inline && start <= key.offset) {
491 new_size = btrfs_file_extent_calc_inline_size(
493 inode_sub_bytes(inode, end - key.offset);
494 btrfs_set_file_extent_ram_bytes(leaf, extent,
496 if (!compression && !encryption)
497 ret = btrfs_truncate_item(trans, root, path,
501 /* create bookend, splitting the extent in two */
502 if (bookend && found_extent) {
503 struct btrfs_key ins;
504 ins.objectid = inode->i_ino;
506 btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
508 btrfs_release_path(root, path);
509 path->leave_spinning = 1;
510 ret = btrfs_insert_empty_item(trans, root, path, &ins,
514 leaf = path->nodes[0];
515 extent = btrfs_item_ptr(leaf, path->slots[0],
516 struct btrfs_file_extent_item);
517 write_extent_buffer(leaf, &old,
518 (unsigned long)extent, sizeof(old));
520 btrfs_set_file_extent_compression(leaf, extent,
522 btrfs_set_file_extent_encryption(leaf, extent,
524 btrfs_set_file_extent_other_encoding(leaf, extent,
526 btrfs_set_file_extent_offset(leaf, extent,
527 le64_to_cpu(old.offset) + end - key.offset);
528 WARN_ON(le64_to_cpu(old.num_bytes) <
530 btrfs_set_file_extent_num_bytes(leaf, extent,
534 * set the ram bytes to the size of the full extent
535 * before splitting. This is a worst case flag,
536 * but its the best we can do because we don't know
537 * how splitting affects compression
539 btrfs_set_file_extent_ram_bytes(leaf, extent,
541 btrfs_set_file_extent_type(leaf, extent, found_type);
543 btrfs_unlock_up_safe(path, 1);
544 btrfs_mark_buffer_dirty(path->nodes[0]);
545 btrfs_set_lock_blocking(path->nodes[0]);
547 path->leave_spinning = 0;
548 btrfs_release_path(root, path);
549 if (disk_bytenr != 0)
550 inode_add_bytes(inode, extent_end - end);
553 if (found_extent && !keep) {
554 u64 old_disk_bytenr = le64_to_cpu(old.disk_bytenr);
556 if (old_disk_bytenr != 0) {
557 inode_sub_bytes(inode,
558 le64_to_cpu(old.num_bytes));
559 ret = btrfs_free_extent(trans, root,
561 le64_to_cpu(old.disk_num_bytes),
562 0, root->root_key.objectid,
563 key.objectid, key.offset -
564 le64_to_cpu(old.offset));
566 *hint_byte = old_disk_bytenr;
570 if (search_start >= end) {
576 btrfs_free_path(path);
577 if (locked_end > orig_locked_end) {
578 unlock_extent(&BTRFS_I(inode)->io_tree, orig_locked_end,
579 locked_end - 1, GFP_NOFS);
584 static int extent_mergeable(struct extent_buffer *leaf, int slot,
585 u64 objectid, u64 bytenr, u64 *start, u64 *end)
587 struct btrfs_file_extent_item *fi;
588 struct btrfs_key key;
591 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
594 btrfs_item_key_to_cpu(leaf, &key, slot);
595 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
598 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
599 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
600 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
601 btrfs_file_extent_compression(leaf, fi) ||
602 btrfs_file_extent_encryption(leaf, fi) ||
603 btrfs_file_extent_other_encoding(leaf, fi))
606 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
607 if ((*start && *start != key.offset) || (*end && *end != extent_end))
616 * Mark extent in the range start - end as written.
618 * This changes extent type from 'pre-allocated' to 'regular'. If only
619 * part of extent is marked as written, the extent will be split into
622 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
623 struct btrfs_root *root,
624 struct inode *inode, u64 start, u64 end)
626 struct extent_buffer *leaf;
627 struct btrfs_path *path;
628 struct btrfs_file_extent_item *fi;
629 struct btrfs_key key;
637 u64 locked_end = end;
642 btrfs_drop_extent_cache(inode, start, end - 1, 0);
644 path = btrfs_alloc_path();
647 key.objectid = inode->i_ino;
648 key.type = BTRFS_EXTENT_DATA_KEY;
652 key.offset = split - 1;
654 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
655 if (ret > 0 && path->slots[0] > 0)
658 leaf = path->nodes[0];
659 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
660 BUG_ON(key.objectid != inode->i_ino ||
661 key.type != BTRFS_EXTENT_DATA_KEY);
662 fi = btrfs_item_ptr(leaf, path->slots[0],
663 struct btrfs_file_extent_item);
664 extent_type = btrfs_file_extent_type(leaf, fi);
665 BUG_ON(extent_type != BTRFS_FILE_EXTENT_PREALLOC);
666 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
667 BUG_ON(key.offset > start || extent_end < end);
669 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
670 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
671 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
673 if (key.offset == start)
676 if (key.offset == start && extent_end == end) {
681 if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
682 bytenr, &other_start, &other_end)) {
683 extent_end = other_end;
684 del_slot = path->slots[0] + 1;
686 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
687 0, root->root_key.objectid,
688 inode->i_ino, orig_offset);
693 if (extent_mergeable(leaf, path->slots[0] - 1, inode->i_ino,
694 bytenr, &other_start, &other_end)) {
695 key.offset = other_start;
696 del_slot = path->slots[0];
698 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
699 0, root->root_key.objectid,
700 inode->i_ino, orig_offset);
705 btrfs_set_file_extent_type(leaf, fi,
706 BTRFS_FILE_EXTENT_REG);
710 fi = btrfs_item_ptr(leaf, del_slot - 1,
711 struct btrfs_file_extent_item);
712 btrfs_set_file_extent_type(leaf, fi, BTRFS_FILE_EXTENT_REG);
713 btrfs_set_file_extent_num_bytes(leaf, fi,
714 extent_end - key.offset);
715 btrfs_mark_buffer_dirty(leaf);
717 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
720 } else if (split == start) {
721 if (locked_end < extent_end) {
722 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
723 locked_end, extent_end - 1, GFP_NOFS);
725 btrfs_release_path(root, path);
726 lock_extent(&BTRFS_I(inode)->io_tree,
727 locked_end, extent_end - 1, GFP_NOFS);
728 locked_end = extent_end;
731 locked_end = extent_end;
733 btrfs_set_file_extent_num_bytes(leaf, fi, split - key.offset);
735 BUG_ON(key.offset != start);
737 btrfs_set_file_extent_offset(leaf, fi, key.offset -
739 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - split);
740 btrfs_set_item_key_safe(trans, root, path, &key);
744 if (extent_end == end) {
746 extent_type = BTRFS_FILE_EXTENT_REG;
748 if (extent_end == end && split == start) {
751 if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
752 bytenr, &other_start, &other_end)) {
754 fi = btrfs_item_ptr(leaf, path->slots[0],
755 struct btrfs_file_extent_item);
757 btrfs_set_item_key_safe(trans, root, path, &key);
758 btrfs_set_file_extent_offset(leaf, fi, key.offset -
760 btrfs_set_file_extent_num_bytes(leaf, fi,
765 if (extent_end == end && split == end) {
768 if (extent_mergeable(leaf, path->slots[0] - 1 , inode->i_ino,
769 bytenr, &other_start, &other_end)) {
771 fi = btrfs_item_ptr(leaf, path->slots[0],
772 struct btrfs_file_extent_item);
773 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end -
779 btrfs_mark_buffer_dirty(leaf);
781 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
782 root->root_key.objectid,
783 inode->i_ino, orig_offset);
785 btrfs_release_path(root, path);
788 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*fi));
791 leaf = path->nodes[0];
792 fi = btrfs_item_ptr(leaf, path->slots[0],
793 struct btrfs_file_extent_item);
794 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
795 btrfs_set_file_extent_type(leaf, fi, extent_type);
796 btrfs_set_file_extent_disk_bytenr(leaf, fi, bytenr);
797 btrfs_set_file_extent_disk_num_bytes(leaf, fi, num_bytes);
798 btrfs_set_file_extent_offset(leaf, fi, key.offset - orig_offset);
799 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - key.offset);
800 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
801 btrfs_set_file_extent_compression(leaf, fi, 0);
802 btrfs_set_file_extent_encryption(leaf, fi, 0);
803 btrfs_set_file_extent_other_encoding(leaf, fi, 0);
805 btrfs_mark_buffer_dirty(leaf);
808 btrfs_release_path(root, path);
809 if (split_end && split == start) {
813 if (locked_end > end) {
814 unlock_extent(&BTRFS_I(inode)->io_tree, end, locked_end - 1,
817 btrfs_free_path(path);
822 * this gets pages into the page cache and locks them down, it also properly
823 * waits for data=ordered extents to finish before allowing the pages to be
826 static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
827 struct page **pages, size_t num_pages,
828 loff_t pos, unsigned long first_index,
829 unsigned long last_index, size_t write_bytes)
832 unsigned long index = pos >> PAGE_CACHE_SHIFT;
833 struct inode *inode = fdentry(file)->d_inode;
838 start_pos = pos & ~((u64)root->sectorsize - 1);
839 last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
841 if (start_pos > inode->i_size) {
842 err = btrfs_cont_expand(inode, start_pos);
847 memset(pages, 0, num_pages * sizeof(struct page *));
849 for (i = 0; i < num_pages; i++) {
850 pages[i] = grab_cache_page(inode->i_mapping, index + i);
855 wait_on_page_writeback(pages[i]);
857 if (start_pos < inode->i_size) {
858 struct btrfs_ordered_extent *ordered;
859 lock_extent(&BTRFS_I(inode)->io_tree,
860 start_pos, last_pos - 1, GFP_NOFS);
861 ordered = btrfs_lookup_first_ordered_extent(inode,
864 ordered->file_offset + ordered->len > start_pos &&
865 ordered->file_offset < last_pos) {
866 btrfs_put_ordered_extent(ordered);
867 unlock_extent(&BTRFS_I(inode)->io_tree,
868 start_pos, last_pos - 1, GFP_NOFS);
869 for (i = 0; i < num_pages; i++) {
870 unlock_page(pages[i]);
871 page_cache_release(pages[i]);
873 btrfs_wait_ordered_range(inode, start_pos,
874 last_pos - start_pos);
878 btrfs_put_ordered_extent(ordered);
880 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
881 last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
882 EXTENT_DO_ACCOUNTING,
884 unlock_extent(&BTRFS_I(inode)->io_tree,
885 start_pos, last_pos - 1, GFP_NOFS);
887 for (i = 0; i < num_pages; i++) {
888 clear_page_dirty_for_io(pages[i]);
889 set_page_extent_mapped(pages[i]);
890 WARN_ON(!PageLocked(pages[i]));
895 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
896 size_t count, loff_t *ppos)
900 ssize_t num_written = 0;
903 struct inode *inode = fdentry(file)->d_inode;
904 struct btrfs_root *root = BTRFS_I(inode)->root;
905 struct page **pages = NULL;
907 struct page *pinned[2];
908 unsigned long first_index;
909 unsigned long last_index;
912 will_write = ((file->f_flags & O_DSYNC) || IS_SYNC(inode) ||
913 (file->f_flags & O_DIRECT));
915 nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
916 PAGE_CACHE_SIZE / (sizeof(struct page *)));
923 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
925 /* do the reserve before the mutex lock in case we have to do some
926 * flushing. We wouldn't deadlock, but this is more polite.
928 err = btrfs_reserve_metadata_for_delalloc(root, inode, 1);
932 mutex_lock(&inode->i_mutex);
934 current->backing_dev_info = inode->i_mapping->backing_dev_info;
935 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
942 err = file_remove_suid(file);
946 file_update_time(file);
948 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
950 /* generic_write_checks can change our pos */
953 BTRFS_I(inode)->sequence++;
954 first_index = pos >> PAGE_CACHE_SHIFT;
955 last_index = (pos + count) >> PAGE_CACHE_SHIFT;
958 * there are lots of better ways to do this, but this code
959 * makes sure the first and last page in the file range are
960 * up to date and ready for cow
962 if ((pos & (PAGE_CACHE_SIZE - 1))) {
963 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
964 if (!PageUptodate(pinned[0])) {
965 ret = btrfs_readpage(NULL, pinned[0]);
967 wait_on_page_locked(pinned[0]);
969 unlock_page(pinned[0]);
972 if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
973 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
974 if (!PageUptodate(pinned[1])) {
975 ret = btrfs_readpage(NULL, pinned[1]);
977 wait_on_page_locked(pinned[1]);
979 unlock_page(pinned[1]);
984 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
985 size_t write_bytes = min(count, nrptrs *
986 (size_t)PAGE_CACHE_SIZE -
988 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
991 WARN_ON(num_pages > nrptrs);
992 memset(pages, 0, sizeof(struct page *) * nrptrs);
994 ret = btrfs_check_data_free_space(root, inode, write_bytes);
998 ret = prepare_pages(root, file, pages, num_pages,
999 pos, first_index, last_index,
1002 btrfs_free_reserved_data_space(root, inode,
1007 ret = btrfs_copy_from_user(pos, num_pages,
1008 write_bytes, pages, buf);
1010 btrfs_free_reserved_data_space(root, inode,
1012 btrfs_drop_pages(pages, num_pages);
1016 ret = dirty_and_release_pages(NULL, root, file, pages,
1017 num_pages, pos, write_bytes);
1018 btrfs_drop_pages(pages, num_pages);
1020 btrfs_free_reserved_data_space(root, inode,
1026 filemap_fdatawrite_range(inode->i_mapping, pos,
1027 pos + write_bytes - 1);
1029 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
1032 (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1033 btrfs_btree_balance_dirty(root, 1);
1034 btrfs_throttle(root);
1038 count -= write_bytes;
1040 num_written += write_bytes;
1045 mutex_unlock(&inode->i_mutex);
1048 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
1053 page_cache_release(pinned[0]);
1055 page_cache_release(pinned[1]);
1059 * we want to make sure fsync finds this change
1060 * but we haven't joined a transaction running right now.
1062 * Later on, someone is sure to update the inode and get the
1063 * real transid recorded.
1065 * We set last_trans now to the fs_info generation + 1,
1066 * this will either be one more than the running transaction
1067 * or the generation used for the next transaction if there isn't
1068 * one running right now.
1070 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1072 if (num_written > 0 && will_write) {
1073 struct btrfs_trans_handle *trans;
1075 err = btrfs_wait_ordered_range(inode, start_pos, num_written);
1079 if ((file->f_flags & O_DSYNC) || IS_SYNC(inode)) {
1080 trans = btrfs_start_transaction(root, 1);
1081 ret = btrfs_log_dentry_safe(trans, root,
1084 ret = btrfs_sync_log(trans, root);
1086 btrfs_end_transaction(trans, root);
1088 btrfs_commit_transaction(trans, root);
1089 } else if (ret != BTRFS_NO_LOG_SYNC) {
1090 btrfs_commit_transaction(trans, root);
1092 btrfs_end_transaction(trans, root);
1095 if (file->f_flags & O_DIRECT) {
1096 invalidate_mapping_pages(inode->i_mapping,
1097 start_pos >> PAGE_CACHE_SHIFT,
1098 (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
1101 current->backing_dev_info = NULL;
1102 return num_written ? num_written : err;
1105 int btrfs_release_file(struct inode *inode, struct file *filp)
1108 * ordered_data_close is set by settattr when we are about to truncate
1109 * a file from a non-zero size to a zero size. This tries to
1110 * flush down new bytes that may have been written if the
1111 * application were using truncate to replace a file in place.
1113 if (BTRFS_I(inode)->ordered_data_close) {
1114 BTRFS_I(inode)->ordered_data_close = 0;
1115 btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode);
1116 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1117 filemap_flush(inode->i_mapping);
1119 if (filp->private_data)
1120 btrfs_ioctl_trans_end(filp);
1125 * fsync call for both files and directories. This logs the inode into
1126 * the tree log instead of forcing full commits whenever possible.
1128 * It needs to call filemap_fdatawait so that all ordered extent updates are
1129 * in the metadata btree are up to date for copying to the log.
1131 * It drops the inode mutex before doing the tree log commit. This is an
1132 * important optimization for directories because holding the mutex prevents
1133 * new operations on the dir while we write to disk.
1135 int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
1137 struct inode *inode = dentry->d_inode;
1138 struct btrfs_root *root = BTRFS_I(inode)->root;
1140 struct btrfs_trans_handle *trans;
1143 /* we wait first, since the writeback may change the inode */
1145 /* the VFS called filemap_fdatawrite for us */
1146 btrfs_wait_ordered_range(inode, 0, (u64)-1);
1150 * check the transaction that last modified this inode
1151 * and see if its already been committed
1153 if (!BTRFS_I(inode)->last_trans)
1157 * if the last transaction that changed this file was before
1158 * the current transaction, we can bail out now without any
1161 mutex_lock(&root->fs_info->trans_mutex);
1162 if (BTRFS_I(inode)->last_trans <=
1163 root->fs_info->last_trans_committed) {
1164 BTRFS_I(inode)->last_trans = 0;
1165 mutex_unlock(&root->fs_info->trans_mutex);
1168 mutex_unlock(&root->fs_info->trans_mutex);
1171 * ok we haven't committed the transaction yet, lets do a commit
1173 if (file && file->private_data)
1174 btrfs_ioctl_trans_end(file);
1176 trans = btrfs_start_transaction(root, 1);
1182 ret = btrfs_log_dentry_safe(trans, root, dentry);
1186 /* we've logged all the items and now have a consistent
1187 * version of the file in the log. It is possible that
1188 * someone will come in and modify the file, but that's
1189 * fine because the log is consistent on disk, and we
1190 * have references to all of the file's extents
1192 * It is possible that someone will come in and log the
1193 * file again, but that will end up using the synchronization
1194 * inside btrfs_sync_log to keep things safe.
1196 mutex_unlock(&dentry->d_inode->i_mutex);
1198 if (ret != BTRFS_NO_LOG_SYNC) {
1200 ret = btrfs_commit_transaction(trans, root);
1202 ret = btrfs_sync_log(trans, root);
1204 ret = btrfs_end_transaction(trans, root);
1206 ret = btrfs_commit_transaction(trans, root);
1209 ret = btrfs_end_transaction(trans, root);
1211 mutex_lock(&dentry->d_inode->i_mutex);
1213 return ret > 0 ? EIO : ret;
1216 static const struct vm_operations_struct btrfs_file_vm_ops = {
1217 .fault = filemap_fault,
1218 .page_mkwrite = btrfs_page_mkwrite,
1221 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
1223 vma->vm_ops = &btrfs_file_vm_ops;
1224 file_accessed(filp);
1228 const struct file_operations btrfs_file_operations = {
1229 .llseek = generic_file_llseek,
1230 .read = do_sync_read,
1231 .aio_read = generic_file_aio_read,
1232 .splice_read = generic_file_splice_read,
1233 .write = btrfs_file_write,
1234 .mmap = btrfs_file_mmap,
1235 .open = generic_file_open,
1236 .release = btrfs_release_file,
1237 .fsync = btrfs_sync_file,
1238 .unlocked_ioctl = btrfs_ioctl,
1239 #ifdef CONFIG_COMPAT
1240 .compat_ioctl = btrfs_ioctl,