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/falloc.h>
28 #include <linux/swap.h>
29 #include <linux/writeback.h>
30 #include <linux/statfs.h>
31 #include <linux/compat.h>
32 #include <linux/slab.h>
35 #include "transaction.h"
36 #include "btrfs_inode.h"
38 #include "print-tree.h"
44 /* simple helper to fault in pages and copy. This should go away
45 * and be replaced with calls into generic code.
47 static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
49 struct page **prepared_pages,
53 size_t total_copied = 0;
55 int offset = pos & (PAGE_CACHE_SIZE - 1);
57 while (write_bytes > 0) {
58 size_t count = min_t(size_t,
59 PAGE_CACHE_SIZE - offset, write_bytes);
60 struct page *page = prepared_pages[pg];
62 * Copy data from userspace to the current page
64 * Disable pagefault to avoid recursive lock since
65 * the pages are already locked
68 copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
71 /* Flush processor's dcache for this page */
72 flush_dcache_page(page);
75 * if we get a partial write, we can end up with
76 * partially up to date pages. These add
77 * a lot of complexity, so make sure they don't
78 * happen by forcing this copy to be retried.
80 * The rest of the btrfs_file_write code will fall
81 * back to page at a time copies after we return 0.
83 if (!PageUptodate(page) && copied < count)
86 iov_iter_advance(i, copied);
87 write_bytes -= copied;
88 total_copied += copied;
90 /* Return to btrfs_file_aio_write to fault page */
91 if (unlikely(copied == 0))
94 if (unlikely(copied < PAGE_CACHE_SIZE - offset)) {
105 * unlocks pages after btrfs_file_write is done with them
107 void btrfs_drop_pages(struct page **pages, size_t num_pages)
110 for (i = 0; i < num_pages; i++) {
111 /* page checked is some magic around finding pages that
112 * have been modified without going through btrfs_set_page_dirty
115 ClearPageChecked(pages[i]);
116 unlock_page(pages[i]);
117 mark_page_accessed(pages[i]);
118 page_cache_release(pages[i]);
123 * after copy_from_user, pages need to be dirtied and we need to make
124 * sure holes are created between the current EOF and the start of
125 * any next extents (if required).
127 * this also makes the decision about creating an inline extent vs
128 * doing real data extents, marking pages dirty and delalloc as required.
130 int btrfs_dirty_pages(struct btrfs_root *root, struct inode *inode,
131 struct page **pages, size_t num_pages,
132 loff_t pos, size_t write_bytes,
133 struct extent_state **cached)
139 u64 end_of_last_block;
140 u64 end_pos = pos + write_bytes;
141 loff_t isize = i_size_read(inode);
143 start_pos = pos & ~((u64)root->sectorsize - 1);
144 num_bytes = (write_bytes + pos - start_pos +
145 root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
147 end_of_last_block = start_pos + num_bytes - 1;
148 err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
153 for (i = 0; i < num_pages; i++) {
154 struct page *p = pages[i];
161 * we've only changed i_size in ram, and we haven't updated
162 * the disk i_size. There is no need to log the inode
166 i_size_write(inode, end_pos);
171 * this drops all the extents in the cache that intersect the range
172 * [start, end]. Existing extents are split as required.
174 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
177 struct extent_map *em;
178 struct extent_map *split = NULL;
179 struct extent_map *split2 = NULL;
180 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
181 u64 len = end - start + 1;
187 WARN_ON(end < start);
188 if (end == (u64)-1) {
194 split = alloc_extent_map(GFP_NOFS);
196 split2 = alloc_extent_map(GFP_NOFS);
197 BUG_ON(!split || !split2);
199 write_lock(&em_tree->lock);
200 em = lookup_extent_mapping(em_tree, start, len);
202 write_unlock(&em_tree->lock);
206 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
207 if (testend && em->start + em->len >= start + len) {
209 write_unlock(&em_tree->lock);
212 start = em->start + em->len;
214 len = start + len - (em->start + em->len);
216 write_unlock(&em_tree->lock);
219 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
220 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
221 remove_extent_mapping(em_tree, em);
223 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
225 split->start = em->start;
226 split->len = start - em->start;
227 split->orig_start = em->orig_start;
228 split->block_start = em->block_start;
231 split->block_len = em->block_len;
233 split->block_len = split->len;
235 split->bdev = em->bdev;
236 split->flags = flags;
237 split->compress_type = em->compress_type;
238 ret = add_extent_mapping(em_tree, split);
240 free_extent_map(split);
244 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
245 testend && em->start + em->len > start + len) {
246 u64 diff = start + len - em->start;
248 split->start = start + len;
249 split->len = em->start + em->len - (start + len);
250 split->bdev = em->bdev;
251 split->flags = flags;
252 split->compress_type = em->compress_type;
255 split->block_len = em->block_len;
256 split->block_start = em->block_start;
257 split->orig_start = em->orig_start;
259 split->block_len = split->len;
260 split->block_start = em->block_start + diff;
261 split->orig_start = split->start;
264 ret = add_extent_mapping(em_tree, split);
266 free_extent_map(split);
269 write_unlock(&em_tree->lock);
273 /* once for the tree*/
277 free_extent_map(split);
279 free_extent_map(split2);
284 * this is very complex, but the basic idea is to drop all extents
285 * in the range start - end. hint_block is filled in with a block number
286 * that would be a good hint to the block allocator for this file.
288 * If an extent intersects the range but is not entirely inside the range
289 * it is either truncated or split. Anything entirely inside the range
290 * is deleted from the tree.
292 int btrfs_drop_extents(struct btrfs_trans_handle *trans, struct inode *inode,
293 u64 start, u64 end, u64 *hint_byte, int drop_cache)
295 struct btrfs_root *root = BTRFS_I(inode)->root;
296 struct extent_buffer *leaf;
297 struct btrfs_file_extent_item *fi;
298 struct btrfs_path *path;
299 struct btrfs_key key;
300 struct btrfs_key new_key;
301 u64 ino = btrfs_ino(inode);
302 u64 search_start = start;
305 u64 extent_offset = 0;
314 btrfs_drop_extent_cache(inode, start, end - 1, 0);
316 path = btrfs_alloc_path();
322 ret = btrfs_lookup_file_extent(trans, root, path, ino,
326 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
327 leaf = path->nodes[0];
328 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
329 if (key.objectid == ino &&
330 key.type == BTRFS_EXTENT_DATA_KEY)
335 leaf = path->nodes[0];
336 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
338 ret = btrfs_next_leaf(root, path);
345 leaf = path->nodes[0];
349 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
350 if (key.objectid > ino ||
351 key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
354 fi = btrfs_item_ptr(leaf, path->slots[0],
355 struct btrfs_file_extent_item);
356 extent_type = btrfs_file_extent_type(leaf, fi);
358 if (extent_type == BTRFS_FILE_EXTENT_REG ||
359 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
360 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
361 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
362 extent_offset = btrfs_file_extent_offset(leaf, fi);
363 extent_end = key.offset +
364 btrfs_file_extent_num_bytes(leaf, fi);
365 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
366 extent_end = key.offset +
367 btrfs_file_extent_inline_len(leaf, fi);
370 extent_end = search_start;
373 if (extent_end <= search_start) {
378 search_start = max(key.offset, start);
380 btrfs_release_path(root, path);
385 * | - range to drop - |
386 * | -------- extent -------- |
388 if (start > key.offset && end < extent_end) {
390 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
392 memcpy(&new_key, &key, sizeof(new_key));
393 new_key.offset = start;
394 ret = btrfs_duplicate_item(trans, root, path,
396 if (ret == -EAGAIN) {
397 btrfs_release_path(root, path);
403 leaf = path->nodes[0];
404 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
405 struct btrfs_file_extent_item);
406 btrfs_set_file_extent_num_bytes(leaf, fi,
409 fi = btrfs_item_ptr(leaf, path->slots[0],
410 struct btrfs_file_extent_item);
412 extent_offset += start - 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);
418 if (disk_bytenr > 0) {
419 ret = btrfs_inc_extent_ref(trans, root,
420 disk_bytenr, num_bytes, 0,
421 root->root_key.objectid,
423 start - extent_offset);
425 *hint_byte = disk_bytenr;
430 * | ---- range to drop ----- |
431 * | -------- extent -------- |
433 if (start <= key.offset && end < extent_end) {
434 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
436 memcpy(&new_key, &key, sizeof(new_key));
437 new_key.offset = end;
438 btrfs_set_item_key_safe(trans, root, path, &new_key);
440 extent_offset += end - key.offset;
441 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
442 btrfs_set_file_extent_num_bytes(leaf, fi,
444 btrfs_mark_buffer_dirty(leaf);
445 if (disk_bytenr > 0) {
446 inode_sub_bytes(inode, end - key.offset);
447 *hint_byte = disk_bytenr;
452 search_start = extent_end;
454 * | ---- range to drop ----- |
455 * | -------- extent -------- |
457 if (start > key.offset && end >= extent_end) {
459 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
461 btrfs_set_file_extent_num_bytes(leaf, fi,
463 btrfs_mark_buffer_dirty(leaf);
464 if (disk_bytenr > 0) {
465 inode_sub_bytes(inode, extent_end - start);
466 *hint_byte = disk_bytenr;
468 if (end == extent_end)
476 * | ---- range to drop ----- |
477 * | ------ extent ------ |
479 if (start <= key.offset && end >= extent_end) {
481 del_slot = path->slots[0];
484 BUG_ON(del_slot + del_nr != path->slots[0]);
488 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
489 inode_sub_bytes(inode,
490 extent_end - key.offset);
491 extent_end = ALIGN(extent_end,
493 } else if (disk_bytenr > 0) {
494 ret = btrfs_free_extent(trans, root,
495 disk_bytenr, num_bytes, 0,
496 root->root_key.objectid,
497 key.objectid, key.offset -
500 inode_sub_bytes(inode,
501 extent_end - key.offset);
502 *hint_byte = disk_bytenr;
505 if (end == extent_end)
508 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
513 ret = btrfs_del_items(trans, root, path, del_slot,
520 btrfs_release_path(root, path);
528 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
532 btrfs_free_path(path);
536 static int extent_mergeable(struct extent_buffer *leaf, int slot,
537 u64 objectid, u64 bytenr, u64 orig_offset,
538 u64 *start, u64 *end)
540 struct btrfs_file_extent_item *fi;
541 struct btrfs_key key;
544 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
547 btrfs_item_key_to_cpu(leaf, &key, slot);
548 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
551 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
552 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
553 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
554 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
555 btrfs_file_extent_compression(leaf, fi) ||
556 btrfs_file_extent_encryption(leaf, fi) ||
557 btrfs_file_extent_other_encoding(leaf, fi))
560 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
561 if ((*start && *start != key.offset) || (*end && *end != extent_end))
570 * Mark extent in the range start - end as written.
572 * This changes extent type from 'pre-allocated' to 'regular'. If only
573 * part of extent is marked as written, the extent will be split into
576 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
577 struct inode *inode, u64 start, u64 end)
579 struct btrfs_root *root = BTRFS_I(inode)->root;
580 struct extent_buffer *leaf;
581 struct btrfs_path *path;
582 struct btrfs_file_extent_item *fi;
583 struct btrfs_key key;
584 struct btrfs_key new_key;
596 u64 ino = btrfs_ino(inode);
598 btrfs_drop_extent_cache(inode, start, end - 1, 0);
600 path = btrfs_alloc_path();
606 key.type = BTRFS_EXTENT_DATA_KEY;
609 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
612 if (ret > 0 && path->slots[0] > 0)
615 leaf = path->nodes[0];
616 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
617 BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY);
618 fi = btrfs_item_ptr(leaf, path->slots[0],
619 struct btrfs_file_extent_item);
620 BUG_ON(btrfs_file_extent_type(leaf, fi) !=
621 BTRFS_FILE_EXTENT_PREALLOC);
622 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
623 BUG_ON(key.offset > start || extent_end < end);
625 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
626 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
627 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
628 memcpy(&new_key, &key, sizeof(new_key));
630 if (start == key.offset && end < extent_end) {
633 if (extent_mergeable(leaf, path->slots[0] - 1,
634 ino, bytenr, orig_offset,
635 &other_start, &other_end)) {
636 new_key.offset = end;
637 btrfs_set_item_key_safe(trans, root, path, &new_key);
638 fi = btrfs_item_ptr(leaf, path->slots[0],
639 struct btrfs_file_extent_item);
640 btrfs_set_file_extent_num_bytes(leaf, fi,
642 btrfs_set_file_extent_offset(leaf, fi,
644 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
645 struct btrfs_file_extent_item);
646 btrfs_set_file_extent_num_bytes(leaf, fi,
648 btrfs_mark_buffer_dirty(leaf);
653 if (start > key.offset && end == extent_end) {
656 if (extent_mergeable(leaf, path->slots[0] + 1,
657 ino, bytenr, orig_offset,
658 &other_start, &other_end)) {
659 fi = btrfs_item_ptr(leaf, path->slots[0],
660 struct btrfs_file_extent_item);
661 btrfs_set_file_extent_num_bytes(leaf, fi,
664 new_key.offset = start;
665 btrfs_set_item_key_safe(trans, root, path, &new_key);
667 fi = btrfs_item_ptr(leaf, path->slots[0],
668 struct btrfs_file_extent_item);
669 btrfs_set_file_extent_num_bytes(leaf, fi,
671 btrfs_set_file_extent_offset(leaf, fi,
672 start - orig_offset);
673 btrfs_mark_buffer_dirty(leaf);
678 while (start > key.offset || end < extent_end) {
679 if (key.offset == start)
682 new_key.offset = split;
683 ret = btrfs_duplicate_item(trans, root, path, &new_key);
684 if (ret == -EAGAIN) {
685 btrfs_release_path(root, path);
690 leaf = path->nodes[0];
691 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
692 struct btrfs_file_extent_item);
693 btrfs_set_file_extent_num_bytes(leaf, fi,
696 fi = btrfs_item_ptr(leaf, path->slots[0],
697 struct btrfs_file_extent_item);
699 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
700 btrfs_set_file_extent_num_bytes(leaf, fi,
702 btrfs_mark_buffer_dirty(leaf);
704 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
705 root->root_key.objectid,
709 if (split == start) {
712 BUG_ON(start != key.offset);
721 if (extent_mergeable(leaf, path->slots[0] + 1,
722 ino, bytenr, orig_offset,
723 &other_start, &other_end)) {
725 btrfs_release_path(root, path);
728 extent_end = other_end;
729 del_slot = path->slots[0] + 1;
731 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
732 0, root->root_key.objectid,
738 if (extent_mergeable(leaf, path->slots[0] - 1,
739 ino, bytenr, orig_offset,
740 &other_start, &other_end)) {
742 btrfs_release_path(root, path);
745 key.offset = other_start;
746 del_slot = path->slots[0];
748 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
749 0, root->root_key.objectid,
754 fi = btrfs_item_ptr(leaf, path->slots[0],
755 struct btrfs_file_extent_item);
756 btrfs_set_file_extent_type(leaf, fi,
757 BTRFS_FILE_EXTENT_REG);
758 btrfs_mark_buffer_dirty(leaf);
760 fi = btrfs_item_ptr(leaf, del_slot - 1,
761 struct btrfs_file_extent_item);
762 btrfs_set_file_extent_type(leaf, fi,
763 BTRFS_FILE_EXTENT_REG);
764 btrfs_set_file_extent_num_bytes(leaf, fi,
765 extent_end - key.offset);
766 btrfs_mark_buffer_dirty(leaf);
768 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
772 btrfs_free_path(path);
777 * on error we return an unlocked page and the error value
778 * on success we return a locked page and 0
780 static int prepare_uptodate_page(struct page *page, u64 pos)
784 if ((pos & (PAGE_CACHE_SIZE - 1)) && !PageUptodate(page)) {
785 ret = btrfs_readpage(NULL, page);
789 if (!PageUptodate(page)) {
798 * this gets pages into the page cache and locks them down, it also properly
799 * waits for data=ordered extents to finish before allowing the pages to be
802 static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
803 struct page **pages, size_t num_pages,
804 loff_t pos, unsigned long first_index,
805 unsigned long last_index, size_t write_bytes)
807 struct extent_state *cached_state = NULL;
809 unsigned long index = pos >> PAGE_CACHE_SHIFT;
810 struct inode *inode = fdentry(file)->d_inode;
816 start_pos = pos & ~((u64)root->sectorsize - 1);
817 last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
819 if (start_pos > inode->i_size) {
820 err = btrfs_cont_expand(inode, i_size_read(inode), start_pos);
826 for (i = 0; i < num_pages; i++) {
827 pages[i] = grab_cache_page(inode->i_mapping, index + i);
835 err = prepare_uptodate_page(pages[i], pos);
836 if (i == num_pages - 1)
837 err = prepare_uptodate_page(pages[i],
840 page_cache_release(pages[i]);
844 wait_on_page_writeback(pages[i]);
847 if (start_pos < inode->i_size) {
848 struct btrfs_ordered_extent *ordered;
849 lock_extent_bits(&BTRFS_I(inode)->io_tree,
850 start_pos, last_pos - 1, 0, &cached_state,
852 ordered = btrfs_lookup_first_ordered_extent(inode,
855 ordered->file_offset + ordered->len > start_pos &&
856 ordered->file_offset < last_pos) {
857 btrfs_put_ordered_extent(ordered);
858 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
859 start_pos, last_pos - 1,
860 &cached_state, GFP_NOFS);
861 for (i = 0; i < num_pages; i++) {
862 unlock_page(pages[i]);
863 page_cache_release(pages[i]);
865 btrfs_wait_ordered_range(inode, start_pos,
866 last_pos - start_pos);
870 btrfs_put_ordered_extent(ordered);
872 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
873 last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
874 EXTENT_DO_ACCOUNTING, 0, 0, &cached_state,
876 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
877 start_pos, last_pos - 1, &cached_state,
880 for (i = 0; i < num_pages; i++) {
881 clear_page_dirty_for_io(pages[i]);
882 set_page_extent_mapped(pages[i]);
883 WARN_ON(!PageLocked(pages[i]));
888 unlock_page(pages[faili]);
889 page_cache_release(pages[faili]);
896 static noinline ssize_t __btrfs_buffered_write(struct file *file,
900 struct inode *inode = fdentry(file)->d_inode;
901 struct btrfs_root *root = BTRFS_I(inode)->root;
902 struct page **pages = NULL;
903 unsigned long first_index;
904 unsigned long last_index;
905 size_t num_written = 0;
909 nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) /
910 PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
911 (sizeof(struct page *)));
912 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
916 first_index = pos >> PAGE_CACHE_SHIFT;
917 last_index = (pos + iov_iter_count(i)) >> PAGE_CACHE_SHIFT;
919 while (iov_iter_count(i) > 0) {
920 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
921 size_t write_bytes = min(iov_iter_count(i),
922 nrptrs * (size_t)PAGE_CACHE_SIZE -
924 size_t num_pages = (write_bytes + offset +
925 PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
929 WARN_ON(num_pages > nrptrs);
932 * Fault pages before locking them in prepare_pages
933 * to avoid recursive lock
935 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
940 ret = btrfs_delalloc_reserve_space(inode,
941 num_pages << PAGE_CACHE_SHIFT);
946 * This is going to setup the pages array with the number of
947 * pages we want, so we don't really need to worry about the
948 * contents of pages from loop to loop
950 ret = prepare_pages(root, file, pages, num_pages,
951 pos, first_index, last_index,
954 btrfs_delalloc_release_space(inode,
955 num_pages << PAGE_CACHE_SHIFT);
959 copied = btrfs_copy_from_user(pos, num_pages,
960 write_bytes, pages, i);
963 * if we have trouble faulting in the pages, fall
964 * back to one page at a time
966 if (copied < write_bytes)
972 dirty_pages = (copied + offset +
973 PAGE_CACHE_SIZE - 1) >>
977 * If we had a short copy we need to release the excess delaloc
978 * bytes we reserved. We need to increment outstanding_extents
979 * because btrfs_delalloc_release_space will decrement it, but
980 * we still have an outstanding extent for the chunk we actually
983 if (num_pages > dirty_pages) {
986 &BTRFS_I(inode)->outstanding_extents);
987 btrfs_delalloc_release_space(inode,
988 (num_pages - dirty_pages) <<
993 ret = btrfs_dirty_pages(root, inode, pages,
994 dirty_pages, pos, copied,
997 btrfs_delalloc_release_space(inode,
998 dirty_pages << PAGE_CACHE_SHIFT);
999 btrfs_drop_pages(pages, num_pages);
1004 btrfs_drop_pages(pages, num_pages);
1008 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
1010 if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1011 btrfs_btree_balance_dirty(root, 1);
1012 btrfs_throttle(root);
1015 num_written += copied;
1020 return num_written ? num_written : ret;
1023 static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1024 const struct iovec *iov,
1025 unsigned long nr_segs, loff_t pos,
1026 loff_t *ppos, size_t count, size_t ocount)
1028 struct file *file = iocb->ki_filp;
1029 struct inode *inode = fdentry(file)->d_inode;
1032 ssize_t written_buffered;
1036 written = generic_file_direct_write(iocb, iov, &nr_segs, pos, ppos,
1040 * the generic O_DIRECT will update in-memory i_size after the
1041 * DIOs are done. But our endio handlers that update the on
1042 * disk i_size never update past the in memory i_size. So we
1043 * need one more update here to catch any additions to the
1046 if (inode->i_size != BTRFS_I(inode)->disk_i_size) {
1047 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
1048 mark_inode_dirty(inode);
1051 if (written < 0 || written == count)
1056 iov_iter_init(&i, iov, nr_segs, count, written);
1057 written_buffered = __btrfs_buffered_write(file, &i, pos);
1058 if (written_buffered < 0) {
1059 err = written_buffered;
1062 endbyte = pos + written_buffered - 1;
1063 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
1066 written += written_buffered;
1067 *ppos = pos + written_buffered;
1068 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
1069 endbyte >> PAGE_CACHE_SHIFT);
1071 return written ? written : err;
1074 static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
1075 const struct iovec *iov,
1076 unsigned long nr_segs, loff_t pos)
1078 struct file *file = iocb->ki_filp;
1079 struct inode *inode = fdentry(file)->d_inode;
1080 struct btrfs_root *root = BTRFS_I(inode)->root;
1081 loff_t *ppos = &iocb->ki_pos;
1082 ssize_t num_written = 0;
1084 size_t count, ocount;
1086 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1088 mutex_lock(&inode->i_mutex);
1090 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
1092 mutex_unlock(&inode->i_mutex);
1097 current->backing_dev_info = inode->i_mapping->backing_dev_info;
1098 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1100 mutex_unlock(&inode->i_mutex);
1105 mutex_unlock(&inode->i_mutex);
1109 err = file_remove_suid(file);
1111 mutex_unlock(&inode->i_mutex);
1116 * If BTRFS flips readonly due to some impossible error
1117 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1118 * although we have opened a file as writable, we have
1119 * to stop this write operation to ensure FS consistency.
1121 if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
1122 mutex_unlock(&inode->i_mutex);
1127 file_update_time(file);
1128 BTRFS_I(inode)->sequence++;
1130 if (unlikely(file->f_flags & O_DIRECT)) {
1131 num_written = __btrfs_direct_write(iocb, iov, nr_segs,
1132 pos, ppos, count, ocount);
1136 iov_iter_init(&i, iov, nr_segs, count, num_written);
1138 num_written = __btrfs_buffered_write(file, &i, pos);
1139 if (num_written > 0)
1140 *ppos = pos + num_written;
1143 mutex_unlock(&inode->i_mutex);
1146 * we want to make sure fsync finds this change
1147 * but we haven't joined a transaction running right now.
1149 * Later on, someone is sure to update the inode and get the
1150 * real transid recorded.
1152 * We set last_trans now to the fs_info generation + 1,
1153 * this will either be one more than the running transaction
1154 * or the generation used for the next transaction if there isn't
1155 * one running right now.
1157 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1158 if (num_written > 0 || num_written == -EIOCBQUEUED) {
1159 err = generic_write_sync(file, pos, num_written);
1160 if (err < 0 && num_written > 0)
1164 current->backing_dev_info = NULL;
1165 return num_written ? num_written : err;
1168 int btrfs_release_file(struct inode *inode, struct file *filp)
1171 * ordered_data_close is set by settattr when we are about to truncate
1172 * a file from a non-zero size to a zero size. This tries to
1173 * flush down new bytes that may have been written if the
1174 * application were using truncate to replace a file in place.
1176 if (BTRFS_I(inode)->ordered_data_close) {
1177 BTRFS_I(inode)->ordered_data_close = 0;
1178 btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode);
1179 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1180 filemap_flush(inode->i_mapping);
1182 if (filp->private_data)
1183 btrfs_ioctl_trans_end(filp);
1188 * fsync call for both files and directories. This logs the inode into
1189 * the tree log instead of forcing full commits whenever possible.
1191 * It needs to call filemap_fdatawait so that all ordered extent updates are
1192 * in the metadata btree are up to date for copying to the log.
1194 * It drops the inode mutex before doing the tree log commit. This is an
1195 * important optimization for directories because holding the mutex prevents
1196 * new operations on the dir while we write to disk.
1198 int btrfs_sync_file(struct file *file, int datasync)
1200 struct dentry *dentry = file->f_path.dentry;
1201 struct inode *inode = dentry->d_inode;
1202 struct btrfs_root *root = BTRFS_I(inode)->root;
1204 struct btrfs_trans_handle *trans;
1206 trace_btrfs_sync_file(file, datasync);
1208 /* we wait first, since the writeback may change the inode */
1210 /* the VFS called filemap_fdatawrite for us */
1211 btrfs_wait_ordered_range(inode, 0, (u64)-1);
1215 * check the transaction that last modified this inode
1216 * and see if its already been committed
1218 if (!BTRFS_I(inode)->last_trans)
1222 * if the last transaction that changed this file was before
1223 * the current transaction, we can bail out now without any
1226 mutex_lock(&root->fs_info->trans_mutex);
1227 if (BTRFS_I(inode)->last_trans <=
1228 root->fs_info->last_trans_committed) {
1229 BTRFS_I(inode)->last_trans = 0;
1230 mutex_unlock(&root->fs_info->trans_mutex);
1233 mutex_unlock(&root->fs_info->trans_mutex);
1236 * ok we haven't committed the transaction yet, lets do a commit
1238 if (file->private_data)
1239 btrfs_ioctl_trans_end(file);
1241 trans = btrfs_start_transaction(root, 0);
1242 if (IS_ERR(trans)) {
1243 ret = PTR_ERR(trans);
1247 ret = btrfs_log_dentry_safe(trans, root, dentry);
1251 /* we've logged all the items and now have a consistent
1252 * version of the file in the log. It is possible that
1253 * someone will come in and modify the file, but that's
1254 * fine because the log is consistent on disk, and we
1255 * have references to all of the file's extents
1257 * It is possible that someone will come in and log the
1258 * file again, but that will end up using the synchronization
1259 * inside btrfs_sync_log to keep things safe.
1261 mutex_unlock(&dentry->d_inode->i_mutex);
1263 if (ret != BTRFS_NO_LOG_SYNC) {
1265 ret = btrfs_commit_transaction(trans, root);
1267 ret = btrfs_sync_log(trans, root);
1269 ret = btrfs_end_transaction(trans, root);
1271 ret = btrfs_commit_transaction(trans, root);
1274 ret = btrfs_end_transaction(trans, root);
1276 mutex_lock(&dentry->d_inode->i_mutex);
1278 return ret > 0 ? -EIO : ret;
1281 static const struct vm_operations_struct btrfs_file_vm_ops = {
1282 .fault = filemap_fault,
1283 .page_mkwrite = btrfs_page_mkwrite,
1286 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
1288 struct address_space *mapping = filp->f_mapping;
1290 if (!mapping->a_ops->readpage)
1293 file_accessed(filp);
1294 vma->vm_ops = &btrfs_file_vm_ops;
1295 vma->vm_flags |= VM_CAN_NONLINEAR;
1300 static long btrfs_fallocate(struct file *file, int mode,
1301 loff_t offset, loff_t len)
1303 struct inode *inode = file->f_path.dentry->d_inode;
1304 struct extent_state *cached_state = NULL;
1311 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
1312 struct extent_map *em;
1315 alloc_start = offset & ~mask;
1316 alloc_end = (offset + len + mask) & ~mask;
1318 /* We only support the FALLOC_FL_KEEP_SIZE mode */
1319 if (mode & ~FALLOC_FL_KEEP_SIZE)
1323 * wait for ordered IO before we have any locks. We'll loop again
1324 * below with the locks held.
1326 btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
1328 mutex_lock(&inode->i_mutex);
1329 ret = inode_newsize_ok(inode, alloc_end);
1333 if (alloc_start > inode->i_size) {
1334 ret = btrfs_cont_expand(inode, i_size_read(inode),
1340 ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
1344 locked_end = alloc_end - 1;
1346 struct btrfs_ordered_extent *ordered;
1348 /* the extent lock is ordered inside the running
1351 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
1352 locked_end, 0, &cached_state, GFP_NOFS);
1353 ordered = btrfs_lookup_first_ordered_extent(inode,
1356 ordered->file_offset + ordered->len > alloc_start &&
1357 ordered->file_offset < alloc_end) {
1358 btrfs_put_ordered_extent(ordered);
1359 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1360 alloc_start, locked_end,
1361 &cached_state, GFP_NOFS);
1363 * we can't wait on the range with the transaction
1364 * running or with the extent lock held
1366 btrfs_wait_ordered_range(inode, alloc_start,
1367 alloc_end - alloc_start);
1370 btrfs_put_ordered_extent(ordered);
1375 cur_offset = alloc_start;
1377 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
1378 alloc_end - cur_offset, 0);
1379 BUG_ON(IS_ERR(em) || !em);
1380 last_byte = min(extent_map_end(em), alloc_end);
1381 last_byte = (last_byte + mask) & ~mask;
1382 if (em->block_start == EXTENT_MAP_HOLE ||
1383 (cur_offset >= inode->i_size &&
1384 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
1385 ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
1386 last_byte - cur_offset,
1387 1 << inode->i_blkbits,
1391 free_extent_map(em);
1395 free_extent_map(em);
1397 cur_offset = last_byte;
1398 if (cur_offset >= alloc_end) {
1403 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
1404 &cached_state, GFP_NOFS);
1406 btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
1408 mutex_unlock(&inode->i_mutex);
1412 const struct file_operations btrfs_file_operations = {
1413 .llseek = generic_file_llseek,
1414 .read = do_sync_read,
1415 .write = do_sync_write,
1416 .aio_read = generic_file_aio_read,
1417 .splice_read = generic_file_splice_read,
1418 .aio_write = btrfs_file_aio_write,
1419 .mmap = btrfs_file_mmap,
1420 .open = generic_file_open,
1421 .release = btrfs_release_file,
1422 .fsync = btrfs_sync_file,
1423 .fallocate = btrfs_fallocate,
1424 .unlocked_ioctl = btrfs_ioctl,
1425 #ifdef CONFIG_COMPAT
1426 .compat_ioctl = btrfs_ioctl,