projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Btrfs: fix regressions in copy_from_user handling
[deliverable/linux.git] / fs / btrfs / file.c
1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
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.
7 *
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.
12 *
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.
17 */
18
19 #include <linux/fs.h>
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>
31 #include <linux/slab.h>
32 #include "ctree.h"
33 #include "disk-io.h"
34 #include "transaction.h"
35 #include "btrfs_inode.h"
36 #include "ioctl.h"
37 #include "print-tree.h"
38 #include "tree-log.h"
39 #include "locking.h"
40 #include "compat.h"
41
42
43 /* simple helper to fault in pages and copy. This should go away
44 * and be replaced with calls into generic code.
45 */
46 static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
47 int write_bytes,
48 struct page **prepared_pages,
49 struct iov_iter *i)
50 {
51 size_t copied = 0;
52 int pg = 0;
53 int offset = pos & (PAGE_CACHE_SIZE - 1);
54 int total_copied = 0;
55
56 while (write_bytes > 0) {
57 size_t count = min_t(size_t,
58 PAGE_CACHE_SIZE - offset, write_bytes);
59 struct page *page = prepared_pages[pg];
60 /*
61 * Copy data from userspace to the current page
62 *
63 * Disable pagefault to avoid recursive lock since
64 * the pages are already locked
65 */
66 pagefault_disable();
67 copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
68 pagefault_enable();
69
70 /* Flush processor's dcache for this page */
71 flush_dcache_page(page);
72 iov_iter_advance(i, copied);
73 write_bytes -= copied;
74 total_copied += copied;
75
76 /* Return to btrfs_file_aio_write to fault page */
77 if (unlikely(copied == 0)) {
78 break;
79 }
80
81 if (unlikely(copied < PAGE_CACHE_SIZE - offset)) {
82 offset += copied;
83 } else {
84 pg++;
85 offset = 0;
86 }
87 }
88 return total_copied;
89 }
90
91 /*
92 * unlocks pages after btrfs_file_write is done with them
93 */
94 static noinline void btrfs_drop_pages(struct page **pages, size_t num_pages)
95 {
96 size_t i;
97 for (i = 0; i < num_pages; i++) {
98 if (!pages[i])
99 break;
100 /* page checked is some magic around finding pages that
101 * have been modified without going through btrfs_set_page_dirty
102 * clear it here
103 */
104 ClearPageChecked(pages[i]);
105 unlock_page(pages[i]);
106 mark_page_accessed(pages[i]);
107 page_cache_release(pages[i]);
108 }
109 }
110
111 /*
112 * after copy_from_user, pages need to be dirtied and we need to make
113 * sure holes are created between the current EOF and the start of
114 * any next extents (if required).
115 *
116 * this also makes the decision about creating an inline extent vs
117 * doing real data extents, marking pages dirty and delalloc as required.
118 */
119 static noinline int dirty_and_release_pages(struct btrfs_trans_handle *trans,
120 struct btrfs_root *root,
121 struct file *file,
122 struct page **pages,
123 size_t num_pages,
124 loff_t pos,
125 size_t write_bytes)
126 {
127 int err = 0;
128 int i;
129 struct inode *inode = fdentry(file)->d_inode;
130 u64 num_bytes;
131 u64 start_pos;
132 u64 end_of_last_block;
133 u64 end_pos = pos + write_bytes;
134 loff_t isize = i_size_read(inode);
135
136 start_pos = pos & ~((u64)root->sectorsize - 1);
137 num_bytes = (write_bytes + pos - start_pos +
138 root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
139
140 end_of_last_block = start_pos + num_bytes - 1;
141 err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
142 NULL);
143 BUG_ON(err);
144
145 for (i = 0; i < num_pages; i++) {
146 struct page *p = pages[i];
147 SetPageUptodate(p);
148 ClearPageChecked(p);
149 set_page_dirty(p);
150 }
151 if (end_pos > isize) {
152 i_size_write(inode, end_pos);
153 /* we've only changed i_size in ram, and we haven't updated
154 * the disk i_size. There is no need to log the inode
155 * at this time.
156 */
157 }
158 return 0;
159 }
160
161 /*
162 * this drops all the extents in the cache that intersect the range
163 * [start, end]. Existing extents are split as required.
164 */
165 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
166 int skip_pinned)
167 {
168 struct extent_map *em;
169 struct extent_map *split = NULL;
170 struct extent_map *split2 = NULL;
171 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
172 u64 len = end - start + 1;
173 int ret;
174 int testend = 1;
175 unsigned long flags;
176 int compressed = 0;
177
178 WARN_ON(end < start);
179 if (end == (u64)-1) {
180 len = (u64)-1;
181 testend = 0;
182 }
183 while (1) {
184 if (!split)
185 split = alloc_extent_map(GFP_NOFS);
186 if (!split2)
187 split2 = alloc_extent_map(GFP_NOFS);
188 BUG_ON(!split || !split2);
189
190 write_lock(&em_tree->lock);
191 em = lookup_extent_mapping(em_tree, start, len);
192 if (!em) {
193 write_unlock(&em_tree->lock);
194 break;
195 }
196 flags = em->flags;
197 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
198 if (testend && em->start + em->len >= start + len) {
199 free_extent_map(em);
200 write_unlock(&em_tree->lock);
201 break;
202 }
203 start = em->start + em->len;
204 if (testend)
205 len = start + len - (em->start + em->len);
206 free_extent_map(em);
207 write_unlock(&em_tree->lock);
208 continue;
209 }
210 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
211 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
212 remove_extent_mapping(em_tree, em);
213
214 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
215 em->start < start) {
216 split->start = em->start;
217 split->len = start - em->start;
218 split->orig_start = em->orig_start;
219 split->block_start = em->block_start;
220
221 if (compressed)
222 split->block_len = em->block_len;
223 else
224 split->block_len = split->len;
225
226 split->bdev = em->bdev;
227 split->flags = flags;
228 split->compress_type = em->compress_type;
229 ret = add_extent_mapping(em_tree, split);
230 BUG_ON(ret);
231 free_extent_map(split);
232 split = split2;
233 split2 = NULL;
234 }
235 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
236 testend && em->start + em->len > start + len) {
237 u64 diff = start + len - em->start;
238
239 split->start = start + len;
240 split->len = em->start + em->len - (start + len);
241 split->bdev = em->bdev;
242 split->flags = flags;
243 split->compress_type = em->compress_type;
244
245 if (compressed) {
246 split->block_len = em->block_len;
247 split->block_start = em->block_start;
248 split->orig_start = em->orig_start;
249 } else {
250 split->block_len = split->len;
251 split->block_start = em->block_start + diff;
252 split->orig_start = split->start;
253 }
254
255 ret = add_extent_mapping(em_tree, split);
256 BUG_ON(ret);
257 free_extent_map(split);
258 split = NULL;
259 }
260 write_unlock(&em_tree->lock);
261
262 /* once for us */
263 free_extent_map(em);
264 /* once for the tree*/
265 free_extent_map(em);
266 }
267 if (split)
268 free_extent_map(split);
269 if (split2)
270 free_extent_map(split2);
271 return 0;
272 }
273
274 /*
275 * this is very complex, but the basic idea is to drop all extents
276 * in the range start - end. hint_block is filled in with a block number
277 * that would be a good hint to the block allocator for this file.
278 *
279 * If an extent intersects the range but is not entirely inside the range
280 * it is either truncated or split. Anything entirely inside the range
281 * is deleted from the tree.
282 */
283 int btrfs_drop_extents(struct btrfs_trans_handle *trans, struct inode *inode,
284 u64 start, u64 end, u64 *hint_byte, int drop_cache)
285 {
286 struct btrfs_root *root = BTRFS_I(inode)->root;
287 struct extent_buffer *leaf;
288 struct btrfs_file_extent_item *fi;
289 struct btrfs_path *path;
290 struct btrfs_key key;
291 struct btrfs_key new_key;
292 u64 search_start = start;
293 u64 disk_bytenr = 0;
294 u64 num_bytes = 0;
295 u64 extent_offset = 0;
296 u64 extent_end = 0;
297 int del_nr = 0;
298 int del_slot = 0;
299 int extent_type;
300 int recow;
301 int ret;
302
303 if (drop_cache)
304 btrfs_drop_extent_cache(inode, start, end - 1, 0);
305
306 path = btrfs_alloc_path();
307 if (!path)
308 return -ENOMEM;
309
310 while (1) {
311 recow = 0;
312 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
313 search_start, -1);
314 if (ret < 0)
315 break;
316 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
317 leaf = path->nodes[0];
318 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
319 if (key.objectid == inode->i_ino &&
320 key.type == BTRFS_EXTENT_DATA_KEY)
321 path->slots[0]--;
322 }
323 ret = 0;
324 next_slot:
325 leaf = path->nodes[0];
326 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
327 BUG_ON(del_nr > 0);
328 ret = btrfs_next_leaf(root, path);
329 if (ret < 0)
330 break;
331 if (ret > 0) {
332 ret = 0;
333 break;
334 }
335 leaf = path->nodes[0];
336 recow = 1;
337 }
338
339 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
340 if (key.objectid > inode->i_ino ||
341 key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
342 break;
343
344 fi = btrfs_item_ptr(leaf, path->slots[0],
345 struct btrfs_file_extent_item);
346 extent_type = btrfs_file_extent_type(leaf, fi);
347
348 if (extent_type == BTRFS_FILE_EXTENT_REG ||
349 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
350 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
351 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
352 extent_offset = btrfs_file_extent_offset(leaf, fi);
353 extent_end = key.offset +
354 btrfs_file_extent_num_bytes(leaf, fi);
355 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
356 extent_end = key.offset +
357 btrfs_file_extent_inline_len(leaf, fi);
358 } else {
359 WARN_ON(1);
360 extent_end = search_start;
361 }
362
363 if (extent_end <= search_start) {
364 path->slots[0]++;
365 goto next_slot;
366 }
367
368 search_start = max(key.offset, start);
369 if (recow) {
370 btrfs_release_path(root, path);
371 continue;
372 }
373
374 /*
375 * | - range to drop - |
376 * | -------- extent -------- |
377 */
378 if (start > key.offset && end < extent_end) {
379 BUG_ON(del_nr > 0);
380 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
381
382 memcpy(&new_key, &key, sizeof(new_key));
383 new_key.offset = start;
384 ret = btrfs_duplicate_item(trans, root, path,
385 &new_key);
386 if (ret == -EAGAIN) {
387 btrfs_release_path(root, path);
388 continue;
389 }
390 if (ret < 0)
391 break;
392
393 leaf = path->nodes[0];
394 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
395 struct btrfs_file_extent_item);
396 btrfs_set_file_extent_num_bytes(leaf, fi,
397 start - key.offset);
398
399 fi = btrfs_item_ptr(leaf, path->slots[0],
400 struct btrfs_file_extent_item);
401
402 extent_offset += start - key.offset;
403 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
404 btrfs_set_file_extent_num_bytes(leaf, fi,
405 extent_end - start);
406 btrfs_mark_buffer_dirty(leaf);
407
408 if (disk_bytenr > 0) {
409 ret = btrfs_inc_extent_ref(trans, root,
410 disk_bytenr, num_bytes, 0,
411 root->root_key.objectid,
412 new_key.objectid,
413 start - extent_offset);
414 BUG_ON(ret);
415 *hint_byte = disk_bytenr;
416 }
417 key.offset = start;
418 }
419 /*
420 * | ---- range to drop ----- |
421 * | -------- extent -------- |
422 */
423 if (start <= key.offset && end < extent_end) {
424 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
425
426 memcpy(&new_key, &key, sizeof(new_key));
427 new_key.offset = end;
428 btrfs_set_item_key_safe(trans, root, path, &new_key);
429
430 extent_offset += end - key.offset;
431 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
432 btrfs_set_file_extent_num_bytes(leaf, fi,
433 extent_end - end);
434 btrfs_mark_buffer_dirty(leaf);
435 if (disk_bytenr > 0) {
436 inode_sub_bytes(inode, end - key.offset);
437 *hint_byte = disk_bytenr;
438 }
439 break;
440 }
441
442 search_start = extent_end;
443 /*
444 * | ---- range to drop ----- |
445 * | -------- extent -------- |
446 */
447 if (start > key.offset && end >= extent_end) {
448 BUG_ON(del_nr > 0);
449 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
450
451 btrfs_set_file_extent_num_bytes(leaf, fi,
452 start - key.offset);
453 btrfs_mark_buffer_dirty(leaf);
454 if (disk_bytenr > 0) {
455 inode_sub_bytes(inode, extent_end - start);
456 *hint_byte = disk_bytenr;
457 }
458 if (end == extent_end)
459 break;
460
461 path->slots[0]++;
462 goto next_slot;
463 }
464
465 /*
466 * | ---- range to drop ----- |
467 * | ------ extent ------ |
468 */
469 if (start <= key.offset && end >= extent_end) {
470 if (del_nr == 0) {
471 del_slot = path->slots[0];
472 del_nr = 1;
473 } else {
474 BUG_ON(del_slot + del_nr != path->slots[0]);
475 del_nr++;
476 }
477
478 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
479 inode_sub_bytes(inode,
480 extent_end - key.offset);
481 extent_end = ALIGN(extent_end,
482 root->sectorsize);
483 } else if (disk_bytenr > 0) {
484 ret = btrfs_free_extent(trans, root,
485 disk_bytenr, num_bytes, 0,
486 root->root_key.objectid,
487 key.objectid, key.offset -
488 extent_offset);
489 BUG_ON(ret);
490 inode_sub_bytes(inode,
491 extent_end - key.offset);
492 *hint_byte = disk_bytenr;
493 }
494
495 if (end == extent_end)
496 break;
497
498 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
499 path->slots[0]++;
500 goto next_slot;
501 }
502
503 ret = btrfs_del_items(trans, root, path, del_slot,
504 del_nr);
505 BUG_ON(ret);
506
507 del_nr = 0;
508 del_slot = 0;
509
510 btrfs_release_path(root, path);
511 continue;
512 }
513
514 BUG_ON(1);
515 }
516
517 if (del_nr > 0) {
518 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
519 BUG_ON(ret);
520 }
521
522 btrfs_free_path(path);
523 return ret;
524 }
525
526 static int extent_mergeable(struct extent_buffer *leaf, int slot,
527 u64 objectid, u64 bytenr, u64 orig_offset,
528 u64 *start, u64 *end)
529 {
530 struct btrfs_file_extent_item *fi;
531 struct btrfs_key key;
532 u64 extent_end;
533
534 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
535 return 0;
536
537 btrfs_item_key_to_cpu(leaf, &key, slot);
538 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
539 return 0;
540
541 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
542 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
543 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
544 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
545 btrfs_file_extent_compression(leaf, fi) ||
546 btrfs_file_extent_encryption(leaf, fi) ||
547 btrfs_file_extent_other_encoding(leaf, fi))
548 return 0;
549
550 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
551 if ((*start && *start != key.offset) || (*end && *end != extent_end))
552 return 0;
553
554 *start = key.offset;
555 *end = extent_end;
556 return 1;
557 }
558
559 /*
560 * Mark extent in the range start - end as written.
561 *
562 * This changes extent type from 'pre-allocated' to 'regular'. If only
563 * part of extent is marked as written, the extent will be split into
564 * two or three.
565 */
566 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
567 struct inode *inode, u64 start, u64 end)
568 {
569 struct btrfs_root *root = BTRFS_I(inode)->root;
570 struct extent_buffer *leaf;
571 struct btrfs_path *path;
572 struct btrfs_file_extent_item *fi;
573 struct btrfs_key key;
574 struct btrfs_key new_key;
575 u64 bytenr;
576 u64 num_bytes;
577 u64 extent_end;
578 u64 orig_offset;
579 u64 other_start;
580 u64 other_end;
581 u64 split;
582 int del_nr = 0;
583 int del_slot = 0;
584 int recow;
585 int ret;
586
587 btrfs_drop_extent_cache(inode, start, end - 1, 0);
588
589 path = btrfs_alloc_path();
590 BUG_ON(!path);
591 again:
592 recow = 0;
593 split = start;
594 key.objectid = inode->i_ino;
595 key.type = BTRFS_EXTENT_DATA_KEY;
596 key.offset = split;
597
598 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
599 if (ret > 0 && path->slots[0] > 0)
600 path->slots[0]--;
601
602 leaf = path->nodes[0];
603 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
604 BUG_ON(key.objectid != inode->i_ino ||
605 key.type != BTRFS_EXTENT_DATA_KEY);
606 fi = btrfs_item_ptr(leaf, path->slots[0],
607 struct btrfs_file_extent_item);
608 BUG_ON(btrfs_file_extent_type(leaf, fi) !=
609 BTRFS_FILE_EXTENT_PREALLOC);
610 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
611 BUG_ON(key.offset > start || extent_end < end);
612
613 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
614 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
615 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
616 memcpy(&new_key, &key, sizeof(new_key));
617
618 if (start == key.offset && end < extent_end) {
619 other_start = 0;
620 other_end = start;
621 if (extent_mergeable(leaf, path->slots[0] - 1,
622 inode->i_ino, bytenr, orig_offset,
623 &other_start, &other_end)) {
624 new_key.offset = end;
625 btrfs_set_item_key_safe(trans, root, path, &new_key);
626 fi = btrfs_item_ptr(leaf, path->slots[0],
627 struct btrfs_file_extent_item);
628 btrfs_set_file_extent_num_bytes(leaf, fi,
629 extent_end - end);
630 btrfs_set_file_extent_offset(leaf, fi,
631 end - orig_offset);
632 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
633 struct btrfs_file_extent_item);
634 btrfs_set_file_extent_num_bytes(leaf, fi,
635 end - other_start);
636 btrfs_mark_buffer_dirty(leaf);
637 goto out;
638 }
639 }
640
641 if (start > key.offset && end == extent_end) {
642 other_start = end;
643 other_end = 0;
644 if (extent_mergeable(leaf, path->slots[0] + 1,
645 inode->i_ino, bytenr, orig_offset,
646 &other_start, &other_end)) {
647 fi = btrfs_item_ptr(leaf, path->slots[0],
648 struct btrfs_file_extent_item);
649 btrfs_set_file_extent_num_bytes(leaf, fi,
650 start - key.offset);
651 path->slots[0]++;
652 new_key.offset = start;
653 btrfs_set_item_key_safe(trans, root, path, &new_key);
654
655 fi = btrfs_item_ptr(leaf, path->slots[0],
656 struct btrfs_file_extent_item);
657 btrfs_set_file_extent_num_bytes(leaf, fi,
658 other_end - start);
659 btrfs_set_file_extent_offset(leaf, fi,
660 start - orig_offset);
661 btrfs_mark_buffer_dirty(leaf);
662 goto out;
663 }
664 }
665
666 while (start > key.offset || end < extent_end) {
667 if (key.offset == start)
668 split = end;
669
670 new_key.offset = split;
671 ret = btrfs_duplicate_item(trans, root, path, &new_key);
672 if (ret == -EAGAIN) {
673 btrfs_release_path(root, path);
674 goto again;
675 }
676 BUG_ON(ret < 0);
677
678 leaf = path->nodes[0];
679 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
680 struct btrfs_file_extent_item);
681 btrfs_set_file_extent_num_bytes(leaf, fi,
682 split - key.offset);
683
684 fi = btrfs_item_ptr(leaf, path->slots[0],
685 struct btrfs_file_extent_item);
686
687 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
688 btrfs_set_file_extent_num_bytes(leaf, fi,
689 extent_end - split);
690 btrfs_mark_buffer_dirty(leaf);
691
692 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
693 root->root_key.objectid,
694 inode->i_ino, orig_offset);
695 BUG_ON(ret);
696
697 if (split == start) {
698 key.offset = start;
699 } else {
700 BUG_ON(start != key.offset);
701 path->slots[0]--;
702 extent_end = end;
703 }
704 recow = 1;
705 }
706
707 other_start = end;
708 other_end = 0;
709 if (extent_mergeable(leaf, path->slots[0] + 1,
710 inode->i_ino, bytenr, orig_offset,
711 &other_start, &other_end)) {
712 if (recow) {
713 btrfs_release_path(root, path);
714 goto again;
715 }
716 extent_end = other_end;
717 del_slot = path->slots[0] + 1;
718 del_nr++;
719 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
720 0, root->root_key.objectid,
721 inode->i_ino, orig_offset);
722 BUG_ON(ret);
723 }
724 other_start = 0;
725 other_end = start;
726 if (extent_mergeable(leaf, path->slots[0] - 1,
727 inode->i_ino, bytenr, orig_offset,
728 &other_start, &other_end)) {
729 if (recow) {
730 btrfs_release_path(root, path);
731 goto again;
732 }
733 key.offset = other_start;
734 del_slot = path->slots[0];
735 del_nr++;
736 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
737 0, root->root_key.objectid,
738 inode->i_ino, orig_offset);
739 BUG_ON(ret);
740 }
741 if (del_nr == 0) {
742 fi = btrfs_item_ptr(leaf, path->slots[0],
743 struct btrfs_file_extent_item);
744 btrfs_set_file_extent_type(leaf, fi,
745 BTRFS_FILE_EXTENT_REG);
746 btrfs_mark_buffer_dirty(leaf);
747 } else {
748 fi = btrfs_item_ptr(leaf, del_slot - 1,
749 struct btrfs_file_extent_item);
750 btrfs_set_file_extent_type(leaf, fi,
751 BTRFS_FILE_EXTENT_REG);
752 btrfs_set_file_extent_num_bytes(leaf, fi,
753 extent_end - key.offset);
754 btrfs_mark_buffer_dirty(leaf);
755
756 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
757 BUG_ON(ret);
758 }
759 out:
760 btrfs_free_path(path);
761 return 0;
762 }
763
764 /*
765 * on error we return an unlocked page and the error value
766 * on success we return a locked page and 0
767 */
768 static int prepare_uptodate_page(struct page *page, u64 pos)
769 {
770 int ret = 0;
771
772 if ((pos & (PAGE_CACHE_SIZE - 1)) && !PageUptodate(page)) {
773 ret = btrfs_readpage(NULL, page);
774 if (ret)
775 return ret;
776 lock_page(page);
777 if (!PageUptodate(page)) {
778 unlock_page(page);
779 return -EIO;
780 }
781 }
782 return 0;
783 }
784
785 /*
786 * this gets pages into the page cache and locks them down, it also properly
787 * waits for data=ordered extents to finish before allowing the pages to be
788 * modified.
789 */
790 static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
791 struct page **pages, size_t num_pages,
792 loff_t pos, unsigned long first_index,
793 unsigned long last_index, size_t write_bytes)
794 {
795 struct extent_state *cached_state = NULL;
796 int i;
797 unsigned long index = pos >> PAGE_CACHE_SHIFT;
798 struct inode *inode = fdentry(file)->d_inode;
799 int err = 0;
800 int faili = 0;
801 u64 start_pos;
802 u64 last_pos;
803
804 start_pos = pos & ~((u64)root->sectorsize - 1);
805 last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
806
807 if (start_pos > inode->i_size) {
808 err = btrfs_cont_expand(inode, start_pos);
809 if (err)
810 return err;
811 }
812
813 memset(pages, 0, num_pages * sizeof(struct page *));
814 again:
815 for (i = 0; i < num_pages; i++) {
816 pages[i] = grab_cache_page(inode->i_mapping, index + i);
817 if (!pages[i]) {
818 faili = i - 1;
819 err = -ENOMEM;
820 goto fail;
821 }
822
823 if (i == 0)
824 err = prepare_uptodate_page(pages[i], pos);
825 if (i == num_pages - 1)
826 err = prepare_uptodate_page(pages[i],
827 pos + write_bytes);
828 if (err) {
829 page_cache_release(pages[i]);
830 faili = i - 1;
831 goto fail;
832 }
833 wait_on_page_writeback(pages[i]);
834 }
835 err = 0;
836 if (start_pos < inode->i_size) {
837 struct btrfs_ordered_extent *ordered;
838 lock_extent_bits(&BTRFS_I(inode)->io_tree,
839 start_pos, last_pos - 1, 0, &cached_state,
840 GFP_NOFS);
841 ordered = btrfs_lookup_first_ordered_extent(inode,
842 last_pos - 1);
843 if (ordered &&
844 ordered->file_offset + ordered->len > start_pos &&
845 ordered->file_offset < last_pos) {
846 btrfs_put_ordered_extent(ordered);
847 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
848 start_pos, last_pos - 1,
849 &cached_state, GFP_NOFS);
850 for (i = 0; i < num_pages; i++) {
851 unlock_page(pages[i]);
852 page_cache_release(pages[i]);
853 }
854 btrfs_wait_ordered_range(inode, start_pos,
855 last_pos - start_pos);
856 goto again;
857 }
858 if (ordered)
859 btrfs_put_ordered_extent(ordered);
860
861 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
862 last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
863 EXTENT_DO_ACCOUNTING, 0, 0, &cached_state,
864 GFP_NOFS);
865 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
866 start_pos, last_pos - 1, &cached_state,
867 GFP_NOFS);
868 }
869 for (i = 0; i < num_pages; i++) {
870 clear_page_dirty_for_io(pages[i]);
871 set_page_extent_mapped(pages[i]);
872 WARN_ON(!PageLocked(pages[i]));
873 }
874 return 0;
875 fail:
876 while (faili >= 0) {
877 unlock_page(pages[faili]);
878 page_cache_release(pages[faili]);
879 faili--;
880 }
881 return err;
882
883 }
884
885 static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
886 const struct iovec *iov,
887 unsigned long nr_segs, loff_t pos)
888 {
889 struct file *file = iocb->ki_filp;
890 struct inode *inode = fdentry(file)->d_inode;
891 struct btrfs_root *root = BTRFS_I(inode)->root;
892 struct page **pages = NULL;
893 struct iov_iter i;
894 loff_t *ppos = &iocb->ki_pos;
895 loff_t start_pos;
896 ssize_t num_written = 0;
897 ssize_t err = 0;
898 size_t count;
899 size_t ocount;
900 int ret = 0;
901 int nrptrs;
902 unsigned long first_index;
903 unsigned long last_index;
904 int will_write;
905 int buffered = 0;
906 int copied = 0;
907 int dirty_pages = 0;
908
909 will_write = ((file->f_flags & O_DSYNC) || IS_SYNC(inode) ||
910 (file->f_flags & O_DIRECT));
911
912 start_pos = pos;
913
914 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
915
916 mutex_lock(&inode->i_mutex);
917
918 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
919 if (err)
920 goto out;
921 count = ocount;
922
923 current->backing_dev_info = inode->i_mapping->backing_dev_info;
924 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
925 if (err)
926 goto out;
927
928 if (count == 0)
929 goto out;
930
931 err = file_remove_suid(file);
932 if (err)
933 goto out;
934
935 /*
936 * If BTRFS flips readonly due to some impossible error
937 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
938 * although we have opened a file as writable, we have
939 * to stop this write operation to ensure FS consistency.
940 */
941 if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
942 err = -EROFS;
943 goto out;
944 }
945
946 file_update_time(file);
947 BTRFS_I(inode)->sequence++;
948
949 if (unlikely(file->f_flags & O_DIRECT)) {
950 num_written = generic_file_direct_write(iocb, iov, &nr_segs,
951 pos, ppos, count,
952 ocount);
953 /*
954 * the generic O_DIRECT will update in-memory i_size after the
955 * DIOs are done. But our endio handlers that update the on
956 * disk i_size never update past the in memory i_size. So we
957 * need one more update here to catch any additions to the
958 * file
959 */
960 if (inode->i_size != BTRFS_I(inode)->disk_i_size) {
961 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
962 mark_inode_dirty(inode);
963 }
964
965 if (num_written < 0) {
966 ret = num_written;
967 num_written = 0;
968 goto out;
969 } else if (num_written == count) {
970 /* pick up pos changes done by the generic code */
971 pos = *ppos;
972 goto out;
973 }
974 /*
975 * We are going to do buffered for the rest of the range, so we
976 * need to make sure to invalidate the buffered pages when we're
977 * done.
978 */
979 buffered = 1;
980 pos += num_written;
981 }
982
983 iov_iter_init(&i, iov, nr_segs, count, num_written);
984 nrptrs = min((iov_iter_count(&i) + PAGE_CACHE_SIZE - 1) /
985 PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
986 (sizeof(struct page *)));
987 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
988 if (!pages) {
989 ret = -ENOMEM;
990 goto out;
991 }
992
993 /* generic_write_checks can change our pos */
994 start_pos = pos;
995
996 first_index = pos >> PAGE_CACHE_SHIFT;
997 last_index = (pos + iov_iter_count(&i)) >> PAGE_CACHE_SHIFT;
998
999 while (iov_iter_count(&i) > 0) {
1000 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1001 size_t write_bytes = min(iov_iter_count(&i),
1002 nrptrs * (size_t)PAGE_CACHE_SIZE -
1003 offset);
1004 size_t num_pages = (write_bytes + offset +
1005 PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1006
1007 WARN_ON(num_pages > nrptrs);
1008 memset(pages, 0, sizeof(struct page *) * nrptrs);
1009
1010 /*
1011 * Fault pages before locking them in prepare_pages
1012 * to avoid recursive lock
1013 */
1014 if (unlikely(iov_iter_fault_in_readable(&i, write_bytes))) {
1015 ret = -EFAULT;
1016 goto out;
1017 }
1018
1019 ret = btrfs_delalloc_reserve_space(inode,
1020 num_pages << PAGE_CACHE_SHIFT);
1021 if (ret)
1022 goto out;
1023
1024 ret = prepare_pages(root, file, pages, num_pages,
1025 pos, first_index, last_index,
1026 write_bytes);
1027 if (ret) {
1028 btrfs_delalloc_release_space(inode,
1029 num_pages << PAGE_CACHE_SHIFT);
1030 goto out;
1031 }
1032
1033 copied = btrfs_copy_from_user(pos, num_pages,
1034 write_bytes, pages, &i);
1035
1036 /*
1037 * if we have trouble faulting in the pages, fall
1038 * back to one page at a time
1039 */
1040 if (copied < write_bytes)
1041 nrptrs = 1;
1042
1043 if (copied == 0)
1044 dirty_pages = 0;
1045 else
1046 dirty_pages = (copied + offset +
1047 PAGE_CACHE_SIZE - 1) >>
1048 PAGE_CACHE_SHIFT;
1049
1050 if (num_pages > dirty_pages) {
1051 if (copied > 0)
1052 atomic_inc(
1053 &BTRFS_I(inode)->outstanding_extents);
1054 btrfs_delalloc_release_space(inode,
1055 (num_pages - dirty_pages) <<
1056 PAGE_CACHE_SHIFT);
1057 }
1058
1059 if (copied > 0) {
1060 dirty_and_release_pages(NULL, root, file, pages,
1061 dirty_pages, pos, copied);
1062 }
1063
1064 btrfs_drop_pages(pages, num_pages);
1065
1066 if (copied > 0) {
1067 if (will_write) {
1068 filemap_fdatawrite_range(inode->i_mapping, pos,
1069 pos + copied - 1);
1070 } else {
1071 balance_dirty_pages_ratelimited_nr(
1072 inode->i_mapping,
1073 dirty_pages);
1074 if (dirty_pages <
1075 (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1076 btrfs_btree_balance_dirty(root, 1);
1077 btrfs_throttle(root);
1078 }
1079 }
1080
1081 pos += copied;
1082 num_written += copied;
1083
1084 cond_resched();
1085 }
1086 out:
1087 mutex_unlock(&inode->i_mutex);
1088 if (ret)
1089 err = ret;
1090
1091 kfree(pages);
1092 *ppos = pos;
1093
1094 /*
1095 * we want to make sure fsync finds this change
1096 * but we haven't joined a transaction running right now.
1097 *
1098 * Later on, someone is sure to update the inode and get the
1099 * real transid recorded.
1100 *
1101 * We set last_trans now to the fs_info generation + 1,
1102 * this will either be one more than the running transaction
1103 * or the generation used for the next transaction if there isn't
1104 * one running right now.
1105 */
1106 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1107
1108 if (num_written > 0 && will_write) {
1109 struct btrfs_trans_handle *trans;
1110
1111 err = btrfs_wait_ordered_range(inode, start_pos, num_written);
1112 if (err)
1113 num_written = err;
1114
1115 if ((file->f_flags & O_DSYNC) || IS_SYNC(inode)) {
1116 trans = btrfs_start_transaction(root, 0);
1117 if (IS_ERR(trans)) {
1118 num_written = PTR_ERR(trans);
1119 goto done;
1120 }
1121 mutex_lock(&inode->i_mutex);
1122 ret = btrfs_log_dentry_safe(trans, root,
1123 file->f_dentry);
1124 mutex_unlock(&inode->i_mutex);
1125 if (ret == 0) {
1126 ret = btrfs_sync_log(trans, root);
1127 if (ret == 0)
1128 btrfs_end_transaction(trans, root);
1129 else
1130 btrfs_commit_transaction(trans, root);
1131 } else if (ret != BTRFS_NO_LOG_SYNC) {
1132 btrfs_commit_transaction(trans, root);
1133 } else {
1134 btrfs_end_transaction(trans, root);
1135 }
1136 }
1137 if (file->f_flags & O_DIRECT && buffered) {
1138 invalidate_mapping_pages(inode->i_mapping,
1139 start_pos >> PAGE_CACHE_SHIFT,
1140 (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
1141 }
1142 }
1143 done:
1144 current->backing_dev_info = NULL;
1145 return num_written ? num_written : err;
1146 }
1147
1148 int btrfs_release_file(struct inode *inode, struct file *filp)
1149 {
1150 /*
1151 * ordered_data_close is set by settattr when we are about to truncate
1152 * a file from a non-zero size to a zero size. This tries to
1153 * flush down new bytes that may have been written if the
1154 * application were using truncate to replace a file in place.
1155 */
1156 if (BTRFS_I(inode)->ordered_data_close) {
1157 BTRFS_I(inode)->ordered_data_close = 0;
1158 btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode);
1159 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1160 filemap_flush(inode->i_mapping);
1161 }
1162 if (filp->private_data)
1163 btrfs_ioctl_trans_end(filp);
1164 return 0;
1165 }
1166
1167 /*
1168 * fsync call for both files and directories. This logs the inode into
1169 * the tree log instead of forcing full commits whenever possible.
1170 *
1171 * It needs to call filemap_fdatawait so that all ordered extent updates are
1172 * in the metadata btree are up to date for copying to the log.
1173 *
1174 * It drops the inode mutex before doing the tree log commit. This is an
1175 * important optimization for directories because holding the mutex prevents
1176 * new operations on the dir while we write to disk.
1177 */
1178 int btrfs_sync_file(struct file *file, int datasync)
1179 {
1180 struct dentry *dentry = file->f_path.dentry;
1181 struct inode *inode = dentry->d_inode;
1182 struct btrfs_root *root = BTRFS_I(inode)->root;
1183 int ret = 0;
1184 struct btrfs_trans_handle *trans;
1185
1186
1187 /* we wait first, since the writeback may change the inode */
1188 root->log_batch++;
1189 /* the VFS called filemap_fdatawrite for us */
1190 btrfs_wait_ordered_range(inode, 0, (u64)-1);
1191 root->log_batch++;
1192
1193 /*
1194 * check the transaction that last modified this inode
1195 * and see if its already been committed
1196 */
1197 if (!BTRFS_I(inode)->last_trans)
1198 goto out;
1199
1200 /*
1201 * if the last transaction that changed this file was before
1202 * the current transaction, we can bail out now without any
1203 * syncing
1204 */
1205 mutex_lock(&root->fs_info->trans_mutex);
1206 if (BTRFS_I(inode)->last_trans <=
1207 root->fs_info->last_trans_committed) {
1208 BTRFS_I(inode)->last_trans = 0;
1209 mutex_unlock(&root->fs_info->trans_mutex);
1210 goto out;
1211 }
1212 mutex_unlock(&root->fs_info->trans_mutex);
1213
1214 /*
1215 * ok we haven't committed the transaction yet, lets do a commit
1216 */
1217 if (file->private_data)
1218 btrfs_ioctl_trans_end(file);
1219
1220 trans = btrfs_start_transaction(root, 0);
1221 if (IS_ERR(trans)) {
1222 ret = PTR_ERR(trans);
1223 goto out;
1224 }
1225
1226 ret = btrfs_log_dentry_safe(trans, root, dentry);
1227 if (ret < 0)
1228 goto out;
1229
1230 /* we've logged all the items and now have a consistent
1231 * version of the file in the log. It is possible that
1232 * someone will come in and modify the file, but that's
1233 * fine because the log is consistent on disk, and we
1234 * have references to all of the file's extents
1235 *
1236 * It is possible that someone will come in and log the
1237 * file again, but that will end up using the synchronization
1238 * inside btrfs_sync_log to keep things safe.
1239 */
1240 mutex_unlock(&dentry->d_inode->i_mutex);
1241
1242 if (ret != BTRFS_NO_LOG_SYNC) {
1243 if (ret > 0) {
1244 ret = btrfs_commit_transaction(trans, root);
1245 } else {
1246 ret = btrfs_sync_log(trans, root);
1247 if (ret == 0)
1248 ret = btrfs_end_transaction(trans, root);
1249 else
1250 ret = btrfs_commit_transaction(trans, root);
1251 }
1252 } else {
1253 ret = btrfs_end_transaction(trans, root);
1254 }
1255 mutex_lock(&dentry->d_inode->i_mutex);
1256 out:
1257 return ret > 0 ? -EIO : ret;
1258 }
1259
1260 static const struct vm_operations_struct btrfs_file_vm_ops = {
1261 .fault = filemap_fault,
1262 .page_mkwrite = btrfs_page_mkwrite,
1263 };
1264
1265 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
1266 {
1267 struct address_space *mapping = filp->f_mapping;
1268
1269 if (!mapping->a_ops->readpage)
1270 return -ENOEXEC;
1271
1272 file_accessed(filp);
1273 vma->vm_ops = &btrfs_file_vm_ops;
1274 vma->vm_flags |= VM_CAN_NONLINEAR;
1275
1276 return 0;
1277 }
1278
1279 const struct file_operations btrfs_file_operations = {
1280 .llseek = generic_file_llseek,
1281 .read = do_sync_read,
1282 .write = do_sync_write,
1283 .aio_read = generic_file_aio_read,
1284 .splice_read = generic_file_splice_read,
1285 .aio_write = btrfs_file_aio_write,
1286 .mmap = btrfs_file_mmap,
1287 .open = generic_file_open,
1288 .release = btrfs_release_file,
1289 .fsync = btrfs_sync_file,
1290 .unlocked_ioctl = btrfs_ioctl,
1291 #ifdef CONFIG_COMPAT
1292 .compat_ioctl = btrfs_ioctl,
1293 #endif
1294 };
Linux kernel - Deliverables
This page took 0.061701 seconds and 5 git commands to generate.