projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Btrfs: fix fallocate deadlock on inode extent lock
[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/smp_lock.h>
26 #include <linux/backing-dev.h>
27 #include <linux/mpage.h>
28 #include <linux/swap.h>
29 #include <linux/writeback.h>
30 #include <linux/statfs.h>
31 #include <linux/compat.h>
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 const char __user *buf)
50 {
51 long page_fault = 0;
52 int i;
53 int offset = pos & (PAGE_CACHE_SIZE - 1);
54
55 for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
56 size_t count = min_t(size_t,
57 PAGE_CACHE_SIZE - offset, write_bytes);
58 struct page *page = prepared_pages[i];
59 fault_in_pages_readable(buf, count);
60
61 /* Copy data from userspace to the current page */
62 kmap(page);
63 page_fault = __copy_from_user(page_address(page) + offset,
64 buf, count);
65 /* Flush processor's dcache for this page */
66 flush_dcache_page(page);
67 kunmap(page);
68 buf += count;
69 write_bytes -= count;
70
71 if (page_fault)
72 break;
73 }
74 return page_fault ? -EFAULT : 0;
75 }
76
77 /*
78 * unlocks pages after btrfs_file_write is done with them
79 */
80 static noinline void btrfs_drop_pages(struct page **pages, size_t num_pages)
81 {
82 size_t i;
83 for (i = 0; i < num_pages; i++) {
84 if (!pages[i])
85 break;
86 /* page checked is some magic around finding pages that
87 * have been modified without going through btrfs_set_page_dirty
88 * clear it here
89 */
90 ClearPageChecked(pages[i]);
91 unlock_page(pages[i]);
92 mark_page_accessed(pages[i]);
93 page_cache_release(pages[i]);
94 }
95 }
96
97 /*
98 * after copy_from_user, pages need to be dirtied and we need to make
99 * sure holes are created between the current EOF and the start of
100 * any next extents (if required).
101 *
102 * this also makes the decision about creating an inline extent vs
103 * doing real data extents, marking pages dirty and delalloc as required.
104 */
105 static noinline int dirty_and_release_pages(struct btrfs_trans_handle *trans,
106 struct btrfs_root *root,
107 struct file *file,
108 struct page **pages,
109 size_t num_pages,
110 loff_t pos,
111 size_t write_bytes)
112 {
113 int err = 0;
114 int i;
115 struct inode *inode = fdentry(file)->d_inode;
116 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
117 u64 hint_byte;
118 u64 num_bytes;
119 u64 start_pos;
120 u64 end_of_last_block;
121 u64 end_pos = pos + write_bytes;
122 loff_t isize = i_size_read(inode);
123
124 start_pos = pos & ~((u64)root->sectorsize - 1);
125 num_bytes = (write_bytes + pos - start_pos +
126 root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
127
128 end_of_last_block = start_pos + num_bytes - 1;
129
130 lock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
131 trans = btrfs_join_transaction(root, 1);
132 if (!trans) {
133 err = -ENOMEM;
134 goto out_unlock;
135 }
136 btrfs_set_trans_block_group(trans, inode);
137 hint_byte = 0;
138
139 set_extent_uptodate(io_tree, start_pos, end_of_last_block, GFP_NOFS);
140
141 /* check for reserved extents on each page, we don't want
142 * to reset the delalloc bit on things that already have
143 * extents reserved.
144 */
145 btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block);
146 for (i = 0; i < num_pages; i++) {
147 struct page *p = pages[i];
148 SetPageUptodate(p);
149 ClearPageChecked(p);
150 set_page_dirty(p);
151 }
152 if (end_pos > isize) {
153 i_size_write(inode, end_pos);
154 btrfs_update_inode(trans, root, inode);
155 }
156 err = btrfs_end_transaction(trans, root);
157 out_unlock:
158 unlock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
159 return err;
160 }
161
162 /*
163 * this drops all the extents in the cache that intersect the range
164 * [start, end]. Existing extents are split as required.
165 */
166 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
167 int skip_pinned)
168 {
169 struct extent_map *em;
170 struct extent_map *split = NULL;
171 struct extent_map *split2 = NULL;
172 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
173 u64 len = end - start + 1;
174 int ret;
175 int testend = 1;
176 unsigned long flags;
177 int compressed = 0;
178
179 WARN_ON(end < start);
180 if (end == (u64)-1) {
181 len = (u64)-1;
182 testend = 0;
183 }
184 while (1) {
185 if (!split)
186 split = alloc_extent_map(GFP_NOFS);
187 if (!split2)
188 split2 = alloc_extent_map(GFP_NOFS);
189
190 spin_lock(&em_tree->lock);
191 em = lookup_extent_mapping(em_tree, start, len);
192 if (!em) {
193 spin_unlock(&em_tree->lock);
194 break;
195 }
196 flags = em->flags;
197 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
198 spin_unlock(&em_tree->lock);
199 if (em->start <= start &&
200 (!testend || em->start + em->len >= start + len)) {
201 free_extent_map(em);
202 break;
203 }
204 if (start < em->start) {
205 len = em->start - start;
206 } else {
207 len = start + len - (em->start + em->len);
208 start = em->start + em->len;
209 }
210 free_extent_map(em);
211 continue;
212 }
213 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
214 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
215 remove_extent_mapping(em_tree, em);
216
217 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
218 em->start < start) {
219 split->start = em->start;
220 split->len = start - em->start;
221 split->orig_start = em->orig_start;
222 split->block_start = em->block_start;
223
224 if (compressed)
225 split->block_len = em->block_len;
226 else
227 split->block_len = split->len;
228
229 split->bdev = em->bdev;
230 split->flags = flags;
231 ret = add_extent_mapping(em_tree, split);
232 BUG_ON(ret);
233 free_extent_map(split);
234 split = split2;
235 split2 = NULL;
236 }
237 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
238 testend && em->start + em->len > start + len) {
239 u64 diff = start + len - em->start;
240
241 split->start = start + len;
242 split->len = em->start + em->len - (start + len);
243 split->bdev = em->bdev;
244 split->flags = flags;
245
246 if (compressed) {
247 split->block_len = em->block_len;
248 split->block_start = em->block_start;
249 split->orig_start = em->orig_start;
250 } else {
251 split->block_len = split->len;
252 split->block_start = em->block_start + diff;
253 split->orig_start = split->start;
254 }
255
256 ret = add_extent_mapping(em_tree, split);
257 BUG_ON(ret);
258 free_extent_map(split);
259 split = NULL;
260 }
261 spin_unlock(&em_tree->lock);
262
263 /* once for us */
264 free_extent_map(em);
265 /* once for the tree*/
266 free_extent_map(em);
267 }
268 if (split)
269 free_extent_map(split);
270 if (split2)
271 free_extent_map(split2);
272 return 0;
273 }
274
275 int btrfs_check_file(struct btrfs_root *root, struct inode *inode)
276 {
277 return 0;
278 #if 0
279 struct btrfs_path *path;
280 struct btrfs_key found_key;
281 struct extent_buffer *leaf;
282 struct btrfs_file_extent_item *extent;
283 u64 last_offset = 0;
284 int nritems;
285 int slot;
286 int found_type;
287 int ret;
288 int err = 0;
289 u64 extent_end = 0;
290
291 path = btrfs_alloc_path();
292 ret = btrfs_lookup_file_extent(NULL, root, path, inode->i_ino,
293 last_offset, 0);
294 while (1) {
295 nritems = btrfs_header_nritems(path->nodes[0]);
296 if (path->slots[0] >= nritems) {
297 ret = btrfs_next_leaf(root, path);
298 if (ret)
299 goto out;
300 nritems = btrfs_header_nritems(path->nodes[0]);
301 }
302 slot = path->slots[0];
303 leaf = path->nodes[0];
304 btrfs_item_key_to_cpu(leaf, &found_key, slot);
305 if (found_key.objectid != inode->i_ino)
306 break;
307 if (found_key.type != BTRFS_EXTENT_DATA_KEY)
308 goto out;
309
310 if (found_key.offset < last_offset) {
311 WARN_ON(1);
312 btrfs_print_leaf(root, leaf);
313 printk(KERN_ERR "inode %lu found offset %llu "
314 "expected %llu\n", inode->i_ino,
315 (unsigned long long)found_key.offset,
316 (unsigned long long)last_offset);
317 err = 1;
318 goto out;
319 }
320 extent = btrfs_item_ptr(leaf, slot,
321 struct btrfs_file_extent_item);
322 found_type = btrfs_file_extent_type(leaf, extent);
323 if (found_type == BTRFS_FILE_EXTENT_REG) {
324 extent_end = found_key.offset +
325 btrfs_file_extent_num_bytes(leaf, extent);
326 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
327 struct btrfs_item *item;
328 item = btrfs_item_nr(leaf, slot);
329 extent_end = found_key.offset +
330 btrfs_file_extent_inline_len(leaf, extent);
331 extent_end = (extent_end + root->sectorsize - 1) &
332 ~((u64)root->sectorsize - 1);
333 }
334 last_offset = extent_end;
335 path->slots[0]++;
336 }
337 if (0 && last_offset < inode->i_size) {
338 WARN_ON(1);
339 btrfs_print_leaf(root, leaf);
340 printk(KERN_ERR "inode %lu found offset %llu size %llu\n",
341 inode->i_ino, (unsigned long long)last_offset,
342 (unsigned long long)inode->i_size);
343 err = 1;
344
345 }
346 out:
347 btrfs_free_path(path);
348 return err;
349 #endif
350 }
351
352 /*
353 * this is very complex, but the basic idea is to drop all extents
354 * in the range start - end. hint_block is filled in with a block number
355 * that would be a good hint to the block allocator for this file.
356 *
357 * If an extent intersects the range but is not entirely inside the range
358 * it is either truncated or split. Anything entirely inside the range
359 * is deleted from the tree.
360 *
361 * inline_limit is used to tell this code which offsets in the file to keep
362 * if they contain inline extents.
363 */
364 noinline int btrfs_drop_extents(struct btrfs_trans_handle *trans,
365 struct btrfs_root *root, struct inode *inode,
366 u64 start, u64 end, u64 locked_end,
367 u64 inline_limit, u64 *hint_byte)
368 {
369 u64 extent_end = 0;
370 u64 search_start = start;
371 u64 leaf_start;
372 u64 ram_bytes = 0;
373 u64 orig_parent = 0;
374 u64 disk_bytenr = 0;
375 u64 orig_locked_end = locked_end;
376 u8 compression;
377 u8 encryption;
378 u16 other_encoding = 0;
379 u64 root_gen;
380 u64 root_owner;
381 struct extent_buffer *leaf;
382 struct btrfs_file_extent_item *extent;
383 struct btrfs_path *path;
384 struct btrfs_key key;
385 struct btrfs_file_extent_item old;
386 int keep;
387 int slot;
388 int bookend;
389 int found_type = 0;
390 int found_extent;
391 int found_inline;
392 int recow;
393 int ret;
394
395 inline_limit = 0;
396 btrfs_drop_extent_cache(inode, start, end - 1, 0);
397
398 path = btrfs_alloc_path();
399 if (!path)
400 return -ENOMEM;
401 while (1) {
402 recow = 0;
403 btrfs_release_path(root, path);
404 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
405 search_start, -1);
406 if (ret < 0)
407 goto out;
408 if (ret > 0) {
409 if (path->slots[0] == 0) {
410 ret = 0;
411 goto out;
412 }
413 path->slots[0]--;
414 }
415 next_slot:
416 keep = 0;
417 bookend = 0;
418 found_extent = 0;
419 found_inline = 0;
420 leaf_start = 0;
421 root_gen = 0;
422 root_owner = 0;
423 compression = 0;
424 encryption = 0;
425 extent = NULL;
426 leaf = path->nodes[0];
427 slot = path->slots[0];
428 ret = 0;
429 btrfs_item_key_to_cpu(leaf, &key, slot);
430 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY &&
431 key.offset >= end) {
432 goto out;
433 }
434 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
435 key.objectid != inode->i_ino) {
436 goto out;
437 }
438 if (recow) {
439 search_start = max(key.offset, start);
440 continue;
441 }
442 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
443 extent = btrfs_item_ptr(leaf, slot,
444 struct btrfs_file_extent_item);
445 found_type = btrfs_file_extent_type(leaf, extent);
446 compression = btrfs_file_extent_compression(leaf,
447 extent);
448 encryption = btrfs_file_extent_encryption(leaf,
449 extent);
450 other_encoding = btrfs_file_extent_other_encoding(leaf,
451 extent);
452 if (found_type == BTRFS_FILE_EXTENT_REG ||
453 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
454 extent_end =
455 btrfs_file_extent_disk_bytenr(leaf,
456 extent);
457 if (extent_end)
458 *hint_byte = extent_end;
459
460 extent_end = key.offset +
461 btrfs_file_extent_num_bytes(leaf, extent);
462 ram_bytes = btrfs_file_extent_ram_bytes(leaf,
463 extent);
464 found_extent = 1;
465 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
466 found_inline = 1;
467 extent_end = key.offset +
468 btrfs_file_extent_inline_len(leaf, extent);
469 }
470 } else {
471 extent_end = search_start;
472 }
473
474 /* we found nothing we can drop */
475 if ((!found_extent && !found_inline) ||
476 search_start >= extent_end) {
477 int nextret;
478 u32 nritems;
479 nritems = btrfs_header_nritems(leaf);
480 if (slot >= nritems - 1) {
481 nextret = btrfs_next_leaf(root, path);
482 if (nextret)
483 goto out;
484 recow = 1;
485 } else {
486 path->slots[0]++;
487 }
488 goto next_slot;
489 }
490
491 if (end <= extent_end && start >= key.offset && found_inline)
492 *hint_byte = EXTENT_MAP_INLINE;
493
494 if (found_extent) {
495 read_extent_buffer(leaf, &old, (unsigned long)extent,
496 sizeof(old));
497 root_gen = btrfs_header_generation(leaf);
498 root_owner = btrfs_header_owner(leaf);
499 leaf_start = leaf->start;
500 }
501
502 if (end < extent_end && end >= key.offset) {
503 bookend = 1;
504 if (found_inline && start <= key.offset)
505 keep = 1;
506 }
507
508 if (bookend && found_extent) {
509 if (locked_end < extent_end) {
510 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
511 locked_end, extent_end - 1,
512 GFP_NOFS);
513 if (!ret) {
514 btrfs_release_path(root, path);
515 lock_extent(&BTRFS_I(inode)->io_tree,
516 locked_end, extent_end - 1,
517 GFP_NOFS);
518 locked_end = extent_end;
519 continue;
520 }
521 locked_end = extent_end;
522 }
523 orig_parent = path->nodes[0]->start;
524 disk_bytenr = le64_to_cpu(old.disk_bytenr);
525 if (disk_bytenr != 0) {
526 ret = btrfs_inc_extent_ref(trans, root,
527 disk_bytenr,
528 le64_to_cpu(old.disk_num_bytes),
529 orig_parent, root->root_key.objectid,
530 trans->transid, inode->i_ino);
531 BUG_ON(ret);
532 }
533 }
534
535 if (found_inline) {
536 u64 mask = root->sectorsize - 1;
537 search_start = (extent_end + mask) & ~mask;
538 } else
539 search_start = extent_end;
540
541 /* truncate existing extent */
542 if (start > key.offset) {
543 u64 new_num;
544 u64 old_num;
545 keep = 1;
546 WARN_ON(start & (root->sectorsize - 1));
547 if (found_extent) {
548 new_num = start - key.offset;
549 old_num = btrfs_file_extent_num_bytes(leaf,
550 extent);
551 *hint_byte =
552 btrfs_file_extent_disk_bytenr(leaf,
553 extent);
554 if (btrfs_file_extent_disk_bytenr(leaf,
555 extent)) {
556 inode_sub_bytes(inode, old_num -
557 new_num);
558 }
559 btrfs_set_file_extent_num_bytes(leaf,
560 extent, new_num);
561 btrfs_mark_buffer_dirty(leaf);
562 } else if (key.offset < inline_limit &&
563 (end > extent_end) &&
564 (inline_limit < extent_end)) {
565 u32 new_size;
566 new_size = btrfs_file_extent_calc_inline_size(
567 inline_limit - key.offset);
568 inode_sub_bytes(inode, extent_end -
569 inline_limit);
570 btrfs_set_file_extent_ram_bytes(leaf, extent,
571 new_size);
572 if (!compression && !encryption) {
573 btrfs_truncate_item(trans, root, path,
574 new_size, 1);
575 }
576 }
577 }
578 /* delete the entire extent */
579 if (!keep) {
580 if (found_inline)
581 inode_sub_bytes(inode, extent_end -
582 key.offset);
583 ret = btrfs_del_item(trans, root, path);
584 /* TODO update progress marker and return */
585 BUG_ON(ret);
586 extent = NULL;
587 btrfs_release_path(root, path);
588 /* the extent will be freed later */
589 }
590 if (bookend && found_inline && start <= key.offset) {
591 u32 new_size;
592 new_size = btrfs_file_extent_calc_inline_size(
593 extent_end - end);
594 inode_sub_bytes(inode, end - key.offset);
595 btrfs_set_file_extent_ram_bytes(leaf, extent,
596 new_size);
597 if (!compression && !encryption)
598 ret = btrfs_truncate_item(trans, root, path,
599 new_size, 0);
600 BUG_ON(ret);
601 }
602 /* create bookend, splitting the extent in two */
603 if (bookend && found_extent) {
604 struct btrfs_key ins;
605 ins.objectid = inode->i_ino;
606 ins.offset = end;
607 btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
608
609 btrfs_release_path(root, path);
610 path->leave_spinning = 1;
611 ret = btrfs_insert_empty_item(trans, root, path, &ins,
612 sizeof(*extent));
613 BUG_ON(ret);
614
615 leaf = path->nodes[0];
616 extent = btrfs_item_ptr(leaf, path->slots[0],
617 struct btrfs_file_extent_item);
618 write_extent_buffer(leaf, &old,
619 (unsigned long)extent, sizeof(old));
620
621 btrfs_set_file_extent_compression(leaf, extent,
622 compression);
623 btrfs_set_file_extent_encryption(leaf, extent,
624 encryption);
625 btrfs_set_file_extent_other_encoding(leaf, extent,
626 other_encoding);
627 btrfs_set_file_extent_offset(leaf, extent,
628 le64_to_cpu(old.offset) + end - key.offset);
629 WARN_ON(le64_to_cpu(old.num_bytes) <
630 (extent_end - end));
631 btrfs_set_file_extent_num_bytes(leaf, extent,
632 extent_end - end);
633
634 /*
635 * set the ram bytes to the size of the full extent
636 * before splitting. This is a worst case flag,
637 * but its the best we can do because we don't know
638 * how splitting affects compression
639 */
640 btrfs_set_file_extent_ram_bytes(leaf, extent,
641 ram_bytes);
642 btrfs_set_file_extent_type(leaf, extent, found_type);
643
644 btrfs_unlock_up_safe(path, 1);
645 btrfs_mark_buffer_dirty(path->nodes[0]);
646 btrfs_set_lock_blocking(path->nodes[0]);
647
648 if (disk_bytenr != 0) {
649 ret = btrfs_update_extent_ref(trans, root,
650 disk_bytenr,
651 le64_to_cpu(old.disk_num_bytes),
652 orig_parent,
653 leaf->start,
654 root->root_key.objectid,
655 trans->transid, ins.objectid);
656
657 BUG_ON(ret);
658 }
659 path->leave_spinning = 0;
660 btrfs_release_path(root, path);
661 if (disk_bytenr != 0)
662 inode_add_bytes(inode, extent_end - end);
663 }
664
665 if (found_extent && !keep) {
666 u64 old_disk_bytenr = le64_to_cpu(old.disk_bytenr);
667
668 if (old_disk_bytenr != 0) {
669 inode_sub_bytes(inode,
670 le64_to_cpu(old.num_bytes));
671 ret = btrfs_free_extent(trans, root,
672 old_disk_bytenr,
673 le64_to_cpu(old.disk_num_bytes),
674 leaf_start, root_owner,
675 root_gen, key.objectid, 0);
676 BUG_ON(ret);
677 *hint_byte = old_disk_bytenr;
678 }
679 }
680
681 if (search_start >= end) {
682 ret = 0;
683 goto out;
684 }
685 }
686 out:
687 btrfs_free_path(path);
688 if (locked_end > orig_locked_end) {
689 unlock_extent(&BTRFS_I(inode)->io_tree, orig_locked_end,
690 locked_end - 1, GFP_NOFS);
691 }
692 btrfs_check_file(root, inode);
693 return ret;
694 }
695
696 static int extent_mergeable(struct extent_buffer *leaf, int slot,
697 u64 objectid, u64 bytenr, u64 *start, u64 *end)
698 {
699 struct btrfs_file_extent_item *fi;
700 struct btrfs_key key;
701 u64 extent_end;
702
703 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
704 return 0;
705
706 btrfs_item_key_to_cpu(leaf, &key, slot);
707 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
708 return 0;
709
710 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
711 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
712 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
713 btrfs_file_extent_compression(leaf, fi) ||
714 btrfs_file_extent_encryption(leaf, fi) ||
715 btrfs_file_extent_other_encoding(leaf, fi))
716 return 0;
717
718 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
719 if ((*start && *start != key.offset) || (*end && *end != extent_end))
720 return 0;
721
722 *start = key.offset;
723 *end = extent_end;
724 return 1;
725 }
726
727 /*
728 * Mark extent in the range start - end as written.
729 *
730 * This changes extent type from 'pre-allocated' to 'regular'. If only
731 * part of extent is marked as written, the extent will be split into
732 * two or three.
733 */
734 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
735 struct btrfs_root *root,
736 struct inode *inode, u64 start, u64 end)
737 {
738 struct extent_buffer *leaf;
739 struct btrfs_path *path;
740 struct btrfs_file_extent_item *fi;
741 struct btrfs_key key;
742 u64 bytenr;
743 u64 num_bytes;
744 u64 extent_end;
745 u64 extent_offset;
746 u64 other_start;
747 u64 other_end;
748 u64 split = start;
749 u64 locked_end = end;
750 u64 orig_parent;
751 int extent_type;
752 int split_end = 1;
753 int ret;
754
755 btrfs_drop_extent_cache(inode, start, end - 1, 0);
756
757 path = btrfs_alloc_path();
758 BUG_ON(!path);
759 again:
760 key.objectid = inode->i_ino;
761 key.type = BTRFS_EXTENT_DATA_KEY;
762 if (split == start)
763 key.offset = split;
764 else
765 key.offset = split - 1;
766
767 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
768 if (ret > 0 && path->slots[0] > 0)
769 path->slots[0]--;
770
771 leaf = path->nodes[0];
772 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
773 BUG_ON(key.objectid != inode->i_ino ||
774 key.type != BTRFS_EXTENT_DATA_KEY);
775 fi = btrfs_item_ptr(leaf, path->slots[0],
776 struct btrfs_file_extent_item);
777 extent_type = btrfs_file_extent_type(leaf, fi);
778 BUG_ON(extent_type != BTRFS_FILE_EXTENT_PREALLOC);
779 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
780 BUG_ON(key.offset > start || extent_end < end);
781
782 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
783 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
784 extent_offset = btrfs_file_extent_offset(leaf, fi);
785
786 if (key.offset == start)
787 split = end;
788
789 if (key.offset == start && extent_end == end) {
790 int del_nr = 0;
791 int del_slot = 0;
792 u64 leaf_owner = btrfs_header_owner(leaf);
793 u64 leaf_gen = btrfs_header_generation(leaf);
794 other_start = end;
795 other_end = 0;
796 if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
797 bytenr, &other_start, &other_end)) {
798 extent_end = other_end;
799 del_slot = path->slots[0] + 1;
800 del_nr++;
801 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
802 leaf->start, leaf_owner,
803 leaf_gen, inode->i_ino, 0);
804 BUG_ON(ret);
805 }
806 other_start = 0;
807 other_end = start;
808 if (extent_mergeable(leaf, path->slots[0] - 1, inode->i_ino,
809 bytenr, &other_start, &other_end)) {
810 key.offset = other_start;
811 del_slot = path->slots[0];
812 del_nr++;
813 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
814 leaf->start, leaf_owner,
815 leaf_gen, inode->i_ino, 0);
816 BUG_ON(ret);
817 }
818 split_end = 0;
819 if (del_nr == 0) {
820 btrfs_set_file_extent_type(leaf, fi,
821 BTRFS_FILE_EXTENT_REG);
822 goto done;
823 }
824
825 fi = btrfs_item_ptr(leaf, del_slot - 1,
826 struct btrfs_file_extent_item);
827 btrfs_set_file_extent_type(leaf, fi, BTRFS_FILE_EXTENT_REG);
828 btrfs_set_file_extent_num_bytes(leaf, fi,
829 extent_end - key.offset);
830 btrfs_mark_buffer_dirty(leaf);
831
832 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
833 BUG_ON(ret);
834 goto release;
835 } else if (split == start) {
836 if (locked_end < extent_end) {
837 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
838 locked_end, extent_end - 1, GFP_NOFS);
839 if (!ret) {
840 btrfs_release_path(root, path);
841 lock_extent(&BTRFS_I(inode)->io_tree,
842 locked_end, extent_end - 1, GFP_NOFS);
843 locked_end = extent_end;
844 goto again;
845 }
846 locked_end = extent_end;
847 }
848 btrfs_set_file_extent_num_bytes(leaf, fi, split - key.offset);
849 extent_offset += split - key.offset;
850 } else {
851 BUG_ON(key.offset != start);
852 btrfs_set_file_extent_offset(leaf, fi, extent_offset +
853 split - key.offset);
854 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - split);
855 key.offset = split;
856 btrfs_set_item_key_safe(trans, root, path, &key);
857 extent_end = split;
858 }
859
860 if (extent_end == end) {
861 split_end = 0;
862 extent_type = BTRFS_FILE_EXTENT_REG;
863 }
864 if (extent_end == end && split == start) {
865 other_start = end;
866 other_end = 0;
867 if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
868 bytenr, &other_start, &other_end)) {
869 path->slots[0]++;
870 fi = btrfs_item_ptr(leaf, path->slots[0],
871 struct btrfs_file_extent_item);
872 key.offset = split;
873 btrfs_set_item_key_safe(trans, root, path, &key);
874 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
875 btrfs_set_file_extent_num_bytes(leaf, fi,
876 other_end - split);
877 goto done;
878 }
879 }
880 if (extent_end == end && split == end) {
881 other_start = 0;
882 other_end = start;
883 if (extent_mergeable(leaf, path->slots[0] - 1 , inode->i_ino,
884 bytenr, &other_start, &other_end)) {
885 path->slots[0]--;
886 fi = btrfs_item_ptr(leaf, path->slots[0],
887 struct btrfs_file_extent_item);
888 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end -
889 other_start);
890 goto done;
891 }
892 }
893
894 btrfs_mark_buffer_dirty(leaf);
895
896 orig_parent = leaf->start;
897 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes,
898 orig_parent, root->root_key.objectid,
899 trans->transid, inode->i_ino);
900 BUG_ON(ret);
901 btrfs_release_path(root, path);
902
903 key.offset = start;
904 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*fi));
905 BUG_ON(ret);
906
907 leaf = path->nodes[0];
908 fi = btrfs_item_ptr(leaf, path->slots[0],
909 struct btrfs_file_extent_item);
910 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
911 btrfs_set_file_extent_type(leaf, fi, extent_type);
912 btrfs_set_file_extent_disk_bytenr(leaf, fi, bytenr);
913 btrfs_set_file_extent_disk_num_bytes(leaf, fi, num_bytes);
914 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
915 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - key.offset);
916 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
917 btrfs_set_file_extent_compression(leaf, fi, 0);
918 btrfs_set_file_extent_encryption(leaf, fi, 0);
919 btrfs_set_file_extent_other_encoding(leaf, fi, 0);
920
921 if (orig_parent != leaf->start) {
922 ret = btrfs_update_extent_ref(trans, root, bytenr, num_bytes,
923 orig_parent, leaf->start,
924 root->root_key.objectid,
925 trans->transid, inode->i_ino);
926 BUG_ON(ret);
927 }
928 done:
929 btrfs_mark_buffer_dirty(leaf);
930
931 release:
932 btrfs_release_path(root, path);
933 if (split_end && split == start) {
934 split = end;
935 goto again;
936 }
937 if (locked_end > end) {
938 unlock_extent(&BTRFS_I(inode)->io_tree, end, locked_end - 1,
939 GFP_NOFS);
940 }
941 btrfs_free_path(path);
942 return 0;
943 }
944
945 /*
946 * this gets pages into the page cache and locks them down, it also properly
947 * waits for data=ordered extents to finish before allowing the pages to be
948 * modified.
949 */
950 static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
951 struct page **pages, size_t num_pages,
952 loff_t pos, unsigned long first_index,
953 unsigned long last_index, size_t write_bytes)
954 {
955 int i;
956 unsigned long index = pos >> PAGE_CACHE_SHIFT;
957 struct inode *inode = fdentry(file)->d_inode;
958 int err = 0;
959 u64 start_pos;
960 u64 last_pos;
961
962 start_pos = pos & ~((u64)root->sectorsize - 1);
963 last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
964
965 if (start_pos > inode->i_size) {
966 err = btrfs_cont_expand(inode, start_pos);
967 if (err)
968 return err;
969 }
970
971 memset(pages, 0, num_pages * sizeof(struct page *));
972 again:
973 for (i = 0; i < num_pages; i++) {
974 pages[i] = grab_cache_page(inode->i_mapping, index + i);
975 if (!pages[i]) {
976 err = -ENOMEM;
977 BUG_ON(1);
978 }
979 wait_on_page_writeback(pages[i]);
980 }
981 if (start_pos < inode->i_size) {
982 struct btrfs_ordered_extent *ordered;
983 lock_extent(&BTRFS_I(inode)->io_tree,
984 start_pos, last_pos - 1, GFP_NOFS);
985 ordered = btrfs_lookup_first_ordered_extent(inode,
986 last_pos - 1);
987 if (ordered &&
988 ordered->file_offset + ordered->len > start_pos &&
989 ordered->file_offset < last_pos) {
990 btrfs_put_ordered_extent(ordered);
991 unlock_extent(&BTRFS_I(inode)->io_tree,
992 start_pos, last_pos - 1, GFP_NOFS);
993 for (i = 0; i < num_pages; i++) {
994 unlock_page(pages[i]);
995 page_cache_release(pages[i]);
996 }
997 btrfs_wait_ordered_range(inode, start_pos,
998 last_pos - start_pos);
999 goto again;
1000 }
1001 if (ordered)
1002 btrfs_put_ordered_extent(ordered);
1003
1004 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
1005 last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC,
1006 GFP_NOFS);
1007 unlock_extent(&BTRFS_I(inode)->io_tree,
1008 start_pos, last_pos - 1, GFP_NOFS);
1009 }
1010 for (i = 0; i < num_pages; i++) {
1011 clear_page_dirty_for_io(pages[i]);
1012 set_page_extent_mapped(pages[i]);
1013 WARN_ON(!PageLocked(pages[i]));
1014 }
1015 return 0;
1016 }
1017
1018 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
1019 size_t count, loff_t *ppos)
1020 {
1021 loff_t pos;
1022 loff_t start_pos;
1023 ssize_t num_written = 0;
1024 ssize_t err = 0;
1025 int ret = 0;
1026 struct inode *inode = fdentry(file)->d_inode;
1027 struct btrfs_root *root = BTRFS_I(inode)->root;
1028 struct page **pages = NULL;
1029 int nrptrs;
1030 struct page *pinned[2];
1031 unsigned long first_index;
1032 unsigned long last_index;
1033 int will_write;
1034
1035 will_write = ((file->f_flags & O_SYNC) || IS_SYNC(inode) ||
1036 (file->f_flags & O_DIRECT));
1037
1038 nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
1039 PAGE_CACHE_SIZE / (sizeof(struct page *)));
1040 pinned[0] = NULL;
1041 pinned[1] = NULL;
1042
1043 pos = *ppos;
1044 start_pos = pos;
1045
1046 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1047 current->backing_dev_info = inode->i_mapping->backing_dev_info;
1048 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1049 if (err)
1050 goto out_nolock;
1051 if (count == 0)
1052 goto out_nolock;
1053
1054 err = file_remove_suid(file);
1055 if (err)
1056 goto out_nolock;
1057 file_update_time(file);
1058
1059 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
1060
1061 mutex_lock(&inode->i_mutex);
1062 BTRFS_I(inode)->sequence++;
1063 first_index = pos >> PAGE_CACHE_SHIFT;
1064 last_index = (pos + count) >> PAGE_CACHE_SHIFT;
1065
1066 /*
1067 * there are lots of better ways to do this, but this code
1068 * makes sure the first and last page in the file range are
1069 * up to date and ready for cow
1070 */
1071 if ((pos & (PAGE_CACHE_SIZE - 1))) {
1072 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
1073 if (!PageUptodate(pinned[0])) {
1074 ret = btrfs_readpage(NULL, pinned[0]);
1075 BUG_ON(ret);
1076 wait_on_page_locked(pinned[0]);
1077 } else {
1078 unlock_page(pinned[0]);
1079 }
1080 }
1081 if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
1082 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
1083 if (!PageUptodate(pinned[1])) {
1084 ret = btrfs_readpage(NULL, pinned[1]);
1085 BUG_ON(ret);
1086 wait_on_page_locked(pinned[1]);
1087 } else {
1088 unlock_page(pinned[1]);
1089 }
1090 }
1091
1092 while (count > 0) {
1093 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1094 size_t write_bytes = min(count, nrptrs *
1095 (size_t)PAGE_CACHE_SIZE -
1096 offset);
1097 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
1098 PAGE_CACHE_SHIFT;
1099
1100 WARN_ON(num_pages > nrptrs);
1101 memset(pages, 0, sizeof(struct page *) * nrptrs);
1102
1103 ret = btrfs_check_data_free_space(root, inode, write_bytes);
1104 if (ret)
1105 goto out;
1106
1107 ret = prepare_pages(root, file, pages, num_pages,
1108 pos, first_index, last_index,
1109 write_bytes);
1110 if (ret) {
1111 btrfs_free_reserved_data_space(root, inode,
1112 write_bytes);
1113 goto out;
1114 }
1115
1116 ret = btrfs_copy_from_user(pos, num_pages,
1117 write_bytes, pages, buf);
1118 if (ret) {
1119 btrfs_free_reserved_data_space(root, inode,
1120 write_bytes);
1121 btrfs_drop_pages(pages, num_pages);
1122 goto out;
1123 }
1124
1125 ret = dirty_and_release_pages(NULL, root, file, pages,
1126 num_pages, pos, write_bytes);
1127 btrfs_drop_pages(pages, num_pages);
1128 if (ret) {
1129 btrfs_free_reserved_data_space(root, inode,
1130 write_bytes);
1131 goto out;
1132 }
1133
1134 if (will_write) {
1135 btrfs_fdatawrite_range(inode->i_mapping, pos,
1136 pos + write_bytes - 1,
1137 WB_SYNC_ALL);
1138 } else {
1139 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
1140 num_pages);
1141 if (num_pages <
1142 (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1143 btrfs_btree_balance_dirty(root, 1);
1144 btrfs_throttle(root);
1145 }
1146
1147 buf += write_bytes;
1148 count -= write_bytes;
1149 pos += write_bytes;
1150 num_written += write_bytes;
1151
1152 cond_resched();
1153 }
1154 out:
1155 mutex_unlock(&inode->i_mutex);
1156 if (ret)
1157 err = ret;
1158
1159 out_nolock:
1160 kfree(pages);
1161 if (pinned[0])
1162 page_cache_release(pinned[0]);
1163 if (pinned[1])
1164 page_cache_release(pinned[1]);
1165 *ppos = pos;
1166
1167 /*
1168 * we want to make sure fsync finds this change
1169 * but we haven't joined a transaction running right now.
1170 *
1171 * Later on, someone is sure to update the inode and get the
1172 * real transid recorded.
1173 *
1174 * We set last_trans now to the fs_info generation + 1,
1175 * this will either be one more than the running transaction
1176 * or the generation used for the next transaction if there isn't
1177 * one running right now.
1178 */
1179 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1180
1181 if (num_written > 0 && will_write) {
1182 struct btrfs_trans_handle *trans;
1183
1184 err = btrfs_wait_ordered_range(inode, start_pos, num_written);
1185 if (err)
1186 num_written = err;
1187
1188 if ((file->f_flags & O_SYNC) || IS_SYNC(inode)) {
1189 trans = btrfs_start_transaction(root, 1);
1190 ret = btrfs_log_dentry_safe(trans, root,
1191 file->f_dentry);
1192 if (ret == 0) {
1193 ret = btrfs_sync_log(trans, root);
1194 if (ret == 0)
1195 btrfs_end_transaction(trans, root);
1196 else
1197 btrfs_commit_transaction(trans, root);
1198 } else {
1199 btrfs_commit_transaction(trans, root);
1200 }
1201 }
1202 if (file->f_flags & O_DIRECT) {
1203 invalidate_mapping_pages(inode->i_mapping,
1204 start_pos >> PAGE_CACHE_SHIFT,
1205 (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
1206 }
1207 }
1208 current->backing_dev_info = NULL;
1209 return num_written ? num_written : err;
1210 }
1211
1212 int btrfs_release_file(struct inode *inode, struct file *filp)
1213 {
1214 /*
1215 * ordered_data_close is set by settattr when we are about to truncate
1216 * a file from a non-zero size to a zero size. This tries to
1217 * flush down new bytes that may have been written if the
1218 * application were using truncate to replace a file in place.
1219 */
1220 if (BTRFS_I(inode)->ordered_data_close) {
1221 BTRFS_I(inode)->ordered_data_close = 0;
1222 btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode);
1223 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1224 filemap_flush(inode->i_mapping);
1225 }
1226 if (filp->private_data)
1227 btrfs_ioctl_trans_end(filp);
1228 return 0;
1229 }
1230
1231 /*
1232 * fsync call for both files and directories. This logs the inode into
1233 * the tree log instead of forcing full commits whenever possible.
1234 *
1235 * It needs to call filemap_fdatawait so that all ordered extent updates are
1236 * in the metadata btree are up to date for copying to the log.
1237 *
1238 * It drops the inode mutex before doing the tree log commit. This is an
1239 * important optimization for directories because holding the mutex prevents
1240 * new operations on the dir while we write to disk.
1241 */
1242 int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
1243 {
1244 struct inode *inode = dentry->d_inode;
1245 struct btrfs_root *root = BTRFS_I(inode)->root;
1246 int ret = 0;
1247 struct btrfs_trans_handle *trans;
1248
1249 /*
1250 * check the transaction that last modified this inode
1251 * and see if its already been committed
1252 */
1253 if (!BTRFS_I(inode)->last_trans)
1254 goto out;
1255
1256 mutex_lock(&root->fs_info->trans_mutex);
1257 if (BTRFS_I(inode)->last_trans <=
1258 root->fs_info->last_trans_committed) {
1259 BTRFS_I(inode)->last_trans = 0;
1260 mutex_unlock(&root->fs_info->trans_mutex);
1261 goto out;
1262 }
1263 mutex_unlock(&root->fs_info->trans_mutex);
1264
1265 root->log_batch++;
1266 filemap_fdatawrite(inode->i_mapping);
1267 btrfs_wait_ordered_range(inode, 0, (u64)-1);
1268 root->log_batch++;
1269
1270 /*
1271 * ok we haven't committed the transaction yet, lets do a commit
1272 */
1273 if (file && file->private_data)
1274 btrfs_ioctl_trans_end(file);
1275
1276 trans = btrfs_start_transaction(root, 1);
1277 if (!trans) {
1278 ret = -ENOMEM;
1279 goto out;
1280 }
1281
1282 ret = btrfs_log_dentry_safe(trans, root, dentry);
1283 if (ret < 0)
1284 goto out;
1285
1286 /* we've logged all the items and now have a consistent
1287 * version of the file in the log. It is possible that
1288 * someone will come in and modify the file, but that's
1289 * fine because the log is consistent on disk, and we
1290 * have references to all of the file's extents
1291 *
1292 * It is possible that someone will come in and log the
1293 * file again, but that will end up using the synchronization
1294 * inside btrfs_sync_log to keep things safe.
1295 */
1296 mutex_unlock(&dentry->d_inode->i_mutex);
1297
1298 if (ret > 0) {
1299 ret = btrfs_commit_transaction(trans, root);
1300 } else {
1301 ret = btrfs_sync_log(trans, root);
1302 if (ret == 0)
1303 ret = btrfs_end_transaction(trans, root);
1304 else
1305 ret = btrfs_commit_transaction(trans, root);
1306 }
1307 mutex_lock(&dentry->d_inode->i_mutex);
1308 out:
1309 return ret > 0 ? EIO : ret;
1310 }
1311
1312 static struct vm_operations_struct btrfs_file_vm_ops = {
1313 .fault = filemap_fault,
1314 .page_mkwrite = btrfs_page_mkwrite,
1315 };
1316
1317 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
1318 {
1319 vma->vm_ops = &btrfs_file_vm_ops;
1320 file_accessed(filp);
1321 return 0;
1322 }
1323
1324 struct file_operations btrfs_file_operations = {
1325 .llseek = generic_file_llseek,
1326 .read = do_sync_read,
1327 .aio_read = generic_file_aio_read,
1328 .splice_read = generic_file_splice_read,
1329 .write = btrfs_file_write,
1330 .mmap = btrfs_file_mmap,
1331 .open = generic_file_open,
1332 .release = btrfs_release_file,
1333 .fsync = btrfs_sync_file,
1334 .unlocked_ioctl = btrfs_ioctl,
1335 #ifdef CONFIG_COMPAT
1336 .compat_ioctl = btrfs_ioctl,
1337 #endif
1338 };
Linux kernel - Deliverables
This page took 0.065584 seconds and 5 git commands to generate.