projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Btrfs: Fix typo in clear_state_cb
[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 <linux/version.h>
33 #include "ctree.h"
34 #include "disk-io.h"
35 #include "transaction.h"
36 #include "btrfs_inode.h"
37 #include "ioctl.h"
38 #include "print-tree.h"
39 #include "tree-log.h"
40 #include "locking.h"
41 #include "compat.h"
42
43
44 /* simple helper to fault in pages and copy. This should go away
45 * and be replaced with calls into generic code.
46 */
47 static int noinline btrfs_copy_from_user(loff_t pos, int num_pages,
48 int write_bytes,
49 struct page **prepared_pages,
50 const char __user * buf)
51 {
52 long page_fault = 0;
53 int i;
54 int offset = pos & (PAGE_CACHE_SIZE - 1);
55
56 for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
57 size_t count = min_t(size_t,
58 PAGE_CACHE_SIZE - offset, write_bytes);
59 struct page *page = prepared_pages[i];
60 fault_in_pages_readable(buf, count);
61
62 /* Copy data from userspace to the current page */
63 kmap(page);
64 page_fault = __copy_from_user(page_address(page) + offset,
65 buf, count);
66 /* Flush processor's dcache for this page */
67 flush_dcache_page(page);
68 kunmap(page);
69 buf += count;
70 write_bytes -= count;
71
72 if (page_fault)
73 break;
74 }
75 return page_fault ? -EFAULT : 0;
76 }
77
78 /*
79 * unlocks pages after btrfs_file_write is done with them
80 */
81 static void noinline btrfs_drop_pages(struct page **pages, size_t num_pages)
82 {
83 size_t i;
84 for (i = 0; i < num_pages; i++) {
85 if (!pages[i])
86 break;
87 /* page checked is some magic around finding pages that
88 * have been modified without going through btrfs_set_page_dirty
89 * clear it here
90 */
91 ClearPageChecked(pages[i]);
92 unlock_page(pages[i]);
93 mark_page_accessed(pages[i]);
94 page_cache_release(pages[i]);
95 }
96 }
97
98 /*
99 * after copy_from_user, pages need to be dirtied and we need to make
100 * sure holes are created between the current EOF and the start of
101 * any next extents (if required).
102 *
103 * this also makes the decision about creating an inline extent vs
104 * doing real data extents, marking pages dirty and delalloc as required.
105 */
106 static int noinline dirty_and_release_pages(struct btrfs_trans_handle *trans,
107 struct btrfs_root *root,
108 struct file *file,
109 struct page **pages,
110 size_t num_pages,
111 loff_t pos,
112 size_t write_bytes)
113 {
114 int err = 0;
115 int i;
116 struct inode *inode = fdentry(file)->d_inode;
117 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
118 u64 hint_byte;
119 u64 num_bytes;
120 u64 start_pos;
121 u64 end_of_last_block;
122 u64 end_pos = pos + write_bytes;
123 loff_t isize = i_size_read(inode);
124
125 start_pos = pos & ~((u64)root->sectorsize - 1);
126 num_bytes = (write_bytes + pos - start_pos +
127 root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
128
129 end_of_last_block = start_pos + num_bytes - 1;
130
131 lock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
132 trans = btrfs_join_transaction(root, 1);
133 if (!trans) {
134 err = -ENOMEM;
135 goto out_unlock;
136 }
137 btrfs_set_trans_block_group(trans, inode);
138 hint_byte = 0;
139
140 if ((end_of_last_block & 4095) == 0) {
141 printk("strange end of last %Lu %zu %Lu\n", start_pos, write_bytes, end_of_last_block);
142 }
143 set_extent_uptodate(io_tree, start_pos, end_of_last_block, GFP_NOFS);
144
145 /* check for reserved extents on each page, we don't want
146 * to reset the delalloc bit on things that already have
147 * extents reserved.
148 */
149 btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block);
150 for (i = 0; i < num_pages; i++) {
151 struct page *p = pages[i];
152 SetPageUptodate(p);
153 ClearPageChecked(p);
154 set_page_dirty(p);
155 }
156 if (end_pos > isize) {
157 i_size_write(inode, end_pos);
158 btrfs_update_inode(trans, root, inode);
159 }
160 err = btrfs_end_transaction(trans, root);
161 out_unlock:
162 unlock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
163 return err;
164 }
165
166 /*
167 * this drops all the extents in the cache that intersect the range
168 * [start, end]. Existing extents are split as required.
169 */
170 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
171 int skip_pinned)
172 {
173 struct extent_map *em;
174 struct extent_map *split = NULL;
175 struct extent_map *split2 = NULL;
176 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
177 u64 len = end - start + 1;
178 int ret;
179 int testend = 1;
180 unsigned long flags;
181 int compressed = 0;
182
183 WARN_ON(end < start);
184 if (end == (u64)-1) {
185 len = (u64)-1;
186 testend = 0;
187 }
188 while(1) {
189 if (!split)
190 split = alloc_extent_map(GFP_NOFS);
191 if (!split2)
192 split2 = alloc_extent_map(GFP_NOFS);
193
194 spin_lock(&em_tree->lock);
195 em = lookup_extent_mapping(em_tree, start, len);
196 if (!em) {
197 spin_unlock(&em_tree->lock);
198 break;
199 }
200 flags = em->flags;
201 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
202 spin_unlock(&em_tree->lock);
203 if (em->start <= start &&
204 (!testend || em->start + em->len >= start + len)) {
205 free_extent_map(em);
206 break;
207 }
208 if (start < em->start) {
209 len = em->start - start;
210 } else {
211 len = start + len - (em->start + em->len);
212 start = em->start + em->len;
213 }
214 free_extent_map(em);
215 continue;
216 }
217 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
218 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
219 remove_extent_mapping(em_tree, em);
220
221 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
222 em->start < start) {
223 split->start = em->start;
224 split->len = start - em->start;
225 split->orig_start = em->orig_start;
226 split->block_start = em->block_start;
227
228 if (compressed)
229 split->block_len = em->block_len;
230 else
231 split->block_len = split->len;
232
233 split->bdev = em->bdev;
234 split->flags = flags;
235 ret = add_extent_mapping(em_tree, split);
236 BUG_ON(ret);
237 free_extent_map(split);
238 split = split2;
239 split2 = NULL;
240 }
241 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
242 testend && em->start + em->len > start + len) {
243 u64 diff = start + len - em->start;
244
245 split->start = start + len;
246 split->len = em->start + em->len - (start + len);
247 split->bdev = em->bdev;
248 split->flags = flags;
249
250 if (compressed) {
251 split->block_len = em->block_len;
252 split->block_start = em->block_start;
253 split->orig_start = em->orig_start;
254 } else {
255 split->block_len = split->len;
256 split->block_start = em->block_start + diff;
257 split->orig_start = split->start;
258 }
259
260 ret = add_extent_mapping(em_tree, split);
261 BUG_ON(ret);
262 free_extent_map(split);
263 split = NULL;
264 }
265 spin_unlock(&em_tree->lock);
266
267 /* once for us */
268 free_extent_map(em);
269 /* once for the tree*/
270 free_extent_map(em);
271 }
272 if (split)
273 free_extent_map(split);
274 if (split2)
275 free_extent_map(split2);
276 return 0;
277 }
278
279 int btrfs_check_file(struct btrfs_root *root, struct inode *inode)
280 {
281 return 0;
282 #if 0
283 struct btrfs_path *path;
284 struct btrfs_key found_key;
285 struct extent_buffer *leaf;
286 struct btrfs_file_extent_item *extent;
287 u64 last_offset = 0;
288 int nritems;
289 int slot;
290 int found_type;
291 int ret;
292 int err = 0;
293 u64 extent_end = 0;
294
295 path = btrfs_alloc_path();
296 ret = btrfs_lookup_file_extent(NULL, root, path, inode->i_ino,
297 last_offset, 0);
298 while(1) {
299 nritems = btrfs_header_nritems(path->nodes[0]);
300 if (path->slots[0] >= nritems) {
301 ret = btrfs_next_leaf(root, path);
302 if (ret)
303 goto out;
304 nritems = btrfs_header_nritems(path->nodes[0]);
305 }
306 slot = path->slots[0];
307 leaf = path->nodes[0];
308 btrfs_item_key_to_cpu(leaf, &found_key, slot);
309 if (found_key.objectid != inode->i_ino)
310 break;
311 if (found_key.type != BTRFS_EXTENT_DATA_KEY)
312 goto out;
313
314 if (found_key.offset < last_offset) {
315 WARN_ON(1);
316 btrfs_print_leaf(root, leaf);
317 printk("inode %lu found offset %Lu expected %Lu\n",
318 inode->i_ino, found_key.offset, last_offset);
319 err = 1;
320 goto out;
321 }
322 extent = btrfs_item_ptr(leaf, slot,
323 struct btrfs_file_extent_item);
324 found_type = btrfs_file_extent_type(leaf, extent);
325 if (found_type == BTRFS_FILE_EXTENT_REG) {
326 extent_end = found_key.offset +
327 btrfs_file_extent_num_bytes(leaf, extent);
328 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
329 struct btrfs_item *item;
330 item = btrfs_item_nr(leaf, slot);
331 extent_end = found_key.offset +
332 btrfs_file_extent_inline_len(leaf, extent);
333 extent_end = (extent_end + root->sectorsize - 1) &
334 ~((u64)root->sectorsize -1 );
335 }
336 last_offset = extent_end;
337 path->slots[0]++;
338 }
339 if (0 && last_offset < inode->i_size) {
340 WARN_ON(1);
341 btrfs_print_leaf(root, leaf);
342 printk("inode %lu found offset %Lu size %Lu\n", inode->i_ino,
343 last_offset, inode->i_size);
344 err = 1;
345
346 }
347 out:
348 btrfs_free_path(path);
349 return err;
350 #endif
351 }
352
353 /*
354 * this is very complex, but the basic idea is to drop all extents
355 * in the range start - end. hint_block is filled in with a block number
356 * that would be a good hint to the block allocator for this file.
357 *
358 * If an extent intersects the range but is not entirely inside the range
359 * it is either truncated or split. Anything entirely inside the range
360 * is deleted from the tree.
361 *
362 * inline_limit is used to tell this code which offsets in the file to keep
363 * if they contain inline extents.
364 */
365 int noinline btrfs_drop_extents(struct btrfs_trans_handle *trans,
366 struct btrfs_root *root, struct inode *inode,
367 u64 start, u64 end, u64 inline_limit, u64 *hint_byte)
368 {
369 u64 extent_end = 0;
370 u64 locked_end = end;
371 u64 search_start = start;
372 u64 leaf_start;
373 u64 ram_bytes = 0;
374 u64 orig_parent = 0;
375 u64 disk_bytenr = 0;
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 if (!compression && !encryption) {
560 btrfs_set_file_extent_ram_bytes(leaf,
561 extent, new_num);
562 }
563 btrfs_set_file_extent_num_bytes(leaf,
564 extent, new_num);
565 btrfs_mark_buffer_dirty(leaf);
566 } else if (key.offset < inline_limit &&
567 (end > extent_end) &&
568 (inline_limit < extent_end)) {
569 u32 new_size;
570 new_size = btrfs_file_extent_calc_inline_size(
571 inline_limit - key.offset);
572 inode_sub_bytes(inode, extent_end -
573 inline_limit);
574 btrfs_set_file_extent_ram_bytes(leaf, extent,
575 new_size);
576 if (!compression && !encryption) {
577 btrfs_truncate_item(trans, root, path,
578 new_size, 1);
579 }
580 }
581 }
582 /* delete the entire extent */
583 if (!keep) {
584 if (found_inline)
585 inode_sub_bytes(inode, extent_end -
586 key.offset);
587 ret = btrfs_del_item(trans, root, path);
588 /* TODO update progress marker and return */
589 BUG_ON(ret);
590 extent = NULL;
591 btrfs_release_path(root, path);
592 /* the extent will be freed later */
593 }
594 if (bookend && found_inline && start <= key.offset) {
595 u32 new_size;
596 new_size = btrfs_file_extent_calc_inline_size(
597 extent_end - end);
598 inode_sub_bytes(inode, end - key.offset);
599 btrfs_set_file_extent_ram_bytes(leaf, extent,
600 new_size);
601 if (!compression && !encryption)
602 ret = btrfs_truncate_item(trans, root, path,
603 new_size, 0);
604 BUG_ON(ret);
605 }
606 /* create bookend, splitting the extent in two */
607 if (bookend && found_extent) {
608 struct btrfs_key ins;
609 ins.objectid = inode->i_ino;
610 ins.offset = end;
611 btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
612
613 btrfs_release_path(root, path);
614 ret = btrfs_insert_empty_item(trans, root, path, &ins,
615 sizeof(*extent));
616 BUG_ON(ret);
617
618 leaf = path->nodes[0];
619 extent = btrfs_item_ptr(leaf, path->slots[0],
620 struct btrfs_file_extent_item);
621 write_extent_buffer(leaf, &old,
622 (unsigned long)extent, sizeof(old));
623
624 btrfs_set_file_extent_compression(leaf, extent,
625 compression);
626 btrfs_set_file_extent_encryption(leaf, extent,
627 encryption);
628 btrfs_set_file_extent_other_encoding(leaf, extent,
629 other_encoding);
630 btrfs_set_file_extent_offset(leaf, extent,
631 le64_to_cpu(old.offset) + end - key.offset);
632 WARN_ON(le64_to_cpu(old.num_bytes) <
633 (extent_end - end));
634 btrfs_set_file_extent_num_bytes(leaf, extent,
635 extent_end - end);
636
637 /*
638 * set the ram bytes to the size of the full extent
639 * before splitting. This is a worst case flag,
640 * but its the best we can do because we don't know
641 * how splitting affects compression
642 */
643 btrfs_set_file_extent_ram_bytes(leaf, extent,
644 ram_bytes);
645 btrfs_set_file_extent_type(leaf, extent, found_type);
646
647 btrfs_mark_buffer_dirty(path->nodes[0]);
648
649 if (disk_bytenr != 0) {
650 ret = btrfs_update_extent_ref(trans, root,
651 disk_bytenr, orig_parent,
652 leaf->start,
653 root->root_key.objectid,
654 trans->transid, ins.objectid);
655
656 BUG_ON(ret);
657 }
658 btrfs_release_path(root, path);
659 if (disk_bytenr != 0) {
660 inode_add_bytes(inode, extent_end - end);
661 }
662 }
663
664 if (found_extent && !keep) {
665 u64 old_disk_bytenr = le64_to_cpu(old.disk_bytenr);
666
667 if (old_disk_bytenr != 0) {
668 inode_sub_bytes(inode,
669 le64_to_cpu(old.num_bytes));
670 ret = btrfs_free_extent(trans, root,
671 old_disk_bytenr,
672 le64_to_cpu(old.disk_num_bytes),
673 leaf_start, root_owner,
674 root_gen, key.objectid, 0);
675 BUG_ON(ret);
676 *hint_byte = old_disk_bytenr;
677 }
678 }
679
680 if (search_start >= end) {
681 ret = 0;
682 goto out;
683 }
684 }
685 out:
686 btrfs_free_path(path);
687 if (locked_end > end) {
688 unlock_extent(&BTRFS_I(inode)->io_tree, end, locked_end - 1,
689 GFP_NOFS);
690 }
691 btrfs_check_file(root, inode);
692 return ret;
693 }
694
695 static int extent_mergeable(struct extent_buffer *leaf, int slot,
696 u64 objectid, u64 bytenr, u64 *start, u64 *end)
697 {
698 struct btrfs_file_extent_item *fi;
699 struct btrfs_key key;
700 u64 extent_end;
701
702 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
703 return 0;
704
705 btrfs_item_key_to_cpu(leaf, &key, slot);
706 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
707 return 0;
708
709 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
710 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
711 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
712 btrfs_file_extent_compression(leaf, fi) ||
713 btrfs_file_extent_encryption(leaf, fi) ||
714 btrfs_file_extent_other_encoding(leaf, fi))
715 return 0;
716
717 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
718 if ((*start && *start != key.offset) || (*end && *end != extent_end))
719 return 0;
720
721 *start = key.offset;
722 *end = extent_end;
723 return 1;
724 }
725
726 /*
727 * Mark extent in the range start - end as written.
728 *
729 * This changes extent type from 'pre-allocated' to 'regular'. If only
730 * part of extent is marked as written, the extent will be split into
731 * two or three.
732 */
733 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
734 struct btrfs_root *root,
735 struct inode *inode, u64 start, u64 end)
736 {
737 struct extent_buffer *leaf;
738 struct btrfs_path *path;
739 struct btrfs_file_extent_item *fi;
740 struct btrfs_key key;
741 u64 bytenr;
742 u64 num_bytes;
743 u64 extent_end;
744 u64 extent_offset;
745 u64 other_start;
746 u64 other_end;
747 u64 split = start;
748 u64 locked_end = end;
749 u64 orig_parent;
750 int extent_type;
751 int split_end = 1;
752 int ret;
753
754 btrfs_drop_extent_cache(inode, start, end - 1, 0);
755
756 path = btrfs_alloc_path();
757 BUG_ON(!path);
758 again:
759 key.objectid = inode->i_ino;
760 key.type = BTRFS_EXTENT_DATA_KEY;
761 if (split == start)
762 key.offset = split;
763 else
764 key.offset = split - 1;
765
766 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
767 if (ret > 0 && path->slots[0] > 0)
768 path->slots[0]--;
769
770 leaf = path->nodes[0];
771 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
772 BUG_ON(key.objectid != inode->i_ino ||
773 key.type != BTRFS_EXTENT_DATA_KEY);
774 fi = btrfs_item_ptr(leaf, path->slots[0],
775 struct btrfs_file_extent_item);
776 extent_type = btrfs_file_extent_type(leaf, fi);
777 BUG_ON(extent_type != BTRFS_FILE_EXTENT_PREALLOC);
778 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
779 BUG_ON(key.offset > start || extent_end < end);
780
781 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
782 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
783 extent_offset = btrfs_file_extent_offset(leaf, fi);
784
785 if (key.offset == start)
786 split = end;
787
788 if (key.offset == start && extent_end == end) {
789 int del_nr = 0;
790 int del_slot = 0;
791 u64 leaf_owner = btrfs_header_owner(leaf);
792 u64 leaf_gen = btrfs_header_generation(leaf);
793 other_start = end;
794 other_end = 0;
795 if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
796 bytenr, &other_start, &other_end)) {
797 extent_end = other_end;
798 del_slot = path->slots[0] + 1;
799 del_nr++;
800 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
801 leaf->start, leaf_owner,
802 leaf_gen, inode->i_ino, 0);
803 BUG_ON(ret);
804 }
805 other_start = 0;
806 other_end = start;
807 if (extent_mergeable(leaf, path->slots[0] - 1, inode->i_ino,
808 bytenr, &other_start, &other_end)) {
809 key.offset = other_start;
810 del_slot = path->slots[0];
811 del_nr++;
812 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
813 leaf->start, leaf_owner,
814 leaf_gen, inode->i_ino, 0);
815 BUG_ON(ret);
816 }
817 split_end = 0;
818 if (del_nr == 0) {
819 btrfs_set_file_extent_type(leaf, fi,
820 BTRFS_FILE_EXTENT_REG);
821 goto done;
822 }
823
824 fi = btrfs_item_ptr(leaf, del_slot - 1,
825 struct btrfs_file_extent_item);
826 btrfs_set_file_extent_type(leaf, fi, BTRFS_FILE_EXTENT_REG);
827 btrfs_set_file_extent_num_bytes(leaf, fi,
828 extent_end - key.offset);
829 btrfs_mark_buffer_dirty(leaf);
830
831 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
832 BUG_ON(ret);
833 goto done;
834 } else if (split == start) {
835 if (locked_end < extent_end) {
836 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
837 locked_end, extent_end - 1, GFP_NOFS);
838 if (!ret) {
839 btrfs_release_path(root, path);
840 lock_extent(&BTRFS_I(inode)->io_tree,
841 locked_end, extent_end - 1, GFP_NOFS);
842 locked_end = extent_end;
843 goto again;
844 }
845 locked_end = extent_end;
846 }
847 btrfs_set_file_extent_num_bytes(leaf, fi, split - key.offset);
848 extent_offset += split - key.offset;
849 } else {
850 BUG_ON(key.offset != start);
851 btrfs_set_file_extent_offset(leaf, fi, extent_offset +
852 split - key.offset);
853 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - split);
854 key.offset = split;
855 btrfs_set_item_key_safe(trans, root, path, &key);
856 extent_end = split;
857 }
858
859 if (extent_end == end) {
860 split_end = 0;
861 extent_type = BTRFS_FILE_EXTENT_REG;
862 }
863 if (extent_end == end && split == start) {
864 other_start = end;
865 other_end = 0;
866 if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
867 bytenr, &other_start, &other_end)) {
868 path->slots[0]++;
869 fi = btrfs_item_ptr(leaf, path->slots[0],
870 struct btrfs_file_extent_item);
871 key.offset = split;
872 btrfs_set_item_key_safe(trans, root, path, &key);
873 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
874 btrfs_set_file_extent_num_bytes(leaf, fi,
875 other_end - split);
876 goto done;
877 }
878 }
879 if (extent_end == end && split == end) {
880 other_start = 0;
881 other_end = start;
882 if (extent_mergeable(leaf, path->slots[0] - 1 , inode->i_ino,
883 bytenr, &other_start, &other_end)) {
884 path->slots[0]--;
885 fi = btrfs_item_ptr(leaf, path->slots[0],
886 struct btrfs_file_extent_item);
887 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end -
888 other_start);
889 goto done;
890 }
891 }
892
893 btrfs_mark_buffer_dirty(leaf);
894
895 orig_parent = leaf->start;
896 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes,
897 orig_parent, root->root_key.objectid,
898 trans->transid, inode->i_ino);
899 BUG_ON(ret);
900 btrfs_release_path(root, path);
901
902 key.offset = start;
903 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*fi));
904 BUG_ON(ret);
905
906 leaf = path->nodes[0];
907 fi = btrfs_item_ptr(leaf, path->slots[0],
908 struct btrfs_file_extent_item);
909 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
910 btrfs_set_file_extent_type(leaf, fi, extent_type);
911 btrfs_set_file_extent_disk_bytenr(leaf, fi, bytenr);
912 btrfs_set_file_extent_disk_num_bytes(leaf, fi, num_bytes);
913 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
914 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - key.offset);
915 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
916 btrfs_set_file_extent_compression(leaf, fi, 0);
917 btrfs_set_file_extent_encryption(leaf, fi, 0);
918 btrfs_set_file_extent_other_encoding(leaf, fi, 0);
919
920 if (orig_parent != leaf->start) {
921 ret = btrfs_update_extent_ref(trans, root, bytenr,
922 orig_parent, leaf->start,
923 root->root_key.objectid,
924 trans->transid, inode->i_ino);
925 BUG_ON(ret);
926 }
927 done:
928 btrfs_mark_buffer_dirty(leaf);
929 btrfs_release_path(root, path);
930 if (split_end && split == start) {
931 split = end;
932 goto again;
933 }
934 if (locked_end > end) {
935 unlock_extent(&BTRFS_I(inode)->io_tree, end, locked_end - 1,
936 GFP_NOFS);
937 }
938 btrfs_free_path(path);
939 return 0;
940 }
941
942 /*
943 * this gets pages into the page cache and locks them down, it also properly
944 * waits for data=ordered extents to finish before allowing the pages to be
945 * modified.
946 */
947 static int noinline prepare_pages(struct btrfs_root *root, struct file *file,
948 struct page **pages, size_t num_pages,
949 loff_t pos, unsigned long first_index,
950 unsigned long last_index, size_t write_bytes)
951 {
952 int i;
953 unsigned long index = pos >> PAGE_CACHE_SHIFT;
954 struct inode *inode = fdentry(file)->d_inode;
955 int err = 0;
956 u64 start_pos;
957 u64 last_pos;
958
959 start_pos = pos & ~((u64)root->sectorsize - 1);
960 last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
961
962 if (start_pos > inode->i_size) {
963 err = btrfs_cont_expand(inode, start_pos);
964 if (err)
965 return err;
966 }
967
968 memset(pages, 0, num_pages * sizeof(struct page *));
969 again:
970 for (i = 0; i < num_pages; i++) {
971 pages[i] = grab_cache_page(inode->i_mapping, index + i);
972 if (!pages[i]) {
973 err = -ENOMEM;
974 BUG_ON(1);
975 }
976 wait_on_page_writeback(pages[i]);
977 }
978 if (start_pos < inode->i_size) {
979 struct btrfs_ordered_extent *ordered;
980 lock_extent(&BTRFS_I(inode)->io_tree,
981 start_pos, last_pos - 1, GFP_NOFS);
982 ordered = btrfs_lookup_first_ordered_extent(inode, last_pos -1);
983 if (ordered &&
984 ordered->file_offset + ordered->len > start_pos &&
985 ordered->file_offset < last_pos) {
986 btrfs_put_ordered_extent(ordered);
987 unlock_extent(&BTRFS_I(inode)->io_tree,
988 start_pos, last_pos - 1, GFP_NOFS);
989 for (i = 0; i < num_pages; i++) {
990 unlock_page(pages[i]);
991 page_cache_release(pages[i]);
992 }
993 btrfs_wait_ordered_range(inode, start_pos,
994 last_pos - start_pos);
995 goto again;
996 }
997 if (ordered)
998 btrfs_put_ordered_extent(ordered);
999
1000 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
1001 last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC,
1002 GFP_NOFS);
1003 unlock_extent(&BTRFS_I(inode)->io_tree,
1004 start_pos, last_pos - 1, GFP_NOFS);
1005 }
1006 for (i = 0; i < num_pages; i++) {
1007 clear_page_dirty_for_io(pages[i]);
1008 set_page_extent_mapped(pages[i]);
1009 WARN_ON(!PageLocked(pages[i]));
1010 }
1011 return 0;
1012 }
1013
1014 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
1015 size_t count, loff_t *ppos)
1016 {
1017 loff_t pos;
1018 loff_t start_pos;
1019 ssize_t num_written = 0;
1020 ssize_t err = 0;
1021 int ret = 0;
1022 struct inode *inode = fdentry(file)->d_inode;
1023 struct btrfs_root *root = BTRFS_I(inode)->root;
1024 struct page **pages = NULL;
1025 int nrptrs;
1026 struct page *pinned[2];
1027 unsigned long first_index;
1028 unsigned long last_index;
1029 int will_write;
1030
1031 will_write = ((file->f_flags & O_SYNC) || IS_SYNC(inode) ||
1032 (file->f_flags & O_DIRECT));
1033
1034 nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
1035 PAGE_CACHE_SIZE / (sizeof(struct page *)));
1036 pinned[0] = NULL;
1037 pinned[1] = NULL;
1038
1039 pos = *ppos;
1040 start_pos = pos;
1041
1042 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1043 current->backing_dev_info = inode->i_mapping->backing_dev_info;
1044 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1045 if (err)
1046 goto out_nolock;
1047 if (count == 0)
1048 goto out_nolock;
1049
1050 err = file_remove_suid(file);
1051 if (err)
1052 goto out_nolock;
1053 file_update_time(file);
1054
1055 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
1056
1057 mutex_lock(&inode->i_mutex);
1058 BTRFS_I(inode)->sequence++;
1059 first_index = pos >> PAGE_CACHE_SHIFT;
1060 last_index = (pos + count) >> PAGE_CACHE_SHIFT;
1061
1062 /*
1063 * there are lots of better ways to do this, but this code
1064 * makes sure the first and last page in the file range are
1065 * up to date and ready for cow
1066 */
1067 if ((pos & (PAGE_CACHE_SIZE - 1))) {
1068 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
1069 if (!PageUptodate(pinned[0])) {
1070 ret = btrfs_readpage(NULL, pinned[0]);
1071 BUG_ON(ret);
1072 wait_on_page_locked(pinned[0]);
1073 } else {
1074 unlock_page(pinned[0]);
1075 }
1076 }
1077 if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
1078 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
1079 if (!PageUptodate(pinned[1])) {
1080 ret = btrfs_readpage(NULL, pinned[1]);
1081 BUG_ON(ret);
1082 wait_on_page_locked(pinned[1]);
1083 } else {
1084 unlock_page(pinned[1]);
1085 }
1086 }
1087
1088 while(count > 0) {
1089 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1090 size_t write_bytes = min(count, nrptrs *
1091 (size_t)PAGE_CACHE_SIZE -
1092 offset);
1093 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
1094 PAGE_CACHE_SHIFT;
1095
1096 WARN_ON(num_pages > nrptrs);
1097 memset(pages, 0, sizeof(struct page *) * nrptrs);
1098
1099 ret = btrfs_check_free_space(root, write_bytes, 0);
1100 if (ret)
1101 goto out;
1102
1103 ret = prepare_pages(root, file, pages, num_pages,
1104 pos, first_index, last_index,
1105 write_bytes);
1106 if (ret)
1107 goto out;
1108
1109 ret = btrfs_copy_from_user(pos, num_pages,
1110 write_bytes, pages, buf);
1111 if (ret) {
1112 btrfs_drop_pages(pages, num_pages);
1113 goto out;
1114 }
1115
1116 ret = dirty_and_release_pages(NULL, root, file, pages,
1117 num_pages, pos, write_bytes);
1118 btrfs_drop_pages(pages, num_pages);
1119 if (ret)
1120 goto out;
1121
1122 if (will_write) {
1123 btrfs_fdatawrite_range(inode->i_mapping, pos,
1124 pos + write_bytes - 1,
1125 WB_SYNC_NONE);
1126 } else {
1127 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
1128 num_pages);
1129 if (num_pages <
1130 (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1131 btrfs_btree_balance_dirty(root, 1);
1132 btrfs_throttle(root);
1133 }
1134
1135 buf += write_bytes;
1136 count -= write_bytes;
1137 pos += write_bytes;
1138 num_written += write_bytes;
1139
1140 cond_resched();
1141 }
1142 out:
1143 mutex_unlock(&inode->i_mutex);
1144
1145 out_nolock:
1146 kfree(pages);
1147 if (pinned[0])
1148 page_cache_release(pinned[0]);
1149 if (pinned[1])
1150 page_cache_release(pinned[1]);
1151 *ppos = pos;
1152
1153 if (num_written > 0 && will_write) {
1154 struct btrfs_trans_handle *trans;
1155
1156 err = btrfs_wait_ordered_range(inode, start_pos, num_written);
1157 if (err)
1158 num_written = err;
1159
1160 if ((file->f_flags & O_SYNC) || IS_SYNC(inode)) {
1161 trans = btrfs_start_transaction(root, 1);
1162 ret = btrfs_log_dentry_safe(trans, root,
1163 file->f_dentry);
1164 if (ret == 0) {
1165 btrfs_sync_log(trans, root);
1166 btrfs_end_transaction(trans, root);
1167 } else {
1168 btrfs_commit_transaction(trans, root);
1169 }
1170 }
1171 if (file->f_flags & O_DIRECT) {
1172 invalidate_mapping_pages(inode->i_mapping,
1173 start_pos >> PAGE_CACHE_SHIFT,
1174 (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
1175 }
1176 }
1177 current->backing_dev_info = NULL;
1178 return num_written ? num_written : err;
1179 }
1180
1181 int btrfs_release_file(struct inode * inode, struct file * filp)
1182 {
1183 if (filp->private_data)
1184 btrfs_ioctl_trans_end(filp);
1185 return 0;
1186 }
1187
1188 /*
1189 * fsync call for both files and directories. This logs the inode into
1190 * the tree log instead of forcing full commits whenever possible.
1191 *
1192 * It needs to call filemap_fdatawait so that all ordered extent updates are
1193 * in the metadata btree are up to date for copying to the log.
1194 *
1195 * It drops the inode mutex before doing the tree log commit. This is an
1196 * important optimization for directories because holding the mutex prevents
1197 * new operations on the dir while we write to disk.
1198 */
1199 int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
1200 {
1201 struct inode *inode = dentry->d_inode;
1202 struct btrfs_root *root = BTRFS_I(inode)->root;
1203 int ret = 0;
1204 struct btrfs_trans_handle *trans;
1205
1206 /*
1207 * check the transaction that last modified this inode
1208 * and see if its already been committed
1209 */
1210 if (!BTRFS_I(inode)->last_trans)
1211 goto out;
1212
1213 mutex_lock(&root->fs_info->trans_mutex);
1214 if (BTRFS_I(inode)->last_trans <=
1215 root->fs_info->last_trans_committed) {
1216 BTRFS_I(inode)->last_trans = 0;
1217 mutex_unlock(&root->fs_info->trans_mutex);
1218 goto out;
1219 }
1220 mutex_unlock(&root->fs_info->trans_mutex);
1221
1222 root->fs_info->tree_log_batch++;
1223 filemap_fdatawrite(inode->i_mapping);
1224 btrfs_wait_ordered_range(inode, 0, (u64)-1);
1225 root->fs_info->tree_log_batch++;
1226
1227 /*
1228 * ok we haven't committed the transaction yet, lets do a commit
1229 */
1230 if (file->private_data)
1231 btrfs_ioctl_trans_end(file);
1232
1233 trans = btrfs_start_transaction(root, 1);
1234 if (!trans) {
1235 ret = -ENOMEM;
1236 goto out;
1237 }
1238
1239 ret = btrfs_log_dentry_safe(trans, root, file->f_dentry);
1240 if (ret < 0) {
1241 goto out;
1242 }
1243
1244 /* we've logged all the items and now have a consistent
1245 * version of the file in the log. It is possible that
1246 * someone will come in and modify the file, but that's
1247 * fine because the log is consistent on disk, and we
1248 * have references to all of the file's extents
1249 *
1250 * It is possible that someone will come in and log the
1251 * file again, but that will end up using the synchronization
1252 * inside btrfs_sync_log to keep things safe.
1253 */
1254 mutex_unlock(&file->f_dentry->d_inode->i_mutex);
1255
1256 if (ret > 0) {
1257 ret = btrfs_commit_transaction(trans, root);
1258 } else {
1259 btrfs_sync_log(trans, root);
1260 ret = btrfs_end_transaction(trans, root);
1261 }
1262 mutex_lock(&file->f_dentry->d_inode->i_mutex);
1263 out:
1264 return ret > 0 ? EIO : ret;
1265 }
1266
1267 static struct vm_operations_struct btrfs_file_vm_ops = {
1268 .fault = filemap_fault,
1269 .page_mkwrite = btrfs_page_mkwrite,
1270 };
1271
1272 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
1273 {
1274 vma->vm_ops = &btrfs_file_vm_ops;
1275 file_accessed(filp);
1276 return 0;
1277 }
1278
1279 struct file_operations btrfs_file_operations = {
1280 .llseek = generic_file_llseek,
1281 .read = do_sync_read,
1282 .aio_read = generic_file_aio_read,
1283 .splice_read = generic_file_splice_read,
1284 .write = btrfs_file_write,
1285 .mmap = btrfs_file_mmap,
1286 .open = generic_file_open,
1287 .release = btrfs_release_file,
1288 .fsync = btrfs_sync_file,
1289 .unlocked_ioctl = btrfs_ioctl,
1290 #ifdef CONFIG_COMPAT
1291 .compat_ioctl = btrfs_ioctl,
1292 #endif
1293 };
Linux kernel - Deliverables
This page took 0.056397 seconds and 5 git commands to generate.