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 */
btrfs_copy_from_user(loff_t pos,int num_pages,int write_bytes,struct page ** prepared_pages,const char __user * buf)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 */
btrfs_drop_pages(struct page ** pages,size_t num_pages)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 */
dirty_and_release_pages(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct file * file,struct page ** pages,size_t num_pages,loff_t pos,size_t write_bytes)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 */
btrfs_drop_extent_cache(struct inode * inode,u64 start,u64 end,int skip_pinned)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
btrfs_check_file(struct btrfs_root * root,struct inode * inode)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 */
btrfs_drop_extents(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct inode * inode,u64 start,u64 end,u64 inline_limit,u64 * hint_byte)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 inline_limit, u64 *hint_byte)
367 {
368 u64 extent_end = 0;
369 u64 locked_end = end;
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 u8 compression;
376 u8 encryption;
377 u16 other_encoding = 0;
378 u64 root_gen;
379 u64 root_owner;
380 struct extent_buffer *leaf;
381 struct btrfs_file_extent_item *extent;
382 struct btrfs_path *path;
383 struct btrfs_key key;
384 struct btrfs_file_extent_item old;
385 int keep;
386 int slot;
387 int bookend;
388 int found_type = 0;
389 int found_extent;
390 int found_inline;
391 int recow;
392 int ret;
393
394 inline_limit = 0;
395 btrfs_drop_extent_cache(inode, start, end - 1, 0);
396
397 path = btrfs_alloc_path();
398 if (!path)
399 return -ENOMEM;
400 while (1) {
401 recow = 0;
402 btrfs_release_path(root, path);
403 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
404 search_start, -1);
405 if (ret < 0)
406 goto out;
407 if (ret > 0) {
408 if (path->slots[0] == 0) {
409 ret = 0;
410 goto out;
411 }
412 path->slots[0]--;
413 }
414 next_slot:
415 keep = 0;
416 bookend = 0;
417 found_extent = 0;
418 found_inline = 0;
419 leaf_start = 0;
420 root_gen = 0;
421 root_owner = 0;
422 compression = 0;
423 encryption = 0;
424 extent = NULL;
425 leaf = path->nodes[0];
426 slot = path->slots[0];
427 ret = 0;
428 btrfs_item_key_to_cpu(leaf, &key, slot);
429 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY &&
430 key.offset >= end) {
431 goto out;
432 }
433 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
434 key.objectid != inode->i_ino) {
435 goto out;
436 }
437 if (recow) {
438 search_start = max(key.offset, start);
439 continue;
440 }
441 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
442 extent = btrfs_item_ptr(leaf, slot,
443 struct btrfs_file_extent_item);
444 found_type = btrfs_file_extent_type(leaf, extent);
445 compression = btrfs_file_extent_compression(leaf,
446 extent);
447 encryption = btrfs_file_extent_encryption(leaf,
448 extent);
449 other_encoding = btrfs_file_extent_other_encoding(leaf,
450 extent);
451 if (found_type == BTRFS_FILE_EXTENT_REG ||
452 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
453 extent_end =
454 btrfs_file_extent_disk_bytenr(leaf,
455 extent);
456 if (extent_end)
457 *hint_byte = extent_end;
458
459 extent_end = key.offset +
460 btrfs_file_extent_num_bytes(leaf, extent);
461 ram_bytes = btrfs_file_extent_ram_bytes(leaf,
462 extent);
463 found_extent = 1;
464 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
465 found_inline = 1;
466 extent_end = key.offset +
467 btrfs_file_extent_inline_len(leaf, extent);
468 }
469 } else {
470 extent_end = search_start;
471 }
472
473 /* we found nothing we can drop */
474 if ((!found_extent && !found_inline) ||
475 search_start >= extent_end) {
476 int nextret;
477 u32 nritems;
478 nritems = btrfs_header_nritems(leaf);
479 if (slot >= nritems - 1) {
480 nextret = btrfs_next_leaf(root, path);
481 if (nextret)
482 goto out;
483 recow = 1;
484 } else {
485 path->slots[0]++;
486 }
487 goto next_slot;
488 }
489
490 if (end <= extent_end && start >= key.offset && found_inline)
491 *hint_byte = EXTENT_MAP_INLINE;
492
493 if (found_extent) {
494 read_extent_buffer(leaf, &old, (unsigned long)extent,
495 sizeof(old));
496 root_gen = btrfs_header_generation(leaf);
497 root_owner = btrfs_header_owner(leaf);
498 leaf_start = leaf->start;
499 }
500
501 if (end < extent_end && end >= key.offset) {
502 bookend = 1;
503 if (found_inline && start <= key.offset)
504 keep = 1;
505 }
506
507 if (bookend && found_extent) {
508 if (locked_end < extent_end) {
509 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
510 locked_end, extent_end - 1,
511 GFP_NOFS);
512 if (!ret) {
513 btrfs_release_path(root, path);
514 lock_extent(&BTRFS_I(inode)->io_tree,
515 locked_end, extent_end - 1,
516 GFP_NOFS);
517 locked_end = extent_end;
518 continue;
519 }
520 locked_end = extent_end;
521 }
522 orig_parent = path->nodes[0]->start;
523 disk_bytenr = le64_to_cpu(old.disk_bytenr);
524 if (disk_bytenr != 0) {
525 ret = btrfs_inc_extent_ref(trans, root,
526 disk_bytenr,
527 le64_to_cpu(old.disk_num_bytes),
528 orig_parent, root->root_key.objectid,
529 trans->transid, inode->i_ino);
530 BUG_ON(ret);
531 }
532 }
533
534 if (found_inline) {
535 u64 mask = root->sectorsize - 1;
536 search_start = (extent_end + mask) & ~mask;
537 } else
538 search_start = extent_end;
539
540 /* truncate existing extent */
541 if (start > key.offset) {
542 u64 new_num;
543 u64 old_num;
544 keep = 1;
545 WARN_ON(start & (root->sectorsize - 1));
546 if (found_extent) {
547 new_num = start - key.offset;
548 old_num = btrfs_file_extent_num_bytes(leaf,
549 extent);
550 *hint_byte =
551 btrfs_file_extent_disk_bytenr(leaf,
552 extent);
553 if (btrfs_file_extent_disk_bytenr(leaf,
554 extent)) {
555 inode_sub_bytes(inode, old_num -
556 new_num);
557 }
558 btrfs_set_file_extent_num_bytes(leaf,
559 extent, new_num);
560 btrfs_mark_buffer_dirty(leaf);
561 } else if (key.offset < inline_limit &&
562 (end > extent_end) &&
563 (inline_limit < extent_end)) {
564 u32 new_size;
565 new_size = btrfs_file_extent_calc_inline_size(
566 inline_limit - key.offset);
567 inode_sub_bytes(inode, extent_end -
568 inline_limit);
569 btrfs_set_file_extent_ram_bytes(leaf, extent,
570 new_size);
571 if (!compression && !encryption) {
572 btrfs_truncate_item(trans, root, path,
573 new_size, 1);
574 }
575 }
576 }
577 /* delete the entire extent */
578 if (!keep) {
579 if (found_inline)
580 inode_sub_bytes(inode, extent_end -
581 key.offset);
582 ret = btrfs_del_item(trans, root, path);
583 /* TODO update progress marker and return */
584 BUG_ON(ret);
585 extent = NULL;
586 btrfs_release_path(root, path);
587 /* the extent will be freed later */
588 }
589 if (bookend && found_inline && start <= key.offset) {
590 u32 new_size;
591 new_size = btrfs_file_extent_calc_inline_size(
592 extent_end - end);
593 inode_sub_bytes(inode, end - key.offset);
594 btrfs_set_file_extent_ram_bytes(leaf, extent,
595 new_size);
596 if (!compression && !encryption)
597 ret = btrfs_truncate_item(trans, root, path,
598 new_size, 0);
599 BUG_ON(ret);
600 }
601 /* create bookend, splitting the extent in two */
602 if (bookend && found_extent) {
603 struct btrfs_key ins;
604 ins.objectid = inode->i_ino;
605 ins.offset = end;
606 btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
607
608 btrfs_release_path(root, path);
609 ret = btrfs_insert_empty_item(trans, root, path, &ins,
610 sizeof(*extent));
611 BUG_ON(ret);
612
613 leaf = path->nodes[0];
614 extent = btrfs_item_ptr(leaf, path->slots[0],
615 struct btrfs_file_extent_item);
616 write_extent_buffer(leaf, &old,
617 (unsigned long)extent, sizeof(old));
618
619 btrfs_set_file_extent_compression(leaf, extent,
620 compression);
621 btrfs_set_file_extent_encryption(leaf, extent,
622 encryption);
623 btrfs_set_file_extent_other_encoding(leaf, extent,
624 other_encoding);
625 btrfs_set_file_extent_offset(leaf, extent,
626 le64_to_cpu(old.offset) + end - key.offset);
627 WARN_ON(le64_to_cpu(old.num_bytes) <
628 (extent_end - end));
629 btrfs_set_file_extent_num_bytes(leaf, extent,
630 extent_end - end);
631
632 /*
633 * set the ram bytes to the size of the full extent
634 * before splitting. This is a worst case flag,
635 * but its the best we can do because we don't know
636 * how splitting affects compression
637 */
638 btrfs_set_file_extent_ram_bytes(leaf, extent,
639 ram_bytes);
640 btrfs_set_file_extent_type(leaf, extent, found_type);
641
642 btrfs_mark_buffer_dirty(path->nodes[0]);
643
644 if (disk_bytenr != 0) {
645 ret = btrfs_update_extent_ref(trans, root,
646 disk_bytenr, orig_parent,
647 leaf->start,
648 root->root_key.objectid,
649 trans->transid, ins.objectid);
650
651 BUG_ON(ret);
652 }
653 btrfs_release_path(root, path);
654 if (disk_bytenr != 0)
655 inode_add_bytes(inode, extent_end - end);
656 }
657
658 if (found_extent && !keep) {
659 u64 old_disk_bytenr = le64_to_cpu(old.disk_bytenr);
660
661 if (old_disk_bytenr != 0) {
662 inode_sub_bytes(inode,
663 le64_to_cpu(old.num_bytes));
664 ret = btrfs_free_extent(trans, root,
665 old_disk_bytenr,
666 le64_to_cpu(old.disk_num_bytes),
667 leaf_start, root_owner,
668 root_gen, key.objectid, 0);
669 BUG_ON(ret);
670 *hint_byte = old_disk_bytenr;
671 }
672 }
673
674 if (search_start >= end) {
675 ret = 0;
676 goto out;
677 }
678 }
679 out:
680 btrfs_free_path(path);
681 if (locked_end > end) {
682 unlock_extent(&BTRFS_I(inode)->io_tree, end, locked_end - 1,
683 GFP_NOFS);
684 }
685 btrfs_check_file(root, inode);
686 return ret;
687 }
688
extent_mergeable(struct extent_buffer * leaf,int slot,u64 objectid,u64 bytenr,u64 * start,u64 * end)689 static int extent_mergeable(struct extent_buffer *leaf, int slot,
690 u64 objectid, u64 bytenr, u64 *start, u64 *end)
691 {
692 struct btrfs_file_extent_item *fi;
693 struct btrfs_key key;
694 u64 extent_end;
695
696 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
697 return 0;
698
699 btrfs_item_key_to_cpu(leaf, &key, slot);
700 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
701 return 0;
702
703 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
704 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
705 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
706 btrfs_file_extent_compression(leaf, fi) ||
707 btrfs_file_extent_encryption(leaf, fi) ||
708 btrfs_file_extent_other_encoding(leaf, fi))
709 return 0;
710
711 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
712 if ((*start && *start != key.offset) || (*end && *end != extent_end))
713 return 0;
714
715 *start = key.offset;
716 *end = extent_end;
717 return 1;
718 }
719
720 /*
721 * Mark extent in the range start - end as written.
722 *
723 * This changes extent type from 'pre-allocated' to 'regular'. If only
724 * part of extent is marked as written, the extent will be split into
725 * two or three.
726 */
btrfs_mark_extent_written(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct inode * inode,u64 start,u64 end)727 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
728 struct btrfs_root *root,
729 struct inode *inode, u64 start, u64 end)
730 {
731 struct extent_buffer *leaf;
732 struct btrfs_path *path;
733 struct btrfs_file_extent_item *fi;
734 struct btrfs_key key;
735 u64 bytenr;
736 u64 num_bytes;
737 u64 extent_end;
738 u64 extent_offset;
739 u64 other_start;
740 u64 other_end;
741 u64 split = start;
742 u64 locked_end = end;
743 u64 orig_parent;
744 int extent_type;
745 int split_end = 1;
746 int ret;
747
748 btrfs_drop_extent_cache(inode, start, end - 1, 0);
749
750 path = btrfs_alloc_path();
751 BUG_ON(!path);
752 again:
753 key.objectid = inode->i_ino;
754 key.type = BTRFS_EXTENT_DATA_KEY;
755 if (split == start)
756 key.offset = split;
757 else
758 key.offset = split - 1;
759
760 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
761 if (ret > 0 && path->slots[0] > 0)
762 path->slots[0]--;
763
764 leaf = path->nodes[0];
765 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
766 BUG_ON(key.objectid != inode->i_ino ||
767 key.type != BTRFS_EXTENT_DATA_KEY);
768 fi = btrfs_item_ptr(leaf, path->slots[0],
769 struct btrfs_file_extent_item);
770 extent_type = btrfs_file_extent_type(leaf, fi);
771 BUG_ON(extent_type != BTRFS_FILE_EXTENT_PREALLOC);
772 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
773 BUG_ON(key.offset > start || extent_end < end);
774
775 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
776 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
777 extent_offset = btrfs_file_extent_offset(leaf, fi);
778
779 if (key.offset == start)
780 split = end;
781
782 if (key.offset == start && extent_end == end) {
783 int del_nr = 0;
784 int del_slot = 0;
785 u64 leaf_owner = btrfs_header_owner(leaf);
786 u64 leaf_gen = btrfs_header_generation(leaf);
787 other_start = end;
788 other_end = 0;
789 if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
790 bytenr, &other_start, &other_end)) {
791 extent_end = other_end;
792 del_slot = path->slots[0] + 1;
793 del_nr++;
794 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
795 leaf->start, leaf_owner,
796 leaf_gen, inode->i_ino, 0);
797 BUG_ON(ret);
798 }
799 other_start = 0;
800 other_end = start;
801 if (extent_mergeable(leaf, path->slots[0] - 1, inode->i_ino,
802 bytenr, &other_start, &other_end)) {
803 key.offset = other_start;
804 del_slot = path->slots[0];
805 del_nr++;
806 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
807 leaf->start, leaf_owner,
808 leaf_gen, inode->i_ino, 0);
809 BUG_ON(ret);
810 }
811 split_end = 0;
812 if (del_nr == 0) {
813 btrfs_set_file_extent_type(leaf, fi,
814 BTRFS_FILE_EXTENT_REG);
815 goto done;
816 }
817
818 fi = btrfs_item_ptr(leaf, del_slot - 1,
819 struct btrfs_file_extent_item);
820 btrfs_set_file_extent_type(leaf, fi, BTRFS_FILE_EXTENT_REG);
821 btrfs_set_file_extent_num_bytes(leaf, fi,
822 extent_end - key.offset);
823 btrfs_mark_buffer_dirty(leaf);
824
825 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
826 BUG_ON(ret);
827 goto done;
828 } else if (split == start) {
829 if (locked_end < extent_end) {
830 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
831 locked_end, extent_end - 1, GFP_NOFS);
832 if (!ret) {
833 btrfs_release_path(root, path);
834 lock_extent(&BTRFS_I(inode)->io_tree,
835 locked_end, extent_end - 1, GFP_NOFS);
836 locked_end = extent_end;
837 goto again;
838 }
839 locked_end = extent_end;
840 }
841 btrfs_set_file_extent_num_bytes(leaf, fi, split - key.offset);
842 extent_offset += split - key.offset;
843 } else {
844 BUG_ON(key.offset != start);
845 btrfs_set_file_extent_offset(leaf, fi, extent_offset +
846 split - key.offset);
847 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - split);
848 key.offset = split;
849 btrfs_set_item_key_safe(trans, root, path, &key);
850 extent_end = split;
851 }
852
853 if (extent_end == end) {
854 split_end = 0;
855 extent_type = BTRFS_FILE_EXTENT_REG;
856 }
857 if (extent_end == end && split == start) {
858 other_start = end;
859 other_end = 0;
860 if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
861 bytenr, &other_start, &other_end)) {
862 path->slots[0]++;
863 fi = btrfs_item_ptr(leaf, path->slots[0],
864 struct btrfs_file_extent_item);
865 key.offset = split;
866 btrfs_set_item_key_safe(trans, root, path, &key);
867 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
868 btrfs_set_file_extent_num_bytes(leaf, fi,
869 other_end - split);
870 goto done;
871 }
872 }
873 if (extent_end == end && split == end) {
874 other_start = 0;
875 other_end = start;
876 if (extent_mergeable(leaf, path->slots[0] - 1 , inode->i_ino,
877 bytenr, &other_start, &other_end)) {
878 path->slots[0]--;
879 fi = btrfs_item_ptr(leaf, path->slots[0],
880 struct btrfs_file_extent_item);
881 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end -
882 other_start);
883 goto done;
884 }
885 }
886
887 btrfs_mark_buffer_dirty(leaf);
888
889 orig_parent = leaf->start;
890 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes,
891 orig_parent, root->root_key.objectid,
892 trans->transid, inode->i_ino);
893 BUG_ON(ret);
894 btrfs_release_path(root, path);
895
896 key.offset = start;
897 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*fi));
898 BUG_ON(ret);
899
900 leaf = path->nodes[0];
901 fi = btrfs_item_ptr(leaf, path->slots[0],
902 struct btrfs_file_extent_item);
903 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
904 btrfs_set_file_extent_type(leaf, fi, extent_type);
905 btrfs_set_file_extent_disk_bytenr(leaf, fi, bytenr);
906 btrfs_set_file_extent_disk_num_bytes(leaf, fi, num_bytes);
907 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
908 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - key.offset);
909 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
910 btrfs_set_file_extent_compression(leaf, fi, 0);
911 btrfs_set_file_extent_encryption(leaf, fi, 0);
912 btrfs_set_file_extent_other_encoding(leaf, fi, 0);
913
914 if (orig_parent != leaf->start) {
915 ret = btrfs_update_extent_ref(trans, root, bytenr,
916 orig_parent, leaf->start,
917 root->root_key.objectid,
918 trans->transid, inode->i_ino);
919 BUG_ON(ret);
920 }
921 done:
922 btrfs_mark_buffer_dirty(leaf);
923 btrfs_release_path(root, path);
924 if (split_end && split == start) {
925 split = end;
926 goto again;
927 }
928 if (locked_end > end) {
929 unlock_extent(&BTRFS_I(inode)->io_tree, end, locked_end - 1,
930 GFP_NOFS);
931 }
932 btrfs_free_path(path);
933 return 0;
934 }
935
936 /*
937 * this gets pages into the page cache and locks them down, it also properly
938 * waits for data=ordered extents to finish before allowing the pages to be
939 * modified.
940 */
prepare_pages(struct btrfs_root * root,struct file * file,struct page ** pages,size_t num_pages,loff_t pos,unsigned long first_index,unsigned long last_index,size_t write_bytes)941 static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
942 struct page **pages, size_t num_pages,
943 loff_t pos, unsigned long first_index,
944 unsigned long last_index, size_t write_bytes)
945 {
946 int i;
947 unsigned long index = pos >> PAGE_CACHE_SHIFT;
948 struct inode *inode = fdentry(file)->d_inode;
949 int err = 0;
950 u64 start_pos;
951 u64 last_pos;
952
953 start_pos = pos & ~((u64)root->sectorsize - 1);
954 last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
955
956 if (start_pos > inode->i_size) {
957 err = btrfs_cont_expand(inode, start_pos);
958 if (err)
959 return err;
960 }
961
962 memset(pages, 0, num_pages * sizeof(struct page *));
963 again:
964 for (i = 0; i < num_pages; i++) {
965 pages[i] = grab_cache_page(inode->i_mapping, index + i);
966 if (!pages[i]) {
967 err = -ENOMEM;
968 BUG_ON(1);
969 }
970 wait_on_page_writeback(pages[i]);
971 }
972 if (start_pos < inode->i_size) {
973 struct btrfs_ordered_extent *ordered;
974 lock_extent(&BTRFS_I(inode)->io_tree,
975 start_pos, last_pos - 1, GFP_NOFS);
976 ordered = btrfs_lookup_first_ordered_extent(inode,
977 last_pos - 1);
978 if (ordered &&
979 ordered->file_offset + ordered->len > start_pos &&
980 ordered->file_offset < last_pos) {
981 btrfs_put_ordered_extent(ordered);
982 unlock_extent(&BTRFS_I(inode)->io_tree,
983 start_pos, last_pos - 1, GFP_NOFS);
984 for (i = 0; i < num_pages; i++) {
985 unlock_page(pages[i]);
986 page_cache_release(pages[i]);
987 }
988 btrfs_wait_ordered_range(inode, start_pos,
989 last_pos - start_pos);
990 goto again;
991 }
992 if (ordered)
993 btrfs_put_ordered_extent(ordered);
994
995 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
996 last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC,
997 GFP_NOFS);
998 unlock_extent(&BTRFS_I(inode)->io_tree,
999 start_pos, last_pos - 1, GFP_NOFS);
1000 }
1001 for (i = 0; i < num_pages; i++) {
1002 clear_page_dirty_for_io(pages[i]);
1003 set_page_extent_mapped(pages[i]);
1004 WARN_ON(!PageLocked(pages[i]));
1005 }
1006 return 0;
1007 }
1008
btrfs_file_write(struct file * file,const char __user * buf,size_t count,loff_t * ppos)1009 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
1010 size_t count, loff_t *ppos)
1011 {
1012 loff_t pos;
1013 loff_t start_pos;
1014 ssize_t num_written = 0;
1015 ssize_t err = 0;
1016 int ret = 0;
1017 struct inode *inode = fdentry(file)->d_inode;
1018 struct btrfs_root *root = BTRFS_I(inode)->root;
1019 struct page **pages = NULL;
1020 int nrptrs;
1021 struct page *pinned[2];
1022 unsigned long first_index;
1023 unsigned long last_index;
1024 int will_write;
1025
1026 will_write = ((file->f_flags & O_SYNC) || IS_SYNC(inode) ||
1027 (file->f_flags & O_DIRECT));
1028
1029 nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
1030 PAGE_CACHE_SIZE / (sizeof(struct page *)));
1031 pinned[0] = NULL;
1032 pinned[1] = NULL;
1033
1034 pos = *ppos;
1035 start_pos = pos;
1036
1037 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1038 current->backing_dev_info = inode->i_mapping->backing_dev_info;
1039 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1040 if (err)
1041 goto out_nolock;
1042 if (count == 0)
1043 goto out_nolock;
1044
1045 err = file_remove_suid(file);
1046 if (err)
1047 goto out_nolock;
1048 file_update_time(file);
1049
1050 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
1051
1052 mutex_lock(&inode->i_mutex);
1053 BTRFS_I(inode)->sequence++;
1054 first_index = pos >> PAGE_CACHE_SHIFT;
1055 last_index = (pos + count) >> PAGE_CACHE_SHIFT;
1056
1057 /*
1058 * there are lots of better ways to do this, but this code
1059 * makes sure the first and last page in the file range are
1060 * up to date and ready for cow
1061 */
1062 if ((pos & (PAGE_CACHE_SIZE - 1))) {
1063 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
1064 if (!PageUptodate(pinned[0])) {
1065 ret = btrfs_readpage(NULL, pinned[0]);
1066 BUG_ON(ret);
1067 wait_on_page_locked(pinned[0]);
1068 } else {
1069 unlock_page(pinned[0]);
1070 }
1071 }
1072 if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
1073 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
1074 if (!PageUptodate(pinned[1])) {
1075 ret = btrfs_readpage(NULL, pinned[1]);
1076 BUG_ON(ret);
1077 wait_on_page_locked(pinned[1]);
1078 } else {
1079 unlock_page(pinned[1]);
1080 }
1081 }
1082
1083 while (count > 0) {
1084 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1085 size_t write_bytes = min(count, nrptrs *
1086 (size_t)PAGE_CACHE_SIZE -
1087 offset);
1088 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
1089 PAGE_CACHE_SHIFT;
1090
1091 WARN_ON(num_pages > nrptrs);
1092 memset(pages, 0, sizeof(struct page *) * nrptrs);
1093
1094 ret = btrfs_check_data_free_space(root, inode, write_bytes);
1095 if (ret)
1096 goto out;
1097
1098 ret = prepare_pages(root, file, pages, num_pages,
1099 pos, first_index, last_index,
1100 write_bytes);
1101 if (ret) {
1102 btrfs_free_reserved_data_space(root, inode,
1103 write_bytes);
1104 goto out;
1105 }
1106
1107 ret = btrfs_copy_from_user(pos, num_pages,
1108 write_bytes, pages, buf);
1109 if (ret) {
1110 btrfs_free_reserved_data_space(root, inode,
1111 write_bytes);
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 btrfs_free_reserved_data_space(root, inode,
1121 write_bytes);
1122 goto out;
1123 }
1124
1125 if (will_write) {
1126 btrfs_fdatawrite_range(inode->i_mapping, pos,
1127 pos + write_bytes - 1,
1128 WB_SYNC_NONE);
1129 } else {
1130 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
1131 num_pages);
1132 if (num_pages <
1133 (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1134 btrfs_btree_balance_dirty(root, 1);
1135 btrfs_throttle(root);
1136 }
1137
1138 buf += write_bytes;
1139 count -= write_bytes;
1140 pos += write_bytes;
1141 num_written += write_bytes;
1142
1143 cond_resched();
1144 }
1145 out:
1146 mutex_unlock(&inode->i_mutex);
1147 if (ret)
1148 err = ret;
1149
1150 out_nolock:
1151 kfree(pages);
1152 if (pinned[0])
1153 page_cache_release(pinned[0]);
1154 if (pinned[1])
1155 page_cache_release(pinned[1]);
1156 *ppos = pos;
1157
1158 if (num_written > 0 && will_write) {
1159 struct btrfs_trans_handle *trans;
1160
1161 err = btrfs_wait_ordered_range(inode, start_pos, num_written);
1162 if (err)
1163 num_written = err;
1164
1165 if ((file->f_flags & O_SYNC) || IS_SYNC(inode)) {
1166 trans = btrfs_start_transaction(root, 1);
1167 ret = btrfs_log_dentry_safe(trans, root,
1168 file->f_dentry);
1169 if (ret == 0) {
1170 btrfs_sync_log(trans, root);
1171 btrfs_end_transaction(trans, root);
1172 } else {
1173 btrfs_commit_transaction(trans, root);
1174 }
1175 }
1176 if (file->f_flags & O_DIRECT) {
1177 invalidate_mapping_pages(inode->i_mapping,
1178 start_pos >> PAGE_CACHE_SHIFT,
1179 (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
1180 }
1181 }
1182 current->backing_dev_info = NULL;
1183 return num_written ? num_written : err;
1184 }
1185
btrfs_release_file(struct inode * inode,struct file * filp)1186 int btrfs_release_file(struct inode *inode, struct file *filp)
1187 {
1188 if (filp->private_data)
1189 btrfs_ioctl_trans_end(filp);
1190 return 0;
1191 }
1192
1193 /*
1194 * fsync call for both files and directories. This logs the inode into
1195 * the tree log instead of forcing full commits whenever possible.
1196 *
1197 * It needs to call filemap_fdatawait so that all ordered extent updates are
1198 * in the metadata btree are up to date for copying to the log.
1199 *
1200 * It drops the inode mutex before doing the tree log commit. This is an
1201 * important optimization for directories because holding the mutex prevents
1202 * new operations on the dir while we write to disk.
1203 */
btrfs_sync_file(struct file * file,struct dentry * dentry,int datasync)1204 int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
1205 {
1206 struct inode *inode = dentry->d_inode;
1207 struct btrfs_root *root = BTRFS_I(inode)->root;
1208 int ret = 0;
1209 struct btrfs_trans_handle *trans;
1210
1211 /*
1212 * check the transaction that last modified this inode
1213 * and see if its already been committed
1214 */
1215 if (!BTRFS_I(inode)->last_trans)
1216 goto out;
1217
1218 mutex_lock(&root->fs_info->trans_mutex);
1219 if (BTRFS_I(inode)->last_trans <=
1220 root->fs_info->last_trans_committed) {
1221 BTRFS_I(inode)->last_trans = 0;
1222 mutex_unlock(&root->fs_info->trans_mutex);
1223 goto out;
1224 }
1225 mutex_unlock(&root->fs_info->trans_mutex);
1226
1227 root->log_batch++;
1228 filemap_fdatawrite(inode->i_mapping);
1229 btrfs_wait_ordered_range(inode, 0, (u64)-1);
1230 root->log_batch++;
1231
1232 /*
1233 * ok we haven't committed the transaction yet, lets do a commit
1234 */
1235 if (file && file->private_data)
1236 btrfs_ioctl_trans_end(file);
1237
1238 trans = btrfs_start_transaction(root, 1);
1239 if (!trans) {
1240 ret = -ENOMEM;
1241 goto out;
1242 }
1243
1244 ret = btrfs_log_dentry_safe(trans, root, dentry);
1245 if (ret < 0)
1246 goto out;
1247
1248 /* we've logged all the items and now have a consistent
1249 * version of the file in the log. It is possible that
1250 * someone will come in and modify the file, but that's
1251 * fine because the log is consistent on disk, and we
1252 * have references to all of the file's extents
1253 *
1254 * It is possible that someone will come in and log the
1255 * file again, but that will end up using the synchronization
1256 * inside btrfs_sync_log to keep things safe.
1257 */
1258 mutex_unlock(&dentry->d_inode->i_mutex);
1259
1260 if (ret > 0) {
1261 ret = btrfs_commit_transaction(trans, root);
1262 } else {
1263 btrfs_sync_log(trans, root);
1264 ret = btrfs_end_transaction(trans, root);
1265 }
1266 mutex_lock(&dentry->d_inode->i_mutex);
1267 out:
1268 return ret > 0 ? EIO : ret;
1269 }
1270
1271 static struct vm_operations_struct btrfs_file_vm_ops = {
1272 .fault = filemap_fault,
1273 .page_mkwrite = btrfs_page_mkwrite,
1274 };
1275
btrfs_file_mmap(struct file * filp,struct vm_area_struct * vma)1276 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
1277 {
1278 vma->vm_ops = &btrfs_file_vm_ops;
1279 file_accessed(filp);
1280 return 0;
1281 }
1282
1283 struct file_operations btrfs_file_operations = {
1284 .llseek = generic_file_llseek,
1285 .read = do_sync_read,
1286 .aio_read = generic_file_aio_read,
1287 .splice_read = generic_file_splice_read,
1288 .write = btrfs_file_write,
1289 .mmap = btrfs_file_mmap,
1290 .open = generic_file_open,
1291 .release = btrfs_release_file,
1292 .fsync = btrfs_sync_file,
1293 .unlocked_ioctl = btrfs_ioctl,
1294 #ifdef CONFIG_COMPAT
1295 .compat_ioctl = btrfs_ioctl,
1296 #endif
1297 };
1298