projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
btrfs_start_transaction: wait for commits in progress to finish
[deliverable/linux.git] / fs / btrfs / inode.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/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/smp_lock.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mpage.h>
32 #include <linux/swap.h>
33 #include <linux/writeback.h>
34 #include <linux/statfs.h>
35 #include <linux/compat.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/version.h>
38 #include <linux/xattr.h>
39 #include "ctree.h"
40 #include "disk-io.h"
41 #include "transaction.h"
42 #include "btrfs_inode.h"
43 #include "ioctl.h"
44 #include "print-tree.h"
45 #include "volumes.h"
46 #include "ordered-data.h"
47
48 struct btrfs_iget_args {
49 u64 ino;
50 struct btrfs_root *root;
51 };
52
53 static struct inode_operations btrfs_dir_inode_operations;
54 static struct inode_operations btrfs_symlink_inode_operations;
55 static struct inode_operations btrfs_dir_ro_inode_operations;
56 static struct inode_operations btrfs_special_inode_operations;
57 static struct inode_operations btrfs_file_inode_operations;
58 static struct address_space_operations btrfs_aops;
59 static struct address_space_operations btrfs_symlink_aops;
60 static struct file_operations btrfs_dir_file_operations;
61 static struct extent_io_ops btrfs_extent_io_ops;
62
63 static struct kmem_cache *btrfs_inode_cachep;
64 struct kmem_cache *btrfs_trans_handle_cachep;
65 struct kmem_cache *btrfs_transaction_cachep;
66 struct kmem_cache *btrfs_bit_radix_cachep;
67 struct kmem_cache *btrfs_path_cachep;
68
69 #define S_SHIFT 12
70 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
71 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
72 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
73 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
74 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
75 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
76 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
77 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
78 };
79
80 int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
81 int for_del)
82 {
83 u64 total;
84 u64 used;
85 u64 thresh;
86 unsigned long flags;
87 int ret = 0;
88
89 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
90 total = btrfs_super_total_bytes(&root->fs_info->super_copy);
91 used = btrfs_super_bytes_used(&root->fs_info->super_copy);
92 if (for_del)
93 thresh = total * 90;
94 else
95 thresh = total * 85;
96
97 do_div(thresh, 100);
98
99 if (used + root->fs_info->delalloc_bytes + num_required > thresh)
100 ret = -ENOSPC;
101 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
102 return ret;
103 }
104
105 static int cow_file_range(struct inode *inode, u64 start, u64 end)
106 {
107 struct btrfs_root *root = BTRFS_I(inode)->root;
108 struct btrfs_trans_handle *trans;
109 u64 alloc_hint = 0;
110 u64 num_bytes;
111 u64 cur_alloc_size;
112 u64 blocksize = root->sectorsize;
113 u64 orig_num_bytes;
114 struct btrfs_key ins;
115 struct extent_map *em;
116 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
117 int ret = 0;
118
119 trans = btrfs_join_transaction(root, 1);
120 BUG_ON(!trans);
121 btrfs_set_trans_block_group(trans, inode);
122
123 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
124 num_bytes = max(blocksize, num_bytes);
125 orig_num_bytes = num_bytes;
126
127 if (alloc_hint == EXTENT_MAP_INLINE)
128 goto out;
129
130 BUG_ON(num_bytes > btrfs_super_total_bytes(&root->fs_info->super_copy));
131 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1);
132
133 while(num_bytes > 0) {
134 cur_alloc_size = min(num_bytes, root->fs_info->max_extent);
135 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
136 root->sectorsize, 0, 0,
137 (u64)-1, &ins, 1);
138 if (ret) {
139 WARN_ON(1);
140 goto out;
141 }
142 em = alloc_extent_map(GFP_NOFS);
143 em->start = start;
144 em->len = ins.offset;
145 em->block_start = ins.objectid;
146 em->bdev = root->fs_info->fs_devices->latest_bdev;
147 while(1) {
148 spin_lock(&em_tree->lock);
149 ret = add_extent_mapping(em_tree, em);
150 spin_unlock(&em_tree->lock);
151 if (ret != -EEXIST) {
152 free_extent_map(em);
153 break;
154 }
155 btrfs_drop_extent_cache(inode, start,
156 start + ins.offset - 1);
157 }
158
159 cur_alloc_size = ins.offset;
160 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
161 ins.offset);
162 BUG_ON(ret);
163 if (num_bytes < cur_alloc_size) {
164 printk("num_bytes %Lu cur_alloc %Lu\n", num_bytes,
165 cur_alloc_size);
166 break;
167 }
168 num_bytes -= cur_alloc_size;
169 alloc_hint = ins.objectid + ins.offset;
170 start += cur_alloc_size;
171 }
172 out:
173 btrfs_end_transaction(trans, root);
174 return ret;
175 }
176
177 static int run_delalloc_nocow(struct inode *inode, u64 start, u64 end)
178 {
179 u64 extent_start;
180 u64 extent_end;
181 u64 bytenr;
182 u64 cow_end;
183 u64 loops = 0;
184 u64 total_fs_bytes;
185 struct btrfs_root *root = BTRFS_I(inode)->root;
186 struct btrfs_block_group_cache *block_group;
187 struct extent_buffer *leaf;
188 int found_type;
189 struct btrfs_path *path;
190 struct btrfs_file_extent_item *item;
191 int ret;
192 int err;
193 struct btrfs_key found_key;
194
195 total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
196 path = btrfs_alloc_path();
197 BUG_ON(!path);
198 again:
199 ret = btrfs_lookup_file_extent(NULL, root, path,
200 inode->i_ino, start, 0);
201 if (ret < 0) {
202 btrfs_free_path(path);
203 return ret;
204 }
205
206 cow_end = end;
207 if (ret != 0) {
208 if (path->slots[0] == 0)
209 goto not_found;
210 path->slots[0]--;
211 }
212
213 leaf = path->nodes[0];
214 item = btrfs_item_ptr(leaf, path->slots[0],
215 struct btrfs_file_extent_item);
216
217 /* are we inside the extent that was found? */
218 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
219 found_type = btrfs_key_type(&found_key);
220 if (found_key.objectid != inode->i_ino ||
221 found_type != BTRFS_EXTENT_DATA_KEY)
222 goto not_found;
223
224 found_type = btrfs_file_extent_type(leaf, item);
225 extent_start = found_key.offset;
226 if (found_type == BTRFS_FILE_EXTENT_REG) {
227 u64 extent_num_bytes;
228
229 extent_num_bytes = btrfs_file_extent_num_bytes(leaf, item);
230 extent_end = extent_start + extent_num_bytes;
231 err = 0;
232
233 if (loops && start != extent_start)
234 goto not_found;
235
236 if (start < extent_start || start >= extent_end)
237 goto not_found;
238
239 cow_end = min(end, extent_end - 1);
240 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
241 if (bytenr == 0)
242 goto not_found;
243
244 if (btrfs_count_snapshots_in_path(root, path, inode->i_ino,
245 bytenr) != 1) {
246 goto not_found;
247 }
248
249 /*
250 * we may be called by the resizer, make sure we're inside
251 * the limits of the FS
252 */
253 block_group = btrfs_lookup_block_group(root->fs_info,
254 bytenr);
255 if (!block_group || block_group->ro)
256 goto not_found;
257
258 start = extent_end;
259 } else {
260 goto not_found;
261 }
262 loop:
263 if (start > end) {
264 btrfs_free_path(path);
265 return 0;
266 }
267 btrfs_release_path(root, path);
268 loops++;
269 goto again;
270
271 not_found:
272 cow_file_range(inode, start, end);
273 start = end + 1;
274 goto loop;
275 }
276
277 static int run_delalloc_range(struct inode *inode, u64 start, u64 end)
278 {
279 struct btrfs_root *root = BTRFS_I(inode)->root;
280 int ret;
281
282 if (btrfs_test_opt(root, NODATACOW) ||
283 btrfs_test_flag(inode, NODATACOW))
284 ret = run_delalloc_nocow(inode, start, end);
285 else
286 ret = cow_file_range(inode, start, end);
287
288 return ret;
289 }
290
291 int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
292 unsigned long old, unsigned long bits)
293 {
294 unsigned long flags;
295 if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
296 struct btrfs_root *root = BTRFS_I(inode)->root;
297 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
298 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
299 root->fs_info->delalloc_bytes += end - start + 1;
300 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
301 }
302 return 0;
303 }
304
305 int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
306 unsigned long old, unsigned long bits)
307 {
308 if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
309 struct btrfs_root *root = BTRFS_I(inode)->root;
310 unsigned long flags;
311
312 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
313 if (end - start + 1 > root->fs_info->delalloc_bytes) {
314 printk("warning: delalloc account %Lu %Lu\n",
315 end - start + 1, root->fs_info->delalloc_bytes);
316 root->fs_info->delalloc_bytes = 0;
317 BTRFS_I(inode)->delalloc_bytes = 0;
318 } else {
319 root->fs_info->delalloc_bytes -= end - start + 1;
320 BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
321 }
322 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
323 }
324 return 0;
325 }
326
327 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
328 size_t size, struct bio *bio)
329 {
330 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
331 struct btrfs_mapping_tree *map_tree;
332 u64 logical = bio->bi_sector << 9;
333 u64 length = 0;
334 u64 map_length;
335 int ret;
336
337 length = bio->bi_size;
338 map_tree = &root->fs_info->mapping_tree;
339 map_length = length;
340 ret = btrfs_map_block(map_tree, READ, logical,
341 &map_length, NULL, 0);
342
343 if (map_length < length + size) {
344 return 1;
345 }
346 return 0;
347 }
348
349 int __btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
350 int mirror_num)
351 {
352 struct btrfs_root *root = BTRFS_I(inode)->root;
353 int ret = 0;
354 struct btrfs_ordered_sum *sums;
355
356 ret = btrfs_csum_one_bio(root, bio, &sums);
357 BUG_ON(ret);
358
359 ret = btrfs_add_ordered_sum(inode, sums);
360 BUG_ON(ret);
361
362 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
363 }
364
365 int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
366 int mirror_num)
367 {
368 struct btrfs_root *root = BTRFS_I(inode)->root;
369 int ret = 0;
370
371 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
372 BUG_ON(ret);
373
374 if (!(rw & (1 << BIO_RW))) {
375 goto mapit;
376 }
377
378 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
379 inode, rw, bio, mirror_num,
380 __btrfs_submit_bio_hook);
381 mapit:
382 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
383 }
384
385 static int add_pending_csums(struct btrfs_trans_handle *trans,
386 struct inode *inode, u64 file_offset,
387 struct list_head *list)
388 {
389 struct list_head *cur;
390 struct btrfs_ordered_sum *sum;
391
392 btrfs_set_trans_block_group(trans, inode);
393 while(!list_empty(list)) {
394 cur = list->next;
395 sum = list_entry(cur, struct btrfs_ordered_sum, list);
396 mutex_lock(&BTRFS_I(inode)->csum_mutex);
397 btrfs_csum_file_blocks(trans, BTRFS_I(inode)->root,
398 inode, sum);
399 mutex_unlock(&BTRFS_I(inode)->csum_mutex);
400 list_del(&sum->list);
401 kfree(sum);
402 }
403 return 0;
404 }
405
406 struct btrfs_writepage_fixup {
407 struct page *page;
408 struct btrfs_work work;
409 };
410
411 /* see btrfs_writepage_start_hook for details on why this is required */
412 void btrfs_writepage_fixup_worker(struct btrfs_work *work)
413 {
414 struct btrfs_writepage_fixup *fixup;
415 struct btrfs_ordered_extent *ordered;
416 struct page *page;
417 struct inode *inode;
418 u64 page_start;
419 u64 page_end;
420
421 fixup = container_of(work, struct btrfs_writepage_fixup, work);
422 page = fixup->page;
423
424 lock_page(page);
425 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
426 ClearPageChecked(page);
427 goto out_page;
428 }
429
430 inode = page->mapping->host;
431 page_start = page_offset(page);
432 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
433
434 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
435 ordered = btrfs_lookup_ordered_extent(inode, page_start);
436 if (ordered)
437 goto out;
438
439 set_extent_delalloc(&BTRFS_I(inode)->io_tree, page_start, page_end,
440 GFP_NOFS);
441 ClearPageChecked(page);
442 out:
443 unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
444 out_page:
445 unlock_page(page);
446 page_cache_release(page);
447 }
448
449 /*
450 * There are a few paths in the higher layers of the kernel that directly
451 * set the page dirty bit without asking the filesystem if it is a
452 * good idea. This causes problems because we want to make sure COW
453 * properly happens and the data=ordered rules are followed.
454 *
455 * In our case any range that doesn't have the EXTENT_ORDERED bit set
456 * hasn't been properly setup for IO. We kick off an async process
457 * to fix it up. The async helper will wait for ordered extents, set
458 * the delalloc bit and make it safe to write the page.
459 */
460 int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
461 {
462 struct inode *inode = page->mapping->host;
463 struct btrfs_writepage_fixup *fixup;
464 struct btrfs_root *root = BTRFS_I(inode)->root;
465 int ret;
466
467 ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
468 EXTENT_ORDERED, 0);
469 if (ret)
470 return 0;
471
472 if (PageChecked(page))
473 return -EAGAIN;
474
475 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
476 if (!fixup)
477 return -EAGAIN;
478 printk("queueing worker to fixup page %lu %Lu\n", inode->i_ino, page_offset(page));
479 SetPageChecked(page);
480 page_cache_get(page);
481 fixup->work.func = btrfs_writepage_fixup_worker;
482 fixup->page = page;
483 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
484 return -EAGAIN;
485 }
486
487 int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
488 struct extent_state *state, int uptodate)
489 {
490 struct inode *inode = page->mapping->host;
491 struct btrfs_root *root = BTRFS_I(inode)->root;
492 struct btrfs_trans_handle *trans;
493 struct btrfs_ordered_extent *ordered_extent;
494 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
495 u64 alloc_hint = 0;
496 struct list_head list;
497 struct btrfs_key ins;
498 int ret;
499
500 ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
501 if (!ret) {
502 return 0;
503 }
504
505 trans = btrfs_join_transaction(root, 1);
506
507 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
508 BUG_ON(!ordered_extent);
509
510 lock_extent(io_tree, ordered_extent->file_offset,
511 ordered_extent->file_offset + ordered_extent->len - 1,
512 GFP_NOFS);
513
514 INIT_LIST_HEAD(&list);
515
516 ins.objectid = ordered_extent->start;
517 ins.offset = ordered_extent->len;
518 ins.type = BTRFS_EXTENT_ITEM_KEY;
519 ret = btrfs_alloc_reserved_extent(trans, root, root->root_key.objectid,
520 trans->transid, inode->i_ino,
521 ordered_extent->file_offset, &ins);
522 BUG_ON(ret);
523 ret = btrfs_drop_extents(trans, root, inode,
524 ordered_extent->file_offset,
525 ordered_extent->file_offset +
526 ordered_extent->len,
527 ordered_extent->file_offset, &alloc_hint);
528 BUG_ON(ret);
529 ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
530 ordered_extent->file_offset,
531 ordered_extent->start,
532 ordered_extent->len,
533 ordered_extent->len, 0);
534 BUG_ON(ret);
535 btrfs_drop_extent_cache(inode, ordered_extent->file_offset,
536 ordered_extent->file_offset +
537 ordered_extent->len - 1);
538 inode->i_blocks += ordered_extent->len >> 9;
539 unlock_extent(io_tree, ordered_extent->file_offset,
540 ordered_extent->file_offset + ordered_extent->len - 1,
541 GFP_NOFS);
542 add_pending_csums(trans, inode, ordered_extent->file_offset,
543 &ordered_extent->list);
544
545 btrfs_ordered_update_i_size(inode, ordered_extent);
546 btrfs_remove_ordered_extent(inode, ordered_extent);
547 /* once for us */
548 btrfs_put_ordered_extent(ordered_extent);
549 /* once for the tree */
550 btrfs_put_ordered_extent(ordered_extent);
551
552 btrfs_update_inode(trans, root, inode);
553 btrfs_end_transaction(trans, root);
554 return 0;
555 }
556
557 int btrfs_readpage_io_hook(struct page *page, u64 start, u64 end)
558 {
559 int ret = 0;
560 struct inode *inode = page->mapping->host;
561 struct btrfs_root *root = BTRFS_I(inode)->root;
562 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
563 struct btrfs_csum_item *item;
564 struct btrfs_path *path = NULL;
565 u32 csum;
566
567 if (btrfs_test_opt(root, NODATASUM) ||
568 btrfs_test_flag(inode, NODATASUM))
569 return 0;
570
571 path = btrfs_alloc_path();
572 item = btrfs_lookup_csum(NULL, root, path, inode->i_ino, start, 0);
573 if (IS_ERR(item)) {
574 ret = PTR_ERR(item);
575 /* a csum that isn't present is a preallocated region. */
576 if (ret == -ENOENT || ret == -EFBIG)
577 ret = 0;
578 csum = 0;
579 printk("no csum found for inode %lu start %Lu\n", inode->i_ino,
580 start);
581 goto out;
582 }
583 read_extent_buffer(path->nodes[0], &csum, (unsigned long)item,
584 BTRFS_CRC32_SIZE);
585 set_state_private(io_tree, start, csum);
586 out:
587 if (path)
588 btrfs_free_path(path);
589 return ret;
590 }
591
592 struct io_failure_record {
593 struct page *page;
594 u64 start;
595 u64 len;
596 u64 logical;
597 int last_mirror;
598 };
599
600 int btrfs_io_failed_hook(struct bio *failed_bio,
601 struct page *page, u64 start, u64 end,
602 struct extent_state *state)
603 {
604 struct io_failure_record *failrec = NULL;
605 u64 private;
606 struct extent_map *em;
607 struct inode *inode = page->mapping->host;
608 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
609 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
610 struct bio *bio;
611 int num_copies;
612 int ret;
613 int rw;
614 u64 logical;
615
616 ret = get_state_private(failure_tree, start, &private);
617 if (ret) {
618 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
619 if (!failrec)
620 return -ENOMEM;
621 failrec->start = start;
622 failrec->len = end - start + 1;
623 failrec->last_mirror = 0;
624
625 spin_lock(&em_tree->lock);
626 em = lookup_extent_mapping(em_tree, start, failrec->len);
627 if (em->start > start || em->start + em->len < start) {
628 free_extent_map(em);
629 em = NULL;
630 }
631 spin_unlock(&em_tree->lock);
632
633 if (!em || IS_ERR(em)) {
634 kfree(failrec);
635 return -EIO;
636 }
637 logical = start - em->start;
638 logical = em->block_start + logical;
639 failrec->logical = logical;
640 free_extent_map(em);
641 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
642 EXTENT_DIRTY, GFP_NOFS);
643 set_state_private(failure_tree, start,
644 (u64)(unsigned long)failrec);
645 } else {
646 failrec = (struct io_failure_record *)(unsigned long)private;
647 }
648 num_copies = btrfs_num_copies(
649 &BTRFS_I(inode)->root->fs_info->mapping_tree,
650 failrec->logical, failrec->len);
651 failrec->last_mirror++;
652 if (!state) {
653 spin_lock_irq(&BTRFS_I(inode)->io_tree.lock);
654 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
655 failrec->start,
656 EXTENT_LOCKED);
657 if (state && state->start != failrec->start)
658 state = NULL;
659 spin_unlock_irq(&BTRFS_I(inode)->io_tree.lock);
660 }
661 if (!state || failrec->last_mirror > num_copies) {
662 set_state_private(failure_tree, failrec->start, 0);
663 clear_extent_bits(failure_tree, failrec->start,
664 failrec->start + failrec->len - 1,
665 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
666 kfree(failrec);
667 return -EIO;
668 }
669 bio = bio_alloc(GFP_NOFS, 1);
670 bio->bi_private = state;
671 bio->bi_end_io = failed_bio->bi_end_io;
672 bio->bi_sector = failrec->logical >> 9;
673 bio->bi_bdev = failed_bio->bi_bdev;
674 bio->bi_size = 0;
675 bio_add_page(bio, page, failrec->len, start - page_offset(page));
676 if (failed_bio->bi_rw & (1 << BIO_RW))
677 rw = WRITE;
678 else
679 rw = READ;
680
681 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
682 failrec->last_mirror);
683 return 0;
684 }
685
686 int btrfs_clean_io_failures(struct inode *inode, u64 start)
687 {
688 u64 private;
689 u64 private_failure;
690 struct io_failure_record *failure;
691 int ret;
692
693 private = 0;
694 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
695 (u64)-1, 1, EXTENT_DIRTY)) {
696 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
697 start, &private_failure);
698 if (ret == 0) {
699 failure = (struct io_failure_record *)(unsigned long)
700 private_failure;
701 set_state_private(&BTRFS_I(inode)->io_failure_tree,
702 failure->start, 0);
703 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
704 failure->start,
705 failure->start + failure->len - 1,
706 EXTENT_DIRTY | EXTENT_LOCKED,
707 GFP_NOFS);
708 kfree(failure);
709 }
710 }
711 return 0;
712 }
713
714 int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
715 struct extent_state *state)
716 {
717 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
718 struct inode *inode = page->mapping->host;
719 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
720 char *kaddr;
721 u64 private = ~(u32)0;
722 int ret;
723 struct btrfs_root *root = BTRFS_I(inode)->root;
724 u32 csum = ~(u32)0;
725 unsigned long flags;
726
727 if (btrfs_test_opt(root, NODATASUM) ||
728 btrfs_test_flag(inode, NODATASUM))
729 return 0;
730 if (state && state->start == start) {
731 private = state->private;
732 ret = 0;
733 } else {
734 ret = get_state_private(io_tree, start, &private);
735 }
736 local_irq_save(flags);
737 kaddr = kmap_atomic(page, KM_IRQ0);
738 if (ret) {
739 goto zeroit;
740 }
741 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
742 btrfs_csum_final(csum, (char *)&csum);
743 if (csum != private) {
744 goto zeroit;
745 }
746 kunmap_atomic(kaddr, KM_IRQ0);
747 local_irq_restore(flags);
748
749 /* if the io failure tree for this inode is non-empty,
750 * check to see if we've recovered from a failed IO
751 */
752 btrfs_clean_io_failures(inode, start);
753 return 0;
754
755 zeroit:
756 printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
757 page->mapping->host->i_ino, (unsigned long long)start, csum,
758 private);
759 memset(kaddr + offset, 1, end - start + 1);
760 flush_dcache_page(page);
761 kunmap_atomic(kaddr, KM_IRQ0);
762 local_irq_restore(flags);
763 if (private == 0)
764 return 0;
765 return -EIO;
766 }
767
768 void btrfs_read_locked_inode(struct inode *inode)
769 {
770 struct btrfs_path *path;
771 struct extent_buffer *leaf;
772 struct btrfs_inode_item *inode_item;
773 struct btrfs_timespec *tspec;
774 struct btrfs_root *root = BTRFS_I(inode)->root;
775 struct btrfs_key location;
776 u64 alloc_group_block;
777 u32 rdev;
778 int ret;
779
780 path = btrfs_alloc_path();
781 BUG_ON(!path);
782 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
783
784 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
785 if (ret)
786 goto make_bad;
787
788 leaf = path->nodes[0];
789 inode_item = btrfs_item_ptr(leaf, path->slots[0],
790 struct btrfs_inode_item);
791
792 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
793 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
794 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
795 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
796 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
797
798 tspec = btrfs_inode_atime(inode_item);
799 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
800 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
801
802 tspec = btrfs_inode_mtime(inode_item);
803 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
804 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
805
806 tspec = btrfs_inode_ctime(inode_item);
807 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
808 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
809
810 inode->i_blocks = btrfs_inode_nblocks(leaf, inode_item);
811 inode->i_generation = btrfs_inode_generation(leaf, inode_item);
812 inode->i_rdev = 0;
813 rdev = btrfs_inode_rdev(leaf, inode_item);
814
815 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
816 BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
817 alloc_group_block);
818 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
819 if (!BTRFS_I(inode)->block_group) {
820 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
821 NULL, 0,
822 BTRFS_BLOCK_GROUP_METADATA, 0);
823 }
824 btrfs_free_path(path);
825 inode_item = NULL;
826
827 switch (inode->i_mode & S_IFMT) {
828 case S_IFREG:
829 inode->i_mapping->a_ops = &btrfs_aops;
830 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
831 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
832 inode->i_fop = &btrfs_file_operations;
833 inode->i_op = &btrfs_file_inode_operations;
834 break;
835 case S_IFDIR:
836 inode->i_fop = &btrfs_dir_file_operations;
837 if (root == root->fs_info->tree_root)
838 inode->i_op = &btrfs_dir_ro_inode_operations;
839 else
840 inode->i_op = &btrfs_dir_inode_operations;
841 break;
842 case S_IFLNK:
843 inode->i_op = &btrfs_symlink_inode_operations;
844 inode->i_mapping->a_ops = &btrfs_symlink_aops;
845 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
846 break;
847 default:
848 init_special_inode(inode, inode->i_mode, rdev);
849 break;
850 }
851 return;
852
853 make_bad:
854 btrfs_free_path(path);
855 make_bad_inode(inode);
856 }
857
858 static void fill_inode_item(struct extent_buffer *leaf,
859 struct btrfs_inode_item *item,
860 struct inode *inode)
861 {
862 btrfs_set_inode_uid(leaf, item, inode->i_uid);
863 btrfs_set_inode_gid(leaf, item, inode->i_gid);
864 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
865 btrfs_set_inode_mode(leaf, item, inode->i_mode);
866 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
867
868 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
869 inode->i_atime.tv_sec);
870 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
871 inode->i_atime.tv_nsec);
872
873 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
874 inode->i_mtime.tv_sec);
875 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
876 inode->i_mtime.tv_nsec);
877
878 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
879 inode->i_ctime.tv_sec);
880 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
881 inode->i_ctime.tv_nsec);
882
883 btrfs_set_inode_nblocks(leaf, item, inode->i_blocks);
884 btrfs_set_inode_generation(leaf, item, inode->i_generation);
885 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
886 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
887 btrfs_set_inode_block_group(leaf, item,
888 BTRFS_I(inode)->block_group->key.objectid);
889 }
890
891 int btrfs_update_inode(struct btrfs_trans_handle *trans,
892 struct btrfs_root *root,
893 struct inode *inode)
894 {
895 struct btrfs_inode_item *inode_item;
896 struct btrfs_path *path;
897 struct extent_buffer *leaf;
898 int ret;
899
900 path = btrfs_alloc_path();
901 BUG_ON(!path);
902 ret = btrfs_lookup_inode(trans, root, path,
903 &BTRFS_I(inode)->location, 1);
904 if (ret) {
905 if (ret > 0)
906 ret = -ENOENT;
907 goto failed;
908 }
909
910 leaf = path->nodes[0];
911 inode_item = btrfs_item_ptr(leaf, path->slots[0],
912 struct btrfs_inode_item);
913
914 fill_inode_item(leaf, inode_item, inode);
915 btrfs_mark_buffer_dirty(leaf);
916 btrfs_set_inode_last_trans(trans, inode);
917 ret = 0;
918 failed:
919 btrfs_free_path(path);
920 return ret;
921 }
922
923
924 static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
925 struct btrfs_root *root,
926 struct inode *dir,
927 struct dentry *dentry)
928 {
929 struct btrfs_path *path;
930 const char *name = dentry->d_name.name;
931 int name_len = dentry->d_name.len;
932 int ret = 0;
933 struct extent_buffer *leaf;
934 struct btrfs_dir_item *di;
935 struct btrfs_key key;
936
937 path = btrfs_alloc_path();
938 if (!path) {
939 ret = -ENOMEM;
940 goto err;
941 }
942
943 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
944 name, name_len, -1);
945 if (IS_ERR(di)) {
946 ret = PTR_ERR(di);
947 goto err;
948 }
949 if (!di) {
950 ret = -ENOENT;
951 goto err;
952 }
953 leaf = path->nodes[0];
954 btrfs_dir_item_key_to_cpu(leaf, di, &key);
955 ret = btrfs_delete_one_dir_name(trans, root, path, di);
956 if (ret)
957 goto err;
958 btrfs_release_path(root, path);
959
960 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
961 key.objectid, name, name_len, -1);
962 if (IS_ERR(di)) {
963 ret = PTR_ERR(di);
964 goto err;
965 }
966 if (!di) {
967 ret = -ENOENT;
968 goto err;
969 }
970 ret = btrfs_delete_one_dir_name(trans, root, path, di);
971 btrfs_release_path(root, path);
972
973 dentry->d_inode->i_ctime = dir->i_ctime;
974 ret = btrfs_del_inode_ref(trans, root, name, name_len,
975 dentry->d_inode->i_ino,
976 dentry->d_parent->d_inode->i_ino);
977 if (ret) {
978 printk("failed to delete reference to %.*s, "
979 "inode %lu parent %lu\n", name_len, name,
980 dentry->d_inode->i_ino,
981 dentry->d_parent->d_inode->i_ino);
982 }
983 err:
984 btrfs_free_path(path);
985 if (!ret) {
986 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
987 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
988 btrfs_update_inode(trans, root, dir);
989 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
990 dentry->d_inode->i_nlink--;
991 #else
992 drop_nlink(dentry->d_inode);
993 #endif
994 ret = btrfs_update_inode(trans, root, dentry->d_inode);
995 dir->i_sb->s_dirt = 1;
996 }
997 return ret;
998 }
999
1000 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
1001 {
1002 struct btrfs_root *root;
1003 struct btrfs_trans_handle *trans;
1004 int ret;
1005 unsigned long nr = 0;
1006
1007 root = BTRFS_I(dir)->root;
1008
1009 ret = btrfs_check_free_space(root, 1, 1);
1010 if (ret)
1011 goto fail;
1012
1013 trans = btrfs_start_transaction(root, 1);
1014
1015 btrfs_set_trans_block_group(trans, dir);
1016 ret = btrfs_unlink_trans(trans, root, dir, dentry);
1017 nr = trans->blocks_used;
1018
1019 btrfs_end_transaction_throttle(trans, root);
1020 fail:
1021 btrfs_btree_balance_dirty(root, nr);
1022 return ret;
1023 }
1024
1025 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
1026 {
1027 struct inode *inode = dentry->d_inode;
1028 int err = 0;
1029 int ret;
1030 struct btrfs_root *root = BTRFS_I(dir)->root;
1031 struct btrfs_trans_handle *trans;
1032 unsigned long nr = 0;
1033
1034 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
1035 return -ENOTEMPTY;
1036 }
1037
1038 ret = btrfs_check_free_space(root, 1, 1);
1039 if (ret)
1040 goto fail;
1041
1042 trans = btrfs_start_transaction(root, 1);
1043 btrfs_set_trans_block_group(trans, dir);
1044
1045 /* now the directory is empty */
1046 err = btrfs_unlink_trans(trans, root, dir, dentry);
1047 if (!err) {
1048 btrfs_i_size_write(inode, 0);
1049 }
1050
1051 nr = trans->blocks_used;
1052 ret = btrfs_end_transaction_throttle(trans, root);
1053 fail:
1054 btrfs_btree_balance_dirty(root, nr);
1055
1056 if (ret && !err)
1057 err = ret;
1058 return err;
1059 }
1060
1061 /*
1062 * this can truncate away extent items, csum items and directory items.
1063 * It starts at a high offset and removes keys until it can't find
1064 * any higher than i_size.
1065 *
1066 * csum items that cross the new i_size are truncated to the new size
1067 * as well.
1068 */
1069 static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
1070 struct btrfs_root *root,
1071 struct inode *inode,
1072 u32 min_type)
1073 {
1074 int ret;
1075 struct btrfs_path *path;
1076 struct btrfs_key key;
1077 struct btrfs_key found_key;
1078 u32 found_type;
1079 struct extent_buffer *leaf;
1080 struct btrfs_file_extent_item *fi;
1081 u64 extent_start = 0;
1082 u64 extent_num_bytes = 0;
1083 u64 item_end = 0;
1084 u64 root_gen = 0;
1085 u64 root_owner = 0;
1086 int found_extent;
1087 int del_item;
1088 int pending_del_nr = 0;
1089 int pending_del_slot = 0;
1090 int extent_type = -1;
1091 u64 mask = root->sectorsize - 1;
1092
1093 btrfs_drop_extent_cache(inode, inode->i_size & (~mask), (u64)-1);
1094 path = btrfs_alloc_path();
1095 path->reada = -1;
1096 BUG_ON(!path);
1097
1098 /* FIXME, add redo link to tree so we don't leak on crash */
1099 key.objectid = inode->i_ino;
1100 key.offset = (u64)-1;
1101 key.type = (u8)-1;
1102
1103 btrfs_init_path(path);
1104 search_again:
1105 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1106 if (ret < 0) {
1107 goto error;
1108 }
1109 if (ret > 0) {
1110 BUG_ON(path->slots[0] == 0);
1111 path->slots[0]--;
1112 }
1113
1114 while(1) {
1115 fi = NULL;
1116 leaf = path->nodes[0];
1117 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1118 found_type = btrfs_key_type(&found_key);
1119
1120 if (found_key.objectid != inode->i_ino)
1121 break;
1122
1123 if (found_type < min_type)
1124 break;
1125
1126 item_end = found_key.offset;
1127 if (found_type == BTRFS_EXTENT_DATA_KEY) {
1128 fi = btrfs_item_ptr(leaf, path->slots[0],
1129 struct btrfs_file_extent_item);
1130 extent_type = btrfs_file_extent_type(leaf, fi);
1131 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
1132 item_end +=
1133 btrfs_file_extent_num_bytes(leaf, fi);
1134 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1135 struct btrfs_item *item = btrfs_item_nr(leaf,
1136 path->slots[0]);
1137 item_end += btrfs_file_extent_inline_len(leaf,
1138 item);
1139 }
1140 item_end--;
1141 }
1142 if (found_type == BTRFS_CSUM_ITEM_KEY) {
1143 ret = btrfs_csum_truncate(trans, root, path,
1144 inode->i_size);
1145 BUG_ON(ret);
1146 }
1147 if (item_end < inode->i_size) {
1148 if (found_type == BTRFS_DIR_ITEM_KEY) {
1149 found_type = BTRFS_INODE_ITEM_KEY;
1150 } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
1151 found_type = BTRFS_CSUM_ITEM_KEY;
1152 } else if (found_type == BTRFS_EXTENT_DATA_KEY) {
1153 found_type = BTRFS_XATTR_ITEM_KEY;
1154 } else if (found_type == BTRFS_XATTR_ITEM_KEY) {
1155 found_type = BTRFS_INODE_REF_KEY;
1156 } else if (found_type) {
1157 found_type--;
1158 } else {
1159 break;
1160 }
1161 btrfs_set_key_type(&key, found_type);
1162 goto next;
1163 }
1164 if (found_key.offset >= inode->i_size)
1165 del_item = 1;
1166 else
1167 del_item = 0;
1168 found_extent = 0;
1169
1170 /* FIXME, shrink the extent if the ref count is only 1 */
1171 if (found_type != BTRFS_EXTENT_DATA_KEY)
1172 goto delete;
1173
1174 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
1175 u64 num_dec;
1176 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
1177 if (!del_item) {
1178 u64 orig_num_bytes =
1179 btrfs_file_extent_num_bytes(leaf, fi);
1180 extent_num_bytes = inode->i_size -
1181 found_key.offset + root->sectorsize - 1;
1182 extent_num_bytes = extent_num_bytes &
1183 ~((u64)root->sectorsize - 1);
1184 btrfs_set_file_extent_num_bytes(leaf, fi,
1185 extent_num_bytes);
1186 num_dec = (orig_num_bytes -
1187 extent_num_bytes);
1188 if (extent_start != 0)
1189 dec_i_blocks(inode, num_dec);
1190 btrfs_mark_buffer_dirty(leaf);
1191 } else {
1192 extent_num_bytes =
1193 btrfs_file_extent_disk_num_bytes(leaf,
1194 fi);
1195 /* FIXME blocksize != 4096 */
1196 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
1197 if (extent_start != 0) {
1198 found_extent = 1;
1199 dec_i_blocks(inode, num_dec);
1200 }
1201 root_gen = btrfs_header_generation(leaf);
1202 root_owner = btrfs_header_owner(leaf);
1203 }
1204 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1205 if (!del_item) {
1206 u32 newsize = inode->i_size - found_key.offset;
1207 dec_i_blocks(inode, item_end + 1 -
1208 found_key.offset - newsize);
1209 newsize =
1210 btrfs_file_extent_calc_inline_size(newsize);
1211 ret = btrfs_truncate_item(trans, root, path,
1212 newsize, 1);
1213 BUG_ON(ret);
1214 } else {
1215 dec_i_blocks(inode, item_end + 1 -
1216 found_key.offset);
1217 }
1218 }
1219 delete:
1220 if (del_item) {
1221 if (!pending_del_nr) {
1222 /* no pending yet, add ourselves */
1223 pending_del_slot = path->slots[0];
1224 pending_del_nr = 1;
1225 } else if (pending_del_nr &&
1226 path->slots[0] + 1 == pending_del_slot) {
1227 /* hop on the pending chunk */
1228 pending_del_nr++;
1229 pending_del_slot = path->slots[0];
1230 } else {
1231 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path->slots[0], pending_del_nr, pending_del_slot);
1232 }
1233 } else {
1234 break;
1235 }
1236 if (found_extent) {
1237 ret = btrfs_free_extent(trans, root, extent_start,
1238 extent_num_bytes,
1239 root_owner,
1240 root_gen, inode->i_ino,
1241 found_key.offset, 0);
1242 BUG_ON(ret);
1243 }
1244 next:
1245 if (path->slots[0] == 0) {
1246 if (pending_del_nr)
1247 goto del_pending;
1248 btrfs_release_path(root, path);
1249 goto search_again;
1250 }
1251
1252 path->slots[0]--;
1253 if (pending_del_nr &&
1254 path->slots[0] + 1 != pending_del_slot) {
1255 struct btrfs_key debug;
1256 del_pending:
1257 btrfs_item_key_to_cpu(path->nodes[0], &debug,
1258 pending_del_slot);
1259 ret = btrfs_del_items(trans, root, path,
1260 pending_del_slot,
1261 pending_del_nr);
1262 BUG_ON(ret);
1263 pending_del_nr = 0;
1264 btrfs_release_path(root, path);
1265 goto search_again;
1266 }
1267 }
1268 ret = 0;
1269 error:
1270 if (pending_del_nr) {
1271 ret = btrfs_del_items(trans, root, path, pending_del_slot,
1272 pending_del_nr);
1273 }
1274 btrfs_free_path(path);
1275 inode->i_sb->s_dirt = 1;
1276 return ret;
1277 }
1278
1279 /*
1280 * taken from block_truncate_page, but does cow as it zeros out
1281 * any bytes left in the last page in the file.
1282 */
1283 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
1284 {
1285 struct inode *inode = mapping->host;
1286 struct btrfs_root *root = BTRFS_I(inode)->root;
1287 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1288 struct btrfs_ordered_extent *ordered;
1289 char *kaddr;
1290 u32 blocksize = root->sectorsize;
1291 pgoff_t index = from >> PAGE_CACHE_SHIFT;
1292 unsigned offset = from & (PAGE_CACHE_SIZE-1);
1293 struct page *page;
1294 int ret = 0;
1295 u64 page_start;
1296 u64 page_end;
1297
1298 if ((offset & (blocksize - 1)) == 0)
1299 goto out;
1300
1301 ret = -ENOMEM;
1302 again:
1303 page = grab_cache_page(mapping, index);
1304 if (!page)
1305 goto out;
1306
1307 page_start = page_offset(page);
1308 page_end = page_start + PAGE_CACHE_SIZE - 1;
1309
1310 if (!PageUptodate(page)) {
1311 ret = btrfs_readpage(NULL, page);
1312 lock_page(page);
1313 if (page->mapping != mapping) {
1314 unlock_page(page);
1315 page_cache_release(page);
1316 goto again;
1317 }
1318 if (!PageUptodate(page)) {
1319 ret = -EIO;
1320 goto out;
1321 }
1322 }
1323 wait_on_page_writeback(page);
1324
1325 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
1326 set_page_extent_mapped(page);
1327
1328 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1329 if (ordered) {
1330 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
1331 unlock_page(page);
1332 page_cache_release(page);
1333 btrfs_wait_ordered_extent(inode, ordered);
1334 btrfs_put_ordered_extent(ordered);
1335 goto again;
1336 }
1337
1338 set_extent_delalloc(&BTRFS_I(inode)->io_tree, page_start,
1339 page_end, GFP_NOFS);
1340 ret = 0;
1341 if (offset != PAGE_CACHE_SIZE) {
1342 kaddr = kmap(page);
1343 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
1344 flush_dcache_page(page);
1345 kunmap(page);
1346 }
1347 ClearPageChecked(page);
1348 set_page_dirty(page);
1349 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
1350
1351 unlock_page(page);
1352 page_cache_release(page);
1353 out:
1354 return ret;
1355 }
1356
1357 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
1358 {
1359 struct inode *inode = dentry->d_inode;
1360 int err;
1361
1362 err = inode_change_ok(inode, attr);
1363 if (err)
1364 return err;
1365
1366 if (S_ISREG(inode->i_mode) &&
1367 attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
1368 struct btrfs_trans_handle *trans;
1369 struct btrfs_root *root = BTRFS_I(inode)->root;
1370 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1371
1372 u64 mask = root->sectorsize - 1;
1373 u64 hole_start = (inode->i_size + mask) & ~mask;
1374 u64 block_end = (attr->ia_size + mask) & ~mask;
1375 u64 hole_size;
1376 u64 alloc_hint = 0;
1377
1378 if (attr->ia_size <= hole_start)
1379 goto out;
1380
1381 err = btrfs_check_free_space(root, 1, 0);
1382 if (err)
1383 goto fail;
1384
1385 btrfs_truncate_page(inode->i_mapping, inode->i_size);
1386
1387 hole_size = block_end - hole_start;
1388 btrfs_wait_ordered_range(inode, hole_start, hole_size);
1389 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1390
1391 trans = btrfs_start_transaction(root, 1);
1392 btrfs_set_trans_block_group(trans, inode);
1393 err = btrfs_drop_extents(trans, root, inode,
1394 hole_start, block_end, hole_start,
1395 &alloc_hint);
1396
1397 if (alloc_hint != EXTENT_MAP_INLINE) {
1398 err = btrfs_insert_file_extent(trans, root,
1399 inode->i_ino,
1400 hole_start, 0, 0,
1401 hole_size, 0);
1402 btrfs_drop_extent_cache(inode, hole_start,
1403 (u64)-1);
1404 btrfs_check_file(root, inode);
1405 }
1406 btrfs_end_transaction(trans, root);
1407 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1408 if (err)
1409 return err;
1410 }
1411 out:
1412 err = inode_setattr(inode, attr);
1413 fail:
1414 return err;
1415 }
1416
1417 void btrfs_delete_inode(struct inode *inode)
1418 {
1419 struct btrfs_trans_handle *trans;
1420 struct btrfs_root *root = BTRFS_I(inode)->root;
1421 unsigned long nr;
1422 int ret;
1423
1424 btrfs_wait_ordered_range(inode, 0, (u64)-1);
1425 truncate_inode_pages(&inode->i_data, 0);
1426 if (is_bad_inode(inode)) {
1427 goto no_delete;
1428 }
1429
1430 btrfs_i_size_write(inode, 0);
1431 trans = btrfs_start_transaction(root, 1);
1432
1433 btrfs_set_trans_block_group(trans, inode);
1434 ret = btrfs_truncate_in_trans(trans, root, inode, 0);
1435 if (ret)
1436 goto no_delete_lock;
1437
1438 nr = trans->blocks_used;
1439 clear_inode(inode);
1440
1441 btrfs_end_transaction(trans, root);
1442 btrfs_btree_balance_dirty(root, nr);
1443 return;
1444
1445 no_delete_lock:
1446 nr = trans->blocks_used;
1447 btrfs_end_transaction(trans, root);
1448 btrfs_btree_balance_dirty(root, nr);
1449 no_delete:
1450 clear_inode(inode);
1451 }
1452
1453 /*
1454 * this returns the key found in the dir entry in the location pointer.
1455 * If no dir entries were found, location->objectid is 0.
1456 */
1457 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
1458 struct btrfs_key *location)
1459 {
1460 const char *name = dentry->d_name.name;
1461 int namelen = dentry->d_name.len;
1462 struct btrfs_dir_item *di;
1463 struct btrfs_path *path;
1464 struct btrfs_root *root = BTRFS_I(dir)->root;
1465 int ret = 0;
1466
1467 if (namelen == 1 && strcmp(name, ".") == 0) {
1468 location->objectid = dir->i_ino;
1469 location->type = BTRFS_INODE_ITEM_KEY;
1470 location->offset = 0;
1471 return 0;
1472 }
1473 path = btrfs_alloc_path();
1474 BUG_ON(!path);
1475
1476 if (namelen == 2 && strcmp(name, "..") == 0) {
1477 struct btrfs_key key;
1478 struct extent_buffer *leaf;
1479 u32 nritems;
1480 int slot;
1481
1482 key.objectid = dir->i_ino;
1483 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1484 key.offset = 0;
1485 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1486 BUG_ON(ret == 0);
1487 ret = 0;
1488
1489 leaf = path->nodes[0];
1490 slot = path->slots[0];
1491 nritems = btrfs_header_nritems(leaf);
1492 if (slot >= nritems)
1493 goto out_err;
1494
1495 btrfs_item_key_to_cpu(leaf, &key, slot);
1496 if (key.objectid != dir->i_ino ||
1497 key.type != BTRFS_INODE_REF_KEY) {
1498 goto out_err;
1499 }
1500 location->objectid = key.offset;
1501 location->type = BTRFS_INODE_ITEM_KEY;
1502 location->offset = 0;
1503 goto out;
1504 }
1505
1506 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
1507 namelen, 0);
1508 if (IS_ERR(di))
1509 ret = PTR_ERR(di);
1510 if (!di || IS_ERR(di)) {
1511 goto out_err;
1512 }
1513 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
1514 out:
1515 btrfs_free_path(path);
1516 return ret;
1517 out_err:
1518 location->objectid = 0;
1519 goto out;
1520 }
1521
1522 /*
1523 * when we hit a tree root in a directory, the btrfs part of the inode
1524 * needs to be changed to reflect the root directory of the tree root. This
1525 * is kind of like crossing a mount point.
1526 */
1527 static int fixup_tree_root_location(struct btrfs_root *root,
1528 struct btrfs_key *location,
1529 struct btrfs_root **sub_root,
1530 struct dentry *dentry)
1531 {
1532 struct btrfs_path *path;
1533 struct btrfs_root_item *ri;
1534
1535 if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
1536 return 0;
1537 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1538 return 0;
1539
1540 path = btrfs_alloc_path();
1541 BUG_ON(!path);
1542
1543 *sub_root = btrfs_read_fs_root(root->fs_info, location,
1544 dentry->d_name.name,
1545 dentry->d_name.len);
1546 if (IS_ERR(*sub_root))
1547 return PTR_ERR(*sub_root);
1548
1549 ri = &(*sub_root)->root_item;
1550 location->objectid = btrfs_root_dirid(ri);
1551 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1552 location->offset = 0;
1553
1554 btrfs_free_path(path);
1555 return 0;
1556 }
1557
1558 static int btrfs_init_locked_inode(struct inode *inode, void *p)
1559 {
1560 struct btrfs_iget_args *args = p;
1561 inode->i_ino = args->ino;
1562 BTRFS_I(inode)->root = args->root;
1563 BTRFS_I(inode)->delalloc_bytes = 0;
1564 BTRFS_I(inode)->disk_i_size = 0;
1565 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1566 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1567 inode->i_mapping, GFP_NOFS);
1568 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
1569 inode->i_mapping, GFP_NOFS);
1570 mutex_init(&BTRFS_I(inode)->csum_mutex);
1571 return 0;
1572 }
1573
1574 static int btrfs_find_actor(struct inode *inode, void *opaque)
1575 {
1576 struct btrfs_iget_args *args = opaque;
1577 return (args->ino == inode->i_ino &&
1578 args->root == BTRFS_I(inode)->root);
1579 }
1580
1581 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
1582 u64 root_objectid)
1583 {
1584 struct btrfs_iget_args args;
1585 args.ino = objectid;
1586 args.root = btrfs_lookup_fs_root(btrfs_sb(s)->fs_info, root_objectid);
1587
1588 if (!args.root)
1589 return NULL;
1590
1591 return ilookup5(s, objectid, btrfs_find_actor, (void *)&args);
1592 }
1593
1594 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
1595 struct btrfs_root *root)
1596 {
1597 struct inode *inode;
1598 struct btrfs_iget_args args;
1599 args.ino = objectid;
1600 args.root = root;
1601
1602 inode = iget5_locked(s, objectid, btrfs_find_actor,
1603 btrfs_init_locked_inode,
1604 (void *)&args);
1605 return inode;
1606 }
1607
1608 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
1609 struct nameidata *nd)
1610 {
1611 struct inode * inode;
1612 struct btrfs_inode *bi = BTRFS_I(dir);
1613 struct btrfs_root *root = bi->root;
1614 struct btrfs_root *sub_root = root;
1615 struct btrfs_key location;
1616 int ret;
1617
1618 if (dentry->d_name.len > BTRFS_NAME_LEN)
1619 return ERR_PTR(-ENAMETOOLONG);
1620
1621 ret = btrfs_inode_by_name(dir, dentry, &location);
1622
1623 if (ret < 0)
1624 return ERR_PTR(ret);
1625
1626 inode = NULL;
1627 if (location.objectid) {
1628 ret = fixup_tree_root_location(root, &location, &sub_root,
1629 dentry);
1630 if (ret < 0)
1631 return ERR_PTR(ret);
1632 if (ret > 0)
1633 return ERR_PTR(-ENOENT);
1634 inode = btrfs_iget_locked(dir->i_sb, location.objectid,
1635 sub_root);
1636 if (!inode)
1637 return ERR_PTR(-EACCES);
1638 if (inode->i_state & I_NEW) {
1639 /* the inode and parent dir are two different roots */
1640 if (sub_root != root) {
1641 igrab(inode);
1642 sub_root->inode = inode;
1643 }
1644 BTRFS_I(inode)->root = sub_root;
1645 memcpy(&BTRFS_I(inode)->location, &location,
1646 sizeof(location));
1647 btrfs_read_locked_inode(inode);
1648 unlock_new_inode(inode);
1649 }
1650 }
1651 return d_splice_alias(inode, dentry);
1652 }
1653
1654 static unsigned char btrfs_filetype_table[] = {
1655 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
1656 };
1657
1658 static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
1659 {
1660 struct inode *inode = filp->f_dentry->d_inode;
1661 struct btrfs_root *root = BTRFS_I(inode)->root;
1662 struct btrfs_item *item;
1663 struct btrfs_dir_item *di;
1664 struct btrfs_key key;
1665 struct btrfs_key found_key;
1666 struct btrfs_path *path;
1667 int ret;
1668 u32 nritems;
1669 struct extent_buffer *leaf;
1670 int slot;
1671 int advance;
1672 unsigned char d_type;
1673 int over = 0;
1674 u32 di_cur;
1675 u32 di_total;
1676 u32 di_len;
1677 int key_type = BTRFS_DIR_INDEX_KEY;
1678 char tmp_name[32];
1679 char *name_ptr;
1680 int name_len;
1681
1682 /* FIXME, use a real flag for deciding about the key type */
1683 if (root->fs_info->tree_root == root)
1684 key_type = BTRFS_DIR_ITEM_KEY;
1685
1686 /* special case for "." */
1687 if (filp->f_pos == 0) {
1688 over = filldir(dirent, ".", 1,
1689 1, inode->i_ino,
1690 DT_DIR);
1691 if (over)
1692 return 0;
1693 filp->f_pos = 1;
1694 }
1695
1696 key.objectid = inode->i_ino;
1697 path = btrfs_alloc_path();
1698 path->reada = 2;
1699
1700 /* special case for .., just use the back ref */
1701 if (filp->f_pos == 1) {
1702 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1703 key.offset = 0;
1704 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1705 BUG_ON(ret == 0);
1706 leaf = path->nodes[0];
1707 slot = path->slots[0];
1708 nritems = btrfs_header_nritems(leaf);
1709 if (slot >= nritems) {
1710 btrfs_release_path(root, path);
1711 goto read_dir_items;
1712 }
1713 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1714 btrfs_release_path(root, path);
1715 if (found_key.objectid != key.objectid ||
1716 found_key.type != BTRFS_INODE_REF_KEY)
1717 goto read_dir_items;
1718 over = filldir(dirent, "..", 2,
1719 2, found_key.offset, DT_DIR);
1720 if (over)
1721 goto nopos;
1722 filp->f_pos = 2;
1723 }
1724
1725 read_dir_items:
1726 btrfs_set_key_type(&key, key_type);
1727 key.offset = filp->f_pos;
1728
1729 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1730 if (ret < 0)
1731 goto err;
1732 advance = 0;
1733 while(1) {
1734 leaf = path->nodes[0];
1735 nritems = btrfs_header_nritems(leaf);
1736 slot = path->slots[0];
1737 if (advance || slot >= nritems) {
1738 if (slot >= nritems -1) {
1739 ret = btrfs_next_leaf(root, path);
1740 if (ret)
1741 break;
1742 leaf = path->nodes[0];
1743 nritems = btrfs_header_nritems(leaf);
1744 slot = path->slots[0];
1745 } else {
1746 slot++;
1747 path->slots[0]++;
1748 }
1749 }
1750 advance = 1;
1751 item = btrfs_item_nr(leaf, slot);
1752 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1753
1754 if (found_key.objectid != key.objectid)
1755 break;
1756 if (btrfs_key_type(&found_key) != key_type)
1757 break;
1758 if (found_key.offset < filp->f_pos)
1759 continue;
1760
1761 filp->f_pos = found_key.offset;
1762 advance = 1;
1763 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
1764 di_cur = 0;
1765 di_total = btrfs_item_size(leaf, item);
1766 while(di_cur < di_total) {
1767 struct btrfs_key location;
1768
1769 name_len = btrfs_dir_name_len(leaf, di);
1770 if (name_len < 32) {
1771 name_ptr = tmp_name;
1772 } else {
1773 name_ptr = kmalloc(name_len, GFP_NOFS);
1774 BUG_ON(!name_ptr);
1775 }
1776 read_extent_buffer(leaf, name_ptr,
1777 (unsigned long)(di + 1), name_len);
1778
1779 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
1780 btrfs_dir_item_key_to_cpu(leaf, di, &location);
1781 over = filldir(dirent, name_ptr, name_len,
1782 found_key.offset,
1783 location.objectid,
1784 d_type);
1785
1786 if (name_ptr != tmp_name)
1787 kfree(name_ptr);
1788
1789 if (over)
1790 goto nopos;
1791 di_len = btrfs_dir_name_len(leaf, di) +
1792 btrfs_dir_data_len(leaf, di) +sizeof(*di);
1793 di_cur += di_len;
1794 di = (struct btrfs_dir_item *)((char *)di + di_len);
1795 }
1796 }
1797 if (key_type == BTRFS_DIR_INDEX_KEY)
1798 filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
1799 else
1800 filp->f_pos++;
1801 nopos:
1802 ret = 0;
1803 err:
1804 btrfs_free_path(path);
1805 return ret;
1806 }
1807
1808 int btrfs_write_inode(struct inode *inode, int wait)
1809 {
1810 struct btrfs_root *root = BTRFS_I(inode)->root;
1811 struct btrfs_trans_handle *trans;
1812 int ret = 0;
1813
1814 if (wait) {
1815 trans = btrfs_join_transaction(root, 1);
1816 btrfs_set_trans_block_group(trans, inode);
1817 ret = btrfs_commit_transaction(trans, root);
1818 }
1819 return ret;
1820 }
1821
1822 /*
1823 * This is somewhat expensive, updating the tree every time the
1824 * inode changes. But, it is most likely to find the inode in cache.
1825 * FIXME, needs more benchmarking...there are no reasons other than performance
1826 * to keep or drop this code.
1827 */
1828 void btrfs_dirty_inode(struct inode *inode)
1829 {
1830 struct btrfs_root *root = BTRFS_I(inode)->root;
1831 struct btrfs_trans_handle *trans;
1832
1833 trans = btrfs_join_transaction(root, 1);
1834 btrfs_set_trans_block_group(trans, inode);
1835 btrfs_update_inode(trans, root, inode);
1836 btrfs_end_transaction(trans, root);
1837 }
1838
1839 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
1840 struct btrfs_root *root,
1841 const char *name, int name_len,
1842 u64 ref_objectid,
1843 u64 objectid,
1844 struct btrfs_block_group_cache *group,
1845 int mode)
1846 {
1847 struct inode *inode;
1848 struct btrfs_inode_item *inode_item;
1849 struct btrfs_block_group_cache *new_inode_group;
1850 struct btrfs_key *location;
1851 struct btrfs_path *path;
1852 struct btrfs_inode_ref *ref;
1853 struct btrfs_key key[2];
1854 u32 sizes[2];
1855 unsigned long ptr;
1856 int ret;
1857 int owner;
1858
1859 path = btrfs_alloc_path();
1860 BUG_ON(!path);
1861
1862 inode = new_inode(root->fs_info->sb);
1863 if (!inode)
1864 return ERR_PTR(-ENOMEM);
1865
1866 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1867 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1868 inode->i_mapping, GFP_NOFS);
1869 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
1870 inode->i_mapping, GFP_NOFS);
1871 mutex_init(&BTRFS_I(inode)->csum_mutex);
1872 BTRFS_I(inode)->delalloc_bytes = 0;
1873 BTRFS_I(inode)->disk_i_size = 0;
1874 BTRFS_I(inode)->root = root;
1875
1876 if (mode & S_IFDIR)
1877 owner = 0;
1878 else
1879 owner = 1;
1880 new_inode_group = btrfs_find_block_group(root, group, 0,
1881 BTRFS_BLOCK_GROUP_METADATA, owner);
1882 if (!new_inode_group) {
1883 printk("find_block group failed\n");
1884 new_inode_group = group;
1885 }
1886 BTRFS_I(inode)->block_group = new_inode_group;
1887 BTRFS_I(inode)->flags = 0;
1888
1889 key[0].objectid = objectid;
1890 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
1891 key[0].offset = 0;
1892
1893 key[1].objectid = objectid;
1894 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
1895 key[1].offset = ref_objectid;
1896
1897 sizes[0] = sizeof(struct btrfs_inode_item);
1898 sizes[1] = name_len + sizeof(*ref);
1899
1900 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
1901 if (ret != 0)
1902 goto fail;
1903
1904 if (objectid > root->highest_inode)
1905 root->highest_inode = objectid;
1906
1907 inode->i_uid = current->fsuid;
1908 inode->i_gid = current->fsgid;
1909 inode->i_mode = mode;
1910 inode->i_ino = objectid;
1911 inode->i_blocks = 0;
1912 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1913 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1914 struct btrfs_inode_item);
1915 fill_inode_item(path->nodes[0], inode_item, inode);
1916
1917 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
1918 struct btrfs_inode_ref);
1919 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
1920 ptr = (unsigned long)(ref + 1);
1921 write_extent_buffer(path->nodes[0], name, ptr, name_len);
1922
1923 btrfs_mark_buffer_dirty(path->nodes[0]);
1924 btrfs_free_path(path);
1925
1926 location = &BTRFS_I(inode)->location;
1927 location->objectid = objectid;
1928 location->offset = 0;
1929 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1930
1931 insert_inode_hash(inode);
1932 return inode;
1933 fail:
1934 btrfs_free_path(path);
1935 return ERR_PTR(ret);
1936 }
1937
1938 static inline u8 btrfs_inode_type(struct inode *inode)
1939 {
1940 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
1941 }
1942
1943 static int btrfs_add_link(struct btrfs_trans_handle *trans,
1944 struct dentry *dentry, struct inode *inode,
1945 int add_backref)
1946 {
1947 int ret;
1948 struct btrfs_key key;
1949 struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
1950 struct inode *parent_inode;
1951
1952 key.objectid = inode->i_ino;
1953 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1954 key.offset = 0;
1955
1956 ret = btrfs_insert_dir_item(trans, root,
1957 dentry->d_name.name, dentry->d_name.len,
1958 dentry->d_parent->d_inode->i_ino,
1959 &key, btrfs_inode_type(inode));
1960 if (ret == 0) {
1961 if (add_backref) {
1962 ret = btrfs_insert_inode_ref(trans, root,
1963 dentry->d_name.name,
1964 dentry->d_name.len,
1965 inode->i_ino,
1966 dentry->d_parent->d_inode->i_ino);
1967 }
1968 parent_inode = dentry->d_parent->d_inode;
1969 btrfs_i_size_write(parent_inode, parent_inode->i_size +
1970 dentry->d_name.len * 2);
1971 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1972 ret = btrfs_update_inode(trans, root,
1973 dentry->d_parent->d_inode);
1974 }
1975 return ret;
1976 }
1977
1978 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
1979 struct dentry *dentry, struct inode *inode,
1980 int backref)
1981 {
1982 int err = btrfs_add_link(trans, dentry, inode, backref);
1983 if (!err) {
1984 d_instantiate(dentry, inode);
1985 return 0;
1986 }
1987 if (err > 0)
1988 err = -EEXIST;
1989 return err;
1990 }
1991
1992 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1993 int mode, dev_t rdev)
1994 {
1995 struct btrfs_trans_handle *trans;
1996 struct btrfs_root *root = BTRFS_I(dir)->root;
1997 struct inode *inode = NULL;
1998 int err;
1999 int drop_inode = 0;
2000 u64 objectid;
2001 unsigned long nr = 0;
2002
2003 if (!new_valid_dev(rdev))
2004 return -EINVAL;
2005
2006 err = btrfs_check_free_space(root, 1, 0);
2007 if (err)
2008 goto fail;
2009
2010 trans = btrfs_start_transaction(root, 1);
2011 btrfs_set_trans_block_group(trans, dir);
2012
2013 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2014 if (err) {
2015 err = -ENOSPC;
2016 goto out_unlock;
2017 }
2018
2019 inode = btrfs_new_inode(trans, root, dentry->d_name.name,
2020 dentry->d_name.len,
2021 dentry->d_parent->d_inode->i_ino, objectid,
2022 BTRFS_I(dir)->block_group, mode);
2023 err = PTR_ERR(inode);
2024 if (IS_ERR(inode))
2025 goto out_unlock;
2026
2027 btrfs_set_trans_block_group(trans, inode);
2028 err = btrfs_add_nondir(trans, dentry, inode, 0);
2029 if (err)
2030 drop_inode = 1;
2031 else {
2032 inode->i_op = &btrfs_special_inode_operations;
2033 init_special_inode(inode, inode->i_mode, rdev);
2034 btrfs_update_inode(trans, root, inode);
2035 }
2036 dir->i_sb->s_dirt = 1;
2037 btrfs_update_inode_block_group(trans, inode);
2038 btrfs_update_inode_block_group(trans, dir);
2039 out_unlock:
2040 nr = trans->blocks_used;
2041 btrfs_end_transaction_throttle(trans, root);
2042 fail:
2043 if (drop_inode) {
2044 inode_dec_link_count(inode);
2045 iput(inode);
2046 }
2047 btrfs_btree_balance_dirty(root, nr);
2048 return err;
2049 }
2050
2051 static int btrfs_create(struct inode *dir, struct dentry *dentry,
2052 int mode, struct nameidata *nd)
2053 {
2054 struct btrfs_trans_handle *trans;
2055 struct btrfs_root *root = BTRFS_I(dir)->root;
2056 struct inode *inode = NULL;
2057 int err;
2058 int drop_inode = 0;
2059 unsigned long nr = 0;
2060 u64 objectid;
2061
2062 err = btrfs_check_free_space(root, 1, 0);
2063 if (err)
2064 goto fail;
2065 trans = btrfs_start_transaction(root, 1);
2066 btrfs_set_trans_block_group(trans, dir);
2067
2068 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2069 if (err) {
2070 err = -ENOSPC;
2071 goto out_unlock;
2072 }
2073
2074 inode = btrfs_new_inode(trans, root, dentry->d_name.name,
2075 dentry->d_name.len,
2076 dentry->d_parent->d_inode->i_ino,
2077 objectid, BTRFS_I(dir)->block_group, mode);
2078 err = PTR_ERR(inode);
2079 if (IS_ERR(inode))
2080 goto out_unlock;
2081
2082 btrfs_set_trans_block_group(trans, inode);
2083 err = btrfs_add_nondir(trans, dentry, inode, 0);
2084 if (err)
2085 drop_inode = 1;
2086 else {
2087 inode->i_mapping->a_ops = &btrfs_aops;
2088 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2089 inode->i_fop = &btrfs_file_operations;
2090 inode->i_op = &btrfs_file_inode_operations;
2091 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2092 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2093 inode->i_mapping, GFP_NOFS);
2094 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2095 inode->i_mapping, GFP_NOFS);
2096 mutex_init(&BTRFS_I(inode)->csum_mutex);
2097 BTRFS_I(inode)->delalloc_bytes = 0;
2098 BTRFS_I(inode)->disk_i_size = 0;
2099 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2100 }
2101 dir->i_sb->s_dirt = 1;
2102 btrfs_update_inode_block_group(trans, inode);
2103 btrfs_update_inode_block_group(trans, dir);
2104 out_unlock:
2105 nr = trans->blocks_used;
2106 btrfs_end_transaction_throttle(trans, root);
2107 fail:
2108 if (drop_inode) {
2109 inode_dec_link_count(inode);
2110 iput(inode);
2111 }
2112 btrfs_btree_balance_dirty(root, nr);
2113 return err;
2114 }
2115
2116 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
2117 struct dentry *dentry)
2118 {
2119 struct btrfs_trans_handle *trans;
2120 struct btrfs_root *root = BTRFS_I(dir)->root;
2121 struct inode *inode = old_dentry->d_inode;
2122 unsigned long nr = 0;
2123 int err;
2124 int drop_inode = 0;
2125
2126 if (inode->i_nlink == 0)
2127 return -ENOENT;
2128
2129 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
2130 inode->i_nlink++;
2131 #else
2132 inc_nlink(inode);
2133 #endif
2134 err = btrfs_check_free_space(root, 1, 0);
2135 if (err)
2136 goto fail;
2137 trans = btrfs_start_transaction(root, 1);
2138
2139 btrfs_set_trans_block_group(trans, dir);
2140 atomic_inc(&inode->i_count);
2141 err = btrfs_add_nondir(trans, dentry, inode, 1);
2142
2143 if (err)
2144 drop_inode = 1;
2145
2146 dir->i_sb->s_dirt = 1;
2147 btrfs_update_inode_block_group(trans, dir);
2148 err = btrfs_update_inode(trans, root, inode);
2149
2150 if (err)
2151 drop_inode = 1;
2152
2153 nr = trans->blocks_used;
2154 btrfs_end_transaction_throttle(trans, root);
2155 fail:
2156 if (drop_inode) {
2157 inode_dec_link_count(inode);
2158 iput(inode);
2159 }
2160 btrfs_btree_balance_dirty(root, nr);
2161 return err;
2162 }
2163
2164 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2165 {
2166 struct inode *inode = NULL;
2167 struct btrfs_trans_handle *trans;
2168 struct btrfs_root *root = BTRFS_I(dir)->root;
2169 int err = 0;
2170 int drop_on_err = 0;
2171 u64 objectid = 0;
2172 unsigned long nr = 1;
2173
2174 err = btrfs_check_free_space(root, 1, 0);
2175 if (err)
2176 goto out_unlock;
2177
2178 trans = btrfs_start_transaction(root, 1);
2179 btrfs_set_trans_block_group(trans, dir);
2180
2181 if (IS_ERR(trans)) {
2182 err = PTR_ERR(trans);
2183 goto out_unlock;
2184 }
2185
2186 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2187 if (err) {
2188 err = -ENOSPC;
2189 goto out_unlock;
2190 }
2191
2192 inode = btrfs_new_inode(trans, root, dentry->d_name.name,
2193 dentry->d_name.len,
2194 dentry->d_parent->d_inode->i_ino, objectid,
2195 BTRFS_I(dir)->block_group, S_IFDIR | mode);
2196 if (IS_ERR(inode)) {
2197 err = PTR_ERR(inode);
2198 goto out_fail;
2199 }
2200
2201 drop_on_err = 1;
2202 inode->i_op = &btrfs_dir_inode_operations;
2203 inode->i_fop = &btrfs_dir_file_operations;
2204 btrfs_set_trans_block_group(trans, inode);
2205
2206 btrfs_i_size_write(inode, 0);
2207 err = btrfs_update_inode(trans, root, inode);
2208 if (err)
2209 goto out_fail;
2210
2211 err = btrfs_add_link(trans, dentry, inode, 0);
2212 if (err)
2213 goto out_fail;
2214
2215 d_instantiate(dentry, inode);
2216 drop_on_err = 0;
2217 dir->i_sb->s_dirt = 1;
2218 btrfs_update_inode_block_group(trans, inode);
2219 btrfs_update_inode_block_group(trans, dir);
2220
2221 out_fail:
2222 nr = trans->blocks_used;
2223 btrfs_end_transaction_throttle(trans, root);
2224
2225 out_unlock:
2226 if (drop_on_err)
2227 iput(inode);
2228 btrfs_btree_balance_dirty(root, nr);
2229 return err;
2230 }
2231
2232 static int merge_extent_mapping(struct extent_map_tree *em_tree,
2233 struct extent_map *existing,
2234 struct extent_map *em,
2235 u64 map_start, u64 map_len)
2236 {
2237 u64 start_diff;
2238
2239 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
2240 start_diff = map_start - em->start;
2241 em->start = map_start;
2242 em->len = map_len;
2243 if (em->block_start < EXTENT_MAP_LAST_BYTE)
2244 em->block_start += start_diff;
2245 return add_extent_mapping(em_tree, em);
2246 }
2247
2248 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
2249 size_t pg_offset, u64 start, u64 len,
2250 int create)
2251 {
2252 int ret;
2253 int err = 0;
2254 u64 bytenr;
2255 u64 extent_start = 0;
2256 u64 extent_end = 0;
2257 u64 objectid = inode->i_ino;
2258 u32 found_type;
2259 struct btrfs_path *path;
2260 struct btrfs_root *root = BTRFS_I(inode)->root;
2261 struct btrfs_file_extent_item *item;
2262 struct extent_buffer *leaf;
2263 struct btrfs_key found_key;
2264 struct extent_map *em = NULL;
2265 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2266 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2267 struct btrfs_trans_handle *trans = NULL;
2268
2269 path = btrfs_alloc_path();
2270 BUG_ON(!path);
2271
2272 again:
2273 spin_lock(&em_tree->lock);
2274 em = lookup_extent_mapping(em_tree, start, len);
2275 if (em)
2276 em->bdev = root->fs_info->fs_devices->latest_bdev;
2277 spin_unlock(&em_tree->lock);
2278
2279 if (em) {
2280 if (em->start > start || em->start + em->len <= start)
2281 free_extent_map(em);
2282 else if (em->block_start == EXTENT_MAP_INLINE && page)
2283 free_extent_map(em);
2284 else
2285 goto out;
2286 }
2287 em = alloc_extent_map(GFP_NOFS);
2288 if (!em) {
2289 err = -ENOMEM;
2290 goto out;
2291 }
2292 em->bdev = root->fs_info->fs_devices->latest_bdev;
2293 em->start = EXTENT_MAP_HOLE;
2294 em->len = (u64)-1;
2295 ret = btrfs_lookup_file_extent(trans, root, path,
2296 objectid, start, trans != NULL);
2297 if (ret < 0) {
2298 err = ret;
2299 goto out;
2300 }
2301
2302 if (ret != 0) {
2303 if (path->slots[0] == 0)
2304 goto not_found;
2305 path->slots[0]--;
2306 }
2307
2308 leaf = path->nodes[0];
2309 item = btrfs_item_ptr(leaf, path->slots[0],
2310 struct btrfs_file_extent_item);
2311 /* are we inside the extent that was found? */
2312 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2313 found_type = btrfs_key_type(&found_key);
2314 if (found_key.objectid != objectid ||
2315 found_type != BTRFS_EXTENT_DATA_KEY) {
2316 goto not_found;
2317 }
2318
2319 found_type = btrfs_file_extent_type(leaf, item);
2320 extent_start = found_key.offset;
2321 if (found_type == BTRFS_FILE_EXTENT_REG) {
2322 extent_end = extent_start +
2323 btrfs_file_extent_num_bytes(leaf, item);
2324 err = 0;
2325 if (start < extent_start || start >= extent_end) {
2326 em->start = start;
2327 if (start < extent_start) {
2328 if (start + len <= extent_start)
2329 goto not_found;
2330 em->len = extent_end - extent_start;
2331 } else {
2332 em->len = len;
2333 }
2334 goto not_found_em;
2335 }
2336 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
2337 if (bytenr == 0) {
2338 em->start = extent_start;
2339 em->len = extent_end - extent_start;
2340 em->block_start = EXTENT_MAP_HOLE;
2341 goto insert;
2342 }
2343 bytenr += btrfs_file_extent_offset(leaf, item);
2344 em->block_start = bytenr;
2345 em->start = extent_start;
2346 em->len = extent_end - extent_start;
2347 goto insert;
2348 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
2349 u64 page_start;
2350 unsigned long ptr;
2351 char *map;
2352 size_t size;
2353 size_t extent_offset;
2354 size_t copy_size;
2355
2356 size = btrfs_file_extent_inline_len(leaf, btrfs_item_nr(leaf,
2357 path->slots[0]));
2358 extent_end = (extent_start + size + root->sectorsize - 1) &
2359 ~((u64)root->sectorsize - 1);
2360 if (start < extent_start || start >= extent_end) {
2361 em->start = start;
2362 if (start < extent_start) {
2363 if (start + len <= extent_start)
2364 goto not_found;
2365 em->len = extent_end - extent_start;
2366 } else {
2367 em->len = len;
2368 }
2369 goto not_found_em;
2370 }
2371 em->block_start = EXTENT_MAP_INLINE;
2372
2373 if (!page) {
2374 em->start = extent_start;
2375 em->len = size;
2376 goto out;
2377 }
2378
2379 page_start = page_offset(page) + pg_offset;
2380 extent_offset = page_start - extent_start;
2381 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
2382 size - extent_offset);
2383 em->start = extent_start + extent_offset;
2384 em->len = (copy_size + root->sectorsize - 1) &
2385 ~((u64)root->sectorsize - 1);
2386 map = kmap(page);
2387 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
2388 if (create == 0 && !PageUptodate(page)) {
2389 read_extent_buffer(leaf, map + pg_offset, ptr,
2390 copy_size);
2391 flush_dcache_page(page);
2392 } else if (create && PageUptodate(page)) {
2393 if (!trans) {
2394 kunmap(page);
2395 free_extent_map(em);
2396 em = NULL;
2397 btrfs_release_path(root, path);
2398 trans = btrfs_join_transaction(root, 1);
2399 goto again;
2400 }
2401 write_extent_buffer(leaf, map + pg_offset, ptr,
2402 copy_size);
2403 btrfs_mark_buffer_dirty(leaf);
2404 }
2405 kunmap(page);
2406 set_extent_uptodate(io_tree, em->start,
2407 extent_map_end(em) - 1, GFP_NOFS);
2408 goto insert;
2409 } else {
2410 printk("unkknown found_type %d\n", found_type);
2411 WARN_ON(1);
2412 }
2413 not_found:
2414 em->start = start;
2415 em->len = len;
2416 not_found_em:
2417 em->block_start = EXTENT_MAP_HOLE;
2418 insert:
2419 btrfs_release_path(root, path);
2420 if (em->start > start || extent_map_end(em) <= start) {
2421 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
2422 err = -EIO;
2423 goto out;
2424 }
2425
2426 err = 0;
2427 spin_lock(&em_tree->lock);
2428 ret = add_extent_mapping(em_tree, em);
2429 /* it is possible that someone inserted the extent into the tree
2430 * while we had the lock dropped. It is also possible that
2431 * an overlapping map exists in the tree
2432 */
2433 if (ret == -EEXIST) {
2434 struct extent_map *existing;
2435
2436 ret = 0;
2437
2438 existing = lookup_extent_mapping(em_tree, start, len);
2439 if (existing && (existing->start > start ||
2440 existing->start + existing->len <= start)) {
2441 free_extent_map(existing);
2442 existing = NULL;
2443 }
2444 if (!existing) {
2445 existing = lookup_extent_mapping(em_tree, em->start,
2446 em->len);
2447 if (existing) {
2448 err = merge_extent_mapping(em_tree, existing,
2449 em, start,
2450 root->sectorsize);
2451 free_extent_map(existing);
2452 if (err) {
2453 free_extent_map(em);
2454 em = NULL;
2455 }
2456 } else {
2457 err = -EIO;
2458 printk("failing to insert %Lu %Lu\n",
2459 start, len);
2460 free_extent_map(em);
2461 em = NULL;
2462 }
2463 } else {
2464 free_extent_map(em);
2465 em = existing;
2466 err = 0;
2467 }
2468 }
2469 spin_unlock(&em_tree->lock);
2470 out:
2471 btrfs_free_path(path);
2472 if (trans) {
2473 ret = btrfs_end_transaction(trans, root);
2474 if (!err) {
2475 err = ret;
2476 }
2477 }
2478 if (err) {
2479 free_extent_map(em);
2480 WARN_ON(1);
2481 return ERR_PTR(err);
2482 }
2483 return em;
2484 }
2485
2486 #if 0 /* waiting for O_DIRECT reads */
2487 static int btrfs_get_block(struct inode *inode, sector_t iblock,
2488 struct buffer_head *bh_result, int create)
2489 {
2490 struct extent_map *em;
2491 u64 start = (u64)iblock << inode->i_blkbits;
2492 struct btrfs_multi_bio *multi = NULL;
2493 struct btrfs_root *root = BTRFS_I(inode)->root;
2494 u64 len;
2495 u64 logical;
2496 u64 map_length;
2497 int ret = 0;
2498
2499 em = btrfs_get_extent(inode, NULL, 0, start, bh_result->b_size, 0);
2500
2501 if (!em || IS_ERR(em))
2502 goto out;
2503
2504 if (em->start > start || em->start + em->len <= start) {
2505 goto out;
2506 }
2507
2508 if (em->block_start == EXTENT_MAP_INLINE) {
2509 ret = -EINVAL;
2510 goto out;
2511 }
2512
2513 len = em->start + em->len - start;
2514 len = min_t(u64, len, INT_LIMIT(typeof(bh_result->b_size)));
2515
2516 if (em->block_start == EXTENT_MAP_HOLE ||
2517 em->block_start == EXTENT_MAP_DELALLOC) {
2518 bh_result->b_size = len;
2519 goto out;
2520 }
2521
2522 logical = start - em->start;
2523 logical = em->block_start + logical;
2524
2525 map_length = len;
2526 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
2527 logical, &map_length, &multi, 0);
2528 BUG_ON(ret);
2529 bh_result->b_blocknr = multi->stripes[0].physical >> inode->i_blkbits;
2530 bh_result->b_size = min(map_length, len);
2531
2532 bh_result->b_bdev = multi->stripes[0].dev->bdev;
2533 set_buffer_mapped(bh_result);
2534 kfree(multi);
2535 out:
2536 free_extent_map(em);
2537 return ret;
2538 }
2539 #endif
2540
2541 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
2542 const struct iovec *iov, loff_t offset,
2543 unsigned long nr_segs)
2544 {
2545 return -EINVAL;
2546 #if 0
2547 struct file *file = iocb->ki_filp;
2548 struct inode *inode = file->f_mapping->host;
2549
2550 if (rw == WRITE)
2551 return -EINVAL;
2552
2553 return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
2554 offset, nr_segs, btrfs_get_block, NULL);
2555 #endif
2556 }
2557
2558 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
2559 {
2560 return extent_bmap(mapping, iblock, btrfs_get_extent);
2561 }
2562
2563 int btrfs_readpage(struct file *file, struct page *page)
2564 {
2565 struct extent_io_tree *tree;
2566 tree = &BTRFS_I(page->mapping->host)->io_tree;
2567 return extent_read_full_page(tree, page, btrfs_get_extent);
2568 }
2569
2570 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
2571 {
2572 struct extent_io_tree *tree;
2573
2574
2575 if (current->flags & PF_MEMALLOC) {
2576 redirty_page_for_writepage(wbc, page);
2577 unlock_page(page);
2578 return 0;
2579 }
2580 tree = &BTRFS_I(page->mapping->host)->io_tree;
2581 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
2582 }
2583
2584 static int btrfs_writepages(struct address_space *mapping,
2585 struct writeback_control *wbc)
2586 {
2587 struct extent_io_tree *tree;
2588 tree = &BTRFS_I(mapping->host)->io_tree;
2589 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
2590 }
2591
2592 static int
2593 btrfs_readpages(struct file *file, struct address_space *mapping,
2594 struct list_head *pages, unsigned nr_pages)
2595 {
2596 struct extent_io_tree *tree;
2597 tree = &BTRFS_I(mapping->host)->io_tree;
2598 return extent_readpages(tree, mapping, pages, nr_pages,
2599 btrfs_get_extent);
2600 }
2601 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
2602 {
2603 struct extent_io_tree *tree;
2604 struct extent_map_tree *map;
2605 int ret;
2606
2607 tree = &BTRFS_I(page->mapping->host)->io_tree;
2608 map = &BTRFS_I(page->mapping->host)->extent_tree;
2609 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
2610 if (ret == 1) {
2611 invalidate_extent_lru(tree, page_offset(page), PAGE_CACHE_SIZE);
2612 ClearPagePrivate(page);
2613 set_page_private(page, 0);
2614 page_cache_release(page);
2615 }
2616 return ret;
2617 }
2618
2619 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
2620 {
2621 struct btrfs_ordered_extent *ordered;
2622
2623 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
2624 page_offset(page));
2625 if (ordered) {
2626 btrfs_put_ordered_extent(ordered);
2627 return 0;
2628 }
2629 return __btrfs_releasepage(page, gfp_flags);
2630 }
2631
2632 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
2633 {
2634 struct extent_io_tree *tree;
2635 struct btrfs_ordered_extent *ordered;
2636 u64 page_start = page_offset(page);
2637 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
2638
2639 wait_on_page_writeback(page);
2640 tree = &BTRFS_I(page->mapping->host)->io_tree;
2641 if (offset) {
2642 btrfs_releasepage(page, GFP_NOFS);
2643 return;
2644 }
2645
2646 lock_extent(tree, page_start, page_end, GFP_NOFS);
2647 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
2648 page_offset(page));
2649 if (ordered) {
2650 clear_extent_bit(tree, page_start, page_end,
2651 EXTENT_DIRTY | EXTENT_DELALLOC |
2652 EXTENT_LOCKED, 1, 0, GFP_NOFS);
2653 btrfs_writepage_end_io_hook(page, page_start,
2654 page_end, NULL, 1);
2655 btrfs_put_ordered_extent(ordered);
2656 lock_extent(tree, page_start, page_end, GFP_NOFS);
2657 }
2658 clear_extent_bit(tree, page_start, page_end,
2659 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
2660 EXTENT_ORDERED,
2661 1, 1, GFP_NOFS);
2662 __btrfs_releasepage(page, GFP_NOFS);
2663
2664 if (PagePrivate(page)) {
2665 invalidate_extent_lru(tree, page_offset(page),
2666 PAGE_CACHE_SIZE);
2667 ClearPagePrivate(page);
2668 set_page_private(page, 0);
2669 page_cache_release(page);
2670 }
2671 }
2672
2673 /*
2674 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
2675 * called from a page fault handler when a page is first dirtied. Hence we must
2676 * be careful to check for EOF conditions here. We set the page up correctly
2677 * for a written page which means we get ENOSPC checking when writing into
2678 * holes and correct delalloc and unwritten extent mapping on filesystems that
2679 * support these features.
2680 *
2681 * We are not allowed to take the i_mutex here so we have to play games to
2682 * protect against truncate races as the page could now be beyond EOF. Because
2683 * vmtruncate() writes the inode size before removing pages, once we have the
2684 * page lock we can determine safely if the page is beyond EOF. If it is not
2685 * beyond EOF, then the page is guaranteed safe against truncation until we
2686 * unlock the page.
2687 */
2688 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
2689 {
2690 struct inode *inode = fdentry(vma->vm_file)->d_inode;
2691 struct btrfs_root *root = BTRFS_I(inode)->root;
2692 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2693 struct btrfs_ordered_extent *ordered;
2694 char *kaddr;
2695 unsigned long zero_start;
2696 loff_t size;
2697 int ret;
2698 u64 page_start;
2699 u64 page_end;
2700
2701 ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
2702 if (ret)
2703 goto out;
2704
2705 ret = -EINVAL;
2706 again:
2707 lock_page(page);
2708 size = i_size_read(inode);
2709 page_start = page_offset(page);
2710 page_end = page_start + PAGE_CACHE_SIZE - 1;
2711
2712 if ((page->mapping != inode->i_mapping) ||
2713 (page_start >= size)) {
2714 /* page got truncated out from underneath us */
2715 goto out_unlock;
2716 }
2717 wait_on_page_writeback(page);
2718
2719 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2720 set_page_extent_mapped(page);
2721
2722 ordered = btrfs_lookup_ordered_extent(inode, page_start);
2723 if (ordered) {
2724 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2725 unlock_page(page);
2726 btrfs_wait_ordered_extent(inode, ordered);
2727 btrfs_put_ordered_extent(ordered);
2728 goto again;
2729 }
2730
2731 set_extent_delalloc(&BTRFS_I(inode)->io_tree, page_start,
2732 page_end, GFP_NOFS);
2733 ret = 0;
2734
2735 /* page is wholly or partially inside EOF */
2736 if (page_start + PAGE_CACHE_SIZE > size)
2737 zero_start = size & ~PAGE_CACHE_MASK;
2738 else
2739 zero_start = PAGE_CACHE_SIZE;
2740
2741 if (zero_start != PAGE_CACHE_SIZE) {
2742 kaddr = kmap(page);
2743 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
2744 flush_dcache_page(page);
2745 kunmap(page);
2746 }
2747 ClearPageChecked(page);
2748 set_page_dirty(page);
2749 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2750
2751 out_unlock:
2752 unlock_page(page);
2753 out:
2754 return ret;
2755 }
2756
2757 static void btrfs_truncate(struct inode *inode)
2758 {
2759 struct btrfs_root *root = BTRFS_I(inode)->root;
2760 int ret;
2761 struct btrfs_trans_handle *trans;
2762 unsigned long nr;
2763 u64 mask = root->sectorsize - 1;
2764
2765 if (!S_ISREG(inode->i_mode))
2766 return;
2767 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2768 return;
2769
2770 btrfs_truncate_page(inode->i_mapping, inode->i_size);
2771
2772 trans = btrfs_start_transaction(root, 1);
2773 btrfs_set_trans_block_group(trans, inode);
2774 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
2775 btrfs_i_size_write(inode, inode->i_size);
2776
2777 /* FIXME, add redo link to tree so we don't leak on crash */
2778 ret = btrfs_truncate_in_trans(trans, root, inode,
2779 BTRFS_EXTENT_DATA_KEY);
2780 btrfs_update_inode(trans, root, inode);
2781 nr = trans->blocks_used;
2782
2783 ret = btrfs_end_transaction_throttle(trans, root);
2784 BUG_ON(ret);
2785 btrfs_btree_balance_dirty(root, nr);
2786 }
2787
2788 /*
2789 * Invalidate a single dcache entry at the root of the filesystem.
2790 * Needed after creation of snapshot or subvolume.
2791 */
2792 void btrfs_invalidate_dcache_root(struct btrfs_root *root, char *name,
2793 int namelen)
2794 {
2795 struct dentry *alias, *entry;
2796 struct qstr qstr;
2797
2798 alias = d_find_alias(root->fs_info->sb->s_root->d_inode);
2799 if (alias) {
2800 qstr.name = name;
2801 qstr.len = namelen;
2802 /* change me if btrfs ever gets a d_hash operation */
2803 qstr.hash = full_name_hash(qstr.name, qstr.len);
2804 entry = d_lookup(alias, &qstr);
2805 dput(alias);
2806 if (entry) {
2807 d_invalidate(entry);
2808 dput(entry);
2809 }
2810 }
2811 }
2812
2813 int btrfs_create_subvol_root(struct btrfs_root *new_root,
2814 struct btrfs_trans_handle *trans, u64 new_dirid,
2815 struct btrfs_block_group_cache *block_group)
2816 {
2817 struct inode *inode;
2818 int ret;
2819
2820 inode = btrfs_new_inode(trans, new_root, "..", 2, new_dirid,
2821 new_dirid, block_group, S_IFDIR | 0700);
2822 if (IS_ERR(inode))
2823 return PTR_ERR(inode);
2824 inode->i_op = &btrfs_dir_inode_operations;
2825 inode->i_fop = &btrfs_dir_file_operations;
2826 new_root->inode = inode;
2827
2828 ret = btrfs_insert_inode_ref(trans, new_root, "..", 2, new_dirid,
2829 new_dirid);
2830 inode->i_nlink = 1;
2831 btrfs_i_size_write(inode, 0);
2832
2833 return btrfs_update_inode(trans, new_root, inode);
2834 }
2835
2836 unsigned long btrfs_force_ra(struct address_space *mapping,
2837 struct file_ra_state *ra, struct file *file,
2838 pgoff_t offset, pgoff_t last_index)
2839 {
2840 pgoff_t req_size = last_index - offset + 1;
2841
2842 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2843 offset = page_cache_readahead(mapping, ra, file, offset, req_size);
2844 return offset;
2845 #else
2846 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
2847 return offset + req_size;
2848 #endif
2849 }
2850
2851 struct inode *btrfs_alloc_inode(struct super_block *sb)
2852 {
2853 struct btrfs_inode *ei;
2854
2855 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
2856 if (!ei)
2857 return NULL;
2858 ei->last_trans = 0;
2859 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
2860 return &ei->vfs_inode;
2861 }
2862
2863 void btrfs_destroy_inode(struct inode *inode)
2864 {
2865 struct btrfs_ordered_extent *ordered;
2866 WARN_ON(!list_empty(&inode->i_dentry));
2867 WARN_ON(inode->i_data.nrpages);
2868
2869 while(1) {
2870 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
2871 if (!ordered)
2872 break;
2873 else {
2874 printk("found ordered extent %Lu %Lu\n",
2875 ordered->file_offset, ordered->len);
2876 btrfs_remove_ordered_extent(inode, ordered);
2877 btrfs_put_ordered_extent(ordered);
2878 btrfs_put_ordered_extent(ordered);
2879 }
2880 }
2881 btrfs_drop_extent_cache(inode, 0, (u64)-1);
2882 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
2883 }
2884
2885 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2886 static void init_once(struct kmem_cache * cachep, void *foo)
2887 #else
2888 static void init_once(void * foo, struct kmem_cache * cachep,
2889 unsigned long flags)
2890 #endif
2891 {
2892 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
2893
2894 inode_init_once(&ei->vfs_inode);
2895 }
2896
2897 void btrfs_destroy_cachep(void)
2898 {
2899 if (btrfs_inode_cachep)
2900 kmem_cache_destroy(btrfs_inode_cachep);
2901 if (btrfs_trans_handle_cachep)
2902 kmem_cache_destroy(btrfs_trans_handle_cachep);
2903 if (btrfs_transaction_cachep)
2904 kmem_cache_destroy(btrfs_transaction_cachep);
2905 if (btrfs_bit_radix_cachep)
2906 kmem_cache_destroy(btrfs_bit_radix_cachep);
2907 if (btrfs_path_cachep)
2908 kmem_cache_destroy(btrfs_path_cachep);
2909 }
2910
2911 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
2912 unsigned long extra_flags,
2913 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2914 void (*ctor)(struct kmem_cache *, void *)
2915 #else
2916 void (*ctor)(void *, struct kmem_cache *,
2917 unsigned long)
2918 #endif
2919 )
2920 {
2921 return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
2922 SLAB_MEM_SPREAD | extra_flags), ctor
2923 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2924 ,NULL
2925 #endif
2926 );
2927 }
2928
2929 int btrfs_init_cachep(void)
2930 {
2931 btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
2932 sizeof(struct btrfs_inode),
2933 0, init_once);
2934 if (!btrfs_inode_cachep)
2935 goto fail;
2936 btrfs_trans_handle_cachep =
2937 btrfs_cache_create("btrfs_trans_handle_cache",
2938 sizeof(struct btrfs_trans_handle),
2939 0, NULL);
2940 if (!btrfs_trans_handle_cachep)
2941 goto fail;
2942 btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
2943 sizeof(struct btrfs_transaction),
2944 0, NULL);
2945 if (!btrfs_transaction_cachep)
2946 goto fail;
2947 btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
2948 sizeof(struct btrfs_path),
2949 0, NULL);
2950 if (!btrfs_path_cachep)
2951 goto fail;
2952 btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
2953 SLAB_DESTROY_BY_RCU, NULL);
2954 if (!btrfs_bit_radix_cachep)
2955 goto fail;
2956 return 0;
2957 fail:
2958 btrfs_destroy_cachep();
2959 return -ENOMEM;
2960 }
2961
2962 static int btrfs_getattr(struct vfsmount *mnt,
2963 struct dentry *dentry, struct kstat *stat)
2964 {
2965 struct inode *inode = dentry->d_inode;
2966 generic_fillattr(inode, stat);
2967 stat->blksize = PAGE_CACHE_SIZE;
2968 stat->blocks = inode->i_blocks + (BTRFS_I(inode)->delalloc_bytes >> 9);
2969 return 0;
2970 }
2971
2972 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
2973 struct inode * new_dir,struct dentry *new_dentry)
2974 {
2975 struct btrfs_trans_handle *trans;
2976 struct btrfs_root *root = BTRFS_I(old_dir)->root;
2977 struct inode *new_inode = new_dentry->d_inode;
2978 struct inode *old_inode = old_dentry->d_inode;
2979 struct timespec ctime = CURRENT_TIME;
2980 int ret;
2981
2982 if (S_ISDIR(old_inode->i_mode) && new_inode &&
2983 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
2984 return -ENOTEMPTY;
2985 }
2986
2987 ret = btrfs_check_free_space(root, 1, 0);
2988 if (ret)
2989 goto out_unlock;
2990
2991 trans = btrfs_start_transaction(root, 1);
2992
2993 btrfs_set_trans_block_group(trans, new_dir);
2994
2995 old_dentry->d_inode->i_nlink++;
2996 old_dir->i_ctime = old_dir->i_mtime = ctime;
2997 new_dir->i_ctime = new_dir->i_mtime = ctime;
2998 old_inode->i_ctime = ctime;
2999
3000 ret = btrfs_unlink_trans(trans, root, old_dir, old_dentry);
3001 if (ret)
3002 goto out_fail;
3003
3004 if (new_inode) {
3005 new_inode->i_ctime = CURRENT_TIME;
3006 ret = btrfs_unlink_trans(trans, root, new_dir, new_dentry);
3007 if (ret)
3008 goto out_fail;
3009 }
3010 ret = btrfs_add_link(trans, new_dentry, old_inode, 1);
3011 if (ret)
3012 goto out_fail;
3013
3014 out_fail:
3015 btrfs_end_transaction(trans, root);
3016 out_unlock:
3017 return ret;
3018 }
3019
3020 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
3021 const char *symname)
3022 {
3023 struct btrfs_trans_handle *trans;
3024 struct btrfs_root *root = BTRFS_I(dir)->root;
3025 struct btrfs_path *path;
3026 struct btrfs_key key;
3027 struct inode *inode = NULL;
3028 int err;
3029 int drop_inode = 0;
3030 u64 objectid;
3031 int name_len;
3032 int datasize;
3033 unsigned long ptr;
3034 struct btrfs_file_extent_item *ei;
3035 struct extent_buffer *leaf;
3036 unsigned long nr = 0;
3037
3038 name_len = strlen(symname) + 1;
3039 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
3040 return -ENAMETOOLONG;
3041
3042 err = btrfs_check_free_space(root, 1, 0);
3043 if (err)
3044 goto out_fail;
3045
3046 trans = btrfs_start_transaction(root, 1);
3047 btrfs_set_trans_block_group(trans, dir);
3048
3049 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3050 if (err) {
3051 err = -ENOSPC;
3052 goto out_unlock;
3053 }
3054
3055 inode = btrfs_new_inode(trans, root, dentry->d_name.name,
3056 dentry->d_name.len,
3057 dentry->d_parent->d_inode->i_ino, objectid,
3058 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO);
3059 err = PTR_ERR(inode);
3060 if (IS_ERR(inode))
3061 goto out_unlock;
3062
3063 btrfs_set_trans_block_group(trans, inode);
3064 err = btrfs_add_nondir(trans, dentry, inode, 0);
3065 if (err)
3066 drop_inode = 1;
3067 else {
3068 inode->i_mapping->a_ops = &btrfs_aops;
3069 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3070 inode->i_fop = &btrfs_file_operations;
3071 inode->i_op = &btrfs_file_inode_operations;
3072 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3073 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
3074 inode->i_mapping, GFP_NOFS);
3075 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3076 inode->i_mapping, GFP_NOFS);
3077 mutex_init(&BTRFS_I(inode)->csum_mutex);
3078 BTRFS_I(inode)->delalloc_bytes = 0;
3079 BTRFS_I(inode)->disk_i_size = 0;
3080 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3081 }
3082 dir->i_sb->s_dirt = 1;
3083 btrfs_update_inode_block_group(trans, inode);
3084 btrfs_update_inode_block_group(trans, dir);
3085 if (drop_inode)
3086 goto out_unlock;
3087
3088 path = btrfs_alloc_path();
3089 BUG_ON(!path);
3090 key.objectid = inode->i_ino;
3091 key.offset = 0;
3092 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
3093 datasize = btrfs_file_extent_calc_inline_size(name_len);
3094 err = btrfs_insert_empty_item(trans, root, path, &key,
3095 datasize);
3096 if (err) {
3097 drop_inode = 1;
3098 goto out_unlock;
3099 }
3100 leaf = path->nodes[0];
3101 ei = btrfs_item_ptr(leaf, path->slots[0],
3102 struct btrfs_file_extent_item);
3103 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
3104 btrfs_set_file_extent_type(leaf, ei,
3105 BTRFS_FILE_EXTENT_INLINE);
3106 ptr = btrfs_file_extent_inline_start(ei);
3107 write_extent_buffer(leaf, symname, ptr, name_len);
3108 btrfs_mark_buffer_dirty(leaf);
3109 btrfs_free_path(path);
3110
3111 inode->i_op = &btrfs_symlink_inode_operations;
3112 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3113 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3114 btrfs_i_size_write(inode, name_len - 1);
3115 err = btrfs_update_inode(trans, root, inode);
3116 if (err)
3117 drop_inode = 1;
3118
3119 out_unlock:
3120 nr = trans->blocks_used;
3121 btrfs_end_transaction_throttle(trans, root);
3122 out_fail:
3123 if (drop_inode) {
3124 inode_dec_link_count(inode);
3125 iput(inode);
3126 }
3127 btrfs_btree_balance_dirty(root, nr);
3128 return err;
3129 }
3130
3131 static int btrfs_set_page_dirty(struct page *page)
3132 {
3133 return __set_page_dirty_nobuffers(page);
3134 }
3135
3136 static int btrfs_permission(struct inode *inode, int mask,
3137 struct nameidata *nd)
3138 {
3139 if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
3140 return -EACCES;
3141 return generic_permission(inode, mask, NULL);
3142 }
3143
3144 static struct inode_operations btrfs_dir_inode_operations = {
3145 .lookup = btrfs_lookup,
3146 .create = btrfs_create,
3147 .unlink = btrfs_unlink,
3148 .link = btrfs_link,
3149 .mkdir = btrfs_mkdir,
3150 .rmdir = btrfs_rmdir,
3151 .rename = btrfs_rename,
3152 .symlink = btrfs_symlink,
3153 .setattr = btrfs_setattr,
3154 .mknod = btrfs_mknod,
3155 .setxattr = generic_setxattr,
3156 .getxattr = generic_getxattr,
3157 .listxattr = btrfs_listxattr,
3158 .removexattr = generic_removexattr,
3159 .permission = btrfs_permission,
3160 };
3161 static struct inode_operations btrfs_dir_ro_inode_operations = {
3162 .lookup = btrfs_lookup,
3163 .permission = btrfs_permission,
3164 };
3165 static struct file_operations btrfs_dir_file_operations = {
3166 .llseek = generic_file_llseek,
3167 .read = generic_read_dir,
3168 .readdir = btrfs_readdir,
3169 .unlocked_ioctl = btrfs_ioctl,
3170 #ifdef CONFIG_COMPAT
3171 .compat_ioctl = btrfs_ioctl,
3172 #endif
3173 .release = btrfs_release_file,
3174 };
3175
3176 static struct extent_io_ops btrfs_extent_io_ops = {
3177 .fill_delalloc = run_delalloc_range,
3178 .submit_bio_hook = btrfs_submit_bio_hook,
3179 .merge_bio_hook = btrfs_merge_bio_hook,
3180 .readpage_io_hook = btrfs_readpage_io_hook,
3181 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
3182 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
3183 .writepage_start_hook = btrfs_writepage_start_hook,
3184 .readpage_io_failed_hook = btrfs_io_failed_hook,
3185 .set_bit_hook = btrfs_set_bit_hook,
3186 .clear_bit_hook = btrfs_clear_bit_hook,
3187 };
3188
3189 static struct address_space_operations btrfs_aops = {
3190 .readpage = btrfs_readpage,
3191 .writepage = btrfs_writepage,
3192 .writepages = btrfs_writepages,
3193 .readpages = btrfs_readpages,
3194 .sync_page = block_sync_page,
3195 .bmap = btrfs_bmap,
3196 .direct_IO = btrfs_direct_IO,
3197 .invalidatepage = btrfs_invalidatepage,
3198 .releasepage = btrfs_releasepage,
3199 .set_page_dirty = btrfs_set_page_dirty,
3200 };
3201
3202 static struct address_space_operations btrfs_symlink_aops = {
3203 .readpage = btrfs_readpage,
3204 .writepage = btrfs_writepage,
3205 .invalidatepage = btrfs_invalidatepage,
3206 .releasepage = btrfs_releasepage,
3207 };
3208
3209 static struct inode_operations btrfs_file_inode_operations = {
3210 .truncate = btrfs_truncate,
3211 .getattr = btrfs_getattr,
3212 .setattr = btrfs_setattr,
3213 .setxattr = generic_setxattr,
3214 .getxattr = generic_getxattr,
3215 .listxattr = btrfs_listxattr,
3216 .removexattr = generic_removexattr,
3217 .permission = btrfs_permission,
3218 };
3219 static struct inode_operations btrfs_special_inode_operations = {
3220 .getattr = btrfs_getattr,
3221 .setattr = btrfs_setattr,
3222 .permission = btrfs_permission,
3223 };
3224 static struct inode_operations btrfs_symlink_inode_operations = {
3225 .readlink = generic_readlink,
3226 .follow_link = page_follow_link_light,
3227 .put_link = page_put_link,
3228 .permission = btrfs_permission,
3229 };
Linux kernel - Deliverables
This page took 0.104755 seconds and 6 git commands to generate.