?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Btrfs: Optimise NFS readdir hack slightly; don't call readdir() again when done
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / btrfs / inode.c
blob10f26f445328b40e442792e4347975bb3f2c0498
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 <linux/posix_acl.h>
40 #include "ctree.h"
41 #include "disk-io.h"
42 #include "transaction.h"
43 #include "btrfs_inode.h"
44 #include "ioctl.h"
45 #include "print-tree.h"
46 #include "volumes.h"
47 #include "ordered-data.h"
48
49 struct btrfs_iget_args {
50 u64 ino;
51 struct btrfs_root *root;
52 };
53
54 static struct inode_operations btrfs_dir_inode_operations;
55 static struct inode_operations btrfs_symlink_inode_operations;
56 static struct inode_operations btrfs_dir_ro_inode_operations;
57 static struct inode_operations btrfs_special_inode_operations;
58 static struct inode_operations btrfs_file_inode_operations;
59 static struct address_space_operations btrfs_aops;
60 static struct address_space_operations btrfs_symlink_aops;
61 static struct file_operations btrfs_dir_file_operations;
62 static struct extent_io_ops btrfs_extent_io_ops;
63
64 static struct kmem_cache *btrfs_inode_cachep;
65 struct kmem_cache *btrfs_trans_handle_cachep;
66 struct kmem_cache *btrfs_transaction_cachep;
67 struct kmem_cache *btrfs_bit_radix_cachep;
68 struct kmem_cache *btrfs_path_cachep;
69
70 #define S_SHIFT 12
71 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
72 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
73 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
74 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
75 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
76 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
77 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
78 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
79 };
80
81 static void btrfs_truncate(struct inode *inode);
82
83 int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
84 int for_del)
85 {
86 u64 total;
87 u64 used;
88 u64 thresh;
89 unsigned long flags;
90 int ret = 0;
91
92 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
93 total = btrfs_super_total_bytes(&root->fs_info->super_copy);
94 used = btrfs_super_bytes_used(&root->fs_info->super_copy);
95 if (for_del)
96 thresh = total * 90;
97 else
98 thresh = total * 85;
99
100 do_div(thresh, 100);
101
102 if (used + root->fs_info->delalloc_bytes + num_required > thresh)
103 ret = -ENOSPC;
104 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
105 return ret;
106 }
107
108 static int cow_file_range(struct inode *inode, u64 start, u64 end)
109 {
110 struct btrfs_root *root = BTRFS_I(inode)->root;
111 struct btrfs_trans_handle *trans;
112 u64 alloc_hint = 0;
113 u64 num_bytes;
114 u64 cur_alloc_size;
115 u64 blocksize = root->sectorsize;
116 u64 orig_num_bytes;
117 struct btrfs_key ins;
118 struct extent_map *em;
119 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
120 int ret = 0;
121
122 trans = btrfs_join_transaction(root, 1);
123 BUG_ON(!trans);
124 btrfs_set_trans_block_group(trans, inode);
125
126 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
127 num_bytes = max(blocksize, num_bytes);
128 orig_num_bytes = num_bytes;
129
130 if (alloc_hint == EXTENT_MAP_INLINE)
131 goto out;
132
133 BUG_ON(num_bytes > btrfs_super_total_bytes(&root->fs_info->super_copy));
134 mutex_lock(&BTRFS_I(inode)->extent_mutex);
135 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1);
136 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
137
138 while(num_bytes > 0) {
139 cur_alloc_size = min(num_bytes, root->fs_info->max_extent);
140 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
141 root->sectorsize, 0, 0,
142 (u64)-1, &ins, 1);
143 if (ret) {
144 WARN_ON(1);
145 goto out;
146 }
147 em = alloc_extent_map(GFP_NOFS);
148 em->start = start;
149 em->len = ins.offset;
150 em->block_start = ins.objectid;
151 em->bdev = root->fs_info->fs_devices->latest_bdev;
152 mutex_lock(&BTRFS_I(inode)->extent_mutex);
153 set_bit(EXTENT_FLAG_PINNED, &em->flags);
154 while(1) {
155 spin_lock(&em_tree->lock);
156 ret = add_extent_mapping(em_tree, em);
157 spin_unlock(&em_tree->lock);
158 if (ret != -EEXIST) {
159 free_extent_map(em);
160 break;
161 }
162 btrfs_drop_extent_cache(inode, start,
163 start + ins.offset - 1);
164 }
165 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
166
167 cur_alloc_size = ins.offset;
168 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
169 ins.offset, 0);
170 BUG_ON(ret);
171 if (num_bytes < cur_alloc_size) {
172 printk("num_bytes %Lu cur_alloc %Lu\n", num_bytes,
173 cur_alloc_size);
174 break;
175 }
176 num_bytes -= cur_alloc_size;
177 alloc_hint = ins.objectid + ins.offset;
178 start += cur_alloc_size;
179 }
180 out:
181 btrfs_end_transaction(trans, root);
182 return ret;
183 }
184
185 static int run_delalloc_nocow(struct inode *inode, u64 start, u64 end)
186 {
187 u64 extent_start;
188 u64 extent_end;
189 u64 bytenr;
190 u64 loops = 0;
191 u64 total_fs_bytes;
192 struct btrfs_root *root = BTRFS_I(inode)->root;
193 struct btrfs_block_group_cache *block_group;
194 struct btrfs_trans_handle *trans;
195 struct extent_buffer *leaf;
196 int found_type;
197 struct btrfs_path *path;
198 struct btrfs_file_extent_item *item;
199 int ret;
200 int err = 0;
201 struct btrfs_key found_key;
202
203 total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
204 path = btrfs_alloc_path();
205 BUG_ON(!path);
206 trans = btrfs_join_transaction(root, 1);
207 BUG_ON(!trans);
208 again:
209 ret = btrfs_lookup_file_extent(NULL, root, path,
210 inode->i_ino, start, 0);
211 if (ret < 0) {
212 err = ret;
213 goto out;
214 }
215
216 if (ret != 0) {
217 if (path->slots[0] == 0)
218 goto not_found;
219 path->slots[0]--;
220 }
221
222 leaf = path->nodes[0];
223 item = btrfs_item_ptr(leaf, path->slots[0],
224 struct btrfs_file_extent_item);
225
226 /* are we inside the extent that was found? */
227 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
228 found_type = btrfs_key_type(&found_key);
229 if (found_key.objectid != inode->i_ino ||
230 found_type != BTRFS_EXTENT_DATA_KEY)
231 goto not_found;
232
233 found_type = btrfs_file_extent_type(leaf, item);
234 extent_start = found_key.offset;
235 if (found_type == BTRFS_FILE_EXTENT_REG) {
236 u64 extent_num_bytes;
237
238 extent_num_bytes = btrfs_file_extent_num_bytes(leaf, item);
239 extent_end = extent_start + extent_num_bytes;
240 err = 0;
241
242 if (loops && start != extent_start)
243 goto not_found;
244
245 if (start < extent_start || start >= extent_end)
246 goto not_found;
247
248 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
249 if (bytenr == 0)
250 goto not_found;
251
252 if (btrfs_cross_ref_exists(trans, root, &found_key, bytenr))
253 goto not_found;
254 /*
255 * we may be called by the resizer, make sure we're inside
256 * the limits of the FS
257 */
258 block_group = btrfs_lookup_block_group(root->fs_info,
259 bytenr);
260 if (!block_group || block_group->ro)
261 goto not_found;
262
263 bytenr += btrfs_file_extent_offset(leaf, item);
264 extent_num_bytes = min(end + 1, extent_end) - start;
265 ret = btrfs_add_ordered_extent(inode, start, bytenr,
266 extent_num_bytes, 1);
267 if (ret) {
268 err = ret;
269 goto out;
270 }
271
272 btrfs_release_path(root, path);
273 start = extent_end;
274 if (start <= end) {
275 loops++;
276 goto again;
277 }
278 } else {
279 not_found:
280 btrfs_end_transaction(trans, root);
281 btrfs_free_path(path);
282 return cow_file_range(inode, start, end);
283 }
284 out:
285 WARN_ON(err);
286 btrfs_end_transaction(trans, root);
287 btrfs_free_path(path);
288 return err;
289 }
290
291 static int run_delalloc_range(struct inode *inode, u64 start, u64 end)
292 {
293 struct btrfs_root *root = BTRFS_I(inode)->root;
294 int ret;
295
296 if (btrfs_test_opt(root, NODATACOW) ||
297 btrfs_test_flag(inode, NODATACOW))
298 ret = run_delalloc_nocow(inode, start, end);
299 else
300 ret = cow_file_range(inode, start, end);
301
302 return ret;
303 }
304
305 int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
306 unsigned long old, unsigned long bits)
307 {
308 unsigned long flags;
309 if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
310 struct btrfs_root *root = BTRFS_I(inode)->root;
311 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
312 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
313 root->fs_info->delalloc_bytes += end - start + 1;
314 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
315 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
316 &root->fs_info->delalloc_inodes);
317 }
318 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
319 }
320 return 0;
321 }
322
323 int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
324 unsigned long old, unsigned long bits)
325 {
326 if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
327 struct btrfs_root *root = BTRFS_I(inode)->root;
328 unsigned long flags;
329
330 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
331 if (end - start + 1 > root->fs_info->delalloc_bytes) {
332 printk("warning: delalloc account %Lu %Lu\n",
333 end - start + 1, root->fs_info->delalloc_bytes);
334 root->fs_info->delalloc_bytes = 0;
335 BTRFS_I(inode)->delalloc_bytes = 0;
336 } else {
337 root->fs_info->delalloc_bytes -= end - start + 1;
338 BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
339 }
340 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
341 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
342 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
343 }
344 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
345 }
346 return 0;
347 }
348
349 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
350 size_t size, struct bio *bio)
351 {
352 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
353 struct btrfs_mapping_tree *map_tree;
354 u64 logical = bio->bi_sector << 9;
355 u64 length = 0;
356 u64 map_length;
357 int ret;
358
359 length = bio->bi_size;
360 map_tree = &root->fs_info->mapping_tree;
361 map_length = length;
362 ret = btrfs_map_block(map_tree, READ, logical,
363 &map_length, NULL, 0);
364
365 if (map_length < length + size) {
366 return 1;
367 }
368 return 0;
369 }
370
371 int __btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
372 int mirror_num)
373 {
374 struct btrfs_root *root = BTRFS_I(inode)->root;
375 int ret = 0;
376
377 ret = btrfs_csum_one_bio(root, inode, bio);
378 BUG_ON(ret);
379
380 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
381 }
382
383 int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
384 int mirror_num)
385 {
386 struct btrfs_root *root = BTRFS_I(inode)->root;
387 int ret = 0;
388
389 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
390 BUG_ON(ret);
391
392 if (btrfs_test_opt(root, NODATASUM) ||
393 btrfs_test_flag(inode, NODATASUM)) {
394 goto mapit;
395 }
396
397 if (!(rw & (1 << BIO_RW))) {
398 btrfs_lookup_bio_sums(root, inode, bio);
399 goto mapit;
400 }
401 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
402 inode, rw, bio, mirror_num,
403 __btrfs_submit_bio_hook);
404 mapit:
405 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
406 }
407
408 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
409 struct inode *inode, u64 file_offset,
410 struct list_head *list)
411 {
412 struct list_head *cur;
413 struct btrfs_ordered_sum *sum;
414
415 btrfs_set_trans_block_group(trans, inode);
416 list_for_each(cur, list) {
417 sum = list_entry(cur, struct btrfs_ordered_sum, list);
418 btrfs_csum_file_blocks(trans, BTRFS_I(inode)->root,
419 inode, sum);
420 }
421 return 0;
422 }
423
424 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
425 {
426 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
427 GFP_NOFS);
428 }
429
430 struct btrfs_writepage_fixup {
431 struct page *page;
432 struct btrfs_work work;
433 };
434
435 /* see btrfs_writepage_start_hook for details on why this is required */
436 void btrfs_writepage_fixup_worker(struct btrfs_work *work)
437 {
438 struct btrfs_writepage_fixup *fixup;
439 struct btrfs_ordered_extent *ordered;
440 struct page *page;
441 struct inode *inode;
442 u64 page_start;
443 u64 page_end;
444
445 fixup = container_of(work, struct btrfs_writepage_fixup, work);
446 page = fixup->page;
447 again:
448 lock_page(page);
449 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
450 ClearPageChecked(page);
451 goto out_page;
452 }
453
454 inode = page->mapping->host;
455 page_start = page_offset(page);
456 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
457
458 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
459
460 /* already ordered? We're done */
461 if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
462 EXTENT_ORDERED, 0)) {
463 goto out;
464 }
465
466 ordered = btrfs_lookup_ordered_extent(inode, page_start);
467 if (ordered) {
468 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
469 page_end, GFP_NOFS);
470 unlock_page(page);
471 btrfs_start_ordered_extent(inode, ordered, 1);
472 goto again;
473 }
474
475 btrfs_set_extent_delalloc(inode, page_start, page_end);
476 ClearPageChecked(page);
477 out:
478 unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
479 out_page:
480 unlock_page(page);
481 page_cache_release(page);
482 }
483
484 /*
485 * There are a few paths in the higher layers of the kernel that directly
486 * set the page dirty bit without asking the filesystem if it is a
487 * good idea. This causes problems because we want to make sure COW
488 * properly happens and the data=ordered rules are followed.
489 *
490 * In our case any range that doesn't have the EXTENT_ORDERED bit set
491 * hasn't been properly setup for IO. We kick off an async process
492 * to fix it up. The async helper will wait for ordered extents, set
493 * the delalloc bit and make it safe to write the page.
494 */
495 int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
496 {
497 struct inode *inode = page->mapping->host;
498 struct btrfs_writepage_fixup *fixup;
499 struct btrfs_root *root = BTRFS_I(inode)->root;
500 int ret;
501
502 ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
503 EXTENT_ORDERED, 0);
504 if (ret)
505 return 0;
506
507 if (PageChecked(page))
508 return -EAGAIN;
509
510 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
511 if (!fixup)
512 return -EAGAIN;
513
514 SetPageChecked(page);
515 page_cache_get(page);
516 fixup->work.func = btrfs_writepage_fixup_worker;
517 fixup->page = page;
518 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
519 return -EAGAIN;
520 }
521
522 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
523 {
524 struct btrfs_root *root = BTRFS_I(inode)->root;
525 struct btrfs_trans_handle *trans;
526 struct btrfs_ordered_extent *ordered_extent;
527 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
528 u64 alloc_hint = 0;
529 struct list_head list;
530 struct btrfs_key ins;
531 int ret;
532
533 ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
534 if (!ret)
535 return 0;
536
537 trans = btrfs_join_transaction(root, 1);
538
539 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
540 BUG_ON(!ordered_extent);
541 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
542 goto nocow;
543
544 lock_extent(io_tree, ordered_extent->file_offset,
545 ordered_extent->file_offset + ordered_extent->len - 1,
546 GFP_NOFS);
547
548 INIT_LIST_HEAD(&list);
549
550 ins.objectid = ordered_extent->start;
551 ins.offset = ordered_extent->len;
552 ins.type = BTRFS_EXTENT_ITEM_KEY;
553
554 ret = btrfs_alloc_reserved_extent(trans, root, root->root_key.objectid,
555 trans->transid, inode->i_ino,
556 ordered_extent->file_offset, &ins);
557 BUG_ON(ret);
558
559 mutex_lock(&BTRFS_I(inode)->extent_mutex);
560
561 ret = btrfs_drop_extents(trans, root, inode,
562 ordered_extent->file_offset,
563 ordered_extent->file_offset +
564 ordered_extent->len,
565 ordered_extent->file_offset, &alloc_hint);
566 BUG_ON(ret);
567 ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
568 ordered_extent->file_offset,
569 ordered_extent->start,
570 ordered_extent->len,
571 ordered_extent->len, 0);
572 BUG_ON(ret);
573
574 btrfs_drop_extent_cache(inode, ordered_extent->file_offset,
575 ordered_extent->file_offset +
576 ordered_extent->len - 1);
577 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
578
579 inode->i_blocks += ordered_extent->len >> 9;
580 unlock_extent(io_tree, ordered_extent->file_offset,
581 ordered_extent->file_offset + ordered_extent->len - 1,
582 GFP_NOFS);
583 nocow:
584 add_pending_csums(trans, inode, ordered_extent->file_offset,
585 &ordered_extent->list);
586
587 btrfs_ordered_update_i_size(inode, ordered_extent);
588 btrfs_remove_ordered_extent(inode, ordered_extent);
589
590 /* once for us */
591 btrfs_put_ordered_extent(ordered_extent);
592 /* once for the tree */
593 btrfs_put_ordered_extent(ordered_extent);
594
595 btrfs_update_inode(trans, root, inode);
596 btrfs_end_transaction(trans, root);
597 return 0;
598 }
599
600 int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
601 struct extent_state *state, int uptodate)
602 {
603 return btrfs_finish_ordered_io(page->mapping->host, start, end);
604 }
605
606 struct io_failure_record {
607 struct page *page;
608 u64 start;
609 u64 len;
610 u64 logical;
611 int last_mirror;
612 };
613
614 int btrfs_io_failed_hook(struct bio *failed_bio,
615 struct page *page, u64 start, u64 end,
616 struct extent_state *state)
617 {
618 struct io_failure_record *failrec = NULL;
619 u64 private;
620 struct extent_map *em;
621 struct inode *inode = page->mapping->host;
622 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
623 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
624 struct bio *bio;
625 int num_copies;
626 int ret;
627 int rw;
628 u64 logical;
629
630 ret = get_state_private(failure_tree, start, &private);
631 if (ret) {
632 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
633 if (!failrec)
634 return -ENOMEM;
635 failrec->start = start;
636 failrec->len = end - start + 1;
637 failrec->last_mirror = 0;
638
639 spin_lock(&em_tree->lock);
640 em = lookup_extent_mapping(em_tree, start, failrec->len);
641 if (em->start > start || em->start + em->len < start) {
642 free_extent_map(em);
643 em = NULL;
644 }
645 spin_unlock(&em_tree->lock);
646
647 if (!em || IS_ERR(em)) {
648 kfree(failrec);
649 return -EIO;
650 }
651 logical = start - em->start;
652 logical = em->block_start + logical;
653 failrec->logical = logical;
654 free_extent_map(em);
655 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
656 EXTENT_DIRTY, GFP_NOFS);
657 set_state_private(failure_tree, start,
658 (u64)(unsigned long)failrec);
659 } else {
660 failrec = (struct io_failure_record *)(unsigned long)private;
661 }
662 num_copies = btrfs_num_copies(
663 &BTRFS_I(inode)->root->fs_info->mapping_tree,
664 failrec->logical, failrec->len);
665 failrec->last_mirror++;
666 if (!state) {
667 spin_lock_irq(&BTRFS_I(inode)->io_tree.lock);
668 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
669 failrec->start,
670 EXTENT_LOCKED);
671 if (state && state->start != failrec->start)
672 state = NULL;
673 spin_unlock_irq(&BTRFS_I(inode)->io_tree.lock);
674 }
675 if (!state || failrec->last_mirror > num_copies) {
676 set_state_private(failure_tree, failrec->start, 0);
677 clear_extent_bits(failure_tree, failrec->start,
678 failrec->start + failrec->len - 1,
679 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
680 kfree(failrec);
681 return -EIO;
682 }
683 bio = bio_alloc(GFP_NOFS, 1);
684 bio->bi_private = state;
685 bio->bi_end_io = failed_bio->bi_end_io;
686 bio->bi_sector = failrec->logical >> 9;
687 bio->bi_bdev = failed_bio->bi_bdev;
688 bio->bi_size = 0;
689 bio_add_page(bio, page, failrec->len, start - page_offset(page));
690 if (failed_bio->bi_rw & (1 << BIO_RW))
691 rw = WRITE;
692 else
693 rw = READ;
694
695 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
696 failrec->last_mirror);
697 return 0;
698 }
699
700 int btrfs_clean_io_failures(struct inode *inode, u64 start)
701 {
702 u64 private;
703 u64 private_failure;
704 struct io_failure_record *failure;
705 int ret;
706
707 private = 0;
708 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
709 (u64)-1, 1, EXTENT_DIRTY)) {
710 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
711 start, &private_failure);
712 if (ret == 0) {
713 failure = (struct io_failure_record *)(unsigned long)
714 private_failure;
715 set_state_private(&BTRFS_I(inode)->io_failure_tree,
716 failure->start, 0);
717 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
718 failure->start,
719 failure->start + failure->len - 1,
720 EXTENT_DIRTY | EXTENT_LOCKED,
721 GFP_NOFS);
722 kfree(failure);
723 }
724 }
725 return 0;
726 }
727
728 int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
729 struct extent_state *state)
730 {
731 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
732 struct inode *inode = page->mapping->host;
733 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
734 char *kaddr;
735 u64 private = ~(u32)0;
736 int ret;
737 struct btrfs_root *root = BTRFS_I(inode)->root;
738 u32 csum = ~(u32)0;
739 unsigned long flags;
740
741 if (btrfs_test_opt(root, NODATASUM) ||
742 btrfs_test_flag(inode, NODATASUM))
743 return 0;
744 if (state && state->start == start) {
745 private = state->private;
746 ret = 0;
747 } else {
748 ret = get_state_private(io_tree, start, &private);
749 }
750 local_irq_save(flags);
751 kaddr = kmap_atomic(page, KM_IRQ0);
752 if (ret) {
753 goto zeroit;
754 }
755 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
756 btrfs_csum_final(csum, (char *)&csum);
757 if (csum != private) {
758 goto zeroit;
759 }
760 kunmap_atomic(kaddr, KM_IRQ0);
761 local_irq_restore(flags);
762
763 /* if the io failure tree for this inode is non-empty,
764 * check to see if we've recovered from a failed IO
765 */
766 btrfs_clean_io_failures(inode, start);
767 return 0;
768
769 zeroit:
770 printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
771 page->mapping->host->i_ino, (unsigned long long)start, csum,
772 private);
773 memset(kaddr + offset, 1, end - start + 1);
774 flush_dcache_page(page);
775 kunmap_atomic(kaddr, KM_IRQ0);
776 local_irq_restore(flags);
777 if (private == 0)
778 return 0;
779 return -EIO;
780 }
781
782 /*
783 * This creates an orphan entry for the given inode in case something goes
784 * wrong in the middle of an unlink/truncate.
785 */
786 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
787 {
788 struct btrfs_root *root = BTRFS_I(inode)->root;
789 int ret = 0;
790
791 spin_lock(&root->list_lock);
792
793 /* already on the orphan list, we're good */
794 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
795 spin_unlock(&root->list_lock);
796 return 0;
797 }
798
799 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
800
801 spin_unlock(&root->list_lock);
802
803 /*
804 * insert an orphan item to track this unlinked/truncated file
805 */
806 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
807
808 return ret;
809 }
810
811 /*
812 * We have done the truncate/delete so we can go ahead and remove the orphan
813 * item for this particular inode.
814 */
815 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
816 {
817 struct btrfs_root *root = BTRFS_I(inode)->root;
818 int ret = 0;
819
820 spin_lock(&root->list_lock);
821
822 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
823 spin_unlock(&root->list_lock);
824 return 0;
825 }
826
827 list_del_init(&BTRFS_I(inode)->i_orphan);
828 if (!trans) {
829 spin_unlock(&root->list_lock);
830 return 0;
831 }
832
833 spin_unlock(&root->list_lock);
834
835 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
836
837 return ret;
838 }
839
840 /*
841 * this cleans up any orphans that may be left on the list from the last use
842 * of this root.
843 */
844 void btrfs_orphan_cleanup(struct btrfs_root *root)
845 {
846 struct btrfs_path *path;
847 struct extent_buffer *leaf;
848 struct btrfs_item *item;
849 struct btrfs_key key, found_key;
850 struct btrfs_trans_handle *trans;
851 struct inode *inode;
852 int ret = 0, nr_unlink = 0, nr_truncate = 0;
853
854 /* don't do orphan cleanup if the fs is readonly. */
855 if (root->inode->i_sb->s_flags & MS_RDONLY)
856 return;
857
858 path = btrfs_alloc_path();
859 if (!path)
860 return;
861 path->reada = -1;
862
863 key.objectid = BTRFS_ORPHAN_OBJECTID;
864 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
865 key.offset = (u64)-1;
866
867 trans = btrfs_start_transaction(root, 1);
868 btrfs_set_trans_block_group(trans, root->inode);
869
870 while (1) {
871 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
872 if (ret < 0) {
873 printk(KERN_ERR "Error searching slot for orphan: %d"
874 "\n", ret);
875 break;
876 }
877
878 /*
879 * if ret == 0 means we found what we were searching for, which
880 * is weird, but possible, so only screw with path if we didnt
881 * find the key and see if we have stuff that matches
882 */
883 if (ret > 0) {
884 if (path->slots[0] == 0)
885 break;
886 path->slots[0]--;
887 }
888
889 /* pull out the item */
890 leaf = path->nodes[0];
891 item = btrfs_item_nr(leaf, path->slots[0]);
892 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
893
894 /* make sure the item matches what we want */
895 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
896 break;
897 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
898 break;
899
900 /* release the path since we're done with it */
901 btrfs_release_path(root, path);
902
903 /*
904 * this is where we are basically btrfs_lookup, without the
905 * crossing root thing. we store the inode number in the
906 * offset of the orphan item.
907 */
908 inode = btrfs_iget_locked(root->inode->i_sb,
909 found_key.offset, root);
910 if (!inode)
911 break;
912
913 if (inode->i_state & I_NEW) {
914 BTRFS_I(inode)->root = root;
915
916 /* have to set the location manually */
917 BTRFS_I(inode)->location.objectid = inode->i_ino;
918 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
919 BTRFS_I(inode)->location.offset = 0;
920
921 btrfs_read_locked_inode(inode);
922 unlock_new_inode(inode);
923 }
924
925 /*
926 * add this inode to the orphan list so btrfs_orphan_del does
927 * the proper thing when we hit it
928 */
929 spin_lock(&root->list_lock);
930 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
931 spin_unlock(&root->list_lock);
932
933 /*
934 * if this is a bad inode, means we actually succeeded in
935 * removing the inode, but not the orphan record, which means
936 * we need to manually delete the orphan since iput will just
937 * do a destroy_inode
938 */
939 if (is_bad_inode(inode)) {
940 btrfs_orphan_del(trans, inode);
941 iput(inode);
942 continue;
943 }
944
945 /* if we have links, this was a truncate, lets do that */
946 if (inode->i_nlink) {
947 nr_truncate++;
948 btrfs_truncate(inode);
949 } else {
950 nr_unlink++;
951 }
952
953 /* this will do delete_inode and everything for us */
954 iput(inode);
955 }
956
957 if (nr_unlink)
958 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
959 if (nr_truncate)
960 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
961
962 btrfs_free_path(path);
963 btrfs_end_transaction(trans, root);
964 }
965
966 void btrfs_read_locked_inode(struct inode *inode)
967 {
968 struct btrfs_path *path;
969 struct extent_buffer *leaf;
970 struct btrfs_inode_item *inode_item;
971 struct btrfs_timespec *tspec;
972 struct btrfs_root *root = BTRFS_I(inode)->root;
973 struct btrfs_key location;
974 u64 alloc_group_block;
975 u32 rdev;
976 int ret;
977
978 path = btrfs_alloc_path();
979 BUG_ON(!path);
980 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
981
982 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
983 if (ret)
984 goto make_bad;
985
986 leaf = path->nodes[0];
987 inode_item = btrfs_item_ptr(leaf, path->slots[0],
988 struct btrfs_inode_item);
989
990 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
991 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
992 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
993 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
994 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
995
996 tspec = btrfs_inode_atime(inode_item);
997 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
998 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
999
1000 tspec = btrfs_inode_mtime(inode_item);
1001 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1002 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1003
1004 tspec = btrfs_inode_ctime(inode_item);
1005 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1006 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1007
1008 inode->i_blocks = btrfs_inode_nblocks(leaf, inode_item);
1009 inode->i_generation = btrfs_inode_generation(leaf, inode_item);
1010 inode->i_rdev = 0;
1011 rdev = btrfs_inode_rdev(leaf, inode_item);
1012
1013 BTRFS_I(inode)->index_cnt = (u64)-1;
1014
1015 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
1016 BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
1017 alloc_group_block);
1018 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
1019 if (!BTRFS_I(inode)->block_group) {
1020 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
1021 NULL, 0,
1022 BTRFS_BLOCK_GROUP_METADATA, 0);
1023 }
1024 btrfs_free_path(path);
1025 inode_item = NULL;
1026
1027 switch (inode->i_mode & S_IFMT) {
1028 case S_IFREG:
1029 inode->i_mapping->a_ops = &btrfs_aops;
1030 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
1031 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
1032 inode->i_fop = &btrfs_file_operations;
1033 inode->i_op = &btrfs_file_inode_operations;
1034 break;
1035 case S_IFDIR:
1036 inode->i_fop = &btrfs_dir_file_operations;
1037 if (root == root->fs_info->tree_root)
1038 inode->i_op = &btrfs_dir_ro_inode_operations;
1039 else
1040 inode->i_op = &btrfs_dir_inode_operations;
1041 break;
1042 case S_IFLNK:
1043 inode->i_op = &btrfs_symlink_inode_operations;
1044 inode->i_mapping->a_ops = &btrfs_symlink_aops;
1045 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
1046 break;
1047 default:
1048 init_special_inode(inode, inode->i_mode, rdev);
1049 break;
1050 }
1051 return;
1052
1053 make_bad:
1054 btrfs_free_path(path);
1055 make_bad_inode(inode);
1056 }
1057
1058 static void fill_inode_item(struct extent_buffer *leaf,
1059 struct btrfs_inode_item *item,
1060 struct inode *inode)
1061 {
1062 btrfs_set_inode_uid(leaf, item, inode->i_uid);
1063 btrfs_set_inode_gid(leaf, item, inode->i_gid);
1064 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
1065 btrfs_set_inode_mode(leaf, item, inode->i_mode);
1066 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
1067
1068 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
1069 inode->i_atime.tv_sec);
1070 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
1071 inode->i_atime.tv_nsec);
1072
1073 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
1074 inode->i_mtime.tv_sec);
1075 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
1076 inode->i_mtime.tv_nsec);
1077
1078 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
1079 inode->i_ctime.tv_sec);
1080 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
1081 inode->i_ctime.tv_nsec);
1082
1083 btrfs_set_inode_nblocks(leaf, item, inode->i_blocks);
1084 btrfs_set_inode_generation(leaf, item, inode->i_generation);
1085 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
1086 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
1087 btrfs_set_inode_block_group(leaf, item,
1088 BTRFS_I(inode)->block_group->key.objectid);
1089 }
1090
1091 int noinline btrfs_update_inode(struct btrfs_trans_handle *trans,
1092 struct btrfs_root *root,
1093 struct inode *inode)
1094 {
1095 struct btrfs_inode_item *inode_item;
1096 struct btrfs_path *path;
1097 struct extent_buffer *leaf;
1098 int ret;
1099
1100 path = btrfs_alloc_path();
1101 BUG_ON(!path);
1102 ret = btrfs_lookup_inode(trans, root, path,
1103 &BTRFS_I(inode)->location, 1);
1104 if (ret) {
1105 if (ret > 0)
1106 ret = -ENOENT;
1107 goto failed;
1108 }
1109
1110 leaf = path->nodes[0];
1111 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1112 struct btrfs_inode_item);
1113
1114 fill_inode_item(leaf, inode_item, inode);
1115 btrfs_mark_buffer_dirty(leaf);
1116 btrfs_set_inode_last_trans(trans, inode);
1117 ret = 0;
1118 failed:
1119 btrfs_free_path(path);
1120 return ret;
1121 }
1122
1123
1124 static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
1125 struct btrfs_root *root,
1126 struct inode *dir,
1127 struct dentry *dentry)
1128 {
1129 struct btrfs_path *path;
1130 const char *name = dentry->d_name.name;
1131 int name_len = dentry->d_name.len;
1132 int ret = 0;
1133 struct extent_buffer *leaf;
1134 struct btrfs_dir_item *di;
1135 struct btrfs_key key;
1136 u64 index;
1137
1138 path = btrfs_alloc_path();
1139 if (!path) {
1140 ret = -ENOMEM;
1141 goto err;
1142 }
1143
1144 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
1145 name, name_len, -1);
1146 if (IS_ERR(di)) {
1147 ret = PTR_ERR(di);
1148 goto err;
1149 }
1150 if (!di) {
1151 ret = -ENOENT;
1152 goto err;
1153 }
1154 leaf = path->nodes[0];
1155 btrfs_dir_item_key_to_cpu(leaf, di, &key);
1156 ret = btrfs_delete_one_dir_name(trans, root, path, di);
1157 if (ret)
1158 goto err;
1159 btrfs_release_path(root, path);
1160
1161 ret = btrfs_del_inode_ref(trans, root, name, name_len,
1162 dentry->d_inode->i_ino,
1163 dentry->d_parent->d_inode->i_ino, &index);
1164 if (ret) {
1165 printk("failed to delete reference to %.*s, "
1166 "inode %lu parent %lu\n", name_len, name,
1167 dentry->d_inode->i_ino,
1168 dentry->d_parent->d_inode->i_ino);
1169 goto err;
1170 }
1171
1172 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
1173 index, name, name_len, -1);
1174 if (IS_ERR(di)) {
1175 ret = PTR_ERR(di);
1176 goto err;
1177 }
1178 if (!di) {
1179 ret = -ENOENT;
1180 goto err;
1181 }
1182 ret = btrfs_delete_one_dir_name(trans, root, path, di);
1183 btrfs_release_path(root, path);
1184
1185 dentry->d_inode->i_ctime = dir->i_ctime;
1186 err:
1187 btrfs_free_path(path);
1188 if (!ret) {
1189 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
1190 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1191 btrfs_update_inode(trans, root, dir);
1192 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1193 dentry->d_inode->i_nlink--;
1194 #else
1195 drop_nlink(dentry->d_inode);
1196 #endif
1197 ret = btrfs_update_inode(trans, root, dentry->d_inode);
1198 dir->i_sb->s_dirt = 1;
1199 }
1200 return ret;
1201 }
1202
1203 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
1204 {
1205 struct btrfs_root *root;
1206 struct btrfs_trans_handle *trans;
1207 struct inode *inode = dentry->d_inode;
1208 int ret;
1209 unsigned long nr = 0;
1210
1211 root = BTRFS_I(dir)->root;
1212
1213 ret = btrfs_check_free_space(root, 1, 1);
1214 if (ret)
1215 goto fail;
1216
1217 trans = btrfs_start_transaction(root, 1);
1218
1219 btrfs_set_trans_block_group(trans, dir);
1220 ret = btrfs_unlink_trans(trans, root, dir, dentry);
1221
1222 if (inode->i_nlink == 0)
1223 ret = btrfs_orphan_add(trans, inode);
1224
1225 nr = trans->blocks_used;
1226
1227 btrfs_end_transaction_throttle(trans, root);
1228 fail:
1229 btrfs_btree_balance_dirty(root, nr);
1230 return ret;
1231 }
1232
1233 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
1234 {
1235 struct inode *inode = dentry->d_inode;
1236 int err = 0;
1237 int ret;
1238 struct btrfs_root *root = BTRFS_I(dir)->root;
1239 struct btrfs_trans_handle *trans;
1240 unsigned long nr = 0;
1241
1242 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
1243 return -ENOTEMPTY;
1244 }
1245
1246 ret = btrfs_check_free_space(root, 1, 1);
1247 if (ret)
1248 goto fail;
1249
1250 trans = btrfs_start_transaction(root, 1);
1251 btrfs_set_trans_block_group(trans, dir);
1252
1253 err = btrfs_orphan_add(trans, inode);
1254 if (err)
1255 goto fail_trans;
1256
1257 /* now the directory is empty */
1258 err = btrfs_unlink_trans(trans, root, dir, dentry);
1259 if (!err) {
1260 btrfs_i_size_write(inode, 0);
1261 }
1262
1263 fail_trans:
1264 nr = trans->blocks_used;
1265 ret = btrfs_end_transaction_throttle(trans, root);
1266 fail:
1267 btrfs_btree_balance_dirty(root, nr);
1268
1269 if (ret && !err)
1270 err = ret;
1271 return err;
1272 }
1273
1274 /*
1275 * this can truncate away extent items, csum items and directory items.
1276 * It starts at a high offset and removes keys until it can't find
1277 * any higher than i_size.
1278 *
1279 * csum items that cross the new i_size are truncated to the new size
1280 * as well.
1281 *
1282 * min_type is the minimum key type to truncate down to. If set to 0, this
1283 * will kill all the items on this inode, including the INODE_ITEM_KEY.
1284 */
1285 static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
1286 struct btrfs_root *root,
1287 struct inode *inode,
1288 u32 min_type)
1289 {
1290 int ret;
1291 struct btrfs_path *path;
1292 struct btrfs_key key;
1293 struct btrfs_key found_key;
1294 u32 found_type;
1295 struct extent_buffer *leaf;
1296 struct btrfs_file_extent_item *fi;
1297 u64 extent_start = 0;
1298 u64 extent_num_bytes = 0;
1299 u64 item_end = 0;
1300 u64 root_gen = 0;
1301 u64 root_owner = 0;
1302 int found_extent;
1303 int del_item;
1304 int pending_del_nr = 0;
1305 int pending_del_slot = 0;
1306 int extent_type = -1;
1307 u64 mask = root->sectorsize - 1;
1308
1309 btrfs_drop_extent_cache(inode, inode->i_size & (~mask), (u64)-1);
1310 path = btrfs_alloc_path();
1311 path->reada = -1;
1312 BUG_ON(!path);
1313
1314 /* FIXME, add redo link to tree so we don't leak on crash */
1315 key.objectid = inode->i_ino;
1316 key.offset = (u64)-1;
1317 key.type = (u8)-1;
1318
1319 btrfs_init_path(path);
1320 search_again:
1321 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1322 if (ret < 0) {
1323 goto error;
1324 }
1325 if (ret > 0) {
1326 BUG_ON(path->slots[0] == 0);
1327 path->slots[0]--;
1328 }
1329
1330 while(1) {
1331 fi = NULL;
1332 leaf = path->nodes[0];
1333 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1334 found_type = btrfs_key_type(&found_key);
1335
1336 if (found_key.objectid != inode->i_ino)
1337 break;
1338
1339 if (found_type < min_type)
1340 break;
1341
1342 item_end = found_key.offset;
1343 if (found_type == BTRFS_EXTENT_DATA_KEY) {
1344 fi = btrfs_item_ptr(leaf, path->slots[0],
1345 struct btrfs_file_extent_item);
1346 extent_type = btrfs_file_extent_type(leaf, fi);
1347 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
1348 item_end +=
1349 btrfs_file_extent_num_bytes(leaf, fi);
1350 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1351 struct btrfs_item *item = btrfs_item_nr(leaf,
1352 path->slots[0]);
1353 item_end += btrfs_file_extent_inline_len(leaf,
1354 item);
1355 }
1356 item_end--;
1357 }
1358 if (found_type == BTRFS_CSUM_ITEM_KEY) {
1359 ret = btrfs_csum_truncate(trans, root, path,
1360 inode->i_size);
1361 BUG_ON(ret);
1362 }
1363 if (item_end < inode->i_size) {
1364 if (found_type == BTRFS_DIR_ITEM_KEY) {
1365 found_type = BTRFS_INODE_ITEM_KEY;
1366 } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
1367 found_type = BTRFS_CSUM_ITEM_KEY;
1368 } else if (found_type == BTRFS_EXTENT_DATA_KEY) {
1369 found_type = BTRFS_XATTR_ITEM_KEY;
1370 } else if (found_type == BTRFS_XATTR_ITEM_KEY) {
1371 found_type = BTRFS_INODE_REF_KEY;
1372 } else if (found_type) {
1373 found_type--;
1374 } else {
1375 break;
1376 }
1377 btrfs_set_key_type(&key, found_type);
1378 goto next;
1379 }
1380 if (found_key.offset >= inode->i_size)
1381 del_item = 1;
1382 else
1383 del_item = 0;
1384 found_extent = 0;
1385
1386 /* FIXME, shrink the extent if the ref count is only 1 */
1387 if (found_type != BTRFS_EXTENT_DATA_KEY)
1388 goto delete;
1389
1390 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
1391 u64 num_dec;
1392 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
1393 if (!del_item) {
1394 u64 orig_num_bytes =
1395 btrfs_file_extent_num_bytes(leaf, fi);
1396 extent_num_bytes = inode->i_size -
1397 found_key.offset + root->sectorsize - 1;
1398 extent_num_bytes = extent_num_bytes &
1399 ~((u64)root->sectorsize - 1);
1400 btrfs_set_file_extent_num_bytes(leaf, fi,
1401 extent_num_bytes);
1402 num_dec = (orig_num_bytes -
1403 extent_num_bytes);
1404 if (extent_start != 0)
1405 dec_i_blocks(inode, num_dec);
1406 btrfs_mark_buffer_dirty(leaf);
1407 } else {
1408 extent_num_bytes =
1409 btrfs_file_extent_disk_num_bytes(leaf,
1410 fi);
1411 /* FIXME blocksize != 4096 */
1412 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
1413 if (extent_start != 0) {
1414 found_extent = 1;
1415 dec_i_blocks(inode, num_dec);
1416 }
1417 root_gen = btrfs_header_generation(leaf);
1418 root_owner = btrfs_header_owner(leaf);
1419 }
1420 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1421 if (!del_item) {
1422 u32 newsize = inode->i_size - found_key.offset;
1423 dec_i_blocks(inode, item_end + 1 -
1424 found_key.offset - newsize);
1425 newsize =
1426 btrfs_file_extent_calc_inline_size(newsize);
1427 ret = btrfs_truncate_item(trans, root, path,
1428 newsize, 1);
1429 BUG_ON(ret);
1430 } else {
1431 dec_i_blocks(inode, item_end + 1 -
1432 found_key.offset);
1433 }
1434 }
1435 delete:
1436 if (del_item) {
1437 if (!pending_del_nr) {
1438 /* no pending yet, add ourselves */
1439 pending_del_slot = path->slots[0];
1440 pending_del_nr = 1;
1441 } else if (pending_del_nr &&
1442 path->slots[0] + 1 == pending_del_slot) {
1443 /* hop on the pending chunk */
1444 pending_del_nr++;
1445 pending_del_slot = path->slots[0];
1446 } else {
1447 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path->slots[0], pending_del_nr, pending_del_slot);
1448 }
1449 } else {
1450 break;
1451 }
1452 if (found_extent) {
1453 ret = btrfs_free_extent(trans, root, extent_start,
1454 extent_num_bytes,
1455 root_owner,
1456 root_gen, inode->i_ino,
1457 found_key.offset, 0);
1458 BUG_ON(ret);
1459 }
1460 next:
1461 if (path->slots[0] == 0) {
1462 if (pending_del_nr)
1463 goto del_pending;
1464 btrfs_release_path(root, path);
1465 goto search_again;
1466 }
1467
1468 path->slots[0]--;
1469 if (pending_del_nr &&
1470 path->slots[0] + 1 != pending_del_slot) {
1471 struct btrfs_key debug;
1472 del_pending:
1473 btrfs_item_key_to_cpu(path->nodes[0], &debug,
1474 pending_del_slot);
1475 ret = btrfs_del_items(trans, root, path,
1476 pending_del_slot,
1477 pending_del_nr);
1478 BUG_ON(ret);
1479 pending_del_nr = 0;
1480 btrfs_release_path(root, path);
1481 goto search_again;
1482 }
1483 }
1484 ret = 0;
1485 error:
1486 if (pending_del_nr) {
1487 ret = btrfs_del_items(trans, root, path, pending_del_slot,
1488 pending_del_nr);
1489 }
1490 btrfs_free_path(path);
1491 inode->i_sb->s_dirt = 1;
1492 return ret;
1493 }
1494
1495 /*
1496 * taken from block_truncate_page, but does cow as it zeros out
1497 * any bytes left in the last page in the file.
1498 */
1499 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
1500 {
1501 struct inode *inode = mapping->host;
1502 struct btrfs_root *root = BTRFS_I(inode)->root;
1503 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1504 struct btrfs_ordered_extent *ordered;
1505 char *kaddr;
1506 u32 blocksize = root->sectorsize;
1507 pgoff_t index = from >> PAGE_CACHE_SHIFT;
1508 unsigned offset = from & (PAGE_CACHE_SIZE-1);
1509 struct page *page;
1510 int ret = 0;
1511 u64 page_start;
1512 u64 page_end;
1513
1514 if ((offset & (blocksize - 1)) == 0)
1515 goto out;
1516
1517 ret = -ENOMEM;
1518 again:
1519 page = grab_cache_page(mapping, index);
1520 if (!page)
1521 goto out;
1522
1523 page_start = page_offset(page);
1524 page_end = page_start + PAGE_CACHE_SIZE - 1;
1525
1526 if (!PageUptodate(page)) {
1527 ret = btrfs_readpage(NULL, page);
1528 lock_page(page);
1529 if (page->mapping != mapping) {
1530 unlock_page(page);
1531 page_cache_release(page);
1532 goto again;
1533 }
1534 if (!PageUptodate(page)) {
1535 ret = -EIO;
1536 goto out_unlock;
1537 }
1538 }
1539 wait_on_page_writeback(page);
1540
1541 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
1542 set_page_extent_mapped(page);
1543
1544 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1545 if (ordered) {
1546 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
1547 unlock_page(page);
1548 page_cache_release(page);
1549 btrfs_start_ordered_extent(inode, ordered, 1);
1550 btrfs_put_ordered_extent(ordered);
1551 goto again;
1552 }
1553
1554 btrfs_set_extent_delalloc(inode, page_start, page_end);
1555 ret = 0;
1556 if (offset != PAGE_CACHE_SIZE) {
1557 kaddr = kmap(page);
1558 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
1559 flush_dcache_page(page);
1560 kunmap(page);
1561 }
1562 ClearPageChecked(page);
1563 set_page_dirty(page);
1564 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
1565
1566 out_unlock:
1567 unlock_page(page);
1568 page_cache_release(page);
1569 out:
1570 return ret;
1571 }
1572
1573 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
1574 {
1575 struct inode *inode = dentry->d_inode;
1576 int err;
1577
1578 err = inode_change_ok(inode, attr);
1579 if (err)
1580 return err;
1581
1582 if (S_ISREG(inode->i_mode) &&
1583 attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
1584 struct btrfs_trans_handle *trans;
1585 struct btrfs_root *root = BTRFS_I(inode)->root;
1586 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1587
1588 u64 mask = root->sectorsize - 1;
1589 u64 hole_start = (inode->i_size + mask) & ~mask;
1590 u64 block_end = (attr->ia_size + mask) & ~mask;
1591 u64 hole_size;
1592 u64 alloc_hint = 0;
1593
1594 if (attr->ia_size <= hole_start)
1595 goto out;
1596
1597 err = btrfs_check_free_space(root, 1, 0);
1598 if (err)
1599 goto fail;
1600
1601 btrfs_truncate_page(inode->i_mapping, inode->i_size);
1602
1603 hole_size = block_end - hole_start;
1604 while(1) {
1605 struct btrfs_ordered_extent *ordered;
1606 btrfs_wait_ordered_range(inode, hole_start, hole_size);
1607
1608 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1609 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
1610 if (ordered) {
1611 unlock_extent(io_tree, hole_start,
1612 block_end - 1, GFP_NOFS);
1613 btrfs_put_ordered_extent(ordered);
1614 } else {
1615 break;
1616 }
1617 }
1618
1619 trans = btrfs_start_transaction(root, 1);
1620 btrfs_set_trans_block_group(trans, inode);
1621 mutex_lock(&BTRFS_I(inode)->extent_mutex);
1622 err = btrfs_drop_extents(trans, root, inode,
1623 hole_start, block_end, hole_start,
1624 &alloc_hint);
1625
1626 if (alloc_hint != EXTENT_MAP_INLINE) {
1627 err = btrfs_insert_file_extent(trans, root,
1628 inode->i_ino,
1629 hole_start, 0, 0,
1630 hole_size, 0);
1631 btrfs_drop_extent_cache(inode, hole_start,
1632 (u64)-1);
1633 btrfs_check_file(root, inode);
1634 }
1635 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1636 btrfs_end_transaction(trans, root);
1637 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1638 if (err)
1639 return err;
1640 }
1641 out:
1642 err = inode_setattr(inode, attr);
1643
1644 if (!err && ((attr->ia_valid & ATTR_MODE)))
1645 err = btrfs_acl_chmod(inode);
1646 fail:
1647 return err;
1648 }
1649
1650 void btrfs_delete_inode(struct inode *inode)
1651 {
1652 struct btrfs_trans_handle *trans;
1653 struct btrfs_root *root = BTRFS_I(inode)->root;
1654 unsigned long nr;
1655 int ret;
1656
1657 truncate_inode_pages(&inode->i_data, 0);
1658 if (is_bad_inode(inode)) {
1659 btrfs_orphan_del(NULL, inode);
1660 goto no_delete;
1661 }
1662 btrfs_wait_ordered_range(inode, 0, (u64)-1);
1663
1664 btrfs_i_size_write(inode, 0);
1665 trans = btrfs_start_transaction(root, 1);
1666
1667 btrfs_set_trans_block_group(trans, inode);
1668 ret = btrfs_truncate_in_trans(trans, root, inode, 0);
1669 if (ret) {
1670 btrfs_orphan_del(NULL, inode);
1671 goto no_delete_lock;
1672 }
1673
1674 btrfs_orphan_del(trans, inode);
1675
1676 nr = trans->blocks_used;
1677 clear_inode(inode);
1678
1679 btrfs_end_transaction(trans, root);
1680 btrfs_btree_balance_dirty(root, nr);
1681 return;
1682
1683 no_delete_lock:
1684 nr = trans->blocks_used;
1685 btrfs_end_transaction(trans, root);
1686 btrfs_btree_balance_dirty(root, nr);
1687 no_delete:
1688 clear_inode(inode);
1689 }
1690
1691 /*
1692 * this returns the key found in the dir entry in the location pointer.
1693 * If no dir entries were found, location->objectid is 0.
1694 */
1695 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
1696 struct btrfs_key *location)
1697 {
1698 const char *name = dentry->d_name.name;
1699 int namelen = dentry->d_name.len;
1700 struct btrfs_dir_item *di;
1701 struct btrfs_path *path;
1702 struct btrfs_root *root = BTRFS_I(dir)->root;
1703 int ret = 0;
1704
1705 path = btrfs_alloc_path();
1706 BUG_ON(!path);
1707
1708 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
1709 namelen, 0);
1710 if (IS_ERR(di))
1711 ret = PTR_ERR(di);
1712 if (!di || IS_ERR(di)) {
1713 goto out_err;
1714 }
1715 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
1716 out:
1717 btrfs_free_path(path);
1718 return ret;
1719 out_err:
1720 location->objectid = 0;
1721 goto out;
1722 }
1723
1724 /*
1725 * when we hit a tree root in a directory, the btrfs part of the inode
1726 * needs to be changed to reflect the root directory of the tree root. This
1727 * is kind of like crossing a mount point.
1728 */
1729 static int fixup_tree_root_location(struct btrfs_root *root,
1730 struct btrfs_key *location,
1731 struct btrfs_root **sub_root,
1732 struct dentry *dentry)
1733 {
1734 struct btrfs_root_item *ri;
1735
1736 if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
1737 return 0;
1738 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1739 return 0;
1740
1741 *sub_root = btrfs_read_fs_root(root->fs_info, location,
1742 dentry->d_name.name,
1743 dentry->d_name.len);
1744 if (IS_ERR(*sub_root))
1745 return PTR_ERR(*sub_root);
1746
1747 ri = &(*sub_root)->root_item;
1748 location->objectid = btrfs_root_dirid(ri);
1749 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1750 location->offset = 0;
1751
1752 return 0;
1753 }
1754
1755 static int btrfs_init_locked_inode(struct inode *inode, void *p)
1756 {
1757 struct btrfs_iget_args *args = p;
1758 inode->i_ino = args->ino;
1759 BTRFS_I(inode)->root = args->root;
1760 BTRFS_I(inode)->delalloc_bytes = 0;
1761 inode->i_mapping->writeback_index = 0;
1762 BTRFS_I(inode)->disk_i_size = 0;
1763 BTRFS_I(inode)->index_cnt = (u64)-1;
1764 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1765 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1766 inode->i_mapping, GFP_NOFS);
1767 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
1768 inode->i_mapping, GFP_NOFS);
1769 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
1770 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
1771 mutex_init(&BTRFS_I(inode)->csum_mutex);
1772 mutex_init(&BTRFS_I(inode)->extent_mutex);
1773 return 0;
1774 }
1775
1776 static int btrfs_find_actor(struct inode *inode, void *opaque)
1777 {
1778 struct btrfs_iget_args *args = opaque;
1779 return (args->ino == inode->i_ino &&
1780 args->root == BTRFS_I(inode)->root);
1781 }
1782
1783 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
1784 u64 root_objectid)
1785 {
1786 struct btrfs_iget_args args;
1787 args.ino = objectid;
1788 args.root = btrfs_lookup_fs_root(btrfs_sb(s)->fs_info, root_objectid);
1789
1790 if (!args.root)
1791 return NULL;
1792
1793 return ilookup5(s, objectid, btrfs_find_actor, (void *)&args);
1794 }
1795
1796 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
1797 struct btrfs_root *root)
1798 {
1799 struct inode *inode;
1800 struct btrfs_iget_args args;
1801 args.ino = objectid;
1802 args.root = root;
1803
1804 inode = iget5_locked(s, objectid, btrfs_find_actor,
1805 btrfs_init_locked_inode,
1806 (void *)&args);
1807 return inode;
1808 }
1809
1810 /* Get an inode object given its location and corresponding root.
1811 * Returns in *is_new if the inode was read from disk
1812 */
1813 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
1814 struct btrfs_root *root, int *is_new)
1815 {
1816 struct inode *inode;
1817
1818 inode = btrfs_iget_locked(s, location->objectid, root);
1819 if (!inode)
1820 return ERR_PTR(-EACCES);
1821
1822 if (inode->i_state & I_NEW) {
1823 BTRFS_I(inode)->root = root;
1824 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
1825 btrfs_read_locked_inode(inode);
1826 unlock_new_inode(inode);
1827 if (is_new)
1828 *is_new = 1;
1829 } else {
1830 if (is_new)
1831 *is_new = 0;
1832 }
1833
1834 return inode;
1835 }
1836
1837 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
1838 struct nameidata *nd)
1839 {
1840 struct inode * inode;
1841 struct btrfs_inode *bi = BTRFS_I(dir);
1842 struct btrfs_root *root = bi->root;
1843 struct btrfs_root *sub_root = root;
1844 struct btrfs_key location;
1845 int ret, new, do_orphan = 0;
1846
1847 if (dentry->d_name.len > BTRFS_NAME_LEN)
1848 return ERR_PTR(-ENAMETOOLONG);
1849
1850 ret = btrfs_inode_by_name(dir, dentry, &location);
1851
1852 if (ret < 0)
1853 return ERR_PTR(ret);
1854
1855 inode = NULL;
1856 if (location.objectid) {
1857 ret = fixup_tree_root_location(root, &location, &sub_root,
1858 dentry);
1859 if (ret < 0)
1860 return ERR_PTR(ret);
1861 if (ret > 0)
1862 return ERR_PTR(-ENOENT);
1863 inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
1864 if (IS_ERR(inode))
1865 return ERR_CAST(inode);
1866
1867 /* the inode and parent dir are two different roots */
1868 if (new && root != sub_root) {
1869 igrab(inode);
1870 sub_root->inode = inode;
1871 do_orphan = 1;
1872 }
1873 }
1874
1875 if (unlikely(do_orphan))
1876 btrfs_orphan_cleanup(sub_root);
1877
1878 return d_splice_alias(inode, dentry);
1879 }
1880
1881 static unsigned char btrfs_filetype_table[] = {
1882 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
1883 };
1884
1885 static int btrfs_real_readdir(struct file *filp, void *dirent,
1886 filldir_t filldir)
1887 {
1888 struct inode *inode = filp->f_dentry->d_inode;
1889 struct btrfs_root *root = BTRFS_I(inode)->root;
1890 struct btrfs_item *item;
1891 struct btrfs_dir_item *di;
1892 struct btrfs_key key;
1893 struct btrfs_key found_key;
1894 struct btrfs_path *path;
1895 int ret;
1896 u32 nritems;
1897 struct extent_buffer *leaf;
1898 int slot;
1899 int advance;
1900 unsigned char d_type;
1901 int over = 0;
1902 u32 di_cur;
1903 u32 di_total;
1904 u32 di_len;
1905 int key_type = BTRFS_DIR_INDEX_KEY;
1906 char tmp_name[32];
1907 char *name_ptr;
1908 int name_len;
1909
1910 /* FIXME, use a real flag for deciding about the key type */
1911 if (root->fs_info->tree_root == root)
1912 key_type = BTRFS_DIR_ITEM_KEY;
1913
1914 /* special case for "." */
1915 if (filp->f_pos == 0) {
1916 over = filldir(dirent, ".", 1,
1917 1, inode->i_ino,
1918 DT_DIR);
1919 if (over)
1920 return 0;
1921 filp->f_pos = 1;
1922 }
1923 /* special case for .., just use the back ref */
1924 if (filp->f_pos == 1) {
1925 u64 pino = parent_ino(filp->f_path.dentry);
1926 over = filldir(dirent, "..", 2,
1927 2, pino, DT_DIR);
1928 if (over)
1929 return 0;
1930 filp->f_pos = 2;
1931 }
1932
1933 path = btrfs_alloc_path();
1934 path->reada = 2;
1935
1936 btrfs_set_key_type(&key, key_type);
1937 key.offset = filp->f_pos;
1938 key.objectid = inode->i_ino;
1939
1940 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1941 if (ret < 0)
1942 goto err;
1943 advance = 0;
1944
1945 while (1) {
1946 leaf = path->nodes[0];
1947 nritems = btrfs_header_nritems(leaf);
1948 slot = path->slots[0];
1949 if (advance || slot >= nritems) {
1950 if (slot >= nritems - 1) {
1951 ret = btrfs_next_leaf(root, path);
1952 if (ret)
1953 break;
1954 leaf = path->nodes[0];
1955 nritems = btrfs_header_nritems(leaf);
1956 slot = path->slots[0];
1957 } else {
1958 slot++;
1959 path->slots[0]++;
1960 }
1961 }
1962 advance = 1;
1963 item = btrfs_item_nr(leaf, slot);
1964 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1965
1966 if (found_key.objectid != key.objectid)
1967 break;
1968 if (btrfs_key_type(&found_key) != key_type)
1969 break;
1970 if (found_key.offset < filp->f_pos)
1971 continue;
1972
1973 filp->f_pos = found_key.offset;
1974
1975 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
1976 di_cur = 0;
1977 di_total = btrfs_item_size(leaf, item);
1978
1979 while (di_cur < di_total) {
1980 struct btrfs_key location;
1981
1982 name_len = btrfs_dir_name_len(leaf, di);
1983 if (name_len <= sizeof(tmp_name)) {
1984 name_ptr = tmp_name;
1985 } else {
1986 name_ptr = kmalloc(name_len, GFP_NOFS);
1987 if (!name_ptr) {
1988 ret = -ENOMEM;
1989 goto err;
1990 }
1991 }
1992 read_extent_buffer(leaf, name_ptr,
1993 (unsigned long)(di + 1), name_len);
1994
1995 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
1996 btrfs_dir_item_key_to_cpu(leaf, di, &location);
1997 over = filldir(dirent, name_ptr, name_len,
1998 found_key.offset, location.objectid,
1999 d_type);
2000
2001 if (name_ptr != tmp_name)
2002 kfree(name_ptr);
2003
2004 if (over)
2005 goto nopos;
2006
2007 di_len = btrfs_dir_name_len(leaf, di) +
2008 btrfs_dir_data_len(leaf, di) + sizeof(*di);
2009 di_cur += di_len;
2010 di = (struct btrfs_dir_item *)((char *)di + di_len);
2011 }
2012 }
2013
2014 /* Reached end of directory/root. Bump pos past the last item. */
2015 if (key_type == BTRFS_DIR_INDEX_KEY)
2016 filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
2017 else
2018 filp->f_pos++;
2019 nopos:
2020 ret = 0;
2021 err:
2022 btrfs_free_path(path);
2023 return ret;
2024 }
2025
2026 /* Kernels earlier than 2.6.28 still have the NFS deadlock where nfsd
2027 will call the file system's ->lookup() method from within its
2028 filldir callback, which in turn was called from the file system's
2029 ->readdir() method. And will deadlock for many file systems. */
2030 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,28)
2031
2032 struct nfshack_dirent {
2033 u64 ino;
2034 loff_t offset;
2035 int namlen;
2036 unsigned int d_type;
2037 char name[];
2038 };
2039
2040 struct nfshack_readdir {
2041 char *dirent;
2042 size_t used;
2043 int full;
2044 };
2045
2046
2047
2048 static int btrfs_nfshack_filldir(void *__buf, const char *name, int namlen,
2049 loff_t offset, u64 ino, unsigned int d_type)
2050 {
2051 struct nfshack_readdir *buf = __buf;
2052 struct nfshack_dirent *de = (void *)(buf->dirent + buf->used);
2053 unsigned int reclen;
2054
2055 reclen = ALIGN(sizeof(struct nfshack_dirent) + namlen, sizeof(u64));
2056 if (buf->used + reclen > PAGE_SIZE) {
2057 buf->full = 1;
2058 return -EINVAL;
2059 }
2060
2061 de->namlen = namlen;
2062 de->offset = offset;
2063 de->ino = ino;
2064 de->d_type = d_type;
2065 memcpy(de->name, name, namlen);
2066 buf->used += reclen;
2067
2068 return 0;
2069 }
2070
2071 static int btrfs_nfshack_readdir(struct file *file, void *dirent,
2072 filldir_t filldir)
2073 {
2074 struct nfshack_readdir buf;
2075 struct nfshack_dirent *de;
2076 int err;
2077 int size;
2078 loff_t offset;
2079
2080 buf.dirent = (void *)__get_free_page(GFP_KERNEL);
2081 if (!buf.dirent)
2082 return -ENOMEM;
2083
2084 offset = file->f_pos;
2085
2086 do {
2087 unsigned int reclen;
2088
2089 buf.used = 0;
2090 buf.full = 0;
2091 err = btrfs_real_readdir(file, &buf, btrfs_nfshack_filldir);
2092 if (err)
2093 break;
2094
2095 size = buf.used;
2096
2097 if (!size)
2098 break;
2099
2100 de = (struct nfshack_dirent *)buf.dirent;
2101 while (size > 0) {
2102 offset = de->offset;
2103
2104 if (filldir(dirent, de->name, de->namlen, de->offset,
2105 de->ino, de->d_type))
2106 goto done;
2107 offset = file->f_pos;
2108
2109 reclen = ALIGN(sizeof(*de) + de->namlen,
2110 sizeof(u64));
2111 size -= reclen;
2112 de = (struct nfshack_dirent *)((char *)de + reclen);
2113 }
2114 } while (buf.full);
2115
2116 done:
2117 free_page((unsigned long)buf.dirent);
2118 file->f_pos = offset;
2119
2120 return err;
2121 }
2122 #endif
2123
2124 int btrfs_write_inode(struct inode *inode, int wait)
2125 {
2126 struct btrfs_root *root = BTRFS_I(inode)->root;
2127 struct btrfs_trans_handle *trans;
2128 int ret = 0;
2129
2130 if (root->fs_info->closing > 1)
2131 return 0;
2132
2133 if (wait) {
2134 trans = btrfs_join_transaction(root, 1);
2135 btrfs_set_trans_block_group(trans, inode);
2136 ret = btrfs_commit_transaction(trans, root);
2137 }
2138 return ret;
2139 }
2140
2141 /*
2142 * This is somewhat expensive, updating the tree every time the
2143 * inode changes. But, it is most likely to find the inode in cache.
2144 * FIXME, needs more benchmarking...there are no reasons other than performance
2145 * to keep or drop this code.
2146 */
2147 void btrfs_dirty_inode(struct inode *inode)
2148 {
2149 struct btrfs_root *root = BTRFS_I(inode)->root;
2150 struct btrfs_trans_handle *trans;
2151
2152 trans = btrfs_join_transaction(root, 1);
2153 btrfs_set_trans_block_group(trans, inode);
2154 btrfs_update_inode(trans, root, inode);
2155 btrfs_end_transaction(trans, root);
2156 }
2157
2158 static int btrfs_set_inode_index_count(struct inode *inode)
2159 {
2160 struct btrfs_root *root = BTRFS_I(inode)->root;
2161 struct btrfs_key key, found_key;
2162 struct btrfs_path *path;
2163 struct extent_buffer *leaf;
2164 int ret;
2165
2166 key.objectid = inode->i_ino;
2167 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
2168 key.offset = (u64)-1;
2169
2170 path = btrfs_alloc_path();
2171 if (!path)
2172 return -ENOMEM;
2173
2174 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2175 if (ret < 0)
2176 goto out;
2177 /* FIXME: we should be able to handle this */
2178 if (ret == 0)
2179 goto out;
2180 ret = 0;
2181
2182 /*
2183 * MAGIC NUMBER EXPLANATION:
2184 * since we search a directory based on f_pos we have to start at 2
2185 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
2186 * else has to start at 2
2187 */
2188 if (path->slots[0] == 0) {
2189 BTRFS_I(inode)->index_cnt = 2;
2190 goto out;
2191 }
2192
2193 path->slots[0]--;
2194
2195 leaf = path->nodes[0];
2196 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2197
2198 if (found_key.objectid != inode->i_ino ||
2199 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
2200 BTRFS_I(inode)->index_cnt = 2;
2201 goto out;
2202 }
2203
2204 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
2205 out:
2206 btrfs_free_path(path);
2207 return ret;
2208 }
2209
2210 static int btrfs_set_inode_index(struct inode *dir, struct inode *inode,
2211 u64 *index)
2212 {
2213 int ret = 0;
2214
2215 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
2216 ret = btrfs_set_inode_index_count(dir);
2217 if (ret)
2218 return ret;
2219 }
2220
2221 *index = BTRFS_I(dir)->index_cnt;
2222 BTRFS_I(dir)->index_cnt++;
2223
2224 return ret;
2225 }
2226
2227 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
2228 struct btrfs_root *root,
2229 struct inode *dir,
2230 const char *name, int name_len,
2231 u64 ref_objectid,
2232 u64 objectid,
2233 struct btrfs_block_group_cache *group,
2234 int mode, u64 *index)
2235 {
2236 struct inode *inode;
2237 struct btrfs_inode_item *inode_item;
2238 struct btrfs_block_group_cache *new_inode_group;
2239 struct btrfs_key *location;
2240 struct btrfs_path *path;
2241 struct btrfs_inode_ref *ref;
2242 struct btrfs_key key[2];
2243 u32 sizes[2];
2244 unsigned long ptr;
2245 int ret;
2246 int owner;
2247
2248 path = btrfs_alloc_path();
2249 BUG_ON(!path);
2250
2251 inode = new_inode(root->fs_info->sb);
2252 if (!inode)
2253 return ERR_PTR(-ENOMEM);
2254
2255 if (dir) {
2256 ret = btrfs_set_inode_index(dir, inode, index);
2257 if (ret)
2258 return ERR_PTR(ret);
2259 }
2260 /*
2261 * index_cnt is ignored for everything but a dir,
2262 * btrfs_get_inode_index_count has an explanation for the magic
2263 * number
2264 */
2265 BTRFS_I(inode)->index_cnt = 2;
2266
2267 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2268 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2269 inode->i_mapping, GFP_NOFS);
2270 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2271 inode->i_mapping, GFP_NOFS);
2272 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
2273 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
2274 mutex_init(&BTRFS_I(inode)->csum_mutex);
2275 mutex_init(&BTRFS_I(inode)->extent_mutex);
2276 BTRFS_I(inode)->delalloc_bytes = 0;
2277 inode->i_mapping->writeback_index = 0;
2278 BTRFS_I(inode)->disk_i_size = 0;
2279 BTRFS_I(inode)->root = root;
2280
2281 if (mode & S_IFDIR)
2282 owner = 0;
2283 else
2284 owner = 1;
2285 new_inode_group = btrfs_find_block_group(root, group, 0,
2286 BTRFS_BLOCK_GROUP_METADATA, owner);
2287 if (!new_inode_group) {
2288 printk("find_block group failed\n");
2289 new_inode_group = group;
2290 }
2291 BTRFS_I(inode)->block_group = new_inode_group;
2292 BTRFS_I(inode)->flags = 0;
2293
2294 key[0].objectid = objectid;
2295 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
2296 key[0].offset = 0;
2297
2298 key[1].objectid = objectid;
2299 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
2300 key[1].offset = ref_objectid;
2301
2302 sizes[0] = sizeof(struct btrfs_inode_item);
2303 sizes[1] = name_len + sizeof(*ref);
2304
2305 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
2306 if (ret != 0)
2307 goto fail;
2308
2309 if (objectid > root->highest_inode)
2310 root->highest_inode = objectid;
2311
2312 inode->i_uid = current->fsuid;
2313 inode->i_gid = current->fsgid;
2314 inode->i_mode = mode;
2315 inode->i_ino = objectid;
2316 inode->i_blocks = 0;
2317 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2318 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2319 struct btrfs_inode_item);
2320 fill_inode_item(path->nodes[0], inode_item, inode);
2321
2322 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2323 struct btrfs_inode_ref);
2324 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
2325 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
2326 ptr = (unsigned long)(ref + 1);
2327 write_extent_buffer(path->nodes[0], name, ptr, name_len);
2328
2329 btrfs_mark_buffer_dirty(path->nodes[0]);
2330 btrfs_free_path(path);
2331
2332 location = &BTRFS_I(inode)->location;
2333 location->objectid = objectid;
2334 location->offset = 0;
2335 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
2336
2337 insert_inode_hash(inode);
2338 return inode;
2339 fail:
2340 if (dir)
2341 BTRFS_I(dir)->index_cnt--;
2342 btrfs_free_path(path);
2343 return ERR_PTR(ret);
2344 }
2345
2346 static inline u8 btrfs_inode_type(struct inode *inode)
2347 {
2348 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
2349 }
2350
2351 static int btrfs_add_link(struct btrfs_trans_handle *trans,
2352 struct dentry *dentry, struct inode *inode,
2353 int add_backref, u64 index)
2354 {
2355 int ret;
2356 struct btrfs_key key;
2357 struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
2358 struct inode *parent_inode = dentry->d_parent->d_inode;
2359
2360 key.objectid = inode->i_ino;
2361 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
2362 key.offset = 0;
2363
2364 ret = btrfs_insert_dir_item(trans, root,
2365 dentry->d_name.name, dentry->d_name.len,
2366 dentry->d_parent->d_inode->i_ino,
2367 &key, btrfs_inode_type(inode),
2368 index);
2369 if (ret == 0) {
2370 if (add_backref) {
2371 ret = btrfs_insert_inode_ref(trans, root,
2372 dentry->d_name.name,
2373 dentry->d_name.len,
2374 inode->i_ino,
2375 parent_inode->i_ino,
2376 index);
2377 }
2378 btrfs_i_size_write(parent_inode, parent_inode->i_size +
2379 dentry->d_name.len * 2);
2380 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
2381 ret = btrfs_update_inode(trans, root,
2382 dentry->d_parent->d_inode);
2383 }
2384 return ret;
2385 }
2386
2387 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
2388 struct dentry *dentry, struct inode *inode,
2389 int backref, u64 index)
2390 {
2391 int err = btrfs_add_link(trans, dentry, inode, backref, index);
2392 if (!err) {
2393 d_instantiate(dentry, inode);
2394 return 0;
2395 }
2396 if (err > 0)
2397 err = -EEXIST;
2398 return err;
2399 }
2400
2401 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
2402 int mode, dev_t rdev)
2403 {
2404 struct btrfs_trans_handle *trans;
2405 struct btrfs_root *root = BTRFS_I(dir)->root;
2406 struct inode *inode = NULL;
2407 int err;
2408 int drop_inode = 0;
2409 u64 objectid;
2410 unsigned long nr = 0;
2411 u64 index = 0;
2412
2413 if (!new_valid_dev(rdev))
2414 return -EINVAL;
2415
2416 err = btrfs_check_free_space(root, 1, 0);
2417 if (err)
2418 goto fail;
2419
2420 trans = btrfs_start_transaction(root, 1);
2421 btrfs_set_trans_block_group(trans, dir);
2422
2423 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2424 if (err) {
2425 err = -ENOSPC;
2426 goto out_unlock;
2427 }
2428
2429 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
2430 dentry->d_name.len,
2431 dentry->d_parent->d_inode->i_ino, objectid,
2432 BTRFS_I(dir)->block_group, mode, &index);
2433 err = PTR_ERR(inode);
2434 if (IS_ERR(inode))
2435 goto out_unlock;
2436
2437 err = btrfs_init_acl(inode, dir);
2438 if (err) {
2439 drop_inode = 1;
2440 goto out_unlock;
2441 }
2442
2443 btrfs_set_trans_block_group(trans, inode);
2444 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
2445 if (err)
2446 drop_inode = 1;
2447 else {
2448 inode->i_op = &btrfs_special_inode_operations;
2449 init_special_inode(inode, inode->i_mode, rdev);
2450 btrfs_update_inode(trans, root, inode);
2451 }
2452 dir->i_sb->s_dirt = 1;
2453 btrfs_update_inode_block_group(trans, inode);
2454 btrfs_update_inode_block_group(trans, dir);
2455 out_unlock:
2456 nr = trans->blocks_used;
2457 btrfs_end_transaction_throttle(trans, root);
2458 fail:
2459 if (drop_inode) {
2460 inode_dec_link_count(inode);
2461 iput(inode);
2462 }
2463 btrfs_btree_balance_dirty(root, nr);
2464 return err;
2465 }
2466
2467 static int btrfs_create(struct inode *dir, struct dentry *dentry,
2468 int mode, struct nameidata *nd)
2469 {
2470 struct btrfs_trans_handle *trans;
2471 struct btrfs_root *root = BTRFS_I(dir)->root;
2472 struct inode *inode = NULL;
2473 int err;
2474 int drop_inode = 0;
2475 unsigned long nr = 0;
2476 u64 objectid;
2477 u64 index = 0;
2478
2479 err = btrfs_check_free_space(root, 1, 0);
2480 if (err)
2481 goto fail;
2482 trans = btrfs_start_transaction(root, 1);
2483 btrfs_set_trans_block_group(trans, dir);
2484
2485 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2486 if (err) {
2487 err = -ENOSPC;
2488 goto out_unlock;
2489 }
2490
2491 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
2492 dentry->d_name.len,
2493 dentry->d_parent->d_inode->i_ino,
2494 objectid, BTRFS_I(dir)->block_group, mode,
2495 &index);
2496 err = PTR_ERR(inode);
2497 if (IS_ERR(inode))
2498 goto out_unlock;
2499
2500 err = btrfs_init_acl(inode, dir);
2501 if (err) {
2502 drop_inode = 1;
2503 goto out_unlock;
2504 }
2505
2506 btrfs_set_trans_block_group(trans, inode);
2507 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
2508 if (err)
2509 drop_inode = 1;
2510 else {
2511 inode->i_mapping->a_ops = &btrfs_aops;
2512 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2513 inode->i_fop = &btrfs_file_operations;
2514 inode->i_op = &btrfs_file_inode_operations;
2515 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2516 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2517 inode->i_mapping, GFP_NOFS);
2518 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2519 inode->i_mapping, GFP_NOFS);
2520 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
2521 mutex_init(&BTRFS_I(inode)->csum_mutex);
2522 mutex_init(&BTRFS_I(inode)->extent_mutex);
2523 BTRFS_I(inode)->delalloc_bytes = 0;
2524 BTRFS_I(inode)->disk_i_size = 0;
2525 inode->i_mapping->writeback_index = 0;
2526 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2527 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
2528 }
2529 dir->i_sb->s_dirt = 1;
2530 btrfs_update_inode_block_group(trans, inode);
2531 btrfs_update_inode_block_group(trans, dir);
2532 out_unlock:
2533 nr = trans->blocks_used;
2534 btrfs_end_transaction_throttle(trans, root);
2535 fail:
2536 if (drop_inode) {
2537 inode_dec_link_count(inode);
2538 iput(inode);
2539 }
2540 btrfs_btree_balance_dirty(root, nr);
2541 return err;
2542 }
2543
2544 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
2545 struct dentry *dentry)
2546 {
2547 struct btrfs_trans_handle *trans;
2548 struct btrfs_root *root = BTRFS_I(dir)->root;
2549 struct inode *inode = old_dentry->d_inode;
2550 u64 index;
2551 unsigned long nr = 0;
2552 int err;
2553 int drop_inode = 0;
2554
2555 if (inode->i_nlink == 0)
2556 return -ENOENT;
2557
2558 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
2559 inode->i_nlink++;
2560 #else
2561 inc_nlink(inode);
2562 #endif
2563 err = btrfs_check_free_space(root, 1, 0);
2564 if (err)
2565 goto fail;
2566 err = btrfs_set_inode_index(dir, inode, &index);
2567 if (err)
2568 goto fail;
2569
2570 trans = btrfs_start_transaction(root, 1);
2571
2572 btrfs_set_trans_block_group(trans, dir);
2573 atomic_inc(&inode->i_count);
2574
2575 err = btrfs_add_nondir(trans, dentry, inode, 1, index);
2576
2577 if (err)
2578 drop_inode = 1;
2579
2580 dir->i_sb->s_dirt = 1;
2581 btrfs_update_inode_block_group(trans, dir);
2582 err = btrfs_update_inode(trans, root, inode);
2583
2584 if (err)
2585 drop_inode = 1;
2586
2587 nr = trans->blocks_used;
2588 btrfs_end_transaction_throttle(trans, root);
2589 fail:
2590 if (drop_inode) {
2591 inode_dec_link_count(inode);
2592 iput(inode);
2593 }
2594 btrfs_btree_balance_dirty(root, nr);
2595 return err;
2596 }
2597
2598 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2599 {
2600 struct inode *inode = NULL;
2601 struct btrfs_trans_handle *trans;
2602 struct btrfs_root *root = BTRFS_I(dir)->root;
2603 int err = 0;
2604 int drop_on_err = 0;
2605 u64 objectid = 0;
2606 u64 index = 0;
2607 unsigned long nr = 1;
2608
2609 err = btrfs_check_free_space(root, 1, 0);
2610 if (err)
2611 goto out_unlock;
2612
2613 trans = btrfs_start_transaction(root, 1);
2614 btrfs_set_trans_block_group(trans, dir);
2615
2616 if (IS_ERR(trans)) {
2617 err = PTR_ERR(trans);
2618 goto out_unlock;
2619 }
2620
2621 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2622 if (err) {
2623 err = -ENOSPC;
2624 goto out_unlock;
2625 }
2626
2627 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
2628 dentry->d_name.len,
2629 dentry->d_parent->d_inode->i_ino, objectid,
2630 BTRFS_I(dir)->block_group, S_IFDIR | mode,
2631 &index);
2632 if (IS_ERR(inode)) {
2633 err = PTR_ERR(inode);
2634 goto out_fail;
2635 }
2636
2637 drop_on_err = 1;
2638
2639 err = btrfs_init_acl(inode, dir);
2640 if (err)
2641 goto out_fail;
2642
2643 inode->i_op = &btrfs_dir_inode_operations;
2644 inode->i_fop = &btrfs_dir_file_operations;
2645 btrfs_set_trans_block_group(trans, inode);
2646
2647 btrfs_i_size_write(inode, 0);
2648 err = btrfs_update_inode(trans, root, inode);
2649 if (err)
2650 goto out_fail;
2651
2652 err = btrfs_add_link(trans, dentry, inode, 0, index);
2653 if (err)
2654 goto out_fail;
2655
2656 d_instantiate(dentry, inode);
2657 drop_on_err = 0;
2658 dir->i_sb->s_dirt = 1;
2659 btrfs_update_inode_block_group(trans, inode);
2660 btrfs_update_inode_block_group(trans, dir);
2661
2662 out_fail:
2663 nr = trans->blocks_used;
2664 btrfs_end_transaction_throttle(trans, root);
2665
2666 out_unlock:
2667 if (drop_on_err)
2668 iput(inode);
2669 btrfs_btree_balance_dirty(root, nr);
2670 return err;
2671 }
2672
2673 static int merge_extent_mapping(struct extent_map_tree *em_tree,
2674 struct extent_map *existing,
2675 struct extent_map *em,
2676 u64 map_start, u64 map_len)
2677 {
2678 u64 start_diff;
2679
2680 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
2681 start_diff = map_start - em->start;
2682 em->start = map_start;
2683 em->len = map_len;
2684 if (em->block_start < EXTENT_MAP_LAST_BYTE)
2685 em->block_start += start_diff;
2686 return add_extent_mapping(em_tree, em);
2687 }
2688
2689 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
2690 size_t pg_offset, u64 start, u64 len,
2691 int create)
2692 {
2693 int ret;
2694 int err = 0;
2695 u64 bytenr;
2696 u64 extent_start = 0;
2697 u64 extent_end = 0;
2698 u64 objectid = inode->i_ino;
2699 u32 found_type;
2700 struct btrfs_path *path = NULL;
2701 struct btrfs_root *root = BTRFS_I(inode)->root;
2702 struct btrfs_file_extent_item *item;
2703 struct extent_buffer *leaf;
2704 struct btrfs_key found_key;
2705 struct extent_map *em = NULL;
2706 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2707 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2708 struct btrfs_trans_handle *trans = NULL;
2709
2710 again:
2711 spin_lock(&em_tree->lock);
2712 em = lookup_extent_mapping(em_tree, start, len);
2713 if (em)
2714 em->bdev = root->fs_info->fs_devices->latest_bdev;
2715 spin_unlock(&em_tree->lock);
2716
2717 if (em) {
2718 if (em->start > start || em->start + em->len <= start)
2719 free_extent_map(em);
2720 else if (em->block_start == EXTENT_MAP_INLINE && page)
2721 free_extent_map(em);
2722 else
2723 goto out;
2724 }
2725 em = alloc_extent_map(GFP_NOFS);
2726 if (!em) {
2727 err = -ENOMEM;
2728 goto out;
2729 }
2730 em->bdev = root->fs_info->fs_devices->latest_bdev;
2731 em->start = EXTENT_MAP_HOLE;
2732 em->len = (u64)-1;
2733
2734 if (!path) {
2735 path = btrfs_alloc_path();
2736 BUG_ON(!path);
2737 }
2738
2739 ret = btrfs_lookup_file_extent(trans, root, path,
2740 objectid, start, trans != NULL);
2741 if (ret < 0) {
2742 err = ret;
2743 goto out;
2744 }
2745
2746 if (ret != 0) {
2747 if (path->slots[0] == 0)
2748 goto not_found;
2749 path->slots[0]--;
2750 }
2751
2752 leaf = path->nodes[0];
2753 item = btrfs_item_ptr(leaf, path->slots[0],
2754 struct btrfs_file_extent_item);
2755 /* are we inside the extent that was found? */
2756 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2757 found_type = btrfs_key_type(&found_key);
2758 if (found_key.objectid != objectid ||
2759 found_type != BTRFS_EXTENT_DATA_KEY) {
2760 goto not_found;
2761 }
2762
2763 found_type = btrfs_file_extent_type(leaf, item);
2764 extent_start = found_key.offset;
2765 if (found_type == BTRFS_FILE_EXTENT_REG) {
2766 extent_end = extent_start +
2767 btrfs_file_extent_num_bytes(leaf, item);
2768 err = 0;
2769 if (start < extent_start || start >= extent_end) {
2770 em->start = start;
2771 if (start < extent_start) {
2772 if (start + len <= extent_start)
2773 goto not_found;
2774 em->len = extent_end - extent_start;
2775 } else {
2776 em->len = len;
2777 }
2778 goto not_found_em;
2779 }
2780 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
2781 if (bytenr == 0) {
2782 em->start = extent_start;
2783 em->len = extent_end - extent_start;
2784 em->block_start = EXTENT_MAP_HOLE;
2785 goto insert;
2786 }
2787 bytenr += btrfs_file_extent_offset(leaf, item);
2788 em->block_start = bytenr;
2789 em->start = extent_start;
2790 em->len = extent_end - extent_start;
2791 goto insert;
2792 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
2793 u64 page_start;
2794 unsigned long ptr;
2795 char *map;
2796 size_t size;
2797 size_t extent_offset;
2798 size_t copy_size;
2799
2800 size = btrfs_file_extent_inline_len(leaf, btrfs_item_nr(leaf,
2801 path->slots[0]));
2802 extent_end = (extent_start + size + root->sectorsize - 1) &
2803 ~((u64)root->sectorsize - 1);
2804 if (start < extent_start || start >= extent_end) {
2805 em->start = start;
2806 if (start < extent_start) {
2807 if (start + len <= extent_start)
2808 goto not_found;
2809 em->len = extent_end - extent_start;
2810 } else {
2811 em->len = len;
2812 }
2813 goto not_found_em;
2814 }
2815 em->block_start = EXTENT_MAP_INLINE;
2816
2817 if (!page) {
2818 em->start = extent_start;
2819 em->len = size;
2820 goto out;
2821 }
2822
2823 page_start = page_offset(page) + pg_offset;
2824 extent_offset = page_start - extent_start;
2825 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
2826 size - extent_offset);
2827 em->start = extent_start + extent_offset;
2828 em->len = (copy_size + root->sectorsize - 1) &
2829 ~((u64)root->sectorsize - 1);
2830 map = kmap(page);
2831 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
2832 if (create == 0 && !PageUptodate(page)) {
2833 read_extent_buffer(leaf, map + pg_offset, ptr,
2834 copy_size);
2835 flush_dcache_page(page);
2836 } else if (create && PageUptodate(page)) {
2837 if (!trans) {
2838 kunmap(page);
2839 free_extent_map(em);
2840 em = NULL;
2841 btrfs_release_path(root, path);
2842 trans = btrfs_join_transaction(root, 1);
2843 goto again;
2844 }
2845 write_extent_buffer(leaf, map + pg_offset, ptr,
2846 copy_size);
2847 btrfs_mark_buffer_dirty(leaf);
2848 }
2849 kunmap(page);
2850 set_extent_uptodate(io_tree, em->start,
2851 extent_map_end(em) - 1, GFP_NOFS);
2852 goto insert;
2853 } else {
2854 printk("unkknown found_type %d\n", found_type);
2855 WARN_ON(1);
2856 }
2857 not_found:
2858 em->start = start;
2859 em->len = len;
2860 not_found_em:
2861 em->block_start = EXTENT_MAP_HOLE;
2862 insert:
2863 btrfs_release_path(root, path);
2864 if (em->start > start || extent_map_end(em) <= start) {
2865 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
2866 err = -EIO;
2867 goto out;
2868 }
2869
2870 err = 0;
2871 spin_lock(&em_tree->lock);
2872 ret = add_extent_mapping(em_tree, em);
2873 /* it is possible that someone inserted the extent into the tree
2874 * while we had the lock dropped. It is also possible that
2875 * an overlapping map exists in the tree
2876 */
2877 if (ret == -EEXIST) {
2878 struct extent_map *existing;
2879
2880 ret = 0;
2881
2882 existing = lookup_extent_mapping(em_tree, start, len);
2883 if (existing && (existing->start > start ||
2884 existing->start + existing->len <= start)) {
2885 free_extent_map(existing);
2886 existing = NULL;
2887 }
2888 if (!existing) {
2889 existing = lookup_extent_mapping(em_tree, em->start,
2890 em->len);
2891 if (existing) {
2892 err = merge_extent_mapping(em_tree, existing,
2893 em, start,
2894 root->sectorsize);
2895 free_extent_map(existing);
2896 if (err) {
2897 free_extent_map(em);
2898 em = NULL;
2899 }
2900 } else {
2901 err = -EIO;
2902 printk("failing to insert %Lu %Lu\n",
2903 start, len);
2904 free_extent_map(em);
2905 em = NULL;
2906 }
2907 } else {
2908 free_extent_map(em);
2909 em = existing;
2910 err = 0;
2911 }
2912 }
2913 spin_unlock(&em_tree->lock);
2914 out:
2915 if (path)
2916 btrfs_free_path(path);
2917 if (trans) {
2918 ret = btrfs_end_transaction(trans, root);
2919 if (!err) {
2920 err = ret;
2921 }
2922 }
2923 if (err) {
2924 free_extent_map(em);
2925 WARN_ON(1);
2926 return ERR_PTR(err);
2927 }
2928 return em;
2929 }
2930
2931 #if 0 /* waiting for O_DIRECT reads */
2932 static int btrfs_get_block(struct inode *inode, sector_t iblock,
2933 struct buffer_head *bh_result, int create)
2934 {
2935 struct extent_map *em;
2936 u64 start = (u64)iblock << inode->i_blkbits;
2937 struct btrfs_multi_bio *multi = NULL;
2938 struct btrfs_root *root = BTRFS_I(inode)->root;
2939 u64 len;
2940 u64 logical;
2941 u64 map_length;
2942 int ret = 0;
2943
2944 em = btrfs_get_extent(inode, NULL, 0, start, bh_result->b_size, 0);
2945
2946 if (!em || IS_ERR(em))
2947 goto out;
2948
2949 if (em->start > start || em->start + em->len <= start) {
2950 goto out;
2951 }
2952
2953 if (em->block_start == EXTENT_MAP_INLINE) {
2954 ret = -EINVAL;
2955 goto out;
2956 }
2957
2958 len = em->start + em->len - start;
2959 len = min_t(u64, len, INT_LIMIT(typeof(bh_result->b_size)));
2960
2961 if (em->block_start == EXTENT_MAP_HOLE ||
2962 em->block_start == EXTENT_MAP_DELALLOC) {
2963 bh_result->b_size = len;
2964 goto out;
2965 }
2966
2967 logical = start - em->start;
2968 logical = em->block_start + logical;
2969
2970 map_length = len;
2971 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
2972 logical, &map_length, &multi, 0);
2973 BUG_ON(ret);
2974 bh_result->b_blocknr = multi->stripes[0].physical >> inode->i_blkbits;
2975 bh_result->b_size = min(map_length, len);
2976
2977 bh_result->b_bdev = multi->stripes[0].dev->bdev;
2978 set_buffer_mapped(bh_result);
2979 kfree(multi);
2980 out:
2981 free_extent_map(em);
2982 return ret;
2983 }
2984 #endif
2985
2986 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
2987 const struct iovec *iov, loff_t offset,
2988 unsigned long nr_segs)
2989 {
2990 return -EINVAL;
2991 #if 0
2992 struct file *file = iocb->ki_filp;
2993 struct inode *inode = file->f_mapping->host;
2994
2995 if (rw == WRITE)
2996 return -EINVAL;
2997
2998 return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
2999 offset, nr_segs, btrfs_get_block, NULL);
3000 #endif
3001 }
3002
3003 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
3004 {
3005 return extent_bmap(mapping, iblock, btrfs_get_extent);
3006 }
3007
3008 int btrfs_readpage(struct file *file, struct page *page)
3009 {
3010 struct extent_io_tree *tree;
3011 tree = &BTRFS_I(page->mapping->host)->io_tree;
3012 return extent_read_full_page(tree, page, btrfs_get_extent);
3013 }
3014
3015 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
3016 {
3017 struct extent_io_tree *tree;
3018
3019
3020 if (current->flags & PF_MEMALLOC) {
3021 redirty_page_for_writepage(wbc, page);
3022 unlock_page(page);
3023 return 0;
3024 }
3025 tree = &BTRFS_I(page->mapping->host)->io_tree;
3026 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
3027 }
3028
3029 int btrfs_writepages(struct address_space *mapping,
3030 struct writeback_control *wbc)
3031 {
3032 struct extent_io_tree *tree;
3033 tree = &BTRFS_I(mapping->host)->io_tree;
3034 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
3035 }
3036
3037 static int
3038 btrfs_readpages(struct file *file, struct address_space *mapping,
3039 struct list_head *pages, unsigned nr_pages)
3040 {
3041 struct extent_io_tree *tree;
3042 tree = &BTRFS_I(mapping->host)->io_tree;
3043 return extent_readpages(tree, mapping, pages, nr_pages,
3044 btrfs_get_extent);
3045 }
3046 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
3047 {
3048 struct extent_io_tree *tree;
3049 struct extent_map_tree *map;
3050 int ret;
3051
3052 tree = &BTRFS_I(page->mapping->host)->io_tree;
3053 map = &BTRFS_I(page->mapping->host)->extent_tree;
3054 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
3055 if (ret == 1) {
3056 ClearPagePrivate(page);
3057 set_page_private(page, 0);
3058 page_cache_release(page);
3059 }
3060 return ret;
3061 }
3062
3063 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
3064 {
3065 return __btrfs_releasepage(page, gfp_flags);
3066 }
3067
3068 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
3069 {
3070 struct extent_io_tree *tree;
3071 struct btrfs_ordered_extent *ordered;
3072 u64 page_start = page_offset(page);
3073 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
3074
3075 wait_on_page_writeback(page);
3076 tree = &BTRFS_I(page->mapping->host)->io_tree;
3077 if (offset) {
3078 btrfs_releasepage(page, GFP_NOFS);
3079 return;
3080 }
3081
3082 lock_extent(tree, page_start, page_end, GFP_NOFS);
3083 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
3084 page_offset(page));
3085 if (ordered) {
3086 /*
3087 * IO on this page will never be started, so we need
3088 * to account for any ordered extents now
3089 */
3090 clear_extent_bit(tree, page_start, page_end,
3091 EXTENT_DIRTY | EXTENT_DELALLOC |
3092 EXTENT_LOCKED, 1, 0, GFP_NOFS);
3093 btrfs_finish_ordered_io(page->mapping->host,
3094 page_start, page_end);
3095 btrfs_put_ordered_extent(ordered);
3096 lock_extent(tree, page_start, page_end, GFP_NOFS);
3097 }
3098 clear_extent_bit(tree, page_start, page_end,
3099 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
3100 EXTENT_ORDERED,
3101 1, 1, GFP_NOFS);
3102 __btrfs_releasepage(page, GFP_NOFS);
3103
3104 ClearPageChecked(page);
3105 if (PagePrivate(page)) {
3106 ClearPagePrivate(page);
3107 set_page_private(page, 0);
3108 page_cache_release(page);
3109 }
3110 }
3111
3112 /*
3113 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
3114 * called from a page fault handler when a page is first dirtied. Hence we must
3115 * be careful to check for EOF conditions here. We set the page up correctly
3116 * for a written page which means we get ENOSPC checking when writing into
3117 * holes and correct delalloc and unwritten extent mapping on filesystems that
3118 * support these features.
3119 *
3120 * We are not allowed to take the i_mutex here so we have to play games to
3121 * protect against truncate races as the page could now be beyond EOF. Because
3122 * vmtruncate() writes the inode size before removing pages, once we have the
3123 * page lock we can determine safely if the page is beyond EOF. If it is not
3124 * beyond EOF, then the page is guaranteed safe against truncation until we
3125 * unlock the page.
3126 */
3127 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
3128 {
3129 struct inode *inode = fdentry(vma->vm_file)->d_inode;
3130 struct btrfs_root *root = BTRFS_I(inode)->root;
3131 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3132 struct btrfs_ordered_extent *ordered;
3133 char *kaddr;
3134 unsigned long zero_start;
3135 loff_t size;
3136 int ret;
3137 u64 page_start;
3138 u64 page_end;
3139
3140 ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
3141 if (ret)
3142 goto out;
3143
3144 ret = -EINVAL;
3145 again:
3146 lock_page(page);
3147 size = i_size_read(inode);
3148 page_start = page_offset(page);
3149 page_end = page_start + PAGE_CACHE_SIZE - 1;
3150
3151 if ((page->mapping != inode->i_mapping) ||
3152 (page_start >= size)) {
3153 /* page got truncated out from underneath us */
3154 goto out_unlock;
3155 }
3156 wait_on_page_writeback(page);
3157
3158 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3159 set_page_extent_mapped(page);
3160
3161 /*
3162 * we can't set the delalloc bits if there are pending ordered
3163 * extents. Drop our locks and wait for them to finish
3164 */
3165 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3166 if (ordered) {
3167 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3168 unlock_page(page);
3169 btrfs_start_ordered_extent(inode, ordered, 1);
3170 btrfs_put_ordered_extent(ordered);
3171 goto again;
3172 }
3173
3174 btrfs_set_extent_delalloc(inode, page_start, page_end);
3175 ret = 0;
3176
3177 /* page is wholly or partially inside EOF */
3178 if (page_start + PAGE_CACHE_SIZE > size)
3179 zero_start = size & ~PAGE_CACHE_MASK;
3180 else
3181 zero_start = PAGE_CACHE_SIZE;
3182
3183 if (zero_start != PAGE_CACHE_SIZE) {
3184 kaddr = kmap(page);
3185 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
3186 flush_dcache_page(page);
3187 kunmap(page);
3188 }
3189 ClearPageChecked(page);
3190 set_page_dirty(page);
3191 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3192
3193 out_unlock:
3194 unlock_page(page);
3195 out:
3196 return ret;
3197 }
3198
3199 static void btrfs_truncate(struct inode *inode)
3200 {
3201 struct btrfs_root *root = BTRFS_I(inode)->root;
3202 int ret;
3203 struct btrfs_trans_handle *trans;
3204 unsigned long nr;
3205 u64 mask = root->sectorsize - 1;
3206
3207 if (!S_ISREG(inode->i_mode))
3208 return;
3209 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3210 return;
3211
3212 btrfs_truncate_page(inode->i_mapping, inode->i_size);
3213 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
3214
3215 trans = btrfs_start_transaction(root, 1);
3216 btrfs_set_trans_block_group(trans, inode);
3217 btrfs_i_size_write(inode, inode->i_size);
3218
3219 ret = btrfs_orphan_add(trans, inode);
3220 if (ret)
3221 goto out;
3222 /* FIXME, add redo link to tree so we don't leak on crash */
3223 ret = btrfs_truncate_in_trans(trans, root, inode,
3224 BTRFS_EXTENT_DATA_KEY);
3225 btrfs_update_inode(trans, root, inode);
3226
3227 ret = btrfs_orphan_del(trans, inode);
3228 BUG_ON(ret);
3229
3230 out:
3231 nr = trans->blocks_used;
3232 ret = btrfs_end_transaction_throttle(trans, root);
3233 BUG_ON(ret);
3234 btrfs_btree_balance_dirty(root, nr);
3235 }
3236
3237 /*
3238 * Invalidate a single dcache entry at the root of the filesystem.
3239 * Needed after creation of snapshot or subvolume.
3240 */
3241 void btrfs_invalidate_dcache_root(struct btrfs_root *root, char *name,
3242 int namelen)
3243 {
3244 struct dentry *alias, *entry;
3245 struct qstr qstr;
3246
3247 alias = d_find_alias(root->fs_info->sb->s_root->d_inode);
3248 if (alias) {
3249 qstr.name = name;
3250 qstr.len = namelen;
3251 /* change me if btrfs ever gets a d_hash operation */
3252 qstr.hash = full_name_hash(qstr.name, qstr.len);
3253 entry = d_lookup(alias, &qstr);
3254 dput(alias);
3255 if (entry) {
3256 d_invalidate(entry);
3257 dput(entry);
3258 }
3259 }
3260 }
3261
3262 int btrfs_create_subvol_root(struct btrfs_root *new_root,
3263 struct btrfs_trans_handle *trans, u64 new_dirid,
3264 struct btrfs_block_group_cache *block_group)
3265 {
3266 struct inode *inode;
3267 u64 index = 0;
3268
3269 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
3270 new_dirid, block_group, S_IFDIR | 0700, &index);
3271 if (IS_ERR(inode))
3272 return PTR_ERR(inode);
3273 inode->i_op = &btrfs_dir_inode_operations;
3274 inode->i_fop = &btrfs_dir_file_operations;
3275 new_root->inode = inode;
3276
3277 inode->i_nlink = 1;
3278 btrfs_i_size_write(inode, 0);
3279
3280 return btrfs_update_inode(trans, new_root, inode);
3281 }
3282
3283 unsigned long btrfs_force_ra(struct address_space *mapping,
3284 struct file_ra_state *ra, struct file *file,
3285 pgoff_t offset, pgoff_t last_index)
3286 {
3287 pgoff_t req_size = last_index - offset + 1;
3288
3289 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
3290 offset = page_cache_readahead(mapping, ra, file, offset, req_size);
3291 return offset;
3292 #else
3293 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
3294 return offset + req_size;
3295 #endif
3296 }
3297
3298 struct inode *btrfs_alloc_inode(struct super_block *sb)
3299 {
3300 struct btrfs_inode *ei;
3301
3302 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
3303 if (!ei)
3304 return NULL;
3305 ei->last_trans = 0;
3306 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
3307 ei->i_acl = BTRFS_ACL_NOT_CACHED;
3308 ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
3309 INIT_LIST_HEAD(&ei->i_orphan);
3310 return &ei->vfs_inode;
3311 }
3312
3313 void btrfs_destroy_inode(struct inode *inode)
3314 {
3315 struct btrfs_ordered_extent *ordered;
3316 WARN_ON(!list_empty(&inode->i_dentry));
3317 WARN_ON(inode->i_data.nrpages);
3318
3319 if (BTRFS_I(inode)->i_acl &&
3320 BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
3321 posix_acl_release(BTRFS_I(inode)->i_acl);
3322 if (BTRFS_I(inode)->i_default_acl &&
3323 BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
3324 posix_acl_release(BTRFS_I(inode)->i_default_acl);
3325
3326 spin_lock(&BTRFS_I(inode)->root->list_lock);
3327 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
3328 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
3329 " list\n", inode->i_ino);
3330 dump_stack();
3331 }
3332 spin_unlock(&BTRFS_I(inode)->root->list_lock);
3333
3334 while(1) {
3335 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
3336 if (!ordered)
3337 break;
3338 else {
3339 printk("found ordered extent %Lu %Lu\n",
3340 ordered->file_offset, ordered->len);
3341 btrfs_remove_ordered_extent(inode, ordered);
3342 btrfs_put_ordered_extent(ordered);
3343 btrfs_put_ordered_extent(ordered);
3344 }
3345 }
3346 btrfs_drop_extent_cache(inode, 0, (u64)-1);
3347 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
3348 }
3349
3350 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,26)
3351 static void init_once(void *foo)
3352 #elif LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
3353 static void init_once(struct kmem_cache * cachep, void *foo)
3354 #else
3355 static void init_once(void * foo, struct kmem_cache * cachep,
3356 unsigned long flags)
3357 #endif
3358 {
3359 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
3360
3361 inode_init_once(&ei->vfs_inode);
3362 }
3363
3364 void btrfs_destroy_cachep(void)
3365 {
3366 if (btrfs_inode_cachep)
3367 kmem_cache_destroy(btrfs_inode_cachep);
3368 if (btrfs_trans_handle_cachep)
3369 kmem_cache_destroy(btrfs_trans_handle_cachep);
3370 if (btrfs_transaction_cachep)
3371 kmem_cache_destroy(btrfs_transaction_cachep);
3372 if (btrfs_bit_radix_cachep)
3373 kmem_cache_destroy(btrfs_bit_radix_cachep);
3374 if (btrfs_path_cachep)
3375 kmem_cache_destroy(btrfs_path_cachep);
3376 }
3377
3378 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
3379 unsigned long extra_flags,
3380 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,26)
3381 void (*ctor)(void *)
3382 #elif LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
3383 void (*ctor)(struct kmem_cache *, void *)
3384 #else
3385 void (*ctor)(void *, struct kmem_cache *,
3386 unsigned long)
3387 #endif
3388 )
3389 {
3390 return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
3391 SLAB_MEM_SPREAD | extra_flags), ctor
3392 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
3393 ,NULL
3394 #endif
3395 );
3396 }
3397
3398 int btrfs_init_cachep(void)
3399 {
3400 btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
3401 sizeof(struct btrfs_inode),
3402 0, init_once);
3403 if (!btrfs_inode_cachep)
3404 goto fail;
3405 btrfs_trans_handle_cachep =
3406 btrfs_cache_create("btrfs_trans_handle_cache",
3407 sizeof(struct btrfs_trans_handle),
3408 0, NULL);
3409 if (!btrfs_trans_handle_cachep)
3410 goto fail;
3411 btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
3412 sizeof(struct btrfs_transaction),
3413 0, NULL);
3414 if (!btrfs_transaction_cachep)
3415 goto fail;
3416 btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
3417 sizeof(struct btrfs_path),
3418 0, NULL);
3419 if (!btrfs_path_cachep)
3420 goto fail;
3421 btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
3422 SLAB_DESTROY_BY_RCU, NULL);
3423 if (!btrfs_bit_radix_cachep)
3424 goto fail;
3425 return 0;
3426 fail:
3427 btrfs_destroy_cachep();
3428 return -ENOMEM;
3429 }
3430
3431 static int btrfs_getattr(struct vfsmount *mnt,
3432 struct dentry *dentry, struct kstat *stat)
3433 {
3434 struct inode *inode = dentry->d_inode;
3435 generic_fillattr(inode, stat);
3436 stat->blksize = PAGE_CACHE_SIZE;
3437 stat->blocks = inode->i_blocks + (BTRFS_I(inode)->delalloc_bytes >> 9);
3438 return 0;
3439 }
3440
3441 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
3442 struct inode * new_dir,struct dentry *new_dentry)
3443 {
3444 struct btrfs_trans_handle *trans;
3445 struct btrfs_root *root = BTRFS_I(old_dir)->root;
3446 struct inode *new_inode = new_dentry->d_inode;
3447 struct inode *old_inode = old_dentry->d_inode;
3448 struct timespec ctime = CURRENT_TIME;
3449 u64 index = 0;
3450 int ret;
3451
3452 if (S_ISDIR(old_inode->i_mode) && new_inode &&
3453 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
3454 return -ENOTEMPTY;
3455 }
3456
3457 ret = btrfs_check_free_space(root, 1, 0);
3458 if (ret)
3459 goto out_unlock;
3460
3461 trans = btrfs_start_transaction(root, 1);
3462
3463 btrfs_set_trans_block_group(trans, new_dir);
3464
3465 old_dentry->d_inode->i_nlink++;
3466 old_dir->i_ctime = old_dir->i_mtime = ctime;
3467 new_dir->i_ctime = new_dir->i_mtime = ctime;
3468 old_inode->i_ctime = ctime;
3469
3470 ret = btrfs_unlink_trans(trans, root, old_dir, old_dentry);
3471 if (ret)
3472 goto out_fail;
3473
3474 if (new_inode) {
3475 new_inode->i_ctime = CURRENT_TIME;
3476 ret = btrfs_unlink_trans(trans, root, new_dir, new_dentry);
3477 if (ret)
3478 goto out_fail;
3479 if (new_inode->i_nlink == 0) {
3480 ret = btrfs_orphan_add(trans, new_inode);
3481 if (ret)
3482 goto out_fail;
3483 }
3484 }
3485 ret = btrfs_set_inode_index(new_dir, old_inode, &index);
3486 if (ret)
3487 goto out_fail;
3488
3489 ret = btrfs_add_link(trans, new_dentry, old_inode, 1, index);
3490 if (ret)
3491 goto out_fail;
3492
3493 out_fail:
3494 btrfs_end_transaction_throttle(trans, root);
3495 out_unlock:
3496 return ret;
3497 }
3498
3499 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
3500 {
3501 struct list_head *head = &root->fs_info->delalloc_inodes;
3502 struct btrfs_inode *binode;
3503 unsigned long flags;
3504
3505 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
3506 while(!list_empty(head)) {
3507 binode = list_entry(head->next, struct btrfs_inode,
3508 delalloc_inodes);
3509 atomic_inc(&binode->vfs_inode.i_count);
3510 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
3511 filemap_write_and_wait(binode->vfs_inode.i_mapping);
3512 iput(&binode->vfs_inode);
3513 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
3514 }
3515 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
3516 return 0;
3517 }
3518
3519 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
3520 const char *symname)
3521 {
3522 struct btrfs_trans_handle *trans;
3523 struct btrfs_root *root = BTRFS_I(dir)->root;
3524 struct btrfs_path *path;
3525 struct btrfs_key key;
3526 struct inode *inode = NULL;
3527 int err;
3528 int drop_inode = 0;
3529 u64 objectid;
3530 u64 index = 0 ;
3531 int name_len;
3532 int datasize;
3533 unsigned long ptr;
3534 struct btrfs_file_extent_item *ei;
3535 struct extent_buffer *leaf;
3536 unsigned long nr = 0;
3537
3538 name_len = strlen(symname) + 1;
3539 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
3540 return -ENAMETOOLONG;
3541
3542 err = btrfs_check_free_space(root, 1, 0);
3543 if (err)
3544 goto out_fail;
3545
3546 trans = btrfs_start_transaction(root, 1);
3547 btrfs_set_trans_block_group(trans, dir);
3548
3549 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3550 if (err) {
3551 err = -ENOSPC;
3552 goto out_unlock;
3553 }
3554
3555 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3556 dentry->d_name.len,
3557 dentry->d_parent->d_inode->i_ino, objectid,
3558 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
3559 &index);
3560 err = PTR_ERR(inode);
3561 if (IS_ERR(inode))
3562 goto out_unlock;
3563
3564 err = btrfs_init_acl(inode, dir);
3565 if (err) {
3566 drop_inode = 1;
3567 goto out_unlock;
3568 }
3569
3570 btrfs_set_trans_block_group(trans, inode);
3571 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3572 if (err)
3573 drop_inode = 1;
3574 else {
3575 inode->i_mapping->a_ops = &btrfs_aops;
3576 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3577 inode->i_fop = &btrfs_file_operations;
3578 inode->i_op = &btrfs_file_inode_operations;
3579 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3580 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
3581 inode->i_mapping, GFP_NOFS);
3582 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3583 inode->i_mapping, GFP_NOFS);
3584 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
3585 mutex_init(&BTRFS_I(inode)->csum_mutex);
3586 mutex_init(&BTRFS_I(inode)->extent_mutex);
3587 BTRFS_I(inode)->delalloc_bytes = 0;
3588 BTRFS_I(inode)->disk_i_size = 0;
3589 inode->i_mapping->writeback_index = 0;
3590 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3591 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
3592 }
3593 dir->i_sb->s_dirt = 1;
3594 btrfs_update_inode_block_group(trans, inode);
3595 btrfs_update_inode_block_group(trans, dir);
3596 if (drop_inode)
3597 goto out_unlock;
3598
3599 path = btrfs_alloc_path();
3600 BUG_ON(!path);
3601 key.objectid = inode->i_ino;
3602 key.offset = 0;
3603 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
3604 datasize = btrfs_file_extent_calc_inline_size(name_len);
3605 err = btrfs_insert_empty_item(trans, root, path, &key,
3606 datasize);
3607 if (err) {
3608 drop_inode = 1;
3609 goto out_unlock;
3610 }
3611 leaf = path->nodes[0];
3612 ei = btrfs_item_ptr(leaf, path->slots[0],
3613 struct btrfs_file_extent_item);
3614 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
3615 btrfs_set_file_extent_type(leaf, ei,
3616 BTRFS_FILE_EXTENT_INLINE);
3617 ptr = btrfs_file_extent_inline_start(ei);
3618 write_extent_buffer(leaf, symname, ptr, name_len);
3619 btrfs_mark_buffer_dirty(leaf);
3620 btrfs_free_path(path);
3621
3622 inode->i_op = &btrfs_symlink_inode_operations;
3623 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3624 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3625 btrfs_i_size_write(inode, name_len - 1);
3626 err = btrfs_update_inode(trans, root, inode);
3627 if (err)
3628 drop_inode = 1;
3629
3630 out_unlock:
3631 nr = trans->blocks_used;
3632 btrfs_end_transaction_throttle(trans, root);
3633 out_fail:
3634 if (drop_inode) {
3635 inode_dec_link_count(inode);
3636 iput(inode);
3637 }
3638 btrfs_btree_balance_dirty(root, nr);
3639 return err;
3640 }
3641
3642 static int btrfs_set_page_dirty(struct page *page)
3643 {
3644 return __set_page_dirty_nobuffers(page);
3645 }
3646
3647 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,26)
3648 static int btrfs_permission(struct inode *inode, int mask)
3649 #else
3650 static int btrfs_permission(struct inode *inode, int mask,
3651 struct nameidata *nd)
3652 #endif
3653 {
3654 if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
3655 return -EACCES;
3656 return generic_permission(inode, mask, btrfs_check_acl);
3657 }
3658
3659 static struct inode_operations btrfs_dir_inode_operations = {
3660 .lookup = btrfs_lookup,
3661 .create = btrfs_create,
3662 .unlink = btrfs_unlink,
3663 .link = btrfs_link,
3664 .mkdir = btrfs_mkdir,
3665 .rmdir = btrfs_rmdir,
3666 .rename = btrfs_rename,
3667 .symlink = btrfs_symlink,
3668 .setattr = btrfs_setattr,
3669 .mknod = btrfs_mknod,
3670 .setxattr = generic_setxattr,
3671 .getxattr = generic_getxattr,
3672 .listxattr = btrfs_listxattr,
3673 .removexattr = generic_removexattr,
3674 .permission = btrfs_permission,
3675 };
3676 static struct inode_operations btrfs_dir_ro_inode_operations = {
3677 .lookup = btrfs_lookup,
3678 .permission = btrfs_permission,
3679 };
3680 static struct file_operations btrfs_dir_file_operations = {
3681 .llseek = generic_file_llseek,
3682 .read = generic_read_dir,
3683 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,28)
3684 .readdir = btrfs_nfshack_readdir,
3685 #else /* NFSd readdir/lookup deadlock is fixed */
3686 .readdir = btrfs_real_readdir,
3687 #endif
3688 .unlocked_ioctl = btrfs_ioctl,
3689 #ifdef CONFIG_COMPAT
3690 .compat_ioctl = btrfs_ioctl,
3691 #endif
3692 .release = btrfs_release_file,
3693 };
3694
3695 static struct extent_io_ops btrfs_extent_io_ops = {
3696 .fill_delalloc = run_delalloc_range,
3697 .submit_bio_hook = btrfs_submit_bio_hook,
3698 .merge_bio_hook = btrfs_merge_bio_hook,
3699 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
3700 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
3701 .writepage_start_hook = btrfs_writepage_start_hook,
3702 .readpage_io_failed_hook = btrfs_io_failed_hook,
3703 .set_bit_hook = btrfs_set_bit_hook,
3704 .clear_bit_hook = btrfs_clear_bit_hook,
3705 };
3706
3707 static struct address_space_operations btrfs_aops = {
3708 .readpage = btrfs_readpage,
3709 .writepage = btrfs_writepage,
3710 .writepages = btrfs_writepages,
3711 .readpages = btrfs_readpages,
3712 .sync_page = block_sync_page,
3713 .bmap = btrfs_bmap,
3714 .direct_IO = btrfs_direct_IO,
3715 .invalidatepage = btrfs_invalidatepage,
3716 .releasepage = btrfs_releasepage,
3717 .set_page_dirty = btrfs_set_page_dirty,
3718 };
3719
3720 static struct address_space_operations btrfs_symlink_aops = {
3721 .readpage = btrfs_readpage,
3722 .writepage = btrfs_writepage,
3723 .invalidatepage = btrfs_invalidatepage,
3724 .releasepage = btrfs_releasepage,
3725 };
3726
3727 static struct inode_operations btrfs_file_inode_operations = {
3728 .truncate = btrfs_truncate,
3729 .getattr = btrfs_getattr,
3730 .setattr = btrfs_setattr,
3731 .setxattr = generic_setxattr,
3732 .getxattr = generic_getxattr,
3733 .listxattr = btrfs_listxattr,
3734 .removexattr = generic_removexattr,
3735 .permission = btrfs_permission,
3736 };
3737 static struct inode_operations btrfs_special_inode_operations = {
3738 .getattr = btrfs_getattr,
3739 .setattr = btrfs_setattr,
3740 .permission = btrfs_permission,
3741 .setxattr = generic_setxattr,
3742 .getxattr = generic_getxattr,
3743 .listxattr = btrfs_listxattr,
3744 .removexattr = generic_removexattr,
3745 };
3746 static struct inode_operations btrfs_symlink_inode_operations = {
3747 .readlink = generic_readlink,
3748 .follow_link = page_follow_link_light,
3749 .put_link = page_put_link,
3750 .permission = btrfs_permission,
3751 };
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree