Fix btrfs_del_ordered_inode to allow forcing the drop during unlinks
[linux-3.10.git] / fs / btrfs / inode.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/smp_lock.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mpage.h>
32 #include <linux/swap.h>
33 #include <linux/writeback.h>
34 #include <linux/statfs.h>
35 #include <linux/compat.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/version.h>
38 #include <linux/xattr.h>
39 #include "ctree.h"
40 #include "disk-io.h"
41 #include "transaction.h"
42 #include "btrfs_inode.h"
43 #include "ioctl.h"
44 #include "print-tree.h"
45 #include "volumes.h"
46
47 struct btrfs_iget_args {
48         u64 ino;
49         struct btrfs_root *root;
50 };
51
52 static struct inode_operations btrfs_dir_inode_operations;
53 static struct inode_operations btrfs_symlink_inode_operations;
54 static struct inode_operations btrfs_dir_ro_inode_operations;
55 static struct inode_operations btrfs_special_inode_operations;
56 static struct inode_operations btrfs_file_inode_operations;
57 static struct address_space_operations btrfs_aops;
58 static struct address_space_operations btrfs_symlink_aops;
59 static struct file_operations btrfs_dir_file_operations;
60 static struct extent_io_ops btrfs_extent_io_ops;
61
62 static struct kmem_cache *btrfs_inode_cachep;
63 struct kmem_cache *btrfs_trans_handle_cachep;
64 struct kmem_cache *btrfs_transaction_cachep;
65 struct kmem_cache *btrfs_bit_radix_cachep;
66 struct kmem_cache *btrfs_path_cachep;
67
68 #define S_SHIFT 12
69 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
70         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
71         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
72         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
73         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
74         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
75         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
76         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
77 };
78
79 int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
80                            int for_del)
81 {
82         u64 total;
83         u64 used;
84         u64 thresh;
85         unsigned long flags;
86         int ret = 0;
87
88         spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
89         total = btrfs_super_total_bytes(&root->fs_info->super_copy);
90         used = btrfs_super_bytes_used(&root->fs_info->super_copy);
91         if (for_del)
92                 thresh = total * 90;
93         else
94                 thresh = total * 85;
95
96         do_div(thresh, 100);
97
98         if (used + root->fs_info->delalloc_bytes + num_required > thresh)
99                 ret = -ENOSPC;
100         spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
101         return ret;
102 }
103
104 static int cow_file_range(struct inode *inode, u64 start, u64 end)
105 {
106         struct btrfs_root *root = BTRFS_I(inode)->root;
107         struct btrfs_trans_handle *trans;
108         u64 alloc_hint = 0;
109         u64 num_bytes;
110         u64 cur_alloc_size;
111         u64 blocksize = root->sectorsize;
112         u64 orig_start = start;
113         u64 orig_num_bytes;
114         struct btrfs_key ins;
115         int ret;
116
117         trans = btrfs_start_transaction(root, 1);
118         BUG_ON(!trans);
119         btrfs_set_trans_block_group(trans, inode);
120
121         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
122         num_bytes = max(blocksize,  num_bytes);
123         ret = btrfs_drop_extents(trans, root, inode,
124                                  start, start + num_bytes, start, &alloc_hint);
125         orig_num_bytes = num_bytes;
126
127         if (alloc_hint == EXTENT_MAP_INLINE)
128                 goto out;
129
130         BUG_ON(num_bytes > btrfs_super_total_bytes(&root->fs_info->super_copy));
131
132         while(num_bytes > 0) {
133                 cur_alloc_size = min(num_bytes, root->fs_info->max_extent);
134                 ret = btrfs_alloc_extent(trans, root, cur_alloc_size,
135                                          root->sectorsize,
136                                          root->root_key.objectid,
137                                          trans->transid,
138                                          inode->i_ino, start, 0,
139                                          alloc_hint, (u64)-1, &ins, 1);
140                 if (ret) {
141                         WARN_ON(1);
142                         goto out;
143                 }
144                 cur_alloc_size = ins.offset;
145                 ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
146                                                start, ins.objectid, ins.offset,
147                                                ins.offset, 0);
148                 inode->i_blocks += ins.offset >> 9;
149                 btrfs_check_file(root, inode);
150                 if (num_bytes < cur_alloc_size) {
151                         printk("num_bytes %Lu cur_alloc %Lu\n", num_bytes,
152                                cur_alloc_size);
153                         break;
154                 }
155                 num_bytes -= cur_alloc_size;
156                 alloc_hint = ins.objectid + ins.offset;
157                 start += cur_alloc_size;
158         }
159         btrfs_drop_extent_cache(inode, orig_start,
160                                 orig_start + orig_num_bytes - 1);
161         btrfs_add_ordered_inode(inode);
162         btrfs_update_inode(trans, root, inode);
163 out:
164         btrfs_end_transaction(trans, root);
165         return ret;
166 }
167
168 static int run_delalloc_nocow(struct inode *inode, u64 start, u64 end)
169 {
170         u64 extent_start;
171         u64 extent_end;
172         u64 bytenr;
173         u64 cow_end;
174         u64 loops = 0;
175         u64 total_fs_bytes;
176         struct btrfs_root *root = BTRFS_I(inode)->root;
177         struct btrfs_block_group_cache *block_group;
178         struct extent_buffer *leaf;
179         int found_type;
180         struct btrfs_path *path;
181         struct btrfs_file_extent_item *item;
182         int ret;
183         int err;
184         struct btrfs_key found_key;
185
186         total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
187         path = btrfs_alloc_path();
188         BUG_ON(!path);
189 again:
190         ret = btrfs_lookup_file_extent(NULL, root, path,
191                                        inode->i_ino, start, 0);
192         if (ret < 0) {
193                 btrfs_free_path(path);
194                 return ret;
195         }
196
197         cow_end = end;
198         if (ret != 0) {
199                 if (path->slots[0] == 0)
200                         goto not_found;
201                 path->slots[0]--;
202         }
203
204         leaf = path->nodes[0];
205         item = btrfs_item_ptr(leaf, path->slots[0],
206                               struct btrfs_file_extent_item);
207
208         /* are we inside the extent that was found? */
209         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
210         found_type = btrfs_key_type(&found_key);
211         if (found_key.objectid != inode->i_ino ||
212             found_type != BTRFS_EXTENT_DATA_KEY)
213                 goto not_found;
214
215         found_type = btrfs_file_extent_type(leaf, item);
216         extent_start = found_key.offset;
217         if (found_type == BTRFS_FILE_EXTENT_REG) {
218                 u64 extent_num_bytes;
219
220                 extent_num_bytes = btrfs_file_extent_num_bytes(leaf, item);
221                 extent_end = extent_start + extent_num_bytes;
222                 err = 0;
223
224                 if (loops && start != extent_start)
225                         goto not_found;
226
227                 if (start < extent_start || start >= extent_end)
228                         goto not_found;
229
230                 cow_end = min(end, extent_end - 1);
231                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
232                 if (bytenr == 0)
233                         goto not_found;
234
235                 if (btrfs_count_snapshots_in_path(root, path, inode->i_ino,
236                                                   bytenr) != 1) {
237                         goto not_found;
238                 }
239
240                 /*
241                  * we may be called by the resizer, make sure we're inside
242                  * the limits of the FS
243                  */
244                 block_group = btrfs_lookup_block_group(root->fs_info,
245                                                        bytenr);
246                 if (!block_group || block_group->ro)
247                         goto not_found;
248
249                 start = extent_end;
250         } else {
251                 goto not_found;
252         }
253 loop:
254         if (start > end) {
255                 btrfs_free_path(path);
256                 return 0;
257         }
258         btrfs_release_path(root, path);
259         loops++;
260         goto again;
261
262 not_found:
263         cow_file_range(inode, start, end);
264         start = end + 1;
265         goto loop;
266 }
267
268 static int run_delalloc_range(struct inode *inode, u64 start, u64 end)
269 {
270         struct btrfs_root *root = BTRFS_I(inode)->root;
271         int ret;
272
273         if (btrfs_test_opt(root, NODATACOW) ||
274             btrfs_test_flag(inode, NODATACOW))
275                 ret = run_delalloc_nocow(inode, start, end);
276         else
277                 ret = cow_file_range(inode, start, end);
278
279         return ret;
280 }
281
282 int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
283                        unsigned long old, unsigned long bits)
284 {
285         unsigned long flags;
286         if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
287                 struct btrfs_root *root = BTRFS_I(inode)->root;
288                 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
289                 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
290                 root->fs_info->delalloc_bytes += end - start + 1;
291                 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
292         }
293         return 0;
294 }
295
296 int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
297                          unsigned long old, unsigned long bits)
298 {
299         if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
300                 struct btrfs_root *root = BTRFS_I(inode)->root;
301                 unsigned long flags;
302
303                 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
304                 if (end - start + 1 > root->fs_info->delalloc_bytes) {
305                         printk("warning: delalloc account %Lu %Lu\n",
306                                end - start + 1, root->fs_info->delalloc_bytes);
307                         root->fs_info->delalloc_bytes = 0;
308                         BTRFS_I(inode)->delalloc_bytes = 0;
309                 } else {
310                         root->fs_info->delalloc_bytes -= end - start + 1;
311                         BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
312                 }
313                 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
314         }
315         return 0;
316 }
317
318 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
319                          size_t size, struct bio *bio)
320 {
321         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
322         struct btrfs_mapping_tree *map_tree;
323         u64 logical = bio->bi_sector << 9;
324         u64 length = 0;
325         u64 map_length;
326         int ret;
327
328         length = bio->bi_size;
329         map_tree = &root->fs_info->mapping_tree;
330         map_length = length;
331         ret = btrfs_map_block(map_tree, READ, logical,
332                               &map_length, NULL, 0);
333
334         if (map_length < length + size) {
335                 return 1;
336         }
337         return 0;
338 }
339
340 int __btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
341                           int mirror_num)
342 {
343         struct btrfs_root *root = BTRFS_I(inode)->root;
344         struct btrfs_trans_handle *trans;
345         int ret = 0;
346         char *sums = NULL;
347
348         ret = btrfs_csum_one_bio(root, bio, &sums);
349         BUG_ON(ret);
350
351         trans = btrfs_start_transaction(root, 1);
352
353         btrfs_set_trans_block_group(trans, inode);
354         btrfs_csum_file_blocks(trans, root, inode, bio, sums);
355
356         ret = btrfs_end_transaction(trans, root);
357         BUG_ON(ret);
358
359         kfree(sums);
360
361         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
362 }
363
364 int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
365                           int mirror_num)
366 {
367         struct btrfs_root *root = BTRFS_I(inode)->root;
368         int ret = 0;
369
370         if (!(rw & (1 << BIO_RW))) {
371                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
372                 BUG_ON(ret);
373                 goto mapit;
374         }
375
376         if (btrfs_test_opt(root, NODATASUM) ||
377             btrfs_test_flag(inode, NODATASUM)) {
378                 goto mapit;
379         }
380
381         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
382                                    inode, rw, bio, mirror_num,
383                                    __btrfs_submit_bio_hook);
384 mapit:
385         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
386 }
387
388 int btrfs_readpage_io_hook(struct page *page, u64 start, u64 end)
389 {
390         int ret = 0;
391         struct inode *inode = page->mapping->host;
392         struct btrfs_root *root = BTRFS_I(inode)->root;
393         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
394         struct btrfs_csum_item *item;
395         struct btrfs_path *path = NULL;
396         u32 csum;
397
398         if (btrfs_test_opt(root, NODATASUM) ||
399             btrfs_test_flag(inode, NODATASUM))
400                 return 0;
401
402         path = btrfs_alloc_path();
403         item = btrfs_lookup_csum(NULL, root, path, inode->i_ino, start, 0);
404         if (IS_ERR(item)) {
405                 ret = PTR_ERR(item);
406                 /* a csum that isn't present is a preallocated region. */
407                 if (ret == -ENOENT || ret == -EFBIG)
408                         ret = 0;
409                 csum = 0;
410                 printk("no csum found for inode %lu start %Lu\n", inode->i_ino, start);
411                 goto out;
412         }
413         read_extent_buffer(path->nodes[0], &csum, (unsigned long)item,
414                            BTRFS_CRC32_SIZE);
415         set_state_private(io_tree, start, csum);
416 out:
417         if (path)
418                 btrfs_free_path(path);
419         return ret;
420 }
421
422 struct io_failure_record {
423         struct page *page;
424         u64 start;
425         u64 len;
426         u64 logical;
427         int last_mirror;
428 };
429
430 int btrfs_io_failed_hook(struct bio *failed_bio,
431                          struct page *page, u64 start, u64 end,
432                          struct extent_state *state)
433 {
434         struct io_failure_record *failrec = NULL;
435         u64 private;
436         struct extent_map *em;
437         struct inode *inode = page->mapping->host;
438         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
439         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
440         struct bio *bio;
441         int num_copies;
442         int ret;
443         int rw;
444         u64 logical;
445
446         ret = get_state_private(failure_tree, start, &private);
447         if (ret) {
448                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
449                 if (!failrec)
450                         return -ENOMEM;
451                 failrec->start = start;
452                 failrec->len = end - start + 1;
453                 failrec->last_mirror = 0;
454
455                 spin_lock(&em_tree->lock);
456                 em = lookup_extent_mapping(em_tree, start, failrec->len);
457                 if (em->start > start || em->start + em->len < start) {
458                         free_extent_map(em);
459                         em = NULL;
460                 }
461                 spin_unlock(&em_tree->lock);
462
463                 if (!em || IS_ERR(em)) {
464                         kfree(failrec);
465                         return -EIO;
466                 }
467                 logical = start - em->start;
468                 logical = em->block_start + logical;
469                 failrec->logical = logical;
470                 free_extent_map(em);
471                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
472                                 EXTENT_DIRTY, GFP_NOFS);
473                 set_state_private(failure_tree, start,
474                                  (u64)(unsigned long)failrec);
475         } else {
476                 failrec = (struct io_failure_record *)(unsigned long)private;
477         }
478         num_copies = btrfs_num_copies(
479                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
480                               failrec->logical, failrec->len);
481         failrec->last_mirror++;
482         if (!state) {
483                 spin_lock_irq(&BTRFS_I(inode)->io_tree.lock);
484                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
485                                                     failrec->start,
486                                                     EXTENT_LOCKED);
487                 if (state && state->start != failrec->start)
488                         state = NULL;
489                 spin_unlock_irq(&BTRFS_I(inode)->io_tree.lock);
490         }
491         if (!state || failrec->last_mirror > num_copies) {
492                 set_state_private(failure_tree, failrec->start, 0);
493                 clear_extent_bits(failure_tree, failrec->start,
494                                   failrec->start + failrec->len - 1,
495                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
496                 kfree(failrec);
497                 return -EIO;
498         }
499         bio = bio_alloc(GFP_NOFS, 1);
500         bio->bi_private = state;
501         bio->bi_end_io = failed_bio->bi_end_io;
502         bio->bi_sector = failrec->logical >> 9;
503         bio->bi_bdev = failed_bio->bi_bdev;
504         bio->bi_size = 0;
505         bio_add_page(bio, page, failrec->len, start - page_offset(page));
506         if (failed_bio->bi_rw & (1 << BIO_RW))
507                 rw = WRITE;
508         else
509                 rw = READ;
510
511         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
512                                                       failrec->last_mirror);
513         return 0;
514 }
515
516 int btrfs_clean_io_failures(struct inode *inode, u64 start)
517 {
518         u64 private;
519         u64 private_failure;
520         struct io_failure_record *failure;
521         int ret;
522
523         private = 0;
524         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
525                              (u64)-1, 1, EXTENT_DIRTY)) {
526                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
527                                         start, &private_failure);
528                 if (ret == 0) {
529                         failure = (struct io_failure_record *)(unsigned long)
530                                    private_failure;
531                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
532                                           failure->start, 0);
533                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
534                                           failure->start,
535                                           failure->start + failure->len - 1,
536                                           EXTENT_DIRTY | EXTENT_LOCKED,
537                                           GFP_NOFS);
538                         kfree(failure);
539                 }
540         }
541         return 0;
542 }
543
544 int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
545                                struct extent_state *state)
546 {
547         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
548         struct inode *inode = page->mapping->host;
549         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
550         char *kaddr;
551         u64 private = ~(u32)0;
552         int ret;
553         struct btrfs_root *root = BTRFS_I(inode)->root;
554         u32 csum = ~(u32)0;
555         unsigned long flags;
556
557         if (btrfs_test_opt(root, NODATASUM) ||
558             btrfs_test_flag(inode, NODATASUM))
559                 return 0;
560         if (state && state->start == start) {
561                 private = state->private;
562                 ret = 0;
563         } else {
564                 ret = get_state_private(io_tree, start, &private);
565         }
566         local_irq_save(flags);
567         kaddr = kmap_atomic(page, KM_IRQ0);
568         if (ret) {
569                 goto zeroit;
570         }
571         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
572         btrfs_csum_final(csum, (char *)&csum);
573         if (csum != private) {
574                 goto zeroit;
575         }
576         kunmap_atomic(kaddr, KM_IRQ0);
577         local_irq_restore(flags);
578
579         /* if the io failure tree for this inode is non-empty,
580          * check to see if we've recovered from a failed IO
581          */
582         btrfs_clean_io_failures(inode, start);
583         return 0;
584
585 zeroit:
586         printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
587                page->mapping->host->i_ino, (unsigned long long)start, csum,
588                private);
589         memset(kaddr + offset, 1, end - start + 1);
590         flush_dcache_page(page);
591         kunmap_atomic(kaddr, KM_IRQ0);
592         local_irq_restore(flags);
593         if (private == 0)
594                 return 0;
595         return -EIO;
596 }
597
598 void btrfs_read_locked_inode(struct inode *inode)
599 {
600         struct btrfs_path *path;
601         struct extent_buffer *leaf;
602         struct btrfs_inode_item *inode_item;
603         struct btrfs_timespec *tspec;
604         struct btrfs_root *root = BTRFS_I(inode)->root;
605         struct btrfs_key location;
606         u64 alloc_group_block;
607         u32 rdev;
608         int ret;
609
610         path = btrfs_alloc_path();
611         BUG_ON(!path);
612         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
613
614         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
615         if (ret)
616                 goto make_bad;
617
618         leaf = path->nodes[0];
619         inode_item = btrfs_item_ptr(leaf, path->slots[0],
620                                     struct btrfs_inode_item);
621
622         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
623         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
624         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
625         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
626         inode->i_size = btrfs_inode_size(leaf, inode_item);
627
628         tspec = btrfs_inode_atime(inode_item);
629         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
630         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
631
632         tspec = btrfs_inode_mtime(inode_item);
633         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
634         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
635
636         tspec = btrfs_inode_ctime(inode_item);
637         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
638         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
639
640         inode->i_blocks = btrfs_inode_nblocks(leaf, inode_item);
641         inode->i_generation = btrfs_inode_generation(leaf, inode_item);
642         inode->i_rdev = 0;
643         rdev = btrfs_inode_rdev(leaf, inode_item);
644
645         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
646         BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
647                                                        alloc_group_block);
648         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
649         if (!BTRFS_I(inode)->block_group) {
650                 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
651                                                  NULL, 0,
652                                                  BTRFS_BLOCK_GROUP_METADATA, 0);
653         }
654         btrfs_free_path(path);
655         inode_item = NULL;
656
657         switch (inode->i_mode & S_IFMT) {
658         case S_IFREG:
659                 inode->i_mapping->a_ops = &btrfs_aops;
660                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
661                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
662                 inode->i_fop = &btrfs_file_operations;
663                 inode->i_op = &btrfs_file_inode_operations;
664                 break;
665         case S_IFDIR:
666                 inode->i_fop = &btrfs_dir_file_operations;
667                 if (root == root->fs_info->tree_root)
668                         inode->i_op = &btrfs_dir_ro_inode_operations;
669                 else
670                         inode->i_op = &btrfs_dir_inode_operations;
671                 break;
672         case S_IFLNK:
673                 inode->i_op = &btrfs_symlink_inode_operations;
674                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
675                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
676                 break;
677         default:
678                 init_special_inode(inode, inode->i_mode, rdev);
679                 break;
680         }
681         return;
682
683 make_bad:
684         btrfs_free_path(path);
685         make_bad_inode(inode);
686 }
687
688 static void fill_inode_item(struct extent_buffer *leaf,
689                             struct btrfs_inode_item *item,
690                             struct inode *inode)
691 {
692         btrfs_set_inode_uid(leaf, item, inode->i_uid);
693         btrfs_set_inode_gid(leaf, item, inode->i_gid);
694         btrfs_set_inode_size(leaf, item, inode->i_size);
695         btrfs_set_inode_mode(leaf, item, inode->i_mode);
696         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
697
698         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
699                                inode->i_atime.tv_sec);
700         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
701                                 inode->i_atime.tv_nsec);
702
703         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
704                                inode->i_mtime.tv_sec);
705         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
706                                 inode->i_mtime.tv_nsec);
707
708         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
709                                inode->i_ctime.tv_sec);
710         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
711                                 inode->i_ctime.tv_nsec);
712
713         btrfs_set_inode_nblocks(leaf, item, inode->i_blocks);
714         btrfs_set_inode_generation(leaf, item, inode->i_generation);
715         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
716         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
717         btrfs_set_inode_block_group(leaf, item,
718                                     BTRFS_I(inode)->block_group->key.objectid);
719 }
720
721 int btrfs_update_inode(struct btrfs_trans_handle *trans,
722                               struct btrfs_root *root,
723                               struct inode *inode)
724 {
725         struct btrfs_inode_item *inode_item;
726         struct btrfs_path *path;
727         struct extent_buffer *leaf;
728         int ret;
729
730         path = btrfs_alloc_path();
731         BUG_ON(!path);
732         ret = btrfs_lookup_inode(trans, root, path,
733                                  &BTRFS_I(inode)->location, 1);
734         if (ret) {
735                 if (ret > 0)
736                         ret = -ENOENT;
737                 goto failed;
738         }
739
740         leaf = path->nodes[0];
741         inode_item = btrfs_item_ptr(leaf, path->slots[0],
742                                   struct btrfs_inode_item);
743
744         fill_inode_item(leaf, inode_item, inode);
745         btrfs_mark_buffer_dirty(leaf);
746         btrfs_set_inode_last_trans(trans, inode);
747         ret = 0;
748 failed:
749         btrfs_free_path(path);
750         return ret;
751 }
752
753
754 static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
755                               struct btrfs_root *root,
756                               struct inode *dir,
757                               struct dentry *dentry)
758 {
759         struct btrfs_path *path;
760         const char *name = dentry->d_name.name;
761         int name_len = dentry->d_name.len;
762         int ret = 0;
763         struct extent_buffer *leaf;
764         struct btrfs_dir_item *di;
765         struct btrfs_key key;
766
767         path = btrfs_alloc_path();
768         if (!path) {
769                 ret = -ENOMEM;
770                 goto err;
771         }
772
773         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
774                                     name, name_len, -1);
775         if (IS_ERR(di)) {
776                 ret = PTR_ERR(di);
777                 goto err;
778         }
779         if (!di) {
780                 ret = -ENOENT;
781                 goto err;
782         }
783         leaf = path->nodes[0];
784         btrfs_dir_item_key_to_cpu(leaf, di, &key);
785         ret = btrfs_delete_one_dir_name(trans, root, path, di);
786         if (ret)
787                 goto err;
788         btrfs_release_path(root, path);
789
790         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
791                                          key.objectid, name, name_len, -1);
792         if (IS_ERR(di)) {
793                 ret = PTR_ERR(di);
794                 goto err;
795         }
796         if (!di) {
797                 ret = -ENOENT;
798                 goto err;
799         }
800         ret = btrfs_delete_one_dir_name(trans, root, path, di);
801         btrfs_release_path(root, path);
802
803         dentry->d_inode->i_ctime = dir->i_ctime;
804         ret = btrfs_del_inode_ref(trans, root, name, name_len,
805                                   dentry->d_inode->i_ino,
806                                   dentry->d_parent->d_inode->i_ino);
807         if (ret) {
808                 printk("failed to delete reference to %.*s, "
809                        "inode %lu parent %lu\n", name_len, name,
810                        dentry->d_inode->i_ino,
811                        dentry->d_parent->d_inode->i_ino);
812         }
813 err:
814         btrfs_free_path(path);
815         if (!ret) {
816                 dir->i_size -= name_len * 2;
817                 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
818                 btrfs_update_inode(trans, root, dir);
819 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
820                 dentry->d_inode->i_nlink--;
821 #else
822                 drop_nlink(dentry->d_inode);
823 #endif
824                 ret = btrfs_update_inode(trans, root, dentry->d_inode);
825                 dir->i_sb->s_dirt = 1;
826         }
827         return ret;
828 }
829
830 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
831 {
832         struct btrfs_root *root;
833         struct btrfs_trans_handle *trans;
834         struct inode *inode = dentry->d_inode;
835         int ret;
836         unsigned long nr = 0;
837
838         root = BTRFS_I(dir)->root;
839
840         ret = btrfs_check_free_space(root, 1, 1);
841         if (ret)
842                 goto fail;
843
844         trans = btrfs_start_transaction(root, 1);
845
846         btrfs_set_trans_block_group(trans, dir);
847         ret = btrfs_unlink_trans(trans, root, dir, dentry);
848         nr = trans->blocks_used;
849
850         if (inode->i_nlink == 0) {
851                 /* if the inode isn't linked anywhere,
852                  * we don't need to worry about
853                  * data=ordered
854                  */
855                 btrfs_del_ordered_inode(inode, 1);
856         }
857
858         btrfs_end_transaction(trans, root);
859 fail:
860         btrfs_btree_balance_dirty(root, nr);
861         btrfs_throttle(root);
862         return ret;
863 }
864
865 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
866 {
867         struct inode *inode = dentry->d_inode;
868         int err = 0;
869         int ret;
870         struct btrfs_root *root = BTRFS_I(dir)->root;
871         struct btrfs_trans_handle *trans;
872         unsigned long nr = 0;
873
874         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
875                 return -ENOTEMPTY;
876         }
877
878         ret = btrfs_check_free_space(root, 1, 1);
879         if (ret)
880                 goto fail;
881
882         trans = btrfs_start_transaction(root, 1);
883         btrfs_set_trans_block_group(trans, dir);
884
885         /* now the directory is empty */
886         err = btrfs_unlink_trans(trans, root, dir, dentry);
887         if (!err) {
888                 inode->i_size = 0;
889         }
890
891         nr = trans->blocks_used;
892         ret = btrfs_end_transaction(trans, root);
893 fail:
894         btrfs_btree_balance_dirty(root, nr);
895         btrfs_throttle(root);
896
897         if (ret && !err)
898                 err = ret;
899         return err;
900 }
901
902 /*
903  * this can truncate away extent items, csum items and directory items.
904  * It starts at a high offset and removes keys until it can't find
905  * any higher than i_size.
906  *
907  * csum items that cross the new i_size are truncated to the new size
908  * as well.
909  */
910 static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
911                                    struct btrfs_root *root,
912                                    struct inode *inode,
913                                    u32 min_type)
914 {
915         int ret;
916         struct btrfs_path *path;
917         struct btrfs_key key;
918         struct btrfs_key found_key;
919         u32 found_type;
920         struct extent_buffer *leaf;
921         struct btrfs_file_extent_item *fi;
922         u64 extent_start = 0;
923         u64 extent_num_bytes = 0;
924         u64 item_end = 0;
925         u64 root_gen = 0;
926         u64 root_owner = 0;
927         int found_extent;
928         int del_item;
929         int pending_del_nr = 0;
930         int pending_del_slot = 0;
931         int extent_type = -1;
932         u64 mask = root->sectorsize - 1;
933
934         btrfs_drop_extent_cache(inode, inode->i_size & (~mask), (u64)-1);
935         path = btrfs_alloc_path();
936         path->reada = -1;
937         BUG_ON(!path);
938
939         /* FIXME, add redo link to tree so we don't leak on crash */
940         key.objectid = inode->i_ino;
941         key.offset = (u64)-1;
942         key.type = (u8)-1;
943
944         btrfs_init_path(path);
945 search_again:
946         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
947         if (ret < 0) {
948                 goto error;
949         }
950         if (ret > 0) {
951                 BUG_ON(path->slots[0] == 0);
952                 path->slots[0]--;
953         }
954
955         while(1) {
956                 fi = NULL;
957                 leaf = path->nodes[0];
958                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
959                 found_type = btrfs_key_type(&found_key);
960
961                 if (found_key.objectid != inode->i_ino)
962                         break;
963
964                 if (found_type < min_type)
965                         break;
966
967                 item_end = found_key.offset;
968                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
969                         fi = btrfs_item_ptr(leaf, path->slots[0],
970                                             struct btrfs_file_extent_item);
971                         extent_type = btrfs_file_extent_type(leaf, fi);
972                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
973                                 item_end +=
974                                     btrfs_file_extent_num_bytes(leaf, fi);
975                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
976                                 struct btrfs_item *item = btrfs_item_nr(leaf,
977                                                                 path->slots[0]);
978                                 item_end += btrfs_file_extent_inline_len(leaf,
979                                                                          item);
980                         }
981                         item_end--;
982                 }
983                 if (found_type == BTRFS_CSUM_ITEM_KEY) {
984                         ret = btrfs_csum_truncate(trans, root, path,
985                                                   inode->i_size);
986                         BUG_ON(ret);
987                 }
988                 if (item_end < inode->i_size) {
989                         if (found_type == BTRFS_DIR_ITEM_KEY) {
990                                 found_type = BTRFS_INODE_ITEM_KEY;
991                         } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
992                                 found_type = BTRFS_CSUM_ITEM_KEY;
993                         } else if (found_type == BTRFS_EXTENT_DATA_KEY) {
994                                 found_type = BTRFS_XATTR_ITEM_KEY;
995                         } else if (found_type == BTRFS_XATTR_ITEM_KEY) {
996                                 found_type = BTRFS_INODE_REF_KEY;
997                         } else if (found_type) {
998                                 found_type--;
999                         } else {
1000                                 break;
1001                         }
1002                         btrfs_set_key_type(&key, found_type);
1003                         goto next;
1004                 }
1005                 if (found_key.offset >= inode->i_size)
1006                         del_item = 1;
1007                 else
1008                         del_item = 0;
1009                 found_extent = 0;
1010
1011                 /* FIXME, shrink the extent if the ref count is only 1 */
1012                 if (found_type != BTRFS_EXTENT_DATA_KEY)
1013                         goto delete;
1014
1015                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
1016                         u64 num_dec;
1017                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
1018                         if (!del_item) {
1019                                 u64 orig_num_bytes =
1020                                         btrfs_file_extent_num_bytes(leaf, fi);
1021                                 extent_num_bytes = inode->i_size -
1022                                         found_key.offset + root->sectorsize - 1;
1023                                 extent_num_bytes = extent_num_bytes &
1024                                         ~((u64)root->sectorsize - 1);
1025                                 btrfs_set_file_extent_num_bytes(leaf, fi,
1026                                                          extent_num_bytes);
1027                                 num_dec = (orig_num_bytes -
1028                                            extent_num_bytes);
1029                                 if (extent_start != 0)
1030                                         dec_i_blocks(inode, num_dec);
1031                                 btrfs_mark_buffer_dirty(leaf);
1032                         } else {
1033                                 extent_num_bytes =
1034                                         btrfs_file_extent_disk_num_bytes(leaf,
1035                                                                          fi);
1036                                 /* FIXME blocksize != 4096 */
1037                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
1038                                 if (extent_start != 0) {
1039                                         found_extent = 1;
1040                                         dec_i_blocks(inode, num_dec);
1041                                 }
1042                                 root_gen = btrfs_header_generation(leaf);
1043                                 root_owner = btrfs_header_owner(leaf);
1044                         }
1045                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1046                         if (!del_item) {
1047                                 u32 newsize = inode->i_size - found_key.offset;
1048                                 dec_i_blocks(inode, item_end + 1 -
1049                                             found_key.offset - newsize);
1050                                 newsize =
1051                                     btrfs_file_extent_calc_inline_size(newsize);
1052                                 ret = btrfs_truncate_item(trans, root, path,
1053                                                           newsize, 1);
1054                                 BUG_ON(ret);
1055                         } else {
1056                                 dec_i_blocks(inode, item_end + 1 -
1057                                              found_key.offset);
1058                         }
1059                 }
1060 delete:
1061                 if (del_item) {
1062                         if (!pending_del_nr) {
1063                                 /* no pending yet, add ourselves */
1064                                 pending_del_slot = path->slots[0];
1065                                 pending_del_nr = 1;
1066                         } else if (pending_del_nr &&
1067                                    path->slots[0] + 1 == pending_del_slot) {
1068                                 /* hop on the pending chunk */
1069                                 pending_del_nr++;
1070                                 pending_del_slot = path->slots[0];
1071                         } else {
1072                                 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path->slots[0], pending_del_nr, pending_del_slot);
1073                         }
1074                 } else {
1075                         break;
1076                 }
1077                 if (found_extent) {
1078                         ret = btrfs_free_extent(trans, root, extent_start,
1079                                                 extent_num_bytes,
1080                                                 root_owner,
1081                                                 root_gen, inode->i_ino,
1082                                                 found_key.offset, 0);
1083                         BUG_ON(ret);
1084                 }
1085 next:
1086                 if (path->slots[0] == 0) {
1087                         if (pending_del_nr)
1088                                 goto del_pending;
1089                         btrfs_release_path(root, path);
1090                         goto search_again;
1091                 }
1092
1093                 path->slots[0]--;
1094                 if (pending_del_nr &&
1095                     path->slots[0] + 1 != pending_del_slot) {
1096                         struct btrfs_key debug;
1097 del_pending:
1098                         btrfs_item_key_to_cpu(path->nodes[0], &debug,
1099                                               pending_del_slot);
1100                         ret = btrfs_del_items(trans, root, path,
1101                                               pending_del_slot,
1102                                               pending_del_nr);
1103                         BUG_ON(ret);
1104                         pending_del_nr = 0;
1105                         btrfs_release_path(root, path);
1106                         goto search_again;
1107                 }
1108         }
1109         ret = 0;
1110 error:
1111         if (pending_del_nr) {
1112                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
1113                                       pending_del_nr);
1114         }
1115         btrfs_free_path(path);
1116         inode->i_sb->s_dirt = 1;
1117         return ret;
1118 }
1119
1120 static int btrfs_cow_one_page(struct inode *inode, struct page *page,
1121                               size_t zero_start)
1122 {
1123         char *kaddr;
1124         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1125         u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
1126         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
1127         int ret = 0;
1128
1129         WARN_ON(!PageLocked(page));
1130         set_page_extent_mapped(page);
1131
1132         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
1133         set_extent_delalloc(&BTRFS_I(inode)->io_tree, page_start,
1134                             page_end, GFP_NOFS);
1135
1136         if (zero_start != PAGE_CACHE_SIZE) {
1137                 kaddr = kmap(page);
1138                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
1139                 flush_dcache_page(page);
1140                 kunmap(page);
1141         }
1142         set_page_dirty(page);
1143         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
1144
1145         return ret;
1146 }
1147
1148 /*
1149  * taken from block_truncate_page, but does cow as it zeros out
1150  * any bytes left in the last page in the file.
1151  */
1152 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
1153 {
1154         struct inode *inode = mapping->host;
1155         struct btrfs_root *root = BTRFS_I(inode)->root;
1156         u32 blocksize = root->sectorsize;
1157         pgoff_t index = from >> PAGE_CACHE_SHIFT;
1158         unsigned offset = from & (PAGE_CACHE_SIZE-1);
1159         struct page *page;
1160         int ret = 0;
1161         u64 page_start;
1162
1163         if ((offset & (blocksize - 1)) == 0)
1164                 goto out;
1165
1166         ret = -ENOMEM;
1167 again:
1168         page = grab_cache_page(mapping, index);
1169         if (!page)
1170                 goto out;
1171         if (!PageUptodate(page)) {
1172                 ret = btrfs_readpage(NULL, page);
1173                 lock_page(page);
1174                 if (page->mapping != mapping) {
1175                         unlock_page(page);
1176                         page_cache_release(page);
1177                         goto again;
1178                 }
1179                 if (!PageUptodate(page)) {
1180                         ret = -EIO;
1181                         goto out;
1182                 }
1183         }
1184
1185         page_start = (u64)page->index << PAGE_CACHE_SHIFT;
1186         wait_on_page_writeback(page);
1187         ret = btrfs_cow_one_page(inode, page, offset);
1188
1189         unlock_page(page);
1190         page_cache_release(page);
1191 out:
1192         return ret;
1193 }
1194
1195 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
1196 {
1197         struct inode *inode = dentry->d_inode;
1198         int err;
1199
1200         err = inode_change_ok(inode, attr);
1201         if (err)
1202                 return err;
1203
1204         if (S_ISREG(inode->i_mode) &&
1205             attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
1206                 struct btrfs_trans_handle *trans;
1207                 struct btrfs_root *root = BTRFS_I(inode)->root;
1208                 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1209
1210                 u64 mask = root->sectorsize - 1;
1211                 u64 hole_start = (inode->i_size + mask) & ~mask;
1212                 u64 block_end = (attr->ia_size + mask) & ~mask;
1213                 u64 hole_size;
1214                 u64 alloc_hint = 0;
1215
1216                 if (attr->ia_size <= hole_start)
1217                         goto out;
1218
1219                 err = btrfs_check_free_space(root, 1, 0);
1220                 if (err)
1221                         goto fail;
1222
1223                 btrfs_truncate_page(inode->i_mapping, inode->i_size);
1224
1225                 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1226                 hole_size = block_end - hole_start;
1227
1228                 trans = btrfs_start_transaction(root, 1);
1229                 btrfs_set_trans_block_group(trans, inode);
1230                 err = btrfs_drop_extents(trans, root, inode,
1231                                          hole_start, block_end, hole_start,
1232                                          &alloc_hint);
1233
1234                 if (alloc_hint != EXTENT_MAP_INLINE) {
1235                         err = btrfs_insert_file_extent(trans, root,
1236                                                        inode->i_ino,
1237                                                        hole_start, 0, 0,
1238                                                        hole_size, 0);
1239                         btrfs_drop_extent_cache(inode, hole_start,
1240                                                 (u64)-1);
1241                         btrfs_check_file(root, inode);
1242                 }
1243                 btrfs_end_transaction(trans, root);
1244                 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1245                 if (err)
1246                         return err;
1247         }
1248 out:
1249         err = inode_setattr(inode, attr);
1250 fail:
1251         return err;
1252 }
1253
1254 void btrfs_delete_inode(struct inode *inode)
1255 {
1256         struct btrfs_trans_handle *trans;
1257         struct btrfs_root *root = BTRFS_I(inode)->root;
1258         unsigned long nr;
1259         int ret;
1260
1261         truncate_inode_pages(&inode->i_data, 0);
1262         if (is_bad_inode(inode)) {
1263                 goto no_delete;
1264         }
1265
1266         inode->i_size = 0;
1267         trans = btrfs_start_transaction(root, 1);
1268
1269         btrfs_set_trans_block_group(trans, inode);
1270         ret = btrfs_truncate_in_trans(trans, root, inode, 0);
1271         if (ret)
1272                 goto no_delete_lock;
1273
1274         nr = trans->blocks_used;
1275         clear_inode(inode);
1276
1277         btrfs_end_transaction(trans, root);
1278         btrfs_btree_balance_dirty(root, nr);
1279         return;
1280
1281 no_delete_lock:
1282         nr = trans->blocks_used;
1283         btrfs_end_transaction(trans, root);
1284         btrfs_btree_balance_dirty(root, nr);
1285 no_delete:
1286         clear_inode(inode);
1287 }
1288
1289 /*
1290  * this returns the key found in the dir entry in the location pointer.
1291  * If no dir entries were found, location->objectid is 0.
1292  */
1293 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
1294                                struct btrfs_key *location)
1295 {
1296         const char *name = dentry->d_name.name;
1297         int namelen = dentry->d_name.len;
1298         struct btrfs_dir_item *di;
1299         struct btrfs_path *path;
1300         struct btrfs_root *root = BTRFS_I(dir)->root;
1301         int ret = 0;
1302
1303         if (namelen == 1 && strcmp(name, ".") == 0) {
1304                 location->objectid = dir->i_ino;
1305                 location->type = BTRFS_INODE_ITEM_KEY;
1306                 location->offset = 0;
1307                 return 0;
1308         }
1309         path = btrfs_alloc_path();
1310         BUG_ON(!path);
1311
1312         if (namelen == 2 && strcmp(name, "..") == 0) {
1313                 struct btrfs_key key;
1314                 struct extent_buffer *leaf;
1315                 u32 nritems;
1316                 int slot;
1317
1318                 key.objectid = dir->i_ino;
1319                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1320                 key.offset = 0;
1321                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1322                 BUG_ON(ret == 0);
1323                 ret = 0;
1324
1325                 leaf = path->nodes[0];
1326                 slot = path->slots[0];
1327                 nritems = btrfs_header_nritems(leaf);
1328                 if (slot >= nritems)
1329                         goto out_err;
1330
1331                 btrfs_item_key_to_cpu(leaf, &key, slot);
1332                 if (key.objectid != dir->i_ino ||
1333                     key.type != BTRFS_INODE_REF_KEY) {
1334                         goto out_err;
1335                 }
1336                 location->objectid = key.offset;
1337                 location->type = BTRFS_INODE_ITEM_KEY;
1338                 location->offset = 0;
1339                 goto out;
1340         }
1341
1342         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
1343                                     namelen, 0);
1344         if (IS_ERR(di))
1345                 ret = PTR_ERR(di);
1346         if (!di || IS_ERR(di)) {
1347                 goto out_err;
1348         }
1349         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
1350 out:
1351         btrfs_free_path(path);
1352         return ret;
1353 out_err:
1354         location->objectid = 0;
1355         goto out;
1356 }
1357
1358 /*
1359  * when we hit a tree root in a directory, the btrfs part of the inode
1360  * needs to be changed to reflect the root directory of the tree root.  This
1361  * is kind of like crossing a mount point.
1362  */
1363 static int fixup_tree_root_location(struct btrfs_root *root,
1364                              struct btrfs_key *location,
1365                              struct btrfs_root **sub_root,
1366                              struct dentry *dentry)
1367 {
1368         struct btrfs_path *path;
1369         struct btrfs_root_item *ri;
1370
1371         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
1372                 return 0;
1373         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1374                 return 0;
1375
1376         path = btrfs_alloc_path();
1377         BUG_ON(!path);
1378
1379         *sub_root = btrfs_read_fs_root(root->fs_info, location,
1380                                         dentry->d_name.name,
1381                                         dentry->d_name.len);
1382         if (IS_ERR(*sub_root))
1383                 return PTR_ERR(*sub_root);
1384
1385         ri = &(*sub_root)->root_item;
1386         location->objectid = btrfs_root_dirid(ri);
1387         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1388         location->offset = 0;
1389
1390         btrfs_free_path(path);
1391         return 0;
1392 }
1393
1394 static int btrfs_init_locked_inode(struct inode *inode, void *p)
1395 {
1396         struct btrfs_iget_args *args = p;
1397         inode->i_ino = args->ino;
1398         BTRFS_I(inode)->root = args->root;
1399         BTRFS_I(inode)->delalloc_bytes = 0;
1400         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1401         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1402                              inode->i_mapping, GFP_NOFS);
1403         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
1404                              inode->i_mapping, GFP_NOFS);
1405         atomic_set(&BTRFS_I(inode)->ordered_writeback, 0);
1406         return 0;
1407 }
1408
1409 static int btrfs_find_actor(struct inode *inode, void *opaque)
1410 {
1411         struct btrfs_iget_args *args = opaque;
1412         return (args->ino == inode->i_ino &&
1413                 args->root == BTRFS_I(inode)->root);
1414 }
1415
1416 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
1417                             u64 root_objectid)
1418 {
1419         struct btrfs_iget_args args;
1420         args.ino = objectid;
1421         args.root = btrfs_lookup_fs_root(btrfs_sb(s)->fs_info, root_objectid);
1422
1423         if (!args.root)
1424                 return NULL;
1425
1426         return ilookup5(s, objectid, btrfs_find_actor, (void *)&args);
1427 }
1428
1429 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
1430                                 struct btrfs_root *root)
1431 {
1432         struct inode *inode;
1433         struct btrfs_iget_args args;
1434         args.ino = objectid;
1435         args.root = root;
1436
1437         inode = iget5_locked(s, objectid, btrfs_find_actor,
1438                              btrfs_init_locked_inode,
1439                              (void *)&args);
1440         return inode;
1441 }
1442
1443 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
1444                                    struct nameidata *nd)
1445 {
1446         struct inode * inode;
1447         struct btrfs_inode *bi = BTRFS_I(dir);
1448         struct btrfs_root *root = bi->root;
1449         struct btrfs_root *sub_root = root;
1450         struct btrfs_key location;
1451         int ret;
1452
1453         if (dentry->d_name.len > BTRFS_NAME_LEN)
1454                 return ERR_PTR(-ENAMETOOLONG);
1455
1456         ret = btrfs_inode_by_name(dir, dentry, &location);
1457
1458         if (ret < 0)
1459                 return ERR_PTR(ret);
1460
1461         inode = NULL;
1462         if (location.objectid) {
1463                 ret = fixup_tree_root_location(root, &location, &sub_root,
1464                                                 dentry);
1465                 if (ret < 0)
1466                         return ERR_PTR(ret);
1467                 if (ret > 0)
1468                         return ERR_PTR(-ENOENT);
1469                 inode = btrfs_iget_locked(dir->i_sb, location.objectid,
1470                                           sub_root);
1471                 if (!inode)
1472                         return ERR_PTR(-EACCES);
1473                 if (inode->i_state & I_NEW) {
1474                         /* the inode and parent dir are two different roots */
1475                         if (sub_root != root) {
1476                                 igrab(inode);
1477                                 sub_root->inode = inode;
1478                         }
1479                         BTRFS_I(inode)->root = sub_root;
1480                         memcpy(&BTRFS_I(inode)->location, &location,
1481                                sizeof(location));
1482                         btrfs_read_locked_inode(inode);
1483                         unlock_new_inode(inode);
1484                 }
1485         }
1486         return d_splice_alias(inode, dentry);
1487 }
1488
1489 static unsigned char btrfs_filetype_table[] = {
1490         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
1491 };
1492
1493 static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
1494 {
1495         struct inode *inode = filp->f_dentry->d_inode;
1496         struct btrfs_root *root = BTRFS_I(inode)->root;
1497         struct btrfs_item *item;
1498         struct btrfs_dir_item *di;
1499         struct btrfs_key key;
1500         struct btrfs_key found_key;
1501         struct btrfs_path *path;
1502         int ret;
1503         u32 nritems;
1504         struct extent_buffer *leaf;
1505         int slot;
1506         int advance;
1507         unsigned char d_type;
1508         int over = 0;
1509         u32 di_cur;
1510         u32 di_total;
1511         u32 di_len;
1512         int key_type = BTRFS_DIR_INDEX_KEY;
1513         char tmp_name[32];
1514         char *name_ptr;
1515         int name_len;
1516
1517         /* FIXME, use a real flag for deciding about the key type */
1518         if (root->fs_info->tree_root == root)
1519                 key_type = BTRFS_DIR_ITEM_KEY;
1520
1521         /* special case for "." */
1522         if (filp->f_pos == 0) {
1523                 over = filldir(dirent, ".", 1,
1524                                1, inode->i_ino,
1525                                DT_DIR);
1526                 if (over)
1527                         return 0;
1528                 filp->f_pos = 1;
1529         }
1530
1531         key.objectid = inode->i_ino;
1532         path = btrfs_alloc_path();
1533         path->reada = 2;
1534
1535         /* special case for .., just use the back ref */
1536         if (filp->f_pos == 1) {
1537                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1538                 key.offset = 0;
1539                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1540                 BUG_ON(ret == 0);
1541                 leaf = path->nodes[0];
1542                 slot = path->slots[0];
1543                 nritems = btrfs_header_nritems(leaf);
1544                 if (slot >= nritems) {
1545                         btrfs_release_path(root, path);
1546                         goto read_dir_items;
1547                 }
1548                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1549                 btrfs_release_path(root, path);
1550                 if (found_key.objectid != key.objectid ||
1551                     found_key.type != BTRFS_INODE_REF_KEY)
1552                         goto read_dir_items;
1553                 over = filldir(dirent, "..", 2,
1554                                2, found_key.offset, DT_DIR);
1555                 if (over)
1556                         goto nopos;
1557                 filp->f_pos = 2;
1558         }
1559
1560 read_dir_items:
1561         btrfs_set_key_type(&key, key_type);
1562         key.offset = filp->f_pos;
1563
1564         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1565         if (ret < 0)
1566                 goto err;
1567         advance = 0;
1568         while(1) {
1569                 leaf = path->nodes[0];
1570                 nritems = btrfs_header_nritems(leaf);
1571                 slot = path->slots[0];
1572                 if (advance || slot >= nritems) {
1573                         if (slot >= nritems -1) {
1574                                 ret = btrfs_next_leaf(root, path);
1575                                 if (ret)
1576                                         break;
1577                                 leaf = path->nodes[0];
1578                                 nritems = btrfs_header_nritems(leaf);
1579                                 slot = path->slots[0];
1580                         } else {
1581                                 slot++;
1582                                 path->slots[0]++;
1583                         }
1584                 }
1585                 advance = 1;
1586                 item = btrfs_item_nr(leaf, slot);
1587                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1588
1589                 if (found_key.objectid != key.objectid)
1590                         break;
1591                 if (btrfs_key_type(&found_key) != key_type)
1592                         break;
1593                 if (found_key.offset < filp->f_pos)
1594                         continue;
1595
1596                 filp->f_pos = found_key.offset;
1597                 advance = 1;
1598                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
1599                 di_cur = 0;
1600                 di_total = btrfs_item_size(leaf, item);
1601                 while(di_cur < di_total) {
1602                         struct btrfs_key location;
1603
1604                         name_len = btrfs_dir_name_len(leaf, di);
1605                         if (name_len < 32) {
1606                                 name_ptr = tmp_name;
1607                         } else {
1608                                 name_ptr = kmalloc(name_len, GFP_NOFS);
1609                                 BUG_ON(!name_ptr);
1610                         }
1611                         read_extent_buffer(leaf, name_ptr,
1612                                            (unsigned long)(di + 1), name_len);
1613
1614                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
1615                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
1616                         over = filldir(dirent, name_ptr, name_len,
1617                                        found_key.offset,
1618                                        location.objectid,
1619                                        d_type);
1620
1621                         if (name_ptr != tmp_name)
1622                                 kfree(name_ptr);
1623
1624                         if (over)
1625                                 goto nopos;
1626                         di_len = btrfs_dir_name_len(leaf, di) +
1627                                 btrfs_dir_data_len(leaf, di) +sizeof(*di);
1628                         di_cur += di_len;
1629                         di = (struct btrfs_dir_item *)((char *)di + di_len);
1630                 }
1631         }
1632         if (key_type == BTRFS_DIR_INDEX_KEY)
1633                 filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
1634         else
1635                 filp->f_pos++;
1636 nopos:
1637         ret = 0;
1638 err:
1639         btrfs_free_path(path);
1640         return ret;
1641 }
1642
1643 int btrfs_write_inode(struct inode *inode, int wait)
1644 {
1645         struct btrfs_root *root = BTRFS_I(inode)->root;
1646         struct btrfs_trans_handle *trans;
1647         int ret = 0;
1648
1649         if (wait) {
1650                 trans = btrfs_start_transaction(root, 1);
1651                 btrfs_set_trans_block_group(trans, inode);
1652                 ret = btrfs_commit_transaction(trans, root);
1653         }
1654         return ret;
1655 }
1656
1657 /*
1658  * This is somewhat expensive, updating the tree every time the
1659  * inode changes.  But, it is most likely to find the inode in cache.
1660  * FIXME, needs more benchmarking...there are no reasons other than performance
1661  * to keep or drop this code.
1662  */
1663 void btrfs_dirty_inode(struct inode *inode)
1664 {
1665         struct btrfs_root *root = BTRFS_I(inode)->root;
1666         struct btrfs_trans_handle *trans;
1667
1668         trans = btrfs_start_transaction(root, 1);
1669         btrfs_set_trans_block_group(trans, inode);
1670         btrfs_update_inode(trans, root, inode);
1671         btrfs_end_transaction(trans, root);
1672 }
1673
1674 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
1675                                      struct btrfs_root *root,
1676                                      const char *name, int name_len,
1677                                      u64 ref_objectid,
1678                                      u64 objectid,
1679                                      struct btrfs_block_group_cache *group,
1680                                      int mode)
1681 {
1682         struct inode *inode;
1683         struct btrfs_inode_item *inode_item;
1684         struct btrfs_block_group_cache *new_inode_group;
1685         struct btrfs_key *location;
1686         struct btrfs_path *path;
1687         struct btrfs_inode_ref *ref;
1688         struct btrfs_key key[2];
1689         u32 sizes[2];
1690         unsigned long ptr;
1691         int ret;
1692         int owner;
1693
1694         path = btrfs_alloc_path();
1695         BUG_ON(!path);
1696
1697         inode = new_inode(root->fs_info->sb);
1698         if (!inode)
1699                 return ERR_PTR(-ENOMEM);
1700
1701         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1702         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1703                              inode->i_mapping, GFP_NOFS);
1704         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
1705                              inode->i_mapping, GFP_NOFS);
1706         atomic_set(&BTRFS_I(inode)->ordered_writeback, 0);
1707         BTRFS_I(inode)->delalloc_bytes = 0;
1708         BTRFS_I(inode)->root = root;
1709
1710         if (mode & S_IFDIR)
1711                 owner = 0;
1712         else
1713                 owner = 1;
1714         new_inode_group = btrfs_find_block_group(root, group, 0,
1715                                        BTRFS_BLOCK_GROUP_METADATA, owner);
1716         if (!new_inode_group) {
1717                 printk("find_block group failed\n");
1718                 new_inode_group = group;
1719         }
1720         BTRFS_I(inode)->block_group = new_inode_group;
1721         BTRFS_I(inode)->flags = 0;
1722
1723         key[0].objectid = objectid;
1724         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
1725         key[0].offset = 0;
1726
1727         key[1].objectid = objectid;
1728         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
1729         key[1].offset = ref_objectid;
1730
1731         sizes[0] = sizeof(struct btrfs_inode_item);
1732         sizes[1] = name_len + sizeof(*ref);
1733
1734         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
1735         if (ret != 0)
1736                 goto fail;
1737
1738         if (objectid > root->highest_inode)
1739                 root->highest_inode = objectid;
1740
1741         inode->i_uid = current->fsuid;
1742         inode->i_gid = current->fsgid;
1743         inode->i_mode = mode;
1744         inode->i_ino = objectid;
1745         inode->i_blocks = 0;
1746         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1747         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1748                                   struct btrfs_inode_item);
1749         fill_inode_item(path->nodes[0], inode_item, inode);
1750
1751         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
1752                              struct btrfs_inode_ref);
1753         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
1754         ptr = (unsigned long)(ref + 1);
1755         write_extent_buffer(path->nodes[0], name, ptr, name_len);
1756
1757         btrfs_mark_buffer_dirty(path->nodes[0]);
1758         btrfs_free_path(path);
1759
1760         location = &BTRFS_I(inode)->location;
1761         location->objectid = objectid;
1762         location->offset = 0;
1763         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1764
1765         insert_inode_hash(inode);
1766         return inode;
1767 fail:
1768         btrfs_free_path(path);
1769         return ERR_PTR(ret);
1770 }
1771
1772 static inline u8 btrfs_inode_type(struct inode *inode)
1773 {
1774         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
1775 }
1776
1777 static int btrfs_add_link(struct btrfs_trans_handle *trans,
1778                             struct dentry *dentry, struct inode *inode,
1779                             int add_backref)
1780 {
1781         int ret;
1782         struct btrfs_key key;
1783         struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
1784         struct inode *parent_inode;
1785
1786         key.objectid = inode->i_ino;
1787         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1788         key.offset = 0;
1789
1790         ret = btrfs_insert_dir_item(trans, root,
1791                                     dentry->d_name.name, dentry->d_name.len,
1792                                     dentry->d_parent->d_inode->i_ino,
1793                                     &key, btrfs_inode_type(inode));
1794         if (ret == 0) {
1795                 if (add_backref) {
1796                         ret = btrfs_insert_inode_ref(trans, root,
1797                                              dentry->d_name.name,
1798                                              dentry->d_name.len,
1799                                              inode->i_ino,
1800                                              dentry->d_parent->d_inode->i_ino);
1801                 }
1802                 parent_inode = dentry->d_parent->d_inode;
1803                 parent_inode->i_size += dentry->d_name.len * 2;
1804                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1805                 ret = btrfs_update_inode(trans, root,
1806                                          dentry->d_parent->d_inode);
1807         }
1808         return ret;
1809 }
1810
1811 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
1812                             struct dentry *dentry, struct inode *inode,
1813                             int backref)
1814 {
1815         int err = btrfs_add_link(trans, dentry, inode, backref);
1816         if (!err) {
1817                 d_instantiate(dentry, inode);
1818                 return 0;
1819         }
1820         if (err > 0)
1821                 err = -EEXIST;
1822         return err;
1823 }
1824
1825 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1826                         int mode, dev_t rdev)
1827 {
1828         struct btrfs_trans_handle *trans;
1829         struct btrfs_root *root = BTRFS_I(dir)->root;
1830         struct inode *inode = NULL;
1831         int err;
1832         int drop_inode = 0;
1833         u64 objectid;
1834         unsigned long nr = 0;
1835
1836         if (!new_valid_dev(rdev))
1837                 return -EINVAL;
1838
1839         err = btrfs_check_free_space(root, 1, 0);
1840         if (err)
1841                 goto fail;
1842
1843         trans = btrfs_start_transaction(root, 1);
1844         btrfs_set_trans_block_group(trans, dir);
1845
1846         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1847         if (err) {
1848                 err = -ENOSPC;
1849                 goto out_unlock;
1850         }
1851
1852         inode = btrfs_new_inode(trans, root, dentry->d_name.name,
1853                                 dentry->d_name.len,
1854                                 dentry->d_parent->d_inode->i_ino, objectid,
1855                                 BTRFS_I(dir)->block_group, mode);
1856         err = PTR_ERR(inode);
1857         if (IS_ERR(inode))
1858                 goto out_unlock;
1859
1860         btrfs_set_trans_block_group(trans, inode);
1861         err = btrfs_add_nondir(trans, dentry, inode, 0);
1862         if (err)
1863                 drop_inode = 1;
1864         else {
1865                 inode->i_op = &btrfs_special_inode_operations;
1866                 init_special_inode(inode, inode->i_mode, rdev);
1867                 btrfs_update_inode(trans, root, inode);
1868         }
1869         dir->i_sb->s_dirt = 1;
1870         btrfs_update_inode_block_group(trans, inode);
1871         btrfs_update_inode_block_group(trans, dir);
1872 out_unlock:
1873         nr = trans->blocks_used;
1874         btrfs_end_transaction(trans, root);
1875 fail:
1876         if (drop_inode) {
1877                 inode_dec_link_count(inode);
1878                 iput(inode);
1879         }
1880         btrfs_btree_balance_dirty(root, nr);
1881         btrfs_throttle(root);
1882         return err;
1883 }
1884
1885 static int btrfs_create(struct inode *dir, struct dentry *dentry,
1886                         int mode, struct nameidata *nd)
1887 {
1888         struct btrfs_trans_handle *trans;
1889         struct btrfs_root *root = BTRFS_I(dir)->root;
1890         struct inode *inode = NULL;
1891         int err;
1892         int drop_inode = 0;
1893         unsigned long nr = 0;
1894         u64 objectid;
1895
1896         err = btrfs_check_free_space(root, 1, 0);
1897         if (err)
1898                 goto fail;
1899         trans = btrfs_start_transaction(root, 1);
1900         btrfs_set_trans_block_group(trans, dir);
1901
1902         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1903         if (err) {
1904                 err = -ENOSPC;
1905                 goto out_unlock;
1906         }
1907
1908         inode = btrfs_new_inode(trans, root, dentry->d_name.name,
1909                                 dentry->d_name.len,
1910                                 dentry->d_parent->d_inode->i_ino,
1911                                 objectid, BTRFS_I(dir)->block_group, mode);
1912         err = PTR_ERR(inode);
1913         if (IS_ERR(inode))
1914                 goto out_unlock;
1915
1916         btrfs_set_trans_block_group(trans, inode);
1917         err = btrfs_add_nondir(trans, dentry, inode, 0);
1918         if (err)
1919                 drop_inode = 1;
1920         else {
1921                 inode->i_mapping->a_ops = &btrfs_aops;
1922                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
1923                 inode->i_fop = &btrfs_file_operations;
1924                 inode->i_op = &btrfs_file_inode_operations;
1925                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1926                 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1927                                      inode->i_mapping, GFP_NOFS);
1928                 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
1929                                      inode->i_mapping, GFP_NOFS);
1930                 BTRFS_I(inode)->delalloc_bytes = 0;
1931                 atomic_set(&BTRFS_I(inode)->ordered_writeback, 0);
1932                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
1933         }
1934         dir->i_sb->s_dirt = 1;
1935         btrfs_update_inode_block_group(trans, inode);
1936         btrfs_update_inode_block_group(trans, dir);
1937 out_unlock:
1938         nr = trans->blocks_used;
1939         btrfs_end_transaction(trans, root);
1940 fail:
1941         if (drop_inode) {
1942                 inode_dec_link_count(inode);
1943                 iput(inode);
1944         }
1945         btrfs_btree_balance_dirty(root, nr);
1946         btrfs_throttle(root);
1947         return err;
1948 }
1949
1950 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
1951                       struct dentry *dentry)
1952 {
1953         struct btrfs_trans_handle *trans;
1954         struct btrfs_root *root = BTRFS_I(dir)->root;
1955         struct inode *inode = old_dentry->d_inode;
1956         unsigned long nr = 0;
1957         int err;
1958         int drop_inode = 0;
1959
1960         if (inode->i_nlink == 0)
1961                 return -ENOENT;
1962
1963 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1964         inode->i_nlink++;
1965 #else
1966         inc_nlink(inode);
1967 #endif
1968         err = btrfs_check_free_space(root, 1, 0);
1969         if (err)
1970                 goto fail;
1971         trans = btrfs_start_transaction(root, 1);
1972
1973         btrfs_set_trans_block_group(trans, dir);
1974         atomic_inc(&inode->i_count);
1975         err = btrfs_add_nondir(trans, dentry, inode, 1);
1976
1977         if (err)
1978                 drop_inode = 1;
1979
1980         dir->i_sb->s_dirt = 1;
1981         btrfs_update_inode_block_group(trans, dir);
1982         err = btrfs_update_inode(trans, root, inode);
1983
1984         if (err)
1985                 drop_inode = 1;
1986
1987         nr = trans->blocks_used;
1988         btrfs_end_transaction(trans, root);
1989 fail:
1990         if (drop_inode) {
1991                 inode_dec_link_count(inode);
1992                 iput(inode);
1993         }
1994         btrfs_btree_balance_dirty(root, nr);
1995         btrfs_throttle(root);
1996         return err;
1997 }
1998
1999 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2000 {
2001         struct inode *inode = NULL;
2002         struct btrfs_trans_handle *trans;
2003         struct btrfs_root *root = BTRFS_I(dir)->root;
2004         int err = 0;
2005         int drop_on_err = 0;
2006         u64 objectid = 0;
2007         unsigned long nr = 1;
2008
2009         err = btrfs_check_free_space(root, 1, 0);
2010         if (err)
2011                 goto out_unlock;
2012
2013         trans = btrfs_start_transaction(root, 1);
2014         btrfs_set_trans_block_group(trans, dir);
2015
2016         if (IS_ERR(trans)) {
2017                 err = PTR_ERR(trans);
2018                 goto out_unlock;
2019         }
2020
2021         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2022         if (err) {
2023                 err = -ENOSPC;
2024                 goto out_unlock;
2025         }
2026
2027         inode = btrfs_new_inode(trans, root, dentry->d_name.name,
2028                                 dentry->d_name.len,
2029                                 dentry->d_parent->d_inode->i_ino, objectid,
2030                                 BTRFS_I(dir)->block_group, S_IFDIR | mode);
2031         if (IS_ERR(inode)) {
2032                 err = PTR_ERR(inode);
2033                 goto out_fail;
2034         }
2035
2036         drop_on_err = 1;
2037         inode->i_op = &btrfs_dir_inode_operations;
2038         inode->i_fop = &btrfs_dir_file_operations;
2039         btrfs_set_trans_block_group(trans, inode);
2040
2041         inode->i_size = 0;
2042         err = btrfs_update_inode(trans, root, inode);
2043         if (err)
2044                 goto out_fail;
2045
2046         err = btrfs_add_link(trans, dentry, inode, 0);
2047         if (err)
2048                 goto out_fail;
2049
2050         d_instantiate(dentry, inode);
2051         drop_on_err = 0;
2052         dir->i_sb->s_dirt = 1;
2053         btrfs_update_inode_block_group(trans, inode);
2054         btrfs_update_inode_block_group(trans, dir);
2055
2056 out_fail:
2057         nr = trans->blocks_used;
2058         btrfs_end_transaction(trans, root);
2059
2060 out_unlock:
2061         if (drop_on_err)
2062                 iput(inode);
2063         btrfs_btree_balance_dirty(root, nr);
2064         btrfs_throttle(root);
2065         return err;
2066 }
2067
2068 static int merge_extent_mapping(struct extent_map_tree *em_tree,
2069                                 struct extent_map *existing,
2070                                 struct extent_map *em)
2071 {
2072         u64 start_diff;
2073         u64 new_end;
2074         int ret = 0;
2075         int real_blocks = existing->block_start < EXTENT_MAP_LAST_BYTE;
2076
2077         if (real_blocks && em->block_start >= EXTENT_MAP_LAST_BYTE)
2078                 goto invalid;
2079
2080         if (!real_blocks && em->block_start != existing->block_start)
2081                 goto invalid;
2082
2083         new_end = max(existing->start + existing->len, em->start + em->len);
2084
2085         if (existing->start >= em->start) {
2086                 if (em->start + em->len < existing->start)
2087                         goto invalid;
2088
2089                 start_diff = existing->start - em->start;
2090                 if (real_blocks && em->block_start + start_diff !=
2091                     existing->block_start)
2092                         goto invalid;
2093
2094                 em->len = new_end - em->start;
2095
2096                 remove_extent_mapping(em_tree, existing);
2097                 /* free for the tree */
2098                 free_extent_map(existing);
2099                 ret = add_extent_mapping(em_tree, em);
2100
2101         } else if (em->start > existing->start) {
2102
2103                 if (existing->start + existing->len < em->start)
2104                         goto invalid;
2105
2106                 start_diff = em->start - existing->start;
2107                 if (real_blocks && existing->block_start + start_diff !=
2108                     em->block_start)
2109                         goto invalid;
2110
2111                 remove_extent_mapping(em_tree, existing);
2112                 em->block_start = existing->block_start;
2113                 em->start = existing->start;
2114                 em->len = new_end - existing->start;
2115                 free_extent_map(existing);
2116
2117                 ret = add_extent_mapping(em_tree, em);
2118         } else {
2119                 goto invalid;
2120         }
2121         return ret;
2122
2123 invalid:
2124         printk("invalid extent map merge [%Lu %Lu %Lu] [%Lu %Lu %Lu]\n",
2125                existing->start, existing->len, existing->block_start,
2126                em->start, em->len, em->block_start);
2127         return -EIO;
2128 }
2129
2130 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
2131                                     size_t pg_offset, u64 start, u64 len,
2132                                     int create)
2133 {
2134         int ret;
2135         int err = 0;
2136         u64 bytenr;
2137         u64 extent_start = 0;
2138         u64 extent_end = 0;
2139         u64 objectid = inode->i_ino;
2140         u32 found_type;
2141         struct btrfs_path *path;
2142         struct btrfs_root *root = BTRFS_I(inode)->root;
2143         struct btrfs_file_extent_item *item;
2144         struct extent_buffer *leaf;
2145         struct btrfs_key found_key;
2146         struct extent_map *em = NULL;
2147         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2148         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2149         struct btrfs_trans_handle *trans = NULL;
2150
2151         path = btrfs_alloc_path();
2152         BUG_ON(!path);
2153
2154 again:
2155         spin_lock(&em_tree->lock);
2156         em = lookup_extent_mapping(em_tree, start, len);
2157         if (em)
2158                 em->bdev = root->fs_info->fs_devices->latest_bdev;
2159         spin_unlock(&em_tree->lock);
2160
2161         if (em) {
2162                 if (em->start > start || em->start + em->len <= start)
2163                         free_extent_map(em);
2164                 else if (em->block_start == EXTENT_MAP_INLINE && page)
2165                         free_extent_map(em);
2166                 else
2167                         goto out;
2168         }
2169         em = alloc_extent_map(GFP_NOFS);
2170         if (!em) {
2171                 err = -ENOMEM;
2172                 goto out;
2173         }
2174
2175         em->start = EXTENT_MAP_HOLE;
2176         em->len = (u64)-1;
2177         em->bdev = root->fs_info->fs_devices->latest_bdev;
2178         ret = btrfs_lookup_file_extent(trans, root, path,
2179                                        objectid, start, trans != NULL);
2180         if (ret < 0) {
2181                 err = ret;
2182                 goto out;
2183         }
2184
2185         if (ret != 0) {
2186                 if (path->slots[0] == 0)
2187                         goto not_found;
2188                 path->slots[0]--;
2189         }
2190
2191         leaf = path->nodes[0];
2192         item = btrfs_item_ptr(leaf, path->slots[0],
2193                               struct btrfs_file_extent_item);
2194         /* are we inside the extent that was found? */
2195         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2196         found_type = btrfs_key_type(&found_key);
2197         if (found_key.objectid != objectid ||
2198             found_type != BTRFS_EXTENT_DATA_KEY) {
2199                 goto not_found;
2200         }
2201
2202         found_type = btrfs_file_extent_type(leaf, item);
2203         extent_start = found_key.offset;
2204         if (found_type == BTRFS_FILE_EXTENT_REG) {
2205                 extent_end = extent_start +
2206                        btrfs_file_extent_num_bytes(leaf, item);
2207                 err = 0;
2208                 if (start < extent_start || start >= extent_end) {
2209                         em->start = start;
2210                         if (start < extent_start) {
2211                                 if (start + len <= extent_start)
2212                                         goto not_found;
2213                                 em->len = extent_end - extent_start;
2214                         } else {
2215                                 em->len = len;
2216                         }
2217                         goto not_found_em;
2218                 }
2219                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
2220                 if (bytenr == 0) {
2221                         em->start = extent_start;
2222                         em->len = extent_end - extent_start;
2223                         em->block_start = EXTENT_MAP_HOLE;
2224                         goto insert;
2225                 }
2226                 bytenr += btrfs_file_extent_offset(leaf, item);
2227                 em->block_start = bytenr;
2228                 em->start = extent_start;
2229                 em->len = extent_end - extent_start;
2230                 goto insert;
2231         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
2232                 u64 page_start;
2233                 unsigned long ptr;
2234                 char *map;
2235                 size_t size;
2236                 size_t extent_offset;
2237                 size_t copy_size;
2238
2239                 size = btrfs_file_extent_inline_len(leaf, btrfs_item_nr(leaf,
2240                                                     path->slots[0]));
2241                 extent_end = (extent_start + size + root->sectorsize - 1) &
2242                         ~((u64)root->sectorsize - 1);
2243                 if (start < extent_start || start >= extent_end) {
2244                         em->start = start;
2245                         if (start < extent_start) {
2246                                 if (start + len <= extent_start)
2247                                         goto not_found;
2248                                 em->len = extent_end - extent_start;
2249                         } else {
2250                                 em->len = len;
2251                         }
2252                         goto not_found_em;
2253                 }
2254                 em->block_start = EXTENT_MAP_INLINE;
2255
2256                 if (!page) {
2257                         em->start = extent_start;
2258                         em->len = size;
2259                         goto out;
2260                 }
2261
2262                 page_start = page_offset(page) + pg_offset;
2263                 extent_offset = page_start - extent_start;
2264                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
2265                                 size - extent_offset);
2266                 em->start = extent_start + extent_offset;
2267                 em->len = (copy_size + root->sectorsize - 1) &
2268                         ~((u64)root->sectorsize - 1);
2269                 map = kmap(page);
2270                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
2271                 if (create == 0 && !PageUptodate(page)) {
2272                         read_extent_buffer(leaf, map + pg_offset, ptr,
2273                                            copy_size);
2274                         flush_dcache_page(page);
2275                 } else if (create && PageUptodate(page)) {
2276                         if (!trans) {
2277                                 kunmap(page);
2278                                 free_extent_map(em);
2279                                 em = NULL;
2280                                 btrfs_release_path(root, path);
2281                                 trans = btrfs_start_transaction(root, 1);
2282                                 goto again;
2283                         }
2284                         write_extent_buffer(leaf, map + pg_offset, ptr,
2285                                             copy_size);
2286                         btrfs_mark_buffer_dirty(leaf);
2287                 }
2288                 kunmap(page);
2289                 set_extent_uptodate(io_tree, em->start,
2290                                     extent_map_end(em) - 1, GFP_NOFS);
2291                 goto insert;
2292         } else {
2293                 printk("unkknown found_type %d\n", found_type);
2294                 WARN_ON(1);
2295         }
2296 not_found:
2297         em->start = start;
2298         em->len = len;
2299 not_found_em:
2300         em->block_start = EXTENT_MAP_HOLE;
2301 insert:
2302         btrfs_release_path(root, path);
2303         if (em->start > start || extent_map_end(em) <= start) {
2304                 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
2305                 err = -EIO;
2306                 goto out;
2307         }
2308
2309         err = 0;
2310         spin_lock(&em_tree->lock);
2311         ret = add_extent_mapping(em_tree, em);
2312         /* it is possible that someone inserted the extent into the tree
2313          * while we had the lock dropped.  It is also possible that
2314          * an overlapping map exists in the tree
2315          */
2316         if (ret == -EEXIST) {
2317                 struct extent_map *existing;
2318                 existing = lookup_extent_mapping(em_tree, start, len);
2319                 if (existing && (existing->start > start ||
2320                     existing->start + existing->len <= start)) {
2321                         free_extent_map(existing);
2322                         existing = NULL;
2323                 }
2324                 if (!existing) {
2325                         existing = lookup_extent_mapping(em_tree, em->start,
2326                                                          em->len);
2327                         if (existing) {
2328                                 err = merge_extent_mapping(em_tree, existing,
2329                                                            em);
2330                                 free_extent_map(existing);
2331                                 if (err) {
2332                                         free_extent_map(em);
2333                                         em = NULL;
2334                                 }
2335                         } else {
2336                                 err = -EIO;
2337                                 printk("failing to insert %Lu %Lu\n",
2338                                        start, len);
2339                                 free_extent_map(em);
2340                                 em = NULL;
2341                         }
2342                 } else {
2343                         free_extent_map(em);
2344                         em = existing;
2345                 }
2346         }
2347         spin_unlock(&em_tree->lock);
2348 out:
2349         btrfs_free_path(path);
2350         if (trans) {
2351                 ret = btrfs_end_transaction(trans, root);
2352                 if (!err)
2353                         err = ret;
2354         }
2355         if (err) {
2356                 free_extent_map(em);
2357                 WARN_ON(1);
2358                 return ERR_PTR(err);
2359         }
2360         return em;
2361 }
2362
2363 #if 0 /* waiting for O_DIRECT reads */
2364 static int btrfs_get_block(struct inode *inode, sector_t iblock,
2365                         struct buffer_head *bh_result, int create)
2366 {
2367         struct extent_map *em;
2368         u64 start = (u64)iblock << inode->i_blkbits;
2369         struct btrfs_multi_bio *multi = NULL;
2370         struct btrfs_root *root = BTRFS_I(inode)->root;
2371         u64 len;
2372         u64 logical;
2373         u64 map_length;
2374         int ret = 0;
2375
2376         em = btrfs_get_extent(inode, NULL, 0, start, bh_result->b_size, 0);
2377
2378         if (!em || IS_ERR(em))
2379                 goto out;
2380
2381         if (em->start > start || em->start + em->len <= start) {
2382             goto out;
2383         }
2384
2385         if (em->block_start == EXTENT_MAP_INLINE) {
2386                 ret = -EINVAL;
2387                 goto out;
2388         }
2389
2390         len = em->start + em->len - start;
2391         len = min_t(u64, len, INT_LIMIT(typeof(bh_result->b_size)));
2392
2393         if (em->block_start == EXTENT_MAP_HOLE ||
2394             em->block_start == EXTENT_MAP_DELALLOC) {
2395                 bh_result->b_size = len;
2396                 goto out;
2397         }
2398
2399         logical = start - em->start;
2400         logical = em->block_start + logical;
2401
2402         map_length = len;
2403         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
2404                               logical, &map_length, &multi, 0);
2405         BUG_ON(ret);
2406         bh_result->b_blocknr = multi->stripes[0].physical >> inode->i_blkbits;
2407         bh_result->b_size = min(map_length, len);
2408
2409         bh_result->b_bdev = multi->stripes[0].dev->bdev;
2410         set_buffer_mapped(bh_result);
2411         kfree(multi);
2412 out:
2413         free_extent_map(em);
2414         return ret;
2415 }
2416 #endif
2417
2418 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
2419                         const struct iovec *iov, loff_t offset,
2420                         unsigned long nr_segs)
2421 {
2422         return -EINVAL;
2423 #if 0
2424         struct file *file = iocb->ki_filp;
2425         struct inode *inode = file->f_mapping->host;
2426
2427         if (rw == WRITE)
2428                 return -EINVAL;
2429
2430         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
2431                                   offset, nr_segs, btrfs_get_block, NULL);
2432 #endif
2433 }
2434
2435 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
2436 {
2437         return extent_bmap(mapping, iblock, btrfs_get_extent);
2438 }
2439
2440 int btrfs_readpage(struct file *file, struct page *page)
2441 {
2442         struct extent_io_tree *tree;
2443         tree = &BTRFS_I(page->mapping->host)->io_tree;
2444         return extent_read_full_page(tree, page, btrfs_get_extent);
2445 }
2446
2447 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
2448 {
2449         struct extent_io_tree *tree;
2450
2451
2452         if (current->flags & PF_MEMALLOC) {
2453                 redirty_page_for_writepage(wbc, page);
2454                 unlock_page(page);
2455                 return 0;
2456         }
2457         tree = &BTRFS_I(page->mapping->host)->io_tree;
2458         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
2459 }
2460
2461 static int btrfs_writepages(struct address_space *mapping,
2462                             struct writeback_control *wbc)
2463 {
2464         struct extent_io_tree *tree;
2465         tree = &BTRFS_I(mapping->host)->io_tree;
2466         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
2467 }
2468
2469 static int
2470 btrfs_readpages(struct file *file, struct address_space *mapping,
2471                 struct list_head *pages, unsigned nr_pages)
2472 {
2473         struct extent_io_tree *tree;
2474         tree = &BTRFS_I(mapping->host)->io_tree;
2475         return extent_readpages(tree, mapping, pages, nr_pages,
2476                                 btrfs_get_extent);
2477 }
2478
2479 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
2480 {
2481         struct extent_io_tree *tree;
2482         struct extent_map_tree *map;
2483         int ret;
2484
2485         tree = &BTRFS_I(page->mapping->host)->io_tree;
2486         map = &BTRFS_I(page->mapping->host)->extent_tree;
2487         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
2488         if (ret == 1) {
2489                 invalidate_extent_lru(tree, page_offset(page), PAGE_CACHE_SIZE);
2490                 ClearPagePrivate(page);
2491                 set_page_private(page, 0);
2492                 page_cache_release(page);
2493         }
2494         return ret;
2495 }
2496
2497 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
2498 {
2499         struct extent_io_tree *tree;
2500
2501         tree = &BTRFS_I(page->mapping->host)->io_tree;
2502         extent_invalidatepage(tree, page, offset);
2503         btrfs_releasepage(page, GFP_NOFS);
2504         if (PagePrivate(page)) {
2505                 invalidate_extent_lru(tree, page_offset(page), PAGE_CACHE_SIZE);
2506                 ClearPagePrivate(page);
2507                 set_page_private(page, 0);
2508                 page_cache_release(page);
2509         }
2510 }
2511
2512 /*
2513  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
2514  * called from a page fault handler when a page is first dirtied. Hence we must
2515  * be careful to check for EOF conditions here. We set the page up correctly
2516  * for a written page which means we get ENOSPC checking when writing into
2517  * holes and correct delalloc and unwritten extent mapping on filesystems that
2518  * support these features.
2519  *
2520  * We are not allowed to take the i_mutex here so we have to play games to
2521  * protect against truncate races as the page could now be beyond EOF.  Because
2522  * vmtruncate() writes the inode size before removing pages, once we have the
2523  * page lock we can determine safely if the page is beyond EOF. If it is not
2524  * beyond EOF, then the page is guaranteed safe against truncation until we
2525  * unlock the page.
2526  */
2527 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
2528 {
2529         struct inode *inode = fdentry(vma->vm_file)->d_inode;
2530         struct btrfs_root *root = BTRFS_I(inode)->root;
2531         unsigned long end;
2532         loff_t size;
2533         int ret;
2534         u64 page_start;
2535
2536         ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
2537         if (ret)
2538                 goto out;
2539
2540         ret = -EINVAL;
2541
2542         lock_page(page);
2543         wait_on_page_writeback(page);
2544         size = i_size_read(inode);
2545         page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2546
2547         if ((page->mapping != inode->i_mapping) ||
2548             (page_start > size)) {
2549                 /* page got truncated out from underneath us */
2550                 goto out_unlock;
2551         }
2552
2553         /* page is wholly or partially inside EOF */
2554         if (page_start + PAGE_CACHE_SIZE > size)
2555                 end = size & ~PAGE_CACHE_MASK;
2556         else
2557                 end = PAGE_CACHE_SIZE;
2558
2559         ret = btrfs_cow_one_page(inode, page, end);
2560
2561 out_unlock:
2562         unlock_page(page);
2563 out:
2564         return ret;
2565 }
2566
2567 static void btrfs_truncate(struct inode *inode)
2568 {
2569         struct btrfs_root *root = BTRFS_I(inode)->root;
2570         int ret;
2571         struct btrfs_trans_handle *trans;
2572         unsigned long nr;
2573
2574         if (!S_ISREG(inode->i_mode))
2575                 return;
2576         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2577                 return;
2578
2579         btrfs_truncate_page(inode->i_mapping, inode->i_size);
2580
2581         trans = btrfs_start_transaction(root, 1);
2582         btrfs_set_trans_block_group(trans, inode);
2583
2584         /* FIXME, add redo link to tree so we don't leak on crash */
2585         ret = btrfs_truncate_in_trans(trans, root, inode,
2586                                       BTRFS_EXTENT_DATA_KEY);
2587         btrfs_update_inode(trans, root, inode);
2588         nr = trans->blocks_used;
2589
2590         ret = btrfs_end_transaction(trans, root);
2591         BUG_ON(ret);
2592         btrfs_btree_balance_dirty(root, nr);
2593         btrfs_throttle(root);
2594 }
2595
2596 /*
2597  * Invalidate a single dcache entry at the root of the filesystem.
2598  * Needed after creation of snapshot or subvolume.
2599  */
2600 void btrfs_invalidate_dcache_root(struct btrfs_root *root, char *name,
2601                                   int namelen)
2602 {
2603         struct dentry *alias, *entry;
2604         struct qstr qstr;
2605
2606         alias = d_find_alias(root->fs_info->sb->s_root->d_inode);
2607         if (alias) {
2608                 qstr.name = name;
2609                 qstr.len = namelen;
2610                 /* change me if btrfs ever gets a d_hash operation */
2611                 qstr.hash = full_name_hash(qstr.name, qstr.len);
2612                 entry = d_lookup(alias, &qstr);
2613                 dput(alias);
2614                 if (entry) {
2615                         d_invalidate(entry);
2616                         dput(entry);
2617                 }
2618         }
2619 }
2620
2621 int btrfs_create_subvol_root(struct btrfs_root *new_root,
2622                 struct btrfs_trans_handle *trans, u64 new_dirid,
2623                 struct btrfs_block_group_cache *block_group)
2624 {
2625         struct inode *inode;
2626         int ret;
2627
2628         inode = btrfs_new_inode(trans, new_root, "..", 2, new_dirid,
2629                                 new_dirid, block_group, S_IFDIR | 0700);
2630         if (IS_ERR(inode))
2631                 return PTR_ERR(inode);
2632         inode->i_op = &btrfs_dir_inode_operations;
2633         inode->i_fop = &btrfs_dir_file_operations;
2634         new_root->inode = inode;
2635
2636         ret = btrfs_insert_inode_ref(trans, new_root, "..", 2, new_dirid,
2637                                      new_dirid);
2638         inode->i_nlink = 1;
2639         inode->i_size = 0;
2640
2641         return btrfs_update_inode(trans, new_root, inode);
2642 }
2643
2644 unsigned long btrfs_force_ra(struct address_space *mapping,
2645                               struct file_ra_state *ra, struct file *file,
2646                               pgoff_t offset, pgoff_t last_index)
2647 {
2648         pgoff_t req_size = last_index - offset + 1;
2649
2650 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2651         offset = page_cache_readahead(mapping, ra, file, offset, req_size);
2652         return offset;
2653 #else
2654         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
2655         return offset + req_size;
2656 #endif
2657 }
2658
2659 struct inode *btrfs_alloc_inode(struct super_block *sb)
2660 {
2661         struct btrfs_inode *ei;
2662
2663         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
2664         if (!ei)
2665                 return NULL;
2666         ei->last_trans = 0;
2667         ei->ordered_trans = 0;
2668         return &ei->vfs_inode;
2669 }
2670
2671 void btrfs_destroy_inode(struct inode *inode)
2672 {
2673         WARN_ON(!list_empty(&inode->i_dentry));
2674         WARN_ON(inode->i_data.nrpages);
2675
2676         btrfs_drop_extent_cache(inode, 0, (u64)-1);
2677         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
2678 }
2679
2680 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2681 static void init_once(struct kmem_cache * cachep, void *foo)
2682 #else
2683 static void init_once(void * foo, struct kmem_cache * cachep,
2684                       unsigned long flags)
2685 #endif
2686 {
2687         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
2688
2689         inode_init_once(&ei->vfs_inode);
2690 }
2691
2692 void btrfs_destroy_cachep(void)
2693 {
2694         if (btrfs_inode_cachep)
2695                 kmem_cache_destroy(btrfs_inode_cachep);
2696         if (btrfs_trans_handle_cachep)
2697                 kmem_cache_destroy(btrfs_trans_handle_cachep);
2698         if (btrfs_transaction_cachep)
2699                 kmem_cache_destroy(btrfs_transaction_cachep);
2700         if (btrfs_bit_radix_cachep)
2701                 kmem_cache_destroy(btrfs_bit_radix_cachep);
2702         if (btrfs_path_cachep)
2703                 kmem_cache_destroy(btrfs_path_cachep);
2704 }
2705
2706 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
2707                                        unsigned long extra_flags,
2708 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2709                                        void (*ctor)(struct kmem_cache *, void *)
2710 #else
2711                                        void (*ctor)(void *, struct kmem_cache *,
2712                                                     unsigned long)
2713 #endif
2714                                      )
2715 {
2716         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
2717                                  SLAB_MEM_SPREAD | extra_flags), ctor
2718 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2719                                  ,NULL
2720 #endif
2721                                 );
2722 }
2723
2724 int btrfs_init_cachep(void)
2725 {
2726         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
2727                                           sizeof(struct btrfs_inode),
2728                                           0, init_once);
2729         if (!btrfs_inode_cachep)
2730                 goto fail;
2731         btrfs_trans_handle_cachep =
2732                         btrfs_cache_create("btrfs_trans_handle_cache",
2733                                            sizeof(struct btrfs_trans_handle),
2734                                            0, NULL);
2735         if (!btrfs_trans_handle_cachep)
2736                 goto fail;
2737         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
2738                                              sizeof(struct btrfs_transaction),
2739                                              0, NULL);
2740         if (!btrfs_transaction_cachep)
2741                 goto fail;
2742         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
2743                                          sizeof(struct btrfs_path),
2744                                          0, NULL);
2745         if (!btrfs_path_cachep)
2746                 goto fail;
2747         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
2748                                               SLAB_DESTROY_BY_RCU, NULL);
2749         if (!btrfs_bit_radix_cachep)
2750                 goto fail;
2751         return 0;
2752 fail:
2753         btrfs_destroy_cachep();
2754         return -ENOMEM;
2755 }
2756
2757 static int btrfs_getattr(struct vfsmount *mnt,
2758                          struct dentry *dentry, struct kstat *stat)
2759 {
2760         struct inode *inode = dentry->d_inode;
2761         generic_fillattr(inode, stat);
2762         stat->blksize = PAGE_CACHE_SIZE;
2763         stat->blocks = inode->i_blocks + (BTRFS_I(inode)->delalloc_bytes >> 9);
2764         return 0;
2765 }
2766
2767 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
2768                            struct inode * new_dir,struct dentry *new_dentry)
2769 {
2770         struct btrfs_trans_handle *trans;
2771         struct btrfs_root *root = BTRFS_I(old_dir)->root;
2772         struct inode *new_inode = new_dentry->d_inode;
2773         struct inode *old_inode = old_dentry->d_inode;
2774         struct timespec ctime = CURRENT_TIME;
2775         int ret;
2776
2777         if (S_ISDIR(old_inode->i_mode) && new_inode &&
2778             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
2779                 return -ENOTEMPTY;
2780         }
2781
2782         ret = btrfs_check_free_space(root, 1, 0);
2783         if (ret)
2784                 goto out_unlock;
2785
2786         trans = btrfs_start_transaction(root, 1);
2787
2788         btrfs_set_trans_block_group(trans, new_dir);
2789
2790         old_dentry->d_inode->i_nlink++;
2791         old_dir->i_ctime = old_dir->i_mtime = ctime;
2792         new_dir->i_ctime = new_dir->i_mtime = ctime;
2793         old_inode->i_ctime = ctime;
2794
2795         ret = btrfs_unlink_trans(trans, root, old_dir, old_dentry);
2796         if (ret)
2797                 goto out_fail;
2798
2799         if (new_inode) {
2800                 new_inode->i_ctime = CURRENT_TIME;
2801                 ret = btrfs_unlink_trans(trans, root, new_dir, new_dentry);
2802                 if (ret)
2803                         goto out_fail;
2804         }
2805         ret = btrfs_add_link(trans, new_dentry, old_inode, 1);
2806         if (ret)
2807                 goto out_fail;
2808
2809 out_fail:
2810         btrfs_end_transaction(trans, root);
2811 out_unlock:
2812         return ret;
2813 }
2814
2815 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
2816                          const char *symname)
2817 {
2818         struct btrfs_trans_handle *trans;
2819         struct btrfs_root *root = BTRFS_I(dir)->root;
2820         struct btrfs_path *path;
2821         struct btrfs_key key;
2822         struct inode *inode = NULL;
2823         int err;
2824         int drop_inode = 0;
2825         u64 objectid;
2826         int name_len;
2827         int datasize;
2828         unsigned long ptr;
2829         struct btrfs_file_extent_item *ei;
2830         struct extent_buffer *leaf;
2831         unsigned long nr = 0;
2832
2833         name_len = strlen(symname) + 1;
2834         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
2835                 return -ENAMETOOLONG;
2836
2837         err = btrfs_check_free_space(root, 1, 0);
2838         if (err)
2839                 goto out_fail;
2840
2841         trans = btrfs_start_transaction(root, 1);
2842         btrfs_set_trans_block_group(trans, dir);
2843
2844         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2845         if (err) {
2846                 err = -ENOSPC;
2847                 goto out_unlock;
2848         }
2849
2850         inode = btrfs_new_inode(trans, root, dentry->d_name.name,
2851                                 dentry->d_name.len,
2852                                 dentry->d_parent->d_inode->i_ino, objectid,
2853                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO);
2854         err = PTR_ERR(inode);
2855         if (IS_ERR(inode))
2856                 goto out_unlock;
2857
2858         btrfs_set_trans_block_group(trans, inode);
2859         err = btrfs_add_nondir(trans, dentry, inode, 0);
2860         if (err)
2861                 drop_inode = 1;
2862         else {
2863                 inode->i_mapping->a_ops = &btrfs_aops;
2864                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2865                 inode->i_fop = &btrfs_file_operations;
2866                 inode->i_op = &btrfs_file_inode_operations;
2867                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2868                 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2869                                      inode->i_mapping, GFP_NOFS);
2870                 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2871                                      inode->i_mapping, GFP_NOFS);
2872                 BTRFS_I(inode)->delalloc_bytes = 0;
2873                 atomic_set(&BTRFS_I(inode)->ordered_writeback, 0);
2874                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2875         }
2876         dir->i_sb->s_dirt = 1;
2877         btrfs_update_inode_block_group(trans, inode);
2878         btrfs_update_inode_block_group(trans, dir);
2879         if (drop_inode)
2880                 goto out_unlock;
2881
2882         path = btrfs_alloc_path();
2883         BUG_ON(!path);
2884         key.objectid = inode->i_ino;
2885         key.offset = 0;
2886         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
2887         datasize = btrfs_file_extent_calc_inline_size(name_len);
2888         err = btrfs_insert_empty_item(trans, root, path, &key,
2889                                       datasize);
2890         if (err) {
2891                 drop_inode = 1;
2892                 goto out_unlock;
2893         }
2894         leaf = path->nodes[0];
2895         ei = btrfs_item_ptr(leaf, path->slots[0],
2896                             struct btrfs_file_extent_item);
2897         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
2898         btrfs_set_file_extent_type(leaf, ei,
2899                                    BTRFS_FILE_EXTENT_INLINE);
2900         ptr = btrfs_file_extent_inline_start(ei);
2901         write_extent_buffer(leaf, symname, ptr, name_len);
2902         btrfs_mark_buffer_dirty(leaf);
2903         btrfs_free_path(path);
2904
2905         inode->i_op = &btrfs_symlink_inode_operations;
2906         inode->i_mapping->a_ops = &btrfs_symlink_aops;
2907         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2908         inode->i_size = name_len - 1;
2909         err = btrfs_update_inode(trans, root, inode);
2910         if (err)
2911                 drop_inode = 1;
2912
2913 out_unlock:
2914         nr = trans->blocks_used;
2915         btrfs_end_transaction(trans, root);
2916 out_fail:
2917         if (drop_inode) {
2918                 inode_dec_link_count(inode);
2919                 iput(inode);
2920         }
2921         btrfs_btree_balance_dirty(root, nr);
2922         btrfs_throttle(root);
2923         return err;
2924 }
2925
2926 static int btrfs_permission(struct inode *inode, int mask,
2927                             struct nameidata *nd)
2928 {
2929         if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
2930                 return -EACCES;
2931         return generic_permission(inode, mask, NULL);
2932 }
2933
2934 static struct inode_operations btrfs_dir_inode_operations = {
2935         .lookup         = btrfs_lookup,
2936         .create         = btrfs_create,
2937         .unlink         = btrfs_unlink,
2938         .link           = btrfs_link,
2939         .mkdir          = btrfs_mkdir,
2940         .rmdir          = btrfs_rmdir,
2941         .rename         = btrfs_rename,
2942         .symlink        = btrfs_symlink,
2943         .setattr        = btrfs_setattr,
2944         .mknod          = btrfs_mknod,
2945         .setxattr       = generic_setxattr,
2946         .getxattr       = generic_getxattr,
2947         .listxattr      = btrfs_listxattr,
2948         .removexattr    = generic_removexattr,
2949         .permission     = btrfs_permission,
2950 };
2951 static struct inode_operations btrfs_dir_ro_inode_operations = {
2952         .lookup         = btrfs_lookup,
2953         .permission     = btrfs_permission,
2954 };
2955 static struct file_operations btrfs_dir_file_operations = {
2956         .llseek         = generic_file_llseek,
2957         .read           = generic_read_dir,
2958         .readdir        = btrfs_readdir,
2959         .unlocked_ioctl = btrfs_ioctl,
2960 #ifdef CONFIG_COMPAT
2961         .compat_ioctl   = btrfs_ioctl,
2962 #endif
2963         .release        = btrfs_release_file,
2964 };
2965
2966 static struct extent_io_ops btrfs_extent_io_ops = {
2967         .fill_delalloc = run_delalloc_range,
2968         .submit_bio_hook = btrfs_submit_bio_hook,
2969         .merge_bio_hook = btrfs_merge_bio_hook,
2970         .readpage_io_hook = btrfs_readpage_io_hook,
2971         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
2972         .readpage_io_failed_hook = btrfs_io_failed_hook,
2973         .set_bit_hook = btrfs_set_bit_hook,
2974         .clear_bit_hook = btrfs_clear_bit_hook,
2975 };
2976
2977 static struct address_space_operations btrfs_aops = {
2978         .readpage       = btrfs_readpage,
2979         .writepage      = btrfs_writepage,
2980         .writepages     = btrfs_writepages,
2981         .readpages      = btrfs_readpages,
2982         .sync_page      = block_sync_page,
2983         .bmap           = btrfs_bmap,
2984         .direct_IO      = btrfs_direct_IO,
2985         .invalidatepage = btrfs_invalidatepage,
2986         .releasepage    = btrfs_releasepage,
2987         .set_page_dirty = __set_page_dirty_nobuffers,
2988 };
2989
2990 static struct address_space_operations btrfs_symlink_aops = {
2991         .readpage       = btrfs_readpage,
2992         .writepage      = btrfs_writepage,
2993         .invalidatepage = btrfs_invalidatepage,
2994         .releasepage    = btrfs_releasepage,
2995 };
2996
2997 static struct inode_operations btrfs_file_inode_operations = {
2998         .truncate       = btrfs_truncate,
2999         .getattr        = btrfs_getattr,
3000         .setattr        = btrfs_setattr,
3001         .setxattr       = generic_setxattr,
3002         .getxattr       = generic_getxattr,
3003         .listxattr      = btrfs_listxattr,
3004         .removexattr    = generic_removexattr,
3005         .permission     = btrfs_permission,
3006 };
3007 static struct inode_operations btrfs_special_inode_operations = {
3008         .getattr        = btrfs_getattr,
3009         .setattr        = btrfs_setattr,
3010         .permission     = btrfs_permission,
3011 };
3012 static struct inode_operations btrfs_symlink_inode_operations = {
3013         .readlink       = generic_readlink,
3014         .follow_link    = page_follow_link_light,
3015         .put_link       = page_put_link,
3016         .permission     = btrfs_permission,
3017 };