Btrfs: online shrinking fixes
[linux-2.6] / 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/buffer_head.h>
20 #include <linux/fs.h>
21 #include <linux/pagemap.h>
22 #include <linux/highmem.h>
23 #include <linux/time.h>
24 #include <linux/init.h>
25 #include <linux/string.h>
26 #include <linux/smp_lock.h>
27 #include <linux/backing-dev.h>
28 #include <linux/mpage.h>
29 #include <linux/swap.h>
30 #include <linux/writeback.h>
31 #include <linux/statfs.h>
32 #include <linux/compat.h>
33 #include <linux/bit_spinlock.h>
34 #include <linux/version.h>
35 #include <linux/xattr.h>
36 #include "ctree.h"
37 #include "disk-io.h"
38 #include "transaction.h"
39 #include "btrfs_inode.h"
40 #include "ioctl.h"
41 #include "print-tree.h"
42
43 struct btrfs_iget_args {
44         u64 ino;
45         struct btrfs_root *root;
46 };
47
48 static struct inode_operations btrfs_dir_inode_operations;
49 static struct inode_operations btrfs_symlink_inode_operations;
50 static struct inode_operations btrfs_dir_ro_inode_operations;
51 static struct inode_operations btrfs_special_inode_operations;
52 static struct inode_operations btrfs_file_inode_operations;
53 static struct address_space_operations btrfs_aops;
54 static struct address_space_operations btrfs_symlink_aops;
55 static struct file_operations btrfs_dir_file_operations;
56 static struct extent_map_ops btrfs_extent_map_ops;
57
58 static struct kmem_cache *btrfs_inode_cachep;
59 struct kmem_cache *btrfs_trans_handle_cachep;
60 struct kmem_cache *btrfs_transaction_cachep;
61 struct kmem_cache *btrfs_bit_radix_cachep;
62 struct kmem_cache *btrfs_path_cachep;
63
64 #define S_SHIFT 12
65 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
66         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
67         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
68         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
69         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
70         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
71         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
72         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
73 };
74
75 int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
76                            int for_del)
77 {
78         u64 total = btrfs_super_total_bytes(&root->fs_info->super_copy);
79         u64 used = btrfs_super_bytes_used(&root->fs_info->super_copy);
80         u64 thresh;
81         int ret = 0;
82
83         if (for_del)
84                 thresh = total * 90;
85         else
86                 thresh = total * 85;
87
88         do_div(thresh, 100);
89
90         spin_lock(&root->fs_info->delalloc_lock);
91         if (used + root->fs_info->delalloc_bytes + num_required > thresh)
92                 ret = -ENOSPC;
93         spin_unlock(&root->fs_info->delalloc_lock);
94         return ret;
95 }
96
97 static int cow_file_range(struct inode *inode, u64 start, u64 end)
98 {
99         struct btrfs_root *root = BTRFS_I(inode)->root;
100         struct btrfs_trans_handle *trans;
101         u64 alloc_hint = 0;
102         u64 num_bytes;
103         u64 cur_alloc_size;
104         u64 blocksize = root->sectorsize;
105         struct btrfs_key ins;
106         int ret;
107
108         trans = btrfs_start_transaction(root, 1);
109         BUG_ON(!trans);
110         btrfs_set_trans_block_group(trans, inode);
111
112         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
113         num_bytes = max(blocksize,  num_bytes);
114         ret = btrfs_drop_extents(trans, root, inode,
115                                  start, start + num_bytes, start, &alloc_hint);
116
117         if (alloc_hint == EXTENT_MAP_INLINE)
118                 goto out;
119
120         while(num_bytes > 0) {
121                 cur_alloc_size = min(num_bytes, root->fs_info->max_extent);
122                 ret = btrfs_alloc_extent(trans, root, cur_alloc_size,
123                                          root->root_key.objectid,
124                                          trans->transid,
125                                          inode->i_ino, start, 0,
126                                          alloc_hint, (u64)-1, &ins, 1);
127                 if (ret) {
128                         WARN_ON(1);
129                         goto out;
130                 }
131                 ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
132                                                start, ins.objectid, ins.offset,
133                                                ins.offset);
134                 num_bytes -= cur_alloc_size;
135                 alloc_hint = ins.objectid + ins.offset;
136                 start += cur_alloc_size;
137         }
138         btrfs_add_ordered_inode(inode);
139 out:
140         btrfs_end_transaction(trans, root);
141         return ret;
142 }
143
144 static int run_delalloc_nocow(struct inode *inode, u64 start, u64 end)
145 {
146         u64 extent_start;
147         u64 extent_end;
148         u64 bytenr;
149         u64 cow_end;
150         u64 loops = 0;
151         u64 total_fs_bytes;
152         struct btrfs_root *root = BTRFS_I(inode)->root;
153         struct extent_buffer *leaf;
154         int found_type;
155         struct btrfs_path *path;
156         struct btrfs_file_extent_item *item;
157         int ret;
158         int err;
159         struct btrfs_key found_key;
160
161         total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
162         path = btrfs_alloc_path();
163         BUG_ON(!path);
164 again:
165         ret = btrfs_lookup_file_extent(NULL, root, path,
166                                        inode->i_ino, start, 0);
167         if (ret < 0) {
168                 btrfs_free_path(path);
169                 return ret;
170         }
171
172         cow_end = end;
173         if (ret != 0) {
174                 if (path->slots[0] == 0)
175                         goto not_found;
176                 path->slots[0]--;
177         }
178
179         leaf = path->nodes[0];
180         item = btrfs_item_ptr(leaf, path->slots[0],
181                               struct btrfs_file_extent_item);
182
183         /* are we inside the extent that was found? */
184         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
185         found_type = btrfs_key_type(&found_key);
186         if (found_key.objectid != inode->i_ino ||
187             found_type != BTRFS_EXTENT_DATA_KEY) {
188                 goto not_found;
189         }
190
191         found_type = btrfs_file_extent_type(leaf, item);
192         extent_start = found_key.offset;
193         if (found_type == BTRFS_FILE_EXTENT_REG) {
194                 u64 extent_num_bytes;
195
196                 extent_num_bytes = btrfs_file_extent_num_bytes(leaf, item);
197                 extent_end = extent_start + extent_num_bytes;
198                 err = 0;
199
200                 if (loops && start != extent_start)
201                         goto not_found;
202
203                 if (start < extent_start || start >= extent_end)
204                         goto not_found;
205
206                 cow_end = min(end, extent_end - 1);
207                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
208                 if (bytenr == 0)
209                         goto not_found;
210
211                 /*
212                  * we may be called by the resizer, make sure we're inside
213                  * the limits of the FS
214                  */
215                 if (bytenr + extent_num_bytes > total_fs_bytes)
216                         goto not_found;
217
218                 if (btrfs_count_snapshots_in_path(root, path, bytenr) != 1) {
219                         goto not_found;
220                 }
221
222                 start = extent_end;
223         } else {
224                 goto not_found;
225         }
226 loop:
227         if (start > end) {
228                 btrfs_free_path(path);
229                 return 0;
230         }
231         btrfs_release_path(root, path);
232         loops++;
233         goto again;
234
235 not_found:
236         cow_file_range(inode, start, cow_end);
237         start = cow_end + 1;
238         goto loop;
239 }
240
241 static int run_delalloc_range(struct inode *inode, u64 start, u64 end)
242 {
243         struct btrfs_root *root = BTRFS_I(inode)->root;
244         u64 num_bytes;
245         int ret;
246
247         mutex_lock(&root->fs_info->fs_mutex);
248         if (btrfs_test_opt(root, NODATACOW))
249                 ret = run_delalloc_nocow(inode, start, end);
250         else
251                 ret = cow_file_range(inode, start, end);
252
253         spin_lock(&root->fs_info->delalloc_lock);
254         num_bytes = end + 1 - start;
255         if (root->fs_info->delalloc_bytes < num_bytes) {
256                 printk("delalloc accounting error total %llu sub %llu\n",
257                        root->fs_info->delalloc_bytes, num_bytes);
258         } else {
259                 root->fs_info->delalloc_bytes -= num_bytes;
260         }
261         spin_unlock(&root->fs_info->delalloc_lock);
262
263         mutex_unlock(&root->fs_info->fs_mutex);
264         return ret;
265 }
266
267 int btrfs_writepage_io_hook(struct page *page, u64 start, u64 end)
268 {
269         struct inode *inode = page->mapping->host;
270         struct btrfs_root *root = BTRFS_I(inode)->root;
271         struct btrfs_trans_handle *trans;
272         char *kaddr;
273         int ret = 0;
274         u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
275         size_t offset = start - page_start;
276
277         if (btrfs_test_opt(root, NODATASUM))
278                 return 0;
279
280         mutex_lock(&root->fs_info->fs_mutex);
281         trans = btrfs_start_transaction(root, 1);
282         btrfs_set_trans_block_group(trans, inode);
283         kaddr = kmap(page);
284         btrfs_csum_file_block(trans, root, inode, inode->i_ino,
285                               start, kaddr + offset, end - start + 1);
286         kunmap(page);
287         ret = btrfs_end_transaction(trans, root);
288         BUG_ON(ret);
289         mutex_unlock(&root->fs_info->fs_mutex);
290         return ret;
291 }
292
293 int btrfs_readpage_io_hook(struct page *page, u64 start, u64 end)
294 {
295         int ret = 0;
296         struct inode *inode = page->mapping->host;
297         struct btrfs_root *root = BTRFS_I(inode)->root;
298         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
299         struct btrfs_csum_item *item;
300         struct btrfs_path *path = NULL;
301         u32 csum;
302
303         if (btrfs_test_opt(root, NODATASUM))
304                 return 0;
305
306         mutex_lock(&root->fs_info->fs_mutex);
307         path = btrfs_alloc_path();
308         item = btrfs_lookup_csum(NULL, root, path, inode->i_ino, start, 0);
309         if (IS_ERR(item)) {
310                 ret = PTR_ERR(item);
311                 /* a csum that isn't present is a preallocated region. */
312                 if (ret == -ENOENT || ret == -EFBIG)
313                         ret = 0;
314                 csum = 0;
315                 goto out;
316         }
317         read_extent_buffer(path->nodes[0], &csum, (unsigned long)item,
318                            BTRFS_CRC32_SIZE);
319         set_state_private(em_tree, start, csum);
320 out:
321         if (path)
322                 btrfs_free_path(path);
323         mutex_unlock(&root->fs_info->fs_mutex);
324         return ret;
325 }
326
327 int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end)
328 {
329         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
330         struct inode *inode = page->mapping->host;
331         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
332         char *kaddr;
333         u64 private;
334         int ret;
335         struct btrfs_root *root = BTRFS_I(inode)->root;
336         u32 csum = ~(u32)0;
337         unsigned long flags;
338
339         if (btrfs_test_opt(root, NODATASUM))
340                 return 0;
341
342         ret = get_state_private(em_tree, start, &private);
343         local_irq_save(flags);
344         kaddr = kmap_atomic(page, KM_IRQ0);
345         if (ret) {
346                 goto zeroit;
347         }
348         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
349         btrfs_csum_final(csum, (char *)&csum);
350         if (csum != private) {
351                 goto zeroit;
352         }
353         kunmap_atomic(kaddr, KM_IRQ0);
354         local_irq_restore(flags);
355         return 0;
356
357 zeroit:
358         printk("btrfs csum failed ino %lu off %llu\n",
359                page->mapping->host->i_ino, (unsigned long long)start);
360         memset(kaddr + offset, 1, end - start + 1);
361         flush_dcache_page(page);
362         kunmap_atomic(kaddr, KM_IRQ0);
363         local_irq_restore(flags);
364         return 0;
365 }
366
367 void btrfs_read_locked_inode(struct inode *inode)
368 {
369         struct btrfs_path *path;
370         struct extent_buffer *leaf;
371         struct btrfs_inode_item *inode_item;
372         struct btrfs_inode_timespec *tspec;
373         struct btrfs_root *root = BTRFS_I(inode)->root;
374         struct btrfs_key location;
375         u64 alloc_group_block;
376         u32 rdev;
377         int ret;
378
379         path = btrfs_alloc_path();
380         BUG_ON(!path);
381         mutex_lock(&root->fs_info->fs_mutex);
382         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
383
384         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
385         if (ret)
386                 goto make_bad;
387
388         leaf = path->nodes[0];
389         inode_item = btrfs_item_ptr(leaf, path->slots[0],
390                                     struct btrfs_inode_item);
391
392         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
393         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
394         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
395         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
396         inode->i_size = btrfs_inode_size(leaf, inode_item);
397
398         tspec = btrfs_inode_atime(inode_item);
399         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
400         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
401
402         tspec = btrfs_inode_mtime(inode_item);
403         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
404         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
405
406         tspec = btrfs_inode_ctime(inode_item);
407         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
408         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
409
410         inode->i_blocks = btrfs_inode_nblocks(leaf, inode_item);
411         inode->i_generation = btrfs_inode_generation(leaf, inode_item);
412         inode->i_rdev = 0;
413         rdev = btrfs_inode_rdev(leaf, inode_item);
414
415         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
416         BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
417                                                        alloc_group_block);
418
419         if (!BTRFS_I(inode)->block_group) {
420                 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
421                                                          NULL, 0, 0, 0);
422         }
423         btrfs_free_path(path);
424         inode_item = NULL;
425
426         mutex_unlock(&root->fs_info->fs_mutex);
427
428         switch (inode->i_mode & S_IFMT) {
429         case S_IFREG:
430                 inode->i_mapping->a_ops = &btrfs_aops;
431                 BTRFS_I(inode)->extent_tree.ops = &btrfs_extent_map_ops;
432                 inode->i_fop = &btrfs_file_operations;
433                 inode->i_op = &btrfs_file_inode_operations;
434                 break;
435         case S_IFDIR:
436                 inode->i_fop = &btrfs_dir_file_operations;
437                 if (root == root->fs_info->tree_root)
438                         inode->i_op = &btrfs_dir_ro_inode_operations;
439                 else
440                         inode->i_op = &btrfs_dir_inode_operations;
441                 break;
442         case S_IFLNK:
443                 inode->i_op = &btrfs_symlink_inode_operations;
444                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
445                 break;
446         default:
447                 init_special_inode(inode, inode->i_mode, rdev);
448                 break;
449         }
450         return;
451
452 make_bad:
453         btrfs_release_path(root, path);
454         btrfs_free_path(path);
455         mutex_unlock(&root->fs_info->fs_mutex);
456         make_bad_inode(inode);
457 }
458
459 static void fill_inode_item(struct extent_buffer *leaf,
460                             struct btrfs_inode_item *item,
461                             struct inode *inode)
462 {
463         btrfs_set_inode_uid(leaf, item, inode->i_uid);
464         btrfs_set_inode_gid(leaf, item, inode->i_gid);
465         btrfs_set_inode_size(leaf, item, inode->i_size);
466         btrfs_set_inode_mode(leaf, item, inode->i_mode);
467         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
468
469         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
470                                inode->i_atime.tv_sec);
471         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
472                                 inode->i_atime.tv_nsec);
473
474         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
475                                inode->i_mtime.tv_sec);
476         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
477                                 inode->i_mtime.tv_nsec);
478
479         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
480                                inode->i_ctime.tv_sec);
481         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
482                                 inode->i_ctime.tv_nsec);
483
484         btrfs_set_inode_nblocks(leaf, item, inode->i_blocks);
485         btrfs_set_inode_generation(leaf, item, inode->i_generation);
486         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
487         btrfs_set_inode_block_group(leaf, item,
488                                     BTRFS_I(inode)->block_group->key.objectid);
489 }
490
491 int btrfs_update_inode(struct btrfs_trans_handle *trans,
492                               struct btrfs_root *root,
493                               struct inode *inode)
494 {
495         struct btrfs_inode_item *inode_item;
496         struct btrfs_path *path;
497         struct extent_buffer *leaf;
498         int ret;
499
500         path = btrfs_alloc_path();
501         BUG_ON(!path);
502         ret = btrfs_lookup_inode(trans, root, path,
503                                  &BTRFS_I(inode)->location, 1);
504         if (ret) {
505                 if (ret > 0)
506                         ret = -ENOENT;
507                 goto failed;
508         }
509
510         leaf = path->nodes[0];
511         inode_item = btrfs_item_ptr(leaf, path->slots[0],
512                                   struct btrfs_inode_item);
513
514         fill_inode_item(leaf, inode_item, inode);
515         btrfs_mark_buffer_dirty(leaf);
516         btrfs_set_inode_last_trans(trans, inode);
517         ret = 0;
518 failed:
519         btrfs_release_path(root, path);
520         btrfs_free_path(path);
521         return ret;
522 }
523
524
525 static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
526                               struct btrfs_root *root,
527                               struct inode *dir,
528                               struct dentry *dentry)
529 {
530         struct btrfs_path *path;
531         const char *name = dentry->d_name.name;
532         int name_len = dentry->d_name.len;
533         int ret = 0;
534         struct extent_buffer *leaf;
535         struct btrfs_dir_item *di;
536         struct btrfs_key key;
537
538         path = btrfs_alloc_path();
539         if (!path) {
540                 ret = -ENOMEM;
541                 goto err;
542         }
543
544         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
545                                     name, name_len, -1);
546         if (IS_ERR(di)) {
547                 ret = PTR_ERR(di);
548                 goto err;
549         }
550         if (!di) {
551                 ret = -ENOENT;
552                 goto err;
553         }
554         leaf = path->nodes[0];
555         btrfs_dir_item_key_to_cpu(leaf, di, &key);
556         ret = btrfs_delete_one_dir_name(trans, root, path, di);
557         if (ret)
558                 goto err;
559         btrfs_release_path(root, path);
560
561         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
562                                          key.objectid, name, name_len, -1);
563         if (IS_ERR(di)) {
564                 ret = PTR_ERR(di);
565                 goto err;
566         }
567         if (!di) {
568                 ret = -ENOENT;
569                 goto err;
570         }
571         ret = btrfs_delete_one_dir_name(trans, root, path, di);
572
573         dentry->d_inode->i_ctime = dir->i_ctime;
574         ret = btrfs_del_inode_ref(trans, root, name, name_len,
575                                   dentry->d_inode->i_ino,
576                                   dentry->d_parent->d_inode->i_ino);
577         if (ret) {
578                 printk("failed to delete reference to %.*s, "
579                        "inode %lu parent %lu\n", name_len, name,
580                        dentry->d_inode->i_ino,
581                        dentry->d_parent->d_inode->i_ino);
582         }
583 err:
584         btrfs_free_path(path);
585         if (!ret) {
586                 dir->i_size -= name_len * 2;
587                 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
588                 btrfs_update_inode(trans, root, dir);
589 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
590                 dentry->d_inode->i_nlink--;
591 #else
592                 drop_nlink(dentry->d_inode);
593 #endif
594                 ret = btrfs_update_inode(trans, root, dentry->d_inode);
595                 dir->i_sb->s_dirt = 1;
596         }
597         return ret;
598 }
599
600 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
601 {
602         struct btrfs_root *root;
603         struct btrfs_trans_handle *trans;
604         int ret;
605         unsigned long nr = 0;
606
607         root = BTRFS_I(dir)->root;
608         mutex_lock(&root->fs_info->fs_mutex);
609
610         ret = btrfs_check_free_space(root, 1, 1);
611         if (ret)
612                 goto fail;
613
614         trans = btrfs_start_transaction(root, 1);
615
616         btrfs_set_trans_block_group(trans, dir);
617         ret = btrfs_unlink_trans(trans, root, dir, dentry);
618         nr = trans->blocks_used;
619
620         btrfs_end_transaction(trans, root);
621 fail:
622         mutex_unlock(&root->fs_info->fs_mutex);
623         btrfs_btree_balance_dirty(root, nr);
624         btrfs_throttle(root);
625         return ret;
626 }
627
628 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
629 {
630         struct inode *inode = dentry->d_inode;
631         int err = 0;
632         int ret;
633         struct btrfs_root *root = BTRFS_I(dir)->root;
634         struct btrfs_trans_handle *trans;
635         unsigned long nr = 0;
636
637         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
638                 return -ENOTEMPTY;
639
640         mutex_lock(&root->fs_info->fs_mutex);
641         ret = btrfs_check_free_space(root, 1, 1);
642         if (ret)
643                 goto fail;
644
645         trans = btrfs_start_transaction(root, 1);
646         btrfs_set_trans_block_group(trans, dir);
647
648         /* now the directory is empty */
649         err = btrfs_unlink_trans(trans, root, dir, dentry);
650         if (!err) {
651                 inode->i_size = 0;
652         }
653
654         nr = trans->blocks_used;
655         ret = btrfs_end_transaction(trans, root);
656 fail:
657         mutex_unlock(&root->fs_info->fs_mutex);
658         btrfs_btree_balance_dirty(root, nr);
659         btrfs_throttle(root);
660
661         if (ret && !err)
662                 err = ret;
663         return err;
664 }
665
666 static int btrfs_free_inode(struct btrfs_trans_handle *trans,
667                             struct btrfs_root *root,
668                             struct inode *inode)
669 {
670         struct btrfs_path *path;
671         int ret;
672
673         clear_inode(inode);
674
675         path = btrfs_alloc_path();
676         BUG_ON(!path);
677         ret = btrfs_lookup_inode(trans, root, path,
678                                  &BTRFS_I(inode)->location, -1);
679         if (ret > 0)
680                 ret = -ENOENT;
681         if (!ret)
682                 ret = btrfs_del_item(trans, root, path);
683         btrfs_free_path(path);
684         return ret;
685 }
686
687 /*
688  * this can truncate away extent items, csum items and directory items.
689  * It starts at a high offset and removes keys until it can't find
690  * any higher than i_size.
691  *
692  * csum items that cross the new i_size are truncated to the new size
693  * as well.
694  */
695 static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
696                                    struct btrfs_root *root,
697                                    struct inode *inode)
698 {
699         int ret;
700         struct btrfs_path *path;
701         struct btrfs_key key;
702         struct btrfs_key found_key;
703         u32 found_type;
704         struct extent_buffer *leaf;
705         struct btrfs_file_extent_item *fi;
706         u64 extent_start = 0;
707         u64 extent_num_bytes = 0;
708         u64 item_end = 0;
709         u64 root_gen = 0;
710         u64 root_owner = 0;
711         int found_extent;
712         int del_item;
713         int extent_type = -1;
714
715         btrfs_drop_extent_cache(inode, inode->i_size, (u64)-1);
716         path = btrfs_alloc_path();
717         path->reada = -1;
718         BUG_ON(!path);
719
720         /* FIXME, add redo link to tree so we don't leak on crash */
721         key.objectid = inode->i_ino;
722         key.offset = (u64)-1;
723         key.type = (u8)-1;
724
725         while(1) {
726                 btrfs_init_path(path);
727                 fi = NULL;
728                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
729                 if (ret < 0) {
730                         goto error;
731                 }
732                 if (ret > 0) {
733                         BUG_ON(path->slots[0] == 0);
734                         path->slots[0]--;
735                 }
736                 leaf = path->nodes[0];
737                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
738                 found_type = btrfs_key_type(&found_key);
739
740                 if (found_key.objectid != inode->i_ino)
741                         break;
742
743                 if (found_type != BTRFS_CSUM_ITEM_KEY &&
744                     found_type != BTRFS_DIR_ITEM_KEY &&
745                     found_type != BTRFS_DIR_INDEX_KEY &&
746                     found_type != BTRFS_EXTENT_DATA_KEY)
747                         break;
748
749                 item_end = found_key.offset;
750                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
751                         fi = btrfs_item_ptr(leaf, path->slots[0],
752                                             struct btrfs_file_extent_item);
753                         extent_type = btrfs_file_extent_type(leaf, fi);
754                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
755                                 item_end +=
756                                     btrfs_file_extent_num_bytes(leaf, fi);
757                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
758                                 struct btrfs_item *item = btrfs_item_nr(leaf,
759                                                                 path->slots[0]);
760                                 item_end += btrfs_file_extent_inline_len(leaf,
761                                                                          item);
762                         }
763                         item_end--;
764                 }
765                 if (found_type == BTRFS_CSUM_ITEM_KEY) {
766                         ret = btrfs_csum_truncate(trans, root, path,
767                                                   inode->i_size);
768                         BUG_ON(ret);
769                 }
770                 if (item_end < inode->i_size) {
771                         if (found_type == BTRFS_DIR_ITEM_KEY) {
772                                 found_type = BTRFS_INODE_ITEM_KEY;
773                         } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
774                                 found_type = BTRFS_CSUM_ITEM_KEY;
775                         } else if (found_type) {
776                                 found_type--;
777                         } else {
778                                 break;
779                         }
780                         btrfs_set_key_type(&key, found_type);
781                         btrfs_release_path(root, path);
782                         continue;
783                 }
784                 if (found_key.offset >= inode->i_size)
785                         del_item = 1;
786                 else
787                         del_item = 0;
788                 found_extent = 0;
789
790                 /* FIXME, shrink the extent if the ref count is only 1 */
791                 if (found_type != BTRFS_EXTENT_DATA_KEY)
792                         goto delete;
793
794                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
795                         u64 num_dec;
796                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
797                         if (!del_item) {
798                                 u64 orig_num_bytes =
799                                         btrfs_file_extent_num_bytes(leaf, fi);
800                                 extent_num_bytes = inode->i_size -
801                                         found_key.offset + root->sectorsize - 1;
802                                 btrfs_set_file_extent_num_bytes(leaf, fi,
803                                                          extent_num_bytes);
804                                 num_dec = (orig_num_bytes -
805                                            extent_num_bytes) >> 9;
806                                 if (extent_start != 0) {
807                                         inode->i_blocks -= num_dec;
808                                 }
809                                 btrfs_mark_buffer_dirty(leaf);
810                         } else {
811                                 extent_num_bytes =
812                                         btrfs_file_extent_disk_num_bytes(leaf,
813                                                                          fi);
814                                 /* FIXME blocksize != 4096 */
815                                 num_dec = btrfs_file_extent_num_bytes(leaf,
816                                                                        fi) >> 9;
817                                 if (extent_start != 0) {
818                                         found_extent = 1;
819                                         inode->i_blocks -= num_dec;
820                                 }
821                                 root_gen = btrfs_header_generation(leaf);
822                                 root_owner = btrfs_header_owner(leaf);
823                         }
824                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE &&
825                            !del_item) {
826                         u32 newsize = inode->i_size - found_key.offset;
827                         newsize = btrfs_file_extent_calc_inline_size(newsize);
828                         ret = btrfs_truncate_item(trans, root, path,
829                                                   newsize, 1);
830                         BUG_ON(ret);
831                 }
832 delete:
833                 if (del_item) {
834                         ret = btrfs_del_item(trans, root, path);
835                         if (ret)
836                                 goto error;
837                 } else {
838                         break;
839                 }
840                 btrfs_release_path(root, path);
841                 if (found_extent) {
842                         ret = btrfs_free_extent(trans, root, extent_start,
843                                                 extent_num_bytes,
844                                                 root_owner,
845                                                 root_gen, inode->i_ino,
846                                                 found_key.offset, 0);
847                         BUG_ON(ret);
848                 }
849         }
850         ret = 0;
851 error:
852         btrfs_release_path(root, path);
853         btrfs_free_path(path);
854         inode->i_sb->s_dirt = 1;
855         return ret;
856 }
857
858 static int btrfs_cow_one_page(struct inode *inode, struct page *page,
859                               size_t zero_start)
860 {
861         char *kaddr;
862         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
863         struct btrfs_root *root = BTRFS_I(inode)->root;
864         u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
865         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
866         u64 existing_delalloc;
867         u64 delalloc_start;
868         int ret = 0;
869
870         WARN_ON(!PageLocked(page));
871         set_page_extent_mapped(page);
872
873         lock_extent(em_tree, page_start, page_end, GFP_NOFS);
874         delalloc_start = page_start;
875         existing_delalloc = count_range_bits(&BTRFS_I(inode)->extent_tree,
876                                              &delalloc_start, page_end,
877                                              PAGE_CACHE_SIZE, EXTENT_DELALLOC);
878         set_extent_delalloc(&BTRFS_I(inode)->extent_tree, page_start,
879                             page_end, GFP_NOFS);
880
881         spin_lock(&root->fs_info->delalloc_lock);
882         root->fs_info->delalloc_bytes += PAGE_CACHE_SIZE - existing_delalloc;
883         spin_unlock(&root->fs_info->delalloc_lock);
884
885         if (zero_start != PAGE_CACHE_SIZE) {
886                 kaddr = kmap(page);
887                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
888                 flush_dcache_page(page);
889                 kunmap(page);
890         }
891         set_page_dirty(page);
892         unlock_extent(em_tree, page_start, page_end, GFP_NOFS);
893
894         return ret;
895 }
896
897 /*
898  * taken from block_truncate_page, but does cow as it zeros out
899  * any bytes left in the last page in the file.
900  */
901 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
902 {
903         struct inode *inode = mapping->host;
904         struct btrfs_root *root = BTRFS_I(inode)->root;
905         u32 blocksize = root->sectorsize;
906         pgoff_t index = from >> PAGE_CACHE_SHIFT;
907         unsigned offset = from & (PAGE_CACHE_SIZE-1);
908         struct page *page;
909         int ret = 0;
910         u64 page_start;
911
912         if ((offset & (blocksize - 1)) == 0)
913                 goto out;
914
915         ret = -ENOMEM;
916         page = grab_cache_page(mapping, index);
917         if (!page)
918                 goto out;
919         if (!PageUptodate(page)) {
920                 ret = btrfs_readpage(NULL, page);
921                 lock_page(page);
922                 if (!PageUptodate(page)) {
923                         ret = -EIO;
924                         goto out;
925                 }
926         }
927         page_start = (u64)page->index << PAGE_CACHE_SHIFT;
928
929         ret = btrfs_cow_one_page(inode, page, offset);
930
931         unlock_page(page);
932         page_cache_release(page);
933 out:
934         return ret;
935 }
936
937 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
938 {
939         struct inode *inode = dentry->d_inode;
940         int err;
941
942         err = inode_change_ok(inode, attr);
943         if (err)
944                 return err;
945
946         if (S_ISREG(inode->i_mode) &&
947             attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
948                 struct btrfs_trans_handle *trans;
949                 struct btrfs_root *root = BTRFS_I(inode)->root;
950                 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
951
952                 u64 mask = root->sectorsize - 1;
953                 u64 pos = (inode->i_size + mask) & ~mask;
954                 u64 block_end = attr->ia_size | mask;
955                 u64 hole_size;
956                 u64 alloc_hint = 0;
957
958                 if (attr->ia_size <= pos)
959                         goto out;
960
961                 mutex_lock(&root->fs_info->fs_mutex);
962                 err = btrfs_check_free_space(root, 1, 0);
963                 mutex_unlock(&root->fs_info->fs_mutex);
964                 if (err)
965                         goto fail;
966
967                 btrfs_truncate_page(inode->i_mapping, inode->i_size);
968
969                 lock_extent(em_tree, pos, block_end, GFP_NOFS);
970                 hole_size = (attr->ia_size - pos + mask) & ~mask;
971
972                 mutex_lock(&root->fs_info->fs_mutex);
973                 trans = btrfs_start_transaction(root, 1);
974                 btrfs_set_trans_block_group(trans, inode);
975                 err = btrfs_drop_extents(trans, root, inode,
976                                          pos, pos + hole_size, pos,
977                                          &alloc_hint);
978
979                 if (alloc_hint != EXTENT_MAP_INLINE) {
980                         err = btrfs_insert_file_extent(trans, root,
981                                                        inode->i_ino,
982                                                        pos, 0, 0, hole_size);
983                 }
984                 btrfs_end_transaction(trans, root);
985                 mutex_unlock(&root->fs_info->fs_mutex);
986                 unlock_extent(em_tree, pos, block_end, GFP_NOFS);
987                 if (err)
988                         return err;
989         }
990 out:
991         err = inode_setattr(inode, attr);
992 fail:
993         return err;
994 }
995 void btrfs_delete_inode(struct inode *inode)
996 {
997         struct btrfs_trans_handle *trans;
998         struct btrfs_root *root = BTRFS_I(inode)->root;
999         unsigned long nr;
1000         int ret;
1001
1002         truncate_inode_pages(&inode->i_data, 0);
1003         if (is_bad_inode(inode)) {
1004                 goto no_delete;
1005         }
1006
1007         inode->i_size = 0;
1008         mutex_lock(&root->fs_info->fs_mutex);
1009         trans = btrfs_start_transaction(root, 1);
1010
1011         btrfs_set_trans_block_group(trans, inode);
1012         ret = btrfs_truncate_in_trans(trans, root, inode);
1013         if (ret)
1014                 goto no_delete_lock;
1015         ret = btrfs_delete_xattrs(trans, root, inode);
1016         if (ret)
1017                 goto no_delete_lock;
1018         ret = btrfs_free_inode(trans, root, inode);
1019         if (ret)
1020                 goto no_delete_lock;
1021         nr = trans->blocks_used;
1022
1023         btrfs_end_transaction(trans, root);
1024         mutex_unlock(&root->fs_info->fs_mutex);
1025         btrfs_btree_balance_dirty(root, nr);
1026         btrfs_throttle(root);
1027         return;
1028
1029 no_delete_lock:
1030         nr = trans->blocks_used;
1031         btrfs_end_transaction(trans, root);
1032         mutex_unlock(&root->fs_info->fs_mutex);
1033         btrfs_btree_balance_dirty(root, nr);
1034         btrfs_throttle(root);
1035 no_delete:
1036         clear_inode(inode);
1037 }
1038
1039 /*
1040  * this returns the key found in the dir entry in the location pointer.
1041  * If no dir entries were found, location->objectid is 0.
1042  */
1043 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
1044                                struct btrfs_key *location)
1045 {
1046         const char *name = dentry->d_name.name;
1047         int namelen = dentry->d_name.len;
1048         struct btrfs_dir_item *di;
1049         struct btrfs_path *path;
1050         struct btrfs_root *root = BTRFS_I(dir)->root;
1051         int ret = 0;
1052
1053         if (namelen == 1 && strcmp(name, ".") == 0) {
1054                 location->objectid = dir->i_ino;
1055                 location->type = BTRFS_INODE_ITEM_KEY;
1056                 location->offset = 0;
1057                 return 0;
1058         }
1059         path = btrfs_alloc_path();
1060         BUG_ON(!path);
1061
1062         if (namelen == 2 && strcmp(name, "..") == 0) {
1063                 struct btrfs_key key;
1064                 struct extent_buffer *leaf;
1065                 u32 nritems;
1066                 int slot;
1067
1068                 key.objectid = dir->i_ino;
1069                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1070                 key.offset = 0;
1071                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1072                 BUG_ON(ret == 0);
1073                 ret = 0;
1074
1075                 leaf = path->nodes[0];
1076                 slot = path->slots[0];
1077                 nritems = btrfs_header_nritems(leaf);
1078                 if (slot >= nritems)
1079                         goto out_err;
1080
1081                 btrfs_item_key_to_cpu(leaf, &key, slot);
1082                 if (key.objectid != dir->i_ino ||
1083                     key.type != BTRFS_INODE_REF_KEY) {
1084                         goto out_err;
1085                 }
1086                 location->objectid = key.offset;
1087                 location->type = BTRFS_INODE_ITEM_KEY;
1088                 location->offset = 0;
1089                 goto out;
1090         }
1091
1092         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
1093                                     namelen, 0);
1094         if (IS_ERR(di))
1095                 ret = PTR_ERR(di);
1096         if (!di || IS_ERR(di)) {
1097                 goto out_err;
1098         }
1099         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
1100 out:
1101         btrfs_free_path(path);
1102         return ret;
1103 out_err:
1104         location->objectid = 0;
1105         goto out;
1106 }
1107
1108 /*
1109  * when we hit a tree root in a directory, the btrfs part of the inode
1110  * needs to be changed to reflect the root directory of the tree root.  This
1111  * is kind of like crossing a mount point.
1112  */
1113 static int fixup_tree_root_location(struct btrfs_root *root,
1114                              struct btrfs_key *location,
1115                              struct btrfs_root **sub_root,
1116                              struct dentry *dentry)
1117 {
1118         struct btrfs_path *path;
1119         struct btrfs_root_item *ri;
1120
1121         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
1122                 return 0;
1123         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1124                 return 0;
1125
1126         path = btrfs_alloc_path();
1127         BUG_ON(!path);
1128         mutex_lock(&root->fs_info->fs_mutex);
1129
1130         *sub_root = btrfs_read_fs_root(root->fs_info, location,
1131                                         dentry->d_name.name,
1132                                         dentry->d_name.len);
1133         if (IS_ERR(*sub_root))
1134                 return PTR_ERR(*sub_root);
1135
1136         ri = &(*sub_root)->root_item;
1137         location->objectid = btrfs_root_dirid(ri);
1138         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1139         location->offset = 0;
1140
1141         btrfs_free_path(path);
1142         mutex_unlock(&root->fs_info->fs_mutex);
1143         return 0;
1144 }
1145
1146 static int btrfs_init_locked_inode(struct inode *inode, void *p)
1147 {
1148         struct btrfs_iget_args *args = p;
1149         inode->i_ino = args->ino;
1150         BTRFS_I(inode)->root = args->root;
1151         extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
1152                              inode->i_mapping, GFP_NOFS);
1153         return 0;
1154 }
1155
1156 static int btrfs_find_actor(struct inode *inode, void *opaque)
1157 {
1158         struct btrfs_iget_args *args = opaque;
1159         return (args->ino == inode->i_ino &&
1160                 args->root == BTRFS_I(inode)->root);
1161 }
1162
1163 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
1164                             u64 root_objectid)
1165 {
1166         struct btrfs_iget_args args;
1167         args.ino = objectid;
1168         args.root = btrfs_lookup_fs_root(btrfs_sb(s)->fs_info, root_objectid);
1169
1170         if (!args.root)
1171                 return NULL;
1172
1173         return ilookup5(s, objectid, btrfs_find_actor, (void *)&args);
1174 }
1175
1176 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
1177                                 struct btrfs_root *root)
1178 {
1179         struct inode *inode;
1180         struct btrfs_iget_args args;
1181         args.ino = objectid;
1182         args.root = root;
1183
1184         inode = iget5_locked(s, objectid, btrfs_find_actor,
1185                              btrfs_init_locked_inode,
1186                              (void *)&args);
1187         return inode;
1188 }
1189
1190 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
1191                                    struct nameidata *nd)
1192 {
1193         struct inode * inode;
1194         struct btrfs_inode *bi = BTRFS_I(dir);
1195         struct btrfs_root *root = bi->root;
1196         struct btrfs_root *sub_root = root;
1197         struct btrfs_key location;
1198         int ret;
1199
1200         if (dentry->d_name.len > BTRFS_NAME_LEN)
1201                 return ERR_PTR(-ENAMETOOLONG);
1202
1203         mutex_lock(&root->fs_info->fs_mutex);
1204         ret = btrfs_inode_by_name(dir, dentry, &location);
1205         mutex_unlock(&root->fs_info->fs_mutex);
1206
1207         if (ret < 0)
1208                 return ERR_PTR(ret);
1209
1210         inode = NULL;
1211         if (location.objectid) {
1212                 ret = fixup_tree_root_location(root, &location, &sub_root,
1213                                                 dentry);
1214                 if (ret < 0)
1215                         return ERR_PTR(ret);
1216                 if (ret > 0)
1217                         return ERR_PTR(-ENOENT);
1218                 inode = btrfs_iget_locked(dir->i_sb, location.objectid,
1219                                           sub_root);
1220                 if (!inode)
1221                         return ERR_PTR(-EACCES);
1222                 if (inode->i_state & I_NEW) {
1223                         /* the inode and parent dir are two different roots */
1224                         if (sub_root != root) {
1225                                 igrab(inode);
1226                                 sub_root->inode = inode;
1227                         }
1228                         BTRFS_I(inode)->root = sub_root;
1229                         memcpy(&BTRFS_I(inode)->location, &location,
1230                                sizeof(location));
1231                         btrfs_read_locked_inode(inode);
1232                         unlock_new_inode(inode);
1233                 }
1234         }
1235         return d_splice_alias(inode, dentry);
1236 }
1237
1238 static unsigned char btrfs_filetype_table[] = {
1239         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
1240 };
1241
1242 static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
1243 {
1244         struct inode *inode = filp->f_dentry->d_inode;
1245         struct btrfs_root *root = BTRFS_I(inode)->root;
1246         struct btrfs_item *item;
1247         struct btrfs_dir_item *di;
1248         struct btrfs_key key;
1249         struct btrfs_key found_key;
1250         struct btrfs_path *path;
1251         int ret;
1252         u32 nritems;
1253         struct extent_buffer *leaf;
1254         int slot;
1255         int advance;
1256         unsigned char d_type;
1257         int over = 0;
1258         u32 di_cur;
1259         u32 di_total;
1260         u32 di_len;
1261         int key_type = BTRFS_DIR_INDEX_KEY;
1262         char tmp_name[32];
1263         char *name_ptr;
1264         int name_len;
1265
1266         /* FIXME, use a real flag for deciding about the key type */
1267         if (root->fs_info->tree_root == root)
1268                 key_type = BTRFS_DIR_ITEM_KEY;
1269
1270         /* special case for "." */
1271         if (filp->f_pos == 0) {
1272                 over = filldir(dirent, ".", 1,
1273                                1, inode->i_ino,
1274                                DT_DIR);
1275                 if (over)
1276                         return 0;
1277                 filp->f_pos = 1;
1278         }
1279
1280         mutex_lock(&root->fs_info->fs_mutex);
1281         key.objectid = inode->i_ino;
1282         path = btrfs_alloc_path();
1283         path->reada = 2;
1284
1285         /* special case for .., just use the back ref */
1286         if (filp->f_pos == 1) {
1287                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1288                 key.offset = 0;
1289                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1290                 BUG_ON(ret == 0);
1291                 leaf = path->nodes[0];
1292                 slot = path->slots[0];
1293                 nritems = btrfs_header_nritems(leaf);
1294                 if (slot >= nritems) {
1295                         btrfs_release_path(root, path);
1296                         goto read_dir_items;
1297                 }
1298                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1299                 btrfs_release_path(root, path);
1300                 if (found_key.objectid != key.objectid ||
1301                     found_key.type != BTRFS_INODE_REF_KEY)
1302                         goto read_dir_items;
1303                 over = filldir(dirent, "..", 2,
1304                                2, found_key.offset, DT_DIR);
1305                 if (over)
1306                         goto nopos;
1307                 filp->f_pos = 2;
1308         }
1309
1310 read_dir_items:
1311         btrfs_set_key_type(&key, key_type);
1312         key.offset = filp->f_pos;
1313
1314         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1315         if (ret < 0)
1316                 goto err;
1317         advance = 0;
1318         while(1) {
1319                 leaf = path->nodes[0];
1320                 nritems = btrfs_header_nritems(leaf);
1321                 slot = path->slots[0];
1322                 if (advance || slot >= nritems) {
1323                         if (slot >= nritems -1) {
1324                                 ret = btrfs_next_leaf(root, path);
1325                                 if (ret)
1326                                         break;
1327                                 leaf = path->nodes[0];
1328                                 nritems = btrfs_header_nritems(leaf);
1329                                 slot = path->slots[0];
1330                         } else {
1331                                 slot++;
1332                                 path->slots[0]++;
1333                         }
1334                 }
1335                 advance = 1;
1336                 item = btrfs_item_nr(leaf, slot);
1337                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1338
1339                 if (found_key.objectid != key.objectid)
1340                         break;
1341                 if (btrfs_key_type(&found_key) != key_type)
1342                         break;
1343                 if (found_key.offset < filp->f_pos)
1344                         continue;
1345
1346                 filp->f_pos = found_key.offset;
1347                 advance = 1;
1348                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
1349                 di_cur = 0;
1350                 di_total = btrfs_item_size(leaf, item);
1351                 while(di_cur < di_total) {
1352                         struct btrfs_key location;
1353
1354                         name_len = btrfs_dir_name_len(leaf, di);
1355                         if (name_len < 32) {
1356                                 name_ptr = tmp_name;
1357                         } else {
1358                                 name_ptr = kmalloc(name_len, GFP_NOFS);
1359                                 BUG_ON(!name_ptr);
1360                         }
1361                         read_extent_buffer(leaf, name_ptr,
1362                                            (unsigned long)(di + 1), name_len);
1363
1364                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
1365                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
1366                         over = filldir(dirent, name_ptr, name_len,
1367                                        found_key.offset,
1368                                        location.objectid,
1369                                        d_type);
1370
1371                         if (name_ptr != tmp_name)
1372                                 kfree(name_ptr);
1373
1374                         if (over)
1375                                 goto nopos;
1376                         di_len = btrfs_dir_name_len(leaf, di) +
1377                                 btrfs_dir_data_len(leaf, di) +sizeof(*di);
1378                         di_cur += di_len;
1379                         di = (struct btrfs_dir_item *)((char *)di + di_len);
1380                 }
1381         }
1382         filp->f_pos++;
1383 nopos:
1384         ret = 0;
1385 err:
1386         btrfs_release_path(root, path);
1387         btrfs_free_path(path);
1388         mutex_unlock(&root->fs_info->fs_mutex);
1389         return ret;
1390 }
1391
1392 int btrfs_write_inode(struct inode *inode, int wait)
1393 {
1394         struct btrfs_root *root = BTRFS_I(inode)->root;
1395         struct btrfs_trans_handle *trans;
1396         int ret = 0;
1397
1398         if (wait) {
1399                 mutex_lock(&root->fs_info->fs_mutex);
1400                 trans = btrfs_start_transaction(root, 1);
1401                 btrfs_set_trans_block_group(trans, inode);
1402                 ret = btrfs_commit_transaction(trans, root);
1403                 mutex_unlock(&root->fs_info->fs_mutex);
1404         }
1405         return ret;
1406 }
1407
1408 /*
1409  * This is somewhat expensive, updating the tree every time the
1410  * inode changes.  But, it is most likely to find the inode in cache.
1411  * FIXME, needs more benchmarking...there are no reasons other than performance
1412  * to keep or drop this code.
1413  */
1414 void btrfs_dirty_inode(struct inode *inode)
1415 {
1416         struct btrfs_root *root = BTRFS_I(inode)->root;
1417         struct btrfs_trans_handle *trans;
1418
1419         mutex_lock(&root->fs_info->fs_mutex);
1420         trans = btrfs_start_transaction(root, 1);
1421         btrfs_set_trans_block_group(trans, inode);
1422         btrfs_update_inode(trans, root, inode);
1423         btrfs_end_transaction(trans, root);
1424         mutex_unlock(&root->fs_info->fs_mutex);
1425 }
1426
1427 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
1428                                      struct btrfs_root *root,
1429                                      u64 objectid,
1430                                      struct btrfs_block_group_cache *group,
1431                                      int mode)
1432 {
1433         struct inode *inode;
1434         struct btrfs_inode_item *inode_item;
1435         struct btrfs_key *location;
1436         struct btrfs_path *path;
1437         int ret;
1438         int owner;
1439
1440         path = btrfs_alloc_path();
1441         BUG_ON(!path);
1442
1443         inode = new_inode(root->fs_info->sb);
1444         if (!inode)
1445                 return ERR_PTR(-ENOMEM);
1446
1447         extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
1448                              inode->i_mapping, GFP_NOFS);
1449         BTRFS_I(inode)->root = root;
1450
1451         if (mode & S_IFDIR)
1452                 owner = 0;
1453         else
1454                 owner = 1;
1455         group = btrfs_find_block_group(root, group, 0, 0, owner);
1456         BTRFS_I(inode)->block_group = group;
1457
1458         ret = btrfs_insert_empty_inode(trans, root, path, objectid);
1459         if (ret)
1460                 goto fail;
1461
1462         inode->i_uid = current->fsuid;
1463         inode->i_gid = current->fsgid;
1464         inode->i_mode = mode;
1465         inode->i_ino = objectid;
1466         inode->i_blocks = 0;
1467         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1468         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1469                                   struct btrfs_inode_item);
1470         fill_inode_item(path->nodes[0], inode_item, inode);
1471         btrfs_mark_buffer_dirty(path->nodes[0]);
1472         btrfs_free_path(path);
1473
1474         location = &BTRFS_I(inode)->location;
1475         location->objectid = objectid;
1476         location->offset = 0;
1477         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1478
1479         insert_inode_hash(inode);
1480         return inode;
1481 fail:
1482         btrfs_free_path(path);
1483         return ERR_PTR(ret);
1484 }
1485
1486 static inline u8 btrfs_inode_type(struct inode *inode)
1487 {
1488         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
1489 }
1490
1491 static int btrfs_add_link(struct btrfs_trans_handle *trans,
1492                             struct dentry *dentry, struct inode *inode)
1493 {
1494         int ret;
1495         struct btrfs_key key;
1496         struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
1497         struct inode *parent_inode;
1498
1499         key.objectid = inode->i_ino;
1500         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1501         key.offset = 0;
1502
1503         ret = btrfs_insert_dir_item(trans, root,
1504                                     dentry->d_name.name, dentry->d_name.len,
1505                                     dentry->d_parent->d_inode->i_ino,
1506                                     &key, btrfs_inode_type(inode));
1507         if (ret == 0) {
1508                 ret = btrfs_insert_inode_ref(trans, root,
1509                                      dentry->d_name.name,
1510                                      dentry->d_name.len,
1511                                      inode->i_ino,
1512                                      dentry->d_parent->d_inode->i_ino);
1513                 parent_inode = dentry->d_parent->d_inode;
1514                 parent_inode->i_size += dentry->d_name.len * 2;
1515                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1516                 ret = btrfs_update_inode(trans, root,
1517                                          dentry->d_parent->d_inode);
1518         }
1519         return ret;
1520 }
1521
1522 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
1523                             struct dentry *dentry, struct inode *inode)
1524 {
1525         int err = btrfs_add_link(trans, dentry, inode);
1526         if (!err) {
1527                 d_instantiate(dentry, inode);
1528                 return 0;
1529         }
1530         if (err > 0)
1531                 err = -EEXIST;
1532         return err;
1533 }
1534
1535 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1536                         int mode, dev_t rdev)
1537 {
1538         struct btrfs_trans_handle *trans;
1539         struct btrfs_root *root = BTRFS_I(dir)->root;
1540         struct inode *inode = NULL;
1541         int err;
1542         int drop_inode = 0;
1543         u64 objectid;
1544         unsigned long nr = 0;
1545
1546         if (!new_valid_dev(rdev))
1547                 return -EINVAL;
1548
1549         mutex_lock(&root->fs_info->fs_mutex);
1550         err = btrfs_check_free_space(root, 1, 0);
1551         if (err)
1552                 goto fail;
1553
1554         trans = btrfs_start_transaction(root, 1);
1555         btrfs_set_trans_block_group(trans, dir);
1556
1557         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1558         if (err) {
1559                 err = -ENOSPC;
1560                 goto out_unlock;
1561         }
1562
1563         inode = btrfs_new_inode(trans, root, objectid,
1564                                 BTRFS_I(dir)->block_group, mode);
1565         err = PTR_ERR(inode);
1566         if (IS_ERR(inode))
1567                 goto out_unlock;
1568
1569         btrfs_set_trans_block_group(trans, inode);
1570         err = btrfs_add_nondir(trans, dentry, inode);
1571         if (err)
1572                 drop_inode = 1;
1573         else {
1574                 inode->i_op = &btrfs_special_inode_operations;
1575                 init_special_inode(inode, inode->i_mode, rdev);
1576                 btrfs_update_inode(trans, root, inode);
1577         }
1578         dir->i_sb->s_dirt = 1;
1579         btrfs_update_inode_block_group(trans, inode);
1580         btrfs_update_inode_block_group(trans, dir);
1581 out_unlock:
1582         nr = trans->blocks_used;
1583         btrfs_end_transaction(trans, root);
1584 fail:
1585         mutex_unlock(&root->fs_info->fs_mutex);
1586
1587         if (drop_inode) {
1588                 inode_dec_link_count(inode);
1589                 iput(inode);
1590         }
1591         btrfs_btree_balance_dirty(root, nr);
1592         btrfs_throttle(root);
1593         return err;
1594 }
1595
1596 static int btrfs_create(struct inode *dir, struct dentry *dentry,
1597                         int mode, struct nameidata *nd)
1598 {
1599         struct btrfs_trans_handle *trans;
1600         struct btrfs_root *root = BTRFS_I(dir)->root;
1601         struct inode *inode = NULL;
1602         int err;
1603         int drop_inode = 0;
1604         unsigned long nr = 0;
1605         u64 objectid;
1606
1607         mutex_lock(&root->fs_info->fs_mutex);
1608         err = btrfs_check_free_space(root, 1, 0);
1609         if (err)
1610                 goto fail;
1611         trans = btrfs_start_transaction(root, 1);
1612         btrfs_set_trans_block_group(trans, dir);
1613
1614         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1615         if (err) {
1616                 err = -ENOSPC;
1617                 goto out_unlock;
1618         }
1619
1620         inode = btrfs_new_inode(trans, root, objectid,
1621                                 BTRFS_I(dir)->block_group, mode);
1622         err = PTR_ERR(inode);
1623         if (IS_ERR(inode))
1624                 goto out_unlock;
1625
1626         btrfs_set_trans_block_group(trans, inode);
1627         err = btrfs_add_nondir(trans, dentry, inode);
1628         if (err)
1629                 drop_inode = 1;
1630         else {
1631                 inode->i_mapping->a_ops = &btrfs_aops;
1632                 inode->i_fop = &btrfs_file_operations;
1633                 inode->i_op = &btrfs_file_inode_operations;
1634                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
1635                                      inode->i_mapping, GFP_NOFS);
1636                 BTRFS_I(inode)->extent_tree.ops = &btrfs_extent_map_ops;
1637         }
1638         dir->i_sb->s_dirt = 1;
1639         btrfs_update_inode_block_group(trans, inode);
1640         btrfs_update_inode_block_group(trans, dir);
1641 out_unlock:
1642         nr = trans->blocks_used;
1643         btrfs_end_transaction(trans, root);
1644 fail:
1645         mutex_unlock(&root->fs_info->fs_mutex);
1646
1647         if (drop_inode) {
1648                 inode_dec_link_count(inode);
1649                 iput(inode);
1650         }
1651         btrfs_btree_balance_dirty(root, nr);
1652         btrfs_throttle(root);
1653         return err;
1654 }
1655
1656 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
1657                       struct dentry *dentry)
1658 {
1659         struct btrfs_trans_handle *trans;
1660         struct btrfs_root *root = BTRFS_I(dir)->root;
1661         struct inode *inode = old_dentry->d_inode;
1662         unsigned long nr = 0;
1663         int err;
1664         int drop_inode = 0;
1665
1666         if (inode->i_nlink == 0)
1667                 return -ENOENT;
1668
1669 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1670         inode->i_nlink++;
1671 #else
1672         inc_nlink(inode);
1673 #endif
1674         mutex_lock(&root->fs_info->fs_mutex);
1675         err = btrfs_check_free_space(root, 1, 0);
1676         if (err)
1677                 goto fail;
1678         trans = btrfs_start_transaction(root, 1);
1679
1680         btrfs_set_trans_block_group(trans, dir);
1681         atomic_inc(&inode->i_count);
1682         err = btrfs_add_nondir(trans, dentry, inode);
1683
1684         if (err)
1685                 drop_inode = 1;
1686
1687         dir->i_sb->s_dirt = 1;
1688         btrfs_update_inode_block_group(trans, dir);
1689         err = btrfs_update_inode(trans, root, inode);
1690
1691         if (err)
1692                 drop_inode = 1;
1693
1694         nr = trans->blocks_used;
1695         btrfs_end_transaction(trans, root);
1696 fail:
1697         mutex_unlock(&root->fs_info->fs_mutex);
1698
1699         if (drop_inode) {
1700                 inode_dec_link_count(inode);
1701                 iput(inode);
1702         }
1703         btrfs_btree_balance_dirty(root, nr);
1704         btrfs_throttle(root);
1705         return err;
1706 }
1707
1708 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1709 {
1710         struct inode *inode;
1711         struct btrfs_trans_handle *trans;
1712         struct btrfs_root *root = BTRFS_I(dir)->root;
1713         int err = 0;
1714         int drop_on_err = 0;
1715         u64 objectid;
1716         unsigned long nr = 1;
1717
1718         mutex_lock(&root->fs_info->fs_mutex);
1719         err = btrfs_check_free_space(root, 1, 0);
1720         if (err)
1721                 goto out_unlock;
1722
1723         trans = btrfs_start_transaction(root, 1);
1724         btrfs_set_trans_block_group(trans, dir);
1725
1726         if (IS_ERR(trans)) {
1727                 err = PTR_ERR(trans);
1728                 goto out_unlock;
1729         }
1730
1731         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1732         if (err) {
1733                 err = -ENOSPC;
1734                 goto out_unlock;
1735         }
1736
1737         inode = btrfs_new_inode(trans, root, objectid,
1738                                 BTRFS_I(dir)->block_group, S_IFDIR | mode);
1739         if (IS_ERR(inode)) {
1740                 err = PTR_ERR(inode);
1741                 goto out_fail;
1742         }
1743
1744         drop_on_err = 1;
1745         inode->i_op = &btrfs_dir_inode_operations;
1746         inode->i_fop = &btrfs_dir_file_operations;
1747         btrfs_set_trans_block_group(trans, inode);
1748
1749         inode->i_size = 0;
1750         err = btrfs_update_inode(trans, root, inode);
1751         if (err)
1752                 goto out_fail;
1753
1754         err = btrfs_add_link(trans, dentry, inode);
1755         if (err)
1756                 goto out_fail;
1757
1758         d_instantiate(dentry, inode);
1759         drop_on_err = 0;
1760         dir->i_sb->s_dirt = 1;
1761         btrfs_update_inode_block_group(trans, inode);
1762         btrfs_update_inode_block_group(trans, dir);
1763
1764 out_fail:
1765         nr = trans->blocks_used;
1766         btrfs_end_transaction(trans, root);
1767
1768 out_unlock:
1769         mutex_unlock(&root->fs_info->fs_mutex);
1770         if (drop_on_err)
1771                 iput(inode);
1772         btrfs_btree_balance_dirty(root, nr);
1773         btrfs_throttle(root);
1774         return err;
1775 }
1776
1777 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
1778                                     size_t page_offset, u64 start, u64 end,
1779                                     int create)
1780 {
1781         int ret;
1782         int err = 0;
1783         u64 bytenr;
1784         u64 extent_start = 0;
1785         u64 extent_end = 0;
1786         u64 objectid = inode->i_ino;
1787         u32 found_type;
1788         int failed_insert = 0;
1789         struct btrfs_path *path;
1790         struct btrfs_root *root = BTRFS_I(inode)->root;
1791         struct btrfs_file_extent_item *item;
1792         struct extent_buffer *leaf;
1793         struct btrfs_key found_key;
1794         struct extent_map *em = NULL;
1795         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1796         struct btrfs_trans_handle *trans = NULL;
1797
1798         path = btrfs_alloc_path();
1799         BUG_ON(!path);
1800         mutex_lock(&root->fs_info->fs_mutex);
1801
1802 again:
1803         em = lookup_extent_mapping(em_tree, start, end);
1804         if (em) {
1805                 if (em->start > start) {
1806                         printk("get_extent start %Lu em start %Lu\n",
1807                                start, em->start);
1808                         WARN_ON(1);
1809                 }
1810                 goto out;
1811         }
1812         if (!em) {
1813                 em = alloc_extent_map(GFP_NOFS);
1814                 if (!em) {
1815                         err = -ENOMEM;
1816                         goto out;
1817                 }
1818                 em->start = EXTENT_MAP_HOLE;
1819                 em->end = EXTENT_MAP_HOLE;
1820         }
1821         em->bdev = inode->i_sb->s_bdev;
1822         ret = btrfs_lookup_file_extent(trans, root, path,
1823                                        objectid, start, trans != NULL);
1824         if (ret < 0) {
1825                 err = ret;
1826                 goto out;
1827         }
1828
1829         if (ret != 0) {
1830                 if (path->slots[0] == 0)
1831                         goto not_found;
1832                 path->slots[0]--;
1833         }
1834
1835         leaf = path->nodes[0];
1836         item = btrfs_item_ptr(leaf, path->slots[0],
1837                               struct btrfs_file_extent_item);
1838         /* are we inside the extent that was found? */
1839         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1840         found_type = btrfs_key_type(&found_key);
1841         if (found_key.objectid != objectid ||
1842             found_type != BTRFS_EXTENT_DATA_KEY) {
1843                 goto not_found;
1844         }
1845
1846         found_type = btrfs_file_extent_type(leaf, item);
1847         extent_start = found_key.offset;
1848         if (found_type == BTRFS_FILE_EXTENT_REG) {
1849                 extent_end = extent_start +
1850                        btrfs_file_extent_num_bytes(leaf, item);
1851                 err = 0;
1852                 if (start < extent_start || start >= extent_end) {
1853                         em->start = start;
1854                         if (start < extent_start) {
1855                                 if (end < extent_start)
1856                                         goto not_found;
1857                                 em->end = extent_end - 1;
1858                         } else {
1859                                 em->end = end;
1860                         }
1861                         goto not_found_em;
1862                 }
1863                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
1864                 if (bytenr == 0) {
1865                         em->start = extent_start;
1866                         em->end = extent_end - 1;
1867                         em->block_start = EXTENT_MAP_HOLE;
1868                         em->block_end = EXTENT_MAP_HOLE;
1869                         goto insert;
1870                 }
1871                 bytenr += btrfs_file_extent_offset(leaf, item);
1872                 em->block_start = bytenr;
1873                 em->block_end = em->block_start +
1874                         btrfs_file_extent_num_bytes(leaf, item) - 1;
1875                 em->start = extent_start;
1876                 em->end = extent_end - 1;
1877                 goto insert;
1878         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
1879                 unsigned long ptr;
1880                 char *map;
1881                 size_t size;
1882                 size_t extent_offset;
1883                 size_t copy_size;
1884
1885                 size = btrfs_file_extent_inline_len(leaf, btrfs_item_nr(leaf,
1886                                                     path->slots[0]));
1887                 extent_end = (extent_start + size - 1) |
1888                         ((u64)root->sectorsize - 1);
1889                 if (start < extent_start || start >= extent_end) {
1890                         em->start = start;
1891                         if (start < extent_start) {
1892                                 if (end < extent_start)
1893                                         goto not_found;
1894                                 em->end = extent_end;
1895                         } else {
1896                                 em->end = end;
1897                         }
1898                         goto not_found_em;
1899                 }
1900                 em->block_start = EXTENT_MAP_INLINE;
1901                 em->block_end = EXTENT_MAP_INLINE;
1902
1903                 if (!page) {
1904                         em->start = extent_start;
1905                         em->end = extent_start + size - 1;
1906                         goto out;
1907                 }
1908
1909                 extent_offset = ((u64)page->index << PAGE_CACHE_SHIFT) -
1910                         extent_start + page_offset;
1911                 copy_size = min_t(u64, PAGE_CACHE_SIZE - page_offset,
1912                                 size - extent_offset);
1913                 em->start = extent_start + extent_offset;
1914                 em->end = (em->start + copy_size -1) |
1915                         ((u64)root->sectorsize -1);
1916                 map = kmap(page);
1917                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
1918                 if (create == 0 && !PageUptodate(page)) {
1919                         read_extent_buffer(leaf, map + page_offset, ptr,
1920                                            copy_size);
1921                         flush_dcache_page(page);
1922                 } else if (create && PageUptodate(page)) {
1923                         if (!trans) {
1924                                 kunmap(page);
1925                                 free_extent_map(em);
1926                                 em = NULL;
1927                                 btrfs_release_path(root, path);
1928                                 trans = btrfs_start_transaction(root, 1);
1929                                 goto again;
1930                         }
1931                         write_extent_buffer(leaf, map + page_offset, ptr,
1932                                             copy_size);
1933                         btrfs_mark_buffer_dirty(leaf);
1934                 }
1935                 kunmap(page);
1936                 set_extent_uptodate(em_tree, em->start, em->end, GFP_NOFS);
1937                 goto insert;
1938         } else {
1939                 printk("unkknown found_type %d\n", found_type);
1940                 WARN_ON(1);
1941         }
1942 not_found:
1943         em->start = start;
1944         em->end = end;
1945 not_found_em:
1946         em->block_start = EXTENT_MAP_HOLE;
1947         em->block_end = EXTENT_MAP_HOLE;
1948 insert:
1949         btrfs_release_path(root, path);
1950         if (em->start > start || em->end < start) {
1951                 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->end, start, end);
1952                 err = -EIO;
1953                 goto out;
1954         }
1955         ret = add_extent_mapping(em_tree, em);
1956         if (ret == -EEXIST) {
1957                 free_extent_map(em);
1958                 em = NULL;
1959                 if (0 && failed_insert == 1) {
1960                         btrfs_drop_extent_cache(inode, start, end);
1961                 }
1962                 failed_insert++;
1963                 if (failed_insert > 5) {
1964                         printk("failing to insert %Lu %Lu\n", start, end);
1965                         err = -EIO;
1966                         goto out;
1967                 }
1968                 goto again;
1969         }
1970         err = 0;
1971 out:
1972         btrfs_free_path(path);
1973         if (trans) {
1974                 ret = btrfs_end_transaction(trans, root);
1975                 if (!err)
1976                         err = ret;
1977         }
1978         mutex_unlock(&root->fs_info->fs_mutex);
1979         if (err) {
1980                 free_extent_map(em);
1981                 WARN_ON(1);
1982                 return ERR_PTR(err);
1983         }
1984         return em;
1985 }
1986
1987 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
1988 {
1989         return extent_bmap(mapping, iblock, btrfs_get_extent);
1990 }
1991
1992 int btrfs_readpage(struct file *file, struct page *page)
1993 {
1994         struct extent_map_tree *tree;
1995         tree = &BTRFS_I(page->mapping->host)->extent_tree;
1996         return extent_read_full_page(tree, page, btrfs_get_extent);
1997 }
1998
1999 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
2000 {
2001         struct extent_map_tree *tree;
2002
2003
2004         if (current->flags & PF_MEMALLOC) {
2005                 redirty_page_for_writepage(wbc, page);
2006                 unlock_page(page);
2007                 return 0;
2008         }
2009         tree = &BTRFS_I(page->mapping->host)->extent_tree;
2010         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
2011 }
2012
2013 static int btrfs_writepages(struct address_space *mapping,
2014                             struct writeback_control *wbc)
2015 {
2016         struct extent_map_tree *tree;
2017         tree = &BTRFS_I(mapping->host)->extent_tree;
2018         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
2019 }
2020
2021 static int
2022 btrfs_readpages(struct file *file, struct address_space *mapping,
2023                 struct list_head *pages, unsigned nr_pages)
2024 {
2025         struct extent_map_tree *tree;
2026         tree = &BTRFS_I(mapping->host)->extent_tree;
2027         return extent_readpages(tree, mapping, pages, nr_pages,
2028                                 btrfs_get_extent);
2029 }
2030
2031 static int btrfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
2032 {
2033         struct extent_map_tree *tree;
2034         int ret;
2035
2036         tree = &BTRFS_I(page->mapping->host)->extent_tree;
2037         ret = try_release_extent_mapping(tree, page);
2038         if (ret == 1) {
2039                 ClearPagePrivate(page);
2040                 set_page_private(page, 0);
2041                 page_cache_release(page);
2042         }
2043         return ret;
2044 }
2045
2046 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
2047 {
2048         struct extent_map_tree *tree;
2049
2050         tree = &BTRFS_I(page->mapping->host)->extent_tree;
2051         extent_invalidatepage(tree, page, offset);
2052         btrfs_releasepage(page, GFP_NOFS);
2053 }
2054
2055 /*
2056  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
2057  * called from a page fault handler when a page is first dirtied. Hence we must
2058  * be careful to check for EOF conditions here. We set the page up correctly
2059  * for a written page which means we get ENOSPC checking when writing into
2060  * holes and correct delalloc and unwritten extent mapping on filesystems that
2061  * support these features.
2062  *
2063  * We are not allowed to take the i_mutex here so we have to play games to
2064  * protect against truncate races as the page could now be beyond EOF.  Because
2065  * vmtruncate() writes the inode size before removing pages, once we have the
2066  * page lock we can determine safely if the page is beyond EOF. If it is not
2067  * beyond EOF, then the page is guaranteed safe against truncation until we
2068  * unlock the page.
2069  */
2070 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
2071 {
2072         struct inode *inode = fdentry(vma->vm_file)->d_inode;
2073         struct btrfs_root *root = BTRFS_I(inode)->root;
2074         unsigned long end;
2075         loff_t size;
2076         int ret;
2077         u64 page_start;
2078
2079         mutex_lock(&root->fs_info->fs_mutex);
2080         ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
2081         mutex_unlock(&root->fs_info->fs_mutex);
2082         if (ret)
2083                 goto out;
2084
2085         ret = -EINVAL;
2086
2087         lock_page(page);
2088         wait_on_page_writeback(page);
2089         size = i_size_read(inode);
2090         page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2091
2092         if ((page->mapping != inode->i_mapping) ||
2093             (page_start > size)) {
2094                 /* page got truncated out from underneath us */
2095                 goto out_unlock;
2096         }
2097
2098         /* page is wholly or partially inside EOF */
2099         if (page_start + PAGE_CACHE_SIZE > size)
2100                 end = size & ~PAGE_CACHE_MASK;
2101         else
2102                 end = PAGE_CACHE_SIZE;
2103
2104         ret = btrfs_cow_one_page(inode, page, end);
2105
2106 out_unlock:
2107         unlock_page(page);
2108 out:
2109         return ret;
2110 }
2111
2112 static void btrfs_truncate(struct inode *inode)
2113 {
2114         struct btrfs_root *root = BTRFS_I(inode)->root;
2115         int ret;
2116         struct btrfs_trans_handle *trans;
2117         unsigned long nr;
2118
2119         if (!S_ISREG(inode->i_mode))
2120                 return;
2121         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2122                 return;
2123
2124         btrfs_truncate_page(inode->i_mapping, inode->i_size);
2125
2126         mutex_lock(&root->fs_info->fs_mutex);
2127         trans = btrfs_start_transaction(root, 1);
2128         btrfs_set_trans_block_group(trans, inode);
2129
2130         /* FIXME, add redo link to tree so we don't leak on crash */
2131         ret = btrfs_truncate_in_trans(trans, root, inode);
2132         btrfs_update_inode(trans, root, inode);
2133         nr = trans->blocks_used;
2134
2135         ret = btrfs_end_transaction(trans, root);
2136         BUG_ON(ret);
2137         mutex_unlock(&root->fs_info->fs_mutex);
2138         btrfs_btree_balance_dirty(root, nr);
2139         btrfs_throttle(root);
2140 }
2141
2142 static int noinline create_subvol(struct btrfs_root *root, char *name,
2143                                   int namelen)
2144 {
2145         struct btrfs_trans_handle *trans;
2146         struct btrfs_key key;
2147         struct btrfs_root_item root_item;
2148         struct btrfs_inode_item *inode_item;
2149         struct extent_buffer *leaf;
2150         struct btrfs_root *new_root = root;
2151         struct inode *inode;
2152         struct inode *dir;
2153         int ret;
2154         int err;
2155         u64 objectid;
2156         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
2157         unsigned long nr = 1;
2158
2159         mutex_lock(&root->fs_info->fs_mutex);
2160         ret = btrfs_check_free_space(root, 1, 0);
2161         if (ret)
2162                 goto fail_commit;
2163
2164         trans = btrfs_start_transaction(root, 1);
2165         BUG_ON(!trans);
2166
2167         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
2168                                        0, &objectid);
2169         if (ret)
2170                 goto fail;
2171
2172         leaf = __btrfs_alloc_free_block(trans, root, root->leafsize,
2173                                         objectid, trans->transid, 0, 0,
2174                                         0, 0);
2175         if (IS_ERR(leaf))
2176                 return PTR_ERR(leaf);
2177
2178         btrfs_set_header_nritems(leaf, 0);
2179         btrfs_set_header_level(leaf, 0);
2180         btrfs_set_header_bytenr(leaf, leaf->start);
2181         btrfs_set_header_generation(leaf, trans->transid);
2182         btrfs_set_header_owner(leaf, objectid);
2183
2184         write_extent_buffer(leaf, root->fs_info->fsid,
2185                             (unsigned long)btrfs_header_fsid(leaf),
2186                             BTRFS_FSID_SIZE);
2187         btrfs_mark_buffer_dirty(leaf);
2188
2189         inode_item = &root_item.inode;
2190         memset(inode_item, 0, sizeof(*inode_item));
2191         inode_item->generation = cpu_to_le64(1);
2192         inode_item->size = cpu_to_le64(3);
2193         inode_item->nlink = cpu_to_le32(1);
2194         inode_item->nblocks = cpu_to_le64(1);
2195         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
2196
2197         btrfs_set_root_bytenr(&root_item, leaf->start);
2198         btrfs_set_root_level(&root_item, 0);
2199         btrfs_set_root_refs(&root_item, 1);
2200         btrfs_set_root_used(&root_item, 0);
2201
2202         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
2203         root_item.drop_level = 0;
2204
2205         free_extent_buffer(leaf);
2206         leaf = NULL;
2207
2208         btrfs_set_root_dirid(&root_item, new_dirid);
2209
2210         key.objectid = objectid;
2211         key.offset = 1;
2212         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2213         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
2214                                 &root_item);
2215         if (ret)
2216                 goto fail;
2217
2218         /*
2219          * insert the directory item
2220          */
2221         key.offset = (u64)-1;
2222         dir = root->fs_info->sb->s_root->d_inode;
2223         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
2224                                     name, namelen, dir->i_ino, &key,
2225                                     BTRFS_FT_DIR);
2226         if (ret)
2227                 goto fail;
2228
2229         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
2230                              name, namelen, objectid,
2231                              root->fs_info->sb->s_root->d_inode->i_ino);
2232         if (ret)
2233                 goto fail;
2234
2235         ret = btrfs_commit_transaction(trans, root);
2236         if (ret)
2237                 goto fail_commit;
2238
2239         new_root = btrfs_read_fs_root(root->fs_info, &key, name, namelen);
2240         BUG_ON(!new_root);
2241
2242         trans = btrfs_start_transaction(new_root, 1);
2243         BUG_ON(!trans);
2244
2245         inode = btrfs_new_inode(trans, new_root, new_dirid,
2246                                 BTRFS_I(dir)->block_group, S_IFDIR | 0700);
2247         if (IS_ERR(inode))
2248                 goto fail;
2249         inode->i_op = &btrfs_dir_inode_operations;
2250         inode->i_fop = &btrfs_dir_file_operations;
2251         new_root->inode = inode;
2252
2253         ret = btrfs_insert_inode_ref(trans, new_root, "..", 2, new_dirid,
2254                                      new_dirid);
2255         inode->i_nlink = 1;
2256         inode->i_size = 0;
2257         ret = btrfs_update_inode(trans, new_root, inode);
2258         if (ret)
2259                 goto fail;
2260 fail:
2261         nr = trans->blocks_used;
2262         err = btrfs_commit_transaction(trans, new_root);
2263         if (err && !ret)
2264                 ret = err;
2265 fail_commit:
2266         mutex_unlock(&root->fs_info->fs_mutex);
2267         btrfs_btree_balance_dirty(root, nr);
2268         btrfs_throttle(root);
2269         return ret;
2270 }
2271
2272 static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
2273 {
2274         struct btrfs_pending_snapshot *pending_snapshot;
2275         struct btrfs_trans_handle *trans;
2276         int ret;
2277         int err;
2278         unsigned long nr = 0;
2279
2280         if (!root->ref_cows)
2281                 return -EINVAL;
2282
2283         mutex_lock(&root->fs_info->fs_mutex);
2284         ret = btrfs_check_free_space(root, 1, 0);
2285         if (ret)
2286                 goto fail_unlock;
2287
2288         pending_snapshot = kmalloc(sizeof(*pending_snapshot), GFP_NOFS);
2289         if (!pending_snapshot) {
2290                 ret = -ENOMEM;
2291                 goto fail_unlock;
2292         }
2293         pending_snapshot->name = kstrndup(name, namelen, GFP_NOFS);
2294         if (!pending_snapshot->name) {
2295                 ret = -ENOMEM;
2296                 kfree(pending_snapshot);
2297                 goto fail_unlock;
2298         }
2299         trans = btrfs_start_transaction(root, 1);
2300         BUG_ON(!trans);
2301
2302         pending_snapshot->root = root;
2303         list_add(&pending_snapshot->list,
2304                  &trans->transaction->pending_snapshots);
2305         ret = btrfs_update_inode(trans, root, root->inode);
2306         err = btrfs_commit_transaction(trans, root);
2307
2308 fail_unlock:
2309         mutex_unlock(&root->fs_info->fs_mutex);
2310         btrfs_btree_balance_dirty(root, nr);
2311         btrfs_throttle(root);
2312         return ret;
2313 }
2314
2315 unsigned long btrfs_force_ra(struct address_space *mapping,
2316                               struct file_ra_state *ra, struct file *file,
2317                               pgoff_t offset, pgoff_t last_index)
2318 {
2319         pgoff_t req_size;
2320
2321 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2322         req_size = last_index - offset + 1;
2323         offset = page_cache_readahead(mapping, ra, file, offset, req_size);
2324         return offset;
2325 #else
2326         req_size = min(last_index - offset + 1, (pgoff_t)128);
2327         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
2328         return offset + req_size;
2329 #endif
2330 }
2331
2332 int btrfs_defrag_file(struct file *file) {
2333         struct inode *inode = fdentry(file)->d_inode;
2334         struct btrfs_root *root = BTRFS_I(inode)->root;
2335         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2336         struct page *page;
2337         unsigned long last_index;
2338         unsigned long ra_index = 0;
2339         u64 page_start;
2340         u64 page_end;
2341         u64 delalloc_start;
2342         u64 existing_delalloc;
2343         unsigned long i;
2344         int ret;
2345
2346         mutex_lock(&root->fs_info->fs_mutex);
2347         ret = btrfs_check_free_space(root, inode->i_size, 0);
2348         mutex_unlock(&root->fs_info->fs_mutex);
2349         if (ret)
2350                 return -ENOSPC;
2351
2352         mutex_lock(&inode->i_mutex);
2353         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
2354         for (i = 0; i <= last_index; i++) {
2355                 if (i == ra_index) {
2356                         ra_index = btrfs_force_ra(inode->i_mapping,
2357                                                   &file->f_ra,
2358                                                   file, ra_index, last_index);
2359                 }
2360                 page = grab_cache_page(inode->i_mapping, i);
2361                 if (!page)
2362                         goto out_unlock;
2363                 if (!PageUptodate(page)) {
2364                         btrfs_readpage(NULL, page);
2365                         lock_page(page);
2366                         if (!PageUptodate(page)) {
2367                                 unlock_page(page);
2368                                 page_cache_release(page);
2369                                 goto out_unlock;
2370                         }
2371                 }
2372                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2373                 page_end = page_start + PAGE_CACHE_SIZE - 1;
2374
2375                 lock_extent(em_tree, page_start, page_end, GFP_NOFS);
2376                 delalloc_start = page_start;
2377                 existing_delalloc =
2378                         count_range_bits(&BTRFS_I(inode)->extent_tree,
2379                                          &delalloc_start, page_end,
2380                                          PAGE_CACHE_SIZE, EXTENT_DELALLOC);
2381                 set_extent_delalloc(em_tree, page_start,
2382                                     page_end, GFP_NOFS);
2383
2384                 spin_lock(&root->fs_info->delalloc_lock);
2385                 root->fs_info->delalloc_bytes += PAGE_CACHE_SIZE -
2386                                                  existing_delalloc;
2387                 spin_unlock(&root->fs_info->delalloc_lock);
2388
2389                 unlock_extent(em_tree, page_start, page_end, GFP_NOFS);
2390                 set_page_dirty(page);
2391                 unlock_page(page);
2392                 page_cache_release(page);
2393                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
2394         }
2395
2396 out_unlock:
2397         mutex_unlock(&inode->i_mutex);
2398         return 0;
2399 }
2400
2401 static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg)
2402 {
2403         u64 new_size;
2404         u64 old_size;
2405         struct btrfs_ioctl_vol_args *vol_args;
2406         struct btrfs_trans_handle *trans;
2407         char *sizestr;
2408         int ret = 0;
2409         int namelen;
2410         int mod = 0;
2411
2412         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
2413
2414         if (!vol_args)
2415                 return -ENOMEM;
2416
2417         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
2418                 ret = -EFAULT;
2419                 goto out;
2420         }
2421         namelen = strlen(vol_args->name);
2422         if (namelen > BTRFS_VOL_NAME_MAX) {
2423                 ret = -EINVAL;
2424                 goto out;
2425         }
2426
2427         sizestr = vol_args->name;
2428         if (!strcmp(sizestr, "max"))
2429                 new_size = root->fs_info->sb->s_bdev->bd_inode->i_size;
2430         else {
2431                 if (sizestr[0] == '-') {
2432                         mod = -1;
2433                         sizestr++;
2434                 } else if (sizestr[0] == '+') {
2435                         mod = 1;
2436                         sizestr++;
2437                 }
2438                 new_size = btrfs_parse_size(sizestr);
2439                 if (new_size == 0) {
2440                         ret = -EINVAL;
2441                         goto out;
2442                 }
2443         }
2444
2445         mutex_lock(&root->fs_info->fs_mutex);
2446         old_size = btrfs_super_total_bytes(&root->fs_info->super_copy);
2447
2448         if (mod < 0) {
2449                 if (new_size > old_size) {
2450                         ret = -EINVAL;
2451                         goto out_unlock;
2452                 }
2453                 new_size = old_size - new_size;
2454         } else if (mod > 0) {
2455                 new_size = old_size + new_size;
2456         }
2457
2458         if (new_size < 256 * 1024 * 1024) {
2459                 ret = -EINVAL;
2460                 goto out_unlock;
2461         }
2462         if (new_size > root->fs_info->sb->s_bdev->bd_inode->i_size) {
2463                 ret = -EFBIG;
2464                 goto out_unlock;
2465         }
2466
2467         do_div(new_size, root->sectorsize);
2468         new_size *= root->sectorsize;
2469
2470 printk("new size is %Lu\n", new_size);
2471         if (new_size > old_size) {
2472                 trans = btrfs_start_transaction(root, 1);
2473                 ret = btrfs_grow_extent_tree(trans, root, new_size);
2474                 btrfs_commit_transaction(trans, root);
2475         } else {
2476                 ret = btrfs_shrink_extent_tree(root, new_size);
2477         }
2478
2479 out_unlock:
2480         mutex_unlock(&root->fs_info->fs_mutex);
2481 out:
2482         kfree(vol_args);
2483         return ret;
2484 }
2485
2486 static int noinline btrfs_ioctl_snap_create(struct btrfs_root *root,
2487                                             void __user *arg)
2488 {
2489         struct btrfs_ioctl_vol_args *vol_args;
2490         struct btrfs_dir_item *di;
2491         struct btrfs_path *path;
2492         u64 root_dirid;
2493         int namelen;
2494         int ret;
2495
2496         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
2497
2498         if (!vol_args)
2499                 return -ENOMEM;
2500
2501         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
2502                 ret = -EFAULT;
2503                 goto out;
2504         }
2505
2506         namelen = strlen(vol_args->name);
2507         if (namelen > BTRFS_VOL_NAME_MAX) {
2508                 ret = -EINVAL;
2509                 goto out;
2510         }
2511         if (strchr(vol_args->name, '/')) {
2512                 ret = -EINVAL;
2513                 goto out;
2514         }
2515
2516         path = btrfs_alloc_path();
2517         if (!path) {
2518                 ret = -ENOMEM;
2519                 goto out;
2520         }
2521
2522         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
2523         mutex_lock(&root->fs_info->fs_mutex);
2524         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
2525                             path, root_dirid,
2526                             vol_args->name, namelen, 0);
2527         mutex_unlock(&root->fs_info->fs_mutex);
2528         btrfs_free_path(path);
2529
2530         if (di && !IS_ERR(di)) {
2531                 ret = -EEXIST;
2532                 goto out;
2533         }
2534
2535         if (IS_ERR(di)) {
2536                 ret = PTR_ERR(di);
2537                 goto out;
2538         }
2539
2540         if (root == root->fs_info->tree_root)
2541                 ret = create_subvol(root, vol_args->name, namelen);
2542         else
2543                 ret = create_snapshot(root, vol_args->name, namelen);
2544 out:
2545         kfree(vol_args);
2546         return ret;
2547 }
2548
2549 static int btrfs_ioctl_defrag(struct file *file)
2550 {
2551         struct inode *inode = fdentry(file)->d_inode;
2552         struct btrfs_root *root = BTRFS_I(inode)->root;
2553
2554         switch (inode->i_mode & S_IFMT) {
2555         case S_IFDIR:
2556                 mutex_lock(&root->fs_info->fs_mutex);
2557                 btrfs_defrag_root(root, 0);
2558                 btrfs_defrag_root(root->fs_info->extent_root, 0);
2559                 mutex_unlock(&root->fs_info->fs_mutex);
2560                 break;
2561         case S_IFREG:
2562                 btrfs_defrag_file(file);
2563                 break;
2564         }
2565
2566         return 0;
2567 }
2568
2569 long btrfs_ioctl(struct file *file, unsigned int
2570                 cmd, unsigned long arg)
2571 {
2572         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
2573
2574         switch (cmd) {
2575         case BTRFS_IOC_SNAP_CREATE:
2576                 return btrfs_ioctl_snap_create(root, (void __user *)arg);
2577         case BTRFS_IOC_DEFRAG:
2578                 return btrfs_ioctl_defrag(file);
2579         case BTRFS_IOC_RESIZE:
2580                 return btrfs_ioctl_resize(root, (void __user *)arg);
2581         }
2582
2583         return -ENOTTY;
2584 }
2585
2586 /*
2587  * Called inside transaction, so use GFP_NOFS
2588  */
2589 struct inode *btrfs_alloc_inode(struct super_block *sb)
2590 {
2591         struct btrfs_inode *ei;
2592
2593         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
2594         if (!ei)
2595                 return NULL;
2596         ei->last_trans = 0;
2597         ei->ordered_trans = 0;
2598         return &ei->vfs_inode;
2599 }
2600
2601 void btrfs_destroy_inode(struct inode *inode)
2602 {
2603         WARN_ON(!list_empty(&inode->i_dentry));
2604         WARN_ON(inode->i_data.nrpages);
2605
2606         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
2607 }
2608
2609 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2610 static void init_once(struct kmem_cache * cachep, void *foo)
2611 #else
2612 static void init_once(void * foo, struct kmem_cache * cachep,
2613                       unsigned long flags)
2614 #endif
2615 {
2616         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
2617
2618         inode_init_once(&ei->vfs_inode);
2619 }
2620
2621 void btrfs_destroy_cachep(void)
2622 {
2623         if (btrfs_inode_cachep)
2624                 kmem_cache_destroy(btrfs_inode_cachep);
2625         if (btrfs_trans_handle_cachep)
2626                 kmem_cache_destroy(btrfs_trans_handle_cachep);
2627         if (btrfs_transaction_cachep)
2628                 kmem_cache_destroy(btrfs_transaction_cachep);
2629         if (btrfs_bit_radix_cachep)
2630                 kmem_cache_destroy(btrfs_bit_radix_cachep);
2631         if (btrfs_path_cachep)
2632                 kmem_cache_destroy(btrfs_path_cachep);
2633 }
2634
2635 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
2636                                        unsigned long extra_flags,
2637 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2638                                        void (*ctor)(struct kmem_cache *, void *)
2639 #else
2640                                        void (*ctor)(void *, struct kmem_cache *,
2641                                                     unsigned long)
2642 #endif
2643                                      )
2644 {
2645         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
2646                                  SLAB_MEM_SPREAD | extra_flags), ctor
2647 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2648                                  ,NULL
2649 #endif
2650                                 );
2651 }
2652
2653 int btrfs_init_cachep(void)
2654 {
2655         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
2656                                           sizeof(struct btrfs_inode),
2657                                           0, init_once);
2658         if (!btrfs_inode_cachep)
2659                 goto fail;
2660         btrfs_trans_handle_cachep =
2661                         btrfs_cache_create("btrfs_trans_handle_cache",
2662                                            sizeof(struct btrfs_trans_handle),
2663                                            0, NULL);
2664         if (!btrfs_trans_handle_cachep)
2665                 goto fail;
2666         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
2667                                              sizeof(struct btrfs_transaction),
2668                                              0, NULL);
2669         if (!btrfs_transaction_cachep)
2670                 goto fail;
2671         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
2672                                          sizeof(struct btrfs_path),
2673                                          0, NULL);
2674         if (!btrfs_path_cachep)
2675                 goto fail;
2676         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
2677                                               SLAB_DESTROY_BY_RCU, NULL);
2678         if (!btrfs_bit_radix_cachep)
2679                 goto fail;
2680         return 0;
2681 fail:
2682         btrfs_destroy_cachep();
2683         return -ENOMEM;
2684 }
2685
2686 static int btrfs_getattr(struct vfsmount *mnt,
2687                          struct dentry *dentry, struct kstat *stat)
2688 {
2689         struct inode *inode = dentry->d_inode;
2690         generic_fillattr(inode, stat);
2691         stat->blksize = PAGE_CACHE_SIZE;
2692         return 0;
2693 }
2694
2695 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
2696                            struct inode * new_dir,struct dentry *new_dentry)
2697 {
2698         struct btrfs_trans_handle *trans;
2699         struct btrfs_root *root = BTRFS_I(old_dir)->root;
2700         struct inode *new_inode = new_dentry->d_inode;
2701         struct inode *old_inode = old_dentry->d_inode;
2702         struct timespec ctime = CURRENT_TIME;
2703         struct btrfs_path *path;
2704         int ret;
2705
2706         if (S_ISDIR(old_inode->i_mode) && new_inode &&
2707             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
2708                 return -ENOTEMPTY;
2709         }
2710
2711         mutex_lock(&root->fs_info->fs_mutex);
2712         ret = btrfs_check_free_space(root, 1, 0);
2713         if (ret)
2714                 goto out_unlock;
2715
2716         trans = btrfs_start_transaction(root, 1);
2717
2718         btrfs_set_trans_block_group(trans, new_dir);
2719         path = btrfs_alloc_path();
2720         if (!path) {
2721                 ret = -ENOMEM;
2722                 goto out_fail;
2723         }
2724
2725         old_dentry->d_inode->i_nlink++;
2726         old_dir->i_ctime = old_dir->i_mtime = ctime;
2727         new_dir->i_ctime = new_dir->i_mtime = ctime;
2728         old_inode->i_ctime = ctime;
2729
2730         ret = btrfs_unlink_trans(trans, root, old_dir, old_dentry);
2731         if (ret)
2732                 goto out_fail;
2733
2734         if (new_inode) {
2735                 new_inode->i_ctime = CURRENT_TIME;
2736                 ret = btrfs_unlink_trans(trans, root, new_dir, new_dentry);
2737                 if (ret)
2738                         goto out_fail;
2739         }
2740         ret = btrfs_add_link(trans, new_dentry, old_inode);
2741         if (ret)
2742                 goto out_fail;
2743
2744 out_fail:
2745         btrfs_free_path(path);
2746         btrfs_end_transaction(trans, root);
2747 out_unlock:
2748         mutex_unlock(&root->fs_info->fs_mutex);
2749         return ret;
2750 }
2751
2752 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
2753                          const char *symname)
2754 {
2755         struct btrfs_trans_handle *trans;
2756         struct btrfs_root *root = BTRFS_I(dir)->root;
2757         struct btrfs_path *path;
2758         struct btrfs_key key;
2759         struct inode *inode = NULL;
2760         int err;
2761         int drop_inode = 0;
2762         u64 objectid;
2763         int name_len;
2764         int datasize;
2765         unsigned long ptr;
2766         struct btrfs_file_extent_item *ei;
2767         struct extent_buffer *leaf;
2768         unsigned long nr = 0;
2769
2770         name_len = strlen(symname) + 1;
2771         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
2772                 return -ENAMETOOLONG;
2773
2774         mutex_lock(&root->fs_info->fs_mutex);
2775         err = btrfs_check_free_space(root, 1, 0);
2776         if (err)
2777                 goto out_fail;
2778
2779         trans = btrfs_start_transaction(root, 1);
2780         btrfs_set_trans_block_group(trans, dir);
2781
2782         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2783         if (err) {
2784                 err = -ENOSPC;
2785                 goto out_unlock;
2786         }
2787
2788         inode = btrfs_new_inode(trans, root, objectid,
2789                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO);
2790         err = PTR_ERR(inode);
2791         if (IS_ERR(inode))
2792                 goto out_unlock;
2793
2794         btrfs_set_trans_block_group(trans, inode);
2795         err = btrfs_add_nondir(trans, dentry, inode);
2796         if (err)
2797                 drop_inode = 1;
2798         else {
2799                 inode->i_mapping->a_ops = &btrfs_aops;
2800                 inode->i_fop = &btrfs_file_operations;
2801                 inode->i_op = &btrfs_file_inode_operations;
2802                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
2803                                      inode->i_mapping, GFP_NOFS);
2804                 BTRFS_I(inode)->extent_tree.ops = &btrfs_extent_map_ops;
2805         }
2806         dir->i_sb->s_dirt = 1;
2807         btrfs_update_inode_block_group(trans, inode);
2808         btrfs_update_inode_block_group(trans, dir);
2809         if (drop_inode)
2810                 goto out_unlock;
2811
2812         path = btrfs_alloc_path();
2813         BUG_ON(!path);
2814         key.objectid = inode->i_ino;
2815         key.offset = 0;
2816         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
2817         datasize = btrfs_file_extent_calc_inline_size(name_len);
2818         err = btrfs_insert_empty_item(trans, root, path, &key,
2819                                       datasize);
2820         if (err) {
2821                 drop_inode = 1;
2822                 goto out_unlock;
2823         }
2824         leaf = path->nodes[0];
2825         ei = btrfs_item_ptr(leaf, path->slots[0],
2826                             struct btrfs_file_extent_item);
2827         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
2828         btrfs_set_file_extent_type(leaf, ei,
2829                                    BTRFS_FILE_EXTENT_INLINE);
2830         ptr = btrfs_file_extent_inline_start(ei);
2831         write_extent_buffer(leaf, symname, ptr, name_len);
2832         btrfs_mark_buffer_dirty(leaf);
2833         btrfs_free_path(path);
2834
2835         inode->i_op = &btrfs_symlink_inode_operations;
2836         inode->i_mapping->a_ops = &btrfs_symlink_aops;
2837         inode->i_size = name_len - 1;
2838         err = btrfs_update_inode(trans, root, inode);
2839         if (err)
2840                 drop_inode = 1;
2841
2842 out_unlock:
2843         nr = trans->blocks_used;
2844         btrfs_end_transaction(trans, root);
2845 out_fail:
2846         mutex_unlock(&root->fs_info->fs_mutex);
2847         if (drop_inode) {
2848                 inode_dec_link_count(inode);
2849                 iput(inode);
2850         }
2851         btrfs_btree_balance_dirty(root, nr);
2852         btrfs_throttle(root);
2853         return err;
2854 }
2855
2856 static struct inode_operations btrfs_dir_inode_operations = {
2857         .lookup         = btrfs_lookup,
2858         .create         = btrfs_create,
2859         .unlink         = btrfs_unlink,
2860         .link           = btrfs_link,
2861         .mkdir          = btrfs_mkdir,
2862         .rmdir          = btrfs_rmdir,
2863         .rename         = btrfs_rename,
2864         .symlink        = btrfs_symlink,
2865         .setattr        = btrfs_setattr,
2866         .mknod          = btrfs_mknod,
2867         .setxattr       = generic_setxattr,
2868         .getxattr       = generic_getxattr,
2869         .listxattr      = btrfs_listxattr,
2870         .removexattr    = generic_removexattr,
2871 };
2872
2873 static struct inode_operations btrfs_dir_ro_inode_operations = {
2874         .lookup         = btrfs_lookup,
2875 };
2876
2877 static struct file_operations btrfs_dir_file_operations = {
2878         .llseek         = generic_file_llseek,
2879         .read           = generic_read_dir,
2880         .readdir        = btrfs_readdir,
2881         .unlocked_ioctl = btrfs_ioctl,
2882 #ifdef CONFIG_COMPAT
2883         .compat_ioctl   = btrfs_ioctl,
2884 #endif
2885 };
2886
2887 static struct extent_map_ops btrfs_extent_map_ops = {
2888         .fill_delalloc = run_delalloc_range,
2889         .writepage_io_hook = btrfs_writepage_io_hook,
2890         .readpage_io_hook = btrfs_readpage_io_hook,
2891         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
2892 };
2893
2894 static struct address_space_operations btrfs_aops = {
2895         .readpage       = btrfs_readpage,
2896         .writepage      = btrfs_writepage,
2897         .writepages     = btrfs_writepages,
2898         .readpages      = btrfs_readpages,
2899         .sync_page      = block_sync_page,
2900         .bmap           = btrfs_bmap,
2901         .invalidatepage = btrfs_invalidatepage,
2902         .releasepage    = btrfs_releasepage,
2903         .set_page_dirty = __set_page_dirty_nobuffers,
2904 };
2905
2906 static struct address_space_operations btrfs_symlink_aops = {
2907         .readpage       = btrfs_readpage,
2908         .writepage      = btrfs_writepage,
2909         .invalidatepage = btrfs_invalidatepage,
2910         .releasepage    = btrfs_releasepage,
2911 };
2912
2913 static struct inode_operations btrfs_file_inode_operations = {
2914         .truncate       = btrfs_truncate,
2915         .getattr        = btrfs_getattr,
2916         .setattr        = btrfs_setattr,
2917         .setxattr       = generic_setxattr,
2918         .getxattr       = generic_getxattr,
2919         .listxattr      = btrfs_listxattr,
2920         .removexattr    = generic_removexattr,
2921 };
2922
2923 static struct inode_operations btrfs_special_inode_operations = {
2924         .getattr        = btrfs_getattr,
2925         .setattr        = btrfs_setattr,
2926 };
2927
2928 static struct inode_operations btrfs_symlink_inode_operations = {
2929         .readlink       = generic_readlink,
2930         .follow_link    = page_follow_link_light,
2931         .put_link       = page_put_link,
2932 };