Btrfs: drop the inode map tree
[safe/jmp/linux-2.6] / fs / btrfs / super.c
1 #include <linux/module.h>
2 #include <linux/buffer_head.h>
3 #include <linux/fs.h>
4 #include <linux/pagemap.h>
5 #include <linux/highmem.h>
6 #include <linux/time.h>
7 #include <linux/init.h>
8 #include <linux/string.h>
9 #include <linux/smp_lock.h>
10 #include <linux/backing-dev.h>
11 #include <linux/mpage.h>
12 #include <linux/swap.h>
13 #include <linux/writeback.h>
14 #include "ctree.h"
15 #include "disk-io.h"
16 #include "transaction.h"
17 #include "btrfs_inode.h"
18 #include "ioctl.h"
19
20 void btrfs_fsinfo_release(struct kobject *obj)
21 {
22         struct btrfs_fs_info *fsinfo = container_of(obj,
23                                             struct btrfs_fs_info, kobj);
24         kfree(fsinfo);
25 }
26
27 struct kobj_type btrfs_fsinfo_ktype = {
28         .release = btrfs_fsinfo_release,
29 };
30
31 struct btrfs_iget_args {
32         u64 ino;
33         struct btrfs_root *root;
34 };
35
36 decl_subsys(btrfs, &btrfs_fsinfo_ktype, NULL);
37
38 #define BTRFS_SUPER_MAGIC 0x9123682E
39
40 static struct inode_operations btrfs_dir_inode_operations;
41 static struct inode_operations btrfs_dir_ro_inode_operations;
42 static struct super_operations btrfs_super_ops;
43 static struct file_operations btrfs_dir_file_operations;
44 static struct inode_operations btrfs_file_inode_operations;
45 static struct address_space_operations btrfs_aops;
46 static struct file_operations btrfs_file_operations;
47
48 static void btrfs_read_locked_inode(struct inode *inode)
49 {
50         struct btrfs_path *path;
51         struct btrfs_inode_item *inode_item;
52         struct btrfs_root *root = BTRFS_I(inode)->root;
53         struct btrfs_key location;
54         int ret;
55
56         path = btrfs_alloc_path();
57         BUG_ON(!path);
58         btrfs_init_path(path);
59         mutex_lock(&root->fs_info->fs_mutex);
60
61         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
62         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
63         if (ret) {
64                 btrfs_free_path(path);
65                 goto make_bad;
66         }
67         inode_item = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
68                                   path->slots[0],
69                                   struct btrfs_inode_item);
70
71         inode->i_mode = btrfs_inode_mode(inode_item);
72         inode->i_nlink = btrfs_inode_nlink(inode_item);
73         inode->i_uid = btrfs_inode_uid(inode_item);
74         inode->i_gid = btrfs_inode_gid(inode_item);
75         inode->i_size = btrfs_inode_size(inode_item);
76         inode->i_atime.tv_sec = btrfs_timespec_sec(&inode_item->atime);
77         inode->i_atime.tv_nsec = btrfs_timespec_nsec(&inode_item->atime);
78         inode->i_mtime.tv_sec = btrfs_timespec_sec(&inode_item->mtime);
79         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(&inode_item->mtime);
80         inode->i_ctime.tv_sec = btrfs_timespec_sec(&inode_item->ctime);
81         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(&inode_item->ctime);
82         inode->i_blocks = btrfs_inode_nblocks(inode_item);
83         inode->i_generation = btrfs_inode_generation(inode_item);
84
85         btrfs_free_path(path);
86         inode_item = NULL;
87
88         mutex_unlock(&root->fs_info->fs_mutex);
89
90         switch (inode->i_mode & S_IFMT) {
91 #if 0
92         default:
93                 init_special_inode(inode, inode->i_mode,
94                                    btrfs_inode_rdev(inode_item));
95                 break;
96 #endif
97         case S_IFREG:
98                 inode->i_mapping->a_ops = &btrfs_aops;
99                 inode->i_fop = &btrfs_file_operations;
100                 inode->i_op = &btrfs_file_inode_operations;
101                 break;
102         case S_IFDIR:
103                 inode->i_fop = &btrfs_dir_file_operations;
104                 if (root == root->fs_info->tree_root)
105                         inode->i_op = &btrfs_dir_ro_inode_operations;
106                 else
107                         inode->i_op = &btrfs_dir_inode_operations;
108                 break;
109         case S_IFLNK:
110                 // inode->i_op = &page_symlink_inode_operations;
111                 break;
112         }
113         return;
114
115 make_bad:
116         btrfs_release_path(root, path);
117         btrfs_free_path(path);
118         mutex_unlock(&root->fs_info->fs_mutex);
119         make_bad_inode(inode);
120 }
121
122 static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
123                               struct btrfs_root *root,
124                               struct inode *dir,
125                               struct dentry *dentry)
126 {
127         struct btrfs_path *path;
128         const char *name = dentry->d_name.name;
129         int name_len = dentry->d_name.len;
130         int ret;
131         u64 objectid;
132         struct btrfs_dir_item *di;
133
134         path = btrfs_alloc_path();
135         BUG_ON(!path);
136         btrfs_init_path(path);
137         ret = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
138                                     name, name_len, -1);
139         if (ret < 0)
140                 goto err;
141         if (ret > 0) {
142                 ret = -ENOENT;
143                 goto err;
144         }
145         di = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]), path->slots[0],
146                             struct btrfs_dir_item);
147         objectid = btrfs_disk_key_objectid(&di->location);
148
149         ret = btrfs_del_item(trans, root, path);
150         BUG_ON(ret);
151
152         btrfs_release_path(root, path);
153         ret = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
154                                           objectid, -1);
155         BUG_ON(ret);
156         ret = btrfs_del_item(trans, root, path);
157         BUG_ON(ret);
158         dentry->d_inode->i_ctime = dir->i_ctime;
159 err:
160         btrfs_release_path(root, path);
161         btrfs_free_path(path);
162         if (ret == 0) {
163                 inode_dec_link_count(dentry->d_inode);
164                 dir->i_size -= name_len * 2;
165                 mark_inode_dirty(dir);
166         }
167         return ret;
168 }
169
170 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
171 {
172         struct btrfs_root *root;
173         struct btrfs_trans_handle *trans;
174         int ret;
175
176         root = BTRFS_I(dir)->root;
177         mutex_lock(&root->fs_info->fs_mutex);
178         trans = btrfs_start_transaction(root, 1);
179         ret = btrfs_unlink_trans(trans, root, dir, dentry);
180         btrfs_end_transaction(trans, root);
181         mutex_unlock(&root->fs_info->fs_mutex);
182         return ret;
183 }
184
185 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
186 {
187         struct inode *inode = dentry->d_inode;
188         int err;
189         int ret;
190         struct btrfs_root *root = BTRFS_I(dir)->root;
191         struct btrfs_path *path;
192         struct btrfs_key key;
193         struct btrfs_trans_handle *trans;
194         struct btrfs_key found_key;
195         int found_type;
196         struct btrfs_leaf *leaf;
197         char *goodnames = "..";
198
199         path = btrfs_alloc_path();
200         BUG_ON(!path);
201         btrfs_init_path(path);
202         mutex_lock(&root->fs_info->fs_mutex);
203         trans = btrfs_start_transaction(root, 1);
204         key.objectid = inode->i_ino;
205         key.offset = (u64)-1;
206         key.flags = (u32)-1;
207         while(1) {
208                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
209                 if (ret < 0) {
210                         err = ret;
211                         goto out;
212                 }
213                 BUG_ON(ret == 0);
214                 if (path->slots[0] == 0) {
215                         err = -ENOENT;
216                         goto out;
217                 }
218                 path->slots[0]--;
219                 leaf = btrfs_buffer_leaf(path->nodes[0]);
220                 btrfs_disk_key_to_cpu(&found_key,
221                                       &leaf->items[path->slots[0]].key);
222                 found_type = btrfs_key_type(&found_key);
223                 if (found_key.objectid != inode->i_ino) {
224                         err = -ENOENT;
225                         goto out;
226                 }
227                 if ((found_type != BTRFS_DIR_ITEM_KEY &&
228                      found_type != BTRFS_DIR_INDEX_KEY) ||
229                     (!btrfs_match_dir_item_name(root, path, goodnames, 2) &&
230                     !btrfs_match_dir_item_name(root, path, goodnames, 1))) {
231                         err = -ENOTEMPTY;
232                         goto out;
233                 }
234                 ret = btrfs_del_item(trans, root, path);
235                 BUG_ON(ret);
236
237                 if (found_type == BTRFS_DIR_ITEM_KEY && found_key.offset == 1)
238                         break;
239                 btrfs_release_path(root, path);
240         }
241         ret = 0;
242         btrfs_release_path(root, path);
243
244         /* now the directory is empty */
245         err = btrfs_unlink_trans(trans, root, dir, dentry);
246         if (!err) {
247                 inode->i_size = 0;
248         }
249 out:
250         btrfs_release_path(root, path);
251         btrfs_free_path(path);
252         mutex_unlock(&root->fs_info->fs_mutex);
253         ret = btrfs_end_transaction(trans, root);
254         if (ret && !err)
255                 err = ret;
256         return err;
257 }
258
259 static int btrfs_free_inode(struct btrfs_trans_handle *trans,
260                             struct btrfs_root *root,
261                             struct inode *inode)
262 {
263         struct btrfs_path *path;
264         int ret;
265
266         clear_inode(inode);
267
268         path = btrfs_alloc_path();
269         BUG_ON(!path);
270         btrfs_init_path(path);
271         ret = btrfs_lookup_inode(trans, root, path,
272                                  &BTRFS_I(inode)->location, -1);
273         BUG_ON(ret);
274         ret = btrfs_del_item(trans, root, path);
275         BUG_ON(ret);
276         btrfs_free_path(path);
277         return ret;
278 }
279
280 static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
281                                    struct btrfs_root *root,
282                                    struct inode *inode)
283 {
284         int ret;
285         struct btrfs_path *path;
286         struct btrfs_key key;
287         struct btrfs_disk_key *found_key;
288         struct btrfs_leaf *leaf;
289         struct btrfs_file_extent_item *fi = NULL;
290         u64 extent_start = 0;
291         u64 extent_num_blocks = 0;
292         int found_extent;
293
294         path = btrfs_alloc_path();
295         BUG_ON(!path);
296         /* FIXME, add redo link to tree so we don't leak on crash */
297         key.objectid = inode->i_ino;
298         key.offset = (u64)-1;
299         key.flags = 0;
300         /*
301          * use BTRFS_CSUM_ITEM_KEY because it is larger than inline keys
302          * or extent data
303          */
304         btrfs_set_key_type(&key, BTRFS_CSUM_ITEM_KEY);
305         while(1) {
306                 btrfs_init_path(path);
307                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
308                 if (ret < 0) {
309                         goto error;
310                 }
311                 if (ret > 0) {
312                         BUG_ON(path->slots[0] == 0);
313                         path->slots[0]--;
314                 }
315                 leaf = btrfs_buffer_leaf(path->nodes[0]);
316                 found_key = &leaf->items[path->slots[0]].key;
317                 if (btrfs_disk_key_objectid(found_key) != inode->i_ino)
318                         break;
319                 if (btrfs_disk_key_type(found_key) != BTRFS_CSUM_ITEM_KEY &&
320                     btrfs_disk_key_type(found_key) != BTRFS_INLINE_DATA_KEY &&
321                     btrfs_disk_key_type(found_key) != BTRFS_EXTENT_DATA_KEY)
322                         break;
323                 if (btrfs_disk_key_offset(found_key) < inode->i_size)
324                         break;
325                 if (btrfs_disk_key_type(found_key) == BTRFS_EXTENT_DATA_KEY) {
326                         fi = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
327                                             path->slots[0],
328                                             struct btrfs_file_extent_item);
329                         extent_start = btrfs_file_extent_disk_blocknr(fi);
330                         extent_num_blocks =
331                                 btrfs_file_extent_disk_num_blocks(fi);
332                         inode->i_blocks -=
333                                 btrfs_file_extent_num_blocks(fi) >> 9;
334                         found_extent = 1;
335                 } else {
336                         found_extent = 0;
337                 }
338                 ret = btrfs_del_item(trans, root, path);
339                 BUG_ON(ret);
340                 btrfs_release_path(root, path);
341                 if (found_extent) {
342                         ret = btrfs_free_extent(trans, root, extent_start,
343                                                 extent_num_blocks, 0);
344                         BUG_ON(ret);
345                 }
346         }
347         ret = 0;
348 error:
349         btrfs_release_path(root, path);
350         btrfs_free_path(path);
351         return ret;
352 }
353
354 static void btrfs_delete_inode(struct inode *inode)
355 {
356         struct btrfs_trans_handle *trans;
357         struct btrfs_root *root = BTRFS_I(inode)->root;
358         int ret;
359
360         truncate_inode_pages(&inode->i_data, 0);
361         if (is_bad_inode(inode)) {
362                 goto no_delete;
363         }
364         inode->i_size = 0;
365         mutex_lock(&root->fs_info->fs_mutex);
366         trans = btrfs_start_transaction(root, 1);
367         if (S_ISREG(inode->i_mode)) {
368                 ret = btrfs_truncate_in_trans(trans, root, inode);
369                 BUG_ON(ret);
370         }
371         btrfs_free_inode(trans, root, inode);
372         btrfs_end_transaction(trans, root);
373         mutex_unlock(&root->fs_info->fs_mutex);
374         return;
375 no_delete:
376         clear_inode(inode);
377 }
378
379 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
380                                struct btrfs_key *location)
381 {
382         const char *name = dentry->d_name.name;
383         int namelen = dentry->d_name.len;
384         struct btrfs_dir_item *di;
385         struct btrfs_path *path;
386         struct btrfs_root *root = BTRFS_I(dir)->root;
387         int ret;
388
389         path = btrfs_alloc_path();
390         BUG_ON(!path);
391         btrfs_init_path(path);
392         ret = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
393                                     namelen, 0);
394         if (ret || !btrfs_match_dir_item_name(root, path, name, namelen)) {
395                 location->objectid = 0;
396                 ret = 0;
397                 goto out;
398         }
399         di = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]), path->slots[0],
400                             struct btrfs_dir_item);
401         btrfs_disk_key_to_cpu(location, &di->location);
402 out:
403         btrfs_release_path(root, path);
404         btrfs_free_path(path);
405         return ret;
406 }
407
408 int fixup_tree_root_location(struct btrfs_root *root,
409                              struct btrfs_key *location,
410                              struct btrfs_root **sub_root)
411 {
412         struct btrfs_path *path;
413         struct btrfs_root_item *ri;
414
415         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
416                 return 0;
417         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
418                 return 0;
419
420         path = btrfs_alloc_path();
421         BUG_ON(!path);
422         mutex_lock(&root->fs_info->fs_mutex);
423
424         *sub_root = btrfs_read_fs_root(root->fs_info, location);
425         if (IS_ERR(*sub_root))
426                 return PTR_ERR(*sub_root);
427
428         ri = &(*sub_root)->root_item;
429         location->objectid = btrfs_root_dirid(ri);
430         location->flags = 0;
431         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
432         location->offset = 0;
433
434         btrfs_free_path(path);
435         mutex_unlock(&root->fs_info->fs_mutex);
436         return 0;
437 }
438
439 int btrfs_init_locked_inode(struct inode *inode, void *p)
440 {
441         struct btrfs_iget_args *args = p;
442         inode->i_ino = args->ino;
443         BTRFS_I(inode)->root = args->root;
444         return 0;
445 }
446
447 int btrfs_find_actor(struct inode *inode, void *opaque)
448 {
449         struct btrfs_iget_args *args = opaque;
450         return (args->ino == inode->i_ino &&
451                 args->root == BTRFS_I(inode)->root);
452 }
453
454 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
455                                 struct btrfs_root *root)
456 {
457         struct inode *inode;
458         struct btrfs_iget_args args;
459         args.ino = objectid;
460         args.root = root;
461
462         inode = iget5_locked(s, objectid, btrfs_find_actor,
463                              btrfs_init_locked_inode,
464                              (void *)&args);
465         return inode;
466 }
467
468 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
469                                    struct nameidata *nd)
470 {
471         struct inode * inode;
472         struct btrfs_inode *bi = BTRFS_I(dir);
473         struct btrfs_root *root = bi->root;
474         struct btrfs_root *sub_root = root;
475         struct btrfs_key location;
476         int ret;
477
478         if (dentry->d_name.len > BTRFS_NAME_LEN)
479                 return ERR_PTR(-ENAMETOOLONG);
480         mutex_lock(&root->fs_info->fs_mutex);
481         ret = btrfs_inode_by_name(dir, dentry, &location);
482         mutex_unlock(&root->fs_info->fs_mutex);
483         if (ret < 0)
484                 return ERR_PTR(ret);
485         inode = NULL;
486         if (location.objectid) {
487                 ret = fixup_tree_root_location(root, &location, &sub_root);
488                 if (ret < 0)
489                         return ERR_PTR(ret);
490                 if (ret > 0)
491                         return ERR_PTR(-ENOENT);
492                 inode = btrfs_iget_locked(dir->i_sb, location.objectid,
493                                           sub_root);
494                 if (!inode)
495                         return ERR_PTR(-EACCES);
496                 if (inode->i_state & I_NEW) {
497                         if (sub_root != root) {
498                                 ret = radix_tree_insert(
499                                                 &root->fs_info->fs_roots_radix,
500                                                 (unsigned long)sub_root,
501                                                 sub_root);
502 printk("adding new root for inode %lu root %p (found %p)\n", inode->i_ino, sub_root, BTRFS_I(inode)->root);
503                                 igrab(inode);
504                                 sub_root->inode = inode;
505                         }
506                         BTRFS_I(inode)->root = sub_root;
507                         memcpy(&BTRFS_I(inode)->location, &location,
508                                sizeof(location));
509                         btrfs_read_locked_inode(inode);
510                         unlock_new_inode(inode);
511                 }
512         }
513         return d_splice_alias(inode, dentry);
514 }
515
516 static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
517 {
518         struct inode *inode = filp->f_path.dentry->d_inode;
519         struct btrfs_root *root = BTRFS_I(inode)->root;
520         struct btrfs_item *item;
521         struct btrfs_dir_item *di;
522         struct btrfs_key key;
523         struct btrfs_path *path;
524         int ret;
525         u32 nritems;
526         struct btrfs_leaf *leaf;
527         int slot;
528         int advance;
529         unsigned char d_type = DT_UNKNOWN;
530         int over = 0;
531         int key_type = BTRFS_DIR_INDEX_KEY;
532
533         /* FIXME, use a real flag for deciding about the key type */
534         if (root->fs_info->tree_root == root)
535                 key_type = BTRFS_DIR_ITEM_KEY;
536         mutex_lock(&root->fs_info->fs_mutex);
537         key.objectid = inode->i_ino;
538         key.flags = 0;
539         btrfs_set_key_type(&key, key_type);
540         key.offset = filp->f_pos;
541         path = btrfs_alloc_path();
542         btrfs_init_path(path);
543         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
544         if (ret < 0)
545                 goto err;
546         advance = 0;
547         while(1) {
548                 leaf = btrfs_buffer_leaf(path->nodes[0]);
549                 nritems = btrfs_header_nritems(&leaf->header);
550                 slot = path->slots[0];
551                 if (advance || slot >= nritems) {
552                         if (slot >= nritems -1) {
553                                 ret = btrfs_next_leaf(root, path);
554                                 if (ret)
555                                         break;
556                                 leaf = btrfs_buffer_leaf(path->nodes[0]);
557                                 nritems = btrfs_header_nritems(&leaf->header);
558                                 slot = path->slots[0];
559                         } else {
560                                 slot++;
561                                 path->slots[0]++;
562                         }
563                 }
564                 advance = 1;
565                 item = leaf->items + slot;
566                 if (btrfs_disk_key_objectid(&item->key) != key.objectid)
567                         break;
568                 if (key_type == BTRFS_DIR_INDEX_KEY &&
569                     btrfs_disk_key_offset(&item->key) > root->highest_inode)
570                         break;
571                 if (btrfs_disk_key_type(&item->key) != key_type)
572                         continue;
573                 if (btrfs_disk_key_offset(&item->key) < filp->f_pos)
574                         continue;
575                 filp->f_pos = btrfs_disk_key_offset(&item->key);
576                 advance = 1;
577                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
578                 over = filldir(dirent, (const char *)(di + 1),
579                                btrfs_dir_name_len(di),
580                                btrfs_disk_key_offset(&item->key),
581                                btrfs_disk_key_objectid(&di->location), d_type);
582                 if (over)
583                         goto nopos;
584         }
585         filp->f_pos++;
586 nopos:
587         ret = 0;
588 err:
589         btrfs_release_path(root, path);
590         btrfs_free_path(path);
591         mutex_unlock(&root->fs_info->fs_mutex);
592         return ret;
593 }
594
595 static void btrfs_put_super (struct super_block * sb)
596 {
597         struct btrfs_root *root = btrfs_sb(sb);
598         int ret;
599
600         ret = close_ctree(root);
601         if (ret) {
602                 printk("close ctree returns %d\n", ret);
603         }
604         sb->s_fs_info = NULL;
605 }
606
607 static int btrfs_fill_super(struct super_block * sb, void * data, int silent)
608 {
609         struct inode * inode;
610         struct dentry * root_dentry;
611         struct btrfs_super_block *disk_super;
612         struct btrfs_root *tree_root;
613         struct btrfs_inode *bi;
614
615         sb->s_maxbytes = MAX_LFS_FILESIZE;
616         sb->s_magic = BTRFS_SUPER_MAGIC;
617         sb->s_op = &btrfs_super_ops;
618         sb->s_time_gran = 1;
619
620         tree_root = open_ctree(sb);
621
622         if (!tree_root) {
623                 printk("btrfs: open_ctree failed\n");
624                 return -EIO;
625         }
626         sb->s_fs_info = tree_root;
627         disk_super = tree_root->fs_info->disk_super;
628         printk("read in super total blocks %Lu root %Lu\n",
629                btrfs_super_total_blocks(disk_super),
630                btrfs_super_root_dir(disk_super));
631
632         inode = btrfs_iget_locked(sb, btrfs_super_root_dir(disk_super),
633                                   tree_root);
634         bi = BTRFS_I(inode);
635         bi->location.objectid = inode->i_ino;
636         bi->location.offset = 0;
637         bi->location.flags = 0;
638         bi->root = tree_root;
639         btrfs_set_key_type(&bi->location, BTRFS_INODE_ITEM_KEY);
640
641         if (!inode)
642                 return -ENOMEM;
643         if (inode->i_state & I_NEW) {
644                 btrfs_read_locked_inode(inode);
645                 unlock_new_inode(inode);
646         }
647
648         root_dentry = d_alloc_root(inode);
649         if (!root_dentry) {
650                 iput(inode);
651                 return -ENOMEM;
652         }
653         sb->s_root = root_dentry;
654
655         return 0;
656 }
657
658 static void fill_inode_item(struct btrfs_inode_item *item,
659                             struct inode *inode)
660 {
661         btrfs_set_inode_uid(item, inode->i_uid);
662         btrfs_set_inode_gid(item, inode->i_gid);
663         btrfs_set_inode_size(item, inode->i_size);
664         btrfs_set_inode_mode(item, inode->i_mode);
665         btrfs_set_inode_nlink(item, inode->i_nlink);
666         btrfs_set_timespec_sec(&item->atime, inode->i_atime.tv_sec);
667         btrfs_set_timespec_nsec(&item->atime, inode->i_atime.tv_nsec);
668         btrfs_set_timespec_sec(&item->mtime, inode->i_mtime.tv_sec);
669         btrfs_set_timespec_nsec(&item->mtime, inode->i_mtime.tv_nsec);
670         btrfs_set_timespec_sec(&item->ctime, inode->i_ctime.tv_sec);
671         btrfs_set_timespec_nsec(&item->ctime, inode->i_ctime.tv_nsec);
672         btrfs_set_inode_nblocks(item, inode->i_blocks);
673         btrfs_set_inode_generation(item, inode->i_generation);
674 }
675
676 static int btrfs_update_inode(struct btrfs_trans_handle *trans,
677                               struct btrfs_root *root,
678                               struct inode *inode)
679 {
680         struct btrfs_inode_item *inode_item;
681         struct btrfs_path *path;
682         int ret;
683
684         path = btrfs_alloc_path();
685         BUG_ON(!path);
686         btrfs_init_path(path);
687         ret = btrfs_lookup_inode(trans, root, path,
688                                  &BTRFS_I(inode)->location, 1);
689         if (ret) {
690                 if (ret > 0)
691                         ret = -ENOENT;
692                 goto failed;
693         }
694
695         inode_item = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
696                                   path->slots[0],
697                                   struct btrfs_inode_item);
698
699         fill_inode_item(inode_item, inode);
700         btrfs_mark_buffer_dirty(path->nodes[0]);
701         ret = 0;
702 failed:
703         btrfs_release_path(root, path);
704         btrfs_free_path(path);
705         return ret;
706 }
707
708 static int btrfs_write_inode(struct inode *inode, int wait)
709 {
710         struct btrfs_root *root = BTRFS_I(inode)->root;
711         struct btrfs_trans_handle *trans;
712         int ret;
713
714         mutex_lock(&root->fs_info->fs_mutex);
715         trans = btrfs_start_transaction(root, 1);
716         ret = btrfs_update_inode(trans, root, inode);
717         if (wait)
718                 btrfs_commit_transaction(trans, root);
719         else
720                 btrfs_end_transaction(trans, root);
721         mutex_unlock(&root->fs_info->fs_mutex);
722         return ret;
723 }
724
725 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
726                                      struct inode *dir, int mode)
727 {
728         struct inode *inode;
729         struct btrfs_inode_item inode_item;
730         struct btrfs_root *root = BTRFS_I(dir)->root;
731         struct btrfs_key *location;
732         int ret;
733         u64 objectid;
734
735         inode = new_inode(dir->i_sb);
736         if (!inode)
737                 return ERR_PTR(-ENOMEM);
738
739         BTRFS_I(inode)->root = BTRFS_I(dir)->root;
740         ret = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
741         BUG_ON(ret);
742
743         inode->i_uid = current->fsuid;
744         inode->i_gid = current->fsgid;
745         inode->i_mode = mode;
746         inode->i_ino = objectid;
747         inode->i_blocks = 0;
748         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
749         fill_inode_item(&inode_item, inode);
750         location = &BTRFS_I(inode)->location;
751         location->objectid = objectid;
752         location->flags = 0;
753         location->offset = 0;
754         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
755
756         ret = btrfs_insert_inode(trans, root, objectid, &inode_item);
757         BUG_ON(ret);
758
759         insert_inode_hash(inode);
760         return inode;
761 }
762
763 static int btrfs_add_link(struct btrfs_trans_handle *trans,
764                             struct dentry *dentry, struct inode *inode)
765 {
766         int ret;
767         struct btrfs_key key;
768         struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
769         key.objectid = inode->i_ino;
770         key.flags = 0;
771         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
772         key.offset = 0;
773
774         ret = btrfs_insert_dir_item(trans, root,
775                                     dentry->d_name.name, dentry->d_name.len,
776                                     dentry->d_parent->d_inode->i_ino,
777                                     &key, 0);
778         if (ret == 0) {
779                 dentry->d_parent->d_inode->i_size += dentry->d_name.len * 2;
780                 ret = btrfs_update_inode(trans, root,
781                                          dentry->d_parent->d_inode);
782         }
783         return ret;
784 }
785
786 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
787                             struct dentry *dentry, struct inode *inode)
788 {
789         int err = btrfs_add_link(trans, dentry, inode);
790         if (!err) {
791                 d_instantiate(dentry, inode);
792                 return 0;
793         }
794         if (err > 0)
795                 err = -EEXIST;
796         return err;
797 }
798
799 static int btrfs_create(struct inode *dir, struct dentry *dentry,
800                         int mode, struct nameidata *nd)
801 {
802         struct btrfs_trans_handle *trans;
803         struct btrfs_root *root = BTRFS_I(dir)->root;
804         struct inode *inode;
805         int err;
806         int drop_inode = 0;
807
808         mutex_lock(&root->fs_info->fs_mutex);
809         trans = btrfs_start_transaction(root, 1);
810         inode = btrfs_new_inode(trans, dir, mode);
811         err = PTR_ERR(inode);
812         if (IS_ERR(inode))
813                 goto out_unlock;
814         // FIXME mark the inode dirty
815         err = btrfs_add_nondir(trans, dentry, inode);
816         if (err)
817                 drop_inode = 1;
818         else {
819                 inode->i_mapping->a_ops = &btrfs_aops;
820                 inode->i_fop = &btrfs_file_operations;
821                 inode->i_op = &btrfs_file_inode_operations;
822         }
823         dir->i_sb->s_dirt = 1;
824 out_unlock:
825         btrfs_end_transaction(trans, root);
826         mutex_unlock(&root->fs_info->fs_mutex);
827
828         if (drop_inode) {
829                 inode_dec_link_count(inode);
830                 iput(inode);
831         }
832         return err;
833 }
834
835 static int btrfs_make_empty_dir(struct btrfs_trans_handle *trans,
836                                 struct inode *inode, struct inode *dir)
837 {
838         struct btrfs_root *root = BTRFS_I(dir)->root;
839         int ret;
840         char buf[2];
841         struct btrfs_key key;
842
843         buf[0] = '.';
844         buf[1] = '.';
845
846         key.objectid = inode->i_ino;
847         key.offset = 0;
848         key.flags = 0;
849         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
850
851         ret = btrfs_insert_dir_item(trans, root, buf, 1, inode->i_ino,
852                                     &key, 1);
853         if (ret)
854                 goto error;
855         key.objectid = dir->i_ino;
856         ret = btrfs_insert_dir_item(trans, root, buf, 2, inode->i_ino,
857                                     &key, 1);
858         if (ret)
859                 goto error;
860         inode->i_size = 6;
861         ret = btrfs_update_inode(trans, root, inode);
862 error:
863         return ret;
864 }
865
866 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
867 {
868         struct inode *inode;
869         struct btrfs_trans_handle *trans;
870         struct btrfs_root *root = BTRFS_I(dir)->root;
871         int err = 0;
872         int drop_on_err = 0;
873
874         mutex_lock(&root->fs_info->fs_mutex);
875         trans = btrfs_start_transaction(root, 1);
876         if (IS_ERR(trans)) {
877                 err = PTR_ERR(trans);
878                 goto out_unlock;
879         }
880         inode = btrfs_new_inode(trans, dir, S_IFDIR | mode);
881         if (IS_ERR(inode)) {
882                 err = PTR_ERR(inode);
883                 goto out_fail;
884         }
885         drop_on_err = 1;
886         inode->i_op = &btrfs_dir_inode_operations;
887         inode->i_fop = &btrfs_dir_file_operations;
888
889         err = btrfs_make_empty_dir(trans, inode, dir);
890         if (err)
891                 goto out_fail;
892         err = btrfs_add_link(trans, dentry, inode);
893         if (err)
894                 goto out_fail;
895         d_instantiate(dentry, inode);
896         drop_on_err = 0;
897
898 out_fail:
899         btrfs_end_transaction(trans, root);
900 out_unlock:
901         mutex_unlock(&root->fs_info->fs_mutex);
902         if (drop_on_err)
903                 iput(inode);
904         return err;
905 }
906
907 static int btrfs_sync_fs(struct super_block *sb, int wait)
908 {
909         struct btrfs_trans_handle *trans;
910         struct btrfs_root *root;
911         int ret;
912         root = btrfs_sb(sb);
913
914         sb->s_dirt = 0;
915         if (!wait) {
916                 filemap_flush(root->fs_info->btree_inode->i_mapping);
917                 return 0;
918         }
919         filemap_write_and_wait(root->fs_info->btree_inode->i_mapping);
920         mutex_lock(&root->fs_info->fs_mutex);
921         trans = btrfs_start_transaction(root, 1);
922         ret = btrfs_commit_transaction(trans, root);
923         sb->s_dirt = 0;
924         BUG_ON(ret);
925 printk("btrfs sync_fs\n");
926         mutex_unlock(&root->fs_info->fs_mutex);
927         return 0;
928 }
929
930 #if 0
931 static int btrfs_get_block_inline(struct inode *inode, sector_t iblock,
932                            struct buffer_head *result, int create)
933 {
934         struct btrfs_root *root = btrfs_sb(inode->i_sb);
935         struct btrfs_path *path;
936         struct btrfs_key key;
937         struct btrfs_leaf *leaf;
938         int num_bytes = result->b_size;
939         int item_size;
940         int ret;
941         u64 pos;
942         char *ptr;
943         int copy_size;
944         int err = 0;
945         char *safe_ptr;
946         char *data_ptr;
947
948         path = btrfs_alloc_path();
949         BUG_ON(!path);
950
951         WARN_ON(create);
952         if (create) {
953                 return 0;
954         }
955         pos = iblock << inode->i_blkbits;
956         key.objectid = inode->i_ino;
957         key.flags = 0;
958         btrfs_set_key_type(&key, BTRFS_INLINE_DATA_KEY);
959         ptr = kmap(result->b_page);
960         safe_ptr = ptr;
961         ptr += (pos & (PAGE_CACHE_SIZE -1));
962 again:
963         key.offset = pos;
964         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
965         if (ret) {
966                 if (ret < 0)
967                         err = ret;
968                 else
969                         err = 0;
970                 goto out;
971         }
972         leaf = btrfs_buffer_leaf(path->nodes[0]);
973         item_size = btrfs_item_size(leaf->items + path->slots[0]);
974         copy_size = min(num_bytes, item_size);
975         data_ptr = btrfs_item_ptr(leaf, path->slots[0], char);
976         WARN_ON(safe_ptr + PAGE_CACHE_SIZE < ptr + copy_size);
977         memcpy(ptr, data_ptr, copy_size);
978         pos += copy_size;
979         num_bytes -= copy_size;
980         WARN_ON(num_bytes < 0);
981         ptr += copy_size;
982         btrfs_release_path(root, path);
983         if (num_bytes != 0) {
984                 if (pos >= i_size_read(inode))
985                         memset(ptr, 0, num_bytes);
986                 else
987                         goto again;
988         }
989         set_buffer_uptodate(result);
990         map_bh(result, inode->i_sb, 0);
991         err = 0;
992 out:
993         btrfs_free_path(path);
994         kunmap(result->b_page);
995         return err;
996 }
997 #endif
998
999 static int btrfs_get_block_lock(struct inode *inode, sector_t iblock,
1000                            struct buffer_head *result, int create)
1001 {
1002         int ret;
1003         int err = 0;
1004         u64 blocknr;
1005         u64 extent_start = 0;
1006         u64 extent_end = 0;
1007         u64 objectid = inode->i_ino;
1008         struct btrfs_path *path;
1009         struct btrfs_root *root = BTRFS_I(inode)->root;
1010         struct btrfs_trans_handle *trans = NULL;
1011         struct btrfs_file_extent_item *item;
1012         struct btrfs_leaf *leaf;
1013         struct btrfs_disk_key *found_key;
1014
1015         path = btrfs_alloc_path();
1016         BUG_ON(!path);
1017         btrfs_init_path(path);
1018         if (create)
1019                 trans = btrfs_start_transaction(root, 1);
1020
1021         ret = btrfs_lookup_file_extent(trans, root, path,
1022                                        inode->i_ino,
1023                                        iblock << inode->i_blkbits, 0);
1024         if (ret < 0) {
1025                 err = ret;
1026                 goto out;
1027         }
1028
1029         if (ret != 0) {
1030                 if (path->slots[0] == 0) {
1031                         btrfs_release_path(root, path);
1032                         goto allocate;
1033                 }
1034                 path->slots[0]--;
1035         }
1036
1037         item = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]), path->slots[0],
1038                               struct btrfs_file_extent_item);
1039         leaf = btrfs_buffer_leaf(path->nodes[0]);
1040         blocknr = btrfs_file_extent_disk_blocknr(item);
1041         blocknr += btrfs_file_extent_offset(item);
1042
1043         /* exact match found, use it */
1044         if (ret == 0) {
1045                 err = 0;
1046                 map_bh(result, inode->i_sb, blocknr);
1047                 goto out;
1048         }
1049
1050         /* are we inside the extent that was found? */
1051         found_key = &leaf->items[path->slots[0]].key;
1052         if (btrfs_disk_key_objectid(found_key) != objectid ||
1053             btrfs_disk_key_type(found_key) != BTRFS_EXTENT_DATA_KEY) {
1054                 extent_end = 0;
1055                 extent_start = 0;
1056                 btrfs_release_path(root, path);
1057                 goto allocate;
1058         }
1059
1060         extent_start = btrfs_disk_key_offset(&leaf->items[path->slots[0]].key);
1061         extent_start = extent_start >> inode->i_blkbits;
1062         extent_start += btrfs_file_extent_offset(item);
1063         extent_end = extent_start + btrfs_file_extent_num_blocks(item);
1064         if (iblock >= extent_start && iblock < extent_end) {
1065                 err = 0;
1066                 map_bh(result, inode->i_sb, blocknr + iblock - extent_start);
1067                 goto out;
1068         }
1069 allocate:
1070         /* ok, create a new extent */
1071         if (!create) {
1072                 err = 0;
1073                 goto out;
1074         }
1075         ret = btrfs_alloc_file_extent(trans, root, objectid,
1076                                       iblock << inode->i_blkbits,
1077                                       1, extent_end, &blocknr);
1078         if (ret) {
1079                 err = ret;
1080                 goto out;
1081         }
1082         inode->i_blocks += inode->i_sb->s_blocksize >> 9;
1083         set_buffer_new(result);
1084         map_bh(result, inode->i_sb, blocknr);
1085
1086 out:
1087         btrfs_release_path(root, path);
1088         btrfs_free_path(path);
1089         if (trans)
1090                 btrfs_end_transaction(trans, root);
1091         return err;
1092 }
1093
1094 static int btrfs_get_block(struct inode *inode, sector_t iblock,
1095                            struct buffer_head *result, int create)
1096 {
1097         int err;
1098         struct btrfs_root *root = BTRFS_I(inode)->root;
1099         mutex_lock(&root->fs_info->fs_mutex);
1100         err = btrfs_get_block_lock(inode, iblock, result, create);
1101         // err = btrfs_get_block_inline(inode, iblock, result, create);
1102         mutex_unlock(&root->fs_info->fs_mutex);
1103         return err;
1104 }
1105
1106 static int btrfs_prepare_write(struct file *file, struct page *page,
1107                                unsigned from, unsigned to)
1108 {
1109         return nobh_prepare_write(page, from, to, btrfs_get_block);
1110 }
1111 static int btrfs_commit_write(struct file *file, struct page *page,
1112                                unsigned from, unsigned to)
1113 {
1114         return nobh_commit_write(file, page, from, to);
1115 }
1116
1117 static void btrfs_write_super(struct super_block *sb)
1118 {
1119         btrfs_sync_fs(sb, 1);
1120 }
1121
1122 static int btrfs_readpage(struct file *file, struct page *page)
1123 {
1124         return mpage_readpage(page, btrfs_get_block);
1125 }
1126
1127 static int btrfs_readpages(struct file *file, struct address_space *mapping,
1128                            struct list_head *pages, unsigned nr_pages)
1129 {
1130         return mpage_readpages(mapping, pages, nr_pages, btrfs_get_block);
1131 }
1132
1133 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
1134 {
1135         return nobh_writepage(page, btrfs_get_block, wbc);
1136 }
1137
1138 static void btrfs_truncate(struct inode *inode)
1139 {
1140         struct btrfs_root *root = BTRFS_I(inode)->root;
1141         int ret;
1142         struct btrfs_trans_handle *trans;
1143
1144         if (!S_ISREG(inode->i_mode))
1145                 return;
1146         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1147                 return;
1148
1149         nobh_truncate_page(inode->i_mapping, inode->i_size);
1150
1151         /* FIXME, add redo link to tree so we don't leak on crash */
1152         mutex_lock(&root->fs_info->fs_mutex);
1153         trans = btrfs_start_transaction(root, 1);
1154         ret = btrfs_truncate_in_trans(trans, root, inode);
1155         BUG_ON(ret);
1156         ret = btrfs_end_transaction(trans, root);
1157         BUG_ON(ret);
1158         mutex_unlock(&root->fs_info->fs_mutex);
1159         mark_inode_dirty(inode);
1160 }
1161
1162 static int btrfs_copy_from_user(loff_t pos, int num_pages, int write_bytes,
1163                                 struct page **prepared_pages,
1164                                 const char __user * buf)
1165 {
1166         long page_fault = 0;
1167         int i;
1168         int offset = pos & (PAGE_CACHE_SIZE - 1);
1169
1170         for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
1171                 size_t count = min_t(size_t,
1172                                      PAGE_CACHE_SIZE - offset, write_bytes);
1173                 struct page *page = prepared_pages[i];
1174                 fault_in_pages_readable(buf, count);
1175
1176                 /* Copy data from userspace to the current page */
1177                 kmap(page);
1178                 page_fault = __copy_from_user(page_address(page) + offset,
1179                                               buf, count);
1180                 /* Flush processor's dcache for this page */
1181                 flush_dcache_page(page);
1182                 kunmap(page);
1183                 buf += count;
1184                 write_bytes -= count;
1185
1186                 if (page_fault)
1187                         break;
1188         }
1189         return page_fault ? -EFAULT : 0;
1190 }
1191
1192 static void btrfs_drop_pages(struct page **pages, size_t num_pages)
1193 {
1194         size_t i;
1195         for (i = 0; i < num_pages; i++) {
1196                 if (!pages[i])
1197                         break;
1198                 unlock_page(pages[i]);
1199                 mark_page_accessed(pages[i]);
1200                 page_cache_release(pages[i]);
1201         }
1202 }
1203 static int dirty_and_release_pages(struct btrfs_trans_handle *trans,
1204                                    struct btrfs_root *root,
1205                                    struct file *file,
1206                                    struct page **pages,
1207                                    size_t num_pages,
1208                                    loff_t pos,
1209                                    size_t write_bytes)
1210 {
1211         int i;
1212         int offset;
1213         int err = 0;
1214         int ret;
1215         int this_write;
1216         struct inode *inode = file->f_path.dentry->d_inode;
1217
1218         for (i = 0; i < num_pages; i++) {
1219                 offset = pos & (PAGE_CACHE_SIZE -1);
1220                 this_write = min(PAGE_CACHE_SIZE - offset, write_bytes);
1221                 /* FIXME, one block at a time */
1222
1223                 mutex_lock(&root->fs_info->fs_mutex);
1224                 trans = btrfs_start_transaction(root, 1);
1225                 btrfs_csum_file_block(trans, root, inode->i_ino,
1226                                       pages[i]->index << PAGE_CACHE_SHIFT,
1227                                       kmap(pages[i]), PAGE_CACHE_SIZE);
1228                 kunmap(pages[i]);
1229                 SetPageChecked(pages[i]);
1230                 ret = btrfs_end_transaction(trans, root);
1231                 BUG_ON(ret);
1232                 mutex_unlock(&root->fs_info->fs_mutex);
1233
1234                 ret = nobh_commit_write(file, pages[i], offset,
1235                                          offset + this_write);
1236                 pos += this_write;
1237                 if (ret) {
1238                         err = ret;
1239                         goto failed;
1240                 }
1241                 WARN_ON(this_write > write_bytes);
1242                 write_bytes -= this_write;
1243         }
1244 failed:
1245         return err;
1246 }
1247
1248 static int prepare_pages(struct btrfs_trans_handle *trans,
1249                          struct btrfs_root *root,
1250                          struct file *file,
1251                          struct page **pages,
1252                          size_t num_pages,
1253                          loff_t pos,
1254                          size_t write_bytes)
1255 {
1256         int i;
1257         unsigned long index = pos >> PAGE_CACHE_SHIFT;
1258         struct inode *inode = file->f_path.dentry->d_inode;
1259         int offset;
1260         int err = 0;
1261         int ret;
1262         int this_write;
1263         loff_t isize = i_size_read(inode);
1264
1265         memset(pages, 0, num_pages * sizeof(struct page *));
1266
1267         for (i = 0; i < num_pages; i++) {
1268                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
1269                 if (!pages[i]) {
1270                         err = -ENOMEM;
1271                         goto failed_release;
1272                 }
1273                 offset = pos & (PAGE_CACHE_SIZE -1);
1274                 this_write = min(PAGE_CACHE_SIZE - offset, write_bytes);
1275                 ret = nobh_prepare_write(pages[i], offset,
1276                                          offset + this_write,
1277                                          btrfs_get_block);
1278                 pos += this_write;
1279                 if (ret) {
1280                         err = ret;
1281                         goto failed_truncate;
1282                 }
1283                 WARN_ON(this_write > write_bytes);
1284                 write_bytes -= this_write;
1285         }
1286         return 0;
1287
1288 failed_release:
1289         btrfs_drop_pages(pages, num_pages);
1290         return err;
1291
1292 failed_truncate:
1293         btrfs_drop_pages(pages, num_pages);
1294         if (pos > isize)
1295                 vmtruncate(inode, isize);
1296         return err;
1297 }
1298
1299 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
1300                                 size_t count, loff_t *ppos)
1301 {
1302         loff_t pos;
1303         size_t num_written = 0;
1304         int err = 0;
1305         int ret = 0;
1306         struct inode *inode = file->f_path.dentry->d_inode;
1307         struct btrfs_root *root = BTRFS_I(inode)->root;
1308         struct page *pages[1];
1309
1310         if (file->f_flags & O_DIRECT)
1311                 return -EINVAL;
1312         pos = *ppos;
1313
1314         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1315         current->backing_dev_info = inode->i_mapping->backing_dev_info;
1316         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1317         if (err)
1318                 goto out;
1319         if (count == 0)
1320                 goto out;
1321         err = remove_suid(file->f_path.dentry);
1322         if (err)
1323                 goto out;
1324         file_update_time(file);
1325         mutex_lock(&inode->i_mutex);
1326         while(count > 0) {
1327                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1328                 size_t write_bytes = min(count, PAGE_CACHE_SIZE - offset);
1329                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
1330                                         PAGE_CACHE_SHIFT;
1331                 ret = prepare_pages(NULL, root, file, pages, num_pages,
1332                                     pos, write_bytes);
1333                 BUG_ON(ret);
1334                 ret = btrfs_copy_from_user(pos, num_pages,
1335                                            write_bytes, pages, buf);
1336                 BUG_ON(ret);
1337
1338                 ret = dirty_and_release_pages(NULL, root, file, pages,
1339                                               num_pages, pos, write_bytes);
1340                 BUG_ON(ret);
1341                 btrfs_drop_pages(pages, num_pages);
1342
1343                 buf += write_bytes;
1344                 count -= write_bytes;
1345                 pos += write_bytes;
1346                 num_written += write_bytes;
1347
1348                 balance_dirty_pages_ratelimited(inode->i_mapping);
1349                 cond_resched();
1350         }
1351         mutex_unlock(&inode->i_mutex);
1352 out:
1353         *ppos = pos;
1354         current->backing_dev_info = NULL;
1355         return num_written ? num_written : err;
1356 }
1357
1358 #if 0
1359 static ssize_t inline_one_page(struct btrfs_root *root, struct inode *inode,
1360                            struct page *page, loff_t pos,
1361                            size_t offset, size_t write_bytes)
1362 {
1363         struct btrfs_path *path;
1364         struct btrfs_trans_handle *trans;
1365         struct btrfs_key key;
1366         struct btrfs_leaf *leaf;
1367         struct btrfs_key found_key;
1368         int ret;
1369         size_t copy_size = 0;
1370         char *dst = NULL;
1371         int err = 0;
1372         size_t num_written = 0;
1373
1374         path = btrfs_alloc_path();
1375         BUG_ON(!path);
1376         mutex_lock(&root->fs_info->fs_mutex);
1377         trans = btrfs_start_transaction(root, 1);
1378         key.objectid = inode->i_ino;
1379         key.flags = 0;
1380         btrfs_set_key_type(&key, BTRFS_INLINE_DATA_KEY);
1381
1382 again:
1383         key.offset = pos;
1384         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1385         if (ret < 0) {
1386                 err = ret;
1387                 goto out;
1388         }
1389         if (ret == 0) {
1390                 leaf = btrfs_buffer_leaf(path->nodes[0]);
1391                 btrfs_disk_key_to_cpu(&found_key,
1392                                       &leaf->items[path->slots[0]].key);
1393                 copy_size = btrfs_item_size(leaf->items + path->slots[0]);
1394                 dst = btrfs_item_ptr(leaf, path->slots[0], char);
1395                 copy_size = min(write_bytes, copy_size);
1396                 goto copyit;
1397         } else {
1398                 int slot = path->slots[0];
1399                 if (slot > 0) {
1400                         slot--;
1401                 }
1402                 // FIXME find max key
1403                 leaf = btrfs_buffer_leaf(path->nodes[0]);
1404                 btrfs_disk_key_to_cpu(&found_key,
1405                                       &leaf->items[slot].key);
1406                 if (found_key.objectid != inode->i_ino)
1407                         goto insert;
1408                 if (btrfs_key_type(&found_key) != BTRFS_INLINE_DATA_KEY)
1409                         goto insert;
1410                 copy_size = btrfs_item_size(leaf->items + slot);
1411                 if (found_key.offset + copy_size <= pos)
1412                         goto insert;
1413                 dst = btrfs_item_ptr(leaf, path->slots[0], char);
1414                 dst += pos - found_key.offset;
1415                 copy_size = copy_size - (pos - found_key.offset);
1416                 BUG_ON(copy_size < 0);
1417                 copy_size = min(write_bytes, copy_size);
1418                 WARN_ON(copy_size == 0);
1419                 goto copyit;
1420         }
1421 insert:
1422         btrfs_release_path(root, path);
1423         copy_size = min(write_bytes,
1424                         (size_t)BTRFS_LEAF_DATA_SIZE(root) -
1425                         sizeof(struct btrfs_item) * 4);
1426         ret = btrfs_insert_empty_item(trans, root, path, &key, copy_size);
1427         BUG_ON(ret);
1428         dst = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
1429                              path->slots[0], char);
1430 copyit:
1431         WARN_ON(copy_size == 0);
1432         WARN_ON(dst + copy_size >
1433                 btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
1434                                                  path->slots[0], char) +
1435                 btrfs_item_size(btrfs_buffer_leaf(path->nodes[0])->items +
1436                                                   path->slots[0]));
1437         btrfs_memcpy(root, path->nodes[0]->b_data, dst,
1438                      page_address(page) + offset, copy_size);
1439         mark_buffer_dirty(path->nodes[0]);
1440         btrfs_release_path(root, path);
1441         pos += copy_size;
1442         offset += copy_size;
1443         num_written += copy_size;
1444         write_bytes -= copy_size;
1445         if (write_bytes)
1446                 goto again;
1447 out:
1448         btrfs_free_path(path);
1449         ret = btrfs_end_transaction(trans, root);
1450         BUG_ON(ret);
1451         mutex_unlock(&root->fs_info->fs_mutex);
1452         return num_written ? num_written : err;
1453 }
1454
1455 static ssize_t btrfs_file_inline_write(struct file *file,
1456                                        const char __user *buf,
1457                                        size_t count, loff_t *ppos)
1458 {
1459         loff_t pos;
1460         size_t num_written = 0;
1461         int err = 0;
1462         int ret = 0;
1463         struct inode *inode = file->f_path.dentry->d_inode;
1464         struct btrfs_root *root = BTRFS_I(inode)->root;
1465         unsigned long page_index;
1466
1467         if (file->f_flags & O_DIRECT)
1468                 return -EINVAL;
1469         pos = *ppos;
1470
1471         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1472         current->backing_dev_info = inode->i_mapping->backing_dev_info;
1473         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1474         if (err)
1475                 goto out;
1476         if (count == 0)
1477                 goto out;
1478         err = remove_suid(file->f_path.dentry);
1479         if (err)
1480                 goto out;
1481         file_update_time(file);
1482         mutex_lock(&inode->i_mutex);
1483         while(count > 0) {
1484                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1485                 size_t write_bytes = min(count, PAGE_CACHE_SIZE - offset);
1486                 struct page *page;
1487
1488                 page_index = pos >> PAGE_CACHE_SHIFT;
1489                 page = grab_cache_page(inode->i_mapping, page_index);
1490                 if (!PageUptodate(page)) {
1491                         ret = mpage_readpage(page, btrfs_get_block);
1492                         BUG_ON(ret);
1493                         lock_page(page);
1494                 }
1495                 ret = btrfs_copy_from_user(pos, 1,
1496                                            write_bytes, &page, buf);
1497                 BUG_ON(ret);
1498                 write_bytes = inline_one_page(root, inode, page, pos,
1499                                       offset, write_bytes);
1500                 SetPageUptodate(page);
1501                 if (write_bytes > 0 && pos + write_bytes > inode->i_size) {
1502                         i_size_write(inode, pos + write_bytes);
1503                         mark_inode_dirty(inode);
1504                 }
1505                 page_cache_release(page);
1506                 unlock_page(page);
1507                 if (write_bytes < 0)
1508                         goto out_unlock;
1509                 buf += write_bytes;
1510                 count -= write_bytes;
1511                 pos += write_bytes;
1512                 num_written += write_bytes;
1513
1514                 balance_dirty_pages_ratelimited(inode->i_mapping);
1515                 cond_resched();
1516         }
1517 out_unlock:
1518         mutex_unlock(&inode->i_mutex);
1519 out:
1520         *ppos = pos;
1521         current->backing_dev_info = NULL;
1522         return num_written ? num_written : err;
1523 }
1524 #endif
1525
1526 static int btrfs_read_actor(read_descriptor_t *desc, struct page *page,
1527                         unsigned long offset, unsigned long size)
1528 {
1529         char *kaddr;
1530         unsigned long left, count = desc->count;
1531         struct inode *inode = page->mapping->host;
1532
1533         if (size > count)
1534                 size = count;
1535
1536         if (!PageChecked(page)) {
1537                 /* FIXME, do it per block */
1538                 struct btrfs_root *root = BTRFS_I(inode)->root;
1539                 int ret = btrfs_csum_verify_file_block(root,
1540                                           page->mapping->host->i_ino,
1541                                           page->index << PAGE_CACHE_SHIFT,
1542                                           kmap(page), PAGE_CACHE_SIZE);
1543                 if (ret) {
1544                         printk("failed to verify ino %lu page %lu\n",
1545                                page->mapping->host->i_ino,
1546                                page->index);
1547                         memset(page_address(page), 0, PAGE_CACHE_SIZE);
1548                 }
1549                 SetPageChecked(page);
1550                 kunmap(page);
1551         }
1552         /*
1553          * Faults on the destination of a read are common, so do it before
1554          * taking the kmap.
1555          */
1556         if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1557                 kaddr = kmap_atomic(page, KM_USER0);
1558                 left = __copy_to_user_inatomic(desc->arg.buf,
1559                                                 kaddr + offset, size);
1560                 kunmap_atomic(kaddr, KM_USER0);
1561                 if (left == 0)
1562                         goto success;
1563         }
1564
1565         /* Do it the slow way */
1566         kaddr = kmap(page);
1567         left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1568         kunmap(page);
1569
1570         if (left) {
1571                 size -= left;
1572                 desc->error = -EFAULT;
1573         }
1574 success:
1575         desc->count = count - size;
1576         desc->written += size;
1577         desc->arg.buf += size;
1578         return size;
1579 }
1580
1581 /**
1582  * btrfs_file_aio_read - filesystem read routine
1583  * @iocb:       kernel I/O control block
1584  * @iov:        io vector request
1585  * @nr_segs:    number of segments in the iovec
1586  * @pos:        current file position
1587  */
1588 static ssize_t btrfs_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1589                                    unsigned long nr_segs, loff_t pos)
1590 {
1591         struct file *filp = iocb->ki_filp;
1592         ssize_t retval;
1593         unsigned long seg;
1594         size_t count;
1595         loff_t *ppos = &iocb->ki_pos;
1596
1597         count = 0;
1598         for (seg = 0; seg < nr_segs; seg++) {
1599                 const struct iovec *iv = &iov[seg];
1600
1601                 /*
1602                  * If any segment has a negative length, or the cumulative
1603                  * length ever wraps negative then return -EINVAL.
1604                  */
1605                 count += iv->iov_len;
1606                 if (unlikely((ssize_t)(count|iv->iov_len) < 0))
1607                         return -EINVAL;
1608                 if (access_ok(VERIFY_WRITE, iv->iov_base, iv->iov_len))
1609                         continue;
1610                 if (seg == 0)
1611                         return -EFAULT;
1612                 nr_segs = seg;
1613                 count -= iv->iov_len;   /* This segment is no good */
1614                 break;
1615         }
1616         retval = 0;
1617         if (count) {
1618                 for (seg = 0; seg < nr_segs; seg++) {
1619                         read_descriptor_t desc;
1620
1621                         desc.written = 0;
1622                         desc.arg.buf = iov[seg].iov_base;
1623                         desc.count = iov[seg].iov_len;
1624                         if (desc.count == 0)
1625                                 continue;
1626                         desc.error = 0;
1627                         do_generic_file_read(filp, ppos, &desc,
1628                                              btrfs_read_actor);
1629                         retval += desc.written;
1630                         if (desc.error) {
1631                                 retval = retval ?: desc.error;
1632                                 break;
1633                         }
1634                 }
1635         }
1636         return retval;
1637 }
1638
1639 static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
1640 {
1641         struct btrfs_trans_handle *trans;
1642         struct btrfs_key key;
1643         struct btrfs_root_item new_root_item;
1644         int ret;
1645         u64 objectid;
1646
1647         mutex_lock(&root->fs_info->fs_mutex);
1648         trans = btrfs_start_transaction(root, 1);
1649         BUG_ON(!trans);
1650
1651         ret = btrfs_update_inode(trans, root, root->inode);
1652         BUG_ON(ret);
1653
1654         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
1655                                        0, &objectid);
1656         BUG_ON(ret);
1657
1658         memset(&new_root_item, 0, sizeof(new_root_item));
1659         memcpy(&new_root_item, &root->root_item,
1660                sizeof(new_root_item));
1661
1662         key.objectid = objectid;
1663         key.offset = 1;
1664         key.flags = 0;
1665         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
1666         btrfs_set_root_blocknr(&new_root_item, root->node->b_blocknr);
1667
1668         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
1669                                 &new_root_item);
1670         BUG_ON(ret);
1671
1672 printk("adding snapshot name %.*s root %Lu %Lu %u\n", namelen, name, key.objectid, key.offset, key.flags);
1673
1674         /*
1675          * insert the directory item
1676          */
1677         key.offset = (u64)-1;
1678         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
1679                                     name, namelen,
1680                                     root->fs_info->sb->s_root->d_inode->i_ino,
1681                                     &key, 0);
1682
1683         BUG_ON(ret);
1684
1685         ret = btrfs_inc_root_ref(trans, root);
1686         BUG_ON(ret);
1687
1688         ret = btrfs_commit_transaction(trans, root);
1689         BUG_ON(ret);
1690         mutex_unlock(&root->fs_info->fs_mutex);
1691         return 0;
1692 }
1693
1694 static int btrfs_ioctl(struct inode *inode, struct file *filp, unsigned int
1695                        cmd, unsigned long arg)
1696 {
1697         struct btrfs_root *root = BTRFS_I(inode)->root;
1698         struct btrfs_ioctl_vol_args vol_args;
1699         int ret;
1700         int namelen;
1701
1702         if (!root->ref_cows)
1703                 return -EINVAL;
1704         switch (cmd) {
1705         case BTRFS_IOC_SNAP_CREATE:
1706                 if (copy_from_user(&vol_args,
1707                                    (struct btrfs_ioctl_vol_args __user *)arg,
1708                                    sizeof(vol_args)))
1709                         return -EFAULT;
1710                 namelen = strlen(vol_args.name);
1711                 if (namelen > BTRFS_VOL_NAME_MAX)
1712                         return -EINVAL;
1713                 ret = create_snapshot(root, vol_args.name, namelen);
1714                 WARN_ON(ret);
1715                 break;
1716         default:
1717                 return -ENOTTY;
1718         }
1719         return 0;
1720 }
1721
1722 static struct kmem_cache *btrfs_inode_cachep;
1723 struct kmem_cache *btrfs_trans_handle_cachep;
1724 struct kmem_cache *btrfs_transaction_cachep;
1725 struct kmem_cache *btrfs_bit_radix_cachep;
1726 struct kmem_cache *btrfs_path_cachep;
1727
1728 /*
1729  * Called inside transaction, so use GFP_NOFS
1730  */
1731 static struct inode *btrfs_alloc_inode(struct super_block *sb)
1732 {
1733         struct btrfs_inode *ei;
1734
1735         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
1736         if (!ei)
1737                 return NULL;
1738         return &ei->vfs_inode;
1739 }
1740
1741 static void btrfs_destroy_inode(struct inode *inode)
1742 {
1743         WARN_ON(!list_empty(&inode->i_dentry));
1744         WARN_ON(inode->i_data.nrpages);
1745
1746         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
1747 }
1748
1749 static void init_once(void * foo, struct kmem_cache * cachep,
1750                       unsigned long flags)
1751 {
1752         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
1753
1754         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
1755             SLAB_CTOR_CONSTRUCTOR) {
1756                 inode_init_once(&ei->vfs_inode);
1757         }
1758 }
1759
1760 static int init_inodecache(void)
1761 {
1762         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
1763                                              sizeof(struct btrfs_inode),
1764                                              0, (SLAB_RECLAIM_ACCOUNT|
1765                                                 SLAB_MEM_SPREAD),
1766                                              init_once, NULL);
1767         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
1768                                              sizeof(struct btrfs_trans_handle),
1769                                              0, (SLAB_RECLAIM_ACCOUNT|
1770                                                 SLAB_MEM_SPREAD),
1771                                              NULL, NULL);
1772         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
1773                                              sizeof(struct btrfs_transaction),
1774                                              0, (SLAB_RECLAIM_ACCOUNT|
1775                                                 SLAB_MEM_SPREAD),
1776                                              NULL, NULL);
1777         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
1778                                              sizeof(struct btrfs_transaction),
1779                                              0, (SLAB_RECLAIM_ACCOUNT|
1780                                                 SLAB_MEM_SPREAD),
1781                                              NULL, NULL);
1782         btrfs_bit_radix_cachep = kmem_cache_create("btrfs_radix",
1783                                              256,
1784                                              0, (SLAB_RECLAIM_ACCOUNT|
1785                                                 SLAB_MEM_SPREAD |
1786                                                 SLAB_DESTROY_BY_RCU),
1787                                              NULL, NULL);
1788         if (btrfs_inode_cachep == NULL || btrfs_trans_handle_cachep == NULL ||
1789             btrfs_transaction_cachep == NULL || btrfs_bit_radix_cachep == NULL)
1790                 return -ENOMEM;
1791         return 0;
1792 }
1793
1794 static void destroy_inodecache(void)
1795 {
1796         kmem_cache_destroy(btrfs_inode_cachep);
1797         kmem_cache_destroy(btrfs_trans_handle_cachep);
1798         kmem_cache_destroy(btrfs_transaction_cachep);
1799         kmem_cache_destroy(btrfs_bit_radix_cachep);
1800         kmem_cache_destroy(btrfs_path_cachep);
1801 }
1802
1803 static int btrfs_get_sb(struct file_system_type *fs_type,
1804         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
1805 {
1806         return get_sb_bdev(fs_type, flags, dev_name, data,
1807                            btrfs_fill_super, mnt);
1808 }
1809
1810 static struct file_system_type btrfs_fs_type = {
1811         .owner          = THIS_MODULE,
1812         .name           = "btrfs",
1813         .get_sb         = btrfs_get_sb,
1814         .kill_sb        = kill_block_super,
1815         .fs_flags       = FS_REQUIRES_DEV,
1816 };
1817
1818 static struct super_operations btrfs_super_ops = {
1819         .statfs         = simple_statfs,
1820         .delete_inode   = btrfs_delete_inode,
1821         .put_super      = btrfs_put_super,
1822         .read_inode     = btrfs_read_locked_inode,
1823         .write_super    = btrfs_write_super,
1824         .sync_fs        = btrfs_sync_fs,
1825         .write_inode    = btrfs_write_inode,
1826         .alloc_inode    = btrfs_alloc_inode,
1827         .destroy_inode  = btrfs_destroy_inode,
1828 };
1829
1830 static struct inode_operations btrfs_dir_inode_operations = {
1831         .lookup         = btrfs_lookup,
1832         .create         = btrfs_create,
1833         .unlink         = btrfs_unlink,
1834         .mkdir          = btrfs_mkdir,
1835         .rmdir          = btrfs_rmdir,
1836 };
1837
1838 static struct inode_operations btrfs_dir_ro_inode_operations = {
1839         .lookup         = btrfs_lookup,
1840 };
1841
1842 static struct file_operations btrfs_dir_file_operations = {
1843         .llseek         = generic_file_llseek,
1844         .read           = generic_read_dir,
1845         .readdir        = btrfs_readdir,
1846         .ioctl          = btrfs_ioctl,
1847 };
1848
1849 static struct address_space_operations btrfs_aops = {
1850         .readpage       = btrfs_readpage,
1851         .readpages      = btrfs_readpages,
1852         .writepage      = btrfs_writepage,
1853         .sync_page      = block_sync_page,
1854         .prepare_write  = btrfs_prepare_write,
1855         .commit_write   = btrfs_commit_write,
1856 };
1857
1858 static struct inode_operations btrfs_file_inode_operations = {
1859         .truncate       = btrfs_truncate,
1860 };
1861
1862 static struct file_operations btrfs_file_operations = {
1863         .llseek         = generic_file_llseek,
1864         .read           = do_sync_read,
1865         .aio_read       = btrfs_file_aio_read,
1866         .write          = btrfs_file_write,
1867         .mmap           = generic_file_mmap,
1868         .open           = generic_file_open,
1869         .ioctl          = btrfs_ioctl,
1870 };
1871
1872 static int __init init_btrfs_fs(void)
1873 {
1874         int err;
1875         printk("btrfs loaded!\n");
1876         err = init_inodecache();
1877         if (err)
1878                 return err;
1879         kset_set_kset_s(&btrfs_subsys, fs_subsys);
1880         err = subsystem_register(&btrfs_subsys);
1881         if (err)
1882                 goto out;
1883         return register_filesystem(&btrfs_fs_type);
1884 out:
1885         destroy_inodecache();
1886         return err;
1887 }
1888
1889 static void __exit exit_btrfs_fs(void)
1890 {
1891         destroy_inodecache();
1892         unregister_filesystem(&btrfs_fs_type);
1893         subsystem_unregister(&btrfs_subsys);
1894         printk("btrfs unloaded\n");
1895 }
1896
1897 module_init(init_btrfs_fs)
1898 module_exit(exit_btrfs_fs)
1899
1900 MODULE_LICENSE("GPL");