Pull percpu-dtc into release branch
[linux-2.6] / fs / libfs.c
1 /*
2  *      fs/libfs.c
3  *      Library for filesystems writers.
4  */
5
6 #include <linux/module.h>
7 #include <linux/pagemap.h>
8 #include <linux/mount.h>
9 #include <linux/vfs.h>
10 #include <linux/mutex.h>
11
12 #include <asm/uaccess.h>
13
14 int simple_getattr(struct vfsmount *mnt, struct dentry *dentry,
15                    struct kstat *stat)
16 {
17         struct inode *inode = dentry->d_inode;
18         generic_fillattr(inode, stat);
19         stat->blocks = inode->i_mapping->nrpages << (PAGE_CACHE_SHIFT - 9);
20         return 0;
21 }
22
23 int simple_statfs(struct dentry *dentry, struct kstatfs *buf)
24 {
25         buf->f_type = dentry->d_sb->s_magic;
26         buf->f_bsize = PAGE_CACHE_SIZE;
27         buf->f_namelen = NAME_MAX;
28         return 0;
29 }
30
31 /*
32  * Retaining negative dentries for an in-memory filesystem just wastes
33  * memory and lookup time: arrange for them to be deleted immediately.
34  */
35 static int simple_delete_dentry(struct dentry *dentry)
36 {
37         return 1;
38 }
39
40 /*
41  * Lookup the data. This is trivial - if the dentry didn't already
42  * exist, we know it is negative.  Set d_op to delete negative dentries.
43  */
44 struct dentry *simple_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
45 {
46         static struct dentry_operations simple_dentry_operations = {
47                 .d_delete = simple_delete_dentry,
48         };
49
50         if (dentry->d_name.len > NAME_MAX)
51                 return ERR_PTR(-ENAMETOOLONG);
52         dentry->d_op = &simple_dentry_operations;
53         d_add(dentry, NULL);
54         return NULL;
55 }
56
57 int simple_sync_file(struct file * file, struct dentry *dentry, int datasync)
58 {
59         return 0;
60 }
61  
62 int dcache_dir_open(struct inode *inode, struct file *file)
63 {
64         static struct qstr cursor_name = {.len = 1, .name = "."};
65
66         file->private_data = d_alloc(file->f_path.dentry, &cursor_name);
67
68         return file->private_data ? 0 : -ENOMEM;
69 }
70
71 int dcache_dir_close(struct inode *inode, struct file *file)
72 {
73         dput(file->private_data);
74         return 0;
75 }
76
77 loff_t dcache_dir_lseek(struct file *file, loff_t offset, int origin)
78 {
79         mutex_lock(&file->f_path.dentry->d_inode->i_mutex);
80         switch (origin) {
81                 case 1:
82                         offset += file->f_pos;
83                 case 0:
84                         if (offset >= 0)
85                                 break;
86                 default:
87                         mutex_unlock(&file->f_path.dentry->d_inode->i_mutex);
88                         return -EINVAL;
89         }
90         if (offset != file->f_pos) {
91                 file->f_pos = offset;
92                 if (file->f_pos >= 2) {
93                         struct list_head *p;
94                         struct dentry *cursor = file->private_data;
95                         loff_t n = file->f_pos - 2;
96
97                         spin_lock(&dcache_lock);
98                         list_del(&cursor->d_u.d_child);
99                         p = file->f_path.dentry->d_subdirs.next;
100                         while (n && p != &file->f_path.dentry->d_subdirs) {
101                                 struct dentry *next;
102                                 next = list_entry(p, struct dentry, d_u.d_child);
103                                 if (!d_unhashed(next) && next->d_inode)
104                                         n--;
105                                 p = p->next;
106                         }
107                         list_add_tail(&cursor->d_u.d_child, p);
108                         spin_unlock(&dcache_lock);
109                 }
110         }
111         mutex_unlock(&file->f_path.dentry->d_inode->i_mutex);
112         return offset;
113 }
114
115 /* Relationship between i_mode and the DT_xxx types */
116 static inline unsigned char dt_type(struct inode *inode)
117 {
118         return (inode->i_mode >> 12) & 15;
119 }
120
121 /*
122  * Directory is locked and all positive dentries in it are safe, since
123  * for ramfs-type trees they can't go away without unlink() or rmdir(),
124  * both impossible due to the lock on directory.
125  */
126
127 int dcache_readdir(struct file * filp, void * dirent, filldir_t filldir)
128 {
129         struct dentry *dentry = filp->f_path.dentry;
130         struct dentry *cursor = filp->private_data;
131         struct list_head *p, *q = &cursor->d_u.d_child;
132         ino_t ino;
133         int i = filp->f_pos;
134
135         switch (i) {
136                 case 0:
137                         ino = dentry->d_inode->i_ino;
138                         if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
139                                 break;
140                         filp->f_pos++;
141                         i++;
142                         /* fallthrough */
143                 case 1:
144                         ino = parent_ino(dentry);
145                         if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
146                                 break;
147                         filp->f_pos++;
148                         i++;
149                         /* fallthrough */
150                 default:
151                         spin_lock(&dcache_lock);
152                         if (filp->f_pos == 2)
153                                 list_move(q, &dentry->d_subdirs);
154
155                         for (p=q->next; p != &dentry->d_subdirs; p=p->next) {
156                                 struct dentry *next;
157                                 next = list_entry(p, struct dentry, d_u.d_child);
158                                 if (d_unhashed(next) || !next->d_inode)
159                                         continue;
160
161                                 spin_unlock(&dcache_lock);
162                                 if (filldir(dirent, next->d_name.name, next->d_name.len, filp->f_pos, next->d_inode->i_ino, dt_type(next->d_inode)) < 0)
163                                         return 0;
164                                 spin_lock(&dcache_lock);
165                                 /* next is still alive */
166                                 list_move(q, p);
167                                 p = q;
168                                 filp->f_pos++;
169                         }
170                         spin_unlock(&dcache_lock);
171         }
172         return 0;
173 }
174
175 ssize_t generic_read_dir(struct file *filp, char __user *buf, size_t siz, loff_t *ppos)
176 {
177         return -EISDIR;
178 }
179
180 const struct file_operations simple_dir_operations = {
181         .open           = dcache_dir_open,
182         .release        = dcache_dir_close,
183         .llseek         = dcache_dir_lseek,
184         .read           = generic_read_dir,
185         .readdir        = dcache_readdir,
186         .fsync          = simple_sync_file,
187 };
188
189 const struct inode_operations simple_dir_inode_operations = {
190         .lookup         = simple_lookup,
191 };
192
193 static const struct super_operations simple_super_operations = {
194         .statfs         = simple_statfs,
195 };
196
197 /*
198  * Common helper for pseudo-filesystems (sockfs, pipefs, bdev - stuff that
199  * will never be mountable)
200  */
201 int get_sb_pseudo(struct file_system_type *fs_type, char *name,
202         const struct super_operations *ops, unsigned long magic,
203         struct vfsmount *mnt)
204 {
205         struct super_block *s = sget(fs_type, NULL, set_anon_super, NULL);
206         struct dentry *dentry;
207         struct inode *root;
208         struct qstr d_name = {.name = name, .len = strlen(name)};
209
210         if (IS_ERR(s))
211                 return PTR_ERR(s);
212
213         s->s_flags = MS_NOUSER;
214         s->s_maxbytes = ~0ULL;
215         s->s_blocksize = 1024;
216         s->s_blocksize_bits = 10;
217         s->s_magic = magic;
218         s->s_op = ops ? ops : &simple_super_operations;
219         s->s_time_gran = 1;
220         root = new_inode(s);
221         if (!root)
222                 goto Enomem;
223         root->i_mode = S_IFDIR | S_IRUSR | S_IWUSR;
224         root->i_uid = root->i_gid = 0;
225         root->i_atime = root->i_mtime = root->i_ctime = CURRENT_TIME;
226         dentry = d_alloc(NULL, &d_name);
227         if (!dentry) {
228                 iput(root);
229                 goto Enomem;
230         }
231         dentry->d_sb = s;
232         dentry->d_parent = dentry;
233         d_instantiate(dentry, root);
234         s->s_root = dentry;
235         s->s_flags |= MS_ACTIVE;
236         return simple_set_mnt(mnt, s);
237
238 Enomem:
239         up_write(&s->s_umount);
240         deactivate_super(s);
241         return -ENOMEM;
242 }
243
244 int simple_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
245 {
246         struct inode *inode = old_dentry->d_inode;
247
248         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
249         inc_nlink(inode);
250         atomic_inc(&inode->i_count);
251         dget(dentry);
252         d_instantiate(dentry, inode);
253         return 0;
254 }
255
256 static inline int simple_positive(struct dentry *dentry)
257 {
258         return dentry->d_inode && !d_unhashed(dentry);
259 }
260
261 int simple_empty(struct dentry *dentry)
262 {
263         struct dentry *child;
264         int ret = 0;
265
266         spin_lock(&dcache_lock);
267         list_for_each_entry(child, &dentry->d_subdirs, d_u.d_child)
268                 if (simple_positive(child))
269                         goto out;
270         ret = 1;
271 out:
272         spin_unlock(&dcache_lock);
273         return ret;
274 }
275
276 int simple_unlink(struct inode *dir, struct dentry *dentry)
277 {
278         struct inode *inode = dentry->d_inode;
279
280         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
281         drop_nlink(inode);
282         dput(dentry);
283         return 0;
284 }
285
286 int simple_rmdir(struct inode *dir, struct dentry *dentry)
287 {
288         if (!simple_empty(dentry))
289                 return -ENOTEMPTY;
290
291         drop_nlink(dentry->d_inode);
292         simple_unlink(dir, dentry);
293         drop_nlink(dir);
294         return 0;
295 }
296
297 int simple_rename(struct inode *old_dir, struct dentry *old_dentry,
298                 struct inode *new_dir, struct dentry *new_dentry)
299 {
300         struct inode *inode = old_dentry->d_inode;
301         int they_are_dirs = S_ISDIR(old_dentry->d_inode->i_mode);
302
303         if (!simple_empty(new_dentry))
304                 return -ENOTEMPTY;
305
306         if (new_dentry->d_inode) {
307                 simple_unlink(new_dir, new_dentry);
308                 if (they_are_dirs)
309                         drop_nlink(old_dir);
310         } else if (they_are_dirs) {
311                 drop_nlink(old_dir);
312                 inc_nlink(new_dir);
313         }
314
315         old_dir->i_ctime = old_dir->i_mtime = new_dir->i_ctime =
316                 new_dir->i_mtime = inode->i_ctime = CURRENT_TIME;
317
318         return 0;
319 }
320
321 int simple_readpage(struct file *file, struct page *page)
322 {
323         clear_highpage(page);
324         flush_dcache_page(page);
325         SetPageUptodate(page);
326         unlock_page(page);
327         return 0;
328 }
329
330 int simple_prepare_write(struct file *file, struct page *page,
331                         unsigned from, unsigned to)
332 {
333         if (!PageUptodate(page)) {
334                 if (to - from != PAGE_CACHE_SIZE) {
335                         void *kaddr = kmap_atomic(page, KM_USER0);
336                         memset(kaddr, 0, from);
337                         memset(kaddr + to, 0, PAGE_CACHE_SIZE - to);
338                         flush_dcache_page(page);
339                         kunmap_atomic(kaddr, KM_USER0);
340                 }
341         }
342         return 0;
343 }
344
345 int simple_commit_write(struct file *file, struct page *page,
346                         unsigned from, unsigned to)
347 {
348         struct inode *inode = page->mapping->host;
349         loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
350
351         if (!PageUptodate(page))
352                 SetPageUptodate(page);
353         /*
354          * No need to use i_size_read() here, the i_size
355          * cannot change under us because we hold the i_mutex.
356          */
357         if (pos > inode->i_size)
358                 i_size_write(inode, pos);
359         set_page_dirty(page);
360         return 0;
361 }
362
363 int simple_fill_super(struct super_block *s, int magic, struct tree_descr *files)
364 {
365         struct inode *inode;
366         struct dentry *root;
367         struct dentry *dentry;
368         int i;
369
370         s->s_blocksize = PAGE_CACHE_SIZE;
371         s->s_blocksize_bits = PAGE_CACHE_SHIFT;
372         s->s_magic = magic;
373         s->s_op = &simple_super_operations;
374         s->s_time_gran = 1;
375
376         inode = new_inode(s);
377         if (!inode)
378                 return -ENOMEM;
379         inode->i_mode = S_IFDIR | 0755;
380         inode->i_uid = inode->i_gid = 0;
381         inode->i_blocks = 0;
382         inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
383         inode->i_op = &simple_dir_inode_operations;
384         inode->i_fop = &simple_dir_operations;
385         inode->i_nlink = 2;
386         root = d_alloc_root(inode);
387         if (!root) {
388                 iput(inode);
389                 return -ENOMEM;
390         }
391         for (i = 0; !files->name || files->name[0]; i++, files++) {
392                 if (!files->name)
393                         continue;
394                 dentry = d_alloc_name(root, files->name);
395                 if (!dentry)
396                         goto out;
397                 inode = new_inode(s);
398                 if (!inode)
399                         goto out;
400                 inode->i_mode = S_IFREG | files->mode;
401                 inode->i_uid = inode->i_gid = 0;
402                 inode->i_blocks = 0;
403                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
404                 inode->i_fop = files->ops;
405                 inode->i_ino = i;
406                 d_add(dentry, inode);
407         }
408         s->s_root = root;
409         return 0;
410 out:
411         d_genocide(root);
412         dput(root);
413         return -ENOMEM;
414 }
415
416 static DEFINE_SPINLOCK(pin_fs_lock);
417
418 int simple_pin_fs(struct file_system_type *type, struct vfsmount **mount, int *count)
419 {
420         struct vfsmount *mnt = NULL;
421         spin_lock(&pin_fs_lock);
422         if (unlikely(!*mount)) {
423                 spin_unlock(&pin_fs_lock);
424                 mnt = vfs_kern_mount(type, 0, type->name, NULL);
425                 if (IS_ERR(mnt))
426                         return PTR_ERR(mnt);
427                 spin_lock(&pin_fs_lock);
428                 if (!*mount)
429                         *mount = mnt;
430         }
431         mntget(*mount);
432         ++*count;
433         spin_unlock(&pin_fs_lock);
434         mntput(mnt);
435         return 0;
436 }
437
438 void simple_release_fs(struct vfsmount **mount, int *count)
439 {
440         struct vfsmount *mnt;
441         spin_lock(&pin_fs_lock);
442         mnt = *mount;
443         if (!--*count)
444                 *mount = NULL;
445         spin_unlock(&pin_fs_lock);
446         mntput(mnt);
447 }
448
449 ssize_t simple_read_from_buffer(void __user *to, size_t count, loff_t *ppos,
450                                 const void *from, size_t available)
451 {
452         loff_t pos = *ppos;
453         if (pos < 0)
454                 return -EINVAL;
455         if (pos >= available)
456                 return 0;
457         if (count > available - pos)
458                 count = available - pos;
459         if (copy_to_user(to, from + pos, count))
460                 return -EFAULT;
461         *ppos = pos + count;
462         return count;
463 }
464
465 /*
466  * Transaction based IO.
467  * The file expects a single write which triggers the transaction, and then
468  * possibly a read which collects the result - which is stored in a
469  * file-local buffer.
470  */
471 char *simple_transaction_get(struct file *file, const char __user *buf, size_t size)
472 {
473         struct simple_transaction_argresp *ar;
474         static DEFINE_SPINLOCK(simple_transaction_lock);
475
476         if (size > SIMPLE_TRANSACTION_LIMIT - 1)
477                 return ERR_PTR(-EFBIG);
478
479         ar = (struct simple_transaction_argresp *)get_zeroed_page(GFP_KERNEL);
480         if (!ar)
481                 return ERR_PTR(-ENOMEM);
482
483         spin_lock(&simple_transaction_lock);
484
485         /* only one write allowed per open */
486         if (file->private_data) {
487                 spin_unlock(&simple_transaction_lock);
488                 free_page((unsigned long)ar);
489                 return ERR_PTR(-EBUSY);
490         }
491
492         file->private_data = ar;
493
494         spin_unlock(&simple_transaction_lock);
495
496         if (copy_from_user(ar->data, buf, size))
497                 return ERR_PTR(-EFAULT);
498
499         return ar->data;
500 }
501
502 ssize_t simple_transaction_read(struct file *file, char __user *buf, size_t size, loff_t *pos)
503 {
504         struct simple_transaction_argresp *ar = file->private_data;
505
506         if (!ar)
507                 return 0;
508         return simple_read_from_buffer(buf, size, pos, ar->data, ar->size);
509 }
510
511 int simple_transaction_release(struct inode *inode, struct file *file)
512 {
513         free_page((unsigned long)file->private_data);
514         return 0;
515 }
516
517 /* Simple attribute files */
518
519 struct simple_attr {
520         u64 (*get)(void *);
521         void (*set)(void *, u64);
522         char get_buf[24];       /* enough to store a u64 and "\n\0" */
523         char set_buf[24];
524         void *data;
525         const char *fmt;        /* format for read operation */
526         struct mutex mutex;     /* protects access to these buffers */
527 };
528
529 /* simple_attr_open is called by an actual attribute open file operation
530  * to set the attribute specific access operations. */
531 int simple_attr_open(struct inode *inode, struct file *file,
532                      u64 (*get)(void *), void (*set)(void *, u64),
533                      const char *fmt)
534 {
535         struct simple_attr *attr;
536
537         attr = kmalloc(sizeof(*attr), GFP_KERNEL);
538         if (!attr)
539                 return -ENOMEM;
540
541         attr->get = get;
542         attr->set = set;
543         attr->data = inode->i_private;
544         attr->fmt = fmt;
545         mutex_init(&attr->mutex);
546
547         file->private_data = attr;
548
549         return nonseekable_open(inode, file);
550 }
551
552 int simple_attr_close(struct inode *inode, struct file *file)
553 {
554         kfree(file->private_data);
555         return 0;
556 }
557
558 /* read from the buffer that is filled with the get function */
559 ssize_t simple_attr_read(struct file *file, char __user *buf,
560                          size_t len, loff_t *ppos)
561 {
562         struct simple_attr *attr;
563         size_t size;
564         ssize_t ret;
565
566         attr = file->private_data;
567
568         if (!attr->get)
569                 return -EACCES;
570
571         mutex_lock(&attr->mutex);
572         if (*ppos) /* continued read */
573                 size = strlen(attr->get_buf);
574         else      /* first read */
575                 size = scnprintf(attr->get_buf, sizeof(attr->get_buf),
576                                  attr->fmt,
577                                  (unsigned long long)attr->get(attr->data));
578
579         ret = simple_read_from_buffer(buf, len, ppos, attr->get_buf, size);
580         mutex_unlock(&attr->mutex);
581         return ret;
582 }
583
584 /* interpret the buffer as a number to call the set function with */
585 ssize_t simple_attr_write(struct file *file, const char __user *buf,
586                           size_t len, loff_t *ppos)
587 {
588         struct simple_attr *attr;
589         u64 val;
590         size_t size;
591         ssize_t ret;
592
593         attr = file->private_data;
594
595         if (!attr->set)
596                 return -EACCES;
597
598         mutex_lock(&attr->mutex);
599         ret = -EFAULT;
600         size = min(sizeof(attr->set_buf) - 1, len);
601         if (copy_from_user(attr->set_buf, buf, size))
602                 goto out;
603
604         ret = len; /* claim we got the whole input */
605         attr->set_buf[size] = '\0';
606         val = simple_strtol(attr->set_buf, NULL, 0);
607         attr->set(attr->data, val);
608 out:
609         mutex_unlock(&attr->mutex);
610         return ret;
611 }
612
613 EXPORT_SYMBOL(dcache_dir_close);
614 EXPORT_SYMBOL(dcache_dir_lseek);
615 EXPORT_SYMBOL(dcache_dir_open);
616 EXPORT_SYMBOL(dcache_readdir);
617 EXPORT_SYMBOL(generic_read_dir);
618 EXPORT_SYMBOL(get_sb_pseudo);
619 EXPORT_SYMBOL(simple_commit_write);
620 EXPORT_SYMBOL(simple_dir_inode_operations);
621 EXPORT_SYMBOL(simple_dir_operations);
622 EXPORT_SYMBOL(simple_empty);
623 EXPORT_SYMBOL(d_alloc_name);
624 EXPORT_SYMBOL(simple_fill_super);
625 EXPORT_SYMBOL(simple_getattr);
626 EXPORT_SYMBOL(simple_link);
627 EXPORT_SYMBOL(simple_lookup);
628 EXPORT_SYMBOL(simple_pin_fs);
629 EXPORT_SYMBOL(simple_prepare_write);
630 EXPORT_SYMBOL(simple_readpage);
631 EXPORT_SYMBOL(simple_release_fs);
632 EXPORT_SYMBOL(simple_rename);
633 EXPORT_SYMBOL(simple_rmdir);
634 EXPORT_SYMBOL(simple_statfs);
635 EXPORT_SYMBOL(simple_sync_file);
636 EXPORT_SYMBOL(simple_unlink);
637 EXPORT_SYMBOL(simple_read_from_buffer);
638 EXPORT_SYMBOL(simple_transaction_get);
639 EXPORT_SYMBOL(simple_transaction_read);
640 EXPORT_SYMBOL(simple_transaction_release);
641 EXPORT_SYMBOL_GPL(simple_attr_open);
642 EXPORT_SYMBOL_GPL(simple_attr_close);
643 EXPORT_SYMBOL_GPL(simple_attr_read);
644 EXPORT_SYMBOL_GPL(simple_attr_write);