[PATCH] sem2mutex: fs/libfs.c
[linux-2.6.git] / 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 super_block *sb, struct kstatfs *buf)
24 {
25         buf->f_type = 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_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_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_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_dentry->d_subdirs.next;
100                         while (n && p != &file->f_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_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_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_del(q);
154                                 list_add(q, &dentry->d_subdirs);
155                         }
156                         for (p=q->next; p != &dentry->d_subdirs; p=p->next) {
157                                 struct dentry *next;
158                                 next = list_entry(p, struct dentry, d_u.d_child);
159                                 if (d_unhashed(next) || !next->d_inode)
160                                         continue;
161
162                                 spin_unlock(&dcache_lock);
163                                 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)
164                                         return 0;
165                                 spin_lock(&dcache_lock);
166                                 /* next is still alive */
167                                 list_del(q);
168                                 list_add(q, p);
169                                 p = q;
170                                 filp->f_pos++;
171                         }
172                         spin_unlock(&dcache_lock);
173         }
174         return 0;
175 }
176
177 ssize_t generic_read_dir(struct file *filp, char __user *buf, size_t siz, loff_t *ppos)
178 {
179         return -EISDIR;
180 }
181
182 struct file_operations simple_dir_operations = {
183         .open           = dcache_dir_open,
184         .release        = dcache_dir_close,
185         .llseek         = dcache_dir_lseek,
186         .read           = generic_read_dir,
187         .readdir        = dcache_readdir,
188         .fsync          = simple_sync_file,
189 };
190
191 struct inode_operations simple_dir_inode_operations = {
192         .lookup         = simple_lookup,
193 };
194
195 /*
196  * Common helper for pseudo-filesystems (sockfs, pipefs, bdev - stuff that
197  * will never be mountable)
198  */
199 struct super_block *
200 get_sb_pseudo(struct file_system_type *fs_type, char *name,
201         struct super_operations *ops, unsigned long magic)
202 {
203         struct super_block *s = sget(fs_type, NULL, set_anon_super, NULL);
204         static struct super_operations default_ops = {.statfs = simple_statfs};
205         struct dentry *dentry;
206         struct inode *root;
207         struct qstr d_name = {.name = name, .len = strlen(name)};
208
209         if (IS_ERR(s))
210                 return s;
211
212         s->s_flags = MS_NOUSER;
213         s->s_maxbytes = ~0ULL;
214         s->s_blocksize = 1024;
215         s->s_blocksize_bits = 10;
216         s->s_magic = magic;
217         s->s_op = ops ? ops : &default_ops;
218         s->s_time_gran = 1;
219         root = new_inode(s);
220         if (!root)
221                 goto Enomem;
222         root->i_mode = S_IFDIR | S_IRUSR | S_IWUSR;
223         root->i_uid = root->i_gid = 0;
224         root->i_atime = root->i_mtime = root->i_ctime = CURRENT_TIME;
225         dentry = d_alloc(NULL, &d_name);
226         if (!dentry) {
227                 iput(root);
228                 goto Enomem;
229         }
230         dentry->d_sb = s;
231         dentry->d_parent = dentry;
232         d_instantiate(dentry, root);
233         s->s_root = dentry;
234         s->s_flags |= MS_ACTIVE;
235         return s;
236
237 Enomem:
238         up_write(&s->s_umount);
239         deactivate_super(s);
240         return ERR_PTR(-ENOMEM);
241 }
242
243 int simple_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
244 {
245         struct inode *inode = old_dentry->d_inode;
246
247         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
248         inode->i_nlink++;
249         atomic_inc(&inode->i_count);
250         dget(dentry);
251         d_instantiate(dentry, inode);
252         return 0;
253 }
254
255 static inline int simple_positive(struct dentry *dentry)
256 {
257         return dentry->d_inode && !d_unhashed(dentry);
258 }
259
260 int simple_empty(struct dentry *dentry)
261 {
262         struct dentry *child;
263         int ret = 0;
264
265         spin_lock(&dcache_lock);
266         list_for_each_entry(child, &dentry->d_subdirs, d_u.d_child)
267                 if (simple_positive(child))
268                         goto out;
269         ret = 1;
270 out:
271         spin_unlock(&dcache_lock);
272         return ret;
273 }
274
275 int simple_unlink(struct inode *dir, struct dentry *dentry)
276 {
277         struct inode *inode = dentry->d_inode;
278
279         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
280         inode->i_nlink--;
281         dput(dentry);
282         return 0;
283 }
284
285 int simple_rmdir(struct inode *dir, struct dentry *dentry)
286 {
287         if (!simple_empty(dentry))
288                 return -ENOTEMPTY;
289
290         dentry->d_inode->i_nlink--;
291         simple_unlink(dir, dentry);
292         dir->i_nlink--;
293         return 0;
294 }
295
296 int simple_rename(struct inode *old_dir, struct dentry *old_dentry,
297                 struct inode *new_dir, struct dentry *new_dentry)
298 {
299         struct inode *inode = old_dentry->d_inode;
300         int they_are_dirs = S_ISDIR(old_dentry->d_inode->i_mode);
301
302         if (!simple_empty(new_dentry))
303                 return -ENOTEMPTY;
304
305         if (new_dentry->d_inode) {
306                 simple_unlink(new_dir, new_dentry);
307                 if (they_are_dirs)
308                         old_dir->i_nlink--;
309         } else if (they_are_dirs) {
310                 old_dir->i_nlink--;
311                 new_dir->i_nlink++;
312         }
313
314         old_dir->i_ctime = old_dir->i_mtime = new_dir->i_ctime =
315                 new_dir->i_mtime = inode->i_ctime = CURRENT_TIME;
316
317         return 0;
318 }
319
320 int simple_readpage(struct file *file, struct page *page)
321 {
322         void *kaddr;
323
324         if (PageUptodate(page))
325                 goto out;
326
327         kaddr = kmap_atomic(page, KM_USER0);
328         memset(kaddr, 0, PAGE_CACHE_SIZE);
329         kunmap_atomic(kaddr, KM_USER0);
330         flush_dcache_page(page);
331         SetPageUptodate(page);
332 out:
333         unlock_page(page);
334         return 0;
335 }
336
337 int simple_prepare_write(struct file *file, struct page *page,
338                         unsigned from, unsigned to)
339 {
340         if (!PageUptodate(page)) {
341                 if (to - from != PAGE_CACHE_SIZE) {
342                         void *kaddr = kmap_atomic(page, KM_USER0);
343                         memset(kaddr, 0, from);
344                         memset(kaddr + to, 0, PAGE_CACHE_SIZE - to);
345                         flush_dcache_page(page);
346                         kunmap_atomic(kaddr, KM_USER0);
347                 }
348                 SetPageUptodate(page);
349         }
350         return 0;
351 }
352
353 int simple_commit_write(struct file *file, struct page *page,
354                         unsigned offset, unsigned to)
355 {
356         struct inode *inode = page->mapping->host;
357         loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
358
359         /*
360          * No need to use i_size_read() here, the i_size
361          * cannot change under us because we hold the i_mutex.
362          */
363         if (pos > inode->i_size)
364                 i_size_write(inode, pos);
365         set_page_dirty(page);
366         return 0;
367 }
368
369 int simple_fill_super(struct super_block *s, int magic, struct tree_descr *files)
370 {
371         static struct super_operations s_ops = {.statfs = simple_statfs};
372         struct inode *inode;
373         struct dentry *root;
374         struct dentry *dentry;
375         int i;
376
377         s->s_blocksize = PAGE_CACHE_SIZE;
378         s->s_blocksize_bits = PAGE_CACHE_SHIFT;
379         s->s_magic = magic;
380         s->s_op = &s_ops;
381         s->s_time_gran = 1;
382
383         inode = new_inode(s);
384         if (!inode)
385                 return -ENOMEM;
386         inode->i_mode = S_IFDIR | 0755;
387         inode->i_uid = inode->i_gid = 0;
388         inode->i_blksize = PAGE_CACHE_SIZE;
389         inode->i_blocks = 0;
390         inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
391         inode->i_op = &simple_dir_inode_operations;
392         inode->i_fop = &simple_dir_operations;
393         inode->i_nlink = 2;
394         root = d_alloc_root(inode);
395         if (!root) {
396                 iput(inode);
397                 return -ENOMEM;
398         }
399         for (i = 0; !files->name || files->name[0]; i++, files++) {
400                 if (!files->name)
401                         continue;
402                 dentry = d_alloc_name(root, files->name);
403                 if (!dentry)
404                         goto out;
405                 inode = new_inode(s);
406                 if (!inode)
407                         goto out;
408                 inode->i_mode = S_IFREG | files->mode;
409                 inode->i_uid = inode->i_gid = 0;
410                 inode->i_blksize = PAGE_CACHE_SIZE;
411                 inode->i_blocks = 0;
412                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
413                 inode->i_fop = files->ops;
414                 inode->i_ino = i;
415                 d_add(dentry, inode);
416         }
417         s->s_root = root;
418         return 0;
419 out:
420         d_genocide(root);
421         dput(root);
422         return -ENOMEM;
423 }
424
425 static DEFINE_SPINLOCK(pin_fs_lock);
426
427 int simple_pin_fs(char *name, struct vfsmount **mount, int *count)
428 {
429         struct vfsmount *mnt = NULL;
430         spin_lock(&pin_fs_lock);
431         if (unlikely(!*mount)) {
432                 spin_unlock(&pin_fs_lock);
433                 mnt = do_kern_mount(name, 0, name, NULL);
434                 if (IS_ERR(mnt))
435                         return PTR_ERR(mnt);
436                 spin_lock(&pin_fs_lock);
437                 if (!*mount)
438                         *mount = mnt;
439         }
440         mntget(*mount);
441         ++*count;
442         spin_unlock(&pin_fs_lock);
443         mntput(mnt);
444         return 0;
445 }
446
447 void simple_release_fs(struct vfsmount **mount, int *count)
448 {
449         struct vfsmount *mnt;
450         spin_lock(&pin_fs_lock);
451         mnt = *mount;
452         if (!--*count)
453                 *mount = NULL;
454         spin_unlock(&pin_fs_lock);
455         mntput(mnt);
456 }
457
458 ssize_t simple_read_from_buffer(void __user *to, size_t count, loff_t *ppos,
459                                 const void *from, size_t available)
460 {
461         loff_t pos = *ppos;
462         if (pos < 0)
463                 return -EINVAL;
464         if (pos >= available)
465                 return 0;
466         if (count > available - pos)
467                 count = available - pos;
468         if (copy_to_user(to, from + pos, count))
469                 return -EFAULT;
470         *ppos = pos + count;
471         return count;
472 }
473
474 /*
475  * Transaction based IO.
476  * The file expects a single write which triggers the transaction, and then
477  * possibly a read which collects the result - which is stored in a
478  * file-local buffer.
479  */
480 char *simple_transaction_get(struct file *file, const char __user *buf, size_t size)
481 {
482         struct simple_transaction_argresp *ar;
483         static DEFINE_SPINLOCK(simple_transaction_lock);
484
485         if (size > SIMPLE_TRANSACTION_LIMIT - 1)
486                 return ERR_PTR(-EFBIG);
487
488         ar = (struct simple_transaction_argresp *)get_zeroed_page(GFP_KERNEL);
489         if (!ar)
490                 return ERR_PTR(-ENOMEM);
491
492         spin_lock(&simple_transaction_lock);
493
494         /* only one write allowed per open */
495         if (file->private_data) {
496                 spin_unlock(&simple_transaction_lock);
497                 free_page((unsigned long)ar);
498                 return ERR_PTR(-EBUSY);
499         }
500
501         file->private_data = ar;
502
503         spin_unlock(&simple_transaction_lock);
504
505         if (copy_from_user(ar->data, buf, size))
506                 return ERR_PTR(-EFAULT);
507
508         return ar->data;
509 }
510
511 ssize_t simple_transaction_read(struct file *file, char __user *buf, size_t size, loff_t *pos)
512 {
513         struct simple_transaction_argresp *ar = file->private_data;
514
515         if (!ar)
516                 return 0;
517         return simple_read_from_buffer(buf, size, pos, ar->data, ar->size);
518 }
519
520 int simple_transaction_release(struct inode *inode, struct file *file)
521 {
522         free_page((unsigned long)file->private_data);
523         return 0;
524 }
525
526 /* Simple attribute files */
527
528 struct simple_attr {
529         u64 (*get)(void *);
530         void (*set)(void *, u64);
531         char get_buf[24];       /* enough to store a u64 and "\n\0" */
532         char set_buf[24];
533         void *data;
534         const char *fmt;        /* format for read operation */
535         struct mutex mutex;     /* protects access to these buffers */
536 };
537
538 /* simple_attr_open is called by an actual attribute open file operation
539  * to set the attribute specific access operations. */
540 int simple_attr_open(struct inode *inode, struct file *file,
541                      u64 (*get)(void *), void (*set)(void *, u64),
542                      const char *fmt)
543 {
544         struct simple_attr *attr;
545
546         attr = kmalloc(sizeof(*attr), GFP_KERNEL);
547         if (!attr)
548                 return -ENOMEM;
549
550         attr->get = get;
551         attr->set = set;
552         attr->data = inode->u.generic_ip;
553         attr->fmt = fmt;
554         mutex_init(&attr->mutex);
555
556         file->private_data = attr;
557
558         return nonseekable_open(inode, file);
559 }
560
561 int simple_attr_close(struct inode *inode, struct file *file)
562 {
563         kfree(file->private_data);
564         return 0;
565 }
566
567 /* read from the buffer that is filled with the get function */
568 ssize_t simple_attr_read(struct file *file, char __user *buf,
569                          size_t len, loff_t *ppos)
570 {
571         struct simple_attr *attr;
572         size_t size;
573         ssize_t ret;
574
575         attr = file->private_data;
576
577         if (!attr->get)
578                 return -EACCES;
579
580         mutex_lock(&attr->mutex);
581         if (*ppos) /* continued read */
582                 size = strlen(attr->get_buf);
583         else      /* first read */
584                 size = scnprintf(attr->get_buf, sizeof(attr->get_buf),
585                                  attr->fmt,
586                                  (unsigned long long)attr->get(attr->data));
587
588         ret = simple_read_from_buffer(buf, len, ppos, attr->get_buf, size);
589         mutex_unlock(&attr->mutex);
590         return ret;
591 }
592
593 /* interpret the buffer as a number to call the set function with */
594 ssize_t simple_attr_write(struct file *file, const char __user *buf,
595                           size_t len, loff_t *ppos)
596 {
597         struct simple_attr *attr;
598         u64 val;
599         size_t size;
600         ssize_t ret;
601
602         attr = file->private_data;
603
604         if (!attr->set)
605                 return -EACCES;
606
607         mutex_lock(&attr->mutex);
608         ret = -EFAULT;
609         size = min(sizeof(attr->set_buf) - 1, len);
610         if (copy_from_user(attr->set_buf, buf, size))
611                 goto out;
612
613         ret = len; /* claim we got the whole input */
614         attr->set_buf[size] = '\0';
615         val = simple_strtol(attr->set_buf, NULL, 0);
616         attr->set(attr->data, val);
617 out:
618         mutex_unlock(&attr->mutex);
619         return ret;
620 }
621
622 EXPORT_SYMBOL(dcache_dir_close);
623 EXPORT_SYMBOL(dcache_dir_lseek);
624 EXPORT_SYMBOL(dcache_dir_open);
625 EXPORT_SYMBOL(dcache_readdir);
626 EXPORT_SYMBOL(generic_read_dir);
627 EXPORT_SYMBOL(get_sb_pseudo);
628 EXPORT_SYMBOL(simple_commit_write);
629 EXPORT_SYMBOL(simple_dir_inode_operations);
630 EXPORT_SYMBOL(simple_dir_operations);
631 EXPORT_SYMBOL(simple_empty);
632 EXPORT_SYMBOL(d_alloc_name);
633 EXPORT_SYMBOL(simple_fill_super);
634 EXPORT_SYMBOL(simple_getattr);
635 EXPORT_SYMBOL(simple_link);
636 EXPORT_SYMBOL(simple_lookup);
637 EXPORT_SYMBOL(simple_pin_fs);
638 EXPORT_SYMBOL(simple_prepare_write);
639 EXPORT_SYMBOL(simple_readpage);
640 EXPORT_SYMBOL(simple_release_fs);
641 EXPORT_SYMBOL(simple_rename);
642 EXPORT_SYMBOL(simple_rmdir);
643 EXPORT_SYMBOL(simple_statfs);
644 EXPORT_SYMBOL(simple_sync_file);
645 EXPORT_SYMBOL(simple_unlink);
646 EXPORT_SYMBOL(simple_read_from_buffer);
647 EXPORT_SYMBOL(simple_transaction_get);
648 EXPORT_SYMBOL(simple_transaction_read);
649 EXPORT_SYMBOL(simple_transaction_release);
650 EXPORT_SYMBOL_GPL(simple_attr_open);
651 EXPORT_SYMBOL_GPL(simple_attr_close);
652 EXPORT_SYMBOL_GPL(simple_attr_read);
653 EXPORT_SYMBOL_GPL(simple_attr_write);