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