Merge branch 'upstream-linus' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik...
[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         /*
224          * since this is the first inode, make it number 1. New inodes created
225          * after this must take care not to collide with it (by passing
226          * max_reserved of 1 to iunique).
227          */
228         root->i_ino = 1;
229         root->i_mode = S_IFDIR | S_IRUSR | S_IWUSR;
230         root->i_uid = root->i_gid = 0;
231         root->i_atime = root->i_mtime = root->i_ctime = CURRENT_TIME;
232         dentry = d_alloc(NULL, &d_name);
233         if (!dentry) {
234                 iput(root);
235                 goto Enomem;
236         }
237         dentry->d_sb = s;
238         dentry->d_parent = dentry;
239         d_instantiate(dentry, root);
240         s->s_root = dentry;
241         s->s_flags |= MS_ACTIVE;
242         return simple_set_mnt(mnt, s);
243
244 Enomem:
245         up_write(&s->s_umount);
246         deactivate_super(s);
247         return -ENOMEM;
248 }
249
250 int simple_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
251 {
252         struct inode *inode = old_dentry->d_inode;
253
254         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
255         inc_nlink(inode);
256         atomic_inc(&inode->i_count);
257         dget(dentry);
258         d_instantiate(dentry, inode);
259         return 0;
260 }
261
262 static inline int simple_positive(struct dentry *dentry)
263 {
264         return dentry->d_inode && !d_unhashed(dentry);
265 }
266
267 int simple_empty(struct dentry *dentry)
268 {
269         struct dentry *child;
270         int ret = 0;
271
272         spin_lock(&dcache_lock);
273         list_for_each_entry(child, &dentry->d_subdirs, d_u.d_child)
274                 if (simple_positive(child))
275                         goto out;
276         ret = 1;
277 out:
278         spin_unlock(&dcache_lock);
279         return ret;
280 }
281
282 int simple_unlink(struct inode *dir, struct dentry *dentry)
283 {
284         struct inode *inode = dentry->d_inode;
285
286         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
287         drop_nlink(inode);
288         dput(dentry);
289         return 0;
290 }
291
292 int simple_rmdir(struct inode *dir, struct dentry *dentry)
293 {
294         if (!simple_empty(dentry))
295                 return -ENOTEMPTY;
296
297         drop_nlink(dentry->d_inode);
298         simple_unlink(dir, dentry);
299         drop_nlink(dir);
300         return 0;
301 }
302
303 int simple_rename(struct inode *old_dir, struct dentry *old_dentry,
304                 struct inode *new_dir, struct dentry *new_dentry)
305 {
306         struct inode *inode = old_dentry->d_inode;
307         int they_are_dirs = S_ISDIR(old_dentry->d_inode->i_mode);
308
309         if (!simple_empty(new_dentry))
310                 return -ENOTEMPTY;
311
312         if (new_dentry->d_inode) {
313                 simple_unlink(new_dir, new_dentry);
314                 if (they_are_dirs)
315                         drop_nlink(old_dir);
316         } else if (they_are_dirs) {
317                 drop_nlink(old_dir);
318                 inc_nlink(new_dir);
319         }
320
321         old_dir->i_ctime = old_dir->i_mtime = new_dir->i_ctime =
322                 new_dir->i_mtime = inode->i_ctime = CURRENT_TIME;
323
324         return 0;
325 }
326
327 int simple_readpage(struct file *file, struct page *page)
328 {
329         clear_highpage(page);
330         flush_dcache_page(page);
331         SetPageUptodate(page);
332         unlock_page(page);
333         return 0;
334 }
335
336 int simple_prepare_write(struct file *file, struct page *page,
337                         unsigned from, unsigned to)
338 {
339         if (!PageUptodate(page)) {
340                 if (to - from != PAGE_CACHE_SIZE) {
341                         void *kaddr = kmap_atomic(page, KM_USER0);
342                         memset(kaddr, 0, from);
343                         memset(kaddr + to, 0, PAGE_CACHE_SIZE - to);
344                         flush_dcache_page(page);
345                         kunmap_atomic(kaddr, KM_USER0);
346                 }
347         }
348         return 0;
349 }
350
351 int simple_commit_write(struct file *file, struct page *page,
352                         unsigned from, unsigned to)
353 {
354         struct inode *inode = page->mapping->host;
355         loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
356
357         if (!PageUptodate(page))
358                 SetPageUptodate(page);
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 /*
370  * the inodes created here are not hashed. If you use iunique to generate
371  * unique inode values later for this filesystem, then you must take care
372  * to pass it an appropriate max_reserved value to avoid collisions.
373  */
374 int simple_fill_super(struct super_block *s, int magic, struct tree_descr *files)
375 {
376         struct inode *inode;
377         struct dentry *root;
378         struct dentry *dentry;
379         int i;
380
381         s->s_blocksize = PAGE_CACHE_SIZE;
382         s->s_blocksize_bits = PAGE_CACHE_SHIFT;
383         s->s_magic = magic;
384         s->s_op = &simple_super_operations;
385         s->s_time_gran = 1;
386
387         inode = new_inode(s);
388         if (!inode)
389                 return -ENOMEM;
390         /*
391          * because the root inode is 1, the files array must not contain an
392          * entry at index 1
393          */
394         inode->i_ino = 1;
395         inode->i_mode = S_IFDIR | 0755;
396         inode->i_uid = inode->i_gid = 0;
397         inode->i_blocks = 0;
398         inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
399         inode->i_op = &simple_dir_inode_operations;
400         inode->i_fop = &simple_dir_operations;
401         inode->i_nlink = 2;
402         root = d_alloc_root(inode);
403         if (!root) {
404                 iput(inode);
405                 return -ENOMEM;
406         }
407         for (i = 0; !files->name || files->name[0]; i++, files++) {
408                 if (!files->name)
409                         continue;
410
411                 /* warn if it tries to conflict with the root inode */
412                 if (unlikely(i == 1))
413                         printk(KERN_WARNING "%s: %s passed in a files array"
414                                 "with an index of 1!\n", __func__,
415                                 s->s_type->name);
416
417                 dentry = d_alloc_name(root, files->name);
418                 if (!dentry)
419                         goto out;
420                 inode = new_inode(s);
421                 if (!inode)
422                         goto out;
423                 inode->i_mode = S_IFREG | files->mode;
424                 inode->i_uid = inode->i_gid = 0;
425                 inode->i_blocks = 0;
426                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
427                 inode->i_fop = files->ops;
428                 inode->i_ino = i;
429                 d_add(dentry, inode);
430         }
431         s->s_root = root;
432         return 0;
433 out:
434         d_genocide(root);
435         dput(root);
436         return -ENOMEM;
437 }
438
439 static DEFINE_SPINLOCK(pin_fs_lock);
440
441 int simple_pin_fs(struct file_system_type *type, struct vfsmount **mount, int *count)
442 {
443         struct vfsmount *mnt = NULL;
444         spin_lock(&pin_fs_lock);
445         if (unlikely(!*mount)) {
446                 spin_unlock(&pin_fs_lock);
447                 mnt = vfs_kern_mount(type, 0, type->name, NULL);
448                 if (IS_ERR(mnt))
449                         return PTR_ERR(mnt);
450                 spin_lock(&pin_fs_lock);
451                 if (!*mount)
452                         *mount = mnt;
453         }
454         mntget(*mount);
455         ++*count;
456         spin_unlock(&pin_fs_lock);
457         mntput(mnt);
458         return 0;
459 }
460
461 void simple_release_fs(struct vfsmount **mount, int *count)
462 {
463         struct vfsmount *mnt;
464         spin_lock(&pin_fs_lock);
465         mnt = *mount;
466         if (!--*count)
467                 *mount = NULL;
468         spin_unlock(&pin_fs_lock);
469         mntput(mnt);
470 }
471
472 ssize_t simple_read_from_buffer(void __user *to, size_t count, loff_t *ppos,
473                                 const void *from, size_t available)
474 {
475         loff_t pos = *ppos;
476         if (pos < 0)
477                 return -EINVAL;
478         if (pos >= available)
479                 return 0;
480         if (count > available - pos)
481                 count = available - pos;
482         if (copy_to_user(to, from + pos, count))
483                 return -EFAULT;
484         *ppos = pos + count;
485         return count;
486 }
487
488 /*
489  * Transaction based IO.
490  * The file expects a single write which triggers the transaction, and then
491  * possibly a read which collects the result - which is stored in a
492  * file-local buffer.
493  */
494 char *simple_transaction_get(struct file *file, const char __user *buf, size_t size)
495 {
496         struct simple_transaction_argresp *ar;
497         static DEFINE_SPINLOCK(simple_transaction_lock);
498
499         if (size > SIMPLE_TRANSACTION_LIMIT - 1)
500                 return ERR_PTR(-EFBIG);
501
502         ar = (struct simple_transaction_argresp *)get_zeroed_page(GFP_KERNEL);
503         if (!ar)
504                 return ERR_PTR(-ENOMEM);
505
506         spin_lock(&simple_transaction_lock);
507
508         /* only one write allowed per open */
509         if (file->private_data) {
510                 spin_unlock(&simple_transaction_lock);
511                 free_page((unsigned long)ar);
512                 return ERR_PTR(-EBUSY);
513         }
514
515         file->private_data = ar;
516
517         spin_unlock(&simple_transaction_lock);
518
519         if (copy_from_user(ar->data, buf, size))
520                 return ERR_PTR(-EFAULT);
521
522         return ar->data;
523 }
524
525 ssize_t simple_transaction_read(struct file *file, char __user *buf, size_t size, loff_t *pos)
526 {
527         struct simple_transaction_argresp *ar = file->private_data;
528
529         if (!ar)
530                 return 0;
531         return simple_read_from_buffer(buf, size, pos, ar->data, ar->size);
532 }
533
534 int simple_transaction_release(struct inode *inode, struct file *file)
535 {
536         free_page((unsigned long)file->private_data);
537         return 0;
538 }
539
540 /* Simple attribute files */
541
542 struct simple_attr {
543         u64 (*get)(void *);
544         void (*set)(void *, u64);
545         char get_buf[24];       /* enough to store a u64 and "\n\0" */
546         char set_buf[24];
547         void *data;
548         const char *fmt;        /* format for read operation */
549         struct mutex mutex;     /* protects access to these buffers */
550 };
551
552 /* simple_attr_open is called by an actual attribute open file operation
553  * to set the attribute specific access operations. */
554 int simple_attr_open(struct inode *inode, struct file *file,
555                      u64 (*get)(void *), void (*set)(void *, u64),
556                      const char *fmt)
557 {
558         struct simple_attr *attr;
559
560         attr = kmalloc(sizeof(*attr), GFP_KERNEL);
561         if (!attr)
562                 return -ENOMEM;
563
564         attr->get = get;
565         attr->set = set;
566         attr->data = inode->i_private;
567         attr->fmt = fmt;
568         mutex_init(&attr->mutex);
569
570         file->private_data = attr;
571
572         return nonseekable_open(inode, file);
573 }
574
575 int simple_attr_close(struct inode *inode, struct file *file)
576 {
577         kfree(file->private_data);
578         return 0;
579 }
580
581 /* read from the buffer that is filled with the get function */
582 ssize_t simple_attr_read(struct file *file, char __user *buf,
583                          size_t len, loff_t *ppos)
584 {
585         struct simple_attr *attr;
586         size_t size;
587         ssize_t ret;
588
589         attr = file->private_data;
590
591         if (!attr->get)
592                 return -EACCES;
593
594         mutex_lock(&attr->mutex);
595         if (*ppos) /* continued read */
596                 size = strlen(attr->get_buf);
597         else      /* first read */
598                 size = scnprintf(attr->get_buf, sizeof(attr->get_buf),
599                                  attr->fmt,
600                                  (unsigned long long)attr->get(attr->data));
601
602         ret = simple_read_from_buffer(buf, len, ppos, attr->get_buf, size);
603         mutex_unlock(&attr->mutex);
604         return ret;
605 }
606
607 /* interpret the buffer as a number to call the set function with */
608 ssize_t simple_attr_write(struct file *file, const char __user *buf,
609                           size_t len, loff_t *ppos)
610 {
611         struct simple_attr *attr;
612         u64 val;
613         size_t size;
614         ssize_t ret;
615
616         attr = file->private_data;
617
618         if (!attr->set)
619                 return -EACCES;
620
621         mutex_lock(&attr->mutex);
622         ret = -EFAULT;
623         size = min(sizeof(attr->set_buf) - 1, len);
624         if (copy_from_user(attr->set_buf, buf, size))
625                 goto out;
626
627         ret = len; /* claim we got the whole input */
628         attr->set_buf[size] = '\0';
629         val = simple_strtol(attr->set_buf, NULL, 0);
630         attr->set(attr->data, val);
631 out:
632         mutex_unlock(&attr->mutex);
633         return ret;
634 }
635
636 EXPORT_SYMBOL(dcache_dir_close);
637 EXPORT_SYMBOL(dcache_dir_lseek);
638 EXPORT_SYMBOL(dcache_dir_open);
639 EXPORT_SYMBOL(dcache_readdir);
640 EXPORT_SYMBOL(generic_read_dir);
641 EXPORT_SYMBOL(get_sb_pseudo);
642 EXPORT_SYMBOL(simple_commit_write);
643 EXPORT_SYMBOL(simple_dir_inode_operations);
644 EXPORT_SYMBOL(simple_dir_operations);
645 EXPORT_SYMBOL(simple_empty);
646 EXPORT_SYMBOL(d_alloc_name);
647 EXPORT_SYMBOL(simple_fill_super);
648 EXPORT_SYMBOL(simple_getattr);
649 EXPORT_SYMBOL(simple_link);
650 EXPORT_SYMBOL(simple_lookup);
651 EXPORT_SYMBOL(simple_pin_fs);
652 EXPORT_SYMBOL(simple_prepare_write);
653 EXPORT_SYMBOL(simple_readpage);
654 EXPORT_SYMBOL(simple_release_fs);
655 EXPORT_SYMBOL(simple_rename);
656 EXPORT_SYMBOL(simple_rmdir);
657 EXPORT_SYMBOL(simple_statfs);
658 EXPORT_SYMBOL(simple_sync_file);
659 EXPORT_SYMBOL(simple_unlink);
660 EXPORT_SYMBOL(simple_read_from_buffer);
661 EXPORT_SYMBOL(simple_transaction_get);
662 EXPORT_SYMBOL(simple_transaction_read);
663 EXPORT_SYMBOL(simple_transaction_release);
664 EXPORT_SYMBOL_GPL(simple_attr_open);
665 EXPORT_SYMBOL_GPL(simple_attr_close);
666 EXPORT_SYMBOL_GPL(simple_attr_read);
667 EXPORT_SYMBOL_GPL(simple_attr_write);