[GFS2] Move pin/unpin into lops.c, clean up locking
[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, 
163                                             next->d_name.len, filp->f_pos, 
164                                             next->d_inode->i_ino, 
165                                             dt_type(next->d_inode)) < 0)
166                                         return 0;
167                                 spin_lock(&dcache_lock);
168                                 /* next is still alive */
169                                 list_move(q, p);
170                                 p = q;
171                                 filp->f_pos++;
172                         }
173                         spin_unlock(&dcache_lock);
174         }
175         return 0;
176 }
177
178 ssize_t generic_read_dir(struct file *filp, char __user *buf, size_t siz, loff_t *ppos)
179 {
180         return -EISDIR;
181 }
182
183 const struct file_operations simple_dir_operations = {
184         .open           = dcache_dir_open,
185         .release        = dcache_dir_close,
186         .llseek         = dcache_dir_lseek,
187         .read           = generic_read_dir,
188         .readdir        = dcache_readdir,
189         .fsync          = simple_sync_file,
190 };
191
192 const struct inode_operations simple_dir_inode_operations = {
193         .lookup         = simple_lookup,
194 };
195
196 static const struct super_operations simple_super_operations = {
197         .statfs         = simple_statfs,
198 };
199
200 /*
201  * Common helper for pseudo-filesystems (sockfs, pipefs, bdev - stuff that
202  * will never be mountable)
203  */
204 int get_sb_pseudo(struct file_system_type *fs_type, char *name,
205         const struct super_operations *ops, unsigned long magic,
206         struct vfsmount *mnt)
207 {
208         struct super_block *s = sget(fs_type, NULL, set_anon_super, NULL);
209         struct dentry *dentry;
210         struct inode *root;
211         struct qstr d_name = {.name = name, .len = strlen(name)};
212
213         if (IS_ERR(s))
214                 return PTR_ERR(s);
215
216         s->s_flags = MS_NOUSER;
217         s->s_maxbytes = ~0ULL;
218         s->s_blocksize = 1024;
219         s->s_blocksize_bits = 10;
220         s->s_magic = magic;
221         s->s_op = ops ? ops : &simple_super_operations;
222         s->s_time_gran = 1;
223         root = new_inode(s);
224         if (!root)
225                 goto Enomem;
226         /*
227          * since this is the first inode, make it number 1. New inodes created
228          * after this must take care not to collide with it (by passing
229          * max_reserved of 1 to iunique).
230          */
231         root->i_ino = 1;
232         root->i_mode = S_IFDIR | S_IRUSR | S_IWUSR;
233         root->i_uid = root->i_gid = 0;
234         root->i_atime = root->i_mtime = root->i_ctime = CURRENT_TIME;
235         dentry = d_alloc(NULL, &d_name);
236         if (!dentry) {
237                 iput(root);
238                 goto Enomem;
239         }
240         dentry->d_sb = s;
241         dentry->d_parent = dentry;
242         d_instantiate(dentry, root);
243         s->s_root = dentry;
244         s->s_flags |= MS_ACTIVE;
245         return simple_set_mnt(mnt, s);
246
247 Enomem:
248         up_write(&s->s_umount);
249         deactivate_super(s);
250         return -ENOMEM;
251 }
252
253 int simple_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
254 {
255         struct inode *inode = old_dentry->d_inode;
256
257         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
258         inc_nlink(inode);
259         atomic_inc(&inode->i_count);
260         dget(dentry);
261         d_instantiate(dentry, inode);
262         return 0;
263 }
264
265 static inline int simple_positive(struct dentry *dentry)
266 {
267         return dentry->d_inode && !d_unhashed(dentry);
268 }
269
270 int simple_empty(struct dentry *dentry)
271 {
272         struct dentry *child;
273         int ret = 0;
274
275         spin_lock(&dcache_lock);
276         list_for_each_entry(child, &dentry->d_subdirs, d_u.d_child)
277                 if (simple_positive(child))
278                         goto out;
279         ret = 1;
280 out:
281         spin_unlock(&dcache_lock);
282         return ret;
283 }
284
285 int simple_unlink(struct inode *dir, struct dentry *dentry)
286 {
287         struct inode *inode = dentry->d_inode;
288
289         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
290         drop_nlink(inode);
291         dput(dentry);
292         return 0;
293 }
294
295 int simple_rmdir(struct inode *dir, struct dentry *dentry)
296 {
297         if (!simple_empty(dentry))
298                 return -ENOTEMPTY;
299
300         drop_nlink(dentry->d_inode);
301         simple_unlink(dir, dentry);
302         drop_nlink(dir);
303         return 0;
304 }
305
306 int simple_rename(struct inode *old_dir, struct dentry *old_dentry,
307                 struct inode *new_dir, struct dentry *new_dentry)
308 {
309         struct inode *inode = old_dentry->d_inode;
310         int they_are_dirs = S_ISDIR(old_dentry->d_inode->i_mode);
311
312         if (!simple_empty(new_dentry))
313                 return -ENOTEMPTY;
314
315         if (new_dentry->d_inode) {
316                 simple_unlink(new_dir, new_dentry);
317                 if (they_are_dirs)
318                         drop_nlink(old_dir);
319         } else if (they_are_dirs) {
320                 drop_nlink(old_dir);
321                 inc_nlink(new_dir);
322         }
323
324         old_dir->i_ctime = old_dir->i_mtime = new_dir->i_ctime =
325                 new_dir->i_mtime = inode->i_ctime = CURRENT_TIME;
326
327         return 0;
328 }
329
330 int simple_readpage(struct file *file, struct page *page)
331 {
332         clear_highpage(page);
333         flush_dcache_page(page);
334         SetPageUptodate(page);
335         unlock_page(page);
336         return 0;
337 }
338
339 int simple_prepare_write(struct file *file, struct page *page,
340                         unsigned from, unsigned to)
341 {
342         if (!PageUptodate(page)) {
343                 if (to - from != PAGE_CACHE_SIZE) {
344                         void *kaddr = kmap_atomic(page, KM_USER0);
345                         memset(kaddr, 0, from);
346                         memset(kaddr + to, 0, PAGE_CACHE_SIZE - to);
347                         flush_dcache_page(page);
348                         kunmap_atomic(kaddr, KM_USER0);
349                 }
350         }
351         return 0;
352 }
353
354 int simple_commit_write(struct file *file, struct page *page,
355                         unsigned from, unsigned to)
356 {
357         struct inode *inode = page->mapping->host;
358         loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
359
360         if (!PageUptodate(page))
361                 SetPageUptodate(page);
362         /*
363          * No need to use i_size_read() here, the i_size
364          * cannot change under us because we hold the i_mutex.
365          */
366         if (pos > inode->i_size)
367                 i_size_write(inode, pos);
368         set_page_dirty(page);
369         return 0;
370 }
371
372 /*
373  * the inodes created here are not hashed. If you use iunique to generate
374  * unique inode values later for this filesystem, then you must take care
375  * to pass it an appropriate max_reserved value to avoid collisions.
376  */
377 int simple_fill_super(struct super_block *s, int magic, struct tree_descr *files)
378 {
379         struct inode *inode;
380         struct dentry *root;
381         struct dentry *dentry;
382         int i;
383
384         s->s_blocksize = PAGE_CACHE_SIZE;
385         s->s_blocksize_bits = PAGE_CACHE_SHIFT;
386         s->s_magic = magic;
387         s->s_op = &simple_super_operations;
388         s->s_time_gran = 1;
389
390         inode = new_inode(s);
391         if (!inode)
392                 return -ENOMEM;
393         /*
394          * because the root inode is 1, the files array must not contain an
395          * entry at index 1
396          */
397         inode->i_ino = 1;
398         inode->i_mode = S_IFDIR | 0755;
399         inode->i_uid = inode->i_gid = 0;
400         inode->i_blocks = 0;
401         inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
402         inode->i_op = &simple_dir_inode_operations;
403         inode->i_fop = &simple_dir_operations;
404         inode->i_nlink = 2;
405         root = d_alloc_root(inode);
406         if (!root) {
407                 iput(inode);
408                 return -ENOMEM;
409         }
410         for (i = 0; !files->name || files->name[0]; i++, files++) {
411                 if (!files->name)
412                         continue;
413
414                 /* warn if it tries to conflict with the root inode */
415                 if (unlikely(i == 1))
416                         printk(KERN_WARNING "%s: %s passed in a files array"
417                                 "with an index of 1!\n", __func__,
418                                 s->s_type->name);
419
420                 dentry = d_alloc_name(root, files->name);
421                 if (!dentry)
422                         goto out;
423                 inode = new_inode(s);
424                 if (!inode)
425                         goto out;
426                 inode->i_mode = S_IFREG | files->mode;
427                 inode->i_uid = inode->i_gid = 0;
428                 inode->i_blocks = 0;
429                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
430                 inode->i_fop = files->ops;
431                 inode->i_ino = i;
432                 d_add(dentry, inode);
433         }
434         s->s_root = root;
435         return 0;
436 out:
437         d_genocide(root);
438         dput(root);
439         return -ENOMEM;
440 }
441
442 static DEFINE_SPINLOCK(pin_fs_lock);
443
444 int simple_pin_fs(struct file_system_type *type, struct vfsmount **mount, int *count)
445 {
446         struct vfsmount *mnt = NULL;
447         spin_lock(&pin_fs_lock);
448         if (unlikely(!*mount)) {
449                 spin_unlock(&pin_fs_lock);
450                 mnt = vfs_kern_mount(type, 0, type->name, NULL);
451                 if (IS_ERR(mnt))
452                         return PTR_ERR(mnt);
453                 spin_lock(&pin_fs_lock);
454                 if (!*mount)
455                         *mount = mnt;
456         }
457         mntget(*mount);
458         ++*count;
459         spin_unlock(&pin_fs_lock);
460         mntput(mnt);
461         return 0;
462 }
463
464 void simple_release_fs(struct vfsmount **mount, int *count)
465 {
466         struct vfsmount *mnt;
467         spin_lock(&pin_fs_lock);
468         mnt = *mount;
469         if (!--*count)
470                 *mount = NULL;
471         spin_unlock(&pin_fs_lock);
472         mntput(mnt);
473 }
474
475 ssize_t simple_read_from_buffer(void __user *to, size_t count, loff_t *ppos,
476                                 const void *from, size_t available)
477 {
478         loff_t pos = *ppos;
479         if (pos < 0)
480                 return -EINVAL;
481         if (pos >= available)
482                 return 0;
483         if (count > available - pos)
484                 count = available - pos;
485         if (copy_to_user(to, from + pos, count))
486                 return -EFAULT;
487         *ppos = pos + count;
488         return count;
489 }
490
491 /*
492  * Transaction based IO.
493  * The file expects a single write which triggers the transaction, and then
494  * possibly a read which collects the result - which is stored in a
495  * file-local buffer.
496  */
497 char *simple_transaction_get(struct file *file, const char __user *buf, size_t size)
498 {
499         struct simple_transaction_argresp *ar;
500         static DEFINE_SPINLOCK(simple_transaction_lock);
501
502         if (size > SIMPLE_TRANSACTION_LIMIT - 1)
503                 return ERR_PTR(-EFBIG);
504
505         ar = (struct simple_transaction_argresp *)get_zeroed_page(GFP_KERNEL);
506         if (!ar)
507                 return ERR_PTR(-ENOMEM);
508
509         spin_lock(&simple_transaction_lock);
510
511         /* only one write allowed per open */
512         if (file->private_data) {
513                 spin_unlock(&simple_transaction_lock);
514                 free_page((unsigned long)ar);
515                 return ERR_PTR(-EBUSY);
516         }
517
518         file->private_data = ar;
519
520         spin_unlock(&simple_transaction_lock);
521
522         if (copy_from_user(ar->data, buf, size))
523                 return ERR_PTR(-EFAULT);
524
525         return ar->data;
526 }
527
528 ssize_t simple_transaction_read(struct file *file, char __user *buf, size_t size, loff_t *pos)
529 {
530         struct simple_transaction_argresp *ar = file->private_data;
531
532         if (!ar)
533                 return 0;
534         return simple_read_from_buffer(buf, size, pos, ar->data, ar->size);
535 }
536
537 int simple_transaction_release(struct inode *inode, struct file *file)
538 {
539         free_page((unsigned long)file->private_data);
540         return 0;
541 }
542
543 /* Simple attribute files */
544
545 struct simple_attr {
546         u64 (*get)(void *);
547         void (*set)(void *, u64);
548         char get_buf[24];       /* enough to store a u64 and "\n\0" */
549         char set_buf[24];
550         void *data;
551         const char *fmt;        /* format for read operation */
552         struct mutex mutex;     /* protects access to these buffers */
553 };
554
555 /* simple_attr_open is called by an actual attribute open file operation
556  * to set the attribute specific access operations. */
557 int simple_attr_open(struct inode *inode, struct file *file,
558                      u64 (*get)(void *), void (*set)(void *, u64),
559                      const char *fmt)
560 {
561         struct simple_attr *attr;
562
563         attr = kmalloc(sizeof(*attr), GFP_KERNEL);
564         if (!attr)
565                 return -ENOMEM;
566
567         attr->get = get;
568         attr->set = set;
569         attr->data = inode->i_private;
570         attr->fmt = fmt;
571         mutex_init(&attr->mutex);
572
573         file->private_data = attr;
574
575         return nonseekable_open(inode, file);
576 }
577
578 int simple_attr_close(struct inode *inode, struct file *file)
579 {
580         kfree(file->private_data);
581         return 0;
582 }
583
584 /* read from the buffer that is filled with the get function */
585 ssize_t simple_attr_read(struct file *file, char __user *buf,
586                          size_t len, loff_t *ppos)
587 {
588         struct simple_attr *attr;
589         size_t size;
590         ssize_t ret;
591
592         attr = file->private_data;
593
594         if (!attr->get)
595                 return -EACCES;
596
597         mutex_lock(&attr->mutex);
598         if (*ppos) /* continued read */
599                 size = strlen(attr->get_buf);
600         else      /* first read */
601                 size = scnprintf(attr->get_buf, sizeof(attr->get_buf),
602                                  attr->fmt,
603                                  (unsigned long long)attr->get(attr->data));
604
605         ret = simple_read_from_buffer(buf, len, ppos, attr->get_buf, size);
606         mutex_unlock(&attr->mutex);
607         return ret;
608 }
609
610 /* interpret the buffer as a number to call the set function with */
611 ssize_t simple_attr_write(struct file *file, const char __user *buf,
612                           size_t len, loff_t *ppos)
613 {
614         struct simple_attr *attr;
615         u64 val;
616         size_t size;
617         ssize_t ret;
618
619         attr = file->private_data;
620
621         if (!attr->set)
622                 return -EACCES;
623
624         mutex_lock(&attr->mutex);
625         ret = -EFAULT;
626         size = min(sizeof(attr->set_buf) - 1, len);
627         if (copy_from_user(attr->set_buf, buf, size))
628                 goto out;
629
630         ret = len; /* claim we got the whole input */
631         attr->set_buf[size] = '\0';
632         val = simple_strtol(attr->set_buf, NULL, 0);
633         attr->set(attr->data, val);
634 out:
635         mutex_unlock(&attr->mutex);
636         return ret;
637 }
638
639 EXPORT_SYMBOL(dcache_dir_close);
640 EXPORT_SYMBOL(dcache_dir_lseek);
641 EXPORT_SYMBOL(dcache_dir_open);
642 EXPORT_SYMBOL(dcache_readdir);
643 EXPORT_SYMBOL(generic_read_dir);
644 EXPORT_SYMBOL(get_sb_pseudo);
645 EXPORT_SYMBOL(simple_commit_write);
646 EXPORT_SYMBOL(simple_dir_inode_operations);
647 EXPORT_SYMBOL(simple_dir_operations);
648 EXPORT_SYMBOL(simple_empty);
649 EXPORT_SYMBOL(d_alloc_name);
650 EXPORT_SYMBOL(simple_fill_super);
651 EXPORT_SYMBOL(simple_getattr);
652 EXPORT_SYMBOL(simple_link);
653 EXPORT_SYMBOL(simple_lookup);
654 EXPORT_SYMBOL(simple_pin_fs);
655 EXPORT_SYMBOL(simple_prepare_write);
656 EXPORT_SYMBOL(simple_readpage);
657 EXPORT_SYMBOL(simple_release_fs);
658 EXPORT_SYMBOL(simple_rename);
659 EXPORT_SYMBOL(simple_rmdir);
660 EXPORT_SYMBOL(simple_statfs);
661 EXPORT_SYMBOL(simple_sync_file);
662 EXPORT_SYMBOL(simple_unlink);
663 EXPORT_SYMBOL(simple_read_from_buffer);
664 EXPORT_SYMBOL(simple_transaction_get);
665 EXPORT_SYMBOL(simple_transaction_read);
666 EXPORT_SYMBOL(simple_transaction_release);
667 EXPORT_SYMBOL_GPL(simple_attr_open);
668 EXPORT_SYMBOL_GPL(simple_attr_close);
669 EXPORT_SYMBOL_GPL(simple_attr_read);
670 EXPORT_SYMBOL_GPL(simple_attr_write);