[PATCH] inode-diet: Eliminate i_blksize from the inode structure
[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 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_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_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 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 int get_sb_pseudo(struct file_system_type *fs_type, char *name,
198         struct super_operations *ops, unsigned long magic,
199         struct vfsmount *mnt)
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 PTR_ERR(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 simple_set_mnt(mnt, s);
234
235 Enomem:
236         up_write(&s->s_umount);
237         deactivate_super(s);
238         return -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_blocks = 0;
387         inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
388         inode->i_op = &simple_dir_inode_operations;
389         inode->i_fop = &simple_dir_operations;
390         inode->i_nlink = 2;
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_blocks = 0;
408                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
409                 inode->i_fop = files->ops;
410                 inode->i_ino = i;
411                 d_add(dentry, inode);
412         }
413         s->s_root = root;
414         return 0;
415 out:
416         d_genocide(root);
417         dput(root);
418         return -ENOMEM;
419 }
420
421 static DEFINE_SPINLOCK(pin_fs_lock);
422
423 int simple_pin_fs(struct file_system_type *type, struct vfsmount **mount, int *count)
424 {
425         struct vfsmount *mnt = NULL;
426         spin_lock(&pin_fs_lock);
427         if (unlikely(!*mount)) {
428                 spin_unlock(&pin_fs_lock);
429                 mnt = vfs_kern_mount(type, 0, type->name, NULL);
430                 if (IS_ERR(mnt))
431                         return PTR_ERR(mnt);
432                 spin_lock(&pin_fs_lock);
433                 if (!*mount)
434                         *mount = mnt;
435         }
436         mntget(*mount);
437         ++*count;
438         spin_unlock(&pin_fs_lock);
439         mntput(mnt);
440         return 0;
441 }
442
443 void simple_release_fs(struct vfsmount **mount, int *count)
444 {
445         struct vfsmount *mnt;
446         spin_lock(&pin_fs_lock);
447         mnt = *mount;
448         if (!--*count)
449                 *mount = NULL;
450         spin_unlock(&pin_fs_lock);
451         mntput(mnt);
452 }
453
454 ssize_t simple_read_from_buffer(void __user *to, size_t count, loff_t *ppos,
455                                 const void *from, size_t available)
456 {
457         loff_t pos = *ppos;
458         if (pos < 0)
459                 return -EINVAL;
460         if (pos >= available)
461                 return 0;
462         if (count > available - pos)
463                 count = available - pos;
464         if (copy_to_user(to, from + pos, count))
465                 return -EFAULT;
466         *ppos = pos + count;
467         return count;
468 }
469
470 /*
471  * Transaction based IO.
472  * The file expects a single write which triggers the transaction, and then
473  * possibly a read which collects the result - which is stored in a
474  * file-local buffer.
475  */
476 char *simple_transaction_get(struct file *file, const char __user *buf, size_t size)
477 {
478         struct simple_transaction_argresp *ar;
479         static DEFINE_SPINLOCK(simple_transaction_lock);
480
481         if (size > SIMPLE_TRANSACTION_LIMIT - 1)
482                 return ERR_PTR(-EFBIG);
483
484         ar = (struct simple_transaction_argresp *)get_zeroed_page(GFP_KERNEL);
485         if (!ar)
486                 return ERR_PTR(-ENOMEM);
487
488         spin_lock(&simple_transaction_lock);
489
490         /* only one write allowed per open */
491         if (file->private_data) {
492                 spin_unlock(&simple_transaction_lock);
493                 free_page((unsigned long)ar);
494                 return ERR_PTR(-EBUSY);
495         }
496
497         file->private_data = ar;
498
499         spin_unlock(&simple_transaction_lock);
500
501         if (copy_from_user(ar->data, buf, size))
502                 return ERR_PTR(-EFAULT);
503
504         return ar->data;
505 }
506
507 ssize_t simple_transaction_read(struct file *file, char __user *buf, size_t size, loff_t *pos)
508 {
509         struct simple_transaction_argresp *ar = file->private_data;
510
511         if (!ar)
512                 return 0;
513         return simple_read_from_buffer(buf, size, pos, ar->data, ar->size);
514 }
515
516 int simple_transaction_release(struct inode *inode, struct file *file)
517 {
518         free_page((unsigned long)file->private_data);
519         return 0;
520 }
521
522 /* Simple attribute files */
523
524 struct simple_attr {
525         u64 (*get)(void *);
526         void (*set)(void *, u64);
527         char get_buf[24];       /* enough to store a u64 and "\n\0" */
528         char set_buf[24];
529         void *data;
530         const char *fmt;        /* format for read operation */
531         struct mutex mutex;     /* protects access to these buffers */
532 };
533
534 /* simple_attr_open is called by an actual attribute open file operation
535  * to set the attribute specific access operations. */
536 int simple_attr_open(struct inode *inode, struct file *file,
537                      u64 (*get)(void *), void (*set)(void *, u64),
538                      const char *fmt)
539 {
540         struct simple_attr *attr;
541
542         attr = kmalloc(sizeof(*attr), GFP_KERNEL);
543         if (!attr)
544                 return -ENOMEM;
545
546         attr->get = get;
547         attr->set = set;
548         attr->data = inode->i_private;
549         attr->fmt = fmt;
550         mutex_init(&attr->mutex);
551
552         file->private_data = attr;
553
554         return nonseekable_open(inode, file);
555 }
556
557 int simple_attr_close(struct inode *inode, struct file *file)
558 {
559         kfree(file->private_data);
560         return 0;
561 }
562
563 /* read from the buffer that is filled with the get function */
564 ssize_t simple_attr_read(struct file *file, char __user *buf,
565                          size_t len, loff_t *ppos)
566 {
567         struct simple_attr *attr;
568         size_t size;
569         ssize_t ret;
570
571         attr = file->private_data;
572
573         if (!attr->get)
574                 return -EACCES;
575
576         mutex_lock(&attr->mutex);
577         if (*ppos) /* continued read */
578                 size = strlen(attr->get_buf);
579         else      /* first read */
580                 size = scnprintf(attr->get_buf, sizeof(attr->get_buf),
581                                  attr->fmt,
582                                  (unsigned long long)attr->get(attr->data));
583
584         ret = simple_read_from_buffer(buf, len, ppos, attr->get_buf, size);
585         mutex_unlock(&attr->mutex);
586         return ret;
587 }
588
589 /* interpret the buffer as a number to call the set function with */
590 ssize_t simple_attr_write(struct file *file, const char __user *buf,
591                           size_t len, loff_t *ppos)
592 {
593         struct simple_attr *attr;
594         u64 val;
595         size_t size;
596         ssize_t ret;
597
598         attr = file->private_data;
599
600         if (!attr->set)
601                 return -EACCES;
602
603         mutex_lock(&attr->mutex);
604         ret = -EFAULT;
605         size = min(sizeof(attr->set_buf) - 1, len);
606         if (copy_from_user(attr->set_buf, buf, size))
607                 goto out;
608
609         ret = len; /* claim we got the whole input */
610         attr->set_buf[size] = '\0';
611         val = simple_strtol(attr->set_buf, NULL, 0);
612         attr->set(attr->data, val);
613 out:
614         mutex_unlock(&attr->mutex);
615         return ret;
616 }
617
618 EXPORT_SYMBOL(dcache_dir_close);
619 EXPORT_SYMBOL(dcache_dir_lseek);
620 EXPORT_SYMBOL(dcache_dir_open);
621 EXPORT_SYMBOL(dcache_readdir);
622 EXPORT_SYMBOL(generic_read_dir);
623 EXPORT_SYMBOL(get_sb_pseudo);
624 EXPORT_SYMBOL(simple_commit_write);
625 EXPORT_SYMBOL(simple_dir_inode_operations);
626 EXPORT_SYMBOL(simple_dir_operations);
627 EXPORT_SYMBOL(simple_empty);
628 EXPORT_SYMBOL(d_alloc_name);
629 EXPORT_SYMBOL(simple_fill_super);
630 EXPORT_SYMBOL(simple_getattr);
631 EXPORT_SYMBOL(simple_link);
632 EXPORT_SYMBOL(simple_lookup);
633 EXPORT_SYMBOL(simple_pin_fs);
634 EXPORT_SYMBOL(simple_prepare_write);
635 EXPORT_SYMBOL(simple_readpage);
636 EXPORT_SYMBOL(simple_release_fs);
637 EXPORT_SYMBOL(simple_rename);
638 EXPORT_SYMBOL(simple_rmdir);
639 EXPORT_SYMBOL(simple_statfs);
640 EXPORT_SYMBOL(simple_sync_file);
641 EXPORT_SYMBOL(simple_unlink);
642 EXPORT_SYMBOL(simple_read_from_buffer);
643 EXPORT_SYMBOL(simple_transaction_get);
644 EXPORT_SYMBOL(simple_transaction_read);
645 EXPORT_SYMBOL(simple_transaction_release);
646 EXPORT_SYMBOL_GPL(simple_attr_open);
647 EXPORT_SYMBOL_GPL(simple_attr_close);
648 EXPORT_SYMBOL_GPL(simple_attr_read);
649 EXPORT_SYMBOL_GPL(simple_attr_write);