hugetlb: fix hugepage allocation with memoryless nodes
[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_write_begin(struct file *file, struct address_space *mapping,
355                         loff_t pos, unsigned len, unsigned flags,
356                         struct page **pagep, void **fsdata)
357 {
358         struct page *page;
359         pgoff_t index;
360         unsigned from;
361
362         index = pos >> PAGE_CACHE_SHIFT;
363         from = pos & (PAGE_CACHE_SIZE - 1);
364
365         page = __grab_cache_page(mapping, index);
366         if (!page)
367                 return -ENOMEM;
368
369         *pagep = page;
370
371         return simple_prepare_write(file, page, from, from+len);
372 }
373
374 int simple_commit_write(struct file *file, struct page *page,
375                         unsigned from, unsigned to)
376 {
377         struct inode *inode = page->mapping->host;
378         loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
379
380         if (!PageUptodate(page))
381                 SetPageUptodate(page);
382         /*
383          * No need to use i_size_read() here, the i_size
384          * cannot change under us because we hold the i_mutex.
385          */
386         if (pos > inode->i_size)
387                 i_size_write(inode, pos);
388         set_page_dirty(page);
389         return 0;
390 }
391
392 int simple_write_end(struct file *file, struct address_space *mapping,
393                         loff_t pos, unsigned len, unsigned copied,
394                         struct page *page, void *fsdata)
395 {
396         unsigned from = pos & (PAGE_CACHE_SIZE - 1);
397
398         /* zero the stale part of the page if we did a short copy */
399         if (copied < len) {
400                 void *kaddr = kmap_atomic(page, KM_USER0);
401                 memset(kaddr + from + copied, 0, len - copied);
402                 flush_dcache_page(page);
403                 kunmap_atomic(kaddr, KM_USER0);
404         }
405
406         simple_commit_write(file, page, from, from+copied);
407
408         unlock_page(page);
409         page_cache_release(page);
410
411         return copied;
412 }
413
414 /*
415  * the inodes created here are not hashed. If you use iunique to generate
416  * unique inode values later for this filesystem, then you must take care
417  * to pass it an appropriate max_reserved value to avoid collisions.
418  */
419 int simple_fill_super(struct super_block *s, int magic, struct tree_descr *files)
420 {
421         struct inode *inode;
422         struct dentry *root;
423         struct dentry *dentry;
424         int i;
425
426         s->s_blocksize = PAGE_CACHE_SIZE;
427         s->s_blocksize_bits = PAGE_CACHE_SHIFT;
428         s->s_magic = magic;
429         s->s_op = &simple_super_operations;
430         s->s_time_gran = 1;
431
432         inode = new_inode(s);
433         if (!inode)
434                 return -ENOMEM;
435         /*
436          * because the root inode is 1, the files array must not contain an
437          * entry at index 1
438          */
439         inode->i_ino = 1;
440         inode->i_mode = S_IFDIR | 0755;
441         inode->i_uid = inode->i_gid = 0;
442         inode->i_blocks = 0;
443         inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
444         inode->i_op = &simple_dir_inode_operations;
445         inode->i_fop = &simple_dir_operations;
446         inode->i_nlink = 2;
447         root = d_alloc_root(inode);
448         if (!root) {
449                 iput(inode);
450                 return -ENOMEM;
451         }
452         for (i = 0; !files->name || files->name[0]; i++, files++) {
453                 if (!files->name)
454                         continue;
455
456                 /* warn if it tries to conflict with the root inode */
457                 if (unlikely(i == 1))
458                         printk(KERN_WARNING "%s: %s passed in a files array"
459                                 "with an index of 1!\n", __func__,
460                                 s->s_type->name);
461
462                 dentry = d_alloc_name(root, files->name);
463                 if (!dentry)
464                         goto out;
465                 inode = new_inode(s);
466                 if (!inode)
467                         goto out;
468                 inode->i_mode = S_IFREG | files->mode;
469                 inode->i_uid = inode->i_gid = 0;
470                 inode->i_blocks = 0;
471                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
472                 inode->i_fop = files->ops;
473                 inode->i_ino = i;
474                 d_add(dentry, inode);
475         }
476         s->s_root = root;
477         return 0;
478 out:
479         d_genocide(root);
480         dput(root);
481         return -ENOMEM;
482 }
483
484 static DEFINE_SPINLOCK(pin_fs_lock);
485
486 int simple_pin_fs(struct file_system_type *type, struct vfsmount **mount, int *count)
487 {
488         struct vfsmount *mnt = NULL;
489         spin_lock(&pin_fs_lock);
490         if (unlikely(!*mount)) {
491                 spin_unlock(&pin_fs_lock);
492                 mnt = vfs_kern_mount(type, 0, type->name, NULL);
493                 if (IS_ERR(mnt))
494                         return PTR_ERR(mnt);
495                 spin_lock(&pin_fs_lock);
496                 if (!*mount)
497                         *mount = mnt;
498         }
499         mntget(*mount);
500         ++*count;
501         spin_unlock(&pin_fs_lock);
502         mntput(mnt);
503         return 0;
504 }
505
506 void simple_release_fs(struct vfsmount **mount, int *count)
507 {
508         struct vfsmount *mnt;
509         spin_lock(&pin_fs_lock);
510         mnt = *mount;
511         if (!--*count)
512                 *mount = NULL;
513         spin_unlock(&pin_fs_lock);
514         mntput(mnt);
515 }
516
517 ssize_t simple_read_from_buffer(void __user *to, size_t count, loff_t *ppos,
518                                 const void *from, size_t available)
519 {
520         loff_t pos = *ppos;
521         if (pos < 0)
522                 return -EINVAL;
523         if (pos >= available)
524                 return 0;
525         if (count > available - pos)
526                 count = available - pos;
527         if (copy_to_user(to, from + pos, count))
528                 return -EFAULT;
529         *ppos = pos + count;
530         return count;
531 }
532
533 /*
534  * Transaction based IO.
535  * The file expects a single write which triggers the transaction, and then
536  * possibly a read which collects the result - which is stored in a
537  * file-local buffer.
538  */
539 char *simple_transaction_get(struct file *file, const char __user *buf, size_t size)
540 {
541         struct simple_transaction_argresp *ar;
542         static DEFINE_SPINLOCK(simple_transaction_lock);
543
544         if (size > SIMPLE_TRANSACTION_LIMIT - 1)
545                 return ERR_PTR(-EFBIG);
546
547         ar = (struct simple_transaction_argresp *)get_zeroed_page(GFP_KERNEL);
548         if (!ar)
549                 return ERR_PTR(-ENOMEM);
550
551         spin_lock(&simple_transaction_lock);
552
553         /* only one write allowed per open */
554         if (file->private_data) {
555                 spin_unlock(&simple_transaction_lock);
556                 free_page((unsigned long)ar);
557                 return ERR_PTR(-EBUSY);
558         }
559
560         file->private_data = ar;
561
562         spin_unlock(&simple_transaction_lock);
563
564         if (copy_from_user(ar->data, buf, size))
565                 return ERR_PTR(-EFAULT);
566
567         return ar->data;
568 }
569
570 ssize_t simple_transaction_read(struct file *file, char __user *buf, size_t size, loff_t *pos)
571 {
572         struct simple_transaction_argresp *ar = file->private_data;
573
574         if (!ar)
575                 return 0;
576         return simple_read_from_buffer(buf, size, pos, ar->data, ar->size);
577 }
578
579 int simple_transaction_release(struct inode *inode, struct file *file)
580 {
581         free_page((unsigned long)file->private_data);
582         return 0;
583 }
584
585 /* Simple attribute files */
586
587 struct simple_attr {
588         u64 (*get)(void *);
589         void (*set)(void *, u64);
590         char get_buf[24];       /* enough to store a u64 and "\n\0" */
591         char set_buf[24];
592         void *data;
593         const char *fmt;        /* format for read operation */
594         struct mutex mutex;     /* protects access to these buffers */
595 };
596
597 /* simple_attr_open is called by an actual attribute open file operation
598  * to set the attribute specific access operations. */
599 int simple_attr_open(struct inode *inode, struct file *file,
600                      u64 (*get)(void *), void (*set)(void *, u64),
601                      const char *fmt)
602 {
603         struct simple_attr *attr;
604
605         attr = kmalloc(sizeof(*attr), GFP_KERNEL);
606         if (!attr)
607                 return -ENOMEM;
608
609         attr->get = get;
610         attr->set = set;
611         attr->data = inode->i_private;
612         attr->fmt = fmt;
613         mutex_init(&attr->mutex);
614
615         file->private_data = attr;
616
617         return nonseekable_open(inode, file);
618 }
619
620 int simple_attr_close(struct inode *inode, struct file *file)
621 {
622         kfree(file->private_data);
623         return 0;
624 }
625
626 /* read from the buffer that is filled with the get function */
627 ssize_t simple_attr_read(struct file *file, char __user *buf,
628                          size_t len, loff_t *ppos)
629 {
630         struct simple_attr *attr;
631         size_t size;
632         ssize_t ret;
633
634         attr = file->private_data;
635
636         if (!attr->get)
637                 return -EACCES;
638
639         mutex_lock(&attr->mutex);
640         if (*ppos) /* continued read */
641                 size = strlen(attr->get_buf);
642         else      /* first read */
643                 size = scnprintf(attr->get_buf, sizeof(attr->get_buf),
644                                  attr->fmt,
645                                  (unsigned long long)attr->get(attr->data));
646
647         ret = simple_read_from_buffer(buf, len, ppos, attr->get_buf, size);
648         mutex_unlock(&attr->mutex);
649         return ret;
650 }
651
652 /* interpret the buffer as a number to call the set function with */
653 ssize_t simple_attr_write(struct file *file, const char __user *buf,
654                           size_t len, loff_t *ppos)
655 {
656         struct simple_attr *attr;
657         u64 val;
658         size_t size;
659         ssize_t ret;
660
661         attr = file->private_data;
662
663         if (!attr->set)
664                 return -EACCES;
665
666         mutex_lock(&attr->mutex);
667         ret = -EFAULT;
668         size = min(sizeof(attr->set_buf) - 1, len);
669         if (copy_from_user(attr->set_buf, buf, size))
670                 goto out;
671
672         ret = len; /* claim we got the whole input */
673         attr->set_buf[size] = '\0';
674         val = simple_strtol(attr->set_buf, NULL, 0);
675         attr->set(attr->data, val);
676 out:
677         mutex_unlock(&attr->mutex);
678         return ret;
679 }
680
681 EXPORT_SYMBOL(dcache_dir_close);
682 EXPORT_SYMBOL(dcache_dir_lseek);
683 EXPORT_SYMBOL(dcache_dir_open);
684 EXPORT_SYMBOL(dcache_readdir);
685 EXPORT_SYMBOL(generic_read_dir);
686 EXPORT_SYMBOL(get_sb_pseudo);
687 EXPORT_SYMBOL(simple_write_begin);
688 EXPORT_SYMBOL(simple_write_end);
689 EXPORT_SYMBOL(simple_commit_write);
690 EXPORT_SYMBOL(simple_dir_inode_operations);
691 EXPORT_SYMBOL(simple_dir_operations);
692 EXPORT_SYMBOL(simple_empty);
693 EXPORT_SYMBOL(d_alloc_name);
694 EXPORT_SYMBOL(simple_fill_super);
695 EXPORT_SYMBOL(simple_getattr);
696 EXPORT_SYMBOL(simple_link);
697 EXPORT_SYMBOL(simple_lookup);
698 EXPORT_SYMBOL(simple_pin_fs);
699 EXPORT_SYMBOL(simple_prepare_write);
700 EXPORT_SYMBOL(simple_readpage);
701 EXPORT_SYMBOL(simple_release_fs);
702 EXPORT_SYMBOL(simple_rename);
703 EXPORT_SYMBOL(simple_rmdir);
704 EXPORT_SYMBOL(simple_statfs);
705 EXPORT_SYMBOL(simple_sync_file);
706 EXPORT_SYMBOL(simple_unlink);
707 EXPORT_SYMBOL(simple_read_from_buffer);
708 EXPORT_SYMBOL(simple_transaction_get);
709 EXPORT_SYMBOL(simple_transaction_read);
710 EXPORT_SYMBOL(simple_transaction_release);
711 EXPORT_SYMBOL_GPL(simple_attr_open);
712 EXPORT_SYMBOL_GPL(simple_attr_close);
713 EXPORT_SYMBOL_GPL(simple_attr_read);
714 EXPORT_SYMBOL_GPL(simple_attr_write);