?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
sysctl: Remove dead register_sysctl_root
[cris-mirror.git] / fs / proc / proc_sysctl.c
blob35eed95b26d57e3d09cc7947b8ab4203c9889b4e
1 /*
2 * /proc/sys support
3 */
4 #include <linux/init.h>
5 #include <linux/sysctl.h>
6 #include <linux/poll.h>
7 #include <linux/proc_fs.h>
8 #include <linux/printk.h>
9 #include <linux/security.h>
10 #include <linux/sched.h>
11 #include <linux/cred.h>
12 #include <linux/namei.h>
13 #include <linux/mm.h>
14 #include <linux/module.h>
15 #include "internal.h"
16
17 static const struct dentry_operations proc_sys_dentry_operations;
18 static const struct file_operations proc_sys_file_operations;
19 static const struct inode_operations proc_sys_inode_operations;
20 static const struct file_operations proc_sys_dir_file_operations;
21 static const struct inode_operations proc_sys_dir_operations;
22
23 /* Support for permanently empty directories */
24
25 struct ctl_table sysctl_mount_point[] = {
26 { }
27 };
28
29 static bool is_empty_dir(struct ctl_table_header *head)
30 {
31 return head->ctl_table[0].child == sysctl_mount_point;
32 }
33
34 static void set_empty_dir(struct ctl_dir *dir)
35 {
36 dir->header.ctl_table[0].child = sysctl_mount_point;
37 }
38
39 static void clear_empty_dir(struct ctl_dir *dir)
40
41 {
42 dir->header.ctl_table[0].child = NULL;
43 }
44
45 void proc_sys_poll_notify(struct ctl_table_poll *poll)
46 {
47 if (!poll)
48 return;
49
50 atomic_inc(&poll->event);
51 wake_up_interruptible(&poll->wait);
52 }
53
54 static struct ctl_table root_table[] = {
55 {
56 .procname = "",
57 .mode = S_IFDIR|S_IRUGO|S_IXUGO,
58 },
59 { }
60 };
61 static struct ctl_table_root sysctl_table_root = {
62 .default_set.dir.header = {
63 {{.count = 1,
64 .nreg = 1,
65 .ctl_table = root_table }},
66 .ctl_table_arg = root_table,
67 .root = &sysctl_table_root,
68 .set = &sysctl_table_root.default_set,
69 },
70 };
71
72 static DEFINE_SPINLOCK(sysctl_lock);
73
74 static void drop_sysctl_table(struct ctl_table_header *header);
75 static int sysctl_follow_link(struct ctl_table_header **phead,
76 struct ctl_table **pentry);
77 static int insert_links(struct ctl_table_header *head);
78 static void put_links(struct ctl_table_header *header);
79
80 static void sysctl_print_dir(struct ctl_dir *dir)
81 {
82 if (dir->header.parent)
83 sysctl_print_dir(dir->header.parent);
84 pr_cont("%s/", dir->header.ctl_table[0].procname);
85 }
86
87 static int namecmp(const char *name1, int len1, const char *name2, int len2)
88 {
89 int minlen;
90 int cmp;
91
92 minlen = len1;
93 if (minlen > len2)
94 minlen = len2;
95
96 cmp = memcmp(name1, name2, minlen);
97 if (cmp == 0)
98 cmp = len1 - len2;
99 return cmp;
100 }
101
102 /* Called under sysctl_lock */
103 static struct ctl_table *find_entry(struct ctl_table_header **phead,
104 struct ctl_dir *dir, const char *name, int namelen)
105 {
106 struct ctl_table_header *head;
107 struct ctl_table *entry;
108 struct rb_node *node = dir->root.rb_node;
109
110 while (node)
111 {
112 struct ctl_node *ctl_node;
113 const char *procname;
114 int cmp;
115
116 ctl_node = rb_entry(node, struct ctl_node, node);
117 head = ctl_node->header;
118 entry = &head->ctl_table[ctl_node - head->node];
119 procname = entry->procname;
120
121 cmp = namecmp(name, namelen, procname, strlen(procname));
122 if (cmp < 0)
123 node = node->rb_left;
124 else if (cmp > 0)
125 node = node->rb_right;
126 else {
127 *phead = head;
128 return entry;
129 }
130 }
131 return NULL;
132 }
133
134 static int insert_entry(struct ctl_table_header *head, struct ctl_table *entry)
135 {
136 struct rb_node *node = &head->node[entry - head->ctl_table].node;
137 struct rb_node **p = &head->parent->root.rb_node;
138 struct rb_node *parent = NULL;
139 const char *name = entry->procname;
140 int namelen = strlen(name);
141
142 while (*p) {
143 struct ctl_table_header *parent_head;
144 struct ctl_table *parent_entry;
145 struct ctl_node *parent_node;
146 const char *parent_name;
147 int cmp;
148
149 parent = *p;
150 parent_node = rb_entry(parent, struct ctl_node, node);
151 parent_head = parent_node->header;
152 parent_entry = &parent_head->ctl_table[parent_node - parent_head->node];
153 parent_name = parent_entry->procname;
154
155 cmp = namecmp(name, namelen, parent_name, strlen(parent_name));
156 if (cmp < 0)
157 p = &(*p)->rb_left;
158 else if (cmp > 0)
159 p = &(*p)->rb_right;
160 else {
161 pr_err("sysctl duplicate entry: ");
162 sysctl_print_dir(head->parent);
163 pr_cont("/%s\n", entry->procname);
164 return -EEXIST;
165 }
166 }
167
168 rb_link_node(node, parent, p);
169 rb_insert_color(node, &head->parent->root);
170 return 0;
171 }
172
173 static void erase_entry(struct ctl_table_header *head, struct ctl_table *entry)
174 {
175 struct rb_node *node = &head->node[entry - head->ctl_table].node;
176
177 rb_erase(node, &head->parent->root);
178 }
179
180 static void init_header(struct ctl_table_header *head,
181 struct ctl_table_root *root, struct ctl_table_set *set,
182 struct ctl_node *node, struct ctl_table *table)
183 {
184 head->ctl_table = table;
185 head->ctl_table_arg = table;
186 head->used = 0;
187 head->count = 1;
188 head->nreg = 1;
189 head->unregistering = NULL;
190 head->root = root;
191 head->set = set;
192 head->parent = NULL;
193 head->node = node;
194 INIT_LIST_HEAD(&head->inodes);
195 if (node) {
196 struct ctl_table *entry;
197 for (entry = table; entry->procname; entry++, node++)
198 node->header = head;
199 }
200 }
201
202 static void erase_header(struct ctl_table_header *head)
203 {
204 struct ctl_table *entry;
205 for (entry = head->ctl_table; entry->procname; entry++)
206 erase_entry(head, entry);
207 }
208
209 static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
210 {
211 struct ctl_table *entry;
212 int err;
213
214 /* Is this a permanently empty directory? */
215 if (is_empty_dir(&dir->header))
216 return -EROFS;
217
218 /* Am I creating a permanently empty directory? */
219 if (header->ctl_table == sysctl_mount_point) {
220 if (!RB_EMPTY_ROOT(&dir->root))
221 return -EINVAL;
222 set_empty_dir(dir);
223 }
224
225 dir->header.nreg++;
226 header->parent = dir;
227 err = insert_links(header);
228 if (err)
229 goto fail_links;
230 for (entry = header->ctl_table; entry->procname; entry++) {
231 err = insert_entry(header, entry);
232 if (err)
233 goto fail;
234 }
235 return 0;
236 fail:
237 erase_header(header);
238 put_links(header);
239 fail_links:
240 if (header->ctl_table == sysctl_mount_point)
241 clear_empty_dir(dir);
242 header->parent = NULL;
243 drop_sysctl_table(&dir->header);
244 return err;
245 }
246
247 /* called under sysctl_lock */
248 static int use_table(struct ctl_table_header *p)
249 {
250 if (unlikely(p->unregistering))
251 return 0;
252 p->used++;
253 return 1;
254 }
255
256 /* called under sysctl_lock */
257 static void unuse_table(struct ctl_table_header *p)
258 {
259 if (!--p->used)
260 if (unlikely(p->unregistering))
261 complete(p->unregistering);
262 }
263
264 /* called under sysctl_lock */
265 static void proc_sys_prune_dcache(struct ctl_table_header *head)
266 {
267 struct inode *inode, *prev = NULL;
268 struct proc_inode *ei;
269
270 rcu_read_lock();
271 list_for_each_entry_rcu(ei, &head->inodes, sysctl_inodes) {
272 inode = igrab(&ei->vfs_inode);
273 if (inode) {
274 rcu_read_unlock();
275 iput(prev);
276 prev = inode;
277 d_prune_aliases(inode);
278 rcu_read_lock();
279 }
280 }
281 rcu_read_unlock();
282 iput(prev);
283 }
284
285 /* called under sysctl_lock, will reacquire if has to wait */
286 static void start_unregistering(struct ctl_table_header *p)
287 {
288 /*
289 * if p->used is 0, nobody will ever touch that entry again;
290 * we'll eliminate all paths to it before dropping sysctl_lock
291 */
292 if (unlikely(p->used)) {
293 struct completion wait;
294 init_completion(&wait);
295 p->unregistering = &wait;
296 spin_unlock(&sysctl_lock);
297 wait_for_completion(&wait);
298 } else {
299 /* anything non-NULL; we'll never dereference it */
300 p->unregistering = ERR_PTR(-EINVAL);
301 spin_unlock(&sysctl_lock);
302 }
303 /*
304 * Prune dentries for unregistered sysctls: namespaced sysctls
305 * can have duplicate names and contaminate dcache very badly.
306 */
307 proc_sys_prune_dcache(p);
308 /*
309 * do not remove from the list until nobody holds it; walking the
310 * list in do_sysctl() relies on that.
311 */
312 spin_lock(&sysctl_lock);
313 erase_header(p);
314 }
315
316 static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
317 {
318 BUG_ON(!head);
319 spin_lock(&sysctl_lock);
320 if (!use_table(head))
321 head = ERR_PTR(-ENOENT);
322 spin_unlock(&sysctl_lock);
323 return head;
324 }
325
326 static void sysctl_head_finish(struct ctl_table_header *head)
327 {
328 if (!head)
329 return;
330 spin_lock(&sysctl_lock);
331 unuse_table(head);
332 spin_unlock(&sysctl_lock);
333 }
334
335 static struct ctl_table_set *
336 lookup_header_set(struct ctl_table_root *root)
337 {
338 struct ctl_table_set *set = &root->default_set;
339 if (root->lookup)
340 set = root->lookup(root);
341 return set;
342 }
343
344 static struct ctl_table *lookup_entry(struct ctl_table_header **phead,
345 struct ctl_dir *dir,
346 const char *name, int namelen)
347 {
348 struct ctl_table_header *head;
349 struct ctl_table *entry;
350
351 spin_lock(&sysctl_lock);
352 entry = find_entry(&head, dir, name, namelen);
353 if (entry && use_table(head))
354 *phead = head;
355 else
356 entry = NULL;
357 spin_unlock(&sysctl_lock);
358 return entry;
359 }
360
361 static struct ctl_node *first_usable_entry(struct rb_node *node)
362 {
363 struct ctl_node *ctl_node;
364
365 for (;node; node = rb_next(node)) {
366 ctl_node = rb_entry(node, struct ctl_node, node);
367 if (use_table(ctl_node->header))
368 return ctl_node;
369 }
370 return NULL;
371 }
372
373 static void first_entry(struct ctl_dir *dir,
374 struct ctl_table_header **phead, struct ctl_table **pentry)
375 {
376 struct ctl_table_header *head = NULL;
377 struct ctl_table *entry = NULL;
378 struct ctl_node *ctl_node;
379
380 spin_lock(&sysctl_lock);
381 ctl_node = first_usable_entry(rb_first(&dir->root));
382 spin_unlock(&sysctl_lock);
383 if (ctl_node) {
384 head = ctl_node->header;
385 entry = &head->ctl_table[ctl_node - head->node];
386 }
387 *phead = head;
388 *pentry = entry;
389 }
390
391 static void next_entry(struct ctl_table_header **phead, struct ctl_table **pentry)
392 {
393 struct ctl_table_header *head = *phead;
394 struct ctl_table *entry = *pentry;
395 struct ctl_node *ctl_node = &head->node[entry - head->ctl_table];
396
397 spin_lock(&sysctl_lock);
398 unuse_table(head);
399
400 ctl_node = first_usable_entry(rb_next(&ctl_node->node));
401 spin_unlock(&sysctl_lock);
402 head = NULL;
403 if (ctl_node) {
404 head = ctl_node->header;
405 entry = &head->ctl_table[ctl_node - head->node];
406 }
407 *phead = head;
408 *pentry = entry;
409 }
410
411 /*
412 * sysctl_perm does NOT grant the superuser all rights automatically, because
413 * some sysctl variables are readonly even to root.
414 */
415
416 static int test_perm(int mode, int op)
417 {
418 if (uid_eq(current_euid(), GLOBAL_ROOT_UID))
419 mode >>= 6;
420 else if (in_egroup_p(GLOBAL_ROOT_GID))
421 mode >>= 3;
422 if ((op & ~mode & (MAY_READ|MAY_WRITE|MAY_EXEC)) == 0)
423 return 0;
424 return -EACCES;
425 }
426
427 static int sysctl_perm(struct ctl_table_header *head, struct ctl_table *table, int op)
428 {
429 struct ctl_table_root *root = head->root;
430 int mode;
431
432 if (root->permissions)
433 mode = root->permissions(head, table);
434 else
435 mode = table->mode;
436
437 return test_perm(mode, op);
438 }
439
440 static struct inode *proc_sys_make_inode(struct super_block *sb,
441 struct ctl_table_header *head, struct ctl_table *table)
442 {
443 struct ctl_table_root *root = head->root;
444 struct inode *inode;
445 struct proc_inode *ei;
446
447 inode = new_inode(sb);
448 if (!inode)
449 goto out;
450
451 inode->i_ino = get_next_ino();
452
453 ei = PROC_I(inode);
454
455 spin_lock(&sysctl_lock);
456 if (unlikely(head->unregistering)) {
457 spin_unlock(&sysctl_lock);
458 iput(inode);
459 inode = NULL;
460 goto out;
461 }
462 ei->sysctl = head;
463 ei->sysctl_entry = table;
464 list_add_rcu(&ei->sysctl_inodes, &head->inodes);
465 head->count++;
466 spin_unlock(&sysctl_lock);
467
468 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
469 inode->i_mode = table->mode;
470 if (!S_ISDIR(table->mode)) {
471 inode->i_mode |= S_IFREG;
472 inode->i_op = &proc_sys_inode_operations;
473 inode->i_fop = &proc_sys_file_operations;
474 } else {
475 inode->i_mode |= S_IFDIR;
476 inode->i_op = &proc_sys_dir_operations;
477 inode->i_fop = &proc_sys_dir_file_operations;
478 if (is_empty_dir(head))
479 make_empty_dir_inode(inode);
480 }
481
482 if (root->set_ownership)
483 root->set_ownership(head, table, &inode->i_uid, &inode->i_gid);
484
485 out:
486 return inode;
487 }
488
489 void proc_sys_evict_inode(struct inode *inode, struct ctl_table_header *head)
490 {
491 spin_lock(&sysctl_lock);
492 list_del_rcu(&PROC_I(inode)->sysctl_inodes);
493 if (!--head->count)
494 kfree_rcu(head, rcu);
495 spin_unlock(&sysctl_lock);
496 }
497
498 static struct ctl_table_header *grab_header(struct inode *inode)
499 {
500 struct ctl_table_header *head = PROC_I(inode)->sysctl;
501 if (!head)
502 head = &sysctl_table_root.default_set.dir.header;
503 return sysctl_head_grab(head);
504 }
505
506 static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
507 unsigned int flags)
508 {
509 struct ctl_table_header *head = grab_header(dir);
510 struct ctl_table_header *h = NULL;
511 const struct qstr *name = &dentry->d_name;
512 struct ctl_table *p;
513 struct inode *inode;
514 struct dentry *err = ERR_PTR(-ENOENT);
515 struct ctl_dir *ctl_dir;
516 int ret;
517
518 if (IS_ERR(head))
519 return ERR_CAST(head);
520
521 ctl_dir = container_of(head, struct ctl_dir, header);
522
523 p = lookup_entry(&h, ctl_dir, name->name, name->len);
524 if (!p)
525 goto out;
526
527 if (S_ISLNK(p->mode)) {
528 ret = sysctl_follow_link(&h, &p);
529 err = ERR_PTR(ret);
530 if (ret)
531 goto out;
532 }
533
534 err = ERR_PTR(-ENOMEM);
535 inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
536 if (!inode)
537 goto out;
538
539 err = NULL;
540 d_set_d_op(dentry, &proc_sys_dentry_operations);
541 d_add(dentry, inode);
542
543 out:
544 if (h)
545 sysctl_head_finish(h);
546 sysctl_head_finish(head);
547 return err;
548 }
549
550 static ssize_t proc_sys_call_handler(struct file *filp, void __user *buf,
551 size_t count, loff_t *ppos, int write)
552 {
553 struct inode *inode = file_inode(filp);
554 struct ctl_table_header *head = grab_header(inode);
555 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
556 ssize_t error;
557 size_t res;
558
559 if (IS_ERR(head))
560 return PTR_ERR(head);
561
562 /*
563 * At this point we know that the sysctl was not unregistered
564 * and won't be until we finish.
565 */
566 error = -EPERM;
567 if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
568 goto out;
569
570 /* if that can happen at all, it should be -EINVAL, not -EISDIR */
571 error = -EINVAL;
572 if (!table->proc_handler)
573 goto out;
574
575 /* careful: calling conventions are nasty here */
576 res = count;
577 error = table->proc_handler(table, write, buf, &res, ppos);
578 if (!error)
579 error = res;
580 out:
581 sysctl_head_finish(head);
582
583 return error;
584 }
585
586 static ssize_t proc_sys_read(struct file *filp, char __user *buf,
587 size_t count, loff_t *ppos)
588 {
589 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 0);
590 }
591
592 static ssize_t proc_sys_write(struct file *filp, const char __user *buf,
593 size_t count, loff_t *ppos)
594 {
595 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 1);
596 }
597
598 static int proc_sys_open(struct inode *inode, struct file *filp)
599 {
600 struct ctl_table_header *head = grab_header(inode);
601 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
602
603 /* sysctl was unregistered */
604 if (IS_ERR(head))
605 return PTR_ERR(head);
606
607 if (table->poll)
608 filp->private_data = proc_sys_poll_event(table->poll);
609
610 sysctl_head_finish(head);
611
612 return 0;
613 }
614
615 static unsigned int proc_sys_poll(struct file *filp, poll_table *wait)
616 {
617 struct inode *inode = file_inode(filp);
618 struct ctl_table_header *head = grab_header(inode);
619 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
620 unsigned int ret = DEFAULT_POLLMASK;
621 unsigned long event;
622
623 /* sysctl was unregistered */
624 if (IS_ERR(head))
625 return POLLERR | POLLHUP;
626
627 if (!table->proc_handler)
628 goto out;
629
630 if (!table->poll)
631 goto out;
632
633 event = (unsigned long)filp->private_data;
634 poll_wait(filp, &table->poll->wait, wait);
635
636 if (event != atomic_read(&table->poll->event)) {
637 filp->private_data = proc_sys_poll_event(table->poll);
638 ret = POLLIN | POLLRDNORM | POLLERR | POLLPRI;
639 }
640
641 out:
642 sysctl_head_finish(head);
643
644 return ret;
645 }
646
647 static bool proc_sys_fill_cache(struct file *file,
648 struct dir_context *ctx,
649 struct ctl_table_header *head,
650 struct ctl_table *table)
651 {
652 struct dentry *child, *dir = file->f_path.dentry;
653 struct inode *inode;
654 struct qstr qname;
655 ino_t ino = 0;
656 unsigned type = DT_UNKNOWN;
657
658 qname.name = table->procname;
659 qname.len = strlen(table->procname);
660 qname.hash = full_name_hash(dir, qname.name, qname.len);
661
662 child = d_lookup(dir, &qname);
663 if (!child) {
664 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
665 child = d_alloc_parallel(dir, &qname, &wq);
666 if (IS_ERR(child))
667 return false;
668 if (d_in_lookup(child)) {
669 inode = proc_sys_make_inode(dir->d_sb, head, table);
670 if (!inode) {
671 d_lookup_done(child);
672 dput(child);
673 return false;
674 }
675 d_set_d_op(child, &proc_sys_dentry_operations);
676 d_add(child, inode);
677 }
678 }
679 inode = d_inode(child);
680 ino = inode->i_ino;
681 type = inode->i_mode >> 12;
682 dput(child);
683 return dir_emit(ctx, qname.name, qname.len, ino, type);
684 }
685
686 static bool proc_sys_link_fill_cache(struct file *file,
687 struct dir_context *ctx,
688 struct ctl_table_header *head,
689 struct ctl_table *table)
690 {
691 bool ret = true;
692 head = sysctl_head_grab(head);
693
694 if (S_ISLNK(table->mode)) {
695 /* It is not an error if we can not follow the link ignore it */
696 int err = sysctl_follow_link(&head, &table);
697 if (err)
698 goto out;
699 }
700
701 ret = proc_sys_fill_cache(file, ctx, head, table);
702 out:
703 sysctl_head_finish(head);
704 return ret;
705 }
706
707 static int scan(struct ctl_table_header *head, struct ctl_table *table,
708 unsigned long *pos, struct file *file,
709 struct dir_context *ctx)
710 {
711 bool res;
712
713 if ((*pos)++ < ctx->pos)
714 return true;
715
716 if (unlikely(S_ISLNK(table->mode)))
717 res = proc_sys_link_fill_cache(file, ctx, head, table);
718 else
719 res = proc_sys_fill_cache(file, ctx, head, table);
720
721 if (res)
722 ctx->pos = *pos;
723
724 return res;
725 }
726
727 static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
728 {
729 struct ctl_table_header *head = grab_header(file_inode(file));
730 struct ctl_table_header *h = NULL;
731 struct ctl_table *entry;
732 struct ctl_dir *ctl_dir;
733 unsigned long pos;
734
735 if (IS_ERR(head))
736 return PTR_ERR(head);
737
738 ctl_dir = container_of(head, struct ctl_dir, header);
739
740 if (!dir_emit_dots(file, ctx))
741 goto out;
742
743 pos = 2;
744
745 for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
746 if (!scan(h, entry, &pos, file, ctx)) {
747 sysctl_head_finish(h);
748 break;
749 }
750 }
751 out:
752 sysctl_head_finish(head);
753 return 0;
754 }
755
756 static int proc_sys_permission(struct inode *inode, int mask)
757 {
758 /*
759 * sysctl entries that are not writeable,
760 * are _NOT_ writeable, capabilities or not.
761 */
762 struct ctl_table_header *head;
763 struct ctl_table *table;
764 int error;
765
766 /* Executable files are not allowed under /proc/sys/ */
767 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
768 return -EACCES;
769
770 head = grab_header(inode);
771 if (IS_ERR(head))
772 return PTR_ERR(head);
773
774 table = PROC_I(inode)->sysctl_entry;
775 if (!table) /* global root - r-xr-xr-x */
776 error = mask & MAY_WRITE ? -EACCES : 0;
777 else /* Use the permissions on the sysctl table entry */
778 error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
779
780 sysctl_head_finish(head);
781 return error;
782 }
783
784 static int proc_sys_setattr(struct dentry *dentry, struct iattr *attr)
785 {
786 struct inode *inode = d_inode(dentry);
787 int error;
788
789 if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
790 return -EPERM;
791
792 error = setattr_prepare(dentry, attr);
793 if (error)
794 return error;
795
796 setattr_copy(inode, attr);
797 mark_inode_dirty(inode);
798 return 0;
799 }
800
801 static int proc_sys_getattr(const struct path *path, struct kstat *stat,
802 u32 request_mask, unsigned int query_flags)
803 {
804 struct inode *inode = d_inode(path->dentry);
805 struct ctl_table_header *head = grab_header(inode);
806 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
807
808 if (IS_ERR(head))
809 return PTR_ERR(head);
810
811 generic_fillattr(inode, stat);
812 if (table)
813 stat->mode = (stat->mode & S_IFMT) | table->mode;
814
815 sysctl_head_finish(head);
816 return 0;
817 }
818
819 static const struct file_operations proc_sys_file_operations = {
820 .open = proc_sys_open,
821 .poll = proc_sys_poll,
822 .read = proc_sys_read,
823 .write = proc_sys_write,
824 .llseek = default_llseek,
825 };
826
827 static const struct file_operations proc_sys_dir_file_operations = {
828 .read = generic_read_dir,
829 .iterate_shared = proc_sys_readdir,
830 .llseek = generic_file_llseek,
831 };
832
833 static const struct inode_operations proc_sys_inode_operations = {
834 .permission = proc_sys_permission,
835 .setattr = proc_sys_setattr,
836 .getattr = proc_sys_getattr,
837 };
838
839 static const struct inode_operations proc_sys_dir_operations = {
840 .lookup = proc_sys_lookup,
841 .permission = proc_sys_permission,
842 .setattr = proc_sys_setattr,
843 .getattr = proc_sys_getattr,
844 };
845
846 static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
847 {
848 if (flags & LOOKUP_RCU)
849 return -ECHILD;
850 return !PROC_I(d_inode(dentry))->sysctl->unregistering;
851 }
852
853 static int proc_sys_delete(const struct dentry *dentry)
854 {
855 return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
856 }
857
858 static int sysctl_is_seen(struct ctl_table_header *p)
859 {
860 struct ctl_table_set *set = p->set;
861 int res;
862 spin_lock(&sysctl_lock);
863 if (p->unregistering)
864 res = 0;
865 else if (!set->is_seen)
866 res = 1;
867 else
868 res = set->is_seen(set);
869 spin_unlock(&sysctl_lock);
870 return res;
871 }
872
873 static int proc_sys_compare(const struct dentry *dentry,
874 unsigned int len, const char *str, const struct qstr *name)
875 {
876 struct ctl_table_header *head;
877 struct inode *inode;
878
879 /* Although proc doesn't have negative dentries, rcu-walk means
880 * that inode here can be NULL */
881 /* AV: can it, indeed? */
882 inode = d_inode_rcu(dentry);
883 if (!inode)
884 return 1;
885 if (name->len != len)
886 return 1;
887 if (memcmp(name->name, str, len))
888 return 1;
889 head = rcu_dereference(PROC_I(inode)->sysctl);
890 return !head || !sysctl_is_seen(head);
891 }
892
893 static const struct dentry_operations proc_sys_dentry_operations = {
894 .d_revalidate = proc_sys_revalidate,
895 .d_delete = proc_sys_delete,
896 .d_compare = proc_sys_compare,
897 };
898
899 static struct ctl_dir *find_subdir(struct ctl_dir *dir,
900 const char *name, int namelen)
901 {
902 struct ctl_table_header *head;
903 struct ctl_table *entry;
904
905 entry = find_entry(&head, dir, name, namelen);
906 if (!entry)
907 return ERR_PTR(-ENOENT);
908 if (!S_ISDIR(entry->mode))
909 return ERR_PTR(-ENOTDIR);
910 return container_of(head, struct ctl_dir, header);
911 }
912
913 static struct ctl_dir *new_dir(struct ctl_table_set *set,
914 const char *name, int namelen)
915 {
916 struct ctl_table *table;
917 struct ctl_dir *new;
918 struct ctl_node *node;
919 char *new_name;
920
921 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
922 sizeof(struct ctl_table)*2 + namelen + 1,
923 GFP_KERNEL);
924 if (!new)
925 return NULL;
926
927 node = (struct ctl_node *)(new + 1);
928 table = (struct ctl_table *)(node + 1);
929 new_name = (char *)(table + 2);
930 memcpy(new_name, name, namelen);
931 new_name[namelen] = '\0';
932 table[0].procname = new_name;
933 table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
934 init_header(&new->header, set->dir.header.root, set, node, table);
935
936 return new;
937 }
938
939 /**
940 * get_subdir - find or create a subdir with the specified name.
941 * @dir: Directory to create the subdirectory in
942 * @name: The name of the subdirectory to find or create
943 * @namelen: The length of name
944 *
945 * Takes a directory with an elevated reference count so we know that
946 * if we drop the lock the directory will not go away. Upon success
947 * the reference is moved from @dir to the returned subdirectory.
948 * Upon error an error code is returned and the reference on @dir is
949 * simply dropped.
950 */
951 static struct ctl_dir *get_subdir(struct ctl_dir *dir,
952 const char *name, int namelen)
953 {
954 struct ctl_table_set *set = dir->header.set;
955 struct ctl_dir *subdir, *new = NULL;
956 int err;
957
958 spin_lock(&sysctl_lock);
959 subdir = find_subdir(dir, name, namelen);
960 if (!IS_ERR(subdir))
961 goto found;
962 if (PTR_ERR(subdir) != -ENOENT)
963 goto failed;
964
965 spin_unlock(&sysctl_lock);
966 new = new_dir(set, name, namelen);
967 spin_lock(&sysctl_lock);
968 subdir = ERR_PTR(-ENOMEM);
969 if (!new)
970 goto failed;
971
972 /* Was the subdir added while we dropped the lock? */
973 subdir = find_subdir(dir, name, namelen);
974 if (!IS_ERR(subdir))
975 goto found;
976 if (PTR_ERR(subdir) != -ENOENT)
977 goto failed;
978
979 /* Nope. Use the our freshly made directory entry. */
980 err = insert_header(dir, &new->header);
981 subdir = ERR_PTR(err);
982 if (err)
983 goto failed;
984 subdir = new;
985 found:
986 subdir->header.nreg++;
987 failed:
988 if (IS_ERR(subdir)) {
989 pr_err("sysctl could not get directory: ");
990 sysctl_print_dir(dir);
991 pr_cont("/%*.*s %ld\n",
992 namelen, namelen, name, PTR_ERR(subdir));
993 }
994 drop_sysctl_table(&dir->header);
995 if (new)
996 drop_sysctl_table(&new->header);
997 spin_unlock(&sysctl_lock);
998 return subdir;
999 }
1000
1001 static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
1002 {
1003 struct ctl_dir *parent;
1004 const char *procname;
1005 if (!dir->header.parent)
1006 return &set->dir;
1007 parent = xlate_dir(set, dir->header.parent);
1008 if (IS_ERR(parent))
1009 return parent;
1010 procname = dir->header.ctl_table[0].procname;
1011 return find_subdir(parent, procname, strlen(procname));
1012 }
1013
1014 static int sysctl_follow_link(struct ctl_table_header **phead,
1015 struct ctl_table **pentry)
1016 {
1017 struct ctl_table_header *head;
1018 struct ctl_table_root *root;
1019 struct ctl_table_set *set;
1020 struct ctl_table *entry;
1021 struct ctl_dir *dir;
1022 int ret;
1023
1024 ret = 0;
1025 spin_lock(&sysctl_lock);
1026 root = (*pentry)->data;
1027 set = lookup_header_set(root);
1028 dir = xlate_dir(set, (*phead)->parent);
1029 if (IS_ERR(dir))
1030 ret = PTR_ERR(dir);
1031 else {
1032 const char *procname = (*pentry)->procname;
1033 head = NULL;
1034 entry = find_entry(&head, dir, procname, strlen(procname));
1035 ret = -ENOENT;
1036 if (entry && use_table(head)) {
1037 unuse_table(*phead);
1038 *phead = head;
1039 *pentry = entry;
1040 ret = 0;
1041 }
1042 }
1043
1044 spin_unlock(&sysctl_lock);
1045 return ret;
1046 }
1047
1048 static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
1049 {
1050 struct va_format vaf;
1051 va_list args;
1052
1053 va_start(args, fmt);
1054 vaf.fmt = fmt;
1055 vaf.va = &args;
1056
1057 pr_err("sysctl table check failed: %s/%s %pV\n",
1058 path, table->procname, &vaf);
1059
1060 va_end(args);
1061 return -EINVAL;
1062 }
1063
1064 static int sysctl_check_table(const char *path, struct ctl_table *table)
1065 {
1066 int err = 0;
1067 for (; table->procname; table++) {
1068 if (table->child)
1069 err = sysctl_err(path, table, "Not a file");
1070
1071 if ((table->proc_handler == proc_dostring) ||
1072 (table->proc_handler == proc_dointvec) ||
1073 (table->proc_handler == proc_dointvec_minmax) ||
1074 (table->proc_handler == proc_dointvec_jiffies) ||
1075 (table->proc_handler == proc_dointvec_userhz_jiffies) ||
1076 (table->proc_handler == proc_dointvec_ms_jiffies) ||
1077 (table->proc_handler == proc_doulongvec_minmax) ||
1078 (table->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1079 if (!table->data)
1080 err = sysctl_err(path, table, "No data");
1081 if (!table->maxlen)
1082 err = sysctl_err(path, table, "No maxlen");
1083 }
1084 if (!table->proc_handler)
1085 err = sysctl_err(path, table, "No proc_handler");
1086
1087 if ((table->mode & (S_IRUGO|S_IWUGO)) != table->mode)
1088 err = sysctl_err(path, table, "bogus .mode 0%o",
1089 table->mode);
1090 }
1091 return err;
1092 }
1093
1094 static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table *table,
1095 struct ctl_table_root *link_root)
1096 {
1097 struct ctl_table *link_table, *entry, *link;
1098 struct ctl_table_header *links;
1099 struct ctl_node *node;
1100 char *link_name;
1101 int nr_entries, name_bytes;
1102
1103 name_bytes = 0;
1104 nr_entries = 0;
1105 for (entry = table; entry->procname; entry++) {
1106 nr_entries++;
1107 name_bytes += strlen(entry->procname) + 1;
1108 }
1109
1110 links = kzalloc(sizeof(struct ctl_table_header) +
1111 sizeof(struct ctl_node)*nr_entries +
1112 sizeof(struct ctl_table)*(nr_entries + 1) +
1113 name_bytes,
1114 GFP_KERNEL);
1115
1116 if (!links)
1117 return NULL;
1118
1119 node = (struct ctl_node *)(links + 1);
1120 link_table = (struct ctl_table *)(node + nr_entries);
1121 link_name = (char *)&link_table[nr_entries + 1];
1122
1123 for (link = link_table, entry = table; entry->procname; link++, entry++) {
1124 int len = strlen(entry->procname) + 1;
1125 memcpy(link_name, entry->procname, len);
1126 link->procname = link_name;
1127 link->mode = S_IFLNK|S_IRWXUGO;
1128 link->data = link_root;
1129 link_name += len;
1130 }
1131 init_header(links, dir->header.root, dir->header.set, node, link_table);
1132 links->nreg = nr_entries;
1133
1134 return links;
1135 }
1136
1137 static bool get_links(struct ctl_dir *dir,
1138 struct ctl_table *table, struct ctl_table_root *link_root)
1139 {
1140 struct ctl_table_header *head;
1141 struct ctl_table *entry, *link;
1142
1143 /* Are there links available for every entry in table? */
1144 for (entry = table; entry->procname; entry++) {
1145 const char *procname = entry->procname;
1146 link = find_entry(&head, dir, procname, strlen(procname));
1147 if (!link)
1148 return false;
1149 if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1150 continue;
1151 if (S_ISLNK(link->mode) && (link->data == link_root))
1152 continue;
1153 return false;
1154 }
1155
1156 /* The checks passed. Increase the registration count on the links */
1157 for (entry = table; entry->procname; entry++) {
1158 const char *procname = entry->procname;
1159 link = find_entry(&head, dir, procname, strlen(procname));
1160 head->nreg++;
1161 }
1162 return true;
1163 }
1164
1165 static int insert_links(struct ctl_table_header *head)
1166 {
1167 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1168 struct ctl_dir *core_parent = NULL;
1169 struct ctl_table_header *links;
1170 int err;
1171
1172 if (head->set == root_set)
1173 return 0;
1174
1175 core_parent = xlate_dir(root_set, head->parent);
1176 if (IS_ERR(core_parent))
1177 return 0;
1178
1179 if (get_links(core_parent, head->ctl_table, head->root))
1180 return 0;
1181
1182 core_parent->header.nreg++;
1183 spin_unlock(&sysctl_lock);
1184
1185 links = new_links(core_parent, head->ctl_table, head->root);
1186
1187 spin_lock(&sysctl_lock);
1188 err = -ENOMEM;
1189 if (!links)
1190 goto out;
1191
1192 err = 0;
1193 if (get_links(core_parent, head->ctl_table, head->root)) {
1194 kfree(links);
1195 goto out;
1196 }
1197
1198 err = insert_header(core_parent, links);
1199 if (err)
1200 kfree(links);
1201 out:
1202 drop_sysctl_table(&core_parent->header);
1203 return err;
1204 }
1205
1206 /**
1207 * __register_sysctl_table - register a leaf sysctl table
1208 * @set: Sysctl tree to register on
1209 * @path: The path to the directory the sysctl table is in.
1210 * @table: the top-level table structure
1211 *
1212 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1213 * array. A completely 0 filled entry terminates the table.
1214 *
1215 * The members of the &struct ctl_table structure are used as follows:
1216 *
1217 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1218 * enter a sysctl file
1219 *
1220 * data - a pointer to data for use by proc_handler
1221 *
1222 * maxlen - the maximum size in bytes of the data
1223 *
1224 * mode - the file permissions for the /proc/sys file
1225 *
1226 * child - must be %NULL.
1227 *
1228 * proc_handler - the text handler routine (described below)
1229 *
1230 * extra1, extra2 - extra pointers usable by the proc handler routines
1231 *
1232 * Leaf nodes in the sysctl tree will be represented by a single file
1233 * under /proc; non-leaf nodes will be represented by directories.
1234 *
1235 * There must be a proc_handler routine for any terminal nodes.
1236 * Several default handlers are available to cover common cases -
1237 *
1238 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1239 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1240 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1241 *
1242 * It is the handler's job to read the input buffer from user memory
1243 * and process it. The handler should return 0 on success.
1244 *
1245 * This routine returns %NULL on a failure to register, and a pointer
1246 * to the table header on success.
1247 */
1248 struct ctl_table_header *__register_sysctl_table(
1249 struct ctl_table_set *set,
1250 const char *path, struct ctl_table *table)
1251 {
1252 struct ctl_table_root *root = set->dir.header.root;
1253 struct ctl_table_header *header;
1254 const char *name, *nextname;
1255 struct ctl_dir *dir;
1256 struct ctl_table *entry;
1257 struct ctl_node *node;
1258 int nr_entries = 0;
1259
1260 for (entry = table; entry->procname; entry++)
1261 nr_entries++;
1262
1263 header = kzalloc(sizeof(struct ctl_table_header) +
1264 sizeof(struct ctl_node)*nr_entries, GFP_KERNEL);
1265 if (!header)
1266 return NULL;
1267
1268 node = (struct ctl_node *)(header + 1);
1269 init_header(header, root, set, node, table);
1270 if (sysctl_check_table(path, table))
1271 goto fail;
1272
1273 spin_lock(&sysctl_lock);
1274 dir = &set->dir;
1275 /* Reference moved down the diretory tree get_subdir */
1276 dir->header.nreg++;
1277 spin_unlock(&sysctl_lock);
1278
1279 /* Find the directory for the ctl_table */
1280 for (name = path; name; name = nextname) {
1281 int namelen;
1282 nextname = strchr(name, '/');
1283 if (nextname) {
1284 namelen = nextname - name;
1285 nextname++;
1286 } else {
1287 namelen = strlen(name);
1288 }
1289 if (namelen == 0)
1290 continue;
1291
1292 dir = get_subdir(dir, name, namelen);
1293 if (IS_ERR(dir))
1294 goto fail;
1295 }
1296
1297 spin_lock(&sysctl_lock);
1298 if (insert_header(dir, header))
1299 goto fail_put_dir_locked;
1300
1301 drop_sysctl_table(&dir->header);
1302 spin_unlock(&sysctl_lock);
1303
1304 return header;
1305
1306 fail_put_dir_locked:
1307 drop_sysctl_table(&dir->header);
1308 spin_unlock(&sysctl_lock);
1309 fail:
1310 kfree(header);
1311 dump_stack();
1312 return NULL;
1313 }
1314
1315 /**
1316 * register_sysctl - register a sysctl table
1317 * @path: The path to the directory the sysctl table is in.
1318 * @table: the table structure
1319 *
1320 * Register a sysctl table. @table should be a filled in ctl_table
1321 * array. A completely 0 filled entry terminates the table.
1322 *
1323 * See __register_sysctl_table for more details.
1324 */
1325 struct ctl_table_header *register_sysctl(const char *path, struct ctl_table *table)
1326 {
1327 return __register_sysctl_table(&sysctl_table_root.default_set,
1328 path, table);
1329 }
1330 EXPORT_SYMBOL(register_sysctl);
1331
1332 static char *append_path(const char *path, char *pos, const char *name)
1333 {
1334 int namelen;
1335 namelen = strlen(name);
1336 if (((pos - path) + namelen + 2) >= PATH_MAX)
1337 return NULL;
1338 memcpy(pos, name, namelen);
1339 pos[namelen] = '/';
1340 pos[namelen + 1] = '\0';
1341 pos += namelen + 1;
1342 return pos;
1343 }
1344
1345 static int count_subheaders(struct ctl_table *table)
1346 {
1347 int has_files = 0;
1348 int nr_subheaders = 0;
1349 struct ctl_table *entry;
1350
1351 /* special case: no directory and empty directory */
1352 if (!table || !table->procname)
1353 return 1;
1354
1355 for (entry = table; entry->procname; entry++) {
1356 if (entry->child)
1357 nr_subheaders += count_subheaders(entry->child);
1358 else
1359 has_files = 1;
1360 }
1361 return nr_subheaders + has_files;
1362 }
1363
1364 static int register_leaf_sysctl_tables(const char *path, char *pos,
1365 struct ctl_table_header ***subheader, struct ctl_table_set *set,
1366 struct ctl_table *table)
1367 {
1368 struct ctl_table *ctl_table_arg = NULL;
1369 struct ctl_table *entry, *files;
1370 int nr_files = 0;
1371 int nr_dirs = 0;
1372 int err = -ENOMEM;
1373
1374 for (entry = table; entry->procname; entry++) {
1375 if (entry->child)
1376 nr_dirs++;
1377 else
1378 nr_files++;
1379 }
1380
1381 files = table;
1382 /* If there are mixed files and directories we need a new table */
1383 if (nr_dirs && nr_files) {
1384 struct ctl_table *new;
1385 files = kzalloc(sizeof(struct ctl_table) * (nr_files + 1),
1386 GFP_KERNEL);
1387 if (!files)
1388 goto out;
1389
1390 ctl_table_arg = files;
1391 for (new = files, entry = table; entry->procname; entry++) {
1392 if (entry->child)
1393 continue;
1394 *new = *entry;
1395 new++;
1396 }
1397 }
1398
1399 /* Register everything except a directory full of subdirectories */
1400 if (nr_files || !nr_dirs) {
1401 struct ctl_table_header *header;
1402 header = __register_sysctl_table(set, path, files);
1403 if (!header) {
1404 kfree(ctl_table_arg);
1405 goto out;
1406 }
1407
1408 /* Remember if we need to free the file table */
1409 header->ctl_table_arg = ctl_table_arg;
1410 **subheader = header;
1411 (*subheader)++;
1412 }
1413
1414 /* Recurse into the subdirectories. */
1415 for (entry = table; entry->procname; entry++) {
1416 char *child_pos;
1417
1418 if (!entry->child)
1419 continue;
1420
1421 err = -ENAMETOOLONG;
1422 child_pos = append_path(path, pos, entry->procname);
1423 if (!child_pos)
1424 goto out;
1425
1426 err = register_leaf_sysctl_tables(path, child_pos, subheader,
1427 set, entry->child);
1428 pos[0] = '\0';
1429 if (err)
1430 goto out;
1431 }
1432 err = 0;
1433 out:
1434 /* On failure our caller will unregister all registered subheaders */
1435 return err;
1436 }
1437
1438 /**
1439 * __register_sysctl_paths - register a sysctl table hierarchy
1440 * @set: Sysctl tree to register on
1441 * @path: The path to the directory the sysctl table is in.
1442 * @table: the top-level table structure
1443 *
1444 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1445 * array. A completely 0 filled entry terminates the table.
1446 *
1447 * See __register_sysctl_table for more details.
1448 */
1449 struct ctl_table_header *__register_sysctl_paths(
1450 struct ctl_table_set *set,
1451 const struct ctl_path *path, struct ctl_table *table)
1452 {
1453 struct ctl_table *ctl_table_arg = table;
1454 int nr_subheaders = count_subheaders(table);
1455 struct ctl_table_header *header = NULL, **subheaders, **subheader;
1456 const struct ctl_path *component;
1457 char *new_path, *pos;
1458
1459 pos = new_path = kmalloc(PATH_MAX, GFP_KERNEL);
1460 if (!new_path)
1461 return NULL;
1462
1463 pos[0] = '\0';
1464 for (component = path; component->procname; component++) {
1465 pos = append_path(new_path, pos, component->procname);
1466 if (!pos)
1467 goto out;
1468 }
1469 while (table->procname && table->child && !table[1].procname) {
1470 pos = append_path(new_path, pos, table->procname);
1471 if (!pos)
1472 goto out;
1473 table = table->child;
1474 }
1475 if (nr_subheaders == 1) {
1476 header = __register_sysctl_table(set, new_path, table);
1477 if (header)
1478 header->ctl_table_arg = ctl_table_arg;
1479 } else {
1480 header = kzalloc(sizeof(*header) +
1481 sizeof(*subheaders)*nr_subheaders, GFP_KERNEL);
1482 if (!header)
1483 goto out;
1484
1485 subheaders = (struct ctl_table_header **) (header + 1);
1486 subheader = subheaders;
1487 header->ctl_table_arg = ctl_table_arg;
1488
1489 if (register_leaf_sysctl_tables(new_path, pos, &subheader,
1490 set, table))
1491 goto err_register_leaves;
1492 }
1493
1494 out:
1495 kfree(new_path);
1496 return header;
1497
1498 err_register_leaves:
1499 while (subheader > subheaders) {
1500 struct ctl_table_header *subh = *(--subheader);
1501 struct ctl_table *table = subh->ctl_table_arg;
1502 unregister_sysctl_table(subh);
1503 kfree(table);
1504 }
1505 kfree(header);
1506 header = NULL;
1507 goto out;
1508 }
1509
1510 /**
1511 * register_sysctl_table_path - register a sysctl table hierarchy
1512 * @path: The path to the directory the sysctl table is in.
1513 * @table: the top-level table structure
1514 *
1515 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1516 * array. A completely 0 filled entry terminates the table.
1517 *
1518 * See __register_sysctl_paths for more details.
1519 */
1520 struct ctl_table_header *register_sysctl_paths(const struct ctl_path *path,
1521 struct ctl_table *table)
1522 {
1523 return __register_sysctl_paths(&sysctl_table_root.default_set,
1524 path, table);
1525 }
1526 EXPORT_SYMBOL(register_sysctl_paths);
1527
1528 /**
1529 * register_sysctl_table - register a sysctl table hierarchy
1530 * @table: the top-level table structure
1531 *
1532 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1533 * array. A completely 0 filled entry terminates the table.
1534 *
1535 * See register_sysctl_paths for more details.
1536 */
1537 struct ctl_table_header *register_sysctl_table(struct ctl_table *table)
1538 {
1539 static const struct ctl_path null_path[] = { {} };
1540
1541 return register_sysctl_paths(null_path, table);
1542 }
1543 EXPORT_SYMBOL(register_sysctl_table);
1544
1545 static void put_links(struct ctl_table_header *header)
1546 {
1547 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1548 struct ctl_table_root *root = header->root;
1549 struct ctl_dir *parent = header->parent;
1550 struct ctl_dir *core_parent;
1551 struct ctl_table *entry;
1552
1553 if (header->set == root_set)
1554 return;
1555
1556 core_parent = xlate_dir(root_set, parent);
1557 if (IS_ERR(core_parent))
1558 return;
1559
1560 for (entry = header->ctl_table; entry->procname; entry++) {
1561 struct ctl_table_header *link_head;
1562 struct ctl_table *link;
1563 const char *name = entry->procname;
1564
1565 link = find_entry(&link_head, core_parent, name, strlen(name));
1566 if (link &&
1567 ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1568 (S_ISLNK(link->mode) && (link->data == root)))) {
1569 drop_sysctl_table(link_head);
1570 }
1571 else {
1572 pr_err("sysctl link missing during unregister: ");
1573 sysctl_print_dir(parent);
1574 pr_cont("/%s\n", name);
1575 }
1576 }
1577 }
1578
1579 static void drop_sysctl_table(struct ctl_table_header *header)
1580 {
1581 struct ctl_dir *parent = header->parent;
1582
1583 if (--header->nreg)
1584 return;
1585
1586 put_links(header);
1587 start_unregistering(header);
1588 if (!--header->count)
1589 kfree_rcu(header, rcu);
1590
1591 if (parent)
1592 drop_sysctl_table(&parent->header);
1593 }
1594
1595 /**
1596 * unregister_sysctl_table - unregister a sysctl table hierarchy
1597 * @header: the header returned from register_sysctl_table
1598 *
1599 * Unregisters the sysctl table and all children. proc entries may not
1600 * actually be removed until they are no longer used by anyone.
1601 */
1602 void unregister_sysctl_table(struct ctl_table_header * header)
1603 {
1604 int nr_subheaders;
1605 might_sleep();
1606
1607 if (header == NULL)
1608 return;
1609
1610 nr_subheaders = count_subheaders(header->ctl_table_arg);
1611 if (unlikely(nr_subheaders > 1)) {
1612 struct ctl_table_header **subheaders;
1613 int i;
1614
1615 subheaders = (struct ctl_table_header **)(header + 1);
1616 for (i = nr_subheaders -1; i >= 0; i--) {
1617 struct ctl_table_header *subh = subheaders[i];
1618 struct ctl_table *table = subh->ctl_table_arg;
1619 unregister_sysctl_table(subh);
1620 kfree(table);
1621 }
1622 kfree(header);
1623 return;
1624 }
1625
1626 spin_lock(&sysctl_lock);
1627 drop_sysctl_table(header);
1628 spin_unlock(&sysctl_lock);
1629 }
1630 EXPORT_SYMBOL(unregister_sysctl_table);
1631
1632 void setup_sysctl_set(struct ctl_table_set *set,
1633 struct ctl_table_root *root,
1634 int (*is_seen)(struct ctl_table_set *))
1635 {
1636 memset(set, 0, sizeof(*set));
1637 set->is_seen = is_seen;
1638 init_header(&set->dir.header, root, set, NULL, root_table);
1639 }
1640
1641 void retire_sysctl_set(struct ctl_table_set *set)
1642 {
1643 WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1644 }
1645
1646 int __init proc_sys_init(void)
1647 {
1648 struct proc_dir_entry *proc_sys_root;
1649
1650 proc_sys_root = proc_mkdir("sys", NULL);
1651 proc_sys_root->proc_iops = &proc_sys_dir_operations;
1652 proc_sys_root->proc_fops = &proc_sys_dir_file_operations;
1653 proc_sys_root->nlink = 0;
1654
1655 return sysctl_init();
1656 }
CRIS linux kernel tree