?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
gpio: rcar: Fix runtime PM imbalance on error
[linux/fpc-iii.git] / fs / proc / proc_sysctl.c
blobb6f5d459b087de348a1f71a9ab08a7809552e36a
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * /proc/sys support
4 */
5 #include <linux/init.h>
6 #include <linux/sysctl.h>
7 #include <linux/poll.h>
8 #include <linux/proc_fs.h>
9 #include <linux/printk.h>
10 #include <linux/security.h>
11 #include <linux/sched.h>
12 #include <linux/cred.h>
13 #include <linux/namei.h>
14 #include <linux/mm.h>
15 #include <linux/module.h>
16 #include <linux/bpf-cgroup.h>
17 #include "internal.h"
18
19 static const struct dentry_operations proc_sys_dentry_operations;
20 static const struct file_operations proc_sys_file_operations;
21 static const struct inode_operations proc_sys_inode_operations;
22 static const struct file_operations proc_sys_dir_file_operations;
23 static const struct inode_operations proc_sys_dir_operations;
24
25 /* shared constants to be used in various sysctls */
26 const int sysctl_vals[] = { 0, 1, INT_MAX };
27 EXPORT_SYMBOL(sysctl_vals);
28
29 /* Support for permanently empty directories */
30
31 struct ctl_table sysctl_mount_point[] = {
32 { }
33 };
34
35 static bool is_empty_dir(struct ctl_table_header *head)
36 {
37 return head->ctl_table[0].child == sysctl_mount_point;
38 }
39
40 static void set_empty_dir(struct ctl_dir *dir)
41 {
42 dir->header.ctl_table[0].child = sysctl_mount_point;
43 }
44
45 static void clear_empty_dir(struct ctl_dir *dir)
46
47 {
48 dir->header.ctl_table[0].child = NULL;
49 }
50
51 void proc_sys_poll_notify(struct ctl_table_poll *poll)
52 {
53 if (!poll)
54 return;
55
56 atomic_inc(&poll->event);
57 wake_up_interruptible(&poll->wait);
58 }
59
60 static struct ctl_table root_table[] = {
61 {
62 .procname = "",
63 .mode = S_IFDIR|S_IRUGO|S_IXUGO,
64 },
65 { }
66 };
67 static struct ctl_table_root sysctl_table_root = {
68 .default_set.dir.header = {
69 {{.count = 1,
70 .nreg = 1,
71 .ctl_table = root_table }},
72 .ctl_table_arg = root_table,
73 .root = &sysctl_table_root,
74 .set = &sysctl_table_root.default_set,
75 },
76 };
77
78 static DEFINE_SPINLOCK(sysctl_lock);
79
80 static void drop_sysctl_table(struct ctl_table_header *header);
81 static int sysctl_follow_link(struct ctl_table_header **phead,
82 struct ctl_table **pentry);
83 static int insert_links(struct ctl_table_header *head);
84 static void put_links(struct ctl_table_header *header);
85
86 static void sysctl_print_dir(struct ctl_dir *dir)
87 {
88 if (dir->header.parent)
89 sysctl_print_dir(dir->header.parent);
90 pr_cont("%s/", dir->header.ctl_table[0].procname);
91 }
92
93 static int namecmp(const char *name1, int len1, const char *name2, int len2)
94 {
95 int minlen;
96 int cmp;
97
98 minlen = len1;
99 if (minlen > len2)
100 minlen = len2;
101
102 cmp = memcmp(name1, name2, minlen);
103 if (cmp == 0)
104 cmp = len1 - len2;
105 return cmp;
106 }
107
108 /* Called under sysctl_lock */
109 static struct ctl_table *find_entry(struct ctl_table_header **phead,
110 struct ctl_dir *dir, const char *name, int namelen)
111 {
112 struct ctl_table_header *head;
113 struct ctl_table *entry;
114 struct rb_node *node = dir->root.rb_node;
115
116 while (node)
117 {
118 struct ctl_node *ctl_node;
119 const char *procname;
120 int cmp;
121
122 ctl_node = rb_entry(node, struct ctl_node, node);
123 head = ctl_node->header;
124 entry = &head->ctl_table[ctl_node - head->node];
125 procname = entry->procname;
126
127 cmp = namecmp(name, namelen, procname, strlen(procname));
128 if (cmp < 0)
129 node = node->rb_left;
130 else if (cmp > 0)
131 node = node->rb_right;
132 else {
133 *phead = head;
134 return entry;
135 }
136 }
137 return NULL;
138 }
139
140 static int insert_entry(struct ctl_table_header *head, struct ctl_table *entry)
141 {
142 struct rb_node *node = &head->node[entry - head->ctl_table].node;
143 struct rb_node **p = &head->parent->root.rb_node;
144 struct rb_node *parent = NULL;
145 const char *name = entry->procname;
146 int namelen = strlen(name);
147
148 while (*p) {
149 struct ctl_table_header *parent_head;
150 struct ctl_table *parent_entry;
151 struct ctl_node *parent_node;
152 const char *parent_name;
153 int cmp;
154
155 parent = *p;
156 parent_node = rb_entry(parent, struct ctl_node, node);
157 parent_head = parent_node->header;
158 parent_entry = &parent_head->ctl_table[parent_node - parent_head->node];
159 parent_name = parent_entry->procname;
160
161 cmp = namecmp(name, namelen, parent_name, strlen(parent_name));
162 if (cmp < 0)
163 p = &(*p)->rb_left;
164 else if (cmp > 0)
165 p = &(*p)->rb_right;
166 else {
167 pr_err("sysctl duplicate entry: ");
168 sysctl_print_dir(head->parent);
169 pr_cont("/%s\n", entry->procname);
170 return -EEXIST;
171 }
172 }
173
174 rb_link_node(node, parent, p);
175 rb_insert_color(node, &head->parent->root);
176 return 0;
177 }
178
179 static void erase_entry(struct ctl_table_header *head, struct ctl_table *entry)
180 {
181 struct rb_node *node = &head->node[entry - head->ctl_table].node;
182
183 rb_erase(node, &head->parent->root);
184 }
185
186 static void init_header(struct ctl_table_header *head,
187 struct ctl_table_root *root, struct ctl_table_set *set,
188 struct ctl_node *node, struct ctl_table *table)
189 {
190 head->ctl_table = table;
191 head->ctl_table_arg = table;
192 head->used = 0;
193 head->count = 1;
194 head->nreg = 1;
195 head->unregistering = NULL;
196 head->root = root;
197 head->set = set;
198 head->parent = NULL;
199 head->node = node;
200 INIT_HLIST_HEAD(&head->inodes);
201 if (node) {
202 struct ctl_table *entry;
203 for (entry = table; entry->procname; entry++, node++)
204 node->header = head;
205 }
206 }
207
208 static void erase_header(struct ctl_table_header *head)
209 {
210 struct ctl_table *entry;
211 for (entry = head->ctl_table; entry->procname; entry++)
212 erase_entry(head, entry);
213 }
214
215 static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
216 {
217 struct ctl_table *entry;
218 int err;
219
220 /* Is this a permanently empty directory? */
221 if (is_empty_dir(&dir->header))
222 return -EROFS;
223
224 /* Am I creating a permanently empty directory? */
225 if (header->ctl_table == sysctl_mount_point) {
226 if (!RB_EMPTY_ROOT(&dir->root))
227 return -EINVAL;
228 set_empty_dir(dir);
229 }
230
231 dir->header.nreg++;
232 header->parent = dir;
233 err = insert_links(header);
234 if (err)
235 goto fail_links;
236 for (entry = header->ctl_table; entry->procname; entry++) {
237 err = insert_entry(header, entry);
238 if (err)
239 goto fail;
240 }
241 return 0;
242 fail:
243 erase_header(header);
244 put_links(header);
245 fail_links:
246 if (header->ctl_table == sysctl_mount_point)
247 clear_empty_dir(dir);
248 header->parent = NULL;
249 drop_sysctl_table(&dir->header);
250 return err;
251 }
252
253 /* called under sysctl_lock */
254 static int use_table(struct ctl_table_header *p)
255 {
256 if (unlikely(p->unregistering))
257 return 0;
258 p->used++;
259 return 1;
260 }
261
262 /* called under sysctl_lock */
263 static void unuse_table(struct ctl_table_header *p)
264 {
265 if (!--p->used)
266 if (unlikely(p->unregistering))
267 complete(p->unregistering);
268 }
269
270 static void proc_sys_invalidate_dcache(struct ctl_table_header *head)
271 {
272 proc_invalidate_siblings_dcache(&head->inodes, &sysctl_lock);
273 }
274
275 /* called under sysctl_lock, will reacquire if has to wait */
276 static void start_unregistering(struct ctl_table_header *p)
277 {
278 /*
279 * if p->used is 0, nobody will ever touch that entry again;
280 * we'll eliminate all paths to it before dropping sysctl_lock
281 */
282 if (unlikely(p->used)) {
283 struct completion wait;
284 init_completion(&wait);
285 p->unregistering = &wait;
286 spin_unlock(&sysctl_lock);
287 wait_for_completion(&wait);
288 } else {
289 /* anything non-NULL; we'll never dereference it */
290 p->unregistering = ERR_PTR(-EINVAL);
291 spin_unlock(&sysctl_lock);
292 }
293 /*
294 * Invalidate dentries for unregistered sysctls: namespaced sysctls
295 * can have duplicate names and contaminate dcache very badly.
296 */
297 proc_sys_invalidate_dcache(p);
298 /*
299 * do not remove from the list until nobody holds it; walking the
300 * list in do_sysctl() relies on that.
301 */
302 spin_lock(&sysctl_lock);
303 erase_header(p);
304 }
305
306 static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
307 {
308 BUG_ON(!head);
309 spin_lock(&sysctl_lock);
310 if (!use_table(head))
311 head = ERR_PTR(-ENOENT);
312 spin_unlock(&sysctl_lock);
313 return head;
314 }
315
316 static void sysctl_head_finish(struct ctl_table_header *head)
317 {
318 if (!head)
319 return;
320 spin_lock(&sysctl_lock);
321 unuse_table(head);
322 spin_unlock(&sysctl_lock);
323 }
324
325 static struct ctl_table_set *
326 lookup_header_set(struct ctl_table_root *root)
327 {
328 struct ctl_table_set *set = &root->default_set;
329 if (root->lookup)
330 set = root->lookup(root);
331 return set;
332 }
333
334 static struct ctl_table *lookup_entry(struct ctl_table_header **phead,
335 struct ctl_dir *dir,
336 const char *name, int namelen)
337 {
338 struct ctl_table_header *head;
339 struct ctl_table *entry;
340
341 spin_lock(&sysctl_lock);
342 entry = find_entry(&head, dir, name, namelen);
343 if (entry && use_table(head))
344 *phead = head;
345 else
346 entry = NULL;
347 spin_unlock(&sysctl_lock);
348 return entry;
349 }
350
351 static struct ctl_node *first_usable_entry(struct rb_node *node)
352 {
353 struct ctl_node *ctl_node;
354
355 for (;node; node = rb_next(node)) {
356 ctl_node = rb_entry(node, struct ctl_node, node);
357 if (use_table(ctl_node->header))
358 return ctl_node;
359 }
360 return NULL;
361 }
362
363 static void first_entry(struct ctl_dir *dir,
364 struct ctl_table_header **phead, struct ctl_table **pentry)
365 {
366 struct ctl_table_header *head = NULL;
367 struct ctl_table *entry = NULL;
368 struct ctl_node *ctl_node;
369
370 spin_lock(&sysctl_lock);
371 ctl_node = first_usable_entry(rb_first(&dir->root));
372 spin_unlock(&sysctl_lock);
373 if (ctl_node) {
374 head = ctl_node->header;
375 entry = &head->ctl_table[ctl_node - head->node];
376 }
377 *phead = head;
378 *pentry = entry;
379 }
380
381 static void next_entry(struct ctl_table_header **phead, struct ctl_table **pentry)
382 {
383 struct ctl_table_header *head = *phead;
384 struct ctl_table *entry = *pentry;
385 struct ctl_node *ctl_node = &head->node[entry - head->ctl_table];
386
387 spin_lock(&sysctl_lock);
388 unuse_table(head);
389
390 ctl_node = first_usable_entry(rb_next(&ctl_node->node));
391 spin_unlock(&sysctl_lock);
392 head = NULL;
393 if (ctl_node) {
394 head = ctl_node->header;
395 entry = &head->ctl_table[ctl_node - head->node];
396 }
397 *phead = head;
398 *pentry = entry;
399 }
400
401 /*
402 * sysctl_perm does NOT grant the superuser all rights automatically, because
403 * some sysctl variables are readonly even to root.
404 */
405
406 static int test_perm(int mode, int op)
407 {
408 if (uid_eq(current_euid(), GLOBAL_ROOT_UID))
409 mode >>= 6;
410 else if (in_egroup_p(GLOBAL_ROOT_GID))
411 mode >>= 3;
412 if ((op & ~mode & (MAY_READ|MAY_WRITE|MAY_EXEC)) == 0)
413 return 0;
414 return -EACCES;
415 }
416
417 static int sysctl_perm(struct ctl_table_header *head, struct ctl_table *table, int op)
418 {
419 struct ctl_table_root *root = head->root;
420 int mode;
421
422 if (root->permissions)
423 mode = root->permissions(head, table);
424 else
425 mode = table->mode;
426
427 return test_perm(mode, op);
428 }
429
430 static struct inode *proc_sys_make_inode(struct super_block *sb,
431 struct ctl_table_header *head, struct ctl_table *table)
432 {
433 struct ctl_table_root *root = head->root;
434 struct inode *inode;
435 struct proc_inode *ei;
436
437 inode = new_inode(sb);
438 if (!inode)
439 return ERR_PTR(-ENOMEM);
440
441 inode->i_ino = get_next_ino();
442
443 ei = PROC_I(inode);
444
445 spin_lock(&sysctl_lock);
446 if (unlikely(head->unregistering)) {
447 spin_unlock(&sysctl_lock);
448 iput(inode);
449 return ERR_PTR(-ENOENT);
450 }
451 ei->sysctl = head;
452 ei->sysctl_entry = table;
453 hlist_add_head_rcu(&ei->sibling_inodes, &head->inodes);
454 head->count++;
455 spin_unlock(&sysctl_lock);
456
457 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
458 inode->i_mode = table->mode;
459 if (!S_ISDIR(table->mode)) {
460 inode->i_mode |= S_IFREG;
461 inode->i_op = &proc_sys_inode_operations;
462 inode->i_fop = &proc_sys_file_operations;
463 } else {
464 inode->i_mode |= S_IFDIR;
465 inode->i_op = &proc_sys_dir_operations;
466 inode->i_fop = &proc_sys_dir_file_operations;
467 if (is_empty_dir(head))
468 make_empty_dir_inode(inode);
469 }
470
471 if (root->set_ownership)
472 root->set_ownership(head, table, &inode->i_uid, &inode->i_gid);
473 else {
474 inode->i_uid = GLOBAL_ROOT_UID;
475 inode->i_gid = GLOBAL_ROOT_GID;
476 }
477
478 return inode;
479 }
480
481 void proc_sys_evict_inode(struct inode *inode, struct ctl_table_header *head)
482 {
483 spin_lock(&sysctl_lock);
484 hlist_del_init_rcu(&PROC_I(inode)->sibling_inodes);
485 if (!--head->count)
486 kfree_rcu(head, rcu);
487 spin_unlock(&sysctl_lock);
488 }
489
490 static struct ctl_table_header *grab_header(struct inode *inode)
491 {
492 struct ctl_table_header *head = PROC_I(inode)->sysctl;
493 if (!head)
494 head = &sysctl_table_root.default_set.dir.header;
495 return sysctl_head_grab(head);
496 }
497
498 static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
499 unsigned int flags)
500 {
501 struct ctl_table_header *head = grab_header(dir);
502 struct ctl_table_header *h = NULL;
503 const struct qstr *name = &dentry->d_name;
504 struct ctl_table *p;
505 struct inode *inode;
506 struct dentry *err = ERR_PTR(-ENOENT);
507 struct ctl_dir *ctl_dir;
508 int ret;
509
510 if (IS_ERR(head))
511 return ERR_CAST(head);
512
513 ctl_dir = container_of(head, struct ctl_dir, header);
514
515 p = lookup_entry(&h, ctl_dir, name->name, name->len);
516 if (!p)
517 goto out;
518
519 if (S_ISLNK(p->mode)) {
520 ret = sysctl_follow_link(&h, &p);
521 err = ERR_PTR(ret);
522 if (ret)
523 goto out;
524 }
525
526 inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
527 if (IS_ERR(inode)) {
528 err = ERR_CAST(inode);
529 goto out;
530 }
531
532 d_set_d_op(dentry, &proc_sys_dentry_operations);
533 err = d_splice_alias(inode, dentry);
534
535 out:
536 if (h)
537 sysctl_head_finish(h);
538 sysctl_head_finish(head);
539 return err;
540 }
541
542 static ssize_t proc_sys_call_handler(struct file *filp, void __user *buf,
543 size_t count, loff_t *ppos, int write)
544 {
545 struct inode *inode = file_inode(filp);
546 struct ctl_table_header *head = grab_header(inode);
547 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
548 void *new_buf = NULL;
549 ssize_t error;
550
551 if (IS_ERR(head))
552 return PTR_ERR(head);
553
554 /*
555 * At this point we know that the sysctl was not unregistered
556 * and won't be until we finish.
557 */
558 error = -EPERM;
559 if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
560 goto out;
561
562 /* if that can happen at all, it should be -EINVAL, not -EISDIR */
563 error = -EINVAL;
564 if (!table->proc_handler)
565 goto out;
566
567 error = BPF_CGROUP_RUN_PROG_SYSCTL(head, table, write, buf, &count,
568 ppos, &new_buf);
569 if (error)
570 goto out;
571
572 /* careful: calling conventions are nasty here */
573 if (new_buf) {
574 mm_segment_t old_fs;
575
576 old_fs = get_fs();
577 set_fs(KERNEL_DS);
578 error = table->proc_handler(table, write, (void __user *)new_buf,
579 &count, ppos);
580 set_fs(old_fs);
581 kfree(new_buf);
582 } else {
583 error = table->proc_handler(table, write, buf, &count, ppos);
584 }
585
586 if (!error)
587 error = count;
588 out:
589 sysctl_head_finish(head);
590
591 return error;
592 }
593
594 static ssize_t proc_sys_read(struct file *filp, char __user *buf,
595 size_t count, loff_t *ppos)
596 {
597 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 0);
598 }
599
600 static ssize_t proc_sys_write(struct file *filp, const char __user *buf,
601 size_t count, loff_t *ppos)
602 {
603 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 1);
604 }
605
606 static int proc_sys_open(struct inode *inode, struct file *filp)
607 {
608 struct ctl_table_header *head = grab_header(inode);
609 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
610
611 /* sysctl was unregistered */
612 if (IS_ERR(head))
613 return PTR_ERR(head);
614
615 if (table->poll)
616 filp->private_data = proc_sys_poll_event(table->poll);
617
618 sysctl_head_finish(head);
619
620 return 0;
621 }
622
623 static __poll_t proc_sys_poll(struct file *filp, poll_table *wait)
624 {
625 struct inode *inode = file_inode(filp);
626 struct ctl_table_header *head = grab_header(inode);
627 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
628 __poll_t ret = DEFAULT_POLLMASK;
629 unsigned long event;
630
631 /* sysctl was unregistered */
632 if (IS_ERR(head))
633 return EPOLLERR | EPOLLHUP;
634
635 if (!table->proc_handler)
636 goto out;
637
638 if (!table->poll)
639 goto out;
640
641 event = (unsigned long)filp->private_data;
642 poll_wait(filp, &table->poll->wait, wait);
643
644 if (event != atomic_read(&table->poll->event)) {
645 filp->private_data = proc_sys_poll_event(table->poll);
646 ret = EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
647 }
648
649 out:
650 sysctl_head_finish(head);
651
652 return ret;
653 }
654
655 static bool proc_sys_fill_cache(struct file *file,
656 struct dir_context *ctx,
657 struct ctl_table_header *head,
658 struct ctl_table *table)
659 {
660 struct dentry *child, *dir = file->f_path.dentry;
661 struct inode *inode;
662 struct qstr qname;
663 ino_t ino = 0;
664 unsigned type = DT_UNKNOWN;
665
666 qname.name = table->procname;
667 qname.len = strlen(table->procname);
668 qname.hash = full_name_hash(dir, qname.name, qname.len);
669
670 child = d_lookup(dir, &qname);
671 if (!child) {
672 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
673 child = d_alloc_parallel(dir, &qname, &wq);
674 if (IS_ERR(child))
675 return false;
676 if (d_in_lookup(child)) {
677 struct dentry *res;
678 inode = proc_sys_make_inode(dir->d_sb, head, table);
679 if (IS_ERR(inode)) {
680 d_lookup_done(child);
681 dput(child);
682 return false;
683 }
684 d_set_d_op(child, &proc_sys_dentry_operations);
685 res = d_splice_alias(inode, child);
686 d_lookup_done(child);
687 if (unlikely(res)) {
688 if (IS_ERR(res)) {
689 dput(child);
690 return false;
691 }
692 dput(child);
693 child = res;
694 }
695 }
696 }
697 inode = d_inode(child);
698 ino = inode->i_ino;
699 type = inode->i_mode >> 12;
700 dput(child);
701 return dir_emit(ctx, qname.name, qname.len, ino, type);
702 }
703
704 static bool proc_sys_link_fill_cache(struct file *file,
705 struct dir_context *ctx,
706 struct ctl_table_header *head,
707 struct ctl_table *table)
708 {
709 bool ret = true;
710
711 head = sysctl_head_grab(head);
712 if (IS_ERR(head))
713 return false;
714
715 /* It is not an error if we can not follow the link ignore it */
716 if (sysctl_follow_link(&head, &table))
717 goto out;
718
719 ret = proc_sys_fill_cache(file, ctx, head, table);
720 out:
721 sysctl_head_finish(head);
722 return ret;
723 }
724
725 static int scan(struct ctl_table_header *head, struct ctl_table *table,
726 unsigned long *pos, struct file *file,
727 struct dir_context *ctx)
728 {
729 bool res;
730
731 if ((*pos)++ < ctx->pos)
732 return true;
733
734 if (unlikely(S_ISLNK(table->mode)))
735 res = proc_sys_link_fill_cache(file, ctx, head, table);
736 else
737 res = proc_sys_fill_cache(file, ctx, head, table);
738
739 if (res)
740 ctx->pos = *pos;
741
742 return res;
743 }
744
745 static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
746 {
747 struct ctl_table_header *head = grab_header(file_inode(file));
748 struct ctl_table_header *h = NULL;
749 struct ctl_table *entry;
750 struct ctl_dir *ctl_dir;
751 unsigned long pos;
752
753 if (IS_ERR(head))
754 return PTR_ERR(head);
755
756 ctl_dir = container_of(head, struct ctl_dir, header);
757
758 if (!dir_emit_dots(file, ctx))
759 goto out;
760
761 pos = 2;
762
763 for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
764 if (!scan(h, entry, &pos, file, ctx)) {
765 sysctl_head_finish(h);
766 break;
767 }
768 }
769 out:
770 sysctl_head_finish(head);
771 return 0;
772 }
773
774 static int proc_sys_permission(struct inode *inode, int mask)
775 {
776 /*
777 * sysctl entries that are not writeable,
778 * are _NOT_ writeable, capabilities or not.
779 */
780 struct ctl_table_header *head;
781 struct ctl_table *table;
782 int error;
783
784 /* Executable files are not allowed under /proc/sys/ */
785 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
786 return -EACCES;
787
788 head = grab_header(inode);
789 if (IS_ERR(head))
790 return PTR_ERR(head);
791
792 table = PROC_I(inode)->sysctl_entry;
793 if (!table) /* global root - r-xr-xr-x */
794 error = mask & MAY_WRITE ? -EACCES : 0;
795 else /* Use the permissions on the sysctl table entry */
796 error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
797
798 sysctl_head_finish(head);
799 return error;
800 }
801
802 static int proc_sys_setattr(struct dentry *dentry, struct iattr *attr)
803 {
804 struct inode *inode = d_inode(dentry);
805 int error;
806
807 if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
808 return -EPERM;
809
810 error = setattr_prepare(dentry, attr);
811 if (error)
812 return error;
813
814 setattr_copy(inode, attr);
815 mark_inode_dirty(inode);
816 return 0;
817 }
818
819 static int proc_sys_getattr(const struct path *path, struct kstat *stat,
820 u32 request_mask, unsigned int query_flags)
821 {
822 struct inode *inode = d_inode(path->dentry);
823 struct ctl_table_header *head = grab_header(inode);
824 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
825
826 if (IS_ERR(head))
827 return PTR_ERR(head);
828
829 generic_fillattr(inode, stat);
830 if (table)
831 stat->mode = (stat->mode & S_IFMT) | table->mode;
832
833 sysctl_head_finish(head);
834 return 0;
835 }
836
837 static const struct file_operations proc_sys_file_operations = {
838 .open = proc_sys_open,
839 .poll = proc_sys_poll,
840 .read = proc_sys_read,
841 .write = proc_sys_write,
842 .llseek = default_llseek,
843 };
844
845 static const struct file_operations proc_sys_dir_file_operations = {
846 .read = generic_read_dir,
847 .iterate_shared = proc_sys_readdir,
848 .llseek = generic_file_llseek,
849 };
850
851 static const struct inode_operations proc_sys_inode_operations = {
852 .permission = proc_sys_permission,
853 .setattr = proc_sys_setattr,
854 .getattr = proc_sys_getattr,
855 };
856
857 static const struct inode_operations proc_sys_dir_operations = {
858 .lookup = proc_sys_lookup,
859 .permission = proc_sys_permission,
860 .setattr = proc_sys_setattr,
861 .getattr = proc_sys_getattr,
862 };
863
864 static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
865 {
866 if (flags & LOOKUP_RCU)
867 return -ECHILD;
868 return !PROC_I(d_inode(dentry))->sysctl->unregistering;
869 }
870
871 static int proc_sys_delete(const struct dentry *dentry)
872 {
873 return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
874 }
875
876 static int sysctl_is_seen(struct ctl_table_header *p)
877 {
878 struct ctl_table_set *set = p->set;
879 int res;
880 spin_lock(&sysctl_lock);
881 if (p->unregistering)
882 res = 0;
883 else if (!set->is_seen)
884 res = 1;
885 else
886 res = set->is_seen(set);
887 spin_unlock(&sysctl_lock);
888 return res;
889 }
890
891 static int proc_sys_compare(const struct dentry *dentry,
892 unsigned int len, const char *str, const struct qstr *name)
893 {
894 struct ctl_table_header *head;
895 struct inode *inode;
896
897 /* Although proc doesn't have negative dentries, rcu-walk means
898 * that inode here can be NULL */
899 /* AV: can it, indeed? */
900 inode = d_inode_rcu(dentry);
901 if (!inode)
902 return 1;
903 if (name->len != len)
904 return 1;
905 if (memcmp(name->name, str, len))
906 return 1;
907 head = rcu_dereference(PROC_I(inode)->sysctl);
908 return !head || !sysctl_is_seen(head);
909 }
910
911 static const struct dentry_operations proc_sys_dentry_operations = {
912 .d_revalidate = proc_sys_revalidate,
913 .d_delete = proc_sys_delete,
914 .d_compare = proc_sys_compare,
915 };
916
917 static struct ctl_dir *find_subdir(struct ctl_dir *dir,
918 const char *name, int namelen)
919 {
920 struct ctl_table_header *head;
921 struct ctl_table *entry;
922
923 entry = find_entry(&head, dir, name, namelen);
924 if (!entry)
925 return ERR_PTR(-ENOENT);
926 if (!S_ISDIR(entry->mode))
927 return ERR_PTR(-ENOTDIR);
928 return container_of(head, struct ctl_dir, header);
929 }
930
931 static struct ctl_dir *new_dir(struct ctl_table_set *set,
932 const char *name, int namelen)
933 {
934 struct ctl_table *table;
935 struct ctl_dir *new;
936 struct ctl_node *node;
937 char *new_name;
938
939 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
940 sizeof(struct ctl_table)*2 + namelen + 1,
941 GFP_KERNEL);
942 if (!new)
943 return NULL;
944
945 node = (struct ctl_node *)(new + 1);
946 table = (struct ctl_table *)(node + 1);
947 new_name = (char *)(table + 2);
948 memcpy(new_name, name, namelen);
949 new_name[namelen] = '\0';
950 table[0].procname = new_name;
951 table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
952 init_header(&new->header, set->dir.header.root, set, node, table);
953
954 return new;
955 }
956
957 /**
958 * get_subdir - find or create a subdir with the specified name.
959 * @dir: Directory to create the subdirectory in
960 * @name: The name of the subdirectory to find or create
961 * @namelen: The length of name
962 *
963 * Takes a directory with an elevated reference count so we know that
964 * if we drop the lock the directory will not go away. Upon success
965 * the reference is moved from @dir to the returned subdirectory.
966 * Upon error an error code is returned and the reference on @dir is
967 * simply dropped.
968 */
969 static struct ctl_dir *get_subdir(struct ctl_dir *dir,
970 const char *name, int namelen)
971 {
972 struct ctl_table_set *set = dir->header.set;
973 struct ctl_dir *subdir, *new = NULL;
974 int err;
975
976 spin_lock(&sysctl_lock);
977 subdir = find_subdir(dir, name, namelen);
978 if (!IS_ERR(subdir))
979 goto found;
980 if (PTR_ERR(subdir) != -ENOENT)
981 goto failed;
982
983 spin_unlock(&sysctl_lock);
984 new = new_dir(set, name, namelen);
985 spin_lock(&sysctl_lock);
986 subdir = ERR_PTR(-ENOMEM);
987 if (!new)
988 goto failed;
989
990 /* Was the subdir added while we dropped the lock? */
991 subdir = find_subdir(dir, name, namelen);
992 if (!IS_ERR(subdir))
993 goto found;
994 if (PTR_ERR(subdir) != -ENOENT)
995 goto failed;
996
997 /* Nope. Use the our freshly made directory entry. */
998 err = insert_header(dir, &new->header);
999 subdir = ERR_PTR(err);
1000 if (err)
1001 goto failed;
1002 subdir = new;
1003 found:
1004 subdir->header.nreg++;
1005 failed:
1006 if (IS_ERR(subdir)) {
1007 pr_err("sysctl could not get directory: ");
1008 sysctl_print_dir(dir);
1009 pr_cont("/%*.*s %ld\n",
1010 namelen, namelen, name, PTR_ERR(subdir));
1011 }
1012 drop_sysctl_table(&dir->header);
1013 if (new)
1014 drop_sysctl_table(&new->header);
1015 spin_unlock(&sysctl_lock);
1016 return subdir;
1017 }
1018
1019 static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
1020 {
1021 struct ctl_dir *parent;
1022 const char *procname;
1023 if (!dir->header.parent)
1024 return &set->dir;
1025 parent = xlate_dir(set, dir->header.parent);
1026 if (IS_ERR(parent))
1027 return parent;
1028 procname = dir->header.ctl_table[0].procname;
1029 return find_subdir(parent, procname, strlen(procname));
1030 }
1031
1032 static int sysctl_follow_link(struct ctl_table_header **phead,
1033 struct ctl_table **pentry)
1034 {
1035 struct ctl_table_header *head;
1036 struct ctl_table_root *root;
1037 struct ctl_table_set *set;
1038 struct ctl_table *entry;
1039 struct ctl_dir *dir;
1040 int ret;
1041
1042 ret = 0;
1043 spin_lock(&sysctl_lock);
1044 root = (*pentry)->data;
1045 set = lookup_header_set(root);
1046 dir = xlate_dir(set, (*phead)->parent);
1047 if (IS_ERR(dir))
1048 ret = PTR_ERR(dir);
1049 else {
1050 const char *procname = (*pentry)->procname;
1051 head = NULL;
1052 entry = find_entry(&head, dir, procname, strlen(procname));
1053 ret = -ENOENT;
1054 if (entry && use_table(head)) {
1055 unuse_table(*phead);
1056 *phead = head;
1057 *pentry = entry;
1058 ret = 0;
1059 }
1060 }
1061
1062 spin_unlock(&sysctl_lock);
1063 return ret;
1064 }
1065
1066 static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
1067 {
1068 struct va_format vaf;
1069 va_list args;
1070
1071 va_start(args, fmt);
1072 vaf.fmt = fmt;
1073 vaf.va = &args;
1074
1075 pr_err("sysctl table check failed: %s/%s %pV\n",
1076 path, table->procname, &vaf);
1077
1078 va_end(args);
1079 return -EINVAL;
1080 }
1081
1082 static int sysctl_check_table_array(const char *path, struct ctl_table *table)
1083 {
1084 int err = 0;
1085
1086 if ((table->proc_handler == proc_douintvec) ||
1087 (table->proc_handler == proc_douintvec_minmax)) {
1088 if (table->maxlen != sizeof(unsigned int))
1089 err |= sysctl_err(path, table, "array not allowed");
1090 }
1091
1092 return err;
1093 }
1094
1095 static int sysctl_check_table(const char *path, struct ctl_table *table)
1096 {
1097 int err = 0;
1098 for (; table->procname; table++) {
1099 if (table->child)
1100 err |= sysctl_err(path, table, "Not a file");
1101
1102 if ((table->proc_handler == proc_dostring) ||
1103 (table->proc_handler == proc_dointvec) ||
1104 (table->proc_handler == proc_douintvec) ||
1105 (table->proc_handler == proc_douintvec_minmax) ||
1106 (table->proc_handler == proc_dointvec_minmax) ||
1107 (table->proc_handler == proc_dointvec_jiffies) ||
1108 (table->proc_handler == proc_dointvec_userhz_jiffies) ||
1109 (table->proc_handler == proc_dointvec_ms_jiffies) ||
1110 (table->proc_handler == proc_doulongvec_minmax) ||
1111 (table->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1112 if (!table->data)
1113 err |= sysctl_err(path, table, "No data");
1114 if (!table->maxlen)
1115 err |= sysctl_err(path, table, "No maxlen");
1116 else
1117 err |= sysctl_check_table_array(path, table);
1118 }
1119 if (!table->proc_handler)
1120 err |= sysctl_err(path, table, "No proc_handler");
1121
1122 if ((table->mode & (S_IRUGO|S_IWUGO)) != table->mode)
1123 err |= sysctl_err(path, table, "bogus .mode 0%o",
1124 table->mode);
1125 }
1126 return err;
1127 }
1128
1129 static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table *table,
1130 struct ctl_table_root *link_root)
1131 {
1132 struct ctl_table *link_table, *entry, *link;
1133 struct ctl_table_header *links;
1134 struct ctl_node *node;
1135 char *link_name;
1136 int nr_entries, name_bytes;
1137
1138 name_bytes = 0;
1139 nr_entries = 0;
1140 for (entry = table; entry->procname; entry++) {
1141 nr_entries++;
1142 name_bytes += strlen(entry->procname) + 1;
1143 }
1144
1145 links = kzalloc(sizeof(struct ctl_table_header) +
1146 sizeof(struct ctl_node)*nr_entries +
1147 sizeof(struct ctl_table)*(nr_entries + 1) +
1148 name_bytes,
1149 GFP_KERNEL);
1150
1151 if (!links)
1152 return NULL;
1153
1154 node = (struct ctl_node *)(links + 1);
1155 link_table = (struct ctl_table *)(node + nr_entries);
1156 link_name = (char *)&link_table[nr_entries + 1];
1157
1158 for (link = link_table, entry = table; entry->procname; link++, entry++) {
1159 int len = strlen(entry->procname) + 1;
1160 memcpy(link_name, entry->procname, len);
1161 link->procname = link_name;
1162 link->mode = S_IFLNK|S_IRWXUGO;
1163 link->data = link_root;
1164 link_name += len;
1165 }
1166 init_header(links, dir->header.root, dir->header.set, node, link_table);
1167 links->nreg = nr_entries;
1168
1169 return links;
1170 }
1171
1172 static bool get_links(struct ctl_dir *dir,
1173 struct ctl_table *table, struct ctl_table_root *link_root)
1174 {
1175 struct ctl_table_header *head;
1176 struct ctl_table *entry, *link;
1177
1178 /* Are there links available for every entry in table? */
1179 for (entry = table; entry->procname; entry++) {
1180 const char *procname = entry->procname;
1181 link = find_entry(&head, dir, procname, strlen(procname));
1182 if (!link)
1183 return false;
1184 if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1185 continue;
1186 if (S_ISLNK(link->mode) && (link->data == link_root))
1187 continue;
1188 return false;
1189 }
1190
1191 /* The checks passed. Increase the registration count on the links */
1192 for (entry = table; entry->procname; entry++) {
1193 const char *procname = entry->procname;
1194 link = find_entry(&head, dir, procname, strlen(procname));
1195 head->nreg++;
1196 }
1197 return true;
1198 }
1199
1200 static int insert_links(struct ctl_table_header *head)
1201 {
1202 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1203 struct ctl_dir *core_parent = NULL;
1204 struct ctl_table_header *links;
1205 int err;
1206
1207 if (head->set == root_set)
1208 return 0;
1209
1210 core_parent = xlate_dir(root_set, head->parent);
1211 if (IS_ERR(core_parent))
1212 return 0;
1213
1214 if (get_links(core_parent, head->ctl_table, head->root))
1215 return 0;
1216
1217 core_parent->header.nreg++;
1218 spin_unlock(&sysctl_lock);
1219
1220 links = new_links(core_parent, head->ctl_table, head->root);
1221
1222 spin_lock(&sysctl_lock);
1223 err = -ENOMEM;
1224 if (!links)
1225 goto out;
1226
1227 err = 0;
1228 if (get_links(core_parent, head->ctl_table, head->root)) {
1229 kfree(links);
1230 goto out;
1231 }
1232
1233 err = insert_header(core_parent, links);
1234 if (err)
1235 kfree(links);
1236 out:
1237 drop_sysctl_table(&core_parent->header);
1238 return err;
1239 }
1240
1241 /**
1242 * __register_sysctl_table - register a leaf sysctl table
1243 * @set: Sysctl tree to register on
1244 * @path: The path to the directory the sysctl table is in.
1245 * @table: the top-level table structure
1246 *
1247 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1248 * array. A completely 0 filled entry terminates the table.
1249 *
1250 * The members of the &struct ctl_table structure are used as follows:
1251 *
1252 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1253 * enter a sysctl file
1254 *
1255 * data - a pointer to data for use by proc_handler
1256 *
1257 * maxlen - the maximum size in bytes of the data
1258 *
1259 * mode - the file permissions for the /proc/sys file
1260 *
1261 * child - must be %NULL.
1262 *
1263 * proc_handler - the text handler routine (described below)
1264 *
1265 * extra1, extra2 - extra pointers usable by the proc handler routines
1266 *
1267 * Leaf nodes in the sysctl tree will be represented by a single file
1268 * under /proc; non-leaf nodes will be represented by directories.
1269 *
1270 * There must be a proc_handler routine for any terminal nodes.
1271 * Several default handlers are available to cover common cases -
1272 *
1273 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1274 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1275 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1276 *
1277 * It is the handler's job to read the input buffer from user memory
1278 * and process it. The handler should return 0 on success.
1279 *
1280 * This routine returns %NULL on a failure to register, and a pointer
1281 * to the table header on success.
1282 */
1283 struct ctl_table_header *__register_sysctl_table(
1284 struct ctl_table_set *set,
1285 const char *path, struct ctl_table *table)
1286 {
1287 struct ctl_table_root *root = set->dir.header.root;
1288 struct ctl_table_header *header;
1289 const char *name, *nextname;
1290 struct ctl_dir *dir;
1291 struct ctl_table *entry;
1292 struct ctl_node *node;
1293 int nr_entries = 0;
1294
1295 for (entry = table; entry->procname; entry++)
1296 nr_entries++;
1297
1298 header = kzalloc(sizeof(struct ctl_table_header) +
1299 sizeof(struct ctl_node)*nr_entries, GFP_KERNEL);
1300 if (!header)
1301 return NULL;
1302
1303 node = (struct ctl_node *)(header + 1);
1304 init_header(header, root, set, node, table);
1305 if (sysctl_check_table(path, table))
1306 goto fail;
1307
1308 spin_lock(&sysctl_lock);
1309 dir = &set->dir;
1310 /* Reference moved down the diretory tree get_subdir */
1311 dir->header.nreg++;
1312 spin_unlock(&sysctl_lock);
1313
1314 /* Find the directory for the ctl_table */
1315 for (name = path; name; name = nextname) {
1316 int namelen;
1317 nextname = strchr(name, '/');
1318 if (nextname) {
1319 namelen = nextname - name;
1320 nextname++;
1321 } else {
1322 namelen = strlen(name);
1323 }
1324 if (namelen == 0)
1325 continue;
1326
1327 dir = get_subdir(dir, name, namelen);
1328 if (IS_ERR(dir))
1329 goto fail;
1330 }
1331
1332 spin_lock(&sysctl_lock);
1333 if (insert_header(dir, header))
1334 goto fail_put_dir_locked;
1335
1336 drop_sysctl_table(&dir->header);
1337 spin_unlock(&sysctl_lock);
1338
1339 return header;
1340
1341 fail_put_dir_locked:
1342 drop_sysctl_table(&dir->header);
1343 spin_unlock(&sysctl_lock);
1344 fail:
1345 kfree(header);
1346 dump_stack();
1347 return NULL;
1348 }
1349
1350 /**
1351 * register_sysctl - register a sysctl table
1352 * @path: The path to the directory the sysctl table is in.
1353 * @table: the table structure
1354 *
1355 * Register a sysctl table. @table should be a filled in ctl_table
1356 * array. A completely 0 filled entry terminates the table.
1357 *
1358 * See __register_sysctl_table for more details.
1359 */
1360 struct ctl_table_header *register_sysctl(const char *path, struct ctl_table *table)
1361 {
1362 return __register_sysctl_table(&sysctl_table_root.default_set,
1363 path, table);
1364 }
1365 EXPORT_SYMBOL(register_sysctl);
1366
1367 static char *append_path(const char *path, char *pos, const char *name)
1368 {
1369 int namelen;
1370 namelen = strlen(name);
1371 if (((pos - path) + namelen + 2) >= PATH_MAX)
1372 return NULL;
1373 memcpy(pos, name, namelen);
1374 pos[namelen] = '/';
1375 pos[namelen + 1] = '\0';
1376 pos += namelen + 1;
1377 return pos;
1378 }
1379
1380 static int count_subheaders(struct ctl_table *table)
1381 {
1382 int has_files = 0;
1383 int nr_subheaders = 0;
1384 struct ctl_table *entry;
1385
1386 /* special case: no directory and empty directory */
1387 if (!table || !table->procname)
1388 return 1;
1389
1390 for (entry = table; entry->procname; entry++) {
1391 if (entry->child)
1392 nr_subheaders += count_subheaders(entry->child);
1393 else
1394 has_files = 1;
1395 }
1396 return nr_subheaders + has_files;
1397 }
1398
1399 static int register_leaf_sysctl_tables(const char *path, char *pos,
1400 struct ctl_table_header ***subheader, struct ctl_table_set *set,
1401 struct ctl_table *table)
1402 {
1403 struct ctl_table *ctl_table_arg = NULL;
1404 struct ctl_table *entry, *files;
1405 int nr_files = 0;
1406 int nr_dirs = 0;
1407 int err = -ENOMEM;
1408
1409 for (entry = table; entry->procname; entry++) {
1410 if (entry->child)
1411 nr_dirs++;
1412 else
1413 nr_files++;
1414 }
1415
1416 files = table;
1417 /* If there are mixed files and directories we need a new table */
1418 if (nr_dirs && nr_files) {
1419 struct ctl_table *new;
1420 files = kcalloc(nr_files + 1, sizeof(struct ctl_table),
1421 GFP_KERNEL);
1422 if (!files)
1423 goto out;
1424
1425 ctl_table_arg = files;
1426 for (new = files, entry = table; entry->procname; entry++) {
1427 if (entry->child)
1428 continue;
1429 *new = *entry;
1430 new++;
1431 }
1432 }
1433
1434 /* Register everything except a directory full of subdirectories */
1435 if (nr_files || !nr_dirs) {
1436 struct ctl_table_header *header;
1437 header = __register_sysctl_table(set, path, files);
1438 if (!header) {
1439 kfree(ctl_table_arg);
1440 goto out;
1441 }
1442
1443 /* Remember if we need to free the file table */
1444 header->ctl_table_arg = ctl_table_arg;
1445 **subheader = header;
1446 (*subheader)++;
1447 }
1448
1449 /* Recurse into the subdirectories. */
1450 for (entry = table; entry->procname; entry++) {
1451 char *child_pos;
1452
1453 if (!entry->child)
1454 continue;
1455
1456 err = -ENAMETOOLONG;
1457 child_pos = append_path(path, pos, entry->procname);
1458 if (!child_pos)
1459 goto out;
1460
1461 err = register_leaf_sysctl_tables(path, child_pos, subheader,
1462 set, entry->child);
1463 pos[0] = '\0';
1464 if (err)
1465 goto out;
1466 }
1467 err = 0;
1468 out:
1469 /* On failure our caller will unregister all registered subheaders */
1470 return err;
1471 }
1472
1473 /**
1474 * __register_sysctl_paths - register a sysctl table hierarchy
1475 * @set: Sysctl tree to register on
1476 * @path: The path to the directory the sysctl table is in.
1477 * @table: the top-level table structure
1478 *
1479 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1480 * array. A completely 0 filled entry terminates the table.
1481 *
1482 * See __register_sysctl_table for more details.
1483 */
1484 struct ctl_table_header *__register_sysctl_paths(
1485 struct ctl_table_set *set,
1486 const struct ctl_path *path, struct ctl_table *table)
1487 {
1488 struct ctl_table *ctl_table_arg = table;
1489 int nr_subheaders = count_subheaders(table);
1490 struct ctl_table_header *header = NULL, **subheaders, **subheader;
1491 const struct ctl_path *component;
1492 char *new_path, *pos;
1493
1494 pos = new_path = kmalloc(PATH_MAX, GFP_KERNEL);
1495 if (!new_path)
1496 return NULL;
1497
1498 pos[0] = '\0';
1499 for (component = path; component->procname; component++) {
1500 pos = append_path(new_path, pos, component->procname);
1501 if (!pos)
1502 goto out;
1503 }
1504 while (table->procname && table->child && !table[1].procname) {
1505 pos = append_path(new_path, pos, table->procname);
1506 if (!pos)
1507 goto out;
1508 table = table->child;
1509 }
1510 if (nr_subheaders == 1) {
1511 header = __register_sysctl_table(set, new_path, table);
1512 if (header)
1513 header->ctl_table_arg = ctl_table_arg;
1514 } else {
1515 header = kzalloc(sizeof(*header) +
1516 sizeof(*subheaders)*nr_subheaders, GFP_KERNEL);
1517 if (!header)
1518 goto out;
1519
1520 subheaders = (struct ctl_table_header **) (header + 1);
1521 subheader = subheaders;
1522 header->ctl_table_arg = ctl_table_arg;
1523
1524 if (register_leaf_sysctl_tables(new_path, pos, &subheader,
1525 set, table))
1526 goto err_register_leaves;
1527 }
1528
1529 out:
1530 kfree(new_path);
1531 return header;
1532
1533 err_register_leaves:
1534 while (subheader > subheaders) {
1535 struct ctl_table_header *subh = *(--subheader);
1536 struct ctl_table *table = subh->ctl_table_arg;
1537 unregister_sysctl_table(subh);
1538 kfree(table);
1539 }
1540 kfree(header);
1541 header = NULL;
1542 goto out;
1543 }
1544
1545 /**
1546 * register_sysctl_table_path - register a sysctl table hierarchy
1547 * @path: The path to the directory the sysctl table is in.
1548 * @table: the top-level table structure
1549 *
1550 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1551 * array. A completely 0 filled entry terminates the table.
1552 *
1553 * See __register_sysctl_paths for more details.
1554 */
1555 struct ctl_table_header *register_sysctl_paths(const struct ctl_path *path,
1556 struct ctl_table *table)
1557 {
1558 return __register_sysctl_paths(&sysctl_table_root.default_set,
1559 path, table);
1560 }
1561 EXPORT_SYMBOL(register_sysctl_paths);
1562
1563 /**
1564 * register_sysctl_table - register a sysctl table hierarchy
1565 * @table: the top-level table structure
1566 *
1567 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1568 * array. A completely 0 filled entry terminates the table.
1569 *
1570 * See register_sysctl_paths for more details.
1571 */
1572 struct ctl_table_header *register_sysctl_table(struct ctl_table *table)
1573 {
1574 static const struct ctl_path null_path[] = { {} };
1575
1576 return register_sysctl_paths(null_path, table);
1577 }
1578 EXPORT_SYMBOL(register_sysctl_table);
1579
1580 static void put_links(struct ctl_table_header *header)
1581 {
1582 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1583 struct ctl_table_root *root = header->root;
1584 struct ctl_dir *parent = header->parent;
1585 struct ctl_dir *core_parent;
1586 struct ctl_table *entry;
1587
1588 if (header->set == root_set)
1589 return;
1590
1591 core_parent = xlate_dir(root_set, parent);
1592 if (IS_ERR(core_parent))
1593 return;
1594
1595 for (entry = header->ctl_table; entry->procname; entry++) {
1596 struct ctl_table_header *link_head;
1597 struct ctl_table *link;
1598 const char *name = entry->procname;
1599
1600 link = find_entry(&link_head, core_parent, name, strlen(name));
1601 if (link &&
1602 ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1603 (S_ISLNK(link->mode) && (link->data == root)))) {
1604 drop_sysctl_table(link_head);
1605 }
1606 else {
1607 pr_err("sysctl link missing during unregister: ");
1608 sysctl_print_dir(parent);
1609 pr_cont("/%s\n", name);
1610 }
1611 }
1612 }
1613
1614 static void drop_sysctl_table(struct ctl_table_header *header)
1615 {
1616 struct ctl_dir *parent = header->parent;
1617
1618 if (--header->nreg)
1619 return;
1620
1621 if (parent) {
1622 put_links(header);
1623 start_unregistering(header);
1624 }
1625
1626 if (!--header->count)
1627 kfree_rcu(header, rcu);
1628
1629 if (parent)
1630 drop_sysctl_table(&parent->header);
1631 }
1632
1633 /**
1634 * unregister_sysctl_table - unregister a sysctl table hierarchy
1635 * @header: the header returned from register_sysctl_table
1636 *
1637 * Unregisters the sysctl table and all children. proc entries may not
1638 * actually be removed until they are no longer used by anyone.
1639 */
1640 void unregister_sysctl_table(struct ctl_table_header * header)
1641 {
1642 int nr_subheaders;
1643 might_sleep();
1644
1645 if (header == NULL)
1646 return;
1647
1648 nr_subheaders = count_subheaders(header->ctl_table_arg);
1649 if (unlikely(nr_subheaders > 1)) {
1650 struct ctl_table_header **subheaders;
1651 int i;
1652
1653 subheaders = (struct ctl_table_header **)(header + 1);
1654 for (i = nr_subheaders -1; i >= 0; i--) {
1655 struct ctl_table_header *subh = subheaders[i];
1656 struct ctl_table *table = subh->ctl_table_arg;
1657 unregister_sysctl_table(subh);
1658 kfree(table);
1659 }
1660 kfree(header);
1661 return;
1662 }
1663
1664 spin_lock(&sysctl_lock);
1665 drop_sysctl_table(header);
1666 spin_unlock(&sysctl_lock);
1667 }
1668 EXPORT_SYMBOL(unregister_sysctl_table);
1669
1670 void setup_sysctl_set(struct ctl_table_set *set,
1671 struct ctl_table_root *root,
1672 int (*is_seen)(struct ctl_table_set *))
1673 {
1674 memset(set, 0, sizeof(*set));
1675 set->is_seen = is_seen;
1676 init_header(&set->dir.header, root, set, NULL, root_table);
1677 }
1678
1679 void retire_sysctl_set(struct ctl_table_set *set)
1680 {
1681 WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1682 }
1683
1684 int __init proc_sys_init(void)
1685 {
1686 struct proc_dir_entry *proc_sys_root;
1687
1688 proc_sys_root = proc_mkdir("sys", NULL);
1689 proc_sys_root->proc_iops = &proc_sys_dir_operations;
1690 proc_sys_root->proc_dir_ops = &proc_sys_dir_file_operations;
1691 proc_sys_root->nlink = 0;
1692
1693 return sysctl_init();
1694 }
Linux with added FPC-III support