Loading...
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/namei.h>
12#include <linux/mm.h>
13#include <linux/module.h>
14#include "internal.h"
15
16static const struct dentry_operations proc_sys_dentry_operations;
17static const struct file_operations proc_sys_file_operations;
18static const struct inode_operations proc_sys_inode_operations;
19static const struct file_operations proc_sys_dir_file_operations;
20static const struct inode_operations proc_sys_dir_operations;
21
22/* Support for permanently empty directories */
23
24struct ctl_table sysctl_mount_point[] = {
25 { }
26};
27
28static bool is_empty_dir(struct ctl_table_header *head)
29{
30 return head->ctl_table[0].child == sysctl_mount_point;
31}
32
33static void set_empty_dir(struct ctl_dir *dir)
34{
35 dir->header.ctl_table[0].child = sysctl_mount_point;
36}
37
38static void clear_empty_dir(struct ctl_dir *dir)
39
40{
41 dir->header.ctl_table[0].child = NULL;
42}
43
44void proc_sys_poll_notify(struct ctl_table_poll *poll)
45{
46 if (!poll)
47 return;
48
49 atomic_inc(&poll->event);
50 wake_up_interruptible(&poll->wait);
51}
52
53static struct ctl_table root_table[] = {
54 {
55 .procname = "",
56 .mode = S_IFDIR|S_IRUGO|S_IXUGO,
57 },
58 { }
59};
60static struct ctl_table_root sysctl_table_root = {
61 .default_set.dir.header = {
62 {{.count = 1,
63 .nreg = 1,
64 .ctl_table = root_table }},
65 .ctl_table_arg = root_table,
66 .root = &sysctl_table_root,
67 .set = &sysctl_table_root.default_set,
68 },
69};
70
71static DEFINE_SPINLOCK(sysctl_lock);
72
73static void drop_sysctl_table(struct ctl_table_header *header);
74static int sysctl_follow_link(struct ctl_table_header **phead,
75 struct ctl_table **pentry, struct nsproxy *namespaces);
76static int insert_links(struct ctl_table_header *head);
77static void put_links(struct ctl_table_header *header);
78
79static void sysctl_print_dir(struct ctl_dir *dir)
80{
81 if (dir->header.parent)
82 sysctl_print_dir(dir->header.parent);
83 pr_cont("%s/", dir->header.ctl_table[0].procname);
84}
85
86static int namecmp(const char *name1, int len1, const char *name2, int len2)
87{
88 int minlen;
89 int cmp;
90
91 minlen = len1;
92 if (minlen > len2)
93 minlen = len2;
94
95 cmp = memcmp(name1, name2, minlen);
96 if (cmp == 0)
97 cmp = len1 - len2;
98 return cmp;
99}
100
101/* Called under sysctl_lock */
102static struct ctl_table *find_entry(struct ctl_table_header **phead,
103 struct ctl_dir *dir, const char *name, int namelen)
104{
105 struct ctl_table_header *head;
106 struct ctl_table *entry;
107 struct rb_node *node = dir->root.rb_node;
108
109 while (node)
110 {
111 struct ctl_node *ctl_node;
112 const char *procname;
113 int cmp;
114
115 ctl_node = rb_entry(node, struct ctl_node, node);
116 head = ctl_node->header;
117 entry = &head->ctl_table[ctl_node - head->node];
118 procname = entry->procname;
119
120 cmp = namecmp(name, namelen, procname, strlen(procname));
121 if (cmp < 0)
122 node = node->rb_left;
123 else if (cmp > 0)
124 node = node->rb_right;
125 else {
126 *phead = head;
127 return entry;
128 }
129 }
130 return NULL;
131}
132
133static int insert_entry(struct ctl_table_header *head, struct ctl_table *entry)
134{
135 struct rb_node *node = &head->node[entry - head->ctl_table].node;
136 struct rb_node **p = &head->parent->root.rb_node;
137 struct rb_node *parent = NULL;
138 const char *name = entry->procname;
139 int namelen = strlen(name);
140
141 while (*p) {
142 struct ctl_table_header *parent_head;
143 struct ctl_table *parent_entry;
144 struct ctl_node *parent_node;
145 const char *parent_name;
146 int cmp;
147
148 parent = *p;
149 parent_node = rb_entry(parent, struct ctl_node, node);
150 parent_head = parent_node->header;
151 parent_entry = &parent_head->ctl_table[parent_node - parent_head->node];
152 parent_name = parent_entry->procname;
153
154 cmp = namecmp(name, namelen, parent_name, strlen(parent_name));
155 if (cmp < 0)
156 p = &(*p)->rb_left;
157 else if (cmp > 0)
158 p = &(*p)->rb_right;
159 else {
160 pr_err("sysctl duplicate entry: ");
161 sysctl_print_dir(head->parent);
162 pr_cont("/%s\n", entry->procname);
163 return -EEXIST;
164 }
165 }
166
167 rb_link_node(node, parent, p);
168 rb_insert_color(node, &head->parent->root);
169 return 0;
170}
171
172static void erase_entry(struct ctl_table_header *head, struct ctl_table *entry)
173{
174 struct rb_node *node = &head->node[entry - head->ctl_table].node;
175
176 rb_erase(node, &head->parent->root);
177}
178
179static void init_header(struct ctl_table_header *head,
180 struct ctl_table_root *root, struct ctl_table_set *set,
181 struct ctl_node *node, struct ctl_table *table)
182{
183 head->ctl_table = table;
184 head->ctl_table_arg = table;
185 head->used = 0;
186 head->count = 1;
187 head->nreg = 1;
188 head->unregistering = NULL;
189 head->root = root;
190 head->set = set;
191 head->parent = NULL;
192 head->node = node;
193 if (node) {
194 struct ctl_table *entry;
195 for (entry = table; entry->procname; entry++, node++)
196 node->header = head;
197 }
198}
199
200static void erase_header(struct ctl_table_header *head)
201{
202 struct ctl_table *entry;
203 for (entry = head->ctl_table; entry->procname; entry++)
204 erase_entry(head, entry);
205}
206
207static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
208{
209 struct ctl_table *entry;
210 int err;
211
212 /* Is this a permanently empty directory? */
213 if (is_empty_dir(&dir->header))
214 return -EROFS;
215
216 /* Am I creating a permanently empty directory? */
217 if (header->ctl_table == sysctl_mount_point) {
218 if (!RB_EMPTY_ROOT(&dir->root))
219 return -EINVAL;
220 set_empty_dir(dir);
221 }
222
223 dir->header.nreg++;
224 header->parent = dir;
225 err = insert_links(header);
226 if (err)
227 goto fail_links;
228 for (entry = header->ctl_table; entry->procname; entry++) {
229 err = insert_entry(header, entry);
230 if (err)
231 goto fail;
232 }
233 return 0;
234fail:
235 erase_header(header);
236 put_links(header);
237fail_links:
238 if (header->ctl_table == sysctl_mount_point)
239 clear_empty_dir(dir);
240 header->parent = NULL;
241 drop_sysctl_table(&dir->header);
242 return err;
243}
244
245/* called under sysctl_lock */
246static int use_table(struct ctl_table_header *p)
247{
248 if (unlikely(p->unregistering))
249 return 0;
250 p->used++;
251 return 1;
252}
253
254/* called under sysctl_lock */
255static void unuse_table(struct ctl_table_header *p)
256{
257 if (!--p->used)
258 if (unlikely(p->unregistering))
259 complete(p->unregistering);
260}
261
262/* called under sysctl_lock, will reacquire if has to wait */
263static void start_unregistering(struct ctl_table_header *p)
264{
265 /*
266 * if p->used is 0, nobody will ever touch that entry again;
267 * we'll eliminate all paths to it before dropping sysctl_lock
268 */
269 if (unlikely(p->used)) {
270 struct completion wait;
271 init_completion(&wait);
272 p->unregistering = &wait;
273 spin_unlock(&sysctl_lock);
274 wait_for_completion(&wait);
275 spin_lock(&sysctl_lock);
276 } else {
277 /* anything non-NULL; we'll never dereference it */
278 p->unregistering = ERR_PTR(-EINVAL);
279 }
280 /*
281 * do not remove from the list until nobody holds it; walking the
282 * list in do_sysctl() relies on that.
283 */
284 erase_header(p);
285}
286
287static void sysctl_head_get(struct ctl_table_header *head)
288{
289 spin_lock(&sysctl_lock);
290 head->count++;
291 spin_unlock(&sysctl_lock);
292}
293
294void sysctl_head_put(struct ctl_table_header *head)
295{
296 spin_lock(&sysctl_lock);
297 if (!--head->count)
298 kfree_rcu(head, rcu);
299 spin_unlock(&sysctl_lock);
300}
301
302static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
303{
304 BUG_ON(!head);
305 spin_lock(&sysctl_lock);
306 if (!use_table(head))
307 head = ERR_PTR(-ENOENT);
308 spin_unlock(&sysctl_lock);
309 return head;
310}
311
312static void sysctl_head_finish(struct ctl_table_header *head)
313{
314 if (!head)
315 return;
316 spin_lock(&sysctl_lock);
317 unuse_table(head);
318 spin_unlock(&sysctl_lock);
319}
320
321static struct ctl_table_set *
322lookup_header_set(struct ctl_table_root *root, struct nsproxy *namespaces)
323{
324 struct ctl_table_set *set = &root->default_set;
325 if (root->lookup)
326 set = root->lookup(root, namespaces);
327 return set;
328}
329
330static struct ctl_table *lookup_entry(struct ctl_table_header **phead,
331 struct ctl_dir *dir,
332 const char *name, int namelen)
333{
334 struct ctl_table_header *head;
335 struct ctl_table *entry;
336
337 spin_lock(&sysctl_lock);
338 entry = find_entry(&head, dir, name, namelen);
339 if (entry && use_table(head))
340 *phead = head;
341 else
342 entry = NULL;
343 spin_unlock(&sysctl_lock);
344 return entry;
345}
346
347static struct ctl_node *first_usable_entry(struct rb_node *node)
348{
349 struct ctl_node *ctl_node;
350
351 for (;node; node = rb_next(node)) {
352 ctl_node = rb_entry(node, struct ctl_node, node);
353 if (use_table(ctl_node->header))
354 return ctl_node;
355 }
356 return NULL;
357}
358
359static void first_entry(struct ctl_dir *dir,
360 struct ctl_table_header **phead, struct ctl_table **pentry)
361{
362 struct ctl_table_header *head = NULL;
363 struct ctl_table *entry = NULL;
364 struct ctl_node *ctl_node;
365
366 spin_lock(&sysctl_lock);
367 ctl_node = first_usable_entry(rb_first(&dir->root));
368 spin_unlock(&sysctl_lock);
369 if (ctl_node) {
370 head = ctl_node->header;
371 entry = &head->ctl_table[ctl_node - head->node];
372 }
373 *phead = head;
374 *pentry = entry;
375}
376
377static void next_entry(struct ctl_table_header **phead, struct ctl_table **pentry)
378{
379 struct ctl_table_header *head = *phead;
380 struct ctl_table *entry = *pentry;
381 struct ctl_node *ctl_node = &head->node[entry - head->ctl_table];
382
383 spin_lock(&sysctl_lock);
384 unuse_table(head);
385
386 ctl_node = first_usable_entry(rb_next(&ctl_node->node));
387 spin_unlock(&sysctl_lock);
388 head = NULL;
389 if (ctl_node) {
390 head = ctl_node->header;
391 entry = &head->ctl_table[ctl_node - head->node];
392 }
393 *phead = head;
394 *pentry = entry;
395}
396
397void register_sysctl_root(struct ctl_table_root *root)
398{
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
406static 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
417static 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
430static struct inode *proc_sys_make_inode(struct super_block *sb,
431 struct ctl_table_header *head, struct ctl_table *table)
432{
433 struct inode *inode;
434 struct proc_inode *ei;
435
436 inode = new_inode(sb);
437 if (!inode)
438 goto out;
439
440 inode->i_ino = get_next_ino();
441
442 sysctl_head_get(head);
443 ei = PROC_I(inode);
444 ei->sysctl = head;
445 ei->sysctl_entry = table;
446
447 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
448 inode->i_mode = table->mode;
449 if (!S_ISDIR(table->mode)) {
450 inode->i_mode |= S_IFREG;
451 inode->i_op = &proc_sys_inode_operations;
452 inode->i_fop = &proc_sys_file_operations;
453 } else {
454 inode->i_mode |= S_IFDIR;
455 inode->i_op = &proc_sys_dir_operations;
456 inode->i_fop = &proc_sys_dir_file_operations;
457 if (is_empty_dir(head))
458 make_empty_dir_inode(inode);
459 }
460out:
461 return inode;
462}
463
464static struct ctl_table_header *grab_header(struct inode *inode)
465{
466 struct ctl_table_header *head = PROC_I(inode)->sysctl;
467 if (!head)
468 head = &sysctl_table_root.default_set.dir.header;
469 return sysctl_head_grab(head);
470}
471
472static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
473 unsigned int flags)
474{
475 struct ctl_table_header *head = grab_header(dir);
476 struct ctl_table_header *h = NULL;
477 struct qstr *name = &dentry->d_name;
478 struct ctl_table *p;
479 struct inode *inode;
480 struct dentry *err = ERR_PTR(-ENOENT);
481 struct ctl_dir *ctl_dir;
482 int ret;
483
484 if (IS_ERR(head))
485 return ERR_CAST(head);
486
487 ctl_dir = container_of(head, struct ctl_dir, header);
488
489 p = lookup_entry(&h, ctl_dir, name->name, name->len);
490 if (!p)
491 goto out;
492
493 if (S_ISLNK(p->mode)) {
494 ret = sysctl_follow_link(&h, &p, current->nsproxy);
495 err = ERR_PTR(ret);
496 if (ret)
497 goto out;
498 }
499
500 err = ERR_PTR(-ENOMEM);
501 inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
502 if (!inode)
503 goto out;
504
505 err = NULL;
506 d_set_d_op(dentry, &proc_sys_dentry_operations);
507 d_add(dentry, inode);
508
509out:
510 if (h)
511 sysctl_head_finish(h);
512 sysctl_head_finish(head);
513 return err;
514}
515
516static ssize_t proc_sys_call_handler(struct file *filp, void __user *buf,
517 size_t count, loff_t *ppos, int write)
518{
519 struct inode *inode = file_inode(filp);
520 struct ctl_table_header *head = grab_header(inode);
521 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
522 ssize_t error;
523 size_t res;
524
525 if (IS_ERR(head))
526 return PTR_ERR(head);
527
528 /*
529 * At this point we know that the sysctl was not unregistered
530 * and won't be until we finish.
531 */
532 error = -EPERM;
533 if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
534 goto out;
535
536 /* if that can happen at all, it should be -EINVAL, not -EISDIR */
537 error = -EINVAL;
538 if (!table->proc_handler)
539 goto out;
540
541 /* careful: calling conventions are nasty here */
542 res = count;
543 error = table->proc_handler(table, write, buf, &res, ppos);
544 if (!error)
545 error = res;
546out:
547 sysctl_head_finish(head);
548
549 return error;
550}
551
552static ssize_t proc_sys_read(struct file *filp, char __user *buf,
553 size_t count, loff_t *ppos)
554{
555 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 0);
556}
557
558static ssize_t proc_sys_write(struct file *filp, const char __user *buf,
559 size_t count, loff_t *ppos)
560{
561 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 1);
562}
563
564static int proc_sys_open(struct inode *inode, struct file *filp)
565{
566 struct ctl_table_header *head = grab_header(inode);
567 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
568
569 /* sysctl was unregistered */
570 if (IS_ERR(head))
571 return PTR_ERR(head);
572
573 if (table->poll)
574 filp->private_data = proc_sys_poll_event(table->poll);
575
576 sysctl_head_finish(head);
577
578 return 0;
579}
580
581static unsigned int proc_sys_poll(struct file *filp, poll_table *wait)
582{
583 struct inode *inode = file_inode(filp);
584 struct ctl_table_header *head = grab_header(inode);
585 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
586 unsigned int ret = DEFAULT_POLLMASK;
587 unsigned long event;
588
589 /* sysctl was unregistered */
590 if (IS_ERR(head))
591 return POLLERR | POLLHUP;
592
593 if (!table->proc_handler)
594 goto out;
595
596 if (!table->poll)
597 goto out;
598
599 event = (unsigned long)filp->private_data;
600 poll_wait(filp, &table->poll->wait, wait);
601
602 if (event != atomic_read(&table->poll->event)) {
603 filp->private_data = proc_sys_poll_event(table->poll);
604 ret = POLLIN | POLLRDNORM | POLLERR | POLLPRI;
605 }
606
607out:
608 sysctl_head_finish(head);
609
610 return ret;
611}
612
613static bool proc_sys_fill_cache(struct file *file,
614 struct dir_context *ctx,
615 struct ctl_table_header *head,
616 struct ctl_table *table)
617{
618 struct dentry *child, *dir = file->f_path.dentry;
619 struct inode *inode;
620 struct qstr qname;
621 ino_t ino = 0;
622 unsigned type = DT_UNKNOWN;
623
624 qname.name = table->procname;
625 qname.len = strlen(table->procname);
626 qname.hash = full_name_hash(qname.name, qname.len);
627
628 child = d_lookup(dir, &qname);
629 if (!child) {
630 child = d_alloc(dir, &qname);
631 if (child) {
632 inode = proc_sys_make_inode(dir->d_sb, head, table);
633 if (!inode) {
634 dput(child);
635 return false;
636 } else {
637 d_set_d_op(child, &proc_sys_dentry_operations);
638 d_add(child, inode);
639 }
640 } else {
641 return false;
642 }
643 }
644 inode = d_inode(child);
645 ino = inode->i_ino;
646 type = inode->i_mode >> 12;
647 dput(child);
648 return dir_emit(ctx, qname.name, qname.len, ino, type);
649}
650
651static bool proc_sys_link_fill_cache(struct file *file,
652 struct dir_context *ctx,
653 struct ctl_table_header *head,
654 struct ctl_table *table)
655{
656 bool ret = true;
657 head = sysctl_head_grab(head);
658
659 if (S_ISLNK(table->mode)) {
660 /* It is not an error if we can not follow the link ignore it */
661 int err = sysctl_follow_link(&head, &table, current->nsproxy);
662 if (err)
663 goto out;
664 }
665
666 ret = proc_sys_fill_cache(file, ctx, head, table);
667out:
668 sysctl_head_finish(head);
669 return ret;
670}
671
672static int scan(struct ctl_table_header *head, struct ctl_table *table,
673 unsigned long *pos, struct file *file,
674 struct dir_context *ctx)
675{
676 bool res;
677
678 if ((*pos)++ < ctx->pos)
679 return true;
680
681 if (unlikely(S_ISLNK(table->mode)))
682 res = proc_sys_link_fill_cache(file, ctx, head, table);
683 else
684 res = proc_sys_fill_cache(file, ctx, head, table);
685
686 if (res)
687 ctx->pos = *pos;
688
689 return res;
690}
691
692static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
693{
694 struct ctl_table_header *head = grab_header(file_inode(file));
695 struct ctl_table_header *h = NULL;
696 struct ctl_table *entry;
697 struct ctl_dir *ctl_dir;
698 unsigned long pos;
699
700 if (IS_ERR(head))
701 return PTR_ERR(head);
702
703 ctl_dir = container_of(head, struct ctl_dir, header);
704
705 if (!dir_emit_dots(file, ctx))
706 return 0;
707
708 pos = 2;
709
710 for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
711 if (!scan(h, entry, &pos, file, ctx)) {
712 sysctl_head_finish(h);
713 break;
714 }
715 }
716 sysctl_head_finish(head);
717 return 0;
718}
719
720static int proc_sys_permission(struct inode *inode, int mask)
721{
722 /*
723 * sysctl entries that are not writeable,
724 * are _NOT_ writeable, capabilities or not.
725 */
726 struct ctl_table_header *head;
727 struct ctl_table *table;
728 int error;
729
730 /* Executable files are not allowed under /proc/sys/ */
731 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
732 return -EACCES;
733
734 head = grab_header(inode);
735 if (IS_ERR(head))
736 return PTR_ERR(head);
737
738 table = PROC_I(inode)->sysctl_entry;
739 if (!table) /* global root - r-xr-xr-x */
740 error = mask & MAY_WRITE ? -EACCES : 0;
741 else /* Use the permissions on the sysctl table entry */
742 error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
743
744 sysctl_head_finish(head);
745 return error;
746}
747
748static int proc_sys_setattr(struct dentry *dentry, struct iattr *attr)
749{
750 struct inode *inode = d_inode(dentry);
751 int error;
752
753 if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
754 return -EPERM;
755
756 error = inode_change_ok(inode, attr);
757 if (error)
758 return error;
759
760 setattr_copy(inode, attr);
761 mark_inode_dirty(inode);
762 return 0;
763}
764
765static int proc_sys_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
766{
767 struct inode *inode = d_inode(dentry);
768 struct ctl_table_header *head = grab_header(inode);
769 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
770
771 if (IS_ERR(head))
772 return PTR_ERR(head);
773
774 generic_fillattr(inode, stat);
775 if (table)
776 stat->mode = (stat->mode & S_IFMT) | table->mode;
777
778 sysctl_head_finish(head);
779 return 0;
780}
781
782static const struct file_operations proc_sys_file_operations = {
783 .open = proc_sys_open,
784 .poll = proc_sys_poll,
785 .read = proc_sys_read,
786 .write = proc_sys_write,
787 .llseek = default_llseek,
788};
789
790static const struct file_operations proc_sys_dir_file_operations = {
791 .read = generic_read_dir,
792 .iterate = proc_sys_readdir,
793 .llseek = generic_file_llseek,
794};
795
796static const struct inode_operations proc_sys_inode_operations = {
797 .permission = proc_sys_permission,
798 .setattr = proc_sys_setattr,
799 .getattr = proc_sys_getattr,
800};
801
802static const struct inode_operations proc_sys_dir_operations = {
803 .lookup = proc_sys_lookup,
804 .permission = proc_sys_permission,
805 .setattr = proc_sys_setattr,
806 .getattr = proc_sys_getattr,
807};
808
809static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
810{
811 if (flags & LOOKUP_RCU)
812 return -ECHILD;
813 return !PROC_I(d_inode(dentry))->sysctl->unregistering;
814}
815
816static int proc_sys_delete(const struct dentry *dentry)
817{
818 return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
819}
820
821static int sysctl_is_seen(struct ctl_table_header *p)
822{
823 struct ctl_table_set *set = p->set;
824 int res;
825 spin_lock(&sysctl_lock);
826 if (p->unregistering)
827 res = 0;
828 else if (!set->is_seen)
829 res = 1;
830 else
831 res = set->is_seen(set);
832 spin_unlock(&sysctl_lock);
833 return res;
834}
835
836static int proc_sys_compare(const struct dentry *parent, const struct dentry *dentry,
837 unsigned int len, const char *str, const struct qstr *name)
838{
839 struct ctl_table_header *head;
840 struct inode *inode;
841
842 /* Although proc doesn't have negative dentries, rcu-walk means
843 * that inode here can be NULL */
844 /* AV: can it, indeed? */
845 inode = d_inode_rcu(dentry);
846 if (!inode)
847 return 1;
848 if (name->len != len)
849 return 1;
850 if (memcmp(name->name, str, len))
851 return 1;
852 head = rcu_dereference(PROC_I(inode)->sysctl);
853 return !head || !sysctl_is_seen(head);
854}
855
856static const struct dentry_operations proc_sys_dentry_operations = {
857 .d_revalidate = proc_sys_revalidate,
858 .d_delete = proc_sys_delete,
859 .d_compare = proc_sys_compare,
860};
861
862static struct ctl_dir *find_subdir(struct ctl_dir *dir,
863 const char *name, int namelen)
864{
865 struct ctl_table_header *head;
866 struct ctl_table *entry;
867
868 entry = find_entry(&head, dir, name, namelen);
869 if (!entry)
870 return ERR_PTR(-ENOENT);
871 if (!S_ISDIR(entry->mode))
872 return ERR_PTR(-ENOTDIR);
873 return container_of(head, struct ctl_dir, header);
874}
875
876static struct ctl_dir *new_dir(struct ctl_table_set *set,
877 const char *name, int namelen)
878{
879 struct ctl_table *table;
880 struct ctl_dir *new;
881 struct ctl_node *node;
882 char *new_name;
883
884 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
885 sizeof(struct ctl_table)*2 + namelen + 1,
886 GFP_KERNEL);
887 if (!new)
888 return NULL;
889
890 node = (struct ctl_node *)(new + 1);
891 table = (struct ctl_table *)(node + 1);
892 new_name = (char *)(table + 2);
893 memcpy(new_name, name, namelen);
894 new_name[namelen] = '\0';
895 table[0].procname = new_name;
896 table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
897 init_header(&new->header, set->dir.header.root, set, node, table);
898
899 return new;
900}
901
902/**
903 * get_subdir - find or create a subdir with the specified name.
904 * @dir: Directory to create the subdirectory in
905 * @name: The name of the subdirectory to find or create
906 * @namelen: The length of name
907 *
908 * Takes a directory with an elevated reference count so we know that
909 * if we drop the lock the directory will not go away. Upon success
910 * the reference is moved from @dir to the returned subdirectory.
911 * Upon error an error code is returned and the reference on @dir is
912 * simply dropped.
913 */
914static struct ctl_dir *get_subdir(struct ctl_dir *dir,
915 const char *name, int namelen)
916{
917 struct ctl_table_set *set = dir->header.set;
918 struct ctl_dir *subdir, *new = NULL;
919 int err;
920
921 spin_lock(&sysctl_lock);
922 subdir = find_subdir(dir, name, namelen);
923 if (!IS_ERR(subdir))
924 goto found;
925 if (PTR_ERR(subdir) != -ENOENT)
926 goto failed;
927
928 spin_unlock(&sysctl_lock);
929 new = new_dir(set, name, namelen);
930 spin_lock(&sysctl_lock);
931 subdir = ERR_PTR(-ENOMEM);
932 if (!new)
933 goto failed;
934
935 /* Was the subdir added while we dropped the lock? */
936 subdir = find_subdir(dir, name, namelen);
937 if (!IS_ERR(subdir))
938 goto found;
939 if (PTR_ERR(subdir) != -ENOENT)
940 goto failed;
941
942 /* Nope. Use the our freshly made directory entry. */
943 err = insert_header(dir, &new->header);
944 subdir = ERR_PTR(err);
945 if (err)
946 goto failed;
947 subdir = new;
948found:
949 subdir->header.nreg++;
950failed:
951 if (IS_ERR(subdir)) {
952 pr_err("sysctl could not get directory: ");
953 sysctl_print_dir(dir);
954 pr_cont("/%*.*s %ld\n",
955 namelen, namelen, name, PTR_ERR(subdir));
956 }
957 drop_sysctl_table(&dir->header);
958 if (new)
959 drop_sysctl_table(&new->header);
960 spin_unlock(&sysctl_lock);
961 return subdir;
962}
963
964static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
965{
966 struct ctl_dir *parent;
967 const char *procname;
968 if (!dir->header.parent)
969 return &set->dir;
970 parent = xlate_dir(set, dir->header.parent);
971 if (IS_ERR(parent))
972 return parent;
973 procname = dir->header.ctl_table[0].procname;
974 return find_subdir(parent, procname, strlen(procname));
975}
976
977static int sysctl_follow_link(struct ctl_table_header **phead,
978 struct ctl_table **pentry, struct nsproxy *namespaces)
979{
980 struct ctl_table_header *head;
981 struct ctl_table_root *root;
982 struct ctl_table_set *set;
983 struct ctl_table *entry;
984 struct ctl_dir *dir;
985 int ret;
986
987 ret = 0;
988 spin_lock(&sysctl_lock);
989 root = (*pentry)->data;
990 set = lookup_header_set(root, namespaces);
991 dir = xlate_dir(set, (*phead)->parent);
992 if (IS_ERR(dir))
993 ret = PTR_ERR(dir);
994 else {
995 const char *procname = (*pentry)->procname;
996 head = NULL;
997 entry = find_entry(&head, dir, procname, strlen(procname));
998 ret = -ENOENT;
999 if (entry && use_table(head)) {
1000 unuse_table(*phead);
1001 *phead = head;
1002 *pentry = entry;
1003 ret = 0;
1004 }
1005 }
1006
1007 spin_unlock(&sysctl_lock);
1008 return ret;
1009}
1010
1011static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
1012{
1013 struct va_format vaf;
1014 va_list args;
1015
1016 va_start(args, fmt);
1017 vaf.fmt = fmt;
1018 vaf.va = &args;
1019
1020 pr_err("sysctl table check failed: %s/%s %pV\n",
1021 path, table->procname, &vaf);
1022
1023 va_end(args);
1024 return -EINVAL;
1025}
1026
1027static int sysctl_check_table(const char *path, struct ctl_table *table)
1028{
1029 int err = 0;
1030 for (; table->procname; table++) {
1031 if (table->child)
1032 err = sysctl_err(path, table, "Not a file");
1033
1034 if ((table->proc_handler == proc_dostring) ||
1035 (table->proc_handler == proc_dointvec) ||
1036 (table->proc_handler == proc_dointvec_minmax) ||
1037 (table->proc_handler == proc_dointvec_jiffies) ||
1038 (table->proc_handler == proc_dointvec_userhz_jiffies) ||
1039 (table->proc_handler == proc_dointvec_ms_jiffies) ||
1040 (table->proc_handler == proc_doulongvec_minmax) ||
1041 (table->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1042 if (!table->data)
1043 err = sysctl_err(path, table, "No data");
1044 if (!table->maxlen)
1045 err = sysctl_err(path, table, "No maxlen");
1046 }
1047 if (!table->proc_handler)
1048 err = sysctl_err(path, table, "No proc_handler");
1049
1050 if ((table->mode & (S_IRUGO|S_IWUGO)) != table->mode)
1051 err = sysctl_err(path, table, "bogus .mode 0%o",
1052 table->mode);
1053 }
1054 return err;
1055}
1056
1057static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table *table,
1058 struct ctl_table_root *link_root)
1059{
1060 struct ctl_table *link_table, *entry, *link;
1061 struct ctl_table_header *links;
1062 struct ctl_node *node;
1063 char *link_name;
1064 int nr_entries, name_bytes;
1065
1066 name_bytes = 0;
1067 nr_entries = 0;
1068 for (entry = table; entry->procname; entry++) {
1069 nr_entries++;
1070 name_bytes += strlen(entry->procname) + 1;
1071 }
1072
1073 links = kzalloc(sizeof(struct ctl_table_header) +
1074 sizeof(struct ctl_node)*nr_entries +
1075 sizeof(struct ctl_table)*(nr_entries + 1) +
1076 name_bytes,
1077 GFP_KERNEL);
1078
1079 if (!links)
1080 return NULL;
1081
1082 node = (struct ctl_node *)(links + 1);
1083 link_table = (struct ctl_table *)(node + nr_entries);
1084 link_name = (char *)&link_table[nr_entries + 1];
1085
1086 for (link = link_table, entry = table; entry->procname; link++, entry++) {
1087 int len = strlen(entry->procname) + 1;
1088 memcpy(link_name, entry->procname, len);
1089 link->procname = link_name;
1090 link->mode = S_IFLNK|S_IRWXUGO;
1091 link->data = link_root;
1092 link_name += len;
1093 }
1094 init_header(links, dir->header.root, dir->header.set, node, link_table);
1095 links->nreg = nr_entries;
1096
1097 return links;
1098}
1099
1100static bool get_links(struct ctl_dir *dir,
1101 struct ctl_table *table, struct ctl_table_root *link_root)
1102{
1103 struct ctl_table_header *head;
1104 struct ctl_table *entry, *link;
1105
1106 /* Are there links available for every entry in table? */
1107 for (entry = table; entry->procname; entry++) {
1108 const char *procname = entry->procname;
1109 link = find_entry(&head, dir, procname, strlen(procname));
1110 if (!link)
1111 return false;
1112 if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1113 continue;
1114 if (S_ISLNK(link->mode) && (link->data == link_root))
1115 continue;
1116 return false;
1117 }
1118
1119 /* The checks passed. Increase the registration count on the links */
1120 for (entry = table; entry->procname; entry++) {
1121 const char *procname = entry->procname;
1122 link = find_entry(&head, dir, procname, strlen(procname));
1123 head->nreg++;
1124 }
1125 return true;
1126}
1127
1128static int insert_links(struct ctl_table_header *head)
1129{
1130 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1131 struct ctl_dir *core_parent = NULL;
1132 struct ctl_table_header *links;
1133 int err;
1134
1135 if (head->set == root_set)
1136 return 0;
1137
1138 core_parent = xlate_dir(root_set, head->parent);
1139 if (IS_ERR(core_parent))
1140 return 0;
1141
1142 if (get_links(core_parent, head->ctl_table, head->root))
1143 return 0;
1144
1145 core_parent->header.nreg++;
1146 spin_unlock(&sysctl_lock);
1147
1148 links = new_links(core_parent, head->ctl_table, head->root);
1149
1150 spin_lock(&sysctl_lock);
1151 err = -ENOMEM;
1152 if (!links)
1153 goto out;
1154
1155 err = 0;
1156 if (get_links(core_parent, head->ctl_table, head->root)) {
1157 kfree(links);
1158 goto out;
1159 }
1160
1161 err = insert_header(core_parent, links);
1162 if (err)
1163 kfree(links);
1164out:
1165 drop_sysctl_table(&core_parent->header);
1166 return err;
1167}
1168
1169/**
1170 * __register_sysctl_table - register a leaf sysctl table
1171 * @set: Sysctl tree to register on
1172 * @path: The path to the directory the sysctl table is in.
1173 * @table: the top-level table structure
1174 *
1175 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1176 * array. A completely 0 filled entry terminates the table.
1177 *
1178 * The members of the &struct ctl_table structure are used as follows:
1179 *
1180 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1181 * enter a sysctl file
1182 *
1183 * data - a pointer to data for use by proc_handler
1184 *
1185 * maxlen - the maximum size in bytes of the data
1186 *
1187 * mode - the file permissions for the /proc/sys file
1188 *
1189 * child - must be %NULL.
1190 *
1191 * proc_handler - the text handler routine (described below)
1192 *
1193 * extra1, extra2 - extra pointers usable by the proc handler routines
1194 *
1195 * Leaf nodes in the sysctl tree will be represented by a single file
1196 * under /proc; non-leaf nodes will be represented by directories.
1197 *
1198 * There must be a proc_handler routine for any terminal nodes.
1199 * Several default handlers are available to cover common cases -
1200 *
1201 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1202 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1203 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1204 *
1205 * It is the handler's job to read the input buffer from user memory
1206 * and process it. The handler should return 0 on success.
1207 *
1208 * This routine returns %NULL on a failure to register, and a pointer
1209 * to the table header on success.
1210 */
1211struct ctl_table_header *__register_sysctl_table(
1212 struct ctl_table_set *set,
1213 const char *path, struct ctl_table *table)
1214{
1215 struct ctl_table_root *root = set->dir.header.root;
1216 struct ctl_table_header *header;
1217 const char *name, *nextname;
1218 struct ctl_dir *dir;
1219 struct ctl_table *entry;
1220 struct ctl_node *node;
1221 int nr_entries = 0;
1222
1223 for (entry = table; entry->procname; entry++)
1224 nr_entries++;
1225
1226 header = kzalloc(sizeof(struct ctl_table_header) +
1227 sizeof(struct ctl_node)*nr_entries, GFP_KERNEL);
1228 if (!header)
1229 return NULL;
1230
1231 node = (struct ctl_node *)(header + 1);
1232 init_header(header, root, set, node, table);
1233 if (sysctl_check_table(path, table))
1234 goto fail;
1235
1236 spin_lock(&sysctl_lock);
1237 dir = &set->dir;
1238 /* Reference moved down the diretory tree get_subdir */
1239 dir->header.nreg++;
1240 spin_unlock(&sysctl_lock);
1241
1242 /* Find the directory for the ctl_table */
1243 for (name = path; name; name = nextname) {
1244 int namelen;
1245 nextname = strchr(name, '/');
1246 if (nextname) {
1247 namelen = nextname - name;
1248 nextname++;
1249 } else {
1250 namelen = strlen(name);
1251 }
1252 if (namelen == 0)
1253 continue;
1254
1255 dir = get_subdir(dir, name, namelen);
1256 if (IS_ERR(dir))
1257 goto fail;
1258 }
1259
1260 spin_lock(&sysctl_lock);
1261 if (insert_header(dir, header))
1262 goto fail_put_dir_locked;
1263
1264 drop_sysctl_table(&dir->header);
1265 spin_unlock(&sysctl_lock);
1266
1267 return header;
1268
1269fail_put_dir_locked:
1270 drop_sysctl_table(&dir->header);
1271 spin_unlock(&sysctl_lock);
1272fail:
1273 kfree(header);
1274 dump_stack();
1275 return NULL;
1276}
1277
1278/**
1279 * register_sysctl - register a sysctl table
1280 * @path: The path to the directory the sysctl table is in.
1281 * @table: the table structure
1282 *
1283 * Register a sysctl table. @table should be a filled in ctl_table
1284 * array. A completely 0 filled entry terminates the table.
1285 *
1286 * See __register_sysctl_table for more details.
1287 */
1288struct ctl_table_header *register_sysctl(const char *path, struct ctl_table *table)
1289{
1290 return __register_sysctl_table(&sysctl_table_root.default_set,
1291 path, table);
1292}
1293EXPORT_SYMBOL(register_sysctl);
1294
1295static char *append_path(const char *path, char *pos, const char *name)
1296{
1297 int namelen;
1298 namelen = strlen(name);
1299 if (((pos - path) + namelen + 2) >= PATH_MAX)
1300 return NULL;
1301 memcpy(pos, name, namelen);
1302 pos[namelen] = '/';
1303 pos[namelen + 1] = '\0';
1304 pos += namelen + 1;
1305 return pos;
1306}
1307
1308static int count_subheaders(struct ctl_table *table)
1309{
1310 int has_files = 0;
1311 int nr_subheaders = 0;
1312 struct ctl_table *entry;
1313
1314 /* special case: no directory and empty directory */
1315 if (!table || !table->procname)
1316 return 1;
1317
1318 for (entry = table; entry->procname; entry++) {
1319 if (entry->child)
1320 nr_subheaders += count_subheaders(entry->child);
1321 else
1322 has_files = 1;
1323 }
1324 return nr_subheaders + has_files;
1325}
1326
1327static int register_leaf_sysctl_tables(const char *path, char *pos,
1328 struct ctl_table_header ***subheader, struct ctl_table_set *set,
1329 struct ctl_table *table)
1330{
1331 struct ctl_table *ctl_table_arg = NULL;
1332 struct ctl_table *entry, *files;
1333 int nr_files = 0;
1334 int nr_dirs = 0;
1335 int err = -ENOMEM;
1336
1337 for (entry = table; entry->procname; entry++) {
1338 if (entry->child)
1339 nr_dirs++;
1340 else
1341 nr_files++;
1342 }
1343
1344 files = table;
1345 /* If there are mixed files and directories we need a new table */
1346 if (nr_dirs && nr_files) {
1347 struct ctl_table *new;
1348 files = kzalloc(sizeof(struct ctl_table) * (nr_files + 1),
1349 GFP_KERNEL);
1350 if (!files)
1351 goto out;
1352
1353 ctl_table_arg = files;
1354 for (new = files, entry = table; entry->procname; entry++) {
1355 if (entry->child)
1356 continue;
1357 *new = *entry;
1358 new++;
1359 }
1360 }
1361
1362 /* Register everything except a directory full of subdirectories */
1363 if (nr_files || !nr_dirs) {
1364 struct ctl_table_header *header;
1365 header = __register_sysctl_table(set, path, files);
1366 if (!header) {
1367 kfree(ctl_table_arg);
1368 goto out;
1369 }
1370
1371 /* Remember if we need to free the file table */
1372 header->ctl_table_arg = ctl_table_arg;
1373 **subheader = header;
1374 (*subheader)++;
1375 }
1376
1377 /* Recurse into the subdirectories. */
1378 for (entry = table; entry->procname; entry++) {
1379 char *child_pos;
1380
1381 if (!entry->child)
1382 continue;
1383
1384 err = -ENAMETOOLONG;
1385 child_pos = append_path(path, pos, entry->procname);
1386 if (!child_pos)
1387 goto out;
1388
1389 err = register_leaf_sysctl_tables(path, child_pos, subheader,
1390 set, entry->child);
1391 pos[0] = '\0';
1392 if (err)
1393 goto out;
1394 }
1395 err = 0;
1396out:
1397 /* On failure our caller will unregister all registered subheaders */
1398 return err;
1399}
1400
1401/**
1402 * __register_sysctl_paths - register a sysctl table hierarchy
1403 * @set: Sysctl tree to register on
1404 * @path: The path to the directory the sysctl table is in.
1405 * @table: the top-level table structure
1406 *
1407 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1408 * array. A completely 0 filled entry terminates the table.
1409 *
1410 * See __register_sysctl_table for more details.
1411 */
1412struct ctl_table_header *__register_sysctl_paths(
1413 struct ctl_table_set *set,
1414 const struct ctl_path *path, struct ctl_table *table)
1415{
1416 struct ctl_table *ctl_table_arg = table;
1417 int nr_subheaders = count_subheaders(table);
1418 struct ctl_table_header *header = NULL, **subheaders, **subheader;
1419 const struct ctl_path *component;
1420 char *new_path, *pos;
1421
1422 pos = new_path = kmalloc(PATH_MAX, GFP_KERNEL);
1423 if (!new_path)
1424 return NULL;
1425
1426 pos[0] = '\0';
1427 for (component = path; component->procname; component++) {
1428 pos = append_path(new_path, pos, component->procname);
1429 if (!pos)
1430 goto out;
1431 }
1432 while (table->procname && table->child && !table[1].procname) {
1433 pos = append_path(new_path, pos, table->procname);
1434 if (!pos)
1435 goto out;
1436 table = table->child;
1437 }
1438 if (nr_subheaders == 1) {
1439 header = __register_sysctl_table(set, new_path, table);
1440 if (header)
1441 header->ctl_table_arg = ctl_table_arg;
1442 } else {
1443 header = kzalloc(sizeof(*header) +
1444 sizeof(*subheaders)*nr_subheaders, GFP_KERNEL);
1445 if (!header)
1446 goto out;
1447
1448 subheaders = (struct ctl_table_header **) (header + 1);
1449 subheader = subheaders;
1450 header->ctl_table_arg = ctl_table_arg;
1451
1452 if (register_leaf_sysctl_tables(new_path, pos, &subheader,
1453 set, table))
1454 goto err_register_leaves;
1455 }
1456
1457out:
1458 kfree(new_path);
1459 return header;
1460
1461err_register_leaves:
1462 while (subheader > subheaders) {
1463 struct ctl_table_header *subh = *(--subheader);
1464 struct ctl_table *table = subh->ctl_table_arg;
1465 unregister_sysctl_table(subh);
1466 kfree(table);
1467 }
1468 kfree(header);
1469 header = NULL;
1470 goto out;
1471}
1472
1473/**
1474 * register_sysctl_table_path - register a sysctl table hierarchy
1475 * @path: The path to the directory the sysctl table is in.
1476 * @table: the top-level table structure
1477 *
1478 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1479 * array. A completely 0 filled entry terminates the table.
1480 *
1481 * See __register_sysctl_paths for more details.
1482 */
1483struct ctl_table_header *register_sysctl_paths(const struct ctl_path *path,
1484 struct ctl_table *table)
1485{
1486 return __register_sysctl_paths(&sysctl_table_root.default_set,
1487 path, table);
1488}
1489EXPORT_SYMBOL(register_sysctl_paths);
1490
1491/**
1492 * register_sysctl_table - register a sysctl table hierarchy
1493 * @table: the top-level table structure
1494 *
1495 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1496 * array. A completely 0 filled entry terminates the table.
1497 *
1498 * See register_sysctl_paths for more details.
1499 */
1500struct ctl_table_header *register_sysctl_table(struct ctl_table *table)
1501{
1502 static const struct ctl_path null_path[] = { {} };
1503
1504 return register_sysctl_paths(null_path, table);
1505}
1506EXPORT_SYMBOL(register_sysctl_table);
1507
1508static void put_links(struct ctl_table_header *header)
1509{
1510 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1511 struct ctl_table_root *root = header->root;
1512 struct ctl_dir *parent = header->parent;
1513 struct ctl_dir *core_parent;
1514 struct ctl_table *entry;
1515
1516 if (header->set == root_set)
1517 return;
1518
1519 core_parent = xlate_dir(root_set, parent);
1520 if (IS_ERR(core_parent))
1521 return;
1522
1523 for (entry = header->ctl_table; entry->procname; entry++) {
1524 struct ctl_table_header *link_head;
1525 struct ctl_table *link;
1526 const char *name = entry->procname;
1527
1528 link = find_entry(&link_head, core_parent, name, strlen(name));
1529 if (link &&
1530 ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1531 (S_ISLNK(link->mode) && (link->data == root)))) {
1532 drop_sysctl_table(link_head);
1533 }
1534 else {
1535 pr_err("sysctl link missing during unregister: ");
1536 sysctl_print_dir(parent);
1537 pr_cont("/%s\n", name);
1538 }
1539 }
1540}
1541
1542static void drop_sysctl_table(struct ctl_table_header *header)
1543{
1544 struct ctl_dir *parent = header->parent;
1545
1546 if (--header->nreg)
1547 return;
1548
1549 put_links(header);
1550 start_unregistering(header);
1551 if (!--header->count)
1552 kfree_rcu(header, rcu);
1553
1554 if (parent)
1555 drop_sysctl_table(&parent->header);
1556}
1557
1558/**
1559 * unregister_sysctl_table - unregister a sysctl table hierarchy
1560 * @header: the header returned from register_sysctl_table
1561 *
1562 * Unregisters the sysctl table and all children. proc entries may not
1563 * actually be removed until they are no longer used by anyone.
1564 */
1565void unregister_sysctl_table(struct ctl_table_header * header)
1566{
1567 int nr_subheaders;
1568 might_sleep();
1569
1570 if (header == NULL)
1571 return;
1572
1573 nr_subheaders = count_subheaders(header->ctl_table_arg);
1574 if (unlikely(nr_subheaders > 1)) {
1575 struct ctl_table_header **subheaders;
1576 int i;
1577
1578 subheaders = (struct ctl_table_header **)(header + 1);
1579 for (i = nr_subheaders -1; i >= 0; i--) {
1580 struct ctl_table_header *subh = subheaders[i];
1581 struct ctl_table *table = subh->ctl_table_arg;
1582 unregister_sysctl_table(subh);
1583 kfree(table);
1584 }
1585 kfree(header);
1586 return;
1587 }
1588
1589 spin_lock(&sysctl_lock);
1590 drop_sysctl_table(header);
1591 spin_unlock(&sysctl_lock);
1592}
1593EXPORT_SYMBOL(unregister_sysctl_table);
1594
1595void setup_sysctl_set(struct ctl_table_set *set,
1596 struct ctl_table_root *root,
1597 int (*is_seen)(struct ctl_table_set *))
1598{
1599 memset(set, 0, sizeof(*set));
1600 set->is_seen = is_seen;
1601 init_header(&set->dir.header, root, set, NULL, root_table);
1602}
1603
1604void retire_sysctl_set(struct ctl_table_set *set)
1605{
1606 WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1607}
1608
1609int __init proc_sys_init(void)
1610{
1611 struct proc_dir_entry *proc_sys_root;
1612
1613 proc_sys_root = proc_mkdir("sys", NULL);
1614 proc_sys_root->proc_iops = &proc_sys_dir_operations;
1615 proc_sys_root->proc_fops = &proc_sys_dir_file_operations;
1616 proc_sys_root->nlink = 0;
1617
1618 return sysctl_init();
1619}
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/uio.h>
16#include <linux/module.h>
17#include <linux/bpf-cgroup.h>
18#include <linux/mount.h>
19#include <linux/kmemleak.h>
20#include <linux/lockdep.h>
21#include "internal.h"
22
23#define list_for_each_table_entry(entry, header) \
24 entry = header->ctl_table; \
25 for (size_t i = 0 ; i < header->ctl_table_size; ++i, entry++)
26
27static const struct dentry_operations proc_sys_dentry_operations;
28static const struct file_operations proc_sys_file_operations;
29static const struct inode_operations proc_sys_inode_operations;
30static const struct file_operations proc_sys_dir_file_operations;
31static const struct inode_operations proc_sys_dir_operations;
32
33/*
34 * Support for permanently empty directories.
35 * Must be non-empty to avoid sharing an address with other tables.
36 */
37static const struct ctl_table sysctl_mount_point[] = {
38 { }
39};
40
41/**
42 * register_sysctl_mount_point() - registers a sysctl mount point
43 * @path: path for the mount point
44 *
45 * Used to create a permanently empty directory to serve as mount point.
46 * There are some subtle but important permission checks this allows in the
47 * case of unprivileged mounts.
48 */
49struct ctl_table_header *register_sysctl_mount_point(const char *path)
50{
51 return register_sysctl_sz(path, sysctl_mount_point, 0);
52}
53EXPORT_SYMBOL(register_sysctl_mount_point);
54
55#define sysctl_is_perm_empty_ctl_header(hptr) \
56 (hptr->type == SYSCTL_TABLE_TYPE_PERMANENTLY_EMPTY)
57#define sysctl_set_perm_empty_ctl_header(hptr) \
58 (hptr->type = SYSCTL_TABLE_TYPE_PERMANENTLY_EMPTY)
59#define sysctl_clear_perm_empty_ctl_header(hptr) \
60 (hptr->type = SYSCTL_TABLE_TYPE_DEFAULT)
61
62void proc_sys_poll_notify(struct ctl_table_poll *poll)
63{
64 if (!poll)
65 return;
66
67 atomic_inc(&poll->event);
68 wake_up_interruptible(&poll->wait);
69}
70
71static const struct ctl_table root_table[] = {
72 {
73 .procname = "",
74 .mode = S_IFDIR|S_IRUGO|S_IXUGO,
75 },
76};
77static struct ctl_table_root sysctl_table_root = {
78 .default_set.dir.header = {
79 {{.count = 1,
80 .nreg = 1,
81 .ctl_table = root_table }},
82 .ctl_table_arg = root_table,
83 .root = &sysctl_table_root,
84 .set = &sysctl_table_root.default_set,
85 },
86};
87
88static DEFINE_SPINLOCK(sysctl_lock);
89
90static void drop_sysctl_table(struct ctl_table_header *header);
91static int sysctl_follow_link(struct ctl_table_header **phead,
92 const struct ctl_table **pentry);
93static int insert_links(struct ctl_table_header *head);
94static void put_links(struct ctl_table_header *header);
95
96static void sysctl_print_dir(struct ctl_dir *dir)
97{
98 if (dir->header.parent)
99 sysctl_print_dir(dir->header.parent);
100 pr_cont("%s/", dir->header.ctl_table[0].procname);
101}
102
103static int namecmp(const char *name1, int len1, const char *name2, int len2)
104{
105 int cmp;
106
107 cmp = memcmp(name1, name2, min(len1, len2));
108 if (cmp == 0)
109 cmp = len1 - len2;
110 return cmp;
111}
112
113static const struct ctl_table *find_entry(struct ctl_table_header **phead,
114 struct ctl_dir *dir, const char *name, int namelen)
115{
116 struct ctl_table_header *head;
117 const struct ctl_table *entry;
118 struct rb_node *node = dir->root.rb_node;
119
120 lockdep_assert_held(&sysctl_lock);
121
122 while (node)
123 {
124 struct ctl_node *ctl_node;
125 const char *procname;
126 int cmp;
127
128 ctl_node = rb_entry(node, struct ctl_node, node);
129 head = ctl_node->header;
130 entry = &head->ctl_table[ctl_node - head->node];
131 procname = entry->procname;
132
133 cmp = namecmp(name, namelen, procname, strlen(procname));
134 if (cmp < 0)
135 node = node->rb_left;
136 else if (cmp > 0)
137 node = node->rb_right;
138 else {
139 *phead = head;
140 return entry;
141 }
142 }
143 return NULL;
144}
145
146static int insert_entry(struct ctl_table_header *head, const struct ctl_table *entry)
147{
148 struct rb_node *node = &head->node[entry - head->ctl_table].node;
149 struct rb_node **p = &head->parent->root.rb_node;
150 struct rb_node *parent = NULL;
151 const char *name = entry->procname;
152 int namelen = strlen(name);
153
154 while (*p) {
155 struct ctl_table_header *parent_head;
156 const struct ctl_table *parent_entry;
157 struct ctl_node *parent_node;
158 const char *parent_name;
159 int cmp;
160
161 parent = *p;
162 parent_node = rb_entry(parent, struct ctl_node, node);
163 parent_head = parent_node->header;
164 parent_entry = &parent_head->ctl_table[parent_node - parent_head->node];
165 parent_name = parent_entry->procname;
166
167 cmp = namecmp(name, namelen, parent_name, strlen(parent_name));
168 if (cmp < 0)
169 p = &(*p)->rb_left;
170 else if (cmp > 0)
171 p = &(*p)->rb_right;
172 else {
173 pr_err("sysctl duplicate entry: ");
174 sysctl_print_dir(head->parent);
175 pr_cont("%s\n", entry->procname);
176 return -EEXIST;
177 }
178 }
179
180 rb_link_node(node, parent, p);
181 rb_insert_color(node, &head->parent->root);
182 return 0;
183}
184
185static void erase_entry(struct ctl_table_header *head, const struct ctl_table *entry)
186{
187 struct rb_node *node = &head->node[entry - head->ctl_table].node;
188
189 rb_erase(node, &head->parent->root);
190}
191
192static void init_header(struct ctl_table_header *head,
193 struct ctl_table_root *root, struct ctl_table_set *set,
194 struct ctl_node *node, const struct ctl_table *table, size_t table_size)
195{
196 head->ctl_table = table;
197 head->ctl_table_size = table_size;
198 head->ctl_table_arg = table;
199 head->used = 0;
200 head->count = 1;
201 head->nreg = 1;
202 head->unregistering = NULL;
203 head->root = root;
204 head->set = set;
205 head->parent = NULL;
206 head->node = node;
207 INIT_HLIST_HEAD(&head->inodes);
208 if (node) {
209 const struct ctl_table *entry;
210
211 list_for_each_table_entry(entry, head) {
212 node->header = head;
213 node++;
214 }
215 }
216 if (table == sysctl_mount_point)
217 sysctl_set_perm_empty_ctl_header(head);
218}
219
220static void erase_header(struct ctl_table_header *head)
221{
222 const struct ctl_table *entry;
223
224 list_for_each_table_entry(entry, head)
225 erase_entry(head, entry);
226}
227
228static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
229{
230 const struct ctl_table *entry;
231 struct ctl_table_header *dir_h = &dir->header;
232 int err;
233
234
235 /* Is this a permanently empty directory? */
236 if (sysctl_is_perm_empty_ctl_header(dir_h))
237 return -EROFS;
238
239 /* Am I creating a permanently empty directory? */
240 if (sysctl_is_perm_empty_ctl_header(header)) {
241 if (!RB_EMPTY_ROOT(&dir->root))
242 return -EINVAL;
243 sysctl_set_perm_empty_ctl_header(dir_h);
244 }
245
246 dir_h->nreg++;
247 header->parent = dir;
248 err = insert_links(header);
249 if (err)
250 goto fail_links;
251 list_for_each_table_entry(entry, header) {
252 err = insert_entry(header, entry);
253 if (err)
254 goto fail;
255 }
256 return 0;
257fail:
258 erase_header(header);
259 put_links(header);
260fail_links:
261 if (header->ctl_table == sysctl_mount_point)
262 sysctl_clear_perm_empty_ctl_header(dir_h);
263 header->parent = NULL;
264 drop_sysctl_table(dir_h);
265 return err;
266}
267
268static int use_table(struct ctl_table_header *p)
269{
270 lockdep_assert_held(&sysctl_lock);
271
272 if (unlikely(p->unregistering))
273 return 0;
274 p->used++;
275 return 1;
276}
277
278static void unuse_table(struct ctl_table_header *p)
279{
280 lockdep_assert_held(&sysctl_lock);
281
282 if (!--p->used)
283 if (unlikely(p->unregistering))
284 complete(p->unregistering);
285}
286
287static void proc_sys_invalidate_dcache(struct ctl_table_header *head)
288{
289 proc_invalidate_siblings_dcache(&head->inodes, &sysctl_lock);
290}
291
292static void start_unregistering(struct ctl_table_header *p)
293{
294 /* will reacquire if has to wait */
295 lockdep_assert_held(&sysctl_lock);
296
297 /*
298 * if p->used is 0, nobody will ever touch that entry again;
299 * we'll eliminate all paths to it before dropping sysctl_lock
300 */
301 if (unlikely(p->used)) {
302 struct completion wait;
303 init_completion(&wait);
304 p->unregistering = &wait;
305 spin_unlock(&sysctl_lock);
306 wait_for_completion(&wait);
307 } else {
308 /* anything non-NULL; we'll never dereference it */
309 p->unregistering = ERR_PTR(-EINVAL);
310 spin_unlock(&sysctl_lock);
311 }
312 /*
313 * Invalidate dentries for unregistered sysctls: namespaced sysctls
314 * can have duplicate names and contaminate dcache very badly.
315 */
316 proc_sys_invalidate_dcache(p);
317 /*
318 * do not remove from the list until nobody holds it; walking the
319 * list in do_sysctl() relies on that.
320 */
321 spin_lock(&sysctl_lock);
322 erase_header(p);
323}
324
325static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
326{
327 BUG_ON(!head);
328 spin_lock(&sysctl_lock);
329 if (!use_table(head))
330 head = ERR_PTR(-ENOENT);
331 spin_unlock(&sysctl_lock);
332 return head;
333}
334
335static void sysctl_head_finish(struct ctl_table_header *head)
336{
337 if (!head)
338 return;
339 spin_lock(&sysctl_lock);
340 unuse_table(head);
341 spin_unlock(&sysctl_lock);
342}
343
344static struct ctl_table_set *
345lookup_header_set(struct ctl_table_root *root)
346{
347 struct ctl_table_set *set = &root->default_set;
348 if (root->lookup)
349 set = root->lookup(root);
350 return set;
351}
352
353static const struct ctl_table *lookup_entry(struct ctl_table_header **phead,
354 struct ctl_dir *dir,
355 const char *name, int namelen)
356{
357 struct ctl_table_header *head;
358 const struct ctl_table *entry;
359
360 spin_lock(&sysctl_lock);
361 entry = find_entry(&head, dir, name, namelen);
362 if (entry && use_table(head))
363 *phead = head;
364 else
365 entry = NULL;
366 spin_unlock(&sysctl_lock);
367 return entry;
368}
369
370static struct ctl_node *first_usable_entry(struct rb_node *node)
371{
372 struct ctl_node *ctl_node;
373
374 for (;node; node = rb_next(node)) {
375 ctl_node = rb_entry(node, struct ctl_node, node);
376 if (use_table(ctl_node->header))
377 return ctl_node;
378 }
379 return NULL;
380}
381
382static void first_entry(struct ctl_dir *dir,
383 struct ctl_table_header **phead, const struct ctl_table **pentry)
384{
385 struct ctl_table_header *head = NULL;
386 const struct ctl_table *entry = NULL;
387 struct ctl_node *ctl_node;
388
389 spin_lock(&sysctl_lock);
390 ctl_node = first_usable_entry(rb_first(&dir->root));
391 spin_unlock(&sysctl_lock);
392 if (ctl_node) {
393 head = ctl_node->header;
394 entry = &head->ctl_table[ctl_node - head->node];
395 }
396 *phead = head;
397 *pentry = entry;
398}
399
400static void next_entry(struct ctl_table_header **phead, const struct ctl_table **pentry)
401{
402 struct ctl_table_header *head = *phead;
403 const struct ctl_table *entry = *pentry;
404 struct ctl_node *ctl_node = &head->node[entry - head->ctl_table];
405
406 spin_lock(&sysctl_lock);
407 unuse_table(head);
408
409 ctl_node = first_usable_entry(rb_next(&ctl_node->node));
410 spin_unlock(&sysctl_lock);
411 head = NULL;
412 if (ctl_node) {
413 head = ctl_node->header;
414 entry = &head->ctl_table[ctl_node - head->node];
415 }
416 *phead = head;
417 *pentry = entry;
418}
419
420/*
421 * sysctl_perm does NOT grant the superuser all rights automatically, because
422 * some sysctl variables are readonly even to root.
423 */
424
425static int test_perm(int mode, int op)
426{
427 if (uid_eq(current_euid(), GLOBAL_ROOT_UID))
428 mode >>= 6;
429 else if (in_egroup_p(GLOBAL_ROOT_GID))
430 mode >>= 3;
431 if ((op & ~mode & (MAY_READ|MAY_WRITE|MAY_EXEC)) == 0)
432 return 0;
433 return -EACCES;
434}
435
436static int sysctl_perm(struct ctl_table_header *head, const struct ctl_table *table, int op)
437{
438 struct ctl_table_root *root = head->root;
439 int mode;
440
441 if (root->permissions)
442 mode = root->permissions(head, table);
443 else
444 mode = table->mode;
445
446 return test_perm(mode, op);
447}
448
449static struct inode *proc_sys_make_inode(struct super_block *sb,
450 struct ctl_table_header *head, const struct ctl_table *table)
451{
452 struct ctl_table_root *root = head->root;
453 struct inode *inode;
454 struct proc_inode *ei;
455
456 inode = new_inode(sb);
457 if (!inode)
458 return ERR_PTR(-ENOMEM);
459
460 inode->i_ino = get_next_ino();
461
462 ei = PROC_I(inode);
463
464 spin_lock(&sysctl_lock);
465 if (unlikely(head->unregistering)) {
466 spin_unlock(&sysctl_lock);
467 iput(inode);
468 return ERR_PTR(-ENOENT);
469 }
470 ei->sysctl = head;
471 ei->sysctl_entry = table;
472 hlist_add_head_rcu(&ei->sibling_inodes, &head->inodes);
473 head->count++;
474 spin_unlock(&sysctl_lock);
475
476 simple_inode_init_ts(inode);
477 inode->i_mode = table->mode;
478 if (!S_ISDIR(table->mode)) {
479 inode->i_mode |= S_IFREG;
480 inode->i_op = &proc_sys_inode_operations;
481 inode->i_fop = &proc_sys_file_operations;
482 } else {
483 inode->i_mode |= S_IFDIR;
484 inode->i_op = &proc_sys_dir_operations;
485 inode->i_fop = &proc_sys_dir_file_operations;
486 if (sysctl_is_perm_empty_ctl_header(head))
487 make_empty_dir_inode(inode);
488 }
489
490 inode->i_uid = GLOBAL_ROOT_UID;
491 inode->i_gid = GLOBAL_ROOT_GID;
492 if (root->set_ownership)
493 root->set_ownership(head, &inode->i_uid, &inode->i_gid);
494
495 return inode;
496}
497
498void proc_sys_evict_inode(struct inode *inode, struct ctl_table_header *head)
499{
500 spin_lock(&sysctl_lock);
501 hlist_del_init_rcu(&PROC_I(inode)->sibling_inodes);
502 if (!--head->count)
503 kfree_rcu(head, rcu);
504 spin_unlock(&sysctl_lock);
505}
506
507static struct ctl_table_header *grab_header(struct inode *inode)
508{
509 struct ctl_table_header *head = PROC_I(inode)->sysctl;
510 if (!head)
511 head = &sysctl_table_root.default_set.dir.header;
512 return sysctl_head_grab(head);
513}
514
515static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
516 unsigned int flags)
517{
518 struct ctl_table_header *head = grab_header(dir);
519 struct ctl_table_header *h = NULL;
520 const struct qstr *name = &dentry->d_name;
521 const struct ctl_table *p;
522 struct inode *inode;
523 struct dentry *err = ERR_PTR(-ENOENT);
524 struct ctl_dir *ctl_dir;
525 int ret;
526
527 if (IS_ERR(head))
528 return ERR_CAST(head);
529
530 ctl_dir = container_of(head, struct ctl_dir, header);
531
532 p = lookup_entry(&h, ctl_dir, name->name, name->len);
533 if (!p)
534 goto out;
535
536 if (S_ISLNK(p->mode)) {
537 ret = sysctl_follow_link(&h, &p);
538 err = ERR_PTR(ret);
539 if (ret)
540 goto out;
541 }
542
543 d_set_d_op(dentry, &proc_sys_dentry_operations);
544 inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
545 err = d_splice_alias(inode, dentry);
546
547out:
548 if (h)
549 sysctl_head_finish(h);
550 sysctl_head_finish(head);
551 return err;
552}
553
554static ssize_t proc_sys_call_handler(struct kiocb *iocb, struct iov_iter *iter,
555 int write)
556{
557 struct inode *inode = file_inode(iocb->ki_filp);
558 struct ctl_table_header *head = grab_header(inode);
559 const struct ctl_table *table = PROC_I(inode)->sysctl_entry;
560 size_t count = iov_iter_count(iter);
561 char *kbuf;
562 ssize_t error;
563
564 if (IS_ERR(head))
565 return PTR_ERR(head);
566
567 /*
568 * At this point we know that the sysctl was not unregistered
569 * and won't be until we finish.
570 */
571 error = -EPERM;
572 if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
573 goto out;
574
575 /* if that can happen at all, it should be -EINVAL, not -EISDIR */
576 error = -EINVAL;
577 if (!table->proc_handler)
578 goto out;
579
580 /* don't even try if the size is too large */
581 error = -ENOMEM;
582 if (count >= KMALLOC_MAX_SIZE)
583 goto out;
584 kbuf = kvzalloc(count + 1, GFP_KERNEL);
585 if (!kbuf)
586 goto out;
587
588 if (write) {
589 error = -EFAULT;
590 if (!copy_from_iter_full(kbuf, count, iter))
591 goto out_free_buf;
592 kbuf[count] = '\0';
593 }
594
595 error = BPF_CGROUP_RUN_PROG_SYSCTL(head, table, write, &kbuf, &count,
596 &iocb->ki_pos);
597 if (error)
598 goto out_free_buf;
599
600 /* careful: calling conventions are nasty here */
601 error = table->proc_handler(table, write, kbuf, &count, &iocb->ki_pos);
602 if (error)
603 goto out_free_buf;
604
605 if (!write) {
606 error = -EFAULT;
607 if (copy_to_iter(kbuf, count, iter) < count)
608 goto out_free_buf;
609 }
610
611 error = count;
612out_free_buf:
613 kvfree(kbuf);
614out:
615 sysctl_head_finish(head);
616
617 return error;
618}
619
620static ssize_t proc_sys_read(struct kiocb *iocb, struct iov_iter *iter)
621{
622 return proc_sys_call_handler(iocb, iter, 0);
623}
624
625static ssize_t proc_sys_write(struct kiocb *iocb, struct iov_iter *iter)
626{
627 return proc_sys_call_handler(iocb, iter, 1);
628}
629
630static int proc_sys_open(struct inode *inode, struct file *filp)
631{
632 struct ctl_table_header *head = grab_header(inode);
633 const struct ctl_table *table = PROC_I(inode)->sysctl_entry;
634
635 /* sysctl was unregistered */
636 if (IS_ERR(head))
637 return PTR_ERR(head);
638
639 if (table->poll)
640 filp->private_data = proc_sys_poll_event(table->poll);
641
642 sysctl_head_finish(head);
643
644 return 0;
645}
646
647static __poll_t proc_sys_poll(struct file *filp, poll_table *wait)
648{
649 struct inode *inode = file_inode(filp);
650 struct ctl_table_header *head = grab_header(inode);
651 const struct ctl_table *table = PROC_I(inode)->sysctl_entry;
652 __poll_t ret = DEFAULT_POLLMASK;
653 unsigned long event;
654
655 /* sysctl was unregistered */
656 if (IS_ERR(head))
657 return EPOLLERR | EPOLLHUP;
658
659 if (!table->proc_handler)
660 goto out;
661
662 if (!table->poll)
663 goto out;
664
665 event = (unsigned long)filp->private_data;
666 poll_wait(filp, &table->poll->wait, wait);
667
668 if (event != atomic_read(&table->poll->event)) {
669 filp->private_data = proc_sys_poll_event(table->poll);
670 ret = EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
671 }
672
673out:
674 sysctl_head_finish(head);
675
676 return ret;
677}
678
679static bool proc_sys_fill_cache(struct file *file,
680 struct dir_context *ctx,
681 struct ctl_table_header *head,
682 const struct ctl_table *table)
683{
684 struct dentry *child, *dir = file->f_path.dentry;
685 struct inode *inode;
686 struct qstr qname;
687 ino_t ino = 0;
688 unsigned type = DT_UNKNOWN;
689
690 qname.name = table->procname;
691 qname.len = strlen(table->procname);
692 qname.hash = full_name_hash(dir, qname.name, qname.len);
693
694 child = d_lookup(dir, &qname);
695 if (!child) {
696 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
697 child = d_alloc_parallel(dir, &qname, &wq);
698 if (IS_ERR(child))
699 return false;
700 if (d_in_lookup(child)) {
701 struct dentry *res;
702 d_set_d_op(child, &proc_sys_dentry_operations);
703 inode = proc_sys_make_inode(dir->d_sb, head, table);
704 res = d_splice_alias(inode, child);
705 d_lookup_done(child);
706 if (unlikely(res)) {
707 dput(child);
708
709 if (IS_ERR(res))
710 return false;
711
712 child = res;
713 }
714 }
715 }
716 inode = d_inode(child);
717 ino = inode->i_ino;
718 type = inode->i_mode >> 12;
719 dput(child);
720 return dir_emit(ctx, qname.name, qname.len, ino, type);
721}
722
723static bool proc_sys_link_fill_cache(struct file *file,
724 struct dir_context *ctx,
725 struct ctl_table_header *head,
726 const struct ctl_table *table)
727{
728 bool ret = true;
729
730 head = sysctl_head_grab(head);
731 if (IS_ERR(head))
732 return false;
733
734 /* It is not an error if we can not follow the link ignore it */
735 if (sysctl_follow_link(&head, &table))
736 goto out;
737
738 ret = proc_sys_fill_cache(file, ctx, head, table);
739out:
740 sysctl_head_finish(head);
741 return ret;
742}
743
744static int scan(struct ctl_table_header *head, const struct ctl_table *table,
745 unsigned long *pos, struct file *file,
746 struct dir_context *ctx)
747{
748 bool res;
749
750 if ((*pos)++ < ctx->pos)
751 return true;
752
753 if (unlikely(S_ISLNK(table->mode)))
754 res = proc_sys_link_fill_cache(file, ctx, head, table);
755 else
756 res = proc_sys_fill_cache(file, ctx, head, table);
757
758 if (res)
759 ctx->pos = *pos;
760
761 return res;
762}
763
764static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
765{
766 struct ctl_table_header *head = grab_header(file_inode(file));
767 struct ctl_table_header *h = NULL;
768 const struct ctl_table *entry;
769 struct ctl_dir *ctl_dir;
770 unsigned long pos;
771
772 if (IS_ERR(head))
773 return PTR_ERR(head);
774
775 ctl_dir = container_of(head, struct ctl_dir, header);
776
777 if (!dir_emit_dots(file, ctx))
778 goto out;
779
780 pos = 2;
781
782 for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
783 if (!scan(h, entry, &pos, file, ctx)) {
784 sysctl_head_finish(h);
785 break;
786 }
787 }
788out:
789 sysctl_head_finish(head);
790 return 0;
791}
792
793static int proc_sys_permission(struct mnt_idmap *idmap,
794 struct inode *inode, int mask)
795{
796 /*
797 * sysctl entries that are not writeable,
798 * are _NOT_ writeable, capabilities or not.
799 */
800 struct ctl_table_header *head;
801 const struct ctl_table *table;
802 int error;
803
804 /* Executable files are not allowed under /proc/sys/ */
805 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
806 return -EACCES;
807
808 head = grab_header(inode);
809 if (IS_ERR(head))
810 return PTR_ERR(head);
811
812 table = PROC_I(inode)->sysctl_entry;
813 if (!table) /* global root - r-xr-xr-x */
814 error = mask & MAY_WRITE ? -EACCES : 0;
815 else /* Use the permissions on the sysctl table entry */
816 error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
817
818 sysctl_head_finish(head);
819 return error;
820}
821
822static int proc_sys_setattr(struct mnt_idmap *idmap,
823 struct dentry *dentry, struct iattr *attr)
824{
825 struct inode *inode = d_inode(dentry);
826 int error;
827
828 if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
829 return -EPERM;
830
831 error = setattr_prepare(&nop_mnt_idmap, dentry, attr);
832 if (error)
833 return error;
834
835 setattr_copy(&nop_mnt_idmap, inode, attr);
836 return 0;
837}
838
839static int proc_sys_getattr(struct mnt_idmap *idmap,
840 const struct path *path, struct kstat *stat,
841 u32 request_mask, unsigned int query_flags)
842{
843 struct inode *inode = d_inode(path->dentry);
844 struct ctl_table_header *head = grab_header(inode);
845 const struct ctl_table *table = PROC_I(inode)->sysctl_entry;
846
847 if (IS_ERR(head))
848 return PTR_ERR(head);
849
850 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
851 if (table)
852 stat->mode = (stat->mode & S_IFMT) | table->mode;
853
854 sysctl_head_finish(head);
855 return 0;
856}
857
858static const struct file_operations proc_sys_file_operations = {
859 .open = proc_sys_open,
860 .poll = proc_sys_poll,
861 .read_iter = proc_sys_read,
862 .write_iter = proc_sys_write,
863 .splice_read = copy_splice_read,
864 .splice_write = iter_file_splice_write,
865 .llseek = default_llseek,
866};
867
868static const struct file_operations proc_sys_dir_file_operations = {
869 .read = generic_read_dir,
870 .iterate_shared = proc_sys_readdir,
871 .llseek = generic_file_llseek,
872};
873
874static const struct inode_operations proc_sys_inode_operations = {
875 .permission = proc_sys_permission,
876 .setattr = proc_sys_setattr,
877 .getattr = proc_sys_getattr,
878};
879
880static const struct inode_operations proc_sys_dir_operations = {
881 .lookup = proc_sys_lookup,
882 .permission = proc_sys_permission,
883 .setattr = proc_sys_setattr,
884 .getattr = proc_sys_getattr,
885};
886
887static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
888{
889 if (flags & LOOKUP_RCU)
890 return -ECHILD;
891 return !PROC_I(d_inode(dentry))->sysctl->unregistering;
892}
893
894static int proc_sys_delete(const struct dentry *dentry)
895{
896 return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
897}
898
899static int sysctl_is_seen(struct ctl_table_header *p)
900{
901 struct ctl_table_set *set = p->set;
902 int res;
903 spin_lock(&sysctl_lock);
904 if (p->unregistering)
905 res = 0;
906 else if (!set->is_seen)
907 res = 1;
908 else
909 res = set->is_seen(set);
910 spin_unlock(&sysctl_lock);
911 return res;
912}
913
914static int proc_sys_compare(const struct dentry *dentry,
915 unsigned int len, const char *str, const struct qstr *name)
916{
917 struct ctl_table_header *head;
918 struct inode *inode;
919
920 /* Although proc doesn't have negative dentries, rcu-walk means
921 * that inode here can be NULL */
922 /* AV: can it, indeed? */
923 inode = d_inode_rcu(dentry);
924 if (!inode)
925 return 1;
926 if (name->len != len)
927 return 1;
928 if (memcmp(name->name, str, len))
929 return 1;
930 head = rcu_dereference(PROC_I(inode)->sysctl);
931 return !head || !sysctl_is_seen(head);
932}
933
934static const struct dentry_operations proc_sys_dentry_operations = {
935 .d_revalidate = proc_sys_revalidate,
936 .d_delete = proc_sys_delete,
937 .d_compare = proc_sys_compare,
938};
939
940static struct ctl_dir *find_subdir(struct ctl_dir *dir,
941 const char *name, int namelen)
942{
943 struct ctl_table_header *head;
944 const struct ctl_table *entry;
945
946 entry = find_entry(&head, dir, name, namelen);
947 if (!entry)
948 return ERR_PTR(-ENOENT);
949 if (!S_ISDIR(entry->mode))
950 return ERR_PTR(-ENOTDIR);
951 return container_of(head, struct ctl_dir, header);
952}
953
954static struct ctl_dir *new_dir(struct ctl_table_set *set,
955 const char *name, int namelen)
956{
957 struct ctl_table *table;
958 struct ctl_dir *new;
959 struct ctl_node *node;
960 char *new_name;
961
962 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
963 sizeof(struct ctl_table) + namelen + 1,
964 GFP_KERNEL);
965 if (!new)
966 return NULL;
967
968 node = (struct ctl_node *)(new + 1);
969 table = (struct ctl_table *)(node + 1);
970 new_name = (char *)(table + 1);
971 memcpy(new_name, name, namelen);
972 table[0].procname = new_name;
973 table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
974 init_header(&new->header, set->dir.header.root, set, node, table, 1);
975
976 return new;
977}
978
979/**
980 * get_subdir - find or create a subdir with the specified name.
981 * @dir: Directory to create the subdirectory in
982 * @name: The name of the subdirectory to find or create
983 * @namelen: The length of name
984 *
985 * Takes a directory with an elevated reference count so we know that
986 * if we drop the lock the directory will not go away. Upon success
987 * the reference is moved from @dir to the returned subdirectory.
988 * Upon error an error code is returned and the reference on @dir is
989 * simply dropped.
990 */
991static struct ctl_dir *get_subdir(struct ctl_dir *dir,
992 const char *name, int namelen)
993{
994 struct ctl_table_set *set = dir->header.set;
995 struct ctl_dir *subdir, *new = NULL;
996 int err;
997
998 spin_lock(&sysctl_lock);
999 subdir = find_subdir(dir, name, namelen);
1000 if (!IS_ERR(subdir))
1001 goto found;
1002 if (PTR_ERR(subdir) != -ENOENT)
1003 goto failed;
1004
1005 spin_unlock(&sysctl_lock);
1006 new = new_dir(set, name, namelen);
1007 spin_lock(&sysctl_lock);
1008 subdir = ERR_PTR(-ENOMEM);
1009 if (!new)
1010 goto failed;
1011
1012 /* Was the subdir added while we dropped the lock? */
1013 subdir = find_subdir(dir, name, namelen);
1014 if (!IS_ERR(subdir))
1015 goto found;
1016 if (PTR_ERR(subdir) != -ENOENT)
1017 goto failed;
1018
1019 /* Nope. Use the our freshly made directory entry. */
1020 err = insert_header(dir, &new->header);
1021 subdir = ERR_PTR(err);
1022 if (err)
1023 goto failed;
1024 subdir = new;
1025found:
1026 subdir->header.nreg++;
1027failed:
1028 if (IS_ERR(subdir)) {
1029 pr_err("sysctl could not get directory: ");
1030 sysctl_print_dir(dir);
1031 pr_cont("%*.*s %ld\n", namelen, namelen, name,
1032 PTR_ERR(subdir));
1033 }
1034 drop_sysctl_table(&dir->header);
1035 if (new)
1036 drop_sysctl_table(&new->header);
1037 spin_unlock(&sysctl_lock);
1038 return subdir;
1039}
1040
1041static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
1042{
1043 struct ctl_dir *parent;
1044 const char *procname;
1045 if (!dir->header.parent)
1046 return &set->dir;
1047 parent = xlate_dir(set, dir->header.parent);
1048 if (IS_ERR(parent))
1049 return parent;
1050 procname = dir->header.ctl_table[0].procname;
1051 return find_subdir(parent, procname, strlen(procname));
1052}
1053
1054static int sysctl_follow_link(struct ctl_table_header **phead,
1055 const struct ctl_table **pentry)
1056{
1057 struct ctl_table_header *head;
1058 const struct ctl_table *entry;
1059 struct ctl_table_root *root;
1060 struct ctl_table_set *set;
1061 struct ctl_dir *dir;
1062 int ret;
1063
1064 spin_lock(&sysctl_lock);
1065 root = (*pentry)->data;
1066 set = lookup_header_set(root);
1067 dir = xlate_dir(set, (*phead)->parent);
1068 if (IS_ERR(dir))
1069 ret = PTR_ERR(dir);
1070 else {
1071 const char *procname = (*pentry)->procname;
1072 head = NULL;
1073 entry = find_entry(&head, dir, procname, strlen(procname));
1074 ret = -ENOENT;
1075 if (entry && use_table(head)) {
1076 unuse_table(*phead);
1077 *phead = head;
1078 *pentry = entry;
1079 ret = 0;
1080 }
1081 }
1082
1083 spin_unlock(&sysctl_lock);
1084 return ret;
1085}
1086
1087static int sysctl_err(const char *path, const struct ctl_table *table, char *fmt, ...)
1088{
1089 struct va_format vaf;
1090 va_list args;
1091
1092 va_start(args, fmt);
1093 vaf.fmt = fmt;
1094 vaf.va = &args;
1095
1096 pr_err("sysctl table check failed: %s/%s %pV\n",
1097 path, table->procname, &vaf);
1098
1099 va_end(args);
1100 return -EINVAL;
1101}
1102
1103static int sysctl_check_table_array(const char *path, const struct ctl_table *table)
1104{
1105 unsigned int extra;
1106 int err = 0;
1107
1108 if ((table->proc_handler == proc_douintvec) ||
1109 (table->proc_handler == proc_douintvec_minmax)) {
1110 if (table->maxlen != sizeof(unsigned int))
1111 err |= sysctl_err(path, table, "array not allowed");
1112 }
1113
1114 if (table->proc_handler == proc_dou8vec_minmax) {
1115 if (table->maxlen != sizeof(u8))
1116 err |= sysctl_err(path, table, "array not allowed");
1117
1118 if (table->extra1) {
1119 extra = *(unsigned int *) table->extra1;
1120 if (extra > 255U)
1121 err |= sysctl_err(path, table,
1122 "range value too large for proc_dou8vec_minmax");
1123 }
1124 if (table->extra2) {
1125 extra = *(unsigned int *) table->extra2;
1126 if (extra > 255U)
1127 err |= sysctl_err(path, table,
1128 "range value too large for proc_dou8vec_minmax");
1129 }
1130 }
1131
1132 if (table->proc_handler == proc_dobool) {
1133 if (table->maxlen != sizeof(bool))
1134 err |= sysctl_err(path, table, "array not allowed");
1135 }
1136
1137 return err;
1138}
1139
1140static int sysctl_check_table(const char *path, struct ctl_table_header *header)
1141{
1142 const struct ctl_table *entry;
1143 int err = 0;
1144 list_for_each_table_entry(entry, header) {
1145 if (!entry->procname)
1146 err |= sysctl_err(path, entry, "procname is null");
1147 if ((entry->proc_handler == proc_dostring) ||
1148 (entry->proc_handler == proc_dobool) ||
1149 (entry->proc_handler == proc_dointvec) ||
1150 (entry->proc_handler == proc_douintvec) ||
1151 (entry->proc_handler == proc_douintvec_minmax) ||
1152 (entry->proc_handler == proc_dointvec_minmax) ||
1153 (entry->proc_handler == proc_dou8vec_minmax) ||
1154 (entry->proc_handler == proc_dointvec_jiffies) ||
1155 (entry->proc_handler == proc_dointvec_userhz_jiffies) ||
1156 (entry->proc_handler == proc_dointvec_ms_jiffies) ||
1157 (entry->proc_handler == proc_doulongvec_minmax) ||
1158 (entry->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1159 if (!entry->data)
1160 err |= sysctl_err(path, entry, "No data");
1161 if (!entry->maxlen)
1162 err |= sysctl_err(path, entry, "No maxlen");
1163 else
1164 err |= sysctl_check_table_array(path, entry);
1165 }
1166 if (!entry->proc_handler)
1167 err |= sysctl_err(path, entry, "No proc_handler");
1168
1169 if ((entry->mode & (S_IRUGO|S_IWUGO)) != entry->mode)
1170 err |= sysctl_err(path, entry, "bogus .mode 0%o",
1171 entry->mode);
1172 }
1173 return err;
1174}
1175
1176static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table_header *head)
1177{
1178 struct ctl_table *link_table, *link;
1179 struct ctl_table_header *links;
1180 const struct ctl_table *entry;
1181 struct ctl_node *node;
1182 char *link_name;
1183 int name_bytes;
1184
1185 name_bytes = 0;
1186 list_for_each_table_entry(entry, head) {
1187 name_bytes += strlen(entry->procname) + 1;
1188 }
1189
1190 links = kzalloc(sizeof(struct ctl_table_header) +
1191 sizeof(struct ctl_node)*head->ctl_table_size +
1192 sizeof(struct ctl_table)*head->ctl_table_size +
1193 name_bytes,
1194 GFP_KERNEL);
1195
1196 if (!links)
1197 return NULL;
1198
1199 node = (struct ctl_node *)(links + 1);
1200 link_table = (struct ctl_table *)(node + head->ctl_table_size);
1201 link_name = (char *)(link_table + head->ctl_table_size);
1202 link = link_table;
1203
1204 list_for_each_table_entry(entry, head) {
1205 int len = strlen(entry->procname) + 1;
1206 memcpy(link_name, entry->procname, len);
1207 link->procname = link_name;
1208 link->mode = S_IFLNK|S_IRWXUGO;
1209 link->data = head->root;
1210 link_name += len;
1211 link++;
1212 }
1213 init_header(links, dir->header.root, dir->header.set, node, link_table,
1214 head->ctl_table_size);
1215 links->nreg = head->ctl_table_size;
1216
1217 return links;
1218}
1219
1220static bool get_links(struct ctl_dir *dir,
1221 struct ctl_table_header *header,
1222 struct ctl_table_root *link_root)
1223{
1224 struct ctl_table_header *tmp_head;
1225 const struct ctl_table *entry, *link;
1226
1227 if (header->ctl_table_size == 0 ||
1228 sysctl_is_perm_empty_ctl_header(header))
1229 return true;
1230
1231 /* Are there links available for every entry in table? */
1232 list_for_each_table_entry(entry, header) {
1233 const char *procname = entry->procname;
1234 link = find_entry(&tmp_head, dir, procname, strlen(procname));
1235 if (!link)
1236 return false;
1237 if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1238 continue;
1239 if (S_ISLNK(link->mode) && (link->data == link_root))
1240 continue;
1241 return false;
1242 }
1243
1244 /* The checks passed. Increase the registration count on the links */
1245 list_for_each_table_entry(entry, header) {
1246 const char *procname = entry->procname;
1247 link = find_entry(&tmp_head, dir, procname, strlen(procname));
1248 tmp_head->nreg++;
1249 }
1250 return true;
1251}
1252
1253static int insert_links(struct ctl_table_header *head)
1254{
1255 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1256 struct ctl_dir *core_parent;
1257 struct ctl_table_header *links;
1258 int err;
1259
1260 if (head->set == root_set)
1261 return 0;
1262
1263 core_parent = xlate_dir(root_set, head->parent);
1264 if (IS_ERR(core_parent))
1265 return 0;
1266
1267 if (get_links(core_parent, head, head->root))
1268 return 0;
1269
1270 core_parent->header.nreg++;
1271 spin_unlock(&sysctl_lock);
1272
1273 links = new_links(core_parent, head);
1274
1275 spin_lock(&sysctl_lock);
1276 err = -ENOMEM;
1277 if (!links)
1278 goto out;
1279
1280 err = 0;
1281 if (get_links(core_parent, head, head->root)) {
1282 kfree(links);
1283 goto out;
1284 }
1285
1286 err = insert_header(core_parent, links);
1287 if (err)
1288 kfree(links);
1289out:
1290 drop_sysctl_table(&core_parent->header);
1291 return err;
1292}
1293
1294/* Find the directory for the ctl_table. If one is not found create it. */
1295static struct ctl_dir *sysctl_mkdir_p(struct ctl_dir *dir, const char *path)
1296{
1297 const char *name, *nextname;
1298
1299 for (name = path; name; name = nextname) {
1300 int namelen;
1301 nextname = strchr(name, '/');
1302 if (nextname) {
1303 namelen = nextname - name;
1304 nextname++;
1305 } else {
1306 namelen = strlen(name);
1307 }
1308 if (namelen == 0)
1309 continue;
1310
1311 /*
1312 * namelen ensures if name is "foo/bar/yay" only foo is
1313 * registered first. We traverse as if using mkdir -p and
1314 * return a ctl_dir for the last directory entry.
1315 */
1316 dir = get_subdir(dir, name, namelen);
1317 if (IS_ERR(dir))
1318 break;
1319 }
1320 return dir;
1321}
1322
1323/**
1324 * __register_sysctl_table - register a leaf sysctl table
1325 * @set: Sysctl tree to register on
1326 * @path: The path to the directory the sysctl table is in.
1327 *
1328 * @table: the top-level table structure. This table should not be free'd
1329 * after registration. So it should not be used on stack. It can either
1330 * be a global or dynamically allocated by the caller and free'd later
1331 * after sysctl unregistration.
1332 * @table_size : The number of elements in table
1333 *
1334 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1335 * array.
1336 *
1337 * The members of the &struct ctl_table structure are used as follows:
1338 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1339 * enter a sysctl file
1340 * data - a pointer to data for use by proc_handler
1341 * maxlen - the maximum size in bytes of the data
1342 * mode - the file permissions for the /proc/sys file
1343 * type - Defines the target type (described in struct definition)
1344 * proc_handler - the text handler routine (described below)
1345 *
1346 * extra1, extra2 - extra pointers usable by the proc handler routines
1347 * XXX: we should eventually modify these to use long min / max [0]
1348 * [0] https://lkml.kernel.org/87zgpte9o4.fsf@email.froward.int.ebiederm.org
1349 *
1350 * Leaf nodes in the sysctl tree will be represented by a single file
1351 * under /proc; non-leaf nodes are not allowed.
1352 *
1353 * There must be a proc_handler routine for any terminal nodes.
1354 * Several default handlers are available to cover common cases -
1355 *
1356 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1357 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1358 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1359 *
1360 * It is the handler's job to read the input buffer from user memory
1361 * and process it. The handler should return 0 on success.
1362 *
1363 * This routine returns %NULL on a failure to register, and a pointer
1364 * to the table header on success.
1365 */
1366struct ctl_table_header *__register_sysctl_table(
1367 struct ctl_table_set *set,
1368 const char *path, const struct ctl_table *table, size_t table_size)
1369{
1370 struct ctl_table_root *root = set->dir.header.root;
1371 struct ctl_table_header *header;
1372 struct ctl_dir *dir;
1373 struct ctl_node *node;
1374
1375 header = kzalloc(sizeof(struct ctl_table_header) +
1376 sizeof(struct ctl_node)*table_size, GFP_KERNEL_ACCOUNT);
1377 if (!header)
1378 return NULL;
1379
1380 node = (struct ctl_node *)(header + 1);
1381 init_header(header, root, set, node, table, table_size);
1382 if (sysctl_check_table(path, header))
1383 goto fail;
1384
1385 spin_lock(&sysctl_lock);
1386 dir = &set->dir;
1387 /* Reference moved down the directory tree get_subdir */
1388 dir->header.nreg++;
1389 spin_unlock(&sysctl_lock);
1390
1391 dir = sysctl_mkdir_p(dir, path);
1392 if (IS_ERR(dir))
1393 goto fail;
1394 spin_lock(&sysctl_lock);
1395 if (insert_header(dir, header))
1396 goto fail_put_dir_locked;
1397
1398 drop_sysctl_table(&dir->header);
1399 spin_unlock(&sysctl_lock);
1400
1401 return header;
1402
1403fail_put_dir_locked:
1404 drop_sysctl_table(&dir->header);
1405 spin_unlock(&sysctl_lock);
1406fail:
1407 kfree(header);
1408 return NULL;
1409}
1410
1411/**
1412 * register_sysctl_sz - register a sysctl table
1413 * @path: The path to the directory the sysctl table is in. If the path
1414 * doesn't exist we will create it for you.
1415 * @table: the table structure. The calller must ensure the life of the @table
1416 * will be kept during the lifetime use of the syctl. It must not be freed
1417 * until unregister_sysctl_table() is called with the given returned table
1418 * with this registration. If your code is non modular then you don't need
1419 * to call unregister_sysctl_table() and can instead use something like
1420 * register_sysctl_init() which does not care for the result of the syctl
1421 * registration.
1422 * @table_size: The number of elements in table.
1423 *
1424 * Register a sysctl table. @table should be a filled in ctl_table
1425 * array. A completely 0 filled entry terminates the table.
1426 *
1427 * See __register_sysctl_table for more details.
1428 */
1429struct ctl_table_header *register_sysctl_sz(const char *path, const struct ctl_table *table,
1430 size_t table_size)
1431{
1432 return __register_sysctl_table(&sysctl_table_root.default_set,
1433 path, table, table_size);
1434}
1435EXPORT_SYMBOL(register_sysctl_sz);
1436
1437/**
1438 * __register_sysctl_init() - register sysctl table to path
1439 * @path: path name for sysctl base. If that path doesn't exist we will create
1440 * it for you.
1441 * @table: This is the sysctl table that needs to be registered to the path.
1442 * The caller must ensure the life of the @table will be kept during the
1443 * lifetime use of the sysctl.
1444 * @table_name: The name of sysctl table, only used for log printing when
1445 * registration fails
1446 * @table_size: The number of elements in table
1447 *
1448 * The sysctl interface is used by userspace to query or modify at runtime
1449 * a predefined value set on a variable. These variables however have default
1450 * values pre-set. Code which depends on these variables will always work even
1451 * if register_sysctl() fails. If register_sysctl() fails you'd just loose the
1452 * ability to query or modify the sysctls dynamically at run time. Chances of
1453 * register_sysctl() failing on init are extremely low, and so for both reasons
1454 * this function does not return any error as it is used by initialization code.
1455 *
1456 * Context: if your base directory does not exist it will be created for you.
1457 */
1458void __init __register_sysctl_init(const char *path, const struct ctl_table *table,
1459 const char *table_name, size_t table_size)
1460{
1461 struct ctl_table_header *hdr = register_sysctl_sz(path, table, table_size);
1462
1463 if (unlikely(!hdr)) {
1464 pr_err("failed when register_sysctl_sz %s to %s\n", table_name, path);
1465 return;
1466 }
1467 kmemleak_not_leak(hdr);
1468}
1469
1470static void put_links(struct ctl_table_header *header)
1471{
1472 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1473 struct ctl_table_root *root = header->root;
1474 struct ctl_dir *parent = header->parent;
1475 struct ctl_dir *core_parent;
1476 const struct ctl_table *entry;
1477
1478 if (header->set == root_set)
1479 return;
1480
1481 core_parent = xlate_dir(root_set, parent);
1482 if (IS_ERR(core_parent))
1483 return;
1484
1485 list_for_each_table_entry(entry, header) {
1486 struct ctl_table_header *link_head;
1487 const struct ctl_table *link;
1488 const char *name = entry->procname;
1489
1490 link = find_entry(&link_head, core_parent, name, strlen(name));
1491 if (link &&
1492 ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1493 (S_ISLNK(link->mode) && (link->data == root)))) {
1494 drop_sysctl_table(link_head);
1495 }
1496 else {
1497 pr_err("sysctl link missing during unregister: ");
1498 sysctl_print_dir(parent);
1499 pr_cont("%s\n", name);
1500 }
1501 }
1502}
1503
1504static void drop_sysctl_table(struct ctl_table_header *header)
1505{
1506 struct ctl_dir *parent = header->parent;
1507
1508 if (--header->nreg)
1509 return;
1510
1511 if (parent) {
1512 put_links(header);
1513 start_unregistering(header);
1514 }
1515
1516 if (!--header->count)
1517 kfree_rcu(header, rcu);
1518
1519 if (parent)
1520 drop_sysctl_table(&parent->header);
1521}
1522
1523/**
1524 * unregister_sysctl_table - unregister a sysctl table hierarchy
1525 * @header: the header returned from register_sysctl or __register_sysctl_table
1526 *
1527 * Unregisters the sysctl table and all children. proc entries may not
1528 * actually be removed until they are no longer used by anyone.
1529 */
1530void unregister_sysctl_table(struct ctl_table_header * header)
1531{
1532 might_sleep();
1533
1534 if (header == NULL)
1535 return;
1536
1537 spin_lock(&sysctl_lock);
1538 drop_sysctl_table(header);
1539 spin_unlock(&sysctl_lock);
1540}
1541EXPORT_SYMBOL(unregister_sysctl_table);
1542
1543void setup_sysctl_set(struct ctl_table_set *set,
1544 struct ctl_table_root *root,
1545 int (*is_seen)(struct ctl_table_set *))
1546{
1547 memset(set, 0, sizeof(*set));
1548 set->is_seen = is_seen;
1549 init_header(&set->dir.header, root, set, NULL, root_table, 1);
1550}
1551
1552void retire_sysctl_set(struct ctl_table_set *set)
1553{
1554 WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1555}
1556
1557int __init proc_sys_init(void)
1558{
1559 struct proc_dir_entry *proc_sys_root;
1560
1561 proc_sys_root = proc_mkdir("sys", NULL);
1562 proc_sys_root->proc_iops = &proc_sys_dir_operations;
1563 proc_sys_root->proc_dir_ops = &proc_sys_dir_file_operations;
1564 proc_sys_root->nlink = 0;
1565
1566 return sysctl_init_bases();
1567}
1568
1569struct sysctl_alias {
1570 const char *kernel_param;
1571 const char *sysctl_param;
1572};
1573
1574/*
1575 * Historically some settings had both sysctl and a command line parameter.
1576 * With the generic sysctl. parameter support, we can handle them at a single
1577 * place and only keep the historical name for compatibility. This is not meant
1578 * to add brand new aliases. When adding existing aliases, consider whether
1579 * the possibly different moment of changing the value (e.g. from early_param
1580 * to the moment do_sysctl_args() is called) is an issue for the specific
1581 * parameter.
1582 */
1583static const struct sysctl_alias sysctl_aliases[] = {
1584 {"hardlockup_all_cpu_backtrace", "kernel.hardlockup_all_cpu_backtrace" },
1585 {"hung_task_panic", "kernel.hung_task_panic" },
1586 {"numa_zonelist_order", "vm.numa_zonelist_order" },
1587 {"softlockup_all_cpu_backtrace", "kernel.softlockup_all_cpu_backtrace" },
1588 { }
1589};
1590
1591static const char *sysctl_find_alias(char *param)
1592{
1593 const struct sysctl_alias *alias;
1594
1595 for (alias = &sysctl_aliases[0]; alias->kernel_param != NULL; alias++) {
1596 if (strcmp(alias->kernel_param, param) == 0)
1597 return alias->sysctl_param;
1598 }
1599
1600 return NULL;
1601}
1602
1603bool sysctl_is_alias(char *param)
1604{
1605 const char *alias = sysctl_find_alias(param);
1606
1607 return alias != NULL;
1608}
1609
1610/* Set sysctl value passed on kernel command line. */
1611static int process_sysctl_arg(char *param, char *val,
1612 const char *unused, void *arg)
1613{
1614 char *path;
1615 struct vfsmount **proc_mnt = arg;
1616 struct file_system_type *proc_fs_type;
1617 struct file *file;
1618 int len;
1619 int err;
1620 loff_t pos = 0;
1621 ssize_t wret;
1622
1623 if (strncmp(param, "sysctl", sizeof("sysctl") - 1) == 0) {
1624 param += sizeof("sysctl") - 1;
1625
1626 if (param[0] != '/' && param[0] != '.')
1627 return 0;
1628
1629 param++;
1630 } else {
1631 param = (char *) sysctl_find_alias(param);
1632 if (!param)
1633 return 0;
1634 }
1635
1636 if (!val)
1637 return -EINVAL;
1638 len = strlen(val);
1639 if (len == 0)
1640 return -EINVAL;
1641
1642 /*
1643 * To set sysctl options, we use a temporary mount of proc, look up the
1644 * respective sys/ file and write to it. To avoid mounting it when no
1645 * options were given, we mount it only when the first sysctl option is
1646 * found. Why not a persistent mount? There are problems with a
1647 * persistent mount of proc in that it forces userspace not to use any
1648 * proc mount options.
1649 */
1650 if (!*proc_mnt) {
1651 proc_fs_type = get_fs_type("proc");
1652 if (!proc_fs_type) {
1653 pr_err("Failed to find procfs to set sysctl from command line\n");
1654 return 0;
1655 }
1656 *proc_mnt = kern_mount(proc_fs_type);
1657 put_filesystem(proc_fs_type);
1658 if (IS_ERR(*proc_mnt)) {
1659 pr_err("Failed to mount procfs to set sysctl from command line\n");
1660 return 0;
1661 }
1662 }
1663
1664 path = kasprintf(GFP_KERNEL, "sys/%s", param);
1665 if (!path)
1666 panic("%s: Failed to allocate path for %s\n", __func__, param);
1667 strreplace(path, '.', '/');
1668
1669 file = file_open_root_mnt(*proc_mnt, path, O_WRONLY, 0);
1670 if (IS_ERR(file)) {
1671 err = PTR_ERR(file);
1672 if (err == -ENOENT)
1673 pr_err("Failed to set sysctl parameter '%s=%s': parameter not found\n",
1674 param, val);
1675 else if (err == -EACCES)
1676 pr_err("Failed to set sysctl parameter '%s=%s': permission denied (read-only?)\n",
1677 param, val);
1678 else
1679 pr_err("Error %pe opening proc file to set sysctl parameter '%s=%s'\n",
1680 file, param, val);
1681 goto out;
1682 }
1683 wret = kernel_write(file, val, len, &pos);
1684 if (wret < 0) {
1685 err = wret;
1686 if (err == -EINVAL)
1687 pr_err("Failed to set sysctl parameter '%s=%s': invalid value\n",
1688 param, val);
1689 else
1690 pr_err("Error %pe writing to proc file to set sysctl parameter '%s=%s'\n",
1691 ERR_PTR(err), param, val);
1692 } else if (wret != len) {
1693 pr_err("Wrote only %zd bytes of %d writing to proc file %s to set sysctl parameter '%s=%s\n",
1694 wret, len, path, param, val);
1695 }
1696
1697 err = filp_close(file, NULL);
1698 if (err)
1699 pr_err("Error %pe closing proc file to set sysctl parameter '%s=%s\n",
1700 ERR_PTR(err), param, val);
1701out:
1702 kfree(path);
1703 return 0;
1704}
1705
1706void do_sysctl_args(void)
1707{
1708 char *command_line;
1709 struct vfsmount *proc_mnt = NULL;
1710
1711 command_line = kstrdup(saved_command_line, GFP_KERNEL);
1712 if (!command_line)
1713 panic("%s: Failed to allocate copy of command line\n", __func__);
1714
1715 parse_args("Setting sysctl args", command_line,
1716 NULL, 0, -1, -1, &proc_mnt, process_sysctl_arg);
1717
1718 if (proc_mnt)
1719 kern_unmount(proc_mnt);
1720
1721 kfree(command_line);
1722}