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

source code of linux/fs/proc/proc_sysctl.c