Loading...
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(&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->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(name1, 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(name, namelen, 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(name, namelen, 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(head->parent);
172 pr_cont("%s\n", entry->procname);
173 return -EEXIST;
174 }
175 }
176
177 rb_link_node(node, parent, 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(header);
245 if (err)
246 goto fail_links;
247 list_for_each_table_entry(entry, header) {
248 err = insert_entry(header, entry);
249 if (err)
250 goto fail;
251 }
252 return 0;
253fail:
254 erase_header(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(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(&head->inodes, &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(&wait);
296 p->unregistering = &wait;
297 spin_unlock(&sysctl_lock);
298 wait_for_completion(&wait);
299 } else {
300 /* anything non-NULL; we'll never dereference it */
301 p->unregistering = ERR_PTR(-EINVAL);
302 spin_unlock(&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(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(&sysctl_lock);
314 erase_header(p);
315}
316
317static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
318{
319 BUG_ON(!head);
320 spin_lock(&sysctl_lock);
321 if (!use_table(head))
322 head = ERR_PTR(-ENOENT);
323 spin_unlock(&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(&sysctl_lock);
332 unuse_table(head);
333 spin_unlock(&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(&sysctl_lock);
353 entry = find_entry(&head, dir, name, namelen);
354 if (entry && use_table(head))
355 *phead = head;
356 else
357 entry = NULL;
358 spin_unlock(&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(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(&sysctl_lock);
382 ctl_node = first_usable_entry(rb_first(&dir->root));
383 spin_unlock(&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(&sysctl_lock);
399 unuse_table(head);
400
401 ctl_node = first_usable_entry(rb_next(&ctl_node->node));
402 spin_unlock(&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(-ENOMEM);
451
452 inode->i_ino = get_next_ino();
453
454 ei = PROC_I(inode);
455
456 spin_lock(&sysctl_lock);
457 if (unlikely(head->unregistering)) {
458 spin_unlock(&sysctl_lock);
459 iput(inode);
460 return ERR_PTR(-ENOENT);
461 }
462 ei->sysctl = head;
463 ei->sysctl_entry = table;
464 hlist_add_head_rcu(&ei->sibling_inodes, &head->inodes);
465 head->count++;
466 spin_unlock(&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(&sysctl_lock);
495 hlist_del_init_rcu(&PROC_I(inode)->sibling_inodes);
496 if (!--head->count)
497 kfree_rcu(head, rcu);
498 spin_unlock(&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(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(-ENOENT);
518 struct ctl_dir *ctl_dir;
519 int ret;
520
521 if (IS_ERR(head))
522 return ERR_CAST(head);
523
524 ctl_dir = container_of(head, struct ctl_dir, header);
525
526 p = lookup_entry(&h, ctl_dir, name->name, name->len);
527 if (!p)
528 goto out;
529
530 if (S_ISLNK(p->mode)) {
531 ret = sysctl_follow_link(&h, &p);
532 err = ERR_PTR(ret);
533 if (ret)
534 goto out;
535 }
536
537 d_set_d_op(dentry, &proc_sys_dentry_operations);
538 inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
539 err = d_splice_alias(inode, dentry);
540
541out:
542 if (h)
543 sysctl_head_finish(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(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(iter);
555 char *kbuf;
556 ssize_t error;
557
558 if (IS_ERR(head))
559 return PTR_ERR(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, 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(count + 1, GFP_KERNEL);
579 if (!kbuf)
580 goto out;
581
582 if (write) {
583 error = -EFAULT;
584 if (!copy_from_iter_full(kbuf, count, 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(kbuf, count, iter) < count)
602 goto out_free_buf;
603 }
604
605 error = count;
606out_free_buf:
607 kvfree(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, 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, 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(head))
631 return PTR_ERR(head);
632
633 if (table->poll)
634 filp->private_data = proc_sys_poll_event(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(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(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, &table->poll->wait, wait);
661
662 if (event != atomic_read(&table->poll->event)) {
663 filp->private_data = proc_sys_poll_event(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(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(child))
693 return false;
694 if (d_in_lookup(child)) {
695 struct dentry *res;
696 d_set_d_op(child, &proc_sys_dentry_operations);
697 inode = proc_sys_make_inode(dir->d_sb, head, table);
698 res = d_splice_alias(inode, child);
699 d_lookup_done(child);
700 if (unlikely(res)) {
701 if (IS_ERR(res)) {
702 dput(child);
703 return false;
704 }
705 dput(child);
706 child = res;
707 }
708 }
709 }
710 inode = d_inode(child);
711 ino = inode->i_ino;
712 type = inode->i_mode >> 12;
713 dput(child);
714 return dir_emit(ctx, qname.name, 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(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(&head, &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(file_inode(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(head))
767 return PTR_ERR(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(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
777 if (!scan(h, entry, &pos, file, ctx)) {
778 sysctl_head_finish(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(head))
804 return PTR_ERR(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, 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(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(head))
842 return PTR_ERR(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(d_inode(dentry))->sysctl->unregistering;
886}
887
888static int proc_sys_delete(const struct dentry *dentry)
889{
890 return !!PROC_I(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(&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(&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(name->name, str, len))
923 return 1;
924 head = rcu_dereference(PROC_I(inode)->sysctl);
925 return !head || !sysctl_is_seen(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(&head, dir, name, namelen);
941 if (!entry)
942 return ERR_PTR(-ENOENT);
943 if (!S_ISDIR(entry->mode))
944 return ERR_PTR(-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(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(&new->header, set->dir.header.root, set, node, table, 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(&sysctl_lock);
993 subdir = find_subdir(dir, name, namelen);
994 if (!IS_ERR(subdir))
995 goto found;
996 if (PTR_ERR(subdir) != -ENOENT)
997 goto failed;
998
999 spin_unlock(&sysctl_lock);
1000 new = new_dir(set, name, namelen);
1001 spin_lock(&sysctl_lock);
1002 subdir = ERR_PTR(-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(subdir))
1009 goto found;
1010 if (PTR_ERR(subdir) != -ENOENT)
1011 goto failed;
1012
1013 /* Nope. Use the our freshly made directory entry. */
1014 err = insert_header(dir, &new->header);
1015 subdir = ERR_PTR(err);
1016 if (err)
1017 goto failed;
1018 subdir = new;
1019found:
1020 subdir->header.nreg++;
1021failed:
1022 if (IS_ERR(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(&dir->header);
1029 if (new)
1030 drop_sysctl_table(&new->header);
1031 spin_unlock(&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->header.parent);
1042 if (IS_ERR(parent))
1043 return parent;
1044 procname = dir->header.ctl_table[0].procname;
1045 return find_subdir(parent, 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(&sysctl_lock);
1059 root = (*pentry)->data;
1060 set = lookup_header_set(root);
1061 dir = xlate_dir(set, (*phead)->parent);
1062 if (IS_ERR(dir))
1063 ret = PTR_ERR(dir);
1064 else {
1065 const char *procname = (*pentry)->procname;
1066 head = NULL;
1067 entry = find_entry(&head, dir, procname, strlen(procname));
1068 ret = -ENOENT;
1069 if (entry && use_table(head)) {
1070 unuse_table(*phead);
1071 *phead = head;
1072 *pentry = entry;
1073 ret = 0;
1074 }
1075 }
1076
1077 spin_unlock(&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, "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, "array not allowed");
1110 }
1111
1112 if (table->proc_handler == proc_dobool) {
1113 if (table->maxlen != sizeof(bool))
1114 err |= sysctl_err(path, table, "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, entry, "No data");
1139 if (!entry->maxlen)
1140 err |= sysctl_err(path, entry, "No maxlen");
1141 else
1142 err |= sysctl_check_table_array(path, entry);
1143 }
1144 if (!entry->proc_handler)
1145 err |= sysctl_err(path, entry, "No proc_handler");
1146
1147 if ((entry->mode & (S_IRUGO|S_IWUGO)) != entry->mode)
1148 err |= sysctl_err(path, entry, "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(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(links, dir->header.root, dir->header.set, node, link_table,
1193 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(&tmp_head, dir, 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(&tmp_head, dir, 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(root_set, head->parent);
1243 if (IS_ERR(core_parent))
1244 return 0;
1245
1246 if (get_links(core_parent, head, head->root))
1247 return 0;
1248
1249 core_parent->header.nreg++;
1250 spin_unlock(&sysctl_lock);
1251
1252 links = new_links(core_parent, head);
1253
1254 spin_lock(&sysctl_lock);
1255 err = -ENOMEM;
1256 if (!links)
1257 goto out;
1258
1259 err = 0;
1260 if (get_links(core_parent, head, head->root)) {
1261 kfree(links);
1262 goto out;
1263 }
1264
1265 err = insert_header(core_parent, links);
1266 if (err)
1267 kfree(links);
1268out:
1269 drop_sysctl_table(&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(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(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(header, root, set, node, table, table_size);
1367 if (sysctl_check_table(path, header))
1368 goto fail;
1369
1370 spin_lock(&sysctl_lock);
1371 dir = &set->dir;
1372 /* Reference moved down the directory tree get_subdir */
1373 dir->header.nreg++;
1374 spin_unlock(&sysctl_lock);
1375
1376 dir = sysctl_mkdir_p(dir, path);
1377 if (IS_ERR(dir))
1378 goto fail;
1379 spin_lock(&sysctl_lock);
1380 if (insert_header(dir, header))
1381 goto fail_put_dir_locked;
1382
1383 drop_sysctl_table(&dir->header);
1384 spin_unlock(&sysctl_lock);
1385
1386 return header;
1387
1388fail_put_dir_locked:
1389 drop_sysctl_table(&dir->header);
1390 spin_unlock(&sysctl_lock);
1391fail:
1392 kfree(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(&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(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(root_set, parent);
1467 if (IS_ERR(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(&link_head, 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(link_head);
1480 }
1481 else {
1482 pr_err("sysctl link missing during unregister: ");
1483 sysctl_print_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(header);
1499 }
1500
1501 if (!--header->count)
1502 kfree_rcu(header, rcu);
1503
1504 if (parent)
1505 drop_sysctl_table(&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(&sysctl_lock);
1523 drop_sysctl_table(header);
1524 spin_unlock(&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(&set->dir.header, root, set, NULL, root_table, 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("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(proc_fs_type);
1643 if (IS_ERR(*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, "sys/%s", param);
1650 if (!path)
1651 panic("%s: Failed to allocate path for %s\n", __func__, param);
1652 strreplace(path, '.', '/');
1653
1654 file = file_open_root_mnt(*proc_mnt, path, O_WRONLY, 0);
1655 if (IS_ERR(file)) {
1656 err = PTR_ERR(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(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(saved_command_line, GFP_KERNEL);
1697 if (!command_line)
1698 panic("%s: Failed to allocate copy of command line\n", __func__);
1699
1700 parse_args("Setting sysctl args", command_line,
1701 NULL, 0, -1, -1, &proc_mnt, process_sysctl_arg);
1702
1703 if (proc_mnt)
1704 kern_unmount(proc_mnt);
1705
1706 kfree(command_line);
1707}
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 "internal.h"
20
21static const struct dentry_operations proc_sys_dentry_operations;
22static const struct file_operations proc_sys_file_operations;
23static const struct inode_operations proc_sys_inode_operations;
24static const struct file_operations proc_sys_dir_file_operations;
25static const struct inode_operations proc_sys_dir_operations;
26
27/* shared constants to be used in various sysctls */
28const int sysctl_vals[] = { 0, 1, INT_MAX };
29EXPORT_SYMBOL(sysctl_vals);
30
31/* Support for permanently empty directories */
32
33struct ctl_table sysctl_mount_point[] = {
34 { }
35};
36
37static bool is_empty_dir(struct ctl_table_header *head)
38{
39 return head->ctl_table[0].child == sysctl_mount_point;
40}
41
42static void set_empty_dir(struct ctl_dir *dir)
43{
44 dir->header.ctl_table[0].child = sysctl_mount_point;
45}
46
47static void clear_empty_dir(struct ctl_dir *dir)
48
49{
50 dir->header.ctl_table[0].child = NULL;
51}
52
53void proc_sys_poll_notify(struct ctl_table_poll *poll)
54{
55 if (!poll)
56 return;
57
58 atomic_inc(&poll->event);
59 wake_up_interruptible(&poll->wait);
60}
61
62static struct ctl_table root_table[] = {
63 {
64 .procname = "",
65 .mode = S_IFDIR|S_IRUGO|S_IXUGO,
66 },
67 { }
68};
69static struct ctl_table_root sysctl_table_root = {
70 .default_set.dir.header = {
71 {{.count = 1,
72 .nreg = 1,
73 .ctl_table = root_table }},
74 .ctl_table_arg = root_table,
75 .root = &sysctl_table_root,
76 .set = &sysctl_table_root.default_set,
77 },
78};
79
80static DEFINE_SPINLOCK(sysctl_lock);
81
82static void drop_sysctl_table(struct ctl_table_header *header);
83static int sysctl_follow_link(struct ctl_table_header **phead,
84 struct ctl_table **pentry);
85static int insert_links(struct ctl_table_header *head);
86static void put_links(struct ctl_table_header *header);
87
88static void sysctl_print_dir(struct ctl_dir *dir)
89{
90 if (dir->header.parent)
91 sysctl_print_dir(dir->header.parent);
92 pr_cont("%s/", dir->header.ctl_table[0].procname);
93}
94
95static int namecmp(const char *name1, int len1, const char *name2, int len2)
96{
97 int cmp;
98
99 cmp = memcmp(name1, name2, min(len1, len2));
100 if (cmp == 0)
101 cmp = len1 - len2;
102 return cmp;
103}
104
105/* Called under sysctl_lock */
106static struct ctl_table *find_entry(struct ctl_table_header **phead,
107 struct ctl_dir *dir, const char *name, int namelen)
108{
109 struct ctl_table_header *head;
110 struct ctl_table *entry;
111 struct rb_node *node = dir->root.rb_node;
112
113 while (node)
114 {
115 struct ctl_node *ctl_node;
116 const char *procname;
117 int cmp;
118
119 ctl_node = rb_entry(node, struct ctl_node, node);
120 head = ctl_node->header;
121 entry = &head->ctl_table[ctl_node - head->node];
122 procname = entry->procname;
123
124 cmp = namecmp(name, namelen, procname, strlen(procname));
125 if (cmp < 0)
126 node = node->rb_left;
127 else if (cmp > 0)
128 node = node->rb_right;
129 else {
130 *phead = head;
131 return entry;
132 }
133 }
134 return NULL;
135}
136
137static int insert_entry(struct ctl_table_header *head, struct ctl_table *entry)
138{
139 struct rb_node *node = &head->node[entry - head->ctl_table].node;
140 struct rb_node **p = &head->parent->root.rb_node;
141 struct rb_node *parent = NULL;
142 const char *name = entry->procname;
143 int namelen = strlen(name);
144
145 while (*p) {
146 struct ctl_table_header *parent_head;
147 struct ctl_table *parent_entry;
148 struct ctl_node *parent_node;
149 const char *parent_name;
150 int cmp;
151
152 parent = *p;
153 parent_node = rb_entry(parent, struct ctl_node, node);
154 parent_head = parent_node->header;
155 parent_entry = &parent_head->ctl_table[parent_node - parent_head->node];
156 parent_name = parent_entry->procname;
157
158 cmp = namecmp(name, namelen, parent_name, strlen(parent_name));
159 if (cmp < 0)
160 p = &(*p)->rb_left;
161 else if (cmp > 0)
162 p = &(*p)->rb_right;
163 else {
164 pr_err("sysctl duplicate entry: ");
165 sysctl_print_dir(head->parent);
166 pr_cont("/%s\n", entry->procname);
167 return -EEXIST;
168 }
169 }
170
171 rb_link_node(node, parent, p);
172 rb_insert_color(node, &head->parent->root);
173 return 0;
174}
175
176static void erase_entry(struct ctl_table_header *head, struct ctl_table *entry)
177{
178 struct rb_node *node = &head->node[entry - head->ctl_table].node;
179
180 rb_erase(node, &head->parent->root);
181}
182
183static void init_header(struct ctl_table_header *head,
184 struct ctl_table_root *root, struct ctl_table_set *set,
185 struct ctl_node *node, struct ctl_table *table)
186{
187 head->ctl_table = table;
188 head->ctl_table_arg = table;
189 head->used = 0;
190 head->count = 1;
191 head->nreg = 1;
192 head->unregistering = NULL;
193 head->root = root;
194 head->set = set;
195 head->parent = NULL;
196 head->node = node;
197 INIT_HLIST_HEAD(&head->inodes);
198 if (node) {
199 struct ctl_table *entry;
200 for (entry = table; entry->procname; entry++, node++)
201 node->header = head;
202 }
203}
204
205static void erase_header(struct ctl_table_header *head)
206{
207 struct ctl_table *entry;
208 for (entry = head->ctl_table; entry->procname; entry++)
209 erase_entry(head, entry);
210}
211
212static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
213{
214 struct ctl_table *entry;
215 int err;
216
217 /* Is this a permanently empty directory? */
218 if (is_empty_dir(&dir->header))
219 return -EROFS;
220
221 /* Am I creating a permanently empty directory? */
222 if (header->ctl_table == sysctl_mount_point) {
223 if (!RB_EMPTY_ROOT(&dir->root))
224 return -EINVAL;
225 set_empty_dir(dir);
226 }
227
228 dir->header.nreg++;
229 header->parent = dir;
230 err = insert_links(header);
231 if (err)
232 goto fail_links;
233 for (entry = header->ctl_table; entry->procname; entry++) {
234 err = insert_entry(header, entry);
235 if (err)
236 goto fail;
237 }
238 return 0;
239fail:
240 erase_header(header);
241 put_links(header);
242fail_links:
243 if (header->ctl_table == sysctl_mount_point)
244 clear_empty_dir(dir);
245 header->parent = NULL;
246 drop_sysctl_table(&dir->header);
247 return err;
248}
249
250/* called under sysctl_lock */
251static int use_table(struct ctl_table_header *p)
252{
253 if (unlikely(p->unregistering))
254 return 0;
255 p->used++;
256 return 1;
257}
258
259/* called under sysctl_lock */
260static void unuse_table(struct ctl_table_header *p)
261{
262 if (!--p->used)
263 if (unlikely(p->unregistering))
264 complete(p->unregistering);
265}
266
267static void proc_sys_invalidate_dcache(struct ctl_table_header *head)
268{
269 proc_invalidate_siblings_dcache(&head->inodes, &sysctl_lock);
270}
271
272/* called under sysctl_lock, will reacquire if has to wait */
273static void start_unregistering(struct ctl_table_header *p)
274{
275 /*
276 * if p->used is 0, nobody will ever touch that entry again;
277 * we'll eliminate all paths to it before dropping sysctl_lock
278 */
279 if (unlikely(p->used)) {
280 struct completion wait;
281 init_completion(&wait);
282 p->unregistering = &wait;
283 spin_unlock(&sysctl_lock);
284 wait_for_completion(&wait);
285 } else {
286 /* anything non-NULL; we'll never dereference it */
287 p->unregistering = ERR_PTR(-EINVAL);
288 spin_unlock(&sysctl_lock);
289 }
290 /*
291 * Invalidate dentries for unregistered sysctls: namespaced sysctls
292 * can have duplicate names and contaminate dcache very badly.
293 */
294 proc_sys_invalidate_dcache(p);
295 /*
296 * do not remove from the list until nobody holds it; walking the
297 * list in do_sysctl() relies on that.
298 */
299 spin_lock(&sysctl_lock);
300 erase_header(p);
301}
302
303static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
304{
305 BUG_ON(!head);
306 spin_lock(&sysctl_lock);
307 if (!use_table(head))
308 head = ERR_PTR(-ENOENT);
309 spin_unlock(&sysctl_lock);
310 return head;
311}
312
313static void sysctl_head_finish(struct ctl_table_header *head)
314{
315 if (!head)
316 return;
317 spin_lock(&sysctl_lock);
318 unuse_table(head);
319 spin_unlock(&sysctl_lock);
320}
321
322static struct ctl_table_set *
323lookup_header_set(struct ctl_table_root *root)
324{
325 struct ctl_table_set *set = &root->default_set;
326 if (root->lookup)
327 set = root->lookup(root);
328 return set;
329}
330
331static struct ctl_table *lookup_entry(struct ctl_table_header **phead,
332 struct ctl_dir *dir,
333 const char *name, int namelen)
334{
335 struct ctl_table_header *head;
336 struct ctl_table *entry;
337
338 spin_lock(&sysctl_lock);
339 entry = find_entry(&head, dir, name, namelen);
340 if (entry && use_table(head))
341 *phead = head;
342 else
343 entry = NULL;
344 spin_unlock(&sysctl_lock);
345 return entry;
346}
347
348static struct ctl_node *first_usable_entry(struct rb_node *node)
349{
350 struct ctl_node *ctl_node;
351
352 for (;node; node = rb_next(node)) {
353 ctl_node = rb_entry(node, struct ctl_node, node);
354 if (use_table(ctl_node->header))
355 return ctl_node;
356 }
357 return NULL;
358}
359
360static void first_entry(struct ctl_dir *dir,
361 struct ctl_table_header **phead, struct ctl_table **pentry)
362{
363 struct ctl_table_header *head = NULL;
364 struct ctl_table *entry = NULL;
365 struct ctl_node *ctl_node;
366
367 spin_lock(&sysctl_lock);
368 ctl_node = first_usable_entry(rb_first(&dir->root));
369 spin_unlock(&sysctl_lock);
370 if (ctl_node) {
371 head = ctl_node->header;
372 entry = &head->ctl_table[ctl_node - head->node];
373 }
374 *phead = head;
375 *pentry = entry;
376}
377
378static void next_entry(struct ctl_table_header **phead, struct ctl_table **pentry)
379{
380 struct ctl_table_header *head = *phead;
381 struct ctl_table *entry = *pentry;
382 struct ctl_node *ctl_node = &head->node[entry - head->ctl_table];
383
384 spin_lock(&sysctl_lock);
385 unuse_table(head);
386
387 ctl_node = first_usable_entry(rb_next(&ctl_node->node));
388 spin_unlock(&sysctl_lock);
389 head = NULL;
390 if (ctl_node) {
391 head = ctl_node->header;
392 entry = &head->ctl_table[ctl_node - head->node];
393 }
394 *phead = head;
395 *pentry = entry;
396}
397
398/*
399 * sysctl_perm does NOT grant the superuser all rights automatically, because
400 * some sysctl variables are readonly even to root.
401 */
402
403static int test_perm(int mode, int op)
404{
405 if (uid_eq(current_euid(), GLOBAL_ROOT_UID))
406 mode >>= 6;
407 else if (in_egroup_p(GLOBAL_ROOT_GID))
408 mode >>= 3;
409 if ((op & ~mode & (MAY_READ|MAY_WRITE|MAY_EXEC)) == 0)
410 return 0;
411 return -EACCES;
412}
413
414static int sysctl_perm(struct ctl_table_header *head, struct ctl_table *table, int op)
415{
416 struct ctl_table_root *root = head->root;
417 int mode;
418
419 if (root->permissions)
420 mode = root->permissions(head, table);
421 else
422 mode = table->mode;
423
424 return test_perm(mode, op);
425}
426
427static struct inode *proc_sys_make_inode(struct super_block *sb,
428 struct ctl_table_header *head, struct ctl_table *table)
429{
430 struct ctl_table_root *root = head->root;
431 struct inode *inode;
432 struct proc_inode *ei;
433
434 inode = new_inode(sb);
435 if (!inode)
436 return ERR_PTR(-ENOMEM);
437
438 inode->i_ino = get_next_ino();
439
440 ei = PROC_I(inode);
441
442 spin_lock(&sysctl_lock);
443 if (unlikely(head->unregistering)) {
444 spin_unlock(&sysctl_lock);
445 iput(inode);
446 return ERR_PTR(-ENOENT);
447 }
448 ei->sysctl = head;
449 ei->sysctl_entry = table;
450 hlist_add_head_rcu(&ei->sibling_inodes, &head->inodes);
451 head->count++;
452 spin_unlock(&sysctl_lock);
453
454 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
455 inode->i_mode = table->mode;
456 if (!S_ISDIR(table->mode)) {
457 inode->i_mode |= S_IFREG;
458 inode->i_op = &proc_sys_inode_operations;
459 inode->i_fop = &proc_sys_file_operations;
460 } else {
461 inode->i_mode |= S_IFDIR;
462 inode->i_op = &proc_sys_dir_operations;
463 inode->i_fop = &proc_sys_dir_file_operations;
464 if (is_empty_dir(head))
465 make_empty_dir_inode(inode);
466 }
467
468 if (root->set_ownership)
469 root->set_ownership(head, table, &inode->i_uid, &inode->i_gid);
470 else {
471 inode->i_uid = GLOBAL_ROOT_UID;
472 inode->i_gid = GLOBAL_ROOT_GID;
473 }
474
475 return inode;
476}
477
478void proc_sys_evict_inode(struct inode *inode, struct ctl_table_header *head)
479{
480 spin_lock(&sysctl_lock);
481 hlist_del_init_rcu(&PROC_I(inode)->sibling_inodes);
482 if (!--head->count)
483 kfree_rcu(head, rcu);
484 spin_unlock(&sysctl_lock);
485}
486
487static struct ctl_table_header *grab_header(struct inode *inode)
488{
489 struct ctl_table_header *head = PROC_I(inode)->sysctl;
490 if (!head)
491 head = &sysctl_table_root.default_set.dir.header;
492 return sysctl_head_grab(head);
493}
494
495static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
496 unsigned int flags)
497{
498 struct ctl_table_header *head = grab_header(dir);
499 struct ctl_table_header *h = NULL;
500 const struct qstr *name = &dentry->d_name;
501 struct ctl_table *p;
502 struct inode *inode;
503 struct dentry *err = ERR_PTR(-ENOENT);
504 struct ctl_dir *ctl_dir;
505 int ret;
506
507 if (IS_ERR(head))
508 return ERR_CAST(head);
509
510 ctl_dir = container_of(head, struct ctl_dir, header);
511
512 p = lookup_entry(&h, ctl_dir, name->name, name->len);
513 if (!p)
514 goto out;
515
516 if (S_ISLNK(p->mode)) {
517 ret = sysctl_follow_link(&h, &p);
518 err = ERR_PTR(ret);
519 if (ret)
520 goto out;
521 }
522
523 inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
524 if (IS_ERR(inode)) {
525 err = ERR_CAST(inode);
526 goto out;
527 }
528
529 d_set_d_op(dentry, &proc_sys_dentry_operations);
530 err = d_splice_alias(inode, dentry);
531
532out:
533 if (h)
534 sysctl_head_finish(h);
535 sysctl_head_finish(head);
536 return err;
537}
538
539static ssize_t proc_sys_call_handler(struct kiocb *iocb, struct iov_iter *iter,
540 int write)
541{
542 struct inode *inode = file_inode(iocb->ki_filp);
543 struct ctl_table_header *head = grab_header(inode);
544 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
545 size_t count = iov_iter_count(iter);
546 char *kbuf;
547 ssize_t error;
548
549 if (IS_ERR(head))
550 return PTR_ERR(head);
551
552 /*
553 * At this point we know that the sysctl was not unregistered
554 * and won't be until we finish.
555 */
556 error = -EPERM;
557 if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
558 goto out;
559
560 /* if that can happen at all, it should be -EINVAL, not -EISDIR */
561 error = -EINVAL;
562 if (!table->proc_handler)
563 goto out;
564
565 /* don't even try if the size is too large */
566 error = -ENOMEM;
567 if (count >= KMALLOC_MAX_SIZE)
568 goto out;
569 kbuf = kvzalloc(count + 1, GFP_KERNEL);
570 if (!kbuf)
571 goto out;
572
573 if (write) {
574 error = -EFAULT;
575 if (!copy_from_iter_full(kbuf, count, iter))
576 goto out_free_buf;
577 kbuf[count] = '\0';
578 }
579
580 error = BPF_CGROUP_RUN_PROG_SYSCTL(head, table, write, &kbuf, &count,
581 &iocb->ki_pos);
582 if (error)
583 goto out_free_buf;
584
585 /* careful: calling conventions are nasty here */
586 error = table->proc_handler(table, write, kbuf, &count, &iocb->ki_pos);
587 if (error)
588 goto out_free_buf;
589
590 if (!write) {
591 error = -EFAULT;
592 if (copy_to_iter(kbuf, count, iter) < count)
593 goto out_free_buf;
594 }
595
596 error = count;
597out_free_buf:
598 kvfree(kbuf);
599out:
600 sysctl_head_finish(head);
601
602 return error;
603}
604
605static ssize_t proc_sys_read(struct kiocb *iocb, struct iov_iter *iter)
606{
607 return proc_sys_call_handler(iocb, iter, 0);
608}
609
610static ssize_t proc_sys_write(struct kiocb *iocb, struct iov_iter *iter)
611{
612 return proc_sys_call_handler(iocb, iter, 1);
613}
614
615static int proc_sys_open(struct inode *inode, struct file *filp)
616{
617 struct ctl_table_header *head = grab_header(inode);
618 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
619
620 /* sysctl was unregistered */
621 if (IS_ERR(head))
622 return PTR_ERR(head);
623
624 if (table->poll)
625 filp->private_data = proc_sys_poll_event(table->poll);
626
627 sysctl_head_finish(head);
628
629 return 0;
630}
631
632static __poll_t proc_sys_poll(struct file *filp, poll_table *wait)
633{
634 struct inode *inode = file_inode(filp);
635 struct ctl_table_header *head = grab_header(inode);
636 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
637 __poll_t ret = DEFAULT_POLLMASK;
638 unsigned long event;
639
640 /* sysctl was unregistered */
641 if (IS_ERR(head))
642 return EPOLLERR | EPOLLHUP;
643
644 if (!table->proc_handler)
645 goto out;
646
647 if (!table->poll)
648 goto out;
649
650 event = (unsigned long)filp->private_data;
651 poll_wait(filp, &table->poll->wait, wait);
652
653 if (event != atomic_read(&table->poll->event)) {
654 filp->private_data = proc_sys_poll_event(table->poll);
655 ret = EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
656 }
657
658out:
659 sysctl_head_finish(head);
660
661 return ret;
662}
663
664static bool proc_sys_fill_cache(struct file *file,
665 struct dir_context *ctx,
666 struct ctl_table_header *head,
667 struct ctl_table *table)
668{
669 struct dentry *child, *dir = file->f_path.dentry;
670 struct inode *inode;
671 struct qstr qname;
672 ino_t ino = 0;
673 unsigned type = DT_UNKNOWN;
674
675 qname.name = table->procname;
676 qname.len = strlen(table->procname);
677 qname.hash = full_name_hash(dir, qname.name, qname.len);
678
679 child = d_lookup(dir, &qname);
680 if (!child) {
681 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
682 child = d_alloc_parallel(dir, &qname, &wq);
683 if (IS_ERR(child))
684 return false;
685 if (d_in_lookup(child)) {
686 struct dentry *res;
687 inode = proc_sys_make_inode(dir->d_sb, head, table);
688 if (IS_ERR(inode)) {
689 d_lookup_done(child);
690 dput(child);
691 return false;
692 }
693 d_set_d_op(child, &proc_sys_dentry_operations);
694 res = d_splice_alias(inode, child);
695 d_lookup_done(child);
696 if (unlikely(res)) {
697 if (IS_ERR(res)) {
698 dput(child);
699 return false;
700 }
701 dput(child);
702 child = res;
703 }
704 }
705 }
706 inode = d_inode(child);
707 ino = inode->i_ino;
708 type = inode->i_mode >> 12;
709 dput(child);
710 return dir_emit(ctx, qname.name, qname.len, ino, type);
711}
712
713static bool proc_sys_link_fill_cache(struct file *file,
714 struct dir_context *ctx,
715 struct ctl_table_header *head,
716 struct ctl_table *table)
717{
718 bool ret = true;
719
720 head = sysctl_head_grab(head);
721 if (IS_ERR(head))
722 return false;
723
724 /* It is not an error if we can not follow the link ignore it */
725 if (sysctl_follow_link(&head, &table))
726 goto out;
727
728 ret = proc_sys_fill_cache(file, ctx, head, table);
729out:
730 sysctl_head_finish(head);
731 return ret;
732}
733
734static int scan(struct ctl_table_header *head, struct ctl_table *table,
735 unsigned long *pos, struct file *file,
736 struct dir_context *ctx)
737{
738 bool res;
739
740 if ((*pos)++ < ctx->pos)
741 return true;
742
743 if (unlikely(S_ISLNK(table->mode)))
744 res = proc_sys_link_fill_cache(file, ctx, head, table);
745 else
746 res = proc_sys_fill_cache(file, ctx, head, table);
747
748 if (res)
749 ctx->pos = *pos;
750
751 return res;
752}
753
754static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
755{
756 struct ctl_table_header *head = grab_header(file_inode(file));
757 struct ctl_table_header *h = NULL;
758 struct ctl_table *entry;
759 struct ctl_dir *ctl_dir;
760 unsigned long pos;
761
762 if (IS_ERR(head))
763 return PTR_ERR(head);
764
765 ctl_dir = container_of(head, struct ctl_dir, header);
766
767 if (!dir_emit_dots(file, ctx))
768 goto out;
769
770 pos = 2;
771
772 for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
773 if (!scan(h, entry, &pos, file, ctx)) {
774 sysctl_head_finish(h);
775 break;
776 }
777 }
778out:
779 sysctl_head_finish(head);
780 return 0;
781}
782
783static int proc_sys_permission(struct user_namespace *mnt_userns,
784 struct inode *inode, int mask)
785{
786 /*
787 * sysctl entries that are not writeable,
788 * are _NOT_ writeable, capabilities or not.
789 */
790 struct ctl_table_header *head;
791 struct ctl_table *table;
792 int error;
793
794 /* Executable files are not allowed under /proc/sys/ */
795 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
796 return -EACCES;
797
798 head = grab_header(inode);
799 if (IS_ERR(head))
800 return PTR_ERR(head);
801
802 table = PROC_I(inode)->sysctl_entry;
803 if (!table) /* global root - r-xr-xr-x */
804 error = mask & MAY_WRITE ? -EACCES : 0;
805 else /* Use the permissions on the sysctl table entry */
806 error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
807
808 sysctl_head_finish(head);
809 return error;
810}
811
812static int proc_sys_setattr(struct user_namespace *mnt_userns,
813 struct dentry *dentry, struct iattr *attr)
814{
815 struct inode *inode = d_inode(dentry);
816 int error;
817
818 if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
819 return -EPERM;
820
821 error = setattr_prepare(&init_user_ns, dentry, attr);
822 if (error)
823 return error;
824
825 setattr_copy(&init_user_ns, inode, attr);
826 mark_inode_dirty(inode);
827 return 0;
828}
829
830static int proc_sys_getattr(struct user_namespace *mnt_userns,
831 const struct path *path, struct kstat *stat,
832 u32 request_mask, unsigned int query_flags)
833{
834 struct inode *inode = d_inode(path->dentry);
835 struct ctl_table_header *head = grab_header(inode);
836 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
837
838 if (IS_ERR(head))
839 return PTR_ERR(head);
840
841 generic_fillattr(&init_user_ns, inode, stat);
842 if (table)
843 stat->mode = (stat->mode & S_IFMT) | table->mode;
844
845 sysctl_head_finish(head);
846 return 0;
847}
848
849static const struct file_operations proc_sys_file_operations = {
850 .open = proc_sys_open,
851 .poll = proc_sys_poll,
852 .read_iter = proc_sys_read,
853 .write_iter = proc_sys_write,
854 .splice_read = generic_file_splice_read,
855 .splice_write = iter_file_splice_write,
856 .llseek = default_llseek,
857};
858
859static const struct file_operations proc_sys_dir_file_operations = {
860 .read = generic_read_dir,
861 .iterate_shared = proc_sys_readdir,
862 .llseek = generic_file_llseek,
863};
864
865static const struct inode_operations proc_sys_inode_operations = {
866 .permission = proc_sys_permission,
867 .setattr = proc_sys_setattr,
868 .getattr = proc_sys_getattr,
869};
870
871static const struct inode_operations proc_sys_dir_operations = {
872 .lookup = proc_sys_lookup,
873 .permission = proc_sys_permission,
874 .setattr = proc_sys_setattr,
875 .getattr = proc_sys_getattr,
876};
877
878static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
879{
880 if (flags & LOOKUP_RCU)
881 return -ECHILD;
882 return !PROC_I(d_inode(dentry))->sysctl->unregistering;
883}
884
885static int proc_sys_delete(const struct dentry *dentry)
886{
887 return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
888}
889
890static int sysctl_is_seen(struct ctl_table_header *p)
891{
892 struct ctl_table_set *set = p->set;
893 int res;
894 spin_lock(&sysctl_lock);
895 if (p->unregistering)
896 res = 0;
897 else if (!set->is_seen)
898 res = 1;
899 else
900 res = set->is_seen(set);
901 spin_unlock(&sysctl_lock);
902 return res;
903}
904
905static int proc_sys_compare(const struct dentry *dentry,
906 unsigned int len, const char *str, const struct qstr *name)
907{
908 struct ctl_table_header *head;
909 struct inode *inode;
910
911 /* Although proc doesn't have negative dentries, rcu-walk means
912 * that inode here can be NULL */
913 /* AV: can it, indeed? */
914 inode = d_inode_rcu(dentry);
915 if (!inode)
916 return 1;
917 if (name->len != len)
918 return 1;
919 if (memcmp(name->name, str, len))
920 return 1;
921 head = rcu_dereference(PROC_I(inode)->sysctl);
922 return !head || !sysctl_is_seen(head);
923}
924
925static const struct dentry_operations proc_sys_dentry_operations = {
926 .d_revalidate = proc_sys_revalidate,
927 .d_delete = proc_sys_delete,
928 .d_compare = proc_sys_compare,
929};
930
931static struct ctl_dir *find_subdir(struct ctl_dir *dir,
932 const char *name, int namelen)
933{
934 struct ctl_table_header *head;
935 struct ctl_table *entry;
936
937 entry = find_entry(&head, dir, name, namelen);
938 if (!entry)
939 return ERR_PTR(-ENOENT);
940 if (!S_ISDIR(entry->mode))
941 return ERR_PTR(-ENOTDIR);
942 return container_of(head, struct ctl_dir, header);
943}
944
945static struct ctl_dir *new_dir(struct ctl_table_set *set,
946 const char *name, int namelen)
947{
948 struct ctl_table *table;
949 struct ctl_dir *new;
950 struct ctl_node *node;
951 char *new_name;
952
953 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
954 sizeof(struct ctl_table)*2 + namelen + 1,
955 GFP_KERNEL);
956 if (!new)
957 return NULL;
958
959 node = (struct ctl_node *)(new + 1);
960 table = (struct ctl_table *)(node + 1);
961 new_name = (char *)(table + 2);
962 memcpy(new_name, name, namelen);
963 new_name[namelen] = '\0';
964 table[0].procname = new_name;
965 table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
966 init_header(&new->header, set->dir.header.root, set, node, table);
967
968 return new;
969}
970
971/**
972 * get_subdir - find or create a subdir with the specified name.
973 * @dir: Directory to create the subdirectory in
974 * @name: The name of the subdirectory to find or create
975 * @namelen: The length of name
976 *
977 * Takes a directory with an elevated reference count so we know that
978 * if we drop the lock the directory will not go away. Upon success
979 * the reference is moved from @dir to the returned subdirectory.
980 * Upon error an error code is returned and the reference on @dir is
981 * simply dropped.
982 */
983static struct ctl_dir *get_subdir(struct ctl_dir *dir,
984 const char *name, int namelen)
985{
986 struct ctl_table_set *set = dir->header.set;
987 struct ctl_dir *subdir, *new = NULL;
988 int err;
989
990 spin_lock(&sysctl_lock);
991 subdir = find_subdir(dir, name, namelen);
992 if (!IS_ERR(subdir))
993 goto found;
994 if (PTR_ERR(subdir) != -ENOENT)
995 goto failed;
996
997 spin_unlock(&sysctl_lock);
998 new = new_dir(set, name, namelen);
999 spin_lock(&sysctl_lock);
1000 subdir = ERR_PTR(-ENOMEM);
1001 if (!new)
1002 goto failed;
1003
1004 /* Was the subdir added while we dropped the lock? */
1005 subdir = find_subdir(dir, name, namelen);
1006 if (!IS_ERR(subdir))
1007 goto found;
1008 if (PTR_ERR(subdir) != -ENOENT)
1009 goto failed;
1010
1011 /* Nope. Use the our freshly made directory entry. */
1012 err = insert_header(dir, &new->header);
1013 subdir = ERR_PTR(err);
1014 if (err)
1015 goto failed;
1016 subdir = new;
1017found:
1018 subdir->header.nreg++;
1019failed:
1020 if (IS_ERR(subdir)) {
1021 pr_err("sysctl could not get directory: ");
1022 sysctl_print_dir(dir);
1023 pr_cont("/%*.*s %ld\n",
1024 namelen, namelen, name, PTR_ERR(subdir));
1025 }
1026 drop_sysctl_table(&dir->header);
1027 if (new)
1028 drop_sysctl_table(&new->header);
1029 spin_unlock(&sysctl_lock);
1030 return subdir;
1031}
1032
1033static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
1034{
1035 struct ctl_dir *parent;
1036 const char *procname;
1037 if (!dir->header.parent)
1038 return &set->dir;
1039 parent = xlate_dir(set, dir->header.parent);
1040 if (IS_ERR(parent))
1041 return parent;
1042 procname = dir->header.ctl_table[0].procname;
1043 return find_subdir(parent, procname, strlen(procname));
1044}
1045
1046static int sysctl_follow_link(struct ctl_table_header **phead,
1047 struct ctl_table **pentry)
1048{
1049 struct ctl_table_header *head;
1050 struct ctl_table_root *root;
1051 struct ctl_table_set *set;
1052 struct ctl_table *entry;
1053 struct ctl_dir *dir;
1054 int ret;
1055
1056 ret = 0;
1057 spin_lock(&sysctl_lock);
1058 root = (*pentry)->data;
1059 set = lookup_header_set(root);
1060 dir = xlate_dir(set, (*phead)->parent);
1061 if (IS_ERR(dir))
1062 ret = PTR_ERR(dir);
1063 else {
1064 const char *procname = (*pentry)->procname;
1065 head = NULL;
1066 entry = find_entry(&head, dir, procname, strlen(procname));
1067 ret = -ENOENT;
1068 if (entry && use_table(head)) {
1069 unuse_table(*phead);
1070 *phead = head;
1071 *pentry = entry;
1072 ret = 0;
1073 }
1074 }
1075
1076 spin_unlock(&sysctl_lock);
1077 return ret;
1078}
1079
1080static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
1081{
1082 struct va_format vaf;
1083 va_list args;
1084
1085 va_start(args, fmt);
1086 vaf.fmt = fmt;
1087 vaf.va = &args;
1088
1089 pr_err("sysctl table check failed: %s/%s %pV\n",
1090 path, table->procname, &vaf);
1091
1092 va_end(args);
1093 return -EINVAL;
1094}
1095
1096static int sysctl_check_table_array(const char *path, struct ctl_table *table)
1097{
1098 int err = 0;
1099
1100 if ((table->proc_handler == proc_douintvec) ||
1101 (table->proc_handler == proc_douintvec_minmax)) {
1102 if (table->maxlen != sizeof(unsigned int))
1103 err |= sysctl_err(path, table, "array not allowed");
1104 }
1105
1106 if (table->proc_handler == proc_dou8vec_minmax) {
1107 if (table->maxlen != sizeof(u8))
1108 err |= sysctl_err(path, table, "array not allowed");
1109 }
1110
1111 return err;
1112}
1113
1114static int sysctl_check_table(const char *path, struct ctl_table *table)
1115{
1116 int err = 0;
1117 for (; table->procname; table++) {
1118 if (table->child)
1119 err |= sysctl_err(path, table, "Not a file");
1120
1121 if ((table->proc_handler == proc_dostring) ||
1122 (table->proc_handler == proc_dointvec) ||
1123 (table->proc_handler == proc_douintvec) ||
1124 (table->proc_handler == proc_douintvec_minmax) ||
1125 (table->proc_handler == proc_dointvec_minmax) ||
1126 (table->proc_handler == proc_dou8vec_minmax) ||
1127 (table->proc_handler == proc_dointvec_jiffies) ||
1128 (table->proc_handler == proc_dointvec_userhz_jiffies) ||
1129 (table->proc_handler == proc_dointvec_ms_jiffies) ||
1130 (table->proc_handler == proc_doulongvec_minmax) ||
1131 (table->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1132 if (!table->data)
1133 err |= sysctl_err(path, table, "No data");
1134 if (!table->maxlen)
1135 err |= sysctl_err(path, table, "No maxlen");
1136 else
1137 err |= sysctl_check_table_array(path, table);
1138 }
1139 if (!table->proc_handler)
1140 err |= sysctl_err(path, table, "No proc_handler");
1141
1142 if ((table->mode & (S_IRUGO|S_IWUGO)) != table->mode)
1143 err |= sysctl_err(path, table, "bogus .mode 0%o",
1144 table->mode);
1145 }
1146 return err;
1147}
1148
1149static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table *table,
1150 struct ctl_table_root *link_root)
1151{
1152 struct ctl_table *link_table, *entry, *link;
1153 struct ctl_table_header *links;
1154 struct ctl_node *node;
1155 char *link_name;
1156 int nr_entries, name_bytes;
1157
1158 name_bytes = 0;
1159 nr_entries = 0;
1160 for (entry = table; entry->procname; entry++) {
1161 nr_entries++;
1162 name_bytes += strlen(entry->procname) + 1;
1163 }
1164
1165 links = kzalloc(sizeof(struct ctl_table_header) +
1166 sizeof(struct ctl_node)*nr_entries +
1167 sizeof(struct ctl_table)*(nr_entries + 1) +
1168 name_bytes,
1169 GFP_KERNEL);
1170
1171 if (!links)
1172 return NULL;
1173
1174 node = (struct ctl_node *)(links + 1);
1175 link_table = (struct ctl_table *)(node + nr_entries);
1176 link_name = (char *)&link_table[nr_entries + 1];
1177
1178 for (link = link_table, entry = table; entry->procname; link++, entry++) {
1179 int len = strlen(entry->procname) + 1;
1180 memcpy(link_name, entry->procname, len);
1181 link->procname = link_name;
1182 link->mode = S_IFLNK|S_IRWXUGO;
1183 link->data = link_root;
1184 link_name += len;
1185 }
1186 init_header(links, dir->header.root, dir->header.set, node, link_table);
1187 links->nreg = nr_entries;
1188
1189 return links;
1190}
1191
1192static bool get_links(struct ctl_dir *dir,
1193 struct ctl_table *table, struct ctl_table_root *link_root)
1194{
1195 struct ctl_table_header *head;
1196 struct ctl_table *entry, *link;
1197
1198 /* Are there links available for every entry in table? */
1199 for (entry = table; entry->procname; entry++) {
1200 const char *procname = entry->procname;
1201 link = find_entry(&head, dir, procname, strlen(procname));
1202 if (!link)
1203 return false;
1204 if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1205 continue;
1206 if (S_ISLNK(link->mode) && (link->data == link_root))
1207 continue;
1208 return false;
1209 }
1210
1211 /* The checks passed. Increase the registration count on the links */
1212 for (entry = table; entry->procname; entry++) {
1213 const char *procname = entry->procname;
1214 link = find_entry(&head, dir, procname, strlen(procname));
1215 head->nreg++;
1216 }
1217 return true;
1218}
1219
1220static int insert_links(struct ctl_table_header *head)
1221{
1222 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1223 struct ctl_dir *core_parent = NULL;
1224 struct ctl_table_header *links;
1225 int err;
1226
1227 if (head->set == root_set)
1228 return 0;
1229
1230 core_parent = xlate_dir(root_set, head->parent);
1231 if (IS_ERR(core_parent))
1232 return 0;
1233
1234 if (get_links(core_parent, head->ctl_table, head->root))
1235 return 0;
1236
1237 core_parent->header.nreg++;
1238 spin_unlock(&sysctl_lock);
1239
1240 links = new_links(core_parent, head->ctl_table, head->root);
1241
1242 spin_lock(&sysctl_lock);
1243 err = -ENOMEM;
1244 if (!links)
1245 goto out;
1246
1247 err = 0;
1248 if (get_links(core_parent, head->ctl_table, head->root)) {
1249 kfree(links);
1250 goto out;
1251 }
1252
1253 err = insert_header(core_parent, links);
1254 if (err)
1255 kfree(links);
1256out:
1257 drop_sysctl_table(&core_parent->header);
1258 return err;
1259}
1260
1261/**
1262 * __register_sysctl_table - register a leaf sysctl table
1263 * @set: Sysctl tree to register on
1264 * @path: The path to the directory the sysctl table is in.
1265 * @table: the top-level table structure
1266 *
1267 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1268 * array. A completely 0 filled entry terminates the table.
1269 *
1270 * The members of the &struct ctl_table structure are used as follows:
1271 *
1272 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1273 * enter a sysctl file
1274 *
1275 * data - a pointer to data for use by proc_handler
1276 *
1277 * maxlen - the maximum size in bytes of the data
1278 *
1279 * mode - the file permissions for the /proc/sys file
1280 *
1281 * child - must be %NULL.
1282 *
1283 * proc_handler - the text handler routine (described below)
1284 *
1285 * extra1, extra2 - extra pointers usable by the proc handler routines
1286 *
1287 * Leaf nodes in the sysctl tree will be represented by a single file
1288 * under /proc; non-leaf nodes will be represented by directories.
1289 *
1290 * There must be a proc_handler routine for any terminal nodes.
1291 * Several default handlers are available to cover common cases -
1292 *
1293 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1294 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1295 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1296 *
1297 * It is the handler's job to read the input buffer from user memory
1298 * and process it. The handler should return 0 on success.
1299 *
1300 * This routine returns %NULL on a failure to register, and a pointer
1301 * to the table header on success.
1302 */
1303struct ctl_table_header *__register_sysctl_table(
1304 struct ctl_table_set *set,
1305 const char *path, struct ctl_table *table)
1306{
1307 struct ctl_table_root *root = set->dir.header.root;
1308 struct ctl_table_header *header;
1309 const char *name, *nextname;
1310 struct ctl_dir *dir;
1311 struct ctl_table *entry;
1312 struct ctl_node *node;
1313 int nr_entries = 0;
1314
1315 for (entry = table; entry->procname; entry++)
1316 nr_entries++;
1317
1318 header = kzalloc(sizeof(struct ctl_table_header) +
1319 sizeof(struct ctl_node)*nr_entries, GFP_KERNEL);
1320 if (!header)
1321 return NULL;
1322
1323 node = (struct ctl_node *)(header + 1);
1324 init_header(header, root, set, node, table);
1325 if (sysctl_check_table(path, table))
1326 goto fail;
1327
1328 spin_lock(&sysctl_lock);
1329 dir = &set->dir;
1330 /* Reference moved down the diretory tree get_subdir */
1331 dir->header.nreg++;
1332 spin_unlock(&sysctl_lock);
1333
1334 /* Find the directory for the ctl_table */
1335 for (name = path; name; name = nextname) {
1336 int namelen;
1337 nextname = strchr(name, '/');
1338 if (nextname) {
1339 namelen = nextname - name;
1340 nextname++;
1341 } else {
1342 namelen = strlen(name);
1343 }
1344 if (namelen == 0)
1345 continue;
1346
1347 dir = get_subdir(dir, name, namelen);
1348 if (IS_ERR(dir))
1349 goto fail;
1350 }
1351
1352 spin_lock(&sysctl_lock);
1353 if (insert_header(dir, header))
1354 goto fail_put_dir_locked;
1355
1356 drop_sysctl_table(&dir->header);
1357 spin_unlock(&sysctl_lock);
1358
1359 return header;
1360
1361fail_put_dir_locked:
1362 drop_sysctl_table(&dir->header);
1363 spin_unlock(&sysctl_lock);
1364fail:
1365 kfree(header);
1366 dump_stack();
1367 return NULL;
1368}
1369
1370/**
1371 * register_sysctl - register a sysctl table
1372 * @path: The path to the directory the sysctl table is in.
1373 * @table: the table structure
1374 *
1375 * Register a sysctl table. @table should be a filled in ctl_table
1376 * array. A completely 0 filled entry terminates the table.
1377 *
1378 * See __register_sysctl_table for more details.
1379 */
1380struct ctl_table_header *register_sysctl(const char *path, struct ctl_table *table)
1381{
1382 return __register_sysctl_table(&sysctl_table_root.default_set,
1383 path, table);
1384}
1385EXPORT_SYMBOL(register_sysctl);
1386
1387static char *append_path(const char *path, char *pos, const char *name)
1388{
1389 int namelen;
1390 namelen = strlen(name);
1391 if (((pos - path) + namelen + 2) >= PATH_MAX)
1392 return NULL;
1393 memcpy(pos, name, namelen);
1394 pos[namelen] = '/';
1395 pos[namelen + 1] = '\0';
1396 pos += namelen + 1;
1397 return pos;
1398}
1399
1400static int count_subheaders(struct ctl_table *table)
1401{
1402 int has_files = 0;
1403 int nr_subheaders = 0;
1404 struct ctl_table *entry;
1405
1406 /* special case: no directory and empty directory */
1407 if (!table || !table->procname)
1408 return 1;
1409
1410 for (entry = table; entry->procname; entry++) {
1411 if (entry->child)
1412 nr_subheaders += count_subheaders(entry->child);
1413 else
1414 has_files = 1;
1415 }
1416 return nr_subheaders + has_files;
1417}
1418
1419static int register_leaf_sysctl_tables(const char *path, char *pos,
1420 struct ctl_table_header ***subheader, struct ctl_table_set *set,
1421 struct ctl_table *table)
1422{
1423 struct ctl_table *ctl_table_arg = NULL;
1424 struct ctl_table *entry, *files;
1425 int nr_files = 0;
1426 int nr_dirs = 0;
1427 int err = -ENOMEM;
1428
1429 for (entry = table; entry->procname; entry++) {
1430 if (entry->child)
1431 nr_dirs++;
1432 else
1433 nr_files++;
1434 }
1435
1436 files = table;
1437 /* If there are mixed files and directories we need a new table */
1438 if (nr_dirs && nr_files) {
1439 struct ctl_table *new;
1440 files = kcalloc(nr_files + 1, sizeof(struct ctl_table),
1441 GFP_KERNEL);
1442 if (!files)
1443 goto out;
1444
1445 ctl_table_arg = files;
1446 for (new = files, entry = table; entry->procname; entry++) {
1447 if (entry->child)
1448 continue;
1449 *new = *entry;
1450 new++;
1451 }
1452 }
1453
1454 /* Register everything except a directory full of subdirectories */
1455 if (nr_files || !nr_dirs) {
1456 struct ctl_table_header *header;
1457 header = __register_sysctl_table(set, path, files);
1458 if (!header) {
1459 kfree(ctl_table_arg);
1460 goto out;
1461 }
1462
1463 /* Remember if we need to free the file table */
1464 header->ctl_table_arg = ctl_table_arg;
1465 **subheader = header;
1466 (*subheader)++;
1467 }
1468
1469 /* Recurse into the subdirectories. */
1470 for (entry = table; entry->procname; entry++) {
1471 char *child_pos;
1472
1473 if (!entry->child)
1474 continue;
1475
1476 err = -ENAMETOOLONG;
1477 child_pos = append_path(path, pos, entry->procname);
1478 if (!child_pos)
1479 goto out;
1480
1481 err = register_leaf_sysctl_tables(path, child_pos, subheader,
1482 set, entry->child);
1483 pos[0] = '\0';
1484 if (err)
1485 goto out;
1486 }
1487 err = 0;
1488out:
1489 /* On failure our caller will unregister all registered subheaders */
1490 return err;
1491}
1492
1493/**
1494 * __register_sysctl_paths - register a sysctl table hierarchy
1495 * @set: Sysctl tree to register on
1496 * @path: The path to the directory the sysctl table is in.
1497 * @table: the top-level table structure
1498 *
1499 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1500 * array. A completely 0 filled entry terminates the table.
1501 *
1502 * See __register_sysctl_table for more details.
1503 */
1504struct ctl_table_header *__register_sysctl_paths(
1505 struct ctl_table_set *set,
1506 const struct ctl_path *path, struct ctl_table *table)
1507{
1508 struct ctl_table *ctl_table_arg = table;
1509 int nr_subheaders = count_subheaders(table);
1510 struct ctl_table_header *header = NULL, **subheaders, **subheader;
1511 const struct ctl_path *component;
1512 char *new_path, *pos;
1513
1514 pos = new_path = kmalloc(PATH_MAX, GFP_KERNEL);
1515 if (!new_path)
1516 return NULL;
1517
1518 pos[0] = '\0';
1519 for (component = path; component->procname; component++) {
1520 pos = append_path(new_path, pos, component->procname);
1521 if (!pos)
1522 goto out;
1523 }
1524 while (table->procname && table->child && !table[1].procname) {
1525 pos = append_path(new_path, pos, table->procname);
1526 if (!pos)
1527 goto out;
1528 table = table->child;
1529 }
1530 if (nr_subheaders == 1) {
1531 header = __register_sysctl_table(set, new_path, table);
1532 if (header)
1533 header->ctl_table_arg = ctl_table_arg;
1534 } else {
1535 header = kzalloc(sizeof(*header) +
1536 sizeof(*subheaders)*nr_subheaders, GFP_KERNEL);
1537 if (!header)
1538 goto out;
1539
1540 subheaders = (struct ctl_table_header **) (header + 1);
1541 subheader = subheaders;
1542 header->ctl_table_arg = ctl_table_arg;
1543
1544 if (register_leaf_sysctl_tables(new_path, pos, &subheader,
1545 set, table))
1546 goto err_register_leaves;
1547 }
1548
1549out:
1550 kfree(new_path);
1551 return header;
1552
1553err_register_leaves:
1554 while (subheader > subheaders) {
1555 struct ctl_table_header *subh = *(--subheader);
1556 struct ctl_table *table = subh->ctl_table_arg;
1557 unregister_sysctl_table(subh);
1558 kfree(table);
1559 }
1560 kfree(header);
1561 header = NULL;
1562 goto out;
1563}
1564
1565/**
1566 * register_sysctl_paths - register a sysctl table hierarchy
1567 * @path: The path to the directory the sysctl table is in.
1568 * @table: the top-level table structure
1569 *
1570 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1571 * array. A completely 0 filled entry terminates the table.
1572 *
1573 * See __register_sysctl_paths for more details.
1574 */
1575struct ctl_table_header *register_sysctl_paths(const struct ctl_path *path,
1576 struct ctl_table *table)
1577{
1578 return __register_sysctl_paths(&sysctl_table_root.default_set,
1579 path, table);
1580}
1581EXPORT_SYMBOL(register_sysctl_paths);
1582
1583/**
1584 * register_sysctl_table - register a sysctl table hierarchy
1585 * @table: the top-level table structure
1586 *
1587 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1588 * array. A completely 0 filled entry terminates the table.
1589 *
1590 * See register_sysctl_paths for more details.
1591 */
1592struct ctl_table_header *register_sysctl_table(struct ctl_table *table)
1593{
1594 static const struct ctl_path null_path[] = { {} };
1595
1596 return register_sysctl_paths(null_path, table);
1597}
1598EXPORT_SYMBOL(register_sysctl_table);
1599
1600static void put_links(struct ctl_table_header *header)
1601{
1602 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1603 struct ctl_table_root *root = header->root;
1604 struct ctl_dir *parent = header->parent;
1605 struct ctl_dir *core_parent;
1606 struct ctl_table *entry;
1607
1608 if (header->set == root_set)
1609 return;
1610
1611 core_parent = xlate_dir(root_set, parent);
1612 if (IS_ERR(core_parent))
1613 return;
1614
1615 for (entry = header->ctl_table; entry->procname; entry++) {
1616 struct ctl_table_header *link_head;
1617 struct ctl_table *link;
1618 const char *name = entry->procname;
1619
1620 link = find_entry(&link_head, core_parent, name, strlen(name));
1621 if (link &&
1622 ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1623 (S_ISLNK(link->mode) && (link->data == root)))) {
1624 drop_sysctl_table(link_head);
1625 }
1626 else {
1627 pr_err("sysctl link missing during unregister: ");
1628 sysctl_print_dir(parent);
1629 pr_cont("/%s\n", name);
1630 }
1631 }
1632}
1633
1634static void drop_sysctl_table(struct ctl_table_header *header)
1635{
1636 struct ctl_dir *parent = header->parent;
1637
1638 if (--header->nreg)
1639 return;
1640
1641 if (parent) {
1642 put_links(header);
1643 start_unregistering(header);
1644 }
1645
1646 if (!--header->count)
1647 kfree_rcu(header, rcu);
1648
1649 if (parent)
1650 drop_sysctl_table(&parent->header);
1651}
1652
1653/**
1654 * unregister_sysctl_table - unregister a sysctl table hierarchy
1655 * @header: the header returned from register_sysctl_table
1656 *
1657 * Unregisters the sysctl table and all children. proc entries may not
1658 * actually be removed until they are no longer used by anyone.
1659 */
1660void unregister_sysctl_table(struct ctl_table_header * header)
1661{
1662 int nr_subheaders;
1663 might_sleep();
1664
1665 if (header == NULL)
1666 return;
1667
1668 nr_subheaders = count_subheaders(header->ctl_table_arg);
1669 if (unlikely(nr_subheaders > 1)) {
1670 struct ctl_table_header **subheaders;
1671 int i;
1672
1673 subheaders = (struct ctl_table_header **)(header + 1);
1674 for (i = nr_subheaders -1; i >= 0; i--) {
1675 struct ctl_table_header *subh = subheaders[i];
1676 struct ctl_table *table = subh->ctl_table_arg;
1677 unregister_sysctl_table(subh);
1678 kfree(table);
1679 }
1680 kfree(header);
1681 return;
1682 }
1683
1684 spin_lock(&sysctl_lock);
1685 drop_sysctl_table(header);
1686 spin_unlock(&sysctl_lock);
1687}
1688EXPORT_SYMBOL(unregister_sysctl_table);
1689
1690void setup_sysctl_set(struct ctl_table_set *set,
1691 struct ctl_table_root *root,
1692 int (*is_seen)(struct ctl_table_set *))
1693{
1694 memset(set, 0, sizeof(*set));
1695 set->is_seen = is_seen;
1696 init_header(&set->dir.header, root, set, NULL, root_table);
1697}
1698
1699void retire_sysctl_set(struct ctl_table_set *set)
1700{
1701 WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1702}
1703
1704int __init proc_sys_init(void)
1705{
1706 struct proc_dir_entry *proc_sys_root;
1707
1708 proc_sys_root = proc_mkdir("sys", NULL);
1709 proc_sys_root->proc_iops = &proc_sys_dir_operations;
1710 proc_sys_root->proc_dir_ops = &proc_sys_dir_file_operations;
1711 proc_sys_root->nlink = 0;
1712
1713 return sysctl_init();
1714}
1715
1716struct sysctl_alias {
1717 const char *kernel_param;
1718 const char *sysctl_param;
1719};
1720
1721/*
1722 * Historically some settings had both sysctl and a command line parameter.
1723 * With the generic sysctl. parameter support, we can handle them at a single
1724 * place and only keep the historical name for compatibility. This is not meant
1725 * to add brand new aliases. When adding existing aliases, consider whether
1726 * the possibly different moment of changing the value (e.g. from early_param
1727 * to the moment do_sysctl_args() is called) is an issue for the specific
1728 * parameter.
1729 */
1730static const struct sysctl_alias sysctl_aliases[] = {
1731 {"hardlockup_all_cpu_backtrace", "kernel.hardlockup_all_cpu_backtrace" },
1732 {"hung_task_panic", "kernel.hung_task_panic" },
1733 {"numa_zonelist_order", "vm.numa_zonelist_order" },
1734 {"softlockup_all_cpu_backtrace", "kernel.softlockup_all_cpu_backtrace" },
1735 {"softlockup_panic", "kernel.softlockup_panic" },
1736 { }
1737};
1738
1739static const char *sysctl_find_alias(char *param)
1740{
1741 const struct sysctl_alias *alias;
1742
1743 for (alias = &sysctl_aliases[0]; alias->kernel_param != NULL; alias++) {
1744 if (strcmp(alias->kernel_param, param) == 0)
1745 return alias->sysctl_param;
1746 }
1747
1748 return NULL;
1749}
1750
1751/* Set sysctl value passed on kernel command line. */
1752static int process_sysctl_arg(char *param, char *val,
1753 const char *unused, void *arg)
1754{
1755 char *path;
1756 struct vfsmount **proc_mnt = arg;
1757 struct file_system_type *proc_fs_type;
1758 struct file *file;
1759 int len;
1760 int err;
1761 loff_t pos = 0;
1762 ssize_t wret;
1763
1764 if (strncmp(param, "sysctl", sizeof("sysctl") - 1) == 0) {
1765 param += sizeof("sysctl") - 1;
1766
1767 if (param[0] != '/' && param[0] != '.')
1768 return 0;
1769
1770 param++;
1771 } else {
1772 param = (char *) sysctl_find_alias(param);
1773 if (!param)
1774 return 0;
1775 }
1776
1777 if (!val)
1778 return -EINVAL;
1779 len = strlen(val);
1780 if (len == 0)
1781 return -EINVAL;
1782
1783 /*
1784 * To set sysctl options, we use a temporary mount of proc, look up the
1785 * respective sys/ file and write to it. To avoid mounting it when no
1786 * options were given, we mount it only when the first sysctl option is
1787 * found. Why not a persistent mount? There are problems with a
1788 * persistent mount of proc in that it forces userspace not to use any
1789 * proc mount options.
1790 */
1791 if (!*proc_mnt) {
1792 proc_fs_type = get_fs_type("proc");
1793 if (!proc_fs_type) {
1794 pr_err("Failed to find procfs to set sysctl from command line\n");
1795 return 0;
1796 }
1797 *proc_mnt = kern_mount(proc_fs_type);
1798 put_filesystem(proc_fs_type);
1799 if (IS_ERR(*proc_mnt)) {
1800 pr_err("Failed to mount procfs to set sysctl from command line\n");
1801 return 0;
1802 }
1803 }
1804
1805 path = kasprintf(GFP_KERNEL, "sys/%s", param);
1806 if (!path)
1807 panic("%s: Failed to allocate path for %s\n", __func__, param);
1808 strreplace(path, '.', '/');
1809
1810 file = file_open_root_mnt(*proc_mnt, path, O_WRONLY, 0);
1811 if (IS_ERR(file)) {
1812 err = PTR_ERR(file);
1813 if (err == -ENOENT)
1814 pr_err("Failed to set sysctl parameter '%s=%s': parameter not found\n",
1815 param, val);
1816 else if (err == -EACCES)
1817 pr_err("Failed to set sysctl parameter '%s=%s': permission denied (read-only?)\n",
1818 param, val);
1819 else
1820 pr_err("Error %pe opening proc file to set sysctl parameter '%s=%s'\n",
1821 file, param, val);
1822 goto out;
1823 }
1824 wret = kernel_write(file, val, len, &pos);
1825 if (wret < 0) {
1826 err = wret;
1827 if (err == -EINVAL)
1828 pr_err("Failed to set sysctl parameter '%s=%s': invalid value\n",
1829 param, val);
1830 else
1831 pr_err("Error %pe writing to proc file to set sysctl parameter '%s=%s'\n",
1832 ERR_PTR(err), param, val);
1833 } else if (wret != len) {
1834 pr_err("Wrote only %zd bytes of %d writing to proc file %s to set sysctl parameter '%s=%s\n",
1835 wret, len, path, param, val);
1836 }
1837
1838 err = filp_close(file, NULL);
1839 if (err)
1840 pr_err("Error %pe closing proc file to set sysctl parameter '%s=%s\n",
1841 ERR_PTR(err), param, val);
1842out:
1843 kfree(path);
1844 return 0;
1845}
1846
1847void do_sysctl_args(void)
1848{
1849 char *command_line;
1850 struct vfsmount *proc_mnt = NULL;
1851
1852 command_line = kstrdup(saved_command_line, GFP_KERNEL);
1853 if (!command_line)
1854 panic("%s: Failed to allocate copy of command line\n", __func__);
1855
1856 parse_args("Setting sysctl args", command_line,
1857 NULL, 0, -1, -1, &proc_mnt, process_sysctl_arg);
1858
1859 if (proc_mnt)
1860 kern_unmount(proc_mnt);
1861
1862 kfree(command_line);
1863}