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