projects/linux·blob master · refs

1725 lines · 44.2 KiB

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 inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
544 err = d_splice_alias_ops(inode, dentry, &proc_sys_dentry_operations);
545
546out:
547 if (h)
548 sysctl_head_finish(h);
549 sysctl_head_finish(head);
550 return err;
551}
552
553static ssize_t proc_sys_call_handler(struct kiocb *iocb, struct iov_iter *iter,
554 int write)
555{
556 struct inode *inode = file_inode(iocb->ki_filp);
557 struct ctl_table_header *head = grab_header(inode);
558 const struct ctl_table *table = PROC_I(inode)->sysctl_entry;
559 size_t count = iov_iter_count(iter);
560 char *kbuf;
561 ssize_t error;
562
563 if (IS_ERR(head))
564 return PTR_ERR(head);
565
566 /*
567 * At this point we know that the sysctl was not unregistered
568 * and won't be until we finish.
569 */
570 error = -EPERM;
571 if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
572 goto out;
573
574 /* if that can happen at all, it should be -EINVAL, not -EISDIR */
575 error = -EINVAL;
576 if (!table->proc_handler)
577 goto out;
578
579 /* don't even try if the size is too large */
580 error = -ENOMEM;
581 if (count >= KMALLOC_MAX_SIZE)
582 goto out;
583 kbuf = kvzalloc(count + 1, GFP_KERNEL);
584 if (!kbuf)
585 goto out;
586
587 if (write) {
588 error = -EFAULT;
589 if (!copy_from_iter_full(kbuf, count, iter))
590 goto out_free_buf;
591 kbuf[count] = '\0';
592 }
593
594 error = BPF_CGROUP_RUN_PROG_SYSCTL(head, table, write, &kbuf, &count,
595 &iocb->ki_pos);
596 if (error)
597 goto out_free_buf;
598
599 /* careful: calling conventions are nasty here */
600 error = table->proc_handler(table, write, kbuf, &count, &iocb->ki_pos);
601 if (error)
602 goto out_free_buf;
603
604 if (!write) {
605 error = -EFAULT;
606 if (copy_to_iter(kbuf, count, iter) < count)
607 goto out_free_buf;
608 }
609
610 error = count;
611out_free_buf:
612 kvfree(kbuf);
613out:
614 sysctl_head_finish(head);
615
616 return error;
617}
618
619static ssize_t proc_sys_read(struct kiocb *iocb, struct iov_iter *iter)
620{
621 return proc_sys_call_handler(iocb, iter, 0);
622}
623
624static ssize_t proc_sys_write(struct kiocb *iocb, struct iov_iter *iter)
625{
626 return proc_sys_call_handler(iocb, iter, 1);
627}
628
629static int proc_sys_open(struct inode *inode, struct file *filp)
630{
631 struct ctl_table_header *head = grab_header(inode);
632 const struct ctl_table *table = PROC_I(inode)->sysctl_entry;
633
634 /* sysctl was unregistered */
635 if (IS_ERR(head))
636 return PTR_ERR(head);
637
638 if (table->poll)
639 filp->private_data = proc_sys_poll_event(table->poll);
640
641 sysctl_head_finish(head);
642
643 return 0;
644}
645
646static __poll_t proc_sys_poll(struct file *filp, poll_table *wait)
647{
648 struct inode *inode = file_inode(filp);
649 struct ctl_table_header *head = grab_header(inode);
650 const struct ctl_table *table = PROC_I(inode)->sysctl_entry;
651 __poll_t ret = DEFAULT_POLLMASK;
652 unsigned long event;
653
654 /* sysctl was unregistered */
655 if (IS_ERR(head))
656 return EPOLLERR | EPOLLHUP;
657
658 if (!table->proc_handler)
659 goto out;
660
661 if (!table->poll)
662 goto out;
663
664 event = (unsigned long)filp->private_data;
665 poll_wait(filp, &table->poll->wait, wait);
666
667 if (event != atomic_read(&table->poll->event)) {
668 filp->private_data = proc_sys_poll_event(table->poll);
669 ret = EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
670 }
671
672out:
673 sysctl_head_finish(head);
674
675 return ret;
676}
677
678static bool proc_sys_fill_cache(struct file *file,
679 struct dir_context *ctx,
680 struct ctl_table_header *head,
681 const struct ctl_table *table)
682{
683 struct dentry *child, *dir = file->f_path.dentry;
684 struct inode *inode;
685 struct qstr qname;
686 ino_t ino = 0;
687 unsigned type = DT_UNKNOWN;
688
689 qname.name = table->procname;
690 qname.len = strlen(table->procname);
691 qname.hash = full_name_hash(dir, qname.name, qname.len);
692
693 child = d_lookup(dir, &qname);
694 if (!child) {
695 child = d_alloc_parallel(dir, &qname);
696 if (IS_ERR(child))
697 return false;
698 if (d_in_lookup(child)) {
699 struct dentry *res;
700 inode = proc_sys_make_inode(dir->d_sb, head, table);
701 res = d_splice_alias_ops(inode, child,
702 &proc_sys_dentry_operations);
703 d_lookup_done(child);
704 if (unlikely(res)) {
705 dput(child);
706
707 if (IS_ERR(res))
708 return false;
709
710 child = res;
711 }
712 }
713 }
714 inode = d_inode(child);
715 ino = inode->i_ino;
716 type = inode->i_mode >> 12;
717 dput(child);
718 return dir_emit(ctx, qname.name, qname.len, ino, type);
719}
720
721static bool proc_sys_link_fill_cache(struct file *file,
722 struct dir_context *ctx,
723 struct ctl_table_header *head,
724 const struct ctl_table *table)
725{
726 bool ret = true;
727
728 head = sysctl_head_grab(head);
729 if (IS_ERR(head))
730 return false;
731
732 /* It is not an error if we can not follow the link ignore it */
733 if (sysctl_follow_link(&head, &table))
734 goto out;
735
736 ret = proc_sys_fill_cache(file, ctx, head, table);
737out:
738 sysctl_head_finish(head);
739 return ret;
740}
741
742static int scan(struct ctl_table_header *head, const struct ctl_table *table,
743 unsigned long *pos, struct file *file,
744 struct dir_context *ctx)
745{
746 bool res;
747
748 if ((*pos)++ < ctx->pos)
749 return true;
750
751 if (unlikely(S_ISLNK(table->mode)))
752 res = proc_sys_link_fill_cache(file, ctx, head, table);
753 else
754 res = proc_sys_fill_cache(file, ctx, head, table);
755
756 if (res)
757 ctx->pos = *pos;
758
759 return res;
760}
761
762static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
763{
764 struct ctl_table_header *head = grab_header(file_inode(file));
765 struct ctl_table_header *h = NULL;
766 const struct ctl_table *entry;
767 struct ctl_dir *ctl_dir;
768 unsigned long pos;
769
770 if (IS_ERR(head))
771 return PTR_ERR(head);
772
773 ctl_dir = container_of(head, struct ctl_dir, header);
774
775 if (!dir_emit_dots(file, ctx))
776 goto out;
777
778 pos = 2;
779
780 for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
781 if (!scan(h, entry, &pos, file, ctx)) {
782 sysctl_head_finish(h);
783 break;
784 }
785 }
786out:
787 sysctl_head_finish(head);
788 return 0;
789}
790
791static int proc_sys_permission(const struct mnt_idmap *idmap,
792 struct inode *inode, int mask)
793{
794 /*
795 * sysctl entries that are not writeable,
796 * are _NOT_ writeable, capabilities or not.
797 */
798 struct ctl_table_header *head;
799 const struct ctl_table *table;
800 int error;
801
802 /* Executable files are not allowed under /proc/sys/ */
803 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
804 return -EACCES;
805
806 head = grab_header(inode);
807 if (IS_ERR(head))
808 return PTR_ERR(head);
809
810 table = PROC_I(inode)->sysctl_entry;
811 if (!table) /* global root - r-xr-xr-x */
812 error = mask & MAY_WRITE ? -EACCES : 0;
813 else /* Use the permissions on the sysctl table entry */
814 error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
815
816 sysctl_head_finish(head);
817 return error;
818}
819
820static int proc_sys_setattr(const struct mnt_idmap *idmap,
821 struct dentry *dentry, struct iattr *attr)
822{
823 struct inode *inode = d_inode(dentry);
824 int error;
825
826 if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
827 return -EPERM;
828
829 error = setattr_prepare(&nop_mnt_idmap, dentry, attr);
830 if (error)
831 return error;
832
833 setattr_copy(&nop_mnt_idmap, inode, attr);
834 return 0;
835}
836
837static int proc_sys_getattr(const struct mnt_idmap *idmap,
838 const struct path *path, struct kstat *stat,
839 u32 request_mask, unsigned int query_flags)
840{
841 struct inode *inode = d_inode(path->dentry);
842 struct ctl_table_header *head = grab_header(inode);
843 const struct ctl_table *table = PROC_I(inode)->sysctl_entry;
844
845 if (IS_ERR(head))
846 return PTR_ERR(head);
847
848 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
849 if (table)
850 stat->mode = (stat->mode & S_IFMT) | table->mode;
851
852 sysctl_head_finish(head);
853 return 0;
854}
855
856static const struct file_operations proc_sys_file_operations = {
857 .open = proc_sys_open,
858 .poll = proc_sys_poll,
859 .read_iter = proc_sys_read,
860 .write_iter = proc_sys_write,
861 .splice_read = copy_splice_read,
862 .splice_write = iter_file_splice_write,
863 .llseek = default_llseek,
864};
865
866static const struct file_operations proc_sys_dir_file_operations = {
867 .read = generic_read_dir,
868 .iterate_shared = proc_sys_readdir,
869 .llseek = generic_file_llseek,
870};
871
872static const struct inode_operations proc_sys_inode_operations = {
873 .permission = proc_sys_permission,
874 .setattr = proc_sys_setattr,
875 .getattr = proc_sys_getattr,
876};
877
878static const struct inode_operations proc_sys_dir_operations = {
879 .lookup = proc_sys_lookup,
880 .permission = proc_sys_permission,
881 .setattr = proc_sys_setattr,
882 .getattr = proc_sys_getattr,
883};
884
885static int proc_sys_revalidate(struct inode *dir, const struct qstr *name,
886 struct dentry *dentry, unsigned int flags)
887{
888 if (flags & LOOKUP_RCU)
889 return -ECHILD;
890 return !PROC_I(d_inode(dentry))->sysctl->unregistering;
891}
892
893static int proc_sys_delete(const struct dentry *dentry)
894{
895 return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
896}
897
898static int sysctl_is_seen(struct ctl_table_header *p)
899{
900 struct ctl_table_set *set = p->set;
901 int res;
902 spin_lock(&sysctl_lock);
903 if (p->unregistering)
904 res = 0;
905 else if (!set->is_seen)
906 res = 1;
907 else
908 res = set->is_seen(set);
909 spin_unlock(&sysctl_lock);
910 return res;
911}
912
913static int proc_sys_compare(const struct dentry *dentry,
914 unsigned int len, const char *str, const struct qstr *name)
915{
916 struct ctl_table_header *head;
917 struct inode *inode;
918
919 if (name->len != len)
920 return 1;
921 if (memcmp(name->name, str, len))
922 return 1;
923
924 // false positive is fine here - we'll recheck anyway
925 if (d_in_lookup(dentry))
926 return 0;
927
928 inode = d_inode_rcu(dentry);
929 // we just might have run into dentry in the middle of __dentry_kill()
930 if (!inode)
931 return 1;
932
933 head = READ_ONCE(PROC_I(inode)->sysctl);
934 return !head || !sysctl_is_seen(head);
935}
936
937static const struct dentry_operations proc_sys_dentry_operations = {
938 .d_revalidate = proc_sys_revalidate,
939 .d_delete = proc_sys_delete,
940 .d_compare = proc_sys_compare,
941};
942
943static struct ctl_dir *find_subdir(struct ctl_dir *dir,
944 const char *name, int namelen)
945{
946 struct ctl_table_header *head;
947 const struct ctl_table *entry;
948
949 entry = find_entry(&head, dir, name, namelen);
950 if (!entry)
951 return ERR_PTR(-ENOENT);
952 if (!S_ISDIR(entry->mode))
953 return ERR_PTR(-ENOTDIR);
954 return container_of(head, struct ctl_dir, header);
955}
956
957static struct ctl_dir *new_dir(struct ctl_table_set *set,
958 const char *name, int namelen)
959{
960 struct ctl_table *table;
961 struct ctl_dir *new;
962 struct ctl_node *node;
963 char *new_name;
964
965 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
966 sizeof(struct ctl_table) + namelen + 1,
967 GFP_KERNEL);
968 if (!new)
969 return NULL;
970
971 node = (struct ctl_node *)(new + 1);
972 table = (struct ctl_table *)(node + 1);
973 new_name = (char *)(table + 1);
974 memcpy(new_name, name, namelen);
975 table[0].procname = new_name;
976 table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
977 init_header(&new->header, set->dir.header.root, set, node, table, 1);
978
979 return new;
980}
981
982/**
983 * get_subdir - find or create a subdir with the specified name.
984 * @dir: Directory to create the subdirectory in
985 * @name: The name of the subdirectory to find or create
986 * @namelen: The length of name
987 *
988 * Takes a directory with an elevated reference count so we know that
989 * if we drop the lock the directory will not go away. Upon success
990 * the reference is moved from @dir to the returned subdirectory.
991 * Upon error an error code is returned and the reference on @dir is
992 * simply dropped.
993 */
994static struct ctl_dir *get_subdir(struct ctl_dir *dir,
995 const char *name, int namelen)
996{
997 struct ctl_table_set *set = dir->header.set;
998 struct ctl_dir *subdir, *new = NULL;
999 int err;
1000
1001 spin_lock(&sysctl_lock);
1002 subdir = find_subdir(dir, name, namelen);
1003 if (!IS_ERR(subdir))
1004 goto found;
1005 if (PTR_ERR(subdir) != -ENOENT)
1006 goto failed;
1007
1008 spin_unlock(&sysctl_lock);
1009 new = new_dir(set, name, namelen);
1010 spin_lock(&sysctl_lock);
1011 subdir = ERR_PTR(-ENOMEM);
1012 if (!new)
1013 goto failed;
1014
1015 /* Was the subdir added while we dropped the lock? */
1016 subdir = find_subdir(dir, name, namelen);
1017 if (!IS_ERR(subdir))
1018 goto found;
1019 if (PTR_ERR(subdir) != -ENOENT)
1020 goto failed;
1021
1022 /* Nope. Use the our freshly made directory entry. */
1023 err = insert_header(dir, &new->header);
1024 subdir = ERR_PTR(err);
1025 if (err)
1026 goto failed;
1027 subdir = new;
1028found:
1029 subdir->header.nreg++;
1030failed:
1031 if (IS_ERR(subdir)) {
1032 pr_err("sysctl could not get directory: ");
1033 sysctl_print_dir(dir);
1034 pr_cont("%*.*s %ld\n", namelen, namelen, name,
1035 PTR_ERR(subdir));
1036 }
1037 drop_sysctl_table(&dir->header);
1038 if (new)
1039 drop_sysctl_table(&new->header);
1040 spin_unlock(&sysctl_lock);
1041 return subdir;
1042}
1043
1044static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
1045{
1046 struct ctl_dir *parent;
1047 const char *procname;
1048 if (!dir->header.parent)
1049 return &set->dir;
1050 parent = xlate_dir(set, dir->header.parent);
1051 if (IS_ERR(parent))
1052 return parent;
1053 procname = dir->header.ctl_table[0].procname;
1054 return find_subdir(parent, procname, strlen(procname));
1055}
1056
1057static int sysctl_follow_link(struct ctl_table_header **phead,
1058 const struct ctl_table **pentry)
1059{
1060 struct ctl_table_header *head;
1061 const struct ctl_table *entry;
1062 struct ctl_table_root *root;
1063 struct ctl_table_set *set;
1064 struct ctl_dir *dir;
1065 int ret;
1066
1067 spin_lock(&sysctl_lock);
1068 root = (*pentry)->data;
1069 set = lookup_header_set(root);
1070 dir = xlate_dir(set, (*phead)->parent);
1071 if (IS_ERR(dir))
1072 ret = PTR_ERR(dir);
1073 else {
1074 const char *procname = (*pentry)->procname;
1075 head = NULL;
1076 entry = find_entry(&head, dir, procname, strlen(procname));
1077 ret = -ENOENT;
1078 if (entry && use_table(head)) {
1079 unuse_table(*phead);
1080 *phead = head;
1081 *pentry = entry;
1082 ret = 0;
1083 }
1084 }
1085
1086 spin_unlock(&sysctl_lock);
1087 return ret;
1088}
1089
1090static int sysctl_err(const char *path, const struct ctl_table *table, char *fmt, ...)
1091{
1092 struct va_format vaf;
1093 va_list args;
1094
1095 va_start(args, fmt);
1096 vaf.fmt = fmt;
1097 vaf.va = &args;
1098
1099 pr_err("sysctl table check failed: %s/%s %pV\n",
1100 path, table->procname, &vaf);
1101
1102 va_end(args);
1103 return -EINVAL;
1104}
1105
1106static int sysctl_check_table_array(const char *path, const struct ctl_table *table)
1107{
1108 unsigned int extra;
1109 int err = 0;
1110
1111 if ((table->proc_handler == proc_douintvec) ||
1112 (table->proc_handler == proc_douintvec_minmax)) {
1113 if (table->maxlen != sizeof(unsigned int))
1114 err |= sysctl_err(path, table, "array not allowed");
1115 }
1116
1117 if (table->proc_handler == proc_dou8vec_minmax) {
1118 if (table->maxlen != sizeof(u8))
1119 err |= sysctl_err(path, table, "array not allowed");
1120
1121 if (table->extra1) {
1122 extra = *(unsigned int *) table->extra1;
1123 if (extra > 255U)
1124 err |= sysctl_err(path, table,
1125 "range value too large for proc_dou8vec_minmax");
1126 }
1127 if (table->extra2) {
1128 extra = *(unsigned int *) table->extra2;
1129 if (extra > 255U)
1130 err |= sysctl_err(path, table,
1131 "range value too large for proc_dou8vec_minmax");
1132 }
1133 }
1134
1135 if (table->proc_handler == proc_dobool) {
1136 if (table->maxlen != sizeof(bool))
1137 err |= sysctl_err(path, table, "array not allowed");
1138 }
1139
1140 return err;
1141}
1142
1143static int sysctl_check_table(const char *path, struct ctl_table_header *header)
1144{
1145 const struct ctl_table *entry;
1146 int err = 0;
1147 list_for_each_table_entry(entry, header) {
1148 if (!entry->procname)
1149 err |= sysctl_err(path, entry, "procname is null");
1150 if ((entry->proc_handler == proc_dostring) ||
1151 (entry->proc_handler == proc_dobool) ||
1152 (entry->proc_handler == proc_dointvec) ||
1153 (entry->proc_handler == proc_douintvec) ||
1154 (entry->proc_handler == proc_douintvec_minmax) ||
1155 (entry->proc_handler == proc_dointvec_minmax) ||
1156 (entry->proc_handler == proc_dou8vec_minmax) ||
1157 (entry->proc_handler == proc_dointvec_jiffies) ||
1158 (entry->proc_handler == proc_dointvec_userhz_jiffies) ||
1159 (entry->proc_handler == proc_dointvec_ms_jiffies) ||
1160 (entry->proc_handler == proc_doulongvec_minmax) ||
1161 (entry->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1162 if (!entry->data)
1163 err |= sysctl_err(path, entry, "No data");
1164 if (!entry->maxlen)
1165 err |= sysctl_err(path, entry, "No maxlen");
1166 else
1167 err |= sysctl_check_table_array(path, entry);
1168 }
1169 if (!entry->proc_handler)
1170 err |= sysctl_err(path, entry, "No proc_handler");
1171
1172 if ((entry->mode & (S_IRUGO|S_IWUGO)) != entry->mode)
1173 err |= sysctl_err(path, entry, "bogus .mode 0%o",
1174 entry->mode);
1175 }
1176 return err;
1177}
1178
1179static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table_header *head)
1180{
1181 struct ctl_table *link_table, *link;
1182 struct ctl_table_header *links;
1183 const struct ctl_table *entry;
1184 struct ctl_node *node;
1185 char *link_name;
1186 int name_bytes;
1187
1188 name_bytes = 0;
1189 list_for_each_table_entry(entry, head) {
1190 name_bytes += strlen(entry->procname) + 1;
1191 }
1192
1193 links = kzalloc(sizeof(struct ctl_table_header) +
1194 sizeof(struct ctl_node)*head->ctl_table_size +
1195 sizeof(struct ctl_table)*head->ctl_table_size +
1196 name_bytes,
1197 GFP_KERNEL);
1198
1199 if (!links)
1200 return NULL;
1201
1202 node = (struct ctl_node *)(links + 1);
1203 link_table = (struct ctl_table *)(node + head->ctl_table_size);
1204 link_name = (char *)(link_table + head->ctl_table_size);
1205 link = link_table;
1206
1207 list_for_each_table_entry(entry, head) {
1208 int len = strlen(entry->procname) + 1;
1209 memcpy(link_name, entry->procname, len);
1210 link->procname = link_name;
1211 link->mode = S_IFLNK|S_IRWXUGO;
1212 link->data = head->root;
1213 link_name += len;
1214 link++;
1215 }
1216 init_header(links, dir->header.root, dir->header.set, node, link_table,
1217 head->ctl_table_size);
1218 links->nreg = head->ctl_table_size;
1219
1220 return links;
1221}
1222
1223static bool get_links(struct ctl_dir *dir,
1224 struct ctl_table_header *header,
1225 struct ctl_table_root *link_root)
1226{
1227 struct ctl_table_header *tmp_head;
1228 const struct ctl_table *entry, *link;
1229
1230 if (header->ctl_table_size == 0 ||
1231 sysctl_is_perm_empty_ctl_header(header))
1232 return true;
1233
1234 /* Are there links available for every entry in table? */
1235 list_for_each_table_entry(entry, header) {
1236 const char *procname = entry->procname;
1237 link = find_entry(&tmp_head, dir, procname, strlen(procname));
1238 if (!link)
1239 return false;
1240 if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1241 continue;
1242 if (S_ISLNK(link->mode) && (link->data == link_root))
1243 continue;
1244 return false;
1245 }
1246
1247 /* The checks passed. Increase the registration count on the links */
1248 list_for_each_table_entry(entry, header) {
1249 const char *procname = entry->procname;
1250 link = find_entry(&tmp_head, dir, procname, strlen(procname));
1251 tmp_head->nreg++;
1252 }
1253 return true;
1254}
1255
1256static int insert_links(struct ctl_table_header *head)
1257{
1258 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1259 struct ctl_dir *core_parent;
1260 struct ctl_table_header *links;
1261 int err;
1262
1263 if (head->set == root_set)
1264 return 0;
1265
1266 core_parent = xlate_dir(root_set, head->parent);
1267 if (IS_ERR(core_parent))
1268 return 0;
1269
1270 if (get_links(core_parent, head, head->root))
1271 return 0;
1272
1273 core_parent->header.nreg++;
1274 spin_unlock(&sysctl_lock);
1275
1276 links = new_links(core_parent, head);
1277
1278 spin_lock(&sysctl_lock);
1279 err = -ENOMEM;
1280 if (!links)
1281 goto out;
1282
1283 err = 0;
1284 if (get_links(core_parent, head, head->root)) {
1285 kfree(links);
1286 goto out;
1287 }
1288
1289 err = insert_header(core_parent, links);
1290 if (err)
1291 kfree(links);
1292out:
1293 drop_sysctl_table(&core_parent->header);
1294 return err;
1295}
1296
1297/* Find the directory for the ctl_table. If one is not found create it. */
1298static struct ctl_dir *sysctl_mkdir_p(struct ctl_dir *dir, const char *path)
1299{
1300 const char *name, *nextname;
1301
1302 for (name = path; name; name = nextname) {
1303 int namelen;
1304 nextname = strchr(name, '/');
1305 if (nextname) {
1306 namelen = nextname - name;
1307 nextname++;
1308 } else {
1309 namelen = strlen(name);
1310 }
1311 if (namelen == 0)
1312 continue;
1313
1314 /*
1315 * namelen ensures if name is "foo/bar/yay" only foo is
1316 * registered first. We traverse as if using mkdir -p and
1317 * return a ctl_dir for the last directory entry.
1318 */
1319 dir = get_subdir(dir, name, namelen);
1320 if (IS_ERR(dir))
1321 break;
1322 }
1323 return dir;
1324}
1325
1326/**
1327 * __register_sysctl_table - register a leaf sysctl table
1328 * @set: Sysctl tree to register on
1329 * @path: The path to the directory the sysctl table is in.
1330 *
1331 * @table: the top-level table structure. This table should not be free'd
1332 * after registration. So it should not be used on stack. It can either
1333 * be a global or dynamically allocated by the caller and free'd later
1334 * after sysctl unregistration.
1335 * @table_size : The number of elements in table
1336 *
1337 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1338 * array.
1339 *
1340 * The members of the &struct ctl_table structure are used as follows:
1341 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1342 * enter a sysctl file
1343 * data - a pointer to data for use by proc_handler
1344 * maxlen - the maximum size in bytes of the data
1345 * mode - the file permissions for the /proc/sys file
1346 * type - Defines the target type (described in struct definition)
1347 * proc_handler - the text handler routine (described below)
1348 *
1349 * extra1, extra2 - extra pointers usable by the proc handler routines
1350 * XXX: we should eventually modify these to use long min / max [0]
1351 * [0] https://lkml.kernel.org/[email protected]
1352 *
1353 * Leaf nodes in the sysctl tree will be represented by a single file
1354 * under /proc; non-leaf nodes are not allowed.
1355 *
1356 * There must be a proc_handler routine for any terminal nodes.
1357 * Several default handlers are available to cover common cases -
1358 *
1359 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1360 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1361 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1362 *
1363 * It is the handler's job to read the input buffer from user memory
1364 * and process it. The handler should return 0 on success.
1365 *
1366 * This routine returns %NULL on a failure to register, and a pointer
1367 * to the table header on success.
1368 */
1369struct ctl_table_header *__register_sysctl_table(
1370 struct ctl_table_set *set,
1371 const char *path, const struct ctl_table *table, size_t table_size)
1372{
1373 struct ctl_table_root *root = set->dir.header.root;
1374 struct ctl_table_header *header;
1375 struct ctl_dir *dir;
1376 struct ctl_node *node;
1377
1378 header = kzalloc(sizeof(struct ctl_table_header) +
1379 sizeof(struct ctl_node)*table_size, GFP_KERNEL_ACCOUNT);
1380 if (!header)
1381 return NULL;
1382
1383 node = (struct ctl_node *)(header + 1);
1384 init_header(header, root, set, node, table, table_size);
1385 if (sysctl_check_table(path, header))
1386 goto fail;
1387
1388 spin_lock(&sysctl_lock);
1389 dir = &set->dir;
1390 /* Reference moved down the directory tree get_subdir */
1391 dir->header.nreg++;
1392 spin_unlock(&sysctl_lock);
1393
1394 dir = sysctl_mkdir_p(dir, path);
1395 if (IS_ERR(dir))
1396 goto fail;
1397 spin_lock(&sysctl_lock);
1398 if (insert_header(dir, header))
1399 goto fail_put_dir_locked;
1400
1401 drop_sysctl_table(&dir->header);
1402 spin_unlock(&sysctl_lock);
1403
1404 return header;
1405
1406fail_put_dir_locked:
1407 drop_sysctl_table(&dir->header);
1408 spin_unlock(&sysctl_lock);
1409fail:
1410 kfree(header);
1411 return NULL;
1412}
1413
1414/**
1415 * register_sysctl_sz - register a sysctl table
1416 * @path: The path to the directory the sysctl table is in. If the path
1417 * doesn't exist we will create it for you.
1418 * @table: the table structure. The calller must ensure the life of the @table
1419 * will be kept during the lifetime use of the syctl. It must not be freed
1420 * until unregister_sysctl_table() is called with the given returned table
1421 * with this registration. If your code is non modular then you don't need
1422 * to call unregister_sysctl_table() and can instead use something like
1423 * register_sysctl_init() which does not care for the result of the syctl
1424 * registration.
1425 * @table_size: The number of elements in table.
1426 *
1427 * Register a sysctl table. @table should be a filled in ctl_table
1428 * array. A completely 0 filled entry terminates the table.
1429 *
1430 * See __register_sysctl_table for more details.
1431 */
1432struct ctl_table_header *register_sysctl_sz(const char *path, const struct ctl_table *table,
1433 size_t table_size)
1434{
1435 return __register_sysctl_table(&sysctl_table_root.default_set,
1436 path, table, table_size);
1437}
1438EXPORT_SYMBOL(register_sysctl_sz);
1439
1440/**
1441 * __register_sysctl_init() - register sysctl table to path
1442 * @path: path name for sysctl base. If that path doesn't exist we will create
1443 * it for you.
1444 * @table: This is the sysctl table that needs to be registered to the path.
1445 * The caller must ensure the life of the @table will be kept during the
1446 * lifetime use of the sysctl.
1447 * @table_name: The name of sysctl table, only used for log printing when
1448 * registration fails
1449 * @table_size: The number of elements in table
1450 *
1451 * The sysctl interface is used by userspace to query or modify at runtime
1452 * a predefined value set on a variable. These variables however have default
1453 * values pre-set. Code which depends on these variables will always work even
1454 * if register_sysctl() fails. If register_sysctl() fails you'd just loose the
1455 * ability to query or modify the sysctls dynamically at run time. Chances of
1456 * register_sysctl() failing on init are extremely low, and so for both reasons
1457 * this function does not return any error as it is used by initialization code.
1458 *
1459 * Context: if your base directory does not exist it will be created for you.
1460 */
1461void __init __register_sysctl_init(const char *path, const struct ctl_table *table,
1462 const char *table_name, size_t table_size)
1463{
1464 struct ctl_table_header *hdr = register_sysctl_sz(path, table, table_size);
1465
1466 if (unlikely(!hdr)) {
1467 pr_err("failed when register_sysctl_sz %s to %s\n", table_name, path);
1468 return;
1469 }
1470 kmemleak_not_leak(hdr);
1471}
1472
1473static void put_links(struct ctl_table_header *header)
1474{
1475 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1476 struct ctl_table_root *root = header->root;
1477 struct ctl_dir *parent = header->parent;
1478 struct ctl_dir *core_parent;
1479 const struct ctl_table *entry;
1480
1481 if (header->set == root_set)
1482 return;
1483
1484 core_parent = xlate_dir(root_set, parent);
1485 if (IS_ERR(core_parent))
1486 return;
1487
1488 list_for_each_table_entry(entry, header) {
1489 struct ctl_table_header *link_head;
1490 const struct ctl_table *link;
1491 const char *name = entry->procname;
1492
1493 link = find_entry(&link_head, core_parent, name, strlen(name));
1494 if (link &&
1495 ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1496 (S_ISLNK(link->mode) && (link->data == root)))) {
1497 drop_sysctl_table(link_head);
1498 }
1499 else {
1500 pr_err("sysctl link missing during unregister: ");
1501 sysctl_print_dir(parent);
1502 pr_cont("%s\n", name);
1503 }
1504 }
1505}
1506
1507static void drop_sysctl_table(struct ctl_table_header *header)
1508{
1509 struct ctl_dir *parent = header->parent;
1510
1511 if (--header->nreg)
1512 return;
1513
1514 if (parent) {
1515 put_links(header);
1516 start_unregistering(header);
1517 }
1518
1519 if (!--header->count)
1520 kfree_rcu(header, rcu);
1521
1522 if (parent)
1523 drop_sysctl_table(&parent->header);
1524}
1525
1526/**
1527 * unregister_sysctl_table - unregister a sysctl table hierarchy
1528 * @header: the header returned from register_sysctl or __register_sysctl_table
1529 *
1530 * Unregisters the sysctl table and all children. proc entries may not
1531 * actually be removed until they are no longer used by anyone.
1532 */
1533void unregister_sysctl_table(struct ctl_table_header * header)
1534{
1535 might_sleep();
1536
1537 if (header == NULL)
1538 return;
1539
1540 spin_lock(&sysctl_lock);
1541 drop_sysctl_table(header);
1542 spin_unlock(&sysctl_lock);
1543}
1544EXPORT_SYMBOL(unregister_sysctl_table);
1545
1546void setup_sysctl_set(struct ctl_table_set *set,
1547 struct ctl_table_root *root,
1548 int (*is_seen)(struct ctl_table_set *))
1549{
1550 memset(set, 0, sizeof(*set));
1551 set->is_seen = is_seen;
1552 init_header(&set->dir.header, root, set, NULL, root_table, 1);
1553}
1554
1555void retire_sysctl_set(struct ctl_table_set *set)
1556{
1557 WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1558}
1559
1560int __init proc_sys_init(void)
1561{
1562 struct proc_dir_entry *proc_sys_root;
1563
1564 proc_sys_root = proc_mkdir("sys", NULL);
1565 proc_sys_root->proc_iops = &proc_sys_dir_operations;
1566 proc_sys_root->proc_dir_ops = &proc_sys_dir_file_operations;
1567 proc_sys_root->nlink = 0;
1568
1569 return sysctl_init_bases();
1570}
1571
1572struct sysctl_alias {
1573 const char *kernel_param;
1574 const char *sysctl_param;
1575};
1576
1577/*
1578 * Historically some settings had both sysctl and a command line parameter.
1579 * With the generic sysctl. parameter support, we can handle them at a single
1580 * place and only keep the historical name for compatibility. This is not meant
1581 * to add brand new aliases. When adding existing aliases, consider whether
1582 * the possibly different moment of changing the value (e.g. from early_param
1583 * to the moment do_sysctl_args() is called) is an issue for the specific
1584 * parameter.
1585 */
1586static const struct sysctl_alias sysctl_aliases[] = {
1587 {"hardlockup_all_cpu_backtrace", "kernel.hardlockup_all_cpu_backtrace" },
1588 {"hung_task_panic", "kernel.hung_task_panic" },
1589 {"numa_zonelist_order", "vm.numa_zonelist_order" },
1590 {"softlockup_all_cpu_backtrace", "kernel.softlockup_all_cpu_backtrace" },
1591 { }
1592};
1593
1594static const char *sysctl_find_alias(char *param)
1595{
1596 const struct sysctl_alias *alias;
1597
1598 for (alias = &sysctl_aliases[0]; alias->kernel_param != NULL; alias++) {
1599 if (strcmp(alias->kernel_param, param) == 0)
1600 return alias->sysctl_param;
1601 }
1602
1603 return NULL;
1604}
1605
1606bool sysctl_is_alias(char *param)
1607{
1608 const char *alias = sysctl_find_alias(param);
1609
1610 return alias != NULL;
1611}
1612
1613/* Set sysctl value passed on kernel command line. */
1614static int process_sysctl_arg(char *param, char *val,
1615 const char *unused, void *arg)
1616{
1617 char *path;
1618 struct vfsmount **proc_mnt = arg;
1619 struct file_system_type *proc_fs_type;
1620 struct file *file;
1621 int len;
1622 int err;
1623 loff_t pos = 0;
1624 ssize_t wret;
1625
1626 if (strncmp(param, "sysctl", sizeof("sysctl") - 1) == 0) {
1627 param += sizeof("sysctl") - 1;
1628
1629 if (param[0] != '/' && param[0] != '.')
1630 return 0;
1631
1632 param++;
1633 } else {
1634 param = (char *) sysctl_find_alias(param);
1635 if (!param)
1636 return 0;
1637 }
1638
1639 if (!val)
1640 return -EINVAL;
1641 len = strlen(val);
1642 if (len == 0)
1643 return -EINVAL;
1644
1645 /*
1646 * To set sysctl options, we use a temporary mount of proc, look up the
1647 * respective sys/ file and write to it. To avoid mounting it when no
1648 * options were given, we mount it only when the first sysctl option is
1649 * found. Why not a persistent mount? There are problems with a
1650 * persistent mount of proc in that it forces userspace not to use any
1651 * proc mount options.
1652 */
1653 if (!*proc_mnt) {
1654 proc_fs_type = get_fs_type("proc");
1655 if (!proc_fs_type) {
1656 pr_err("Failed to find procfs to set sysctl from command line\n");
1657 return 0;
1658 }
1659 *proc_mnt = kern_mount(proc_fs_type);
1660 put_filesystem(proc_fs_type);
1661 if (IS_ERR(*proc_mnt)) {
1662 pr_err("Failed to mount procfs to set sysctl from command line\n");
1663 return 0;
1664 }
1665 }
1666
1667 path = kasprintf(GFP_KERNEL, "sys/%s", param);
1668 if (!path)
1669 panic("%s: Failed to allocate path for %s\n", __func__, param);
1670 strreplace(path, '.', '/');
1671
1672 file = file_open_root_mnt(*proc_mnt, path, O_WRONLY, 0);
1673 if (IS_ERR(file)) {
1674 err = PTR_ERR(file);
1675 if (err == -ENOENT)
1676 pr_err("Failed to set sysctl parameter '%s=%s': parameter not found\n",
1677 param, val);
1678 else if (err == -EACCES)
1679 pr_err("Failed to set sysctl parameter '%s=%s': permission denied (read-only?)\n",
1680 param, val);
1681 else
1682 pr_err("Error %pe opening proc file to set sysctl parameter '%s=%s'\n",
1683 file, param, val);
1684 goto out;
1685 }
1686 wret = kernel_write(file, val, len, &pos);
1687 if (wret < 0) {
1688 err = wret;
1689 if (err == -EINVAL)
1690 pr_err("Failed to set sysctl parameter '%s=%s': invalid value\n",
1691 param, val);
1692 else
1693 pr_err("Error %pe writing to proc file to set sysctl parameter '%s=%s'\n",
1694 ERR_PTR(err), param, val);
1695 } else if (wret != len) {
1696 pr_err("Wrote only %zd bytes of %d writing to proc file %s to set sysctl parameter '%s=%s\n",
1697 wret, len, path, param, val);
1698 }
1699
1700 err = filp_close(file, NULL);
1701 if (err)
1702 pr_err("Error %pe closing proc file to set sysctl parameter '%s=%s\n",
1703 ERR_PTR(err), param, val);
1704out:
1705 kfree(path);
1706 return 0;
1707}
1708
1709void do_sysctl_args(void)
1710{
1711 char *command_line;
1712 struct vfsmount *proc_mnt = NULL;
1713
1714 command_line = kstrdup(saved_command_line, GFP_KERNEL);
1715 if (!command_line)
1716 panic("%s: Failed to allocate copy of command line\n", __func__);
1717
1718 parse_args("Setting sysctl args", command_line,
1719 NULL, 0, -1, -1, &proc_mnt, process_sysctl_arg);
1720
1721 if (proc_mnt)
1722 kern_unmount(proc_mnt);
1723
1724 kfree(command_line);
1725}