Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0
2 : /*
3 : * /proc/sys support
4 : */
5 : #include <linux/init.h>
6 : #include <linux/sysctl.h>
7 : #include <linux/poll.h>
8 : #include <linux/proc_fs.h>
9 : #include <linux/printk.h>
10 : #include <linux/security.h>
11 : #include <linux/sched.h>
12 : #include <linux/cred.h>
13 : #include <linux/namei.h>
14 : #include <linux/mm.h>
15 : #include <linux/uio.h>
16 : #include <linux/module.h>
17 : #include <linux/bpf-cgroup.h>
18 : #include <linux/mount.h>
19 : #include <linux/kmemleak.h>
20 : #include "internal.h"
21 :
22 : #define list_for_each_table_entry(entry, table) \
23 : for ((entry) = (table); (entry)->procname; (entry)++)
24 :
25 : static const struct dentry_operations proc_sys_dentry_operations;
26 : static const struct file_operations proc_sys_file_operations;
27 : static const struct inode_operations proc_sys_inode_operations;
28 : static const struct file_operations proc_sys_dir_file_operations;
29 : static const struct inode_operations proc_sys_dir_operations;
30 :
31 : /* Support for permanently empty directories */
32 :
33 : struct ctl_table sysctl_mount_point[] = {
34 : { }
35 : };
36 :
37 : /**
38 : * register_sysctl_mount_point() - registers a sysctl mount point
39 : * @path: path for the mount point
40 : *
41 : * Used to create a permanently empty directory to serve as mount point.
42 : * There are some subtle but important permission checks this allows in the
43 : * case of unprivileged mounts.
44 : */
45 0 : struct ctl_table_header *register_sysctl_mount_point(const char *path)
46 : {
47 0 : return register_sysctl(path, sysctl_mount_point);
48 : }
49 : EXPORT_SYMBOL(register_sysctl_mount_point);
50 :
51 : static bool is_empty_dir(struct ctl_table_header *head)
52 : {
53 52 : return head->ctl_table[0].child == sysctl_mount_point;
54 : }
55 :
56 : static void set_empty_dir(struct ctl_dir *dir)
57 : {
58 0 : dir->header.ctl_table[0].child = sysctl_mount_point;
59 : }
60 :
61 : static void clear_empty_dir(struct ctl_dir *dir)
62 :
63 : {
64 0 : dir->header.ctl_table[0].child = NULL;
65 : }
66 :
67 0 : void proc_sys_poll_notify(struct ctl_table_poll *poll)
68 : {
69 0 : if (!poll)
70 : return;
71 :
72 0 : atomic_inc(&poll->event);
73 0 : wake_up_interruptible(&poll->wait);
74 : }
75 :
76 : static struct ctl_table root_table[] = {
77 : {
78 : .procname = "",
79 : .mode = S_IFDIR|S_IRUGO|S_IXUGO,
80 : },
81 : { }
82 : };
83 : static struct ctl_table_root sysctl_table_root = {
84 : .default_set.dir.header = {
85 : {{.count = 1,
86 : .nreg = 1,
87 : .ctl_table = root_table }},
88 : .ctl_table_arg = root_table,
89 : .root = &sysctl_table_root,
90 : .set = &sysctl_table_root.default_set,
91 : },
92 : };
93 :
94 : static DEFINE_SPINLOCK(sysctl_lock);
95 :
96 : static void drop_sysctl_table(struct ctl_table_header *header);
97 : static int sysctl_follow_link(struct ctl_table_header **phead,
98 : struct ctl_table **pentry);
99 : static int insert_links(struct ctl_table_header *head);
100 : static void put_links(struct ctl_table_header *header);
101 :
102 0 : static void sysctl_print_dir(struct ctl_dir *dir)
103 : {
104 0 : if (dir->header.parent)
105 0 : sysctl_print_dir(dir->header.parent);
106 0 : pr_cont("%s/", dir->header.ctl_table[0].procname);
107 0 : }
108 :
109 : static int namecmp(const char *name1, int len1, const char *name2, int len2)
110 : {
111 : int cmp;
112 :
113 644 : cmp = memcmp(name1, name2, min(len1, len2));
114 644 : if (cmp == 0)
115 44 : cmp = len1 - len2;
116 : return cmp;
117 : }
118 :
119 : /* Called under sysctl_lock */
120 63 : static struct ctl_table *find_entry(struct ctl_table_header **phead,
121 : struct ctl_dir *dir, const char *name, int namelen)
122 : {
123 : struct ctl_table_header *head;
124 : struct ctl_table *entry;
125 63 : struct rb_node *node = dir->root.rb_node;
126 :
127 167 : while (node)
128 : {
129 : struct ctl_node *ctl_node;
130 : const char *procname;
131 : int cmp;
132 :
133 137 : ctl_node = rb_entry(node, struct ctl_node, node);
134 137 : head = ctl_node->header;
135 137 : entry = &head->ctl_table[ctl_node - head->node];
136 137 : procname = entry->procname;
137 :
138 274 : cmp = namecmp(name, namelen, procname, strlen(procname));
139 137 : if (cmp < 0)
140 51 : node = node->rb_left;
141 86 : else if (cmp > 0)
142 53 : node = node->rb_right;
143 : else {
144 33 : *phead = head;
145 : return entry;
146 : }
147 : }
148 : return NULL;
149 : }
150 :
151 149 : static int insert_entry(struct ctl_table_header *head, struct ctl_table *entry)
152 : {
153 149 : struct rb_node *node = &head->node[entry - head->ctl_table].node;
154 149 : struct rb_node **p = &head->parent->root.rb_node;
155 149 : struct rb_node *parent = NULL;
156 149 : const char *name = entry->procname;
157 149 : int namelen = strlen(name);
158 :
159 805 : while (*p) {
160 : struct ctl_table_header *parent_head;
161 : struct ctl_table *parent_entry;
162 : struct ctl_node *parent_node;
163 : const char *parent_name;
164 : int cmp;
165 :
166 507 : parent = *p;
167 507 : parent_node = rb_entry(parent, struct ctl_node, node);
168 507 : parent_head = parent_node->header;
169 507 : parent_entry = &parent_head->ctl_table[parent_node - parent_head->node];
170 507 : parent_name = parent_entry->procname;
171 :
172 1014 : cmp = namecmp(name, namelen, parent_name, strlen(parent_name));
173 507 : if (cmp < 0)
174 235 : p = &(*p)->rb_left;
175 272 : else if (cmp > 0)
176 272 : p = &(*p)->rb_right;
177 : else {
178 0 : pr_err("sysctl duplicate entry: ");
179 0 : sysctl_print_dir(head->parent);
180 0 : pr_cont("%s\n", entry->procname);
181 0 : return -EEXIST;
182 : }
183 : }
184 :
185 149 : rb_link_node(node, parent, p);
186 149 : rb_insert_color(node, &head->parent->root);
187 149 : return 0;
188 : }
189 :
190 : static void erase_entry(struct ctl_table_header *head, struct ctl_table *entry)
191 : {
192 0 : struct rb_node *node = &head->node[entry - head->ctl_table].node;
193 :
194 0 : rb_erase(node, &head->parent->root);
195 : }
196 :
197 : static void init_header(struct ctl_table_header *head,
198 : struct ctl_table_root *root, struct ctl_table_set *set,
199 : struct ctl_node *node, struct ctl_table *table)
200 : {
201 53 : head->ctl_table = table;
202 53 : head->ctl_table_arg = table;
203 53 : head->used = 0;
204 53 : head->count = 1;
205 53 : head->nreg = 1;
206 53 : head->unregistering = NULL;
207 53 : head->root = root;
208 53 : head->set = set;
209 53 : head->parent = NULL;
210 53 : head->node = node;
211 53 : INIT_HLIST_HEAD(&head->inodes);
212 52 : if (node) {
213 : struct ctl_table *entry;
214 :
215 149 : list_for_each_table_entry(entry, table) {
216 149 : node->header = head;
217 149 : node++;
218 : }
219 : }
220 : }
221 :
222 0 : static void erase_header(struct ctl_table_header *head)
223 : {
224 : struct ctl_table *entry;
225 :
226 0 : list_for_each_table_entry(entry, head->ctl_table)
227 0 : erase_entry(head, entry);
228 0 : }
229 :
230 52 : static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
231 : {
232 : struct ctl_table *entry;
233 : int err;
234 :
235 : /* Is this a permanently empty directory? */
236 104 : if (is_empty_dir(&dir->header))
237 : return -EROFS;
238 :
239 : /* Am I creating a permanently empty directory? */
240 52 : if (header->ctl_table == sysctl_mount_point) {
241 0 : if (!RB_EMPTY_ROOT(&dir->root))
242 : return -EINVAL;
243 0 : set_empty_dir(dir);
244 : }
245 :
246 52 : dir->header.nreg++;
247 52 : header->parent = dir;
248 52 : err = insert_links(header);
249 52 : if (err)
250 : goto fail_links;
251 201 : list_for_each_table_entry(entry, header->ctl_table) {
252 149 : err = insert_entry(header, entry);
253 149 : if (err)
254 : goto fail;
255 : }
256 : return 0;
257 : fail:
258 0 : erase_header(header);
259 0 : put_links(header);
260 : fail_links:
261 0 : if (header->ctl_table == sysctl_mount_point)
262 0 : clear_empty_dir(dir);
263 0 : header->parent = NULL;
264 0 : drop_sysctl_table(&dir->header);
265 0 : return err;
266 : }
267 :
268 : /* called under sysctl_lock */
269 : static int use_table(struct ctl_table_header *p)
270 : {
271 0 : if (unlikely(p->unregistering))
272 : return 0;
273 0 : p->used++;
274 : return 1;
275 : }
276 :
277 : /* called under sysctl_lock */
278 : static void unuse_table(struct ctl_table_header *p)
279 : {
280 0 : if (!--p->used)
281 0 : if (unlikely(p->unregistering))
282 0 : complete(p->unregistering);
283 : }
284 :
285 : static void proc_sys_invalidate_dcache(struct ctl_table_header *head)
286 : {
287 0 : proc_invalidate_siblings_dcache(&head->inodes, &sysctl_lock);
288 : }
289 :
290 : /* called under sysctl_lock, will reacquire if has to wait */
291 0 : static void start_unregistering(struct ctl_table_header *p)
292 : {
293 : /*
294 : * if p->used is 0, nobody will ever touch that entry again;
295 : * we'll eliminate all paths to it before dropping sysctl_lock
296 : */
297 0 : if (unlikely(p->used)) {
298 : struct completion wait;
299 0 : init_completion(&wait);
300 0 : p->unregistering = &wait;
301 0 : spin_unlock(&sysctl_lock);
302 0 : wait_for_completion(&wait);
303 : } else {
304 : /* anything non-NULL; we'll never dereference it */
305 0 : p->unregistering = ERR_PTR(-EINVAL);
306 : spin_unlock(&sysctl_lock);
307 : }
308 : /*
309 : * Invalidate dentries for unregistered sysctls: namespaced sysctls
310 : * can have duplicate names and contaminate dcache very badly.
311 : */
312 0 : proc_sys_invalidate_dcache(p);
313 : /*
314 : * do not remove from the list until nobody holds it; walking the
315 : * list in do_sysctl() relies on that.
316 : */
317 0 : spin_lock(&sysctl_lock);
318 0 : erase_header(p);
319 0 : }
320 :
321 0 : static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
322 : {
323 0 : BUG_ON(!head);
324 0 : spin_lock(&sysctl_lock);
325 0 : if (!use_table(head))
326 0 : head = ERR_PTR(-ENOENT);
327 0 : spin_unlock(&sysctl_lock);
328 0 : return head;
329 : }
330 :
331 0 : static void sysctl_head_finish(struct ctl_table_header *head)
332 : {
333 0 : if (!head)
334 : return;
335 0 : spin_lock(&sysctl_lock);
336 0 : unuse_table(head);
337 : spin_unlock(&sysctl_lock);
338 : }
339 :
340 : static struct ctl_table_set *
341 : lookup_header_set(struct ctl_table_root *root)
342 : {
343 0 : struct ctl_table_set *set = &root->default_set;
344 0 : if (root->lookup)
345 0 : set = root->lookup(root);
346 : return set;
347 : }
348 :
349 0 : static struct ctl_table *lookup_entry(struct ctl_table_header **phead,
350 : struct ctl_dir *dir,
351 : const char *name, int namelen)
352 : {
353 : struct ctl_table_header *head;
354 : struct ctl_table *entry;
355 :
356 0 : spin_lock(&sysctl_lock);
357 0 : entry = find_entry(&head, dir, name, namelen);
358 0 : if (entry && use_table(head))
359 0 : *phead = head;
360 : else
361 : entry = NULL;
362 0 : spin_unlock(&sysctl_lock);
363 0 : return entry;
364 : }
365 :
366 0 : static struct ctl_node *first_usable_entry(struct rb_node *node)
367 : {
368 : struct ctl_node *ctl_node;
369 :
370 0 : for (;node; node = rb_next(node)) {
371 0 : ctl_node = rb_entry(node, struct ctl_node, node);
372 0 : if (use_table(ctl_node->header))
373 : return ctl_node;
374 : }
375 : return NULL;
376 : }
377 :
378 0 : static void first_entry(struct ctl_dir *dir,
379 : struct ctl_table_header **phead, struct ctl_table **pentry)
380 : {
381 0 : struct ctl_table_header *head = NULL;
382 0 : struct ctl_table *entry = NULL;
383 : struct ctl_node *ctl_node;
384 :
385 0 : spin_lock(&sysctl_lock);
386 0 : ctl_node = first_usable_entry(rb_first(&dir->root));
387 0 : spin_unlock(&sysctl_lock);
388 0 : if (ctl_node) {
389 0 : head = ctl_node->header;
390 0 : entry = &head->ctl_table[ctl_node - head->node];
391 : }
392 0 : *phead = head;
393 0 : *pentry = entry;
394 0 : }
395 :
396 0 : static void next_entry(struct ctl_table_header **phead, struct ctl_table **pentry)
397 : {
398 0 : struct ctl_table_header *head = *phead;
399 0 : struct ctl_table *entry = *pentry;
400 0 : struct ctl_node *ctl_node = &head->node[entry - head->ctl_table];
401 :
402 0 : spin_lock(&sysctl_lock);
403 0 : unuse_table(head);
404 :
405 0 : ctl_node = first_usable_entry(rb_next(&ctl_node->node));
406 0 : spin_unlock(&sysctl_lock);
407 0 : head = NULL;
408 0 : if (ctl_node) {
409 0 : head = ctl_node->header;
410 0 : entry = &head->ctl_table[ctl_node - head->node];
411 : }
412 0 : *phead = head;
413 0 : *pentry = entry;
414 0 : }
415 :
416 : /*
417 : * sysctl_perm does NOT grant the superuser all rights automatically, because
418 : * some sysctl variables are readonly even to root.
419 : */
420 :
421 0 : static int test_perm(int mode, int op)
422 : {
423 0 : if (uid_eq(current_euid(), GLOBAL_ROOT_UID))
424 0 : mode >>= 6;
425 0 : else if (in_egroup_p(GLOBAL_ROOT_GID))
426 0 : mode >>= 3;
427 0 : if ((op & ~mode & (MAY_READ|MAY_WRITE|MAY_EXEC)) == 0)
428 : return 0;
429 0 : return -EACCES;
430 : }
431 :
432 0 : static int sysctl_perm(struct ctl_table_header *head, struct ctl_table *table, int op)
433 : {
434 0 : struct ctl_table_root *root = head->root;
435 : int mode;
436 :
437 0 : if (root->permissions)
438 0 : mode = root->permissions(head, table);
439 : else
440 0 : mode = table->mode;
441 :
442 0 : return test_perm(mode, op);
443 : }
444 :
445 0 : static struct inode *proc_sys_make_inode(struct super_block *sb,
446 : struct ctl_table_header *head, struct ctl_table *table)
447 : {
448 0 : struct ctl_table_root *root = head->root;
449 : struct inode *inode;
450 : struct proc_inode *ei;
451 :
452 0 : inode = new_inode(sb);
453 0 : if (!inode)
454 : return ERR_PTR(-ENOMEM);
455 :
456 0 : inode->i_ino = get_next_ino();
457 :
458 0 : ei = PROC_I(inode);
459 :
460 0 : spin_lock(&sysctl_lock);
461 0 : if (unlikely(head->unregistering)) {
462 0 : spin_unlock(&sysctl_lock);
463 0 : iput(inode);
464 0 : return ERR_PTR(-ENOENT);
465 : }
466 0 : ei->sysctl = head;
467 0 : ei->sysctl_entry = table;
468 0 : hlist_add_head_rcu(&ei->sibling_inodes, &head->inodes);
469 0 : head->count++;
470 0 : spin_unlock(&sysctl_lock);
471 :
472 0 : inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
473 0 : inode->i_mode = table->mode;
474 0 : if (!S_ISDIR(table->mode)) {
475 0 : inode->i_mode |= S_IFREG;
476 0 : inode->i_op = &proc_sys_inode_operations;
477 0 : inode->i_fop = &proc_sys_file_operations;
478 : } else {
479 0 : inode->i_mode |= S_IFDIR;
480 0 : inode->i_op = &proc_sys_dir_operations;
481 0 : inode->i_fop = &proc_sys_dir_file_operations;
482 0 : if (is_empty_dir(head))
483 0 : make_empty_dir_inode(inode);
484 : }
485 :
486 0 : if (root->set_ownership)
487 0 : root->set_ownership(head, table, &inode->i_uid, &inode->i_gid);
488 : else {
489 0 : inode->i_uid = GLOBAL_ROOT_UID;
490 0 : inode->i_gid = GLOBAL_ROOT_GID;
491 : }
492 :
493 : return inode;
494 : }
495 :
496 0 : void proc_sys_evict_inode(struct inode *inode, struct ctl_table_header *head)
497 : {
498 0 : spin_lock(&sysctl_lock);
499 0 : hlist_del_init_rcu(&PROC_I(inode)->sibling_inodes);
500 0 : if (!--head->count)
501 0 : kfree_rcu(head, rcu);
502 0 : spin_unlock(&sysctl_lock);
503 0 : }
504 :
505 : static struct ctl_table_header *grab_header(struct inode *inode)
506 : {
507 0 : struct ctl_table_header *head = PROC_I(inode)->sysctl;
508 0 : if (!head)
509 0 : head = &sysctl_table_root.default_set.dir.header;
510 0 : return sysctl_head_grab(head);
511 : }
512 :
513 0 : static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
514 : unsigned int flags)
515 : {
516 0 : struct ctl_table_header *head = grab_header(dir);
517 0 : struct ctl_table_header *h = NULL;
518 0 : const struct qstr *name = &dentry->d_name;
519 : struct ctl_table *p;
520 : struct inode *inode;
521 0 : struct dentry *err = ERR_PTR(-ENOENT);
522 : struct ctl_dir *ctl_dir;
523 : int ret;
524 :
525 0 : if (IS_ERR(head))
526 : return ERR_CAST(head);
527 :
528 0 : ctl_dir = container_of(head, struct ctl_dir, header);
529 :
530 0 : p = lookup_entry(&h, ctl_dir, name->name, name->len);
531 0 : if (!p)
532 : goto out;
533 :
534 0 : if (S_ISLNK(p->mode)) {
535 0 : ret = sysctl_follow_link(&h, &p);
536 0 : err = ERR_PTR(ret);
537 0 : if (ret)
538 : goto out;
539 : }
540 :
541 0 : inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
542 0 : if (IS_ERR(inode)) {
543 : err = ERR_CAST(inode);
544 : goto out;
545 : }
546 :
547 0 : d_set_d_op(dentry, &proc_sys_dentry_operations);
548 0 : err = d_splice_alias(inode, dentry);
549 :
550 : out:
551 0 : if (h)
552 0 : sysctl_head_finish(h);
553 0 : sysctl_head_finish(head);
554 0 : return err;
555 : }
556 :
557 0 : static ssize_t proc_sys_call_handler(struct kiocb *iocb, struct iov_iter *iter,
558 : int write)
559 : {
560 0 : struct inode *inode = file_inode(iocb->ki_filp);
561 0 : struct ctl_table_header *head = grab_header(inode);
562 0 : struct ctl_table *table = PROC_I(inode)->sysctl_entry;
563 0 : size_t count = iov_iter_count(iter);
564 : char *kbuf;
565 : ssize_t error;
566 :
567 0 : if (IS_ERR(head))
568 0 : return PTR_ERR(head);
569 :
570 : /*
571 : * At this point we know that the sysctl was not unregistered
572 : * and won't be until we finish.
573 : */
574 0 : error = -EPERM;
575 0 : if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
576 : goto out;
577 :
578 : /* if that can happen at all, it should be -EINVAL, not -EISDIR */
579 0 : error = -EINVAL;
580 0 : if (!table->proc_handler)
581 : goto out;
582 :
583 : /* don't even try if the size is too large */
584 0 : error = -ENOMEM;
585 0 : if (count >= KMALLOC_MAX_SIZE)
586 : goto out;
587 0 : kbuf = kvzalloc(count + 1, GFP_KERNEL);
588 0 : if (!kbuf)
589 : goto out;
590 :
591 0 : if (write) {
592 0 : error = -EFAULT;
593 0 : if (!copy_from_iter_full(kbuf, count, iter))
594 : goto out_free_buf;
595 0 : kbuf[count] = '\0';
596 : }
597 :
598 0 : error = BPF_CGROUP_RUN_PROG_SYSCTL(head, table, write, &kbuf, &count,
599 : &iocb->ki_pos);
600 : if (error)
601 : goto out_free_buf;
602 :
603 : /* careful: calling conventions are nasty here */
604 0 : error = table->proc_handler(table, write, kbuf, &count, &iocb->ki_pos);
605 0 : if (error)
606 : goto out_free_buf;
607 :
608 0 : if (!write) {
609 0 : error = -EFAULT;
610 0 : if (copy_to_iter(kbuf, count, iter) < count)
611 : goto out_free_buf;
612 : }
613 :
614 0 : error = count;
615 : out_free_buf:
616 0 : kvfree(kbuf);
617 : out:
618 0 : sysctl_head_finish(head);
619 :
620 0 : return error;
621 : }
622 :
623 0 : static ssize_t proc_sys_read(struct kiocb *iocb, struct iov_iter *iter)
624 : {
625 0 : return proc_sys_call_handler(iocb, iter, 0);
626 : }
627 :
628 0 : static ssize_t proc_sys_write(struct kiocb *iocb, struct iov_iter *iter)
629 : {
630 0 : return proc_sys_call_handler(iocb, iter, 1);
631 : }
632 :
633 0 : static int proc_sys_open(struct inode *inode, struct file *filp)
634 : {
635 0 : struct ctl_table_header *head = grab_header(inode);
636 0 : struct ctl_table *table = PROC_I(inode)->sysctl_entry;
637 :
638 : /* sysctl was unregistered */
639 0 : if (IS_ERR(head))
640 0 : return PTR_ERR(head);
641 :
642 0 : if (table->poll)
643 0 : filp->private_data = proc_sys_poll_event(table->poll);
644 :
645 0 : sysctl_head_finish(head);
646 :
647 0 : return 0;
648 : }
649 :
650 0 : static __poll_t proc_sys_poll(struct file *filp, poll_table *wait)
651 : {
652 0 : struct inode *inode = file_inode(filp);
653 0 : struct ctl_table_header *head = grab_header(inode);
654 0 : struct ctl_table *table = PROC_I(inode)->sysctl_entry;
655 0 : __poll_t ret = DEFAULT_POLLMASK;
656 : unsigned long event;
657 :
658 : /* sysctl was unregistered */
659 0 : if (IS_ERR(head))
660 : return EPOLLERR | EPOLLHUP;
661 :
662 0 : if (!table->proc_handler)
663 : goto out;
664 :
665 0 : if (!table->poll)
666 : goto out;
667 :
668 0 : event = (unsigned long)filp->private_data;
669 0 : poll_wait(filp, &table->poll->wait, wait);
670 :
671 0 : if (event != atomic_read(&table->poll->event)) {
672 0 : filp->private_data = proc_sys_poll_event(table->poll);
673 0 : ret = EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
674 : }
675 :
676 : out:
677 0 : sysctl_head_finish(head);
678 :
679 0 : return ret;
680 : }
681 :
682 0 : static bool proc_sys_fill_cache(struct file *file,
683 : struct dir_context *ctx,
684 : struct ctl_table_header *head,
685 : struct ctl_table *table)
686 : {
687 0 : struct dentry *child, *dir = file->f_path.dentry;
688 : struct inode *inode;
689 : struct qstr qname;
690 0 : ino_t ino = 0;
691 0 : unsigned type = DT_UNKNOWN;
692 :
693 0 : qname.name = table->procname;
694 0 : qname.len = strlen(table->procname);
695 0 : qname.hash = full_name_hash(dir, qname.name, qname.len);
696 :
697 0 : child = d_lookup(dir, &qname);
698 0 : if (!child) {
699 0 : DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
700 0 : child = d_alloc_parallel(dir, &qname, &wq);
701 0 : if (IS_ERR(child))
702 0 : return false;
703 0 : if (d_in_lookup(child)) {
704 : struct dentry *res;
705 0 : inode = proc_sys_make_inode(dir->d_sb, head, table);
706 0 : if (IS_ERR(inode)) {
707 0 : d_lookup_done(child);
708 0 : dput(child);
709 : return false;
710 : }
711 0 : d_set_d_op(child, &proc_sys_dentry_operations);
712 0 : res = d_splice_alias(inode, child);
713 0 : d_lookup_done(child);
714 0 : if (unlikely(res)) {
715 0 : if (IS_ERR(res)) {
716 0 : dput(child);
717 : return false;
718 : }
719 0 : dput(child);
720 0 : child = res;
721 : }
722 : }
723 : }
724 0 : inode = d_inode(child);
725 0 : ino = inode->i_ino;
726 0 : type = inode->i_mode >> 12;
727 0 : dput(child);
728 0 : return dir_emit(ctx, qname.name, qname.len, ino, type);
729 : }
730 :
731 0 : static bool proc_sys_link_fill_cache(struct file *file,
732 : struct dir_context *ctx,
733 : struct ctl_table_header *head,
734 : struct ctl_table *table)
735 : {
736 0 : bool ret = true;
737 :
738 0 : head = sysctl_head_grab(head);
739 0 : if (IS_ERR(head))
740 : return false;
741 :
742 : /* It is not an error if we can not follow the link ignore it */
743 0 : if (sysctl_follow_link(&head, &table))
744 : goto out;
745 :
746 0 : ret = proc_sys_fill_cache(file, ctx, head, table);
747 : out:
748 0 : sysctl_head_finish(head);
749 0 : return ret;
750 : }
751 :
752 0 : static int scan(struct ctl_table_header *head, struct ctl_table *table,
753 : unsigned long *pos, struct file *file,
754 : struct dir_context *ctx)
755 : {
756 : bool res;
757 :
758 0 : if ((*pos)++ < ctx->pos)
759 : return true;
760 :
761 0 : if (unlikely(S_ISLNK(table->mode)))
762 0 : res = proc_sys_link_fill_cache(file, ctx, head, table);
763 : else
764 0 : res = proc_sys_fill_cache(file, ctx, head, table);
765 :
766 0 : if (res)
767 0 : ctx->pos = *pos;
768 :
769 0 : return res;
770 : }
771 :
772 0 : static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
773 : {
774 0 : struct ctl_table_header *head = grab_header(file_inode(file));
775 0 : struct ctl_table_header *h = NULL;
776 : struct ctl_table *entry;
777 : struct ctl_dir *ctl_dir;
778 : unsigned long pos;
779 :
780 0 : if (IS_ERR(head))
781 0 : return PTR_ERR(head);
782 :
783 0 : ctl_dir = container_of(head, struct ctl_dir, header);
784 :
785 0 : if (!dir_emit_dots(file, ctx))
786 : goto out;
787 :
788 0 : pos = 2;
789 :
790 0 : for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
791 0 : if (!scan(h, entry, &pos, file, ctx)) {
792 0 : sysctl_head_finish(h);
793 0 : break;
794 : }
795 : }
796 : out:
797 0 : sysctl_head_finish(head);
798 0 : return 0;
799 : }
800 :
801 0 : static int proc_sys_permission(struct mnt_idmap *idmap,
802 : struct inode *inode, int mask)
803 : {
804 : /*
805 : * sysctl entries that are not writeable,
806 : * are _NOT_ writeable, capabilities or not.
807 : */
808 : struct ctl_table_header *head;
809 : struct ctl_table *table;
810 : int error;
811 :
812 : /* Executable files are not allowed under /proc/sys/ */
813 0 : if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
814 : return -EACCES;
815 :
816 0 : head = grab_header(inode);
817 0 : if (IS_ERR(head))
818 0 : return PTR_ERR(head);
819 :
820 0 : table = PROC_I(inode)->sysctl_entry;
821 0 : if (!table) /* global root - r-xr-xr-x */
822 0 : error = mask & MAY_WRITE ? -EACCES : 0;
823 : else /* Use the permissions on the sysctl table entry */
824 0 : error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
825 :
826 0 : sysctl_head_finish(head);
827 0 : return error;
828 : }
829 :
830 0 : static int proc_sys_setattr(struct mnt_idmap *idmap,
831 : struct dentry *dentry, struct iattr *attr)
832 : {
833 0 : struct inode *inode = d_inode(dentry);
834 : int error;
835 :
836 0 : if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
837 : return -EPERM;
838 :
839 0 : error = setattr_prepare(&nop_mnt_idmap, dentry, attr);
840 0 : if (error)
841 : return error;
842 :
843 0 : setattr_copy(&nop_mnt_idmap, inode, attr);
844 0 : return 0;
845 : }
846 :
847 0 : static int proc_sys_getattr(struct mnt_idmap *idmap,
848 : const struct path *path, struct kstat *stat,
849 : u32 request_mask, unsigned int query_flags)
850 : {
851 0 : struct inode *inode = d_inode(path->dentry);
852 0 : struct ctl_table_header *head = grab_header(inode);
853 0 : struct ctl_table *table = PROC_I(inode)->sysctl_entry;
854 :
855 0 : if (IS_ERR(head))
856 0 : return PTR_ERR(head);
857 :
858 0 : generic_fillattr(&nop_mnt_idmap, inode, stat);
859 0 : if (table)
860 0 : stat->mode = (stat->mode & S_IFMT) | table->mode;
861 :
862 0 : sysctl_head_finish(head);
863 0 : return 0;
864 : }
865 :
866 : static const struct file_operations proc_sys_file_operations = {
867 : .open = proc_sys_open,
868 : .poll = proc_sys_poll,
869 : .read_iter = proc_sys_read,
870 : .write_iter = proc_sys_write,
871 : .splice_read = generic_file_splice_read,
872 : .splice_write = iter_file_splice_write,
873 : .llseek = default_llseek,
874 : };
875 :
876 : static const struct file_operations proc_sys_dir_file_operations = {
877 : .read = generic_read_dir,
878 : .iterate_shared = proc_sys_readdir,
879 : .llseek = generic_file_llseek,
880 : };
881 :
882 : static const struct inode_operations proc_sys_inode_operations = {
883 : .permission = proc_sys_permission,
884 : .setattr = proc_sys_setattr,
885 : .getattr = proc_sys_getattr,
886 : };
887 :
888 : static const struct inode_operations proc_sys_dir_operations = {
889 : .lookup = proc_sys_lookup,
890 : .permission = proc_sys_permission,
891 : .setattr = proc_sys_setattr,
892 : .getattr = proc_sys_getattr,
893 : };
894 :
895 0 : static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
896 : {
897 0 : if (flags & LOOKUP_RCU)
898 : return -ECHILD;
899 0 : return !PROC_I(d_inode(dentry))->sysctl->unregistering;
900 : }
901 :
902 0 : static int proc_sys_delete(const struct dentry *dentry)
903 : {
904 0 : return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
905 : }
906 :
907 : static int sysctl_is_seen(struct ctl_table_header *p)
908 : {
909 0 : struct ctl_table_set *set = p->set;
910 : int res;
911 0 : spin_lock(&sysctl_lock);
912 0 : if (p->unregistering)
913 : res = 0;
914 0 : else if (!set->is_seen)
915 : res = 1;
916 : else
917 0 : res = set->is_seen(set);
918 0 : spin_unlock(&sysctl_lock);
919 : return res;
920 : }
921 :
922 0 : static int proc_sys_compare(const struct dentry *dentry,
923 : unsigned int len, const char *str, const struct qstr *name)
924 : {
925 : struct ctl_table_header *head;
926 : struct inode *inode;
927 :
928 : /* Although proc doesn't have negative dentries, rcu-walk means
929 : * that inode here can be NULL */
930 : /* AV: can it, indeed? */
931 0 : inode = d_inode_rcu(dentry);
932 0 : if (!inode)
933 : return 1;
934 0 : if (name->len != len)
935 : return 1;
936 0 : if (memcmp(name->name, str, len))
937 : return 1;
938 0 : head = rcu_dereference(PROC_I(inode)->sysctl);
939 0 : return !head || !sysctl_is_seen(head);
940 : }
941 :
942 : static const struct dentry_operations proc_sys_dentry_operations = {
943 : .d_revalidate = proc_sys_revalidate,
944 : .d_delete = proc_sys_delete,
945 : .d_compare = proc_sys_compare,
946 : };
947 :
948 59 : static struct ctl_dir *find_subdir(struct ctl_dir *dir,
949 : const char *name, int namelen)
950 : {
951 : struct ctl_table_header *head;
952 : struct ctl_table *entry;
953 :
954 59 : entry = find_entry(&head, dir, name, namelen);
955 59 : if (!entry)
956 : return ERR_PTR(-ENOENT);
957 31 : if (!S_ISDIR(entry->mode))
958 : return ERR_PTR(-ENOTDIR);
959 31 : return container_of(head, struct ctl_dir, header);
960 : }
961 :
962 14 : static struct ctl_dir *new_dir(struct ctl_table_set *set,
963 : const char *name, int namelen)
964 : {
965 : struct ctl_table *table;
966 : struct ctl_dir *new;
967 : struct ctl_node *node;
968 : char *new_name;
969 :
970 14 : new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
971 : sizeof(struct ctl_table)*2 + namelen + 1,
972 : GFP_KERNEL);
973 14 : if (!new)
974 : return NULL;
975 :
976 14 : node = (struct ctl_node *)(new + 1);
977 14 : table = (struct ctl_table *)(node + 1);
978 14 : new_name = (char *)(table + 2);
979 14 : memcpy(new_name, name, namelen);
980 14 : table[0].procname = new_name;
981 14 : table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
982 14 : init_header(&new->header, set->dir.header.root, set, node, table);
983 :
984 : return new;
985 : }
986 :
987 : /**
988 : * get_subdir - find or create a subdir with the specified name.
989 : * @dir: Directory to create the subdirectory in
990 : * @name: The name of the subdirectory to find or create
991 : * @namelen: The length of name
992 : *
993 : * Takes a directory with an elevated reference count so we know that
994 : * if we drop the lock the directory will not go away. Upon success
995 : * the reference is moved from @dir to the returned subdirectory.
996 : * Upon error an error code is returned and the reference on @dir is
997 : * simply dropped.
998 : */
999 44 : static struct ctl_dir *get_subdir(struct ctl_dir *dir,
1000 : const char *name, int namelen)
1001 : {
1002 44 : struct ctl_table_set *set = dir->header.set;
1003 44 : struct ctl_dir *subdir, *new = NULL;
1004 : int err;
1005 :
1006 44 : spin_lock(&sysctl_lock);
1007 44 : subdir = find_subdir(dir, name, namelen);
1008 44 : if (!IS_ERR(subdir))
1009 : goto found;
1010 14 : if (PTR_ERR(subdir) != -ENOENT)
1011 : goto failed;
1012 :
1013 14 : spin_unlock(&sysctl_lock);
1014 14 : new = new_dir(set, name, namelen);
1015 14 : spin_lock(&sysctl_lock);
1016 14 : subdir = ERR_PTR(-ENOMEM);
1017 14 : if (!new)
1018 : goto failed;
1019 :
1020 : /* Was the subdir added while we dropped the lock? */
1021 14 : subdir = find_subdir(dir, name, namelen);
1022 14 : if (!IS_ERR(subdir))
1023 : goto found;
1024 14 : if (PTR_ERR(subdir) != -ENOENT)
1025 : goto failed;
1026 :
1027 : /* Nope. Use the our freshly made directory entry. */
1028 14 : err = insert_header(dir, &new->header);
1029 28 : subdir = ERR_PTR(err);
1030 14 : if (err)
1031 : goto failed;
1032 : subdir = new;
1033 : found:
1034 44 : subdir->header.nreg++;
1035 : failed:
1036 44 : if (IS_ERR(subdir)) {
1037 0 : pr_err("sysctl could not get directory: ");
1038 0 : sysctl_print_dir(dir);
1039 0 : pr_cont("%*.*s %ld\n", namelen, namelen, name,
1040 : PTR_ERR(subdir));
1041 : }
1042 44 : drop_sysctl_table(&dir->header);
1043 44 : if (new)
1044 14 : drop_sysctl_table(&new->header);
1045 44 : spin_unlock(&sysctl_lock);
1046 44 : return subdir;
1047 : }
1048 :
1049 3 : static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
1050 : {
1051 : struct ctl_dir *parent;
1052 : const char *procname;
1053 3 : if (!dir->header.parent)
1054 2 : return &set->dir;
1055 1 : parent = xlate_dir(set, dir->header.parent);
1056 1 : if (IS_ERR(parent))
1057 : return parent;
1058 1 : procname = dir->header.ctl_table[0].procname;
1059 1 : return find_subdir(parent, procname, strlen(procname));
1060 : }
1061 :
1062 0 : static int sysctl_follow_link(struct ctl_table_header **phead,
1063 : struct ctl_table **pentry)
1064 : {
1065 : struct ctl_table_header *head;
1066 : struct ctl_table_root *root;
1067 : struct ctl_table_set *set;
1068 : struct ctl_table *entry;
1069 : struct ctl_dir *dir;
1070 : int ret;
1071 :
1072 0 : spin_lock(&sysctl_lock);
1073 0 : root = (*pentry)->data;
1074 0 : set = lookup_header_set(root);
1075 0 : dir = xlate_dir(set, (*phead)->parent);
1076 0 : if (IS_ERR(dir))
1077 0 : ret = PTR_ERR(dir);
1078 : else {
1079 0 : const char *procname = (*pentry)->procname;
1080 0 : head = NULL;
1081 0 : entry = find_entry(&head, dir, procname, strlen(procname));
1082 0 : ret = -ENOENT;
1083 0 : if (entry && use_table(head)) {
1084 0 : unuse_table(*phead);
1085 0 : *phead = head;
1086 0 : *pentry = entry;
1087 0 : ret = 0;
1088 : }
1089 : }
1090 :
1091 0 : spin_unlock(&sysctl_lock);
1092 0 : return ret;
1093 : }
1094 :
1095 0 : static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
1096 : {
1097 : struct va_format vaf;
1098 : va_list args;
1099 :
1100 0 : va_start(args, fmt);
1101 0 : vaf.fmt = fmt;
1102 0 : vaf.va = &args;
1103 :
1104 0 : pr_err("sysctl table check failed: %s/%s %pV\n",
1105 : path, table->procname, &vaf);
1106 :
1107 0 : va_end(args);
1108 0 : return -EINVAL;
1109 : }
1110 :
1111 78 : static int sysctl_check_table_array(const char *path, struct ctl_table *table)
1112 : {
1113 78 : int err = 0;
1114 :
1115 78 : if ((table->proc_handler == proc_douintvec) ||
1116 : (table->proc_handler == proc_douintvec_minmax)) {
1117 4 : if (table->maxlen != sizeof(unsigned int))
1118 0 : err |= sysctl_err(path, table, "array not allowed");
1119 : }
1120 :
1121 78 : if (table->proc_handler == proc_dou8vec_minmax) {
1122 0 : if (table->maxlen != sizeof(u8))
1123 0 : err |= sysctl_err(path, table, "array not allowed");
1124 : }
1125 :
1126 78 : if (table->proc_handler == proc_dobool) {
1127 1 : if (table->maxlen != sizeof(bool))
1128 0 : err |= sysctl_err(path, table, "array not allowed");
1129 : }
1130 :
1131 78 : return err;
1132 : }
1133 :
1134 37 : static int sysctl_check_table(const char *path, struct ctl_table *table)
1135 : {
1136 : struct ctl_table *entry;
1137 37 : int err = 0;
1138 162 : list_for_each_table_entry(entry, table) {
1139 125 : if (entry->child)
1140 0 : err |= sysctl_err(path, entry, "Not a file");
1141 :
1142 125 : if ((entry->proc_handler == proc_dostring) ||
1143 122 : (entry->proc_handler == proc_dobool) ||
1144 100 : (entry->proc_handler == proc_dointvec) ||
1145 98 : (entry->proc_handler == proc_douintvec) ||
1146 96 : (entry->proc_handler == proc_douintvec_minmax) ||
1147 70 : (entry->proc_handler == proc_dointvec_minmax) ||
1148 70 : (entry->proc_handler == proc_dou8vec_minmax) ||
1149 68 : (entry->proc_handler == proc_dointvec_jiffies) ||
1150 68 : (entry->proc_handler == proc_dointvec_userhz_jiffies) ||
1151 68 : (entry->proc_handler == proc_dointvec_ms_jiffies) ||
1152 47 : (entry->proc_handler == proc_doulongvec_minmax) ||
1153 : (entry->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1154 78 : if (!entry->data)
1155 0 : err |= sysctl_err(path, entry, "No data");
1156 78 : if (!entry->maxlen)
1157 0 : err |= sysctl_err(path, entry, "No maxlen");
1158 : else
1159 78 : err |= sysctl_check_table_array(path, entry);
1160 : }
1161 125 : if (!entry->proc_handler)
1162 0 : err |= sysctl_err(path, entry, "No proc_handler");
1163 :
1164 125 : if ((entry->mode & (S_IRUGO|S_IWUGO)) != entry->mode)
1165 0 : err |= sysctl_err(path, entry, "bogus .mode 0%o",
1166 : entry->mode);
1167 : }
1168 37 : return err;
1169 : }
1170 :
1171 1 : static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table *table,
1172 : struct ctl_table_root *link_root)
1173 : {
1174 : struct ctl_table *link_table, *entry, *link;
1175 : struct ctl_table_header *links;
1176 : struct ctl_node *node;
1177 : char *link_name;
1178 : int nr_entries, name_bytes;
1179 :
1180 1 : name_bytes = 0;
1181 1 : nr_entries = 0;
1182 11 : list_for_each_table_entry(entry, table) {
1183 10 : nr_entries++;
1184 10 : name_bytes += strlen(entry->procname) + 1;
1185 : }
1186 :
1187 1 : links = kzalloc(sizeof(struct ctl_table_header) +
1188 2 : sizeof(struct ctl_node)*nr_entries +
1189 2 : sizeof(struct ctl_table)*(nr_entries + 1) +
1190 : name_bytes,
1191 : GFP_KERNEL);
1192 :
1193 1 : if (!links)
1194 : return NULL;
1195 :
1196 1 : node = (struct ctl_node *)(links + 1);
1197 1 : link_table = (struct ctl_table *)(node + nr_entries);
1198 1 : link_name = (char *)&link_table[nr_entries + 1];
1199 1 : link = link_table;
1200 :
1201 11 : list_for_each_table_entry(entry, table) {
1202 10 : int len = strlen(entry->procname) + 1;
1203 10 : memcpy(link_name, entry->procname, len);
1204 10 : link->procname = link_name;
1205 10 : link->mode = S_IFLNK|S_IRWXUGO;
1206 10 : link->data = link_root;
1207 10 : link_name += len;
1208 10 : link++;
1209 : }
1210 2 : init_header(links, dir->header.root, dir->header.set, node, link_table);
1211 1 : links->nreg = nr_entries;
1212 :
1213 : return links;
1214 : }
1215 :
1216 3 : static bool get_links(struct ctl_dir *dir,
1217 : struct ctl_table *table, struct ctl_table_root *link_root)
1218 : {
1219 : struct ctl_table_header *head;
1220 : struct ctl_table *entry, *link;
1221 :
1222 : /* Are there links available for every entry in table? */
1223 4 : list_for_each_table_entry(entry, table) {
1224 3 : const char *procname = entry->procname;
1225 3 : link = find_entry(&head, dir, procname, strlen(procname));
1226 3 : if (!link)
1227 : return false;
1228 1 : if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1229 1 : continue;
1230 0 : if (S_ISLNK(link->mode) && (link->data == link_root))
1231 0 : continue;
1232 : return false;
1233 : }
1234 :
1235 : /* The checks passed. Increase the registration count on the links */
1236 1 : list_for_each_table_entry(entry, table) {
1237 1 : const char *procname = entry->procname;
1238 1 : link = find_entry(&head, dir, procname, strlen(procname));
1239 1 : head->nreg++;
1240 : }
1241 : return true;
1242 : }
1243 :
1244 52 : static int insert_links(struct ctl_table_header *head)
1245 : {
1246 52 : struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1247 : struct ctl_dir *core_parent;
1248 : struct ctl_table_header *links;
1249 : int err;
1250 :
1251 52 : if (head->set == root_set)
1252 : return 0;
1253 :
1254 2 : core_parent = xlate_dir(root_set, head->parent);
1255 2 : if (IS_ERR(core_parent))
1256 : return 0;
1257 :
1258 2 : if (get_links(core_parent, head->ctl_table, head->root))
1259 : return 0;
1260 :
1261 1 : core_parent->header.nreg++;
1262 1 : spin_unlock(&sysctl_lock);
1263 :
1264 1 : links = new_links(core_parent, head->ctl_table, head->root);
1265 :
1266 1 : spin_lock(&sysctl_lock);
1267 1 : err = -ENOMEM;
1268 1 : if (!links)
1269 : goto out;
1270 :
1271 1 : err = 0;
1272 1 : if (get_links(core_parent, head->ctl_table, head->root)) {
1273 0 : kfree(links);
1274 0 : goto out;
1275 : }
1276 :
1277 1 : err = insert_header(core_parent, links);
1278 1 : if (err)
1279 0 : kfree(links);
1280 : out:
1281 1 : drop_sysctl_table(&core_parent->header);
1282 1 : return err;
1283 : }
1284 :
1285 : /* Find the directory for the ctl_table. If one is not found create it. */
1286 37 : static struct ctl_dir *sysctl_mkdir_p(struct ctl_dir *dir, const char *path)
1287 : {
1288 : const char *name, *nextname;
1289 :
1290 124 : for (name = path; name; name = nextname) {
1291 : int namelen;
1292 50 : nextname = strchr(name, '/');
1293 50 : if (nextname) {
1294 13 : namelen = nextname - name;
1295 13 : nextname++;
1296 : } else {
1297 37 : namelen = strlen(name);
1298 : }
1299 50 : if (namelen == 0)
1300 6 : continue;
1301 :
1302 : /*
1303 : * namelen ensures if name is "foo/bar/yay" only foo is
1304 : * registered first. We traverse as if using mkdir -p and
1305 : * return a ctl_dir for the last directory entry.
1306 : */
1307 44 : dir = get_subdir(dir, name, namelen);
1308 44 : if (IS_ERR(dir))
1309 : break;
1310 : }
1311 37 : return dir;
1312 : }
1313 :
1314 : /**
1315 : * __register_sysctl_table - register a leaf sysctl table
1316 : * @set: Sysctl tree to register on
1317 : * @path: The path to the directory the sysctl table is in.
1318 : * @table: the top-level table structure without any child. This table
1319 : * should not be free'd after registration. So it should not be
1320 : * used on stack. It can either be a global or dynamically allocated
1321 : * by the caller and free'd later after sysctl unregistration.
1322 : *
1323 : * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1324 : * array. A completely 0 filled entry terminates the table.
1325 : *
1326 : * The members of the &struct ctl_table structure are used as follows:
1327 : *
1328 : * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1329 : * enter a sysctl file
1330 : *
1331 : * data - a pointer to data for use by proc_handler
1332 : *
1333 : * maxlen - the maximum size in bytes of the data
1334 : *
1335 : * mode - the file permissions for the /proc/sys file
1336 : *
1337 : * child - must be %NULL.
1338 : *
1339 : * proc_handler - the text handler routine (described below)
1340 : *
1341 : * extra1, extra2 - extra pointers usable by the proc handler routines
1342 : * XXX: we should eventually modify these to use long min / max [0]
1343 : * [0] https://lkml.kernel.org/87zgpte9o4.fsf@email.froward.int.ebiederm.org
1344 : *
1345 : * Leaf nodes in the sysctl tree will be represented by a single file
1346 : * under /proc; non-leaf nodes (where child is not NULL) are not allowed,
1347 : * sysctl_check_table() verifies this.
1348 : *
1349 : * There must be a proc_handler routine for any terminal nodes.
1350 : * Several default handlers are available to cover common cases -
1351 : *
1352 : * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1353 : * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1354 : * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1355 : *
1356 : * It is the handler's job to read the input buffer from user memory
1357 : * and process it. The handler should return 0 on success.
1358 : *
1359 : * This routine returns %NULL on a failure to register, and a pointer
1360 : * to the table header on success.
1361 : */
1362 37 : struct ctl_table_header *__register_sysctl_table(
1363 : struct ctl_table_set *set,
1364 : const char *path, struct ctl_table *table)
1365 : {
1366 37 : struct ctl_table_root *root = set->dir.header.root;
1367 : struct ctl_table_header *header;
1368 : struct ctl_dir *dir;
1369 : struct ctl_table *entry;
1370 : struct ctl_node *node;
1371 37 : int nr_entries = 0;
1372 :
1373 162 : list_for_each_table_entry(entry, table)
1374 125 : nr_entries++;
1375 :
1376 37 : header = kzalloc(sizeof(struct ctl_table_header) +
1377 37 : sizeof(struct ctl_node)*nr_entries, GFP_KERNEL_ACCOUNT);
1378 37 : if (!header)
1379 : return NULL;
1380 :
1381 37 : node = (struct ctl_node *)(header + 1);
1382 37 : init_header(header, root, set, node, table);
1383 37 : if (sysctl_check_table(path, table))
1384 : goto fail;
1385 :
1386 37 : spin_lock(&sysctl_lock);
1387 37 : dir = &set->dir;
1388 : /* Reference moved down the directory tree get_subdir */
1389 37 : dir->header.nreg++;
1390 37 : spin_unlock(&sysctl_lock);
1391 :
1392 37 : dir = sysctl_mkdir_p(dir, path);
1393 37 : if (IS_ERR(dir))
1394 : goto fail;
1395 37 : spin_lock(&sysctl_lock);
1396 37 : if (insert_header(dir, header))
1397 : goto fail_put_dir_locked;
1398 :
1399 37 : drop_sysctl_table(&dir->header);
1400 37 : spin_unlock(&sysctl_lock);
1401 :
1402 37 : return header;
1403 :
1404 : fail_put_dir_locked:
1405 0 : drop_sysctl_table(&dir->header);
1406 : spin_unlock(&sysctl_lock);
1407 : fail:
1408 0 : kfree(header);
1409 0 : dump_stack();
1410 0 : return NULL;
1411 : }
1412 :
1413 : /**
1414 : * register_sysctl - register a sysctl table
1415 : * @path: The path to the directory the sysctl table is in. If the path
1416 : * doesn't exist we will create it for you.
1417 : * @table: the table structure. The calller must ensure the life of the @table
1418 : * will be kept during the lifetime use of the syctl. It must not be freed
1419 : * until unregister_sysctl_table() is called with the given returned table
1420 : * with this registration. If your code is non modular then you don't need
1421 : * to call unregister_sysctl_table() and can instead use something like
1422 : * register_sysctl_init() which does not care for the result of the syctl
1423 : * registration.
1424 : *
1425 : * Register a sysctl table. @table should be a filled in ctl_table
1426 : * array. A completely 0 filled entry terminates the table.
1427 : *
1428 : * See __register_sysctl_table for more details.
1429 : */
1430 5 : struct ctl_table_header *register_sysctl(const char *path, struct ctl_table *table)
1431 : {
1432 30 : return __register_sysctl_table(&sysctl_table_root.default_set,
1433 : path, table);
1434 : }
1435 : EXPORT_SYMBOL(register_sysctl);
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 : *
1447 : * The sysctl interface is used by userspace to query or modify at runtime
1448 : * a predefined value set on a variable. These variables however have default
1449 : * values pre-set. Code which depends on these variables will always work even
1450 : * if register_sysctl() fails. If register_sysctl() fails you'd just loose the
1451 : * ability to query or modify the sysctls dynamically at run time. Chances of
1452 : * register_sysctl() failing on init are extremely low, and so for both reasons
1453 : * this function does not return any error as it is used by initialization code.
1454 : *
1455 : * Context: if your base directory does not exist it will be created for you.
1456 : */
1457 25 : void __init __register_sysctl_init(const char *path, struct ctl_table *table,
1458 : const char *table_name)
1459 : {
1460 25 : struct ctl_table_header *hdr = register_sysctl(path, table);
1461 :
1462 25 : if (unlikely(!hdr)) {
1463 0 : pr_err("failed when register_sysctl %s to %s\n", table_name, path);
1464 0 : return;
1465 : }
1466 : kmemleak_not_leak(hdr);
1467 : }
1468 :
1469 6 : static char *append_path(const char *path, char *pos, const char *name)
1470 : {
1471 : int namelen;
1472 6 : namelen = strlen(name);
1473 6 : if (((pos - path) + namelen + 2) >= PATH_MAX)
1474 : return NULL;
1475 6 : memcpy(pos, name, namelen);
1476 6 : pos[namelen] = '/';
1477 6 : pos[namelen + 1] = '\0';
1478 6 : pos += namelen + 1;
1479 6 : return pos;
1480 : }
1481 :
1482 11 : static int count_subheaders(struct ctl_table *table)
1483 : {
1484 11 : int has_files = 0;
1485 11 : int nr_subheaders = 0;
1486 : struct ctl_table *entry;
1487 :
1488 : /* special case: no directory and empty directory */
1489 11 : if (!table || !table->procname)
1490 : return 1;
1491 :
1492 44 : list_for_each_table_entry(entry, table) {
1493 44 : if (entry->child)
1494 6 : nr_subheaders += count_subheaders(entry->child);
1495 : else
1496 : has_files = 1;
1497 : }
1498 9 : return nr_subheaders + has_files;
1499 : }
1500 :
1501 2 : static int register_leaf_sysctl_tables(const char *path, char *pos,
1502 : struct ctl_table_header ***subheader, struct ctl_table_set *set,
1503 : struct ctl_table *table)
1504 : {
1505 2 : struct ctl_table *ctl_table_arg = NULL;
1506 : struct ctl_table *entry, *files;
1507 2 : int nr_files = 0;
1508 2 : int nr_dirs = 0;
1509 2 : int err = -ENOMEM;
1510 :
1511 21 : list_for_each_table_entry(entry, table) {
1512 19 : if (entry->child)
1513 1 : nr_dirs++;
1514 : else
1515 18 : nr_files++;
1516 : }
1517 :
1518 2 : files = table;
1519 : /* If there are mixed files and directories we need a new table */
1520 2 : if (nr_dirs && nr_files) {
1521 : struct ctl_table *new;
1522 2 : files = kcalloc(nr_files + 1, sizeof(struct ctl_table),
1523 : GFP_KERNEL);
1524 1 : if (!files)
1525 : goto out;
1526 :
1527 : ctl_table_arg = files;
1528 : new = files;
1529 :
1530 17 : list_for_each_table_entry(entry, table) {
1531 17 : if (entry->child)
1532 1 : continue;
1533 16 : *new = *entry;
1534 16 : new++;
1535 : }
1536 : }
1537 :
1538 : /* Register everything except a directory full of subdirectories */
1539 2 : if (nr_files || !nr_dirs) {
1540 : struct ctl_table_header *header;
1541 2 : header = __register_sysctl_table(set, path, files);
1542 2 : if (!header) {
1543 0 : kfree(ctl_table_arg);
1544 0 : goto out;
1545 : }
1546 :
1547 : /* Remember if we need to free the file table */
1548 2 : header->ctl_table_arg = ctl_table_arg;
1549 2 : **subheader = header;
1550 2 : (*subheader)++;
1551 : }
1552 :
1553 : /* Recurse into the subdirectories. */
1554 19 : list_for_each_table_entry(entry, table) {
1555 : char *child_pos;
1556 :
1557 19 : if (!entry->child)
1558 18 : continue;
1559 :
1560 1 : err = -ENAMETOOLONG;
1561 1 : child_pos = append_path(path, pos, entry->procname);
1562 1 : if (!child_pos)
1563 : goto out;
1564 :
1565 1 : err = register_leaf_sysctl_tables(path, child_pos, subheader,
1566 : set, entry->child);
1567 1 : pos[0] = '\0';
1568 1 : if (err)
1569 : goto out;
1570 : }
1571 : err = 0;
1572 : out:
1573 : /* On failure our caller will unregister all registered subheaders */
1574 2 : return err;
1575 : }
1576 :
1577 : /**
1578 : * register_sysctl_table - register a sysctl table hierarchy
1579 : * @table: the top-level table structure
1580 : *
1581 : * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1582 : * array. A completely 0 filled entry terminates the table.
1583 : * We are slowly deprecating this call so avoid its use.
1584 : */
1585 5 : struct ctl_table_header *register_sysctl_table(struct ctl_table *table)
1586 : {
1587 5 : struct ctl_table *ctl_table_arg = table;
1588 5 : int nr_subheaders = count_subheaders(table);
1589 5 : struct ctl_table_header *header = NULL, **subheaders, **subheader;
1590 : char *new_path, *pos;
1591 :
1592 5 : pos = new_path = kmalloc(PATH_MAX, GFP_KERNEL);
1593 5 : if (!new_path)
1594 : return NULL;
1595 :
1596 5 : pos[0] = '\0';
1597 15 : while (table->procname && table->child && !table[1].procname) {
1598 5 : pos = append_path(new_path, pos, table->procname);
1599 5 : if (!pos)
1600 : goto out;
1601 5 : table = table->child;
1602 : }
1603 5 : if (nr_subheaders == 1) {
1604 4 : header = __register_sysctl_table(&sysctl_table_root.default_set, new_path, table);
1605 4 : if (header)
1606 4 : header->ctl_table_arg = ctl_table_arg;
1607 : } else {
1608 1 : header = kzalloc(sizeof(*header) +
1609 : sizeof(*subheaders)*nr_subheaders, GFP_KERNEL);
1610 1 : if (!header)
1611 : goto out;
1612 :
1613 1 : subheaders = (struct ctl_table_header **) (header + 1);
1614 1 : subheader = subheaders;
1615 1 : header->ctl_table_arg = ctl_table_arg;
1616 :
1617 1 : if (register_leaf_sysctl_tables(new_path, pos, &subheader,
1618 : &sysctl_table_root.default_set, table))
1619 : goto err_register_leaves;
1620 : }
1621 :
1622 : out:
1623 5 : kfree(new_path);
1624 5 : return header;
1625 :
1626 : err_register_leaves:
1627 0 : while (subheader > subheaders) {
1628 0 : struct ctl_table_header *subh = *(--subheader);
1629 0 : struct ctl_table *table = subh->ctl_table_arg;
1630 0 : unregister_sysctl_table(subh);
1631 0 : kfree(table);
1632 : }
1633 0 : kfree(header);
1634 0 : header = NULL;
1635 0 : goto out;
1636 : }
1637 : EXPORT_SYMBOL(register_sysctl_table);
1638 :
1639 5 : int __register_sysctl_base(struct ctl_table *base_table)
1640 : {
1641 : struct ctl_table_header *hdr;
1642 :
1643 5 : hdr = register_sysctl_table(base_table);
1644 5 : kmemleak_not_leak(hdr);
1645 5 : return 0;
1646 : }
1647 :
1648 0 : static void put_links(struct ctl_table_header *header)
1649 : {
1650 0 : struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1651 0 : struct ctl_table_root *root = header->root;
1652 0 : struct ctl_dir *parent = header->parent;
1653 : struct ctl_dir *core_parent;
1654 : struct ctl_table *entry;
1655 :
1656 0 : if (header->set == root_set)
1657 : return;
1658 :
1659 0 : core_parent = xlate_dir(root_set, parent);
1660 0 : if (IS_ERR(core_parent))
1661 : return;
1662 :
1663 0 : list_for_each_table_entry(entry, header->ctl_table) {
1664 : struct ctl_table_header *link_head;
1665 : struct ctl_table *link;
1666 0 : const char *name = entry->procname;
1667 :
1668 0 : link = find_entry(&link_head, core_parent, name, strlen(name));
1669 0 : if (link &&
1670 0 : ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1671 0 : (S_ISLNK(link->mode) && (link->data == root)))) {
1672 0 : drop_sysctl_table(link_head);
1673 : }
1674 : else {
1675 0 : pr_err("sysctl link missing during unregister: ");
1676 0 : sysctl_print_dir(parent);
1677 0 : pr_cont("%s\n", name);
1678 : }
1679 : }
1680 : }
1681 :
1682 96 : static void drop_sysctl_table(struct ctl_table_header *header)
1683 : {
1684 96 : struct ctl_dir *parent = header->parent;
1685 :
1686 96 : if (--header->nreg)
1687 : return;
1688 :
1689 0 : if (parent) {
1690 0 : put_links(header);
1691 0 : start_unregistering(header);
1692 : }
1693 :
1694 0 : if (!--header->count)
1695 0 : kfree_rcu(header, rcu);
1696 :
1697 0 : if (parent)
1698 0 : drop_sysctl_table(&parent->header);
1699 : }
1700 :
1701 : /**
1702 : * unregister_sysctl_table - unregister a sysctl table hierarchy
1703 : * @header: the header returned from register_sysctl_table
1704 : *
1705 : * Unregisters the sysctl table and all children. proc entries may not
1706 : * actually be removed until they are no longer used by anyone.
1707 : */
1708 0 : void unregister_sysctl_table(struct ctl_table_header * header)
1709 : {
1710 : int nr_subheaders;
1711 : might_sleep();
1712 :
1713 0 : if (header == NULL)
1714 : return;
1715 :
1716 0 : nr_subheaders = count_subheaders(header->ctl_table_arg);
1717 0 : if (unlikely(nr_subheaders > 1)) {
1718 : struct ctl_table_header **subheaders;
1719 : int i;
1720 :
1721 0 : subheaders = (struct ctl_table_header **)(header + 1);
1722 0 : for (i = nr_subheaders -1; i >= 0; i--) {
1723 0 : struct ctl_table_header *subh = subheaders[i];
1724 0 : struct ctl_table *table = subh->ctl_table_arg;
1725 0 : unregister_sysctl_table(subh);
1726 0 : kfree(table);
1727 : }
1728 0 : kfree(header);
1729 0 : return;
1730 : }
1731 :
1732 0 : spin_lock(&sysctl_lock);
1733 0 : drop_sysctl_table(header);
1734 : spin_unlock(&sysctl_lock);
1735 : }
1736 : EXPORT_SYMBOL(unregister_sysctl_table);
1737 :
1738 1 : void setup_sysctl_set(struct ctl_table_set *set,
1739 : struct ctl_table_root *root,
1740 : int (*is_seen)(struct ctl_table_set *))
1741 : {
1742 1 : memset(set, 0, sizeof(*set));
1743 1 : set->is_seen = is_seen;
1744 2 : init_header(&set->dir.header, root, set, NULL, root_table);
1745 1 : }
1746 :
1747 0 : void retire_sysctl_set(struct ctl_table_set *set)
1748 : {
1749 0 : WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1750 0 : }
1751 :
1752 1 : int __init proc_sys_init(void)
1753 : {
1754 : struct proc_dir_entry *proc_sys_root;
1755 :
1756 1 : proc_sys_root = proc_mkdir("sys", NULL);
1757 1 : proc_sys_root->proc_iops = &proc_sys_dir_operations;
1758 1 : proc_sys_root->proc_dir_ops = &proc_sys_dir_file_operations;
1759 1 : proc_sys_root->nlink = 0;
1760 :
1761 1 : return sysctl_init_bases();
1762 : }
1763 :
1764 : struct sysctl_alias {
1765 : const char *kernel_param;
1766 : const char *sysctl_param;
1767 : };
1768 :
1769 : /*
1770 : * Historically some settings had both sysctl and a command line parameter.
1771 : * With the generic sysctl. parameter support, we can handle them at a single
1772 : * place and only keep the historical name for compatibility. This is not meant
1773 : * to add brand new aliases. When adding existing aliases, consider whether
1774 : * the possibly different moment of changing the value (e.g. from early_param
1775 : * to the moment do_sysctl_args() is called) is an issue for the specific
1776 : * parameter.
1777 : */
1778 : static const struct sysctl_alias sysctl_aliases[] = {
1779 : {"hardlockup_all_cpu_backtrace", "kernel.hardlockup_all_cpu_backtrace" },
1780 : {"hung_task_panic", "kernel.hung_task_panic" },
1781 : {"numa_zonelist_order", "vm.numa_zonelist_order" },
1782 : {"softlockup_all_cpu_backtrace", "kernel.softlockup_all_cpu_backtrace" },
1783 : {"softlockup_panic", "kernel.softlockup_panic" },
1784 : { }
1785 : };
1786 :
1787 0 : static const char *sysctl_find_alias(char *param)
1788 : {
1789 : const struct sysctl_alias *alias;
1790 :
1791 0 : for (alias = &sysctl_aliases[0]; alias->kernel_param != NULL; alias++) {
1792 0 : if (strcmp(alias->kernel_param, param) == 0)
1793 0 : return alias->sysctl_param;
1794 : }
1795 :
1796 : return NULL;
1797 : }
1798 :
1799 : /* Set sysctl value passed on kernel command line. */
1800 0 : static int process_sysctl_arg(char *param, char *val,
1801 : const char *unused, void *arg)
1802 : {
1803 : char *path;
1804 0 : struct vfsmount **proc_mnt = arg;
1805 : struct file_system_type *proc_fs_type;
1806 : struct file *file;
1807 : int len;
1808 : int err;
1809 0 : loff_t pos = 0;
1810 : ssize_t wret;
1811 :
1812 0 : if (strncmp(param, "sysctl", sizeof("sysctl") - 1) == 0) {
1813 0 : param += sizeof("sysctl") - 1;
1814 :
1815 0 : if (param[0] != '/' && param[0] != '.')
1816 : return 0;
1817 :
1818 0 : param++;
1819 : } else {
1820 0 : param = (char *) sysctl_find_alias(param);
1821 0 : if (!param)
1822 : return 0;
1823 : }
1824 :
1825 0 : if (!val)
1826 : return -EINVAL;
1827 0 : len = strlen(val);
1828 0 : if (len == 0)
1829 : return -EINVAL;
1830 :
1831 : /*
1832 : * To set sysctl options, we use a temporary mount of proc, look up the
1833 : * respective sys/ file and write to it. To avoid mounting it when no
1834 : * options were given, we mount it only when the first sysctl option is
1835 : * found. Why not a persistent mount? There are problems with a
1836 : * persistent mount of proc in that it forces userspace not to use any
1837 : * proc mount options.
1838 : */
1839 0 : if (!*proc_mnt) {
1840 0 : proc_fs_type = get_fs_type("proc");
1841 0 : if (!proc_fs_type) {
1842 0 : pr_err("Failed to find procfs to set sysctl from command line\n");
1843 0 : return 0;
1844 : }
1845 0 : *proc_mnt = kern_mount(proc_fs_type);
1846 0 : put_filesystem(proc_fs_type);
1847 0 : if (IS_ERR(*proc_mnt)) {
1848 0 : pr_err("Failed to mount procfs to set sysctl from command line\n");
1849 0 : return 0;
1850 : }
1851 : }
1852 :
1853 0 : path = kasprintf(GFP_KERNEL, "sys/%s", param);
1854 0 : if (!path)
1855 0 : panic("%s: Failed to allocate path for %s\n", __func__, param);
1856 0 : strreplace(path, '.', '/');
1857 :
1858 0 : file = file_open_root_mnt(*proc_mnt, path, O_WRONLY, 0);
1859 0 : if (IS_ERR(file)) {
1860 0 : err = PTR_ERR(file);
1861 0 : if (err == -ENOENT)
1862 0 : pr_err("Failed to set sysctl parameter '%s=%s': parameter not found\n",
1863 : param, val);
1864 0 : else if (err == -EACCES)
1865 0 : pr_err("Failed to set sysctl parameter '%s=%s': permission denied (read-only?)\n",
1866 : param, val);
1867 : else
1868 0 : pr_err("Error %pe opening proc file to set sysctl parameter '%s=%s'\n",
1869 : file, param, val);
1870 : goto out;
1871 : }
1872 0 : wret = kernel_write(file, val, len, &pos);
1873 0 : if (wret < 0) {
1874 0 : err = wret;
1875 0 : if (err == -EINVAL)
1876 0 : pr_err("Failed to set sysctl parameter '%s=%s': invalid value\n",
1877 : param, val);
1878 : else
1879 0 : pr_err("Error %pe writing to proc file to set sysctl parameter '%s=%s'\n",
1880 : ERR_PTR(err), param, val);
1881 0 : } else if (wret != len) {
1882 0 : pr_err("Wrote only %zd bytes of %d writing to proc file %s to set sysctl parameter '%s=%s\n",
1883 : wret, len, path, param, val);
1884 : }
1885 :
1886 0 : err = filp_close(file, NULL);
1887 0 : if (err)
1888 0 : pr_err("Error %pe closing proc file to set sysctl parameter '%s=%s\n",
1889 : ERR_PTR(err), param, val);
1890 : out:
1891 0 : kfree(path);
1892 0 : return 0;
1893 : }
1894 :
1895 0 : void do_sysctl_args(void)
1896 : {
1897 : char *command_line;
1898 0 : struct vfsmount *proc_mnt = NULL;
1899 :
1900 0 : command_line = kstrdup(saved_command_line, GFP_KERNEL);
1901 0 : if (!command_line)
1902 0 : panic("%s: Failed to allocate copy of command line\n", __func__);
1903 :
1904 0 : parse_args("Setting sysctl args", command_line,
1905 : NULL, 0, -1, -1, &proc_mnt, process_sysctl_arg);
1906 :
1907 0 : if (proc_mnt)
1908 0 : kern_unmount(proc_mnt);
1909 :
1910 0 : kfree(command_line);
1911 0 : }
|
