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 : static const struct dentry_operations proc_sys_dentry_operations;
23 : static const struct file_operations proc_sys_file_operations;
24 : static const struct inode_operations proc_sys_inode_operations;
25 : static const struct file_operations proc_sys_dir_file_operations;
26 : static const struct inode_operations proc_sys_dir_operations;
27 :
28 : /* shared constants to be used in various sysctls */
29 : const int sysctl_vals[] = { -1, 0, 1, 2, 4, 100, 200, 1000, 3000, INT_MAX, 65535 };
30 : EXPORT_SYMBOL(sysctl_vals);
31 :
32 : const unsigned long sysctl_long_vals[] = { 0, 1, LONG_MAX };
33 : EXPORT_SYMBOL_GPL(sysctl_long_vals);
34 :
35 : /* Support for permanently empty directories */
36 :
37 : 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 : */
49 0 : struct ctl_table_header *register_sysctl_mount_point(const char *path)
50 : {
51 0 : return register_sysctl(path, sysctl_mount_point);
52 : }
53 : EXPORT_SYMBOL(register_sysctl_mount_point);
54 :
55 : static bool is_empty_dir(struct ctl_table_header *head)
56 : {
57 41 : return head->ctl_table[0].child == sysctl_mount_point;
58 : }
59 :
60 : static void set_empty_dir(struct ctl_dir *dir)
61 : {
62 0 : dir->header.ctl_table[0].child = sysctl_mount_point;
63 : }
64 :
65 : static void clear_empty_dir(struct ctl_dir *dir)
66 :
67 : {
68 0 : dir->header.ctl_table[0].child = NULL;
69 : }
70 :
71 0 : void proc_sys_poll_notify(struct ctl_table_poll *poll)
72 : {
73 0 : if (!poll)
74 : return;
75 :
76 0 : atomic_inc(&poll->event);
77 0 : wake_up_interruptible(&poll->wait);
78 : }
79 :
80 : static struct ctl_table root_table[] = {
81 : {
82 : .procname = "",
83 : .mode = S_IFDIR|S_IRUGO|S_IXUGO,
84 : },
85 : { }
86 : };
87 : static struct ctl_table_root sysctl_table_root = {
88 : .default_set.dir.header = {
89 : {{.count = 1,
90 : .nreg = 1,
91 : .ctl_table = root_table }},
92 : .ctl_table_arg = root_table,
93 : .root = &sysctl_table_root,
94 : .set = &sysctl_table_root.default_set,
95 : },
96 : };
97 :
98 : static DEFINE_SPINLOCK(sysctl_lock);
99 :
100 : static void drop_sysctl_table(struct ctl_table_header *header);
101 : static int sysctl_follow_link(struct ctl_table_header **phead,
102 : struct ctl_table **pentry);
103 : static int insert_links(struct ctl_table_header *head);
104 : static void put_links(struct ctl_table_header *header);
105 :
106 0 : static void sysctl_print_dir(struct ctl_dir *dir)
107 : {
108 0 : if (dir->header.parent)
109 0 : sysctl_print_dir(dir->header.parent);
110 0 : pr_cont("%s/", dir->header.ctl_table[0].procname);
111 0 : }
112 :
113 : static int namecmp(const char *name1, int len1, const char *name2, int len2)
114 : {
115 : int cmp;
116 :
117 595 : cmp = memcmp(name1, name2, min(len1, len2));
118 595 : if (cmp == 0)
119 31 : cmp = len1 - len2;
120 : return cmp;
121 : }
122 :
123 : /* Called under sysctl_lock */
124 52 : static struct ctl_table *find_entry(struct ctl_table_header **phead,
125 : struct ctl_dir *dir, const char *name, int namelen)
126 : {
127 : struct ctl_table_header *head;
128 : struct ctl_table *entry;
129 52 : struct rb_node *node = dir->root.rb_node;
130 :
131 157 : while (node)
132 : {
133 : struct ctl_node *ctl_node;
134 : const char *procname;
135 : int cmp;
136 :
137 127 : ctl_node = rb_entry(node, struct ctl_node, node);
138 127 : head = ctl_node->header;
139 127 : entry = &head->ctl_table[ctl_node - head->node];
140 127 : procname = entry->procname;
141 :
142 254 : cmp = namecmp(name, namelen, procname, strlen(procname));
143 127 : if (cmp < 0)
144 47 : node = node->rb_left;
145 80 : else if (cmp > 0)
146 58 : node = node->rb_right;
147 : else {
148 22 : *phead = head;
149 : return entry;
150 : }
151 : }
152 : return NULL;
153 : }
154 :
155 143 : static int insert_entry(struct ctl_table_header *head, struct ctl_table *entry)
156 : {
157 143 : struct rb_node *node = &head->node[entry - head->ctl_table].node;
158 143 : struct rb_node **p = &head->parent->root.rb_node;
159 143 : struct rb_node *parent = NULL;
160 143 : const char *name = entry->procname;
161 143 : int namelen = strlen(name);
162 :
163 754 : while (*p) {
164 : struct ctl_table_header *parent_head;
165 : struct ctl_table *parent_entry;
166 : struct ctl_node *parent_node;
167 : const char *parent_name;
168 : int cmp;
169 :
170 468 : parent = *p;
171 468 : parent_node = rb_entry(parent, struct ctl_node, node);
172 468 : parent_head = parent_node->header;
173 468 : parent_entry = &parent_head->ctl_table[parent_node - parent_head->node];
174 468 : parent_name = parent_entry->procname;
175 :
176 936 : cmp = namecmp(name, namelen, parent_name, strlen(parent_name));
177 468 : if (cmp < 0)
178 255 : p = &(*p)->rb_left;
179 213 : else if (cmp > 0)
180 213 : p = &(*p)->rb_right;
181 : else {
182 0 : pr_err("sysctl duplicate entry: ");
183 0 : sysctl_print_dir(head->parent);
184 0 : pr_cont("%s\n", entry->procname);
185 0 : return -EEXIST;
186 : }
187 : }
188 :
189 143 : rb_link_node(node, parent, p);
190 143 : rb_insert_color(node, &head->parent->root);
191 143 : return 0;
192 : }
193 :
194 : static void erase_entry(struct ctl_table_header *head, struct ctl_table *entry)
195 : {
196 0 : struct rb_node *node = &head->node[entry - head->ctl_table].node;
197 :
198 0 : rb_erase(node, &head->parent->root);
199 : }
200 :
201 : static void init_header(struct ctl_table_header *head,
202 : struct ctl_table_root *root, struct ctl_table_set *set,
203 : struct ctl_node *node, struct ctl_table *table)
204 : {
205 42 : head->ctl_table = table;
206 42 : head->ctl_table_arg = table;
207 42 : head->used = 0;
208 42 : head->count = 1;
209 42 : head->nreg = 1;
210 42 : head->unregistering = NULL;
211 42 : head->root = root;
212 42 : head->set = set;
213 42 : head->parent = NULL;
214 42 : head->node = node;
215 42 : INIT_HLIST_HEAD(&head->inodes);
216 41 : if (node) {
217 : struct ctl_table *entry;
218 143 : for (entry = table; entry->procname; entry++, node++)
219 143 : node->header = head;
220 : }
221 : }
222 :
223 0 : static void erase_header(struct ctl_table_header *head)
224 : {
225 : struct ctl_table *entry;
226 0 : for (entry = head->ctl_table; entry->procname; entry++)
227 0 : erase_entry(head, entry);
228 0 : }
229 :
230 41 : 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 82 : if (is_empty_dir(&dir->header))
237 : return -EROFS;
238 :
239 : /* Am I creating a permanently empty directory? */
240 41 : 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 41 : dir->header.nreg++;
247 41 : header->parent = dir;
248 41 : err = insert_links(header);
249 41 : if (err)
250 : goto fail_links;
251 184 : for (entry = header->ctl_table; entry->procname; entry++) {
252 143 : err = insert_entry(header, entry);
253 143 : 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 user_namespace *mnt_userns,
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 user_namespace *mnt_userns,
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(&init_user_ns, dentry, attr);
840 0 : if (error)
841 : return error;
842 :
843 0 : setattr_copy(&init_user_ns, inode, attr);
844 0 : mark_inode_dirty(inode);
845 0 : return 0;
846 : }
847 :
848 0 : static int proc_sys_getattr(struct user_namespace *mnt_userns,
849 : const struct path *path, struct kstat *stat,
850 : u32 request_mask, unsigned int query_flags)
851 : {
852 0 : struct inode *inode = d_inode(path->dentry);
853 0 : struct ctl_table_header *head = grab_header(inode);
854 0 : struct ctl_table *table = PROC_I(inode)->sysctl_entry;
855 :
856 0 : if (IS_ERR(head))
857 0 : return PTR_ERR(head);
858 :
859 0 : generic_fillattr(&init_user_ns, inode, stat);
860 0 : if (table)
861 0 : stat->mode = (stat->mode & S_IFMT) | table->mode;
862 :
863 0 : sysctl_head_finish(head);
864 0 : return 0;
865 : }
866 :
867 : static const struct file_operations proc_sys_file_operations = {
868 : .open = proc_sys_open,
869 : .poll = proc_sys_poll,
870 : .read_iter = proc_sys_read,
871 : .write_iter = proc_sys_write,
872 : .splice_read = generic_file_splice_read,
873 : .splice_write = iter_file_splice_write,
874 : .llseek = default_llseek,
875 : };
876 :
877 : static const struct file_operations proc_sys_dir_file_operations = {
878 : .read = generic_read_dir,
879 : .iterate_shared = proc_sys_readdir,
880 : .llseek = generic_file_llseek,
881 : };
882 :
883 : static const struct inode_operations proc_sys_inode_operations = {
884 : .permission = proc_sys_permission,
885 : .setattr = proc_sys_setattr,
886 : .getattr = proc_sys_getattr,
887 : };
888 :
889 : static const struct inode_operations proc_sys_dir_operations = {
890 : .lookup = proc_sys_lookup,
891 : .permission = proc_sys_permission,
892 : .setattr = proc_sys_setattr,
893 : .getattr = proc_sys_getattr,
894 : };
895 :
896 0 : static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
897 : {
898 0 : if (flags & LOOKUP_RCU)
899 : return -ECHILD;
900 0 : return !PROC_I(d_inode(dentry))->sysctl->unregistering;
901 : }
902 :
903 0 : static int proc_sys_delete(const struct dentry *dentry)
904 : {
905 0 : return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
906 : }
907 :
908 : static int sysctl_is_seen(struct ctl_table_header *p)
909 : {
910 0 : struct ctl_table_set *set = p->set;
911 : int res;
912 0 : spin_lock(&sysctl_lock);
913 0 : if (p->unregistering)
914 : res = 0;
915 0 : else if (!set->is_seen)
916 : res = 1;
917 : else
918 0 : res = set->is_seen(set);
919 0 : spin_unlock(&sysctl_lock);
920 : return res;
921 : }
922 :
923 0 : static int proc_sys_compare(const struct dentry *dentry,
924 : unsigned int len, const char *str, const struct qstr *name)
925 : {
926 : struct ctl_table_header *head;
927 : struct inode *inode;
928 :
929 : /* Although proc doesn't have negative dentries, rcu-walk means
930 : * that inode here can be NULL */
931 : /* AV: can it, indeed? */
932 0 : inode = d_inode_rcu(dentry);
933 0 : if (!inode)
934 : return 1;
935 0 : if (name->len != len)
936 : return 1;
937 0 : if (memcmp(name->name, str, len))
938 : return 1;
939 0 : head = rcu_dereference(PROC_I(inode)->sysctl);
940 0 : return !head || !sysctl_is_seen(head);
941 : }
942 :
943 : static const struct dentry_operations proc_sys_dentry_operations = {
944 : .d_revalidate = proc_sys_revalidate,
945 : .d_delete = proc_sys_delete,
946 : .d_compare = proc_sys_compare,
947 : };
948 :
949 48 : static struct ctl_dir *find_subdir(struct ctl_dir *dir,
950 : const char *name, int namelen)
951 : {
952 : struct ctl_table_header *head;
953 : struct ctl_table *entry;
954 :
955 48 : entry = find_entry(&head, dir, name, namelen);
956 48 : if (!entry)
957 : return ERR_PTR(-ENOENT);
958 20 : if (!S_ISDIR(entry->mode))
959 : return ERR_PTR(-ENOTDIR);
960 20 : return container_of(head, struct ctl_dir, header);
961 : }
962 :
963 14 : static struct ctl_dir *new_dir(struct ctl_table_set *set,
964 : const char *name, int namelen)
965 : {
966 : struct ctl_table *table;
967 : struct ctl_dir *new;
968 : struct ctl_node *node;
969 : char *new_name;
970 :
971 14 : new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
972 : sizeof(struct ctl_table)*2 + namelen + 1,
973 : GFP_KERNEL);
974 14 : if (!new)
975 : return NULL;
976 :
977 14 : node = (struct ctl_node *)(new + 1);
978 14 : table = (struct ctl_table *)(node + 1);
979 14 : new_name = (char *)(table + 2);
980 14 : memcpy(new_name, name, namelen);
981 14 : new_name[namelen] = '\0';
982 14 : table[0].procname = new_name;
983 14 : table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
984 14 : init_header(&new->header, set->dir.header.root, set, node, table);
985 :
986 : return new;
987 : }
988 :
989 : /**
990 : * get_subdir - find or create a subdir with the specified name.
991 : * @dir: Directory to create the subdirectory in
992 : * @name: The name of the subdirectory to find or create
993 : * @namelen: The length of name
994 : *
995 : * Takes a directory with an elevated reference count so we know that
996 : * if we drop the lock the directory will not go away. Upon success
997 : * the reference is moved from @dir to the returned subdirectory.
998 : * Upon error an error code is returned and the reference on @dir is
999 : * simply dropped.
1000 : */
1001 33 : static struct ctl_dir *get_subdir(struct ctl_dir *dir,
1002 : const char *name, int namelen)
1003 : {
1004 33 : struct ctl_table_set *set = dir->header.set;
1005 33 : struct ctl_dir *subdir, *new = NULL;
1006 : int err;
1007 :
1008 33 : spin_lock(&sysctl_lock);
1009 33 : subdir = find_subdir(dir, name, namelen);
1010 33 : if (!IS_ERR(subdir))
1011 : goto found;
1012 14 : if (PTR_ERR(subdir) != -ENOENT)
1013 : goto failed;
1014 :
1015 14 : spin_unlock(&sysctl_lock);
1016 14 : new = new_dir(set, name, namelen);
1017 14 : spin_lock(&sysctl_lock);
1018 14 : subdir = ERR_PTR(-ENOMEM);
1019 14 : if (!new)
1020 : goto failed;
1021 :
1022 : /* Was the subdir added while we dropped the lock? */
1023 14 : subdir = find_subdir(dir, name, namelen);
1024 14 : if (!IS_ERR(subdir))
1025 : goto found;
1026 14 : if (PTR_ERR(subdir) != -ENOENT)
1027 : goto failed;
1028 :
1029 : /* Nope. Use the our freshly made directory entry. */
1030 14 : err = insert_header(dir, &new->header);
1031 28 : subdir = ERR_PTR(err);
1032 14 : if (err)
1033 : goto failed;
1034 : subdir = new;
1035 : found:
1036 33 : subdir->header.nreg++;
1037 : failed:
1038 33 : if (IS_ERR(subdir)) {
1039 0 : pr_err("sysctl could not get directory: ");
1040 0 : sysctl_print_dir(dir);
1041 0 : pr_cont("%*.*s %ld\n", namelen, namelen, name,
1042 : PTR_ERR(subdir));
1043 : }
1044 33 : drop_sysctl_table(&dir->header);
1045 33 : if (new)
1046 14 : drop_sysctl_table(&new->header);
1047 33 : spin_unlock(&sysctl_lock);
1048 33 : return subdir;
1049 : }
1050 :
1051 3 : static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
1052 : {
1053 : struct ctl_dir *parent;
1054 : const char *procname;
1055 3 : if (!dir->header.parent)
1056 2 : return &set->dir;
1057 1 : parent = xlate_dir(set, dir->header.parent);
1058 1 : if (IS_ERR(parent))
1059 : return parent;
1060 1 : procname = dir->header.ctl_table[0].procname;
1061 1 : return find_subdir(parent, procname, strlen(procname));
1062 : }
1063 :
1064 0 : static int sysctl_follow_link(struct ctl_table_header **phead,
1065 : struct ctl_table **pentry)
1066 : {
1067 : struct ctl_table_header *head;
1068 : struct ctl_table_root *root;
1069 : struct ctl_table_set *set;
1070 : struct ctl_table *entry;
1071 : struct ctl_dir *dir;
1072 : int ret;
1073 :
1074 0 : spin_lock(&sysctl_lock);
1075 0 : root = (*pentry)->data;
1076 0 : set = lookup_header_set(root);
1077 0 : dir = xlate_dir(set, (*phead)->parent);
1078 0 : if (IS_ERR(dir))
1079 0 : ret = PTR_ERR(dir);
1080 : else {
1081 0 : const char *procname = (*pentry)->procname;
1082 0 : head = NULL;
1083 0 : entry = find_entry(&head, dir, procname, strlen(procname));
1084 0 : ret = -ENOENT;
1085 0 : if (entry && use_table(head)) {
1086 0 : unuse_table(*phead);
1087 0 : *phead = head;
1088 0 : *pentry = entry;
1089 0 : ret = 0;
1090 : }
1091 : }
1092 :
1093 0 : spin_unlock(&sysctl_lock);
1094 0 : return ret;
1095 : }
1096 :
1097 0 : static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
1098 : {
1099 : struct va_format vaf;
1100 : va_list args;
1101 :
1102 0 : va_start(args, fmt);
1103 0 : vaf.fmt = fmt;
1104 0 : vaf.va = &args;
1105 :
1106 0 : pr_err("sysctl table check failed: %s/%s %pV\n",
1107 : path, table->procname, &vaf);
1108 :
1109 0 : va_end(args);
1110 0 : return -EINVAL;
1111 : }
1112 :
1113 73 : static int sysctl_check_table_array(const char *path, struct ctl_table *table)
1114 : {
1115 73 : int err = 0;
1116 :
1117 73 : if ((table->proc_handler == proc_douintvec) ||
1118 : (table->proc_handler == proc_douintvec_minmax)) {
1119 0 : if (table->maxlen != sizeof(unsigned int))
1120 0 : err |= sysctl_err(path, table, "array not allowed");
1121 : }
1122 :
1123 73 : if (table->proc_handler == proc_dou8vec_minmax) {
1124 0 : if (table->maxlen != sizeof(u8))
1125 0 : err |= sysctl_err(path, table, "array not allowed");
1126 : }
1127 :
1128 73 : return err;
1129 : }
1130 :
1131 26 : static int sysctl_check_table(const char *path, struct ctl_table *table)
1132 : {
1133 26 : int err = 0;
1134 145 : for (; table->procname; table++) {
1135 119 : if (table->child)
1136 0 : err |= sysctl_err(path, table, "Not a file");
1137 :
1138 119 : if ((table->proc_handler == proc_dostring) ||
1139 93 : (table->proc_handler == proc_dointvec) ||
1140 93 : (table->proc_handler == proc_douintvec) ||
1141 93 : (table->proc_handler == proc_douintvec_minmax) ||
1142 69 : (table->proc_handler == proc_dointvec_minmax) ||
1143 69 : (table->proc_handler == proc_dou8vec_minmax) ||
1144 67 : (table->proc_handler == proc_dointvec_jiffies) ||
1145 67 : (table->proc_handler == proc_dointvec_userhz_jiffies) ||
1146 67 : (table->proc_handler == proc_dointvec_ms_jiffies) ||
1147 46 : (table->proc_handler == proc_doulongvec_minmax) ||
1148 : (table->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1149 73 : if (!table->data)
1150 0 : err |= sysctl_err(path, table, "No data");
1151 73 : if (!table->maxlen)
1152 0 : err |= sysctl_err(path, table, "No maxlen");
1153 : else
1154 73 : err |= sysctl_check_table_array(path, table);
1155 : }
1156 119 : if (!table->proc_handler)
1157 0 : err |= sysctl_err(path, table, "No proc_handler");
1158 :
1159 119 : if ((table->mode & (S_IRUGO|S_IWUGO)) != table->mode)
1160 0 : err |= sysctl_err(path, table, "bogus .mode 0%o",
1161 : table->mode);
1162 : }
1163 26 : return err;
1164 : }
1165 :
1166 1 : static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table *table,
1167 : struct ctl_table_root *link_root)
1168 : {
1169 : struct ctl_table *link_table, *entry, *link;
1170 : struct ctl_table_header *links;
1171 : struct ctl_node *node;
1172 : char *link_name;
1173 : int nr_entries, name_bytes;
1174 :
1175 1 : name_bytes = 0;
1176 1 : nr_entries = 0;
1177 11 : for (entry = table; entry->procname; entry++) {
1178 10 : nr_entries++;
1179 10 : name_bytes += strlen(entry->procname) + 1;
1180 : }
1181 :
1182 1 : links = kzalloc(sizeof(struct ctl_table_header) +
1183 2 : sizeof(struct ctl_node)*nr_entries +
1184 2 : sizeof(struct ctl_table)*(nr_entries + 1) +
1185 : name_bytes,
1186 : GFP_KERNEL);
1187 :
1188 1 : if (!links)
1189 : return NULL;
1190 :
1191 1 : node = (struct ctl_node *)(links + 1);
1192 1 : link_table = (struct ctl_table *)(node + nr_entries);
1193 1 : link_name = (char *)&link_table[nr_entries + 1];
1194 :
1195 11 : for (link = link_table, entry = table; entry->procname; link++, entry++) {
1196 10 : int len = strlen(entry->procname) + 1;
1197 10 : memcpy(link_name, entry->procname, len);
1198 10 : link->procname = link_name;
1199 10 : link->mode = S_IFLNK|S_IRWXUGO;
1200 10 : link->data = link_root;
1201 10 : link_name += len;
1202 : }
1203 2 : init_header(links, dir->header.root, dir->header.set, node, link_table);
1204 1 : links->nreg = nr_entries;
1205 :
1206 : return links;
1207 : }
1208 :
1209 3 : static bool get_links(struct ctl_dir *dir,
1210 : struct ctl_table *table, struct ctl_table_root *link_root)
1211 : {
1212 : struct ctl_table_header *head;
1213 : struct ctl_table *entry, *link;
1214 :
1215 : /* Are there links available for every entry in table? */
1216 4 : for (entry = table; entry->procname; entry++) {
1217 3 : const char *procname = entry->procname;
1218 3 : link = find_entry(&head, dir, procname, strlen(procname));
1219 3 : if (!link)
1220 : return false;
1221 1 : if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1222 1 : continue;
1223 0 : if (S_ISLNK(link->mode) && (link->data == link_root))
1224 0 : continue;
1225 : return false;
1226 : }
1227 :
1228 : /* The checks passed. Increase the registration count on the links */
1229 1 : for (entry = table; entry->procname; entry++) {
1230 1 : const char *procname = entry->procname;
1231 1 : link = find_entry(&head, dir, procname, strlen(procname));
1232 1 : head->nreg++;
1233 : }
1234 : return true;
1235 : }
1236 :
1237 41 : static int insert_links(struct ctl_table_header *head)
1238 : {
1239 41 : struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1240 41 : struct ctl_dir *core_parent = NULL;
1241 : struct ctl_table_header *links;
1242 : int err;
1243 :
1244 41 : if (head->set == root_set)
1245 : return 0;
1246 :
1247 2 : core_parent = xlate_dir(root_set, head->parent);
1248 2 : if (IS_ERR(core_parent))
1249 : return 0;
1250 :
1251 2 : if (get_links(core_parent, head->ctl_table, head->root))
1252 : return 0;
1253 :
1254 1 : core_parent->header.nreg++;
1255 1 : spin_unlock(&sysctl_lock);
1256 :
1257 1 : links = new_links(core_parent, head->ctl_table, head->root);
1258 :
1259 1 : spin_lock(&sysctl_lock);
1260 1 : err = -ENOMEM;
1261 1 : if (!links)
1262 : goto out;
1263 :
1264 1 : err = 0;
1265 1 : if (get_links(core_parent, head->ctl_table, head->root)) {
1266 0 : kfree(links);
1267 0 : goto out;
1268 : }
1269 :
1270 1 : err = insert_header(core_parent, links);
1271 1 : if (err)
1272 0 : kfree(links);
1273 : out:
1274 1 : drop_sysctl_table(&core_parent->header);
1275 1 : return err;
1276 : }
1277 :
1278 : /**
1279 : * __register_sysctl_table - register a leaf sysctl table
1280 : * @set: Sysctl tree to register on
1281 : * @path: The path to the directory the sysctl table is in.
1282 : * @table: the top-level table structure
1283 : *
1284 : * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1285 : * array. A completely 0 filled entry terminates the table.
1286 : *
1287 : * The members of the &struct ctl_table structure are used as follows:
1288 : *
1289 : * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1290 : * enter a sysctl file
1291 : *
1292 : * data - a pointer to data for use by proc_handler
1293 : *
1294 : * maxlen - the maximum size in bytes of the data
1295 : *
1296 : * mode - the file permissions for the /proc/sys file
1297 : *
1298 : * child - must be %NULL.
1299 : *
1300 : * proc_handler - the text handler routine (described below)
1301 : *
1302 : * extra1, extra2 - extra pointers usable by the proc handler routines
1303 : *
1304 : * Leaf nodes in the sysctl tree will be represented by a single file
1305 : * under /proc; non-leaf nodes will be represented by directories.
1306 : *
1307 : * There must be a proc_handler routine for any terminal nodes.
1308 : * Several default handlers are available to cover common cases -
1309 : *
1310 : * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1311 : * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1312 : * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1313 : *
1314 : * It is the handler's job to read the input buffer from user memory
1315 : * and process it. The handler should return 0 on success.
1316 : *
1317 : * This routine returns %NULL on a failure to register, and a pointer
1318 : * to the table header on success.
1319 : */
1320 26 : struct ctl_table_header *__register_sysctl_table(
1321 : struct ctl_table_set *set,
1322 : const char *path, struct ctl_table *table)
1323 : {
1324 26 : struct ctl_table_root *root = set->dir.header.root;
1325 : struct ctl_table_header *header;
1326 : const char *name, *nextname;
1327 : struct ctl_dir *dir;
1328 : struct ctl_table *entry;
1329 : struct ctl_node *node;
1330 26 : int nr_entries = 0;
1331 :
1332 145 : for (entry = table; entry->procname; entry++)
1333 119 : nr_entries++;
1334 :
1335 26 : header = kzalloc(sizeof(struct ctl_table_header) +
1336 26 : sizeof(struct ctl_node)*nr_entries, GFP_KERNEL);
1337 26 : if (!header)
1338 : return NULL;
1339 :
1340 26 : node = (struct ctl_node *)(header + 1);
1341 26 : init_header(header, root, set, node, table);
1342 26 : if (sysctl_check_table(path, table))
1343 : goto fail;
1344 :
1345 26 : spin_lock(&sysctl_lock);
1346 26 : dir = &set->dir;
1347 : /* Reference moved down the diretory tree get_subdir */
1348 26 : dir->header.nreg++;
1349 26 : spin_unlock(&sysctl_lock);
1350 :
1351 : /* Find the directory for the ctl_table */
1352 94 : for (name = path; name; name = nextname) {
1353 : int namelen;
1354 42 : nextname = strchr(name, '/');
1355 42 : if (nextname) {
1356 16 : namelen = nextname - name;
1357 16 : nextname++;
1358 : } else {
1359 26 : namelen = strlen(name);
1360 : }
1361 42 : if (namelen == 0)
1362 9 : continue;
1363 :
1364 33 : dir = get_subdir(dir, name, namelen);
1365 33 : if (IS_ERR(dir))
1366 : goto fail;
1367 : }
1368 :
1369 26 : spin_lock(&sysctl_lock);
1370 26 : if (insert_header(dir, header))
1371 : goto fail_put_dir_locked;
1372 :
1373 26 : drop_sysctl_table(&dir->header);
1374 26 : spin_unlock(&sysctl_lock);
1375 :
1376 26 : return header;
1377 :
1378 : fail_put_dir_locked:
1379 0 : drop_sysctl_table(&dir->header);
1380 : spin_unlock(&sysctl_lock);
1381 : fail:
1382 0 : kfree(header);
1383 0 : dump_stack();
1384 0 : return NULL;
1385 : }
1386 :
1387 : /**
1388 : * register_sysctl - register a sysctl table
1389 : * @path: The path to the directory the sysctl table is in.
1390 : * @table: the table structure
1391 : *
1392 : * Register a sysctl table. @table should be a filled in ctl_table
1393 : * array. A completely 0 filled entry terminates the table.
1394 : *
1395 : * See __register_sysctl_table for more details.
1396 : */
1397 3 : struct ctl_table_header *register_sysctl(const char *path, struct ctl_table *table)
1398 : {
1399 16 : return __register_sysctl_table(&sysctl_table_root.default_set,
1400 : path, table);
1401 : }
1402 : EXPORT_SYMBOL(register_sysctl);
1403 :
1404 : /**
1405 : * __register_sysctl_init() - register sysctl table to path
1406 : * @path: path name for sysctl base
1407 : * @table: This is the sysctl table that needs to be registered to the path
1408 : * @table_name: The name of sysctl table, only used for log printing when
1409 : * registration fails
1410 : *
1411 : * The sysctl interface is used by userspace to query or modify at runtime
1412 : * a predefined value set on a variable. These variables however have default
1413 : * values pre-set. Code which depends on these variables will always work even
1414 : * if register_sysctl() fails. If register_sysctl() fails you'd just loose the
1415 : * ability to query or modify the sysctls dynamically at run time. Chances of
1416 : * register_sysctl() failing on init are extremely low, and so for both reasons
1417 : * this function does not return any error as it is used by initialization code.
1418 : *
1419 : * Context: Can only be called after your respective sysctl base path has been
1420 : * registered. So for instance, most base directories are registered early on
1421 : * init before init levels are processed through proc_sys_init() and
1422 : * sysctl_init_bases().
1423 : */
1424 13 : void __init __register_sysctl_init(const char *path, struct ctl_table *table,
1425 : const char *table_name)
1426 : {
1427 13 : struct ctl_table_header *hdr = register_sysctl(path, table);
1428 :
1429 13 : if (unlikely(!hdr)) {
1430 0 : pr_err("failed when register_sysctl %s to %s\n", table_name, path);
1431 0 : return;
1432 : }
1433 : kmemleak_not_leak(hdr);
1434 : }
1435 :
1436 12 : static char *append_path(const char *path, char *pos, const char *name)
1437 : {
1438 : int namelen;
1439 12 : namelen = strlen(name);
1440 12 : if (((pos - path) + namelen + 2) >= PATH_MAX)
1441 : return NULL;
1442 12 : memcpy(pos, name, namelen);
1443 12 : pos[namelen] = '/';
1444 12 : pos[namelen + 1] = '\0';
1445 12 : pos += namelen + 1;
1446 12 : return pos;
1447 : }
1448 :
1449 18 : static int count_subheaders(struct ctl_table *table)
1450 : {
1451 18 : int has_files = 0;
1452 18 : int nr_subheaders = 0;
1453 : struct ctl_table *entry;
1454 :
1455 : /* special case: no directory and empty directory */
1456 18 : if (!table || !table->procname)
1457 : return 1;
1458 :
1459 79 : for (entry = table; entry->procname; entry++) {
1460 79 : if (entry->child)
1461 10 : nr_subheaders += count_subheaders(entry->child);
1462 : else
1463 : has_files = 1;
1464 : }
1465 16 : return nr_subheaders + has_files;
1466 : }
1467 :
1468 2 : static int register_leaf_sysctl_tables(const char *path, char *pos,
1469 : struct ctl_table_header ***subheader, struct ctl_table_set *set,
1470 : struct ctl_table *table)
1471 : {
1472 2 : struct ctl_table *ctl_table_arg = NULL;
1473 : struct ctl_table *entry, *files;
1474 2 : int nr_files = 0;
1475 2 : int nr_dirs = 0;
1476 2 : int err = -ENOMEM;
1477 :
1478 29 : for (entry = table; entry->procname; entry++) {
1479 27 : if (entry->child)
1480 1 : nr_dirs++;
1481 : else
1482 26 : nr_files++;
1483 : }
1484 :
1485 2 : files = table;
1486 : /* If there are mixed files and directories we need a new table */
1487 2 : if (nr_dirs && nr_files) {
1488 : struct ctl_table *new;
1489 2 : files = kcalloc(nr_files + 1, sizeof(struct ctl_table),
1490 : GFP_KERNEL);
1491 1 : if (!files)
1492 : goto out;
1493 :
1494 : ctl_table_arg = files;
1495 25 : for (new = files, entry = table; entry->procname; entry++) {
1496 25 : if (entry->child)
1497 1 : continue;
1498 24 : *new = *entry;
1499 24 : new++;
1500 : }
1501 : }
1502 :
1503 : /* Register everything except a directory full of subdirectories */
1504 2 : if (nr_files || !nr_dirs) {
1505 : struct ctl_table_header *header;
1506 2 : header = __register_sysctl_table(set, path, files);
1507 2 : if (!header) {
1508 0 : kfree(ctl_table_arg);
1509 0 : goto out;
1510 : }
1511 :
1512 : /* Remember if we need to free the file table */
1513 2 : header->ctl_table_arg = ctl_table_arg;
1514 2 : **subheader = header;
1515 2 : (*subheader)++;
1516 : }
1517 :
1518 : /* Recurse into the subdirectories. */
1519 27 : for (entry = table; entry->procname; entry++) {
1520 : char *child_pos;
1521 :
1522 27 : if (!entry->child)
1523 26 : continue;
1524 :
1525 1 : err = -ENAMETOOLONG;
1526 1 : child_pos = append_path(path, pos, entry->procname);
1527 1 : if (!child_pos)
1528 : goto out;
1529 :
1530 1 : err = register_leaf_sysctl_tables(path, child_pos, subheader,
1531 : set, entry->child);
1532 1 : pos[0] = '\0';
1533 1 : if (err)
1534 : goto out;
1535 : }
1536 : err = 0;
1537 : out:
1538 : /* On failure our caller will unregister all registered subheaders */
1539 2 : return err;
1540 : }
1541 :
1542 : /**
1543 : * __register_sysctl_paths - register a sysctl table hierarchy
1544 : * @set: Sysctl tree to register on
1545 : * @path: The path to the directory the sysctl table is in.
1546 : * @table: the top-level table structure
1547 : *
1548 : * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1549 : * array. A completely 0 filled entry terminates the table.
1550 : *
1551 : * See __register_sysctl_table for more details.
1552 : */
1553 8 : struct ctl_table_header *__register_sysctl_paths(
1554 : struct ctl_table_set *set,
1555 : const struct ctl_path *path, struct ctl_table *table)
1556 : {
1557 8 : struct ctl_table *ctl_table_arg = table;
1558 8 : int nr_subheaders = count_subheaders(table);
1559 8 : struct ctl_table_header *header = NULL, **subheaders, **subheader;
1560 : const struct ctl_path *component;
1561 : char *new_path, *pos;
1562 :
1563 8 : pos = new_path = kmalloc(PATH_MAX, GFP_KERNEL);
1564 8 : if (!new_path)
1565 : return NULL;
1566 :
1567 8 : pos[0] = '\0';
1568 10 : for (component = path; component->procname; component++) {
1569 2 : pos = append_path(new_path, pos, component->procname);
1570 2 : if (!pos)
1571 : goto out;
1572 : }
1573 17 : while (table->procname && table->child && !table[1].procname) {
1574 9 : pos = append_path(new_path, pos, table->procname);
1575 9 : if (!pos)
1576 : goto out;
1577 9 : table = table->child;
1578 : }
1579 8 : if (nr_subheaders == 1) {
1580 7 : header = __register_sysctl_table(set, new_path, table);
1581 7 : if (header)
1582 7 : header->ctl_table_arg = ctl_table_arg;
1583 : } else {
1584 1 : header = kzalloc(sizeof(*header) +
1585 : sizeof(*subheaders)*nr_subheaders, GFP_KERNEL);
1586 1 : if (!header)
1587 : goto out;
1588 :
1589 1 : subheaders = (struct ctl_table_header **) (header + 1);
1590 1 : subheader = subheaders;
1591 1 : header->ctl_table_arg = ctl_table_arg;
1592 :
1593 1 : if (register_leaf_sysctl_tables(new_path, pos, &subheader,
1594 : set, table))
1595 : goto err_register_leaves;
1596 : }
1597 :
1598 : out:
1599 8 : kfree(new_path);
1600 8 : return header;
1601 :
1602 : err_register_leaves:
1603 0 : while (subheader > subheaders) {
1604 0 : struct ctl_table_header *subh = *(--subheader);
1605 0 : struct ctl_table *table = subh->ctl_table_arg;
1606 0 : unregister_sysctl_table(subh);
1607 0 : kfree(table);
1608 : }
1609 0 : kfree(header);
1610 0 : header = NULL;
1611 0 : goto out;
1612 : }
1613 :
1614 : /**
1615 : * register_sysctl_paths - register a sysctl table hierarchy
1616 : * @path: The path to the directory the sysctl table is in.
1617 : * @table: the top-level table structure
1618 : *
1619 : * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1620 : * array. A completely 0 filled entry terminates the table.
1621 : *
1622 : * See __register_sysctl_paths for more details.
1623 : */
1624 1 : struct ctl_table_header *register_sysctl_paths(const struct ctl_path *path,
1625 : struct ctl_table *table)
1626 : {
1627 8 : return __register_sysctl_paths(&sysctl_table_root.default_set,
1628 : path, table);
1629 : }
1630 : EXPORT_SYMBOL(register_sysctl_paths);
1631 :
1632 : /**
1633 : * register_sysctl_table - register a sysctl table hierarchy
1634 : * @table: the top-level table structure
1635 : *
1636 : * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1637 : * array. A completely 0 filled entry terminates the table.
1638 : *
1639 : * See register_sysctl_paths for more details.
1640 : */
1641 3 : struct ctl_table_header *register_sysctl_table(struct ctl_table *table)
1642 : {
1643 : static const struct ctl_path null_path[] = { {} };
1644 :
1645 7 : return register_sysctl_paths(null_path, table);
1646 : }
1647 : EXPORT_SYMBOL(register_sysctl_table);
1648 :
1649 4 : int __register_sysctl_base(struct ctl_table *base_table)
1650 : {
1651 : struct ctl_table_header *hdr;
1652 :
1653 4 : hdr = register_sysctl_table(base_table);
1654 : kmemleak_not_leak(hdr);
1655 4 : return 0;
1656 : }
1657 :
1658 0 : static void put_links(struct ctl_table_header *header)
1659 : {
1660 0 : struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1661 0 : struct ctl_table_root *root = header->root;
1662 0 : struct ctl_dir *parent = header->parent;
1663 : struct ctl_dir *core_parent;
1664 : struct ctl_table *entry;
1665 :
1666 0 : if (header->set == root_set)
1667 : return;
1668 :
1669 0 : core_parent = xlate_dir(root_set, parent);
1670 0 : if (IS_ERR(core_parent))
1671 : return;
1672 :
1673 0 : for (entry = header->ctl_table; entry->procname; entry++) {
1674 : struct ctl_table_header *link_head;
1675 : struct ctl_table *link;
1676 0 : const char *name = entry->procname;
1677 :
1678 0 : link = find_entry(&link_head, core_parent, name, strlen(name));
1679 0 : if (link &&
1680 0 : ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1681 0 : (S_ISLNK(link->mode) && (link->data == root)))) {
1682 0 : drop_sysctl_table(link_head);
1683 : }
1684 : else {
1685 0 : pr_err("sysctl link missing during unregister: ");
1686 0 : sysctl_print_dir(parent);
1687 0 : pr_cont("%s\n", name);
1688 : }
1689 : }
1690 : }
1691 :
1692 74 : static void drop_sysctl_table(struct ctl_table_header *header)
1693 : {
1694 74 : struct ctl_dir *parent = header->parent;
1695 :
1696 74 : if (--header->nreg)
1697 : return;
1698 :
1699 0 : if (parent) {
1700 0 : put_links(header);
1701 0 : start_unregistering(header);
1702 : }
1703 :
1704 0 : if (!--header->count)
1705 0 : kfree_rcu(header, rcu);
1706 :
1707 0 : if (parent)
1708 0 : drop_sysctl_table(&parent->header);
1709 : }
1710 :
1711 : /**
1712 : * unregister_sysctl_table - unregister a sysctl table hierarchy
1713 : * @header: the header returned from register_sysctl_table
1714 : *
1715 : * Unregisters the sysctl table and all children. proc entries may not
1716 : * actually be removed until they are no longer used by anyone.
1717 : */
1718 0 : void unregister_sysctl_table(struct ctl_table_header * header)
1719 : {
1720 : int nr_subheaders;
1721 : might_sleep();
1722 :
1723 0 : if (header == NULL)
1724 : return;
1725 :
1726 0 : nr_subheaders = count_subheaders(header->ctl_table_arg);
1727 0 : if (unlikely(nr_subheaders > 1)) {
1728 : struct ctl_table_header **subheaders;
1729 : int i;
1730 :
1731 0 : subheaders = (struct ctl_table_header **)(header + 1);
1732 0 : for (i = nr_subheaders -1; i >= 0; i--) {
1733 0 : struct ctl_table_header *subh = subheaders[i];
1734 0 : struct ctl_table *table = subh->ctl_table_arg;
1735 0 : unregister_sysctl_table(subh);
1736 0 : kfree(table);
1737 : }
1738 0 : kfree(header);
1739 0 : return;
1740 : }
1741 :
1742 0 : spin_lock(&sysctl_lock);
1743 0 : drop_sysctl_table(header);
1744 : spin_unlock(&sysctl_lock);
1745 : }
1746 : EXPORT_SYMBOL(unregister_sysctl_table);
1747 :
1748 1 : void setup_sysctl_set(struct ctl_table_set *set,
1749 : struct ctl_table_root *root,
1750 : int (*is_seen)(struct ctl_table_set *))
1751 : {
1752 1 : memset(set, 0, sizeof(*set));
1753 1 : set->is_seen = is_seen;
1754 2 : init_header(&set->dir.header, root, set, NULL, root_table);
1755 1 : }
1756 :
1757 0 : void retire_sysctl_set(struct ctl_table_set *set)
1758 : {
1759 0 : WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1760 0 : }
1761 :
1762 1 : int __init proc_sys_init(void)
1763 : {
1764 : struct proc_dir_entry *proc_sys_root;
1765 :
1766 1 : proc_sys_root = proc_mkdir("sys", NULL);
1767 1 : proc_sys_root->proc_iops = &proc_sys_dir_operations;
1768 1 : proc_sys_root->proc_dir_ops = &proc_sys_dir_file_operations;
1769 1 : proc_sys_root->nlink = 0;
1770 :
1771 1 : return sysctl_init_bases();
1772 : }
1773 :
1774 : struct sysctl_alias {
1775 : const char *kernel_param;
1776 : const char *sysctl_param;
1777 : };
1778 :
1779 : /*
1780 : * Historically some settings had both sysctl and a command line parameter.
1781 : * With the generic sysctl. parameter support, we can handle them at a single
1782 : * place and only keep the historical name for compatibility. This is not meant
1783 : * to add brand new aliases. When adding existing aliases, consider whether
1784 : * the possibly different moment of changing the value (e.g. from early_param
1785 : * to the moment do_sysctl_args() is called) is an issue for the specific
1786 : * parameter.
1787 : */
1788 : static const struct sysctl_alias sysctl_aliases[] = {
1789 : {"hardlockup_all_cpu_backtrace", "kernel.hardlockup_all_cpu_backtrace" },
1790 : {"hung_task_panic", "kernel.hung_task_panic" },
1791 : {"numa_zonelist_order", "vm.numa_zonelist_order" },
1792 : {"softlockup_all_cpu_backtrace", "kernel.softlockup_all_cpu_backtrace" },
1793 : {"softlockup_panic", "kernel.softlockup_panic" },
1794 : { }
1795 : };
1796 :
1797 0 : static const char *sysctl_find_alias(char *param)
1798 : {
1799 : const struct sysctl_alias *alias;
1800 :
1801 0 : for (alias = &sysctl_aliases[0]; alias->kernel_param != NULL; alias++) {
1802 0 : if (strcmp(alias->kernel_param, param) == 0)
1803 0 : return alias->sysctl_param;
1804 : }
1805 :
1806 : return NULL;
1807 : }
1808 :
1809 : /* Set sysctl value passed on kernel command line. */
1810 0 : static int process_sysctl_arg(char *param, char *val,
1811 : const char *unused, void *arg)
1812 : {
1813 : char *path;
1814 0 : struct vfsmount **proc_mnt = arg;
1815 : struct file_system_type *proc_fs_type;
1816 : struct file *file;
1817 : int len;
1818 : int err;
1819 0 : loff_t pos = 0;
1820 : ssize_t wret;
1821 :
1822 0 : if (strncmp(param, "sysctl", sizeof("sysctl") - 1) == 0) {
1823 0 : param += sizeof("sysctl") - 1;
1824 :
1825 0 : if (param[0] != '/' && param[0] != '.')
1826 : return 0;
1827 :
1828 0 : param++;
1829 : } else {
1830 0 : param = (char *) sysctl_find_alias(param);
1831 0 : if (!param)
1832 : return 0;
1833 : }
1834 :
1835 0 : if (!val)
1836 : return -EINVAL;
1837 0 : len = strlen(val);
1838 0 : if (len == 0)
1839 : return -EINVAL;
1840 :
1841 : /*
1842 : * To set sysctl options, we use a temporary mount of proc, look up the
1843 : * respective sys/ file and write to it. To avoid mounting it when no
1844 : * options were given, we mount it only when the first sysctl option is
1845 : * found. Why not a persistent mount? There are problems with a
1846 : * persistent mount of proc in that it forces userspace not to use any
1847 : * proc mount options.
1848 : */
1849 0 : if (!*proc_mnt) {
1850 0 : proc_fs_type = get_fs_type("proc");
1851 0 : if (!proc_fs_type) {
1852 0 : pr_err("Failed to find procfs to set sysctl from command line\n");
1853 0 : return 0;
1854 : }
1855 0 : *proc_mnt = kern_mount(proc_fs_type);
1856 0 : put_filesystem(proc_fs_type);
1857 0 : if (IS_ERR(*proc_mnt)) {
1858 0 : pr_err("Failed to mount procfs to set sysctl from command line\n");
1859 0 : return 0;
1860 : }
1861 : }
1862 :
1863 0 : path = kasprintf(GFP_KERNEL, "sys/%s", param);
1864 0 : if (!path)
1865 0 : panic("%s: Failed to allocate path for %s\n", __func__, param);
1866 0 : strreplace(path, '.', '/');
1867 :
1868 0 : file = file_open_root_mnt(*proc_mnt, path, O_WRONLY, 0);
1869 0 : if (IS_ERR(file)) {
1870 0 : err = PTR_ERR(file);
1871 0 : if (err == -ENOENT)
1872 0 : pr_err("Failed to set sysctl parameter '%s=%s': parameter not found\n",
1873 : param, val);
1874 0 : else if (err == -EACCES)
1875 0 : pr_err("Failed to set sysctl parameter '%s=%s': permission denied (read-only?)\n",
1876 : param, val);
1877 : else
1878 0 : pr_err("Error %pe opening proc file to set sysctl parameter '%s=%s'\n",
1879 : file, param, val);
1880 : goto out;
1881 : }
1882 0 : wret = kernel_write(file, val, len, &pos);
1883 0 : if (wret < 0) {
1884 0 : err = wret;
1885 0 : if (err == -EINVAL)
1886 0 : pr_err("Failed to set sysctl parameter '%s=%s': invalid value\n",
1887 : param, val);
1888 : else
1889 0 : pr_err("Error %pe writing to proc file to set sysctl parameter '%s=%s'\n",
1890 : ERR_PTR(err), param, val);
1891 0 : } else if (wret != len) {
1892 0 : pr_err("Wrote only %zd bytes of %d writing to proc file %s to set sysctl parameter '%s=%s\n",
1893 : wret, len, path, param, val);
1894 : }
1895 :
1896 0 : err = filp_close(file, NULL);
1897 0 : if (err)
1898 0 : pr_err("Error %pe closing proc file to set sysctl parameter '%s=%s\n",
1899 : ERR_PTR(err), param, val);
1900 : out:
1901 0 : kfree(path);
1902 0 : return 0;
1903 : }
1904 :
1905 0 : void do_sysctl_args(void)
1906 : {
1907 : char *command_line;
1908 0 : struct vfsmount *proc_mnt = NULL;
1909 :
1910 0 : command_line = kstrdup(saved_command_line, GFP_KERNEL);
1911 0 : if (!command_line)
1912 0 : panic("%s: Failed to allocate copy of command line\n", __func__);
1913 :
1914 0 : parse_args("Setting sysctl args", command_line,
1915 : NULL, 0, -1, -1, &proc_mnt, process_sysctl_arg);
1916 :
1917 0 : if (proc_mnt)
1918 0 : kern_unmount(proc_mnt);
1919 :
1920 0 : kfree(command_line);
1921 0 : }
|
