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 : static struct ctl_table sysctl_mount_point[] = {
33 : {.type = SYSCTL_TABLE_TYPE_PERMANENTLY_EMPTY }
34 : };
35 :
36 : /**
37 : * register_sysctl_mount_point() - registers a sysctl mount point
38 : * @path: path for the mount point
39 : *
40 : * Used to create a permanently empty directory to serve as mount point.
41 : * There are some subtle but important permission checks this allows in the
42 : * case of unprivileged mounts.
43 : */
44 0 : struct ctl_table_header *register_sysctl_mount_point(const char *path)
45 : {
46 0 : return register_sysctl(path, sysctl_mount_point);
47 : }
48 : EXPORT_SYMBOL(register_sysctl_mount_point);
49 :
50 : #define sysctl_is_perm_empty_ctl_table(tptr) \
51 : (tptr[0].type == SYSCTL_TABLE_TYPE_PERMANENTLY_EMPTY)
52 : #define sysctl_is_perm_empty_ctl_header(hptr) \
53 : (sysctl_is_perm_empty_ctl_table(hptr->ctl_table))
54 : #define sysctl_set_perm_empty_ctl_header(hptr) \
55 : (hptr->ctl_table[0].type = SYSCTL_TABLE_TYPE_PERMANENTLY_EMPTY)
56 : #define sysctl_clear_perm_empty_ctl_header(hptr) \
57 : (hptr->ctl_table[0].type = SYSCTL_TABLE_TYPE_DEFAULT)
58 :
59 0 : void proc_sys_poll_notify(struct ctl_table_poll *poll)
60 : {
61 0 : if (!poll)
62 : return;
63 :
64 0 : atomic_inc(&poll->event);
65 0 : wake_up_interruptible(&poll->wait);
66 : }
67 :
68 : static struct ctl_table root_table[] = {
69 : {
70 : .procname = "",
71 : .mode = S_IFDIR|S_IRUGO|S_IXUGO,
72 : },
73 : { }
74 : };
75 : static struct ctl_table_root sysctl_table_root = {
76 : .default_set.dir.header = {
77 : {{.count = 1,
78 : .nreg = 1,
79 : .ctl_table = root_table }},
80 : .ctl_table_arg = root_table,
81 : .root = &sysctl_table_root,
82 : .set = &sysctl_table_root.default_set,
83 : },
84 : };
85 :
86 : static DEFINE_SPINLOCK(sysctl_lock);
87 :
88 : static void drop_sysctl_table(struct ctl_table_header *header);
89 : static int sysctl_follow_link(struct ctl_table_header **phead,
90 : struct ctl_table **pentry);
91 : static int insert_links(struct ctl_table_header *head);
92 : static void put_links(struct ctl_table_header *header);
93 :
94 0 : static void sysctl_print_dir(struct ctl_dir *dir)
95 : {
96 0 : if (dir->header.parent)
97 0 : sysctl_print_dir(dir->header.parent);
98 0 : pr_cont("%s/", dir->header.ctl_table[0].procname);
99 0 : }
100 :
101 638 : static int namecmp(const char *name1, int len1, const char *name2, int len2)
102 : {
103 : int cmp;
104 :
105 1276 : cmp = memcmp(name1, name2, min(len1, len2));
106 638 : if (cmp == 0)
107 44 : cmp = len1 - len2;
108 638 : return cmp;
109 : }
110 :
111 : /* Called under sysctl_lock */
112 63 : static struct ctl_table *find_entry(struct ctl_table_header **phead,
113 : struct ctl_dir *dir, const char *name, int namelen)
114 : {
115 : struct ctl_table_header *head;
116 : struct ctl_table *entry;
117 63 : struct rb_node *node = dir->root.rb_node;
118 :
119 166 : while (node)
120 : {
121 : struct ctl_node *ctl_node;
122 : const char *procname;
123 : int cmp;
124 :
125 136 : ctl_node = rb_entry(node, struct ctl_node, node);
126 136 : head = ctl_node->header;
127 136 : entry = &head->ctl_table[ctl_node - head->node];
128 136 : procname = entry->procname;
129 :
130 136 : cmp = namecmp(name, namelen, procname, strlen(procname));
131 136 : if (cmp < 0)
132 51 : node = node->rb_left;
133 85 : else if (cmp > 0)
134 52 : node = node->rb_right;
135 : else {
136 33 : *phead = head;
137 : return entry;
138 : }
139 : }
140 : return NULL;
141 : }
142 :
143 149 : static int insert_entry(struct ctl_table_header *head, struct ctl_table *entry)
144 : {
145 149 : struct rb_node *node = &head->node[entry - head->ctl_table].node;
146 149 : struct rb_node **p = &head->parent->root.rb_node;
147 149 : struct rb_node *parent = NULL;
148 149 : const char *name = entry->procname;
149 149 : int namelen = strlen(name);
150 :
151 800 : while (*p) {
152 : struct ctl_table_header *parent_head;
153 : struct ctl_table *parent_entry;
154 : struct ctl_node *parent_node;
155 : const char *parent_name;
156 : int cmp;
157 :
158 502 : parent = *p;
159 502 : parent_node = rb_entry(parent, struct ctl_node, node);
160 502 : parent_head = parent_node->header;
161 502 : parent_entry = &parent_head->ctl_table[parent_node - parent_head->node];
162 502 : parent_name = parent_entry->procname;
163 :
164 502 : cmp = namecmp(name, namelen, parent_name, strlen(parent_name));
165 502 : if (cmp < 0)
166 232 : p = &(*p)->rb_left;
167 270 : else if (cmp > 0)
168 270 : p = &(*p)->rb_right;
169 : else {
170 0 : pr_err("sysctl duplicate entry: ");
171 0 : sysctl_print_dir(head->parent);
172 0 : pr_cont("%s\n", entry->procname);
173 0 : return -EEXIST;
174 : }
175 : }
176 :
177 149 : rb_link_node(node, parent, p);
178 149 : rb_insert_color(node, &head->parent->root);
179 149 : return 0;
180 : }
181 :
182 : static void erase_entry(struct ctl_table_header *head, struct ctl_table *entry)
183 : {
184 0 : struct rb_node *node = &head->node[entry - head->ctl_table].node;
185 :
186 0 : rb_erase(node, &head->parent->root);
187 : }
188 :
189 : static void init_header(struct ctl_table_header *head,
190 : struct ctl_table_root *root, struct ctl_table_set *set,
191 : struct ctl_node *node, struct ctl_table *table)
192 : {
193 53 : head->ctl_table = table;
194 53 : head->ctl_table_arg = table;
195 53 : head->used = 0;
196 53 : head->count = 1;
197 53 : head->nreg = 1;
198 53 : head->unregistering = NULL;
199 53 : head->root = root;
200 53 : head->set = set;
201 53 : head->parent = NULL;
202 53 : head->node = node;
203 53 : INIT_HLIST_HEAD(&head->inodes);
204 52 : if (node) {
205 : struct ctl_table *entry;
206 :
207 149 : list_for_each_table_entry(entry, table) {
208 149 : node->header = head;
209 149 : node++;
210 : }
211 : }
212 : }
213 :
214 0 : static void erase_header(struct ctl_table_header *head)
215 : {
216 : struct ctl_table *entry;
217 :
218 0 : list_for_each_table_entry(entry, head->ctl_table)
219 0 : erase_entry(head, entry);
220 0 : }
221 :
222 52 : static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
223 : {
224 : struct ctl_table *entry;
225 52 : struct ctl_table_header *dir_h = &dir->header;
226 : int err;
227 :
228 :
229 : /* Is this a permanently empty directory? */
230 52 : if (sysctl_is_perm_empty_ctl_header(dir_h))
231 : return -EROFS;
232 :
233 : /* Am I creating a permanently empty directory? */
234 52 : if (sysctl_is_perm_empty_ctl_table(header->ctl_table)) {
235 0 : if (!RB_EMPTY_ROOT(&dir->root))
236 : return -EINVAL;
237 0 : sysctl_set_perm_empty_ctl_header(dir_h);
238 : }
239 :
240 52 : dir_h->nreg++;
241 52 : header->parent = dir;
242 52 : err = insert_links(header);
243 52 : if (err)
244 : goto fail_links;
245 201 : list_for_each_table_entry(entry, header->ctl_table) {
246 149 : err = insert_entry(header, entry);
247 149 : if (err)
248 : goto fail;
249 : }
250 : return 0;
251 : fail:
252 0 : erase_header(header);
253 0 : put_links(header);
254 : fail_links:
255 0 : if (header->ctl_table == sysctl_mount_point)
256 0 : sysctl_clear_perm_empty_ctl_header(dir_h);
257 0 : header->parent = NULL;
258 0 : drop_sysctl_table(dir_h);
259 0 : return err;
260 : }
261 :
262 : /* called under sysctl_lock */
263 : static int use_table(struct ctl_table_header *p)
264 : {
265 0 : if (unlikely(p->unregistering))
266 : return 0;
267 0 : p->used++;
268 : return 1;
269 : }
270 :
271 : /* called under sysctl_lock */
272 : static void unuse_table(struct ctl_table_header *p)
273 : {
274 0 : if (!--p->used)
275 0 : if (unlikely(p->unregistering))
276 0 : complete(p->unregistering);
277 : }
278 :
279 : static void proc_sys_invalidate_dcache(struct ctl_table_header *head)
280 : {
281 0 : proc_invalidate_siblings_dcache(&head->inodes, &sysctl_lock);
282 : }
283 :
284 : /* called under sysctl_lock, will reacquire if has to wait */
285 0 : static void start_unregistering(struct ctl_table_header *p)
286 : {
287 : /*
288 : * if p->used is 0, nobody will ever touch that entry again;
289 : * we'll eliminate all paths to it before dropping sysctl_lock
290 : */
291 0 : if (unlikely(p->used)) {
292 : struct completion wait;
293 0 : init_completion(&wait);
294 0 : p->unregistering = &wait;
295 0 : spin_unlock(&sysctl_lock);
296 0 : wait_for_completion(&wait);
297 : } else {
298 : /* anything non-NULL; we'll never dereference it */
299 0 : p->unregistering = ERR_PTR(-EINVAL);
300 : spin_unlock(&sysctl_lock);
301 : }
302 : /*
303 : * Invalidate dentries for unregistered sysctls: namespaced sysctls
304 : * can have duplicate names and contaminate dcache very badly.
305 : */
306 0 : proc_sys_invalidate_dcache(p);
307 : /*
308 : * do not remove from the list until nobody holds it; walking the
309 : * list in do_sysctl() relies on that.
310 : */
311 0 : spin_lock(&sysctl_lock);
312 0 : erase_header(p);
313 0 : }
314 :
315 0 : static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
316 : {
317 0 : BUG_ON(!head);
318 0 : spin_lock(&sysctl_lock);
319 0 : if (!use_table(head))
320 0 : head = ERR_PTR(-ENOENT);
321 0 : spin_unlock(&sysctl_lock);
322 0 : return head;
323 : }
324 :
325 0 : static void sysctl_head_finish(struct ctl_table_header *head)
326 : {
327 0 : if (!head)
328 : return;
329 0 : spin_lock(&sysctl_lock);
330 0 : unuse_table(head);
331 : spin_unlock(&sysctl_lock);
332 : }
333 :
334 : static struct ctl_table_set *
335 : lookup_header_set(struct ctl_table_root *root)
336 : {
337 0 : struct ctl_table_set *set = &root->default_set;
338 0 : if (root->lookup)
339 0 : set = root->lookup(root);
340 : return set;
341 : }
342 :
343 0 : static struct ctl_table *lookup_entry(struct ctl_table_header **phead,
344 : struct ctl_dir *dir,
345 : const char *name, int namelen)
346 : {
347 : struct ctl_table_header *head;
348 : struct ctl_table *entry;
349 :
350 0 : spin_lock(&sysctl_lock);
351 0 : entry = find_entry(&head, dir, name, namelen);
352 0 : if (entry && use_table(head))
353 0 : *phead = head;
354 : else
355 : entry = NULL;
356 0 : spin_unlock(&sysctl_lock);
357 0 : return entry;
358 : }
359 :
360 0 : static struct ctl_node *first_usable_entry(struct rb_node *node)
361 : {
362 : struct ctl_node *ctl_node;
363 :
364 0 : for (;node; node = rb_next(node)) {
365 0 : ctl_node = rb_entry(node, struct ctl_node, node);
366 0 : if (use_table(ctl_node->header))
367 : return ctl_node;
368 : }
369 : return NULL;
370 : }
371 :
372 0 : static void first_entry(struct ctl_dir *dir,
373 : struct ctl_table_header **phead, struct ctl_table **pentry)
374 : {
375 0 : struct ctl_table_header *head = NULL;
376 0 : struct ctl_table *entry = NULL;
377 : struct ctl_node *ctl_node;
378 :
379 0 : spin_lock(&sysctl_lock);
380 0 : ctl_node = first_usable_entry(rb_first(&dir->root));
381 0 : spin_unlock(&sysctl_lock);
382 0 : if (ctl_node) {
383 0 : head = ctl_node->header;
384 0 : entry = &head->ctl_table[ctl_node - head->node];
385 : }
386 0 : *phead = head;
387 0 : *pentry = entry;
388 0 : }
389 :
390 0 : static void next_entry(struct ctl_table_header **phead, struct ctl_table **pentry)
391 : {
392 0 : struct ctl_table_header *head = *phead;
393 0 : struct ctl_table *entry = *pentry;
394 0 : struct ctl_node *ctl_node = &head->node[entry - head->ctl_table];
395 :
396 0 : spin_lock(&sysctl_lock);
397 0 : unuse_table(head);
398 :
399 0 : ctl_node = first_usable_entry(rb_next(&ctl_node->node));
400 0 : spin_unlock(&sysctl_lock);
401 0 : head = NULL;
402 0 : if (ctl_node) {
403 0 : head = ctl_node->header;
404 0 : entry = &head->ctl_table[ctl_node - head->node];
405 : }
406 0 : *phead = head;
407 0 : *pentry = entry;
408 0 : }
409 :
410 : /*
411 : * sysctl_perm does NOT grant the superuser all rights automatically, because
412 : * some sysctl variables are readonly even to root.
413 : */
414 :
415 0 : static int test_perm(int mode, int op)
416 : {
417 0 : if (uid_eq(current_euid(), GLOBAL_ROOT_UID))
418 0 : mode >>= 6;
419 0 : else if (in_egroup_p(GLOBAL_ROOT_GID))
420 0 : mode >>= 3;
421 0 : if ((op & ~mode & (MAY_READ|MAY_WRITE|MAY_EXEC)) == 0)
422 : return 0;
423 0 : return -EACCES;
424 : }
425 :
426 0 : static int sysctl_perm(struct ctl_table_header *head, struct ctl_table *table, int op)
427 : {
428 0 : struct ctl_table_root *root = head->root;
429 : int mode;
430 :
431 0 : if (root->permissions)
432 0 : mode = root->permissions(head, table);
433 : else
434 0 : mode = table->mode;
435 :
436 0 : return test_perm(mode, op);
437 : }
438 :
439 0 : static struct inode *proc_sys_make_inode(struct super_block *sb,
440 : struct ctl_table_header *head, struct ctl_table *table)
441 : {
442 0 : struct ctl_table_root *root = head->root;
443 : struct inode *inode;
444 : struct proc_inode *ei;
445 :
446 0 : inode = new_inode(sb);
447 0 : if (!inode)
448 : return ERR_PTR(-ENOMEM);
449 :
450 0 : inode->i_ino = get_next_ino();
451 :
452 0 : ei = PROC_I(inode);
453 :
454 0 : spin_lock(&sysctl_lock);
455 0 : if (unlikely(head->unregistering)) {
456 0 : spin_unlock(&sysctl_lock);
457 0 : iput(inode);
458 0 : return ERR_PTR(-ENOENT);
459 : }
460 0 : ei->sysctl = head;
461 0 : ei->sysctl_entry = table;
462 0 : hlist_add_head_rcu(&ei->sibling_inodes, &head->inodes);
463 0 : head->count++;
464 0 : spin_unlock(&sysctl_lock);
465 :
466 0 : inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
467 0 : inode->i_mode = table->mode;
468 0 : if (!S_ISDIR(table->mode)) {
469 0 : inode->i_mode |= S_IFREG;
470 0 : inode->i_op = &proc_sys_inode_operations;
471 0 : inode->i_fop = &proc_sys_file_operations;
472 : } else {
473 0 : inode->i_mode |= S_IFDIR;
474 0 : inode->i_op = &proc_sys_dir_operations;
475 0 : inode->i_fop = &proc_sys_dir_file_operations;
476 0 : if (sysctl_is_perm_empty_ctl_header(head))
477 0 : make_empty_dir_inode(inode);
478 : }
479 :
480 0 : if (root->set_ownership)
481 0 : root->set_ownership(head, table, &inode->i_uid, &inode->i_gid);
482 : else {
483 0 : inode->i_uid = GLOBAL_ROOT_UID;
484 0 : inode->i_gid = GLOBAL_ROOT_GID;
485 : }
486 :
487 : return inode;
488 : }
489 :
490 0 : void proc_sys_evict_inode(struct inode *inode, struct ctl_table_header *head)
491 : {
492 0 : spin_lock(&sysctl_lock);
493 0 : hlist_del_init_rcu(&PROC_I(inode)->sibling_inodes);
494 0 : if (!--head->count)
495 0 : kfree_rcu(head, rcu);
496 0 : spin_unlock(&sysctl_lock);
497 0 : }
498 :
499 : static struct ctl_table_header *grab_header(struct inode *inode)
500 : {
501 0 : struct ctl_table_header *head = PROC_I(inode)->sysctl;
502 0 : if (!head)
503 0 : head = &sysctl_table_root.default_set.dir.header;
504 0 : return sysctl_head_grab(head);
505 : }
506 :
507 0 : static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
508 : unsigned int flags)
509 : {
510 0 : struct ctl_table_header *head = grab_header(dir);
511 0 : struct ctl_table_header *h = NULL;
512 0 : const struct qstr *name = &dentry->d_name;
513 : struct ctl_table *p;
514 : struct inode *inode;
515 0 : struct dentry *err = ERR_PTR(-ENOENT);
516 : struct ctl_dir *ctl_dir;
517 : int ret;
518 :
519 0 : if (IS_ERR(head))
520 : return ERR_CAST(head);
521 :
522 0 : ctl_dir = container_of(head, struct ctl_dir, header);
523 :
524 0 : p = lookup_entry(&h, ctl_dir, name->name, name->len);
525 0 : if (!p)
526 : goto out;
527 :
528 0 : if (S_ISLNK(p->mode)) {
529 0 : ret = sysctl_follow_link(&h, &p);
530 0 : err = ERR_PTR(ret);
531 0 : if (ret)
532 : goto out;
533 : }
534 :
535 0 : inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
536 0 : if (IS_ERR(inode)) {
537 : err = ERR_CAST(inode);
538 : goto out;
539 : }
540 :
541 0 : d_set_d_op(dentry, &proc_sys_dentry_operations);
542 0 : err = d_splice_alias(inode, dentry);
543 :
544 : out:
545 0 : if (h)
546 0 : sysctl_head_finish(h);
547 0 : sysctl_head_finish(head);
548 0 : return err;
549 : }
550 :
551 0 : static ssize_t proc_sys_call_handler(struct kiocb *iocb, struct iov_iter *iter,
552 : int write)
553 : {
554 0 : struct inode *inode = file_inode(iocb->ki_filp);
555 0 : struct ctl_table_header *head = grab_header(inode);
556 0 : struct ctl_table *table = PROC_I(inode)->sysctl_entry;
557 0 : size_t count = iov_iter_count(iter);
558 : char *kbuf;
559 : ssize_t error;
560 :
561 0 : if (IS_ERR(head))
562 0 : return PTR_ERR(head);
563 :
564 : /*
565 : * At this point we know that the sysctl was not unregistered
566 : * and won't be until we finish.
567 : */
568 0 : error = -EPERM;
569 0 : if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
570 : goto out;
571 :
572 : /* if that can happen at all, it should be -EINVAL, not -EISDIR */
573 0 : error = -EINVAL;
574 0 : if (!table->proc_handler)
575 : goto out;
576 :
577 : /* don't even try if the size is too large */
578 0 : error = -ENOMEM;
579 0 : if (count >= KMALLOC_MAX_SIZE)
580 : goto out;
581 0 : kbuf = kvzalloc(count + 1, GFP_KERNEL);
582 0 : if (!kbuf)
583 : goto out;
584 :
585 0 : if (write) {
586 0 : error = -EFAULT;
587 0 : if (!copy_from_iter_full(kbuf, count, iter))
588 : goto out_free_buf;
589 0 : kbuf[count] = '\0';
590 : }
591 :
592 0 : error = BPF_CGROUP_RUN_PROG_SYSCTL(head, table, write, &kbuf, &count,
593 : &iocb->ki_pos);
594 : if (error)
595 : goto out_free_buf;
596 :
597 : /* careful: calling conventions are nasty here */
598 0 : error = table->proc_handler(table, write, kbuf, &count, &iocb->ki_pos);
599 0 : if (error)
600 : goto out_free_buf;
601 :
602 0 : if (!write) {
603 0 : error = -EFAULT;
604 0 : if (copy_to_iter(kbuf, count, iter) < count)
605 : goto out_free_buf;
606 : }
607 :
608 0 : error = count;
609 : out_free_buf:
610 0 : kvfree(kbuf);
611 : out:
612 0 : sysctl_head_finish(head);
613 :
614 0 : return error;
615 : }
616 :
617 0 : static ssize_t proc_sys_read(struct kiocb *iocb, struct iov_iter *iter)
618 : {
619 0 : return proc_sys_call_handler(iocb, iter, 0);
620 : }
621 :
622 0 : static ssize_t proc_sys_write(struct kiocb *iocb, struct iov_iter *iter)
623 : {
624 0 : return proc_sys_call_handler(iocb, iter, 1);
625 : }
626 :
627 0 : static int proc_sys_open(struct inode *inode, struct file *filp)
628 : {
629 0 : struct ctl_table_header *head = grab_header(inode);
630 0 : struct ctl_table *table = PROC_I(inode)->sysctl_entry;
631 :
632 : /* sysctl was unregistered */
633 0 : if (IS_ERR(head))
634 0 : return PTR_ERR(head);
635 :
636 0 : if (table->poll)
637 0 : filp->private_data = proc_sys_poll_event(table->poll);
638 :
639 0 : sysctl_head_finish(head);
640 :
641 0 : return 0;
642 : }
643 :
644 0 : static __poll_t proc_sys_poll(struct file *filp, poll_table *wait)
645 : {
646 0 : struct inode *inode = file_inode(filp);
647 0 : struct ctl_table_header *head = grab_header(inode);
648 0 : struct ctl_table *table = PROC_I(inode)->sysctl_entry;
649 0 : __poll_t ret = DEFAULT_POLLMASK;
650 : unsigned long event;
651 :
652 : /* sysctl was unregistered */
653 0 : if (IS_ERR(head))
654 : return EPOLLERR | EPOLLHUP;
655 :
656 0 : if (!table->proc_handler)
657 : goto out;
658 :
659 0 : if (!table->poll)
660 : goto out;
661 :
662 0 : event = (unsigned long)filp->private_data;
663 0 : poll_wait(filp, &table->poll->wait, wait);
664 :
665 0 : if (event != atomic_read(&table->poll->event)) {
666 0 : filp->private_data = proc_sys_poll_event(table->poll);
667 0 : ret = EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
668 : }
669 :
670 : out:
671 0 : sysctl_head_finish(head);
672 :
673 0 : return ret;
674 : }
675 :
676 0 : static bool proc_sys_fill_cache(struct file *file,
677 : struct dir_context *ctx,
678 : struct ctl_table_header *head,
679 : struct ctl_table *table)
680 : {
681 0 : struct dentry *child, *dir = file->f_path.dentry;
682 : struct inode *inode;
683 : struct qstr qname;
684 0 : ino_t ino = 0;
685 0 : unsigned type = DT_UNKNOWN;
686 :
687 0 : qname.name = table->procname;
688 0 : qname.len = strlen(table->procname);
689 0 : qname.hash = full_name_hash(dir, qname.name, qname.len);
690 :
691 0 : child = d_lookup(dir, &qname);
692 0 : if (!child) {
693 0 : DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
694 0 : child = d_alloc_parallel(dir, &qname, &wq);
695 0 : if (IS_ERR(child))
696 0 : return false;
697 0 : if (d_in_lookup(child)) {
698 : struct dentry *res;
699 0 : inode = proc_sys_make_inode(dir->d_sb, head, table);
700 0 : if (IS_ERR(inode)) {
701 0 : d_lookup_done(child);
702 0 : dput(child);
703 : return false;
704 : }
705 0 : d_set_d_op(child, &proc_sys_dentry_operations);
706 0 : res = d_splice_alias(inode, child);
707 0 : d_lookup_done(child);
708 0 : if (unlikely(res)) {
709 0 : if (IS_ERR(res)) {
710 0 : dput(child);
711 : return false;
712 : }
713 0 : dput(child);
714 0 : child = res;
715 : }
716 : }
717 : }
718 0 : inode = d_inode(child);
719 0 : ino = inode->i_ino;
720 0 : type = inode->i_mode >> 12;
721 0 : dput(child);
722 0 : return dir_emit(ctx, qname.name, qname.len, ino, type);
723 : }
724 :
725 0 : static bool proc_sys_link_fill_cache(struct file *file,
726 : struct dir_context *ctx,
727 : struct ctl_table_header *head,
728 : struct ctl_table *table)
729 : {
730 0 : bool ret = true;
731 :
732 0 : head = sysctl_head_grab(head);
733 0 : if (IS_ERR(head))
734 : return false;
735 :
736 : /* It is not an error if we can not follow the link ignore it */
737 0 : if (sysctl_follow_link(&head, &table))
738 : goto out;
739 :
740 0 : ret = proc_sys_fill_cache(file, ctx, head, table);
741 : out:
742 0 : sysctl_head_finish(head);
743 0 : return ret;
744 : }
745 :
746 0 : static int scan(struct ctl_table_header *head, struct ctl_table *table,
747 : unsigned long *pos, struct file *file,
748 : struct dir_context *ctx)
749 : {
750 : bool res;
751 :
752 0 : if ((*pos)++ < ctx->pos)
753 : return true;
754 :
755 0 : if (unlikely(S_ISLNK(table->mode)))
756 0 : res = proc_sys_link_fill_cache(file, ctx, head, table);
757 : else
758 0 : res = proc_sys_fill_cache(file, ctx, head, table);
759 :
760 0 : if (res)
761 0 : ctx->pos = *pos;
762 :
763 0 : return res;
764 : }
765 :
766 0 : static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
767 : {
768 0 : struct ctl_table_header *head = grab_header(file_inode(file));
769 0 : struct ctl_table_header *h = NULL;
770 : struct ctl_table *entry;
771 : struct ctl_dir *ctl_dir;
772 : unsigned long pos;
773 :
774 0 : if (IS_ERR(head))
775 0 : return PTR_ERR(head);
776 :
777 0 : ctl_dir = container_of(head, struct ctl_dir, header);
778 :
779 0 : if (!dir_emit_dots(file, ctx))
780 : goto out;
781 :
782 0 : pos = 2;
783 :
784 0 : for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
785 0 : if (!scan(h, entry, &pos, file, ctx)) {
786 0 : sysctl_head_finish(h);
787 0 : break;
788 : }
789 : }
790 : out:
791 0 : sysctl_head_finish(head);
792 0 : return 0;
793 : }
794 :
795 0 : static int proc_sys_permission(struct mnt_idmap *idmap,
796 : struct inode *inode, int mask)
797 : {
798 : /*
799 : * sysctl entries that are not writeable,
800 : * are _NOT_ writeable, capabilities or not.
801 : */
802 : struct ctl_table_header *head;
803 : struct ctl_table *table;
804 : int error;
805 :
806 : /* Executable files are not allowed under /proc/sys/ */
807 0 : if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
808 : return -EACCES;
809 :
810 0 : head = grab_header(inode);
811 0 : if (IS_ERR(head))
812 0 : return PTR_ERR(head);
813 :
814 0 : table = PROC_I(inode)->sysctl_entry;
815 0 : if (!table) /* global root - r-xr-xr-x */
816 0 : error = mask & MAY_WRITE ? -EACCES : 0;
817 : else /* Use the permissions on the sysctl table entry */
818 0 : error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
819 :
820 0 : sysctl_head_finish(head);
821 0 : return error;
822 : }
823 :
824 0 : static int proc_sys_setattr(struct mnt_idmap *idmap,
825 : struct dentry *dentry, struct iattr *attr)
826 : {
827 0 : struct inode *inode = d_inode(dentry);
828 : int error;
829 :
830 0 : if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
831 : return -EPERM;
832 :
833 0 : error = setattr_prepare(&nop_mnt_idmap, dentry, attr);
834 0 : if (error)
835 : return error;
836 :
837 0 : setattr_copy(&nop_mnt_idmap, inode, attr);
838 0 : return 0;
839 : }
840 :
841 0 : static int proc_sys_getattr(struct mnt_idmap *idmap,
842 : const struct path *path, struct kstat *stat,
843 : u32 request_mask, unsigned int query_flags)
844 : {
845 0 : struct inode *inode = d_inode(path->dentry);
846 0 : struct ctl_table_header *head = grab_header(inode);
847 0 : struct ctl_table *table = PROC_I(inode)->sysctl_entry;
848 :
849 0 : if (IS_ERR(head))
850 0 : return PTR_ERR(head);
851 :
852 0 : generic_fillattr(&nop_mnt_idmap, inode, stat);
853 0 : if (table)
854 0 : stat->mode = (stat->mode & S_IFMT) | table->mode;
855 :
856 0 : sysctl_head_finish(head);
857 0 : return 0;
858 : }
859 :
860 : static const struct file_operations proc_sys_file_operations = {
861 : .open = proc_sys_open,
862 : .poll = proc_sys_poll,
863 : .read_iter = proc_sys_read,
864 : .write_iter = proc_sys_write,
865 : .splice_read = copy_splice_read,
866 : .splice_write = iter_file_splice_write,
867 : .llseek = default_llseek,
868 : };
869 :
870 : static const struct file_operations proc_sys_dir_file_operations = {
871 : .read = generic_read_dir,
872 : .iterate_shared = proc_sys_readdir,
873 : .llseek = generic_file_llseek,
874 : };
875 :
876 : static const struct inode_operations proc_sys_inode_operations = {
877 : .permission = proc_sys_permission,
878 : .setattr = proc_sys_setattr,
879 : .getattr = proc_sys_getattr,
880 : };
881 :
882 : static const struct inode_operations proc_sys_dir_operations = {
883 : .lookup = proc_sys_lookup,
884 : .permission = proc_sys_permission,
885 : .setattr = proc_sys_setattr,
886 : .getattr = proc_sys_getattr,
887 : };
888 :
889 0 : static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
890 : {
891 0 : if (flags & LOOKUP_RCU)
892 : return -ECHILD;
893 0 : return !PROC_I(d_inode(dentry))->sysctl->unregistering;
894 : }
895 :
896 0 : static int proc_sys_delete(const struct dentry *dentry)
897 : {
898 0 : return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
899 : }
900 :
901 : static int sysctl_is_seen(struct ctl_table_header *p)
902 : {
903 0 : struct ctl_table_set *set = p->set;
904 : int res;
905 0 : spin_lock(&sysctl_lock);
906 0 : if (p->unregistering)
907 : res = 0;
908 0 : else if (!set->is_seen)
909 : res = 1;
910 : else
911 0 : res = set->is_seen(set);
912 0 : spin_unlock(&sysctl_lock);
913 : return res;
914 : }
915 :
916 0 : static int proc_sys_compare(const struct dentry *dentry,
917 : unsigned int len, const char *str, const struct qstr *name)
918 : {
919 : struct ctl_table_header *head;
920 : struct inode *inode;
921 :
922 : /* Although proc doesn't have negative dentries, rcu-walk means
923 : * that inode here can be NULL */
924 : /* AV: can it, indeed? */
925 0 : inode = d_inode_rcu(dentry);
926 0 : if (!inode)
927 : return 1;
928 0 : if (name->len != len)
929 : return 1;
930 0 : if (memcmp(name->name, str, len))
931 : return 1;
932 0 : head = rcu_dereference(PROC_I(inode)->sysctl);
933 0 : return !head || !sysctl_is_seen(head);
934 : }
935 :
936 : static const struct dentry_operations proc_sys_dentry_operations = {
937 : .d_revalidate = proc_sys_revalidate,
938 : .d_delete = proc_sys_delete,
939 : .d_compare = proc_sys_compare,
940 : };
941 :
942 59 : static struct ctl_dir *find_subdir(struct ctl_dir *dir,
943 : const char *name, int namelen)
944 : {
945 : struct ctl_table_header *head;
946 : struct ctl_table *entry;
947 :
948 59 : entry = find_entry(&head, dir, name, namelen);
949 59 : if (!entry)
950 : return ERR_PTR(-ENOENT);
951 31 : if (!S_ISDIR(entry->mode))
952 : return ERR_PTR(-ENOTDIR);
953 31 : return container_of(head, struct ctl_dir, header);
954 : }
955 :
956 14 : static struct ctl_dir *new_dir(struct ctl_table_set *set,
957 : const char *name, int namelen)
958 : {
959 : struct ctl_table *table;
960 : struct ctl_dir *new;
961 : struct ctl_node *node;
962 : char *new_name;
963 :
964 14 : new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
965 : sizeof(struct ctl_table)*2 + namelen + 1,
966 : GFP_KERNEL);
967 14 : if (!new)
968 : return NULL;
969 :
970 14 : node = (struct ctl_node *)(new + 1);
971 14 : table = (struct ctl_table *)(node + 1);
972 14 : new_name = (char *)(table + 2);
973 28 : memcpy(new_name, name, namelen);
974 14 : table[0].procname = new_name;
975 14 : table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
976 14 : init_header(&new->header, set->dir.header.root, set, node, table);
977 :
978 : return new;
979 : }
980 :
981 : /**
982 : * get_subdir - find or create a subdir with the specified name.
983 : * @dir: Directory to create the subdirectory in
984 : * @name: The name of the subdirectory to find or create
985 : * @namelen: The length of name
986 : *
987 : * Takes a directory with an elevated reference count so we know that
988 : * if we drop the lock the directory will not go away. Upon success
989 : * the reference is moved from @dir to the returned subdirectory.
990 : * Upon error an error code is returned and the reference on @dir is
991 : * simply dropped.
992 : */
993 44 : static struct ctl_dir *get_subdir(struct ctl_dir *dir,
994 : const char *name, int namelen)
995 : {
996 44 : struct ctl_table_set *set = dir->header.set;
997 44 : struct ctl_dir *subdir, *new = NULL;
998 : int err;
999 :
1000 44 : spin_lock(&sysctl_lock);
1001 44 : subdir = find_subdir(dir, name, namelen);
1002 44 : if (!IS_ERR(subdir))
1003 : goto found;
1004 14 : if (PTR_ERR(subdir) != -ENOENT)
1005 : goto failed;
1006 :
1007 14 : spin_unlock(&sysctl_lock);
1008 14 : new = new_dir(set, name, namelen);
1009 14 : spin_lock(&sysctl_lock);
1010 14 : subdir = ERR_PTR(-ENOMEM);
1011 14 : if (!new)
1012 : goto failed;
1013 :
1014 : /* Was the subdir added while we dropped the lock? */
1015 14 : subdir = find_subdir(dir, name, namelen);
1016 14 : if (!IS_ERR(subdir))
1017 : goto found;
1018 14 : if (PTR_ERR(subdir) != -ENOENT)
1019 : goto failed;
1020 :
1021 : /* Nope. Use the our freshly made directory entry. */
1022 14 : err = insert_header(dir, &new->header);
1023 28 : subdir = ERR_PTR(err);
1024 14 : if (err)
1025 : goto failed;
1026 : subdir = new;
1027 : found:
1028 44 : subdir->header.nreg++;
1029 : failed:
1030 44 : if (IS_ERR(subdir)) {
1031 0 : pr_err("sysctl could not get directory: ");
1032 0 : sysctl_print_dir(dir);
1033 0 : pr_cont("%*.*s %ld\n", namelen, namelen, name,
1034 : PTR_ERR(subdir));
1035 : }
1036 44 : drop_sysctl_table(&dir->header);
1037 44 : if (new)
1038 14 : drop_sysctl_table(&new->header);
1039 44 : spin_unlock(&sysctl_lock);
1040 44 : return subdir;
1041 : }
1042 :
1043 3 : static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
1044 : {
1045 : struct ctl_dir *parent;
1046 : const char *procname;
1047 3 : if (!dir->header.parent)
1048 2 : return &set->dir;
1049 1 : parent = xlate_dir(set, dir->header.parent);
1050 1 : if (IS_ERR(parent))
1051 : return parent;
1052 1 : procname = dir->header.ctl_table[0].procname;
1053 1 : return find_subdir(parent, procname, strlen(procname));
1054 : }
1055 :
1056 0 : static int sysctl_follow_link(struct ctl_table_header **phead,
1057 : struct ctl_table **pentry)
1058 : {
1059 : struct ctl_table_header *head;
1060 : struct ctl_table_root *root;
1061 : struct ctl_table_set *set;
1062 : struct ctl_table *entry;
1063 : struct ctl_dir *dir;
1064 : int ret;
1065 :
1066 0 : spin_lock(&sysctl_lock);
1067 0 : root = (*pentry)->data;
1068 0 : set = lookup_header_set(root);
1069 0 : dir = xlate_dir(set, (*phead)->parent);
1070 0 : if (IS_ERR(dir))
1071 0 : ret = PTR_ERR(dir);
1072 : else {
1073 0 : const char *procname = (*pentry)->procname;
1074 0 : head = NULL;
1075 0 : entry = find_entry(&head, dir, procname, strlen(procname));
1076 0 : ret = -ENOENT;
1077 0 : if (entry && use_table(head)) {
1078 0 : unuse_table(*phead);
1079 0 : *phead = head;
1080 0 : *pentry = entry;
1081 0 : ret = 0;
1082 : }
1083 : }
1084 :
1085 0 : spin_unlock(&sysctl_lock);
1086 0 : return ret;
1087 : }
1088 :
1089 0 : static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
1090 : {
1091 : struct va_format vaf;
1092 : va_list args;
1093 :
1094 0 : va_start(args, fmt);
1095 0 : vaf.fmt = fmt;
1096 0 : vaf.va = &args;
1097 :
1098 0 : pr_err("sysctl table check failed: %s/%s %pV\n",
1099 : path, table->procname, &vaf);
1100 :
1101 0 : va_end(args);
1102 0 : return -EINVAL;
1103 : }
1104 :
1105 78 : static int sysctl_check_table_array(const char *path, struct ctl_table *table)
1106 : {
1107 78 : int err = 0;
1108 :
1109 78 : if ((table->proc_handler == proc_douintvec) ||
1110 : (table->proc_handler == proc_douintvec_minmax)) {
1111 4 : if (table->maxlen != sizeof(unsigned int))
1112 0 : err |= sysctl_err(path, table, "array not allowed");
1113 : }
1114 :
1115 78 : if (table->proc_handler == proc_dou8vec_minmax) {
1116 0 : if (table->maxlen != sizeof(u8))
1117 0 : err |= sysctl_err(path, table, "array not allowed");
1118 : }
1119 :
1120 78 : if (table->proc_handler == proc_dobool) {
1121 1 : if (table->maxlen != sizeof(bool))
1122 0 : err |= sysctl_err(path, table, "array not allowed");
1123 : }
1124 :
1125 78 : return err;
1126 : }
1127 :
1128 37 : static int sysctl_check_table(const char *path, struct ctl_table *table)
1129 : {
1130 : struct ctl_table *entry;
1131 37 : int err = 0;
1132 162 : list_for_each_table_entry(entry, table) {
1133 125 : if ((entry->proc_handler == proc_dostring) ||
1134 122 : (entry->proc_handler == proc_dobool) ||
1135 100 : (entry->proc_handler == proc_dointvec) ||
1136 98 : (entry->proc_handler == proc_douintvec) ||
1137 96 : (entry->proc_handler == proc_douintvec_minmax) ||
1138 70 : (entry->proc_handler == proc_dointvec_minmax) ||
1139 70 : (entry->proc_handler == proc_dou8vec_minmax) ||
1140 68 : (entry->proc_handler == proc_dointvec_jiffies) ||
1141 68 : (entry->proc_handler == proc_dointvec_userhz_jiffies) ||
1142 68 : (entry->proc_handler == proc_dointvec_ms_jiffies) ||
1143 47 : (entry->proc_handler == proc_doulongvec_minmax) ||
1144 : (entry->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1145 78 : if (!entry->data)
1146 0 : err |= sysctl_err(path, entry, "No data");
1147 78 : if (!entry->maxlen)
1148 0 : err |= sysctl_err(path, entry, "No maxlen");
1149 : else
1150 78 : err |= sysctl_check_table_array(path, entry);
1151 : }
1152 125 : if (!entry->proc_handler)
1153 0 : err |= sysctl_err(path, entry, "No proc_handler");
1154 :
1155 125 : if ((entry->mode & (S_IRUGO|S_IWUGO)) != entry->mode)
1156 0 : err |= sysctl_err(path, entry, "bogus .mode 0%o",
1157 : entry->mode);
1158 : }
1159 37 : return err;
1160 : }
1161 :
1162 1 : static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table *table,
1163 : struct ctl_table_root *link_root)
1164 : {
1165 : struct ctl_table *link_table, *entry, *link;
1166 : struct ctl_table_header *links;
1167 : struct ctl_node *node;
1168 : char *link_name;
1169 : int nr_entries, name_bytes;
1170 :
1171 : name_bytes = 0;
1172 : nr_entries = 0;
1173 10 : list_for_each_table_entry(entry, table) {
1174 10 : nr_entries++;
1175 20 : name_bytes += strlen(entry->procname) + 1;
1176 : }
1177 :
1178 1 : links = kzalloc(sizeof(struct ctl_table_header) +
1179 2 : sizeof(struct ctl_node)*nr_entries +
1180 2 : sizeof(struct ctl_table)*(nr_entries + 1) +
1181 : name_bytes,
1182 : GFP_KERNEL);
1183 :
1184 1 : if (!links)
1185 : return NULL;
1186 :
1187 1 : node = (struct ctl_node *)(links + 1);
1188 1 : link_table = (struct ctl_table *)(node + nr_entries);
1189 1 : link_name = (char *)&link_table[nr_entries + 1];
1190 1 : link = link_table;
1191 :
1192 11 : list_for_each_table_entry(entry, table) {
1193 20 : int len = strlen(entry->procname) + 1;
1194 20 : memcpy(link_name, entry->procname, len);
1195 10 : link->procname = link_name;
1196 10 : link->mode = S_IFLNK|S_IRWXUGO;
1197 10 : link->data = link_root;
1198 10 : link_name += len;
1199 10 : link++;
1200 : }
1201 2 : init_header(links, dir->header.root, dir->header.set, node, link_table);
1202 1 : links->nreg = nr_entries;
1203 :
1204 : return links;
1205 : }
1206 :
1207 3 : static bool get_links(struct ctl_dir *dir,
1208 : struct ctl_table *table, struct ctl_table_root *link_root)
1209 : {
1210 : struct ctl_table_header *head;
1211 : struct ctl_table *entry, *link;
1212 :
1213 : /* Are there links available for every entry in table? */
1214 4 : list_for_each_table_entry(entry, table) {
1215 3 : const char *procname = entry->procname;
1216 3 : link = find_entry(&head, dir, procname, strlen(procname));
1217 3 : if (!link)
1218 : return false;
1219 1 : if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1220 1 : continue;
1221 0 : if (S_ISLNK(link->mode) && (link->data == link_root))
1222 0 : continue;
1223 : return false;
1224 : }
1225 :
1226 : /* The checks passed. Increase the registration count on the links */
1227 1 : list_for_each_table_entry(entry, table) {
1228 1 : const char *procname = entry->procname;
1229 1 : link = find_entry(&head, dir, procname, strlen(procname));
1230 1 : head->nreg++;
1231 : }
1232 : return true;
1233 : }
1234 :
1235 52 : static int insert_links(struct ctl_table_header *head)
1236 : {
1237 52 : struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1238 : struct ctl_dir *core_parent;
1239 : struct ctl_table_header *links;
1240 : int err;
1241 :
1242 52 : if (head->set == root_set)
1243 : return 0;
1244 :
1245 2 : core_parent = xlate_dir(root_set, head->parent);
1246 2 : if (IS_ERR(core_parent))
1247 : return 0;
1248 :
1249 2 : if (get_links(core_parent, head->ctl_table, head->root))
1250 : return 0;
1251 :
1252 1 : core_parent->header.nreg++;
1253 1 : spin_unlock(&sysctl_lock);
1254 :
1255 1 : links = new_links(core_parent, head->ctl_table, head->root);
1256 :
1257 1 : spin_lock(&sysctl_lock);
1258 1 : err = -ENOMEM;
1259 1 : if (!links)
1260 : goto out;
1261 :
1262 1 : err = 0;
1263 1 : if (get_links(core_parent, head->ctl_table, head->root)) {
1264 0 : kfree(links);
1265 0 : goto out;
1266 : }
1267 :
1268 1 : err = insert_header(core_parent, links);
1269 1 : if (err)
1270 0 : kfree(links);
1271 : out:
1272 1 : drop_sysctl_table(&core_parent->header);
1273 1 : return err;
1274 : }
1275 :
1276 : /* Find the directory for the ctl_table. If one is not found create it. */
1277 37 : static struct ctl_dir *sysctl_mkdir_p(struct ctl_dir *dir, const char *path)
1278 : {
1279 : const char *name, *nextname;
1280 :
1281 118 : for (name = path; name; name = nextname) {
1282 : int namelen;
1283 44 : nextname = strchr(name, '/');
1284 44 : if (nextname) {
1285 7 : namelen = nextname - name;
1286 7 : nextname++;
1287 : } else {
1288 37 : namelen = strlen(name);
1289 : }
1290 44 : if (namelen == 0)
1291 0 : continue;
1292 :
1293 : /*
1294 : * namelen ensures if name is "foo/bar/yay" only foo is
1295 : * registered first. We traverse as if using mkdir -p and
1296 : * return a ctl_dir for the last directory entry.
1297 : */
1298 44 : dir = get_subdir(dir, name, namelen);
1299 44 : if (IS_ERR(dir))
1300 : break;
1301 : }
1302 37 : return dir;
1303 : }
1304 :
1305 : /**
1306 : * __register_sysctl_table - register a leaf sysctl table
1307 : * @set: Sysctl tree to register on
1308 : * @path: The path to the directory the sysctl table is in.
1309 : * @table: the top-level table structure without any child. This table
1310 : * should not be free'd after registration. So it should not be
1311 : * used on stack. It can either be a global or dynamically allocated
1312 : * by the caller and free'd later after sysctl unregistration.
1313 : *
1314 : * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1315 : * array. A completely 0 filled entry terminates the table.
1316 : *
1317 : * The members of the &struct ctl_table structure are used as follows:
1318 : *
1319 : * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1320 : * enter a sysctl file
1321 : *
1322 : * data - a pointer to data for use by proc_handler
1323 : *
1324 : * maxlen - the maximum size in bytes of the data
1325 : *
1326 : * mode - the file permissions for the /proc/sys file
1327 : *
1328 : * child - must be %NULL.
1329 : *
1330 : * proc_handler - the text handler routine (described below)
1331 : *
1332 : * extra1, extra2 - extra pointers usable by the proc handler routines
1333 : * XXX: we should eventually modify these to use long min / max [0]
1334 : * [0] https://lkml.kernel.org/87zgpte9o4.fsf@email.froward.int.ebiederm.org
1335 : *
1336 : * Leaf nodes in the sysctl tree will be represented by a single file
1337 : * under /proc; non-leaf nodes (where child is not NULL) are not allowed,
1338 : * sysctl_check_table() verifies this.
1339 : *
1340 : * There must be a proc_handler routine for any terminal nodes.
1341 : * Several default handlers are available to cover common cases -
1342 : *
1343 : * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1344 : * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1345 : * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1346 : *
1347 : * It is the handler's job to read the input buffer from user memory
1348 : * and process it. The handler should return 0 on success.
1349 : *
1350 : * This routine returns %NULL on a failure to register, and a pointer
1351 : * to the table header on success.
1352 : */
1353 37 : struct ctl_table_header *__register_sysctl_table(
1354 : struct ctl_table_set *set,
1355 : const char *path, struct ctl_table *table)
1356 : {
1357 37 : struct ctl_table_root *root = set->dir.header.root;
1358 : struct ctl_table_header *header;
1359 : struct ctl_dir *dir;
1360 : struct ctl_table *entry;
1361 : struct ctl_node *node;
1362 37 : int nr_entries = 0;
1363 :
1364 162 : list_for_each_table_entry(entry, table)
1365 125 : nr_entries++;
1366 :
1367 37 : header = kzalloc(sizeof(struct ctl_table_header) +
1368 37 : sizeof(struct ctl_node)*nr_entries, GFP_KERNEL_ACCOUNT);
1369 37 : if (!header)
1370 : return NULL;
1371 :
1372 37 : node = (struct ctl_node *)(header + 1);
1373 37 : init_header(header, root, set, node, table);
1374 37 : if (sysctl_check_table(path, table))
1375 : goto fail;
1376 :
1377 37 : spin_lock(&sysctl_lock);
1378 37 : dir = &set->dir;
1379 : /* Reference moved down the directory tree get_subdir */
1380 37 : dir->header.nreg++;
1381 37 : spin_unlock(&sysctl_lock);
1382 :
1383 37 : dir = sysctl_mkdir_p(dir, path);
1384 37 : if (IS_ERR(dir))
1385 : goto fail;
1386 37 : spin_lock(&sysctl_lock);
1387 37 : if (insert_header(dir, header))
1388 : goto fail_put_dir_locked;
1389 :
1390 37 : drop_sysctl_table(&dir->header);
1391 37 : spin_unlock(&sysctl_lock);
1392 :
1393 37 : return header;
1394 :
1395 : fail_put_dir_locked:
1396 0 : drop_sysctl_table(&dir->header);
1397 : spin_unlock(&sysctl_lock);
1398 : fail:
1399 0 : kfree(header);
1400 0 : return NULL;
1401 : }
1402 :
1403 : /**
1404 : * register_sysctl - register a sysctl table
1405 : * @path: The path to the directory the sysctl table is in. If the path
1406 : * doesn't exist we will create it for you.
1407 : * @table: the table structure. The calller must ensure the life of the @table
1408 : * will be kept during the lifetime use of the syctl. It must not be freed
1409 : * until unregister_sysctl_table() is called with the given returned table
1410 : * with this registration. If your code is non modular then you don't need
1411 : * to call unregister_sysctl_table() and can instead use something like
1412 : * register_sysctl_init() which does not care for the result of the syctl
1413 : * registration.
1414 : *
1415 : * Register a sysctl table. @table should be a filled in ctl_table
1416 : * array. A completely 0 filled entry terminates the table.
1417 : *
1418 : * See __register_sysctl_table for more details.
1419 : */
1420 5 : struct ctl_table_header *register_sysctl(const char *path, struct ctl_table *table)
1421 : {
1422 36 : return __register_sysctl_table(&sysctl_table_root.default_set,
1423 : path, table);
1424 : }
1425 : EXPORT_SYMBOL(register_sysctl);
1426 :
1427 : /**
1428 : * __register_sysctl_init() - register sysctl table to path
1429 : * @path: path name for sysctl base. If that path doesn't exist we will create
1430 : * it for you.
1431 : * @table: This is the sysctl table that needs to be registered to the path.
1432 : * The caller must ensure the life of the @table will be kept during the
1433 : * lifetime use of the sysctl.
1434 : * @table_name: The name of sysctl table, only used for log printing when
1435 : * registration fails
1436 : *
1437 : * The sysctl interface is used by userspace to query or modify at runtime
1438 : * a predefined value set on a variable. These variables however have default
1439 : * values pre-set. Code which depends on these variables will always work even
1440 : * if register_sysctl() fails. If register_sysctl() fails you'd just loose the
1441 : * ability to query or modify the sysctls dynamically at run time. Chances of
1442 : * register_sysctl() failing on init are extremely low, and so for both reasons
1443 : * this function does not return any error as it is used by initialization code.
1444 : *
1445 : * Context: if your base directory does not exist it will be created for you.
1446 : */
1447 31 : void __init __register_sysctl_init(const char *path, struct ctl_table *table,
1448 : const char *table_name)
1449 : {
1450 31 : struct ctl_table_header *hdr = register_sysctl(path, table);
1451 :
1452 31 : if (unlikely(!hdr)) {
1453 0 : pr_err("failed when register_sysctl %s to %s\n", table_name, path);
1454 0 : return;
1455 : }
1456 : kmemleak_not_leak(hdr);
1457 : }
1458 :
1459 0 : static void put_links(struct ctl_table_header *header)
1460 : {
1461 0 : struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1462 0 : struct ctl_table_root *root = header->root;
1463 0 : struct ctl_dir *parent = header->parent;
1464 : struct ctl_dir *core_parent;
1465 : struct ctl_table *entry;
1466 :
1467 0 : if (header->set == root_set)
1468 : return;
1469 :
1470 0 : core_parent = xlate_dir(root_set, parent);
1471 0 : if (IS_ERR(core_parent))
1472 : return;
1473 :
1474 0 : list_for_each_table_entry(entry, header->ctl_table) {
1475 : struct ctl_table_header *link_head;
1476 : struct ctl_table *link;
1477 0 : const char *name = entry->procname;
1478 :
1479 0 : link = find_entry(&link_head, core_parent, name, strlen(name));
1480 0 : if (link &&
1481 0 : ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1482 0 : (S_ISLNK(link->mode) && (link->data == root)))) {
1483 0 : drop_sysctl_table(link_head);
1484 : }
1485 : else {
1486 0 : pr_err("sysctl link missing during unregister: ");
1487 0 : sysctl_print_dir(parent);
1488 0 : pr_cont("%s\n", name);
1489 : }
1490 : }
1491 : }
1492 :
1493 96 : static void drop_sysctl_table(struct ctl_table_header *header)
1494 : {
1495 96 : struct ctl_dir *parent = header->parent;
1496 :
1497 96 : if (--header->nreg)
1498 : return;
1499 :
1500 0 : if (parent) {
1501 0 : put_links(header);
1502 0 : start_unregistering(header);
1503 : }
1504 :
1505 0 : if (!--header->count)
1506 0 : kfree_rcu(header, rcu);
1507 :
1508 0 : if (parent)
1509 0 : drop_sysctl_table(&parent->header);
1510 : }
1511 :
1512 : /**
1513 : * unregister_sysctl_table - unregister a sysctl table hierarchy
1514 : * @header: the header returned from register_sysctl or __register_sysctl_table
1515 : *
1516 : * Unregisters the sysctl table and all children. proc entries may not
1517 : * actually be removed until they are no longer used by anyone.
1518 : */
1519 0 : void unregister_sysctl_table(struct ctl_table_header * header)
1520 : {
1521 : might_sleep();
1522 :
1523 0 : if (header == NULL)
1524 : return;
1525 :
1526 0 : spin_lock(&sysctl_lock);
1527 0 : drop_sysctl_table(header);
1528 : spin_unlock(&sysctl_lock);
1529 : }
1530 : EXPORT_SYMBOL(unregister_sysctl_table);
1531 :
1532 1 : void setup_sysctl_set(struct ctl_table_set *set,
1533 : struct ctl_table_root *root,
1534 : int (*is_seen)(struct ctl_table_set *))
1535 : {
1536 2 : memset(set, 0, sizeof(*set));
1537 1 : set->is_seen = is_seen;
1538 2 : init_header(&set->dir.header, root, set, NULL, root_table);
1539 1 : }
1540 :
1541 0 : void retire_sysctl_set(struct ctl_table_set *set)
1542 : {
1543 0 : WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1544 0 : }
1545 :
1546 1 : int __init proc_sys_init(void)
1547 : {
1548 : struct proc_dir_entry *proc_sys_root;
1549 :
1550 1 : proc_sys_root = proc_mkdir("sys", NULL);
1551 1 : proc_sys_root->proc_iops = &proc_sys_dir_operations;
1552 1 : proc_sys_root->proc_dir_ops = &proc_sys_dir_file_operations;
1553 1 : proc_sys_root->nlink = 0;
1554 :
1555 1 : return sysctl_init_bases();
1556 : }
1557 :
1558 : struct sysctl_alias {
1559 : const char *kernel_param;
1560 : const char *sysctl_param;
1561 : };
1562 :
1563 : /*
1564 : * Historically some settings had both sysctl and a command line parameter.
1565 : * With the generic sysctl. parameter support, we can handle them at a single
1566 : * place and only keep the historical name for compatibility. This is not meant
1567 : * to add brand new aliases. When adding existing aliases, consider whether
1568 : * the possibly different moment of changing the value (e.g. from early_param
1569 : * to the moment do_sysctl_args() is called) is an issue for the specific
1570 : * parameter.
1571 : */
1572 : static const struct sysctl_alias sysctl_aliases[] = {
1573 : {"hardlockup_all_cpu_backtrace", "kernel.hardlockup_all_cpu_backtrace" },
1574 : {"hung_task_panic", "kernel.hung_task_panic" },
1575 : {"numa_zonelist_order", "vm.numa_zonelist_order" },
1576 : {"softlockup_all_cpu_backtrace", "kernel.softlockup_all_cpu_backtrace" },
1577 : {"softlockup_panic", "kernel.softlockup_panic" },
1578 : { }
1579 : };
1580 :
1581 0 : static const char *sysctl_find_alias(char *param)
1582 : {
1583 : const struct sysctl_alias *alias;
1584 :
1585 0 : for (alias = &sysctl_aliases[0]; alias->kernel_param != NULL; alias++) {
1586 0 : if (strcmp(alias->kernel_param, param) == 0)
1587 0 : return alias->sysctl_param;
1588 : }
1589 :
1590 : return NULL;
1591 : }
1592 :
1593 : /* Set sysctl value passed on kernel command line. */
1594 0 : static int process_sysctl_arg(char *param, char *val,
1595 : const char *unused, void *arg)
1596 : {
1597 : char *path;
1598 0 : struct vfsmount **proc_mnt = arg;
1599 : struct file_system_type *proc_fs_type;
1600 : struct file *file;
1601 : int len;
1602 : int err;
1603 0 : loff_t pos = 0;
1604 : ssize_t wret;
1605 :
1606 0 : if (strncmp(param, "sysctl", sizeof("sysctl") - 1) == 0) {
1607 0 : param += sizeof("sysctl") - 1;
1608 :
1609 0 : if (param[0] != '/' && param[0] != '.')
1610 : return 0;
1611 :
1612 0 : param++;
1613 : } else {
1614 0 : param = (char *) sysctl_find_alias(param);
1615 0 : if (!param)
1616 : return 0;
1617 : }
1618 :
1619 0 : if (!val)
1620 : return -EINVAL;
1621 0 : len = strlen(val);
1622 0 : if (len == 0)
1623 : return -EINVAL;
1624 :
1625 : /*
1626 : * To set sysctl options, we use a temporary mount of proc, look up the
1627 : * respective sys/ file and write to it. To avoid mounting it when no
1628 : * options were given, we mount it only when the first sysctl option is
1629 : * found. Why not a persistent mount? There are problems with a
1630 : * persistent mount of proc in that it forces userspace not to use any
1631 : * proc mount options.
1632 : */
1633 0 : if (!*proc_mnt) {
1634 0 : proc_fs_type = get_fs_type("proc");
1635 0 : if (!proc_fs_type) {
1636 0 : pr_err("Failed to find procfs to set sysctl from command line\n");
1637 0 : return 0;
1638 : }
1639 0 : *proc_mnt = kern_mount(proc_fs_type);
1640 0 : put_filesystem(proc_fs_type);
1641 0 : if (IS_ERR(*proc_mnt)) {
1642 0 : pr_err("Failed to mount procfs to set sysctl from command line\n");
1643 0 : return 0;
1644 : }
1645 : }
1646 :
1647 0 : path = kasprintf(GFP_KERNEL, "sys/%s", param);
1648 0 : if (!path)
1649 0 : panic("%s: Failed to allocate path for %s\n", __func__, param);
1650 0 : strreplace(path, '.', '/');
1651 :
1652 0 : file = file_open_root_mnt(*proc_mnt, path, O_WRONLY, 0);
1653 0 : if (IS_ERR(file)) {
1654 0 : err = PTR_ERR(file);
1655 0 : if (err == -ENOENT)
1656 0 : pr_err("Failed to set sysctl parameter '%s=%s': parameter not found\n",
1657 : param, val);
1658 0 : else if (err == -EACCES)
1659 0 : pr_err("Failed to set sysctl parameter '%s=%s': permission denied (read-only?)\n",
1660 : param, val);
1661 : else
1662 0 : pr_err("Error %pe opening proc file to set sysctl parameter '%s=%s'\n",
1663 : file, param, val);
1664 : goto out;
1665 : }
1666 0 : wret = kernel_write(file, val, len, &pos);
1667 0 : if (wret < 0) {
1668 0 : err = wret;
1669 0 : if (err == -EINVAL)
1670 0 : pr_err("Failed to set sysctl parameter '%s=%s': invalid value\n",
1671 : param, val);
1672 : else
1673 0 : pr_err("Error %pe writing to proc file to set sysctl parameter '%s=%s'\n",
1674 : ERR_PTR(err), param, val);
1675 0 : } else if (wret != len) {
1676 0 : pr_err("Wrote only %zd bytes of %d writing to proc file %s to set sysctl parameter '%s=%s\n",
1677 : wret, len, path, param, val);
1678 : }
1679 :
1680 0 : err = filp_close(file, NULL);
1681 0 : if (err)
1682 0 : pr_err("Error %pe closing proc file to set sysctl parameter '%s=%s\n",
1683 : ERR_PTR(err), param, val);
1684 : out:
1685 0 : kfree(path);
1686 0 : return 0;
1687 : }
1688 :
1689 0 : void do_sysctl_args(void)
1690 : {
1691 : char *command_line;
1692 0 : struct vfsmount *proc_mnt = NULL;
1693 :
1694 0 : command_line = kstrdup(saved_command_line, GFP_KERNEL);
1695 0 : if (!command_line)
1696 0 : panic("%s: Failed to allocate copy of command line\n", __func__);
1697 :
1698 0 : parse_args("Setting sysctl args", command_line,
1699 : NULL, 0, -1, -1, &proc_mnt, process_sysctl_arg);
1700 :
1701 0 : if (proc_mnt)
1702 0 : kern_unmount(proc_mnt);
1703 :
1704 0 : kfree(command_line);
1705 0 : }
|
