]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - net/netlink/af_netlink.c
projects / mirror_ubuntu-jammy-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
af_netlink: Fix shift out of bounds in group mask calculation
[mirror_ubuntu-jammy-kernel.git] / net / netlink / af_netlink.c
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * NETLINK Kernel-user communication protocol.
4 *
5 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
6 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
7 * Patrick McHardy <kaber@trash.net>
8 *
9 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
10 * added netlink_proto_exit
11 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
12 * use nlk_sk, as sk->protinfo is on a diet 8)
13 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
14 * - inc module use count of module that owns
15 * the kernel socket in case userspace opens
16 * socket of same protocol
17 * - remove all module support, since netlink is
18 * mandatory if CONFIG_NET=y these days
19 */
20
21 #include <linux/module.h>
22
23 #include <linux/capability.h>
24 #include <linux/kernel.h>
25 #include <linux/init.h>
26 #include <linux/signal.h>
27 #include <linux/sched.h>
28 #include <linux/errno.h>
29 #include <linux/string.h>
30 #include <linux/stat.h>
31 #include <linux/socket.h>
32 #include <linux/un.h>
33 #include <linux/fcntl.h>
34 #include <linux/termios.h>
35 #include <linux/sockios.h>
36 #include <linux/net.h>
37 #include <linux/fs.h>
38 #include <linux/slab.h>
39 #include <linux/uaccess.h>
40 #include <linux/skbuff.h>
41 #include <linux/netdevice.h>
42 #include <linux/rtnetlink.h>
43 #include <linux/proc_fs.h>
44 #include <linux/seq_file.h>
45 #include <linux/notifier.h>
46 #include <linux/security.h>
47 #include <linux/jhash.h>
48 #include <linux/jiffies.h>
49 #include <linux/random.h>
50 #include <linux/bitops.h>
51 #include <linux/mm.h>
52 #include <linux/types.h>
53 #include <linux/audit.h>
54 #include <linux/mutex.h>
55 #include <linux/vmalloc.h>
56 #include <linux/if_arp.h>
57 #include <linux/rhashtable.h>
58 #include <asm/cacheflush.h>
59 #include <linux/hash.h>
60 #include <linux/genetlink.h>
61 #include <linux/net_namespace.h>
62 #include <linux/nospec.h>
63 #include <linux/btf_ids.h>
64
65 #include <net/net_namespace.h>
66 #include <net/netns/generic.h>
67 #include <net/sock.h>
68 #include <net/scm.h>
69 #include <net/netlink.h>
70 #define CREATE_TRACE_POINTS
71 #include <trace/events/netlink.h>
72
73 #include "af_netlink.h"
74
75 struct listeners {
76 struct rcu_head rcu;
77 unsigned long masks[];
78 };
79
80 /* state bits */
81 #define NETLINK_S_CONGESTED 0x0
82
83 static inline int netlink_is_kernel(struct sock *sk)
84 {
85 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
86 }
87
88 struct netlink_table *nl_table __read_mostly;
89 EXPORT_SYMBOL_GPL(nl_table);
90
91 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
92
93 static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
94
95 static const char *const nlk_cb_mutex_key_strings[MAX_LINKS + 1] = {
96 "nlk_cb_mutex-ROUTE",
97 "nlk_cb_mutex-1",
98 "nlk_cb_mutex-USERSOCK",
99 "nlk_cb_mutex-FIREWALL",
100 "nlk_cb_mutex-SOCK_DIAG",
101 "nlk_cb_mutex-NFLOG",
102 "nlk_cb_mutex-XFRM",
103 "nlk_cb_mutex-SELINUX",
104 "nlk_cb_mutex-ISCSI",
105 "nlk_cb_mutex-AUDIT",
106 "nlk_cb_mutex-FIB_LOOKUP",
107 "nlk_cb_mutex-CONNECTOR",
108 "nlk_cb_mutex-NETFILTER",
109 "nlk_cb_mutex-IP6_FW",
110 "nlk_cb_mutex-DNRTMSG",
111 "nlk_cb_mutex-KOBJECT_UEVENT",
112 "nlk_cb_mutex-GENERIC",
113 "nlk_cb_mutex-17",
114 "nlk_cb_mutex-SCSITRANSPORT",
115 "nlk_cb_mutex-ECRYPTFS",
116 "nlk_cb_mutex-RDMA",
117 "nlk_cb_mutex-CRYPTO",
118 "nlk_cb_mutex-SMC",
119 "nlk_cb_mutex-23",
120 "nlk_cb_mutex-24",
121 "nlk_cb_mutex-25",
122 "nlk_cb_mutex-26",
123 "nlk_cb_mutex-27",
124 "nlk_cb_mutex-28",
125 "nlk_cb_mutex-29",
126 "nlk_cb_mutex-30",
127 "nlk_cb_mutex-31",
128 "nlk_cb_mutex-MAX_LINKS"
129 };
130
131 static int netlink_dump(struct sock *sk);
132
133 /* nl_table locking explained:
134 * Lookup and traversal are protected with an RCU read-side lock. Insertion
135 * and removal are protected with per bucket lock while using RCU list
136 * modification primitives and may run in parallel to RCU protected lookups.
137 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
138 * been acquired * either during or after the socket has been removed from
139 * the list and after an RCU grace period.
140 */
141 DEFINE_RWLOCK(nl_table_lock);
142 EXPORT_SYMBOL_GPL(nl_table_lock);
143 static atomic_t nl_table_users = ATOMIC_INIT(0);
144
145 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
146
147 static BLOCKING_NOTIFIER_HEAD(netlink_chain);
148
149
150 static const struct rhashtable_params netlink_rhashtable_params;
151
152 void do_trace_netlink_extack(const char *msg)
153 {
154 trace_netlink_extack(msg);
155 }
156 EXPORT_SYMBOL(do_trace_netlink_extack);
157
158 static inline u32 netlink_group_mask(u32 group)
159 {
160 if (group > 32)
161 return 0;
162 return group ? 1 << (group - 1) : 0;
163 }
164
165 static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
166 gfp_t gfp_mask)
167 {
168 unsigned int len = skb_end_offset(skb);
169 struct sk_buff *new;
170
171 new = alloc_skb(len, gfp_mask);
172 if (new == NULL)
173 return NULL;
174
175 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
176 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
177 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
178
179 skb_put_data(new, skb->data, len);
180 return new;
181 }
182
183 static unsigned int netlink_tap_net_id;
184
185 struct netlink_tap_net {
186 struct list_head netlink_tap_all;
187 struct mutex netlink_tap_lock;
188 };
189
190 int netlink_add_tap(struct netlink_tap *nt)
191 {
192 struct net *net = dev_net(nt->dev);
193 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
194
195 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
196 return -EINVAL;
197
198 mutex_lock(&nn->netlink_tap_lock);
199 list_add_rcu(&nt->list, &nn->netlink_tap_all);
200 mutex_unlock(&nn->netlink_tap_lock);
201
202 __module_get(nt->module);
203
204 return 0;
205 }
206 EXPORT_SYMBOL_GPL(netlink_add_tap);
207
208 static int __netlink_remove_tap(struct netlink_tap *nt)
209 {
210 struct net *net = dev_net(nt->dev);
211 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
212 bool found = false;
213 struct netlink_tap *tmp;
214
215 mutex_lock(&nn->netlink_tap_lock);
216
217 list_for_each_entry(tmp, &nn->netlink_tap_all, list) {
218 if (nt == tmp) {
219 list_del_rcu(&nt->list);
220 found = true;
221 goto out;
222 }
223 }
224
225 pr_warn("__netlink_remove_tap: %p not found\n", nt);
226 out:
227 mutex_unlock(&nn->netlink_tap_lock);
228
229 if (found)
230 module_put(nt->module);
231
232 return found ? 0 : -ENODEV;
233 }
234
235 int netlink_remove_tap(struct netlink_tap *nt)
236 {
237 int ret;
238
239 ret = __netlink_remove_tap(nt);
240 synchronize_net();
241
242 return ret;
243 }
244 EXPORT_SYMBOL_GPL(netlink_remove_tap);
245
246 static __net_init int netlink_tap_init_net(struct net *net)
247 {
248 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
249
250 INIT_LIST_HEAD(&nn->netlink_tap_all);
251 mutex_init(&nn->netlink_tap_lock);
252 return 0;
253 }
254
255 static struct pernet_operations netlink_tap_net_ops = {
256 .init = netlink_tap_init_net,
257 .id = &netlink_tap_net_id,
258 .size = sizeof(struct netlink_tap_net),
259 };
260
261 static bool netlink_filter_tap(const struct sk_buff *skb)
262 {
263 struct sock *sk = skb->sk;
264
265 /* We take the more conservative approach and
266 * whitelist socket protocols that may pass.
267 */
268 switch (sk->sk_protocol) {
269 case NETLINK_ROUTE:
270 case NETLINK_USERSOCK:
271 case NETLINK_SOCK_DIAG:
272 case NETLINK_NFLOG:
273 case NETLINK_XFRM:
274 case NETLINK_FIB_LOOKUP:
275 case NETLINK_NETFILTER:
276 case NETLINK_GENERIC:
277 return true;
278 }
279
280 return false;
281 }
282
283 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
284 struct net_device *dev)
285 {
286 struct sk_buff *nskb;
287 struct sock *sk = skb->sk;
288 int ret = -ENOMEM;
289
290 if (!net_eq(dev_net(dev), sock_net(sk)))
291 return 0;
292
293 dev_hold(dev);
294
295 if (is_vmalloc_addr(skb->head))
296 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
297 else
298 nskb = skb_clone(skb, GFP_ATOMIC);
299 if (nskb) {
300 nskb->dev = dev;
301 nskb->protocol = htons((u16) sk->sk_protocol);
302 nskb->pkt_type = netlink_is_kernel(sk) ?
303 PACKET_KERNEL : PACKET_USER;
304 skb_reset_network_header(nskb);
305 ret = dev_queue_xmit(nskb);
306 if (unlikely(ret > 0))
307 ret = net_xmit_errno(ret);
308 }
309
310 dev_put(dev);
311 return ret;
312 }
313
314 static void __netlink_deliver_tap(struct sk_buff *skb, struct netlink_tap_net *nn)
315 {
316 int ret;
317 struct netlink_tap *tmp;
318
319 if (!netlink_filter_tap(skb))
320 return;
321
322 list_for_each_entry_rcu(tmp, &nn->netlink_tap_all, list) {
323 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
324 if (unlikely(ret))
325 break;
326 }
327 }
328
329 static void netlink_deliver_tap(struct net *net, struct sk_buff *skb)
330 {
331 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
332
333 rcu_read_lock();
334
335 if (unlikely(!list_empty(&nn->netlink_tap_all)))
336 __netlink_deliver_tap(skb, nn);
337
338 rcu_read_unlock();
339 }
340
341 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
342 struct sk_buff *skb)
343 {
344 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
345 netlink_deliver_tap(sock_net(dst), skb);
346 }
347
348 static void netlink_overrun(struct sock *sk)
349 {
350 struct netlink_sock *nlk = nlk_sk(sk);
351
352 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
353 if (!test_and_set_bit(NETLINK_S_CONGESTED,
354 &nlk_sk(sk)->state)) {
355 sk->sk_err = ENOBUFS;
356 sk_error_report(sk);
357 }
358 }
359 atomic_inc(&sk->sk_drops);
360 }
361
362 static void netlink_rcv_wake(struct sock *sk)
363 {
364 struct netlink_sock *nlk = nlk_sk(sk);
365
366 if (skb_queue_empty_lockless(&sk->sk_receive_queue))
367 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
368 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
369 wake_up_interruptible(&nlk->wait);
370 }
371
372 static void netlink_skb_destructor(struct sk_buff *skb)
373 {
374 if (is_vmalloc_addr(skb->head)) {
375 if (!skb->cloned ||
376 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
377 vfree(skb->head);
378
379 skb->head = NULL;
380 }
381 if (skb->sk != NULL)
382 sock_rfree(skb);
383 }
384
385 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
386 {
387 WARN_ON(skb->sk != NULL);
388 skb->sk = sk;
389 skb->destructor = netlink_skb_destructor;
390 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
391 sk_mem_charge(sk, skb->truesize);
392 }
393
394 static void netlink_sock_destruct(struct sock *sk)
395 {
396 struct netlink_sock *nlk = nlk_sk(sk);
397
398 if (nlk->cb_running) {
399 if (nlk->cb.done)
400 nlk->cb.done(&nlk->cb);
401 module_put(nlk->cb.module);
402 kfree_skb(nlk->cb.skb);
403 }
404
405 skb_queue_purge(&sk->sk_receive_queue);
406
407 if (!sock_flag(sk, SOCK_DEAD)) {
408 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
409 return;
410 }
411
412 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
413 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
414 WARN_ON(nlk_sk(sk)->groups);
415 }
416
417 static void netlink_sock_destruct_work(struct work_struct *work)
418 {
419 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
420 work);
421
422 sk_free(&nlk->sk);
423 }
424
425 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
426 * SMP. Look, when several writers sleep and reader wakes them up, all but one
427 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
428 * this, _but_ remember, it adds useless work on UP machines.
429 */
430
431 void netlink_table_grab(void)
432 __acquires(nl_table_lock)
433 {
434 might_sleep();
435
436 write_lock_irq(&nl_table_lock);
437
438 if (atomic_read(&nl_table_users)) {
439 DECLARE_WAITQUEUE(wait, current);
440
441 add_wait_queue_exclusive(&nl_table_wait, &wait);
442 for (;;) {
443 set_current_state(TASK_UNINTERRUPTIBLE);
444 if (atomic_read(&nl_table_users) == 0)
445 break;
446 write_unlock_irq(&nl_table_lock);
447 schedule();
448 write_lock_irq(&nl_table_lock);
449 }
450
451 __set_current_state(TASK_RUNNING);
452 remove_wait_queue(&nl_table_wait, &wait);
453 }
454 }
455
456 void netlink_table_ungrab(void)
457 __releases(nl_table_lock)
458 {
459 write_unlock_irq(&nl_table_lock);
460 wake_up(&nl_table_wait);
461 }
462
463 static inline void
464 netlink_lock_table(void)
465 {
466 unsigned long flags;
467
468 /* read_lock() synchronizes us to netlink_table_grab */
469
470 read_lock_irqsave(&nl_table_lock, flags);
471 atomic_inc(&nl_table_users);
472 read_unlock_irqrestore(&nl_table_lock, flags);
473 }
474
475 static inline void
476 netlink_unlock_table(void)
477 {
478 if (atomic_dec_and_test(&nl_table_users))
479 wake_up(&nl_table_wait);
480 }
481
482 struct netlink_compare_arg
483 {
484 possible_net_t pnet;
485 u32 portid;
486 };
487
488 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
489 #define netlink_compare_arg_len \
490 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
491
492 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
493 const void *ptr)
494 {
495 const struct netlink_compare_arg *x = arg->key;
496 const struct netlink_sock *nlk = ptr;
497
498 return nlk->portid != x->portid ||
499 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
500 }
501
502 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
503 struct net *net, u32 portid)
504 {
505 memset(arg, 0, sizeof(*arg));
506 write_pnet(&arg->pnet, net);
507 arg->portid = portid;
508 }
509
510 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
511 struct net *net)
512 {
513 struct netlink_compare_arg arg;
514
515 netlink_compare_arg_init(&arg, net, portid);
516 return rhashtable_lookup_fast(&table->hash, &arg,
517 netlink_rhashtable_params);
518 }
519
520 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
521 {
522 struct netlink_compare_arg arg;
523
524 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
525 return rhashtable_lookup_insert_key(&table->hash, &arg,
526 &nlk_sk(sk)->node,
527 netlink_rhashtable_params);
528 }
529
530 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
531 {
532 struct netlink_table *table = &nl_table[protocol];
533 struct sock *sk;
534
535 rcu_read_lock();
536 sk = __netlink_lookup(table, portid, net);
537 if (sk)
538 sock_hold(sk);
539 rcu_read_unlock();
540
541 return sk;
542 }
543
544 static const struct proto_ops netlink_ops;
545
546 static void
547 netlink_update_listeners(struct sock *sk)
548 {
549 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
550 unsigned long mask;
551 unsigned int i;
552 struct listeners *listeners;
553
554 listeners = nl_deref_protected(tbl->listeners);
555 if (!listeners)
556 return;
557
558 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
559 mask = 0;
560 sk_for_each_bound(sk, &tbl->mc_list) {
561 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
562 mask |= nlk_sk(sk)->groups[i];
563 }
564 listeners->masks[i] = mask;
565 }
566 /* this function is only called with the netlink table "grabbed", which
567 * makes sure updates are visible before bind or setsockopt return. */
568 }
569
570 static int netlink_insert(struct sock *sk, u32 portid)
571 {
572 struct netlink_table *table = &nl_table[sk->sk_protocol];
573 int err;
574
575 lock_sock(sk);
576
577 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
578 if (nlk_sk(sk)->bound)
579 goto err;
580
581 nlk_sk(sk)->portid = portid;
582 sock_hold(sk);
583
584 err = __netlink_insert(table, sk);
585 if (err) {
586 /* In case the hashtable backend returns with -EBUSY
587 * from here, it must not escape to the caller.
588 */
589 if (unlikely(err == -EBUSY))
590 err = -EOVERFLOW;
591 if (err == -EEXIST)
592 err = -EADDRINUSE;
593 sock_put(sk);
594 goto err;
595 }
596
597 /* We need to ensure that the socket is hashed and visible. */
598 smp_wmb();
599 /* Paired with lockless reads from netlink_bind(),
600 * netlink_connect() and netlink_sendmsg().
601 */
602 WRITE_ONCE(nlk_sk(sk)->bound, portid);
603
604 err:
605 release_sock(sk);
606 return err;
607 }
608
609 static void netlink_remove(struct sock *sk)
610 {
611 struct netlink_table *table;
612
613 table = &nl_table[sk->sk_protocol];
614 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
615 netlink_rhashtable_params)) {
616 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
617 __sock_put(sk);
618 }
619
620 netlink_table_grab();
621 if (nlk_sk(sk)->subscriptions) {
622 __sk_del_bind_node(sk);
623 netlink_update_listeners(sk);
624 }
625 if (sk->sk_protocol == NETLINK_GENERIC)
626 atomic_inc(&genl_sk_destructing_cnt);
627 netlink_table_ungrab();
628 }
629
630 static struct proto netlink_proto = {
631 .name = "NETLINK",
632 .owner = THIS_MODULE,
633 .obj_size = sizeof(struct netlink_sock),
634 };
635
636 static int __netlink_create(struct net *net, struct socket *sock,
637 struct mutex *cb_mutex, int protocol,
638 int kern)
639 {
640 struct sock *sk;
641 struct netlink_sock *nlk;
642
643 sock->ops = &netlink_ops;
644
645 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
646 if (!sk)
647 return -ENOMEM;
648
649 sock_init_data(sock, sk);
650
651 nlk = nlk_sk(sk);
652 if (cb_mutex) {
653 nlk->cb_mutex = cb_mutex;
654 } else {
655 nlk->cb_mutex = &nlk->cb_def_mutex;
656 mutex_init(nlk->cb_mutex);
657 lockdep_set_class_and_name(nlk->cb_mutex,
658 nlk_cb_mutex_keys + protocol,
659 nlk_cb_mutex_key_strings[protocol]);
660 }
661 init_waitqueue_head(&nlk->wait);
662
663 sk->sk_destruct = netlink_sock_destruct;
664 sk->sk_protocol = protocol;
665 return 0;
666 }
667
668 static int netlink_create(struct net *net, struct socket *sock, int protocol,
669 int kern)
670 {
671 struct module *module = NULL;
672 struct mutex *cb_mutex;
673 struct netlink_sock *nlk;
674 int (*bind)(struct net *net, int group);
675 void (*unbind)(struct net *net, int group);
676 int err = 0;
677
678 sock->state = SS_UNCONNECTED;
679
680 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
681 return -ESOCKTNOSUPPORT;
682
683 if (protocol < 0 || protocol >= MAX_LINKS)
684 return -EPROTONOSUPPORT;
685 protocol = array_index_nospec(protocol, MAX_LINKS);
686
687 netlink_lock_table();
688 #ifdef CONFIG_MODULES
689 if (!nl_table[protocol].registered) {
690 netlink_unlock_table();
691 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
692 netlink_lock_table();
693 }
694 #endif
695 if (nl_table[protocol].registered &&
696 try_module_get(nl_table[protocol].module))
697 module = nl_table[protocol].module;
698 else
699 err = -EPROTONOSUPPORT;
700 cb_mutex = nl_table[protocol].cb_mutex;
701 bind = nl_table[protocol].bind;
702 unbind = nl_table[protocol].unbind;
703 netlink_unlock_table();
704
705 if (err < 0)
706 goto out;
707
708 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
709 if (err < 0)
710 goto out_module;
711
712 local_bh_disable();
713 sock_prot_inuse_add(net, &netlink_proto, 1);
714 local_bh_enable();
715
716 nlk = nlk_sk(sock->sk);
717 nlk->module = module;
718 nlk->netlink_bind = bind;
719 nlk->netlink_unbind = unbind;
720 out:
721 return err;
722
723 out_module:
724 module_put(module);
725 goto out;
726 }
727
728 static void deferred_put_nlk_sk(struct rcu_head *head)
729 {
730 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
731 struct sock *sk = &nlk->sk;
732
733 kfree(nlk->groups);
734 nlk->groups = NULL;
735
736 if (!refcount_dec_and_test(&sk->sk_refcnt))
737 return;
738
739 if (nlk->cb_running && nlk->cb.done) {
740 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
741 schedule_work(&nlk->work);
742 return;
743 }
744
745 sk_free(sk);
746 }
747
748 static int netlink_release(struct socket *sock)
749 {
750 struct sock *sk = sock->sk;
751 struct netlink_sock *nlk;
752
753 if (!sk)
754 return 0;
755
756 netlink_remove(sk);
757 sock_orphan(sk);
758 nlk = nlk_sk(sk);
759
760 /*
761 * OK. Socket is unlinked, any packets that arrive now
762 * will be purged.
763 */
764
765 /* must not acquire netlink_table_lock in any way again before unbind
766 * and notifying genetlink is done as otherwise it might deadlock
767 */
768 if (nlk->netlink_unbind) {
769 int i;
770
771 for (i = 0; i < nlk->ngroups; i++)
772 if (test_bit(i, nlk->groups))
773 nlk->netlink_unbind(sock_net(sk), i + 1);
774 }
775 if (sk->sk_protocol == NETLINK_GENERIC &&
776 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
777 wake_up(&genl_sk_destructing_waitq);
778
779 sock->sk = NULL;
780 wake_up_interruptible_all(&nlk->wait);
781
782 skb_queue_purge(&sk->sk_write_queue);
783
784 if (nlk->portid && nlk->bound) {
785 struct netlink_notify n = {
786 .net = sock_net(sk),
787 .protocol = sk->sk_protocol,
788 .portid = nlk->portid,
789 };
790 blocking_notifier_call_chain(&netlink_chain,
791 NETLINK_URELEASE, &n);
792 }
793
794 module_put(nlk->module);
795
796 if (netlink_is_kernel(sk)) {
797 netlink_table_grab();
798 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
799 if (--nl_table[sk->sk_protocol].registered == 0) {
800 struct listeners *old;
801
802 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
803 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
804 kfree_rcu(old, rcu);
805 nl_table[sk->sk_protocol].module = NULL;
806 nl_table[sk->sk_protocol].bind = NULL;
807 nl_table[sk->sk_protocol].unbind = NULL;
808 nl_table[sk->sk_protocol].flags = 0;
809 nl_table[sk->sk_protocol].registered = 0;
810 }
811 netlink_table_ungrab();
812 }
813
814 local_bh_disable();
815 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
816 local_bh_enable();
817 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
818 return 0;
819 }
820
821 static int netlink_autobind(struct socket *sock)
822 {
823 struct sock *sk = sock->sk;
824 struct net *net = sock_net(sk);
825 struct netlink_table *table = &nl_table[sk->sk_protocol];
826 s32 portid = task_tgid_vnr(current);
827 int err;
828 s32 rover = -4096;
829 bool ok;
830
831 retry:
832 cond_resched();
833 rcu_read_lock();
834 ok = !__netlink_lookup(table, portid, net);
835 rcu_read_unlock();
836 if (!ok) {
837 /* Bind collision, search negative portid values. */
838 if (rover == -4096)
839 /* rover will be in range [S32_MIN, -4097] */
840 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
841 else if (rover >= -4096)
842 rover = -4097;
843 portid = rover--;
844 goto retry;
845 }
846
847 err = netlink_insert(sk, portid);
848 if (err == -EADDRINUSE)
849 goto retry;
850
851 /* If 2 threads race to autobind, that is fine. */
852 if (err == -EBUSY)
853 err = 0;
854
855 return err;
856 }
857
858 /**
859 * __netlink_ns_capable - General netlink message capability test
860 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
861 * @user_ns: The user namespace of the capability to use
862 * @cap: The capability to use
863 *
864 * Test to see if the opener of the socket we received the message
865 * from had when the netlink socket was created and the sender of the
866 * message has the capability @cap in the user namespace @user_ns.
867 */
868 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
869 struct user_namespace *user_ns, int cap)
870 {
871 return ((nsp->flags & NETLINK_SKB_DST) ||
872 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
873 ns_capable(user_ns, cap);
874 }
875 EXPORT_SYMBOL(__netlink_ns_capable);
876
877 /**
878 * netlink_ns_capable - General netlink message capability test
879 * @skb: socket buffer holding a netlink command from userspace
880 * @user_ns: The user namespace of the capability to use
881 * @cap: The capability to use
882 *
883 * Test to see if the opener of the socket we received the message
884 * from had when the netlink socket was created and the sender of the
885 * message has the capability @cap in the user namespace @user_ns.
886 */
887 bool netlink_ns_capable(const struct sk_buff *skb,
888 struct user_namespace *user_ns, int cap)
889 {
890 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
891 }
892 EXPORT_SYMBOL(netlink_ns_capable);
893
894 /**
895 * netlink_capable - Netlink global message capability test
896 * @skb: socket buffer holding a netlink command from userspace
897 * @cap: The capability to use
898 *
899 * Test to see if the opener of the socket we received the message
900 * from had when the netlink socket was created and the sender of the
901 * message has the capability @cap in all user namespaces.
902 */
903 bool netlink_capable(const struct sk_buff *skb, int cap)
904 {
905 return netlink_ns_capable(skb, &init_user_ns, cap);
906 }
907 EXPORT_SYMBOL(netlink_capable);
908
909 /**
910 * netlink_net_capable - Netlink network namespace message capability test
911 * @skb: socket buffer holding a netlink command from userspace
912 * @cap: The capability to use
913 *
914 * Test to see if the opener of the socket we received the message
915 * from had when the netlink socket was created and the sender of the
916 * message has the capability @cap over the network namespace of
917 * the socket we received the message from.
918 */
919 bool netlink_net_capable(const struct sk_buff *skb, int cap)
920 {
921 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
922 }
923 EXPORT_SYMBOL(netlink_net_capable);
924
925 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
926 {
927 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
928 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
929 }
930
931 static void
932 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
933 {
934 struct netlink_sock *nlk = nlk_sk(sk);
935
936 if (nlk->subscriptions && !subscriptions)
937 __sk_del_bind_node(sk);
938 else if (!nlk->subscriptions && subscriptions)
939 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
940 nlk->subscriptions = subscriptions;
941 }
942
943 static int netlink_realloc_groups(struct sock *sk)
944 {
945 struct netlink_sock *nlk = nlk_sk(sk);
946 unsigned int groups;
947 unsigned long *new_groups;
948 int err = 0;
949
950 netlink_table_grab();
951
952 groups = nl_table[sk->sk_protocol].groups;
953 if (!nl_table[sk->sk_protocol].registered) {
954 err = -ENOENT;
955 goto out_unlock;
956 }
957
958 if (nlk->ngroups >= groups)
959 goto out_unlock;
960
961 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
962 if (new_groups == NULL) {
963 err = -ENOMEM;
964 goto out_unlock;
965 }
966 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
967 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
968
969 nlk->groups = new_groups;
970 nlk->ngroups = groups;
971 out_unlock:
972 netlink_table_ungrab();
973 return err;
974 }
975
976 static void netlink_undo_bind(int group, long unsigned int groups,
977 struct sock *sk)
978 {
979 struct netlink_sock *nlk = nlk_sk(sk);
980 int undo;
981
982 if (!nlk->netlink_unbind)
983 return;
984
985 for (undo = 0; undo < group; undo++)
986 if (test_bit(undo, &groups))
987 nlk->netlink_unbind(sock_net(sk), undo + 1);
988 }
989
990 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
991 int addr_len)
992 {
993 struct sock *sk = sock->sk;
994 struct net *net = sock_net(sk);
995 struct netlink_sock *nlk = nlk_sk(sk);
996 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
997 int err = 0;
998 unsigned long groups;
999 bool bound;
1000
1001 if (addr_len < sizeof(struct sockaddr_nl))
1002 return -EINVAL;
1003
1004 if (nladdr->nl_family != AF_NETLINK)
1005 return -EINVAL;
1006 groups = nladdr->nl_groups;
1007
1008 /* Only superuser is allowed to listen multicasts */
1009 if (groups) {
1010 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1011 return -EPERM;
1012 err = netlink_realloc_groups(sk);
1013 if (err)
1014 return err;
1015 }
1016
1017 if (nlk->ngroups < BITS_PER_LONG)
1018 groups &= (1UL << nlk->ngroups) - 1;
1019
1020 /* Paired with WRITE_ONCE() in netlink_insert() */
1021 bound = READ_ONCE(nlk->bound);
1022 if (bound) {
1023 /* Ensure nlk->portid is up-to-date. */
1024 smp_rmb();
1025
1026 if (nladdr->nl_pid != nlk->portid)
1027 return -EINVAL;
1028 }
1029
1030 if (nlk->netlink_bind && groups) {
1031 int group;
1032
1033 /* nl_groups is a u32, so cap the maximum groups we can bind */
1034 for (group = 0; group < BITS_PER_TYPE(u32); group++) {
1035 if (!test_bit(group, &groups))
1036 continue;
1037 err = nlk->netlink_bind(net, group + 1);
1038 if (!err)
1039 continue;
1040 netlink_undo_bind(group, groups, sk);
1041 return err;
1042 }
1043 }
1044
1045 /* No need for barriers here as we return to user-space without
1046 * using any of the bound attributes.
1047 */
1048 netlink_lock_table();
1049 if (!bound) {
1050 err = nladdr->nl_pid ?
1051 netlink_insert(sk, nladdr->nl_pid) :
1052 netlink_autobind(sock);
1053 if (err) {
1054 netlink_undo_bind(BITS_PER_TYPE(u32), groups, sk);
1055 goto unlock;
1056 }
1057 }
1058
1059 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1060 goto unlock;
1061 netlink_unlock_table();
1062
1063 netlink_table_grab();
1064 netlink_update_subscriptions(sk, nlk->subscriptions +
1065 hweight32(groups) -
1066 hweight32(nlk->groups[0]));
1067 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1068 netlink_update_listeners(sk);
1069 netlink_table_ungrab();
1070
1071 return 0;
1072
1073 unlock:
1074 netlink_unlock_table();
1075 return err;
1076 }
1077
1078 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1079 int alen, int flags)
1080 {
1081 int err = 0;
1082 struct sock *sk = sock->sk;
1083 struct netlink_sock *nlk = nlk_sk(sk);
1084 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1085
1086 if (alen < sizeof(addr->sa_family))
1087 return -EINVAL;
1088
1089 if (addr->sa_family == AF_UNSPEC) {
1090 sk->sk_state = NETLINK_UNCONNECTED;
1091 nlk->dst_portid = 0;
1092 nlk->dst_group = 0;
1093 return 0;
1094 }
1095 if (addr->sa_family != AF_NETLINK)
1096 return -EINVAL;
1097
1098 if (alen < sizeof(struct sockaddr_nl))
1099 return -EINVAL;
1100
1101 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1102 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1103 return -EPERM;
1104
1105 /* No need for barriers here as we return to user-space without
1106 * using any of the bound attributes.
1107 * Paired with WRITE_ONCE() in netlink_insert().
1108 */
1109 if (!READ_ONCE(nlk->bound))
1110 err = netlink_autobind(sock);
1111
1112 if (err == 0) {
1113 sk->sk_state = NETLINK_CONNECTED;
1114 nlk->dst_portid = nladdr->nl_pid;
1115 nlk->dst_group = ffs(nladdr->nl_groups);
1116 }
1117
1118 return err;
1119 }
1120
1121 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1122 int peer)
1123 {
1124 struct sock *sk = sock->sk;
1125 struct netlink_sock *nlk = nlk_sk(sk);
1126 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1127
1128 nladdr->nl_family = AF_NETLINK;
1129 nladdr->nl_pad = 0;
1130
1131 if (peer) {
1132 nladdr->nl_pid = nlk->dst_portid;
1133 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1134 } else {
1135 nladdr->nl_pid = nlk->portid;
1136 netlink_lock_table();
1137 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1138 netlink_unlock_table();
1139 }
1140 return sizeof(*nladdr);
1141 }
1142
1143 static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1144 unsigned long arg)
1145 {
1146 /* try to hand this ioctl down to the NIC drivers.
1147 */
1148 return -ENOIOCTLCMD;
1149 }
1150
1151 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1152 {
1153 struct sock *sock;
1154 struct netlink_sock *nlk;
1155
1156 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1157 if (!sock)
1158 return ERR_PTR(-ECONNREFUSED);
1159
1160 /* Don't bother queuing skb if kernel socket has no input function */
1161 nlk = nlk_sk(sock);
1162 if (sock->sk_state == NETLINK_CONNECTED &&
1163 nlk->dst_portid != nlk_sk(ssk)->portid) {
1164 sock_put(sock);
1165 return ERR_PTR(-ECONNREFUSED);
1166 }
1167 return sock;
1168 }
1169
1170 struct sock *netlink_getsockbyfilp(struct file *filp)
1171 {
1172 struct inode *inode = file_inode(filp);
1173 struct sock *sock;
1174
1175 if (!S_ISSOCK(inode->i_mode))
1176 return ERR_PTR(-ENOTSOCK);
1177
1178 sock = SOCKET_I(inode)->sk;
1179 if (sock->sk_family != AF_NETLINK)
1180 return ERR_PTR(-EINVAL);
1181
1182 sock_hold(sock);
1183 return sock;
1184 }
1185
1186 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1187 int broadcast)
1188 {
1189 struct sk_buff *skb;
1190 void *data;
1191
1192 if (size <= NLMSG_GOODSIZE || broadcast)
1193 return alloc_skb(size, GFP_KERNEL);
1194
1195 size = SKB_DATA_ALIGN(size) +
1196 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1197
1198 data = vmalloc(size);
1199 if (data == NULL)
1200 return NULL;
1201
1202 skb = __build_skb(data, size);
1203 if (skb == NULL)
1204 vfree(data);
1205 else
1206 skb->destructor = netlink_skb_destructor;
1207
1208 return skb;
1209 }
1210
1211 /*
1212 * Attach a skb to a netlink socket.
1213 * The caller must hold a reference to the destination socket. On error, the
1214 * reference is dropped. The skb is not send to the destination, just all
1215 * all error checks are performed and memory in the queue is reserved.
1216 * Return values:
1217 * < 0: error. skb freed, reference to sock dropped.
1218 * 0: continue
1219 * 1: repeat lookup - reference dropped while waiting for socket memory.
1220 */
1221 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1222 long *timeo, struct sock *ssk)
1223 {
1224 struct netlink_sock *nlk;
1225
1226 nlk = nlk_sk(sk);
1227
1228 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1229 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1230 DECLARE_WAITQUEUE(wait, current);
1231 if (!*timeo) {
1232 if (!ssk || netlink_is_kernel(ssk))
1233 netlink_overrun(sk);
1234 sock_put(sk);
1235 kfree_skb(skb);
1236 return -EAGAIN;
1237 }
1238
1239 __set_current_state(TASK_INTERRUPTIBLE);
1240 add_wait_queue(&nlk->wait, &wait);
1241
1242 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1243 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1244 !sock_flag(sk, SOCK_DEAD))
1245 *timeo = schedule_timeout(*timeo);
1246
1247 __set_current_state(TASK_RUNNING);
1248 remove_wait_queue(&nlk->wait, &wait);
1249 sock_put(sk);
1250
1251 if (signal_pending(current)) {
1252 kfree_skb(skb);
1253 return sock_intr_errno(*timeo);
1254 }
1255 return 1;
1256 }
1257 netlink_skb_set_owner_r(skb, sk);
1258 return 0;
1259 }
1260
1261 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1262 {
1263 int len = skb->len;
1264
1265 netlink_deliver_tap(sock_net(sk), skb);
1266
1267 skb_queue_tail(&sk->sk_receive_queue, skb);
1268 sk->sk_data_ready(sk);
1269 return len;
1270 }
1271
1272 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1273 {
1274 int len = __netlink_sendskb(sk, skb);
1275
1276 sock_put(sk);
1277 return len;
1278 }
1279
1280 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1281 {
1282 kfree_skb(skb);
1283 sock_put(sk);
1284 }
1285
1286 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1287 {
1288 int delta;
1289
1290 WARN_ON(skb->sk != NULL);
1291 delta = skb->end - skb->tail;
1292 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1293 return skb;
1294
1295 if (skb_shared(skb)) {
1296 struct sk_buff *nskb = skb_clone(skb, allocation);
1297 if (!nskb)
1298 return skb;
1299 consume_skb(skb);
1300 skb = nskb;
1301 }
1302
1303 pskb_expand_head(skb, 0, -delta,
1304 (allocation & ~__GFP_DIRECT_RECLAIM) |
1305 __GFP_NOWARN | __GFP_NORETRY);
1306 return skb;
1307 }
1308
1309 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1310 struct sock *ssk)
1311 {
1312 int ret;
1313 struct netlink_sock *nlk = nlk_sk(sk);
1314
1315 ret = -ECONNREFUSED;
1316 if (nlk->netlink_rcv != NULL) {
1317 ret = skb->len;
1318 netlink_skb_set_owner_r(skb, sk);
1319 NETLINK_CB(skb).sk = ssk;
1320 netlink_deliver_tap_kernel(sk, ssk, skb);
1321 nlk->netlink_rcv(skb);
1322 consume_skb(skb);
1323 } else {
1324 kfree_skb(skb);
1325 }
1326 sock_put(sk);
1327 return ret;
1328 }
1329
1330 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1331 u32 portid, int nonblock)
1332 {
1333 struct sock *sk;
1334 int err;
1335 long timeo;
1336
1337 skb = netlink_trim(skb, gfp_any());
1338
1339 timeo = sock_sndtimeo(ssk, nonblock);
1340 retry:
1341 sk = netlink_getsockbyportid(ssk, portid);
1342 if (IS_ERR(sk)) {
1343 kfree_skb(skb);
1344 return PTR_ERR(sk);
1345 }
1346 if (netlink_is_kernel(sk))
1347 return netlink_unicast_kernel(sk, skb, ssk);
1348
1349 if (sk_filter(sk, skb)) {
1350 err = skb->len;
1351 kfree_skb(skb);
1352 sock_put(sk);
1353 return err;
1354 }
1355
1356 err = netlink_attachskb(sk, skb, &timeo, ssk);
1357 if (err == 1)
1358 goto retry;
1359 if (err)
1360 return err;
1361
1362 return netlink_sendskb(sk, skb);
1363 }
1364 EXPORT_SYMBOL(netlink_unicast);
1365
1366 int netlink_has_listeners(struct sock *sk, unsigned int group)
1367 {
1368 int res = 0;
1369 struct listeners *listeners;
1370
1371 BUG_ON(!netlink_is_kernel(sk));
1372
1373 rcu_read_lock();
1374 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1375
1376 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1377 res = test_bit(group - 1, listeners->masks);
1378
1379 rcu_read_unlock();
1380
1381 return res;
1382 }
1383 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1384
1385 bool netlink_strict_get_check(struct sk_buff *skb)
1386 {
1387 const struct netlink_sock *nlk = nlk_sk(NETLINK_CB(skb).sk);
1388
1389 return nlk->flags & NETLINK_F_STRICT_CHK;
1390 }
1391 EXPORT_SYMBOL_GPL(netlink_strict_get_check);
1392
1393 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1394 {
1395 struct netlink_sock *nlk = nlk_sk(sk);
1396
1397 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1398 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1399 netlink_skb_set_owner_r(skb, sk);
1400 __netlink_sendskb(sk, skb);
1401 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1402 }
1403 return -1;
1404 }
1405
1406 struct netlink_broadcast_data {
1407 struct sock *exclude_sk;
1408 struct net *net;
1409 u32 portid;
1410 u32 group;
1411 int failure;
1412 int delivery_failure;
1413 int congested;
1414 int delivered;
1415 gfp_t allocation;
1416 struct sk_buff *skb, *skb2;
1417 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1418 void *tx_data;
1419 };
1420
1421 static void do_one_broadcast(struct sock *sk,
1422 struct netlink_broadcast_data *p)
1423 {
1424 struct netlink_sock *nlk = nlk_sk(sk);
1425 int val;
1426
1427 if (p->exclude_sk == sk)
1428 return;
1429
1430 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1431 !test_bit(p->group - 1, nlk->groups))
1432 return;
1433
1434 if (!net_eq(sock_net(sk), p->net)) {
1435 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1436 return;
1437
1438 if (!peernet_has_id(sock_net(sk), p->net))
1439 return;
1440
1441 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1442 CAP_NET_BROADCAST))
1443 return;
1444 }
1445
1446 if (p->failure) {
1447 netlink_overrun(sk);
1448 return;
1449 }
1450
1451 sock_hold(sk);
1452 if (p->skb2 == NULL) {
1453 if (skb_shared(p->skb)) {
1454 p->skb2 = skb_clone(p->skb, p->allocation);
1455 } else {
1456 p->skb2 = skb_get(p->skb);
1457 /*
1458 * skb ownership may have been set when
1459 * delivered to a previous socket.
1460 */
1461 skb_orphan(p->skb2);
1462 }
1463 }
1464 if (p->skb2 == NULL) {
1465 netlink_overrun(sk);
1466 /* Clone failed. Notify ALL listeners. */
1467 p->failure = 1;
1468 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1469 p->delivery_failure = 1;
1470 goto out;
1471 }
1472 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1473 kfree_skb(p->skb2);
1474 p->skb2 = NULL;
1475 goto out;
1476 }
1477 if (sk_filter(sk, p->skb2)) {
1478 kfree_skb(p->skb2);
1479 p->skb2 = NULL;
1480 goto out;
1481 }
1482 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1483 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1484 NETLINK_CB(p->skb2).nsid_is_set = true;
1485 val = netlink_broadcast_deliver(sk, p->skb2);
1486 if (val < 0) {
1487 netlink_overrun(sk);
1488 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1489 p->delivery_failure = 1;
1490 } else {
1491 p->congested |= val;
1492 p->delivered = 1;
1493 p->skb2 = NULL;
1494 }
1495 out:
1496 sock_put(sk);
1497 }
1498
1499 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1500 u32 group, gfp_t allocation,
1501 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1502 void *filter_data)
1503 {
1504 struct net *net = sock_net(ssk);
1505 struct netlink_broadcast_data info;
1506 struct sock *sk;
1507
1508 skb = netlink_trim(skb, allocation);
1509
1510 info.exclude_sk = ssk;
1511 info.net = net;
1512 info.portid = portid;
1513 info.group = group;
1514 info.failure = 0;
1515 info.delivery_failure = 0;
1516 info.congested = 0;
1517 info.delivered = 0;
1518 info.allocation = allocation;
1519 info.skb = skb;
1520 info.skb2 = NULL;
1521 info.tx_filter = filter;
1522 info.tx_data = filter_data;
1523
1524 /* While we sleep in clone, do not allow to change socket list */
1525
1526 netlink_lock_table();
1527
1528 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1529 do_one_broadcast(sk, &info);
1530
1531 consume_skb(skb);
1532
1533 netlink_unlock_table();
1534
1535 if (info.delivery_failure) {
1536 kfree_skb(info.skb2);
1537 return -ENOBUFS;
1538 }
1539 consume_skb(info.skb2);
1540
1541 if (info.delivered) {
1542 if (info.congested && gfpflags_allow_blocking(allocation))
1543 yield();
1544 return 0;
1545 }
1546 return -ESRCH;
1547 }
1548 EXPORT_SYMBOL(netlink_broadcast_filtered);
1549
1550 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1551 u32 group, gfp_t allocation)
1552 {
1553 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1554 NULL, NULL);
1555 }
1556 EXPORT_SYMBOL(netlink_broadcast);
1557
1558 struct netlink_set_err_data {
1559 struct sock *exclude_sk;
1560 u32 portid;
1561 u32 group;
1562 int code;
1563 };
1564
1565 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1566 {
1567 struct netlink_sock *nlk = nlk_sk(sk);
1568 int ret = 0;
1569
1570 if (sk == p->exclude_sk)
1571 goto out;
1572
1573 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1574 goto out;
1575
1576 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1577 !test_bit(p->group - 1, nlk->groups))
1578 goto out;
1579
1580 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1581 ret = 1;
1582 goto out;
1583 }
1584
1585 sk->sk_err = p->code;
1586 sk_error_report(sk);
1587 out:
1588 return ret;
1589 }
1590
1591 /**
1592 * netlink_set_err - report error to broadcast listeners
1593 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1594 * @portid: the PORTID of a process that we want to skip (if any)
1595 * @group: the broadcast group that will notice the error
1596 * @code: error code, must be negative (as usual in kernelspace)
1597 *
1598 * This function returns the number of broadcast listeners that have set the
1599 * NETLINK_NO_ENOBUFS socket option.
1600 */
1601 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1602 {
1603 struct netlink_set_err_data info;
1604 struct sock *sk;
1605 int ret = 0;
1606
1607 info.exclude_sk = ssk;
1608 info.portid = portid;
1609 info.group = group;
1610 /* sk->sk_err wants a positive error value */
1611 info.code = -code;
1612
1613 read_lock(&nl_table_lock);
1614
1615 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1616 ret += do_one_set_err(sk, &info);
1617
1618 read_unlock(&nl_table_lock);
1619 return ret;
1620 }
1621 EXPORT_SYMBOL(netlink_set_err);
1622
1623 /* must be called with netlink table grabbed */
1624 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1625 unsigned int group,
1626 int is_new)
1627 {
1628 int old, new = !!is_new, subscriptions;
1629
1630 old = test_bit(group - 1, nlk->groups);
1631 subscriptions = nlk->subscriptions - old + new;
1632 if (new)
1633 __set_bit(group - 1, nlk->groups);
1634 else
1635 __clear_bit(group - 1, nlk->groups);
1636 netlink_update_subscriptions(&nlk->sk, subscriptions);
1637 netlink_update_listeners(&nlk->sk);
1638 }
1639
1640 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1641 sockptr_t optval, unsigned int optlen)
1642 {
1643 struct sock *sk = sock->sk;
1644 struct netlink_sock *nlk = nlk_sk(sk);
1645 unsigned int val = 0;
1646 int err;
1647
1648 if (level != SOL_NETLINK)
1649 return -ENOPROTOOPT;
1650
1651 if (optlen >= sizeof(int) &&
1652 copy_from_sockptr(&val, optval, sizeof(val)))
1653 return -EFAULT;
1654
1655 switch (optname) {
1656 case NETLINK_PKTINFO:
1657 if (val)
1658 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1659 else
1660 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1661 err = 0;
1662 break;
1663 case NETLINK_ADD_MEMBERSHIP:
1664 case NETLINK_DROP_MEMBERSHIP: {
1665 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1666 return -EPERM;
1667 err = netlink_realloc_groups(sk);
1668 if (err)
1669 return err;
1670 if (!val || val - 1 >= nlk->ngroups)
1671 return -EINVAL;
1672 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1673 err = nlk->netlink_bind(sock_net(sk), val);
1674 if (err)
1675 return err;
1676 }
1677 netlink_table_grab();
1678 netlink_update_socket_mc(nlk, val,
1679 optname == NETLINK_ADD_MEMBERSHIP);
1680 netlink_table_ungrab();
1681 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1682 nlk->netlink_unbind(sock_net(sk), val);
1683
1684 err = 0;
1685 break;
1686 }
1687 case NETLINK_BROADCAST_ERROR:
1688 if (val)
1689 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1690 else
1691 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1692 err = 0;
1693 break;
1694 case NETLINK_NO_ENOBUFS:
1695 if (val) {
1696 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1697 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1698 wake_up_interruptible(&nlk->wait);
1699 } else {
1700 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1701 }
1702 err = 0;
1703 break;
1704 case NETLINK_LISTEN_ALL_NSID:
1705 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1706 return -EPERM;
1707
1708 if (val)
1709 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1710 else
1711 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1712 err = 0;
1713 break;
1714 case NETLINK_CAP_ACK:
1715 if (val)
1716 nlk->flags |= NETLINK_F_CAP_ACK;
1717 else
1718 nlk->flags &= ~NETLINK_F_CAP_ACK;
1719 err = 0;
1720 break;
1721 case NETLINK_EXT_ACK:
1722 if (val)
1723 nlk->flags |= NETLINK_F_EXT_ACK;
1724 else
1725 nlk->flags &= ~NETLINK_F_EXT_ACK;
1726 err = 0;
1727 break;
1728 case NETLINK_GET_STRICT_CHK:
1729 if (val)
1730 nlk->flags |= NETLINK_F_STRICT_CHK;
1731 else
1732 nlk->flags &= ~NETLINK_F_STRICT_CHK;
1733 err = 0;
1734 break;
1735 default:
1736 err = -ENOPROTOOPT;
1737 }
1738 return err;
1739 }
1740
1741 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1742 char __user *optval, int __user *optlen)
1743 {
1744 struct sock *sk = sock->sk;
1745 struct netlink_sock *nlk = nlk_sk(sk);
1746 int len, val, err;
1747
1748 if (level != SOL_NETLINK)
1749 return -ENOPROTOOPT;
1750
1751 if (get_user(len, optlen))
1752 return -EFAULT;
1753 if (len < 0)
1754 return -EINVAL;
1755
1756 switch (optname) {
1757 case NETLINK_PKTINFO:
1758 if (len < sizeof(int))
1759 return -EINVAL;
1760 len = sizeof(int);
1761 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1762 if (put_user(len, optlen) ||
1763 put_user(val, optval))
1764 return -EFAULT;
1765 err = 0;
1766 break;
1767 case NETLINK_BROADCAST_ERROR:
1768 if (len < sizeof(int))
1769 return -EINVAL;
1770 len = sizeof(int);
1771 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1772 if (put_user(len, optlen) ||
1773 put_user(val, optval))
1774 return -EFAULT;
1775 err = 0;
1776 break;
1777 case NETLINK_NO_ENOBUFS:
1778 if (len < sizeof(int))
1779 return -EINVAL;
1780 len = sizeof(int);
1781 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1782 if (put_user(len, optlen) ||
1783 put_user(val, optval))
1784 return -EFAULT;
1785 err = 0;
1786 break;
1787 case NETLINK_LIST_MEMBERSHIPS: {
1788 int pos, idx, shift;
1789
1790 err = 0;
1791 netlink_lock_table();
1792 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1793 if (len - pos < sizeof(u32))
1794 break;
1795
1796 idx = pos / sizeof(unsigned long);
1797 shift = (pos % sizeof(unsigned long)) * 8;
1798 if (put_user((u32)(nlk->groups[idx] >> shift),
1799 (u32 __user *)(optval + pos))) {
1800 err = -EFAULT;
1801 break;
1802 }
1803 }
1804 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1805 err = -EFAULT;
1806 netlink_unlock_table();
1807 break;
1808 }
1809 case NETLINK_CAP_ACK:
1810 if (len < sizeof(int))
1811 return -EINVAL;
1812 len = sizeof(int);
1813 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1814 if (put_user(len, optlen) ||
1815 put_user(val, optval))
1816 return -EFAULT;
1817 err = 0;
1818 break;
1819 case NETLINK_EXT_ACK:
1820 if (len < sizeof(int))
1821 return -EINVAL;
1822 len = sizeof(int);
1823 val = nlk->flags & NETLINK_F_EXT_ACK ? 1 : 0;
1824 if (put_user(len, optlen) || put_user(val, optval))
1825 return -EFAULT;
1826 err = 0;
1827 break;
1828 case NETLINK_GET_STRICT_CHK:
1829 if (len < sizeof(int))
1830 return -EINVAL;
1831 len = sizeof(int);
1832 val = nlk->flags & NETLINK_F_STRICT_CHK ? 1 : 0;
1833 if (put_user(len, optlen) || put_user(val, optval))
1834 return -EFAULT;
1835 err = 0;
1836 break;
1837 default:
1838 err = -ENOPROTOOPT;
1839 }
1840 return err;
1841 }
1842
1843 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1844 {
1845 struct nl_pktinfo info;
1846
1847 info.group = NETLINK_CB(skb).dst_group;
1848 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1849 }
1850
1851 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1852 struct sk_buff *skb)
1853 {
1854 if (!NETLINK_CB(skb).nsid_is_set)
1855 return;
1856
1857 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1858 &NETLINK_CB(skb).nsid);
1859 }
1860
1861 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1862 {
1863 struct sock *sk = sock->sk;
1864 struct netlink_sock *nlk = nlk_sk(sk);
1865 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1866 u32 dst_portid;
1867 u32 dst_group;
1868 struct sk_buff *skb;
1869 int err;
1870 struct scm_cookie scm;
1871 u32 netlink_skb_flags = 0;
1872
1873 if (msg->msg_flags & MSG_OOB)
1874 return -EOPNOTSUPP;
1875
1876 if (len == 0) {
1877 pr_warn_once("Zero length message leads to an empty skb\n");
1878 return -ENODATA;
1879 }
1880
1881 err = scm_send(sock, msg, &scm, true);
1882 if (err < 0)
1883 return err;
1884
1885 if (msg->msg_namelen) {
1886 err = -EINVAL;
1887 if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1888 goto out;
1889 if (addr->nl_family != AF_NETLINK)
1890 goto out;
1891 dst_portid = addr->nl_pid;
1892 dst_group = ffs(addr->nl_groups);
1893 err = -EPERM;
1894 if ((dst_group || dst_portid) &&
1895 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1896 goto out;
1897 netlink_skb_flags |= NETLINK_SKB_DST;
1898 } else {
1899 dst_portid = nlk->dst_portid;
1900 dst_group = nlk->dst_group;
1901 }
1902
1903 /* Paired with WRITE_ONCE() in netlink_insert() */
1904 if (!READ_ONCE(nlk->bound)) {
1905 err = netlink_autobind(sock);
1906 if (err)
1907 goto out;
1908 } else {
1909 /* Ensure nlk is hashed and visible. */
1910 smp_rmb();
1911 }
1912
1913 err = -EMSGSIZE;
1914 if (len > sk->sk_sndbuf - 32)
1915 goto out;
1916 err = -ENOBUFS;
1917 skb = netlink_alloc_large_skb(len, dst_group);
1918 if (skb == NULL)
1919 goto out;
1920
1921 NETLINK_CB(skb).portid = nlk->portid;
1922 NETLINK_CB(skb).dst_group = dst_group;
1923 NETLINK_CB(skb).creds = scm.creds;
1924 NETLINK_CB(skb).flags = netlink_skb_flags;
1925
1926 err = -EFAULT;
1927 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1928 kfree_skb(skb);
1929 goto out;
1930 }
1931
1932 err = security_netlink_send(sk, skb);
1933 if (err) {
1934 kfree_skb(skb);
1935 goto out;
1936 }
1937
1938 if (dst_group) {
1939 refcount_inc(&skb->users);
1940 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1941 }
1942 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags & MSG_DONTWAIT);
1943
1944 out:
1945 scm_destroy(&scm);
1946 return err;
1947 }
1948
1949 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1950 int flags)
1951 {
1952 struct scm_cookie scm;
1953 struct sock *sk = sock->sk;
1954 struct netlink_sock *nlk = nlk_sk(sk);
1955 int noblock = flags & MSG_DONTWAIT;
1956 size_t copied;
1957 struct sk_buff *skb, *data_skb;
1958 int err, ret;
1959
1960 if (flags & MSG_OOB)
1961 return -EOPNOTSUPP;
1962
1963 copied = 0;
1964
1965 skb = skb_recv_datagram(sk, flags, noblock, &err);
1966 if (skb == NULL)
1967 goto out;
1968
1969 data_skb = skb;
1970
1971 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1972 if (unlikely(skb_shinfo(skb)->frag_list)) {
1973 /*
1974 * If this skb has a frag_list, then here that means that we
1975 * will have to use the frag_list skb's data for compat tasks
1976 * and the regular skb's data for normal (non-compat) tasks.
1977 *
1978 * If we need to send the compat skb, assign it to the
1979 * 'data_skb' variable so that it will be used below for data
1980 * copying. We keep 'skb' for everything else, including
1981 * freeing both later.
1982 */
1983 if (flags & MSG_CMSG_COMPAT)
1984 data_skb = skb_shinfo(skb)->frag_list;
1985 }
1986 #endif
1987
1988 /* Record the max length of recvmsg() calls for future allocations */
1989 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1990 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1991 SKB_WITH_OVERHEAD(32768));
1992
1993 copied = data_skb->len;
1994 if (len < copied) {
1995 msg->msg_flags |= MSG_TRUNC;
1996 copied = len;
1997 }
1998
1999 skb_reset_transport_header(data_skb);
2000 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
2001
2002 if (msg->msg_name) {
2003 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
2004 addr->nl_family = AF_NETLINK;
2005 addr->nl_pad = 0;
2006 addr->nl_pid = NETLINK_CB(skb).portid;
2007 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
2008 msg->msg_namelen = sizeof(*addr);
2009 }
2010
2011 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
2012 netlink_cmsg_recv_pktinfo(msg, skb);
2013 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
2014 netlink_cmsg_listen_all_nsid(sk, msg, skb);
2015
2016 memset(&scm, 0, sizeof(scm));
2017 scm.creds = *NETLINK_CREDS(skb);
2018 if (flags & MSG_TRUNC)
2019 copied = data_skb->len;
2020
2021 skb_free_datagram(sk, skb);
2022
2023 if (nlk->cb_running &&
2024 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2025 ret = netlink_dump(sk);
2026 if (ret) {
2027 sk->sk_err = -ret;
2028 sk_error_report(sk);
2029 }
2030 }
2031
2032 scm_recv(sock, msg, &scm, flags);
2033 out:
2034 netlink_rcv_wake(sk);
2035 return err ? : copied;
2036 }
2037
2038 static void netlink_data_ready(struct sock *sk)
2039 {
2040 BUG();
2041 }
2042
2043 /*
2044 * We export these functions to other modules. They provide a
2045 * complete set of kernel non-blocking support for message
2046 * queueing.
2047 */
2048
2049 struct sock *
2050 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2051 struct netlink_kernel_cfg *cfg)
2052 {
2053 struct socket *sock;
2054 struct sock *sk;
2055 struct netlink_sock *nlk;
2056 struct listeners *listeners = NULL;
2057 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2058 unsigned int groups;
2059
2060 BUG_ON(!nl_table);
2061
2062 if (unit < 0 || unit >= MAX_LINKS)
2063 return NULL;
2064
2065 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2066 return NULL;
2067
2068 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2069 goto out_sock_release_nosk;
2070
2071 sk = sock->sk;
2072
2073 if (!cfg || cfg->groups < 32)
2074 groups = 32;
2075 else
2076 groups = cfg->groups;
2077
2078 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2079 if (!listeners)
2080 goto out_sock_release;
2081
2082 sk->sk_data_ready = netlink_data_ready;
2083 if (cfg && cfg->input)
2084 nlk_sk(sk)->netlink_rcv = cfg->input;
2085
2086 if (netlink_insert(sk, 0))
2087 goto out_sock_release;
2088
2089 nlk = nlk_sk(sk);
2090 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2091
2092 netlink_table_grab();
2093 if (!nl_table[unit].registered) {
2094 nl_table[unit].groups = groups;
2095 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2096 nl_table[unit].cb_mutex = cb_mutex;
2097 nl_table[unit].module = module;
2098 if (cfg) {
2099 nl_table[unit].bind = cfg->bind;
2100 nl_table[unit].unbind = cfg->unbind;
2101 nl_table[unit].flags = cfg->flags;
2102 if (cfg->compare)
2103 nl_table[unit].compare = cfg->compare;
2104 }
2105 nl_table[unit].registered = 1;
2106 } else {
2107 kfree(listeners);
2108 nl_table[unit].registered++;
2109 }
2110 netlink_table_ungrab();
2111 return sk;
2112
2113 out_sock_release:
2114 kfree(listeners);
2115 netlink_kernel_release(sk);
2116 return NULL;
2117
2118 out_sock_release_nosk:
2119 sock_release(sock);
2120 return NULL;
2121 }
2122 EXPORT_SYMBOL(__netlink_kernel_create);
2123
2124 void
2125 netlink_kernel_release(struct sock *sk)
2126 {
2127 if (sk == NULL || sk->sk_socket == NULL)
2128 return;
2129
2130 sock_release(sk->sk_socket);
2131 }
2132 EXPORT_SYMBOL(netlink_kernel_release);
2133
2134 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2135 {
2136 struct listeners *new, *old;
2137 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2138
2139 if (groups < 32)
2140 groups = 32;
2141
2142 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2143 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2144 if (!new)
2145 return -ENOMEM;
2146 old = nl_deref_protected(tbl->listeners);
2147 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2148 rcu_assign_pointer(tbl->listeners, new);
2149
2150 kfree_rcu(old, rcu);
2151 }
2152 tbl->groups = groups;
2153
2154 return 0;
2155 }
2156
2157 /**
2158 * netlink_change_ngroups - change number of multicast groups
2159 *
2160 * This changes the number of multicast groups that are available
2161 * on a certain netlink family. Note that it is not possible to
2162 * change the number of groups to below 32. Also note that it does
2163 * not implicitly call netlink_clear_multicast_users() when the
2164 * number of groups is reduced.
2165 *
2166 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2167 * @groups: The new number of groups.
2168 */
2169 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2170 {
2171 int err;
2172
2173 netlink_table_grab();
2174 err = __netlink_change_ngroups(sk, groups);
2175 netlink_table_ungrab();
2176
2177 return err;
2178 }
2179
2180 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2181 {
2182 struct sock *sk;
2183 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2184
2185 sk_for_each_bound(sk, &tbl->mc_list)
2186 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2187 }
2188
2189 struct nlmsghdr *
2190 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2191 {
2192 struct nlmsghdr *nlh;
2193 int size = nlmsg_msg_size(len);
2194
2195 nlh = skb_put(skb, NLMSG_ALIGN(size));
2196 nlh->nlmsg_type = type;
2197 nlh->nlmsg_len = size;
2198 nlh->nlmsg_flags = flags;
2199 nlh->nlmsg_pid = portid;
2200 nlh->nlmsg_seq = seq;
2201 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2202 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2203 return nlh;
2204 }
2205 EXPORT_SYMBOL(__nlmsg_put);
2206
2207 /*
2208 * It looks a bit ugly.
2209 * It would be better to create kernel thread.
2210 */
2211
2212 static int netlink_dump_done(struct netlink_sock *nlk, struct sk_buff *skb,
2213 struct netlink_callback *cb,
2214 struct netlink_ext_ack *extack)
2215 {
2216 struct nlmsghdr *nlh;
2217
2218 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(nlk->dump_done_errno),
2219 NLM_F_MULTI | cb->answer_flags);
2220 if (WARN_ON(!nlh))
2221 return -ENOBUFS;
2222
2223 nl_dump_check_consistent(cb, nlh);
2224 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno, sizeof(nlk->dump_done_errno));
2225
2226 if (extack->_msg && nlk->flags & NETLINK_F_EXT_ACK) {
2227 nlh->nlmsg_flags |= NLM_F_ACK_TLVS;
2228 if (!nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg))
2229 nlmsg_end(skb, nlh);
2230 }
2231
2232 return 0;
2233 }
2234
2235 static int netlink_dump(struct sock *sk)
2236 {
2237 struct netlink_sock *nlk = nlk_sk(sk);
2238 struct netlink_ext_ack extack = {};
2239 struct netlink_callback *cb;
2240 struct sk_buff *skb = NULL;
2241 struct module *module;
2242 int err = -ENOBUFS;
2243 int alloc_min_size;
2244 int alloc_size;
2245
2246 mutex_lock(nlk->cb_mutex);
2247 if (!nlk->cb_running) {
2248 err = -EINVAL;
2249 goto errout_skb;
2250 }
2251
2252 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2253 goto errout_skb;
2254
2255 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2256 * required, but it makes sense to _attempt_ a 16K bytes allocation
2257 * to reduce number of system calls on dump operations, if user
2258 * ever provided a big enough buffer.
2259 */
2260 cb = &nlk->cb;
2261 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2262
2263 if (alloc_min_size < nlk->max_recvmsg_len) {
2264 alloc_size = nlk->max_recvmsg_len;
2265 skb = alloc_skb(alloc_size,
2266 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2267 __GFP_NOWARN | __GFP_NORETRY);
2268 }
2269 if (!skb) {
2270 alloc_size = alloc_min_size;
2271 skb = alloc_skb(alloc_size, GFP_KERNEL);
2272 }
2273 if (!skb)
2274 goto errout_skb;
2275
2276 /* Trim skb to allocated size. User is expected to provide buffer as
2277 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2278 * netlink_recvmsg())). dump will pack as many smaller messages as
2279 * could fit within the allocated skb. skb is typically allocated
2280 * with larger space than required (could be as much as near 2x the
2281 * requested size with align to next power of 2 approach). Allowing
2282 * dump to use the excess space makes it difficult for a user to have a
2283 * reasonable static buffer based on the expected largest dump of a
2284 * single netdev. The outcome is MSG_TRUNC error.
2285 */
2286 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2287 netlink_skb_set_owner_r(skb, sk);
2288
2289 if (nlk->dump_done_errno > 0) {
2290 cb->extack = &extack;
2291 nlk->dump_done_errno = cb->dump(skb, cb);
2292 cb->extack = NULL;
2293 }
2294
2295 if (nlk->dump_done_errno > 0 ||
2296 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2297 mutex_unlock(nlk->cb_mutex);
2298
2299 if (sk_filter(sk, skb))
2300 kfree_skb(skb);
2301 else
2302 __netlink_sendskb(sk, skb);
2303 return 0;
2304 }
2305
2306 if (netlink_dump_done(nlk, skb, cb, &extack))
2307 goto errout_skb;
2308
2309 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2310 /* frag_list skb's data is used for compat tasks
2311 * and the regular skb's data for normal (non-compat) tasks.
2312 * See netlink_recvmsg().
2313 */
2314 if (unlikely(skb_shinfo(skb)->frag_list)) {
2315 if (netlink_dump_done(nlk, skb_shinfo(skb)->frag_list, cb, &extack))
2316 goto errout_skb;
2317 }
2318 #endif
2319
2320 if (sk_filter(sk, skb))
2321 kfree_skb(skb);
2322 else
2323 __netlink_sendskb(sk, skb);
2324
2325 if (cb->done)
2326 cb->done(cb);
2327
2328 nlk->cb_running = false;
2329 module = cb->module;
2330 skb = cb->skb;
2331 mutex_unlock(nlk->cb_mutex);
2332 module_put(module);
2333 consume_skb(skb);
2334 return 0;
2335
2336 errout_skb:
2337 mutex_unlock(nlk->cb_mutex);
2338 kfree_skb(skb);
2339 return err;
2340 }
2341
2342 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2343 const struct nlmsghdr *nlh,
2344 struct netlink_dump_control *control)
2345 {
2346 struct netlink_sock *nlk, *nlk2;
2347 struct netlink_callback *cb;
2348 struct sock *sk;
2349 int ret;
2350
2351 refcount_inc(&skb->users);
2352
2353 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2354 if (sk == NULL) {
2355 ret = -ECONNREFUSED;
2356 goto error_free;
2357 }
2358
2359 nlk = nlk_sk(sk);
2360 mutex_lock(nlk->cb_mutex);
2361 /* A dump is in progress... */
2362 if (nlk->cb_running) {
2363 ret = -EBUSY;
2364 goto error_unlock;
2365 }
2366 /* add reference of module which cb->dump belongs to */
2367 if (!try_module_get(control->module)) {
2368 ret = -EPROTONOSUPPORT;
2369 goto error_unlock;
2370 }
2371
2372 cb = &nlk->cb;
2373 memset(cb, 0, sizeof(*cb));
2374 cb->dump = control->dump;
2375 cb->done = control->done;
2376 cb->nlh = nlh;
2377 cb->data = control->data;
2378 cb->module = control->module;
2379 cb->min_dump_alloc = control->min_dump_alloc;
2380 cb->skb = skb;
2381
2382 nlk2 = nlk_sk(NETLINK_CB(skb).sk);
2383 cb->strict_check = !!(nlk2->flags & NETLINK_F_STRICT_CHK);
2384
2385 if (control->start) {
2386 ret = control->start(cb);
2387 if (ret)
2388 goto error_put;
2389 }
2390
2391 nlk->cb_running = true;
2392 nlk->dump_done_errno = INT_MAX;
2393
2394 mutex_unlock(nlk->cb_mutex);
2395
2396 ret = netlink_dump(sk);
2397
2398 sock_put(sk);
2399
2400 if (ret)
2401 return ret;
2402
2403 /* We successfully started a dump, by returning -EINTR we
2404 * signal not to send ACK even if it was requested.
2405 */
2406 return -EINTR;
2407
2408 error_put:
2409 module_put(control->module);
2410 error_unlock:
2411 sock_put(sk);
2412 mutex_unlock(nlk->cb_mutex);
2413 error_free:
2414 kfree_skb(skb);
2415 return ret;
2416 }
2417 EXPORT_SYMBOL(__netlink_dump_start);
2418
2419 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2420 const struct netlink_ext_ack *extack)
2421 {
2422 struct sk_buff *skb;
2423 struct nlmsghdr *rep;
2424 struct nlmsgerr *errmsg;
2425 size_t payload = sizeof(*errmsg);
2426 size_t tlvlen = 0;
2427 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2428 unsigned int flags = 0;
2429 bool nlk_has_extack = nlk->flags & NETLINK_F_EXT_ACK;
2430
2431 /* Error messages get the original request appened, unless the user
2432 * requests to cap the error message, and get extra error data if
2433 * requested.
2434 */
2435 if (nlk_has_extack && extack && extack->_msg)
2436 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2437
2438 if (err && !(nlk->flags & NETLINK_F_CAP_ACK))
2439 payload += nlmsg_len(nlh);
2440 else
2441 flags |= NLM_F_CAPPED;
2442 if (err && nlk_has_extack && extack && extack->bad_attr)
2443 tlvlen += nla_total_size(sizeof(u32));
2444 if (nlk_has_extack && extack && extack->cookie_len)
2445 tlvlen += nla_total_size(extack->cookie_len);
2446 if (err && nlk_has_extack && extack && extack->policy)
2447 tlvlen += netlink_policy_dump_attr_size_estimate(extack->policy);
2448
2449 if (tlvlen)
2450 flags |= NLM_F_ACK_TLVS;
2451
2452 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2453 if (!skb) {
2454 NETLINK_CB(in_skb).sk->sk_err = ENOBUFS;
2455 sk_error_report(NETLINK_CB(in_skb).sk);
2456 return;
2457 }
2458
2459 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2460 NLMSG_ERROR, payload, flags);
2461 errmsg = nlmsg_data(rep);
2462 errmsg->error = err;
2463 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2464
2465 if (nlk_has_extack && extack) {
2466 if (extack->_msg) {
2467 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG,
2468 extack->_msg));
2469 }
2470 if (err && extack->bad_attr &&
2471 !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2472 (u8 *)extack->bad_attr >= in_skb->data +
2473 in_skb->len))
2474 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2475 (u8 *)extack->bad_attr -
2476 (u8 *)nlh));
2477 if (extack->cookie_len)
2478 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2479 extack->cookie_len, extack->cookie));
2480 if (extack->policy)
2481 netlink_policy_dump_write_attr(skb, extack->policy,
2482 NLMSGERR_ATTR_POLICY);
2483 }
2484
2485 nlmsg_end(skb, rep);
2486
2487 nlmsg_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid);
2488 }
2489 EXPORT_SYMBOL(netlink_ack);
2490
2491 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2492 struct nlmsghdr *,
2493 struct netlink_ext_ack *))
2494 {
2495 struct netlink_ext_ack extack;
2496 struct nlmsghdr *nlh;
2497 int err;
2498
2499 while (skb->len >= nlmsg_total_size(0)) {
2500 int msglen;
2501
2502 memset(&extack, 0, sizeof(extack));
2503 nlh = nlmsg_hdr(skb);
2504 err = 0;
2505
2506 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2507 return 0;
2508
2509 /* Only requests are handled by the kernel */
2510 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2511 goto ack;
2512
2513 /* Skip control messages */
2514 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2515 goto ack;
2516
2517 err = cb(skb, nlh, &extack);
2518 if (err == -EINTR)
2519 goto skip;
2520
2521 ack:
2522 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2523 netlink_ack(skb, nlh, err, &extack);
2524
2525 skip:
2526 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2527 if (msglen > skb->len)
2528 msglen = skb->len;
2529 skb_pull(skb, msglen);
2530 }
2531
2532 return 0;
2533 }
2534 EXPORT_SYMBOL(netlink_rcv_skb);
2535
2536 /**
2537 * nlmsg_notify - send a notification netlink message
2538 * @sk: netlink socket to use
2539 * @skb: notification message
2540 * @portid: destination netlink portid for reports or 0
2541 * @group: destination multicast group or 0
2542 * @report: 1 to report back, 0 to disable
2543 * @flags: allocation flags
2544 */
2545 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2546 unsigned int group, int report, gfp_t flags)
2547 {
2548 int err = 0;
2549
2550 if (group) {
2551 int exclude_portid = 0;
2552
2553 if (report) {
2554 refcount_inc(&skb->users);
2555 exclude_portid = portid;
2556 }
2557
2558 /* errors reported via destination sk->sk_err, but propagate
2559 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2560 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2561 if (err == -ESRCH)
2562 err = 0;
2563 }
2564
2565 if (report) {
2566 int err2;
2567
2568 err2 = nlmsg_unicast(sk, skb, portid);
2569 if (!err)
2570 err = err2;
2571 }
2572
2573 return err;
2574 }
2575 EXPORT_SYMBOL(nlmsg_notify);
2576
2577 #ifdef CONFIG_PROC_FS
2578 struct nl_seq_iter {
2579 struct seq_net_private p;
2580 struct rhashtable_iter hti;
2581 int link;
2582 };
2583
2584 static void netlink_walk_start(struct nl_seq_iter *iter)
2585 {
2586 rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti);
2587 rhashtable_walk_start(&iter->hti);
2588 }
2589
2590 static void netlink_walk_stop(struct nl_seq_iter *iter)
2591 {
2592 rhashtable_walk_stop(&iter->hti);
2593 rhashtable_walk_exit(&iter->hti);
2594 }
2595
2596 static void *__netlink_seq_next(struct seq_file *seq)
2597 {
2598 struct nl_seq_iter *iter = seq->private;
2599 struct netlink_sock *nlk;
2600
2601 do {
2602 for (;;) {
2603 nlk = rhashtable_walk_next(&iter->hti);
2604
2605 if (IS_ERR(nlk)) {
2606 if (PTR_ERR(nlk) == -EAGAIN)
2607 continue;
2608
2609 return nlk;
2610 }
2611
2612 if (nlk)
2613 break;
2614
2615 netlink_walk_stop(iter);
2616 if (++iter->link >= MAX_LINKS)
2617 return NULL;
2618
2619 netlink_walk_start(iter);
2620 }
2621 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2622
2623 return nlk;
2624 }
2625
2626 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2627 __acquires(RCU)
2628 {
2629 struct nl_seq_iter *iter = seq->private;
2630 void *obj = SEQ_START_TOKEN;
2631 loff_t pos;
2632
2633 iter->link = 0;
2634
2635 netlink_walk_start(iter);
2636
2637 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2638 obj = __netlink_seq_next(seq);
2639
2640 return obj;
2641 }
2642
2643 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2644 {
2645 ++*pos;
2646 return __netlink_seq_next(seq);
2647 }
2648
2649 static void netlink_native_seq_stop(struct seq_file *seq, void *v)
2650 {
2651 struct nl_seq_iter *iter = seq->private;
2652
2653 if (iter->link >= MAX_LINKS)
2654 return;
2655
2656 netlink_walk_stop(iter);
2657 }
2658
2659
2660 static int netlink_native_seq_show(struct seq_file *seq, void *v)
2661 {
2662 if (v == SEQ_START_TOKEN) {
2663 seq_puts(seq,
2664 "sk Eth Pid Groups "
2665 "Rmem Wmem Dump Locks Drops Inode\n");
2666 } else {
2667 struct sock *s = v;
2668 struct netlink_sock *nlk = nlk_sk(s);
2669
2670 seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n",
2671 s,
2672 s->sk_protocol,
2673 nlk->portid,
2674 nlk->groups ? (u32)nlk->groups[0] : 0,
2675 sk_rmem_alloc_get(s),
2676 sk_wmem_alloc_get(s),
2677 nlk->cb_running,
2678 refcount_read(&s->sk_refcnt),
2679 atomic_read(&s->sk_drops),
2680 sock_i_ino(s)
2681 );
2682
2683 }
2684 return 0;
2685 }
2686
2687 #ifdef CONFIG_BPF_SYSCALL
2688 struct bpf_iter__netlink {
2689 __bpf_md_ptr(struct bpf_iter_meta *, meta);
2690 __bpf_md_ptr(struct netlink_sock *, sk);
2691 };
2692
2693 DEFINE_BPF_ITER_FUNC(netlink, struct bpf_iter_meta *meta, struct netlink_sock *sk)
2694
2695 static int netlink_prog_seq_show(struct bpf_prog *prog,
2696 struct bpf_iter_meta *meta,
2697 void *v)
2698 {
2699 struct bpf_iter__netlink ctx;
2700
2701 meta->seq_num--; /* skip SEQ_START_TOKEN */
2702 ctx.meta = meta;
2703 ctx.sk = nlk_sk((struct sock *)v);
2704 return bpf_iter_run_prog(prog, &ctx);
2705 }
2706
2707 static int netlink_seq_show(struct seq_file *seq, void *v)
2708 {
2709 struct bpf_iter_meta meta;
2710 struct bpf_prog *prog;
2711
2712 meta.seq = seq;
2713 prog = bpf_iter_get_info(&meta, false);
2714 if (!prog)
2715 return netlink_native_seq_show(seq, v);
2716
2717 if (v != SEQ_START_TOKEN)
2718 return netlink_prog_seq_show(prog, &meta, v);
2719
2720 return 0;
2721 }
2722
2723 static void netlink_seq_stop(struct seq_file *seq, void *v)
2724 {
2725 struct bpf_iter_meta meta;
2726 struct bpf_prog *prog;
2727
2728 if (!v) {
2729 meta.seq = seq;
2730 prog = bpf_iter_get_info(&meta, true);
2731 if (prog)
2732 (void)netlink_prog_seq_show(prog, &meta, v);
2733 }
2734
2735 netlink_native_seq_stop(seq, v);
2736 }
2737 #else
2738 static int netlink_seq_show(struct seq_file *seq, void *v)
2739 {
2740 return netlink_native_seq_show(seq, v);
2741 }
2742
2743 static void netlink_seq_stop(struct seq_file *seq, void *v)
2744 {
2745 netlink_native_seq_stop(seq, v);
2746 }
2747 #endif
2748
2749 static const struct seq_operations netlink_seq_ops = {
2750 .start = netlink_seq_start,
2751 .next = netlink_seq_next,
2752 .stop = netlink_seq_stop,
2753 .show = netlink_seq_show,
2754 };
2755 #endif
2756
2757 int netlink_register_notifier(struct notifier_block *nb)
2758 {
2759 return blocking_notifier_chain_register(&netlink_chain, nb);
2760 }
2761 EXPORT_SYMBOL(netlink_register_notifier);
2762
2763 int netlink_unregister_notifier(struct notifier_block *nb)
2764 {
2765 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2766 }
2767 EXPORT_SYMBOL(netlink_unregister_notifier);
2768
2769 static const struct proto_ops netlink_ops = {
2770 .family = PF_NETLINK,
2771 .owner = THIS_MODULE,
2772 .release = netlink_release,
2773 .bind = netlink_bind,
2774 .connect = netlink_connect,
2775 .socketpair = sock_no_socketpair,
2776 .accept = sock_no_accept,
2777 .getname = netlink_getname,
2778 .poll = datagram_poll,
2779 .ioctl = netlink_ioctl,
2780 .listen = sock_no_listen,
2781 .shutdown = sock_no_shutdown,
2782 .setsockopt = netlink_setsockopt,
2783 .getsockopt = netlink_getsockopt,
2784 .sendmsg = netlink_sendmsg,
2785 .recvmsg = netlink_recvmsg,
2786 .mmap = sock_no_mmap,
2787 .sendpage = sock_no_sendpage,
2788 };
2789
2790 static const struct net_proto_family netlink_family_ops = {
2791 .family = PF_NETLINK,
2792 .create = netlink_create,
2793 .owner = THIS_MODULE, /* for consistency 8) */
2794 };
2795
2796 static int __net_init netlink_net_init(struct net *net)
2797 {
2798 #ifdef CONFIG_PROC_FS
2799 if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2800 sizeof(struct nl_seq_iter)))
2801 return -ENOMEM;
2802 #endif
2803 return 0;
2804 }
2805
2806 static void __net_exit netlink_net_exit(struct net *net)
2807 {
2808 #ifdef CONFIG_PROC_FS
2809 remove_proc_entry("netlink", net->proc_net);
2810 #endif
2811 }
2812
2813 static void __init netlink_add_usersock_entry(void)
2814 {
2815 struct listeners *listeners;
2816 int groups = 32;
2817
2818 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2819 if (!listeners)
2820 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2821
2822 netlink_table_grab();
2823
2824 nl_table[NETLINK_USERSOCK].groups = groups;
2825 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2826 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2827 nl_table[NETLINK_USERSOCK].registered = 1;
2828 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2829
2830 netlink_table_ungrab();
2831 }
2832
2833 static struct pernet_operations __net_initdata netlink_net_ops = {
2834 .init = netlink_net_init,
2835 .exit = netlink_net_exit,
2836 };
2837
2838 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2839 {
2840 const struct netlink_sock *nlk = data;
2841 struct netlink_compare_arg arg;
2842
2843 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2844 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2845 }
2846
2847 static const struct rhashtable_params netlink_rhashtable_params = {
2848 .head_offset = offsetof(struct netlink_sock, node),
2849 .key_len = netlink_compare_arg_len,
2850 .obj_hashfn = netlink_hash,
2851 .obj_cmpfn = netlink_compare,
2852 .automatic_shrinking = true,
2853 };
2854
2855 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2856 BTF_ID_LIST(btf_netlink_sock_id)
2857 BTF_ID(struct, netlink_sock)
2858
2859 static const struct bpf_iter_seq_info netlink_seq_info = {
2860 .seq_ops = &netlink_seq_ops,
2861 .init_seq_private = bpf_iter_init_seq_net,
2862 .fini_seq_private = bpf_iter_fini_seq_net,
2863 .seq_priv_size = sizeof(struct nl_seq_iter),
2864 };
2865
2866 static struct bpf_iter_reg netlink_reg_info = {
2867 .target = "netlink",
2868 .ctx_arg_info_size = 1,
2869 .ctx_arg_info = {
2870 { offsetof(struct bpf_iter__netlink, sk),
2871 PTR_TO_BTF_ID_OR_NULL },
2872 },
2873 .seq_info = &netlink_seq_info,
2874 };
2875
2876 static int __init bpf_iter_register(void)
2877 {
2878 netlink_reg_info.ctx_arg_info[0].btf_id = *btf_netlink_sock_id;
2879 return bpf_iter_reg_target(&netlink_reg_info);
2880 }
2881 #endif
2882
2883 static int __init netlink_proto_init(void)
2884 {
2885 int i;
2886 int err = proto_register(&netlink_proto, 0);
2887
2888 if (err != 0)
2889 goto out;
2890
2891 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2892 err = bpf_iter_register();
2893 if (err)
2894 goto out;
2895 #endif
2896
2897 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof_field(struct sk_buff, cb));
2898
2899 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2900 if (!nl_table)
2901 goto panic;
2902
2903 for (i = 0; i < MAX_LINKS; i++) {
2904 if (rhashtable_init(&nl_table[i].hash,
2905 &netlink_rhashtable_params) < 0) {
2906 while (--i > 0)
2907 rhashtable_destroy(&nl_table[i].hash);
2908 kfree(nl_table);
2909 goto panic;
2910 }
2911 }
2912
2913 netlink_add_usersock_entry();
2914
2915 sock_register(&netlink_family_ops);
2916 register_pernet_subsys(&netlink_net_ops);
2917 register_pernet_subsys(&netlink_tap_net_ops);
2918 /* The netlink device handler may be needed early. */
2919 rtnetlink_init();
2920 out:
2921 return err;
2922 panic:
2923 panic("netlink_init: Cannot allocate nl_table\n");
2924 }
2925
2926 core_initcall(netlink_proto_init);
Ubuntu Jammy Linux kernel mirror