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/bpf.h>
24 #include <linux/capability.h>
25 #include <linux/kernel.h>
26 #include <linux/filter.h>
27 #include <linux/init.h>
28 #include <linux/signal.h>
29 #include <linux/sched.h>
30 #include <linux/errno.h>
31 #include <linux/string.h>
32 #include <linux/stat.h>
33 #include <linux/socket.h>
34 #include <linux/un.h>
35 #include <linux/fcntl.h>
36 #include <linux/termios.h>
37 #include <linux/sockios.h>
38 #include <linux/net.h>
39 #include <linux/fs.h>
40 #include <linux/slab.h>
41 #include <linux/uaccess.h>
42 #include <linux/skbuff.h>
43 #include <linux/netdevice.h>
44 #include <linux/rtnetlink.h>
45 #include <linux/proc_fs.h>
46 #include <linux/seq_file.h>
47 #include <linux/notifier.h>
48 #include <linux/security.h>
49 #include <linux/jhash.h>
50 #include <linux/jiffies.h>
51 #include <linux/random.h>
52 #include <linux/bitops.h>
53 #include <linux/mm.h>
54 #include <linux/types.h>
55 #include <linux/audit.h>
56 #include <linux/mutex.h>
57 #include <linux/vmalloc.h>
58 #include <linux/if_arp.h>
59 #include <linux/rhashtable.h>
60 #include <asm/cacheflush.h>
61 #include <linux/hash.h>
62 #include <linux/net_namespace.h>
63 #include <linux/nospec.h>
64 #include <linux/btf_ids.h>
65
66 #include <net/net_namespace.h>
67 #include <net/netns/generic.h>
68 #include <net/sock.h>
69 #include <net/scm.h>
70 #include <net/netlink.h>
71 #define CREATE_TRACE_POINTS
72 #include <trace/events/netlink.h>
73
74 #include "af_netlink.h"
75 #include "genetlink.h"
76
77 struct listeners {
78 struct rcu_head rcu;
79 unsigned long masks[];
80 };
81
82 /* state bits */
83 #define NETLINK_S_CONGESTED 0x0
84
netlink_is_kernel(struct sock * sk)85 static inline int netlink_is_kernel(struct sock *sk)
86 {
87 return nlk_test_bit(KERNEL_SOCKET, sk);
88 }
89
90 struct netlink_table *nl_table __read_mostly;
91 EXPORT_SYMBOL_GPL(nl_table);
92
93 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
94
95 static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
96
97 static const char *const nlk_cb_mutex_key_strings[MAX_LINKS + 1] = {
98 "nlk_cb_mutex-ROUTE",
99 "nlk_cb_mutex-1",
100 "nlk_cb_mutex-USERSOCK",
101 "nlk_cb_mutex-FIREWALL",
102 "nlk_cb_mutex-SOCK_DIAG",
103 "nlk_cb_mutex-NFLOG",
104 "nlk_cb_mutex-XFRM",
105 "nlk_cb_mutex-SELINUX",
106 "nlk_cb_mutex-ISCSI",
107 "nlk_cb_mutex-AUDIT",
108 "nlk_cb_mutex-FIB_LOOKUP",
109 "nlk_cb_mutex-CONNECTOR",
110 "nlk_cb_mutex-NETFILTER",
111 "nlk_cb_mutex-IP6_FW",
112 "nlk_cb_mutex-DNRTMSG",
113 "nlk_cb_mutex-KOBJECT_UEVENT",
114 "nlk_cb_mutex-GENERIC",
115 "nlk_cb_mutex-17",
116 "nlk_cb_mutex-SCSITRANSPORT",
117 "nlk_cb_mutex-ECRYPTFS",
118 "nlk_cb_mutex-RDMA",
119 "nlk_cb_mutex-CRYPTO",
120 "nlk_cb_mutex-SMC",
121 "nlk_cb_mutex-23",
122 "nlk_cb_mutex-24",
123 "nlk_cb_mutex-25",
124 "nlk_cb_mutex-26",
125 "nlk_cb_mutex-27",
126 "nlk_cb_mutex-28",
127 "nlk_cb_mutex-29",
128 "nlk_cb_mutex-30",
129 "nlk_cb_mutex-31",
130 "nlk_cb_mutex-MAX_LINKS"
131 };
132
133 static int netlink_dump(struct sock *sk, bool lock_taken);
134
135 /* nl_table locking explained:
136 * Lookup and traversal are protected with an RCU read-side lock. Insertion
137 * and removal are protected with per bucket lock while using RCU list
138 * modification primitives and may run in parallel to RCU protected lookups.
139 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
140 * been acquired * either during or after the socket has been removed from
141 * the list and after an RCU grace period.
142 */
143 DEFINE_RWLOCK(nl_table_lock);
144 EXPORT_SYMBOL_GPL(nl_table_lock);
145 static atomic_t nl_table_users = ATOMIC_INIT(0);
146
147 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
148
149 static BLOCKING_NOTIFIER_HEAD(netlink_chain);
150
151
152 static const struct rhashtable_params netlink_rhashtable_params;
153
do_trace_netlink_extack(const char * msg)154 void do_trace_netlink_extack(const char *msg)
155 {
156 trace_netlink_extack(msg);
157 }
158 EXPORT_SYMBOL(do_trace_netlink_extack);
159
netlink_group_mask(u32 group)160 static inline u32 netlink_group_mask(u32 group)
161 {
162 if (group > 32)
163 return 0;
164 return group ? 1 << (group - 1) : 0;
165 }
166
netlink_to_full_skb(const struct sk_buff * skb,gfp_t gfp_mask)167 static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
168 gfp_t gfp_mask)
169 {
170 unsigned int len = skb->len;
171 struct sk_buff *new;
172
173 new = alloc_skb(len, gfp_mask);
174 if (new == NULL)
175 return NULL;
176
177 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
178 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
179 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
180
181 skb_put_data(new, skb->data, len);
182 return new;
183 }
184
185 static unsigned int netlink_tap_net_id;
186
187 struct netlink_tap_net {
188 struct list_head netlink_tap_all;
189 struct mutex netlink_tap_lock;
190 };
191
netlink_add_tap(struct netlink_tap * nt)192 int netlink_add_tap(struct netlink_tap *nt)
193 {
194 struct net *net = dev_net(nt->dev);
195 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
196
197 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
198 return -EINVAL;
199
200 mutex_lock(&nn->netlink_tap_lock);
201 list_add_rcu(&nt->list, &nn->netlink_tap_all);
202 mutex_unlock(&nn->netlink_tap_lock);
203
204 __module_get(nt->module);
205
206 return 0;
207 }
208 EXPORT_SYMBOL_GPL(netlink_add_tap);
209
__netlink_remove_tap(struct netlink_tap * nt)210 static int __netlink_remove_tap(struct netlink_tap *nt)
211 {
212 struct net *net = dev_net(nt->dev);
213 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
214 bool found = false;
215 struct netlink_tap *tmp;
216
217 mutex_lock(&nn->netlink_tap_lock);
218
219 list_for_each_entry(tmp, &nn->netlink_tap_all, list) {
220 if (nt == tmp) {
221 list_del_rcu(&nt->list);
222 found = true;
223 goto out;
224 }
225 }
226
227 pr_warn("__netlink_remove_tap: %p not found\n", nt);
228 out:
229 mutex_unlock(&nn->netlink_tap_lock);
230
231 if (found)
232 module_put(nt->module);
233
234 return found ? 0 : -ENODEV;
235 }
236
netlink_remove_tap(struct netlink_tap * nt)237 int netlink_remove_tap(struct netlink_tap *nt)
238 {
239 int ret;
240
241 ret = __netlink_remove_tap(nt);
242 synchronize_net();
243
244 return ret;
245 }
246 EXPORT_SYMBOL_GPL(netlink_remove_tap);
247
netlink_tap_init_net(struct net * net)248 static __net_init int netlink_tap_init_net(struct net *net)
249 {
250 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
251
252 INIT_LIST_HEAD(&nn->netlink_tap_all);
253 mutex_init(&nn->netlink_tap_lock);
254 return 0;
255 }
256
257 static struct pernet_operations netlink_tap_net_ops = {
258 .init = netlink_tap_init_net,
259 .id = &netlink_tap_net_id,
260 .size = sizeof(struct netlink_tap_net),
261 };
262
netlink_filter_tap(const struct sk_buff * skb)263 static bool netlink_filter_tap(const struct sk_buff *skb)
264 {
265 struct sock *sk = skb->sk;
266
267 /* We take the more conservative approach and
268 * whitelist socket protocols that may pass.
269 */
270 switch (sk->sk_protocol) {
271 case NETLINK_ROUTE:
272 case NETLINK_USERSOCK:
273 case NETLINK_SOCK_DIAG:
274 case NETLINK_NFLOG:
275 case NETLINK_XFRM:
276 case NETLINK_FIB_LOOKUP:
277 case NETLINK_NETFILTER:
278 case NETLINK_GENERIC:
279 return true;
280 }
281
282 return false;
283 }
284
__netlink_deliver_tap_skb(struct sk_buff * skb,struct net_device * dev)285 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
286 struct net_device *dev)
287 {
288 struct sk_buff *nskb;
289 struct sock *sk = skb->sk;
290 int ret = -ENOMEM;
291
292 if (!net_eq(dev_net(dev), sock_net(sk)))
293 return 0;
294
295 dev_hold(dev);
296
297 if (is_vmalloc_addr(skb->head))
298 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
299 else
300 nskb = skb_clone(skb, GFP_ATOMIC);
301 if (nskb) {
302 nskb->dev = dev;
303 nskb->protocol = htons((u16) sk->sk_protocol);
304 nskb->pkt_type = netlink_is_kernel(sk) ?
305 PACKET_KERNEL : PACKET_USER;
306 skb_reset_network_header(nskb);
307 ret = dev_queue_xmit(nskb);
308 if (unlikely(ret > 0))
309 ret = net_xmit_errno(ret);
310 }
311
312 dev_put(dev);
313 return ret;
314 }
315
__netlink_deliver_tap(struct sk_buff * skb,struct netlink_tap_net * nn)316 static void __netlink_deliver_tap(struct sk_buff *skb, struct netlink_tap_net *nn)
317 {
318 int ret;
319 struct netlink_tap *tmp;
320
321 if (!netlink_filter_tap(skb))
322 return;
323
324 list_for_each_entry_rcu(tmp, &nn->netlink_tap_all, list) {
325 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
326 if (unlikely(ret))
327 break;
328 }
329 }
330
netlink_deliver_tap(struct net * net,struct sk_buff * skb)331 static void netlink_deliver_tap(struct net *net, struct sk_buff *skb)
332 {
333 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
334
335 rcu_read_lock();
336
337 if (unlikely(!list_empty(&nn->netlink_tap_all)))
338 __netlink_deliver_tap(skb, nn);
339
340 rcu_read_unlock();
341 }
342
netlink_deliver_tap_kernel(struct sock * dst,struct sock * src,struct sk_buff * skb)343 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
344 struct sk_buff *skb)
345 {
346 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
347 netlink_deliver_tap(sock_net(dst), skb);
348 }
349
netlink_overrun(struct sock * sk)350 static void netlink_overrun(struct sock *sk)
351 {
352 if (!nlk_test_bit(RECV_NO_ENOBUFS, sk)) {
353 if (!test_and_set_bit(NETLINK_S_CONGESTED,
354 &nlk_sk(sk)->state)) {
355 WRITE_ONCE(sk->sk_err, ENOBUFS);
356 sk_error_report(sk);
357 }
358 }
359 atomic_inc(&sk->sk_drops);
360 }
361
netlink_rcv_wake(struct sock * sk)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
netlink_skb_destructor(struct sk_buff * skb)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_atomic(skb->head);
378
379 skb->head = NULL;
380 }
381 if (skb->sk != NULL)
382 sock_rfree(skb);
383 }
384
netlink_skb_set_owner_r(struct sk_buff * skb,struct sock * sk)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 sk_mem_charge(sk, skb->truesize);
391 }
392
netlink_sock_destruct(struct sock * sk)393 static void netlink_sock_destruct(struct sock *sk)
394 {
395 skb_queue_purge(&sk->sk_receive_queue);
396
397 if (!sock_flag(sk, SOCK_DEAD)) {
398 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
399 return;
400 }
401
402 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
403 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
404 WARN_ON(nlk_sk(sk)->groups);
405 }
406
407 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
408 * SMP. Look, when several writers sleep and reader wakes them up, all but one
409 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
410 * this, _but_ remember, it adds useless work on UP machines.
411 */
412
netlink_table_grab(void)413 void netlink_table_grab(void)
414 __acquires(nl_table_lock)
415 {
416 might_sleep();
417
418 write_lock_irq(&nl_table_lock);
419
420 if (atomic_read(&nl_table_users)) {
421 DECLARE_WAITQUEUE(wait, current);
422
423 add_wait_queue_exclusive(&nl_table_wait, &wait);
424 for (;;) {
425 set_current_state(TASK_UNINTERRUPTIBLE);
426 if (atomic_read(&nl_table_users) == 0)
427 break;
428 write_unlock_irq(&nl_table_lock);
429 schedule();
430 write_lock_irq(&nl_table_lock);
431 }
432
433 __set_current_state(TASK_RUNNING);
434 remove_wait_queue(&nl_table_wait, &wait);
435 }
436 }
437
netlink_table_ungrab(void)438 void netlink_table_ungrab(void)
439 __releases(nl_table_lock)
440 {
441 write_unlock_irq(&nl_table_lock);
442 wake_up(&nl_table_wait);
443 }
444
445 static inline void
netlink_lock_table(void)446 netlink_lock_table(void)
447 {
448 unsigned long flags;
449
450 /* read_lock() synchronizes us to netlink_table_grab */
451
452 read_lock_irqsave(&nl_table_lock, flags);
453 atomic_inc(&nl_table_users);
454 read_unlock_irqrestore(&nl_table_lock, flags);
455 }
456
457 static inline void
netlink_unlock_table(void)458 netlink_unlock_table(void)
459 {
460 if (atomic_dec_and_test(&nl_table_users))
461 wake_up(&nl_table_wait);
462 }
463
464 struct netlink_compare_arg
465 {
466 possible_net_t pnet;
467 u32 portid;
468 };
469
470 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
471 #define netlink_compare_arg_len \
472 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
473
netlink_compare(struct rhashtable_compare_arg * arg,const void * ptr)474 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
475 const void *ptr)
476 {
477 const struct netlink_compare_arg *x = arg->key;
478 const struct netlink_sock *nlk = ptr;
479
480 return nlk->portid != x->portid ||
481 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
482 }
483
netlink_compare_arg_init(struct netlink_compare_arg * arg,struct net * net,u32 portid)484 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
485 struct net *net, u32 portid)
486 {
487 memset(arg, 0, sizeof(*arg));
488 write_pnet(&arg->pnet, net);
489 arg->portid = portid;
490 }
491
__netlink_lookup(struct netlink_table * table,u32 portid,struct net * net)492 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
493 struct net *net)
494 {
495 struct netlink_compare_arg arg;
496
497 netlink_compare_arg_init(&arg, net, portid);
498 return rhashtable_lookup_fast(&table->hash, &arg,
499 netlink_rhashtable_params);
500 }
501
__netlink_insert(struct netlink_table * table,struct sock * sk)502 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
503 {
504 struct netlink_compare_arg arg;
505
506 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
507 return rhashtable_lookup_insert_key(&table->hash, &arg,
508 &nlk_sk(sk)->node,
509 netlink_rhashtable_params);
510 }
511
netlink_lookup(struct net * net,int protocol,u32 portid)512 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
513 {
514 struct netlink_table *table = &nl_table[protocol];
515 struct sock *sk;
516
517 rcu_read_lock();
518 sk = __netlink_lookup(table, portid, net);
519 if (sk)
520 sock_hold(sk);
521 rcu_read_unlock();
522
523 return sk;
524 }
525
526 static const struct proto_ops netlink_ops;
527
528 static void
netlink_update_listeners(struct sock * sk)529 netlink_update_listeners(struct sock *sk)
530 {
531 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
532 unsigned long mask;
533 unsigned int i;
534 struct listeners *listeners;
535
536 listeners = nl_deref_protected(tbl->listeners);
537 if (!listeners)
538 return;
539
540 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
541 mask = 0;
542 sk_for_each_bound(sk, &tbl->mc_list) {
543 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
544 mask |= nlk_sk(sk)->groups[i];
545 }
546 listeners->masks[i] = mask;
547 }
548 /* this function is only called with the netlink table "grabbed", which
549 * makes sure updates are visible before bind or setsockopt return. */
550 }
551
netlink_insert(struct sock * sk,u32 portid)552 static int netlink_insert(struct sock *sk, u32 portid)
553 {
554 struct netlink_table *table = &nl_table[sk->sk_protocol];
555 int err;
556
557 lock_sock(sk);
558
559 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
560 if (nlk_sk(sk)->bound)
561 goto err;
562
563 /* portid can be read locklessly from netlink_getname(). */
564 WRITE_ONCE(nlk_sk(sk)->portid, portid);
565
566 sock_hold(sk);
567
568 err = __netlink_insert(table, sk);
569 if (err) {
570 /* In case the hashtable backend returns with -EBUSY
571 * from here, it must not escape to the caller.
572 */
573 if (unlikely(err == -EBUSY))
574 err = -EOVERFLOW;
575 if (err == -EEXIST)
576 err = -EADDRINUSE;
577 sock_put(sk);
578 goto err;
579 }
580
581 /* We need to ensure that the socket is hashed and visible. */
582 smp_wmb();
583 /* Paired with lockless reads from netlink_bind(),
584 * netlink_connect() and netlink_sendmsg().
585 */
586 WRITE_ONCE(nlk_sk(sk)->bound, portid);
587
588 err:
589 release_sock(sk);
590 return err;
591 }
592
netlink_remove(struct sock * sk)593 static void netlink_remove(struct sock *sk)
594 {
595 struct netlink_table *table;
596
597 table = &nl_table[sk->sk_protocol];
598 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
599 netlink_rhashtable_params)) {
600 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
601 __sock_put(sk);
602 }
603
604 netlink_table_grab();
605 if (nlk_sk(sk)->subscriptions) {
606 __sk_del_bind_node(sk);
607 netlink_update_listeners(sk);
608 }
609 if (sk->sk_protocol == NETLINK_GENERIC)
610 atomic_inc(&genl_sk_destructing_cnt);
611 netlink_table_ungrab();
612 }
613
614 static struct proto netlink_proto = {
615 .name = "NETLINK",
616 .owner = THIS_MODULE,
617 .obj_size = sizeof(struct netlink_sock),
618 };
619
__netlink_create(struct net * net,struct socket * sock,int protocol,int kern)620 static int __netlink_create(struct net *net, struct socket *sock,
621 int protocol, int kern)
622 {
623 struct sock *sk;
624 struct netlink_sock *nlk;
625
626 sock->ops = &netlink_ops;
627
628 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
629 if (!sk)
630 return -ENOMEM;
631
632 sock_init_data(sock, sk);
633
634 nlk = nlk_sk(sk);
635 mutex_init(&nlk->nl_cb_mutex);
636 lockdep_set_class_and_name(&nlk->nl_cb_mutex,
637 nlk_cb_mutex_keys + protocol,
638 nlk_cb_mutex_key_strings[protocol]);
639 init_waitqueue_head(&nlk->wait);
640
641 sk->sk_destruct = netlink_sock_destruct;
642 sk->sk_protocol = protocol;
643 return 0;
644 }
645
netlink_create(struct net * net,struct socket * sock,int protocol,int kern)646 static int netlink_create(struct net *net, struct socket *sock, int protocol,
647 int kern)
648 {
649 struct module *module = NULL;
650 struct netlink_sock *nlk;
651 int (*bind)(struct net *net, int group);
652 void (*unbind)(struct net *net, int group);
653 void (*release)(struct sock *sock, unsigned long *groups);
654 int err = 0;
655
656 sock->state = SS_UNCONNECTED;
657
658 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
659 return -ESOCKTNOSUPPORT;
660
661 if (protocol < 0 || protocol >= MAX_LINKS)
662 return -EPROTONOSUPPORT;
663 protocol = array_index_nospec(protocol, MAX_LINKS);
664
665 netlink_lock_table();
666 #ifdef CONFIG_MODULES
667 if (!nl_table[protocol].registered) {
668 netlink_unlock_table();
669 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
670 netlink_lock_table();
671 }
672 #endif
673 if (nl_table[protocol].registered &&
674 try_module_get(nl_table[protocol].module))
675 module = nl_table[protocol].module;
676 else
677 err = -EPROTONOSUPPORT;
678 bind = nl_table[protocol].bind;
679 unbind = nl_table[protocol].unbind;
680 release = nl_table[protocol].release;
681 netlink_unlock_table();
682
683 if (err < 0)
684 goto out;
685
686 err = __netlink_create(net, sock, protocol, kern);
687 if (err < 0)
688 goto out_module;
689
690 sock_prot_inuse_add(net, &netlink_proto, 1);
691
692 nlk = nlk_sk(sock->sk);
693 nlk->module = module;
694 nlk->netlink_bind = bind;
695 nlk->netlink_unbind = unbind;
696 nlk->netlink_release = release;
697 out:
698 return err;
699
700 out_module:
701 module_put(module);
702 goto out;
703 }
704
deferred_put_nlk_sk(struct rcu_head * head)705 static void deferred_put_nlk_sk(struct rcu_head *head)
706 {
707 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
708 struct sock *sk = &nlk->sk;
709
710 kfree(nlk->groups);
711 nlk->groups = NULL;
712
713 if (!refcount_dec_and_test(&sk->sk_refcnt))
714 return;
715
716 sk_free(sk);
717 }
718
netlink_release(struct socket * sock)719 static int netlink_release(struct socket *sock)
720 {
721 struct sock *sk = sock->sk;
722 struct netlink_sock *nlk;
723
724 if (!sk)
725 return 0;
726
727 netlink_remove(sk);
728 sock_orphan(sk);
729 nlk = nlk_sk(sk);
730
731 /*
732 * OK. Socket is unlinked, any packets that arrive now
733 * will be purged.
734 */
735 if (nlk->netlink_release)
736 nlk->netlink_release(sk, nlk->groups);
737
738 /* must not acquire netlink_table_lock in any way again before unbind
739 * and notifying genetlink is done as otherwise it might deadlock
740 */
741 if (nlk->netlink_unbind) {
742 int i;
743
744 for (i = 0; i < nlk->ngroups; i++)
745 if (test_bit(i, nlk->groups))
746 nlk->netlink_unbind(sock_net(sk), i + 1);
747 }
748 if (sk->sk_protocol == NETLINK_GENERIC &&
749 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
750 wake_up(&genl_sk_destructing_waitq);
751
752 sock->sk = NULL;
753 wake_up_interruptible_all(&nlk->wait);
754
755 skb_queue_purge(&sk->sk_write_queue);
756
757 if (nlk->portid && nlk->bound) {
758 struct netlink_notify n = {
759 .net = sock_net(sk),
760 .protocol = sk->sk_protocol,
761 .portid = nlk->portid,
762 };
763 blocking_notifier_call_chain(&netlink_chain,
764 NETLINK_URELEASE, &n);
765 }
766
767 /* Terminate any outstanding dump */
768 if (nlk->cb_running) {
769 if (nlk->cb.done)
770 nlk->cb.done(&nlk->cb);
771 module_put(nlk->cb.module);
772 kfree_skb(nlk->cb.skb);
773 }
774
775 module_put(nlk->module);
776
777 if (netlink_is_kernel(sk)) {
778 netlink_table_grab();
779 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
780 if (--nl_table[sk->sk_protocol].registered == 0) {
781 struct listeners *old;
782
783 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
784 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
785 kfree_rcu(old, rcu);
786 nl_table[sk->sk_protocol].module = NULL;
787 nl_table[sk->sk_protocol].bind = NULL;
788 nl_table[sk->sk_protocol].unbind = NULL;
789 nl_table[sk->sk_protocol].flags = 0;
790 nl_table[sk->sk_protocol].registered = 0;
791 }
792 netlink_table_ungrab();
793 }
794
795 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
796
797 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
798 return 0;
799 }
800
netlink_autobind(struct socket * sock)801 static int netlink_autobind(struct socket *sock)
802 {
803 struct sock *sk = sock->sk;
804 struct net *net = sock_net(sk);
805 struct netlink_table *table = &nl_table[sk->sk_protocol];
806 s32 portid = task_tgid_vnr(current);
807 int err;
808 s32 rover = -4096;
809 bool ok;
810
811 retry:
812 cond_resched();
813 rcu_read_lock();
814 ok = !__netlink_lookup(table, portid, net);
815 rcu_read_unlock();
816 if (!ok) {
817 /* Bind collision, search negative portid values. */
818 if (rover == -4096)
819 /* rover will be in range [S32_MIN, -4097] */
820 rover = S32_MIN + get_random_u32_below(-4096 - S32_MIN);
821 else if (rover >= -4096)
822 rover = -4097;
823 portid = rover--;
824 goto retry;
825 }
826
827 err = netlink_insert(sk, portid);
828 if (err == -EADDRINUSE)
829 goto retry;
830
831 /* If 2 threads race to autobind, that is fine. */
832 if (err == -EBUSY)
833 err = 0;
834
835 return err;
836 }
837
838 /**
839 * __netlink_ns_capable - General netlink message capability test
840 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
841 * @user_ns: The user namespace of the capability to use
842 * @cap: The capability to use
843 *
844 * Test to see if the opener of the socket we received the message
845 * from had when the netlink socket was created and the sender of the
846 * message has the capability @cap in the user namespace @user_ns.
847 */
__netlink_ns_capable(const struct netlink_skb_parms * nsp,struct user_namespace * user_ns,int cap)848 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
849 struct user_namespace *user_ns, int cap)
850 {
851 return ((nsp->flags & NETLINK_SKB_DST) ||
852 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
853 ns_capable(user_ns, cap);
854 }
855 EXPORT_SYMBOL(__netlink_ns_capable);
856
857 /**
858 * netlink_ns_capable - General netlink message capability test
859 * @skb: socket buffer holding a netlink command from userspace
860 * @user_ns: The user namespace of the capability to use
861 * @cap: The capability to use
862 *
863 * Test to see if the opener of the socket we received the message
864 * from had when the netlink socket was created and the sender of the
865 * message has the capability @cap in the user namespace @user_ns.
866 */
netlink_ns_capable(const struct sk_buff * skb,struct user_namespace * user_ns,int cap)867 bool netlink_ns_capable(const struct sk_buff *skb,
868 struct user_namespace *user_ns, int cap)
869 {
870 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
871 }
872 EXPORT_SYMBOL(netlink_ns_capable);
873
874 /**
875 * netlink_capable - Netlink global message capability test
876 * @skb: socket buffer holding a netlink command from userspace
877 * @cap: The capability to use
878 *
879 * Test to see if the opener of the socket we received the message
880 * from had when the netlink socket was created and the sender of the
881 * message has the capability @cap in all user namespaces.
882 */
netlink_capable(const struct sk_buff * skb,int cap)883 bool netlink_capable(const struct sk_buff *skb, int cap)
884 {
885 return netlink_ns_capable(skb, &init_user_ns, cap);
886 }
887 EXPORT_SYMBOL(netlink_capable);
888
889 /**
890 * netlink_net_capable - Netlink network namespace message capability test
891 * @skb: socket buffer holding a netlink command from userspace
892 * @cap: The capability to use
893 *
894 * Test to see if the opener of the socket we received the message
895 * from had when the netlink socket was created and the sender of the
896 * message has the capability @cap over the network namespace of
897 * the socket we received the message from.
898 */
netlink_net_capable(const struct sk_buff * skb,int cap)899 bool netlink_net_capable(const struct sk_buff *skb, int cap)
900 {
901 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
902 }
903 EXPORT_SYMBOL(netlink_net_capable);
904
netlink_allowed(const struct socket * sock,unsigned int flag)905 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
906 {
907 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
908 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
909 }
910
911 static void
netlink_update_subscriptions(struct sock * sk,unsigned int subscriptions)912 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
913 {
914 struct netlink_sock *nlk = nlk_sk(sk);
915
916 if (nlk->subscriptions && !subscriptions)
917 __sk_del_bind_node(sk);
918 else if (!nlk->subscriptions && subscriptions)
919 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
920 nlk->subscriptions = subscriptions;
921 }
922
netlink_realloc_groups(struct sock * sk)923 static int netlink_realloc_groups(struct sock *sk)
924 {
925 struct netlink_sock *nlk = nlk_sk(sk);
926 unsigned int groups;
927 unsigned long *new_groups;
928 int err = 0;
929
930 netlink_table_grab();
931
932 groups = nl_table[sk->sk_protocol].groups;
933 if (!nl_table[sk->sk_protocol].registered) {
934 err = -ENOENT;
935 goto out_unlock;
936 }
937
938 if (nlk->ngroups >= groups)
939 goto out_unlock;
940
941 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
942 if (new_groups == NULL) {
943 err = -ENOMEM;
944 goto out_unlock;
945 }
946 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
947 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
948
949 nlk->groups = new_groups;
950 nlk->ngroups = groups;
951 out_unlock:
952 netlink_table_ungrab();
953 return err;
954 }
955
netlink_undo_bind(int group,long unsigned int groups,struct sock * sk)956 static void netlink_undo_bind(int group, long unsigned int groups,
957 struct sock *sk)
958 {
959 struct netlink_sock *nlk = nlk_sk(sk);
960 int undo;
961
962 if (!nlk->netlink_unbind)
963 return;
964
965 for (undo = 0; undo < group; undo++)
966 if (test_bit(undo, &groups))
967 nlk->netlink_unbind(sock_net(sk), undo + 1);
968 }
969
netlink_bind(struct socket * sock,struct sockaddr * addr,int addr_len)970 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
971 int addr_len)
972 {
973 struct sock *sk = sock->sk;
974 struct net *net = sock_net(sk);
975 struct netlink_sock *nlk = nlk_sk(sk);
976 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
977 int err = 0;
978 unsigned long groups;
979 bool bound;
980
981 if (addr_len < sizeof(struct sockaddr_nl))
982 return -EINVAL;
983
984 if (nladdr->nl_family != AF_NETLINK)
985 return -EINVAL;
986 groups = nladdr->nl_groups;
987
988 /* Only superuser is allowed to listen multicasts */
989 if (groups) {
990 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
991 return -EPERM;
992 err = netlink_realloc_groups(sk);
993 if (err)
994 return err;
995 }
996
997 if (nlk->ngroups < BITS_PER_LONG)
998 groups &= (1UL << nlk->ngroups) - 1;
999
1000 /* Paired with WRITE_ONCE() in netlink_insert() */
1001 bound = READ_ONCE(nlk->bound);
1002 if (bound) {
1003 /* Ensure nlk->portid is up-to-date. */
1004 smp_rmb();
1005
1006 if (nladdr->nl_pid != nlk->portid)
1007 return -EINVAL;
1008 }
1009
1010 if (nlk->netlink_bind && groups) {
1011 int group;
1012
1013 /* nl_groups is a u32, so cap the maximum groups we can bind */
1014 for (group = 0; group < BITS_PER_TYPE(u32); group++) {
1015 if (!test_bit(group, &groups))
1016 continue;
1017 err = nlk->netlink_bind(net, group + 1);
1018 if (!err)
1019 continue;
1020 netlink_undo_bind(group, groups, sk);
1021 return err;
1022 }
1023 }
1024
1025 /* No need for barriers here as we return to user-space without
1026 * using any of the bound attributes.
1027 */
1028 netlink_lock_table();
1029 if (!bound) {
1030 err = nladdr->nl_pid ?
1031 netlink_insert(sk, nladdr->nl_pid) :
1032 netlink_autobind(sock);
1033 if (err) {
1034 netlink_undo_bind(BITS_PER_TYPE(u32), groups, sk);
1035 goto unlock;
1036 }
1037 }
1038
1039 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1040 goto unlock;
1041 netlink_unlock_table();
1042
1043 netlink_table_grab();
1044 netlink_update_subscriptions(sk, nlk->subscriptions +
1045 hweight32(groups) -
1046 hweight32(nlk->groups[0]));
1047 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1048 netlink_update_listeners(sk);
1049 netlink_table_ungrab();
1050
1051 return 0;
1052
1053 unlock:
1054 netlink_unlock_table();
1055 return err;
1056 }
1057
netlink_connect(struct socket * sock,struct sockaddr * addr,int alen,int flags)1058 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1059 int alen, int flags)
1060 {
1061 int err = 0;
1062 struct sock *sk = sock->sk;
1063 struct netlink_sock *nlk = nlk_sk(sk);
1064 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1065
1066 if (alen < sizeof(addr->sa_family))
1067 return -EINVAL;
1068
1069 if (addr->sa_family == AF_UNSPEC) {
1070 /* paired with READ_ONCE() in netlink_getsockbyportid() */
1071 WRITE_ONCE(sk->sk_state, NETLINK_UNCONNECTED);
1072 /* dst_portid and dst_group can be read locklessly */
1073 WRITE_ONCE(nlk->dst_portid, 0);
1074 WRITE_ONCE(nlk->dst_group, 0);
1075 return 0;
1076 }
1077 if (addr->sa_family != AF_NETLINK)
1078 return -EINVAL;
1079
1080 if (alen < sizeof(struct sockaddr_nl))
1081 return -EINVAL;
1082
1083 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1084 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1085 return -EPERM;
1086
1087 /* No need for barriers here as we return to user-space without
1088 * using any of the bound attributes.
1089 * Paired with WRITE_ONCE() in netlink_insert().
1090 */
1091 if (!READ_ONCE(nlk->bound))
1092 err = netlink_autobind(sock);
1093
1094 if (err == 0) {
1095 /* paired with READ_ONCE() in netlink_getsockbyportid() */
1096 WRITE_ONCE(sk->sk_state, NETLINK_CONNECTED);
1097 /* dst_portid and dst_group can be read locklessly */
1098 WRITE_ONCE(nlk->dst_portid, nladdr->nl_pid);
1099 WRITE_ONCE(nlk->dst_group, ffs(nladdr->nl_groups));
1100 }
1101
1102 return err;
1103 }
1104
netlink_getname(struct socket * sock,struct sockaddr * addr,int peer)1105 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1106 int peer)
1107 {
1108 struct sock *sk = sock->sk;
1109 struct netlink_sock *nlk = nlk_sk(sk);
1110 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1111
1112 nladdr->nl_family = AF_NETLINK;
1113 nladdr->nl_pad = 0;
1114
1115 if (peer) {
1116 /* Paired with WRITE_ONCE() in netlink_connect() */
1117 nladdr->nl_pid = READ_ONCE(nlk->dst_portid);
1118 nladdr->nl_groups = netlink_group_mask(READ_ONCE(nlk->dst_group));
1119 } else {
1120 /* Paired with WRITE_ONCE() in netlink_insert() */
1121 nladdr->nl_pid = READ_ONCE(nlk->portid);
1122 netlink_lock_table();
1123 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1124 netlink_unlock_table();
1125 }
1126 return sizeof(*nladdr);
1127 }
1128
netlink_ioctl(struct socket * sock,unsigned int cmd,unsigned long arg)1129 static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1130 unsigned long arg)
1131 {
1132 /* try to hand this ioctl down to the NIC drivers.
1133 */
1134 return -ENOIOCTLCMD;
1135 }
1136
netlink_getsockbyportid(struct sock * ssk,u32 portid)1137 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1138 {
1139 struct sock *sock;
1140 struct netlink_sock *nlk;
1141
1142 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1143 if (!sock)
1144 return ERR_PTR(-ECONNREFUSED);
1145
1146 /* Don't bother queuing skb if kernel socket has no input function */
1147 nlk = nlk_sk(sock);
1148 /* dst_portid and sk_state can be changed in netlink_connect() */
1149 if (READ_ONCE(sock->sk_state) == NETLINK_CONNECTED &&
1150 READ_ONCE(nlk->dst_portid) != nlk_sk(ssk)->portid) {
1151 sock_put(sock);
1152 return ERR_PTR(-ECONNREFUSED);
1153 }
1154 return sock;
1155 }
1156
netlink_getsockbyfilp(struct file * filp)1157 struct sock *netlink_getsockbyfilp(struct file *filp)
1158 {
1159 struct inode *inode = file_inode(filp);
1160 struct sock *sock;
1161
1162 if (!S_ISSOCK(inode->i_mode))
1163 return ERR_PTR(-ENOTSOCK);
1164
1165 sock = SOCKET_I(inode)->sk;
1166 if (sock->sk_family != AF_NETLINK)
1167 return ERR_PTR(-EINVAL);
1168
1169 sock_hold(sock);
1170 return sock;
1171 }
1172
netlink_alloc_large_skb(unsigned int size,int broadcast)1173 struct sk_buff *netlink_alloc_large_skb(unsigned int size, int broadcast)
1174 {
1175 size_t head_size = SKB_HEAD_ALIGN(size);
1176 struct sk_buff *skb;
1177 void *data;
1178
1179 if (head_size <= PAGE_SIZE || broadcast)
1180 return alloc_skb(size, GFP_KERNEL);
1181
1182 data = kvmalloc(head_size, GFP_KERNEL);
1183 if (!data)
1184 return NULL;
1185
1186 skb = __build_skb(data, head_size);
1187 if (!skb)
1188 kvfree(data);
1189 else if (is_vmalloc_addr(data))
1190 skb->destructor = netlink_skb_destructor;
1191
1192 return skb;
1193 }
1194
1195 /*
1196 * Attach a skb to a netlink socket.
1197 * The caller must hold a reference to the destination socket. On error, the
1198 * reference is dropped. The skb is not send to the destination, just all
1199 * all error checks are performed and memory in the queue is reserved.
1200 * Return values:
1201 * < 0: error. skb freed, reference to sock dropped.
1202 * 0: continue
1203 * 1: repeat lookup - reference dropped while waiting for socket memory.
1204 */
netlink_attachskb(struct sock * sk,struct sk_buff * skb,long * timeo,struct sock * ssk)1205 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1206 long *timeo, struct sock *ssk)
1207 {
1208 DECLARE_WAITQUEUE(wait, current);
1209 struct netlink_sock *nlk;
1210 unsigned int rmem;
1211
1212 nlk = nlk_sk(sk);
1213 rmem = atomic_add_return(skb->truesize, &sk->sk_rmem_alloc);
1214
1215 if ((rmem == skb->truesize || rmem <= READ_ONCE(sk->sk_rcvbuf)) &&
1216 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1217 netlink_skb_set_owner_r(skb, sk);
1218 return 0;
1219 }
1220
1221 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1222
1223 if (!*timeo) {
1224 if (!ssk || netlink_is_kernel(ssk))
1225 netlink_overrun(sk);
1226 sock_put(sk);
1227 kfree_skb(skb);
1228 return -EAGAIN;
1229 }
1230
1231 __set_current_state(TASK_INTERRUPTIBLE);
1232 add_wait_queue(&nlk->wait, &wait);
1233 rmem = atomic_read(&sk->sk_rmem_alloc);
1234
1235 if (((rmem && rmem + skb->truesize > READ_ONCE(sk->sk_rcvbuf)) ||
1236 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1237 !sock_flag(sk, SOCK_DEAD))
1238 *timeo = schedule_timeout(*timeo);
1239
1240 __set_current_state(TASK_RUNNING);
1241 remove_wait_queue(&nlk->wait, &wait);
1242 sock_put(sk);
1243
1244 if (signal_pending(current)) {
1245 kfree_skb(skb);
1246 return sock_intr_errno(*timeo);
1247 }
1248
1249 return 1;
1250 }
1251
__netlink_sendskb(struct sock * sk,struct sk_buff * skb)1252 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1253 {
1254 int len = skb->len;
1255
1256 netlink_deliver_tap(sock_net(sk), skb);
1257
1258 skb_queue_tail(&sk->sk_receive_queue, skb);
1259 sk->sk_data_ready(sk);
1260 return len;
1261 }
1262
netlink_sendskb(struct sock * sk,struct sk_buff * skb)1263 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1264 {
1265 int len = __netlink_sendskb(sk, skb);
1266
1267 sock_put(sk);
1268 return len;
1269 }
1270
netlink_detachskb(struct sock * sk,struct sk_buff * skb)1271 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1272 {
1273 kfree_skb(skb);
1274 sock_put(sk);
1275 }
1276
netlink_trim(struct sk_buff * skb,gfp_t allocation)1277 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1278 {
1279 int delta;
1280
1281 WARN_ON(skb->sk != NULL);
1282 delta = skb->end - skb->tail;
1283 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1284 return skb;
1285
1286 if (skb_shared(skb)) {
1287 struct sk_buff *nskb = skb_clone(skb, allocation);
1288 if (!nskb)
1289 return skb;
1290 consume_skb(skb);
1291 skb = nskb;
1292 }
1293
1294 pskb_expand_head(skb, 0, -delta,
1295 (allocation & ~__GFP_DIRECT_RECLAIM) |
1296 __GFP_NOWARN | __GFP_NORETRY);
1297 return skb;
1298 }
1299
netlink_unicast_kernel(struct sock * sk,struct sk_buff * skb,struct sock * ssk)1300 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1301 struct sock *ssk)
1302 {
1303 int ret;
1304 struct netlink_sock *nlk = nlk_sk(sk);
1305
1306 ret = -ECONNREFUSED;
1307 if (nlk->netlink_rcv != NULL) {
1308 ret = skb->len;
1309 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
1310 netlink_skb_set_owner_r(skb, sk);
1311 NETLINK_CB(skb).sk = ssk;
1312 netlink_deliver_tap_kernel(sk, ssk, skb);
1313 nlk->netlink_rcv(skb);
1314 consume_skb(skb);
1315 } else {
1316 kfree_skb(skb);
1317 }
1318 sock_put(sk);
1319 return ret;
1320 }
1321
netlink_unicast(struct sock * ssk,struct sk_buff * skb,u32 portid,int nonblock)1322 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1323 u32 portid, int nonblock)
1324 {
1325 struct sock *sk;
1326 int err;
1327 long timeo;
1328
1329 skb = netlink_trim(skb, gfp_any());
1330
1331 timeo = sock_sndtimeo(ssk, nonblock);
1332 retry:
1333 sk = netlink_getsockbyportid(ssk, portid);
1334 if (IS_ERR(sk)) {
1335 kfree_skb(skb);
1336 return PTR_ERR(sk);
1337 }
1338 if (netlink_is_kernel(sk))
1339 return netlink_unicast_kernel(sk, skb, ssk);
1340
1341 if (sk_filter(sk, skb)) {
1342 err = skb->len;
1343 kfree_skb(skb);
1344 sock_put(sk);
1345 return err;
1346 }
1347
1348 err = netlink_attachskb(sk, skb, &timeo, ssk);
1349 if (err == 1)
1350 goto retry;
1351 if (err)
1352 return err;
1353
1354 return netlink_sendskb(sk, skb);
1355 }
1356 EXPORT_SYMBOL(netlink_unicast);
1357
netlink_has_listeners(struct sock * sk,unsigned int group)1358 int netlink_has_listeners(struct sock *sk, unsigned int group)
1359 {
1360 int res = 0;
1361 struct listeners *listeners;
1362
1363 BUG_ON(!netlink_is_kernel(sk));
1364
1365 rcu_read_lock();
1366 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1367
1368 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1369 res = test_bit(group - 1, listeners->masks);
1370
1371 rcu_read_unlock();
1372
1373 return res;
1374 }
1375 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1376
netlink_strict_get_check(struct sk_buff * skb)1377 bool netlink_strict_get_check(struct sk_buff *skb)
1378 {
1379 return nlk_test_bit(STRICT_CHK, NETLINK_CB(skb).sk);
1380 }
1381 EXPORT_SYMBOL_GPL(netlink_strict_get_check);
1382
netlink_broadcast_deliver(struct sock * sk,struct sk_buff * skb)1383 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1384 {
1385 struct netlink_sock *nlk = nlk_sk(sk);
1386 unsigned int rmem, rcvbuf;
1387
1388 rmem = atomic_add_return(skb->truesize, &sk->sk_rmem_alloc);
1389 rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1390
1391 if ((rmem == skb->truesize || rmem <= rcvbuf) &&
1392 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1393 netlink_skb_set_owner_r(skb, sk);
1394 __netlink_sendskb(sk, skb);
1395 return rmem > (rcvbuf >> 1);
1396 }
1397
1398 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1399 return -1;
1400 }
1401
1402 struct netlink_broadcast_data {
1403 struct sock *exclude_sk;
1404 struct net *net;
1405 u32 portid;
1406 u32 group;
1407 int failure;
1408 int delivery_failure;
1409 int congested;
1410 int delivered;
1411 gfp_t allocation;
1412 struct sk_buff *skb, *skb2;
1413 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1414 void *tx_data;
1415 };
1416
do_one_broadcast(struct sock * sk,struct netlink_broadcast_data * p)1417 static void do_one_broadcast(struct sock *sk,
1418 struct netlink_broadcast_data *p)
1419 {
1420 struct netlink_sock *nlk = nlk_sk(sk);
1421 int val;
1422
1423 if (p->exclude_sk == sk)
1424 return;
1425
1426 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1427 !test_bit(p->group - 1, nlk->groups))
1428 return;
1429
1430 if (!net_eq(sock_net(sk), p->net)) {
1431 if (!nlk_test_bit(LISTEN_ALL_NSID, sk))
1432 return;
1433
1434 if (!peernet_has_id(sock_net(sk), p->net))
1435 return;
1436
1437 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1438 CAP_NET_BROADCAST))
1439 return;
1440 }
1441
1442 if (p->failure) {
1443 netlink_overrun(sk);
1444 return;
1445 }
1446
1447 sock_hold(sk);
1448 if (p->skb2 == NULL) {
1449 if (skb_shared(p->skb)) {
1450 p->skb2 = skb_clone(p->skb, p->allocation);
1451 } else {
1452 p->skb2 = skb_get(p->skb);
1453 /*
1454 * skb ownership may have been set when
1455 * delivered to a previous socket.
1456 */
1457 skb_orphan(p->skb2);
1458 }
1459 }
1460 if (p->skb2 == NULL) {
1461 netlink_overrun(sk);
1462 /* Clone failed. Notify ALL listeners. */
1463 p->failure = 1;
1464 if (nlk_test_bit(BROADCAST_SEND_ERROR, sk))
1465 p->delivery_failure = 1;
1466 goto out;
1467 }
1468
1469 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1470 kfree_skb(p->skb2);
1471 p->skb2 = NULL;
1472 goto out;
1473 }
1474
1475 if (sk_filter(sk, p->skb2)) {
1476 kfree_skb(p->skb2);
1477 p->skb2 = NULL;
1478 goto out;
1479 }
1480 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1481 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1482 NETLINK_CB(p->skb2).nsid_is_set = true;
1483 val = netlink_broadcast_deliver(sk, p->skb2);
1484 if (val < 0) {
1485 netlink_overrun(sk);
1486 if (nlk_test_bit(BROADCAST_SEND_ERROR, sk))
1487 p->delivery_failure = 1;
1488 } else {
1489 p->congested |= val;
1490 p->delivered = 1;
1491 p->skb2 = NULL;
1492 }
1493 out:
1494 sock_put(sk);
1495 }
1496
netlink_broadcast_filtered(struct sock * ssk,struct sk_buff * skb,u32 portid,u32 group,gfp_t allocation,netlink_filter_fn filter,void * filter_data)1497 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb,
1498 u32 portid,
1499 u32 group, gfp_t allocation,
1500 netlink_filter_fn filter,
1501 void *filter_data)
1502 {
1503 struct net *net = sock_net(ssk);
1504 struct netlink_broadcast_data info;
1505 struct sock *sk;
1506
1507 skb = netlink_trim(skb, allocation);
1508
1509 info.exclude_sk = ssk;
1510 info.net = net;
1511 info.portid = portid;
1512 info.group = group;
1513 info.failure = 0;
1514 info.delivery_failure = 0;
1515 info.congested = 0;
1516 info.delivered = 0;
1517 info.allocation = allocation;
1518 info.skb = skb;
1519 info.skb2 = NULL;
1520 info.tx_filter = filter;
1521 info.tx_data = filter_data;
1522
1523 /* While we sleep in clone, do not allow to change socket list */
1524
1525 netlink_lock_table();
1526
1527 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1528 do_one_broadcast(sk, &info);
1529
1530 consume_skb(skb);
1531
1532 netlink_unlock_table();
1533
1534 if (info.delivery_failure) {
1535 kfree_skb(info.skb2);
1536 return -ENOBUFS;
1537 }
1538 consume_skb(info.skb2);
1539
1540 if (info.delivered) {
1541 if (info.congested && gfpflags_allow_blocking(allocation))
1542 yield();
1543 return 0;
1544 }
1545 return -ESRCH;
1546 }
1547 EXPORT_SYMBOL(netlink_broadcast_filtered);
1548
netlink_broadcast(struct sock * ssk,struct sk_buff * skb,u32 portid,u32 group,gfp_t allocation)1549 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1550 u32 group, gfp_t allocation)
1551 {
1552 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1553 NULL, NULL);
1554 }
1555 EXPORT_SYMBOL(netlink_broadcast);
1556
1557 struct netlink_set_err_data {
1558 struct sock *exclude_sk;
1559 u32 portid;
1560 u32 group;
1561 int code;
1562 };
1563
do_one_set_err(struct sock * sk,struct netlink_set_err_data * p)1564 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1565 {
1566 struct netlink_sock *nlk = nlk_sk(sk);
1567 int ret = 0;
1568
1569 if (sk == p->exclude_sk)
1570 goto out;
1571
1572 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1573 goto out;
1574
1575 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1576 !test_bit(p->group - 1, nlk->groups))
1577 goto out;
1578
1579 if (p->code == ENOBUFS && nlk_test_bit(RECV_NO_ENOBUFS, sk)) {
1580 ret = 1;
1581 goto out;
1582 }
1583
1584 WRITE_ONCE(sk->sk_err, p->code);
1585 sk_error_report(sk);
1586 out:
1587 return ret;
1588 }
1589
1590 /**
1591 * netlink_set_err - report error to broadcast listeners
1592 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1593 * @portid: the PORTID of a process that we want to skip (if any)
1594 * @group: the broadcast group that will notice the error
1595 * @code: error code, must be negative (as usual in kernelspace)
1596 *
1597 * This function returns the number of broadcast listeners that have set the
1598 * NETLINK_NO_ENOBUFS socket option.
1599 */
netlink_set_err(struct sock * ssk,u32 portid,u32 group,int code)1600 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1601 {
1602 struct netlink_set_err_data info;
1603 unsigned long flags;
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_irqsave(&nl_table_lock, flags);
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_irqrestore(&nl_table_lock, flags);
1619 return ret;
1620 }
1621 EXPORT_SYMBOL(netlink_set_err);
1622
1623 /* must be called with netlink table grabbed */
netlink_update_socket_mc(struct netlink_sock * nlk,unsigned int group,int is_new)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 __assign_bit(group - 1, nlk->groups, new);
1633 netlink_update_subscriptions(&nlk->sk, subscriptions);
1634 netlink_update_listeners(&nlk->sk);
1635 }
1636
netlink_setsockopt(struct socket * sock,int level,int optname,sockptr_t optval,unsigned int optlen)1637 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1638 sockptr_t optval, unsigned int optlen)
1639 {
1640 struct sock *sk = sock->sk;
1641 struct netlink_sock *nlk = nlk_sk(sk);
1642 unsigned int val = 0;
1643 int nr = -1;
1644
1645 if (level != SOL_NETLINK)
1646 return -ENOPROTOOPT;
1647
1648 if (optlen >= sizeof(int) &&
1649 copy_from_sockptr(&val, optval, sizeof(val)))
1650 return -EFAULT;
1651
1652 switch (optname) {
1653 case NETLINK_PKTINFO:
1654 nr = NETLINK_F_RECV_PKTINFO;
1655 break;
1656 case NETLINK_ADD_MEMBERSHIP:
1657 case NETLINK_DROP_MEMBERSHIP: {
1658 int err;
1659
1660 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1661 return -EPERM;
1662 err = netlink_realloc_groups(sk);
1663 if (err)
1664 return err;
1665 if (!val || val - 1 >= nlk->ngroups)
1666 return -EINVAL;
1667 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1668 err = nlk->netlink_bind(sock_net(sk), val);
1669 if (err)
1670 return err;
1671 }
1672 netlink_table_grab();
1673 netlink_update_socket_mc(nlk, val,
1674 optname == NETLINK_ADD_MEMBERSHIP);
1675 netlink_table_ungrab();
1676 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1677 nlk->netlink_unbind(sock_net(sk), val);
1678
1679 break;
1680 }
1681 case NETLINK_BROADCAST_ERROR:
1682 nr = NETLINK_F_BROADCAST_SEND_ERROR;
1683 break;
1684 case NETLINK_NO_ENOBUFS:
1685 assign_bit(NETLINK_F_RECV_NO_ENOBUFS, &nlk->flags, val);
1686 if (val) {
1687 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1688 wake_up_interruptible(&nlk->wait);
1689 }
1690 break;
1691 case NETLINK_LISTEN_ALL_NSID:
1692 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1693 return -EPERM;
1694 nr = NETLINK_F_LISTEN_ALL_NSID;
1695 break;
1696 case NETLINK_CAP_ACK:
1697 nr = NETLINK_F_CAP_ACK;
1698 break;
1699 case NETLINK_EXT_ACK:
1700 nr = NETLINK_F_EXT_ACK;
1701 break;
1702 case NETLINK_GET_STRICT_CHK:
1703 nr = NETLINK_F_STRICT_CHK;
1704 break;
1705 default:
1706 return -ENOPROTOOPT;
1707 }
1708 if (nr >= 0)
1709 assign_bit(nr, &nlk->flags, val);
1710 return 0;
1711 }
1712
netlink_getsockopt(struct socket * sock,int level,int optname,char __user * optval,int __user * optlen)1713 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1714 char __user *optval, int __user *optlen)
1715 {
1716 struct sock *sk = sock->sk;
1717 struct netlink_sock *nlk = nlk_sk(sk);
1718 unsigned int flag;
1719 int len, val;
1720
1721 if (level != SOL_NETLINK)
1722 return -ENOPROTOOPT;
1723
1724 if (get_user(len, optlen))
1725 return -EFAULT;
1726 if (len < 0)
1727 return -EINVAL;
1728
1729 switch (optname) {
1730 case NETLINK_PKTINFO:
1731 flag = NETLINK_F_RECV_PKTINFO;
1732 break;
1733 case NETLINK_BROADCAST_ERROR:
1734 flag = NETLINK_F_BROADCAST_SEND_ERROR;
1735 break;
1736 case NETLINK_NO_ENOBUFS:
1737 flag = NETLINK_F_RECV_NO_ENOBUFS;
1738 break;
1739 case NETLINK_LIST_MEMBERSHIPS: {
1740 int pos, idx, shift, err = 0;
1741
1742 netlink_lock_table();
1743 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1744 if (len - pos < sizeof(u32))
1745 break;
1746
1747 idx = pos / sizeof(unsigned long);
1748 shift = (pos % sizeof(unsigned long)) * 8;
1749 if (put_user((u32)(nlk->groups[idx] >> shift),
1750 (u32 __user *)(optval + pos))) {
1751 err = -EFAULT;
1752 break;
1753 }
1754 }
1755 if (put_user(ALIGN(BITS_TO_BYTES(nlk->ngroups), sizeof(u32)), optlen))
1756 err = -EFAULT;
1757 netlink_unlock_table();
1758 return err;
1759 }
1760 case NETLINK_LISTEN_ALL_NSID:
1761 flag = NETLINK_F_LISTEN_ALL_NSID;
1762 break;
1763 case NETLINK_CAP_ACK:
1764 flag = NETLINK_F_CAP_ACK;
1765 break;
1766 case NETLINK_EXT_ACK:
1767 flag = NETLINK_F_EXT_ACK;
1768 break;
1769 case NETLINK_GET_STRICT_CHK:
1770 flag = NETLINK_F_STRICT_CHK;
1771 break;
1772 default:
1773 return -ENOPROTOOPT;
1774 }
1775
1776 if (len < sizeof(int))
1777 return -EINVAL;
1778
1779 len = sizeof(int);
1780 val = test_bit(flag, &nlk->flags);
1781
1782 if (put_user(len, optlen) ||
1783 copy_to_user(optval, &val, len))
1784 return -EFAULT;
1785
1786 return 0;
1787 }
1788
netlink_cmsg_recv_pktinfo(struct msghdr * msg,struct sk_buff * skb)1789 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1790 {
1791 struct nl_pktinfo info;
1792
1793 info.group = NETLINK_CB(skb).dst_group;
1794 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1795 }
1796
netlink_cmsg_listen_all_nsid(struct sock * sk,struct msghdr * msg,struct sk_buff * skb)1797 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1798 struct sk_buff *skb)
1799 {
1800 if (!NETLINK_CB(skb).nsid_is_set)
1801 return;
1802
1803 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1804 &NETLINK_CB(skb).nsid);
1805 }
1806
netlink_sendmsg(struct socket * sock,struct msghdr * msg,size_t len)1807 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1808 {
1809 struct sock *sk = sock->sk;
1810 struct netlink_sock *nlk = nlk_sk(sk);
1811 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1812 u32 dst_portid;
1813 u32 dst_group;
1814 struct sk_buff *skb;
1815 int err;
1816 struct scm_cookie scm;
1817 u32 netlink_skb_flags = 0;
1818
1819 if (msg->msg_flags & MSG_OOB)
1820 return -EOPNOTSUPP;
1821
1822 if (len == 0) {
1823 pr_warn_once("Zero length message leads to an empty skb\n");
1824 return -ENODATA;
1825 }
1826
1827 err = scm_send(sock, msg, &scm, true);
1828 if (err < 0)
1829 return err;
1830
1831 if (msg->msg_namelen) {
1832 err = -EINVAL;
1833 if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1834 goto out;
1835 if (addr->nl_family != AF_NETLINK)
1836 goto out;
1837 dst_portid = addr->nl_pid;
1838 dst_group = ffs(addr->nl_groups);
1839 err = -EPERM;
1840 if ((dst_group || dst_portid) &&
1841 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1842 goto out;
1843 netlink_skb_flags |= NETLINK_SKB_DST;
1844 } else {
1845 /* Paired with WRITE_ONCE() in netlink_connect() */
1846 dst_portid = READ_ONCE(nlk->dst_portid);
1847 dst_group = READ_ONCE(nlk->dst_group);
1848 }
1849
1850 /* Paired with WRITE_ONCE() in netlink_insert() */
1851 if (!READ_ONCE(nlk->bound)) {
1852 err = netlink_autobind(sock);
1853 if (err)
1854 goto out;
1855 } else {
1856 /* Ensure nlk is hashed and visible. */
1857 smp_rmb();
1858 }
1859
1860 err = -EMSGSIZE;
1861 if (len > sk->sk_sndbuf - 32)
1862 goto out;
1863 err = -ENOBUFS;
1864 skb = netlink_alloc_large_skb(len, dst_group);
1865 if (skb == NULL)
1866 goto out;
1867
1868 NETLINK_CB(skb).portid = nlk->portid;
1869 NETLINK_CB(skb).dst_group = dst_group;
1870 NETLINK_CB(skb).creds = scm.creds;
1871 NETLINK_CB(skb).flags = netlink_skb_flags;
1872
1873 err = -EFAULT;
1874 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1875 kfree_skb(skb);
1876 goto out;
1877 }
1878
1879 err = security_netlink_send(sk, skb);
1880 if (err) {
1881 kfree_skb(skb);
1882 goto out;
1883 }
1884
1885 if (dst_group) {
1886 refcount_inc(&skb->users);
1887 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1888 }
1889 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags & MSG_DONTWAIT);
1890
1891 out:
1892 scm_destroy(&scm);
1893 return err;
1894 }
1895
netlink_recvmsg(struct socket * sock,struct msghdr * msg,size_t len,int flags)1896 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1897 int flags)
1898 {
1899 struct scm_cookie scm;
1900 struct sock *sk = sock->sk;
1901 struct netlink_sock *nlk = nlk_sk(sk);
1902 size_t copied, max_recvmsg_len;
1903 struct sk_buff *skb, *data_skb;
1904 int err, ret;
1905
1906 if (flags & MSG_OOB)
1907 return -EOPNOTSUPP;
1908
1909 copied = 0;
1910
1911 skb = skb_recv_datagram(sk, flags, &err);
1912 if (skb == NULL)
1913 goto out;
1914
1915 data_skb = skb;
1916
1917 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1918 if (unlikely(skb_shinfo(skb)->frag_list)) {
1919 /*
1920 * If this skb has a frag_list, then here that means that we
1921 * will have to use the frag_list skb's data for compat tasks
1922 * and the regular skb's data for normal (non-compat) tasks.
1923 *
1924 * If we need to send the compat skb, assign it to the
1925 * 'data_skb' variable so that it will be used below for data
1926 * copying. We keep 'skb' for everything else, including
1927 * freeing both later.
1928 */
1929 if (flags & MSG_CMSG_COMPAT)
1930 data_skb = skb_shinfo(skb)->frag_list;
1931 }
1932 #endif
1933
1934 /* Record the max length of recvmsg() calls for future allocations */
1935 max_recvmsg_len = max(READ_ONCE(nlk->max_recvmsg_len), len);
1936 max_recvmsg_len = min_t(size_t, max_recvmsg_len,
1937 SKB_WITH_OVERHEAD(32768));
1938 WRITE_ONCE(nlk->max_recvmsg_len, max_recvmsg_len);
1939
1940 copied = data_skb->len;
1941 if (len < copied) {
1942 msg->msg_flags |= MSG_TRUNC;
1943 copied = len;
1944 }
1945
1946 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1947
1948 if (msg->msg_name) {
1949 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1950 addr->nl_family = AF_NETLINK;
1951 addr->nl_pad = 0;
1952 addr->nl_pid = NETLINK_CB(skb).portid;
1953 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1954 msg->msg_namelen = sizeof(*addr);
1955 }
1956
1957 if (nlk_test_bit(RECV_PKTINFO, sk))
1958 netlink_cmsg_recv_pktinfo(msg, skb);
1959 if (nlk_test_bit(LISTEN_ALL_NSID, sk))
1960 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1961
1962 memset(&scm, 0, sizeof(scm));
1963 scm.creds = *NETLINK_CREDS(skb);
1964 if (flags & MSG_TRUNC)
1965 copied = data_skb->len;
1966
1967 skb_free_datagram(sk, skb);
1968
1969 if (READ_ONCE(nlk->cb_running) &&
1970 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1971 ret = netlink_dump(sk, false);
1972 if (ret) {
1973 WRITE_ONCE(sk->sk_err, -ret);
1974 sk_error_report(sk);
1975 }
1976 }
1977
1978 scm_recv(sock, msg, &scm, flags);
1979 out:
1980 netlink_rcv_wake(sk);
1981 return err ? : copied;
1982 }
1983
netlink_data_ready(struct sock * sk)1984 static void netlink_data_ready(struct sock *sk)
1985 {
1986 BUG();
1987 }
1988
1989 /*
1990 * We export these functions to other modules. They provide a
1991 * complete set of kernel non-blocking support for message
1992 * queueing.
1993 */
1994
1995 struct sock *
__netlink_kernel_create(struct net * net,int unit,struct module * module,struct netlink_kernel_cfg * cfg)1996 __netlink_kernel_create(struct net *net, int unit, struct module *module,
1997 struct netlink_kernel_cfg *cfg)
1998 {
1999 struct socket *sock;
2000 struct sock *sk;
2001 struct netlink_sock *nlk;
2002 struct listeners *listeners = NULL;
2003 unsigned int groups;
2004
2005 BUG_ON(!nl_table);
2006
2007 if (unit < 0 || unit >= MAX_LINKS)
2008 return NULL;
2009
2010 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2011 return NULL;
2012
2013 if (__netlink_create(net, sock, unit, 1) < 0)
2014 goto out_sock_release_nosk;
2015
2016 sk = sock->sk;
2017
2018 if (!cfg || cfg->groups < 32)
2019 groups = 32;
2020 else
2021 groups = cfg->groups;
2022
2023 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2024 if (!listeners)
2025 goto out_sock_release;
2026
2027 sk->sk_data_ready = netlink_data_ready;
2028 if (cfg && cfg->input)
2029 nlk_sk(sk)->netlink_rcv = cfg->input;
2030
2031 if (netlink_insert(sk, 0))
2032 goto out_sock_release;
2033
2034 nlk = nlk_sk(sk);
2035 set_bit(NETLINK_F_KERNEL_SOCKET, &nlk->flags);
2036
2037 netlink_table_grab();
2038 if (!nl_table[unit].registered) {
2039 nl_table[unit].groups = groups;
2040 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2041 nl_table[unit].module = module;
2042 if (cfg) {
2043 nl_table[unit].bind = cfg->bind;
2044 nl_table[unit].unbind = cfg->unbind;
2045 nl_table[unit].release = cfg->release;
2046 nl_table[unit].flags = cfg->flags;
2047 }
2048 nl_table[unit].registered = 1;
2049 } else {
2050 kfree(listeners);
2051 nl_table[unit].registered++;
2052 }
2053 netlink_table_ungrab();
2054 return sk;
2055
2056 out_sock_release:
2057 kfree(listeners);
2058 netlink_kernel_release(sk);
2059 return NULL;
2060
2061 out_sock_release_nosk:
2062 sock_release(sock);
2063 return NULL;
2064 }
2065 EXPORT_SYMBOL(__netlink_kernel_create);
2066
2067 void
netlink_kernel_release(struct sock * sk)2068 netlink_kernel_release(struct sock *sk)
2069 {
2070 if (sk == NULL || sk->sk_socket == NULL)
2071 return;
2072
2073 sock_release(sk->sk_socket);
2074 }
2075 EXPORT_SYMBOL(netlink_kernel_release);
2076
__netlink_change_ngroups(struct sock * sk,unsigned int groups)2077 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2078 {
2079 struct listeners *new, *old;
2080 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2081
2082 if (groups < 32)
2083 groups = 32;
2084
2085 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2086 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2087 if (!new)
2088 return -ENOMEM;
2089 old = nl_deref_protected(tbl->listeners);
2090 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2091 rcu_assign_pointer(tbl->listeners, new);
2092
2093 kfree_rcu(old, rcu);
2094 }
2095 tbl->groups = groups;
2096
2097 return 0;
2098 }
2099
2100 /**
2101 * netlink_change_ngroups - change number of multicast groups
2102 *
2103 * This changes the number of multicast groups that are available
2104 * on a certain netlink family. Note that it is not possible to
2105 * change the number of groups to below 32. Also note that it does
2106 * not implicitly call netlink_clear_multicast_users() when the
2107 * number of groups is reduced.
2108 *
2109 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2110 * @groups: The new number of groups.
2111 */
netlink_change_ngroups(struct sock * sk,unsigned int groups)2112 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2113 {
2114 int err;
2115
2116 netlink_table_grab();
2117 err = __netlink_change_ngroups(sk, groups);
2118 netlink_table_ungrab();
2119
2120 return err;
2121 }
2122
__netlink_clear_multicast_users(struct sock * ksk,unsigned int group)2123 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2124 {
2125 struct sock *sk;
2126 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2127 struct hlist_node *tmp;
2128
2129 sk_for_each_bound_safe(sk, tmp, &tbl->mc_list)
2130 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2131 }
2132
2133 struct nlmsghdr *
__nlmsg_put(struct sk_buff * skb,u32 portid,u32 seq,int type,int len,int flags)2134 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2135 {
2136 struct nlmsghdr *nlh;
2137 int size = nlmsg_msg_size(len);
2138
2139 nlh = skb_put(skb, NLMSG_ALIGN(size));
2140 nlh->nlmsg_type = type;
2141 nlh->nlmsg_len = size;
2142 nlh->nlmsg_flags = flags;
2143 nlh->nlmsg_pid = portid;
2144 nlh->nlmsg_seq = seq;
2145 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2146 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2147 return nlh;
2148 }
2149 EXPORT_SYMBOL(__nlmsg_put);
2150
2151 static size_t
netlink_ack_tlv_len(struct netlink_sock * nlk,int err,const struct netlink_ext_ack * extack)2152 netlink_ack_tlv_len(struct netlink_sock *nlk, int err,
2153 const struct netlink_ext_ack *extack)
2154 {
2155 size_t tlvlen;
2156
2157 if (!extack || !test_bit(NETLINK_F_EXT_ACK, &nlk->flags))
2158 return 0;
2159
2160 tlvlen = 0;
2161 if (extack->_msg)
2162 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2163 if (extack->cookie_len)
2164 tlvlen += nla_total_size(extack->cookie_len);
2165
2166 /* Following attributes are only reported as error (not warning) */
2167 if (!err)
2168 return tlvlen;
2169
2170 if (extack->bad_attr)
2171 tlvlen += nla_total_size(sizeof(u32));
2172 if (extack->policy)
2173 tlvlen += netlink_policy_dump_attr_size_estimate(extack->policy);
2174 if (extack->miss_type)
2175 tlvlen += nla_total_size(sizeof(u32));
2176 if (extack->miss_nest)
2177 tlvlen += nla_total_size(sizeof(u32));
2178
2179 return tlvlen;
2180 }
2181
nlmsg_check_in_payload(const struct nlmsghdr * nlh,const void * addr)2182 static bool nlmsg_check_in_payload(const struct nlmsghdr *nlh, const void *addr)
2183 {
2184 return !WARN_ON(addr < nlmsg_data(nlh) ||
2185 addr - (const void *) nlh >= nlh->nlmsg_len);
2186 }
2187
2188 static void
netlink_ack_tlv_fill(struct sk_buff * skb,const struct nlmsghdr * nlh,int err,const struct netlink_ext_ack * extack)2189 netlink_ack_tlv_fill(struct sk_buff *skb, const struct nlmsghdr *nlh, int err,
2190 const struct netlink_ext_ack *extack)
2191 {
2192 if (extack->_msg)
2193 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg));
2194 if (extack->cookie_len)
2195 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2196 extack->cookie_len, extack->cookie));
2197
2198 if (!err)
2199 return;
2200
2201 if (extack->bad_attr && nlmsg_check_in_payload(nlh, extack->bad_attr))
2202 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2203 (u8 *)extack->bad_attr - (const u8 *)nlh));
2204 if (extack->policy)
2205 netlink_policy_dump_write_attr(skb, extack->policy,
2206 NLMSGERR_ATTR_POLICY);
2207 if (extack->miss_type)
2208 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_TYPE,
2209 extack->miss_type));
2210 if (extack->miss_nest && nlmsg_check_in_payload(nlh, extack->miss_nest))
2211 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_NEST,
2212 (u8 *)extack->miss_nest - (const u8 *)nlh));
2213 }
2214
2215 /*
2216 * It looks a bit ugly.
2217 * It would be better to create kernel thread.
2218 */
2219
netlink_dump_done(struct netlink_sock * nlk,struct sk_buff * skb,struct netlink_callback * cb,struct netlink_ext_ack * extack)2220 static int netlink_dump_done(struct netlink_sock *nlk, struct sk_buff *skb,
2221 struct netlink_callback *cb,
2222 struct netlink_ext_ack *extack)
2223 {
2224 struct nlmsghdr *nlh;
2225 size_t extack_len;
2226
2227 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(nlk->dump_done_errno),
2228 NLM_F_MULTI | cb->answer_flags);
2229 if (WARN_ON(!nlh))
2230 return -ENOBUFS;
2231
2232 nl_dump_check_consistent(cb, nlh);
2233 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno, sizeof(nlk->dump_done_errno));
2234
2235 extack_len = netlink_ack_tlv_len(nlk, nlk->dump_done_errno, extack);
2236 if (extack_len) {
2237 nlh->nlmsg_flags |= NLM_F_ACK_TLVS;
2238 if (skb_tailroom(skb) >= extack_len) {
2239 netlink_ack_tlv_fill(skb, cb->nlh,
2240 nlk->dump_done_errno, extack);
2241 nlmsg_end(skb, nlh);
2242 }
2243 }
2244
2245 return 0;
2246 }
2247
netlink_dump(struct sock * sk,bool lock_taken)2248 static int netlink_dump(struct sock *sk, bool lock_taken)
2249 {
2250 struct netlink_sock *nlk = nlk_sk(sk);
2251 struct netlink_ext_ack extack = {};
2252 struct netlink_callback *cb;
2253 struct sk_buff *skb = NULL;
2254 unsigned int rmem, rcvbuf;
2255 size_t max_recvmsg_len;
2256 struct module *module;
2257 int err = -ENOBUFS;
2258 int alloc_min_size;
2259 int alloc_size;
2260
2261 if (!lock_taken)
2262 mutex_lock(&nlk->nl_cb_mutex);
2263 if (!nlk->cb_running) {
2264 err = -EINVAL;
2265 goto errout_skb;
2266 }
2267
2268 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2269 * required, but it makes sense to _attempt_ a 16K bytes allocation
2270 * to reduce number of system calls on dump operations, if user
2271 * ever provided a big enough buffer.
2272 */
2273 cb = &nlk->cb;
2274 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2275
2276 max_recvmsg_len = READ_ONCE(nlk->max_recvmsg_len);
2277 if (alloc_min_size < max_recvmsg_len) {
2278 alloc_size = max_recvmsg_len;
2279 skb = alloc_skb(alloc_size,
2280 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2281 __GFP_NOWARN | __GFP_NORETRY);
2282 }
2283 if (!skb) {
2284 alloc_size = alloc_min_size;
2285 skb = alloc_skb(alloc_size, GFP_KERNEL);
2286 }
2287 if (!skb)
2288 goto errout_skb;
2289
2290 rcvbuf = READ_ONCE(sk->sk_rcvbuf);
2291 rmem = atomic_add_return(skb->truesize, &sk->sk_rmem_alloc);
2292 if (rmem != skb->truesize && rmem >= rcvbuf) {
2293 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
2294 goto errout_skb;
2295 }
2296
2297 /* Trim skb to allocated size. User is expected to provide buffer as
2298 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2299 * netlink_recvmsg())). dump will pack as many smaller messages as
2300 * could fit within the allocated skb. skb is typically allocated
2301 * with larger space than required (could be as much as near 2x the
2302 * requested size with align to next power of 2 approach). Allowing
2303 * dump to use the excess space makes it difficult for a user to have a
2304 * reasonable static buffer based on the expected largest dump of a
2305 * single netdev. The outcome is MSG_TRUNC error.
2306 */
2307 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2308
2309 /* Make sure malicious BPF programs can not read unitialized memory
2310 * from skb->head -> skb->data
2311 */
2312 skb_reset_network_header(skb);
2313 skb_reset_mac_header(skb);
2314
2315 netlink_skb_set_owner_r(skb, sk);
2316
2317 if (nlk->dump_done_errno > 0) {
2318 cb->extack = &extack;
2319
2320 nlk->dump_done_errno = cb->dump(skb, cb);
2321
2322 /* EMSGSIZE plus something already in the skb means
2323 * that there's more to dump but current skb has filled up.
2324 * If the callback really wants to return EMSGSIZE to user space
2325 * it needs to do so again, on the next cb->dump() call,
2326 * without putting data in the skb.
2327 */
2328 if (nlk->dump_done_errno == -EMSGSIZE && skb->len)
2329 nlk->dump_done_errno = skb->len;
2330
2331 cb->extack = NULL;
2332 }
2333
2334 if (nlk->dump_done_errno > 0 ||
2335 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2336 mutex_unlock(&nlk->nl_cb_mutex);
2337
2338 if (sk_filter(sk, skb))
2339 kfree_skb(skb);
2340 else
2341 __netlink_sendskb(sk, skb);
2342 return 0;
2343 }
2344
2345 if (netlink_dump_done(nlk, skb, cb, &extack))
2346 goto errout_skb;
2347
2348 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2349 /* frag_list skb's data is used for compat tasks
2350 * and the regular skb's data for normal (non-compat) tasks.
2351 * See netlink_recvmsg().
2352 */
2353 if (unlikely(skb_shinfo(skb)->frag_list)) {
2354 if (netlink_dump_done(nlk, skb_shinfo(skb)->frag_list, cb, &extack))
2355 goto errout_skb;
2356 }
2357 #endif
2358
2359 if (sk_filter(sk, skb))
2360 kfree_skb(skb);
2361 else
2362 __netlink_sendskb(sk, skb);
2363
2364 if (cb->done)
2365 cb->done(cb);
2366
2367 WRITE_ONCE(nlk->cb_running, false);
2368 module = cb->module;
2369 skb = cb->skb;
2370 mutex_unlock(&nlk->nl_cb_mutex);
2371 module_put(module);
2372 consume_skb(skb);
2373 return 0;
2374
2375 errout_skb:
2376 mutex_unlock(&nlk->nl_cb_mutex);
2377 kfree_skb(skb);
2378 return err;
2379 }
2380
__netlink_dump_start(struct sock * ssk,struct sk_buff * skb,const struct nlmsghdr * nlh,struct netlink_dump_control * control)2381 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2382 const struct nlmsghdr *nlh,
2383 struct netlink_dump_control *control)
2384 {
2385 struct netlink_callback *cb;
2386 struct netlink_sock *nlk;
2387 struct sock *sk;
2388 int ret;
2389
2390 refcount_inc(&skb->users);
2391
2392 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2393 if (sk == NULL) {
2394 ret = -ECONNREFUSED;
2395 goto error_free;
2396 }
2397
2398 nlk = nlk_sk(sk);
2399 mutex_lock(&nlk->nl_cb_mutex);
2400 /* A dump is in progress... */
2401 if (nlk->cb_running) {
2402 ret = -EBUSY;
2403 goto error_unlock;
2404 }
2405 /* add reference of module which cb->dump belongs to */
2406 if (!try_module_get(control->module)) {
2407 ret = -EPROTONOSUPPORT;
2408 goto error_unlock;
2409 }
2410
2411 cb = &nlk->cb;
2412 memset(cb, 0, sizeof(*cb));
2413 cb->dump = control->dump;
2414 cb->done = control->done;
2415 cb->nlh = nlh;
2416 cb->data = control->data;
2417 cb->module = control->module;
2418 cb->min_dump_alloc = control->min_dump_alloc;
2419 cb->flags = control->flags;
2420 cb->skb = skb;
2421
2422 cb->strict_check = nlk_test_bit(STRICT_CHK, NETLINK_CB(skb).sk);
2423
2424 if (control->start) {
2425 cb->extack = control->extack;
2426 ret = control->start(cb);
2427 cb->extack = NULL;
2428 if (ret)
2429 goto error_put;
2430 }
2431
2432 WRITE_ONCE(nlk->cb_running, true);
2433 nlk->dump_done_errno = INT_MAX;
2434
2435 ret = netlink_dump(sk, true);
2436
2437 sock_put(sk);
2438
2439 if (ret)
2440 return ret;
2441
2442 /* We successfully started a dump, by returning -EINTR we
2443 * signal not to send ACK even if it was requested.
2444 */
2445 return -EINTR;
2446
2447 error_put:
2448 module_put(control->module);
2449 error_unlock:
2450 sock_put(sk);
2451 mutex_unlock(&nlk->nl_cb_mutex);
2452 error_free:
2453 kfree_skb(skb);
2454 return ret;
2455 }
2456 EXPORT_SYMBOL(__netlink_dump_start);
2457
netlink_ack(struct sk_buff * in_skb,struct nlmsghdr * nlh,int err,const struct netlink_ext_ack * extack)2458 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2459 const struct netlink_ext_ack *extack)
2460 {
2461 struct sk_buff *skb;
2462 struct nlmsghdr *rep;
2463 struct nlmsgerr *errmsg;
2464 size_t payload = sizeof(*errmsg);
2465 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2466 unsigned int flags = 0;
2467 size_t tlvlen;
2468
2469 /* Error messages get the original request appened, unless the user
2470 * requests to cap the error message, and get extra error data if
2471 * requested.
2472 */
2473 if (err && !test_bit(NETLINK_F_CAP_ACK, &nlk->flags))
2474 payload += nlmsg_len(nlh);
2475 else
2476 flags |= NLM_F_CAPPED;
2477
2478 tlvlen = netlink_ack_tlv_len(nlk, err, extack);
2479 if (tlvlen)
2480 flags |= NLM_F_ACK_TLVS;
2481
2482 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2483 if (!skb)
2484 goto err_skb;
2485
2486 rep = nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2487 NLMSG_ERROR, sizeof(*errmsg), flags);
2488 if (!rep)
2489 goto err_bad_put;
2490 errmsg = nlmsg_data(rep);
2491 errmsg->error = err;
2492 errmsg->msg = *nlh;
2493
2494 if (!(flags & NLM_F_CAPPED)) {
2495 if (!nlmsg_append(skb, nlmsg_len(nlh)))
2496 goto err_bad_put;
2497
2498 memcpy(nlmsg_data(&errmsg->msg), nlmsg_data(nlh),
2499 nlmsg_len(nlh));
2500 }
2501
2502 if (tlvlen)
2503 netlink_ack_tlv_fill(skb, nlh, err, extack);
2504
2505 nlmsg_end(skb, rep);
2506
2507 nlmsg_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid);
2508
2509 return;
2510
2511 err_bad_put:
2512 nlmsg_free(skb);
2513 err_skb:
2514 WRITE_ONCE(NETLINK_CB(in_skb).sk->sk_err, ENOBUFS);
2515 sk_error_report(NETLINK_CB(in_skb).sk);
2516 }
2517 EXPORT_SYMBOL(netlink_ack);
2518
netlink_rcv_skb(struct sk_buff * skb,int (* cb)(struct sk_buff *,struct nlmsghdr *,struct netlink_ext_ack *))2519 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2520 struct nlmsghdr *,
2521 struct netlink_ext_ack *))
2522 {
2523 struct netlink_ext_ack extack;
2524 struct nlmsghdr *nlh;
2525 int err;
2526
2527 while (skb->len >= nlmsg_total_size(0)) {
2528 int msglen;
2529
2530 memset(&extack, 0, sizeof(extack));
2531 nlh = nlmsg_hdr(skb);
2532 err = 0;
2533
2534 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2535 return 0;
2536
2537 /* Only requests are handled by the kernel */
2538 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2539 goto ack;
2540
2541 /* Skip control messages */
2542 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2543 goto ack;
2544
2545 err = cb(skb, nlh, &extack);
2546 if (err == -EINTR)
2547 goto skip;
2548
2549 ack:
2550 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2551 netlink_ack(skb, nlh, err, &extack);
2552
2553 skip:
2554 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2555 if (msglen > skb->len)
2556 msglen = skb->len;
2557 skb_pull(skb, msglen);
2558 }
2559
2560 return 0;
2561 }
2562 EXPORT_SYMBOL(netlink_rcv_skb);
2563
2564 /**
2565 * nlmsg_notify - send a notification netlink message
2566 * @sk: netlink socket to use
2567 * @skb: notification message
2568 * @portid: destination netlink portid for reports or 0
2569 * @group: destination multicast group or 0
2570 * @report: 1 to report back, 0 to disable
2571 * @flags: allocation flags
2572 */
nlmsg_notify(struct sock * sk,struct sk_buff * skb,u32 portid,unsigned int group,int report,gfp_t flags)2573 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2574 unsigned int group, int report, gfp_t flags)
2575 {
2576 int err = 0;
2577
2578 if (group) {
2579 int exclude_portid = 0;
2580
2581 if (report) {
2582 refcount_inc(&skb->users);
2583 exclude_portid = portid;
2584 }
2585
2586 /* errors reported via destination sk->sk_err, but propagate
2587 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2588 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2589 if (err == -ESRCH)
2590 err = 0;
2591 }
2592
2593 if (report) {
2594 int err2;
2595
2596 err2 = nlmsg_unicast(sk, skb, portid);
2597 if (!err)
2598 err = err2;
2599 }
2600
2601 return err;
2602 }
2603 EXPORT_SYMBOL(nlmsg_notify);
2604
2605 #ifdef CONFIG_PROC_FS
2606 struct nl_seq_iter {
2607 struct seq_net_private p;
2608 struct rhashtable_iter hti;
2609 int link;
2610 };
2611
netlink_walk_start(struct nl_seq_iter * iter)2612 static void netlink_walk_start(struct nl_seq_iter *iter)
2613 {
2614 rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti);
2615 rhashtable_walk_start(&iter->hti);
2616 }
2617
netlink_walk_stop(struct nl_seq_iter * iter)2618 static void netlink_walk_stop(struct nl_seq_iter *iter)
2619 {
2620 rhashtable_walk_stop(&iter->hti);
2621 rhashtable_walk_exit(&iter->hti);
2622 }
2623
__netlink_seq_next(struct seq_file * seq)2624 static void *__netlink_seq_next(struct seq_file *seq)
2625 {
2626 struct nl_seq_iter *iter = seq->private;
2627 struct netlink_sock *nlk;
2628
2629 do {
2630 for (;;) {
2631 nlk = rhashtable_walk_next(&iter->hti);
2632
2633 if (IS_ERR(nlk)) {
2634 if (PTR_ERR(nlk) == -EAGAIN)
2635 continue;
2636
2637 return nlk;
2638 }
2639
2640 if (nlk)
2641 break;
2642
2643 netlink_walk_stop(iter);
2644 if (++iter->link >= MAX_LINKS)
2645 return NULL;
2646
2647 netlink_walk_start(iter);
2648 }
2649 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2650
2651 return nlk;
2652 }
2653
netlink_seq_start(struct seq_file * seq,loff_t * posp)2654 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2655 __acquires(RCU)
2656 {
2657 struct nl_seq_iter *iter = seq->private;
2658 void *obj = SEQ_START_TOKEN;
2659 loff_t pos;
2660
2661 iter->link = 0;
2662
2663 netlink_walk_start(iter);
2664
2665 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2666 obj = __netlink_seq_next(seq);
2667
2668 return obj;
2669 }
2670
netlink_seq_next(struct seq_file * seq,void * v,loff_t * pos)2671 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2672 {
2673 ++*pos;
2674 return __netlink_seq_next(seq);
2675 }
2676
netlink_native_seq_stop(struct seq_file * seq,void * v)2677 static void netlink_native_seq_stop(struct seq_file *seq, void *v)
2678 {
2679 struct nl_seq_iter *iter = seq->private;
2680
2681 if (iter->link >= MAX_LINKS)
2682 return;
2683
2684 netlink_walk_stop(iter);
2685 }
2686
2687
netlink_native_seq_show(struct seq_file * seq,void * v)2688 static int netlink_native_seq_show(struct seq_file *seq, void *v)
2689 {
2690 if (v == SEQ_START_TOKEN) {
2691 seq_puts(seq,
2692 "sk Eth Pid Groups "
2693 "Rmem Wmem Dump Locks Drops Inode\n");
2694 } else {
2695 struct sock *s = v;
2696 struct netlink_sock *nlk = nlk_sk(s);
2697
2698 seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n",
2699 s,
2700 s->sk_protocol,
2701 nlk->portid,
2702 nlk->groups ? (u32)nlk->groups[0] : 0,
2703 sk_rmem_alloc_get(s),
2704 sk_wmem_alloc_get(s),
2705 READ_ONCE(nlk->cb_running),
2706 refcount_read(&s->sk_refcnt),
2707 atomic_read(&s->sk_drops),
2708 sock_i_ino(s)
2709 );
2710
2711 }
2712 return 0;
2713 }
2714
2715 #ifdef CONFIG_BPF_SYSCALL
2716 struct bpf_iter__netlink {
2717 __bpf_md_ptr(struct bpf_iter_meta *, meta);
2718 __bpf_md_ptr(struct netlink_sock *, sk);
2719 };
2720
DEFINE_BPF_ITER_FUNC(netlink,struct bpf_iter_meta * meta,struct netlink_sock * sk)2721 DEFINE_BPF_ITER_FUNC(netlink, struct bpf_iter_meta *meta, struct netlink_sock *sk)
2722
2723 static int netlink_prog_seq_show(struct bpf_prog *prog,
2724 struct bpf_iter_meta *meta,
2725 void *v)
2726 {
2727 struct bpf_iter__netlink ctx;
2728
2729 meta->seq_num--; /* skip SEQ_START_TOKEN */
2730 ctx.meta = meta;
2731 ctx.sk = nlk_sk((struct sock *)v);
2732 return bpf_iter_run_prog(prog, &ctx);
2733 }
2734
netlink_seq_show(struct seq_file * seq,void * v)2735 static int netlink_seq_show(struct seq_file *seq, void *v)
2736 {
2737 struct bpf_iter_meta meta;
2738 struct bpf_prog *prog;
2739
2740 meta.seq = seq;
2741 prog = bpf_iter_get_info(&meta, false);
2742 if (!prog)
2743 return netlink_native_seq_show(seq, v);
2744
2745 if (v != SEQ_START_TOKEN)
2746 return netlink_prog_seq_show(prog, &meta, v);
2747
2748 return 0;
2749 }
2750
netlink_seq_stop(struct seq_file * seq,void * v)2751 static void netlink_seq_stop(struct seq_file *seq, void *v)
2752 {
2753 struct bpf_iter_meta meta;
2754 struct bpf_prog *prog;
2755
2756 if (!v) {
2757 meta.seq = seq;
2758 prog = bpf_iter_get_info(&meta, true);
2759 if (prog)
2760 (void)netlink_prog_seq_show(prog, &meta, v);
2761 }
2762
2763 netlink_native_seq_stop(seq, v);
2764 }
2765 #else
netlink_seq_show(struct seq_file * seq,void * v)2766 static int netlink_seq_show(struct seq_file *seq, void *v)
2767 {
2768 return netlink_native_seq_show(seq, v);
2769 }
2770
netlink_seq_stop(struct seq_file * seq,void * v)2771 static void netlink_seq_stop(struct seq_file *seq, void *v)
2772 {
2773 netlink_native_seq_stop(seq, v);
2774 }
2775 #endif
2776
2777 static const struct seq_operations netlink_seq_ops = {
2778 .start = netlink_seq_start,
2779 .next = netlink_seq_next,
2780 .stop = netlink_seq_stop,
2781 .show = netlink_seq_show,
2782 };
2783 #endif
2784
netlink_register_notifier(struct notifier_block * nb)2785 int netlink_register_notifier(struct notifier_block *nb)
2786 {
2787 return blocking_notifier_chain_register(&netlink_chain, nb);
2788 }
2789 EXPORT_SYMBOL(netlink_register_notifier);
2790
netlink_unregister_notifier(struct notifier_block * nb)2791 int netlink_unregister_notifier(struct notifier_block *nb)
2792 {
2793 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2794 }
2795 EXPORT_SYMBOL(netlink_unregister_notifier);
2796
2797 static const struct proto_ops netlink_ops = {
2798 .family = PF_NETLINK,
2799 .owner = THIS_MODULE,
2800 .release = netlink_release,
2801 .bind = netlink_bind,
2802 .connect = netlink_connect,
2803 .socketpair = sock_no_socketpair,
2804 .accept = sock_no_accept,
2805 .getname = netlink_getname,
2806 .poll = datagram_poll,
2807 .ioctl = netlink_ioctl,
2808 .listen = sock_no_listen,
2809 .shutdown = sock_no_shutdown,
2810 .setsockopt = netlink_setsockopt,
2811 .getsockopt = netlink_getsockopt,
2812 .sendmsg = netlink_sendmsg,
2813 .recvmsg = netlink_recvmsg,
2814 .mmap = sock_no_mmap,
2815 };
2816
2817 static const struct net_proto_family netlink_family_ops = {
2818 .family = PF_NETLINK,
2819 .create = netlink_create,
2820 .owner = THIS_MODULE, /* for consistency 8) */
2821 };
2822
netlink_net_init(struct net * net)2823 static int __net_init netlink_net_init(struct net *net)
2824 {
2825 #ifdef CONFIG_PROC_FS
2826 if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2827 sizeof(struct nl_seq_iter)))
2828 return -ENOMEM;
2829 #endif
2830 return 0;
2831 }
2832
netlink_net_exit(struct net * net)2833 static void __net_exit netlink_net_exit(struct net *net)
2834 {
2835 #ifdef CONFIG_PROC_FS
2836 remove_proc_entry("netlink", net->proc_net);
2837 #endif
2838 }
2839
netlink_add_usersock_entry(void)2840 static void __init netlink_add_usersock_entry(void)
2841 {
2842 struct listeners *listeners;
2843 int groups = 32;
2844
2845 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2846 if (!listeners)
2847 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2848
2849 netlink_table_grab();
2850
2851 nl_table[NETLINK_USERSOCK].groups = groups;
2852 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2853 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2854 nl_table[NETLINK_USERSOCK].registered = 1;
2855 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2856
2857 netlink_table_ungrab();
2858 }
2859
2860 static struct pernet_operations __net_initdata netlink_net_ops = {
2861 .init = netlink_net_init,
2862 .exit = netlink_net_exit,
2863 };
2864
netlink_hash(const void * data,u32 len,u32 seed)2865 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2866 {
2867 const struct netlink_sock *nlk = data;
2868 struct netlink_compare_arg arg;
2869
2870 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2871 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2872 }
2873
2874 static const struct rhashtable_params netlink_rhashtable_params = {
2875 .head_offset = offsetof(struct netlink_sock, node),
2876 .key_len = netlink_compare_arg_len,
2877 .obj_hashfn = netlink_hash,
2878 .obj_cmpfn = netlink_compare,
2879 .automatic_shrinking = true,
2880 };
2881
2882 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2883 BTF_ID_LIST(btf_netlink_sock_id)
2884 BTF_ID(struct, netlink_sock)
2885
2886 static const struct bpf_iter_seq_info netlink_seq_info = {
2887 .seq_ops = &netlink_seq_ops,
2888 .init_seq_private = bpf_iter_init_seq_net,
2889 .fini_seq_private = bpf_iter_fini_seq_net,
2890 .seq_priv_size = sizeof(struct nl_seq_iter),
2891 };
2892
2893 static struct bpf_iter_reg netlink_reg_info = {
2894 .target = "netlink",
2895 .ctx_arg_info_size = 1,
2896 .ctx_arg_info = {
2897 { offsetof(struct bpf_iter__netlink, sk),
2898 PTR_TO_BTF_ID_OR_NULL },
2899 },
2900 .seq_info = &netlink_seq_info,
2901 };
2902
bpf_iter_register(void)2903 static int __init bpf_iter_register(void)
2904 {
2905 netlink_reg_info.ctx_arg_info[0].btf_id = *btf_netlink_sock_id;
2906 return bpf_iter_reg_target(&netlink_reg_info);
2907 }
2908 #endif
2909
netlink_proto_init(void)2910 static int __init netlink_proto_init(void)
2911 {
2912 int i;
2913 int err = proto_register(&netlink_proto, 0);
2914
2915 if (err != 0)
2916 goto out;
2917
2918 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2919 err = bpf_iter_register();
2920 if (err)
2921 goto out;
2922 #endif
2923
2924 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof_field(struct sk_buff, cb));
2925
2926 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2927 if (!nl_table)
2928 goto panic;
2929
2930 for (i = 0; i < MAX_LINKS; i++) {
2931 if (rhashtable_init(&nl_table[i].hash,
2932 &netlink_rhashtable_params) < 0) {
2933 while (--i > 0)
2934 rhashtable_destroy(&nl_table[i].hash);
2935 kfree(nl_table);
2936 goto panic;
2937 }
2938 }
2939
2940 netlink_add_usersock_entry();
2941
2942 sock_register(&netlink_family_ops);
2943 register_pernet_subsys(&netlink_net_ops);
2944 register_pernet_subsys(&netlink_tap_net_ops);
2945 /* The netlink device handler may be needed early. */
2946 rtnetlink_init();
2947 out:
2948 return err;
2949 panic:
2950 panic("netlink_init: Cannot allocate nl_table\n");
2951 }
2952
2953 core_initcall(netlink_proto_init);
2954