Loading...
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
77struct listeners {
78 struct rcu_head rcu;
79 unsigned long masks[];
80};
81
82/* state bits */
83#define NETLINK_S_CONGESTED 0x0
84
85static inline int netlink_is_kernel(struct sock *sk)
86{
87 return nlk_test_bit(KERNEL_SOCKET, sk);
88}
89
90struct netlink_table *nl_table __read_mostly;
91EXPORT_SYMBOL_GPL(nl_table);
92
93static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
94
95static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
96
97static 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
133static 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 */
143DEFINE_RWLOCK(nl_table_lock);
144EXPORT_SYMBOL_GPL(nl_table_lock);
145static 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
149static BLOCKING_NOTIFIER_HEAD(netlink_chain);
150
151
152static const struct rhashtable_params netlink_rhashtable_params;
153
154void do_trace_netlink_extack(const char *msg)
155{
156 trace_netlink_extack(msg);
157}
158EXPORT_SYMBOL(do_trace_netlink_extack);
159
160static inline u32 netlink_group_mask(u32 group)
161{
162 if (group > 32)
163 return 0;
164 return group ? 1 << (group - 1) : 0;
165}
166
167static 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
185static unsigned int netlink_tap_net_id;
186
187struct netlink_tap_net {
188 struct list_head netlink_tap_all;
189 struct mutex netlink_tap_lock;
190};
191
192int 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}
208EXPORT_SYMBOL_GPL(netlink_add_tap);
209
210static 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);
228out:
229 mutex_unlock(&nn->netlink_tap_lock);
230
231 if (found)
232 module_put(nt->module);
233
234 return found ? 0 : -ENODEV;
235}
236
237int 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}
246EXPORT_SYMBOL_GPL(netlink_remove_tap);
247
248static __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
257static 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
263static 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
285static 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
316static 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
331static 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
343static 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
350static 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
362static 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
372static 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
385static 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
394static void netlink_sock_destruct(struct sock *sk)
395{
396 skb_queue_purge(&sk->sk_receive_queue);
397
398 if (!sock_flag(sk, SOCK_DEAD)) {
399 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
400 return;
401 }
402
403 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
404 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
405 WARN_ON(nlk_sk(sk)->groups);
406}
407
408/* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
409 * SMP. Look, when several writers sleep and reader wakes them up, all but one
410 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
411 * this, _but_ remember, it adds useless work on UP machines.
412 */
413
414void netlink_table_grab(void)
415 __acquires(nl_table_lock)
416{
417 might_sleep();
418
419 write_lock_irq(&nl_table_lock);
420
421 if (atomic_read(&nl_table_users)) {
422 DECLARE_WAITQUEUE(wait, current);
423
424 add_wait_queue_exclusive(&nl_table_wait, &wait);
425 for (;;) {
426 set_current_state(TASK_UNINTERRUPTIBLE);
427 if (atomic_read(&nl_table_users) == 0)
428 break;
429 write_unlock_irq(&nl_table_lock);
430 schedule();
431 write_lock_irq(&nl_table_lock);
432 }
433
434 __set_current_state(TASK_RUNNING);
435 remove_wait_queue(&nl_table_wait, &wait);
436 }
437}
438
439void netlink_table_ungrab(void)
440 __releases(nl_table_lock)
441{
442 write_unlock_irq(&nl_table_lock);
443 wake_up(&nl_table_wait);
444}
445
446static inline void
447netlink_lock_table(void)
448{
449 unsigned long flags;
450
451 /* read_lock() synchronizes us to netlink_table_grab */
452
453 read_lock_irqsave(&nl_table_lock, flags);
454 atomic_inc(&nl_table_users);
455 read_unlock_irqrestore(&nl_table_lock, flags);
456}
457
458static inline void
459netlink_unlock_table(void)
460{
461 if (atomic_dec_and_test(&nl_table_users))
462 wake_up(&nl_table_wait);
463}
464
465struct netlink_compare_arg
466{
467 possible_net_t pnet;
468 u32 portid;
469};
470
471/* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
472#define netlink_compare_arg_len \
473 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
474
475static inline int netlink_compare(struct rhashtable_compare_arg *arg,
476 const void *ptr)
477{
478 const struct netlink_compare_arg *x = arg->key;
479 const struct netlink_sock *nlk = ptr;
480
481 return nlk->portid != x->portid ||
482 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
483}
484
485static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
486 struct net *net, u32 portid)
487{
488 memset(arg, 0, sizeof(*arg));
489 write_pnet(&arg->pnet, net);
490 arg->portid = portid;
491}
492
493static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
494 struct net *net)
495{
496 struct netlink_compare_arg arg;
497
498 netlink_compare_arg_init(&arg, net, portid);
499 return rhashtable_lookup_fast(&table->hash, &arg,
500 netlink_rhashtable_params);
501}
502
503static int __netlink_insert(struct netlink_table *table, struct sock *sk)
504{
505 struct netlink_compare_arg arg;
506
507 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
508 return rhashtable_lookup_insert_key(&table->hash, &arg,
509 &nlk_sk(sk)->node,
510 netlink_rhashtable_params);
511}
512
513static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
514{
515 struct netlink_table *table = &nl_table[protocol];
516 struct sock *sk;
517
518 rcu_read_lock();
519 sk = __netlink_lookup(table, portid, net);
520 if (sk)
521 sock_hold(sk);
522 rcu_read_unlock();
523
524 return sk;
525}
526
527static const struct proto_ops netlink_ops;
528
529static void
530netlink_update_listeners(struct sock *sk)
531{
532 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
533 unsigned long mask;
534 unsigned int i;
535 struct listeners *listeners;
536
537 listeners = nl_deref_protected(tbl->listeners);
538 if (!listeners)
539 return;
540
541 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
542 mask = 0;
543 sk_for_each_bound(sk, &tbl->mc_list) {
544 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
545 mask |= nlk_sk(sk)->groups[i];
546 }
547 listeners->masks[i] = mask;
548 }
549 /* this function is only called with the netlink table "grabbed", which
550 * makes sure updates are visible before bind or setsockopt return. */
551}
552
553static int netlink_insert(struct sock *sk, u32 portid)
554{
555 struct netlink_table *table = &nl_table[sk->sk_protocol];
556 int err;
557
558 lock_sock(sk);
559
560 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
561 if (nlk_sk(sk)->bound)
562 goto err;
563
564 /* portid can be read locklessly from netlink_getname(). */
565 WRITE_ONCE(nlk_sk(sk)->portid, portid);
566
567 sock_hold(sk);
568
569 err = __netlink_insert(table, sk);
570 if (err) {
571 /* In case the hashtable backend returns with -EBUSY
572 * from here, it must not escape to the caller.
573 */
574 if (unlikely(err == -EBUSY))
575 err = -EOVERFLOW;
576 if (err == -EEXIST)
577 err = -EADDRINUSE;
578 sock_put(sk);
579 goto err;
580 }
581
582 /* We need to ensure that the socket is hashed and visible. */
583 smp_wmb();
584 /* Paired with lockless reads from netlink_bind(),
585 * netlink_connect() and netlink_sendmsg().
586 */
587 WRITE_ONCE(nlk_sk(sk)->bound, portid);
588
589err:
590 release_sock(sk);
591 return err;
592}
593
594static void netlink_remove(struct sock *sk)
595{
596 struct netlink_table *table;
597
598 table = &nl_table[sk->sk_protocol];
599 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
600 netlink_rhashtable_params)) {
601 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
602 __sock_put(sk);
603 }
604
605 netlink_table_grab();
606 if (nlk_sk(sk)->subscriptions) {
607 __sk_del_bind_node(sk);
608 netlink_update_listeners(sk);
609 }
610 if (sk->sk_protocol == NETLINK_GENERIC)
611 atomic_inc(&genl_sk_destructing_cnt);
612 netlink_table_ungrab();
613}
614
615static struct proto netlink_proto = {
616 .name = "NETLINK",
617 .owner = THIS_MODULE,
618 .obj_size = sizeof(struct netlink_sock),
619};
620
621static int __netlink_create(struct net *net, struct socket *sock,
622 int protocol, int kern)
623{
624 struct sock *sk;
625 struct netlink_sock *nlk;
626
627 sock->ops = &netlink_ops;
628
629 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
630 if (!sk)
631 return -ENOMEM;
632
633 sock_init_data(sock, sk);
634
635 nlk = nlk_sk(sk);
636 mutex_init(&nlk->nl_cb_mutex);
637 lockdep_set_class_and_name(&nlk->nl_cb_mutex,
638 nlk_cb_mutex_keys + protocol,
639 nlk_cb_mutex_key_strings[protocol]);
640 init_waitqueue_head(&nlk->wait);
641
642 sk->sk_destruct = netlink_sock_destruct;
643 sk->sk_protocol = protocol;
644 return 0;
645}
646
647static int netlink_create(struct net *net, struct socket *sock, int protocol,
648 int kern)
649{
650 struct module *module = NULL;
651 struct netlink_sock *nlk;
652 int (*bind)(struct net *net, int group);
653 void (*unbind)(struct net *net, int group);
654 void (*release)(struct sock *sock, unsigned long *groups);
655 int err = 0;
656
657 sock->state = SS_UNCONNECTED;
658
659 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
660 return -ESOCKTNOSUPPORT;
661
662 if (protocol < 0 || protocol >= MAX_LINKS)
663 return -EPROTONOSUPPORT;
664 protocol = array_index_nospec(protocol, MAX_LINKS);
665
666 netlink_lock_table();
667#ifdef CONFIG_MODULES
668 if (!nl_table[protocol].registered) {
669 netlink_unlock_table();
670 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
671 netlink_lock_table();
672 }
673#endif
674 if (nl_table[protocol].registered &&
675 try_module_get(nl_table[protocol].module))
676 module = nl_table[protocol].module;
677 else
678 err = -EPROTONOSUPPORT;
679 bind = nl_table[protocol].bind;
680 unbind = nl_table[protocol].unbind;
681 release = nl_table[protocol].release;
682 netlink_unlock_table();
683
684 if (err < 0)
685 goto out;
686
687 err = __netlink_create(net, sock, protocol, kern);
688 if (err < 0)
689 goto out_module;
690
691 sock_prot_inuse_add(net, &netlink_proto, 1);
692
693 nlk = nlk_sk(sock->sk);
694 nlk->module = module;
695 nlk->netlink_bind = bind;
696 nlk->netlink_unbind = unbind;
697 nlk->netlink_release = release;
698out:
699 return err;
700
701out_module:
702 module_put(module);
703 goto out;
704}
705
706static void deferred_put_nlk_sk(struct rcu_head *head)
707{
708 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
709 struct sock *sk = &nlk->sk;
710
711 kfree(nlk->groups);
712 nlk->groups = NULL;
713
714 if (!refcount_dec_and_test(&sk->sk_refcnt))
715 return;
716
717 sk_free(sk);
718}
719
720static int netlink_release(struct socket *sock)
721{
722 struct sock *sk = sock->sk;
723 struct netlink_sock *nlk;
724
725 if (!sk)
726 return 0;
727
728 netlink_remove(sk);
729 sock_orphan(sk);
730 nlk = nlk_sk(sk);
731
732 /*
733 * OK. Socket is unlinked, any packets that arrive now
734 * will be purged.
735 */
736 if (nlk->netlink_release)
737 nlk->netlink_release(sk, nlk->groups);
738
739 /* must not acquire netlink_table_lock in any way again before unbind
740 * and notifying genetlink is done as otherwise it might deadlock
741 */
742 if (nlk->netlink_unbind) {
743 int i;
744
745 for (i = 0; i < nlk->ngroups; i++)
746 if (test_bit(i, nlk->groups))
747 nlk->netlink_unbind(sock_net(sk), i + 1);
748 }
749 if (sk->sk_protocol == NETLINK_GENERIC &&
750 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
751 wake_up(&genl_sk_destructing_waitq);
752
753 sock->sk = NULL;
754 wake_up_interruptible_all(&nlk->wait);
755
756 skb_queue_purge(&sk->sk_write_queue);
757
758 if (nlk->portid && nlk->bound) {
759 struct netlink_notify n = {
760 .net = sock_net(sk),
761 .protocol = sk->sk_protocol,
762 .portid = nlk->portid,
763 };
764 blocking_notifier_call_chain(&netlink_chain,
765 NETLINK_URELEASE, &n);
766 }
767
768 /* Terminate any outstanding dump */
769 if (nlk->cb_running) {
770 if (nlk->cb.done)
771 nlk->cb.done(&nlk->cb);
772 module_put(nlk->cb.module);
773 kfree_skb(nlk->cb.skb);
774 }
775
776 module_put(nlk->module);
777
778 if (netlink_is_kernel(sk)) {
779 netlink_table_grab();
780 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
781 if (--nl_table[sk->sk_protocol].registered == 0) {
782 struct listeners *old;
783
784 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
785 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
786 kfree_rcu(old, rcu);
787 nl_table[sk->sk_protocol].module = NULL;
788 nl_table[sk->sk_protocol].bind = NULL;
789 nl_table[sk->sk_protocol].unbind = NULL;
790 nl_table[sk->sk_protocol].flags = 0;
791 nl_table[sk->sk_protocol].registered = 0;
792 }
793 netlink_table_ungrab();
794 }
795
796 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
797
798 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
799 return 0;
800}
801
802static int netlink_autobind(struct socket *sock)
803{
804 struct sock *sk = sock->sk;
805 struct net *net = sock_net(sk);
806 struct netlink_table *table = &nl_table[sk->sk_protocol];
807 s32 portid = task_tgid_vnr(current);
808 int err;
809 s32 rover = -4096;
810 bool ok;
811
812retry:
813 cond_resched();
814 rcu_read_lock();
815 ok = !__netlink_lookup(table, portid, net);
816 rcu_read_unlock();
817 if (!ok) {
818 /* Bind collision, search negative portid values. */
819 if (rover == -4096)
820 /* rover will be in range [S32_MIN, -4097] */
821 rover = S32_MIN + get_random_u32_below(-4096 - S32_MIN);
822 else if (rover >= -4096)
823 rover = -4097;
824 portid = rover--;
825 goto retry;
826 }
827
828 err = netlink_insert(sk, portid);
829 if (err == -EADDRINUSE)
830 goto retry;
831
832 /* If 2 threads race to autobind, that is fine. */
833 if (err == -EBUSY)
834 err = 0;
835
836 return err;
837}
838
839/**
840 * __netlink_ns_capable - General netlink message capability test
841 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
842 * @user_ns: The user namespace of the capability to use
843 * @cap: The capability to use
844 *
845 * Test to see if the opener of the socket we received the message
846 * from had when the netlink socket was created and the sender of the
847 * message has the capability @cap in the user namespace @user_ns.
848 */
849bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
850 struct user_namespace *user_ns, int cap)
851{
852 return ((nsp->flags & NETLINK_SKB_DST) ||
853 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
854 ns_capable(user_ns, cap);
855}
856EXPORT_SYMBOL(__netlink_ns_capable);
857
858/**
859 * netlink_ns_capable - General netlink message capability test
860 * @skb: 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 */
868bool netlink_ns_capable(const struct sk_buff *skb,
869 struct user_namespace *user_ns, int cap)
870{
871 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
872}
873EXPORT_SYMBOL(netlink_ns_capable);
874
875/**
876 * netlink_capable - Netlink global message capability test
877 * @skb: socket buffer holding a netlink command from userspace
878 * @cap: The capability to use
879 *
880 * Test to see if the opener of the socket we received the message
881 * from had when the netlink socket was created and the sender of the
882 * message has the capability @cap in all user namespaces.
883 */
884bool netlink_capable(const struct sk_buff *skb, int cap)
885{
886 return netlink_ns_capable(skb, &init_user_ns, cap);
887}
888EXPORT_SYMBOL(netlink_capable);
889
890/**
891 * netlink_net_capable - Netlink network namespace message capability test
892 * @skb: socket buffer holding a netlink command from userspace
893 * @cap: The capability to use
894 *
895 * Test to see if the opener of the socket we received the message
896 * from had when the netlink socket was created and the sender of the
897 * message has the capability @cap over the network namespace of
898 * the socket we received the message from.
899 */
900bool netlink_net_capable(const struct sk_buff *skb, int cap)
901{
902 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
903}
904EXPORT_SYMBOL(netlink_net_capable);
905
906static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
907{
908 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
909 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
910}
911
912static void
913netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
914{
915 struct netlink_sock *nlk = nlk_sk(sk);
916
917 if (nlk->subscriptions && !subscriptions)
918 __sk_del_bind_node(sk);
919 else if (!nlk->subscriptions && subscriptions)
920 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
921 nlk->subscriptions = subscriptions;
922}
923
924static int netlink_realloc_groups(struct sock *sk)
925{
926 struct netlink_sock *nlk = nlk_sk(sk);
927 unsigned int groups;
928 unsigned long *new_groups;
929 int err = 0;
930
931 netlink_table_grab();
932
933 groups = nl_table[sk->sk_protocol].groups;
934 if (!nl_table[sk->sk_protocol].registered) {
935 err = -ENOENT;
936 goto out_unlock;
937 }
938
939 if (nlk->ngroups >= groups)
940 goto out_unlock;
941
942 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
943 if (new_groups == NULL) {
944 err = -ENOMEM;
945 goto out_unlock;
946 }
947 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
948 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
949
950 nlk->groups = new_groups;
951 nlk->ngroups = groups;
952 out_unlock:
953 netlink_table_ungrab();
954 return err;
955}
956
957static void netlink_undo_bind(int group, long unsigned int groups,
958 struct sock *sk)
959{
960 struct netlink_sock *nlk = nlk_sk(sk);
961 int undo;
962
963 if (!nlk->netlink_unbind)
964 return;
965
966 for (undo = 0; undo < group; undo++)
967 if (test_bit(undo, &groups))
968 nlk->netlink_unbind(sock_net(sk), undo + 1);
969}
970
971static int netlink_bind(struct socket *sock, struct sockaddr *addr,
972 int addr_len)
973{
974 struct sock *sk = sock->sk;
975 struct net *net = sock_net(sk);
976 struct netlink_sock *nlk = nlk_sk(sk);
977 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
978 int err = 0;
979 unsigned long groups;
980 bool bound;
981
982 if (addr_len < sizeof(struct sockaddr_nl))
983 return -EINVAL;
984
985 if (nladdr->nl_family != AF_NETLINK)
986 return -EINVAL;
987 groups = nladdr->nl_groups;
988
989 /* Only superuser is allowed to listen multicasts */
990 if (groups) {
991 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
992 return -EPERM;
993 err = netlink_realloc_groups(sk);
994 if (err)
995 return err;
996 }
997
998 if (nlk->ngroups < BITS_PER_LONG)
999 groups &= (1UL << nlk->ngroups) - 1;
1000
1001 /* Paired with WRITE_ONCE() in netlink_insert() */
1002 bound = READ_ONCE(nlk->bound);
1003 if (bound) {
1004 /* Ensure nlk->portid is up-to-date. */
1005 smp_rmb();
1006
1007 if (nladdr->nl_pid != nlk->portid)
1008 return -EINVAL;
1009 }
1010
1011 if (nlk->netlink_bind && groups) {
1012 int group;
1013
1014 /* nl_groups is a u32, so cap the maximum groups we can bind */
1015 for (group = 0; group < BITS_PER_TYPE(u32); group++) {
1016 if (!test_bit(group, &groups))
1017 continue;
1018 err = nlk->netlink_bind(net, group + 1);
1019 if (!err)
1020 continue;
1021 netlink_undo_bind(group, groups, sk);
1022 return err;
1023 }
1024 }
1025
1026 /* No need for barriers here as we return to user-space without
1027 * using any of the bound attributes.
1028 */
1029 netlink_lock_table();
1030 if (!bound) {
1031 err = nladdr->nl_pid ?
1032 netlink_insert(sk, nladdr->nl_pid) :
1033 netlink_autobind(sock);
1034 if (err) {
1035 netlink_undo_bind(BITS_PER_TYPE(u32), groups, sk);
1036 goto unlock;
1037 }
1038 }
1039
1040 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1041 goto unlock;
1042 netlink_unlock_table();
1043
1044 netlink_table_grab();
1045 netlink_update_subscriptions(sk, nlk->subscriptions +
1046 hweight32(groups) -
1047 hweight32(nlk->groups[0]));
1048 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1049 netlink_update_listeners(sk);
1050 netlink_table_ungrab();
1051
1052 return 0;
1053
1054unlock:
1055 netlink_unlock_table();
1056 return err;
1057}
1058
1059static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1060 int alen, int flags)
1061{
1062 int err = 0;
1063 struct sock *sk = sock->sk;
1064 struct netlink_sock *nlk = nlk_sk(sk);
1065 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1066
1067 if (alen < sizeof(addr->sa_family))
1068 return -EINVAL;
1069
1070 if (addr->sa_family == AF_UNSPEC) {
1071 /* paired with READ_ONCE() in netlink_getsockbyportid() */
1072 WRITE_ONCE(sk->sk_state, NETLINK_UNCONNECTED);
1073 /* dst_portid and dst_group can be read locklessly */
1074 WRITE_ONCE(nlk->dst_portid, 0);
1075 WRITE_ONCE(nlk->dst_group, 0);
1076 return 0;
1077 }
1078 if (addr->sa_family != AF_NETLINK)
1079 return -EINVAL;
1080
1081 if (alen < sizeof(struct sockaddr_nl))
1082 return -EINVAL;
1083
1084 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1085 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1086 return -EPERM;
1087
1088 /* No need for barriers here as we return to user-space without
1089 * using any of the bound attributes.
1090 * Paired with WRITE_ONCE() in netlink_insert().
1091 */
1092 if (!READ_ONCE(nlk->bound))
1093 err = netlink_autobind(sock);
1094
1095 if (err == 0) {
1096 /* paired with READ_ONCE() in netlink_getsockbyportid() */
1097 WRITE_ONCE(sk->sk_state, NETLINK_CONNECTED);
1098 /* dst_portid and dst_group can be read locklessly */
1099 WRITE_ONCE(nlk->dst_portid, nladdr->nl_pid);
1100 WRITE_ONCE(nlk->dst_group, ffs(nladdr->nl_groups));
1101 }
1102
1103 return err;
1104}
1105
1106static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1107 int peer)
1108{
1109 struct sock *sk = sock->sk;
1110 struct netlink_sock *nlk = nlk_sk(sk);
1111 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1112
1113 nladdr->nl_family = AF_NETLINK;
1114 nladdr->nl_pad = 0;
1115
1116 if (peer) {
1117 /* Paired with WRITE_ONCE() in netlink_connect() */
1118 nladdr->nl_pid = READ_ONCE(nlk->dst_portid);
1119 nladdr->nl_groups = netlink_group_mask(READ_ONCE(nlk->dst_group));
1120 } else {
1121 /* Paired with WRITE_ONCE() in netlink_insert() */
1122 nladdr->nl_pid = READ_ONCE(nlk->portid);
1123 netlink_lock_table();
1124 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1125 netlink_unlock_table();
1126 }
1127 return sizeof(*nladdr);
1128}
1129
1130static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1131 unsigned long arg)
1132{
1133 /* try to hand this ioctl down to the NIC drivers.
1134 */
1135 return -ENOIOCTLCMD;
1136}
1137
1138static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1139{
1140 struct sock *sock;
1141 struct netlink_sock *nlk;
1142
1143 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1144 if (!sock)
1145 return ERR_PTR(-ECONNREFUSED);
1146
1147 /* Don't bother queuing skb if kernel socket has no input function */
1148 nlk = nlk_sk(sock);
1149 /* dst_portid and sk_state can be changed in netlink_connect() */
1150 if (READ_ONCE(sock->sk_state) == NETLINK_CONNECTED &&
1151 READ_ONCE(nlk->dst_portid) != nlk_sk(ssk)->portid) {
1152 sock_put(sock);
1153 return ERR_PTR(-ECONNREFUSED);
1154 }
1155 return sock;
1156}
1157
1158struct sock *netlink_getsockbyfd(int fd)
1159{
1160 CLASS(fd, f)(fd);
1161 struct inode *inode;
1162 struct sock *sock;
1163
1164 if (fd_empty(f))
1165 return ERR_PTR(-EBADF);
1166
1167 inode = file_inode(fd_file(f));
1168 if (!S_ISSOCK(inode->i_mode))
1169 return ERR_PTR(-ENOTSOCK);
1170
1171 sock = SOCKET_I(inode)->sk;
1172 if (sock->sk_family != AF_NETLINK)
1173 return ERR_PTR(-EINVAL);
1174
1175 sock_hold(sock);
1176 return sock;
1177}
1178
1179struct sk_buff *netlink_alloc_large_skb(unsigned int size, int broadcast)
1180{
1181 size_t head_size = SKB_HEAD_ALIGN(size);
1182 struct sk_buff *skb;
1183 void *data;
1184
1185 if (head_size <= PAGE_SIZE || broadcast)
1186 return alloc_skb(size, GFP_KERNEL);
1187
1188 data = kvmalloc(head_size, GFP_KERNEL);
1189 if (!data)
1190 return NULL;
1191
1192 skb = __build_skb(data, head_size);
1193 if (!skb)
1194 kvfree(data);
1195 else if (is_vmalloc_addr(data))
1196 skb->destructor = netlink_skb_destructor;
1197
1198 return skb;
1199}
1200
1201/*
1202 * Attach a skb to a netlink socket.
1203 * The caller must hold a reference to the destination socket. On error, the
1204 * reference is dropped. The skb is not send to the destination, just all
1205 * all error checks are performed and memory in the queue is reserved.
1206 * Return values:
1207 * < 0: error. skb freed, reference to sock dropped.
1208 * 0: continue
1209 * 1: repeat lookup - reference dropped while waiting for socket memory.
1210 */
1211int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1212 long *timeo, struct sock *ssk)
1213{
1214 struct netlink_sock *nlk;
1215
1216 nlk = nlk_sk(sk);
1217
1218 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1219 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1220 DECLARE_WAITQUEUE(wait, current);
1221 if (!*timeo) {
1222 if (!ssk || netlink_is_kernel(ssk))
1223 netlink_overrun(sk);
1224 sock_put(sk);
1225 kfree_skb(skb);
1226 return -EAGAIN;
1227 }
1228
1229 __set_current_state(TASK_INTERRUPTIBLE);
1230 add_wait_queue(&nlk->wait, &wait);
1231
1232 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1233 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1234 !sock_flag(sk, SOCK_DEAD))
1235 *timeo = schedule_timeout(*timeo);
1236
1237 __set_current_state(TASK_RUNNING);
1238 remove_wait_queue(&nlk->wait, &wait);
1239 sock_put(sk);
1240
1241 if (signal_pending(current)) {
1242 kfree_skb(skb);
1243 return sock_intr_errno(*timeo);
1244 }
1245 return 1;
1246 }
1247 netlink_skb_set_owner_r(skb, sk);
1248 return 0;
1249}
1250
1251static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1252{
1253 int len = skb->len;
1254
1255 netlink_deliver_tap(sock_net(sk), skb);
1256
1257 skb_queue_tail(&sk->sk_receive_queue, skb);
1258 sk->sk_data_ready(sk);
1259 return len;
1260}
1261
1262int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1263{
1264 int len = __netlink_sendskb(sk, skb);
1265
1266 sock_put(sk);
1267 return len;
1268}
1269
1270void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1271{
1272 kfree_skb(skb);
1273 sock_put(sk);
1274}
1275
1276static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1277{
1278 int delta;
1279
1280 WARN_ON(skb->sk != NULL);
1281 delta = skb->end - skb->tail;
1282 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1283 return skb;
1284
1285 if (skb_shared(skb)) {
1286 struct sk_buff *nskb = skb_clone(skb, allocation);
1287 if (!nskb)
1288 return skb;
1289 consume_skb(skb);
1290 skb = nskb;
1291 }
1292
1293 pskb_expand_head(skb, 0, -delta,
1294 (allocation & ~__GFP_DIRECT_RECLAIM) |
1295 __GFP_NOWARN | __GFP_NORETRY);
1296 return skb;
1297}
1298
1299static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1300 struct sock *ssk)
1301{
1302 int ret;
1303 struct netlink_sock *nlk = nlk_sk(sk);
1304
1305 ret = -ECONNREFUSED;
1306 if (nlk->netlink_rcv != NULL) {
1307 ret = skb->len;
1308 netlink_skb_set_owner_r(skb, sk);
1309 NETLINK_CB(skb).sk = ssk;
1310 netlink_deliver_tap_kernel(sk, ssk, skb);
1311 nlk->netlink_rcv(skb);
1312 consume_skb(skb);
1313 } else {
1314 kfree_skb(skb);
1315 }
1316 sock_put(sk);
1317 return ret;
1318}
1319
1320int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1321 u32 portid, int nonblock)
1322{
1323 struct sock *sk;
1324 int err;
1325 long timeo;
1326
1327 skb = netlink_trim(skb, gfp_any());
1328
1329 timeo = sock_sndtimeo(ssk, nonblock);
1330retry:
1331 sk = netlink_getsockbyportid(ssk, portid);
1332 if (IS_ERR(sk)) {
1333 kfree_skb(skb);
1334 return PTR_ERR(sk);
1335 }
1336 if (netlink_is_kernel(sk))
1337 return netlink_unicast_kernel(sk, skb, ssk);
1338
1339 if (sk_filter(sk, skb)) {
1340 err = skb->len;
1341 kfree_skb(skb);
1342 sock_put(sk);
1343 return err;
1344 }
1345
1346 err = netlink_attachskb(sk, skb, &timeo, ssk);
1347 if (err == 1)
1348 goto retry;
1349 if (err)
1350 return err;
1351
1352 return netlink_sendskb(sk, skb);
1353}
1354EXPORT_SYMBOL(netlink_unicast);
1355
1356int netlink_has_listeners(struct sock *sk, unsigned int group)
1357{
1358 int res = 0;
1359 struct listeners *listeners;
1360
1361 BUG_ON(!netlink_is_kernel(sk));
1362
1363 rcu_read_lock();
1364 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1365
1366 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1367 res = test_bit(group - 1, listeners->masks);
1368
1369 rcu_read_unlock();
1370
1371 return res;
1372}
1373EXPORT_SYMBOL_GPL(netlink_has_listeners);
1374
1375bool netlink_strict_get_check(struct sk_buff *skb)
1376{
1377 return nlk_test_bit(STRICT_CHK, NETLINK_CB(skb).sk);
1378}
1379EXPORT_SYMBOL_GPL(netlink_strict_get_check);
1380
1381static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1382{
1383 struct netlink_sock *nlk = nlk_sk(sk);
1384
1385 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1386 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1387 netlink_skb_set_owner_r(skb, sk);
1388 __netlink_sendskb(sk, skb);
1389 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1390 }
1391 return -1;
1392}
1393
1394struct netlink_broadcast_data {
1395 struct sock *exclude_sk;
1396 struct net *net;
1397 u32 portid;
1398 u32 group;
1399 int failure;
1400 int delivery_failure;
1401 int congested;
1402 int delivered;
1403 gfp_t allocation;
1404 struct sk_buff *skb, *skb2;
1405 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1406 void *tx_data;
1407};
1408
1409static void do_one_broadcast(struct sock *sk,
1410 struct netlink_broadcast_data *p)
1411{
1412 struct netlink_sock *nlk = nlk_sk(sk);
1413 int val;
1414
1415 if (p->exclude_sk == sk)
1416 return;
1417
1418 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1419 !test_bit(p->group - 1, nlk->groups))
1420 return;
1421
1422 if (!net_eq(sock_net(sk), p->net)) {
1423 if (!nlk_test_bit(LISTEN_ALL_NSID, sk))
1424 return;
1425
1426 if (!peernet_has_id(sock_net(sk), p->net))
1427 return;
1428
1429 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1430 CAP_NET_BROADCAST))
1431 return;
1432 }
1433
1434 if (p->failure) {
1435 netlink_overrun(sk);
1436 return;
1437 }
1438
1439 sock_hold(sk);
1440 if (p->skb2 == NULL) {
1441 if (skb_shared(p->skb)) {
1442 p->skb2 = skb_clone(p->skb, p->allocation);
1443 } else {
1444 p->skb2 = skb_get(p->skb);
1445 /*
1446 * skb ownership may have been set when
1447 * delivered to a previous socket.
1448 */
1449 skb_orphan(p->skb2);
1450 }
1451 }
1452 if (p->skb2 == NULL) {
1453 netlink_overrun(sk);
1454 /* Clone failed. Notify ALL listeners. */
1455 p->failure = 1;
1456 if (nlk_test_bit(BROADCAST_SEND_ERROR, sk))
1457 p->delivery_failure = 1;
1458 goto out;
1459 }
1460
1461 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1462 kfree_skb(p->skb2);
1463 p->skb2 = NULL;
1464 goto out;
1465 }
1466
1467 if (sk_filter(sk, p->skb2)) {
1468 kfree_skb(p->skb2);
1469 p->skb2 = NULL;
1470 goto out;
1471 }
1472 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1473 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1474 NETLINK_CB(p->skb2).nsid_is_set = true;
1475 val = netlink_broadcast_deliver(sk, p->skb2);
1476 if (val < 0) {
1477 netlink_overrun(sk);
1478 if (nlk_test_bit(BROADCAST_SEND_ERROR, sk))
1479 p->delivery_failure = 1;
1480 } else {
1481 p->congested |= val;
1482 p->delivered = 1;
1483 p->skb2 = NULL;
1484 }
1485out:
1486 sock_put(sk);
1487}
1488
1489int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb,
1490 u32 portid,
1491 u32 group, gfp_t allocation,
1492 netlink_filter_fn filter,
1493 void *filter_data)
1494{
1495 struct net *net = sock_net(ssk);
1496 struct netlink_broadcast_data info;
1497 struct sock *sk;
1498
1499 skb = netlink_trim(skb, allocation);
1500
1501 info.exclude_sk = ssk;
1502 info.net = net;
1503 info.portid = portid;
1504 info.group = group;
1505 info.failure = 0;
1506 info.delivery_failure = 0;
1507 info.congested = 0;
1508 info.delivered = 0;
1509 info.allocation = allocation;
1510 info.skb = skb;
1511 info.skb2 = NULL;
1512 info.tx_filter = filter;
1513 info.tx_data = filter_data;
1514
1515 /* While we sleep in clone, do not allow to change socket list */
1516
1517 netlink_lock_table();
1518
1519 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1520 do_one_broadcast(sk, &info);
1521
1522 consume_skb(skb);
1523
1524 netlink_unlock_table();
1525
1526 if (info.delivery_failure) {
1527 kfree_skb(info.skb2);
1528 return -ENOBUFS;
1529 }
1530 consume_skb(info.skb2);
1531
1532 if (info.delivered) {
1533 if (info.congested && gfpflags_allow_blocking(allocation))
1534 yield();
1535 return 0;
1536 }
1537 return -ESRCH;
1538}
1539EXPORT_SYMBOL(netlink_broadcast_filtered);
1540
1541int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1542 u32 group, gfp_t allocation)
1543{
1544 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1545 NULL, NULL);
1546}
1547EXPORT_SYMBOL(netlink_broadcast);
1548
1549struct netlink_set_err_data {
1550 struct sock *exclude_sk;
1551 u32 portid;
1552 u32 group;
1553 int code;
1554};
1555
1556static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1557{
1558 struct netlink_sock *nlk = nlk_sk(sk);
1559 int ret = 0;
1560
1561 if (sk == p->exclude_sk)
1562 goto out;
1563
1564 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1565 goto out;
1566
1567 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1568 !test_bit(p->group - 1, nlk->groups))
1569 goto out;
1570
1571 if (p->code == ENOBUFS && nlk_test_bit(RECV_NO_ENOBUFS, sk)) {
1572 ret = 1;
1573 goto out;
1574 }
1575
1576 WRITE_ONCE(sk->sk_err, p->code);
1577 sk_error_report(sk);
1578out:
1579 return ret;
1580}
1581
1582/**
1583 * netlink_set_err - report error to broadcast listeners
1584 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1585 * @portid: the PORTID of a process that we want to skip (if any)
1586 * @group: the broadcast group that will notice the error
1587 * @code: error code, must be negative (as usual in kernelspace)
1588 *
1589 * This function returns the number of broadcast listeners that have set the
1590 * NETLINK_NO_ENOBUFS socket option.
1591 */
1592int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1593{
1594 struct netlink_set_err_data info;
1595 unsigned long flags;
1596 struct sock *sk;
1597 int ret = 0;
1598
1599 info.exclude_sk = ssk;
1600 info.portid = portid;
1601 info.group = group;
1602 /* sk->sk_err wants a positive error value */
1603 info.code = -code;
1604
1605 read_lock_irqsave(&nl_table_lock, flags);
1606
1607 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1608 ret += do_one_set_err(sk, &info);
1609
1610 read_unlock_irqrestore(&nl_table_lock, flags);
1611 return ret;
1612}
1613EXPORT_SYMBOL(netlink_set_err);
1614
1615/* must be called with netlink table grabbed */
1616static void netlink_update_socket_mc(struct netlink_sock *nlk,
1617 unsigned int group,
1618 int is_new)
1619{
1620 int old, new = !!is_new, subscriptions;
1621
1622 old = test_bit(group - 1, nlk->groups);
1623 subscriptions = nlk->subscriptions - old + new;
1624 __assign_bit(group - 1, nlk->groups, new);
1625 netlink_update_subscriptions(&nlk->sk, subscriptions);
1626 netlink_update_listeners(&nlk->sk);
1627}
1628
1629static int netlink_setsockopt(struct socket *sock, int level, int optname,
1630 sockptr_t optval, unsigned int optlen)
1631{
1632 struct sock *sk = sock->sk;
1633 struct netlink_sock *nlk = nlk_sk(sk);
1634 unsigned int val = 0;
1635 int nr = -1;
1636
1637 if (level != SOL_NETLINK)
1638 return -ENOPROTOOPT;
1639
1640 if (optlen >= sizeof(int) &&
1641 copy_from_sockptr(&val, optval, sizeof(val)))
1642 return -EFAULT;
1643
1644 switch (optname) {
1645 case NETLINK_PKTINFO:
1646 nr = NETLINK_F_RECV_PKTINFO;
1647 break;
1648 case NETLINK_ADD_MEMBERSHIP:
1649 case NETLINK_DROP_MEMBERSHIP: {
1650 int err;
1651
1652 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1653 return -EPERM;
1654 err = netlink_realloc_groups(sk);
1655 if (err)
1656 return err;
1657 if (!val || val - 1 >= nlk->ngroups)
1658 return -EINVAL;
1659 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1660 err = nlk->netlink_bind(sock_net(sk), val);
1661 if (err)
1662 return err;
1663 }
1664 netlink_table_grab();
1665 netlink_update_socket_mc(nlk, val,
1666 optname == NETLINK_ADD_MEMBERSHIP);
1667 netlink_table_ungrab();
1668 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1669 nlk->netlink_unbind(sock_net(sk), val);
1670
1671 break;
1672 }
1673 case NETLINK_BROADCAST_ERROR:
1674 nr = NETLINK_F_BROADCAST_SEND_ERROR;
1675 break;
1676 case NETLINK_NO_ENOBUFS:
1677 assign_bit(NETLINK_F_RECV_NO_ENOBUFS, &nlk->flags, val);
1678 if (val) {
1679 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1680 wake_up_interruptible(&nlk->wait);
1681 }
1682 break;
1683 case NETLINK_LISTEN_ALL_NSID:
1684 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1685 return -EPERM;
1686 nr = NETLINK_F_LISTEN_ALL_NSID;
1687 break;
1688 case NETLINK_CAP_ACK:
1689 nr = NETLINK_F_CAP_ACK;
1690 break;
1691 case NETLINK_EXT_ACK:
1692 nr = NETLINK_F_EXT_ACK;
1693 break;
1694 case NETLINK_GET_STRICT_CHK:
1695 nr = NETLINK_F_STRICT_CHK;
1696 break;
1697 default:
1698 return -ENOPROTOOPT;
1699 }
1700 if (nr >= 0)
1701 assign_bit(nr, &nlk->flags, val);
1702 return 0;
1703}
1704
1705static int netlink_getsockopt(struct socket *sock, int level, int optname,
1706 char __user *optval, int __user *optlen)
1707{
1708 struct sock *sk = sock->sk;
1709 struct netlink_sock *nlk = nlk_sk(sk);
1710 unsigned int flag;
1711 int len, val;
1712
1713 if (level != SOL_NETLINK)
1714 return -ENOPROTOOPT;
1715
1716 if (get_user(len, optlen))
1717 return -EFAULT;
1718 if (len < 0)
1719 return -EINVAL;
1720
1721 switch (optname) {
1722 case NETLINK_PKTINFO:
1723 flag = NETLINK_F_RECV_PKTINFO;
1724 break;
1725 case NETLINK_BROADCAST_ERROR:
1726 flag = NETLINK_F_BROADCAST_SEND_ERROR;
1727 break;
1728 case NETLINK_NO_ENOBUFS:
1729 flag = NETLINK_F_RECV_NO_ENOBUFS;
1730 break;
1731 case NETLINK_LIST_MEMBERSHIPS: {
1732 int pos, idx, shift, err = 0;
1733
1734 netlink_lock_table();
1735 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1736 if (len - pos < sizeof(u32))
1737 break;
1738
1739 idx = pos / sizeof(unsigned long);
1740 shift = (pos % sizeof(unsigned long)) * 8;
1741 if (put_user((u32)(nlk->groups[idx] >> shift),
1742 (u32 __user *)(optval + pos))) {
1743 err = -EFAULT;
1744 break;
1745 }
1746 }
1747 if (put_user(ALIGN(BITS_TO_BYTES(nlk->ngroups), sizeof(u32)), optlen))
1748 err = -EFAULT;
1749 netlink_unlock_table();
1750 return err;
1751 }
1752 case NETLINK_LISTEN_ALL_NSID:
1753 flag = NETLINK_F_LISTEN_ALL_NSID;
1754 break;
1755 case NETLINK_CAP_ACK:
1756 flag = NETLINK_F_CAP_ACK;
1757 break;
1758 case NETLINK_EXT_ACK:
1759 flag = NETLINK_F_EXT_ACK;
1760 break;
1761 case NETLINK_GET_STRICT_CHK:
1762 flag = NETLINK_F_STRICT_CHK;
1763 break;
1764 default:
1765 return -ENOPROTOOPT;
1766 }
1767
1768 if (len < sizeof(int))
1769 return -EINVAL;
1770
1771 len = sizeof(int);
1772 val = test_bit(flag, &nlk->flags);
1773
1774 if (put_user(len, optlen) ||
1775 copy_to_user(optval, &val, len))
1776 return -EFAULT;
1777
1778 return 0;
1779}
1780
1781static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1782{
1783 struct nl_pktinfo info;
1784
1785 info.group = NETLINK_CB(skb).dst_group;
1786 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1787}
1788
1789static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1790 struct sk_buff *skb)
1791{
1792 if (!NETLINK_CB(skb).nsid_is_set)
1793 return;
1794
1795 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1796 &NETLINK_CB(skb).nsid);
1797}
1798
1799static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1800{
1801 struct sock *sk = sock->sk;
1802 struct netlink_sock *nlk = nlk_sk(sk);
1803 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1804 u32 dst_portid;
1805 u32 dst_group;
1806 struct sk_buff *skb;
1807 int err;
1808 struct scm_cookie scm;
1809 u32 netlink_skb_flags = 0;
1810
1811 if (msg->msg_flags & MSG_OOB)
1812 return -EOPNOTSUPP;
1813
1814 if (len == 0) {
1815 pr_warn_once("Zero length message leads to an empty skb\n");
1816 return -ENODATA;
1817 }
1818
1819 err = scm_send(sock, msg, &scm, true);
1820 if (err < 0)
1821 return err;
1822
1823 if (msg->msg_namelen) {
1824 err = -EINVAL;
1825 if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1826 goto out;
1827 if (addr->nl_family != AF_NETLINK)
1828 goto out;
1829 dst_portid = addr->nl_pid;
1830 dst_group = ffs(addr->nl_groups);
1831 err = -EPERM;
1832 if ((dst_group || dst_portid) &&
1833 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1834 goto out;
1835 netlink_skb_flags |= NETLINK_SKB_DST;
1836 } else {
1837 /* Paired with WRITE_ONCE() in netlink_connect() */
1838 dst_portid = READ_ONCE(nlk->dst_portid);
1839 dst_group = READ_ONCE(nlk->dst_group);
1840 }
1841
1842 /* Paired with WRITE_ONCE() in netlink_insert() */
1843 if (!READ_ONCE(nlk->bound)) {
1844 err = netlink_autobind(sock);
1845 if (err)
1846 goto out;
1847 } else {
1848 /* Ensure nlk is hashed and visible. */
1849 smp_rmb();
1850 }
1851
1852 err = -EMSGSIZE;
1853 if (len > sk->sk_sndbuf - 32)
1854 goto out;
1855 err = -ENOBUFS;
1856 skb = netlink_alloc_large_skb(len, dst_group);
1857 if (skb == NULL)
1858 goto out;
1859
1860 NETLINK_CB(skb).portid = nlk->portid;
1861 NETLINK_CB(skb).dst_group = dst_group;
1862 NETLINK_CB(skb).creds = scm.creds;
1863 NETLINK_CB(skb).flags = netlink_skb_flags;
1864
1865 err = -EFAULT;
1866 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1867 kfree_skb(skb);
1868 goto out;
1869 }
1870
1871 err = security_netlink_send(sk, skb);
1872 if (err) {
1873 kfree_skb(skb);
1874 goto out;
1875 }
1876
1877 if (dst_group) {
1878 refcount_inc(&skb->users);
1879 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1880 }
1881 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags & MSG_DONTWAIT);
1882
1883out:
1884 scm_destroy(&scm);
1885 return err;
1886}
1887
1888static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1889 int flags)
1890{
1891 struct scm_cookie scm;
1892 struct sock *sk = sock->sk;
1893 struct netlink_sock *nlk = nlk_sk(sk);
1894 size_t copied, max_recvmsg_len;
1895 struct sk_buff *skb, *data_skb;
1896 int err, ret;
1897
1898 if (flags & MSG_OOB)
1899 return -EOPNOTSUPP;
1900
1901 copied = 0;
1902
1903 skb = skb_recv_datagram(sk, flags, &err);
1904 if (skb == NULL)
1905 goto out;
1906
1907 data_skb = skb;
1908
1909#ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1910 if (unlikely(skb_shinfo(skb)->frag_list)) {
1911 /*
1912 * If this skb has a frag_list, then here that means that we
1913 * will have to use the frag_list skb's data for compat tasks
1914 * and the regular skb's data for normal (non-compat) tasks.
1915 *
1916 * If we need to send the compat skb, assign it to the
1917 * 'data_skb' variable so that it will be used below for data
1918 * copying. We keep 'skb' for everything else, including
1919 * freeing both later.
1920 */
1921 if (flags & MSG_CMSG_COMPAT)
1922 data_skb = skb_shinfo(skb)->frag_list;
1923 }
1924#endif
1925
1926 /* Record the max length of recvmsg() calls for future allocations */
1927 max_recvmsg_len = max(READ_ONCE(nlk->max_recvmsg_len), len);
1928 max_recvmsg_len = min_t(size_t, max_recvmsg_len,
1929 SKB_WITH_OVERHEAD(32768));
1930 WRITE_ONCE(nlk->max_recvmsg_len, max_recvmsg_len);
1931
1932 copied = data_skb->len;
1933 if (len < copied) {
1934 msg->msg_flags |= MSG_TRUNC;
1935 copied = len;
1936 }
1937
1938 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1939
1940 if (msg->msg_name) {
1941 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1942 addr->nl_family = AF_NETLINK;
1943 addr->nl_pad = 0;
1944 addr->nl_pid = NETLINK_CB(skb).portid;
1945 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1946 msg->msg_namelen = sizeof(*addr);
1947 }
1948
1949 if (nlk_test_bit(RECV_PKTINFO, sk))
1950 netlink_cmsg_recv_pktinfo(msg, skb);
1951 if (nlk_test_bit(LISTEN_ALL_NSID, sk))
1952 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1953
1954 memset(&scm, 0, sizeof(scm));
1955 scm.creds = *NETLINK_CREDS(skb);
1956 if (flags & MSG_TRUNC)
1957 copied = data_skb->len;
1958
1959 skb_free_datagram(sk, skb);
1960
1961 if (READ_ONCE(nlk->cb_running) &&
1962 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1963 ret = netlink_dump(sk, false);
1964 if (ret) {
1965 WRITE_ONCE(sk->sk_err, -ret);
1966 sk_error_report(sk);
1967 }
1968 }
1969
1970 scm_recv(sock, msg, &scm, flags);
1971out:
1972 netlink_rcv_wake(sk);
1973 return err ? : copied;
1974}
1975
1976static void netlink_data_ready(struct sock *sk)
1977{
1978 BUG();
1979}
1980
1981/*
1982 * We export these functions to other modules. They provide a
1983 * complete set of kernel non-blocking support for message
1984 * queueing.
1985 */
1986
1987struct sock *
1988__netlink_kernel_create(struct net *net, int unit, struct module *module,
1989 struct netlink_kernel_cfg *cfg)
1990{
1991 struct socket *sock;
1992 struct sock *sk;
1993 struct netlink_sock *nlk;
1994 struct listeners *listeners = NULL;
1995 unsigned int groups;
1996
1997 BUG_ON(!nl_table);
1998
1999 if (unit < 0 || unit >= MAX_LINKS)
2000 return NULL;
2001
2002 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2003 return NULL;
2004
2005 if (__netlink_create(net, sock, unit, 1) < 0)
2006 goto out_sock_release_nosk;
2007
2008 sk = sock->sk;
2009
2010 if (!cfg || cfg->groups < 32)
2011 groups = 32;
2012 else
2013 groups = cfg->groups;
2014
2015 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2016 if (!listeners)
2017 goto out_sock_release;
2018
2019 sk->sk_data_ready = netlink_data_ready;
2020 if (cfg && cfg->input)
2021 nlk_sk(sk)->netlink_rcv = cfg->input;
2022
2023 if (netlink_insert(sk, 0))
2024 goto out_sock_release;
2025
2026 nlk = nlk_sk(sk);
2027 set_bit(NETLINK_F_KERNEL_SOCKET, &nlk->flags);
2028
2029 netlink_table_grab();
2030 if (!nl_table[unit].registered) {
2031 nl_table[unit].groups = groups;
2032 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2033 nl_table[unit].module = module;
2034 if (cfg) {
2035 nl_table[unit].bind = cfg->bind;
2036 nl_table[unit].unbind = cfg->unbind;
2037 nl_table[unit].release = cfg->release;
2038 nl_table[unit].flags = cfg->flags;
2039 }
2040 nl_table[unit].registered = 1;
2041 } else {
2042 kfree(listeners);
2043 nl_table[unit].registered++;
2044 }
2045 netlink_table_ungrab();
2046 return sk;
2047
2048out_sock_release:
2049 kfree(listeners);
2050 netlink_kernel_release(sk);
2051 return NULL;
2052
2053out_sock_release_nosk:
2054 sock_release(sock);
2055 return NULL;
2056}
2057EXPORT_SYMBOL(__netlink_kernel_create);
2058
2059void
2060netlink_kernel_release(struct sock *sk)
2061{
2062 if (sk == NULL || sk->sk_socket == NULL)
2063 return;
2064
2065 sock_release(sk->sk_socket);
2066}
2067EXPORT_SYMBOL(netlink_kernel_release);
2068
2069int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2070{
2071 struct listeners *new, *old;
2072 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2073
2074 if (groups < 32)
2075 groups = 32;
2076
2077 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2078 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2079 if (!new)
2080 return -ENOMEM;
2081 old = nl_deref_protected(tbl->listeners);
2082 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2083 rcu_assign_pointer(tbl->listeners, new);
2084
2085 kfree_rcu(old, rcu);
2086 }
2087 tbl->groups = groups;
2088
2089 return 0;
2090}
2091
2092/**
2093 * netlink_change_ngroups - change number of multicast groups
2094 *
2095 * This changes the number of multicast groups that are available
2096 * on a certain netlink family. Note that it is not possible to
2097 * change the number of groups to below 32. Also note that it does
2098 * not implicitly call netlink_clear_multicast_users() when the
2099 * number of groups is reduced.
2100 *
2101 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2102 * @groups: The new number of groups.
2103 */
2104int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2105{
2106 int err;
2107
2108 netlink_table_grab();
2109 err = __netlink_change_ngroups(sk, groups);
2110 netlink_table_ungrab();
2111
2112 return err;
2113}
2114
2115void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2116{
2117 struct sock *sk;
2118 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2119 struct hlist_node *tmp;
2120
2121 sk_for_each_bound_safe(sk, tmp, &tbl->mc_list)
2122 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2123}
2124
2125struct nlmsghdr *
2126__nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2127{
2128 struct nlmsghdr *nlh;
2129 int size = nlmsg_msg_size(len);
2130
2131 nlh = skb_put(skb, NLMSG_ALIGN(size));
2132 nlh->nlmsg_type = type;
2133 nlh->nlmsg_len = size;
2134 nlh->nlmsg_flags = flags;
2135 nlh->nlmsg_pid = portid;
2136 nlh->nlmsg_seq = seq;
2137 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2138 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2139 return nlh;
2140}
2141EXPORT_SYMBOL(__nlmsg_put);
2142
2143static size_t
2144netlink_ack_tlv_len(struct netlink_sock *nlk, int err,
2145 const struct netlink_ext_ack *extack)
2146{
2147 size_t tlvlen;
2148
2149 if (!extack || !test_bit(NETLINK_F_EXT_ACK, &nlk->flags))
2150 return 0;
2151
2152 tlvlen = 0;
2153 if (extack->_msg)
2154 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2155 if (extack->cookie_len)
2156 tlvlen += nla_total_size(extack->cookie_len);
2157
2158 /* Following attributes are only reported as error (not warning) */
2159 if (!err)
2160 return tlvlen;
2161
2162 if (extack->bad_attr)
2163 tlvlen += nla_total_size(sizeof(u32));
2164 if (extack->policy)
2165 tlvlen += netlink_policy_dump_attr_size_estimate(extack->policy);
2166 if (extack->miss_type)
2167 tlvlen += nla_total_size(sizeof(u32));
2168 if (extack->miss_nest)
2169 tlvlen += nla_total_size(sizeof(u32));
2170
2171 return tlvlen;
2172}
2173
2174static bool nlmsg_check_in_payload(const struct nlmsghdr *nlh, const void *addr)
2175{
2176 return !WARN_ON(addr < nlmsg_data(nlh) ||
2177 addr - (const void *) nlh >= nlh->nlmsg_len);
2178}
2179
2180static void
2181netlink_ack_tlv_fill(struct sk_buff *skb, const struct nlmsghdr *nlh, int err,
2182 const struct netlink_ext_ack *extack)
2183{
2184 if (extack->_msg)
2185 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg));
2186 if (extack->cookie_len)
2187 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2188 extack->cookie_len, extack->cookie));
2189
2190 if (!err)
2191 return;
2192
2193 if (extack->bad_attr && nlmsg_check_in_payload(nlh, extack->bad_attr))
2194 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2195 (u8 *)extack->bad_attr - (const u8 *)nlh));
2196 if (extack->policy)
2197 netlink_policy_dump_write_attr(skb, extack->policy,
2198 NLMSGERR_ATTR_POLICY);
2199 if (extack->miss_type)
2200 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_TYPE,
2201 extack->miss_type));
2202 if (extack->miss_nest && nlmsg_check_in_payload(nlh, extack->miss_nest))
2203 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_NEST,
2204 (u8 *)extack->miss_nest - (const u8 *)nlh));
2205}
2206
2207/*
2208 * It looks a bit ugly.
2209 * It would be better to create kernel thread.
2210 */
2211
2212static 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 size_t extack_len;
2218
2219 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(nlk->dump_done_errno),
2220 NLM_F_MULTI | cb->answer_flags);
2221 if (WARN_ON(!nlh))
2222 return -ENOBUFS;
2223
2224 nl_dump_check_consistent(cb, nlh);
2225 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno, sizeof(nlk->dump_done_errno));
2226
2227 extack_len = netlink_ack_tlv_len(nlk, nlk->dump_done_errno, extack);
2228 if (extack_len) {
2229 nlh->nlmsg_flags |= NLM_F_ACK_TLVS;
2230 if (skb_tailroom(skb) >= extack_len) {
2231 netlink_ack_tlv_fill(skb, cb->nlh,
2232 nlk->dump_done_errno, extack);
2233 nlmsg_end(skb, nlh);
2234 }
2235 }
2236
2237 return 0;
2238}
2239
2240static int netlink_dump(struct sock *sk, bool lock_taken)
2241{
2242 struct netlink_sock *nlk = nlk_sk(sk);
2243 struct netlink_ext_ack extack = {};
2244 struct netlink_callback *cb;
2245 struct sk_buff *skb = NULL;
2246 size_t max_recvmsg_len;
2247 struct module *module;
2248 int err = -ENOBUFS;
2249 int alloc_min_size;
2250 int alloc_size;
2251
2252 if (!lock_taken)
2253 mutex_lock(&nlk->nl_cb_mutex);
2254 if (!nlk->cb_running) {
2255 err = -EINVAL;
2256 goto errout_skb;
2257 }
2258
2259 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2260 goto errout_skb;
2261
2262 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2263 * required, but it makes sense to _attempt_ a 32KiB allocation
2264 * to reduce number of system calls on dump operations, if user
2265 * ever provided a big enough buffer.
2266 */
2267 cb = &nlk->cb;
2268 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2269
2270 max_recvmsg_len = READ_ONCE(nlk->max_recvmsg_len);
2271 if (alloc_min_size < max_recvmsg_len) {
2272 alloc_size = max_recvmsg_len;
2273 skb = alloc_skb(alloc_size,
2274 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2275 __GFP_NOWARN | __GFP_NORETRY);
2276 }
2277 if (!skb) {
2278 alloc_size = alloc_min_size;
2279 skb = alloc_skb(alloc_size, GFP_KERNEL);
2280 }
2281 if (!skb)
2282 goto errout_skb;
2283
2284 /* Trim skb to allocated size. User is expected to provide buffer as
2285 * large as max(min_dump_alloc, 32KiB (max_recvmsg_len capped at
2286 * netlink_recvmsg())). dump will pack as many smaller messages as
2287 * could fit within the allocated skb. skb is typically allocated
2288 * with larger space than required (could be as much as near 2x the
2289 * requested size with align to next power of 2 approach). Allowing
2290 * dump to use the excess space makes it difficult for a user to have a
2291 * reasonable static buffer based on the expected largest dump of a
2292 * single netdev. The outcome is MSG_TRUNC error.
2293 */
2294 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2295
2296 /* Make sure malicious BPF programs can not read unitialized memory
2297 * from skb->head -> skb->data
2298 */
2299 skb_reset_network_header(skb);
2300 skb_reset_mac_header(skb);
2301
2302 netlink_skb_set_owner_r(skb, sk);
2303
2304 if (nlk->dump_done_errno > 0) {
2305 cb->extack = &extack;
2306
2307 nlk->dump_done_errno = cb->dump(skb, cb);
2308
2309 /* EMSGSIZE plus something already in the skb means
2310 * that there's more to dump but current skb has filled up.
2311 * If the callback really wants to return EMSGSIZE to user space
2312 * it needs to do so again, on the next cb->dump() call,
2313 * without putting data in the skb.
2314 */
2315 if (nlk->dump_done_errno == -EMSGSIZE && skb->len)
2316 nlk->dump_done_errno = skb->len;
2317
2318 cb->extack = NULL;
2319 }
2320
2321 if (nlk->dump_done_errno > 0 ||
2322 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2323 mutex_unlock(&nlk->nl_cb_mutex);
2324
2325 if (sk_filter(sk, skb))
2326 kfree_skb(skb);
2327 else
2328 __netlink_sendskb(sk, skb);
2329 return 0;
2330 }
2331
2332 if (netlink_dump_done(nlk, skb, cb, &extack))
2333 goto errout_skb;
2334
2335#ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2336 /* frag_list skb's data is used for compat tasks
2337 * and the regular skb's data for normal (non-compat) tasks.
2338 * See netlink_recvmsg().
2339 */
2340 if (unlikely(skb_shinfo(skb)->frag_list)) {
2341 if (netlink_dump_done(nlk, skb_shinfo(skb)->frag_list, cb, &extack))
2342 goto errout_skb;
2343 }
2344#endif
2345
2346 if (sk_filter(sk, skb))
2347 kfree_skb(skb);
2348 else
2349 __netlink_sendskb(sk, skb);
2350
2351 if (cb->done)
2352 cb->done(cb);
2353
2354 WRITE_ONCE(nlk->cb_running, false);
2355 module = cb->module;
2356 skb = cb->skb;
2357 mutex_unlock(&nlk->nl_cb_mutex);
2358 module_put(module);
2359 consume_skb(skb);
2360 return 0;
2361
2362errout_skb:
2363 mutex_unlock(&nlk->nl_cb_mutex);
2364 kfree_skb(skb);
2365 return err;
2366}
2367
2368int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2369 const struct nlmsghdr *nlh,
2370 struct netlink_dump_control *control)
2371{
2372 struct netlink_callback *cb;
2373 struct netlink_sock *nlk;
2374 struct sock *sk;
2375 int ret;
2376
2377 refcount_inc(&skb->users);
2378
2379 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2380 if (sk == NULL) {
2381 ret = -ECONNREFUSED;
2382 goto error_free;
2383 }
2384
2385 nlk = nlk_sk(sk);
2386 mutex_lock(&nlk->nl_cb_mutex);
2387 /* A dump is in progress... */
2388 if (nlk->cb_running) {
2389 ret = -EBUSY;
2390 goto error_unlock;
2391 }
2392 /* add reference of module which cb->dump belongs to */
2393 if (!try_module_get(control->module)) {
2394 ret = -EPROTONOSUPPORT;
2395 goto error_unlock;
2396 }
2397
2398 cb = &nlk->cb;
2399 memset(cb, 0, sizeof(*cb));
2400 cb->dump = control->dump;
2401 cb->done = control->done;
2402 cb->nlh = nlh;
2403 cb->data = control->data;
2404 cb->module = control->module;
2405 cb->min_dump_alloc = control->min_dump_alloc;
2406 cb->flags = control->flags;
2407 cb->skb = skb;
2408
2409 cb->strict_check = nlk_test_bit(STRICT_CHK, NETLINK_CB(skb).sk);
2410
2411 if (control->start) {
2412 cb->extack = control->extack;
2413 ret = control->start(cb);
2414 cb->extack = NULL;
2415 if (ret)
2416 goto error_put;
2417 }
2418
2419 WRITE_ONCE(nlk->cb_running, true);
2420 nlk->dump_done_errno = INT_MAX;
2421
2422 ret = netlink_dump(sk, true);
2423
2424 sock_put(sk);
2425
2426 if (ret)
2427 return ret;
2428
2429 /* We successfully started a dump, by returning -EINTR we
2430 * signal not to send ACK even if it was requested.
2431 */
2432 return -EINTR;
2433
2434error_put:
2435 module_put(control->module);
2436error_unlock:
2437 sock_put(sk);
2438 mutex_unlock(&nlk->nl_cb_mutex);
2439error_free:
2440 kfree_skb(skb);
2441 return ret;
2442}
2443EXPORT_SYMBOL(__netlink_dump_start);
2444
2445void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2446 const struct netlink_ext_ack *extack)
2447{
2448 struct sk_buff *skb;
2449 struct nlmsghdr *rep;
2450 struct nlmsgerr *errmsg;
2451 size_t payload = sizeof(*errmsg);
2452 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2453 unsigned int flags = 0;
2454 size_t tlvlen;
2455
2456 /* Error messages get the original request appened, unless the user
2457 * requests to cap the error message, and get extra error data if
2458 * requested.
2459 */
2460 if (err && !test_bit(NETLINK_F_CAP_ACK, &nlk->flags))
2461 payload += nlmsg_len(nlh);
2462 else
2463 flags |= NLM_F_CAPPED;
2464
2465 tlvlen = netlink_ack_tlv_len(nlk, err, extack);
2466 if (tlvlen)
2467 flags |= NLM_F_ACK_TLVS;
2468
2469 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2470 if (!skb)
2471 goto err_skb;
2472
2473 rep = nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2474 NLMSG_ERROR, sizeof(*errmsg), flags);
2475 if (!rep)
2476 goto err_bad_put;
2477 errmsg = nlmsg_data(rep);
2478 errmsg->error = err;
2479 errmsg->msg = *nlh;
2480
2481 if (!(flags & NLM_F_CAPPED)) {
2482 if (!nlmsg_append(skb, nlmsg_len(nlh)))
2483 goto err_bad_put;
2484
2485 memcpy(nlmsg_data(&errmsg->msg), nlmsg_data(nlh),
2486 nlmsg_len(nlh));
2487 }
2488
2489 if (tlvlen)
2490 netlink_ack_tlv_fill(skb, nlh, err, extack);
2491
2492 nlmsg_end(skb, rep);
2493
2494 nlmsg_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid);
2495
2496 return;
2497
2498err_bad_put:
2499 nlmsg_free(skb);
2500err_skb:
2501 WRITE_ONCE(NETLINK_CB(in_skb).sk->sk_err, ENOBUFS);
2502 sk_error_report(NETLINK_CB(in_skb).sk);
2503}
2504EXPORT_SYMBOL(netlink_ack);
2505
2506int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2507 struct nlmsghdr *,
2508 struct netlink_ext_ack *))
2509{
2510 struct netlink_ext_ack extack;
2511 struct nlmsghdr *nlh;
2512 int err;
2513
2514 while (skb->len >= nlmsg_total_size(0)) {
2515 int msglen;
2516
2517 memset(&extack, 0, sizeof(extack));
2518 nlh = nlmsg_hdr(skb);
2519 err = 0;
2520
2521 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2522 return 0;
2523
2524 /* Only requests are handled by the kernel */
2525 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2526 goto ack;
2527
2528 /* Skip control messages */
2529 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2530 goto ack;
2531
2532 err = cb(skb, nlh, &extack);
2533 if (err == -EINTR)
2534 goto skip;
2535
2536ack:
2537 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2538 netlink_ack(skb, nlh, err, &extack);
2539
2540skip:
2541 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2542 if (msglen > skb->len)
2543 msglen = skb->len;
2544 skb_pull(skb, msglen);
2545 }
2546
2547 return 0;
2548}
2549EXPORT_SYMBOL(netlink_rcv_skb);
2550
2551/**
2552 * nlmsg_notify - send a notification netlink message
2553 * @sk: netlink socket to use
2554 * @skb: notification message
2555 * @portid: destination netlink portid for reports or 0
2556 * @group: destination multicast group or 0
2557 * @report: 1 to report back, 0 to disable
2558 * @flags: allocation flags
2559 */
2560int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2561 unsigned int group, int report, gfp_t flags)
2562{
2563 int err = 0;
2564
2565 if (group) {
2566 int exclude_portid = 0;
2567
2568 if (report) {
2569 refcount_inc(&skb->users);
2570 exclude_portid = portid;
2571 }
2572
2573 /* errors reported via destination sk->sk_err, but propagate
2574 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2575 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2576 if (err == -ESRCH)
2577 err = 0;
2578 }
2579
2580 if (report) {
2581 int err2;
2582
2583 err2 = nlmsg_unicast(sk, skb, portid);
2584 if (!err)
2585 err = err2;
2586 }
2587
2588 return err;
2589}
2590EXPORT_SYMBOL(nlmsg_notify);
2591
2592#ifdef CONFIG_PROC_FS
2593struct nl_seq_iter {
2594 struct seq_net_private p;
2595 struct rhashtable_iter hti;
2596 int link;
2597};
2598
2599static void netlink_walk_start(struct nl_seq_iter *iter)
2600{
2601 rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti);
2602 rhashtable_walk_start(&iter->hti);
2603}
2604
2605static void netlink_walk_stop(struct nl_seq_iter *iter)
2606{
2607 rhashtable_walk_stop(&iter->hti);
2608 rhashtable_walk_exit(&iter->hti);
2609}
2610
2611static void *__netlink_seq_next(struct seq_file *seq)
2612{
2613 struct nl_seq_iter *iter = seq->private;
2614 struct netlink_sock *nlk;
2615
2616 do {
2617 for (;;) {
2618 nlk = rhashtable_walk_next(&iter->hti);
2619
2620 if (IS_ERR(nlk)) {
2621 if (PTR_ERR(nlk) == -EAGAIN)
2622 continue;
2623
2624 return nlk;
2625 }
2626
2627 if (nlk)
2628 break;
2629
2630 netlink_walk_stop(iter);
2631 if (++iter->link >= MAX_LINKS)
2632 return NULL;
2633
2634 netlink_walk_start(iter);
2635 }
2636 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2637
2638 return nlk;
2639}
2640
2641static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2642 __acquires(RCU)
2643{
2644 struct nl_seq_iter *iter = seq->private;
2645 void *obj = SEQ_START_TOKEN;
2646 loff_t pos;
2647
2648 iter->link = 0;
2649
2650 netlink_walk_start(iter);
2651
2652 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2653 obj = __netlink_seq_next(seq);
2654
2655 return obj;
2656}
2657
2658static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2659{
2660 ++*pos;
2661 return __netlink_seq_next(seq);
2662}
2663
2664static void netlink_native_seq_stop(struct seq_file *seq, void *v)
2665{
2666 struct nl_seq_iter *iter = seq->private;
2667
2668 if (iter->link >= MAX_LINKS)
2669 return;
2670
2671 netlink_walk_stop(iter);
2672}
2673
2674
2675static int netlink_native_seq_show(struct seq_file *seq, void *v)
2676{
2677 if (v == SEQ_START_TOKEN) {
2678 seq_puts(seq,
2679 "sk Eth Pid Groups "
2680 "Rmem Wmem Dump Locks Drops Inode\n");
2681 } else {
2682 struct sock *s = v;
2683 struct netlink_sock *nlk = nlk_sk(s);
2684
2685 seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n",
2686 s,
2687 s->sk_protocol,
2688 nlk->portid,
2689 nlk->groups ? (u32)nlk->groups[0] : 0,
2690 sk_rmem_alloc_get(s),
2691 sk_wmem_alloc_get(s),
2692 READ_ONCE(nlk->cb_running),
2693 refcount_read(&s->sk_refcnt),
2694 atomic_read(&s->sk_drops),
2695 sock_i_ino(s)
2696 );
2697
2698 }
2699 return 0;
2700}
2701
2702#ifdef CONFIG_BPF_SYSCALL
2703struct bpf_iter__netlink {
2704 __bpf_md_ptr(struct bpf_iter_meta *, meta);
2705 __bpf_md_ptr(struct netlink_sock *, sk);
2706};
2707
2708DEFINE_BPF_ITER_FUNC(netlink, struct bpf_iter_meta *meta, struct netlink_sock *sk)
2709
2710static int netlink_prog_seq_show(struct bpf_prog *prog,
2711 struct bpf_iter_meta *meta,
2712 void *v)
2713{
2714 struct bpf_iter__netlink ctx;
2715
2716 meta->seq_num--; /* skip SEQ_START_TOKEN */
2717 ctx.meta = meta;
2718 ctx.sk = nlk_sk((struct sock *)v);
2719 return bpf_iter_run_prog(prog, &ctx);
2720}
2721
2722static int netlink_seq_show(struct seq_file *seq, void *v)
2723{
2724 struct bpf_iter_meta meta;
2725 struct bpf_prog *prog;
2726
2727 meta.seq = seq;
2728 prog = bpf_iter_get_info(&meta, false);
2729 if (!prog)
2730 return netlink_native_seq_show(seq, v);
2731
2732 if (v != SEQ_START_TOKEN)
2733 return netlink_prog_seq_show(prog, &meta, v);
2734
2735 return 0;
2736}
2737
2738static void netlink_seq_stop(struct seq_file *seq, void *v)
2739{
2740 struct bpf_iter_meta meta;
2741 struct bpf_prog *prog;
2742
2743 if (!v) {
2744 meta.seq = seq;
2745 prog = bpf_iter_get_info(&meta, true);
2746 if (prog)
2747 (void)netlink_prog_seq_show(prog, &meta, v);
2748 }
2749
2750 netlink_native_seq_stop(seq, v);
2751}
2752#else
2753static int netlink_seq_show(struct seq_file *seq, void *v)
2754{
2755 return netlink_native_seq_show(seq, v);
2756}
2757
2758static void netlink_seq_stop(struct seq_file *seq, void *v)
2759{
2760 netlink_native_seq_stop(seq, v);
2761}
2762#endif
2763
2764static const struct seq_operations netlink_seq_ops = {
2765 .start = netlink_seq_start,
2766 .next = netlink_seq_next,
2767 .stop = netlink_seq_stop,
2768 .show = netlink_seq_show,
2769};
2770#endif
2771
2772int netlink_register_notifier(struct notifier_block *nb)
2773{
2774 return blocking_notifier_chain_register(&netlink_chain, nb);
2775}
2776EXPORT_SYMBOL(netlink_register_notifier);
2777
2778int netlink_unregister_notifier(struct notifier_block *nb)
2779{
2780 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2781}
2782EXPORT_SYMBOL(netlink_unregister_notifier);
2783
2784static const struct proto_ops netlink_ops = {
2785 .family = PF_NETLINK,
2786 .owner = THIS_MODULE,
2787 .release = netlink_release,
2788 .bind = netlink_bind,
2789 .connect = netlink_connect,
2790 .socketpair = sock_no_socketpair,
2791 .accept = sock_no_accept,
2792 .getname = netlink_getname,
2793 .poll = datagram_poll,
2794 .ioctl = netlink_ioctl,
2795 .listen = sock_no_listen,
2796 .shutdown = sock_no_shutdown,
2797 .setsockopt = netlink_setsockopt,
2798 .getsockopt = netlink_getsockopt,
2799 .sendmsg = netlink_sendmsg,
2800 .recvmsg = netlink_recvmsg,
2801 .mmap = sock_no_mmap,
2802};
2803
2804static const struct net_proto_family netlink_family_ops = {
2805 .family = PF_NETLINK,
2806 .create = netlink_create,
2807 .owner = THIS_MODULE, /* for consistency 8) */
2808};
2809
2810static int __net_init netlink_net_init(struct net *net)
2811{
2812#ifdef CONFIG_PROC_FS
2813 if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2814 sizeof(struct nl_seq_iter)))
2815 return -ENOMEM;
2816#endif
2817 return 0;
2818}
2819
2820static void __net_exit netlink_net_exit(struct net *net)
2821{
2822#ifdef CONFIG_PROC_FS
2823 remove_proc_entry("netlink", net->proc_net);
2824#endif
2825}
2826
2827static void __init netlink_add_usersock_entry(void)
2828{
2829 struct listeners *listeners;
2830 int groups = 32;
2831
2832 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2833 if (!listeners)
2834 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2835
2836 netlink_table_grab();
2837
2838 nl_table[NETLINK_USERSOCK].groups = groups;
2839 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2840 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2841 nl_table[NETLINK_USERSOCK].registered = 1;
2842 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2843
2844 netlink_table_ungrab();
2845}
2846
2847static struct pernet_operations __net_initdata netlink_net_ops = {
2848 .init = netlink_net_init,
2849 .exit = netlink_net_exit,
2850};
2851
2852static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2853{
2854 const struct netlink_sock *nlk = data;
2855 struct netlink_compare_arg arg;
2856
2857 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2858 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2859}
2860
2861static const struct rhashtable_params netlink_rhashtable_params = {
2862 .head_offset = offsetof(struct netlink_sock, node),
2863 .key_len = netlink_compare_arg_len,
2864 .obj_hashfn = netlink_hash,
2865 .obj_cmpfn = netlink_compare,
2866 .automatic_shrinking = true,
2867};
2868
2869#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2870BTF_ID_LIST(btf_netlink_sock_id)
2871BTF_ID(struct, netlink_sock)
2872
2873static const struct bpf_iter_seq_info netlink_seq_info = {
2874 .seq_ops = &netlink_seq_ops,
2875 .init_seq_private = bpf_iter_init_seq_net,
2876 .fini_seq_private = bpf_iter_fini_seq_net,
2877 .seq_priv_size = sizeof(struct nl_seq_iter),
2878};
2879
2880static struct bpf_iter_reg netlink_reg_info = {
2881 .target = "netlink",
2882 .ctx_arg_info_size = 1,
2883 .ctx_arg_info = {
2884 { offsetof(struct bpf_iter__netlink, sk),
2885 PTR_TO_BTF_ID_OR_NULL },
2886 },
2887 .seq_info = &netlink_seq_info,
2888};
2889
2890static int __init bpf_iter_register(void)
2891{
2892 netlink_reg_info.ctx_arg_info[0].btf_id = *btf_netlink_sock_id;
2893 return bpf_iter_reg_target(&netlink_reg_info);
2894}
2895#endif
2896
2897static int __init netlink_proto_init(void)
2898{
2899 int i;
2900 int err = proto_register(&netlink_proto, 0);
2901
2902 if (err != 0)
2903 goto out;
2904
2905#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2906 err = bpf_iter_register();
2907 if (err)
2908 goto out;
2909#endif
2910
2911 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof_field(struct sk_buff, cb));
2912
2913 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2914 if (!nl_table)
2915 goto panic;
2916
2917 for (i = 0; i < MAX_LINKS; i++) {
2918 if (rhashtable_init(&nl_table[i].hash,
2919 &netlink_rhashtable_params) < 0)
2920 goto panic;
2921 }
2922
2923 netlink_add_usersock_entry();
2924
2925 sock_register(&netlink_family_ops);
2926 register_pernet_subsys(&netlink_net_ops);
2927 register_pernet_subsys(&netlink_tap_net_ops);
2928 /* The netlink device handler may be needed early. */
2929 rtnetlink_init();
2930out:
2931 return err;
2932panic:
2933 panic("netlink_init: Cannot allocate nl_table\n");
2934}
2935
2936core_initcall(netlink_proto_init);
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/genetlink.h>
63#include <linux/net_namespace.h>
64#include <linux/nospec.h>
65#include <linux/btf_ids.h>
66
67#include <net/net_namespace.h>
68#include <net/netns/generic.h>
69#include <net/sock.h>
70#include <net/scm.h>
71#include <net/netlink.h>
72#define CREATE_TRACE_POINTS
73#include <trace/events/netlink.h>
74
75#include "af_netlink.h"
76
77struct listeners {
78 struct rcu_head rcu;
79 unsigned long masks[];
80};
81
82/* state bits */
83#define NETLINK_S_CONGESTED 0x0
84
85static inline int netlink_is_kernel(struct sock *sk)
86{
87 return nlk_test_bit(KERNEL_SOCKET, sk);
88}
89
90struct netlink_table *nl_table __read_mostly;
91EXPORT_SYMBOL_GPL(nl_table);
92
93static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
94
95static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
96
97static 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
133static 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 */
143DEFINE_RWLOCK(nl_table_lock);
144EXPORT_SYMBOL_GPL(nl_table_lock);
145static 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
149static BLOCKING_NOTIFIER_HEAD(netlink_chain);
150
151
152static const struct rhashtable_params netlink_rhashtable_params;
153
154void do_trace_netlink_extack(const char *msg)
155{
156 trace_netlink_extack(msg);
157}
158EXPORT_SYMBOL(do_trace_netlink_extack);
159
160static inline u32 netlink_group_mask(u32 group)
161{
162 if (group > 32)
163 return 0;
164 return group ? 1 << (group - 1) : 0;
165}
166
167static 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
185static unsigned int netlink_tap_net_id;
186
187struct netlink_tap_net {
188 struct list_head netlink_tap_all;
189 struct mutex netlink_tap_lock;
190};
191
192int 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}
208EXPORT_SYMBOL_GPL(netlink_add_tap);
209
210static 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);
228out:
229 mutex_unlock(&nn->netlink_tap_lock);
230
231 if (found)
232 module_put(nt->module);
233
234 return found ? 0 : -ENODEV;
235}
236
237int 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}
246EXPORT_SYMBOL_GPL(netlink_remove_tap);
247
248static __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
257static 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
263static 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
285static 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
316static 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
331static 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
343static 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
350static 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
362static 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
372static 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
385static 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
394static 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
417static 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
431void 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
456void 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
463static inline void
464netlink_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
475static inline void
476netlink_unlock_table(void)
477{
478 if (atomic_dec_and_test(&nl_table_users))
479 wake_up(&nl_table_wait);
480}
481
482struct 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
492static 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
502static 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
510static 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
520static 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
530static 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
544static const struct proto_ops netlink_ops;
545
546static void
547netlink_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
570static 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 /* portid can be read locklessly from netlink_getname(). */
582 WRITE_ONCE(nlk_sk(sk)->portid, portid);
583
584 sock_hold(sk);
585
586 err = __netlink_insert(table, sk);
587 if (err) {
588 /* In case the hashtable backend returns with -EBUSY
589 * from here, it must not escape to the caller.
590 */
591 if (unlikely(err == -EBUSY))
592 err = -EOVERFLOW;
593 if (err == -EEXIST)
594 err = -EADDRINUSE;
595 sock_put(sk);
596 goto err;
597 }
598
599 /* We need to ensure that the socket is hashed and visible. */
600 smp_wmb();
601 /* Paired with lockless reads from netlink_bind(),
602 * netlink_connect() and netlink_sendmsg().
603 */
604 WRITE_ONCE(nlk_sk(sk)->bound, portid);
605
606err:
607 release_sock(sk);
608 return err;
609}
610
611static void netlink_remove(struct sock *sk)
612{
613 struct netlink_table *table;
614
615 table = &nl_table[sk->sk_protocol];
616 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
617 netlink_rhashtable_params)) {
618 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
619 __sock_put(sk);
620 }
621
622 netlink_table_grab();
623 if (nlk_sk(sk)->subscriptions) {
624 __sk_del_bind_node(sk);
625 netlink_update_listeners(sk);
626 }
627 if (sk->sk_protocol == NETLINK_GENERIC)
628 atomic_inc(&genl_sk_destructing_cnt);
629 netlink_table_ungrab();
630}
631
632static struct proto netlink_proto = {
633 .name = "NETLINK",
634 .owner = THIS_MODULE,
635 .obj_size = sizeof(struct netlink_sock),
636};
637
638static int __netlink_create(struct net *net, struct socket *sock,
639 struct mutex *dump_cb_mutex, int protocol,
640 int kern)
641{
642 struct sock *sk;
643 struct netlink_sock *nlk;
644
645 sock->ops = &netlink_ops;
646
647 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
648 if (!sk)
649 return -ENOMEM;
650
651 sock_init_data(sock, sk);
652
653 nlk = nlk_sk(sk);
654 mutex_init(&nlk->nl_cb_mutex);
655 lockdep_set_class_and_name(&nlk->nl_cb_mutex,
656 nlk_cb_mutex_keys + protocol,
657 nlk_cb_mutex_key_strings[protocol]);
658 nlk->dump_cb_mutex = dump_cb_mutex;
659 init_waitqueue_head(&nlk->wait);
660
661 sk->sk_destruct = netlink_sock_destruct;
662 sk->sk_protocol = protocol;
663 return 0;
664}
665
666static int netlink_create(struct net *net, struct socket *sock, int protocol,
667 int kern)
668{
669 struct module *module = NULL;
670 struct mutex *cb_mutex;
671 struct netlink_sock *nlk;
672 int (*bind)(struct net *net, int group);
673 void (*unbind)(struct net *net, int group);
674 void (*release)(struct sock *sock, unsigned long *groups);
675 int err = 0;
676
677 sock->state = SS_UNCONNECTED;
678
679 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
680 return -ESOCKTNOSUPPORT;
681
682 if (protocol < 0 || protocol >= MAX_LINKS)
683 return -EPROTONOSUPPORT;
684 protocol = array_index_nospec(protocol, MAX_LINKS);
685
686 netlink_lock_table();
687#ifdef CONFIG_MODULES
688 if (!nl_table[protocol].registered) {
689 netlink_unlock_table();
690 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
691 netlink_lock_table();
692 }
693#endif
694 if (nl_table[protocol].registered &&
695 try_module_get(nl_table[protocol].module))
696 module = nl_table[protocol].module;
697 else
698 err = -EPROTONOSUPPORT;
699 cb_mutex = nl_table[protocol].cb_mutex;
700 bind = nl_table[protocol].bind;
701 unbind = nl_table[protocol].unbind;
702 release = nl_table[protocol].release;
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 sock_prot_inuse_add(net, &netlink_proto, 1);
713
714 nlk = nlk_sk(sock->sk);
715 nlk->module = module;
716 nlk->netlink_bind = bind;
717 nlk->netlink_unbind = unbind;
718 nlk->netlink_release = release;
719out:
720 return err;
721
722out_module:
723 module_put(module);
724 goto out;
725}
726
727static void deferred_put_nlk_sk(struct rcu_head *head)
728{
729 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
730 struct sock *sk = &nlk->sk;
731
732 kfree(nlk->groups);
733 nlk->groups = NULL;
734
735 if (!refcount_dec_and_test(&sk->sk_refcnt))
736 return;
737
738 if (nlk->cb_running && nlk->cb.done) {
739 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
740 schedule_work(&nlk->work);
741 return;
742 }
743
744 sk_free(sk);
745}
746
747static int netlink_release(struct socket *sock)
748{
749 struct sock *sk = sock->sk;
750 struct netlink_sock *nlk;
751
752 if (!sk)
753 return 0;
754
755 netlink_remove(sk);
756 sock_orphan(sk);
757 nlk = nlk_sk(sk);
758
759 /*
760 * OK. Socket is unlinked, any packets that arrive now
761 * will be purged.
762 */
763 if (nlk->netlink_release)
764 nlk->netlink_release(sk, nlk->groups);
765
766 /* must not acquire netlink_table_lock in any way again before unbind
767 * and notifying genetlink is done as otherwise it might deadlock
768 */
769 if (nlk->netlink_unbind) {
770 int i;
771
772 for (i = 0; i < nlk->ngroups; i++)
773 if (test_bit(i, nlk->groups))
774 nlk->netlink_unbind(sock_net(sk), i + 1);
775 }
776 if (sk->sk_protocol == NETLINK_GENERIC &&
777 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
778 wake_up(&genl_sk_destructing_waitq);
779
780 sock->sk = NULL;
781 wake_up_interruptible_all(&nlk->wait);
782
783 skb_queue_purge(&sk->sk_write_queue);
784
785 if (nlk->portid && nlk->bound) {
786 struct netlink_notify n = {
787 .net = sock_net(sk),
788 .protocol = sk->sk_protocol,
789 .portid = nlk->portid,
790 };
791 blocking_notifier_call_chain(&netlink_chain,
792 NETLINK_URELEASE, &n);
793 }
794
795 module_put(nlk->module);
796
797 if (netlink_is_kernel(sk)) {
798 netlink_table_grab();
799 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
800 if (--nl_table[sk->sk_protocol].registered == 0) {
801 struct listeners *old;
802
803 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
804 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
805 kfree_rcu(old, rcu);
806 nl_table[sk->sk_protocol].module = NULL;
807 nl_table[sk->sk_protocol].bind = NULL;
808 nl_table[sk->sk_protocol].unbind = NULL;
809 nl_table[sk->sk_protocol].flags = 0;
810 nl_table[sk->sk_protocol].registered = 0;
811 }
812 netlink_table_ungrab();
813 }
814
815 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
816
817 /* Because struct net might disappear soon, do not keep a pointer. */
818 if (!sk->sk_net_refcnt && sock_net(sk) != &init_net) {
819 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false);
820 /* Because of deferred_put_nlk_sk and use of work queue,
821 * it is possible netns will be freed before this socket.
822 */
823 sock_net_set(sk, &init_net);
824 __netns_tracker_alloc(&init_net, &sk->ns_tracker,
825 false, GFP_KERNEL);
826 }
827 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
828 return 0;
829}
830
831static int netlink_autobind(struct socket *sock)
832{
833 struct sock *sk = sock->sk;
834 struct net *net = sock_net(sk);
835 struct netlink_table *table = &nl_table[sk->sk_protocol];
836 s32 portid = task_tgid_vnr(current);
837 int err;
838 s32 rover = -4096;
839 bool ok;
840
841retry:
842 cond_resched();
843 rcu_read_lock();
844 ok = !__netlink_lookup(table, portid, net);
845 rcu_read_unlock();
846 if (!ok) {
847 /* Bind collision, search negative portid values. */
848 if (rover == -4096)
849 /* rover will be in range [S32_MIN, -4097] */
850 rover = S32_MIN + get_random_u32_below(-4096 - S32_MIN);
851 else if (rover >= -4096)
852 rover = -4097;
853 portid = rover--;
854 goto retry;
855 }
856
857 err = netlink_insert(sk, portid);
858 if (err == -EADDRINUSE)
859 goto retry;
860
861 /* If 2 threads race to autobind, that is fine. */
862 if (err == -EBUSY)
863 err = 0;
864
865 return err;
866}
867
868/**
869 * __netlink_ns_capable - General netlink message capability test
870 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
871 * @user_ns: The user namespace of the capability to use
872 * @cap: The capability to use
873 *
874 * Test to see if the opener of the socket we received the message
875 * from had when the netlink socket was created and the sender of the
876 * message has the capability @cap in the user namespace @user_ns.
877 */
878bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
879 struct user_namespace *user_ns, int cap)
880{
881 return ((nsp->flags & NETLINK_SKB_DST) ||
882 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
883 ns_capable(user_ns, cap);
884}
885EXPORT_SYMBOL(__netlink_ns_capable);
886
887/**
888 * netlink_ns_capable - General netlink message capability test
889 * @skb: socket buffer holding a netlink command from userspace
890 * @user_ns: The user namespace of the capability to use
891 * @cap: The capability to use
892 *
893 * Test to see if the opener of the socket we received the message
894 * from had when the netlink socket was created and the sender of the
895 * message has the capability @cap in the user namespace @user_ns.
896 */
897bool netlink_ns_capable(const struct sk_buff *skb,
898 struct user_namespace *user_ns, int cap)
899{
900 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
901}
902EXPORT_SYMBOL(netlink_ns_capable);
903
904/**
905 * netlink_capable - Netlink global message capability test
906 * @skb: socket buffer holding a netlink command from userspace
907 * @cap: The capability to use
908 *
909 * Test to see if the opener of the socket we received the message
910 * from had when the netlink socket was created and the sender of the
911 * message has the capability @cap in all user namespaces.
912 */
913bool netlink_capable(const struct sk_buff *skb, int cap)
914{
915 return netlink_ns_capable(skb, &init_user_ns, cap);
916}
917EXPORT_SYMBOL(netlink_capable);
918
919/**
920 * netlink_net_capable - Netlink network namespace message capability test
921 * @skb: socket buffer holding a netlink command from userspace
922 * @cap: The capability to use
923 *
924 * Test to see if the opener of the socket we received the message
925 * from had when the netlink socket was created and the sender of the
926 * message has the capability @cap over the network namespace of
927 * the socket we received the message from.
928 */
929bool netlink_net_capable(const struct sk_buff *skb, int cap)
930{
931 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
932}
933EXPORT_SYMBOL(netlink_net_capable);
934
935static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
936{
937 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
938 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
939}
940
941static void
942netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
943{
944 struct netlink_sock *nlk = nlk_sk(sk);
945
946 if (nlk->subscriptions && !subscriptions)
947 __sk_del_bind_node(sk);
948 else if (!nlk->subscriptions && subscriptions)
949 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
950 nlk->subscriptions = subscriptions;
951}
952
953static int netlink_realloc_groups(struct sock *sk)
954{
955 struct netlink_sock *nlk = nlk_sk(sk);
956 unsigned int groups;
957 unsigned long *new_groups;
958 int err = 0;
959
960 netlink_table_grab();
961
962 groups = nl_table[sk->sk_protocol].groups;
963 if (!nl_table[sk->sk_protocol].registered) {
964 err = -ENOENT;
965 goto out_unlock;
966 }
967
968 if (nlk->ngroups >= groups)
969 goto out_unlock;
970
971 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
972 if (new_groups == NULL) {
973 err = -ENOMEM;
974 goto out_unlock;
975 }
976 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
977 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
978
979 nlk->groups = new_groups;
980 nlk->ngroups = groups;
981 out_unlock:
982 netlink_table_ungrab();
983 return err;
984}
985
986static void netlink_undo_bind(int group, long unsigned int groups,
987 struct sock *sk)
988{
989 struct netlink_sock *nlk = nlk_sk(sk);
990 int undo;
991
992 if (!nlk->netlink_unbind)
993 return;
994
995 for (undo = 0; undo < group; undo++)
996 if (test_bit(undo, &groups))
997 nlk->netlink_unbind(sock_net(sk), undo + 1);
998}
999
1000static int netlink_bind(struct socket *sock, struct sockaddr *addr,
1001 int addr_len)
1002{
1003 struct sock *sk = sock->sk;
1004 struct net *net = sock_net(sk);
1005 struct netlink_sock *nlk = nlk_sk(sk);
1006 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1007 int err = 0;
1008 unsigned long groups;
1009 bool bound;
1010
1011 if (addr_len < sizeof(struct sockaddr_nl))
1012 return -EINVAL;
1013
1014 if (nladdr->nl_family != AF_NETLINK)
1015 return -EINVAL;
1016 groups = nladdr->nl_groups;
1017
1018 /* Only superuser is allowed to listen multicasts */
1019 if (groups) {
1020 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1021 return -EPERM;
1022 err = netlink_realloc_groups(sk);
1023 if (err)
1024 return err;
1025 }
1026
1027 if (nlk->ngroups < BITS_PER_LONG)
1028 groups &= (1UL << nlk->ngroups) - 1;
1029
1030 /* Paired with WRITE_ONCE() in netlink_insert() */
1031 bound = READ_ONCE(nlk->bound);
1032 if (bound) {
1033 /* Ensure nlk->portid is up-to-date. */
1034 smp_rmb();
1035
1036 if (nladdr->nl_pid != nlk->portid)
1037 return -EINVAL;
1038 }
1039
1040 if (nlk->netlink_bind && groups) {
1041 int group;
1042
1043 /* nl_groups is a u32, so cap the maximum groups we can bind */
1044 for (group = 0; group < BITS_PER_TYPE(u32); group++) {
1045 if (!test_bit(group, &groups))
1046 continue;
1047 err = nlk->netlink_bind(net, group + 1);
1048 if (!err)
1049 continue;
1050 netlink_undo_bind(group, groups, sk);
1051 return err;
1052 }
1053 }
1054
1055 /* No need for barriers here as we return to user-space without
1056 * using any of the bound attributes.
1057 */
1058 netlink_lock_table();
1059 if (!bound) {
1060 err = nladdr->nl_pid ?
1061 netlink_insert(sk, nladdr->nl_pid) :
1062 netlink_autobind(sock);
1063 if (err) {
1064 netlink_undo_bind(BITS_PER_TYPE(u32), groups, sk);
1065 goto unlock;
1066 }
1067 }
1068
1069 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1070 goto unlock;
1071 netlink_unlock_table();
1072
1073 netlink_table_grab();
1074 netlink_update_subscriptions(sk, nlk->subscriptions +
1075 hweight32(groups) -
1076 hweight32(nlk->groups[0]));
1077 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1078 netlink_update_listeners(sk);
1079 netlink_table_ungrab();
1080
1081 return 0;
1082
1083unlock:
1084 netlink_unlock_table();
1085 return err;
1086}
1087
1088static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1089 int alen, int flags)
1090{
1091 int err = 0;
1092 struct sock *sk = sock->sk;
1093 struct netlink_sock *nlk = nlk_sk(sk);
1094 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1095
1096 if (alen < sizeof(addr->sa_family))
1097 return -EINVAL;
1098
1099 if (addr->sa_family == AF_UNSPEC) {
1100 /* paired with READ_ONCE() in netlink_getsockbyportid() */
1101 WRITE_ONCE(sk->sk_state, NETLINK_UNCONNECTED);
1102 /* dst_portid and dst_group can be read locklessly */
1103 WRITE_ONCE(nlk->dst_portid, 0);
1104 WRITE_ONCE(nlk->dst_group, 0);
1105 return 0;
1106 }
1107 if (addr->sa_family != AF_NETLINK)
1108 return -EINVAL;
1109
1110 if (alen < sizeof(struct sockaddr_nl))
1111 return -EINVAL;
1112
1113 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1114 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1115 return -EPERM;
1116
1117 /* No need for barriers here as we return to user-space without
1118 * using any of the bound attributes.
1119 * Paired with WRITE_ONCE() in netlink_insert().
1120 */
1121 if (!READ_ONCE(nlk->bound))
1122 err = netlink_autobind(sock);
1123
1124 if (err == 0) {
1125 /* paired with READ_ONCE() in netlink_getsockbyportid() */
1126 WRITE_ONCE(sk->sk_state, NETLINK_CONNECTED);
1127 /* dst_portid and dst_group can be read locklessly */
1128 WRITE_ONCE(nlk->dst_portid, nladdr->nl_pid);
1129 WRITE_ONCE(nlk->dst_group, ffs(nladdr->nl_groups));
1130 }
1131
1132 return err;
1133}
1134
1135static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1136 int peer)
1137{
1138 struct sock *sk = sock->sk;
1139 struct netlink_sock *nlk = nlk_sk(sk);
1140 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1141
1142 nladdr->nl_family = AF_NETLINK;
1143 nladdr->nl_pad = 0;
1144
1145 if (peer) {
1146 /* Paired with WRITE_ONCE() in netlink_connect() */
1147 nladdr->nl_pid = READ_ONCE(nlk->dst_portid);
1148 nladdr->nl_groups = netlink_group_mask(READ_ONCE(nlk->dst_group));
1149 } else {
1150 /* Paired with WRITE_ONCE() in netlink_insert() */
1151 nladdr->nl_pid = READ_ONCE(nlk->portid);
1152 netlink_lock_table();
1153 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1154 netlink_unlock_table();
1155 }
1156 return sizeof(*nladdr);
1157}
1158
1159static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1160 unsigned long arg)
1161{
1162 /* try to hand this ioctl down to the NIC drivers.
1163 */
1164 return -ENOIOCTLCMD;
1165}
1166
1167static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1168{
1169 struct sock *sock;
1170 struct netlink_sock *nlk;
1171
1172 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1173 if (!sock)
1174 return ERR_PTR(-ECONNREFUSED);
1175
1176 /* Don't bother queuing skb if kernel socket has no input function */
1177 nlk = nlk_sk(sock);
1178 /* dst_portid and sk_state can be changed in netlink_connect() */
1179 if (READ_ONCE(sock->sk_state) == NETLINK_CONNECTED &&
1180 READ_ONCE(nlk->dst_portid) != nlk_sk(ssk)->portid) {
1181 sock_put(sock);
1182 return ERR_PTR(-ECONNREFUSED);
1183 }
1184 return sock;
1185}
1186
1187struct sock *netlink_getsockbyfilp(struct file *filp)
1188{
1189 struct inode *inode = file_inode(filp);
1190 struct sock *sock;
1191
1192 if (!S_ISSOCK(inode->i_mode))
1193 return ERR_PTR(-ENOTSOCK);
1194
1195 sock = SOCKET_I(inode)->sk;
1196 if (sock->sk_family != AF_NETLINK)
1197 return ERR_PTR(-EINVAL);
1198
1199 sock_hold(sock);
1200 return sock;
1201}
1202
1203struct sk_buff *netlink_alloc_large_skb(unsigned int size, int broadcast)
1204{
1205 size_t head_size = SKB_HEAD_ALIGN(size);
1206 struct sk_buff *skb;
1207 void *data;
1208
1209 if (head_size <= PAGE_SIZE || broadcast)
1210 return alloc_skb(size, GFP_KERNEL);
1211
1212 data = kvmalloc(head_size, GFP_KERNEL);
1213 if (!data)
1214 return NULL;
1215
1216 skb = __build_skb(data, head_size);
1217 if (!skb)
1218 kvfree(data);
1219 else if (is_vmalloc_addr(data))
1220 skb->destructor = netlink_skb_destructor;
1221
1222 return skb;
1223}
1224
1225/*
1226 * Attach a skb to a netlink socket.
1227 * The caller must hold a reference to the destination socket. On error, the
1228 * reference is dropped. The skb is not send to the destination, just all
1229 * all error checks are performed and memory in the queue is reserved.
1230 * Return values:
1231 * < 0: error. skb freed, reference to sock dropped.
1232 * 0: continue
1233 * 1: repeat lookup - reference dropped while waiting for socket memory.
1234 */
1235int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1236 long *timeo, struct sock *ssk)
1237{
1238 struct netlink_sock *nlk;
1239
1240 nlk = nlk_sk(sk);
1241
1242 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1243 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1244 DECLARE_WAITQUEUE(wait, current);
1245 if (!*timeo) {
1246 if (!ssk || netlink_is_kernel(ssk))
1247 netlink_overrun(sk);
1248 sock_put(sk);
1249 kfree_skb(skb);
1250 return -EAGAIN;
1251 }
1252
1253 __set_current_state(TASK_INTERRUPTIBLE);
1254 add_wait_queue(&nlk->wait, &wait);
1255
1256 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1257 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1258 !sock_flag(sk, SOCK_DEAD))
1259 *timeo = schedule_timeout(*timeo);
1260
1261 __set_current_state(TASK_RUNNING);
1262 remove_wait_queue(&nlk->wait, &wait);
1263 sock_put(sk);
1264
1265 if (signal_pending(current)) {
1266 kfree_skb(skb);
1267 return sock_intr_errno(*timeo);
1268 }
1269 return 1;
1270 }
1271 netlink_skb_set_owner_r(skb, sk);
1272 return 0;
1273}
1274
1275static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1276{
1277 int len = skb->len;
1278
1279 netlink_deliver_tap(sock_net(sk), skb);
1280
1281 skb_queue_tail(&sk->sk_receive_queue, skb);
1282 sk->sk_data_ready(sk);
1283 return len;
1284}
1285
1286int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1287{
1288 int len = __netlink_sendskb(sk, skb);
1289
1290 sock_put(sk);
1291 return len;
1292}
1293
1294void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1295{
1296 kfree_skb(skb);
1297 sock_put(sk);
1298}
1299
1300static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1301{
1302 int delta;
1303
1304 WARN_ON(skb->sk != NULL);
1305 delta = skb->end - skb->tail;
1306 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1307 return skb;
1308
1309 if (skb_shared(skb)) {
1310 struct sk_buff *nskb = skb_clone(skb, allocation);
1311 if (!nskb)
1312 return skb;
1313 consume_skb(skb);
1314 skb = nskb;
1315 }
1316
1317 pskb_expand_head(skb, 0, -delta,
1318 (allocation & ~__GFP_DIRECT_RECLAIM) |
1319 __GFP_NOWARN | __GFP_NORETRY);
1320 return skb;
1321}
1322
1323static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1324 struct sock *ssk)
1325{
1326 int ret;
1327 struct netlink_sock *nlk = nlk_sk(sk);
1328
1329 ret = -ECONNREFUSED;
1330 if (nlk->netlink_rcv != NULL) {
1331 ret = skb->len;
1332 netlink_skb_set_owner_r(skb, sk);
1333 NETLINK_CB(skb).sk = ssk;
1334 netlink_deliver_tap_kernel(sk, ssk, skb);
1335 nlk->netlink_rcv(skb);
1336 consume_skb(skb);
1337 } else {
1338 kfree_skb(skb);
1339 }
1340 sock_put(sk);
1341 return ret;
1342}
1343
1344int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1345 u32 portid, int nonblock)
1346{
1347 struct sock *sk;
1348 int err;
1349 long timeo;
1350
1351 skb = netlink_trim(skb, gfp_any());
1352
1353 timeo = sock_sndtimeo(ssk, nonblock);
1354retry:
1355 sk = netlink_getsockbyportid(ssk, portid);
1356 if (IS_ERR(sk)) {
1357 kfree_skb(skb);
1358 return PTR_ERR(sk);
1359 }
1360 if (netlink_is_kernel(sk))
1361 return netlink_unicast_kernel(sk, skb, ssk);
1362
1363 if (sk_filter(sk, skb)) {
1364 err = skb->len;
1365 kfree_skb(skb);
1366 sock_put(sk);
1367 return err;
1368 }
1369
1370 err = netlink_attachskb(sk, skb, &timeo, ssk);
1371 if (err == 1)
1372 goto retry;
1373 if (err)
1374 return err;
1375
1376 return netlink_sendskb(sk, skb);
1377}
1378EXPORT_SYMBOL(netlink_unicast);
1379
1380int netlink_has_listeners(struct sock *sk, unsigned int group)
1381{
1382 int res = 0;
1383 struct listeners *listeners;
1384
1385 BUG_ON(!netlink_is_kernel(sk));
1386
1387 rcu_read_lock();
1388 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1389
1390 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1391 res = test_bit(group - 1, listeners->masks);
1392
1393 rcu_read_unlock();
1394
1395 return res;
1396}
1397EXPORT_SYMBOL_GPL(netlink_has_listeners);
1398
1399bool netlink_strict_get_check(struct sk_buff *skb)
1400{
1401 return nlk_test_bit(STRICT_CHK, NETLINK_CB(skb).sk);
1402}
1403EXPORT_SYMBOL_GPL(netlink_strict_get_check);
1404
1405static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1406{
1407 struct netlink_sock *nlk = nlk_sk(sk);
1408
1409 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1410 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1411 netlink_skb_set_owner_r(skb, sk);
1412 __netlink_sendskb(sk, skb);
1413 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1414 }
1415 return -1;
1416}
1417
1418struct netlink_broadcast_data {
1419 struct sock *exclude_sk;
1420 struct net *net;
1421 u32 portid;
1422 u32 group;
1423 int failure;
1424 int delivery_failure;
1425 int congested;
1426 int delivered;
1427 gfp_t allocation;
1428 struct sk_buff *skb, *skb2;
1429 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1430 void *tx_data;
1431};
1432
1433static void do_one_broadcast(struct sock *sk,
1434 struct netlink_broadcast_data *p)
1435{
1436 struct netlink_sock *nlk = nlk_sk(sk);
1437 int val;
1438
1439 if (p->exclude_sk == sk)
1440 return;
1441
1442 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1443 !test_bit(p->group - 1, nlk->groups))
1444 return;
1445
1446 if (!net_eq(sock_net(sk), p->net)) {
1447 if (!nlk_test_bit(LISTEN_ALL_NSID, sk))
1448 return;
1449
1450 if (!peernet_has_id(sock_net(sk), p->net))
1451 return;
1452
1453 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1454 CAP_NET_BROADCAST))
1455 return;
1456 }
1457
1458 if (p->failure) {
1459 netlink_overrun(sk);
1460 return;
1461 }
1462
1463 sock_hold(sk);
1464 if (p->skb2 == NULL) {
1465 if (skb_shared(p->skb)) {
1466 p->skb2 = skb_clone(p->skb, p->allocation);
1467 } else {
1468 p->skb2 = skb_get(p->skb);
1469 /*
1470 * skb ownership may have been set when
1471 * delivered to a previous socket.
1472 */
1473 skb_orphan(p->skb2);
1474 }
1475 }
1476 if (p->skb2 == NULL) {
1477 netlink_overrun(sk);
1478 /* Clone failed. Notify ALL listeners. */
1479 p->failure = 1;
1480 if (nlk_test_bit(BROADCAST_SEND_ERROR, sk))
1481 p->delivery_failure = 1;
1482 goto out;
1483 }
1484
1485 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1486 kfree_skb(p->skb2);
1487 p->skb2 = NULL;
1488 goto out;
1489 }
1490
1491 if (sk_filter(sk, p->skb2)) {
1492 kfree_skb(p->skb2);
1493 p->skb2 = NULL;
1494 goto out;
1495 }
1496 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1497 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1498 NETLINK_CB(p->skb2).nsid_is_set = true;
1499 val = netlink_broadcast_deliver(sk, p->skb2);
1500 if (val < 0) {
1501 netlink_overrun(sk);
1502 if (nlk_test_bit(BROADCAST_SEND_ERROR, sk))
1503 p->delivery_failure = 1;
1504 } else {
1505 p->congested |= val;
1506 p->delivered = 1;
1507 p->skb2 = NULL;
1508 }
1509out:
1510 sock_put(sk);
1511}
1512
1513int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb,
1514 u32 portid,
1515 u32 group, gfp_t allocation,
1516 netlink_filter_fn filter,
1517 void *filter_data)
1518{
1519 struct net *net = sock_net(ssk);
1520 struct netlink_broadcast_data info;
1521 struct sock *sk;
1522
1523 skb = netlink_trim(skb, allocation);
1524
1525 info.exclude_sk = ssk;
1526 info.net = net;
1527 info.portid = portid;
1528 info.group = group;
1529 info.failure = 0;
1530 info.delivery_failure = 0;
1531 info.congested = 0;
1532 info.delivered = 0;
1533 info.allocation = allocation;
1534 info.skb = skb;
1535 info.skb2 = NULL;
1536 info.tx_filter = filter;
1537 info.tx_data = filter_data;
1538
1539 /* While we sleep in clone, do not allow to change socket list */
1540
1541 netlink_lock_table();
1542
1543 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1544 do_one_broadcast(sk, &info);
1545
1546 consume_skb(skb);
1547
1548 netlink_unlock_table();
1549
1550 if (info.delivery_failure) {
1551 kfree_skb(info.skb2);
1552 return -ENOBUFS;
1553 }
1554 consume_skb(info.skb2);
1555
1556 if (info.delivered) {
1557 if (info.congested && gfpflags_allow_blocking(allocation))
1558 yield();
1559 return 0;
1560 }
1561 return -ESRCH;
1562}
1563EXPORT_SYMBOL(netlink_broadcast_filtered);
1564
1565int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1566 u32 group, gfp_t allocation)
1567{
1568 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1569 NULL, NULL);
1570}
1571EXPORT_SYMBOL(netlink_broadcast);
1572
1573struct netlink_set_err_data {
1574 struct sock *exclude_sk;
1575 u32 portid;
1576 u32 group;
1577 int code;
1578};
1579
1580static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1581{
1582 struct netlink_sock *nlk = nlk_sk(sk);
1583 int ret = 0;
1584
1585 if (sk == p->exclude_sk)
1586 goto out;
1587
1588 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1589 goto out;
1590
1591 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1592 !test_bit(p->group - 1, nlk->groups))
1593 goto out;
1594
1595 if (p->code == ENOBUFS && nlk_test_bit(RECV_NO_ENOBUFS, sk)) {
1596 ret = 1;
1597 goto out;
1598 }
1599
1600 WRITE_ONCE(sk->sk_err, p->code);
1601 sk_error_report(sk);
1602out:
1603 return ret;
1604}
1605
1606/**
1607 * netlink_set_err - report error to broadcast listeners
1608 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1609 * @portid: the PORTID of a process that we want to skip (if any)
1610 * @group: the broadcast group that will notice the error
1611 * @code: error code, must be negative (as usual in kernelspace)
1612 *
1613 * This function returns the number of broadcast listeners that have set the
1614 * NETLINK_NO_ENOBUFS socket option.
1615 */
1616int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1617{
1618 struct netlink_set_err_data info;
1619 unsigned long flags;
1620 struct sock *sk;
1621 int ret = 0;
1622
1623 info.exclude_sk = ssk;
1624 info.portid = portid;
1625 info.group = group;
1626 /* sk->sk_err wants a positive error value */
1627 info.code = -code;
1628
1629 read_lock_irqsave(&nl_table_lock, flags);
1630
1631 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1632 ret += do_one_set_err(sk, &info);
1633
1634 read_unlock_irqrestore(&nl_table_lock, flags);
1635 return ret;
1636}
1637EXPORT_SYMBOL(netlink_set_err);
1638
1639/* must be called with netlink table grabbed */
1640static void netlink_update_socket_mc(struct netlink_sock *nlk,
1641 unsigned int group,
1642 int is_new)
1643{
1644 int old, new = !!is_new, subscriptions;
1645
1646 old = test_bit(group - 1, nlk->groups);
1647 subscriptions = nlk->subscriptions - old + new;
1648 __assign_bit(group - 1, nlk->groups, new);
1649 netlink_update_subscriptions(&nlk->sk, subscriptions);
1650 netlink_update_listeners(&nlk->sk);
1651}
1652
1653static int netlink_setsockopt(struct socket *sock, int level, int optname,
1654 sockptr_t optval, unsigned int optlen)
1655{
1656 struct sock *sk = sock->sk;
1657 struct netlink_sock *nlk = nlk_sk(sk);
1658 unsigned int val = 0;
1659 int nr = -1;
1660
1661 if (level != SOL_NETLINK)
1662 return -ENOPROTOOPT;
1663
1664 if (optlen >= sizeof(int) &&
1665 copy_from_sockptr(&val, optval, sizeof(val)))
1666 return -EFAULT;
1667
1668 switch (optname) {
1669 case NETLINK_PKTINFO:
1670 nr = NETLINK_F_RECV_PKTINFO;
1671 break;
1672 case NETLINK_ADD_MEMBERSHIP:
1673 case NETLINK_DROP_MEMBERSHIP: {
1674 int err;
1675
1676 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1677 return -EPERM;
1678 err = netlink_realloc_groups(sk);
1679 if (err)
1680 return err;
1681 if (!val || val - 1 >= nlk->ngroups)
1682 return -EINVAL;
1683 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1684 err = nlk->netlink_bind(sock_net(sk), val);
1685 if (err)
1686 return err;
1687 }
1688 netlink_table_grab();
1689 netlink_update_socket_mc(nlk, val,
1690 optname == NETLINK_ADD_MEMBERSHIP);
1691 netlink_table_ungrab();
1692 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1693 nlk->netlink_unbind(sock_net(sk), val);
1694
1695 break;
1696 }
1697 case NETLINK_BROADCAST_ERROR:
1698 nr = NETLINK_F_BROADCAST_SEND_ERROR;
1699 break;
1700 case NETLINK_NO_ENOBUFS:
1701 assign_bit(NETLINK_F_RECV_NO_ENOBUFS, &nlk->flags, val);
1702 if (val) {
1703 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1704 wake_up_interruptible(&nlk->wait);
1705 }
1706 break;
1707 case NETLINK_LISTEN_ALL_NSID:
1708 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1709 return -EPERM;
1710 nr = NETLINK_F_LISTEN_ALL_NSID;
1711 break;
1712 case NETLINK_CAP_ACK:
1713 nr = NETLINK_F_CAP_ACK;
1714 break;
1715 case NETLINK_EXT_ACK:
1716 nr = NETLINK_F_EXT_ACK;
1717 break;
1718 case NETLINK_GET_STRICT_CHK:
1719 nr = NETLINK_F_STRICT_CHK;
1720 break;
1721 default:
1722 return -ENOPROTOOPT;
1723 }
1724 if (nr >= 0)
1725 assign_bit(nr, &nlk->flags, val);
1726 return 0;
1727}
1728
1729static int netlink_getsockopt(struct socket *sock, int level, int optname,
1730 char __user *optval, int __user *optlen)
1731{
1732 struct sock *sk = sock->sk;
1733 struct netlink_sock *nlk = nlk_sk(sk);
1734 unsigned int flag;
1735 int len, val;
1736
1737 if (level != SOL_NETLINK)
1738 return -ENOPROTOOPT;
1739
1740 if (get_user(len, optlen))
1741 return -EFAULT;
1742 if (len < 0)
1743 return -EINVAL;
1744
1745 switch (optname) {
1746 case NETLINK_PKTINFO:
1747 flag = NETLINK_F_RECV_PKTINFO;
1748 break;
1749 case NETLINK_BROADCAST_ERROR:
1750 flag = NETLINK_F_BROADCAST_SEND_ERROR;
1751 break;
1752 case NETLINK_NO_ENOBUFS:
1753 flag = NETLINK_F_RECV_NO_ENOBUFS;
1754 break;
1755 case NETLINK_LIST_MEMBERSHIPS: {
1756 int pos, idx, shift, err = 0;
1757
1758 netlink_lock_table();
1759 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1760 if (len - pos < sizeof(u32))
1761 break;
1762
1763 idx = pos / sizeof(unsigned long);
1764 shift = (pos % sizeof(unsigned long)) * 8;
1765 if (put_user((u32)(nlk->groups[idx] >> shift),
1766 (u32 __user *)(optval + pos))) {
1767 err = -EFAULT;
1768 break;
1769 }
1770 }
1771 if (put_user(ALIGN(BITS_TO_BYTES(nlk->ngroups), sizeof(u32)), optlen))
1772 err = -EFAULT;
1773 netlink_unlock_table();
1774 return err;
1775 }
1776 case NETLINK_LISTEN_ALL_NSID:
1777 flag = NETLINK_F_LISTEN_ALL_NSID;
1778 break;
1779 case NETLINK_CAP_ACK:
1780 flag = NETLINK_F_CAP_ACK;
1781 break;
1782 case NETLINK_EXT_ACK:
1783 flag = NETLINK_F_EXT_ACK;
1784 break;
1785 case NETLINK_GET_STRICT_CHK:
1786 flag = NETLINK_F_STRICT_CHK;
1787 break;
1788 default:
1789 return -ENOPROTOOPT;
1790 }
1791
1792 if (len < sizeof(int))
1793 return -EINVAL;
1794
1795 len = sizeof(int);
1796 val = test_bit(flag, &nlk->flags);
1797
1798 if (put_user(len, optlen) ||
1799 copy_to_user(optval, &val, len))
1800 return -EFAULT;
1801
1802 return 0;
1803}
1804
1805static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1806{
1807 struct nl_pktinfo info;
1808
1809 info.group = NETLINK_CB(skb).dst_group;
1810 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1811}
1812
1813static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1814 struct sk_buff *skb)
1815{
1816 if (!NETLINK_CB(skb).nsid_is_set)
1817 return;
1818
1819 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1820 &NETLINK_CB(skb).nsid);
1821}
1822
1823static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1824{
1825 struct sock *sk = sock->sk;
1826 struct netlink_sock *nlk = nlk_sk(sk);
1827 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1828 u32 dst_portid;
1829 u32 dst_group;
1830 struct sk_buff *skb;
1831 int err;
1832 struct scm_cookie scm;
1833 u32 netlink_skb_flags = 0;
1834
1835 if (msg->msg_flags & MSG_OOB)
1836 return -EOPNOTSUPP;
1837
1838 if (len == 0) {
1839 pr_warn_once("Zero length message leads to an empty skb\n");
1840 return -ENODATA;
1841 }
1842
1843 err = scm_send(sock, msg, &scm, true);
1844 if (err < 0)
1845 return err;
1846
1847 if (msg->msg_namelen) {
1848 err = -EINVAL;
1849 if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1850 goto out;
1851 if (addr->nl_family != AF_NETLINK)
1852 goto out;
1853 dst_portid = addr->nl_pid;
1854 dst_group = ffs(addr->nl_groups);
1855 err = -EPERM;
1856 if ((dst_group || dst_portid) &&
1857 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1858 goto out;
1859 netlink_skb_flags |= NETLINK_SKB_DST;
1860 } else {
1861 /* Paired with WRITE_ONCE() in netlink_connect() */
1862 dst_portid = READ_ONCE(nlk->dst_portid);
1863 dst_group = READ_ONCE(nlk->dst_group);
1864 }
1865
1866 /* Paired with WRITE_ONCE() in netlink_insert() */
1867 if (!READ_ONCE(nlk->bound)) {
1868 err = netlink_autobind(sock);
1869 if (err)
1870 goto out;
1871 } else {
1872 /* Ensure nlk is hashed and visible. */
1873 smp_rmb();
1874 }
1875
1876 err = -EMSGSIZE;
1877 if (len > sk->sk_sndbuf - 32)
1878 goto out;
1879 err = -ENOBUFS;
1880 skb = netlink_alloc_large_skb(len, dst_group);
1881 if (skb == NULL)
1882 goto out;
1883
1884 NETLINK_CB(skb).portid = nlk->portid;
1885 NETLINK_CB(skb).dst_group = dst_group;
1886 NETLINK_CB(skb).creds = scm.creds;
1887 NETLINK_CB(skb).flags = netlink_skb_flags;
1888
1889 err = -EFAULT;
1890 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1891 kfree_skb(skb);
1892 goto out;
1893 }
1894
1895 err = security_netlink_send(sk, skb);
1896 if (err) {
1897 kfree_skb(skb);
1898 goto out;
1899 }
1900
1901 if (dst_group) {
1902 refcount_inc(&skb->users);
1903 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1904 }
1905 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags & MSG_DONTWAIT);
1906
1907out:
1908 scm_destroy(&scm);
1909 return err;
1910}
1911
1912static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1913 int flags)
1914{
1915 struct scm_cookie scm;
1916 struct sock *sk = sock->sk;
1917 struct netlink_sock *nlk = nlk_sk(sk);
1918 size_t copied, max_recvmsg_len;
1919 struct sk_buff *skb, *data_skb;
1920 int err, ret;
1921
1922 if (flags & MSG_OOB)
1923 return -EOPNOTSUPP;
1924
1925 copied = 0;
1926
1927 skb = skb_recv_datagram(sk, flags, &err);
1928 if (skb == NULL)
1929 goto out;
1930
1931 data_skb = skb;
1932
1933#ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1934 if (unlikely(skb_shinfo(skb)->frag_list)) {
1935 /*
1936 * If this skb has a frag_list, then here that means that we
1937 * will have to use the frag_list skb's data for compat tasks
1938 * and the regular skb's data for normal (non-compat) tasks.
1939 *
1940 * If we need to send the compat skb, assign it to the
1941 * 'data_skb' variable so that it will be used below for data
1942 * copying. We keep 'skb' for everything else, including
1943 * freeing both later.
1944 */
1945 if (flags & MSG_CMSG_COMPAT)
1946 data_skb = skb_shinfo(skb)->frag_list;
1947 }
1948#endif
1949
1950 /* Record the max length of recvmsg() calls for future allocations */
1951 max_recvmsg_len = max(READ_ONCE(nlk->max_recvmsg_len), len);
1952 max_recvmsg_len = min_t(size_t, max_recvmsg_len,
1953 SKB_WITH_OVERHEAD(32768));
1954 WRITE_ONCE(nlk->max_recvmsg_len, max_recvmsg_len);
1955
1956 copied = data_skb->len;
1957 if (len < copied) {
1958 msg->msg_flags |= MSG_TRUNC;
1959 copied = len;
1960 }
1961
1962 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1963
1964 if (msg->msg_name) {
1965 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1966 addr->nl_family = AF_NETLINK;
1967 addr->nl_pad = 0;
1968 addr->nl_pid = NETLINK_CB(skb).portid;
1969 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1970 msg->msg_namelen = sizeof(*addr);
1971 }
1972
1973 if (nlk_test_bit(RECV_PKTINFO, sk))
1974 netlink_cmsg_recv_pktinfo(msg, skb);
1975 if (nlk_test_bit(LISTEN_ALL_NSID, sk))
1976 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1977
1978 memset(&scm, 0, sizeof(scm));
1979 scm.creds = *NETLINK_CREDS(skb);
1980 if (flags & MSG_TRUNC)
1981 copied = data_skb->len;
1982
1983 skb_free_datagram(sk, skb);
1984
1985 if (READ_ONCE(nlk->cb_running) &&
1986 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1987 ret = netlink_dump(sk, false);
1988 if (ret) {
1989 WRITE_ONCE(sk->sk_err, -ret);
1990 sk_error_report(sk);
1991 }
1992 }
1993
1994 scm_recv(sock, msg, &scm, flags);
1995out:
1996 netlink_rcv_wake(sk);
1997 return err ? : copied;
1998}
1999
2000static void netlink_data_ready(struct sock *sk)
2001{
2002 BUG();
2003}
2004
2005/*
2006 * We export these functions to other modules. They provide a
2007 * complete set of kernel non-blocking support for message
2008 * queueing.
2009 */
2010
2011struct sock *
2012__netlink_kernel_create(struct net *net, int unit, struct module *module,
2013 struct netlink_kernel_cfg *cfg)
2014{
2015 struct socket *sock;
2016 struct sock *sk;
2017 struct netlink_sock *nlk;
2018 struct listeners *listeners = NULL;
2019 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2020 unsigned int groups;
2021
2022 BUG_ON(!nl_table);
2023
2024 if (unit < 0 || unit >= MAX_LINKS)
2025 return NULL;
2026
2027 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2028 return NULL;
2029
2030 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2031 goto out_sock_release_nosk;
2032
2033 sk = sock->sk;
2034
2035 if (!cfg || cfg->groups < 32)
2036 groups = 32;
2037 else
2038 groups = cfg->groups;
2039
2040 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2041 if (!listeners)
2042 goto out_sock_release;
2043
2044 sk->sk_data_ready = netlink_data_ready;
2045 if (cfg && cfg->input)
2046 nlk_sk(sk)->netlink_rcv = cfg->input;
2047
2048 if (netlink_insert(sk, 0))
2049 goto out_sock_release;
2050
2051 nlk = nlk_sk(sk);
2052 set_bit(NETLINK_F_KERNEL_SOCKET, &nlk->flags);
2053
2054 netlink_table_grab();
2055 if (!nl_table[unit].registered) {
2056 nl_table[unit].groups = groups;
2057 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2058 nl_table[unit].cb_mutex = cb_mutex;
2059 nl_table[unit].module = module;
2060 if (cfg) {
2061 nl_table[unit].bind = cfg->bind;
2062 nl_table[unit].unbind = cfg->unbind;
2063 nl_table[unit].release = cfg->release;
2064 nl_table[unit].flags = cfg->flags;
2065 }
2066 nl_table[unit].registered = 1;
2067 } else {
2068 kfree(listeners);
2069 nl_table[unit].registered++;
2070 }
2071 netlink_table_ungrab();
2072 return sk;
2073
2074out_sock_release:
2075 kfree(listeners);
2076 netlink_kernel_release(sk);
2077 return NULL;
2078
2079out_sock_release_nosk:
2080 sock_release(sock);
2081 return NULL;
2082}
2083EXPORT_SYMBOL(__netlink_kernel_create);
2084
2085void
2086netlink_kernel_release(struct sock *sk)
2087{
2088 if (sk == NULL || sk->sk_socket == NULL)
2089 return;
2090
2091 sock_release(sk->sk_socket);
2092}
2093EXPORT_SYMBOL(netlink_kernel_release);
2094
2095int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2096{
2097 struct listeners *new, *old;
2098 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2099
2100 if (groups < 32)
2101 groups = 32;
2102
2103 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2104 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2105 if (!new)
2106 return -ENOMEM;
2107 old = nl_deref_protected(tbl->listeners);
2108 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2109 rcu_assign_pointer(tbl->listeners, new);
2110
2111 kfree_rcu(old, rcu);
2112 }
2113 tbl->groups = groups;
2114
2115 return 0;
2116}
2117
2118/**
2119 * netlink_change_ngroups - change number of multicast groups
2120 *
2121 * This changes the number of multicast groups that are available
2122 * on a certain netlink family. Note that it is not possible to
2123 * change the number of groups to below 32. Also note that it does
2124 * not implicitly call netlink_clear_multicast_users() when the
2125 * number of groups is reduced.
2126 *
2127 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2128 * @groups: The new number of groups.
2129 */
2130int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2131{
2132 int err;
2133
2134 netlink_table_grab();
2135 err = __netlink_change_ngroups(sk, groups);
2136 netlink_table_ungrab();
2137
2138 return err;
2139}
2140
2141void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2142{
2143 struct sock *sk;
2144 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2145
2146 sk_for_each_bound(sk, &tbl->mc_list)
2147 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2148}
2149
2150struct nlmsghdr *
2151__nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2152{
2153 struct nlmsghdr *nlh;
2154 int size = nlmsg_msg_size(len);
2155
2156 nlh = skb_put(skb, NLMSG_ALIGN(size));
2157 nlh->nlmsg_type = type;
2158 nlh->nlmsg_len = size;
2159 nlh->nlmsg_flags = flags;
2160 nlh->nlmsg_pid = portid;
2161 nlh->nlmsg_seq = seq;
2162 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2163 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2164 return nlh;
2165}
2166EXPORT_SYMBOL(__nlmsg_put);
2167
2168/*
2169 * It looks a bit ugly.
2170 * It would be better to create kernel thread.
2171 */
2172
2173static int netlink_dump_done(struct netlink_sock *nlk, struct sk_buff *skb,
2174 struct netlink_callback *cb,
2175 struct netlink_ext_ack *extack)
2176{
2177 struct nlmsghdr *nlh;
2178
2179 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(nlk->dump_done_errno),
2180 NLM_F_MULTI | cb->answer_flags);
2181 if (WARN_ON(!nlh))
2182 return -ENOBUFS;
2183
2184 nl_dump_check_consistent(cb, nlh);
2185 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno, sizeof(nlk->dump_done_errno));
2186
2187 if (extack->_msg && test_bit(NETLINK_F_EXT_ACK, &nlk->flags)) {
2188 nlh->nlmsg_flags |= NLM_F_ACK_TLVS;
2189 if (!nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg))
2190 nlmsg_end(skb, nlh);
2191 }
2192
2193 return 0;
2194}
2195
2196static int netlink_dump(struct sock *sk, bool lock_taken)
2197{
2198 struct netlink_sock *nlk = nlk_sk(sk);
2199 struct netlink_ext_ack extack = {};
2200 struct netlink_callback *cb;
2201 struct sk_buff *skb = NULL;
2202 size_t max_recvmsg_len;
2203 struct module *module;
2204 int err = -ENOBUFS;
2205 int alloc_min_size;
2206 int alloc_size;
2207
2208 if (!lock_taken)
2209 mutex_lock(&nlk->nl_cb_mutex);
2210 if (!nlk->cb_running) {
2211 err = -EINVAL;
2212 goto errout_skb;
2213 }
2214
2215 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2216 goto errout_skb;
2217
2218 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2219 * required, but it makes sense to _attempt_ a 16K bytes allocation
2220 * to reduce number of system calls on dump operations, if user
2221 * ever provided a big enough buffer.
2222 */
2223 cb = &nlk->cb;
2224 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2225
2226 max_recvmsg_len = READ_ONCE(nlk->max_recvmsg_len);
2227 if (alloc_min_size < max_recvmsg_len) {
2228 alloc_size = max_recvmsg_len;
2229 skb = alloc_skb(alloc_size,
2230 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2231 __GFP_NOWARN | __GFP_NORETRY);
2232 }
2233 if (!skb) {
2234 alloc_size = alloc_min_size;
2235 skb = alloc_skb(alloc_size, GFP_KERNEL);
2236 }
2237 if (!skb)
2238 goto errout_skb;
2239
2240 /* Trim skb to allocated size. User is expected to provide buffer as
2241 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2242 * netlink_recvmsg())). dump will pack as many smaller messages as
2243 * could fit within the allocated skb. skb is typically allocated
2244 * with larger space than required (could be as much as near 2x the
2245 * requested size with align to next power of 2 approach). Allowing
2246 * dump to use the excess space makes it difficult for a user to have a
2247 * reasonable static buffer based on the expected largest dump of a
2248 * single netdev. The outcome is MSG_TRUNC error.
2249 */
2250 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2251
2252 /* Make sure malicious BPF programs can not read unitialized memory
2253 * from skb->head -> skb->data
2254 */
2255 skb_reset_network_header(skb);
2256 skb_reset_mac_header(skb);
2257
2258 netlink_skb_set_owner_r(skb, sk);
2259
2260 if (nlk->dump_done_errno > 0) {
2261 struct mutex *extra_mutex = nlk->dump_cb_mutex;
2262
2263 cb->extack = &extack;
2264
2265 if (cb->flags & RTNL_FLAG_DUMP_UNLOCKED)
2266 extra_mutex = NULL;
2267 if (extra_mutex)
2268 mutex_lock(extra_mutex);
2269 nlk->dump_done_errno = cb->dump(skb, cb);
2270 if (extra_mutex)
2271 mutex_unlock(extra_mutex);
2272
2273 /* EMSGSIZE plus something already in the skb means
2274 * that there's more to dump but current skb has filled up.
2275 * If the callback really wants to return EMSGSIZE to user space
2276 * it needs to do so again, on the next cb->dump() call,
2277 * without putting data in the skb.
2278 */
2279 if (nlk->dump_done_errno == -EMSGSIZE && skb->len)
2280 nlk->dump_done_errno = skb->len;
2281
2282 cb->extack = NULL;
2283 }
2284
2285 if (nlk->dump_done_errno > 0 ||
2286 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2287 mutex_unlock(&nlk->nl_cb_mutex);
2288
2289 if (sk_filter(sk, skb))
2290 kfree_skb(skb);
2291 else
2292 __netlink_sendskb(sk, skb);
2293 return 0;
2294 }
2295
2296 if (netlink_dump_done(nlk, skb, cb, &extack))
2297 goto errout_skb;
2298
2299#ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2300 /* frag_list skb's data is used for compat tasks
2301 * and the regular skb's data for normal (non-compat) tasks.
2302 * See netlink_recvmsg().
2303 */
2304 if (unlikely(skb_shinfo(skb)->frag_list)) {
2305 if (netlink_dump_done(nlk, skb_shinfo(skb)->frag_list, cb, &extack))
2306 goto errout_skb;
2307 }
2308#endif
2309
2310 if (sk_filter(sk, skb))
2311 kfree_skb(skb);
2312 else
2313 __netlink_sendskb(sk, skb);
2314
2315 if (cb->done)
2316 cb->done(cb);
2317
2318 WRITE_ONCE(nlk->cb_running, false);
2319 module = cb->module;
2320 skb = cb->skb;
2321 mutex_unlock(&nlk->nl_cb_mutex);
2322 module_put(module);
2323 consume_skb(skb);
2324 return 0;
2325
2326errout_skb:
2327 mutex_unlock(&nlk->nl_cb_mutex);
2328 kfree_skb(skb);
2329 return err;
2330}
2331
2332int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2333 const struct nlmsghdr *nlh,
2334 struct netlink_dump_control *control)
2335{
2336 struct netlink_callback *cb;
2337 struct netlink_sock *nlk;
2338 struct sock *sk;
2339 int ret;
2340
2341 refcount_inc(&skb->users);
2342
2343 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2344 if (sk == NULL) {
2345 ret = -ECONNREFUSED;
2346 goto error_free;
2347 }
2348
2349 nlk = nlk_sk(sk);
2350 mutex_lock(&nlk->nl_cb_mutex);
2351 /* A dump is in progress... */
2352 if (nlk->cb_running) {
2353 ret = -EBUSY;
2354 goto error_unlock;
2355 }
2356 /* add reference of module which cb->dump belongs to */
2357 if (!try_module_get(control->module)) {
2358 ret = -EPROTONOSUPPORT;
2359 goto error_unlock;
2360 }
2361
2362 cb = &nlk->cb;
2363 memset(cb, 0, sizeof(*cb));
2364 cb->dump = control->dump;
2365 cb->done = control->done;
2366 cb->nlh = nlh;
2367 cb->data = control->data;
2368 cb->module = control->module;
2369 cb->min_dump_alloc = control->min_dump_alloc;
2370 cb->flags = control->flags;
2371 cb->skb = skb;
2372
2373 cb->strict_check = nlk_test_bit(STRICT_CHK, NETLINK_CB(skb).sk);
2374
2375 if (control->start) {
2376 cb->extack = control->extack;
2377 ret = control->start(cb);
2378 cb->extack = NULL;
2379 if (ret)
2380 goto error_put;
2381 }
2382
2383 WRITE_ONCE(nlk->cb_running, true);
2384 nlk->dump_done_errno = INT_MAX;
2385
2386 ret = netlink_dump(sk, true);
2387
2388 sock_put(sk);
2389
2390 if (ret)
2391 return ret;
2392
2393 /* We successfully started a dump, by returning -EINTR we
2394 * signal not to send ACK even if it was requested.
2395 */
2396 return -EINTR;
2397
2398error_put:
2399 module_put(control->module);
2400error_unlock:
2401 sock_put(sk);
2402 mutex_unlock(&nlk->nl_cb_mutex);
2403error_free:
2404 kfree_skb(skb);
2405 return ret;
2406}
2407EXPORT_SYMBOL(__netlink_dump_start);
2408
2409static size_t
2410netlink_ack_tlv_len(struct netlink_sock *nlk, int err,
2411 const struct netlink_ext_ack *extack)
2412{
2413 size_t tlvlen;
2414
2415 if (!extack || !test_bit(NETLINK_F_EXT_ACK, &nlk->flags))
2416 return 0;
2417
2418 tlvlen = 0;
2419 if (extack->_msg)
2420 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2421 if (extack->cookie_len)
2422 tlvlen += nla_total_size(extack->cookie_len);
2423
2424 /* Following attributes are only reported as error (not warning) */
2425 if (!err)
2426 return tlvlen;
2427
2428 if (extack->bad_attr)
2429 tlvlen += nla_total_size(sizeof(u32));
2430 if (extack->policy)
2431 tlvlen += netlink_policy_dump_attr_size_estimate(extack->policy);
2432 if (extack->miss_type)
2433 tlvlen += nla_total_size(sizeof(u32));
2434 if (extack->miss_nest)
2435 tlvlen += nla_total_size(sizeof(u32));
2436
2437 return tlvlen;
2438}
2439
2440static void
2441netlink_ack_tlv_fill(struct sk_buff *in_skb, struct sk_buff *skb,
2442 struct nlmsghdr *nlh, int err,
2443 const struct netlink_ext_ack *extack)
2444{
2445 if (extack->_msg)
2446 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg));
2447 if (extack->cookie_len)
2448 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2449 extack->cookie_len, extack->cookie));
2450
2451 if (!err)
2452 return;
2453
2454 if (extack->bad_attr &&
2455 !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2456 (u8 *)extack->bad_attr >= in_skb->data + in_skb->len))
2457 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2458 (u8 *)extack->bad_attr - (u8 *)nlh));
2459 if (extack->policy)
2460 netlink_policy_dump_write_attr(skb, extack->policy,
2461 NLMSGERR_ATTR_POLICY);
2462 if (extack->miss_type)
2463 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_TYPE,
2464 extack->miss_type));
2465 if (extack->miss_nest &&
2466 !WARN_ON((u8 *)extack->miss_nest < in_skb->data ||
2467 (u8 *)extack->miss_nest > in_skb->data + in_skb->len))
2468 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_NEST,
2469 (u8 *)extack->miss_nest - (u8 *)nlh));
2470}
2471
2472void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2473 const struct netlink_ext_ack *extack)
2474{
2475 struct sk_buff *skb;
2476 struct nlmsghdr *rep;
2477 struct nlmsgerr *errmsg;
2478 size_t payload = sizeof(*errmsg);
2479 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2480 unsigned int flags = 0;
2481 size_t tlvlen;
2482
2483 /* Error messages get the original request appened, unless the user
2484 * requests to cap the error message, and get extra error data if
2485 * requested.
2486 */
2487 if (err && !test_bit(NETLINK_F_CAP_ACK, &nlk->flags))
2488 payload += nlmsg_len(nlh);
2489 else
2490 flags |= NLM_F_CAPPED;
2491
2492 tlvlen = netlink_ack_tlv_len(nlk, err, extack);
2493 if (tlvlen)
2494 flags |= NLM_F_ACK_TLVS;
2495
2496 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2497 if (!skb)
2498 goto err_skb;
2499
2500 rep = nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2501 NLMSG_ERROR, sizeof(*errmsg), flags);
2502 if (!rep)
2503 goto err_bad_put;
2504 errmsg = nlmsg_data(rep);
2505 errmsg->error = err;
2506 errmsg->msg = *nlh;
2507
2508 if (!(flags & NLM_F_CAPPED)) {
2509 if (!nlmsg_append(skb, nlmsg_len(nlh)))
2510 goto err_bad_put;
2511
2512 memcpy(nlmsg_data(&errmsg->msg), nlmsg_data(nlh),
2513 nlmsg_len(nlh));
2514 }
2515
2516 if (tlvlen)
2517 netlink_ack_tlv_fill(in_skb, skb, nlh, err, extack);
2518
2519 nlmsg_end(skb, rep);
2520
2521 nlmsg_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid);
2522
2523 return;
2524
2525err_bad_put:
2526 nlmsg_free(skb);
2527err_skb:
2528 WRITE_ONCE(NETLINK_CB(in_skb).sk->sk_err, ENOBUFS);
2529 sk_error_report(NETLINK_CB(in_skb).sk);
2530}
2531EXPORT_SYMBOL(netlink_ack);
2532
2533int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2534 struct nlmsghdr *,
2535 struct netlink_ext_ack *))
2536{
2537 struct netlink_ext_ack extack;
2538 struct nlmsghdr *nlh;
2539 int err;
2540
2541 while (skb->len >= nlmsg_total_size(0)) {
2542 int msglen;
2543
2544 memset(&extack, 0, sizeof(extack));
2545 nlh = nlmsg_hdr(skb);
2546 err = 0;
2547
2548 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2549 return 0;
2550
2551 /* Only requests are handled by the kernel */
2552 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2553 goto ack;
2554
2555 /* Skip control messages */
2556 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2557 goto ack;
2558
2559 err = cb(skb, nlh, &extack);
2560 if (err == -EINTR)
2561 goto skip;
2562
2563ack:
2564 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2565 netlink_ack(skb, nlh, err, &extack);
2566
2567skip:
2568 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2569 if (msglen > skb->len)
2570 msglen = skb->len;
2571 skb_pull(skb, msglen);
2572 }
2573
2574 return 0;
2575}
2576EXPORT_SYMBOL(netlink_rcv_skb);
2577
2578/**
2579 * nlmsg_notify - send a notification netlink message
2580 * @sk: netlink socket to use
2581 * @skb: notification message
2582 * @portid: destination netlink portid for reports or 0
2583 * @group: destination multicast group or 0
2584 * @report: 1 to report back, 0 to disable
2585 * @flags: allocation flags
2586 */
2587int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2588 unsigned int group, int report, gfp_t flags)
2589{
2590 int err = 0;
2591
2592 if (group) {
2593 int exclude_portid = 0;
2594
2595 if (report) {
2596 refcount_inc(&skb->users);
2597 exclude_portid = portid;
2598 }
2599
2600 /* errors reported via destination sk->sk_err, but propagate
2601 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2602 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2603 if (err == -ESRCH)
2604 err = 0;
2605 }
2606
2607 if (report) {
2608 int err2;
2609
2610 err2 = nlmsg_unicast(sk, skb, portid);
2611 if (!err)
2612 err = err2;
2613 }
2614
2615 return err;
2616}
2617EXPORT_SYMBOL(nlmsg_notify);
2618
2619#ifdef CONFIG_PROC_FS
2620struct nl_seq_iter {
2621 struct seq_net_private p;
2622 struct rhashtable_iter hti;
2623 int link;
2624};
2625
2626static void netlink_walk_start(struct nl_seq_iter *iter)
2627{
2628 rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti);
2629 rhashtable_walk_start(&iter->hti);
2630}
2631
2632static void netlink_walk_stop(struct nl_seq_iter *iter)
2633{
2634 rhashtable_walk_stop(&iter->hti);
2635 rhashtable_walk_exit(&iter->hti);
2636}
2637
2638static void *__netlink_seq_next(struct seq_file *seq)
2639{
2640 struct nl_seq_iter *iter = seq->private;
2641 struct netlink_sock *nlk;
2642
2643 do {
2644 for (;;) {
2645 nlk = rhashtable_walk_next(&iter->hti);
2646
2647 if (IS_ERR(nlk)) {
2648 if (PTR_ERR(nlk) == -EAGAIN)
2649 continue;
2650
2651 return nlk;
2652 }
2653
2654 if (nlk)
2655 break;
2656
2657 netlink_walk_stop(iter);
2658 if (++iter->link >= MAX_LINKS)
2659 return NULL;
2660
2661 netlink_walk_start(iter);
2662 }
2663 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2664
2665 return nlk;
2666}
2667
2668static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2669 __acquires(RCU)
2670{
2671 struct nl_seq_iter *iter = seq->private;
2672 void *obj = SEQ_START_TOKEN;
2673 loff_t pos;
2674
2675 iter->link = 0;
2676
2677 netlink_walk_start(iter);
2678
2679 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2680 obj = __netlink_seq_next(seq);
2681
2682 return obj;
2683}
2684
2685static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2686{
2687 ++*pos;
2688 return __netlink_seq_next(seq);
2689}
2690
2691static void netlink_native_seq_stop(struct seq_file *seq, void *v)
2692{
2693 struct nl_seq_iter *iter = seq->private;
2694
2695 if (iter->link >= MAX_LINKS)
2696 return;
2697
2698 netlink_walk_stop(iter);
2699}
2700
2701
2702static int netlink_native_seq_show(struct seq_file *seq, void *v)
2703{
2704 if (v == SEQ_START_TOKEN) {
2705 seq_puts(seq,
2706 "sk Eth Pid Groups "
2707 "Rmem Wmem Dump Locks Drops Inode\n");
2708 } else {
2709 struct sock *s = v;
2710 struct netlink_sock *nlk = nlk_sk(s);
2711
2712 seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n",
2713 s,
2714 s->sk_protocol,
2715 nlk->portid,
2716 nlk->groups ? (u32)nlk->groups[0] : 0,
2717 sk_rmem_alloc_get(s),
2718 sk_wmem_alloc_get(s),
2719 READ_ONCE(nlk->cb_running),
2720 refcount_read(&s->sk_refcnt),
2721 atomic_read(&s->sk_drops),
2722 sock_i_ino(s)
2723 );
2724
2725 }
2726 return 0;
2727}
2728
2729#ifdef CONFIG_BPF_SYSCALL
2730struct bpf_iter__netlink {
2731 __bpf_md_ptr(struct bpf_iter_meta *, meta);
2732 __bpf_md_ptr(struct netlink_sock *, sk);
2733};
2734
2735DEFINE_BPF_ITER_FUNC(netlink, struct bpf_iter_meta *meta, struct netlink_sock *sk)
2736
2737static int netlink_prog_seq_show(struct bpf_prog *prog,
2738 struct bpf_iter_meta *meta,
2739 void *v)
2740{
2741 struct bpf_iter__netlink ctx;
2742
2743 meta->seq_num--; /* skip SEQ_START_TOKEN */
2744 ctx.meta = meta;
2745 ctx.sk = nlk_sk((struct sock *)v);
2746 return bpf_iter_run_prog(prog, &ctx);
2747}
2748
2749static int netlink_seq_show(struct seq_file *seq, void *v)
2750{
2751 struct bpf_iter_meta meta;
2752 struct bpf_prog *prog;
2753
2754 meta.seq = seq;
2755 prog = bpf_iter_get_info(&meta, false);
2756 if (!prog)
2757 return netlink_native_seq_show(seq, v);
2758
2759 if (v != SEQ_START_TOKEN)
2760 return netlink_prog_seq_show(prog, &meta, v);
2761
2762 return 0;
2763}
2764
2765static void netlink_seq_stop(struct seq_file *seq, void *v)
2766{
2767 struct bpf_iter_meta meta;
2768 struct bpf_prog *prog;
2769
2770 if (!v) {
2771 meta.seq = seq;
2772 prog = bpf_iter_get_info(&meta, true);
2773 if (prog)
2774 (void)netlink_prog_seq_show(prog, &meta, v);
2775 }
2776
2777 netlink_native_seq_stop(seq, v);
2778}
2779#else
2780static int netlink_seq_show(struct seq_file *seq, void *v)
2781{
2782 return netlink_native_seq_show(seq, v);
2783}
2784
2785static void netlink_seq_stop(struct seq_file *seq, void *v)
2786{
2787 netlink_native_seq_stop(seq, v);
2788}
2789#endif
2790
2791static const struct seq_operations netlink_seq_ops = {
2792 .start = netlink_seq_start,
2793 .next = netlink_seq_next,
2794 .stop = netlink_seq_stop,
2795 .show = netlink_seq_show,
2796};
2797#endif
2798
2799int netlink_register_notifier(struct notifier_block *nb)
2800{
2801 return blocking_notifier_chain_register(&netlink_chain, nb);
2802}
2803EXPORT_SYMBOL(netlink_register_notifier);
2804
2805int netlink_unregister_notifier(struct notifier_block *nb)
2806{
2807 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2808}
2809EXPORT_SYMBOL(netlink_unregister_notifier);
2810
2811static const struct proto_ops netlink_ops = {
2812 .family = PF_NETLINK,
2813 .owner = THIS_MODULE,
2814 .release = netlink_release,
2815 .bind = netlink_bind,
2816 .connect = netlink_connect,
2817 .socketpair = sock_no_socketpair,
2818 .accept = sock_no_accept,
2819 .getname = netlink_getname,
2820 .poll = datagram_poll,
2821 .ioctl = netlink_ioctl,
2822 .listen = sock_no_listen,
2823 .shutdown = sock_no_shutdown,
2824 .setsockopt = netlink_setsockopt,
2825 .getsockopt = netlink_getsockopt,
2826 .sendmsg = netlink_sendmsg,
2827 .recvmsg = netlink_recvmsg,
2828 .mmap = sock_no_mmap,
2829};
2830
2831static const struct net_proto_family netlink_family_ops = {
2832 .family = PF_NETLINK,
2833 .create = netlink_create,
2834 .owner = THIS_MODULE, /* for consistency 8) */
2835};
2836
2837static int __net_init netlink_net_init(struct net *net)
2838{
2839#ifdef CONFIG_PROC_FS
2840 if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2841 sizeof(struct nl_seq_iter)))
2842 return -ENOMEM;
2843#endif
2844 return 0;
2845}
2846
2847static void __net_exit netlink_net_exit(struct net *net)
2848{
2849#ifdef CONFIG_PROC_FS
2850 remove_proc_entry("netlink", net->proc_net);
2851#endif
2852}
2853
2854static void __init netlink_add_usersock_entry(void)
2855{
2856 struct listeners *listeners;
2857 int groups = 32;
2858
2859 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2860 if (!listeners)
2861 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2862
2863 netlink_table_grab();
2864
2865 nl_table[NETLINK_USERSOCK].groups = groups;
2866 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2867 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2868 nl_table[NETLINK_USERSOCK].registered = 1;
2869 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2870
2871 netlink_table_ungrab();
2872}
2873
2874static struct pernet_operations __net_initdata netlink_net_ops = {
2875 .init = netlink_net_init,
2876 .exit = netlink_net_exit,
2877};
2878
2879static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2880{
2881 const struct netlink_sock *nlk = data;
2882 struct netlink_compare_arg arg;
2883
2884 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2885 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2886}
2887
2888static const struct rhashtable_params netlink_rhashtable_params = {
2889 .head_offset = offsetof(struct netlink_sock, node),
2890 .key_len = netlink_compare_arg_len,
2891 .obj_hashfn = netlink_hash,
2892 .obj_cmpfn = netlink_compare,
2893 .automatic_shrinking = true,
2894};
2895
2896#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2897BTF_ID_LIST(btf_netlink_sock_id)
2898BTF_ID(struct, netlink_sock)
2899
2900static const struct bpf_iter_seq_info netlink_seq_info = {
2901 .seq_ops = &netlink_seq_ops,
2902 .init_seq_private = bpf_iter_init_seq_net,
2903 .fini_seq_private = bpf_iter_fini_seq_net,
2904 .seq_priv_size = sizeof(struct nl_seq_iter),
2905};
2906
2907static struct bpf_iter_reg netlink_reg_info = {
2908 .target = "netlink",
2909 .ctx_arg_info_size = 1,
2910 .ctx_arg_info = {
2911 { offsetof(struct bpf_iter__netlink, sk),
2912 PTR_TO_BTF_ID_OR_NULL },
2913 },
2914 .seq_info = &netlink_seq_info,
2915};
2916
2917static int __init bpf_iter_register(void)
2918{
2919 netlink_reg_info.ctx_arg_info[0].btf_id = *btf_netlink_sock_id;
2920 return bpf_iter_reg_target(&netlink_reg_info);
2921}
2922#endif
2923
2924static int __init netlink_proto_init(void)
2925{
2926 int i;
2927 int err = proto_register(&netlink_proto, 0);
2928
2929 if (err != 0)
2930 goto out;
2931
2932#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2933 err = bpf_iter_register();
2934 if (err)
2935 goto out;
2936#endif
2937
2938 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof_field(struct sk_buff, cb));
2939
2940 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2941 if (!nl_table)
2942 goto panic;
2943
2944 for (i = 0; i < MAX_LINKS; i++) {
2945 if (rhashtable_init(&nl_table[i].hash,
2946 &netlink_rhashtable_params) < 0) {
2947 while (--i > 0)
2948 rhashtable_destroy(&nl_table[i].hash);
2949 kfree(nl_table);
2950 goto panic;
2951 }
2952 }
2953
2954 netlink_add_usersock_entry();
2955
2956 sock_register(&netlink_family_ops);
2957 register_pernet_subsys(&netlink_net_ops);
2958 register_pernet_subsys(&netlink_tap_net_ops);
2959 /* The netlink device handler may be needed early. */
2960 rtnetlink_init();
2961out:
2962 return err;
2963panic:
2964 panic("netlink_init: Cannot allocate nl_table\n");
2965}
2966
2967core_initcall(netlink_proto_init);