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