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/*
2 * NETLINK Kernel-user communication protocol.
3 *
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
6 * Patrick McHardy <kaber@trash.net>
7 *
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
12 *
13 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
14 * added netlink_proto_exit
15 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
16 * use nlk_sk, as sk->protinfo is on a diet 8)
17 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
18 * - inc module use count of module that owns
19 * the kernel socket in case userspace opens
20 * socket of same protocol
21 * - remove all module support, since netlink is
22 * mandatory if CONFIG_NET=y these days
23 */
24
25#include <linux/module.h>
26
27#include <linux/capability.h>
28#include <linux/kernel.h>
29#include <linux/init.h>
30#include <linux/signal.h>
31#include <linux/sched.h>
32#include <linux/errno.h>
33#include <linux/string.h>
34#include <linux/stat.h>
35#include <linux/socket.h>
36#include <linux/un.h>
37#include <linux/fcntl.h>
38#include <linux/termios.h>
39#include <linux/sockios.h>
40#include <linux/net.h>
41#include <linux/fs.h>
42#include <linux/slab.h>
43#include <asm/uaccess.h>
44#include <linux/skbuff.h>
45#include <linux/netdevice.h>
46#include <linux/rtnetlink.h>
47#include <linux/proc_fs.h>
48#include <linux/seq_file.h>
49#include <linux/notifier.h>
50#include <linux/security.h>
51#include <linux/jhash.h>
52#include <linux/jiffies.h>
53#include <linux/random.h>
54#include <linux/bitops.h>
55#include <linux/mm.h>
56#include <linux/types.h>
57#include <linux/audit.h>
58#include <linux/mutex.h>
59#include <linux/vmalloc.h>
60#include <linux/if_arp.h>
61#include <linux/rhashtable.h>
62#include <asm/cacheflush.h>
63#include <linux/hash.h>
64#include <linux/genetlink.h>
65
66#include <net/net_namespace.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[0];
76};
77
78/* state bits */
79#define NETLINK_S_CONGESTED 0x0
80
81/* flags */
82#define NETLINK_F_KERNEL_SOCKET 0x1
83#define NETLINK_F_RECV_PKTINFO 0x2
84#define NETLINK_F_BROADCAST_SEND_ERROR 0x4
85#define NETLINK_F_RECV_NO_ENOBUFS 0x8
86#define NETLINK_F_LISTEN_ALL_NSID 0x10
87#define NETLINK_F_CAP_ACK 0x20
88
89static inline int netlink_is_kernel(struct sock *sk)
90{
91 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
92}
93
94struct netlink_table *nl_table __read_mostly;
95EXPORT_SYMBOL_GPL(nl_table);
96
97static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
98
99static int netlink_dump(struct sock *sk);
100static void netlink_skb_destructor(struct sk_buff *skb);
101
102/* nl_table locking explained:
103 * Lookup and traversal are protected with an RCU read-side lock. Insertion
104 * and removal are protected with per bucket lock while using RCU list
105 * modification primitives and may run in parallel to RCU protected lookups.
106 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
107 * been acquired * either during or after the socket has been removed from
108 * the list and after an RCU grace period.
109 */
110DEFINE_RWLOCK(nl_table_lock);
111EXPORT_SYMBOL_GPL(nl_table_lock);
112static atomic_t nl_table_users = ATOMIC_INIT(0);
113
114#define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
115
116static ATOMIC_NOTIFIER_HEAD(netlink_chain);
117
118static DEFINE_SPINLOCK(netlink_tap_lock);
119static struct list_head netlink_tap_all __read_mostly;
120
121static const struct rhashtable_params netlink_rhashtable_params;
122
123static inline u32 netlink_group_mask(u32 group)
124{
125 return group ? 1 << (group - 1) : 0;
126}
127
128static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
129 gfp_t gfp_mask)
130{
131 unsigned int len = skb_end_offset(skb);
132 struct sk_buff *new;
133
134 new = alloc_skb(len, gfp_mask);
135 if (new == NULL)
136 return NULL;
137
138 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
139 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
140 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
141
142 memcpy(skb_put(new, len), skb->data, len);
143 return new;
144}
145
146int netlink_add_tap(struct netlink_tap *nt)
147{
148 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
149 return -EINVAL;
150
151 spin_lock(&netlink_tap_lock);
152 list_add_rcu(&nt->list, &netlink_tap_all);
153 spin_unlock(&netlink_tap_lock);
154
155 __module_get(nt->module);
156
157 return 0;
158}
159EXPORT_SYMBOL_GPL(netlink_add_tap);
160
161static int __netlink_remove_tap(struct netlink_tap *nt)
162{
163 bool found = false;
164 struct netlink_tap *tmp;
165
166 spin_lock(&netlink_tap_lock);
167
168 list_for_each_entry(tmp, &netlink_tap_all, list) {
169 if (nt == tmp) {
170 list_del_rcu(&nt->list);
171 found = true;
172 goto out;
173 }
174 }
175
176 pr_warn("__netlink_remove_tap: %p not found\n", nt);
177out:
178 spin_unlock(&netlink_tap_lock);
179
180 if (found)
181 module_put(nt->module);
182
183 return found ? 0 : -ENODEV;
184}
185
186int netlink_remove_tap(struct netlink_tap *nt)
187{
188 int ret;
189
190 ret = __netlink_remove_tap(nt);
191 synchronize_net();
192
193 return ret;
194}
195EXPORT_SYMBOL_GPL(netlink_remove_tap);
196
197static bool netlink_filter_tap(const struct sk_buff *skb)
198{
199 struct sock *sk = skb->sk;
200
201 /* We take the more conservative approach and
202 * whitelist socket protocols that may pass.
203 */
204 switch (sk->sk_protocol) {
205 case NETLINK_ROUTE:
206 case NETLINK_USERSOCK:
207 case NETLINK_SOCK_DIAG:
208 case NETLINK_NFLOG:
209 case NETLINK_XFRM:
210 case NETLINK_FIB_LOOKUP:
211 case NETLINK_NETFILTER:
212 case NETLINK_GENERIC:
213 return true;
214 }
215
216 return false;
217}
218
219static int __netlink_deliver_tap_skb(struct sk_buff *skb,
220 struct net_device *dev)
221{
222 struct sk_buff *nskb;
223 struct sock *sk = skb->sk;
224 int ret = -ENOMEM;
225
226 dev_hold(dev);
227
228 if (is_vmalloc_addr(skb->head))
229 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
230 else
231 nskb = skb_clone(skb, GFP_ATOMIC);
232 if (nskb) {
233 nskb->dev = dev;
234 nskb->protocol = htons((u16) sk->sk_protocol);
235 nskb->pkt_type = netlink_is_kernel(sk) ?
236 PACKET_KERNEL : PACKET_USER;
237 skb_reset_network_header(nskb);
238 ret = dev_queue_xmit(nskb);
239 if (unlikely(ret > 0))
240 ret = net_xmit_errno(ret);
241 }
242
243 dev_put(dev);
244 return ret;
245}
246
247static void __netlink_deliver_tap(struct sk_buff *skb)
248{
249 int ret;
250 struct netlink_tap *tmp;
251
252 if (!netlink_filter_tap(skb))
253 return;
254
255 list_for_each_entry_rcu(tmp, &netlink_tap_all, list) {
256 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
257 if (unlikely(ret))
258 break;
259 }
260}
261
262static void netlink_deliver_tap(struct sk_buff *skb)
263{
264 rcu_read_lock();
265
266 if (unlikely(!list_empty(&netlink_tap_all)))
267 __netlink_deliver_tap(skb);
268
269 rcu_read_unlock();
270}
271
272static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
273 struct sk_buff *skb)
274{
275 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
276 netlink_deliver_tap(skb);
277}
278
279static void netlink_overrun(struct sock *sk)
280{
281 struct netlink_sock *nlk = nlk_sk(sk);
282
283 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
284 if (!test_and_set_bit(NETLINK_S_CONGESTED,
285 &nlk_sk(sk)->state)) {
286 sk->sk_err = ENOBUFS;
287 sk->sk_error_report(sk);
288 }
289 }
290 atomic_inc(&sk->sk_drops);
291}
292
293static void netlink_rcv_wake(struct sock *sk)
294{
295 struct netlink_sock *nlk = nlk_sk(sk);
296
297 if (skb_queue_empty(&sk->sk_receive_queue))
298 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
299 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
300 wake_up_interruptible(&nlk->wait);
301}
302
303static void netlink_skb_destructor(struct sk_buff *skb)
304{
305 if (is_vmalloc_addr(skb->head)) {
306 if (!skb->cloned ||
307 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
308 vfree(skb->head);
309
310 skb->head = NULL;
311 }
312 if (skb->sk != NULL)
313 sock_rfree(skb);
314}
315
316static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
317{
318 WARN_ON(skb->sk != NULL);
319 skb->sk = sk;
320 skb->destructor = netlink_skb_destructor;
321 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
322 sk_mem_charge(sk, skb->truesize);
323}
324
325static void netlink_sock_destruct(struct sock *sk)
326{
327 struct netlink_sock *nlk = nlk_sk(sk);
328
329 if (nlk->cb_running) {
330 if (nlk->cb.done)
331 nlk->cb.done(&nlk->cb);
332
333 module_put(nlk->cb.module);
334 kfree_skb(nlk->cb.skb);
335 }
336
337 skb_queue_purge(&sk->sk_receive_queue);
338
339 if (!sock_flag(sk, SOCK_DEAD)) {
340 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
341 return;
342 }
343
344 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
345 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
346 WARN_ON(nlk_sk(sk)->groups);
347}
348
349/* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
350 * SMP. Look, when several writers sleep and reader wakes them up, all but one
351 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
352 * this, _but_ remember, it adds useless work on UP machines.
353 */
354
355void netlink_table_grab(void)
356 __acquires(nl_table_lock)
357{
358 might_sleep();
359
360 write_lock_irq(&nl_table_lock);
361
362 if (atomic_read(&nl_table_users)) {
363 DECLARE_WAITQUEUE(wait, current);
364
365 add_wait_queue_exclusive(&nl_table_wait, &wait);
366 for (;;) {
367 set_current_state(TASK_UNINTERRUPTIBLE);
368 if (atomic_read(&nl_table_users) == 0)
369 break;
370 write_unlock_irq(&nl_table_lock);
371 schedule();
372 write_lock_irq(&nl_table_lock);
373 }
374
375 __set_current_state(TASK_RUNNING);
376 remove_wait_queue(&nl_table_wait, &wait);
377 }
378}
379
380void netlink_table_ungrab(void)
381 __releases(nl_table_lock)
382{
383 write_unlock_irq(&nl_table_lock);
384 wake_up(&nl_table_wait);
385}
386
387static inline void
388netlink_lock_table(void)
389{
390 /* read_lock() synchronizes us to netlink_table_grab */
391
392 read_lock(&nl_table_lock);
393 atomic_inc(&nl_table_users);
394 read_unlock(&nl_table_lock);
395}
396
397static inline void
398netlink_unlock_table(void)
399{
400 if (atomic_dec_and_test(&nl_table_users))
401 wake_up(&nl_table_wait);
402}
403
404struct netlink_compare_arg
405{
406 possible_net_t pnet;
407 u32 portid;
408};
409
410/* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
411#define netlink_compare_arg_len \
412 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
413
414static inline int netlink_compare(struct rhashtable_compare_arg *arg,
415 const void *ptr)
416{
417 const struct netlink_compare_arg *x = arg->key;
418 const struct netlink_sock *nlk = ptr;
419
420 return nlk->portid != x->portid ||
421 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
422}
423
424static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
425 struct net *net, u32 portid)
426{
427 memset(arg, 0, sizeof(*arg));
428 write_pnet(&arg->pnet, net);
429 arg->portid = portid;
430}
431
432static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
433 struct net *net)
434{
435 struct netlink_compare_arg arg;
436
437 netlink_compare_arg_init(&arg, net, portid);
438 return rhashtable_lookup_fast(&table->hash, &arg,
439 netlink_rhashtable_params);
440}
441
442static int __netlink_insert(struct netlink_table *table, struct sock *sk)
443{
444 struct netlink_compare_arg arg;
445
446 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
447 return rhashtable_lookup_insert_key(&table->hash, &arg,
448 &nlk_sk(sk)->node,
449 netlink_rhashtable_params);
450}
451
452static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
453{
454 struct netlink_table *table = &nl_table[protocol];
455 struct sock *sk;
456
457 rcu_read_lock();
458 sk = __netlink_lookup(table, portid, net);
459 if (sk)
460 sock_hold(sk);
461 rcu_read_unlock();
462
463 return sk;
464}
465
466static const struct proto_ops netlink_ops;
467
468static void
469netlink_update_listeners(struct sock *sk)
470{
471 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
472 unsigned long mask;
473 unsigned int i;
474 struct listeners *listeners;
475
476 listeners = nl_deref_protected(tbl->listeners);
477 if (!listeners)
478 return;
479
480 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
481 mask = 0;
482 sk_for_each_bound(sk, &tbl->mc_list) {
483 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
484 mask |= nlk_sk(sk)->groups[i];
485 }
486 listeners->masks[i] = mask;
487 }
488 /* this function is only called with the netlink table "grabbed", which
489 * makes sure updates are visible before bind or setsockopt return. */
490}
491
492static int netlink_insert(struct sock *sk, u32 portid)
493{
494 struct netlink_table *table = &nl_table[sk->sk_protocol];
495 int err;
496
497 lock_sock(sk);
498
499 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
500 if (nlk_sk(sk)->bound)
501 goto err;
502
503 err = -ENOMEM;
504 if (BITS_PER_LONG > 32 &&
505 unlikely(atomic_read(&table->hash.nelems) >= UINT_MAX))
506 goto err;
507
508 nlk_sk(sk)->portid = portid;
509 sock_hold(sk);
510
511 err = __netlink_insert(table, sk);
512 if (err) {
513 /* In case the hashtable backend returns with -EBUSY
514 * from here, it must not escape to the caller.
515 */
516 if (unlikely(err == -EBUSY))
517 err = -EOVERFLOW;
518 if (err == -EEXIST)
519 err = -EADDRINUSE;
520 sock_put(sk);
521 goto err;
522 }
523
524 /* We need to ensure that the socket is hashed and visible. */
525 smp_wmb();
526 nlk_sk(sk)->bound = portid;
527
528err:
529 release_sock(sk);
530 return err;
531}
532
533static void netlink_remove(struct sock *sk)
534{
535 struct netlink_table *table;
536
537 table = &nl_table[sk->sk_protocol];
538 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
539 netlink_rhashtable_params)) {
540 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
541 __sock_put(sk);
542 }
543
544 netlink_table_grab();
545 if (nlk_sk(sk)->subscriptions) {
546 __sk_del_bind_node(sk);
547 netlink_update_listeners(sk);
548 }
549 if (sk->sk_protocol == NETLINK_GENERIC)
550 atomic_inc(&genl_sk_destructing_cnt);
551 netlink_table_ungrab();
552}
553
554static struct proto netlink_proto = {
555 .name = "NETLINK",
556 .owner = THIS_MODULE,
557 .obj_size = sizeof(struct netlink_sock),
558};
559
560static int __netlink_create(struct net *net, struct socket *sock,
561 struct mutex *cb_mutex, int protocol,
562 int kern)
563{
564 struct sock *sk;
565 struct netlink_sock *nlk;
566
567 sock->ops = &netlink_ops;
568
569 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
570 if (!sk)
571 return -ENOMEM;
572
573 sock_init_data(sock, sk);
574
575 nlk = nlk_sk(sk);
576 if (cb_mutex) {
577 nlk->cb_mutex = cb_mutex;
578 } else {
579 nlk->cb_mutex = &nlk->cb_def_mutex;
580 mutex_init(nlk->cb_mutex);
581 }
582 init_waitqueue_head(&nlk->wait);
583
584 sk->sk_destruct = netlink_sock_destruct;
585 sk->sk_protocol = protocol;
586 return 0;
587}
588
589static int netlink_create(struct net *net, struct socket *sock, int protocol,
590 int kern)
591{
592 struct module *module = NULL;
593 struct mutex *cb_mutex;
594 struct netlink_sock *nlk;
595 int (*bind)(struct net *net, int group);
596 void (*unbind)(struct net *net, int group);
597 int err = 0;
598
599 sock->state = SS_UNCONNECTED;
600
601 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
602 return -ESOCKTNOSUPPORT;
603
604 if (protocol < 0 || protocol >= MAX_LINKS)
605 return -EPROTONOSUPPORT;
606
607 netlink_lock_table();
608#ifdef CONFIG_MODULES
609 if (!nl_table[protocol].registered) {
610 netlink_unlock_table();
611 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
612 netlink_lock_table();
613 }
614#endif
615 if (nl_table[protocol].registered &&
616 try_module_get(nl_table[protocol].module))
617 module = nl_table[protocol].module;
618 else
619 err = -EPROTONOSUPPORT;
620 cb_mutex = nl_table[protocol].cb_mutex;
621 bind = nl_table[protocol].bind;
622 unbind = nl_table[protocol].unbind;
623 netlink_unlock_table();
624
625 if (err < 0)
626 goto out;
627
628 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
629 if (err < 0)
630 goto out_module;
631
632 local_bh_disable();
633 sock_prot_inuse_add(net, &netlink_proto, 1);
634 local_bh_enable();
635
636 nlk = nlk_sk(sock->sk);
637 nlk->module = module;
638 nlk->netlink_bind = bind;
639 nlk->netlink_unbind = unbind;
640out:
641 return err;
642
643out_module:
644 module_put(module);
645 goto out;
646}
647
648static void deferred_put_nlk_sk(struct rcu_head *head)
649{
650 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
651
652 sock_put(&nlk->sk);
653}
654
655static int netlink_release(struct socket *sock)
656{
657 struct sock *sk = sock->sk;
658 struct netlink_sock *nlk;
659
660 if (!sk)
661 return 0;
662
663 netlink_remove(sk);
664 sock_orphan(sk);
665 nlk = nlk_sk(sk);
666
667 /*
668 * OK. Socket is unlinked, any packets that arrive now
669 * will be purged.
670 */
671
672 /* must not acquire netlink_table_lock in any way again before unbind
673 * and notifying genetlink is done as otherwise it might deadlock
674 */
675 if (nlk->netlink_unbind) {
676 int i;
677
678 for (i = 0; i < nlk->ngroups; i++)
679 if (test_bit(i, nlk->groups))
680 nlk->netlink_unbind(sock_net(sk), i + 1);
681 }
682 if (sk->sk_protocol == NETLINK_GENERIC &&
683 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
684 wake_up(&genl_sk_destructing_waitq);
685
686 sock->sk = NULL;
687 wake_up_interruptible_all(&nlk->wait);
688
689 skb_queue_purge(&sk->sk_write_queue);
690
691 if (nlk->portid && nlk->bound) {
692 struct netlink_notify n = {
693 .net = sock_net(sk),
694 .protocol = sk->sk_protocol,
695 .portid = nlk->portid,
696 };
697 atomic_notifier_call_chain(&netlink_chain,
698 NETLINK_URELEASE, &n);
699 }
700
701 module_put(nlk->module);
702
703 if (netlink_is_kernel(sk)) {
704 netlink_table_grab();
705 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
706 if (--nl_table[sk->sk_protocol].registered == 0) {
707 struct listeners *old;
708
709 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
710 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
711 kfree_rcu(old, rcu);
712 nl_table[sk->sk_protocol].module = NULL;
713 nl_table[sk->sk_protocol].bind = NULL;
714 nl_table[sk->sk_protocol].unbind = NULL;
715 nl_table[sk->sk_protocol].flags = 0;
716 nl_table[sk->sk_protocol].registered = 0;
717 }
718 netlink_table_ungrab();
719 }
720
721 kfree(nlk->groups);
722 nlk->groups = NULL;
723
724 local_bh_disable();
725 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
726 local_bh_enable();
727 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
728 return 0;
729}
730
731static int netlink_autobind(struct socket *sock)
732{
733 struct sock *sk = sock->sk;
734 struct net *net = sock_net(sk);
735 struct netlink_table *table = &nl_table[sk->sk_protocol];
736 s32 portid = task_tgid_vnr(current);
737 int err;
738 s32 rover = -4096;
739 bool ok;
740
741retry:
742 cond_resched();
743 rcu_read_lock();
744 ok = !__netlink_lookup(table, portid, net);
745 rcu_read_unlock();
746 if (!ok) {
747 /* Bind collision, search negative portid values. */
748 if (rover == -4096)
749 /* rover will be in range [S32_MIN, -4097] */
750 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
751 else if (rover >= -4096)
752 rover = -4097;
753 portid = rover--;
754 goto retry;
755 }
756
757 err = netlink_insert(sk, portid);
758 if (err == -EADDRINUSE)
759 goto retry;
760
761 /* If 2 threads race to autobind, that is fine. */
762 if (err == -EBUSY)
763 err = 0;
764
765 return err;
766}
767
768/**
769 * __netlink_ns_capable - General netlink message capability test
770 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
771 * @user_ns: The user namespace of the capability to use
772 * @cap: The capability to use
773 *
774 * Test to see if the opener of the socket we received the message
775 * from had when the netlink socket was created and the sender of the
776 * message has has the capability @cap in the user namespace @user_ns.
777 */
778bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
779 struct user_namespace *user_ns, int cap)
780{
781 return ((nsp->flags & NETLINK_SKB_DST) ||
782 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
783 ns_capable(user_ns, cap);
784}
785EXPORT_SYMBOL(__netlink_ns_capable);
786
787/**
788 * netlink_ns_capable - General netlink message capability test
789 * @skb: socket buffer holding a netlink command from userspace
790 * @user_ns: The user namespace of the capability to use
791 * @cap: The capability to use
792 *
793 * Test to see if the opener of the socket we received the message
794 * from had when the netlink socket was created and the sender of the
795 * message has has the capability @cap in the user namespace @user_ns.
796 */
797bool netlink_ns_capable(const struct sk_buff *skb,
798 struct user_namespace *user_ns, int cap)
799{
800 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
801}
802EXPORT_SYMBOL(netlink_ns_capable);
803
804/**
805 * netlink_capable - Netlink global message capability test
806 * @skb: socket buffer holding a netlink command from userspace
807 * @cap: The capability to use
808 *
809 * Test to see if the opener of the socket we received the message
810 * from had when the netlink socket was created and the sender of the
811 * message has has the capability @cap in all user namespaces.
812 */
813bool netlink_capable(const struct sk_buff *skb, int cap)
814{
815 return netlink_ns_capable(skb, &init_user_ns, cap);
816}
817EXPORT_SYMBOL(netlink_capable);
818
819/**
820 * netlink_net_capable - Netlink network namespace message capability test
821 * @skb: socket buffer holding a netlink command from userspace
822 * @cap: The capability to use
823 *
824 * Test to see if the opener of the socket we received the message
825 * from had when the netlink socket was created and the sender of the
826 * message has has the capability @cap over the network namespace of
827 * the socket we received the message from.
828 */
829bool netlink_net_capable(const struct sk_buff *skb, int cap)
830{
831 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
832}
833EXPORT_SYMBOL(netlink_net_capable);
834
835static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
836{
837 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
838 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
839}
840
841static void
842netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
843{
844 struct netlink_sock *nlk = nlk_sk(sk);
845
846 if (nlk->subscriptions && !subscriptions)
847 __sk_del_bind_node(sk);
848 else if (!nlk->subscriptions && subscriptions)
849 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
850 nlk->subscriptions = subscriptions;
851}
852
853static int netlink_realloc_groups(struct sock *sk)
854{
855 struct netlink_sock *nlk = nlk_sk(sk);
856 unsigned int groups;
857 unsigned long *new_groups;
858 int err = 0;
859
860 netlink_table_grab();
861
862 groups = nl_table[sk->sk_protocol].groups;
863 if (!nl_table[sk->sk_protocol].registered) {
864 err = -ENOENT;
865 goto out_unlock;
866 }
867
868 if (nlk->ngroups >= groups)
869 goto out_unlock;
870
871 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
872 if (new_groups == NULL) {
873 err = -ENOMEM;
874 goto out_unlock;
875 }
876 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
877 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
878
879 nlk->groups = new_groups;
880 nlk->ngroups = groups;
881 out_unlock:
882 netlink_table_ungrab();
883 return err;
884}
885
886static void netlink_undo_bind(int group, long unsigned int groups,
887 struct sock *sk)
888{
889 struct netlink_sock *nlk = nlk_sk(sk);
890 int undo;
891
892 if (!nlk->netlink_unbind)
893 return;
894
895 for (undo = 0; undo < group; undo++)
896 if (test_bit(undo, &groups))
897 nlk->netlink_unbind(sock_net(sk), undo + 1);
898}
899
900static int netlink_bind(struct socket *sock, struct sockaddr *addr,
901 int addr_len)
902{
903 struct sock *sk = sock->sk;
904 struct net *net = sock_net(sk);
905 struct netlink_sock *nlk = nlk_sk(sk);
906 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
907 int err;
908 long unsigned int groups = nladdr->nl_groups;
909 bool bound;
910
911 if (addr_len < sizeof(struct sockaddr_nl))
912 return -EINVAL;
913
914 if (nladdr->nl_family != AF_NETLINK)
915 return -EINVAL;
916
917 /* Only superuser is allowed to listen multicasts */
918 if (groups) {
919 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
920 return -EPERM;
921 err = netlink_realloc_groups(sk);
922 if (err)
923 return err;
924 }
925
926 bound = nlk->bound;
927 if (bound) {
928 /* Ensure nlk->portid is up-to-date. */
929 smp_rmb();
930
931 if (nladdr->nl_pid != nlk->portid)
932 return -EINVAL;
933 }
934
935 if (nlk->netlink_bind && groups) {
936 int group;
937
938 for (group = 0; group < nlk->ngroups; group++) {
939 if (!test_bit(group, &groups))
940 continue;
941 err = nlk->netlink_bind(net, group + 1);
942 if (!err)
943 continue;
944 netlink_undo_bind(group, groups, sk);
945 return err;
946 }
947 }
948
949 /* No need for barriers here as we return to user-space without
950 * using any of the bound attributes.
951 */
952 if (!bound) {
953 err = nladdr->nl_pid ?
954 netlink_insert(sk, nladdr->nl_pid) :
955 netlink_autobind(sock);
956 if (err) {
957 netlink_undo_bind(nlk->ngroups, groups, sk);
958 return err;
959 }
960 }
961
962 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
963 return 0;
964
965 netlink_table_grab();
966 netlink_update_subscriptions(sk, nlk->subscriptions +
967 hweight32(groups) -
968 hweight32(nlk->groups[0]));
969 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
970 netlink_update_listeners(sk);
971 netlink_table_ungrab();
972
973 return 0;
974}
975
976static int netlink_connect(struct socket *sock, struct sockaddr *addr,
977 int alen, int flags)
978{
979 int err = 0;
980 struct sock *sk = sock->sk;
981 struct netlink_sock *nlk = nlk_sk(sk);
982 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
983
984 if (alen < sizeof(addr->sa_family))
985 return -EINVAL;
986
987 if (addr->sa_family == AF_UNSPEC) {
988 sk->sk_state = NETLINK_UNCONNECTED;
989 nlk->dst_portid = 0;
990 nlk->dst_group = 0;
991 return 0;
992 }
993 if (addr->sa_family != AF_NETLINK)
994 return -EINVAL;
995
996 if ((nladdr->nl_groups || nladdr->nl_pid) &&
997 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
998 return -EPERM;
999
1000 /* No need for barriers here as we return to user-space without
1001 * using any of the bound attributes.
1002 */
1003 if (!nlk->bound)
1004 err = netlink_autobind(sock);
1005
1006 if (err == 0) {
1007 sk->sk_state = NETLINK_CONNECTED;
1008 nlk->dst_portid = nladdr->nl_pid;
1009 nlk->dst_group = ffs(nladdr->nl_groups);
1010 }
1011
1012 return err;
1013}
1014
1015static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1016 int *addr_len, int peer)
1017{
1018 struct sock *sk = sock->sk;
1019 struct netlink_sock *nlk = nlk_sk(sk);
1020 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1021
1022 nladdr->nl_family = AF_NETLINK;
1023 nladdr->nl_pad = 0;
1024 *addr_len = sizeof(*nladdr);
1025
1026 if (peer) {
1027 nladdr->nl_pid = nlk->dst_portid;
1028 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1029 } else {
1030 nladdr->nl_pid = nlk->portid;
1031 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1032 }
1033 return 0;
1034}
1035
1036static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1037 unsigned long arg)
1038{
1039 /* try to hand this ioctl down to the NIC drivers.
1040 */
1041 return -ENOIOCTLCMD;
1042}
1043
1044static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1045{
1046 struct sock *sock;
1047 struct netlink_sock *nlk;
1048
1049 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1050 if (!sock)
1051 return ERR_PTR(-ECONNREFUSED);
1052
1053 /* Don't bother queuing skb if kernel socket has no input function */
1054 nlk = nlk_sk(sock);
1055 if (sock->sk_state == NETLINK_CONNECTED &&
1056 nlk->dst_portid != nlk_sk(ssk)->portid) {
1057 sock_put(sock);
1058 return ERR_PTR(-ECONNREFUSED);
1059 }
1060 return sock;
1061}
1062
1063struct sock *netlink_getsockbyfilp(struct file *filp)
1064{
1065 struct inode *inode = file_inode(filp);
1066 struct sock *sock;
1067
1068 if (!S_ISSOCK(inode->i_mode))
1069 return ERR_PTR(-ENOTSOCK);
1070
1071 sock = SOCKET_I(inode)->sk;
1072 if (sock->sk_family != AF_NETLINK)
1073 return ERR_PTR(-EINVAL);
1074
1075 sock_hold(sock);
1076 return sock;
1077}
1078
1079static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1080 int broadcast)
1081{
1082 struct sk_buff *skb;
1083 void *data;
1084
1085 if (size <= NLMSG_GOODSIZE || broadcast)
1086 return alloc_skb(size, GFP_KERNEL);
1087
1088 size = SKB_DATA_ALIGN(size) +
1089 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1090
1091 data = vmalloc(size);
1092 if (data == NULL)
1093 return NULL;
1094
1095 skb = __build_skb(data, size);
1096 if (skb == NULL)
1097 vfree(data);
1098 else
1099 skb->destructor = netlink_skb_destructor;
1100
1101 return skb;
1102}
1103
1104/*
1105 * Attach a skb to a netlink socket.
1106 * The caller must hold a reference to the destination socket. On error, the
1107 * reference is dropped. The skb is not send to the destination, just all
1108 * all error checks are performed and memory in the queue is reserved.
1109 * Return values:
1110 * < 0: error. skb freed, reference to sock dropped.
1111 * 0: continue
1112 * 1: repeat lookup - reference dropped while waiting for socket memory.
1113 */
1114int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1115 long *timeo, struct sock *ssk)
1116{
1117 struct netlink_sock *nlk;
1118
1119 nlk = nlk_sk(sk);
1120
1121 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1122 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1123 DECLARE_WAITQUEUE(wait, current);
1124 if (!*timeo) {
1125 if (!ssk || netlink_is_kernel(ssk))
1126 netlink_overrun(sk);
1127 sock_put(sk);
1128 kfree_skb(skb);
1129 return -EAGAIN;
1130 }
1131
1132 __set_current_state(TASK_INTERRUPTIBLE);
1133 add_wait_queue(&nlk->wait, &wait);
1134
1135 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1136 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1137 !sock_flag(sk, SOCK_DEAD))
1138 *timeo = schedule_timeout(*timeo);
1139
1140 __set_current_state(TASK_RUNNING);
1141 remove_wait_queue(&nlk->wait, &wait);
1142 sock_put(sk);
1143
1144 if (signal_pending(current)) {
1145 kfree_skb(skb);
1146 return sock_intr_errno(*timeo);
1147 }
1148 return 1;
1149 }
1150 netlink_skb_set_owner_r(skb, sk);
1151 return 0;
1152}
1153
1154static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1155{
1156 int len = skb->len;
1157
1158 netlink_deliver_tap(skb);
1159
1160 skb_queue_tail(&sk->sk_receive_queue, skb);
1161 sk->sk_data_ready(sk);
1162 return len;
1163}
1164
1165int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1166{
1167 int len = __netlink_sendskb(sk, skb);
1168
1169 sock_put(sk);
1170 return len;
1171}
1172
1173void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1174{
1175 kfree_skb(skb);
1176 sock_put(sk);
1177}
1178
1179static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1180{
1181 int delta;
1182
1183 WARN_ON(skb->sk != NULL);
1184 delta = skb->end - skb->tail;
1185 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1186 return skb;
1187
1188 if (skb_shared(skb)) {
1189 struct sk_buff *nskb = skb_clone(skb, allocation);
1190 if (!nskb)
1191 return skb;
1192 consume_skb(skb);
1193 skb = nskb;
1194 }
1195
1196 if (!pskb_expand_head(skb, 0, -delta, allocation))
1197 skb->truesize -= delta;
1198
1199 return skb;
1200}
1201
1202static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1203 struct sock *ssk)
1204{
1205 int ret;
1206 struct netlink_sock *nlk = nlk_sk(sk);
1207
1208 ret = -ECONNREFUSED;
1209 if (nlk->netlink_rcv != NULL) {
1210 ret = skb->len;
1211 netlink_skb_set_owner_r(skb, sk);
1212 NETLINK_CB(skb).sk = ssk;
1213 netlink_deliver_tap_kernel(sk, ssk, skb);
1214 nlk->netlink_rcv(skb);
1215 consume_skb(skb);
1216 } else {
1217 kfree_skb(skb);
1218 }
1219 sock_put(sk);
1220 return ret;
1221}
1222
1223int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1224 u32 portid, int nonblock)
1225{
1226 struct sock *sk;
1227 int err;
1228 long timeo;
1229
1230 skb = netlink_trim(skb, gfp_any());
1231
1232 timeo = sock_sndtimeo(ssk, nonblock);
1233retry:
1234 sk = netlink_getsockbyportid(ssk, portid);
1235 if (IS_ERR(sk)) {
1236 kfree_skb(skb);
1237 return PTR_ERR(sk);
1238 }
1239 if (netlink_is_kernel(sk))
1240 return netlink_unicast_kernel(sk, skb, ssk);
1241
1242 if (sk_filter(sk, skb)) {
1243 err = skb->len;
1244 kfree_skb(skb);
1245 sock_put(sk);
1246 return err;
1247 }
1248
1249 err = netlink_attachskb(sk, skb, &timeo, ssk);
1250 if (err == 1)
1251 goto retry;
1252 if (err)
1253 return err;
1254
1255 return netlink_sendskb(sk, skb);
1256}
1257EXPORT_SYMBOL(netlink_unicast);
1258
1259int netlink_has_listeners(struct sock *sk, unsigned int group)
1260{
1261 int res = 0;
1262 struct listeners *listeners;
1263
1264 BUG_ON(!netlink_is_kernel(sk));
1265
1266 rcu_read_lock();
1267 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1268
1269 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1270 res = test_bit(group - 1, listeners->masks);
1271
1272 rcu_read_unlock();
1273
1274 return res;
1275}
1276EXPORT_SYMBOL_GPL(netlink_has_listeners);
1277
1278static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1279{
1280 struct netlink_sock *nlk = nlk_sk(sk);
1281
1282 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1283 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1284 netlink_skb_set_owner_r(skb, sk);
1285 __netlink_sendskb(sk, skb);
1286 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1287 }
1288 return -1;
1289}
1290
1291struct netlink_broadcast_data {
1292 struct sock *exclude_sk;
1293 struct net *net;
1294 u32 portid;
1295 u32 group;
1296 int failure;
1297 int delivery_failure;
1298 int congested;
1299 int delivered;
1300 gfp_t allocation;
1301 struct sk_buff *skb, *skb2;
1302 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1303 void *tx_data;
1304};
1305
1306static void do_one_broadcast(struct sock *sk,
1307 struct netlink_broadcast_data *p)
1308{
1309 struct netlink_sock *nlk = nlk_sk(sk);
1310 int val;
1311
1312 if (p->exclude_sk == sk)
1313 return;
1314
1315 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1316 !test_bit(p->group - 1, nlk->groups))
1317 return;
1318
1319 if (!net_eq(sock_net(sk), p->net)) {
1320 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1321 return;
1322
1323 if (!peernet_has_id(sock_net(sk), p->net))
1324 return;
1325
1326 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1327 CAP_NET_BROADCAST))
1328 return;
1329 }
1330
1331 if (p->failure) {
1332 netlink_overrun(sk);
1333 return;
1334 }
1335
1336 sock_hold(sk);
1337 if (p->skb2 == NULL) {
1338 if (skb_shared(p->skb)) {
1339 p->skb2 = skb_clone(p->skb, p->allocation);
1340 } else {
1341 p->skb2 = skb_get(p->skb);
1342 /*
1343 * skb ownership may have been set when
1344 * delivered to a previous socket.
1345 */
1346 skb_orphan(p->skb2);
1347 }
1348 }
1349 if (p->skb2 == NULL) {
1350 netlink_overrun(sk);
1351 /* Clone failed. Notify ALL listeners. */
1352 p->failure = 1;
1353 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1354 p->delivery_failure = 1;
1355 goto out;
1356 }
1357 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1358 kfree_skb(p->skb2);
1359 p->skb2 = NULL;
1360 goto out;
1361 }
1362 if (sk_filter(sk, p->skb2)) {
1363 kfree_skb(p->skb2);
1364 p->skb2 = NULL;
1365 goto out;
1366 }
1367 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1368 NETLINK_CB(p->skb2).nsid_is_set = true;
1369 val = netlink_broadcast_deliver(sk, p->skb2);
1370 if (val < 0) {
1371 netlink_overrun(sk);
1372 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1373 p->delivery_failure = 1;
1374 } else {
1375 p->congested |= val;
1376 p->delivered = 1;
1377 p->skb2 = NULL;
1378 }
1379out:
1380 sock_put(sk);
1381}
1382
1383int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1384 u32 group, gfp_t allocation,
1385 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1386 void *filter_data)
1387{
1388 struct net *net = sock_net(ssk);
1389 struct netlink_broadcast_data info;
1390 struct sock *sk;
1391
1392 skb = netlink_trim(skb, allocation);
1393
1394 info.exclude_sk = ssk;
1395 info.net = net;
1396 info.portid = portid;
1397 info.group = group;
1398 info.failure = 0;
1399 info.delivery_failure = 0;
1400 info.congested = 0;
1401 info.delivered = 0;
1402 info.allocation = allocation;
1403 info.skb = skb;
1404 info.skb2 = NULL;
1405 info.tx_filter = filter;
1406 info.tx_data = filter_data;
1407
1408 /* While we sleep in clone, do not allow to change socket list */
1409
1410 netlink_lock_table();
1411
1412 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1413 do_one_broadcast(sk, &info);
1414
1415 consume_skb(skb);
1416
1417 netlink_unlock_table();
1418
1419 if (info.delivery_failure) {
1420 kfree_skb(info.skb2);
1421 return -ENOBUFS;
1422 }
1423 consume_skb(info.skb2);
1424
1425 if (info.delivered) {
1426 if (info.congested && gfpflags_allow_blocking(allocation))
1427 yield();
1428 return 0;
1429 }
1430 return -ESRCH;
1431}
1432EXPORT_SYMBOL(netlink_broadcast_filtered);
1433
1434int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1435 u32 group, gfp_t allocation)
1436{
1437 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1438 NULL, NULL);
1439}
1440EXPORT_SYMBOL(netlink_broadcast);
1441
1442struct netlink_set_err_data {
1443 struct sock *exclude_sk;
1444 u32 portid;
1445 u32 group;
1446 int code;
1447};
1448
1449static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1450{
1451 struct netlink_sock *nlk = nlk_sk(sk);
1452 int ret = 0;
1453
1454 if (sk == p->exclude_sk)
1455 goto out;
1456
1457 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1458 goto out;
1459
1460 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1461 !test_bit(p->group - 1, nlk->groups))
1462 goto out;
1463
1464 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1465 ret = 1;
1466 goto out;
1467 }
1468
1469 sk->sk_err = p->code;
1470 sk->sk_error_report(sk);
1471out:
1472 return ret;
1473}
1474
1475/**
1476 * netlink_set_err - report error to broadcast listeners
1477 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1478 * @portid: the PORTID of a process that we want to skip (if any)
1479 * @group: the broadcast group that will notice the error
1480 * @code: error code, must be negative (as usual in kernelspace)
1481 *
1482 * This function returns the number of broadcast listeners that have set the
1483 * NETLINK_NO_ENOBUFS socket option.
1484 */
1485int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1486{
1487 struct netlink_set_err_data info;
1488 struct sock *sk;
1489 int ret = 0;
1490
1491 info.exclude_sk = ssk;
1492 info.portid = portid;
1493 info.group = group;
1494 /* sk->sk_err wants a positive error value */
1495 info.code = -code;
1496
1497 read_lock(&nl_table_lock);
1498
1499 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1500 ret += do_one_set_err(sk, &info);
1501
1502 read_unlock(&nl_table_lock);
1503 return ret;
1504}
1505EXPORT_SYMBOL(netlink_set_err);
1506
1507/* must be called with netlink table grabbed */
1508static void netlink_update_socket_mc(struct netlink_sock *nlk,
1509 unsigned int group,
1510 int is_new)
1511{
1512 int old, new = !!is_new, subscriptions;
1513
1514 old = test_bit(group - 1, nlk->groups);
1515 subscriptions = nlk->subscriptions - old + new;
1516 if (new)
1517 __set_bit(group - 1, nlk->groups);
1518 else
1519 __clear_bit(group - 1, nlk->groups);
1520 netlink_update_subscriptions(&nlk->sk, subscriptions);
1521 netlink_update_listeners(&nlk->sk);
1522}
1523
1524static int netlink_setsockopt(struct socket *sock, int level, int optname,
1525 char __user *optval, unsigned int optlen)
1526{
1527 struct sock *sk = sock->sk;
1528 struct netlink_sock *nlk = nlk_sk(sk);
1529 unsigned int val = 0;
1530 int err;
1531
1532 if (level != SOL_NETLINK)
1533 return -ENOPROTOOPT;
1534
1535 if (optlen >= sizeof(int) &&
1536 get_user(val, (unsigned int __user *)optval))
1537 return -EFAULT;
1538
1539 switch (optname) {
1540 case NETLINK_PKTINFO:
1541 if (val)
1542 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1543 else
1544 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1545 err = 0;
1546 break;
1547 case NETLINK_ADD_MEMBERSHIP:
1548 case NETLINK_DROP_MEMBERSHIP: {
1549 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1550 return -EPERM;
1551 err = netlink_realloc_groups(sk);
1552 if (err)
1553 return err;
1554 if (!val || val - 1 >= nlk->ngroups)
1555 return -EINVAL;
1556 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1557 err = nlk->netlink_bind(sock_net(sk), val);
1558 if (err)
1559 return err;
1560 }
1561 netlink_table_grab();
1562 netlink_update_socket_mc(nlk, val,
1563 optname == NETLINK_ADD_MEMBERSHIP);
1564 netlink_table_ungrab();
1565 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1566 nlk->netlink_unbind(sock_net(sk), val);
1567
1568 err = 0;
1569 break;
1570 }
1571 case NETLINK_BROADCAST_ERROR:
1572 if (val)
1573 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1574 else
1575 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1576 err = 0;
1577 break;
1578 case NETLINK_NO_ENOBUFS:
1579 if (val) {
1580 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1581 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1582 wake_up_interruptible(&nlk->wait);
1583 } else {
1584 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1585 }
1586 err = 0;
1587 break;
1588 case NETLINK_LISTEN_ALL_NSID:
1589 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1590 return -EPERM;
1591
1592 if (val)
1593 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1594 else
1595 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1596 err = 0;
1597 break;
1598 case NETLINK_CAP_ACK:
1599 if (val)
1600 nlk->flags |= NETLINK_F_CAP_ACK;
1601 else
1602 nlk->flags &= ~NETLINK_F_CAP_ACK;
1603 err = 0;
1604 break;
1605 default:
1606 err = -ENOPROTOOPT;
1607 }
1608 return err;
1609}
1610
1611static int netlink_getsockopt(struct socket *sock, int level, int optname,
1612 char __user *optval, int __user *optlen)
1613{
1614 struct sock *sk = sock->sk;
1615 struct netlink_sock *nlk = nlk_sk(sk);
1616 int len, val, err;
1617
1618 if (level != SOL_NETLINK)
1619 return -ENOPROTOOPT;
1620
1621 if (get_user(len, optlen))
1622 return -EFAULT;
1623 if (len < 0)
1624 return -EINVAL;
1625
1626 switch (optname) {
1627 case NETLINK_PKTINFO:
1628 if (len < sizeof(int))
1629 return -EINVAL;
1630 len = sizeof(int);
1631 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1632 if (put_user(len, optlen) ||
1633 put_user(val, optval))
1634 return -EFAULT;
1635 err = 0;
1636 break;
1637 case NETLINK_BROADCAST_ERROR:
1638 if (len < sizeof(int))
1639 return -EINVAL;
1640 len = sizeof(int);
1641 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1642 if (put_user(len, optlen) ||
1643 put_user(val, optval))
1644 return -EFAULT;
1645 err = 0;
1646 break;
1647 case NETLINK_NO_ENOBUFS:
1648 if (len < sizeof(int))
1649 return -EINVAL;
1650 len = sizeof(int);
1651 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1652 if (put_user(len, optlen) ||
1653 put_user(val, optval))
1654 return -EFAULT;
1655 err = 0;
1656 break;
1657 case NETLINK_LIST_MEMBERSHIPS: {
1658 int pos, idx, shift;
1659
1660 err = 0;
1661 netlink_lock_table();
1662 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1663 if (len - pos < sizeof(u32))
1664 break;
1665
1666 idx = pos / sizeof(unsigned long);
1667 shift = (pos % sizeof(unsigned long)) * 8;
1668 if (put_user((u32)(nlk->groups[idx] >> shift),
1669 (u32 __user *)(optval + pos))) {
1670 err = -EFAULT;
1671 break;
1672 }
1673 }
1674 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1675 err = -EFAULT;
1676 netlink_unlock_table();
1677 break;
1678 }
1679 case NETLINK_CAP_ACK:
1680 if (len < sizeof(int))
1681 return -EINVAL;
1682 len = sizeof(int);
1683 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1684 if (put_user(len, optlen) ||
1685 put_user(val, optval))
1686 return -EFAULT;
1687 err = 0;
1688 break;
1689 default:
1690 err = -ENOPROTOOPT;
1691 }
1692 return err;
1693}
1694
1695static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1696{
1697 struct nl_pktinfo info;
1698
1699 info.group = NETLINK_CB(skb).dst_group;
1700 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1701}
1702
1703static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1704 struct sk_buff *skb)
1705{
1706 if (!NETLINK_CB(skb).nsid_is_set)
1707 return;
1708
1709 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1710 &NETLINK_CB(skb).nsid);
1711}
1712
1713static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1714{
1715 struct sock *sk = sock->sk;
1716 struct netlink_sock *nlk = nlk_sk(sk);
1717 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1718 u32 dst_portid;
1719 u32 dst_group;
1720 struct sk_buff *skb;
1721 int err;
1722 struct scm_cookie scm;
1723 u32 netlink_skb_flags = 0;
1724
1725 if (msg->msg_flags&MSG_OOB)
1726 return -EOPNOTSUPP;
1727
1728 err = scm_send(sock, msg, &scm, true);
1729 if (err < 0)
1730 return err;
1731
1732 if (msg->msg_namelen) {
1733 err = -EINVAL;
1734 if (addr->nl_family != AF_NETLINK)
1735 goto out;
1736 dst_portid = addr->nl_pid;
1737 dst_group = ffs(addr->nl_groups);
1738 err = -EPERM;
1739 if ((dst_group || dst_portid) &&
1740 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1741 goto out;
1742 netlink_skb_flags |= NETLINK_SKB_DST;
1743 } else {
1744 dst_portid = nlk->dst_portid;
1745 dst_group = nlk->dst_group;
1746 }
1747
1748 if (!nlk->bound) {
1749 err = netlink_autobind(sock);
1750 if (err)
1751 goto out;
1752 } else {
1753 /* Ensure nlk is hashed and visible. */
1754 smp_rmb();
1755 }
1756
1757 err = -EMSGSIZE;
1758 if (len > sk->sk_sndbuf - 32)
1759 goto out;
1760 err = -ENOBUFS;
1761 skb = netlink_alloc_large_skb(len, dst_group);
1762 if (skb == NULL)
1763 goto out;
1764
1765 NETLINK_CB(skb).portid = nlk->portid;
1766 NETLINK_CB(skb).dst_group = dst_group;
1767 NETLINK_CB(skb).creds = scm.creds;
1768 NETLINK_CB(skb).flags = netlink_skb_flags;
1769
1770 err = -EFAULT;
1771 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1772 kfree_skb(skb);
1773 goto out;
1774 }
1775
1776 err = security_netlink_send(sk, skb);
1777 if (err) {
1778 kfree_skb(skb);
1779 goto out;
1780 }
1781
1782 if (dst_group) {
1783 atomic_inc(&skb->users);
1784 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1785 }
1786 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
1787
1788out:
1789 scm_destroy(&scm);
1790 return err;
1791}
1792
1793static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1794 int flags)
1795{
1796 struct scm_cookie scm;
1797 struct sock *sk = sock->sk;
1798 struct netlink_sock *nlk = nlk_sk(sk);
1799 int noblock = flags&MSG_DONTWAIT;
1800 size_t copied;
1801 struct sk_buff *skb, *data_skb;
1802 int err, ret;
1803
1804 if (flags&MSG_OOB)
1805 return -EOPNOTSUPP;
1806
1807 copied = 0;
1808
1809 skb = skb_recv_datagram(sk, flags, noblock, &err);
1810 if (skb == NULL)
1811 goto out;
1812
1813 data_skb = skb;
1814
1815#ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1816 if (unlikely(skb_shinfo(skb)->frag_list)) {
1817 /*
1818 * If this skb has a frag_list, then here that means that we
1819 * will have to use the frag_list skb's data for compat tasks
1820 * and the regular skb's data for normal (non-compat) tasks.
1821 *
1822 * If we need to send the compat skb, assign it to the
1823 * 'data_skb' variable so that it will be used below for data
1824 * copying. We keep 'skb' for everything else, including
1825 * freeing both later.
1826 */
1827 if (flags & MSG_CMSG_COMPAT)
1828 data_skb = skb_shinfo(skb)->frag_list;
1829 }
1830#endif
1831
1832 /* Record the max length of recvmsg() calls for future allocations */
1833 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1834 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1835 16384);
1836
1837 copied = data_skb->len;
1838 if (len < copied) {
1839 msg->msg_flags |= MSG_TRUNC;
1840 copied = len;
1841 }
1842
1843 skb_reset_transport_header(data_skb);
1844 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1845
1846 if (msg->msg_name) {
1847 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1848 addr->nl_family = AF_NETLINK;
1849 addr->nl_pad = 0;
1850 addr->nl_pid = NETLINK_CB(skb).portid;
1851 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1852 msg->msg_namelen = sizeof(*addr);
1853 }
1854
1855 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1856 netlink_cmsg_recv_pktinfo(msg, skb);
1857 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1858 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1859
1860 memset(&scm, 0, sizeof(scm));
1861 scm.creds = *NETLINK_CREDS(skb);
1862 if (flags & MSG_TRUNC)
1863 copied = data_skb->len;
1864
1865 skb_free_datagram(sk, skb);
1866
1867 if (nlk->cb_running &&
1868 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1869 ret = netlink_dump(sk);
1870 if (ret) {
1871 sk->sk_err = -ret;
1872 sk->sk_error_report(sk);
1873 }
1874 }
1875
1876 scm_recv(sock, msg, &scm, flags);
1877out:
1878 netlink_rcv_wake(sk);
1879 return err ? : copied;
1880}
1881
1882static void netlink_data_ready(struct sock *sk)
1883{
1884 BUG();
1885}
1886
1887/*
1888 * We export these functions to other modules. They provide a
1889 * complete set of kernel non-blocking support for message
1890 * queueing.
1891 */
1892
1893struct sock *
1894__netlink_kernel_create(struct net *net, int unit, struct module *module,
1895 struct netlink_kernel_cfg *cfg)
1896{
1897 struct socket *sock;
1898 struct sock *sk;
1899 struct netlink_sock *nlk;
1900 struct listeners *listeners = NULL;
1901 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
1902 unsigned int groups;
1903
1904 BUG_ON(!nl_table);
1905
1906 if (unit < 0 || unit >= MAX_LINKS)
1907 return NULL;
1908
1909 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1910 return NULL;
1911
1912 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
1913 goto out_sock_release_nosk;
1914
1915 sk = sock->sk;
1916
1917 if (!cfg || cfg->groups < 32)
1918 groups = 32;
1919 else
1920 groups = cfg->groups;
1921
1922 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
1923 if (!listeners)
1924 goto out_sock_release;
1925
1926 sk->sk_data_ready = netlink_data_ready;
1927 if (cfg && cfg->input)
1928 nlk_sk(sk)->netlink_rcv = cfg->input;
1929
1930 if (netlink_insert(sk, 0))
1931 goto out_sock_release;
1932
1933 nlk = nlk_sk(sk);
1934 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
1935
1936 netlink_table_grab();
1937 if (!nl_table[unit].registered) {
1938 nl_table[unit].groups = groups;
1939 rcu_assign_pointer(nl_table[unit].listeners, listeners);
1940 nl_table[unit].cb_mutex = cb_mutex;
1941 nl_table[unit].module = module;
1942 if (cfg) {
1943 nl_table[unit].bind = cfg->bind;
1944 nl_table[unit].unbind = cfg->unbind;
1945 nl_table[unit].flags = cfg->flags;
1946 if (cfg->compare)
1947 nl_table[unit].compare = cfg->compare;
1948 }
1949 nl_table[unit].registered = 1;
1950 } else {
1951 kfree(listeners);
1952 nl_table[unit].registered++;
1953 }
1954 netlink_table_ungrab();
1955 return sk;
1956
1957out_sock_release:
1958 kfree(listeners);
1959 netlink_kernel_release(sk);
1960 return NULL;
1961
1962out_sock_release_nosk:
1963 sock_release(sock);
1964 return NULL;
1965}
1966EXPORT_SYMBOL(__netlink_kernel_create);
1967
1968void
1969netlink_kernel_release(struct sock *sk)
1970{
1971 if (sk == NULL || sk->sk_socket == NULL)
1972 return;
1973
1974 sock_release(sk->sk_socket);
1975}
1976EXPORT_SYMBOL(netlink_kernel_release);
1977
1978int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
1979{
1980 struct listeners *new, *old;
1981 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1982
1983 if (groups < 32)
1984 groups = 32;
1985
1986 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1987 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
1988 if (!new)
1989 return -ENOMEM;
1990 old = nl_deref_protected(tbl->listeners);
1991 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
1992 rcu_assign_pointer(tbl->listeners, new);
1993
1994 kfree_rcu(old, rcu);
1995 }
1996 tbl->groups = groups;
1997
1998 return 0;
1999}
2000
2001/**
2002 * netlink_change_ngroups - change number of multicast groups
2003 *
2004 * This changes the number of multicast groups that are available
2005 * on a certain netlink family. Note that it is not possible to
2006 * change the number of groups to below 32. Also note that it does
2007 * not implicitly call netlink_clear_multicast_users() when the
2008 * number of groups is reduced.
2009 *
2010 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2011 * @groups: The new number of groups.
2012 */
2013int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2014{
2015 int err;
2016
2017 netlink_table_grab();
2018 err = __netlink_change_ngroups(sk, groups);
2019 netlink_table_ungrab();
2020
2021 return err;
2022}
2023
2024void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2025{
2026 struct sock *sk;
2027 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2028
2029 sk_for_each_bound(sk, &tbl->mc_list)
2030 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2031}
2032
2033struct nlmsghdr *
2034__nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2035{
2036 struct nlmsghdr *nlh;
2037 int size = nlmsg_msg_size(len);
2038
2039 nlh = (struct nlmsghdr *)skb_put(skb, NLMSG_ALIGN(size));
2040 nlh->nlmsg_type = type;
2041 nlh->nlmsg_len = size;
2042 nlh->nlmsg_flags = flags;
2043 nlh->nlmsg_pid = portid;
2044 nlh->nlmsg_seq = seq;
2045 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2046 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2047 return nlh;
2048}
2049EXPORT_SYMBOL(__nlmsg_put);
2050
2051/*
2052 * It looks a bit ugly.
2053 * It would be better to create kernel thread.
2054 */
2055
2056static int netlink_dump(struct sock *sk)
2057{
2058 struct netlink_sock *nlk = nlk_sk(sk);
2059 struct netlink_callback *cb;
2060 struct sk_buff *skb = NULL;
2061 struct nlmsghdr *nlh;
2062 int len, err = -ENOBUFS;
2063 int alloc_min_size;
2064 int alloc_size;
2065
2066 mutex_lock(nlk->cb_mutex);
2067 if (!nlk->cb_running) {
2068 err = -EINVAL;
2069 goto errout_skb;
2070 }
2071
2072 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2073 goto errout_skb;
2074
2075 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2076 * required, but it makes sense to _attempt_ a 16K bytes allocation
2077 * to reduce number of system calls on dump operations, if user
2078 * ever provided a big enough buffer.
2079 */
2080 cb = &nlk->cb;
2081 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2082
2083 if (alloc_min_size < nlk->max_recvmsg_len) {
2084 alloc_size = nlk->max_recvmsg_len;
2085 skb = alloc_skb(alloc_size, GFP_KERNEL |
2086 __GFP_NOWARN | __GFP_NORETRY);
2087 }
2088 if (!skb) {
2089 alloc_size = alloc_min_size;
2090 skb = alloc_skb(alloc_size, GFP_KERNEL);
2091 }
2092 if (!skb)
2093 goto errout_skb;
2094
2095 /* Trim skb to allocated size. User is expected to provide buffer as
2096 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2097 * netlink_recvmsg())). dump will pack as many smaller messages as
2098 * could fit within the allocated skb. skb is typically allocated
2099 * with larger space than required (could be as much as near 2x the
2100 * requested size with align to next power of 2 approach). Allowing
2101 * dump to use the excess space makes it difficult for a user to have a
2102 * reasonable static buffer based on the expected largest dump of a
2103 * single netdev. The outcome is MSG_TRUNC error.
2104 */
2105 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2106 netlink_skb_set_owner_r(skb, sk);
2107
2108 len = cb->dump(skb, cb);
2109
2110 if (len > 0) {
2111 mutex_unlock(nlk->cb_mutex);
2112
2113 if (sk_filter(sk, skb))
2114 kfree_skb(skb);
2115 else
2116 __netlink_sendskb(sk, skb);
2117 return 0;
2118 }
2119
2120 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
2121 if (!nlh)
2122 goto errout_skb;
2123
2124 nl_dump_check_consistent(cb, nlh);
2125
2126 memcpy(nlmsg_data(nlh), &len, sizeof(len));
2127
2128 if (sk_filter(sk, skb))
2129 kfree_skb(skb);
2130 else
2131 __netlink_sendskb(sk, skb);
2132
2133 if (cb->done)
2134 cb->done(cb);
2135
2136 nlk->cb_running = false;
2137 mutex_unlock(nlk->cb_mutex);
2138 module_put(cb->module);
2139 consume_skb(cb->skb);
2140 return 0;
2141
2142errout_skb:
2143 mutex_unlock(nlk->cb_mutex);
2144 kfree_skb(skb);
2145 return err;
2146}
2147
2148int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2149 const struct nlmsghdr *nlh,
2150 struct netlink_dump_control *control)
2151{
2152 struct netlink_callback *cb;
2153 struct sock *sk;
2154 struct netlink_sock *nlk;
2155 int ret;
2156
2157 atomic_inc(&skb->users);
2158
2159 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2160 if (sk == NULL) {
2161 ret = -ECONNREFUSED;
2162 goto error_free;
2163 }
2164
2165 nlk = nlk_sk(sk);
2166 mutex_lock(nlk->cb_mutex);
2167 /* A dump is in progress... */
2168 if (nlk->cb_running) {
2169 ret = -EBUSY;
2170 goto error_unlock;
2171 }
2172 /* add reference of module which cb->dump belongs to */
2173 if (!try_module_get(control->module)) {
2174 ret = -EPROTONOSUPPORT;
2175 goto error_unlock;
2176 }
2177
2178 cb = &nlk->cb;
2179 memset(cb, 0, sizeof(*cb));
2180 cb->start = control->start;
2181 cb->dump = control->dump;
2182 cb->done = control->done;
2183 cb->nlh = nlh;
2184 cb->data = control->data;
2185 cb->module = control->module;
2186 cb->min_dump_alloc = control->min_dump_alloc;
2187 cb->skb = skb;
2188
2189 nlk->cb_running = true;
2190
2191 mutex_unlock(nlk->cb_mutex);
2192
2193 if (cb->start)
2194 cb->start(cb);
2195
2196 ret = netlink_dump(sk);
2197 sock_put(sk);
2198
2199 if (ret)
2200 return ret;
2201
2202 /* We successfully started a dump, by returning -EINTR we
2203 * signal not to send ACK even if it was requested.
2204 */
2205 return -EINTR;
2206
2207error_unlock:
2208 sock_put(sk);
2209 mutex_unlock(nlk->cb_mutex);
2210error_free:
2211 kfree_skb(skb);
2212 return ret;
2213}
2214EXPORT_SYMBOL(__netlink_dump_start);
2215
2216void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
2217{
2218 struct sk_buff *skb;
2219 struct nlmsghdr *rep;
2220 struct nlmsgerr *errmsg;
2221 size_t payload = sizeof(*errmsg);
2222 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2223
2224 /* Error messages get the original request appened, unless the user
2225 * requests to cap the error message.
2226 */
2227 if (!(nlk->flags & NETLINK_F_CAP_ACK) && err)
2228 payload += nlmsg_len(nlh);
2229
2230 skb = nlmsg_new(payload, GFP_KERNEL);
2231 if (!skb) {
2232 struct sock *sk;
2233
2234 sk = netlink_lookup(sock_net(in_skb->sk),
2235 in_skb->sk->sk_protocol,
2236 NETLINK_CB(in_skb).portid);
2237 if (sk) {
2238 sk->sk_err = ENOBUFS;
2239 sk->sk_error_report(sk);
2240 sock_put(sk);
2241 }
2242 return;
2243 }
2244
2245 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2246 NLMSG_ERROR, payload, 0);
2247 errmsg = nlmsg_data(rep);
2248 errmsg->error = err;
2249 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2250 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2251}
2252EXPORT_SYMBOL(netlink_ack);
2253
2254int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2255 struct nlmsghdr *))
2256{
2257 struct nlmsghdr *nlh;
2258 int err;
2259
2260 while (skb->len >= nlmsg_total_size(0)) {
2261 int msglen;
2262
2263 nlh = nlmsg_hdr(skb);
2264 err = 0;
2265
2266 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2267 return 0;
2268
2269 /* Only requests are handled by the kernel */
2270 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2271 goto ack;
2272
2273 /* Skip control messages */
2274 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2275 goto ack;
2276
2277 err = cb(skb, nlh);
2278 if (err == -EINTR)
2279 goto skip;
2280
2281ack:
2282 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2283 netlink_ack(skb, nlh, err);
2284
2285skip:
2286 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2287 if (msglen > skb->len)
2288 msglen = skb->len;
2289 skb_pull(skb, msglen);
2290 }
2291
2292 return 0;
2293}
2294EXPORT_SYMBOL(netlink_rcv_skb);
2295
2296/**
2297 * nlmsg_notify - send a notification netlink message
2298 * @sk: netlink socket to use
2299 * @skb: notification message
2300 * @portid: destination netlink portid for reports or 0
2301 * @group: destination multicast group or 0
2302 * @report: 1 to report back, 0 to disable
2303 * @flags: allocation flags
2304 */
2305int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2306 unsigned int group, int report, gfp_t flags)
2307{
2308 int err = 0;
2309
2310 if (group) {
2311 int exclude_portid = 0;
2312
2313 if (report) {
2314 atomic_inc(&skb->users);
2315 exclude_portid = portid;
2316 }
2317
2318 /* errors reported via destination sk->sk_err, but propagate
2319 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2320 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2321 }
2322
2323 if (report) {
2324 int err2;
2325
2326 err2 = nlmsg_unicast(sk, skb, portid);
2327 if (!err || err == -ESRCH)
2328 err = err2;
2329 }
2330
2331 return err;
2332}
2333EXPORT_SYMBOL(nlmsg_notify);
2334
2335#ifdef CONFIG_PROC_FS
2336struct nl_seq_iter {
2337 struct seq_net_private p;
2338 struct rhashtable_iter hti;
2339 int link;
2340};
2341
2342static int netlink_walk_start(struct nl_seq_iter *iter)
2343{
2344 int err;
2345
2346 err = rhashtable_walk_init(&nl_table[iter->link].hash, &iter->hti);
2347 if (err) {
2348 iter->link = MAX_LINKS;
2349 return err;
2350 }
2351
2352 err = rhashtable_walk_start(&iter->hti);
2353 return err == -EAGAIN ? 0 : err;
2354}
2355
2356static void netlink_walk_stop(struct nl_seq_iter *iter)
2357{
2358 rhashtable_walk_stop(&iter->hti);
2359 rhashtable_walk_exit(&iter->hti);
2360}
2361
2362static void *__netlink_seq_next(struct seq_file *seq)
2363{
2364 struct nl_seq_iter *iter = seq->private;
2365 struct netlink_sock *nlk;
2366
2367 do {
2368 for (;;) {
2369 int err;
2370
2371 nlk = rhashtable_walk_next(&iter->hti);
2372
2373 if (IS_ERR(nlk)) {
2374 if (PTR_ERR(nlk) == -EAGAIN)
2375 continue;
2376
2377 return nlk;
2378 }
2379
2380 if (nlk)
2381 break;
2382
2383 netlink_walk_stop(iter);
2384 if (++iter->link >= MAX_LINKS)
2385 return NULL;
2386
2387 err = netlink_walk_start(iter);
2388 if (err)
2389 return ERR_PTR(err);
2390 }
2391 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2392
2393 return nlk;
2394}
2395
2396static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2397{
2398 struct nl_seq_iter *iter = seq->private;
2399 void *obj = SEQ_START_TOKEN;
2400 loff_t pos;
2401 int err;
2402
2403 iter->link = 0;
2404
2405 err = netlink_walk_start(iter);
2406 if (err)
2407 return ERR_PTR(err);
2408
2409 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2410 obj = __netlink_seq_next(seq);
2411
2412 return obj;
2413}
2414
2415static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2416{
2417 ++*pos;
2418 return __netlink_seq_next(seq);
2419}
2420
2421static void netlink_seq_stop(struct seq_file *seq, void *v)
2422{
2423 struct nl_seq_iter *iter = seq->private;
2424
2425 if (iter->link >= MAX_LINKS)
2426 return;
2427
2428 netlink_walk_stop(iter);
2429}
2430
2431
2432static int netlink_seq_show(struct seq_file *seq, void *v)
2433{
2434 if (v == SEQ_START_TOKEN) {
2435 seq_puts(seq,
2436 "sk Eth Pid Groups "
2437 "Rmem Wmem Dump Locks Drops Inode\n");
2438 } else {
2439 struct sock *s = v;
2440 struct netlink_sock *nlk = nlk_sk(s);
2441
2442 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
2443 s,
2444 s->sk_protocol,
2445 nlk->portid,
2446 nlk->groups ? (u32)nlk->groups[0] : 0,
2447 sk_rmem_alloc_get(s),
2448 sk_wmem_alloc_get(s),
2449 nlk->cb_running,
2450 atomic_read(&s->sk_refcnt),
2451 atomic_read(&s->sk_drops),
2452 sock_i_ino(s)
2453 );
2454
2455 }
2456 return 0;
2457}
2458
2459static const struct seq_operations netlink_seq_ops = {
2460 .start = netlink_seq_start,
2461 .next = netlink_seq_next,
2462 .stop = netlink_seq_stop,
2463 .show = netlink_seq_show,
2464};
2465
2466
2467static int netlink_seq_open(struct inode *inode, struct file *file)
2468{
2469 return seq_open_net(inode, file, &netlink_seq_ops,
2470 sizeof(struct nl_seq_iter));
2471}
2472
2473static const struct file_operations netlink_seq_fops = {
2474 .owner = THIS_MODULE,
2475 .open = netlink_seq_open,
2476 .read = seq_read,
2477 .llseek = seq_lseek,
2478 .release = seq_release_net,
2479};
2480
2481#endif
2482
2483int netlink_register_notifier(struct notifier_block *nb)
2484{
2485 return atomic_notifier_chain_register(&netlink_chain, nb);
2486}
2487EXPORT_SYMBOL(netlink_register_notifier);
2488
2489int netlink_unregister_notifier(struct notifier_block *nb)
2490{
2491 return atomic_notifier_chain_unregister(&netlink_chain, nb);
2492}
2493EXPORT_SYMBOL(netlink_unregister_notifier);
2494
2495static const struct proto_ops netlink_ops = {
2496 .family = PF_NETLINK,
2497 .owner = THIS_MODULE,
2498 .release = netlink_release,
2499 .bind = netlink_bind,
2500 .connect = netlink_connect,
2501 .socketpair = sock_no_socketpair,
2502 .accept = sock_no_accept,
2503 .getname = netlink_getname,
2504 .poll = datagram_poll,
2505 .ioctl = netlink_ioctl,
2506 .listen = sock_no_listen,
2507 .shutdown = sock_no_shutdown,
2508 .setsockopt = netlink_setsockopt,
2509 .getsockopt = netlink_getsockopt,
2510 .sendmsg = netlink_sendmsg,
2511 .recvmsg = netlink_recvmsg,
2512 .mmap = sock_no_mmap,
2513 .sendpage = sock_no_sendpage,
2514};
2515
2516static const struct net_proto_family netlink_family_ops = {
2517 .family = PF_NETLINK,
2518 .create = netlink_create,
2519 .owner = THIS_MODULE, /* for consistency 8) */
2520};
2521
2522static int __net_init netlink_net_init(struct net *net)
2523{
2524#ifdef CONFIG_PROC_FS
2525 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
2526 return -ENOMEM;
2527#endif
2528 return 0;
2529}
2530
2531static void __net_exit netlink_net_exit(struct net *net)
2532{
2533#ifdef CONFIG_PROC_FS
2534 remove_proc_entry("netlink", net->proc_net);
2535#endif
2536}
2537
2538static void __init netlink_add_usersock_entry(void)
2539{
2540 struct listeners *listeners;
2541 int groups = 32;
2542
2543 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2544 if (!listeners)
2545 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2546
2547 netlink_table_grab();
2548
2549 nl_table[NETLINK_USERSOCK].groups = groups;
2550 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2551 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2552 nl_table[NETLINK_USERSOCK].registered = 1;
2553 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2554
2555 netlink_table_ungrab();
2556}
2557
2558static struct pernet_operations __net_initdata netlink_net_ops = {
2559 .init = netlink_net_init,
2560 .exit = netlink_net_exit,
2561};
2562
2563static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2564{
2565 const struct netlink_sock *nlk = data;
2566 struct netlink_compare_arg arg;
2567
2568 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2569 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2570}
2571
2572static const struct rhashtable_params netlink_rhashtable_params = {
2573 .head_offset = offsetof(struct netlink_sock, node),
2574 .key_len = netlink_compare_arg_len,
2575 .obj_hashfn = netlink_hash,
2576 .obj_cmpfn = netlink_compare,
2577 .automatic_shrinking = true,
2578};
2579
2580static int __init netlink_proto_init(void)
2581{
2582 int i;
2583 int err = proto_register(&netlink_proto, 0);
2584
2585 if (err != 0)
2586 goto out;
2587
2588 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2589
2590 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2591 if (!nl_table)
2592 goto panic;
2593
2594 for (i = 0; i < MAX_LINKS; i++) {
2595 if (rhashtable_init(&nl_table[i].hash,
2596 &netlink_rhashtable_params) < 0) {
2597 while (--i > 0)
2598 rhashtable_destroy(&nl_table[i].hash);
2599 kfree(nl_table);
2600 goto panic;
2601 }
2602 }
2603
2604 INIT_LIST_HEAD(&netlink_tap_all);
2605
2606 netlink_add_usersock_entry();
2607
2608 sock_register(&netlink_family_ops);
2609 register_pernet_subsys(&netlink_net_ops);
2610 /* The netlink device handler may be needed early. */
2611 rtnetlink_init();
2612out:
2613 return err;
2614panic:
2615 panic("netlink_init: Cannot allocate nl_table\n");
2616}
2617
2618core_initcall(netlink_proto_init);