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/capability.h>
24#include <linux/kernel.h>
25#include <linux/init.h>
26#include <linux/signal.h>
27#include <linux/sched.h>
28#include <linux/errno.h>
29#include <linux/string.h>
30#include <linux/stat.h>
31#include <linux/socket.h>
32#include <linux/un.h>
33#include <linux/fcntl.h>
34#include <linux/termios.h>
35#include <linux/sockios.h>
36#include <linux/net.h>
37#include <linux/fs.h>
38#include <linux/slab.h>
39#include <linux/uaccess.h>
40#include <linux/skbuff.h>
41#include <linux/netdevice.h>
42#include <linux/rtnetlink.h>
43#include <linux/proc_fs.h>
44#include <linux/seq_file.h>
45#include <linux/notifier.h>
46#include <linux/security.h>
47#include <linux/jhash.h>
48#include <linux/jiffies.h>
49#include <linux/random.h>
50#include <linux/bitops.h>
51#include <linux/mm.h>
52#include <linux/types.h>
53#include <linux/audit.h>
54#include <linux/mutex.h>
55#include <linux/vmalloc.h>
56#include <linux/if_arp.h>
57#include <linux/rhashtable.h>
58#include <asm/cacheflush.h>
59#include <linux/hash.h>
60#include <linux/genetlink.h>
61#include <linux/net_namespace.h>
62#include <linux/nospec.h>
63
64#include <net/net_namespace.h>
65#include <net/netns/generic.h>
66#include <net/sock.h>
67#include <net/scm.h>
68#include <net/netlink.h>
69
70#include "af_netlink.h"
71
72struct listeners {
73 struct rcu_head rcu;
74 unsigned long masks[0];
75};
76
77/* state bits */
78#define NETLINK_S_CONGESTED 0x0
79
80static inline int netlink_is_kernel(struct sock *sk)
81{
82 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
83}
84
85struct netlink_table *nl_table __read_mostly;
86EXPORT_SYMBOL_GPL(nl_table);
87
88static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
89
90static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
91
92static const char *const nlk_cb_mutex_key_strings[MAX_LINKS + 1] = {
93 "nlk_cb_mutex-ROUTE",
94 "nlk_cb_mutex-1",
95 "nlk_cb_mutex-USERSOCK",
96 "nlk_cb_mutex-FIREWALL",
97 "nlk_cb_mutex-SOCK_DIAG",
98 "nlk_cb_mutex-NFLOG",
99 "nlk_cb_mutex-XFRM",
100 "nlk_cb_mutex-SELINUX",
101 "nlk_cb_mutex-ISCSI",
102 "nlk_cb_mutex-AUDIT",
103 "nlk_cb_mutex-FIB_LOOKUP",
104 "nlk_cb_mutex-CONNECTOR",
105 "nlk_cb_mutex-NETFILTER",
106 "nlk_cb_mutex-IP6_FW",
107 "nlk_cb_mutex-DNRTMSG",
108 "nlk_cb_mutex-KOBJECT_UEVENT",
109 "nlk_cb_mutex-GENERIC",
110 "nlk_cb_mutex-17",
111 "nlk_cb_mutex-SCSITRANSPORT",
112 "nlk_cb_mutex-ECRYPTFS",
113 "nlk_cb_mutex-RDMA",
114 "nlk_cb_mutex-CRYPTO",
115 "nlk_cb_mutex-SMC",
116 "nlk_cb_mutex-23",
117 "nlk_cb_mutex-24",
118 "nlk_cb_mutex-25",
119 "nlk_cb_mutex-26",
120 "nlk_cb_mutex-27",
121 "nlk_cb_mutex-28",
122 "nlk_cb_mutex-29",
123 "nlk_cb_mutex-30",
124 "nlk_cb_mutex-31",
125 "nlk_cb_mutex-MAX_LINKS"
126};
127
128static int netlink_dump(struct sock *sk);
129
130/* nl_table locking explained:
131 * Lookup and traversal are protected with an RCU read-side lock. Insertion
132 * and removal are protected with per bucket lock while using RCU list
133 * modification primitives and may run in parallel to RCU protected lookups.
134 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
135 * been acquired * either during or after the socket has been removed from
136 * the list and after an RCU grace period.
137 */
138DEFINE_RWLOCK(nl_table_lock);
139EXPORT_SYMBOL_GPL(nl_table_lock);
140static atomic_t nl_table_users = ATOMIC_INIT(0);
141
142#define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
143
144static BLOCKING_NOTIFIER_HEAD(netlink_chain);
145
146
147static const struct rhashtable_params netlink_rhashtable_params;
148
149static inline u32 netlink_group_mask(u32 group)
150{
151 return group ? 1 << (group - 1) : 0;
152}
153
154static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
155 gfp_t gfp_mask)
156{
157 unsigned int len = skb_end_offset(skb);
158 struct sk_buff *new;
159
160 new = alloc_skb(len, gfp_mask);
161 if (new == NULL)
162 return NULL;
163
164 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
165 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
166 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
167
168 skb_put_data(new, skb->data, len);
169 return new;
170}
171
172static unsigned int netlink_tap_net_id;
173
174struct netlink_tap_net {
175 struct list_head netlink_tap_all;
176 struct mutex netlink_tap_lock;
177};
178
179int netlink_add_tap(struct netlink_tap *nt)
180{
181 struct net *net = dev_net(nt->dev);
182 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
183
184 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
185 return -EINVAL;
186
187 mutex_lock(&nn->netlink_tap_lock);
188 list_add_rcu(&nt->list, &nn->netlink_tap_all);
189 mutex_unlock(&nn->netlink_tap_lock);
190
191 __module_get(nt->module);
192
193 return 0;
194}
195EXPORT_SYMBOL_GPL(netlink_add_tap);
196
197static int __netlink_remove_tap(struct netlink_tap *nt)
198{
199 struct net *net = dev_net(nt->dev);
200 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
201 bool found = false;
202 struct netlink_tap *tmp;
203
204 mutex_lock(&nn->netlink_tap_lock);
205
206 list_for_each_entry(tmp, &nn->netlink_tap_all, list) {
207 if (nt == tmp) {
208 list_del_rcu(&nt->list);
209 found = true;
210 goto out;
211 }
212 }
213
214 pr_warn("__netlink_remove_tap: %p not found\n", nt);
215out:
216 mutex_unlock(&nn->netlink_tap_lock);
217
218 if (found)
219 module_put(nt->module);
220
221 return found ? 0 : -ENODEV;
222}
223
224int netlink_remove_tap(struct netlink_tap *nt)
225{
226 int ret;
227
228 ret = __netlink_remove_tap(nt);
229 synchronize_net();
230
231 return ret;
232}
233EXPORT_SYMBOL_GPL(netlink_remove_tap);
234
235static __net_init int netlink_tap_init_net(struct net *net)
236{
237 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
238
239 INIT_LIST_HEAD(&nn->netlink_tap_all);
240 mutex_init(&nn->netlink_tap_lock);
241 return 0;
242}
243
244static struct pernet_operations netlink_tap_net_ops = {
245 .init = netlink_tap_init_net,
246 .id = &netlink_tap_net_id,
247 .size = sizeof(struct netlink_tap_net),
248};
249
250static bool netlink_filter_tap(const struct sk_buff *skb)
251{
252 struct sock *sk = skb->sk;
253
254 /* We take the more conservative approach and
255 * whitelist socket protocols that may pass.
256 */
257 switch (sk->sk_protocol) {
258 case NETLINK_ROUTE:
259 case NETLINK_USERSOCK:
260 case NETLINK_SOCK_DIAG:
261 case NETLINK_NFLOG:
262 case NETLINK_XFRM:
263 case NETLINK_FIB_LOOKUP:
264 case NETLINK_NETFILTER:
265 case NETLINK_GENERIC:
266 return true;
267 }
268
269 return false;
270}
271
272static int __netlink_deliver_tap_skb(struct sk_buff *skb,
273 struct net_device *dev)
274{
275 struct sk_buff *nskb;
276 struct sock *sk = skb->sk;
277 int ret = -ENOMEM;
278
279 if (!net_eq(dev_net(dev), sock_net(sk)))
280 return 0;
281
282 dev_hold(dev);
283
284 if (is_vmalloc_addr(skb->head))
285 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
286 else
287 nskb = skb_clone(skb, GFP_ATOMIC);
288 if (nskb) {
289 nskb->dev = dev;
290 nskb->protocol = htons((u16) sk->sk_protocol);
291 nskb->pkt_type = netlink_is_kernel(sk) ?
292 PACKET_KERNEL : PACKET_USER;
293 skb_reset_network_header(nskb);
294 ret = dev_queue_xmit(nskb);
295 if (unlikely(ret > 0))
296 ret = net_xmit_errno(ret);
297 }
298
299 dev_put(dev);
300 return ret;
301}
302
303static void __netlink_deliver_tap(struct sk_buff *skb, struct netlink_tap_net *nn)
304{
305 int ret;
306 struct netlink_tap *tmp;
307
308 if (!netlink_filter_tap(skb))
309 return;
310
311 list_for_each_entry_rcu(tmp, &nn->netlink_tap_all, list) {
312 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
313 if (unlikely(ret))
314 break;
315 }
316}
317
318static void netlink_deliver_tap(struct net *net, struct sk_buff *skb)
319{
320 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
321
322 rcu_read_lock();
323
324 if (unlikely(!list_empty(&nn->netlink_tap_all)))
325 __netlink_deliver_tap(skb, nn);
326
327 rcu_read_unlock();
328}
329
330static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
331 struct sk_buff *skb)
332{
333 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
334 netlink_deliver_tap(sock_net(dst), skb);
335}
336
337static void netlink_overrun(struct sock *sk)
338{
339 struct netlink_sock *nlk = nlk_sk(sk);
340
341 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
342 if (!test_and_set_bit(NETLINK_S_CONGESTED,
343 &nlk_sk(sk)->state)) {
344 sk->sk_err = ENOBUFS;
345 sk->sk_error_report(sk);
346 }
347 }
348 atomic_inc(&sk->sk_drops);
349}
350
351static void netlink_rcv_wake(struct sock *sk)
352{
353 struct netlink_sock *nlk = nlk_sk(sk);
354
355 if (skb_queue_empty(&sk->sk_receive_queue))
356 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
357 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
358 wake_up_interruptible(&nlk->wait);
359}
360
361static void netlink_skb_destructor(struct sk_buff *skb)
362{
363 if (is_vmalloc_addr(skb->head)) {
364 if (!skb->cloned ||
365 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
366 vfree(skb->head);
367
368 skb->head = NULL;
369 }
370 if (skb->sk != NULL)
371 sock_rfree(skb);
372}
373
374static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
375{
376 WARN_ON(skb->sk != NULL);
377 skb->sk = sk;
378 skb->destructor = netlink_skb_destructor;
379 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
380 sk_mem_charge(sk, skb->truesize);
381}
382
383static void netlink_sock_destruct(struct sock *sk)
384{
385 struct netlink_sock *nlk = nlk_sk(sk);
386
387 if (nlk->cb_running) {
388 if (nlk->cb.done)
389 nlk->cb.done(&nlk->cb);
390 module_put(nlk->cb.module);
391 kfree_skb(nlk->cb.skb);
392 }
393
394 skb_queue_purge(&sk->sk_receive_queue);
395
396 if (!sock_flag(sk, SOCK_DEAD)) {
397 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
398 return;
399 }
400
401 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
402 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
403 WARN_ON(nlk_sk(sk)->groups);
404}
405
406static void netlink_sock_destruct_work(struct work_struct *work)
407{
408 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
409 work);
410
411 sk_free(&nlk->sk);
412}
413
414/* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
415 * SMP. Look, when several writers sleep and reader wakes them up, all but one
416 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
417 * this, _but_ remember, it adds useless work on UP machines.
418 */
419
420void netlink_table_grab(void)
421 __acquires(nl_table_lock)
422{
423 might_sleep();
424
425 write_lock_irq(&nl_table_lock);
426
427 if (atomic_read(&nl_table_users)) {
428 DECLARE_WAITQUEUE(wait, current);
429
430 add_wait_queue_exclusive(&nl_table_wait, &wait);
431 for (;;) {
432 set_current_state(TASK_UNINTERRUPTIBLE);
433 if (atomic_read(&nl_table_users) == 0)
434 break;
435 write_unlock_irq(&nl_table_lock);
436 schedule();
437 write_lock_irq(&nl_table_lock);
438 }
439
440 __set_current_state(TASK_RUNNING);
441 remove_wait_queue(&nl_table_wait, &wait);
442 }
443}
444
445void netlink_table_ungrab(void)
446 __releases(nl_table_lock)
447{
448 write_unlock_irq(&nl_table_lock);
449 wake_up(&nl_table_wait);
450}
451
452static inline void
453netlink_lock_table(void)
454{
455 /* read_lock() synchronizes us to netlink_table_grab */
456
457 read_lock(&nl_table_lock);
458 atomic_inc(&nl_table_users);
459 read_unlock(&nl_table_lock);
460}
461
462static inline void
463netlink_unlock_table(void)
464{
465 if (atomic_dec_and_test(&nl_table_users))
466 wake_up(&nl_table_wait);
467}
468
469struct netlink_compare_arg
470{
471 possible_net_t pnet;
472 u32 portid;
473};
474
475/* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
476#define netlink_compare_arg_len \
477 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
478
479static inline int netlink_compare(struct rhashtable_compare_arg *arg,
480 const void *ptr)
481{
482 const struct netlink_compare_arg *x = arg->key;
483 const struct netlink_sock *nlk = ptr;
484
485 return nlk->portid != x->portid ||
486 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
487}
488
489static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
490 struct net *net, u32 portid)
491{
492 memset(arg, 0, sizeof(*arg));
493 write_pnet(&arg->pnet, net);
494 arg->portid = portid;
495}
496
497static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
498 struct net *net)
499{
500 struct netlink_compare_arg arg;
501
502 netlink_compare_arg_init(&arg, net, portid);
503 return rhashtable_lookup_fast(&table->hash, &arg,
504 netlink_rhashtable_params);
505}
506
507static int __netlink_insert(struct netlink_table *table, struct sock *sk)
508{
509 struct netlink_compare_arg arg;
510
511 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
512 return rhashtable_lookup_insert_key(&table->hash, &arg,
513 &nlk_sk(sk)->node,
514 netlink_rhashtable_params);
515}
516
517static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
518{
519 struct netlink_table *table = &nl_table[protocol];
520 struct sock *sk;
521
522 rcu_read_lock();
523 sk = __netlink_lookup(table, portid, net);
524 if (sk)
525 sock_hold(sk);
526 rcu_read_unlock();
527
528 return sk;
529}
530
531static const struct proto_ops netlink_ops;
532
533static void
534netlink_update_listeners(struct sock *sk)
535{
536 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
537 unsigned long mask;
538 unsigned int i;
539 struct listeners *listeners;
540
541 listeners = nl_deref_protected(tbl->listeners);
542 if (!listeners)
543 return;
544
545 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
546 mask = 0;
547 sk_for_each_bound(sk, &tbl->mc_list) {
548 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
549 mask |= nlk_sk(sk)->groups[i];
550 }
551 listeners->masks[i] = mask;
552 }
553 /* this function is only called with the netlink table "grabbed", which
554 * makes sure updates are visible before bind or setsockopt return. */
555}
556
557static int netlink_insert(struct sock *sk, u32 portid)
558{
559 struct netlink_table *table = &nl_table[sk->sk_protocol];
560 int err;
561
562 lock_sock(sk);
563
564 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
565 if (nlk_sk(sk)->bound)
566 goto err;
567
568 nlk_sk(sk)->portid = portid;
569 sock_hold(sk);
570
571 err = __netlink_insert(table, sk);
572 if (err) {
573 /* In case the hashtable backend returns with -EBUSY
574 * from here, it must not escape to the caller.
575 */
576 if (unlikely(err == -EBUSY))
577 err = -EOVERFLOW;
578 if (err == -EEXIST)
579 err = -EADDRINUSE;
580 sock_put(sk);
581 goto err;
582 }
583
584 /* We need to ensure that the socket is hashed and visible. */
585 smp_wmb();
586 nlk_sk(sk)->bound = portid;
587
588err:
589 release_sock(sk);
590 return err;
591}
592
593static void netlink_remove(struct sock *sk)
594{
595 struct netlink_table *table;
596
597 table = &nl_table[sk->sk_protocol];
598 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
599 netlink_rhashtable_params)) {
600 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
601 __sock_put(sk);
602 }
603
604 netlink_table_grab();
605 if (nlk_sk(sk)->subscriptions) {
606 __sk_del_bind_node(sk);
607 netlink_update_listeners(sk);
608 }
609 if (sk->sk_protocol == NETLINK_GENERIC)
610 atomic_inc(&genl_sk_destructing_cnt);
611 netlink_table_ungrab();
612}
613
614static struct proto netlink_proto = {
615 .name = "NETLINK",
616 .owner = THIS_MODULE,
617 .obj_size = sizeof(struct netlink_sock),
618};
619
620static int __netlink_create(struct net *net, struct socket *sock,
621 struct mutex *cb_mutex, int protocol,
622 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 if (cb_mutex) {
637 nlk->cb_mutex = cb_mutex;
638 } else {
639 nlk->cb_mutex = &nlk->cb_def_mutex;
640 mutex_init(nlk->cb_mutex);
641 lockdep_set_class_and_name(nlk->cb_mutex,
642 nlk_cb_mutex_keys + protocol,
643 nlk_cb_mutex_key_strings[protocol]);
644 }
645 init_waitqueue_head(&nlk->wait);
646
647 sk->sk_destruct = netlink_sock_destruct;
648 sk->sk_protocol = protocol;
649 return 0;
650}
651
652static int netlink_create(struct net *net, struct socket *sock, int protocol,
653 int kern)
654{
655 struct module *module = NULL;
656 struct mutex *cb_mutex;
657 struct netlink_sock *nlk;
658 int (*bind)(struct net *net, int group);
659 void (*unbind)(struct net *net, int group);
660 int err = 0;
661
662 sock->state = SS_UNCONNECTED;
663
664 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
665 return -ESOCKTNOSUPPORT;
666
667 if (protocol < 0 || protocol >= MAX_LINKS)
668 return -EPROTONOSUPPORT;
669 protocol = array_index_nospec(protocol, MAX_LINKS);
670
671 netlink_lock_table();
672#ifdef CONFIG_MODULES
673 if (!nl_table[protocol].registered) {
674 netlink_unlock_table();
675 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
676 netlink_lock_table();
677 }
678#endif
679 if (nl_table[protocol].registered &&
680 try_module_get(nl_table[protocol].module))
681 module = nl_table[protocol].module;
682 else
683 err = -EPROTONOSUPPORT;
684 cb_mutex = nl_table[protocol].cb_mutex;
685 bind = nl_table[protocol].bind;
686 unbind = nl_table[protocol].unbind;
687 netlink_unlock_table();
688
689 if (err < 0)
690 goto out;
691
692 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
693 if (err < 0)
694 goto out_module;
695
696 local_bh_disable();
697 sock_prot_inuse_add(net, &netlink_proto, 1);
698 local_bh_enable();
699
700 nlk = nlk_sk(sock->sk);
701 nlk->module = module;
702 nlk->netlink_bind = bind;
703 nlk->netlink_unbind = unbind;
704out:
705 return err;
706
707out_module:
708 module_put(module);
709 goto out;
710}
711
712static void deferred_put_nlk_sk(struct rcu_head *head)
713{
714 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
715 struct sock *sk = &nlk->sk;
716
717 kfree(nlk->groups);
718 nlk->groups = NULL;
719
720 if (!refcount_dec_and_test(&sk->sk_refcnt))
721 return;
722
723 if (nlk->cb_running && nlk->cb.done) {
724 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
725 schedule_work(&nlk->work);
726 return;
727 }
728
729 sk_free(sk);
730}
731
732static int netlink_release(struct socket *sock)
733{
734 struct sock *sk = sock->sk;
735 struct netlink_sock *nlk;
736
737 if (!sk)
738 return 0;
739
740 netlink_remove(sk);
741 sock_orphan(sk);
742 nlk = nlk_sk(sk);
743
744 /*
745 * OK. Socket is unlinked, any packets that arrive now
746 * will be purged.
747 */
748
749 /* must not acquire netlink_table_lock in any way again before unbind
750 * and notifying genetlink is done as otherwise it might deadlock
751 */
752 if (nlk->netlink_unbind) {
753 int i;
754
755 for (i = 0; i < nlk->ngroups; i++)
756 if (test_bit(i, nlk->groups))
757 nlk->netlink_unbind(sock_net(sk), i + 1);
758 }
759 if (sk->sk_protocol == NETLINK_GENERIC &&
760 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
761 wake_up(&genl_sk_destructing_waitq);
762
763 sock->sk = NULL;
764 wake_up_interruptible_all(&nlk->wait);
765
766 skb_queue_purge(&sk->sk_write_queue);
767
768 if (nlk->portid && nlk->bound) {
769 struct netlink_notify n = {
770 .net = sock_net(sk),
771 .protocol = sk->sk_protocol,
772 .portid = nlk->portid,
773 };
774 blocking_notifier_call_chain(&netlink_chain,
775 NETLINK_URELEASE, &n);
776 }
777
778 module_put(nlk->module);
779
780 if (netlink_is_kernel(sk)) {
781 netlink_table_grab();
782 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
783 if (--nl_table[sk->sk_protocol].registered == 0) {
784 struct listeners *old;
785
786 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
787 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
788 kfree_rcu(old, rcu);
789 nl_table[sk->sk_protocol].module = NULL;
790 nl_table[sk->sk_protocol].bind = NULL;
791 nl_table[sk->sk_protocol].unbind = NULL;
792 nl_table[sk->sk_protocol].flags = 0;
793 nl_table[sk->sk_protocol].registered = 0;
794 }
795 netlink_table_ungrab();
796 }
797
798 local_bh_disable();
799 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
800 local_bh_enable();
801 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
802 return 0;
803}
804
805static int netlink_autobind(struct socket *sock)
806{
807 struct sock *sk = sock->sk;
808 struct net *net = sock_net(sk);
809 struct netlink_table *table = &nl_table[sk->sk_protocol];
810 s32 portid = task_tgid_vnr(current);
811 int err;
812 s32 rover = -4096;
813 bool ok;
814
815retry:
816 cond_resched();
817 rcu_read_lock();
818 ok = !__netlink_lookup(table, portid, net);
819 rcu_read_unlock();
820 if (!ok) {
821 /* Bind collision, search negative portid values. */
822 if (rover == -4096)
823 /* rover will be in range [S32_MIN, -4097] */
824 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
825 else if (rover >= -4096)
826 rover = -4097;
827 portid = rover--;
828 goto retry;
829 }
830
831 err = netlink_insert(sk, portid);
832 if (err == -EADDRINUSE)
833 goto retry;
834
835 /* If 2 threads race to autobind, that is fine. */
836 if (err == -EBUSY)
837 err = 0;
838
839 return err;
840}
841
842/**
843 * __netlink_ns_capable - General netlink message capability test
844 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
845 * @user_ns: The user namespace of the capability to use
846 * @cap: The capability to use
847 *
848 * Test to see if the opener of the socket we received the message
849 * from had when the netlink socket was created and the sender of the
850 * message has has the capability @cap in the user namespace @user_ns.
851 */
852bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
853 struct user_namespace *user_ns, int cap)
854{
855 return ((nsp->flags & NETLINK_SKB_DST) ||
856 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
857 ns_capable(user_ns, cap);
858}
859EXPORT_SYMBOL(__netlink_ns_capable);
860
861/**
862 * netlink_ns_capable - General netlink message capability test
863 * @skb: socket buffer holding a netlink command from userspace
864 * @user_ns: The user namespace of the capability to use
865 * @cap: The capability to use
866 *
867 * Test to see if the opener of the socket we received the message
868 * from had when the netlink socket was created and the sender of the
869 * message has has the capability @cap in the user namespace @user_ns.
870 */
871bool netlink_ns_capable(const struct sk_buff *skb,
872 struct user_namespace *user_ns, int cap)
873{
874 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
875}
876EXPORT_SYMBOL(netlink_ns_capable);
877
878/**
879 * netlink_capable - Netlink global message capability test
880 * @skb: socket buffer holding a netlink command from userspace
881 * @cap: The capability to use
882 *
883 * Test to see if the opener of the socket we received the message
884 * from had when the netlink socket was created and the sender of the
885 * message has has the capability @cap in all user namespaces.
886 */
887bool netlink_capable(const struct sk_buff *skb, int cap)
888{
889 return netlink_ns_capable(skb, &init_user_ns, cap);
890}
891EXPORT_SYMBOL(netlink_capable);
892
893/**
894 * netlink_net_capable - Netlink network namespace message capability test
895 * @skb: socket buffer holding a netlink command from userspace
896 * @cap: The capability to use
897 *
898 * Test to see if the opener of the socket we received the message
899 * from had when the netlink socket was created and the sender of the
900 * message has has the capability @cap over the network namespace of
901 * the socket we received the message from.
902 */
903bool netlink_net_capable(const struct sk_buff *skb, int cap)
904{
905 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
906}
907EXPORT_SYMBOL(netlink_net_capable);
908
909static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
910{
911 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
912 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
913}
914
915static void
916netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
917{
918 struct netlink_sock *nlk = nlk_sk(sk);
919
920 if (nlk->subscriptions && !subscriptions)
921 __sk_del_bind_node(sk);
922 else if (!nlk->subscriptions && subscriptions)
923 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
924 nlk->subscriptions = subscriptions;
925}
926
927static int netlink_realloc_groups(struct sock *sk)
928{
929 struct netlink_sock *nlk = nlk_sk(sk);
930 unsigned int groups;
931 unsigned long *new_groups;
932 int err = 0;
933
934 netlink_table_grab();
935
936 groups = nl_table[sk->sk_protocol].groups;
937 if (!nl_table[sk->sk_protocol].registered) {
938 err = -ENOENT;
939 goto out_unlock;
940 }
941
942 if (nlk->ngroups >= groups)
943 goto out_unlock;
944
945 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
946 if (new_groups == NULL) {
947 err = -ENOMEM;
948 goto out_unlock;
949 }
950 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
951 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
952
953 nlk->groups = new_groups;
954 nlk->ngroups = groups;
955 out_unlock:
956 netlink_table_ungrab();
957 return err;
958}
959
960static void netlink_undo_bind(int group, long unsigned int groups,
961 struct sock *sk)
962{
963 struct netlink_sock *nlk = nlk_sk(sk);
964 int undo;
965
966 if (!nlk->netlink_unbind)
967 return;
968
969 for (undo = 0; undo < group; undo++)
970 if (test_bit(undo, &groups))
971 nlk->netlink_unbind(sock_net(sk), undo + 1);
972}
973
974static int netlink_bind(struct socket *sock, struct sockaddr *addr,
975 int addr_len)
976{
977 struct sock *sk = sock->sk;
978 struct net *net = sock_net(sk);
979 struct netlink_sock *nlk = nlk_sk(sk);
980 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
981 int err = 0;
982 unsigned long groups;
983 bool bound;
984
985 if (addr_len < sizeof(struct sockaddr_nl))
986 return -EINVAL;
987
988 if (nladdr->nl_family != AF_NETLINK)
989 return -EINVAL;
990 groups = nladdr->nl_groups;
991
992 /* Only superuser is allowed to listen multicasts */
993 if (groups) {
994 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
995 return -EPERM;
996 err = netlink_realloc_groups(sk);
997 if (err)
998 return err;
999 }
1000
1001 if (nlk->ngroups < BITS_PER_LONG)
1002 groups &= (1UL << nlk->ngroups) - 1;
1003
1004 bound = nlk->bound;
1005 if (bound) {
1006 /* Ensure nlk->portid is up-to-date. */
1007 smp_rmb();
1008
1009 if (nladdr->nl_pid != nlk->portid)
1010 return -EINVAL;
1011 }
1012
1013 netlink_lock_table();
1014 if (nlk->netlink_bind && groups) {
1015 int group;
1016
1017 for (group = 0; group < nlk->ngroups; group++) {
1018 if (!test_bit(group, &groups))
1019 continue;
1020 err = nlk->netlink_bind(net, group + 1);
1021 if (!err)
1022 continue;
1023 netlink_undo_bind(group, groups, sk);
1024 goto unlock;
1025 }
1026 }
1027
1028 /* No need for barriers here as we return to user-space without
1029 * using any of the bound attributes.
1030 */
1031 if (!bound) {
1032 err = nladdr->nl_pid ?
1033 netlink_insert(sk, nladdr->nl_pid) :
1034 netlink_autobind(sock);
1035 if (err) {
1036 netlink_undo_bind(nlk->ngroups, groups, sk);
1037 goto unlock;
1038 }
1039 }
1040
1041 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1042 goto unlock;
1043 netlink_unlock_table();
1044
1045 netlink_table_grab();
1046 netlink_update_subscriptions(sk, nlk->subscriptions +
1047 hweight32(groups) -
1048 hweight32(nlk->groups[0]));
1049 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1050 netlink_update_listeners(sk);
1051 netlink_table_ungrab();
1052
1053 return 0;
1054
1055unlock:
1056 netlink_unlock_table();
1057 return err;
1058}
1059
1060static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1061 int alen, int flags)
1062{
1063 int err = 0;
1064 struct sock *sk = sock->sk;
1065 struct netlink_sock *nlk = nlk_sk(sk);
1066 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1067
1068 if (alen < sizeof(addr->sa_family))
1069 return -EINVAL;
1070
1071 if (addr->sa_family == AF_UNSPEC) {
1072 sk->sk_state = NETLINK_UNCONNECTED;
1073 nlk->dst_portid = 0;
1074 nlk->dst_group = 0;
1075 return 0;
1076 }
1077 if (addr->sa_family != AF_NETLINK)
1078 return -EINVAL;
1079
1080 if (alen < sizeof(struct sockaddr_nl))
1081 return -EINVAL;
1082
1083 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1084 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1085 return -EPERM;
1086
1087 /* No need for barriers here as we return to user-space without
1088 * using any of the bound attributes.
1089 */
1090 if (!nlk->bound)
1091 err = netlink_autobind(sock);
1092
1093 if (err == 0) {
1094 sk->sk_state = NETLINK_CONNECTED;
1095 nlk->dst_portid = nladdr->nl_pid;
1096 nlk->dst_group = ffs(nladdr->nl_groups);
1097 }
1098
1099 return err;
1100}
1101
1102static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1103 int peer)
1104{
1105 struct sock *sk = sock->sk;
1106 struct netlink_sock *nlk = nlk_sk(sk);
1107 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1108
1109 nladdr->nl_family = AF_NETLINK;
1110 nladdr->nl_pad = 0;
1111
1112 if (peer) {
1113 nladdr->nl_pid = nlk->dst_portid;
1114 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1115 } else {
1116 nladdr->nl_pid = nlk->portid;
1117 netlink_lock_table();
1118 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1119 netlink_unlock_table();
1120 }
1121 return sizeof(*nladdr);
1122}
1123
1124static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1125 unsigned long arg)
1126{
1127 /* try to hand this ioctl down to the NIC drivers.
1128 */
1129 return -ENOIOCTLCMD;
1130}
1131
1132static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1133{
1134 struct sock *sock;
1135 struct netlink_sock *nlk;
1136
1137 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1138 if (!sock)
1139 return ERR_PTR(-ECONNREFUSED);
1140
1141 /* Don't bother queuing skb if kernel socket has no input function */
1142 nlk = nlk_sk(sock);
1143 if (sock->sk_state == NETLINK_CONNECTED &&
1144 nlk->dst_portid != nlk_sk(ssk)->portid) {
1145 sock_put(sock);
1146 return ERR_PTR(-ECONNREFUSED);
1147 }
1148 return sock;
1149}
1150
1151struct sock *netlink_getsockbyfilp(struct file *filp)
1152{
1153 struct inode *inode = file_inode(filp);
1154 struct sock *sock;
1155
1156 if (!S_ISSOCK(inode->i_mode))
1157 return ERR_PTR(-ENOTSOCK);
1158
1159 sock = SOCKET_I(inode)->sk;
1160 if (sock->sk_family != AF_NETLINK)
1161 return ERR_PTR(-EINVAL);
1162
1163 sock_hold(sock);
1164 return sock;
1165}
1166
1167static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1168 int broadcast)
1169{
1170 struct sk_buff *skb;
1171 void *data;
1172
1173 if (size <= NLMSG_GOODSIZE || broadcast)
1174 return alloc_skb(size, GFP_KERNEL);
1175
1176 size = SKB_DATA_ALIGN(size) +
1177 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1178
1179 data = vmalloc(size);
1180 if (data == NULL)
1181 return NULL;
1182
1183 skb = __build_skb(data, size);
1184 if (skb == NULL)
1185 vfree(data);
1186 else
1187 skb->destructor = netlink_skb_destructor;
1188
1189 return skb;
1190}
1191
1192/*
1193 * Attach a skb to a netlink socket.
1194 * The caller must hold a reference to the destination socket. On error, the
1195 * reference is dropped. The skb is not send to the destination, just all
1196 * all error checks are performed and memory in the queue is reserved.
1197 * Return values:
1198 * < 0: error. skb freed, reference to sock dropped.
1199 * 0: continue
1200 * 1: repeat lookup - reference dropped while waiting for socket memory.
1201 */
1202int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1203 long *timeo, struct sock *ssk)
1204{
1205 struct netlink_sock *nlk;
1206
1207 nlk = nlk_sk(sk);
1208
1209 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1210 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1211 DECLARE_WAITQUEUE(wait, current);
1212 if (!*timeo) {
1213 if (!ssk || netlink_is_kernel(ssk))
1214 netlink_overrun(sk);
1215 sock_put(sk);
1216 kfree_skb(skb);
1217 return -EAGAIN;
1218 }
1219
1220 __set_current_state(TASK_INTERRUPTIBLE);
1221 add_wait_queue(&nlk->wait, &wait);
1222
1223 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1224 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1225 !sock_flag(sk, SOCK_DEAD))
1226 *timeo = schedule_timeout(*timeo);
1227
1228 __set_current_state(TASK_RUNNING);
1229 remove_wait_queue(&nlk->wait, &wait);
1230 sock_put(sk);
1231
1232 if (signal_pending(current)) {
1233 kfree_skb(skb);
1234 return sock_intr_errno(*timeo);
1235 }
1236 return 1;
1237 }
1238 netlink_skb_set_owner_r(skb, sk);
1239 return 0;
1240}
1241
1242static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1243{
1244 int len = skb->len;
1245
1246 netlink_deliver_tap(sock_net(sk), skb);
1247
1248 skb_queue_tail(&sk->sk_receive_queue, skb);
1249 sk->sk_data_ready(sk);
1250 return len;
1251}
1252
1253int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1254{
1255 int len = __netlink_sendskb(sk, skb);
1256
1257 sock_put(sk);
1258 return len;
1259}
1260
1261void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1262{
1263 kfree_skb(skb);
1264 sock_put(sk);
1265}
1266
1267static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1268{
1269 int delta;
1270
1271 WARN_ON(skb->sk != NULL);
1272 delta = skb->end - skb->tail;
1273 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1274 return skb;
1275
1276 if (skb_shared(skb)) {
1277 struct sk_buff *nskb = skb_clone(skb, allocation);
1278 if (!nskb)
1279 return skb;
1280 consume_skb(skb);
1281 skb = nskb;
1282 }
1283
1284 pskb_expand_head(skb, 0, -delta,
1285 (allocation & ~__GFP_DIRECT_RECLAIM) |
1286 __GFP_NOWARN | __GFP_NORETRY);
1287 return skb;
1288}
1289
1290static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1291 struct sock *ssk)
1292{
1293 int ret;
1294 struct netlink_sock *nlk = nlk_sk(sk);
1295
1296 ret = -ECONNREFUSED;
1297 if (nlk->netlink_rcv != NULL) {
1298 ret = skb->len;
1299 netlink_skb_set_owner_r(skb, sk);
1300 NETLINK_CB(skb).sk = ssk;
1301 netlink_deliver_tap_kernel(sk, ssk, skb);
1302 nlk->netlink_rcv(skb);
1303 consume_skb(skb);
1304 } else {
1305 kfree_skb(skb);
1306 }
1307 sock_put(sk);
1308 return ret;
1309}
1310
1311int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1312 u32 portid, int nonblock)
1313{
1314 struct sock *sk;
1315 int err;
1316 long timeo;
1317
1318 skb = netlink_trim(skb, gfp_any());
1319
1320 timeo = sock_sndtimeo(ssk, nonblock);
1321retry:
1322 sk = netlink_getsockbyportid(ssk, portid);
1323 if (IS_ERR(sk)) {
1324 kfree_skb(skb);
1325 return PTR_ERR(sk);
1326 }
1327 if (netlink_is_kernel(sk))
1328 return netlink_unicast_kernel(sk, skb, ssk);
1329
1330 if (sk_filter(sk, skb)) {
1331 err = skb->len;
1332 kfree_skb(skb);
1333 sock_put(sk);
1334 return err;
1335 }
1336
1337 err = netlink_attachskb(sk, skb, &timeo, ssk);
1338 if (err == 1)
1339 goto retry;
1340 if (err)
1341 return err;
1342
1343 return netlink_sendskb(sk, skb);
1344}
1345EXPORT_SYMBOL(netlink_unicast);
1346
1347int netlink_has_listeners(struct sock *sk, unsigned int group)
1348{
1349 int res = 0;
1350 struct listeners *listeners;
1351
1352 BUG_ON(!netlink_is_kernel(sk));
1353
1354 rcu_read_lock();
1355 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1356
1357 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1358 res = test_bit(group - 1, listeners->masks);
1359
1360 rcu_read_unlock();
1361
1362 return res;
1363}
1364EXPORT_SYMBOL_GPL(netlink_has_listeners);
1365
1366bool netlink_strict_get_check(struct sk_buff *skb)
1367{
1368 const struct netlink_sock *nlk = nlk_sk(NETLINK_CB(skb).sk);
1369
1370 return nlk->flags & NETLINK_F_STRICT_CHK;
1371}
1372EXPORT_SYMBOL_GPL(netlink_strict_get_check);
1373
1374static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1375{
1376 struct netlink_sock *nlk = nlk_sk(sk);
1377
1378 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1379 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1380 netlink_skb_set_owner_r(skb, sk);
1381 __netlink_sendskb(sk, skb);
1382 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1383 }
1384 return -1;
1385}
1386
1387struct netlink_broadcast_data {
1388 struct sock *exclude_sk;
1389 struct net *net;
1390 u32 portid;
1391 u32 group;
1392 int failure;
1393 int delivery_failure;
1394 int congested;
1395 int delivered;
1396 gfp_t allocation;
1397 struct sk_buff *skb, *skb2;
1398 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1399 void *tx_data;
1400};
1401
1402static void do_one_broadcast(struct sock *sk,
1403 struct netlink_broadcast_data *p)
1404{
1405 struct netlink_sock *nlk = nlk_sk(sk);
1406 int val;
1407
1408 if (p->exclude_sk == sk)
1409 return;
1410
1411 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1412 !test_bit(p->group - 1, nlk->groups))
1413 return;
1414
1415 if (!net_eq(sock_net(sk), p->net)) {
1416 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1417 return;
1418
1419 if (!peernet_has_id(sock_net(sk), p->net))
1420 return;
1421
1422 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1423 CAP_NET_BROADCAST))
1424 return;
1425 }
1426
1427 if (p->failure) {
1428 netlink_overrun(sk);
1429 return;
1430 }
1431
1432 sock_hold(sk);
1433 if (p->skb2 == NULL) {
1434 if (skb_shared(p->skb)) {
1435 p->skb2 = skb_clone(p->skb, p->allocation);
1436 } else {
1437 p->skb2 = skb_get(p->skb);
1438 /*
1439 * skb ownership may have been set when
1440 * delivered to a previous socket.
1441 */
1442 skb_orphan(p->skb2);
1443 }
1444 }
1445 if (p->skb2 == NULL) {
1446 netlink_overrun(sk);
1447 /* Clone failed. Notify ALL listeners. */
1448 p->failure = 1;
1449 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1450 p->delivery_failure = 1;
1451 goto out;
1452 }
1453 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1454 kfree_skb(p->skb2);
1455 p->skb2 = NULL;
1456 goto out;
1457 }
1458 if (sk_filter(sk, p->skb2)) {
1459 kfree_skb(p->skb2);
1460 p->skb2 = NULL;
1461 goto out;
1462 }
1463 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1464 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1465 NETLINK_CB(p->skb2).nsid_is_set = true;
1466 val = netlink_broadcast_deliver(sk, p->skb2);
1467 if (val < 0) {
1468 netlink_overrun(sk);
1469 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1470 p->delivery_failure = 1;
1471 } else {
1472 p->congested |= val;
1473 p->delivered = 1;
1474 p->skb2 = NULL;
1475 }
1476out:
1477 sock_put(sk);
1478}
1479
1480int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1481 u32 group, gfp_t allocation,
1482 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1483 void *filter_data)
1484{
1485 struct net *net = sock_net(ssk);
1486 struct netlink_broadcast_data info;
1487 struct sock *sk;
1488
1489 skb = netlink_trim(skb, allocation);
1490
1491 info.exclude_sk = ssk;
1492 info.net = net;
1493 info.portid = portid;
1494 info.group = group;
1495 info.failure = 0;
1496 info.delivery_failure = 0;
1497 info.congested = 0;
1498 info.delivered = 0;
1499 info.allocation = allocation;
1500 info.skb = skb;
1501 info.skb2 = NULL;
1502 info.tx_filter = filter;
1503 info.tx_data = filter_data;
1504
1505 /* While we sleep in clone, do not allow to change socket list */
1506
1507 netlink_lock_table();
1508
1509 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1510 do_one_broadcast(sk, &info);
1511
1512 consume_skb(skb);
1513
1514 netlink_unlock_table();
1515
1516 if (info.delivery_failure) {
1517 kfree_skb(info.skb2);
1518 return -ENOBUFS;
1519 }
1520 consume_skb(info.skb2);
1521
1522 if (info.delivered) {
1523 if (info.congested && gfpflags_allow_blocking(allocation))
1524 yield();
1525 return 0;
1526 }
1527 return -ESRCH;
1528}
1529EXPORT_SYMBOL(netlink_broadcast_filtered);
1530
1531int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1532 u32 group, gfp_t allocation)
1533{
1534 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1535 NULL, NULL);
1536}
1537EXPORT_SYMBOL(netlink_broadcast);
1538
1539struct netlink_set_err_data {
1540 struct sock *exclude_sk;
1541 u32 portid;
1542 u32 group;
1543 int code;
1544};
1545
1546static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1547{
1548 struct netlink_sock *nlk = nlk_sk(sk);
1549 int ret = 0;
1550
1551 if (sk == p->exclude_sk)
1552 goto out;
1553
1554 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1555 goto out;
1556
1557 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1558 !test_bit(p->group - 1, nlk->groups))
1559 goto out;
1560
1561 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1562 ret = 1;
1563 goto out;
1564 }
1565
1566 sk->sk_err = p->code;
1567 sk->sk_error_report(sk);
1568out:
1569 return ret;
1570}
1571
1572/**
1573 * netlink_set_err - report error to broadcast listeners
1574 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1575 * @portid: the PORTID of a process that we want to skip (if any)
1576 * @group: the broadcast group that will notice the error
1577 * @code: error code, must be negative (as usual in kernelspace)
1578 *
1579 * This function returns the number of broadcast listeners that have set the
1580 * NETLINK_NO_ENOBUFS socket option.
1581 */
1582int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1583{
1584 struct netlink_set_err_data info;
1585 struct sock *sk;
1586 int ret = 0;
1587
1588 info.exclude_sk = ssk;
1589 info.portid = portid;
1590 info.group = group;
1591 /* sk->sk_err wants a positive error value */
1592 info.code = -code;
1593
1594 read_lock(&nl_table_lock);
1595
1596 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1597 ret += do_one_set_err(sk, &info);
1598
1599 read_unlock(&nl_table_lock);
1600 return ret;
1601}
1602EXPORT_SYMBOL(netlink_set_err);
1603
1604/* must be called with netlink table grabbed */
1605static void netlink_update_socket_mc(struct netlink_sock *nlk,
1606 unsigned int group,
1607 int is_new)
1608{
1609 int old, new = !!is_new, subscriptions;
1610
1611 old = test_bit(group - 1, nlk->groups);
1612 subscriptions = nlk->subscriptions - old + new;
1613 if (new)
1614 __set_bit(group - 1, nlk->groups);
1615 else
1616 __clear_bit(group - 1, nlk->groups);
1617 netlink_update_subscriptions(&nlk->sk, subscriptions);
1618 netlink_update_listeners(&nlk->sk);
1619}
1620
1621static int netlink_setsockopt(struct socket *sock, int level, int optname,
1622 char __user *optval, unsigned int optlen)
1623{
1624 struct sock *sk = sock->sk;
1625 struct netlink_sock *nlk = nlk_sk(sk);
1626 unsigned int val = 0;
1627 int err;
1628
1629 if (level != SOL_NETLINK)
1630 return -ENOPROTOOPT;
1631
1632 if (optlen >= sizeof(int) &&
1633 get_user(val, (unsigned int __user *)optval))
1634 return -EFAULT;
1635
1636 switch (optname) {
1637 case NETLINK_PKTINFO:
1638 if (val)
1639 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1640 else
1641 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1642 err = 0;
1643 break;
1644 case NETLINK_ADD_MEMBERSHIP:
1645 case NETLINK_DROP_MEMBERSHIP: {
1646 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1647 return -EPERM;
1648 err = netlink_realloc_groups(sk);
1649 if (err)
1650 return err;
1651 if (!val || val - 1 >= nlk->ngroups)
1652 return -EINVAL;
1653 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1654 err = nlk->netlink_bind(sock_net(sk), val);
1655 if (err)
1656 return err;
1657 }
1658 netlink_table_grab();
1659 netlink_update_socket_mc(nlk, val,
1660 optname == NETLINK_ADD_MEMBERSHIP);
1661 netlink_table_ungrab();
1662 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1663 nlk->netlink_unbind(sock_net(sk), val);
1664
1665 err = 0;
1666 break;
1667 }
1668 case NETLINK_BROADCAST_ERROR:
1669 if (val)
1670 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1671 else
1672 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1673 err = 0;
1674 break;
1675 case NETLINK_NO_ENOBUFS:
1676 if (val) {
1677 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1678 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1679 wake_up_interruptible(&nlk->wait);
1680 } else {
1681 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1682 }
1683 err = 0;
1684 break;
1685 case NETLINK_LISTEN_ALL_NSID:
1686 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1687 return -EPERM;
1688
1689 if (val)
1690 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1691 else
1692 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1693 err = 0;
1694 break;
1695 case NETLINK_CAP_ACK:
1696 if (val)
1697 nlk->flags |= NETLINK_F_CAP_ACK;
1698 else
1699 nlk->flags &= ~NETLINK_F_CAP_ACK;
1700 err = 0;
1701 break;
1702 case NETLINK_EXT_ACK:
1703 if (val)
1704 nlk->flags |= NETLINK_F_EXT_ACK;
1705 else
1706 nlk->flags &= ~NETLINK_F_EXT_ACK;
1707 err = 0;
1708 break;
1709 case NETLINK_GET_STRICT_CHK:
1710 if (val)
1711 nlk->flags |= NETLINK_F_STRICT_CHK;
1712 else
1713 nlk->flags &= ~NETLINK_F_STRICT_CHK;
1714 err = 0;
1715 break;
1716 default:
1717 err = -ENOPROTOOPT;
1718 }
1719 return err;
1720}
1721
1722static int netlink_getsockopt(struct socket *sock, int level, int optname,
1723 char __user *optval, int __user *optlen)
1724{
1725 struct sock *sk = sock->sk;
1726 struct netlink_sock *nlk = nlk_sk(sk);
1727 int len, val, err;
1728
1729 if (level != SOL_NETLINK)
1730 return -ENOPROTOOPT;
1731
1732 if (get_user(len, optlen))
1733 return -EFAULT;
1734 if (len < 0)
1735 return -EINVAL;
1736
1737 switch (optname) {
1738 case NETLINK_PKTINFO:
1739 if (len < sizeof(int))
1740 return -EINVAL;
1741 len = sizeof(int);
1742 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1743 if (put_user(len, optlen) ||
1744 put_user(val, optval))
1745 return -EFAULT;
1746 err = 0;
1747 break;
1748 case NETLINK_BROADCAST_ERROR:
1749 if (len < sizeof(int))
1750 return -EINVAL;
1751 len = sizeof(int);
1752 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1753 if (put_user(len, optlen) ||
1754 put_user(val, optval))
1755 return -EFAULT;
1756 err = 0;
1757 break;
1758 case NETLINK_NO_ENOBUFS:
1759 if (len < sizeof(int))
1760 return -EINVAL;
1761 len = sizeof(int);
1762 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1763 if (put_user(len, optlen) ||
1764 put_user(val, optval))
1765 return -EFAULT;
1766 err = 0;
1767 break;
1768 case NETLINK_LIST_MEMBERSHIPS: {
1769 int pos, idx, shift;
1770
1771 err = 0;
1772 netlink_lock_table();
1773 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1774 if (len - pos < sizeof(u32))
1775 break;
1776
1777 idx = pos / sizeof(unsigned long);
1778 shift = (pos % sizeof(unsigned long)) * 8;
1779 if (put_user((u32)(nlk->groups[idx] >> shift),
1780 (u32 __user *)(optval + pos))) {
1781 err = -EFAULT;
1782 break;
1783 }
1784 }
1785 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1786 err = -EFAULT;
1787 netlink_unlock_table();
1788 break;
1789 }
1790 case NETLINK_CAP_ACK:
1791 if (len < sizeof(int))
1792 return -EINVAL;
1793 len = sizeof(int);
1794 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1795 if (put_user(len, optlen) ||
1796 put_user(val, optval))
1797 return -EFAULT;
1798 err = 0;
1799 break;
1800 case NETLINK_EXT_ACK:
1801 if (len < sizeof(int))
1802 return -EINVAL;
1803 len = sizeof(int);
1804 val = nlk->flags & NETLINK_F_EXT_ACK ? 1 : 0;
1805 if (put_user(len, optlen) || put_user(val, optval))
1806 return -EFAULT;
1807 err = 0;
1808 break;
1809 case NETLINK_GET_STRICT_CHK:
1810 if (len < sizeof(int))
1811 return -EINVAL;
1812 len = sizeof(int);
1813 val = nlk->flags & NETLINK_F_STRICT_CHK ? 1 : 0;
1814 if (put_user(len, optlen) || put_user(val, optval))
1815 return -EFAULT;
1816 err = 0;
1817 break;
1818 default:
1819 err = -ENOPROTOOPT;
1820 }
1821 return err;
1822}
1823
1824static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1825{
1826 struct nl_pktinfo info;
1827
1828 info.group = NETLINK_CB(skb).dst_group;
1829 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1830}
1831
1832static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1833 struct sk_buff *skb)
1834{
1835 if (!NETLINK_CB(skb).nsid_is_set)
1836 return;
1837
1838 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1839 &NETLINK_CB(skb).nsid);
1840}
1841
1842static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1843{
1844 struct sock *sk = sock->sk;
1845 struct netlink_sock *nlk = nlk_sk(sk);
1846 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1847 u32 dst_portid;
1848 u32 dst_group;
1849 struct sk_buff *skb;
1850 int err;
1851 struct scm_cookie scm;
1852 u32 netlink_skb_flags = 0;
1853
1854 if (msg->msg_flags&MSG_OOB)
1855 return -EOPNOTSUPP;
1856
1857 err = scm_send(sock, msg, &scm, true);
1858 if (err < 0)
1859 return err;
1860
1861 if (msg->msg_namelen) {
1862 err = -EINVAL;
1863 if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1864 goto out;
1865 if (addr->nl_family != AF_NETLINK)
1866 goto out;
1867 dst_portid = addr->nl_pid;
1868 dst_group = ffs(addr->nl_groups);
1869 err = -EPERM;
1870 if ((dst_group || dst_portid) &&
1871 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1872 goto out;
1873 netlink_skb_flags |= NETLINK_SKB_DST;
1874 } else {
1875 dst_portid = nlk->dst_portid;
1876 dst_group = nlk->dst_group;
1877 }
1878
1879 if (!nlk->bound) {
1880 err = netlink_autobind(sock);
1881 if (err)
1882 goto out;
1883 } else {
1884 /* Ensure nlk is hashed and visible. */
1885 smp_rmb();
1886 }
1887
1888 err = -EMSGSIZE;
1889 if (len > sk->sk_sndbuf - 32)
1890 goto out;
1891 err = -ENOBUFS;
1892 skb = netlink_alloc_large_skb(len, dst_group);
1893 if (skb == NULL)
1894 goto out;
1895
1896 NETLINK_CB(skb).portid = nlk->portid;
1897 NETLINK_CB(skb).dst_group = dst_group;
1898 NETLINK_CB(skb).creds = scm.creds;
1899 NETLINK_CB(skb).flags = netlink_skb_flags;
1900
1901 err = -EFAULT;
1902 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1903 kfree_skb(skb);
1904 goto out;
1905 }
1906
1907 err = security_netlink_send(sk, skb);
1908 if (err) {
1909 kfree_skb(skb);
1910 goto out;
1911 }
1912
1913 if (dst_group) {
1914 refcount_inc(&skb->users);
1915 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1916 }
1917 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
1918
1919out:
1920 scm_destroy(&scm);
1921 return err;
1922}
1923
1924static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1925 int flags)
1926{
1927 struct scm_cookie scm;
1928 struct sock *sk = sock->sk;
1929 struct netlink_sock *nlk = nlk_sk(sk);
1930 int noblock = flags&MSG_DONTWAIT;
1931 size_t copied;
1932 struct sk_buff *skb, *data_skb;
1933 int err, ret;
1934
1935 if (flags&MSG_OOB)
1936 return -EOPNOTSUPP;
1937
1938 copied = 0;
1939
1940 skb = skb_recv_datagram(sk, flags, noblock, &err);
1941 if (skb == NULL)
1942 goto out;
1943
1944 data_skb = skb;
1945
1946#ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1947 if (unlikely(skb_shinfo(skb)->frag_list)) {
1948 /*
1949 * If this skb has a frag_list, then here that means that we
1950 * will have to use the frag_list skb's data for compat tasks
1951 * and the regular skb's data for normal (non-compat) tasks.
1952 *
1953 * If we need to send the compat skb, assign it to the
1954 * 'data_skb' variable so that it will be used below for data
1955 * copying. We keep 'skb' for everything else, including
1956 * freeing both later.
1957 */
1958 if (flags & MSG_CMSG_COMPAT)
1959 data_skb = skb_shinfo(skb)->frag_list;
1960 }
1961#endif
1962
1963 /* Record the max length of recvmsg() calls for future allocations */
1964 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1965 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1966 SKB_WITH_OVERHEAD(32768));
1967
1968 copied = data_skb->len;
1969 if (len < copied) {
1970 msg->msg_flags |= MSG_TRUNC;
1971 copied = len;
1972 }
1973
1974 skb_reset_transport_header(data_skb);
1975 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1976
1977 if (msg->msg_name) {
1978 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1979 addr->nl_family = AF_NETLINK;
1980 addr->nl_pad = 0;
1981 addr->nl_pid = NETLINK_CB(skb).portid;
1982 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1983 msg->msg_namelen = sizeof(*addr);
1984 }
1985
1986 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1987 netlink_cmsg_recv_pktinfo(msg, skb);
1988 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1989 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1990
1991 memset(&scm, 0, sizeof(scm));
1992 scm.creds = *NETLINK_CREDS(skb);
1993 if (flags & MSG_TRUNC)
1994 copied = data_skb->len;
1995
1996 skb_free_datagram(sk, skb);
1997
1998 if (nlk->cb_running &&
1999 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2000 ret = netlink_dump(sk);
2001 if (ret) {
2002 sk->sk_err = -ret;
2003 sk->sk_error_report(sk);
2004 }
2005 }
2006
2007 scm_recv(sock, msg, &scm, flags);
2008out:
2009 netlink_rcv_wake(sk);
2010 return err ? : copied;
2011}
2012
2013static void netlink_data_ready(struct sock *sk)
2014{
2015 BUG();
2016}
2017
2018/*
2019 * We export these functions to other modules. They provide a
2020 * complete set of kernel non-blocking support for message
2021 * queueing.
2022 */
2023
2024struct sock *
2025__netlink_kernel_create(struct net *net, int unit, struct module *module,
2026 struct netlink_kernel_cfg *cfg)
2027{
2028 struct socket *sock;
2029 struct sock *sk;
2030 struct netlink_sock *nlk;
2031 struct listeners *listeners = NULL;
2032 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2033 unsigned int groups;
2034
2035 BUG_ON(!nl_table);
2036
2037 if (unit < 0 || unit >= MAX_LINKS)
2038 return NULL;
2039
2040 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2041 return NULL;
2042
2043 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2044 goto out_sock_release_nosk;
2045
2046 sk = sock->sk;
2047
2048 if (!cfg || cfg->groups < 32)
2049 groups = 32;
2050 else
2051 groups = cfg->groups;
2052
2053 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2054 if (!listeners)
2055 goto out_sock_release;
2056
2057 sk->sk_data_ready = netlink_data_ready;
2058 if (cfg && cfg->input)
2059 nlk_sk(sk)->netlink_rcv = cfg->input;
2060
2061 if (netlink_insert(sk, 0))
2062 goto out_sock_release;
2063
2064 nlk = nlk_sk(sk);
2065 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2066
2067 netlink_table_grab();
2068 if (!nl_table[unit].registered) {
2069 nl_table[unit].groups = groups;
2070 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2071 nl_table[unit].cb_mutex = cb_mutex;
2072 nl_table[unit].module = module;
2073 if (cfg) {
2074 nl_table[unit].bind = cfg->bind;
2075 nl_table[unit].unbind = cfg->unbind;
2076 nl_table[unit].flags = cfg->flags;
2077 if (cfg->compare)
2078 nl_table[unit].compare = cfg->compare;
2079 }
2080 nl_table[unit].registered = 1;
2081 } else {
2082 kfree(listeners);
2083 nl_table[unit].registered++;
2084 }
2085 netlink_table_ungrab();
2086 return sk;
2087
2088out_sock_release:
2089 kfree(listeners);
2090 netlink_kernel_release(sk);
2091 return NULL;
2092
2093out_sock_release_nosk:
2094 sock_release(sock);
2095 return NULL;
2096}
2097EXPORT_SYMBOL(__netlink_kernel_create);
2098
2099void
2100netlink_kernel_release(struct sock *sk)
2101{
2102 if (sk == NULL || sk->sk_socket == NULL)
2103 return;
2104
2105 sock_release(sk->sk_socket);
2106}
2107EXPORT_SYMBOL(netlink_kernel_release);
2108
2109int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2110{
2111 struct listeners *new, *old;
2112 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2113
2114 if (groups < 32)
2115 groups = 32;
2116
2117 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2118 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2119 if (!new)
2120 return -ENOMEM;
2121 old = nl_deref_protected(tbl->listeners);
2122 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2123 rcu_assign_pointer(tbl->listeners, new);
2124
2125 kfree_rcu(old, rcu);
2126 }
2127 tbl->groups = groups;
2128
2129 return 0;
2130}
2131
2132/**
2133 * netlink_change_ngroups - change number of multicast groups
2134 *
2135 * This changes the number of multicast groups that are available
2136 * on a certain netlink family. Note that it is not possible to
2137 * change the number of groups to below 32. Also note that it does
2138 * not implicitly call netlink_clear_multicast_users() when the
2139 * number of groups is reduced.
2140 *
2141 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2142 * @groups: The new number of groups.
2143 */
2144int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2145{
2146 int err;
2147
2148 netlink_table_grab();
2149 err = __netlink_change_ngroups(sk, groups);
2150 netlink_table_ungrab();
2151
2152 return err;
2153}
2154
2155void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2156{
2157 struct sock *sk;
2158 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2159
2160 sk_for_each_bound(sk, &tbl->mc_list)
2161 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2162}
2163
2164struct nlmsghdr *
2165__nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2166{
2167 struct nlmsghdr *nlh;
2168 int size = nlmsg_msg_size(len);
2169
2170 nlh = skb_put(skb, NLMSG_ALIGN(size));
2171 nlh->nlmsg_type = type;
2172 nlh->nlmsg_len = size;
2173 nlh->nlmsg_flags = flags;
2174 nlh->nlmsg_pid = portid;
2175 nlh->nlmsg_seq = seq;
2176 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2177 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2178 return nlh;
2179}
2180EXPORT_SYMBOL(__nlmsg_put);
2181
2182/*
2183 * It looks a bit ugly.
2184 * It would be better to create kernel thread.
2185 */
2186
2187static int netlink_dump(struct sock *sk)
2188{
2189 struct netlink_sock *nlk = nlk_sk(sk);
2190 struct netlink_ext_ack extack = {};
2191 struct netlink_callback *cb;
2192 struct sk_buff *skb = NULL;
2193 struct nlmsghdr *nlh;
2194 struct module *module;
2195 int err = -ENOBUFS;
2196 int alloc_min_size;
2197 int alloc_size;
2198
2199 mutex_lock(nlk->cb_mutex);
2200 if (!nlk->cb_running) {
2201 err = -EINVAL;
2202 goto errout_skb;
2203 }
2204
2205 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2206 goto errout_skb;
2207
2208 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2209 * required, but it makes sense to _attempt_ a 16K bytes allocation
2210 * to reduce number of system calls on dump operations, if user
2211 * ever provided a big enough buffer.
2212 */
2213 cb = &nlk->cb;
2214 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2215
2216 if (alloc_min_size < nlk->max_recvmsg_len) {
2217 alloc_size = nlk->max_recvmsg_len;
2218 skb = alloc_skb(alloc_size,
2219 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2220 __GFP_NOWARN | __GFP_NORETRY);
2221 }
2222 if (!skb) {
2223 alloc_size = alloc_min_size;
2224 skb = alloc_skb(alloc_size, GFP_KERNEL);
2225 }
2226 if (!skb)
2227 goto errout_skb;
2228
2229 /* Trim skb to allocated size. User is expected to provide buffer as
2230 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2231 * netlink_recvmsg())). dump will pack as many smaller messages as
2232 * could fit within the allocated skb. skb is typically allocated
2233 * with larger space than required (could be as much as near 2x the
2234 * requested size with align to next power of 2 approach). Allowing
2235 * dump to use the excess space makes it difficult for a user to have a
2236 * reasonable static buffer based on the expected largest dump of a
2237 * single netdev. The outcome is MSG_TRUNC error.
2238 */
2239 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2240 netlink_skb_set_owner_r(skb, sk);
2241
2242 if (nlk->dump_done_errno > 0) {
2243 cb->extack = &extack;
2244 nlk->dump_done_errno = cb->dump(skb, cb);
2245 cb->extack = NULL;
2246 }
2247
2248 if (nlk->dump_done_errno > 0 ||
2249 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2250 mutex_unlock(nlk->cb_mutex);
2251
2252 if (sk_filter(sk, skb))
2253 kfree_skb(skb);
2254 else
2255 __netlink_sendskb(sk, skb);
2256 return 0;
2257 }
2258
2259 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE,
2260 sizeof(nlk->dump_done_errno),
2261 NLM_F_MULTI | cb->answer_flags);
2262 if (WARN_ON(!nlh))
2263 goto errout_skb;
2264
2265 nl_dump_check_consistent(cb, nlh);
2266
2267 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno,
2268 sizeof(nlk->dump_done_errno));
2269
2270 if (extack._msg && nlk->flags & NETLINK_F_EXT_ACK) {
2271 nlh->nlmsg_flags |= NLM_F_ACK_TLVS;
2272 if (!nla_put_string(skb, NLMSGERR_ATTR_MSG, extack._msg))
2273 nlmsg_end(skb, nlh);
2274 }
2275
2276 if (sk_filter(sk, skb))
2277 kfree_skb(skb);
2278 else
2279 __netlink_sendskb(sk, skb);
2280
2281 if (cb->done)
2282 cb->done(cb);
2283
2284 nlk->cb_running = false;
2285 module = cb->module;
2286 skb = cb->skb;
2287 mutex_unlock(nlk->cb_mutex);
2288 module_put(module);
2289 consume_skb(skb);
2290 return 0;
2291
2292errout_skb:
2293 mutex_unlock(nlk->cb_mutex);
2294 kfree_skb(skb);
2295 return err;
2296}
2297
2298int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2299 const struct nlmsghdr *nlh,
2300 struct netlink_dump_control *control)
2301{
2302 struct netlink_sock *nlk, *nlk2;
2303 struct netlink_callback *cb;
2304 struct sock *sk;
2305 int ret;
2306
2307 refcount_inc(&skb->users);
2308
2309 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2310 if (sk == NULL) {
2311 ret = -ECONNREFUSED;
2312 goto error_free;
2313 }
2314
2315 nlk = nlk_sk(sk);
2316 mutex_lock(nlk->cb_mutex);
2317 /* A dump is in progress... */
2318 if (nlk->cb_running) {
2319 ret = -EBUSY;
2320 goto error_unlock;
2321 }
2322 /* add reference of module which cb->dump belongs to */
2323 if (!try_module_get(control->module)) {
2324 ret = -EPROTONOSUPPORT;
2325 goto error_unlock;
2326 }
2327
2328 cb = &nlk->cb;
2329 memset(cb, 0, sizeof(*cb));
2330 cb->dump = control->dump;
2331 cb->done = control->done;
2332 cb->nlh = nlh;
2333 cb->data = control->data;
2334 cb->module = control->module;
2335 cb->min_dump_alloc = control->min_dump_alloc;
2336 cb->skb = skb;
2337
2338 nlk2 = nlk_sk(NETLINK_CB(skb).sk);
2339 cb->strict_check = !!(nlk2->flags & NETLINK_F_STRICT_CHK);
2340
2341 if (control->start) {
2342 ret = control->start(cb);
2343 if (ret)
2344 goto error_put;
2345 }
2346
2347 nlk->cb_running = true;
2348 nlk->dump_done_errno = INT_MAX;
2349
2350 mutex_unlock(nlk->cb_mutex);
2351
2352 ret = netlink_dump(sk);
2353
2354 sock_put(sk);
2355
2356 if (ret)
2357 return ret;
2358
2359 /* We successfully started a dump, by returning -EINTR we
2360 * signal not to send ACK even if it was requested.
2361 */
2362 return -EINTR;
2363
2364error_put:
2365 module_put(control->module);
2366error_unlock:
2367 sock_put(sk);
2368 mutex_unlock(nlk->cb_mutex);
2369error_free:
2370 kfree_skb(skb);
2371 return ret;
2372}
2373EXPORT_SYMBOL(__netlink_dump_start);
2374
2375void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2376 const struct netlink_ext_ack *extack)
2377{
2378 struct sk_buff *skb;
2379 struct nlmsghdr *rep;
2380 struct nlmsgerr *errmsg;
2381 size_t payload = sizeof(*errmsg);
2382 size_t tlvlen = 0;
2383 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2384 unsigned int flags = 0;
2385 bool nlk_has_extack = nlk->flags & NETLINK_F_EXT_ACK;
2386
2387 /* Error messages get the original request appened, unless the user
2388 * requests to cap the error message, and get extra error data if
2389 * requested.
2390 */
2391 if (nlk_has_extack && extack && extack->_msg)
2392 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2393
2394 if (err) {
2395 if (!(nlk->flags & NETLINK_F_CAP_ACK))
2396 payload += nlmsg_len(nlh);
2397 else
2398 flags |= NLM_F_CAPPED;
2399 if (nlk_has_extack && extack && extack->bad_attr)
2400 tlvlen += nla_total_size(sizeof(u32));
2401 } else {
2402 flags |= NLM_F_CAPPED;
2403
2404 if (nlk_has_extack && extack && extack->cookie_len)
2405 tlvlen += nla_total_size(extack->cookie_len);
2406 }
2407
2408 if (tlvlen)
2409 flags |= NLM_F_ACK_TLVS;
2410
2411 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2412 if (!skb) {
2413 NETLINK_CB(in_skb).sk->sk_err = ENOBUFS;
2414 NETLINK_CB(in_skb).sk->sk_error_report(NETLINK_CB(in_skb).sk);
2415 return;
2416 }
2417
2418 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2419 NLMSG_ERROR, payload, flags);
2420 errmsg = nlmsg_data(rep);
2421 errmsg->error = err;
2422 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2423
2424 if (nlk_has_extack && extack) {
2425 if (extack->_msg) {
2426 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG,
2427 extack->_msg));
2428 }
2429 if (err) {
2430 if (extack->bad_attr &&
2431 !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2432 (u8 *)extack->bad_attr >= in_skb->data +
2433 in_skb->len))
2434 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2435 (u8 *)extack->bad_attr -
2436 in_skb->data));
2437 } else {
2438 if (extack->cookie_len)
2439 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2440 extack->cookie_len,
2441 extack->cookie));
2442 }
2443 }
2444
2445 nlmsg_end(skb, rep);
2446
2447 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2448}
2449EXPORT_SYMBOL(netlink_ack);
2450
2451int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2452 struct nlmsghdr *,
2453 struct netlink_ext_ack *))
2454{
2455 struct netlink_ext_ack extack;
2456 struct nlmsghdr *nlh;
2457 int err;
2458
2459 while (skb->len >= nlmsg_total_size(0)) {
2460 int msglen;
2461
2462 memset(&extack, 0, sizeof(extack));
2463 nlh = nlmsg_hdr(skb);
2464 err = 0;
2465
2466 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2467 return 0;
2468
2469 /* Only requests are handled by the kernel */
2470 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2471 goto ack;
2472
2473 /* Skip control messages */
2474 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2475 goto ack;
2476
2477 err = cb(skb, nlh, &extack);
2478 if (err == -EINTR)
2479 goto skip;
2480
2481ack:
2482 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2483 netlink_ack(skb, nlh, err, &extack);
2484
2485skip:
2486 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2487 if (msglen > skb->len)
2488 msglen = skb->len;
2489 skb_pull(skb, msglen);
2490 }
2491
2492 return 0;
2493}
2494EXPORT_SYMBOL(netlink_rcv_skb);
2495
2496/**
2497 * nlmsg_notify - send a notification netlink message
2498 * @sk: netlink socket to use
2499 * @skb: notification message
2500 * @portid: destination netlink portid for reports or 0
2501 * @group: destination multicast group or 0
2502 * @report: 1 to report back, 0 to disable
2503 * @flags: allocation flags
2504 */
2505int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2506 unsigned int group, int report, gfp_t flags)
2507{
2508 int err = 0;
2509
2510 if (group) {
2511 int exclude_portid = 0;
2512
2513 if (report) {
2514 refcount_inc(&skb->users);
2515 exclude_portid = portid;
2516 }
2517
2518 /* errors reported via destination sk->sk_err, but propagate
2519 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2520 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2521 }
2522
2523 if (report) {
2524 int err2;
2525
2526 err2 = nlmsg_unicast(sk, skb, portid);
2527 if (!err || err == -ESRCH)
2528 err = err2;
2529 }
2530
2531 return err;
2532}
2533EXPORT_SYMBOL(nlmsg_notify);
2534
2535#ifdef CONFIG_PROC_FS
2536struct nl_seq_iter {
2537 struct seq_net_private p;
2538 struct rhashtable_iter hti;
2539 int link;
2540};
2541
2542static void netlink_walk_start(struct nl_seq_iter *iter)
2543{
2544 rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti);
2545 rhashtable_walk_start(&iter->hti);
2546}
2547
2548static void netlink_walk_stop(struct nl_seq_iter *iter)
2549{
2550 rhashtable_walk_stop(&iter->hti);
2551 rhashtable_walk_exit(&iter->hti);
2552}
2553
2554static void *__netlink_seq_next(struct seq_file *seq)
2555{
2556 struct nl_seq_iter *iter = seq->private;
2557 struct netlink_sock *nlk;
2558
2559 do {
2560 for (;;) {
2561 nlk = rhashtable_walk_next(&iter->hti);
2562
2563 if (IS_ERR(nlk)) {
2564 if (PTR_ERR(nlk) == -EAGAIN)
2565 continue;
2566
2567 return nlk;
2568 }
2569
2570 if (nlk)
2571 break;
2572
2573 netlink_walk_stop(iter);
2574 if (++iter->link >= MAX_LINKS)
2575 return NULL;
2576
2577 netlink_walk_start(iter);
2578 }
2579 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2580
2581 return nlk;
2582}
2583
2584static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2585{
2586 struct nl_seq_iter *iter = seq->private;
2587 void *obj = SEQ_START_TOKEN;
2588 loff_t pos;
2589
2590 iter->link = 0;
2591
2592 netlink_walk_start(iter);
2593
2594 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2595 obj = __netlink_seq_next(seq);
2596
2597 return obj;
2598}
2599
2600static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2601{
2602 ++*pos;
2603 return __netlink_seq_next(seq);
2604}
2605
2606static void netlink_seq_stop(struct seq_file *seq, void *v)
2607{
2608 struct nl_seq_iter *iter = seq->private;
2609
2610 if (iter->link >= MAX_LINKS)
2611 return;
2612
2613 netlink_walk_stop(iter);
2614}
2615
2616
2617static int netlink_seq_show(struct seq_file *seq, void *v)
2618{
2619 if (v == SEQ_START_TOKEN) {
2620 seq_puts(seq,
2621 "sk Eth Pid Groups "
2622 "Rmem Wmem Dump Locks Drops Inode\n");
2623 } else {
2624 struct sock *s = v;
2625 struct netlink_sock *nlk = nlk_sk(s);
2626
2627 seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n",
2628 s,
2629 s->sk_protocol,
2630 nlk->portid,
2631 nlk->groups ? (u32)nlk->groups[0] : 0,
2632 sk_rmem_alloc_get(s),
2633 sk_wmem_alloc_get(s),
2634 nlk->cb_running,
2635 refcount_read(&s->sk_refcnt),
2636 atomic_read(&s->sk_drops),
2637 sock_i_ino(s)
2638 );
2639
2640 }
2641 return 0;
2642}
2643
2644static const struct seq_operations netlink_seq_ops = {
2645 .start = netlink_seq_start,
2646 .next = netlink_seq_next,
2647 .stop = netlink_seq_stop,
2648 .show = netlink_seq_show,
2649};
2650#endif
2651
2652int netlink_register_notifier(struct notifier_block *nb)
2653{
2654 return blocking_notifier_chain_register(&netlink_chain, nb);
2655}
2656EXPORT_SYMBOL(netlink_register_notifier);
2657
2658int netlink_unregister_notifier(struct notifier_block *nb)
2659{
2660 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2661}
2662EXPORT_SYMBOL(netlink_unregister_notifier);
2663
2664static const struct proto_ops netlink_ops = {
2665 .family = PF_NETLINK,
2666 .owner = THIS_MODULE,
2667 .release = netlink_release,
2668 .bind = netlink_bind,
2669 .connect = netlink_connect,
2670 .socketpair = sock_no_socketpair,
2671 .accept = sock_no_accept,
2672 .getname = netlink_getname,
2673 .poll = datagram_poll,
2674 .ioctl = netlink_ioctl,
2675 .listen = sock_no_listen,
2676 .shutdown = sock_no_shutdown,
2677 .setsockopt = netlink_setsockopt,
2678 .getsockopt = netlink_getsockopt,
2679 .sendmsg = netlink_sendmsg,
2680 .recvmsg = netlink_recvmsg,
2681 .mmap = sock_no_mmap,
2682 .sendpage = sock_no_sendpage,
2683};
2684
2685static const struct net_proto_family netlink_family_ops = {
2686 .family = PF_NETLINK,
2687 .create = netlink_create,
2688 .owner = THIS_MODULE, /* for consistency 8) */
2689};
2690
2691static int __net_init netlink_net_init(struct net *net)
2692{
2693#ifdef CONFIG_PROC_FS
2694 if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2695 sizeof(struct nl_seq_iter)))
2696 return -ENOMEM;
2697#endif
2698 return 0;
2699}
2700
2701static void __net_exit netlink_net_exit(struct net *net)
2702{
2703#ifdef CONFIG_PROC_FS
2704 remove_proc_entry("netlink", net->proc_net);
2705#endif
2706}
2707
2708static void __init netlink_add_usersock_entry(void)
2709{
2710 struct listeners *listeners;
2711 int groups = 32;
2712
2713 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2714 if (!listeners)
2715 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2716
2717 netlink_table_grab();
2718
2719 nl_table[NETLINK_USERSOCK].groups = groups;
2720 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2721 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2722 nl_table[NETLINK_USERSOCK].registered = 1;
2723 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2724
2725 netlink_table_ungrab();
2726}
2727
2728static struct pernet_operations __net_initdata netlink_net_ops = {
2729 .init = netlink_net_init,
2730 .exit = netlink_net_exit,
2731};
2732
2733static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2734{
2735 const struct netlink_sock *nlk = data;
2736 struct netlink_compare_arg arg;
2737
2738 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2739 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2740}
2741
2742static const struct rhashtable_params netlink_rhashtable_params = {
2743 .head_offset = offsetof(struct netlink_sock, node),
2744 .key_len = netlink_compare_arg_len,
2745 .obj_hashfn = netlink_hash,
2746 .obj_cmpfn = netlink_compare,
2747 .automatic_shrinking = true,
2748};
2749
2750static int __init netlink_proto_init(void)
2751{
2752 int i;
2753 int err = proto_register(&netlink_proto, 0);
2754
2755 if (err != 0)
2756 goto out;
2757
2758 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2759
2760 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2761 if (!nl_table)
2762 goto panic;
2763
2764 for (i = 0; i < MAX_LINKS; i++) {
2765 if (rhashtable_init(&nl_table[i].hash,
2766 &netlink_rhashtable_params) < 0) {
2767 while (--i > 0)
2768 rhashtable_destroy(&nl_table[i].hash);
2769 kfree(nl_table);
2770 goto panic;
2771 }
2772 }
2773
2774 netlink_add_usersock_entry();
2775
2776 sock_register(&netlink_family_ops);
2777 register_pernet_subsys(&netlink_net_ops);
2778 register_pernet_subsys(&netlink_tap_net_ops);
2779 /* The netlink device handler may be needed early. */
2780 rtnetlink_init();
2781out:
2782 return err;
2783panic:
2784 panic("netlink_init: Cannot allocate nl_table\n");
2785}
2786
2787core_initcall(netlink_proto_init);
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * NETLINK Kernel-user communication protocol.
4 *
5 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
6 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
7 * Patrick McHardy <kaber@trash.net>
8 *
9 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
10 * added netlink_proto_exit
11 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
12 * use nlk_sk, as sk->protinfo is on a diet 8)
13 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
14 * - inc module use count of module that owns
15 * the kernel socket in case userspace opens
16 * socket of same protocol
17 * - remove all module support, since netlink is
18 * mandatory if CONFIG_NET=y these days
19 */
20
21#include <linux/module.h>
22
23#include <linux/bpf.h>
24#include <linux/capability.h>
25#include <linux/kernel.h>
26#include <linux/filter.h>
27#include <linux/init.h>
28#include <linux/signal.h>
29#include <linux/sched.h>
30#include <linux/errno.h>
31#include <linux/string.h>
32#include <linux/stat.h>
33#include <linux/socket.h>
34#include <linux/un.h>
35#include <linux/fcntl.h>
36#include <linux/termios.h>
37#include <linux/sockios.h>
38#include <linux/net.h>
39#include <linux/fs.h>
40#include <linux/slab.h>
41#include <linux/uaccess.h>
42#include <linux/skbuff.h>
43#include <linux/netdevice.h>
44#include <linux/rtnetlink.h>
45#include <linux/proc_fs.h>
46#include <linux/seq_file.h>
47#include <linux/notifier.h>
48#include <linux/security.h>
49#include <linux/jhash.h>
50#include <linux/jiffies.h>
51#include <linux/random.h>
52#include <linux/bitops.h>
53#include <linux/mm.h>
54#include <linux/types.h>
55#include <linux/audit.h>
56#include <linux/mutex.h>
57#include <linux/vmalloc.h>
58#include <linux/if_arp.h>
59#include <linux/rhashtable.h>
60#include <asm/cacheflush.h>
61#include <linux/hash.h>
62#include <linux/genetlink.h>
63#include <linux/net_namespace.h>
64#include <linux/nospec.h>
65#include <linux/btf_ids.h>
66
67#include <net/net_namespace.h>
68#include <net/netns/generic.h>
69#include <net/sock.h>
70#include <net/scm.h>
71#include <net/netlink.h>
72#define CREATE_TRACE_POINTS
73#include <trace/events/netlink.h>
74
75#include "af_netlink.h"
76
77struct listeners {
78 struct rcu_head rcu;
79 unsigned long masks[];
80};
81
82/* state bits */
83#define NETLINK_S_CONGESTED 0x0
84
85static inline int netlink_is_kernel(struct sock *sk)
86{
87 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
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);
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_end_offset(skb);
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 struct netlink_sock *nlk = nlk_sk(sk);
353
354 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
355 if (!test_and_set_bit(NETLINK_S_CONGESTED,
356 &nlk_sk(sk)->state)) {
357 sk->sk_err = ENOBUFS;
358 sk_error_report(sk);
359 }
360 }
361 atomic_inc(&sk->sk_drops);
362}
363
364static void netlink_rcv_wake(struct sock *sk)
365{
366 struct netlink_sock *nlk = nlk_sk(sk);
367
368 if (skb_queue_empty_lockless(&sk->sk_receive_queue))
369 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
370 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
371 wake_up_interruptible(&nlk->wait);
372}
373
374static void netlink_skb_destructor(struct sk_buff *skb)
375{
376 if (is_vmalloc_addr(skb->head)) {
377 if (!skb->cloned ||
378 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
379 vfree(skb->head);
380
381 skb->head = NULL;
382 }
383 if (skb->sk != NULL)
384 sock_rfree(skb);
385}
386
387static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
388{
389 WARN_ON(skb->sk != NULL);
390 skb->sk = sk;
391 skb->destructor = netlink_skb_destructor;
392 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
393 sk_mem_charge(sk, skb->truesize);
394}
395
396static void netlink_sock_destruct(struct sock *sk)
397{
398 struct netlink_sock *nlk = nlk_sk(sk);
399
400 if (nlk->cb_running) {
401 if (nlk->cb.done)
402 nlk->cb.done(&nlk->cb);
403 module_put(nlk->cb.module);
404 kfree_skb(nlk->cb.skb);
405 }
406
407 skb_queue_purge(&sk->sk_receive_queue);
408
409 if (!sock_flag(sk, SOCK_DEAD)) {
410 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
411 return;
412 }
413
414 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
415 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
416 WARN_ON(nlk_sk(sk)->groups);
417}
418
419static void netlink_sock_destruct_work(struct work_struct *work)
420{
421 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
422 work);
423
424 sk_free(&nlk->sk);
425}
426
427/* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
428 * SMP. Look, when several writers sleep and reader wakes them up, all but one
429 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
430 * this, _but_ remember, it adds useless work on UP machines.
431 */
432
433void netlink_table_grab(void)
434 __acquires(nl_table_lock)
435{
436 might_sleep();
437
438 write_lock_irq(&nl_table_lock);
439
440 if (atomic_read(&nl_table_users)) {
441 DECLARE_WAITQUEUE(wait, current);
442
443 add_wait_queue_exclusive(&nl_table_wait, &wait);
444 for (;;) {
445 set_current_state(TASK_UNINTERRUPTIBLE);
446 if (atomic_read(&nl_table_users) == 0)
447 break;
448 write_unlock_irq(&nl_table_lock);
449 schedule();
450 write_lock_irq(&nl_table_lock);
451 }
452
453 __set_current_state(TASK_RUNNING);
454 remove_wait_queue(&nl_table_wait, &wait);
455 }
456}
457
458void netlink_table_ungrab(void)
459 __releases(nl_table_lock)
460{
461 write_unlock_irq(&nl_table_lock);
462 wake_up(&nl_table_wait);
463}
464
465static inline void
466netlink_lock_table(void)
467{
468 unsigned long flags;
469
470 /* read_lock() synchronizes us to netlink_table_grab */
471
472 read_lock_irqsave(&nl_table_lock, flags);
473 atomic_inc(&nl_table_users);
474 read_unlock_irqrestore(&nl_table_lock, flags);
475}
476
477static inline void
478netlink_unlock_table(void)
479{
480 if (atomic_dec_and_test(&nl_table_users))
481 wake_up(&nl_table_wait);
482}
483
484struct netlink_compare_arg
485{
486 possible_net_t pnet;
487 u32 portid;
488};
489
490/* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
491#define netlink_compare_arg_len \
492 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
493
494static inline int netlink_compare(struct rhashtable_compare_arg *arg,
495 const void *ptr)
496{
497 const struct netlink_compare_arg *x = arg->key;
498 const struct netlink_sock *nlk = ptr;
499
500 return nlk->portid != x->portid ||
501 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
502}
503
504static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
505 struct net *net, u32 portid)
506{
507 memset(arg, 0, sizeof(*arg));
508 write_pnet(&arg->pnet, net);
509 arg->portid = portid;
510}
511
512static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
513 struct net *net)
514{
515 struct netlink_compare_arg arg;
516
517 netlink_compare_arg_init(&arg, net, portid);
518 return rhashtable_lookup_fast(&table->hash, &arg,
519 netlink_rhashtable_params);
520}
521
522static int __netlink_insert(struct netlink_table *table, struct sock *sk)
523{
524 struct netlink_compare_arg arg;
525
526 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
527 return rhashtable_lookup_insert_key(&table->hash, &arg,
528 &nlk_sk(sk)->node,
529 netlink_rhashtable_params);
530}
531
532static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
533{
534 struct netlink_table *table = &nl_table[protocol];
535 struct sock *sk;
536
537 rcu_read_lock();
538 sk = __netlink_lookup(table, portid, net);
539 if (sk)
540 sock_hold(sk);
541 rcu_read_unlock();
542
543 return sk;
544}
545
546static const struct proto_ops netlink_ops;
547
548static void
549netlink_update_listeners(struct sock *sk)
550{
551 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
552 unsigned long mask;
553 unsigned int i;
554 struct listeners *listeners;
555
556 listeners = nl_deref_protected(tbl->listeners);
557 if (!listeners)
558 return;
559
560 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
561 mask = 0;
562 sk_for_each_bound(sk, &tbl->mc_list) {
563 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
564 mask |= nlk_sk(sk)->groups[i];
565 }
566 listeners->masks[i] = mask;
567 }
568 /* this function is only called with the netlink table "grabbed", which
569 * makes sure updates are visible before bind or setsockopt return. */
570}
571
572static int netlink_insert(struct sock *sk, u32 portid)
573{
574 struct netlink_table *table = &nl_table[sk->sk_protocol];
575 int err;
576
577 lock_sock(sk);
578
579 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
580 if (nlk_sk(sk)->bound)
581 goto err;
582
583 /* portid can be read locklessly from netlink_getname(). */
584 WRITE_ONCE(nlk_sk(sk)->portid, portid);
585
586 sock_hold(sk);
587
588 err = __netlink_insert(table, sk);
589 if (err) {
590 /* In case the hashtable backend returns with -EBUSY
591 * from here, it must not escape to the caller.
592 */
593 if (unlikely(err == -EBUSY))
594 err = -EOVERFLOW;
595 if (err == -EEXIST)
596 err = -EADDRINUSE;
597 sock_put(sk);
598 goto err;
599 }
600
601 /* We need to ensure that the socket is hashed and visible. */
602 smp_wmb();
603 /* Paired with lockless reads from netlink_bind(),
604 * netlink_connect() and netlink_sendmsg().
605 */
606 WRITE_ONCE(nlk_sk(sk)->bound, portid);
607
608err:
609 release_sock(sk);
610 return err;
611}
612
613static void netlink_remove(struct sock *sk)
614{
615 struct netlink_table *table;
616
617 table = &nl_table[sk->sk_protocol];
618 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
619 netlink_rhashtable_params)) {
620 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
621 __sock_put(sk);
622 }
623
624 netlink_table_grab();
625 if (nlk_sk(sk)->subscriptions) {
626 __sk_del_bind_node(sk);
627 netlink_update_listeners(sk);
628 }
629 if (sk->sk_protocol == NETLINK_GENERIC)
630 atomic_inc(&genl_sk_destructing_cnt);
631 netlink_table_ungrab();
632}
633
634static struct proto netlink_proto = {
635 .name = "NETLINK",
636 .owner = THIS_MODULE,
637 .obj_size = sizeof(struct netlink_sock),
638};
639
640static int __netlink_create(struct net *net, struct socket *sock,
641 struct mutex *cb_mutex, int protocol,
642 int kern)
643{
644 struct sock *sk;
645 struct netlink_sock *nlk;
646
647 sock->ops = &netlink_ops;
648
649 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
650 if (!sk)
651 return -ENOMEM;
652
653 sock_init_data(sock, sk);
654
655 nlk = nlk_sk(sk);
656 if (cb_mutex) {
657 nlk->cb_mutex = cb_mutex;
658 } else {
659 nlk->cb_mutex = &nlk->cb_def_mutex;
660 mutex_init(nlk->cb_mutex);
661 lockdep_set_class_and_name(nlk->cb_mutex,
662 nlk_cb_mutex_keys + protocol,
663 nlk_cb_mutex_key_strings[protocol]);
664 }
665 init_waitqueue_head(&nlk->wait);
666
667 sk->sk_destruct = netlink_sock_destruct;
668 sk->sk_protocol = protocol;
669 return 0;
670}
671
672static int netlink_create(struct net *net, struct socket *sock, int protocol,
673 int kern)
674{
675 struct module *module = NULL;
676 struct mutex *cb_mutex;
677 struct netlink_sock *nlk;
678 int (*bind)(struct net *net, int group);
679 void (*unbind)(struct net *net, int group);
680 int err = 0;
681
682 sock->state = SS_UNCONNECTED;
683
684 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
685 return -ESOCKTNOSUPPORT;
686
687 if (protocol < 0 || protocol >= MAX_LINKS)
688 return -EPROTONOSUPPORT;
689 protocol = array_index_nospec(protocol, MAX_LINKS);
690
691 netlink_lock_table();
692#ifdef CONFIG_MODULES
693 if (!nl_table[protocol].registered) {
694 netlink_unlock_table();
695 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
696 netlink_lock_table();
697 }
698#endif
699 if (nl_table[protocol].registered &&
700 try_module_get(nl_table[protocol].module))
701 module = nl_table[protocol].module;
702 else
703 err = -EPROTONOSUPPORT;
704 cb_mutex = nl_table[protocol].cb_mutex;
705 bind = nl_table[protocol].bind;
706 unbind = nl_table[protocol].unbind;
707 netlink_unlock_table();
708
709 if (err < 0)
710 goto out;
711
712 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
713 if (err < 0)
714 goto out_module;
715
716 sock_prot_inuse_add(net, &netlink_proto, 1);
717
718 nlk = nlk_sk(sock->sk);
719 nlk->module = module;
720 nlk->netlink_bind = bind;
721 nlk->netlink_unbind = unbind;
722out:
723 return err;
724
725out_module:
726 module_put(module);
727 goto out;
728}
729
730static void deferred_put_nlk_sk(struct rcu_head *head)
731{
732 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
733 struct sock *sk = &nlk->sk;
734
735 kfree(nlk->groups);
736 nlk->groups = NULL;
737
738 if (!refcount_dec_and_test(&sk->sk_refcnt))
739 return;
740
741 if (nlk->cb_running && nlk->cb.done) {
742 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
743 schedule_work(&nlk->work);
744 return;
745 }
746
747 sk_free(sk);
748}
749
750static int netlink_release(struct socket *sock)
751{
752 struct sock *sk = sock->sk;
753 struct netlink_sock *nlk;
754
755 if (!sk)
756 return 0;
757
758 netlink_remove(sk);
759 sock_orphan(sk);
760 nlk = nlk_sk(sk);
761
762 /*
763 * OK. Socket is unlinked, any packets that arrive now
764 * will be purged.
765 */
766
767 /* must not acquire netlink_table_lock in any way again before unbind
768 * and notifying genetlink is done as otherwise it might deadlock
769 */
770 if (nlk->netlink_unbind) {
771 int i;
772
773 for (i = 0; i < nlk->ngroups; i++)
774 if (test_bit(i, nlk->groups))
775 nlk->netlink_unbind(sock_net(sk), i + 1);
776 }
777 if (sk->sk_protocol == NETLINK_GENERIC &&
778 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
779 wake_up(&genl_sk_destructing_waitq);
780
781 sock->sk = NULL;
782 wake_up_interruptible_all(&nlk->wait);
783
784 skb_queue_purge(&sk->sk_write_queue);
785
786 if (nlk->portid && nlk->bound) {
787 struct netlink_notify n = {
788 .net = sock_net(sk),
789 .protocol = sk->sk_protocol,
790 .portid = nlk->portid,
791 };
792 blocking_notifier_call_chain(&netlink_chain,
793 NETLINK_URELEASE, &n);
794 }
795
796 module_put(nlk->module);
797
798 if (netlink_is_kernel(sk)) {
799 netlink_table_grab();
800 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
801 if (--nl_table[sk->sk_protocol].registered == 0) {
802 struct listeners *old;
803
804 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
805 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
806 kfree_rcu(old, rcu);
807 nl_table[sk->sk_protocol].module = NULL;
808 nl_table[sk->sk_protocol].bind = NULL;
809 nl_table[sk->sk_protocol].unbind = NULL;
810 nl_table[sk->sk_protocol].flags = 0;
811 nl_table[sk->sk_protocol].registered = 0;
812 }
813 netlink_table_ungrab();
814 }
815
816 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
817
818 /* Because struct net might disappear soon, do not keep a pointer. */
819 if (!sk->sk_net_refcnt && sock_net(sk) != &init_net) {
820 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false);
821 /* Because of deferred_put_nlk_sk and use of work queue,
822 * it is possible netns will be freed before this socket.
823 */
824 sock_net_set(sk, &init_net);
825 __netns_tracker_alloc(&init_net, &sk->ns_tracker,
826 false, GFP_KERNEL);
827 }
828 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
829 return 0;
830}
831
832static int netlink_autobind(struct socket *sock)
833{
834 struct sock *sk = sock->sk;
835 struct net *net = sock_net(sk);
836 struct netlink_table *table = &nl_table[sk->sk_protocol];
837 s32 portid = task_tgid_vnr(current);
838 int err;
839 s32 rover = -4096;
840 bool ok;
841
842retry:
843 cond_resched();
844 rcu_read_lock();
845 ok = !__netlink_lookup(table, portid, net);
846 rcu_read_unlock();
847 if (!ok) {
848 /* Bind collision, search negative portid values. */
849 if (rover == -4096)
850 /* rover will be in range [S32_MIN, -4097] */
851 rover = S32_MIN + get_random_u32_below(-4096 - S32_MIN);
852 else if (rover >= -4096)
853 rover = -4097;
854 portid = rover--;
855 goto retry;
856 }
857
858 err = netlink_insert(sk, portid);
859 if (err == -EADDRINUSE)
860 goto retry;
861
862 /* If 2 threads race to autobind, that is fine. */
863 if (err == -EBUSY)
864 err = 0;
865
866 return err;
867}
868
869/**
870 * __netlink_ns_capable - General netlink message capability test
871 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
872 * @user_ns: The user namespace of the capability to use
873 * @cap: The capability to use
874 *
875 * Test to see if the opener of the socket we received the message
876 * from had when the netlink socket was created and the sender of the
877 * message has the capability @cap in the user namespace @user_ns.
878 */
879bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
880 struct user_namespace *user_ns, int cap)
881{
882 return ((nsp->flags & NETLINK_SKB_DST) ||
883 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
884 ns_capable(user_ns, cap);
885}
886EXPORT_SYMBOL(__netlink_ns_capable);
887
888/**
889 * netlink_ns_capable - General netlink message capability test
890 * @skb: socket buffer holding a netlink command from userspace
891 * @user_ns: The user namespace of the capability to use
892 * @cap: The capability to use
893 *
894 * Test to see if the opener of the socket we received the message
895 * from had when the netlink socket was created and the sender of the
896 * message has the capability @cap in the user namespace @user_ns.
897 */
898bool netlink_ns_capable(const struct sk_buff *skb,
899 struct user_namespace *user_ns, int cap)
900{
901 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
902}
903EXPORT_SYMBOL(netlink_ns_capable);
904
905/**
906 * netlink_capable - Netlink global message capability test
907 * @skb: socket buffer holding a netlink command from userspace
908 * @cap: The capability to use
909 *
910 * Test to see if the opener of the socket we received the message
911 * from had when the netlink socket was created and the sender of the
912 * message has the capability @cap in all user namespaces.
913 */
914bool netlink_capable(const struct sk_buff *skb, int cap)
915{
916 return netlink_ns_capable(skb, &init_user_ns, cap);
917}
918EXPORT_SYMBOL(netlink_capable);
919
920/**
921 * netlink_net_capable - Netlink network namespace message capability test
922 * @skb: socket buffer holding a netlink command from userspace
923 * @cap: The capability to use
924 *
925 * Test to see if the opener of the socket we received the message
926 * from had when the netlink socket was created and the sender of the
927 * message has the capability @cap over the network namespace of
928 * the socket we received the message from.
929 */
930bool netlink_net_capable(const struct sk_buff *skb, int cap)
931{
932 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
933}
934EXPORT_SYMBOL(netlink_net_capable);
935
936static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
937{
938 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
939 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
940}
941
942static void
943netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
944{
945 struct netlink_sock *nlk = nlk_sk(sk);
946
947 if (nlk->subscriptions && !subscriptions)
948 __sk_del_bind_node(sk);
949 else if (!nlk->subscriptions && subscriptions)
950 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
951 nlk->subscriptions = subscriptions;
952}
953
954static int netlink_realloc_groups(struct sock *sk)
955{
956 struct netlink_sock *nlk = nlk_sk(sk);
957 unsigned int groups;
958 unsigned long *new_groups;
959 int err = 0;
960
961 netlink_table_grab();
962
963 groups = nl_table[sk->sk_protocol].groups;
964 if (!nl_table[sk->sk_protocol].registered) {
965 err = -ENOENT;
966 goto out_unlock;
967 }
968
969 if (nlk->ngroups >= groups)
970 goto out_unlock;
971
972 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
973 if (new_groups == NULL) {
974 err = -ENOMEM;
975 goto out_unlock;
976 }
977 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
978 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
979
980 nlk->groups = new_groups;
981 nlk->ngroups = groups;
982 out_unlock:
983 netlink_table_ungrab();
984 return err;
985}
986
987static void netlink_undo_bind(int group, long unsigned int groups,
988 struct sock *sk)
989{
990 struct netlink_sock *nlk = nlk_sk(sk);
991 int undo;
992
993 if (!nlk->netlink_unbind)
994 return;
995
996 for (undo = 0; undo < group; undo++)
997 if (test_bit(undo, &groups))
998 nlk->netlink_unbind(sock_net(sk), undo + 1);
999}
1000
1001static int netlink_bind(struct socket *sock, struct sockaddr *addr,
1002 int addr_len)
1003{
1004 struct sock *sk = sock->sk;
1005 struct net *net = sock_net(sk);
1006 struct netlink_sock *nlk = nlk_sk(sk);
1007 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1008 int err = 0;
1009 unsigned long groups;
1010 bool bound;
1011
1012 if (addr_len < sizeof(struct sockaddr_nl))
1013 return -EINVAL;
1014
1015 if (nladdr->nl_family != AF_NETLINK)
1016 return -EINVAL;
1017 groups = nladdr->nl_groups;
1018
1019 /* Only superuser is allowed to listen multicasts */
1020 if (groups) {
1021 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1022 return -EPERM;
1023 err = netlink_realloc_groups(sk);
1024 if (err)
1025 return err;
1026 }
1027
1028 if (nlk->ngroups < BITS_PER_LONG)
1029 groups &= (1UL << nlk->ngroups) - 1;
1030
1031 /* Paired with WRITE_ONCE() in netlink_insert() */
1032 bound = READ_ONCE(nlk->bound);
1033 if (bound) {
1034 /* Ensure nlk->portid is up-to-date. */
1035 smp_rmb();
1036
1037 if (nladdr->nl_pid != nlk->portid)
1038 return -EINVAL;
1039 }
1040
1041 if (nlk->netlink_bind && groups) {
1042 int group;
1043
1044 /* nl_groups is a u32, so cap the maximum groups we can bind */
1045 for (group = 0; group < BITS_PER_TYPE(u32); group++) {
1046 if (!test_bit(group, &groups))
1047 continue;
1048 err = nlk->netlink_bind(net, group + 1);
1049 if (!err)
1050 continue;
1051 netlink_undo_bind(group, groups, sk);
1052 return err;
1053 }
1054 }
1055
1056 /* No need for barriers here as we return to user-space without
1057 * using any of the bound attributes.
1058 */
1059 netlink_lock_table();
1060 if (!bound) {
1061 err = nladdr->nl_pid ?
1062 netlink_insert(sk, nladdr->nl_pid) :
1063 netlink_autobind(sock);
1064 if (err) {
1065 netlink_undo_bind(BITS_PER_TYPE(u32), groups, sk);
1066 goto unlock;
1067 }
1068 }
1069
1070 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1071 goto unlock;
1072 netlink_unlock_table();
1073
1074 netlink_table_grab();
1075 netlink_update_subscriptions(sk, nlk->subscriptions +
1076 hweight32(groups) -
1077 hweight32(nlk->groups[0]));
1078 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1079 netlink_update_listeners(sk);
1080 netlink_table_ungrab();
1081
1082 return 0;
1083
1084unlock:
1085 netlink_unlock_table();
1086 return err;
1087}
1088
1089static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1090 int alen, int flags)
1091{
1092 int err = 0;
1093 struct sock *sk = sock->sk;
1094 struct netlink_sock *nlk = nlk_sk(sk);
1095 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1096
1097 if (alen < sizeof(addr->sa_family))
1098 return -EINVAL;
1099
1100 if (addr->sa_family == AF_UNSPEC) {
1101 /* paired with READ_ONCE() in netlink_getsockbyportid() */
1102 WRITE_ONCE(sk->sk_state, NETLINK_UNCONNECTED);
1103 /* dst_portid and dst_group can be read locklessly */
1104 WRITE_ONCE(nlk->dst_portid, 0);
1105 WRITE_ONCE(nlk->dst_group, 0);
1106 return 0;
1107 }
1108 if (addr->sa_family != AF_NETLINK)
1109 return -EINVAL;
1110
1111 if (alen < sizeof(struct sockaddr_nl))
1112 return -EINVAL;
1113
1114 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1115 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1116 return -EPERM;
1117
1118 /* No need for barriers here as we return to user-space without
1119 * using any of the bound attributes.
1120 * Paired with WRITE_ONCE() in netlink_insert().
1121 */
1122 if (!READ_ONCE(nlk->bound))
1123 err = netlink_autobind(sock);
1124
1125 if (err == 0) {
1126 /* paired with READ_ONCE() in netlink_getsockbyportid() */
1127 WRITE_ONCE(sk->sk_state, NETLINK_CONNECTED);
1128 /* dst_portid and dst_group can be read locklessly */
1129 WRITE_ONCE(nlk->dst_portid, nladdr->nl_pid);
1130 WRITE_ONCE(nlk->dst_group, ffs(nladdr->nl_groups));
1131 }
1132
1133 return err;
1134}
1135
1136static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1137 int peer)
1138{
1139 struct sock *sk = sock->sk;
1140 struct netlink_sock *nlk = nlk_sk(sk);
1141 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1142
1143 nladdr->nl_family = AF_NETLINK;
1144 nladdr->nl_pad = 0;
1145
1146 if (peer) {
1147 /* Paired with WRITE_ONCE() in netlink_connect() */
1148 nladdr->nl_pid = READ_ONCE(nlk->dst_portid);
1149 nladdr->nl_groups = netlink_group_mask(READ_ONCE(nlk->dst_group));
1150 } else {
1151 /* Paired with WRITE_ONCE() in netlink_insert() */
1152 nladdr->nl_pid = READ_ONCE(nlk->portid);
1153 netlink_lock_table();
1154 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1155 netlink_unlock_table();
1156 }
1157 return sizeof(*nladdr);
1158}
1159
1160static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1161 unsigned long arg)
1162{
1163 /* try to hand this ioctl down to the NIC drivers.
1164 */
1165 return -ENOIOCTLCMD;
1166}
1167
1168static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1169{
1170 struct sock *sock;
1171 struct netlink_sock *nlk;
1172
1173 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1174 if (!sock)
1175 return ERR_PTR(-ECONNREFUSED);
1176
1177 /* Don't bother queuing skb if kernel socket has no input function */
1178 nlk = nlk_sk(sock);
1179 /* dst_portid and sk_state can be changed in netlink_connect() */
1180 if (READ_ONCE(sock->sk_state) == NETLINK_CONNECTED &&
1181 READ_ONCE(nlk->dst_portid) != nlk_sk(ssk)->portid) {
1182 sock_put(sock);
1183 return ERR_PTR(-ECONNREFUSED);
1184 }
1185 return sock;
1186}
1187
1188struct sock *netlink_getsockbyfilp(struct file *filp)
1189{
1190 struct inode *inode = file_inode(filp);
1191 struct sock *sock;
1192
1193 if (!S_ISSOCK(inode->i_mode))
1194 return ERR_PTR(-ENOTSOCK);
1195
1196 sock = SOCKET_I(inode)->sk;
1197 if (sock->sk_family != AF_NETLINK)
1198 return ERR_PTR(-EINVAL);
1199
1200 sock_hold(sock);
1201 return sock;
1202}
1203
1204static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1205 int broadcast)
1206{
1207 struct sk_buff *skb;
1208 void *data;
1209
1210 if (size <= NLMSG_GOODSIZE || broadcast)
1211 return alloc_skb(size, GFP_KERNEL);
1212
1213 size = SKB_DATA_ALIGN(size) +
1214 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1215
1216 data = vmalloc(size);
1217 if (data == NULL)
1218 return NULL;
1219
1220 skb = __build_skb(data, size);
1221 if (skb == NULL)
1222 vfree(data);
1223 else
1224 skb->destructor = netlink_skb_destructor;
1225
1226 return skb;
1227}
1228
1229/*
1230 * Attach a skb to a netlink socket.
1231 * The caller must hold a reference to the destination socket. On error, the
1232 * reference is dropped. The skb is not send to the destination, just all
1233 * all error checks are performed and memory in the queue is reserved.
1234 * Return values:
1235 * < 0: error. skb freed, reference to sock dropped.
1236 * 0: continue
1237 * 1: repeat lookup - reference dropped while waiting for socket memory.
1238 */
1239int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1240 long *timeo, struct sock *ssk)
1241{
1242 struct netlink_sock *nlk;
1243
1244 nlk = nlk_sk(sk);
1245
1246 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1247 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1248 DECLARE_WAITQUEUE(wait, current);
1249 if (!*timeo) {
1250 if (!ssk || netlink_is_kernel(ssk))
1251 netlink_overrun(sk);
1252 sock_put(sk);
1253 kfree_skb(skb);
1254 return -EAGAIN;
1255 }
1256
1257 __set_current_state(TASK_INTERRUPTIBLE);
1258 add_wait_queue(&nlk->wait, &wait);
1259
1260 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1261 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1262 !sock_flag(sk, SOCK_DEAD))
1263 *timeo = schedule_timeout(*timeo);
1264
1265 __set_current_state(TASK_RUNNING);
1266 remove_wait_queue(&nlk->wait, &wait);
1267 sock_put(sk);
1268
1269 if (signal_pending(current)) {
1270 kfree_skb(skb);
1271 return sock_intr_errno(*timeo);
1272 }
1273 return 1;
1274 }
1275 netlink_skb_set_owner_r(skb, sk);
1276 return 0;
1277}
1278
1279static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1280{
1281 int len = skb->len;
1282
1283 netlink_deliver_tap(sock_net(sk), skb);
1284
1285 skb_queue_tail(&sk->sk_receive_queue, skb);
1286 sk->sk_data_ready(sk);
1287 return len;
1288}
1289
1290int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1291{
1292 int len = __netlink_sendskb(sk, skb);
1293
1294 sock_put(sk);
1295 return len;
1296}
1297
1298void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1299{
1300 kfree_skb(skb);
1301 sock_put(sk);
1302}
1303
1304static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1305{
1306 int delta;
1307
1308 WARN_ON(skb->sk != NULL);
1309 delta = skb->end - skb->tail;
1310 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1311 return skb;
1312
1313 if (skb_shared(skb)) {
1314 struct sk_buff *nskb = skb_clone(skb, allocation);
1315 if (!nskb)
1316 return skb;
1317 consume_skb(skb);
1318 skb = nskb;
1319 }
1320
1321 pskb_expand_head(skb, 0, -delta,
1322 (allocation & ~__GFP_DIRECT_RECLAIM) |
1323 __GFP_NOWARN | __GFP_NORETRY);
1324 return skb;
1325}
1326
1327static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1328 struct sock *ssk)
1329{
1330 int ret;
1331 struct netlink_sock *nlk = nlk_sk(sk);
1332
1333 ret = -ECONNREFUSED;
1334 if (nlk->netlink_rcv != NULL) {
1335 ret = skb->len;
1336 netlink_skb_set_owner_r(skb, sk);
1337 NETLINK_CB(skb).sk = ssk;
1338 netlink_deliver_tap_kernel(sk, ssk, skb);
1339 nlk->netlink_rcv(skb);
1340 consume_skb(skb);
1341 } else {
1342 kfree_skb(skb);
1343 }
1344 sock_put(sk);
1345 return ret;
1346}
1347
1348int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1349 u32 portid, int nonblock)
1350{
1351 struct sock *sk;
1352 int err;
1353 long timeo;
1354
1355 skb = netlink_trim(skb, gfp_any());
1356
1357 timeo = sock_sndtimeo(ssk, nonblock);
1358retry:
1359 sk = netlink_getsockbyportid(ssk, portid);
1360 if (IS_ERR(sk)) {
1361 kfree_skb(skb);
1362 return PTR_ERR(sk);
1363 }
1364 if (netlink_is_kernel(sk))
1365 return netlink_unicast_kernel(sk, skb, ssk);
1366
1367 if (sk_filter(sk, skb)) {
1368 err = skb->len;
1369 kfree_skb(skb);
1370 sock_put(sk);
1371 return err;
1372 }
1373
1374 err = netlink_attachskb(sk, skb, &timeo, ssk);
1375 if (err == 1)
1376 goto retry;
1377 if (err)
1378 return err;
1379
1380 return netlink_sendskb(sk, skb);
1381}
1382EXPORT_SYMBOL(netlink_unicast);
1383
1384int netlink_has_listeners(struct sock *sk, unsigned int group)
1385{
1386 int res = 0;
1387 struct listeners *listeners;
1388
1389 BUG_ON(!netlink_is_kernel(sk));
1390
1391 rcu_read_lock();
1392 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1393
1394 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1395 res = test_bit(group - 1, listeners->masks);
1396
1397 rcu_read_unlock();
1398
1399 return res;
1400}
1401EXPORT_SYMBOL_GPL(netlink_has_listeners);
1402
1403bool netlink_strict_get_check(struct sk_buff *skb)
1404{
1405 const struct netlink_sock *nlk = nlk_sk(NETLINK_CB(skb).sk);
1406
1407 return nlk->flags & NETLINK_F_STRICT_CHK;
1408}
1409EXPORT_SYMBOL_GPL(netlink_strict_get_check);
1410
1411static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1412{
1413 struct netlink_sock *nlk = nlk_sk(sk);
1414
1415 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1416 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1417 netlink_skb_set_owner_r(skb, sk);
1418 __netlink_sendskb(sk, skb);
1419 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1420 }
1421 return -1;
1422}
1423
1424struct netlink_broadcast_data {
1425 struct sock *exclude_sk;
1426 struct net *net;
1427 u32 portid;
1428 u32 group;
1429 int failure;
1430 int delivery_failure;
1431 int congested;
1432 int delivered;
1433 gfp_t allocation;
1434 struct sk_buff *skb, *skb2;
1435};
1436
1437static void do_one_broadcast(struct sock *sk,
1438 struct netlink_broadcast_data *p)
1439{
1440 struct netlink_sock *nlk = nlk_sk(sk);
1441 int val;
1442
1443 if (p->exclude_sk == sk)
1444 return;
1445
1446 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1447 !test_bit(p->group - 1, nlk->groups))
1448 return;
1449
1450 if (!net_eq(sock_net(sk), p->net)) {
1451 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1452 return;
1453
1454 if (!peernet_has_id(sock_net(sk), p->net))
1455 return;
1456
1457 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1458 CAP_NET_BROADCAST))
1459 return;
1460 }
1461
1462 if (p->failure) {
1463 netlink_overrun(sk);
1464 return;
1465 }
1466
1467 sock_hold(sk);
1468 if (p->skb2 == NULL) {
1469 if (skb_shared(p->skb)) {
1470 p->skb2 = skb_clone(p->skb, p->allocation);
1471 } else {
1472 p->skb2 = skb_get(p->skb);
1473 /*
1474 * skb ownership may have been set when
1475 * delivered to a previous socket.
1476 */
1477 skb_orphan(p->skb2);
1478 }
1479 }
1480 if (p->skb2 == NULL) {
1481 netlink_overrun(sk);
1482 /* Clone failed. Notify ALL listeners. */
1483 p->failure = 1;
1484 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1485 p->delivery_failure = 1;
1486 goto out;
1487 }
1488 if (sk_filter(sk, p->skb2)) {
1489 kfree_skb(p->skb2);
1490 p->skb2 = NULL;
1491 goto out;
1492 }
1493 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1494 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1495 NETLINK_CB(p->skb2).nsid_is_set = true;
1496 val = netlink_broadcast_deliver(sk, p->skb2);
1497 if (val < 0) {
1498 netlink_overrun(sk);
1499 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1500 p->delivery_failure = 1;
1501 } else {
1502 p->congested |= val;
1503 p->delivered = 1;
1504 p->skb2 = NULL;
1505 }
1506out:
1507 sock_put(sk);
1508}
1509
1510int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1511 u32 group, gfp_t allocation)
1512{
1513 struct net *net = sock_net(ssk);
1514 struct netlink_broadcast_data info;
1515 struct sock *sk;
1516
1517 skb = netlink_trim(skb, allocation);
1518
1519 info.exclude_sk = ssk;
1520 info.net = net;
1521 info.portid = portid;
1522 info.group = group;
1523 info.failure = 0;
1524 info.delivery_failure = 0;
1525 info.congested = 0;
1526 info.delivered = 0;
1527 info.allocation = allocation;
1528 info.skb = skb;
1529 info.skb2 = NULL;
1530
1531 /* While we sleep in clone, do not allow to change socket list */
1532
1533 netlink_lock_table();
1534
1535 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1536 do_one_broadcast(sk, &info);
1537
1538 consume_skb(skb);
1539
1540 netlink_unlock_table();
1541
1542 if (info.delivery_failure) {
1543 kfree_skb(info.skb2);
1544 return -ENOBUFS;
1545 }
1546 consume_skb(info.skb2);
1547
1548 if (info.delivered) {
1549 if (info.congested && gfpflags_allow_blocking(allocation))
1550 yield();
1551 return 0;
1552 }
1553 return -ESRCH;
1554}
1555EXPORT_SYMBOL(netlink_broadcast);
1556
1557struct netlink_set_err_data {
1558 struct sock *exclude_sk;
1559 u32 portid;
1560 u32 group;
1561 int code;
1562};
1563
1564static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1565{
1566 struct netlink_sock *nlk = nlk_sk(sk);
1567 int ret = 0;
1568
1569 if (sk == p->exclude_sk)
1570 goto out;
1571
1572 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1573 goto out;
1574
1575 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1576 !test_bit(p->group - 1, nlk->groups))
1577 goto out;
1578
1579 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1580 ret = 1;
1581 goto out;
1582 }
1583
1584 sk->sk_err = p->code;
1585 sk_error_report(sk);
1586out:
1587 return ret;
1588}
1589
1590/**
1591 * netlink_set_err - report error to broadcast listeners
1592 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1593 * @portid: the PORTID of a process that we want to skip (if any)
1594 * @group: the broadcast group that will notice the error
1595 * @code: error code, must be negative (as usual in kernelspace)
1596 *
1597 * This function returns the number of broadcast listeners that have set the
1598 * NETLINK_NO_ENOBUFS socket option.
1599 */
1600int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1601{
1602 struct netlink_set_err_data info;
1603 struct sock *sk;
1604 int ret = 0;
1605
1606 info.exclude_sk = ssk;
1607 info.portid = portid;
1608 info.group = group;
1609 /* sk->sk_err wants a positive error value */
1610 info.code = -code;
1611
1612 read_lock(&nl_table_lock);
1613
1614 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1615 ret += do_one_set_err(sk, &info);
1616
1617 read_unlock(&nl_table_lock);
1618 return ret;
1619}
1620EXPORT_SYMBOL(netlink_set_err);
1621
1622/* must be called with netlink table grabbed */
1623static void netlink_update_socket_mc(struct netlink_sock *nlk,
1624 unsigned int group,
1625 int is_new)
1626{
1627 int old, new = !!is_new, subscriptions;
1628
1629 old = test_bit(group - 1, nlk->groups);
1630 subscriptions = nlk->subscriptions - old + new;
1631 if (new)
1632 __set_bit(group - 1, nlk->groups);
1633 else
1634 __clear_bit(group - 1, nlk->groups);
1635 netlink_update_subscriptions(&nlk->sk, subscriptions);
1636 netlink_update_listeners(&nlk->sk);
1637}
1638
1639static int netlink_setsockopt(struct socket *sock, int level, int optname,
1640 sockptr_t optval, unsigned int optlen)
1641{
1642 struct sock *sk = sock->sk;
1643 struct netlink_sock *nlk = nlk_sk(sk);
1644 unsigned int val = 0;
1645 int err;
1646
1647 if (level != SOL_NETLINK)
1648 return -ENOPROTOOPT;
1649
1650 if (optlen >= sizeof(int) &&
1651 copy_from_sockptr(&val, optval, sizeof(val)))
1652 return -EFAULT;
1653
1654 switch (optname) {
1655 case NETLINK_PKTINFO:
1656 if (val)
1657 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1658 else
1659 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1660 err = 0;
1661 break;
1662 case NETLINK_ADD_MEMBERSHIP:
1663 case NETLINK_DROP_MEMBERSHIP: {
1664 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1665 return -EPERM;
1666 err = netlink_realloc_groups(sk);
1667 if (err)
1668 return err;
1669 if (!val || val - 1 >= nlk->ngroups)
1670 return -EINVAL;
1671 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1672 err = nlk->netlink_bind(sock_net(sk), val);
1673 if (err)
1674 return err;
1675 }
1676 netlink_table_grab();
1677 netlink_update_socket_mc(nlk, val,
1678 optname == NETLINK_ADD_MEMBERSHIP);
1679 netlink_table_ungrab();
1680 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1681 nlk->netlink_unbind(sock_net(sk), val);
1682
1683 err = 0;
1684 break;
1685 }
1686 case NETLINK_BROADCAST_ERROR:
1687 if (val)
1688 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1689 else
1690 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1691 err = 0;
1692 break;
1693 case NETLINK_NO_ENOBUFS:
1694 if (val) {
1695 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1696 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1697 wake_up_interruptible(&nlk->wait);
1698 } else {
1699 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1700 }
1701 err = 0;
1702 break;
1703 case NETLINK_LISTEN_ALL_NSID:
1704 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1705 return -EPERM;
1706
1707 if (val)
1708 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1709 else
1710 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1711 err = 0;
1712 break;
1713 case NETLINK_CAP_ACK:
1714 if (val)
1715 nlk->flags |= NETLINK_F_CAP_ACK;
1716 else
1717 nlk->flags &= ~NETLINK_F_CAP_ACK;
1718 err = 0;
1719 break;
1720 case NETLINK_EXT_ACK:
1721 if (val)
1722 nlk->flags |= NETLINK_F_EXT_ACK;
1723 else
1724 nlk->flags &= ~NETLINK_F_EXT_ACK;
1725 err = 0;
1726 break;
1727 case NETLINK_GET_STRICT_CHK:
1728 if (val)
1729 nlk->flags |= NETLINK_F_STRICT_CHK;
1730 else
1731 nlk->flags &= ~NETLINK_F_STRICT_CHK;
1732 err = 0;
1733 break;
1734 default:
1735 err = -ENOPROTOOPT;
1736 }
1737 return err;
1738}
1739
1740static int netlink_getsockopt(struct socket *sock, int level, int optname,
1741 char __user *optval, int __user *optlen)
1742{
1743 struct sock *sk = sock->sk;
1744 struct netlink_sock *nlk = nlk_sk(sk);
1745 int len, val, err;
1746
1747 if (level != SOL_NETLINK)
1748 return -ENOPROTOOPT;
1749
1750 if (get_user(len, optlen))
1751 return -EFAULT;
1752 if (len < 0)
1753 return -EINVAL;
1754
1755 switch (optname) {
1756 case NETLINK_PKTINFO:
1757 if (len < sizeof(int))
1758 return -EINVAL;
1759 len = sizeof(int);
1760 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1761 if (put_user(len, optlen) ||
1762 put_user(val, optval))
1763 return -EFAULT;
1764 err = 0;
1765 break;
1766 case NETLINK_BROADCAST_ERROR:
1767 if (len < sizeof(int))
1768 return -EINVAL;
1769 len = sizeof(int);
1770 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1771 if (put_user(len, optlen) ||
1772 put_user(val, optval))
1773 return -EFAULT;
1774 err = 0;
1775 break;
1776 case NETLINK_NO_ENOBUFS:
1777 if (len < sizeof(int))
1778 return -EINVAL;
1779 len = sizeof(int);
1780 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1781 if (put_user(len, optlen) ||
1782 put_user(val, optval))
1783 return -EFAULT;
1784 err = 0;
1785 break;
1786 case NETLINK_LIST_MEMBERSHIPS: {
1787 int pos, idx, shift;
1788
1789 err = 0;
1790 netlink_lock_table();
1791 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1792 if (len - pos < sizeof(u32))
1793 break;
1794
1795 idx = pos / sizeof(unsigned long);
1796 shift = (pos % sizeof(unsigned long)) * 8;
1797 if (put_user((u32)(nlk->groups[idx] >> shift),
1798 (u32 __user *)(optval + pos))) {
1799 err = -EFAULT;
1800 break;
1801 }
1802 }
1803 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1804 err = -EFAULT;
1805 netlink_unlock_table();
1806 break;
1807 }
1808 case NETLINK_CAP_ACK:
1809 if (len < sizeof(int))
1810 return -EINVAL;
1811 len = sizeof(int);
1812 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1813 if (put_user(len, optlen) ||
1814 put_user(val, optval))
1815 return -EFAULT;
1816 err = 0;
1817 break;
1818 case NETLINK_EXT_ACK:
1819 if (len < sizeof(int))
1820 return -EINVAL;
1821 len = sizeof(int);
1822 val = nlk->flags & NETLINK_F_EXT_ACK ? 1 : 0;
1823 if (put_user(len, optlen) || put_user(val, optval))
1824 return -EFAULT;
1825 err = 0;
1826 break;
1827 case NETLINK_GET_STRICT_CHK:
1828 if (len < sizeof(int))
1829 return -EINVAL;
1830 len = sizeof(int);
1831 val = nlk->flags & NETLINK_F_STRICT_CHK ? 1 : 0;
1832 if (put_user(len, optlen) || put_user(val, optval))
1833 return -EFAULT;
1834 err = 0;
1835 break;
1836 default:
1837 err = -ENOPROTOOPT;
1838 }
1839 return err;
1840}
1841
1842static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1843{
1844 struct nl_pktinfo info;
1845
1846 info.group = NETLINK_CB(skb).dst_group;
1847 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1848}
1849
1850static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1851 struct sk_buff *skb)
1852{
1853 if (!NETLINK_CB(skb).nsid_is_set)
1854 return;
1855
1856 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1857 &NETLINK_CB(skb).nsid);
1858}
1859
1860static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1861{
1862 struct sock *sk = sock->sk;
1863 struct netlink_sock *nlk = nlk_sk(sk);
1864 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1865 u32 dst_portid;
1866 u32 dst_group;
1867 struct sk_buff *skb;
1868 int err;
1869 struct scm_cookie scm;
1870 u32 netlink_skb_flags = 0;
1871
1872 if (msg->msg_flags & MSG_OOB)
1873 return -EOPNOTSUPP;
1874
1875 if (len == 0) {
1876 pr_warn_once("Zero length message leads to an empty skb\n");
1877 return -ENODATA;
1878 }
1879
1880 err = scm_send(sock, msg, &scm, true);
1881 if (err < 0)
1882 return err;
1883
1884 if (msg->msg_namelen) {
1885 err = -EINVAL;
1886 if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1887 goto out;
1888 if (addr->nl_family != AF_NETLINK)
1889 goto out;
1890 dst_portid = addr->nl_pid;
1891 dst_group = ffs(addr->nl_groups);
1892 err = -EPERM;
1893 if ((dst_group || dst_portid) &&
1894 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1895 goto out;
1896 netlink_skb_flags |= NETLINK_SKB_DST;
1897 } else {
1898 /* Paired with WRITE_ONCE() in netlink_connect() */
1899 dst_portid = READ_ONCE(nlk->dst_portid);
1900 dst_group = READ_ONCE(nlk->dst_group);
1901 }
1902
1903 /* Paired with WRITE_ONCE() in netlink_insert() */
1904 if (!READ_ONCE(nlk->bound)) {
1905 err = netlink_autobind(sock);
1906 if (err)
1907 goto out;
1908 } else {
1909 /* Ensure nlk is hashed and visible. */
1910 smp_rmb();
1911 }
1912
1913 err = -EMSGSIZE;
1914 if (len > sk->sk_sndbuf - 32)
1915 goto out;
1916 err = -ENOBUFS;
1917 skb = netlink_alloc_large_skb(len, dst_group);
1918 if (skb == NULL)
1919 goto out;
1920
1921 NETLINK_CB(skb).portid = nlk->portid;
1922 NETLINK_CB(skb).dst_group = dst_group;
1923 NETLINK_CB(skb).creds = scm.creds;
1924 NETLINK_CB(skb).flags = netlink_skb_flags;
1925
1926 err = -EFAULT;
1927 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1928 kfree_skb(skb);
1929 goto out;
1930 }
1931
1932 err = security_netlink_send(sk, skb);
1933 if (err) {
1934 kfree_skb(skb);
1935 goto out;
1936 }
1937
1938 if (dst_group) {
1939 refcount_inc(&skb->users);
1940 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1941 }
1942 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags & MSG_DONTWAIT);
1943
1944out:
1945 scm_destroy(&scm);
1946 return err;
1947}
1948
1949static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1950 int flags)
1951{
1952 struct scm_cookie scm;
1953 struct sock *sk = sock->sk;
1954 struct netlink_sock *nlk = nlk_sk(sk);
1955 size_t copied;
1956 struct sk_buff *skb, *data_skb;
1957 int err, ret;
1958
1959 if (flags & MSG_OOB)
1960 return -EOPNOTSUPP;
1961
1962 copied = 0;
1963
1964 skb = skb_recv_datagram(sk, flags, &err);
1965 if (skb == NULL)
1966 goto out;
1967
1968 data_skb = skb;
1969
1970#ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1971 if (unlikely(skb_shinfo(skb)->frag_list)) {
1972 /*
1973 * If this skb has a frag_list, then here that means that we
1974 * will have to use the frag_list skb's data for compat tasks
1975 * and the regular skb's data for normal (non-compat) tasks.
1976 *
1977 * If we need to send the compat skb, assign it to the
1978 * 'data_skb' variable so that it will be used below for data
1979 * copying. We keep 'skb' for everything else, including
1980 * freeing both later.
1981 */
1982 if (flags & MSG_CMSG_COMPAT)
1983 data_skb = skb_shinfo(skb)->frag_list;
1984 }
1985#endif
1986
1987 /* Record the max length of recvmsg() calls for future allocations */
1988 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1989 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1990 SKB_WITH_OVERHEAD(32768));
1991
1992 copied = data_skb->len;
1993 if (len < copied) {
1994 msg->msg_flags |= MSG_TRUNC;
1995 copied = len;
1996 }
1997
1998 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1999
2000 if (msg->msg_name) {
2001 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
2002 addr->nl_family = AF_NETLINK;
2003 addr->nl_pad = 0;
2004 addr->nl_pid = NETLINK_CB(skb).portid;
2005 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
2006 msg->msg_namelen = sizeof(*addr);
2007 }
2008
2009 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
2010 netlink_cmsg_recv_pktinfo(msg, skb);
2011 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
2012 netlink_cmsg_listen_all_nsid(sk, msg, skb);
2013
2014 memset(&scm, 0, sizeof(scm));
2015 scm.creds = *NETLINK_CREDS(skb);
2016 if (flags & MSG_TRUNC)
2017 copied = data_skb->len;
2018
2019 skb_free_datagram(sk, skb);
2020
2021 if (nlk->cb_running &&
2022 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2023 ret = netlink_dump(sk);
2024 if (ret) {
2025 sk->sk_err = -ret;
2026 sk_error_report(sk);
2027 }
2028 }
2029
2030 scm_recv(sock, msg, &scm, flags);
2031out:
2032 netlink_rcv_wake(sk);
2033 return err ? : copied;
2034}
2035
2036static void netlink_data_ready(struct sock *sk)
2037{
2038 BUG();
2039}
2040
2041/*
2042 * We export these functions to other modules. They provide a
2043 * complete set of kernel non-blocking support for message
2044 * queueing.
2045 */
2046
2047struct sock *
2048__netlink_kernel_create(struct net *net, int unit, struct module *module,
2049 struct netlink_kernel_cfg *cfg)
2050{
2051 struct socket *sock;
2052 struct sock *sk;
2053 struct netlink_sock *nlk;
2054 struct listeners *listeners = NULL;
2055 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2056 unsigned int groups;
2057
2058 BUG_ON(!nl_table);
2059
2060 if (unit < 0 || unit >= MAX_LINKS)
2061 return NULL;
2062
2063 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2064 return NULL;
2065
2066 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2067 goto out_sock_release_nosk;
2068
2069 sk = sock->sk;
2070
2071 if (!cfg || cfg->groups < 32)
2072 groups = 32;
2073 else
2074 groups = cfg->groups;
2075
2076 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2077 if (!listeners)
2078 goto out_sock_release;
2079
2080 sk->sk_data_ready = netlink_data_ready;
2081 if (cfg && cfg->input)
2082 nlk_sk(sk)->netlink_rcv = cfg->input;
2083
2084 if (netlink_insert(sk, 0))
2085 goto out_sock_release;
2086
2087 nlk = nlk_sk(sk);
2088 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2089
2090 netlink_table_grab();
2091 if (!nl_table[unit].registered) {
2092 nl_table[unit].groups = groups;
2093 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2094 nl_table[unit].cb_mutex = cb_mutex;
2095 nl_table[unit].module = module;
2096 if (cfg) {
2097 nl_table[unit].bind = cfg->bind;
2098 nl_table[unit].unbind = cfg->unbind;
2099 nl_table[unit].flags = cfg->flags;
2100 if (cfg->compare)
2101 nl_table[unit].compare = cfg->compare;
2102 }
2103 nl_table[unit].registered = 1;
2104 } else {
2105 kfree(listeners);
2106 nl_table[unit].registered++;
2107 }
2108 netlink_table_ungrab();
2109 return sk;
2110
2111out_sock_release:
2112 kfree(listeners);
2113 netlink_kernel_release(sk);
2114 return NULL;
2115
2116out_sock_release_nosk:
2117 sock_release(sock);
2118 return NULL;
2119}
2120EXPORT_SYMBOL(__netlink_kernel_create);
2121
2122void
2123netlink_kernel_release(struct sock *sk)
2124{
2125 if (sk == NULL || sk->sk_socket == NULL)
2126 return;
2127
2128 sock_release(sk->sk_socket);
2129}
2130EXPORT_SYMBOL(netlink_kernel_release);
2131
2132int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2133{
2134 struct listeners *new, *old;
2135 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2136
2137 if (groups < 32)
2138 groups = 32;
2139
2140 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2141 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2142 if (!new)
2143 return -ENOMEM;
2144 old = nl_deref_protected(tbl->listeners);
2145 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2146 rcu_assign_pointer(tbl->listeners, new);
2147
2148 kfree_rcu(old, rcu);
2149 }
2150 tbl->groups = groups;
2151
2152 return 0;
2153}
2154
2155/**
2156 * netlink_change_ngroups - change number of multicast groups
2157 *
2158 * This changes the number of multicast groups that are available
2159 * on a certain netlink family. Note that it is not possible to
2160 * change the number of groups to below 32. Also note that it does
2161 * not implicitly call netlink_clear_multicast_users() when the
2162 * number of groups is reduced.
2163 *
2164 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2165 * @groups: The new number of groups.
2166 */
2167int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2168{
2169 int err;
2170
2171 netlink_table_grab();
2172 err = __netlink_change_ngroups(sk, groups);
2173 netlink_table_ungrab();
2174
2175 return err;
2176}
2177
2178void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2179{
2180 struct sock *sk;
2181 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2182
2183 sk_for_each_bound(sk, &tbl->mc_list)
2184 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2185}
2186
2187struct nlmsghdr *
2188__nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2189{
2190 struct nlmsghdr *nlh;
2191 int size = nlmsg_msg_size(len);
2192
2193 nlh = skb_put(skb, NLMSG_ALIGN(size));
2194 nlh->nlmsg_type = type;
2195 nlh->nlmsg_len = size;
2196 nlh->nlmsg_flags = flags;
2197 nlh->nlmsg_pid = portid;
2198 nlh->nlmsg_seq = seq;
2199 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2200 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2201 return nlh;
2202}
2203EXPORT_SYMBOL(__nlmsg_put);
2204
2205/*
2206 * It looks a bit ugly.
2207 * It would be better to create kernel thread.
2208 */
2209
2210static int netlink_dump_done(struct netlink_sock *nlk, struct sk_buff *skb,
2211 struct netlink_callback *cb,
2212 struct netlink_ext_ack *extack)
2213{
2214 struct nlmsghdr *nlh;
2215
2216 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(nlk->dump_done_errno),
2217 NLM_F_MULTI | cb->answer_flags);
2218 if (WARN_ON(!nlh))
2219 return -ENOBUFS;
2220
2221 nl_dump_check_consistent(cb, nlh);
2222 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno, sizeof(nlk->dump_done_errno));
2223
2224 if (extack->_msg && nlk->flags & NETLINK_F_EXT_ACK) {
2225 nlh->nlmsg_flags |= NLM_F_ACK_TLVS;
2226 if (!nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg))
2227 nlmsg_end(skb, nlh);
2228 }
2229
2230 return 0;
2231}
2232
2233static int netlink_dump(struct sock *sk)
2234{
2235 struct netlink_sock *nlk = nlk_sk(sk);
2236 struct netlink_ext_ack extack = {};
2237 struct netlink_callback *cb;
2238 struct sk_buff *skb = NULL;
2239 struct module *module;
2240 int err = -ENOBUFS;
2241 int alloc_min_size;
2242 int alloc_size;
2243
2244 mutex_lock(nlk->cb_mutex);
2245 if (!nlk->cb_running) {
2246 err = -EINVAL;
2247 goto errout_skb;
2248 }
2249
2250 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2251 goto errout_skb;
2252
2253 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2254 * required, but it makes sense to _attempt_ a 16K bytes allocation
2255 * to reduce number of system calls on dump operations, if user
2256 * ever provided a big enough buffer.
2257 */
2258 cb = &nlk->cb;
2259 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2260
2261 if (alloc_min_size < nlk->max_recvmsg_len) {
2262 alloc_size = nlk->max_recvmsg_len;
2263 skb = alloc_skb(alloc_size,
2264 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2265 __GFP_NOWARN | __GFP_NORETRY);
2266 }
2267 if (!skb) {
2268 alloc_size = alloc_min_size;
2269 skb = alloc_skb(alloc_size, GFP_KERNEL);
2270 }
2271 if (!skb)
2272 goto errout_skb;
2273
2274 /* Trim skb to allocated size. User is expected to provide buffer as
2275 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2276 * netlink_recvmsg())). dump will pack as many smaller messages as
2277 * could fit within the allocated skb. skb is typically allocated
2278 * with larger space than required (could be as much as near 2x the
2279 * requested size with align to next power of 2 approach). Allowing
2280 * dump to use the excess space makes it difficult for a user to have a
2281 * reasonable static buffer based on the expected largest dump of a
2282 * single netdev. The outcome is MSG_TRUNC error.
2283 */
2284 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2285
2286 /* Make sure malicious BPF programs can not read unitialized memory
2287 * from skb->head -> skb->data
2288 */
2289 skb_reset_network_header(skb);
2290 skb_reset_mac_header(skb);
2291
2292 netlink_skb_set_owner_r(skb, sk);
2293
2294 if (nlk->dump_done_errno > 0) {
2295 cb->extack = &extack;
2296 nlk->dump_done_errno = cb->dump(skb, cb);
2297 cb->extack = NULL;
2298 }
2299
2300 if (nlk->dump_done_errno > 0 ||
2301 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2302 mutex_unlock(nlk->cb_mutex);
2303
2304 if (sk_filter(sk, skb))
2305 kfree_skb(skb);
2306 else
2307 __netlink_sendskb(sk, skb);
2308 return 0;
2309 }
2310
2311 if (netlink_dump_done(nlk, skb, cb, &extack))
2312 goto errout_skb;
2313
2314#ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2315 /* frag_list skb's data is used for compat tasks
2316 * and the regular skb's data for normal (non-compat) tasks.
2317 * See netlink_recvmsg().
2318 */
2319 if (unlikely(skb_shinfo(skb)->frag_list)) {
2320 if (netlink_dump_done(nlk, skb_shinfo(skb)->frag_list, cb, &extack))
2321 goto errout_skb;
2322 }
2323#endif
2324
2325 if (sk_filter(sk, skb))
2326 kfree_skb(skb);
2327 else
2328 __netlink_sendskb(sk, skb);
2329
2330 if (cb->done)
2331 cb->done(cb);
2332
2333 nlk->cb_running = false;
2334 module = cb->module;
2335 skb = cb->skb;
2336 mutex_unlock(nlk->cb_mutex);
2337 module_put(module);
2338 consume_skb(skb);
2339 return 0;
2340
2341errout_skb:
2342 mutex_unlock(nlk->cb_mutex);
2343 kfree_skb(skb);
2344 return err;
2345}
2346
2347int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2348 const struct nlmsghdr *nlh,
2349 struct netlink_dump_control *control)
2350{
2351 struct netlink_sock *nlk, *nlk2;
2352 struct netlink_callback *cb;
2353 struct sock *sk;
2354 int ret;
2355
2356 refcount_inc(&skb->users);
2357
2358 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2359 if (sk == NULL) {
2360 ret = -ECONNREFUSED;
2361 goto error_free;
2362 }
2363
2364 nlk = nlk_sk(sk);
2365 mutex_lock(nlk->cb_mutex);
2366 /* A dump is in progress... */
2367 if (nlk->cb_running) {
2368 ret = -EBUSY;
2369 goto error_unlock;
2370 }
2371 /* add reference of module which cb->dump belongs to */
2372 if (!try_module_get(control->module)) {
2373 ret = -EPROTONOSUPPORT;
2374 goto error_unlock;
2375 }
2376
2377 cb = &nlk->cb;
2378 memset(cb, 0, sizeof(*cb));
2379 cb->dump = control->dump;
2380 cb->done = control->done;
2381 cb->nlh = nlh;
2382 cb->data = control->data;
2383 cb->module = control->module;
2384 cb->min_dump_alloc = control->min_dump_alloc;
2385 cb->skb = skb;
2386
2387 nlk2 = nlk_sk(NETLINK_CB(skb).sk);
2388 cb->strict_check = !!(nlk2->flags & NETLINK_F_STRICT_CHK);
2389
2390 if (control->start) {
2391 ret = control->start(cb);
2392 if (ret)
2393 goto error_put;
2394 }
2395
2396 nlk->cb_running = true;
2397 nlk->dump_done_errno = INT_MAX;
2398
2399 mutex_unlock(nlk->cb_mutex);
2400
2401 ret = netlink_dump(sk);
2402
2403 sock_put(sk);
2404
2405 if (ret)
2406 return ret;
2407
2408 /* We successfully started a dump, by returning -EINTR we
2409 * signal not to send ACK even if it was requested.
2410 */
2411 return -EINTR;
2412
2413error_put:
2414 module_put(control->module);
2415error_unlock:
2416 sock_put(sk);
2417 mutex_unlock(nlk->cb_mutex);
2418error_free:
2419 kfree_skb(skb);
2420 return ret;
2421}
2422EXPORT_SYMBOL(__netlink_dump_start);
2423
2424static size_t
2425netlink_ack_tlv_len(struct netlink_sock *nlk, int err,
2426 const struct netlink_ext_ack *extack)
2427{
2428 size_t tlvlen;
2429
2430 if (!extack || !(nlk->flags & NETLINK_F_EXT_ACK))
2431 return 0;
2432
2433 tlvlen = 0;
2434 if (extack->_msg)
2435 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2436 if (extack->cookie_len)
2437 tlvlen += nla_total_size(extack->cookie_len);
2438
2439 /* Following attributes are only reported as error (not warning) */
2440 if (!err)
2441 return tlvlen;
2442
2443 if (extack->bad_attr)
2444 tlvlen += nla_total_size(sizeof(u32));
2445 if (extack->policy)
2446 tlvlen += netlink_policy_dump_attr_size_estimate(extack->policy);
2447 if (extack->miss_type)
2448 tlvlen += nla_total_size(sizeof(u32));
2449 if (extack->miss_nest)
2450 tlvlen += nla_total_size(sizeof(u32));
2451
2452 return tlvlen;
2453}
2454
2455static void
2456netlink_ack_tlv_fill(struct sk_buff *in_skb, struct sk_buff *skb,
2457 struct nlmsghdr *nlh, int err,
2458 const struct netlink_ext_ack *extack)
2459{
2460 if (extack->_msg)
2461 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg));
2462 if (extack->cookie_len)
2463 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2464 extack->cookie_len, extack->cookie));
2465
2466 if (!err)
2467 return;
2468
2469 if (extack->bad_attr &&
2470 !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2471 (u8 *)extack->bad_attr >= in_skb->data + in_skb->len))
2472 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2473 (u8 *)extack->bad_attr - (u8 *)nlh));
2474 if (extack->policy)
2475 netlink_policy_dump_write_attr(skb, extack->policy,
2476 NLMSGERR_ATTR_POLICY);
2477 if (extack->miss_type)
2478 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_TYPE,
2479 extack->miss_type));
2480 if (extack->miss_nest &&
2481 !WARN_ON((u8 *)extack->miss_nest < in_skb->data ||
2482 (u8 *)extack->miss_nest > in_skb->data + in_skb->len))
2483 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_NEST,
2484 (u8 *)extack->miss_nest - (u8 *)nlh));
2485}
2486
2487void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2488 const struct netlink_ext_ack *extack)
2489{
2490 struct sk_buff *skb;
2491 struct nlmsghdr *rep;
2492 struct nlmsgerr *errmsg;
2493 size_t payload = sizeof(*errmsg);
2494 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2495 unsigned int flags = 0;
2496 size_t tlvlen;
2497
2498 /* Error messages get the original request appened, unless the user
2499 * requests to cap the error message, and get extra error data if
2500 * requested.
2501 */
2502 if (err && !(nlk->flags & NETLINK_F_CAP_ACK))
2503 payload += nlmsg_len(nlh);
2504 else
2505 flags |= NLM_F_CAPPED;
2506
2507 tlvlen = netlink_ack_tlv_len(nlk, err, extack);
2508 if (tlvlen)
2509 flags |= NLM_F_ACK_TLVS;
2510
2511 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2512 if (!skb)
2513 goto err_skb;
2514
2515 rep = nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2516 NLMSG_ERROR, sizeof(*errmsg), flags);
2517 if (!rep)
2518 goto err_bad_put;
2519 errmsg = nlmsg_data(rep);
2520 errmsg->error = err;
2521 errmsg->msg = *nlh;
2522
2523 if (!(flags & NLM_F_CAPPED)) {
2524 if (!nlmsg_append(skb, nlmsg_len(nlh)))
2525 goto err_bad_put;
2526
2527 memcpy(nlmsg_data(&errmsg->msg), nlmsg_data(nlh),
2528 nlmsg_len(nlh));
2529 }
2530
2531 if (tlvlen)
2532 netlink_ack_tlv_fill(in_skb, skb, nlh, err, extack);
2533
2534 nlmsg_end(skb, rep);
2535
2536 nlmsg_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid);
2537
2538 return;
2539
2540err_bad_put:
2541 nlmsg_free(skb);
2542err_skb:
2543 NETLINK_CB(in_skb).sk->sk_err = ENOBUFS;
2544 sk_error_report(NETLINK_CB(in_skb).sk);
2545}
2546EXPORT_SYMBOL(netlink_ack);
2547
2548int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2549 struct nlmsghdr *,
2550 struct netlink_ext_ack *))
2551{
2552 struct netlink_ext_ack extack;
2553 struct nlmsghdr *nlh;
2554 int err;
2555
2556 while (skb->len >= nlmsg_total_size(0)) {
2557 int msglen;
2558
2559 memset(&extack, 0, sizeof(extack));
2560 nlh = nlmsg_hdr(skb);
2561 err = 0;
2562
2563 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2564 return 0;
2565
2566 /* Only requests are handled by the kernel */
2567 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2568 goto ack;
2569
2570 /* Skip control messages */
2571 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2572 goto ack;
2573
2574 err = cb(skb, nlh, &extack);
2575 if (err == -EINTR)
2576 goto skip;
2577
2578ack:
2579 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2580 netlink_ack(skb, nlh, err, &extack);
2581
2582skip:
2583 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2584 if (msglen > skb->len)
2585 msglen = skb->len;
2586 skb_pull(skb, msglen);
2587 }
2588
2589 return 0;
2590}
2591EXPORT_SYMBOL(netlink_rcv_skb);
2592
2593/**
2594 * nlmsg_notify - send a notification netlink message
2595 * @sk: netlink socket to use
2596 * @skb: notification message
2597 * @portid: destination netlink portid for reports or 0
2598 * @group: destination multicast group or 0
2599 * @report: 1 to report back, 0 to disable
2600 * @flags: allocation flags
2601 */
2602int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2603 unsigned int group, int report, gfp_t flags)
2604{
2605 int err = 0;
2606
2607 if (group) {
2608 int exclude_portid = 0;
2609
2610 if (report) {
2611 refcount_inc(&skb->users);
2612 exclude_portid = portid;
2613 }
2614
2615 /* errors reported via destination sk->sk_err, but propagate
2616 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2617 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2618 if (err == -ESRCH)
2619 err = 0;
2620 }
2621
2622 if (report) {
2623 int err2;
2624
2625 err2 = nlmsg_unicast(sk, skb, portid);
2626 if (!err)
2627 err = err2;
2628 }
2629
2630 return err;
2631}
2632EXPORT_SYMBOL(nlmsg_notify);
2633
2634#ifdef CONFIG_PROC_FS
2635struct nl_seq_iter {
2636 struct seq_net_private p;
2637 struct rhashtable_iter hti;
2638 int link;
2639};
2640
2641static void netlink_walk_start(struct nl_seq_iter *iter)
2642{
2643 rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti);
2644 rhashtable_walk_start(&iter->hti);
2645}
2646
2647static void netlink_walk_stop(struct nl_seq_iter *iter)
2648{
2649 rhashtable_walk_stop(&iter->hti);
2650 rhashtable_walk_exit(&iter->hti);
2651}
2652
2653static void *__netlink_seq_next(struct seq_file *seq)
2654{
2655 struct nl_seq_iter *iter = seq->private;
2656 struct netlink_sock *nlk;
2657
2658 do {
2659 for (;;) {
2660 nlk = rhashtable_walk_next(&iter->hti);
2661
2662 if (IS_ERR(nlk)) {
2663 if (PTR_ERR(nlk) == -EAGAIN)
2664 continue;
2665
2666 return nlk;
2667 }
2668
2669 if (nlk)
2670 break;
2671
2672 netlink_walk_stop(iter);
2673 if (++iter->link >= MAX_LINKS)
2674 return NULL;
2675
2676 netlink_walk_start(iter);
2677 }
2678 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2679
2680 return nlk;
2681}
2682
2683static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2684 __acquires(RCU)
2685{
2686 struct nl_seq_iter *iter = seq->private;
2687 void *obj = SEQ_START_TOKEN;
2688 loff_t pos;
2689
2690 iter->link = 0;
2691
2692 netlink_walk_start(iter);
2693
2694 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2695 obj = __netlink_seq_next(seq);
2696
2697 return obj;
2698}
2699
2700static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2701{
2702 ++*pos;
2703 return __netlink_seq_next(seq);
2704}
2705
2706static void netlink_native_seq_stop(struct seq_file *seq, void *v)
2707{
2708 struct nl_seq_iter *iter = seq->private;
2709
2710 if (iter->link >= MAX_LINKS)
2711 return;
2712
2713 netlink_walk_stop(iter);
2714}
2715
2716
2717static int netlink_native_seq_show(struct seq_file *seq, void *v)
2718{
2719 if (v == SEQ_START_TOKEN) {
2720 seq_puts(seq,
2721 "sk Eth Pid Groups "
2722 "Rmem Wmem Dump Locks Drops Inode\n");
2723 } else {
2724 struct sock *s = v;
2725 struct netlink_sock *nlk = nlk_sk(s);
2726
2727 seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n",
2728 s,
2729 s->sk_protocol,
2730 nlk->portid,
2731 nlk->groups ? (u32)nlk->groups[0] : 0,
2732 sk_rmem_alloc_get(s),
2733 sk_wmem_alloc_get(s),
2734 nlk->cb_running,
2735 refcount_read(&s->sk_refcnt),
2736 atomic_read(&s->sk_drops),
2737 sock_i_ino(s)
2738 );
2739
2740 }
2741 return 0;
2742}
2743
2744#ifdef CONFIG_BPF_SYSCALL
2745struct bpf_iter__netlink {
2746 __bpf_md_ptr(struct bpf_iter_meta *, meta);
2747 __bpf_md_ptr(struct netlink_sock *, sk);
2748};
2749
2750DEFINE_BPF_ITER_FUNC(netlink, struct bpf_iter_meta *meta, struct netlink_sock *sk)
2751
2752static int netlink_prog_seq_show(struct bpf_prog *prog,
2753 struct bpf_iter_meta *meta,
2754 void *v)
2755{
2756 struct bpf_iter__netlink ctx;
2757
2758 meta->seq_num--; /* skip SEQ_START_TOKEN */
2759 ctx.meta = meta;
2760 ctx.sk = nlk_sk((struct sock *)v);
2761 return bpf_iter_run_prog(prog, &ctx);
2762}
2763
2764static int netlink_seq_show(struct seq_file *seq, void *v)
2765{
2766 struct bpf_iter_meta meta;
2767 struct bpf_prog *prog;
2768
2769 meta.seq = seq;
2770 prog = bpf_iter_get_info(&meta, false);
2771 if (!prog)
2772 return netlink_native_seq_show(seq, v);
2773
2774 if (v != SEQ_START_TOKEN)
2775 return netlink_prog_seq_show(prog, &meta, v);
2776
2777 return 0;
2778}
2779
2780static void netlink_seq_stop(struct seq_file *seq, void *v)
2781{
2782 struct bpf_iter_meta meta;
2783 struct bpf_prog *prog;
2784
2785 if (!v) {
2786 meta.seq = seq;
2787 prog = bpf_iter_get_info(&meta, true);
2788 if (prog)
2789 (void)netlink_prog_seq_show(prog, &meta, v);
2790 }
2791
2792 netlink_native_seq_stop(seq, v);
2793}
2794#else
2795static int netlink_seq_show(struct seq_file *seq, void *v)
2796{
2797 return netlink_native_seq_show(seq, v);
2798}
2799
2800static void netlink_seq_stop(struct seq_file *seq, void *v)
2801{
2802 netlink_native_seq_stop(seq, v);
2803}
2804#endif
2805
2806static const struct seq_operations netlink_seq_ops = {
2807 .start = netlink_seq_start,
2808 .next = netlink_seq_next,
2809 .stop = netlink_seq_stop,
2810 .show = netlink_seq_show,
2811};
2812#endif
2813
2814int netlink_register_notifier(struct notifier_block *nb)
2815{
2816 return blocking_notifier_chain_register(&netlink_chain, nb);
2817}
2818EXPORT_SYMBOL(netlink_register_notifier);
2819
2820int netlink_unregister_notifier(struct notifier_block *nb)
2821{
2822 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2823}
2824EXPORT_SYMBOL(netlink_unregister_notifier);
2825
2826static const struct proto_ops netlink_ops = {
2827 .family = PF_NETLINK,
2828 .owner = THIS_MODULE,
2829 .release = netlink_release,
2830 .bind = netlink_bind,
2831 .connect = netlink_connect,
2832 .socketpair = sock_no_socketpair,
2833 .accept = sock_no_accept,
2834 .getname = netlink_getname,
2835 .poll = datagram_poll,
2836 .ioctl = netlink_ioctl,
2837 .listen = sock_no_listen,
2838 .shutdown = sock_no_shutdown,
2839 .setsockopt = netlink_setsockopt,
2840 .getsockopt = netlink_getsockopt,
2841 .sendmsg = netlink_sendmsg,
2842 .recvmsg = netlink_recvmsg,
2843 .mmap = sock_no_mmap,
2844 .sendpage = sock_no_sendpage,
2845};
2846
2847static const struct net_proto_family netlink_family_ops = {
2848 .family = PF_NETLINK,
2849 .create = netlink_create,
2850 .owner = THIS_MODULE, /* for consistency 8) */
2851};
2852
2853static int __net_init netlink_net_init(struct net *net)
2854{
2855#ifdef CONFIG_PROC_FS
2856 if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2857 sizeof(struct nl_seq_iter)))
2858 return -ENOMEM;
2859#endif
2860 return 0;
2861}
2862
2863static void __net_exit netlink_net_exit(struct net *net)
2864{
2865#ifdef CONFIG_PROC_FS
2866 remove_proc_entry("netlink", net->proc_net);
2867#endif
2868}
2869
2870static void __init netlink_add_usersock_entry(void)
2871{
2872 struct listeners *listeners;
2873 int groups = 32;
2874
2875 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2876 if (!listeners)
2877 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2878
2879 netlink_table_grab();
2880
2881 nl_table[NETLINK_USERSOCK].groups = groups;
2882 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2883 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2884 nl_table[NETLINK_USERSOCK].registered = 1;
2885 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2886
2887 netlink_table_ungrab();
2888}
2889
2890static struct pernet_operations __net_initdata netlink_net_ops = {
2891 .init = netlink_net_init,
2892 .exit = netlink_net_exit,
2893};
2894
2895static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2896{
2897 const struct netlink_sock *nlk = data;
2898 struct netlink_compare_arg arg;
2899
2900 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2901 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2902}
2903
2904static const struct rhashtable_params netlink_rhashtable_params = {
2905 .head_offset = offsetof(struct netlink_sock, node),
2906 .key_len = netlink_compare_arg_len,
2907 .obj_hashfn = netlink_hash,
2908 .obj_cmpfn = netlink_compare,
2909 .automatic_shrinking = true,
2910};
2911
2912#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2913BTF_ID_LIST(btf_netlink_sock_id)
2914BTF_ID(struct, netlink_sock)
2915
2916static const struct bpf_iter_seq_info netlink_seq_info = {
2917 .seq_ops = &netlink_seq_ops,
2918 .init_seq_private = bpf_iter_init_seq_net,
2919 .fini_seq_private = bpf_iter_fini_seq_net,
2920 .seq_priv_size = sizeof(struct nl_seq_iter),
2921};
2922
2923static struct bpf_iter_reg netlink_reg_info = {
2924 .target = "netlink",
2925 .ctx_arg_info_size = 1,
2926 .ctx_arg_info = {
2927 { offsetof(struct bpf_iter__netlink, sk),
2928 PTR_TO_BTF_ID_OR_NULL },
2929 },
2930 .seq_info = &netlink_seq_info,
2931};
2932
2933static int __init bpf_iter_register(void)
2934{
2935 netlink_reg_info.ctx_arg_info[0].btf_id = *btf_netlink_sock_id;
2936 return bpf_iter_reg_target(&netlink_reg_info);
2937}
2938#endif
2939
2940static int __init netlink_proto_init(void)
2941{
2942 int i;
2943 int err = proto_register(&netlink_proto, 0);
2944
2945 if (err != 0)
2946 goto out;
2947
2948#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2949 err = bpf_iter_register();
2950 if (err)
2951 goto out;
2952#endif
2953
2954 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof_field(struct sk_buff, cb));
2955
2956 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2957 if (!nl_table)
2958 goto panic;
2959
2960 for (i = 0; i < MAX_LINKS; i++) {
2961 if (rhashtable_init(&nl_table[i].hash,
2962 &netlink_rhashtable_params) < 0) {
2963 while (--i > 0)
2964 rhashtable_destroy(&nl_table[i].hash);
2965 kfree(nl_table);
2966 goto panic;
2967 }
2968 }
2969
2970 netlink_add_usersock_entry();
2971
2972 sock_register(&netlink_family_ops);
2973 register_pernet_subsys(&netlink_net_ops);
2974 register_pernet_subsys(&netlink_tap_net_ops);
2975 /* The netlink device handler may be needed early. */
2976 rtnetlink_init();
2977out:
2978 return err;
2979panic:
2980 panic("netlink_init: Cannot allocate nl_table\n");
2981}
2982
2983core_initcall(netlink_proto_init);