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