Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * NETLINK Kernel-user communication protocol.
4 *
5 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
6 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
7 * Patrick McHardy <kaber@trash.net>
8 *
9 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
10 * added netlink_proto_exit
11 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
12 * use nlk_sk, as sk->protinfo is on a diet 8)
13 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
14 * - inc module use count of module that owns
15 * the kernel socket in case userspace opens
16 * socket of same protocol
17 * - remove all module support, since netlink is
18 * mandatory if CONFIG_NET=y these days
19 */
20
21#include <linux/module.h>
22
23#include <linux/capability.h>
24#include <linux/kernel.h>
25#include <linux/init.h>
26#include <linux/signal.h>
27#include <linux/sched.h>
28#include <linux/errno.h>
29#include <linux/string.h>
30#include <linux/stat.h>
31#include <linux/socket.h>
32#include <linux/un.h>
33#include <linux/fcntl.h>
34#include <linux/termios.h>
35#include <linux/sockios.h>
36#include <linux/net.h>
37#include <linux/fs.h>
38#include <linux/slab.h>
39#include <linux/uaccess.h>
40#include <linux/skbuff.h>
41#include <linux/netdevice.h>
42#include <linux/rtnetlink.h>
43#include <linux/proc_fs.h>
44#include <linux/seq_file.h>
45#include <linux/notifier.h>
46#include <linux/security.h>
47#include <linux/jhash.h>
48#include <linux/jiffies.h>
49#include <linux/random.h>
50#include <linux/bitops.h>
51#include <linux/mm.h>
52#include <linux/types.h>
53#include <linux/audit.h>
54#include <linux/mutex.h>
55#include <linux/vmalloc.h>
56#include <linux/if_arp.h>
57#include <linux/rhashtable.h>
58#include <asm/cacheflush.h>
59#include <linux/hash.h>
60#include <linux/genetlink.h>
61#include <linux/net_namespace.h>
62#include <linux/nospec.h>
63
64#include <net/net_namespace.h>
65#include <net/netns/generic.h>
66#include <net/sock.h>
67#include <net/scm.h>
68#include <net/netlink.h>
69
70#include "af_netlink.h"
71
72struct listeners {
73 struct rcu_head rcu;
74 unsigned long masks[0];
75};
76
77/* state bits */
78#define NETLINK_S_CONGESTED 0x0
79
80static inline int netlink_is_kernel(struct sock *sk)
81{
82 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
83}
84
85struct netlink_table *nl_table __read_mostly;
86EXPORT_SYMBOL_GPL(nl_table);
87
88static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
89
90static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
91
92static const char *const nlk_cb_mutex_key_strings[MAX_LINKS + 1] = {
93 "nlk_cb_mutex-ROUTE",
94 "nlk_cb_mutex-1",
95 "nlk_cb_mutex-USERSOCK",
96 "nlk_cb_mutex-FIREWALL",
97 "nlk_cb_mutex-SOCK_DIAG",
98 "nlk_cb_mutex-NFLOG",
99 "nlk_cb_mutex-XFRM",
100 "nlk_cb_mutex-SELINUX",
101 "nlk_cb_mutex-ISCSI",
102 "nlk_cb_mutex-AUDIT",
103 "nlk_cb_mutex-FIB_LOOKUP",
104 "nlk_cb_mutex-CONNECTOR",
105 "nlk_cb_mutex-NETFILTER",
106 "nlk_cb_mutex-IP6_FW",
107 "nlk_cb_mutex-DNRTMSG",
108 "nlk_cb_mutex-KOBJECT_UEVENT",
109 "nlk_cb_mutex-GENERIC",
110 "nlk_cb_mutex-17",
111 "nlk_cb_mutex-SCSITRANSPORT",
112 "nlk_cb_mutex-ECRYPTFS",
113 "nlk_cb_mutex-RDMA",
114 "nlk_cb_mutex-CRYPTO",
115 "nlk_cb_mutex-SMC",
116 "nlk_cb_mutex-23",
117 "nlk_cb_mutex-24",
118 "nlk_cb_mutex-25",
119 "nlk_cb_mutex-26",
120 "nlk_cb_mutex-27",
121 "nlk_cb_mutex-28",
122 "nlk_cb_mutex-29",
123 "nlk_cb_mutex-30",
124 "nlk_cb_mutex-31",
125 "nlk_cb_mutex-MAX_LINKS"
126};
127
128static int netlink_dump(struct sock *sk);
129
130/* nl_table locking explained:
131 * Lookup and traversal are protected with an RCU read-side lock. Insertion
132 * and removal are protected with per bucket lock while using RCU list
133 * modification primitives and may run in parallel to RCU protected lookups.
134 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
135 * been acquired * either during or after the socket has been removed from
136 * the list and after an RCU grace period.
137 */
138DEFINE_RWLOCK(nl_table_lock);
139EXPORT_SYMBOL_GPL(nl_table_lock);
140static atomic_t nl_table_users = ATOMIC_INIT(0);
141
142#define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
143
144static BLOCKING_NOTIFIER_HEAD(netlink_chain);
145
146
147static const struct rhashtable_params netlink_rhashtable_params;
148
149static inline u32 netlink_group_mask(u32 group)
150{
151 return group ? 1 << (group - 1) : 0;
152}
153
154static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
155 gfp_t gfp_mask)
156{
157 unsigned int len = skb_end_offset(skb);
158 struct sk_buff *new;
159
160 new = alloc_skb(len, gfp_mask);
161 if (new == NULL)
162 return NULL;
163
164 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
165 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
166 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
167
168 skb_put_data(new, skb->data, len);
169 return new;
170}
171
172static unsigned int netlink_tap_net_id;
173
174struct netlink_tap_net {
175 struct list_head netlink_tap_all;
176 struct mutex netlink_tap_lock;
177};
178
179int netlink_add_tap(struct netlink_tap *nt)
180{
181 struct net *net = dev_net(nt->dev);
182 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
183
184 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
185 return -EINVAL;
186
187 mutex_lock(&nn->netlink_tap_lock);
188 list_add_rcu(&nt->list, &nn->netlink_tap_all);
189 mutex_unlock(&nn->netlink_tap_lock);
190
191 __module_get(nt->module);
192
193 return 0;
194}
195EXPORT_SYMBOL_GPL(netlink_add_tap);
196
197static int __netlink_remove_tap(struct netlink_tap *nt)
198{
199 struct net *net = dev_net(nt->dev);
200 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
201 bool found = false;
202 struct netlink_tap *tmp;
203
204 mutex_lock(&nn->netlink_tap_lock);
205
206 list_for_each_entry(tmp, &nn->netlink_tap_all, list) {
207 if (nt == tmp) {
208 list_del_rcu(&nt->list);
209 found = true;
210 goto out;
211 }
212 }
213
214 pr_warn("__netlink_remove_tap: %p not found\n", nt);
215out:
216 mutex_unlock(&nn->netlink_tap_lock);
217
218 if (found)
219 module_put(nt->module);
220
221 return found ? 0 : -ENODEV;
222}
223
224int netlink_remove_tap(struct netlink_tap *nt)
225{
226 int ret;
227
228 ret = __netlink_remove_tap(nt);
229 synchronize_net();
230
231 return ret;
232}
233EXPORT_SYMBOL_GPL(netlink_remove_tap);
234
235static __net_init int netlink_tap_init_net(struct net *net)
236{
237 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
238
239 INIT_LIST_HEAD(&nn->netlink_tap_all);
240 mutex_init(&nn->netlink_tap_lock);
241 return 0;
242}
243
244static struct pernet_operations netlink_tap_net_ops = {
245 .init = netlink_tap_init_net,
246 .id = &netlink_tap_net_id,
247 .size = sizeof(struct netlink_tap_net),
248};
249
250static bool netlink_filter_tap(const struct sk_buff *skb)
251{
252 struct sock *sk = skb->sk;
253
254 /* We take the more conservative approach and
255 * whitelist socket protocols that may pass.
256 */
257 switch (sk->sk_protocol) {
258 case NETLINK_ROUTE:
259 case NETLINK_USERSOCK:
260 case NETLINK_SOCK_DIAG:
261 case NETLINK_NFLOG:
262 case NETLINK_XFRM:
263 case NETLINK_FIB_LOOKUP:
264 case NETLINK_NETFILTER:
265 case NETLINK_GENERIC:
266 return true;
267 }
268
269 return false;
270}
271
272static int __netlink_deliver_tap_skb(struct sk_buff *skb,
273 struct net_device *dev)
274{
275 struct sk_buff *nskb;
276 struct sock *sk = skb->sk;
277 int ret = -ENOMEM;
278
279 if (!net_eq(dev_net(dev), sock_net(sk)))
280 return 0;
281
282 dev_hold(dev);
283
284 if (is_vmalloc_addr(skb->head))
285 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
286 else
287 nskb = skb_clone(skb, GFP_ATOMIC);
288 if (nskb) {
289 nskb->dev = dev;
290 nskb->protocol = htons((u16) sk->sk_protocol);
291 nskb->pkt_type = netlink_is_kernel(sk) ?
292 PACKET_KERNEL : PACKET_USER;
293 skb_reset_network_header(nskb);
294 ret = dev_queue_xmit(nskb);
295 if (unlikely(ret > 0))
296 ret = net_xmit_errno(ret);
297 }
298
299 dev_put(dev);
300 return ret;
301}
302
303static void __netlink_deliver_tap(struct sk_buff *skb, struct netlink_tap_net *nn)
304{
305 int ret;
306 struct netlink_tap *tmp;
307
308 if (!netlink_filter_tap(skb))
309 return;
310
311 list_for_each_entry_rcu(tmp, &nn->netlink_tap_all, list) {
312 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
313 if (unlikely(ret))
314 break;
315 }
316}
317
318static void netlink_deliver_tap(struct net *net, struct sk_buff *skb)
319{
320 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
321
322 rcu_read_lock();
323
324 if (unlikely(!list_empty(&nn->netlink_tap_all)))
325 __netlink_deliver_tap(skb, nn);
326
327 rcu_read_unlock();
328}
329
330static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
331 struct sk_buff *skb)
332{
333 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
334 netlink_deliver_tap(sock_net(dst), skb);
335}
336
337static void netlink_overrun(struct sock *sk)
338{
339 struct netlink_sock *nlk = nlk_sk(sk);
340
341 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
342 if (!test_and_set_bit(NETLINK_S_CONGESTED,
343 &nlk_sk(sk)->state)) {
344 sk->sk_err = ENOBUFS;
345 sk->sk_error_report(sk);
346 }
347 }
348 atomic_inc(&sk->sk_drops);
349}
350
351static void netlink_rcv_wake(struct sock *sk)
352{
353 struct netlink_sock *nlk = nlk_sk(sk);
354
355 if (skb_queue_empty(&sk->sk_receive_queue))
356 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
357 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
358 wake_up_interruptible(&nlk->wait);
359}
360
361static void netlink_skb_destructor(struct sk_buff *skb)
362{
363 if (is_vmalloc_addr(skb->head)) {
364 if (!skb->cloned ||
365 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
366 vfree(skb->head);
367
368 skb->head = NULL;
369 }
370 if (skb->sk != NULL)
371 sock_rfree(skb);
372}
373
374static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
375{
376 WARN_ON(skb->sk != NULL);
377 skb->sk = sk;
378 skb->destructor = netlink_skb_destructor;
379 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
380 sk_mem_charge(sk, skb->truesize);
381}
382
383static void netlink_sock_destruct(struct sock *sk)
384{
385 struct netlink_sock *nlk = nlk_sk(sk);
386
387 if (nlk->cb_running) {
388 if (nlk->cb.done)
389 nlk->cb.done(&nlk->cb);
390 module_put(nlk->cb.module);
391 kfree_skb(nlk->cb.skb);
392 }
393
394 skb_queue_purge(&sk->sk_receive_queue);
395
396 if (!sock_flag(sk, SOCK_DEAD)) {
397 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
398 return;
399 }
400
401 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
402 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
403 WARN_ON(nlk_sk(sk)->groups);
404}
405
406static void netlink_sock_destruct_work(struct work_struct *work)
407{
408 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
409 work);
410
411 sk_free(&nlk->sk);
412}
413
414/* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
415 * SMP. Look, when several writers sleep and reader wakes them up, all but one
416 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
417 * this, _but_ remember, it adds useless work on UP machines.
418 */
419
420void netlink_table_grab(void)
421 __acquires(nl_table_lock)
422{
423 might_sleep();
424
425 write_lock_irq(&nl_table_lock);
426
427 if (atomic_read(&nl_table_users)) {
428 DECLARE_WAITQUEUE(wait, current);
429
430 add_wait_queue_exclusive(&nl_table_wait, &wait);
431 for (;;) {
432 set_current_state(TASK_UNINTERRUPTIBLE);
433 if (atomic_read(&nl_table_users) == 0)
434 break;
435 write_unlock_irq(&nl_table_lock);
436 schedule();
437 write_lock_irq(&nl_table_lock);
438 }
439
440 __set_current_state(TASK_RUNNING);
441 remove_wait_queue(&nl_table_wait, &wait);
442 }
443}
444
445void netlink_table_ungrab(void)
446 __releases(nl_table_lock)
447{
448 write_unlock_irq(&nl_table_lock);
449 wake_up(&nl_table_wait);
450}
451
452static inline void
453netlink_lock_table(void)
454{
455 /* read_lock() synchronizes us to netlink_table_grab */
456
457 read_lock(&nl_table_lock);
458 atomic_inc(&nl_table_users);
459 read_unlock(&nl_table_lock);
460}
461
462static inline void
463netlink_unlock_table(void)
464{
465 if (atomic_dec_and_test(&nl_table_users))
466 wake_up(&nl_table_wait);
467}
468
469struct netlink_compare_arg
470{
471 possible_net_t pnet;
472 u32 portid;
473};
474
475/* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
476#define netlink_compare_arg_len \
477 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
478
479static inline int netlink_compare(struct rhashtable_compare_arg *arg,
480 const void *ptr)
481{
482 const struct netlink_compare_arg *x = arg->key;
483 const struct netlink_sock *nlk = ptr;
484
485 return nlk->portid != x->portid ||
486 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
487}
488
489static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
490 struct net *net, u32 portid)
491{
492 memset(arg, 0, sizeof(*arg));
493 write_pnet(&arg->pnet, net);
494 arg->portid = portid;
495}
496
497static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
498 struct net *net)
499{
500 struct netlink_compare_arg arg;
501
502 netlink_compare_arg_init(&arg, net, portid);
503 return rhashtable_lookup_fast(&table->hash, &arg,
504 netlink_rhashtable_params);
505}
506
507static int __netlink_insert(struct netlink_table *table, struct sock *sk)
508{
509 struct netlink_compare_arg arg;
510
511 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
512 return rhashtable_lookup_insert_key(&table->hash, &arg,
513 &nlk_sk(sk)->node,
514 netlink_rhashtable_params);
515}
516
517static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
518{
519 struct netlink_table *table = &nl_table[protocol];
520 struct sock *sk;
521
522 rcu_read_lock();
523 sk = __netlink_lookup(table, portid, net);
524 if (sk)
525 sock_hold(sk);
526 rcu_read_unlock();
527
528 return sk;
529}
530
531static const struct proto_ops netlink_ops;
532
533static void
534netlink_update_listeners(struct sock *sk)
535{
536 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
537 unsigned long mask;
538 unsigned int i;
539 struct listeners *listeners;
540
541 listeners = nl_deref_protected(tbl->listeners);
542 if (!listeners)
543 return;
544
545 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
546 mask = 0;
547 sk_for_each_bound(sk, &tbl->mc_list) {
548 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
549 mask |= nlk_sk(sk)->groups[i];
550 }
551 listeners->masks[i] = mask;
552 }
553 /* this function is only called with the netlink table "grabbed", which
554 * makes sure updates are visible before bind or setsockopt return. */
555}
556
557static int netlink_insert(struct sock *sk, u32 portid)
558{
559 struct netlink_table *table = &nl_table[sk->sk_protocol];
560 int err;
561
562 lock_sock(sk);
563
564 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
565 if (nlk_sk(sk)->bound)
566 goto err;
567
568 nlk_sk(sk)->portid = portid;
569 sock_hold(sk);
570
571 err = __netlink_insert(table, sk);
572 if (err) {
573 /* In case the hashtable backend returns with -EBUSY
574 * from here, it must not escape to the caller.
575 */
576 if (unlikely(err == -EBUSY))
577 err = -EOVERFLOW;
578 if (err == -EEXIST)
579 err = -EADDRINUSE;
580 sock_put(sk);
581 goto err;
582 }
583
584 /* We need to ensure that the socket is hashed and visible. */
585 smp_wmb();
586 nlk_sk(sk)->bound = portid;
587
588err:
589 release_sock(sk);
590 return err;
591}
592
593static void netlink_remove(struct sock *sk)
594{
595 struct netlink_table *table;
596
597 table = &nl_table[sk->sk_protocol];
598 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
599 netlink_rhashtable_params)) {
600 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
601 __sock_put(sk);
602 }
603
604 netlink_table_grab();
605 if (nlk_sk(sk)->subscriptions) {
606 __sk_del_bind_node(sk);
607 netlink_update_listeners(sk);
608 }
609 if (sk->sk_protocol == NETLINK_GENERIC)
610 atomic_inc(&genl_sk_destructing_cnt);
611 netlink_table_ungrab();
612}
613
614static struct proto netlink_proto = {
615 .name = "NETLINK",
616 .owner = THIS_MODULE,
617 .obj_size = sizeof(struct netlink_sock),
618};
619
620static int __netlink_create(struct net *net, struct socket *sock,
621 struct mutex *cb_mutex, int protocol,
622 int kern)
623{
624 struct sock *sk;
625 struct netlink_sock *nlk;
626
627 sock->ops = &netlink_ops;
628
629 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
630 if (!sk)
631 return -ENOMEM;
632
633 sock_init_data(sock, sk);
634
635 nlk = nlk_sk(sk);
636 if (cb_mutex) {
637 nlk->cb_mutex = cb_mutex;
638 } else {
639 nlk->cb_mutex = &nlk->cb_def_mutex;
640 mutex_init(nlk->cb_mutex);
641 lockdep_set_class_and_name(nlk->cb_mutex,
642 nlk_cb_mutex_keys + protocol,
643 nlk_cb_mutex_key_strings[protocol]);
644 }
645 init_waitqueue_head(&nlk->wait);
646
647 sk->sk_destruct = netlink_sock_destruct;
648 sk->sk_protocol = protocol;
649 return 0;
650}
651
652static int netlink_create(struct net *net, struct socket *sock, int protocol,
653 int kern)
654{
655 struct module *module = NULL;
656 struct mutex *cb_mutex;
657 struct netlink_sock *nlk;
658 int (*bind)(struct net *net, int group);
659 void (*unbind)(struct net *net, int group);
660 int err = 0;
661
662 sock->state = SS_UNCONNECTED;
663
664 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
665 return -ESOCKTNOSUPPORT;
666
667 if (protocol < 0 || protocol >= MAX_LINKS)
668 return -EPROTONOSUPPORT;
669 protocol = array_index_nospec(protocol, MAX_LINKS);
670
671 netlink_lock_table();
672#ifdef CONFIG_MODULES
673 if (!nl_table[protocol].registered) {
674 netlink_unlock_table();
675 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
676 netlink_lock_table();
677 }
678#endif
679 if (nl_table[protocol].registered &&
680 try_module_get(nl_table[protocol].module))
681 module = nl_table[protocol].module;
682 else
683 err = -EPROTONOSUPPORT;
684 cb_mutex = nl_table[protocol].cb_mutex;
685 bind = nl_table[protocol].bind;
686 unbind = nl_table[protocol].unbind;
687 netlink_unlock_table();
688
689 if (err < 0)
690 goto out;
691
692 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
693 if (err < 0)
694 goto out_module;
695
696 local_bh_disable();
697 sock_prot_inuse_add(net, &netlink_proto, 1);
698 local_bh_enable();
699
700 nlk = nlk_sk(sock->sk);
701 nlk->module = module;
702 nlk->netlink_bind = bind;
703 nlk->netlink_unbind = unbind;
704out:
705 return err;
706
707out_module:
708 module_put(module);
709 goto out;
710}
711
712static void deferred_put_nlk_sk(struct rcu_head *head)
713{
714 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
715 struct sock *sk = &nlk->sk;
716
717 kfree(nlk->groups);
718 nlk->groups = NULL;
719
720 if (!refcount_dec_and_test(&sk->sk_refcnt))
721 return;
722
723 if (nlk->cb_running && nlk->cb.done) {
724 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
725 schedule_work(&nlk->work);
726 return;
727 }
728
729 sk_free(sk);
730}
731
732static int netlink_release(struct socket *sock)
733{
734 struct sock *sk = sock->sk;
735 struct netlink_sock *nlk;
736
737 if (!sk)
738 return 0;
739
740 netlink_remove(sk);
741 sock_orphan(sk);
742 nlk = nlk_sk(sk);
743
744 /*
745 * OK. Socket is unlinked, any packets that arrive now
746 * will be purged.
747 */
748
749 /* must not acquire netlink_table_lock in any way again before unbind
750 * and notifying genetlink is done as otherwise it might deadlock
751 */
752 if (nlk->netlink_unbind) {
753 int i;
754
755 for (i = 0; i < nlk->ngroups; i++)
756 if (test_bit(i, nlk->groups))
757 nlk->netlink_unbind(sock_net(sk), i + 1);
758 }
759 if (sk->sk_protocol == NETLINK_GENERIC &&
760 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
761 wake_up(&genl_sk_destructing_waitq);
762
763 sock->sk = NULL;
764 wake_up_interruptible_all(&nlk->wait);
765
766 skb_queue_purge(&sk->sk_write_queue);
767
768 if (nlk->portid && nlk->bound) {
769 struct netlink_notify n = {
770 .net = sock_net(sk),
771 .protocol = sk->sk_protocol,
772 .portid = nlk->portid,
773 };
774 blocking_notifier_call_chain(&netlink_chain,
775 NETLINK_URELEASE, &n);
776 }
777
778 module_put(nlk->module);
779
780 if (netlink_is_kernel(sk)) {
781 netlink_table_grab();
782 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
783 if (--nl_table[sk->sk_protocol].registered == 0) {
784 struct listeners *old;
785
786 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
787 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
788 kfree_rcu(old, rcu);
789 nl_table[sk->sk_protocol].module = NULL;
790 nl_table[sk->sk_protocol].bind = NULL;
791 nl_table[sk->sk_protocol].unbind = NULL;
792 nl_table[sk->sk_protocol].flags = 0;
793 nl_table[sk->sk_protocol].registered = 0;
794 }
795 netlink_table_ungrab();
796 }
797
798 local_bh_disable();
799 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
800 local_bh_enable();
801 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
802 return 0;
803}
804
805static int netlink_autobind(struct socket *sock)
806{
807 struct sock *sk = sock->sk;
808 struct net *net = sock_net(sk);
809 struct netlink_table *table = &nl_table[sk->sk_protocol];
810 s32 portid = task_tgid_vnr(current);
811 int err;
812 s32 rover = -4096;
813 bool ok;
814
815retry:
816 cond_resched();
817 rcu_read_lock();
818 ok = !__netlink_lookup(table, portid, net);
819 rcu_read_unlock();
820 if (!ok) {
821 /* Bind collision, search negative portid values. */
822 if (rover == -4096)
823 /* rover will be in range [S32_MIN, -4097] */
824 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
825 else if (rover >= -4096)
826 rover = -4097;
827 portid = rover--;
828 goto retry;
829 }
830
831 err = netlink_insert(sk, portid);
832 if (err == -EADDRINUSE)
833 goto retry;
834
835 /* If 2 threads race to autobind, that is fine. */
836 if (err == -EBUSY)
837 err = 0;
838
839 return err;
840}
841
842/**
843 * __netlink_ns_capable - General netlink message capability test
844 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
845 * @user_ns: The user namespace of the capability to use
846 * @cap: The capability to use
847 *
848 * Test to see if the opener of the socket we received the message
849 * from had when the netlink socket was created and the sender of the
850 * message has has the capability @cap in the user namespace @user_ns.
851 */
852bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
853 struct user_namespace *user_ns, int cap)
854{
855 return ((nsp->flags & NETLINK_SKB_DST) ||
856 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
857 ns_capable(user_ns, cap);
858}
859EXPORT_SYMBOL(__netlink_ns_capable);
860
861/**
862 * netlink_ns_capable - General netlink message capability test
863 * @skb: socket buffer holding a netlink command from userspace
864 * @user_ns: The user namespace of the capability to use
865 * @cap: The capability to use
866 *
867 * Test to see if the opener of the socket we received the message
868 * from had when the netlink socket was created and the sender of the
869 * message has has the capability @cap in the user namespace @user_ns.
870 */
871bool netlink_ns_capable(const struct sk_buff *skb,
872 struct user_namespace *user_ns, int cap)
873{
874 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
875}
876EXPORT_SYMBOL(netlink_ns_capable);
877
878/**
879 * netlink_capable - Netlink global message capability test
880 * @skb: socket buffer holding a netlink command from userspace
881 * @cap: The capability to use
882 *
883 * Test to see if the opener of the socket we received the message
884 * from had when the netlink socket was created and the sender of the
885 * message has has the capability @cap in all user namespaces.
886 */
887bool netlink_capable(const struct sk_buff *skb, int cap)
888{
889 return netlink_ns_capable(skb, &init_user_ns, cap);
890}
891EXPORT_SYMBOL(netlink_capable);
892
893/**
894 * netlink_net_capable - Netlink network namespace message capability test
895 * @skb: socket buffer holding a netlink command from userspace
896 * @cap: The capability to use
897 *
898 * Test to see if the opener of the socket we received the message
899 * from had when the netlink socket was created and the sender of the
900 * message has has the capability @cap over the network namespace of
901 * the socket we received the message from.
902 */
903bool netlink_net_capable(const struct sk_buff *skb, int cap)
904{
905 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
906}
907EXPORT_SYMBOL(netlink_net_capable);
908
909static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
910{
911 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
912 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
913}
914
915static void
916netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
917{
918 struct netlink_sock *nlk = nlk_sk(sk);
919
920 if (nlk->subscriptions && !subscriptions)
921 __sk_del_bind_node(sk);
922 else if (!nlk->subscriptions && subscriptions)
923 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
924 nlk->subscriptions = subscriptions;
925}
926
927static int netlink_realloc_groups(struct sock *sk)
928{
929 struct netlink_sock *nlk = nlk_sk(sk);
930 unsigned int groups;
931 unsigned long *new_groups;
932 int err = 0;
933
934 netlink_table_grab();
935
936 groups = nl_table[sk->sk_protocol].groups;
937 if (!nl_table[sk->sk_protocol].registered) {
938 err = -ENOENT;
939 goto out_unlock;
940 }
941
942 if (nlk->ngroups >= groups)
943 goto out_unlock;
944
945 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
946 if (new_groups == NULL) {
947 err = -ENOMEM;
948 goto out_unlock;
949 }
950 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
951 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
952
953 nlk->groups = new_groups;
954 nlk->ngroups = groups;
955 out_unlock:
956 netlink_table_ungrab();
957 return err;
958}
959
960static void netlink_undo_bind(int group, long unsigned int groups,
961 struct sock *sk)
962{
963 struct netlink_sock *nlk = nlk_sk(sk);
964 int undo;
965
966 if (!nlk->netlink_unbind)
967 return;
968
969 for (undo = 0; undo < group; undo++)
970 if (test_bit(undo, &groups))
971 nlk->netlink_unbind(sock_net(sk), undo + 1);
972}
973
974static int netlink_bind(struct socket *sock, struct sockaddr *addr,
975 int addr_len)
976{
977 struct sock *sk = sock->sk;
978 struct net *net = sock_net(sk);
979 struct netlink_sock *nlk = nlk_sk(sk);
980 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
981 int err = 0;
982 unsigned long groups;
983 bool bound;
984
985 if (addr_len < sizeof(struct sockaddr_nl))
986 return -EINVAL;
987
988 if (nladdr->nl_family != AF_NETLINK)
989 return -EINVAL;
990 groups = nladdr->nl_groups;
991
992 /* Only superuser is allowed to listen multicasts */
993 if (groups) {
994 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
995 return -EPERM;
996 err = netlink_realloc_groups(sk);
997 if (err)
998 return err;
999 }
1000
1001 if (nlk->ngroups < BITS_PER_LONG)
1002 groups &= (1UL << nlk->ngroups) - 1;
1003
1004 bound = nlk->bound;
1005 if (bound) {
1006 /* Ensure nlk->portid is up-to-date. */
1007 smp_rmb();
1008
1009 if (nladdr->nl_pid != nlk->portid)
1010 return -EINVAL;
1011 }
1012
1013 netlink_lock_table();
1014 if (nlk->netlink_bind && groups) {
1015 int group;
1016
1017 for (group = 0; group < nlk->ngroups; group++) {
1018 if (!test_bit(group, &groups))
1019 continue;
1020 err = nlk->netlink_bind(net, group + 1);
1021 if (!err)
1022 continue;
1023 netlink_undo_bind(group, groups, sk);
1024 goto unlock;
1025 }
1026 }
1027
1028 /* No need for barriers here as we return to user-space without
1029 * using any of the bound attributes.
1030 */
1031 if (!bound) {
1032 err = nladdr->nl_pid ?
1033 netlink_insert(sk, nladdr->nl_pid) :
1034 netlink_autobind(sock);
1035 if (err) {
1036 netlink_undo_bind(nlk->ngroups, groups, sk);
1037 goto unlock;
1038 }
1039 }
1040
1041 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1042 goto unlock;
1043 netlink_unlock_table();
1044
1045 netlink_table_grab();
1046 netlink_update_subscriptions(sk, nlk->subscriptions +
1047 hweight32(groups) -
1048 hweight32(nlk->groups[0]));
1049 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1050 netlink_update_listeners(sk);
1051 netlink_table_ungrab();
1052
1053 return 0;
1054
1055unlock:
1056 netlink_unlock_table();
1057 return err;
1058}
1059
1060static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1061 int alen, int flags)
1062{
1063 int err = 0;
1064 struct sock *sk = sock->sk;
1065 struct netlink_sock *nlk = nlk_sk(sk);
1066 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1067
1068 if (alen < sizeof(addr->sa_family))
1069 return -EINVAL;
1070
1071 if (addr->sa_family == AF_UNSPEC) {
1072 sk->sk_state = NETLINK_UNCONNECTED;
1073 nlk->dst_portid = 0;
1074 nlk->dst_group = 0;
1075 return 0;
1076 }
1077 if (addr->sa_family != AF_NETLINK)
1078 return -EINVAL;
1079
1080 if (alen < sizeof(struct sockaddr_nl))
1081 return -EINVAL;
1082
1083 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1084 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1085 return -EPERM;
1086
1087 /* No need for barriers here as we return to user-space without
1088 * using any of the bound attributes.
1089 */
1090 if (!nlk->bound)
1091 err = netlink_autobind(sock);
1092
1093 if (err == 0) {
1094 sk->sk_state = NETLINK_CONNECTED;
1095 nlk->dst_portid = nladdr->nl_pid;
1096 nlk->dst_group = ffs(nladdr->nl_groups);
1097 }
1098
1099 return err;
1100}
1101
1102static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1103 int peer)
1104{
1105 struct sock *sk = sock->sk;
1106 struct netlink_sock *nlk = nlk_sk(sk);
1107 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1108
1109 nladdr->nl_family = AF_NETLINK;
1110 nladdr->nl_pad = 0;
1111
1112 if (peer) {
1113 nladdr->nl_pid = nlk->dst_portid;
1114 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1115 } else {
1116 nladdr->nl_pid = nlk->portid;
1117 netlink_lock_table();
1118 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1119 netlink_unlock_table();
1120 }
1121 return sizeof(*nladdr);
1122}
1123
1124static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1125 unsigned long arg)
1126{
1127 /* try to hand this ioctl down to the NIC drivers.
1128 */
1129 return -ENOIOCTLCMD;
1130}
1131
1132static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1133{
1134 struct sock *sock;
1135 struct netlink_sock *nlk;
1136
1137 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1138 if (!sock)
1139 return ERR_PTR(-ECONNREFUSED);
1140
1141 /* Don't bother queuing skb if kernel socket has no input function */
1142 nlk = nlk_sk(sock);
1143 if (sock->sk_state == NETLINK_CONNECTED &&
1144 nlk->dst_portid != nlk_sk(ssk)->portid) {
1145 sock_put(sock);
1146 return ERR_PTR(-ECONNREFUSED);
1147 }
1148 return sock;
1149}
1150
1151struct sock *netlink_getsockbyfilp(struct file *filp)
1152{
1153 struct inode *inode = file_inode(filp);
1154 struct sock *sock;
1155
1156 if (!S_ISSOCK(inode->i_mode))
1157 return ERR_PTR(-ENOTSOCK);
1158
1159 sock = SOCKET_I(inode)->sk;
1160 if (sock->sk_family != AF_NETLINK)
1161 return ERR_PTR(-EINVAL);
1162
1163 sock_hold(sock);
1164 return sock;
1165}
1166
1167static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1168 int broadcast)
1169{
1170 struct sk_buff *skb;
1171 void *data;
1172
1173 if (size <= NLMSG_GOODSIZE || broadcast)
1174 return alloc_skb(size, GFP_KERNEL);
1175
1176 size = SKB_DATA_ALIGN(size) +
1177 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1178
1179 data = vmalloc(size);
1180 if (data == NULL)
1181 return NULL;
1182
1183 skb = __build_skb(data, size);
1184 if (skb == NULL)
1185 vfree(data);
1186 else
1187 skb->destructor = netlink_skb_destructor;
1188
1189 return skb;
1190}
1191
1192/*
1193 * Attach a skb to a netlink socket.
1194 * The caller must hold a reference to the destination socket. On error, the
1195 * reference is dropped. The skb is not send to the destination, just all
1196 * all error checks are performed and memory in the queue is reserved.
1197 * Return values:
1198 * < 0: error. skb freed, reference to sock dropped.
1199 * 0: continue
1200 * 1: repeat lookup - reference dropped while waiting for socket memory.
1201 */
1202int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1203 long *timeo, struct sock *ssk)
1204{
1205 struct netlink_sock *nlk;
1206
1207 nlk = nlk_sk(sk);
1208
1209 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1210 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1211 DECLARE_WAITQUEUE(wait, current);
1212 if (!*timeo) {
1213 if (!ssk || netlink_is_kernel(ssk))
1214 netlink_overrun(sk);
1215 sock_put(sk);
1216 kfree_skb(skb);
1217 return -EAGAIN;
1218 }
1219
1220 __set_current_state(TASK_INTERRUPTIBLE);
1221 add_wait_queue(&nlk->wait, &wait);
1222
1223 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1224 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1225 !sock_flag(sk, SOCK_DEAD))
1226 *timeo = schedule_timeout(*timeo);
1227
1228 __set_current_state(TASK_RUNNING);
1229 remove_wait_queue(&nlk->wait, &wait);
1230 sock_put(sk);
1231
1232 if (signal_pending(current)) {
1233 kfree_skb(skb);
1234 return sock_intr_errno(*timeo);
1235 }
1236 return 1;
1237 }
1238 netlink_skb_set_owner_r(skb, sk);
1239 return 0;
1240}
1241
1242static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1243{
1244 int len = skb->len;
1245
1246 netlink_deliver_tap(sock_net(sk), skb);
1247
1248 skb_queue_tail(&sk->sk_receive_queue, skb);
1249 sk->sk_data_ready(sk);
1250 return len;
1251}
1252
1253int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1254{
1255 int len = __netlink_sendskb(sk, skb);
1256
1257 sock_put(sk);
1258 return len;
1259}
1260
1261void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1262{
1263 kfree_skb(skb);
1264 sock_put(sk);
1265}
1266
1267static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1268{
1269 int delta;
1270
1271 WARN_ON(skb->sk != NULL);
1272 delta = skb->end - skb->tail;
1273 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1274 return skb;
1275
1276 if (skb_shared(skb)) {
1277 struct sk_buff *nskb = skb_clone(skb, allocation);
1278 if (!nskb)
1279 return skb;
1280 consume_skb(skb);
1281 skb = nskb;
1282 }
1283
1284 pskb_expand_head(skb, 0, -delta,
1285 (allocation & ~__GFP_DIRECT_RECLAIM) |
1286 __GFP_NOWARN | __GFP_NORETRY);
1287 return skb;
1288}
1289
1290static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1291 struct sock *ssk)
1292{
1293 int ret;
1294 struct netlink_sock *nlk = nlk_sk(sk);
1295
1296 ret = -ECONNREFUSED;
1297 if (nlk->netlink_rcv != NULL) {
1298 ret = skb->len;
1299 netlink_skb_set_owner_r(skb, sk);
1300 NETLINK_CB(skb).sk = ssk;
1301 netlink_deliver_tap_kernel(sk, ssk, skb);
1302 nlk->netlink_rcv(skb);
1303 consume_skb(skb);
1304 } else {
1305 kfree_skb(skb);
1306 }
1307 sock_put(sk);
1308 return ret;
1309}
1310
1311int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1312 u32 portid, int nonblock)
1313{
1314 struct sock *sk;
1315 int err;
1316 long timeo;
1317
1318 skb = netlink_trim(skb, gfp_any());
1319
1320 timeo = sock_sndtimeo(ssk, nonblock);
1321retry:
1322 sk = netlink_getsockbyportid(ssk, portid);
1323 if (IS_ERR(sk)) {
1324 kfree_skb(skb);
1325 return PTR_ERR(sk);
1326 }
1327 if (netlink_is_kernel(sk))
1328 return netlink_unicast_kernel(sk, skb, ssk);
1329
1330 if (sk_filter(sk, skb)) {
1331 err = skb->len;
1332 kfree_skb(skb);
1333 sock_put(sk);
1334 return err;
1335 }
1336
1337 err = netlink_attachskb(sk, skb, &timeo, ssk);
1338 if (err == 1)
1339 goto retry;
1340 if (err)
1341 return err;
1342
1343 return netlink_sendskb(sk, skb);
1344}
1345EXPORT_SYMBOL(netlink_unicast);
1346
1347int netlink_has_listeners(struct sock *sk, unsigned int group)
1348{
1349 int res = 0;
1350 struct listeners *listeners;
1351
1352 BUG_ON(!netlink_is_kernel(sk));
1353
1354 rcu_read_lock();
1355 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1356
1357 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1358 res = test_bit(group - 1, listeners->masks);
1359
1360 rcu_read_unlock();
1361
1362 return res;
1363}
1364EXPORT_SYMBOL_GPL(netlink_has_listeners);
1365
1366bool netlink_strict_get_check(struct sk_buff *skb)
1367{
1368 const struct netlink_sock *nlk = nlk_sk(NETLINK_CB(skb).sk);
1369
1370 return nlk->flags & NETLINK_F_STRICT_CHK;
1371}
1372EXPORT_SYMBOL_GPL(netlink_strict_get_check);
1373
1374static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1375{
1376 struct netlink_sock *nlk = nlk_sk(sk);
1377
1378 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1379 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1380 netlink_skb_set_owner_r(skb, sk);
1381 __netlink_sendskb(sk, skb);
1382 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1383 }
1384 return -1;
1385}
1386
1387struct netlink_broadcast_data {
1388 struct sock *exclude_sk;
1389 struct net *net;
1390 u32 portid;
1391 u32 group;
1392 int failure;
1393 int delivery_failure;
1394 int congested;
1395 int delivered;
1396 gfp_t allocation;
1397 struct sk_buff *skb, *skb2;
1398 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1399 void *tx_data;
1400};
1401
1402static void do_one_broadcast(struct sock *sk,
1403 struct netlink_broadcast_data *p)
1404{
1405 struct netlink_sock *nlk = nlk_sk(sk);
1406 int val;
1407
1408 if (p->exclude_sk == sk)
1409 return;
1410
1411 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1412 !test_bit(p->group - 1, nlk->groups))
1413 return;
1414
1415 if (!net_eq(sock_net(sk), p->net)) {
1416 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1417 return;
1418
1419 if (!peernet_has_id(sock_net(sk), p->net))
1420 return;
1421
1422 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1423 CAP_NET_BROADCAST))
1424 return;
1425 }
1426
1427 if (p->failure) {
1428 netlink_overrun(sk);
1429 return;
1430 }
1431
1432 sock_hold(sk);
1433 if (p->skb2 == NULL) {
1434 if (skb_shared(p->skb)) {
1435 p->skb2 = skb_clone(p->skb, p->allocation);
1436 } else {
1437 p->skb2 = skb_get(p->skb);
1438 /*
1439 * skb ownership may have been set when
1440 * delivered to a previous socket.
1441 */
1442 skb_orphan(p->skb2);
1443 }
1444 }
1445 if (p->skb2 == NULL) {
1446 netlink_overrun(sk);
1447 /* Clone failed. Notify ALL listeners. */
1448 p->failure = 1;
1449 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1450 p->delivery_failure = 1;
1451 goto out;
1452 }
1453 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1454 kfree_skb(p->skb2);
1455 p->skb2 = NULL;
1456 goto out;
1457 }
1458 if (sk_filter(sk, p->skb2)) {
1459 kfree_skb(p->skb2);
1460 p->skb2 = NULL;
1461 goto out;
1462 }
1463 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1464 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1465 NETLINK_CB(p->skb2).nsid_is_set = true;
1466 val = netlink_broadcast_deliver(sk, p->skb2);
1467 if (val < 0) {
1468 netlink_overrun(sk);
1469 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1470 p->delivery_failure = 1;
1471 } else {
1472 p->congested |= val;
1473 p->delivered = 1;
1474 p->skb2 = NULL;
1475 }
1476out:
1477 sock_put(sk);
1478}
1479
1480int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1481 u32 group, gfp_t allocation,
1482 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1483 void *filter_data)
1484{
1485 struct net *net = sock_net(ssk);
1486 struct netlink_broadcast_data info;
1487 struct sock *sk;
1488
1489 skb = netlink_trim(skb, allocation);
1490
1491 info.exclude_sk = ssk;
1492 info.net = net;
1493 info.portid = portid;
1494 info.group = group;
1495 info.failure = 0;
1496 info.delivery_failure = 0;
1497 info.congested = 0;
1498 info.delivered = 0;
1499 info.allocation = allocation;
1500 info.skb = skb;
1501 info.skb2 = NULL;
1502 info.tx_filter = filter;
1503 info.tx_data = filter_data;
1504
1505 /* While we sleep in clone, do not allow to change socket list */
1506
1507 netlink_lock_table();
1508
1509 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1510 do_one_broadcast(sk, &info);
1511
1512 consume_skb(skb);
1513
1514 netlink_unlock_table();
1515
1516 if (info.delivery_failure) {
1517 kfree_skb(info.skb2);
1518 return -ENOBUFS;
1519 }
1520 consume_skb(info.skb2);
1521
1522 if (info.delivered) {
1523 if (info.congested && gfpflags_allow_blocking(allocation))
1524 yield();
1525 return 0;
1526 }
1527 return -ESRCH;
1528}
1529EXPORT_SYMBOL(netlink_broadcast_filtered);
1530
1531int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1532 u32 group, gfp_t allocation)
1533{
1534 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1535 NULL, NULL);
1536}
1537EXPORT_SYMBOL(netlink_broadcast);
1538
1539struct netlink_set_err_data {
1540 struct sock *exclude_sk;
1541 u32 portid;
1542 u32 group;
1543 int code;
1544};
1545
1546static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1547{
1548 struct netlink_sock *nlk = nlk_sk(sk);
1549 int ret = 0;
1550
1551 if (sk == p->exclude_sk)
1552 goto out;
1553
1554 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1555 goto out;
1556
1557 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1558 !test_bit(p->group - 1, nlk->groups))
1559 goto out;
1560
1561 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1562 ret = 1;
1563 goto out;
1564 }
1565
1566 sk->sk_err = p->code;
1567 sk->sk_error_report(sk);
1568out:
1569 return ret;
1570}
1571
1572/**
1573 * netlink_set_err - report error to broadcast listeners
1574 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1575 * @portid: the PORTID of a process that we want to skip (if any)
1576 * @group: the broadcast group that will notice the error
1577 * @code: error code, must be negative (as usual in kernelspace)
1578 *
1579 * This function returns the number of broadcast listeners that have set the
1580 * NETLINK_NO_ENOBUFS socket option.
1581 */
1582int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1583{
1584 struct netlink_set_err_data info;
1585 struct sock *sk;
1586 int ret = 0;
1587
1588 info.exclude_sk = ssk;
1589 info.portid = portid;
1590 info.group = group;
1591 /* sk->sk_err wants a positive error value */
1592 info.code = -code;
1593
1594 read_lock(&nl_table_lock);
1595
1596 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1597 ret += do_one_set_err(sk, &info);
1598
1599 read_unlock(&nl_table_lock);
1600 return ret;
1601}
1602EXPORT_SYMBOL(netlink_set_err);
1603
1604/* must be called with netlink table grabbed */
1605static void netlink_update_socket_mc(struct netlink_sock *nlk,
1606 unsigned int group,
1607 int is_new)
1608{
1609 int old, new = !!is_new, subscriptions;
1610
1611 old = test_bit(group - 1, nlk->groups);
1612 subscriptions = nlk->subscriptions - old + new;
1613 if (new)
1614 __set_bit(group - 1, nlk->groups);
1615 else
1616 __clear_bit(group - 1, nlk->groups);
1617 netlink_update_subscriptions(&nlk->sk, subscriptions);
1618 netlink_update_listeners(&nlk->sk);
1619}
1620
1621static int netlink_setsockopt(struct socket *sock, int level, int optname,
1622 char __user *optval, unsigned int optlen)
1623{
1624 struct sock *sk = sock->sk;
1625 struct netlink_sock *nlk = nlk_sk(sk);
1626 unsigned int val = 0;
1627 int err;
1628
1629 if (level != SOL_NETLINK)
1630 return -ENOPROTOOPT;
1631
1632 if (optlen >= sizeof(int) &&
1633 get_user(val, (unsigned int __user *)optval))
1634 return -EFAULT;
1635
1636 switch (optname) {
1637 case NETLINK_PKTINFO:
1638 if (val)
1639 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1640 else
1641 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1642 err = 0;
1643 break;
1644 case NETLINK_ADD_MEMBERSHIP:
1645 case NETLINK_DROP_MEMBERSHIP: {
1646 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1647 return -EPERM;
1648 err = netlink_realloc_groups(sk);
1649 if (err)
1650 return err;
1651 if (!val || val - 1 >= nlk->ngroups)
1652 return -EINVAL;
1653 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1654 err = nlk->netlink_bind(sock_net(sk), val);
1655 if (err)
1656 return err;
1657 }
1658 netlink_table_grab();
1659 netlink_update_socket_mc(nlk, val,
1660 optname == NETLINK_ADD_MEMBERSHIP);
1661 netlink_table_ungrab();
1662 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1663 nlk->netlink_unbind(sock_net(sk), val);
1664
1665 err = 0;
1666 break;
1667 }
1668 case NETLINK_BROADCAST_ERROR:
1669 if (val)
1670 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1671 else
1672 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1673 err = 0;
1674 break;
1675 case NETLINK_NO_ENOBUFS:
1676 if (val) {
1677 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1678 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1679 wake_up_interruptible(&nlk->wait);
1680 } else {
1681 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1682 }
1683 err = 0;
1684 break;
1685 case NETLINK_LISTEN_ALL_NSID:
1686 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1687 return -EPERM;
1688
1689 if (val)
1690 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1691 else
1692 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1693 err = 0;
1694 break;
1695 case NETLINK_CAP_ACK:
1696 if (val)
1697 nlk->flags |= NETLINK_F_CAP_ACK;
1698 else
1699 nlk->flags &= ~NETLINK_F_CAP_ACK;
1700 err = 0;
1701 break;
1702 case NETLINK_EXT_ACK:
1703 if (val)
1704 nlk->flags |= NETLINK_F_EXT_ACK;
1705 else
1706 nlk->flags &= ~NETLINK_F_EXT_ACK;
1707 err = 0;
1708 break;
1709 case NETLINK_GET_STRICT_CHK:
1710 if (val)
1711 nlk->flags |= NETLINK_F_STRICT_CHK;
1712 else
1713 nlk->flags &= ~NETLINK_F_STRICT_CHK;
1714 err = 0;
1715 break;
1716 default:
1717 err = -ENOPROTOOPT;
1718 }
1719 return err;
1720}
1721
1722static int netlink_getsockopt(struct socket *sock, int level, int optname,
1723 char __user *optval, int __user *optlen)
1724{
1725 struct sock *sk = sock->sk;
1726 struct netlink_sock *nlk = nlk_sk(sk);
1727 int len, val, err;
1728
1729 if (level != SOL_NETLINK)
1730 return -ENOPROTOOPT;
1731
1732 if (get_user(len, optlen))
1733 return -EFAULT;
1734 if (len < 0)
1735 return -EINVAL;
1736
1737 switch (optname) {
1738 case NETLINK_PKTINFO:
1739 if (len < sizeof(int))
1740 return -EINVAL;
1741 len = sizeof(int);
1742 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1743 if (put_user(len, optlen) ||
1744 put_user(val, optval))
1745 return -EFAULT;
1746 err = 0;
1747 break;
1748 case NETLINK_BROADCAST_ERROR:
1749 if (len < sizeof(int))
1750 return -EINVAL;
1751 len = sizeof(int);
1752 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1753 if (put_user(len, optlen) ||
1754 put_user(val, optval))
1755 return -EFAULT;
1756 err = 0;
1757 break;
1758 case NETLINK_NO_ENOBUFS:
1759 if (len < sizeof(int))
1760 return -EINVAL;
1761 len = sizeof(int);
1762 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1763 if (put_user(len, optlen) ||
1764 put_user(val, optval))
1765 return -EFAULT;
1766 err = 0;
1767 break;
1768 case NETLINK_LIST_MEMBERSHIPS: {
1769 int pos, idx, shift;
1770
1771 err = 0;
1772 netlink_lock_table();
1773 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1774 if (len - pos < sizeof(u32))
1775 break;
1776
1777 idx = pos / sizeof(unsigned long);
1778 shift = (pos % sizeof(unsigned long)) * 8;
1779 if (put_user((u32)(nlk->groups[idx] >> shift),
1780 (u32 __user *)(optval + pos))) {
1781 err = -EFAULT;
1782 break;
1783 }
1784 }
1785 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1786 err = -EFAULT;
1787 netlink_unlock_table();
1788 break;
1789 }
1790 case NETLINK_CAP_ACK:
1791 if (len < sizeof(int))
1792 return -EINVAL;
1793 len = sizeof(int);
1794 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1795 if (put_user(len, optlen) ||
1796 put_user(val, optval))
1797 return -EFAULT;
1798 err = 0;
1799 break;
1800 case NETLINK_EXT_ACK:
1801 if (len < sizeof(int))
1802 return -EINVAL;
1803 len = sizeof(int);
1804 val = nlk->flags & NETLINK_F_EXT_ACK ? 1 : 0;
1805 if (put_user(len, optlen) || put_user(val, optval))
1806 return -EFAULT;
1807 err = 0;
1808 break;
1809 case NETLINK_GET_STRICT_CHK:
1810 if (len < sizeof(int))
1811 return -EINVAL;
1812 len = sizeof(int);
1813 val = nlk->flags & NETLINK_F_STRICT_CHK ? 1 : 0;
1814 if (put_user(len, optlen) || put_user(val, optval))
1815 return -EFAULT;
1816 err = 0;
1817 break;
1818 default:
1819 err = -ENOPROTOOPT;
1820 }
1821 return err;
1822}
1823
1824static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1825{
1826 struct nl_pktinfo info;
1827
1828 info.group = NETLINK_CB(skb).dst_group;
1829 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1830}
1831
1832static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1833 struct sk_buff *skb)
1834{
1835 if (!NETLINK_CB(skb).nsid_is_set)
1836 return;
1837
1838 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1839 &NETLINK_CB(skb).nsid);
1840}
1841
1842static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1843{
1844 struct sock *sk = sock->sk;
1845 struct netlink_sock *nlk = nlk_sk(sk);
1846 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1847 u32 dst_portid;
1848 u32 dst_group;
1849 struct sk_buff *skb;
1850 int err;
1851 struct scm_cookie scm;
1852 u32 netlink_skb_flags = 0;
1853
1854 if (msg->msg_flags&MSG_OOB)
1855 return -EOPNOTSUPP;
1856
1857 err = scm_send(sock, msg, &scm, true);
1858 if (err < 0)
1859 return err;
1860
1861 if (msg->msg_namelen) {
1862 err = -EINVAL;
1863 if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1864 goto out;
1865 if (addr->nl_family != AF_NETLINK)
1866 goto out;
1867 dst_portid = addr->nl_pid;
1868 dst_group = ffs(addr->nl_groups);
1869 err = -EPERM;
1870 if ((dst_group || dst_portid) &&
1871 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1872 goto out;
1873 netlink_skb_flags |= NETLINK_SKB_DST;
1874 } else {
1875 dst_portid = nlk->dst_portid;
1876 dst_group = nlk->dst_group;
1877 }
1878
1879 if (!nlk->bound) {
1880 err = netlink_autobind(sock);
1881 if (err)
1882 goto out;
1883 } else {
1884 /* Ensure nlk is hashed and visible. */
1885 smp_rmb();
1886 }
1887
1888 err = -EMSGSIZE;
1889 if (len > sk->sk_sndbuf - 32)
1890 goto out;
1891 err = -ENOBUFS;
1892 skb = netlink_alloc_large_skb(len, dst_group);
1893 if (skb == NULL)
1894 goto out;
1895
1896 NETLINK_CB(skb).portid = nlk->portid;
1897 NETLINK_CB(skb).dst_group = dst_group;
1898 NETLINK_CB(skb).creds = scm.creds;
1899 NETLINK_CB(skb).flags = netlink_skb_flags;
1900
1901 err = -EFAULT;
1902 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1903 kfree_skb(skb);
1904 goto out;
1905 }
1906
1907 err = security_netlink_send(sk, skb);
1908 if (err) {
1909 kfree_skb(skb);
1910 goto out;
1911 }
1912
1913 if (dst_group) {
1914 refcount_inc(&skb->users);
1915 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1916 }
1917 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
1918
1919out:
1920 scm_destroy(&scm);
1921 return err;
1922}
1923
1924static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1925 int flags)
1926{
1927 struct scm_cookie scm;
1928 struct sock *sk = sock->sk;
1929 struct netlink_sock *nlk = nlk_sk(sk);
1930 int noblock = flags&MSG_DONTWAIT;
1931 size_t copied;
1932 struct sk_buff *skb, *data_skb;
1933 int err, ret;
1934
1935 if (flags&MSG_OOB)
1936 return -EOPNOTSUPP;
1937
1938 copied = 0;
1939
1940 skb = skb_recv_datagram(sk, flags, noblock, &err);
1941 if (skb == NULL)
1942 goto out;
1943
1944 data_skb = skb;
1945
1946#ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1947 if (unlikely(skb_shinfo(skb)->frag_list)) {
1948 /*
1949 * If this skb has a frag_list, then here that means that we
1950 * will have to use the frag_list skb's data for compat tasks
1951 * and the regular skb's data for normal (non-compat) tasks.
1952 *
1953 * If we need to send the compat skb, assign it to the
1954 * 'data_skb' variable so that it will be used below for data
1955 * copying. We keep 'skb' for everything else, including
1956 * freeing both later.
1957 */
1958 if (flags & MSG_CMSG_COMPAT)
1959 data_skb = skb_shinfo(skb)->frag_list;
1960 }
1961#endif
1962
1963 /* Record the max length of recvmsg() calls for future allocations */
1964 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1965 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1966 SKB_WITH_OVERHEAD(32768));
1967
1968 copied = data_skb->len;
1969 if (len < copied) {
1970 msg->msg_flags |= MSG_TRUNC;
1971 copied = len;
1972 }
1973
1974 skb_reset_transport_header(data_skb);
1975 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1976
1977 if (msg->msg_name) {
1978 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1979 addr->nl_family = AF_NETLINK;
1980 addr->nl_pad = 0;
1981 addr->nl_pid = NETLINK_CB(skb).portid;
1982 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1983 msg->msg_namelen = sizeof(*addr);
1984 }
1985
1986 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1987 netlink_cmsg_recv_pktinfo(msg, skb);
1988 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1989 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1990
1991 memset(&scm, 0, sizeof(scm));
1992 scm.creds = *NETLINK_CREDS(skb);
1993 if (flags & MSG_TRUNC)
1994 copied = data_skb->len;
1995
1996 skb_free_datagram(sk, skb);
1997
1998 if (nlk->cb_running &&
1999 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2000 ret = netlink_dump(sk);
2001 if (ret) {
2002 sk->sk_err = -ret;
2003 sk->sk_error_report(sk);
2004 }
2005 }
2006
2007 scm_recv(sock, msg, &scm, flags);
2008out:
2009 netlink_rcv_wake(sk);
2010 return err ? : copied;
2011}
2012
2013static void netlink_data_ready(struct sock *sk)
2014{
2015 BUG();
2016}
2017
2018/*
2019 * We export these functions to other modules. They provide a
2020 * complete set of kernel non-blocking support for message
2021 * queueing.
2022 */
2023
2024struct sock *
2025__netlink_kernel_create(struct net *net, int unit, struct module *module,
2026 struct netlink_kernel_cfg *cfg)
2027{
2028 struct socket *sock;
2029 struct sock *sk;
2030 struct netlink_sock *nlk;
2031 struct listeners *listeners = NULL;
2032 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2033 unsigned int groups;
2034
2035 BUG_ON(!nl_table);
2036
2037 if (unit < 0 || unit >= MAX_LINKS)
2038 return NULL;
2039
2040 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2041 return NULL;
2042
2043 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2044 goto out_sock_release_nosk;
2045
2046 sk = sock->sk;
2047
2048 if (!cfg || cfg->groups < 32)
2049 groups = 32;
2050 else
2051 groups = cfg->groups;
2052
2053 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2054 if (!listeners)
2055 goto out_sock_release;
2056
2057 sk->sk_data_ready = netlink_data_ready;
2058 if (cfg && cfg->input)
2059 nlk_sk(sk)->netlink_rcv = cfg->input;
2060
2061 if (netlink_insert(sk, 0))
2062 goto out_sock_release;
2063
2064 nlk = nlk_sk(sk);
2065 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2066
2067 netlink_table_grab();
2068 if (!nl_table[unit].registered) {
2069 nl_table[unit].groups = groups;
2070 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2071 nl_table[unit].cb_mutex = cb_mutex;
2072 nl_table[unit].module = module;
2073 if (cfg) {
2074 nl_table[unit].bind = cfg->bind;
2075 nl_table[unit].unbind = cfg->unbind;
2076 nl_table[unit].flags = cfg->flags;
2077 if (cfg->compare)
2078 nl_table[unit].compare = cfg->compare;
2079 }
2080 nl_table[unit].registered = 1;
2081 } else {
2082 kfree(listeners);
2083 nl_table[unit].registered++;
2084 }
2085 netlink_table_ungrab();
2086 return sk;
2087
2088out_sock_release:
2089 kfree(listeners);
2090 netlink_kernel_release(sk);
2091 return NULL;
2092
2093out_sock_release_nosk:
2094 sock_release(sock);
2095 return NULL;
2096}
2097EXPORT_SYMBOL(__netlink_kernel_create);
2098
2099void
2100netlink_kernel_release(struct sock *sk)
2101{
2102 if (sk == NULL || sk->sk_socket == NULL)
2103 return;
2104
2105 sock_release(sk->sk_socket);
2106}
2107EXPORT_SYMBOL(netlink_kernel_release);
2108
2109int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2110{
2111 struct listeners *new, *old;
2112 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2113
2114 if (groups < 32)
2115 groups = 32;
2116
2117 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2118 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2119 if (!new)
2120 return -ENOMEM;
2121 old = nl_deref_protected(tbl->listeners);
2122 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2123 rcu_assign_pointer(tbl->listeners, new);
2124
2125 kfree_rcu(old, rcu);
2126 }
2127 tbl->groups = groups;
2128
2129 return 0;
2130}
2131
2132/**
2133 * netlink_change_ngroups - change number of multicast groups
2134 *
2135 * This changes the number of multicast groups that are available
2136 * on a certain netlink family. Note that it is not possible to
2137 * change the number of groups to below 32. Also note that it does
2138 * not implicitly call netlink_clear_multicast_users() when the
2139 * number of groups is reduced.
2140 *
2141 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2142 * @groups: The new number of groups.
2143 */
2144int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2145{
2146 int err;
2147
2148 netlink_table_grab();
2149 err = __netlink_change_ngroups(sk, groups);
2150 netlink_table_ungrab();
2151
2152 return err;
2153}
2154
2155void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2156{
2157 struct sock *sk;
2158 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2159
2160 sk_for_each_bound(sk, &tbl->mc_list)
2161 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2162}
2163
2164struct nlmsghdr *
2165__nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2166{
2167 struct nlmsghdr *nlh;
2168 int size = nlmsg_msg_size(len);
2169
2170 nlh = skb_put(skb, NLMSG_ALIGN(size));
2171 nlh->nlmsg_type = type;
2172 nlh->nlmsg_len = size;
2173 nlh->nlmsg_flags = flags;
2174 nlh->nlmsg_pid = portid;
2175 nlh->nlmsg_seq = seq;
2176 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2177 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2178 return nlh;
2179}
2180EXPORT_SYMBOL(__nlmsg_put);
2181
2182/*
2183 * It looks a bit ugly.
2184 * It would be better to create kernel thread.
2185 */
2186
2187static int netlink_dump(struct sock *sk)
2188{
2189 struct netlink_sock *nlk = nlk_sk(sk);
2190 struct netlink_ext_ack extack = {};
2191 struct netlink_callback *cb;
2192 struct sk_buff *skb = NULL;
2193 struct nlmsghdr *nlh;
2194 struct module *module;
2195 int err = -ENOBUFS;
2196 int alloc_min_size;
2197 int alloc_size;
2198
2199 mutex_lock(nlk->cb_mutex);
2200 if (!nlk->cb_running) {
2201 err = -EINVAL;
2202 goto errout_skb;
2203 }
2204
2205 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2206 goto errout_skb;
2207
2208 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2209 * required, but it makes sense to _attempt_ a 16K bytes allocation
2210 * to reduce number of system calls on dump operations, if user
2211 * ever provided a big enough buffer.
2212 */
2213 cb = &nlk->cb;
2214 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2215
2216 if (alloc_min_size < nlk->max_recvmsg_len) {
2217 alloc_size = nlk->max_recvmsg_len;
2218 skb = alloc_skb(alloc_size,
2219 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2220 __GFP_NOWARN | __GFP_NORETRY);
2221 }
2222 if (!skb) {
2223 alloc_size = alloc_min_size;
2224 skb = alloc_skb(alloc_size, GFP_KERNEL);
2225 }
2226 if (!skb)
2227 goto errout_skb;
2228
2229 /* Trim skb to allocated size. User is expected to provide buffer as
2230 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2231 * netlink_recvmsg())). dump will pack as many smaller messages as
2232 * could fit within the allocated skb. skb is typically allocated
2233 * with larger space than required (could be as much as near 2x the
2234 * requested size with align to next power of 2 approach). Allowing
2235 * dump to use the excess space makes it difficult for a user to have a
2236 * reasonable static buffer based on the expected largest dump of a
2237 * single netdev. The outcome is MSG_TRUNC error.
2238 */
2239 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2240 netlink_skb_set_owner_r(skb, sk);
2241
2242 if (nlk->dump_done_errno > 0) {
2243 cb->extack = &extack;
2244 nlk->dump_done_errno = cb->dump(skb, cb);
2245 cb->extack = NULL;
2246 }
2247
2248 if (nlk->dump_done_errno > 0 ||
2249 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2250 mutex_unlock(nlk->cb_mutex);
2251
2252 if (sk_filter(sk, skb))
2253 kfree_skb(skb);
2254 else
2255 __netlink_sendskb(sk, skb);
2256 return 0;
2257 }
2258
2259 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE,
2260 sizeof(nlk->dump_done_errno),
2261 NLM_F_MULTI | cb->answer_flags);
2262 if (WARN_ON(!nlh))
2263 goto errout_skb;
2264
2265 nl_dump_check_consistent(cb, nlh);
2266
2267 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno,
2268 sizeof(nlk->dump_done_errno));
2269
2270 if (extack._msg && nlk->flags & NETLINK_F_EXT_ACK) {
2271 nlh->nlmsg_flags |= NLM_F_ACK_TLVS;
2272 if (!nla_put_string(skb, NLMSGERR_ATTR_MSG, extack._msg))
2273 nlmsg_end(skb, nlh);
2274 }
2275
2276 if (sk_filter(sk, skb))
2277 kfree_skb(skb);
2278 else
2279 __netlink_sendskb(sk, skb);
2280
2281 if (cb->done)
2282 cb->done(cb);
2283
2284 nlk->cb_running = false;
2285 module = cb->module;
2286 skb = cb->skb;
2287 mutex_unlock(nlk->cb_mutex);
2288 module_put(module);
2289 consume_skb(skb);
2290 return 0;
2291
2292errout_skb:
2293 mutex_unlock(nlk->cb_mutex);
2294 kfree_skb(skb);
2295 return err;
2296}
2297
2298int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2299 const struct nlmsghdr *nlh,
2300 struct netlink_dump_control *control)
2301{
2302 struct netlink_sock *nlk, *nlk2;
2303 struct netlink_callback *cb;
2304 struct sock *sk;
2305 int ret;
2306
2307 refcount_inc(&skb->users);
2308
2309 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2310 if (sk == NULL) {
2311 ret = -ECONNREFUSED;
2312 goto error_free;
2313 }
2314
2315 nlk = nlk_sk(sk);
2316 mutex_lock(nlk->cb_mutex);
2317 /* A dump is in progress... */
2318 if (nlk->cb_running) {
2319 ret = -EBUSY;
2320 goto error_unlock;
2321 }
2322 /* add reference of module which cb->dump belongs to */
2323 if (!try_module_get(control->module)) {
2324 ret = -EPROTONOSUPPORT;
2325 goto error_unlock;
2326 }
2327
2328 cb = &nlk->cb;
2329 memset(cb, 0, sizeof(*cb));
2330 cb->dump = control->dump;
2331 cb->done = control->done;
2332 cb->nlh = nlh;
2333 cb->data = control->data;
2334 cb->module = control->module;
2335 cb->min_dump_alloc = control->min_dump_alloc;
2336 cb->skb = skb;
2337
2338 nlk2 = nlk_sk(NETLINK_CB(skb).sk);
2339 cb->strict_check = !!(nlk2->flags & NETLINK_F_STRICT_CHK);
2340
2341 if (control->start) {
2342 ret = control->start(cb);
2343 if (ret)
2344 goto error_put;
2345 }
2346
2347 nlk->cb_running = true;
2348 nlk->dump_done_errno = INT_MAX;
2349
2350 mutex_unlock(nlk->cb_mutex);
2351
2352 ret = netlink_dump(sk);
2353
2354 sock_put(sk);
2355
2356 if (ret)
2357 return ret;
2358
2359 /* We successfully started a dump, by returning -EINTR we
2360 * signal not to send ACK even if it was requested.
2361 */
2362 return -EINTR;
2363
2364error_put:
2365 module_put(control->module);
2366error_unlock:
2367 sock_put(sk);
2368 mutex_unlock(nlk->cb_mutex);
2369error_free:
2370 kfree_skb(skb);
2371 return ret;
2372}
2373EXPORT_SYMBOL(__netlink_dump_start);
2374
2375void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2376 const struct netlink_ext_ack *extack)
2377{
2378 struct sk_buff *skb;
2379 struct nlmsghdr *rep;
2380 struct nlmsgerr *errmsg;
2381 size_t payload = sizeof(*errmsg);
2382 size_t tlvlen = 0;
2383 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2384 unsigned int flags = 0;
2385 bool nlk_has_extack = nlk->flags & NETLINK_F_EXT_ACK;
2386
2387 /* Error messages get the original request appened, unless the user
2388 * requests to cap the error message, and get extra error data if
2389 * requested.
2390 */
2391 if (nlk_has_extack && extack && extack->_msg)
2392 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2393
2394 if (err) {
2395 if (!(nlk->flags & NETLINK_F_CAP_ACK))
2396 payload += nlmsg_len(nlh);
2397 else
2398 flags |= NLM_F_CAPPED;
2399 if (nlk_has_extack && extack && extack->bad_attr)
2400 tlvlen += nla_total_size(sizeof(u32));
2401 } else {
2402 flags |= NLM_F_CAPPED;
2403
2404 if (nlk_has_extack && extack && extack->cookie_len)
2405 tlvlen += nla_total_size(extack->cookie_len);
2406 }
2407
2408 if (tlvlen)
2409 flags |= NLM_F_ACK_TLVS;
2410
2411 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2412 if (!skb) {
2413 NETLINK_CB(in_skb).sk->sk_err = ENOBUFS;
2414 NETLINK_CB(in_skb).sk->sk_error_report(NETLINK_CB(in_skb).sk);
2415 return;
2416 }
2417
2418 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2419 NLMSG_ERROR, payload, flags);
2420 errmsg = nlmsg_data(rep);
2421 errmsg->error = err;
2422 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2423
2424 if (nlk_has_extack && extack) {
2425 if (extack->_msg) {
2426 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG,
2427 extack->_msg));
2428 }
2429 if (err) {
2430 if (extack->bad_attr &&
2431 !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2432 (u8 *)extack->bad_attr >= in_skb->data +
2433 in_skb->len))
2434 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2435 (u8 *)extack->bad_attr -
2436 in_skb->data));
2437 } else {
2438 if (extack->cookie_len)
2439 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2440 extack->cookie_len,
2441 extack->cookie));
2442 }
2443 }
2444
2445 nlmsg_end(skb, rep);
2446
2447 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2448}
2449EXPORT_SYMBOL(netlink_ack);
2450
2451int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2452 struct nlmsghdr *,
2453 struct netlink_ext_ack *))
2454{
2455 struct netlink_ext_ack extack;
2456 struct nlmsghdr *nlh;
2457 int err;
2458
2459 while (skb->len >= nlmsg_total_size(0)) {
2460 int msglen;
2461
2462 memset(&extack, 0, sizeof(extack));
2463 nlh = nlmsg_hdr(skb);
2464 err = 0;
2465
2466 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2467 return 0;
2468
2469 /* Only requests are handled by the kernel */
2470 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2471 goto ack;
2472
2473 /* Skip control messages */
2474 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2475 goto ack;
2476
2477 err = cb(skb, nlh, &extack);
2478 if (err == -EINTR)
2479 goto skip;
2480
2481ack:
2482 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2483 netlink_ack(skb, nlh, err, &extack);
2484
2485skip:
2486 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2487 if (msglen > skb->len)
2488 msglen = skb->len;
2489 skb_pull(skb, msglen);
2490 }
2491
2492 return 0;
2493}
2494EXPORT_SYMBOL(netlink_rcv_skb);
2495
2496/**
2497 * nlmsg_notify - send a notification netlink message
2498 * @sk: netlink socket to use
2499 * @skb: notification message
2500 * @portid: destination netlink portid for reports or 0
2501 * @group: destination multicast group or 0
2502 * @report: 1 to report back, 0 to disable
2503 * @flags: allocation flags
2504 */
2505int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2506 unsigned int group, int report, gfp_t flags)
2507{
2508 int err = 0;
2509
2510 if (group) {
2511 int exclude_portid = 0;
2512
2513 if (report) {
2514 refcount_inc(&skb->users);
2515 exclude_portid = portid;
2516 }
2517
2518 /* errors reported via destination sk->sk_err, but propagate
2519 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2520 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2521 }
2522
2523 if (report) {
2524 int err2;
2525
2526 err2 = nlmsg_unicast(sk, skb, portid);
2527 if (!err || err == -ESRCH)
2528 err = err2;
2529 }
2530
2531 return err;
2532}
2533EXPORT_SYMBOL(nlmsg_notify);
2534
2535#ifdef CONFIG_PROC_FS
2536struct nl_seq_iter {
2537 struct seq_net_private p;
2538 struct rhashtable_iter hti;
2539 int link;
2540};
2541
2542static void netlink_walk_start(struct nl_seq_iter *iter)
2543{
2544 rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti);
2545 rhashtable_walk_start(&iter->hti);
2546}
2547
2548static void netlink_walk_stop(struct nl_seq_iter *iter)
2549{
2550 rhashtable_walk_stop(&iter->hti);
2551 rhashtable_walk_exit(&iter->hti);
2552}
2553
2554static void *__netlink_seq_next(struct seq_file *seq)
2555{
2556 struct nl_seq_iter *iter = seq->private;
2557 struct netlink_sock *nlk;
2558
2559 do {
2560 for (;;) {
2561 nlk = rhashtable_walk_next(&iter->hti);
2562
2563 if (IS_ERR(nlk)) {
2564 if (PTR_ERR(nlk) == -EAGAIN)
2565 continue;
2566
2567 return nlk;
2568 }
2569
2570 if (nlk)
2571 break;
2572
2573 netlink_walk_stop(iter);
2574 if (++iter->link >= MAX_LINKS)
2575 return NULL;
2576
2577 netlink_walk_start(iter);
2578 }
2579 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2580
2581 return nlk;
2582}
2583
2584static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2585{
2586 struct nl_seq_iter *iter = seq->private;
2587 void *obj = SEQ_START_TOKEN;
2588 loff_t pos;
2589
2590 iter->link = 0;
2591
2592 netlink_walk_start(iter);
2593
2594 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2595 obj = __netlink_seq_next(seq);
2596
2597 return obj;
2598}
2599
2600static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2601{
2602 ++*pos;
2603 return __netlink_seq_next(seq);
2604}
2605
2606static void netlink_seq_stop(struct seq_file *seq, void *v)
2607{
2608 struct nl_seq_iter *iter = seq->private;
2609
2610 if (iter->link >= MAX_LINKS)
2611 return;
2612
2613 netlink_walk_stop(iter);
2614}
2615
2616
2617static int netlink_seq_show(struct seq_file *seq, void *v)
2618{
2619 if (v == SEQ_START_TOKEN) {
2620 seq_puts(seq,
2621 "sk Eth Pid Groups "
2622 "Rmem Wmem Dump Locks Drops Inode\n");
2623 } else {
2624 struct sock *s = v;
2625 struct netlink_sock *nlk = nlk_sk(s);
2626
2627 seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n",
2628 s,
2629 s->sk_protocol,
2630 nlk->portid,
2631 nlk->groups ? (u32)nlk->groups[0] : 0,
2632 sk_rmem_alloc_get(s),
2633 sk_wmem_alloc_get(s),
2634 nlk->cb_running,
2635 refcount_read(&s->sk_refcnt),
2636 atomic_read(&s->sk_drops),
2637 sock_i_ino(s)
2638 );
2639
2640 }
2641 return 0;
2642}
2643
2644static const struct seq_operations netlink_seq_ops = {
2645 .start = netlink_seq_start,
2646 .next = netlink_seq_next,
2647 .stop = netlink_seq_stop,
2648 .show = netlink_seq_show,
2649};
2650#endif
2651
2652int netlink_register_notifier(struct notifier_block *nb)
2653{
2654 return blocking_notifier_chain_register(&netlink_chain, nb);
2655}
2656EXPORT_SYMBOL(netlink_register_notifier);
2657
2658int netlink_unregister_notifier(struct notifier_block *nb)
2659{
2660 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2661}
2662EXPORT_SYMBOL(netlink_unregister_notifier);
2663
2664static const struct proto_ops netlink_ops = {
2665 .family = PF_NETLINK,
2666 .owner = THIS_MODULE,
2667 .release = netlink_release,
2668 .bind = netlink_bind,
2669 .connect = netlink_connect,
2670 .socketpair = sock_no_socketpair,
2671 .accept = sock_no_accept,
2672 .getname = netlink_getname,
2673 .poll = datagram_poll,
2674 .ioctl = netlink_ioctl,
2675 .listen = sock_no_listen,
2676 .shutdown = sock_no_shutdown,
2677 .setsockopt = netlink_setsockopt,
2678 .getsockopt = netlink_getsockopt,
2679 .sendmsg = netlink_sendmsg,
2680 .recvmsg = netlink_recvmsg,
2681 .mmap = sock_no_mmap,
2682 .sendpage = sock_no_sendpage,
2683};
2684
2685static const struct net_proto_family netlink_family_ops = {
2686 .family = PF_NETLINK,
2687 .create = netlink_create,
2688 .owner = THIS_MODULE, /* for consistency 8) */
2689};
2690
2691static int __net_init netlink_net_init(struct net *net)
2692{
2693#ifdef CONFIG_PROC_FS
2694 if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2695 sizeof(struct nl_seq_iter)))
2696 return -ENOMEM;
2697#endif
2698 return 0;
2699}
2700
2701static void __net_exit netlink_net_exit(struct net *net)
2702{
2703#ifdef CONFIG_PROC_FS
2704 remove_proc_entry("netlink", net->proc_net);
2705#endif
2706}
2707
2708static void __init netlink_add_usersock_entry(void)
2709{
2710 struct listeners *listeners;
2711 int groups = 32;
2712
2713 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2714 if (!listeners)
2715 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2716
2717 netlink_table_grab();
2718
2719 nl_table[NETLINK_USERSOCK].groups = groups;
2720 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2721 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2722 nl_table[NETLINK_USERSOCK].registered = 1;
2723 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2724
2725 netlink_table_ungrab();
2726}
2727
2728static struct pernet_operations __net_initdata netlink_net_ops = {
2729 .init = netlink_net_init,
2730 .exit = netlink_net_exit,
2731};
2732
2733static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2734{
2735 const struct netlink_sock *nlk = data;
2736 struct netlink_compare_arg arg;
2737
2738 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2739 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2740}
2741
2742static const struct rhashtable_params netlink_rhashtable_params = {
2743 .head_offset = offsetof(struct netlink_sock, node),
2744 .key_len = netlink_compare_arg_len,
2745 .obj_hashfn = netlink_hash,
2746 .obj_cmpfn = netlink_compare,
2747 .automatic_shrinking = true,
2748};
2749
2750static int __init netlink_proto_init(void)
2751{
2752 int i;
2753 int err = proto_register(&netlink_proto, 0);
2754
2755 if (err != 0)
2756 goto out;
2757
2758 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2759
2760 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2761 if (!nl_table)
2762 goto panic;
2763
2764 for (i = 0; i < MAX_LINKS; i++) {
2765 if (rhashtable_init(&nl_table[i].hash,
2766 &netlink_rhashtable_params) < 0) {
2767 while (--i > 0)
2768 rhashtable_destroy(&nl_table[i].hash);
2769 kfree(nl_table);
2770 goto panic;
2771 }
2772 }
2773
2774 netlink_add_usersock_entry();
2775
2776 sock_register(&netlink_family_ops);
2777 register_pernet_subsys(&netlink_net_ops);
2778 register_pernet_subsys(&netlink_tap_net_ops);
2779 /* The netlink device handler may be needed early. */
2780 rtnetlink_init();
2781out:
2782 return err;
2783panic:
2784 panic("netlink_init: Cannot allocate nl_table\n");
2785}
2786
2787core_initcall(netlink_proto_init);
1/*
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
66#include <net/net_namespace.h>
67#include <net/sock.h>
68#include <net/scm.h>
69#include <net/netlink.h>
70
71#include "af_netlink.h"
72
73struct listeners {
74 struct rcu_head rcu;
75 unsigned long masks[0];
76};
77
78/* state bits */
79#define NETLINK_S_CONGESTED 0x0
80
81/* flags */
82#define NETLINK_F_KERNEL_SOCKET 0x1
83#define NETLINK_F_RECV_PKTINFO 0x2
84#define NETLINK_F_BROADCAST_SEND_ERROR 0x4
85#define NETLINK_F_RECV_NO_ENOBUFS 0x8
86#define NETLINK_F_LISTEN_ALL_NSID 0x10
87#define NETLINK_F_CAP_ACK 0x20
88
89static inline int netlink_is_kernel(struct sock *sk)
90{
91 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
92}
93
94struct netlink_table *nl_table __read_mostly;
95EXPORT_SYMBOL_GPL(nl_table);
96
97static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
98
99static int netlink_dump(struct sock *sk);
100static void netlink_skb_destructor(struct sk_buff *skb);
101
102/* nl_table locking explained:
103 * Lookup and traversal are protected with an RCU read-side lock. Insertion
104 * and removal are protected with per bucket lock while using RCU list
105 * modification primitives and may run in parallel to RCU protected lookups.
106 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
107 * been acquired * either during or after the socket has been removed from
108 * the list and after an RCU grace period.
109 */
110DEFINE_RWLOCK(nl_table_lock);
111EXPORT_SYMBOL_GPL(nl_table_lock);
112static atomic_t nl_table_users = ATOMIC_INIT(0);
113
114#define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
115
116static BLOCKING_NOTIFIER_HEAD(netlink_chain);
117
118static DEFINE_SPINLOCK(netlink_tap_lock);
119static struct list_head netlink_tap_all __read_mostly;
120
121static const struct rhashtable_params netlink_rhashtable_params;
122
123static inline u32 netlink_group_mask(u32 group)
124{
125 return group ? 1 << (group - 1) : 0;
126}
127
128static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
129 gfp_t gfp_mask)
130{
131 unsigned int len = skb_end_offset(skb);
132 struct sk_buff *new;
133
134 new = alloc_skb(len, gfp_mask);
135 if (new == NULL)
136 return NULL;
137
138 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
139 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
140 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
141
142 memcpy(skb_put(new, len), skb->data, len);
143 return new;
144}
145
146int netlink_add_tap(struct netlink_tap *nt)
147{
148 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
149 return -EINVAL;
150
151 spin_lock(&netlink_tap_lock);
152 list_add_rcu(&nt->list, &netlink_tap_all);
153 spin_unlock(&netlink_tap_lock);
154
155 __module_get(nt->module);
156
157 return 0;
158}
159EXPORT_SYMBOL_GPL(netlink_add_tap);
160
161static int __netlink_remove_tap(struct netlink_tap *nt)
162{
163 bool found = false;
164 struct netlink_tap *tmp;
165
166 spin_lock(&netlink_tap_lock);
167
168 list_for_each_entry(tmp, &netlink_tap_all, list) {
169 if (nt == tmp) {
170 list_del_rcu(&nt->list);
171 found = true;
172 goto out;
173 }
174 }
175
176 pr_warn("__netlink_remove_tap: %p not found\n", nt);
177out:
178 spin_unlock(&netlink_tap_lock);
179
180 if (found)
181 module_put(nt->module);
182
183 return found ? 0 : -ENODEV;
184}
185
186int netlink_remove_tap(struct netlink_tap *nt)
187{
188 int ret;
189
190 ret = __netlink_remove_tap(nt);
191 synchronize_net();
192
193 return ret;
194}
195EXPORT_SYMBOL_GPL(netlink_remove_tap);
196
197static bool netlink_filter_tap(const struct sk_buff *skb)
198{
199 struct sock *sk = skb->sk;
200
201 /* We take the more conservative approach and
202 * whitelist socket protocols that may pass.
203 */
204 switch (sk->sk_protocol) {
205 case NETLINK_ROUTE:
206 case NETLINK_USERSOCK:
207 case NETLINK_SOCK_DIAG:
208 case NETLINK_NFLOG:
209 case NETLINK_XFRM:
210 case NETLINK_FIB_LOOKUP:
211 case NETLINK_NETFILTER:
212 case NETLINK_GENERIC:
213 return true;
214 }
215
216 return false;
217}
218
219static int __netlink_deliver_tap_skb(struct sk_buff *skb,
220 struct net_device *dev)
221{
222 struct sk_buff *nskb;
223 struct sock *sk = skb->sk;
224 int ret = -ENOMEM;
225
226 dev_hold(dev);
227
228 if (is_vmalloc_addr(skb->head))
229 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
230 else
231 nskb = skb_clone(skb, GFP_ATOMIC);
232 if (nskb) {
233 nskb->dev = dev;
234 nskb->protocol = htons((u16) sk->sk_protocol);
235 nskb->pkt_type = netlink_is_kernel(sk) ?
236 PACKET_KERNEL : PACKET_USER;
237 skb_reset_network_header(nskb);
238 ret = dev_queue_xmit(nskb);
239 if (unlikely(ret > 0))
240 ret = net_xmit_errno(ret);
241 }
242
243 dev_put(dev);
244 return ret;
245}
246
247static void __netlink_deliver_tap(struct sk_buff *skb)
248{
249 int ret;
250 struct netlink_tap *tmp;
251
252 if (!netlink_filter_tap(skb))
253 return;
254
255 list_for_each_entry_rcu(tmp, &netlink_tap_all, list) {
256 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
257 if (unlikely(ret))
258 break;
259 }
260}
261
262static void netlink_deliver_tap(struct sk_buff *skb)
263{
264 rcu_read_lock();
265
266 if (unlikely(!list_empty(&netlink_tap_all)))
267 __netlink_deliver_tap(skb);
268
269 rcu_read_unlock();
270}
271
272static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
273 struct sk_buff *skb)
274{
275 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
276 netlink_deliver_tap(skb);
277}
278
279static void netlink_overrun(struct sock *sk)
280{
281 struct netlink_sock *nlk = nlk_sk(sk);
282
283 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
284 if (!test_and_set_bit(NETLINK_S_CONGESTED,
285 &nlk_sk(sk)->state)) {
286 sk->sk_err = ENOBUFS;
287 sk->sk_error_report(sk);
288 }
289 }
290 atomic_inc(&sk->sk_drops);
291}
292
293static void netlink_rcv_wake(struct sock *sk)
294{
295 struct netlink_sock *nlk = nlk_sk(sk);
296
297 if (skb_queue_empty(&sk->sk_receive_queue))
298 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
299 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
300 wake_up_interruptible(&nlk->wait);
301}
302
303static void netlink_skb_destructor(struct sk_buff *skb)
304{
305 if (is_vmalloc_addr(skb->head)) {
306 if (!skb->cloned ||
307 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
308 vfree(skb->head);
309
310 skb->head = NULL;
311 }
312 if (skb->sk != NULL)
313 sock_rfree(skb);
314}
315
316static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
317{
318 WARN_ON(skb->sk != NULL);
319 skb->sk = sk;
320 skb->destructor = netlink_skb_destructor;
321 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
322 sk_mem_charge(sk, skb->truesize);
323}
324
325static void netlink_sock_destruct(struct sock *sk)
326{
327 struct netlink_sock *nlk = nlk_sk(sk);
328
329 if (nlk->cb_running) {
330 if (nlk->cb.done)
331 nlk->cb.done(&nlk->cb);
332 module_put(nlk->cb.module);
333 kfree_skb(nlk->cb.skb);
334 }
335
336 skb_queue_purge(&sk->sk_receive_queue);
337
338 if (!sock_flag(sk, SOCK_DEAD)) {
339 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
340 return;
341 }
342
343 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
344 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
345 WARN_ON(nlk_sk(sk)->groups);
346}
347
348static void netlink_sock_destruct_work(struct work_struct *work)
349{
350 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
351 work);
352
353 sk_free(&nlk->sk);
354}
355
356/* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
357 * SMP. Look, when several writers sleep and reader wakes them up, all but one
358 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
359 * this, _but_ remember, it adds useless work on UP machines.
360 */
361
362void netlink_table_grab(void)
363 __acquires(nl_table_lock)
364{
365 might_sleep();
366
367 write_lock_irq(&nl_table_lock);
368
369 if (atomic_read(&nl_table_users)) {
370 DECLARE_WAITQUEUE(wait, current);
371
372 add_wait_queue_exclusive(&nl_table_wait, &wait);
373 for (;;) {
374 set_current_state(TASK_UNINTERRUPTIBLE);
375 if (atomic_read(&nl_table_users) == 0)
376 break;
377 write_unlock_irq(&nl_table_lock);
378 schedule();
379 write_lock_irq(&nl_table_lock);
380 }
381
382 __set_current_state(TASK_RUNNING);
383 remove_wait_queue(&nl_table_wait, &wait);
384 }
385}
386
387void netlink_table_ungrab(void)
388 __releases(nl_table_lock)
389{
390 write_unlock_irq(&nl_table_lock);
391 wake_up(&nl_table_wait);
392}
393
394static inline void
395netlink_lock_table(void)
396{
397 /* read_lock() synchronizes us to netlink_table_grab */
398
399 read_lock(&nl_table_lock);
400 atomic_inc(&nl_table_users);
401 read_unlock(&nl_table_lock);
402}
403
404static inline void
405netlink_unlock_table(void)
406{
407 if (atomic_dec_and_test(&nl_table_users))
408 wake_up(&nl_table_wait);
409}
410
411struct netlink_compare_arg
412{
413 possible_net_t pnet;
414 u32 portid;
415};
416
417/* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
418#define netlink_compare_arg_len \
419 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
420
421static inline int netlink_compare(struct rhashtable_compare_arg *arg,
422 const void *ptr)
423{
424 const struct netlink_compare_arg *x = arg->key;
425 const struct netlink_sock *nlk = ptr;
426
427 return nlk->portid != x->portid ||
428 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
429}
430
431static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
432 struct net *net, u32 portid)
433{
434 memset(arg, 0, sizeof(*arg));
435 write_pnet(&arg->pnet, net);
436 arg->portid = portid;
437}
438
439static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
440 struct net *net)
441{
442 struct netlink_compare_arg arg;
443
444 netlink_compare_arg_init(&arg, net, portid);
445 return rhashtable_lookup_fast(&table->hash, &arg,
446 netlink_rhashtable_params);
447}
448
449static int __netlink_insert(struct netlink_table *table, struct sock *sk)
450{
451 struct netlink_compare_arg arg;
452
453 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
454 return rhashtable_lookup_insert_key(&table->hash, &arg,
455 &nlk_sk(sk)->node,
456 netlink_rhashtable_params);
457}
458
459static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
460{
461 struct netlink_table *table = &nl_table[protocol];
462 struct sock *sk;
463
464 rcu_read_lock();
465 sk = __netlink_lookup(table, portid, net);
466 if (sk)
467 sock_hold(sk);
468 rcu_read_unlock();
469
470 return sk;
471}
472
473static const struct proto_ops netlink_ops;
474
475static void
476netlink_update_listeners(struct sock *sk)
477{
478 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
479 unsigned long mask;
480 unsigned int i;
481 struct listeners *listeners;
482
483 listeners = nl_deref_protected(tbl->listeners);
484 if (!listeners)
485 return;
486
487 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
488 mask = 0;
489 sk_for_each_bound(sk, &tbl->mc_list) {
490 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
491 mask |= nlk_sk(sk)->groups[i];
492 }
493 listeners->masks[i] = mask;
494 }
495 /* this function is only called with the netlink table "grabbed", which
496 * makes sure updates are visible before bind or setsockopt return. */
497}
498
499static int netlink_insert(struct sock *sk, u32 portid)
500{
501 struct netlink_table *table = &nl_table[sk->sk_protocol];
502 int err;
503
504 lock_sock(sk);
505
506 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
507 if (nlk_sk(sk)->bound)
508 goto err;
509
510 err = -ENOMEM;
511 if (BITS_PER_LONG > 32 &&
512 unlikely(atomic_read(&table->hash.nelems) >= UINT_MAX))
513 goto err;
514
515 nlk_sk(sk)->portid = portid;
516 sock_hold(sk);
517
518 err = __netlink_insert(table, sk);
519 if (err) {
520 /* In case the hashtable backend returns with -EBUSY
521 * from here, it must not escape to the caller.
522 */
523 if (unlikely(err == -EBUSY))
524 err = -EOVERFLOW;
525 if (err == -EEXIST)
526 err = -EADDRINUSE;
527 sock_put(sk);
528 goto err;
529 }
530
531 /* We need to ensure that the socket is hashed and visible. */
532 smp_wmb();
533 nlk_sk(sk)->bound = portid;
534
535err:
536 release_sock(sk);
537 return err;
538}
539
540static void netlink_remove(struct sock *sk)
541{
542 struct netlink_table *table;
543
544 table = &nl_table[sk->sk_protocol];
545 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
546 netlink_rhashtable_params)) {
547 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
548 __sock_put(sk);
549 }
550
551 netlink_table_grab();
552 if (nlk_sk(sk)->subscriptions) {
553 __sk_del_bind_node(sk);
554 netlink_update_listeners(sk);
555 }
556 if (sk->sk_protocol == NETLINK_GENERIC)
557 atomic_inc(&genl_sk_destructing_cnt);
558 netlink_table_ungrab();
559}
560
561static struct proto netlink_proto = {
562 .name = "NETLINK",
563 .owner = THIS_MODULE,
564 .obj_size = sizeof(struct netlink_sock),
565};
566
567static int __netlink_create(struct net *net, struct socket *sock,
568 struct mutex *cb_mutex, int protocol,
569 int kern)
570{
571 struct sock *sk;
572 struct netlink_sock *nlk;
573
574 sock->ops = &netlink_ops;
575
576 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
577 if (!sk)
578 return -ENOMEM;
579
580 sock_init_data(sock, sk);
581
582 nlk = nlk_sk(sk);
583 if (cb_mutex) {
584 nlk->cb_mutex = cb_mutex;
585 } else {
586 nlk->cb_mutex = &nlk->cb_def_mutex;
587 mutex_init(nlk->cb_mutex);
588 }
589 init_waitqueue_head(&nlk->wait);
590
591 sk->sk_destruct = netlink_sock_destruct;
592 sk->sk_protocol = protocol;
593 return 0;
594}
595
596static int netlink_create(struct net *net, struct socket *sock, int protocol,
597 int kern)
598{
599 struct module *module = NULL;
600 struct mutex *cb_mutex;
601 struct netlink_sock *nlk;
602 int (*bind)(struct net *net, int group);
603 void (*unbind)(struct net *net, int group);
604 int err = 0;
605
606 sock->state = SS_UNCONNECTED;
607
608 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
609 return -ESOCKTNOSUPPORT;
610
611 if (protocol < 0 || protocol >= MAX_LINKS)
612 return -EPROTONOSUPPORT;
613
614 netlink_lock_table();
615#ifdef CONFIG_MODULES
616 if (!nl_table[protocol].registered) {
617 netlink_unlock_table();
618 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
619 netlink_lock_table();
620 }
621#endif
622 if (nl_table[protocol].registered &&
623 try_module_get(nl_table[protocol].module))
624 module = nl_table[protocol].module;
625 else
626 err = -EPROTONOSUPPORT;
627 cb_mutex = nl_table[protocol].cb_mutex;
628 bind = nl_table[protocol].bind;
629 unbind = nl_table[protocol].unbind;
630 netlink_unlock_table();
631
632 if (err < 0)
633 goto out;
634
635 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
636 if (err < 0)
637 goto out_module;
638
639 local_bh_disable();
640 sock_prot_inuse_add(net, &netlink_proto, 1);
641 local_bh_enable();
642
643 nlk = nlk_sk(sock->sk);
644 nlk->module = module;
645 nlk->netlink_bind = bind;
646 nlk->netlink_unbind = unbind;
647out:
648 return err;
649
650out_module:
651 module_put(module);
652 goto out;
653}
654
655static void deferred_put_nlk_sk(struct rcu_head *head)
656{
657 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
658 struct sock *sk = &nlk->sk;
659
660 if (!atomic_dec_and_test(&sk->sk_refcnt))
661 return;
662
663 if (nlk->cb_running && nlk->cb.done) {
664 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
665 schedule_work(&nlk->work);
666 return;
667 }
668
669 sk_free(sk);
670}
671
672static int netlink_release(struct socket *sock)
673{
674 struct sock *sk = sock->sk;
675 struct netlink_sock *nlk;
676
677 if (!sk)
678 return 0;
679
680 netlink_remove(sk);
681 sock_orphan(sk);
682 nlk = nlk_sk(sk);
683
684 /*
685 * OK. Socket is unlinked, any packets that arrive now
686 * will be purged.
687 */
688
689 /* must not acquire netlink_table_lock in any way again before unbind
690 * and notifying genetlink is done as otherwise it might deadlock
691 */
692 if (nlk->netlink_unbind) {
693 int i;
694
695 for (i = 0; i < nlk->ngroups; i++)
696 if (test_bit(i, nlk->groups))
697 nlk->netlink_unbind(sock_net(sk), i + 1);
698 }
699 if (sk->sk_protocol == NETLINK_GENERIC &&
700 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
701 wake_up(&genl_sk_destructing_waitq);
702
703 sock->sk = NULL;
704 wake_up_interruptible_all(&nlk->wait);
705
706 skb_queue_purge(&sk->sk_write_queue);
707
708 if (nlk->portid && nlk->bound) {
709 struct netlink_notify n = {
710 .net = sock_net(sk),
711 .protocol = sk->sk_protocol,
712 .portid = nlk->portid,
713 };
714 blocking_notifier_call_chain(&netlink_chain,
715 NETLINK_URELEASE, &n);
716 }
717
718 module_put(nlk->module);
719
720 if (netlink_is_kernel(sk)) {
721 netlink_table_grab();
722 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
723 if (--nl_table[sk->sk_protocol].registered == 0) {
724 struct listeners *old;
725
726 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
727 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
728 kfree_rcu(old, rcu);
729 nl_table[sk->sk_protocol].module = NULL;
730 nl_table[sk->sk_protocol].bind = NULL;
731 nl_table[sk->sk_protocol].unbind = NULL;
732 nl_table[sk->sk_protocol].flags = 0;
733 nl_table[sk->sk_protocol].registered = 0;
734 }
735 netlink_table_ungrab();
736 }
737
738 kfree(nlk->groups);
739 nlk->groups = NULL;
740
741 local_bh_disable();
742 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
743 local_bh_enable();
744 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
745 return 0;
746}
747
748static int netlink_autobind(struct socket *sock)
749{
750 struct sock *sk = sock->sk;
751 struct net *net = sock_net(sk);
752 struct netlink_table *table = &nl_table[sk->sk_protocol];
753 s32 portid = task_tgid_vnr(current);
754 int err;
755 s32 rover = -4096;
756 bool ok;
757
758retry:
759 cond_resched();
760 rcu_read_lock();
761 ok = !__netlink_lookup(table, portid, net);
762 rcu_read_unlock();
763 if (!ok) {
764 /* Bind collision, search negative portid values. */
765 if (rover == -4096)
766 /* rover will be in range [S32_MIN, -4097] */
767 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
768 else if (rover >= -4096)
769 rover = -4097;
770 portid = rover--;
771 goto retry;
772 }
773
774 err = netlink_insert(sk, portid);
775 if (err == -EADDRINUSE)
776 goto retry;
777
778 /* If 2 threads race to autobind, that is fine. */
779 if (err == -EBUSY)
780 err = 0;
781
782 return err;
783}
784
785/**
786 * __netlink_ns_capable - General netlink message capability test
787 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
788 * @user_ns: The user namespace of the capability to use
789 * @cap: The capability to use
790 *
791 * Test to see if the opener of the socket we received the message
792 * from had when the netlink socket was created and the sender of the
793 * message has has the capability @cap in the user namespace @user_ns.
794 */
795bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
796 struct user_namespace *user_ns, int cap)
797{
798 return ((nsp->flags & NETLINK_SKB_DST) ||
799 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
800 ns_capable(user_ns, cap);
801}
802EXPORT_SYMBOL(__netlink_ns_capable);
803
804/**
805 * netlink_ns_capable - General netlink message capability test
806 * @skb: socket buffer holding a netlink command from userspace
807 * @user_ns: The user namespace of the capability to use
808 * @cap: The capability to use
809 *
810 * Test to see if the opener of the socket we received the message
811 * from had when the netlink socket was created and the sender of the
812 * message has has the capability @cap in the user namespace @user_ns.
813 */
814bool netlink_ns_capable(const struct sk_buff *skb,
815 struct user_namespace *user_ns, int cap)
816{
817 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
818}
819EXPORT_SYMBOL(netlink_ns_capable);
820
821/**
822 * netlink_capable - Netlink global message capability test
823 * @skb: socket buffer holding a netlink command from userspace
824 * @cap: The capability to use
825 *
826 * Test to see if the opener of the socket we received the message
827 * from had when the netlink socket was created and the sender of the
828 * message has has the capability @cap in all user namespaces.
829 */
830bool netlink_capable(const struct sk_buff *skb, int cap)
831{
832 return netlink_ns_capable(skb, &init_user_ns, cap);
833}
834EXPORT_SYMBOL(netlink_capable);
835
836/**
837 * netlink_net_capable - Netlink network namespace message capability test
838 * @skb: socket buffer holding a netlink command from userspace
839 * @cap: The capability to use
840 *
841 * Test to see if the opener of the socket we received the message
842 * from had when the netlink socket was created and the sender of the
843 * message has has the capability @cap over the network namespace of
844 * the socket we received the message from.
845 */
846bool netlink_net_capable(const struct sk_buff *skb, int cap)
847{
848 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
849}
850EXPORT_SYMBOL(netlink_net_capable);
851
852static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
853{
854 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
855 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
856}
857
858static void
859netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
860{
861 struct netlink_sock *nlk = nlk_sk(sk);
862
863 if (nlk->subscriptions && !subscriptions)
864 __sk_del_bind_node(sk);
865 else if (!nlk->subscriptions && subscriptions)
866 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
867 nlk->subscriptions = subscriptions;
868}
869
870static int netlink_realloc_groups(struct sock *sk)
871{
872 struct netlink_sock *nlk = nlk_sk(sk);
873 unsigned int groups;
874 unsigned long *new_groups;
875 int err = 0;
876
877 netlink_table_grab();
878
879 groups = nl_table[sk->sk_protocol].groups;
880 if (!nl_table[sk->sk_protocol].registered) {
881 err = -ENOENT;
882 goto out_unlock;
883 }
884
885 if (nlk->ngroups >= groups)
886 goto out_unlock;
887
888 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
889 if (new_groups == NULL) {
890 err = -ENOMEM;
891 goto out_unlock;
892 }
893 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
894 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
895
896 nlk->groups = new_groups;
897 nlk->ngroups = groups;
898 out_unlock:
899 netlink_table_ungrab();
900 return err;
901}
902
903static void netlink_undo_bind(int group, long unsigned int groups,
904 struct sock *sk)
905{
906 struct netlink_sock *nlk = nlk_sk(sk);
907 int undo;
908
909 if (!nlk->netlink_unbind)
910 return;
911
912 for (undo = 0; undo < group; undo++)
913 if (test_bit(undo, &groups))
914 nlk->netlink_unbind(sock_net(sk), undo + 1);
915}
916
917static int netlink_bind(struct socket *sock, struct sockaddr *addr,
918 int addr_len)
919{
920 struct sock *sk = sock->sk;
921 struct net *net = sock_net(sk);
922 struct netlink_sock *nlk = nlk_sk(sk);
923 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
924 int err;
925 long unsigned int groups = nladdr->nl_groups;
926 bool bound;
927
928 if (addr_len < sizeof(struct sockaddr_nl))
929 return -EINVAL;
930
931 if (nladdr->nl_family != AF_NETLINK)
932 return -EINVAL;
933
934 /* Only superuser is allowed to listen multicasts */
935 if (groups) {
936 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
937 return -EPERM;
938 err = netlink_realloc_groups(sk);
939 if (err)
940 return err;
941 }
942
943 bound = nlk->bound;
944 if (bound) {
945 /* Ensure nlk->portid is up-to-date. */
946 smp_rmb();
947
948 if (nladdr->nl_pid != nlk->portid)
949 return -EINVAL;
950 }
951
952 if (nlk->netlink_bind && groups) {
953 int group;
954
955 for (group = 0; group < nlk->ngroups; group++) {
956 if (!test_bit(group, &groups))
957 continue;
958 err = nlk->netlink_bind(net, group + 1);
959 if (!err)
960 continue;
961 netlink_undo_bind(group, groups, sk);
962 return err;
963 }
964 }
965
966 /* No need for barriers here as we return to user-space without
967 * using any of the bound attributes.
968 */
969 if (!bound) {
970 err = nladdr->nl_pid ?
971 netlink_insert(sk, nladdr->nl_pid) :
972 netlink_autobind(sock);
973 if (err) {
974 netlink_undo_bind(nlk->ngroups, groups, sk);
975 return err;
976 }
977 }
978
979 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
980 return 0;
981
982 netlink_table_grab();
983 netlink_update_subscriptions(sk, nlk->subscriptions +
984 hweight32(groups) -
985 hweight32(nlk->groups[0]));
986 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
987 netlink_update_listeners(sk);
988 netlink_table_ungrab();
989
990 return 0;
991}
992
993static int netlink_connect(struct socket *sock, struct sockaddr *addr,
994 int alen, int flags)
995{
996 int err = 0;
997 struct sock *sk = sock->sk;
998 struct netlink_sock *nlk = nlk_sk(sk);
999 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1000
1001 if (alen < sizeof(addr->sa_family))
1002 return -EINVAL;
1003
1004 if (addr->sa_family == AF_UNSPEC) {
1005 sk->sk_state = NETLINK_UNCONNECTED;
1006 nlk->dst_portid = 0;
1007 nlk->dst_group = 0;
1008 return 0;
1009 }
1010 if (addr->sa_family != AF_NETLINK)
1011 return -EINVAL;
1012
1013 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1014 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1015 return -EPERM;
1016
1017 /* No need for barriers here as we return to user-space without
1018 * using any of the bound attributes.
1019 */
1020 if (!nlk->bound)
1021 err = netlink_autobind(sock);
1022
1023 if (err == 0) {
1024 sk->sk_state = NETLINK_CONNECTED;
1025 nlk->dst_portid = nladdr->nl_pid;
1026 nlk->dst_group = ffs(nladdr->nl_groups);
1027 }
1028
1029 return err;
1030}
1031
1032static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1033 int *addr_len, int peer)
1034{
1035 struct sock *sk = sock->sk;
1036 struct netlink_sock *nlk = nlk_sk(sk);
1037 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1038
1039 nladdr->nl_family = AF_NETLINK;
1040 nladdr->nl_pad = 0;
1041 *addr_len = sizeof(*nladdr);
1042
1043 if (peer) {
1044 nladdr->nl_pid = nlk->dst_portid;
1045 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1046 } else {
1047 nladdr->nl_pid = nlk->portid;
1048 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1049 }
1050 return 0;
1051}
1052
1053static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1054 unsigned long arg)
1055{
1056 /* try to hand this ioctl down to the NIC drivers.
1057 */
1058 return -ENOIOCTLCMD;
1059}
1060
1061static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1062{
1063 struct sock *sock;
1064 struct netlink_sock *nlk;
1065
1066 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1067 if (!sock)
1068 return ERR_PTR(-ECONNREFUSED);
1069
1070 /* Don't bother queuing skb if kernel socket has no input function */
1071 nlk = nlk_sk(sock);
1072 if (sock->sk_state == NETLINK_CONNECTED &&
1073 nlk->dst_portid != nlk_sk(ssk)->portid) {
1074 sock_put(sock);
1075 return ERR_PTR(-ECONNREFUSED);
1076 }
1077 return sock;
1078}
1079
1080struct sock *netlink_getsockbyfilp(struct file *filp)
1081{
1082 struct inode *inode = file_inode(filp);
1083 struct sock *sock;
1084
1085 if (!S_ISSOCK(inode->i_mode))
1086 return ERR_PTR(-ENOTSOCK);
1087
1088 sock = SOCKET_I(inode)->sk;
1089 if (sock->sk_family != AF_NETLINK)
1090 return ERR_PTR(-EINVAL);
1091
1092 sock_hold(sock);
1093 return sock;
1094}
1095
1096static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1097 int broadcast)
1098{
1099 struct sk_buff *skb;
1100 void *data;
1101
1102 if (size <= NLMSG_GOODSIZE || broadcast)
1103 return alloc_skb(size, GFP_KERNEL);
1104
1105 size = SKB_DATA_ALIGN(size) +
1106 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1107
1108 data = vmalloc(size);
1109 if (data == NULL)
1110 return NULL;
1111
1112 skb = __build_skb(data, size);
1113 if (skb == NULL)
1114 vfree(data);
1115 else
1116 skb->destructor = netlink_skb_destructor;
1117
1118 return skb;
1119}
1120
1121/*
1122 * Attach a skb to a netlink socket.
1123 * The caller must hold a reference to the destination socket. On error, the
1124 * reference is dropped. The skb is not send to the destination, just all
1125 * all error checks are performed and memory in the queue is reserved.
1126 * Return values:
1127 * < 0: error. skb freed, reference to sock dropped.
1128 * 0: continue
1129 * 1: repeat lookup - reference dropped while waiting for socket memory.
1130 */
1131int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1132 long *timeo, struct sock *ssk)
1133{
1134 struct netlink_sock *nlk;
1135
1136 nlk = nlk_sk(sk);
1137
1138 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1139 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1140 DECLARE_WAITQUEUE(wait, current);
1141 if (!*timeo) {
1142 if (!ssk || netlink_is_kernel(ssk))
1143 netlink_overrun(sk);
1144 sock_put(sk);
1145 kfree_skb(skb);
1146 return -EAGAIN;
1147 }
1148
1149 __set_current_state(TASK_INTERRUPTIBLE);
1150 add_wait_queue(&nlk->wait, &wait);
1151
1152 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1153 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1154 !sock_flag(sk, SOCK_DEAD))
1155 *timeo = schedule_timeout(*timeo);
1156
1157 __set_current_state(TASK_RUNNING);
1158 remove_wait_queue(&nlk->wait, &wait);
1159 sock_put(sk);
1160
1161 if (signal_pending(current)) {
1162 kfree_skb(skb);
1163 return sock_intr_errno(*timeo);
1164 }
1165 return 1;
1166 }
1167 netlink_skb_set_owner_r(skb, sk);
1168 return 0;
1169}
1170
1171static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1172{
1173 int len = skb->len;
1174
1175 netlink_deliver_tap(skb);
1176
1177 skb_queue_tail(&sk->sk_receive_queue, skb);
1178 sk->sk_data_ready(sk);
1179 return len;
1180}
1181
1182int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1183{
1184 int len = __netlink_sendskb(sk, skb);
1185
1186 sock_put(sk);
1187 return len;
1188}
1189
1190void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1191{
1192 kfree_skb(skb);
1193 sock_put(sk);
1194}
1195
1196static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1197{
1198 int delta;
1199
1200 WARN_ON(skb->sk != NULL);
1201 delta = skb->end - skb->tail;
1202 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1203 return skb;
1204
1205 if (skb_shared(skb)) {
1206 struct sk_buff *nskb = skb_clone(skb, allocation);
1207 if (!nskb)
1208 return skb;
1209 consume_skb(skb);
1210 skb = nskb;
1211 }
1212
1213 if (!pskb_expand_head(skb, 0, -delta, allocation))
1214 skb->truesize -= delta;
1215
1216 return skb;
1217}
1218
1219static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1220 struct sock *ssk)
1221{
1222 int ret;
1223 struct netlink_sock *nlk = nlk_sk(sk);
1224
1225 ret = -ECONNREFUSED;
1226 if (nlk->netlink_rcv != NULL) {
1227 ret = skb->len;
1228 netlink_skb_set_owner_r(skb, sk);
1229 NETLINK_CB(skb).sk = ssk;
1230 netlink_deliver_tap_kernel(sk, ssk, skb);
1231 nlk->netlink_rcv(skb);
1232 consume_skb(skb);
1233 } else {
1234 kfree_skb(skb);
1235 }
1236 sock_put(sk);
1237 return ret;
1238}
1239
1240int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1241 u32 portid, int nonblock)
1242{
1243 struct sock *sk;
1244 int err;
1245 long timeo;
1246
1247 skb = netlink_trim(skb, gfp_any());
1248
1249 timeo = sock_sndtimeo(ssk, nonblock);
1250retry:
1251 sk = netlink_getsockbyportid(ssk, portid);
1252 if (IS_ERR(sk)) {
1253 kfree_skb(skb);
1254 return PTR_ERR(sk);
1255 }
1256 if (netlink_is_kernel(sk))
1257 return netlink_unicast_kernel(sk, skb, ssk);
1258
1259 if (sk_filter(sk, skb)) {
1260 err = skb->len;
1261 kfree_skb(skb);
1262 sock_put(sk);
1263 return err;
1264 }
1265
1266 err = netlink_attachskb(sk, skb, &timeo, ssk);
1267 if (err == 1)
1268 goto retry;
1269 if (err)
1270 return err;
1271
1272 return netlink_sendskb(sk, skb);
1273}
1274EXPORT_SYMBOL(netlink_unicast);
1275
1276int netlink_has_listeners(struct sock *sk, unsigned int group)
1277{
1278 int res = 0;
1279 struct listeners *listeners;
1280
1281 BUG_ON(!netlink_is_kernel(sk));
1282
1283 rcu_read_lock();
1284 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1285
1286 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1287 res = test_bit(group - 1, listeners->masks);
1288
1289 rcu_read_unlock();
1290
1291 return res;
1292}
1293EXPORT_SYMBOL_GPL(netlink_has_listeners);
1294
1295static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1296{
1297 struct netlink_sock *nlk = nlk_sk(sk);
1298
1299 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1300 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1301 netlink_skb_set_owner_r(skb, sk);
1302 __netlink_sendskb(sk, skb);
1303 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1304 }
1305 return -1;
1306}
1307
1308struct netlink_broadcast_data {
1309 struct sock *exclude_sk;
1310 struct net *net;
1311 u32 portid;
1312 u32 group;
1313 int failure;
1314 int delivery_failure;
1315 int congested;
1316 int delivered;
1317 gfp_t allocation;
1318 struct sk_buff *skb, *skb2;
1319 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1320 void *tx_data;
1321};
1322
1323static void do_one_broadcast(struct sock *sk,
1324 struct netlink_broadcast_data *p)
1325{
1326 struct netlink_sock *nlk = nlk_sk(sk);
1327 int val;
1328
1329 if (p->exclude_sk == sk)
1330 return;
1331
1332 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1333 !test_bit(p->group - 1, nlk->groups))
1334 return;
1335
1336 if (!net_eq(sock_net(sk), p->net)) {
1337 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1338 return;
1339
1340 if (!peernet_has_id(sock_net(sk), p->net))
1341 return;
1342
1343 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1344 CAP_NET_BROADCAST))
1345 return;
1346 }
1347
1348 if (p->failure) {
1349 netlink_overrun(sk);
1350 return;
1351 }
1352
1353 sock_hold(sk);
1354 if (p->skb2 == NULL) {
1355 if (skb_shared(p->skb)) {
1356 p->skb2 = skb_clone(p->skb, p->allocation);
1357 } else {
1358 p->skb2 = skb_get(p->skb);
1359 /*
1360 * skb ownership may have been set when
1361 * delivered to a previous socket.
1362 */
1363 skb_orphan(p->skb2);
1364 }
1365 }
1366 if (p->skb2 == NULL) {
1367 netlink_overrun(sk);
1368 /* Clone failed. Notify ALL listeners. */
1369 p->failure = 1;
1370 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1371 p->delivery_failure = 1;
1372 goto out;
1373 }
1374 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1375 kfree_skb(p->skb2);
1376 p->skb2 = NULL;
1377 goto out;
1378 }
1379 if (sk_filter(sk, p->skb2)) {
1380 kfree_skb(p->skb2);
1381 p->skb2 = NULL;
1382 goto out;
1383 }
1384 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1385 NETLINK_CB(p->skb2).nsid_is_set = true;
1386 val = netlink_broadcast_deliver(sk, p->skb2);
1387 if (val < 0) {
1388 netlink_overrun(sk);
1389 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1390 p->delivery_failure = 1;
1391 } else {
1392 p->congested |= val;
1393 p->delivered = 1;
1394 p->skb2 = NULL;
1395 }
1396out:
1397 sock_put(sk);
1398}
1399
1400int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1401 u32 group, gfp_t allocation,
1402 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1403 void *filter_data)
1404{
1405 struct net *net = sock_net(ssk);
1406 struct netlink_broadcast_data info;
1407 struct sock *sk;
1408
1409 skb = netlink_trim(skb, allocation);
1410
1411 info.exclude_sk = ssk;
1412 info.net = net;
1413 info.portid = portid;
1414 info.group = group;
1415 info.failure = 0;
1416 info.delivery_failure = 0;
1417 info.congested = 0;
1418 info.delivered = 0;
1419 info.allocation = allocation;
1420 info.skb = skb;
1421 info.skb2 = NULL;
1422 info.tx_filter = filter;
1423 info.tx_data = filter_data;
1424
1425 /* While we sleep in clone, do not allow to change socket list */
1426
1427 netlink_lock_table();
1428
1429 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1430 do_one_broadcast(sk, &info);
1431
1432 consume_skb(skb);
1433
1434 netlink_unlock_table();
1435
1436 if (info.delivery_failure) {
1437 kfree_skb(info.skb2);
1438 return -ENOBUFS;
1439 }
1440 consume_skb(info.skb2);
1441
1442 if (info.delivered) {
1443 if (info.congested && gfpflags_allow_blocking(allocation))
1444 yield();
1445 return 0;
1446 }
1447 return -ESRCH;
1448}
1449EXPORT_SYMBOL(netlink_broadcast_filtered);
1450
1451int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1452 u32 group, gfp_t allocation)
1453{
1454 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1455 NULL, NULL);
1456}
1457EXPORT_SYMBOL(netlink_broadcast);
1458
1459struct netlink_set_err_data {
1460 struct sock *exclude_sk;
1461 u32 portid;
1462 u32 group;
1463 int code;
1464};
1465
1466static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1467{
1468 struct netlink_sock *nlk = nlk_sk(sk);
1469 int ret = 0;
1470
1471 if (sk == p->exclude_sk)
1472 goto out;
1473
1474 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1475 goto out;
1476
1477 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1478 !test_bit(p->group - 1, nlk->groups))
1479 goto out;
1480
1481 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1482 ret = 1;
1483 goto out;
1484 }
1485
1486 sk->sk_err = p->code;
1487 sk->sk_error_report(sk);
1488out:
1489 return ret;
1490}
1491
1492/**
1493 * netlink_set_err - report error to broadcast listeners
1494 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1495 * @portid: the PORTID of a process that we want to skip (if any)
1496 * @group: the broadcast group that will notice the error
1497 * @code: error code, must be negative (as usual in kernelspace)
1498 *
1499 * This function returns the number of broadcast listeners that have set the
1500 * NETLINK_NO_ENOBUFS socket option.
1501 */
1502int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1503{
1504 struct netlink_set_err_data info;
1505 struct sock *sk;
1506 int ret = 0;
1507
1508 info.exclude_sk = ssk;
1509 info.portid = portid;
1510 info.group = group;
1511 /* sk->sk_err wants a positive error value */
1512 info.code = -code;
1513
1514 read_lock(&nl_table_lock);
1515
1516 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1517 ret += do_one_set_err(sk, &info);
1518
1519 read_unlock(&nl_table_lock);
1520 return ret;
1521}
1522EXPORT_SYMBOL(netlink_set_err);
1523
1524/* must be called with netlink table grabbed */
1525static void netlink_update_socket_mc(struct netlink_sock *nlk,
1526 unsigned int group,
1527 int is_new)
1528{
1529 int old, new = !!is_new, subscriptions;
1530
1531 old = test_bit(group - 1, nlk->groups);
1532 subscriptions = nlk->subscriptions - old + new;
1533 if (new)
1534 __set_bit(group - 1, nlk->groups);
1535 else
1536 __clear_bit(group - 1, nlk->groups);
1537 netlink_update_subscriptions(&nlk->sk, subscriptions);
1538 netlink_update_listeners(&nlk->sk);
1539}
1540
1541static int netlink_setsockopt(struct socket *sock, int level, int optname,
1542 char __user *optval, unsigned int optlen)
1543{
1544 struct sock *sk = sock->sk;
1545 struct netlink_sock *nlk = nlk_sk(sk);
1546 unsigned int val = 0;
1547 int err;
1548
1549 if (level != SOL_NETLINK)
1550 return -ENOPROTOOPT;
1551
1552 if (optlen >= sizeof(int) &&
1553 get_user(val, (unsigned int __user *)optval))
1554 return -EFAULT;
1555
1556 switch (optname) {
1557 case NETLINK_PKTINFO:
1558 if (val)
1559 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1560 else
1561 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1562 err = 0;
1563 break;
1564 case NETLINK_ADD_MEMBERSHIP:
1565 case NETLINK_DROP_MEMBERSHIP: {
1566 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1567 return -EPERM;
1568 err = netlink_realloc_groups(sk);
1569 if (err)
1570 return err;
1571 if (!val || val - 1 >= nlk->ngroups)
1572 return -EINVAL;
1573 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1574 err = nlk->netlink_bind(sock_net(sk), val);
1575 if (err)
1576 return err;
1577 }
1578 netlink_table_grab();
1579 netlink_update_socket_mc(nlk, val,
1580 optname == NETLINK_ADD_MEMBERSHIP);
1581 netlink_table_ungrab();
1582 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1583 nlk->netlink_unbind(sock_net(sk), val);
1584
1585 err = 0;
1586 break;
1587 }
1588 case NETLINK_BROADCAST_ERROR:
1589 if (val)
1590 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1591 else
1592 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1593 err = 0;
1594 break;
1595 case NETLINK_NO_ENOBUFS:
1596 if (val) {
1597 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1598 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1599 wake_up_interruptible(&nlk->wait);
1600 } else {
1601 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1602 }
1603 err = 0;
1604 break;
1605 case NETLINK_LISTEN_ALL_NSID:
1606 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1607 return -EPERM;
1608
1609 if (val)
1610 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1611 else
1612 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1613 err = 0;
1614 break;
1615 case NETLINK_CAP_ACK:
1616 if (val)
1617 nlk->flags |= NETLINK_F_CAP_ACK;
1618 else
1619 nlk->flags &= ~NETLINK_F_CAP_ACK;
1620 err = 0;
1621 break;
1622 default:
1623 err = -ENOPROTOOPT;
1624 }
1625 return err;
1626}
1627
1628static int netlink_getsockopt(struct socket *sock, int level, int optname,
1629 char __user *optval, int __user *optlen)
1630{
1631 struct sock *sk = sock->sk;
1632 struct netlink_sock *nlk = nlk_sk(sk);
1633 int len, val, err;
1634
1635 if (level != SOL_NETLINK)
1636 return -ENOPROTOOPT;
1637
1638 if (get_user(len, optlen))
1639 return -EFAULT;
1640 if (len < 0)
1641 return -EINVAL;
1642
1643 switch (optname) {
1644 case NETLINK_PKTINFO:
1645 if (len < sizeof(int))
1646 return -EINVAL;
1647 len = sizeof(int);
1648 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1649 if (put_user(len, optlen) ||
1650 put_user(val, optval))
1651 return -EFAULT;
1652 err = 0;
1653 break;
1654 case NETLINK_BROADCAST_ERROR:
1655 if (len < sizeof(int))
1656 return -EINVAL;
1657 len = sizeof(int);
1658 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1659 if (put_user(len, optlen) ||
1660 put_user(val, optval))
1661 return -EFAULT;
1662 err = 0;
1663 break;
1664 case NETLINK_NO_ENOBUFS:
1665 if (len < sizeof(int))
1666 return -EINVAL;
1667 len = sizeof(int);
1668 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1669 if (put_user(len, optlen) ||
1670 put_user(val, optval))
1671 return -EFAULT;
1672 err = 0;
1673 break;
1674 case NETLINK_LIST_MEMBERSHIPS: {
1675 int pos, idx, shift;
1676
1677 err = 0;
1678 netlink_lock_table();
1679 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1680 if (len - pos < sizeof(u32))
1681 break;
1682
1683 idx = pos / sizeof(unsigned long);
1684 shift = (pos % sizeof(unsigned long)) * 8;
1685 if (put_user((u32)(nlk->groups[idx] >> shift),
1686 (u32 __user *)(optval + pos))) {
1687 err = -EFAULT;
1688 break;
1689 }
1690 }
1691 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1692 err = -EFAULT;
1693 netlink_unlock_table();
1694 break;
1695 }
1696 case NETLINK_CAP_ACK:
1697 if (len < sizeof(int))
1698 return -EINVAL;
1699 len = sizeof(int);
1700 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1701 if (put_user(len, optlen) ||
1702 put_user(val, optval))
1703 return -EFAULT;
1704 err = 0;
1705 break;
1706 default:
1707 err = -ENOPROTOOPT;
1708 }
1709 return err;
1710}
1711
1712static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1713{
1714 struct nl_pktinfo info;
1715
1716 info.group = NETLINK_CB(skb).dst_group;
1717 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1718}
1719
1720static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1721 struct sk_buff *skb)
1722{
1723 if (!NETLINK_CB(skb).nsid_is_set)
1724 return;
1725
1726 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1727 &NETLINK_CB(skb).nsid);
1728}
1729
1730static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1731{
1732 struct sock *sk = sock->sk;
1733 struct netlink_sock *nlk = nlk_sk(sk);
1734 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1735 u32 dst_portid;
1736 u32 dst_group;
1737 struct sk_buff *skb;
1738 int err;
1739 struct scm_cookie scm;
1740 u32 netlink_skb_flags = 0;
1741
1742 if (msg->msg_flags&MSG_OOB)
1743 return -EOPNOTSUPP;
1744
1745 err = scm_send(sock, msg, &scm, true);
1746 if (err < 0)
1747 return err;
1748
1749 if (msg->msg_namelen) {
1750 err = -EINVAL;
1751 if (addr->nl_family != AF_NETLINK)
1752 goto out;
1753 dst_portid = addr->nl_pid;
1754 dst_group = ffs(addr->nl_groups);
1755 err = -EPERM;
1756 if ((dst_group || dst_portid) &&
1757 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1758 goto out;
1759 netlink_skb_flags |= NETLINK_SKB_DST;
1760 } else {
1761 dst_portid = nlk->dst_portid;
1762 dst_group = nlk->dst_group;
1763 }
1764
1765 if (!nlk->bound) {
1766 err = netlink_autobind(sock);
1767 if (err)
1768 goto out;
1769 } else {
1770 /* Ensure nlk is hashed and visible. */
1771 smp_rmb();
1772 }
1773
1774 err = -EMSGSIZE;
1775 if (len > sk->sk_sndbuf - 32)
1776 goto out;
1777 err = -ENOBUFS;
1778 skb = netlink_alloc_large_skb(len, dst_group);
1779 if (skb == NULL)
1780 goto out;
1781
1782 NETLINK_CB(skb).portid = nlk->portid;
1783 NETLINK_CB(skb).dst_group = dst_group;
1784 NETLINK_CB(skb).creds = scm.creds;
1785 NETLINK_CB(skb).flags = netlink_skb_flags;
1786
1787 err = -EFAULT;
1788 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1789 kfree_skb(skb);
1790 goto out;
1791 }
1792
1793 err = security_netlink_send(sk, skb);
1794 if (err) {
1795 kfree_skb(skb);
1796 goto out;
1797 }
1798
1799 if (dst_group) {
1800 atomic_inc(&skb->users);
1801 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1802 }
1803 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
1804
1805out:
1806 scm_destroy(&scm);
1807 return err;
1808}
1809
1810static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1811 int flags)
1812{
1813 struct scm_cookie scm;
1814 struct sock *sk = sock->sk;
1815 struct netlink_sock *nlk = nlk_sk(sk);
1816 int noblock = flags&MSG_DONTWAIT;
1817 size_t copied;
1818 struct sk_buff *skb, *data_skb;
1819 int err, ret;
1820
1821 if (flags&MSG_OOB)
1822 return -EOPNOTSUPP;
1823
1824 copied = 0;
1825
1826 skb = skb_recv_datagram(sk, flags, noblock, &err);
1827 if (skb == NULL)
1828 goto out;
1829
1830 data_skb = skb;
1831
1832#ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1833 if (unlikely(skb_shinfo(skb)->frag_list)) {
1834 /*
1835 * If this skb has a frag_list, then here that means that we
1836 * will have to use the frag_list skb's data for compat tasks
1837 * and the regular skb's data for normal (non-compat) tasks.
1838 *
1839 * If we need to send the compat skb, assign it to the
1840 * 'data_skb' variable so that it will be used below for data
1841 * copying. We keep 'skb' for everything else, including
1842 * freeing both later.
1843 */
1844 if (flags & MSG_CMSG_COMPAT)
1845 data_skb = skb_shinfo(skb)->frag_list;
1846 }
1847#endif
1848
1849 /* Record the max length of recvmsg() calls for future allocations */
1850 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1851 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1852 SKB_WITH_OVERHEAD(32768));
1853
1854 copied = data_skb->len;
1855 if (len < copied) {
1856 msg->msg_flags |= MSG_TRUNC;
1857 copied = len;
1858 }
1859
1860 skb_reset_transport_header(data_skb);
1861 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1862
1863 if (msg->msg_name) {
1864 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1865 addr->nl_family = AF_NETLINK;
1866 addr->nl_pad = 0;
1867 addr->nl_pid = NETLINK_CB(skb).portid;
1868 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1869 msg->msg_namelen = sizeof(*addr);
1870 }
1871
1872 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1873 netlink_cmsg_recv_pktinfo(msg, skb);
1874 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1875 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1876
1877 memset(&scm, 0, sizeof(scm));
1878 scm.creds = *NETLINK_CREDS(skb);
1879 if (flags & MSG_TRUNC)
1880 copied = data_skb->len;
1881
1882 skb_free_datagram(sk, skb);
1883
1884 if (nlk->cb_running &&
1885 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1886 ret = netlink_dump(sk);
1887 if (ret) {
1888 sk->sk_err = -ret;
1889 sk->sk_error_report(sk);
1890 }
1891 }
1892
1893 scm_recv(sock, msg, &scm, flags);
1894out:
1895 netlink_rcv_wake(sk);
1896 return err ? : copied;
1897}
1898
1899static void netlink_data_ready(struct sock *sk)
1900{
1901 BUG();
1902}
1903
1904/*
1905 * We export these functions to other modules. They provide a
1906 * complete set of kernel non-blocking support for message
1907 * queueing.
1908 */
1909
1910struct sock *
1911__netlink_kernel_create(struct net *net, int unit, struct module *module,
1912 struct netlink_kernel_cfg *cfg)
1913{
1914 struct socket *sock;
1915 struct sock *sk;
1916 struct netlink_sock *nlk;
1917 struct listeners *listeners = NULL;
1918 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
1919 unsigned int groups;
1920
1921 BUG_ON(!nl_table);
1922
1923 if (unit < 0 || unit >= MAX_LINKS)
1924 return NULL;
1925
1926 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1927 return NULL;
1928
1929 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
1930 goto out_sock_release_nosk;
1931
1932 sk = sock->sk;
1933
1934 if (!cfg || cfg->groups < 32)
1935 groups = 32;
1936 else
1937 groups = cfg->groups;
1938
1939 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
1940 if (!listeners)
1941 goto out_sock_release;
1942
1943 sk->sk_data_ready = netlink_data_ready;
1944 if (cfg && cfg->input)
1945 nlk_sk(sk)->netlink_rcv = cfg->input;
1946
1947 if (netlink_insert(sk, 0))
1948 goto out_sock_release;
1949
1950 nlk = nlk_sk(sk);
1951 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
1952
1953 netlink_table_grab();
1954 if (!nl_table[unit].registered) {
1955 nl_table[unit].groups = groups;
1956 rcu_assign_pointer(nl_table[unit].listeners, listeners);
1957 nl_table[unit].cb_mutex = cb_mutex;
1958 nl_table[unit].module = module;
1959 if (cfg) {
1960 nl_table[unit].bind = cfg->bind;
1961 nl_table[unit].unbind = cfg->unbind;
1962 nl_table[unit].flags = cfg->flags;
1963 if (cfg->compare)
1964 nl_table[unit].compare = cfg->compare;
1965 }
1966 nl_table[unit].registered = 1;
1967 } else {
1968 kfree(listeners);
1969 nl_table[unit].registered++;
1970 }
1971 netlink_table_ungrab();
1972 return sk;
1973
1974out_sock_release:
1975 kfree(listeners);
1976 netlink_kernel_release(sk);
1977 return NULL;
1978
1979out_sock_release_nosk:
1980 sock_release(sock);
1981 return NULL;
1982}
1983EXPORT_SYMBOL(__netlink_kernel_create);
1984
1985void
1986netlink_kernel_release(struct sock *sk)
1987{
1988 if (sk == NULL || sk->sk_socket == NULL)
1989 return;
1990
1991 sock_release(sk->sk_socket);
1992}
1993EXPORT_SYMBOL(netlink_kernel_release);
1994
1995int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
1996{
1997 struct listeners *new, *old;
1998 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1999
2000 if (groups < 32)
2001 groups = 32;
2002
2003 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2004 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2005 if (!new)
2006 return -ENOMEM;
2007 old = nl_deref_protected(tbl->listeners);
2008 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2009 rcu_assign_pointer(tbl->listeners, new);
2010
2011 kfree_rcu(old, rcu);
2012 }
2013 tbl->groups = groups;
2014
2015 return 0;
2016}
2017
2018/**
2019 * netlink_change_ngroups - change number of multicast groups
2020 *
2021 * This changes the number of multicast groups that are available
2022 * on a certain netlink family. Note that it is not possible to
2023 * change the number of groups to below 32. Also note that it does
2024 * not implicitly call netlink_clear_multicast_users() when the
2025 * number of groups is reduced.
2026 *
2027 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2028 * @groups: The new number of groups.
2029 */
2030int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2031{
2032 int err;
2033
2034 netlink_table_grab();
2035 err = __netlink_change_ngroups(sk, groups);
2036 netlink_table_ungrab();
2037
2038 return err;
2039}
2040
2041void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2042{
2043 struct sock *sk;
2044 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2045
2046 sk_for_each_bound(sk, &tbl->mc_list)
2047 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2048}
2049
2050struct nlmsghdr *
2051__nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2052{
2053 struct nlmsghdr *nlh;
2054 int size = nlmsg_msg_size(len);
2055
2056 nlh = (struct nlmsghdr *)skb_put(skb, NLMSG_ALIGN(size));
2057 nlh->nlmsg_type = type;
2058 nlh->nlmsg_len = size;
2059 nlh->nlmsg_flags = flags;
2060 nlh->nlmsg_pid = portid;
2061 nlh->nlmsg_seq = seq;
2062 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2063 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2064 return nlh;
2065}
2066EXPORT_SYMBOL(__nlmsg_put);
2067
2068/*
2069 * It looks a bit ugly.
2070 * It would be better to create kernel thread.
2071 */
2072
2073static int netlink_dump(struct sock *sk)
2074{
2075 struct netlink_sock *nlk = nlk_sk(sk);
2076 struct netlink_callback *cb;
2077 struct sk_buff *skb = NULL;
2078 struct nlmsghdr *nlh;
2079 struct module *module;
2080 int len, err = -ENOBUFS;
2081 int alloc_min_size;
2082 int alloc_size;
2083
2084 mutex_lock(nlk->cb_mutex);
2085 if (!nlk->cb_running) {
2086 err = -EINVAL;
2087 goto errout_skb;
2088 }
2089
2090 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2091 goto errout_skb;
2092
2093 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2094 * required, but it makes sense to _attempt_ a 16K bytes allocation
2095 * to reduce number of system calls on dump operations, if user
2096 * ever provided a big enough buffer.
2097 */
2098 cb = &nlk->cb;
2099 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2100
2101 if (alloc_min_size < nlk->max_recvmsg_len) {
2102 alloc_size = nlk->max_recvmsg_len;
2103 skb = alloc_skb(alloc_size,
2104 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2105 __GFP_NOWARN | __GFP_NORETRY);
2106 }
2107 if (!skb) {
2108 alloc_size = alloc_min_size;
2109 skb = alloc_skb(alloc_size, GFP_KERNEL);
2110 }
2111 if (!skb)
2112 goto errout_skb;
2113
2114 /* Trim skb to allocated size. User is expected to provide buffer as
2115 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2116 * netlink_recvmsg())). dump will pack as many smaller messages as
2117 * could fit within the allocated skb. skb is typically allocated
2118 * with larger space than required (could be as much as near 2x the
2119 * requested size with align to next power of 2 approach). Allowing
2120 * dump to use the excess space makes it difficult for a user to have a
2121 * reasonable static buffer based on the expected largest dump of a
2122 * single netdev. The outcome is MSG_TRUNC error.
2123 */
2124 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2125 netlink_skb_set_owner_r(skb, sk);
2126
2127 len = cb->dump(skb, cb);
2128
2129 if (len > 0) {
2130 mutex_unlock(nlk->cb_mutex);
2131
2132 if (sk_filter(sk, skb))
2133 kfree_skb(skb);
2134 else
2135 __netlink_sendskb(sk, skb);
2136 return 0;
2137 }
2138
2139 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
2140 if (!nlh)
2141 goto errout_skb;
2142
2143 nl_dump_check_consistent(cb, nlh);
2144
2145 memcpy(nlmsg_data(nlh), &len, sizeof(len));
2146
2147 if (sk_filter(sk, skb))
2148 kfree_skb(skb);
2149 else
2150 __netlink_sendskb(sk, skb);
2151
2152 if (cb->done)
2153 cb->done(cb);
2154
2155 nlk->cb_running = false;
2156 module = cb->module;
2157 skb = cb->skb;
2158 mutex_unlock(nlk->cb_mutex);
2159 module_put(module);
2160 consume_skb(skb);
2161 return 0;
2162
2163errout_skb:
2164 mutex_unlock(nlk->cb_mutex);
2165 kfree_skb(skb);
2166 return err;
2167}
2168
2169int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2170 const struct nlmsghdr *nlh,
2171 struct netlink_dump_control *control)
2172{
2173 struct netlink_callback *cb;
2174 struct sock *sk;
2175 struct netlink_sock *nlk;
2176 int ret;
2177
2178 atomic_inc(&skb->users);
2179
2180 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2181 if (sk == NULL) {
2182 ret = -ECONNREFUSED;
2183 goto error_free;
2184 }
2185
2186 nlk = nlk_sk(sk);
2187 mutex_lock(nlk->cb_mutex);
2188 /* A dump is in progress... */
2189 if (nlk->cb_running) {
2190 ret = -EBUSY;
2191 goto error_unlock;
2192 }
2193 /* add reference of module which cb->dump belongs to */
2194 if (!try_module_get(control->module)) {
2195 ret = -EPROTONOSUPPORT;
2196 goto error_unlock;
2197 }
2198
2199 cb = &nlk->cb;
2200 memset(cb, 0, sizeof(*cb));
2201 cb->start = control->start;
2202 cb->dump = control->dump;
2203 cb->done = control->done;
2204 cb->nlh = nlh;
2205 cb->data = control->data;
2206 cb->module = control->module;
2207 cb->min_dump_alloc = control->min_dump_alloc;
2208 cb->skb = skb;
2209
2210 nlk->cb_running = true;
2211
2212 mutex_unlock(nlk->cb_mutex);
2213
2214 if (cb->start)
2215 cb->start(cb);
2216
2217 ret = netlink_dump(sk);
2218 sock_put(sk);
2219
2220 if (ret)
2221 return ret;
2222
2223 /* We successfully started a dump, by returning -EINTR we
2224 * signal not to send ACK even if it was requested.
2225 */
2226 return -EINTR;
2227
2228error_unlock:
2229 sock_put(sk);
2230 mutex_unlock(nlk->cb_mutex);
2231error_free:
2232 kfree_skb(skb);
2233 return ret;
2234}
2235EXPORT_SYMBOL(__netlink_dump_start);
2236
2237void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
2238{
2239 struct sk_buff *skb;
2240 struct nlmsghdr *rep;
2241 struct nlmsgerr *errmsg;
2242 size_t payload = sizeof(*errmsg);
2243 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2244
2245 /* Error messages get the original request appened, unless the user
2246 * requests to cap the error message.
2247 */
2248 if (!(nlk->flags & NETLINK_F_CAP_ACK) && err)
2249 payload += nlmsg_len(nlh);
2250
2251 skb = nlmsg_new(payload, GFP_KERNEL);
2252 if (!skb) {
2253 struct sock *sk;
2254
2255 sk = netlink_lookup(sock_net(in_skb->sk),
2256 in_skb->sk->sk_protocol,
2257 NETLINK_CB(in_skb).portid);
2258 if (sk) {
2259 sk->sk_err = ENOBUFS;
2260 sk->sk_error_report(sk);
2261 sock_put(sk);
2262 }
2263 return;
2264 }
2265
2266 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2267 NLMSG_ERROR, payload, 0);
2268 errmsg = nlmsg_data(rep);
2269 errmsg->error = err;
2270 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2271 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2272}
2273EXPORT_SYMBOL(netlink_ack);
2274
2275int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2276 struct nlmsghdr *))
2277{
2278 struct nlmsghdr *nlh;
2279 int err;
2280
2281 while (skb->len >= nlmsg_total_size(0)) {
2282 int msglen;
2283
2284 nlh = nlmsg_hdr(skb);
2285 err = 0;
2286
2287 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2288 return 0;
2289
2290 /* Only requests are handled by the kernel */
2291 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2292 goto ack;
2293
2294 /* Skip control messages */
2295 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2296 goto ack;
2297
2298 err = cb(skb, nlh);
2299 if (err == -EINTR)
2300 goto skip;
2301
2302ack:
2303 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2304 netlink_ack(skb, nlh, err);
2305
2306skip:
2307 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2308 if (msglen > skb->len)
2309 msglen = skb->len;
2310 skb_pull(skb, msglen);
2311 }
2312
2313 return 0;
2314}
2315EXPORT_SYMBOL(netlink_rcv_skb);
2316
2317/**
2318 * nlmsg_notify - send a notification netlink message
2319 * @sk: netlink socket to use
2320 * @skb: notification message
2321 * @portid: destination netlink portid for reports or 0
2322 * @group: destination multicast group or 0
2323 * @report: 1 to report back, 0 to disable
2324 * @flags: allocation flags
2325 */
2326int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2327 unsigned int group, int report, gfp_t flags)
2328{
2329 int err = 0;
2330
2331 if (group) {
2332 int exclude_portid = 0;
2333
2334 if (report) {
2335 atomic_inc(&skb->users);
2336 exclude_portid = portid;
2337 }
2338
2339 /* errors reported via destination sk->sk_err, but propagate
2340 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2341 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2342 }
2343
2344 if (report) {
2345 int err2;
2346
2347 err2 = nlmsg_unicast(sk, skb, portid);
2348 if (!err || err == -ESRCH)
2349 err = err2;
2350 }
2351
2352 return err;
2353}
2354EXPORT_SYMBOL(nlmsg_notify);
2355
2356#ifdef CONFIG_PROC_FS
2357struct nl_seq_iter {
2358 struct seq_net_private p;
2359 struct rhashtable_iter hti;
2360 int link;
2361};
2362
2363static int netlink_walk_start(struct nl_seq_iter *iter)
2364{
2365 int err;
2366
2367 err = rhashtable_walk_init(&nl_table[iter->link].hash, &iter->hti,
2368 GFP_KERNEL);
2369 if (err) {
2370 iter->link = MAX_LINKS;
2371 return err;
2372 }
2373
2374 err = rhashtable_walk_start(&iter->hti);
2375 return err == -EAGAIN ? 0 : err;
2376}
2377
2378static void netlink_walk_stop(struct nl_seq_iter *iter)
2379{
2380 rhashtable_walk_stop(&iter->hti);
2381 rhashtable_walk_exit(&iter->hti);
2382}
2383
2384static void *__netlink_seq_next(struct seq_file *seq)
2385{
2386 struct nl_seq_iter *iter = seq->private;
2387 struct netlink_sock *nlk;
2388
2389 do {
2390 for (;;) {
2391 int err;
2392
2393 nlk = rhashtable_walk_next(&iter->hti);
2394
2395 if (IS_ERR(nlk)) {
2396 if (PTR_ERR(nlk) == -EAGAIN)
2397 continue;
2398
2399 return nlk;
2400 }
2401
2402 if (nlk)
2403 break;
2404
2405 netlink_walk_stop(iter);
2406 if (++iter->link >= MAX_LINKS)
2407 return NULL;
2408
2409 err = netlink_walk_start(iter);
2410 if (err)
2411 return ERR_PTR(err);
2412 }
2413 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2414
2415 return nlk;
2416}
2417
2418static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2419{
2420 struct nl_seq_iter *iter = seq->private;
2421 void *obj = SEQ_START_TOKEN;
2422 loff_t pos;
2423 int err;
2424
2425 iter->link = 0;
2426
2427 err = netlink_walk_start(iter);
2428 if (err)
2429 return ERR_PTR(err);
2430
2431 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2432 obj = __netlink_seq_next(seq);
2433
2434 return obj;
2435}
2436
2437static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2438{
2439 ++*pos;
2440 return __netlink_seq_next(seq);
2441}
2442
2443static void netlink_seq_stop(struct seq_file *seq, void *v)
2444{
2445 struct nl_seq_iter *iter = seq->private;
2446
2447 if (iter->link >= MAX_LINKS)
2448 return;
2449
2450 netlink_walk_stop(iter);
2451}
2452
2453
2454static int netlink_seq_show(struct seq_file *seq, void *v)
2455{
2456 if (v == SEQ_START_TOKEN) {
2457 seq_puts(seq,
2458 "sk Eth Pid Groups "
2459 "Rmem Wmem Dump Locks Drops Inode\n");
2460 } else {
2461 struct sock *s = v;
2462 struct netlink_sock *nlk = nlk_sk(s);
2463
2464 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
2465 s,
2466 s->sk_protocol,
2467 nlk->portid,
2468 nlk->groups ? (u32)nlk->groups[0] : 0,
2469 sk_rmem_alloc_get(s),
2470 sk_wmem_alloc_get(s),
2471 nlk->cb_running,
2472 atomic_read(&s->sk_refcnt),
2473 atomic_read(&s->sk_drops),
2474 sock_i_ino(s)
2475 );
2476
2477 }
2478 return 0;
2479}
2480
2481static const struct seq_operations netlink_seq_ops = {
2482 .start = netlink_seq_start,
2483 .next = netlink_seq_next,
2484 .stop = netlink_seq_stop,
2485 .show = netlink_seq_show,
2486};
2487
2488
2489static int netlink_seq_open(struct inode *inode, struct file *file)
2490{
2491 return seq_open_net(inode, file, &netlink_seq_ops,
2492 sizeof(struct nl_seq_iter));
2493}
2494
2495static const struct file_operations netlink_seq_fops = {
2496 .owner = THIS_MODULE,
2497 .open = netlink_seq_open,
2498 .read = seq_read,
2499 .llseek = seq_lseek,
2500 .release = seq_release_net,
2501};
2502
2503#endif
2504
2505int netlink_register_notifier(struct notifier_block *nb)
2506{
2507 return blocking_notifier_chain_register(&netlink_chain, nb);
2508}
2509EXPORT_SYMBOL(netlink_register_notifier);
2510
2511int netlink_unregister_notifier(struct notifier_block *nb)
2512{
2513 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2514}
2515EXPORT_SYMBOL(netlink_unregister_notifier);
2516
2517static const struct proto_ops netlink_ops = {
2518 .family = PF_NETLINK,
2519 .owner = THIS_MODULE,
2520 .release = netlink_release,
2521 .bind = netlink_bind,
2522 .connect = netlink_connect,
2523 .socketpair = sock_no_socketpair,
2524 .accept = sock_no_accept,
2525 .getname = netlink_getname,
2526 .poll = datagram_poll,
2527 .ioctl = netlink_ioctl,
2528 .listen = sock_no_listen,
2529 .shutdown = sock_no_shutdown,
2530 .setsockopt = netlink_setsockopt,
2531 .getsockopt = netlink_getsockopt,
2532 .sendmsg = netlink_sendmsg,
2533 .recvmsg = netlink_recvmsg,
2534 .mmap = sock_no_mmap,
2535 .sendpage = sock_no_sendpage,
2536};
2537
2538static const struct net_proto_family netlink_family_ops = {
2539 .family = PF_NETLINK,
2540 .create = netlink_create,
2541 .owner = THIS_MODULE, /* for consistency 8) */
2542};
2543
2544static int __net_init netlink_net_init(struct net *net)
2545{
2546#ifdef CONFIG_PROC_FS
2547 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
2548 return -ENOMEM;
2549#endif
2550 return 0;
2551}
2552
2553static void __net_exit netlink_net_exit(struct net *net)
2554{
2555#ifdef CONFIG_PROC_FS
2556 remove_proc_entry("netlink", net->proc_net);
2557#endif
2558}
2559
2560static void __init netlink_add_usersock_entry(void)
2561{
2562 struct listeners *listeners;
2563 int groups = 32;
2564
2565 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2566 if (!listeners)
2567 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2568
2569 netlink_table_grab();
2570
2571 nl_table[NETLINK_USERSOCK].groups = groups;
2572 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2573 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2574 nl_table[NETLINK_USERSOCK].registered = 1;
2575 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2576
2577 netlink_table_ungrab();
2578}
2579
2580static struct pernet_operations __net_initdata netlink_net_ops = {
2581 .init = netlink_net_init,
2582 .exit = netlink_net_exit,
2583};
2584
2585static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2586{
2587 const struct netlink_sock *nlk = data;
2588 struct netlink_compare_arg arg;
2589
2590 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2591 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2592}
2593
2594static const struct rhashtable_params netlink_rhashtable_params = {
2595 .head_offset = offsetof(struct netlink_sock, node),
2596 .key_len = netlink_compare_arg_len,
2597 .obj_hashfn = netlink_hash,
2598 .obj_cmpfn = netlink_compare,
2599 .automatic_shrinking = true,
2600};
2601
2602static int __init netlink_proto_init(void)
2603{
2604 int i;
2605 int err = proto_register(&netlink_proto, 0);
2606
2607 if (err != 0)
2608 goto out;
2609
2610 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2611
2612 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2613 if (!nl_table)
2614 goto panic;
2615
2616 for (i = 0; i < MAX_LINKS; i++) {
2617 if (rhashtable_init(&nl_table[i].hash,
2618 &netlink_rhashtable_params) < 0) {
2619 while (--i > 0)
2620 rhashtable_destroy(&nl_table[i].hash);
2621 kfree(nl_table);
2622 goto panic;
2623 }
2624 }
2625
2626 INIT_LIST_HEAD(&netlink_tap_all);
2627
2628 netlink_add_usersock_entry();
2629
2630 sock_register(&netlink_family_ops);
2631 register_pernet_subsys(&netlink_net_ops);
2632 /* The netlink device handler may be needed early. */
2633 rtnetlink_init();
2634out:
2635 return err;
2636panic:
2637 panic("netlink_init: Cannot allocate nl_table\n");
2638}
2639
2640core_initcall(netlink_proto_init);