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 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
11 *
12 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
13 * added netlink_proto_exit
14 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
15 * use nlk_sk, as sk->protinfo is on a diet 8)
16 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
17 * - inc module use count of module that owns
18 * the kernel socket in case userspace opens
19 * socket of same protocol
20 * - remove all module support, since netlink is
21 * mandatory if CONFIG_NET=y these days
22 */
23
24#include <linux/module.h>
25
26#include <linux/capability.h>
27#include <linux/kernel.h>
28#include <linux/init.h>
29#include <linux/signal.h>
30#include <linux/sched.h>
31#include <linux/errno.h>
32#include <linux/string.h>
33#include <linux/stat.h>
34#include <linux/socket.h>
35#include <linux/un.h>
36#include <linux/fcntl.h>
37#include <linux/termios.h>
38#include <linux/sockios.h>
39#include <linux/net.h>
40#include <linux/fs.h>
41#include <linux/slab.h>
42#include <asm/uaccess.h>
43#include <linux/skbuff.h>
44#include <linux/netdevice.h>
45#include <linux/rtnetlink.h>
46#include <linux/proc_fs.h>
47#include <linux/seq_file.h>
48#include <linux/notifier.h>
49#include <linux/security.h>
50#include <linux/jhash.h>
51#include <linux/jiffies.h>
52#include <linux/random.h>
53#include <linux/bitops.h>
54#include <linux/mm.h>
55#include <linux/types.h>
56#include <linux/audit.h>
57#include <linux/mutex.h>
58
59#include <net/net_namespace.h>
60#include <net/sock.h>
61#include <net/scm.h>
62#include <net/netlink.h>
63
64#define NLGRPSZ(x) (ALIGN(x, sizeof(unsigned long) * 8) / 8)
65#define NLGRPLONGS(x) (NLGRPSZ(x)/sizeof(unsigned long))
66
67struct netlink_sock {
68 /* struct sock has to be the first member of netlink_sock */
69 struct sock sk;
70 u32 pid;
71 u32 dst_pid;
72 u32 dst_group;
73 u32 flags;
74 u32 subscriptions;
75 u32 ngroups;
76 unsigned long *groups;
77 unsigned long state;
78 wait_queue_head_t wait;
79 struct netlink_callback *cb;
80 struct mutex *cb_mutex;
81 struct mutex cb_def_mutex;
82 void (*netlink_rcv)(struct sk_buff *skb);
83 struct module *module;
84};
85
86struct listeners {
87 struct rcu_head rcu;
88 unsigned long masks[0];
89};
90
91#define NETLINK_KERNEL_SOCKET 0x1
92#define NETLINK_RECV_PKTINFO 0x2
93#define NETLINK_BROADCAST_SEND_ERROR 0x4
94#define NETLINK_RECV_NO_ENOBUFS 0x8
95
96static inline struct netlink_sock *nlk_sk(struct sock *sk)
97{
98 return container_of(sk, struct netlink_sock, sk);
99}
100
101static inline int netlink_is_kernel(struct sock *sk)
102{
103 return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET;
104}
105
106struct nl_pid_hash {
107 struct hlist_head *table;
108 unsigned long rehash_time;
109
110 unsigned int mask;
111 unsigned int shift;
112
113 unsigned int entries;
114 unsigned int max_shift;
115
116 u32 rnd;
117};
118
119struct netlink_table {
120 struct nl_pid_hash hash;
121 struct hlist_head mc_list;
122 struct listeners __rcu *listeners;
123 unsigned int nl_nonroot;
124 unsigned int groups;
125 struct mutex *cb_mutex;
126 struct module *module;
127 int registered;
128};
129
130static struct netlink_table *nl_table;
131
132static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
133
134static int netlink_dump(struct sock *sk);
135
136static DEFINE_RWLOCK(nl_table_lock);
137static atomic_t nl_table_users = ATOMIC_INIT(0);
138
139static ATOMIC_NOTIFIER_HEAD(netlink_chain);
140
141static inline u32 netlink_group_mask(u32 group)
142{
143 return group ? 1 << (group - 1) : 0;
144}
145
146static inline struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
147{
148 return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
149}
150
151static void netlink_destroy_callback(struct netlink_callback *cb)
152{
153 kfree_skb(cb->skb);
154 kfree(cb);
155}
156
157static void netlink_consume_callback(struct netlink_callback *cb)
158{
159 consume_skb(cb->skb);
160 kfree(cb);
161}
162
163static void netlink_sock_destruct(struct sock *sk)
164{
165 struct netlink_sock *nlk = nlk_sk(sk);
166
167 if (nlk->cb) {
168 if (nlk->cb->done)
169 nlk->cb->done(nlk->cb);
170 netlink_destroy_callback(nlk->cb);
171 }
172
173 skb_queue_purge(&sk->sk_receive_queue);
174
175 if (!sock_flag(sk, SOCK_DEAD)) {
176 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
177 return;
178 }
179
180 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
181 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
182 WARN_ON(nlk_sk(sk)->groups);
183}
184
185/* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
186 * SMP. Look, when several writers sleep and reader wakes them up, all but one
187 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
188 * this, _but_ remember, it adds useless work on UP machines.
189 */
190
191void netlink_table_grab(void)
192 __acquires(nl_table_lock)
193{
194 might_sleep();
195
196 write_lock_irq(&nl_table_lock);
197
198 if (atomic_read(&nl_table_users)) {
199 DECLARE_WAITQUEUE(wait, current);
200
201 add_wait_queue_exclusive(&nl_table_wait, &wait);
202 for (;;) {
203 set_current_state(TASK_UNINTERRUPTIBLE);
204 if (atomic_read(&nl_table_users) == 0)
205 break;
206 write_unlock_irq(&nl_table_lock);
207 schedule();
208 write_lock_irq(&nl_table_lock);
209 }
210
211 __set_current_state(TASK_RUNNING);
212 remove_wait_queue(&nl_table_wait, &wait);
213 }
214}
215
216void netlink_table_ungrab(void)
217 __releases(nl_table_lock)
218{
219 write_unlock_irq(&nl_table_lock);
220 wake_up(&nl_table_wait);
221}
222
223static inline void
224netlink_lock_table(void)
225{
226 /* read_lock() synchronizes us to netlink_table_grab */
227
228 read_lock(&nl_table_lock);
229 atomic_inc(&nl_table_users);
230 read_unlock(&nl_table_lock);
231}
232
233static inline void
234netlink_unlock_table(void)
235{
236 if (atomic_dec_and_test(&nl_table_users))
237 wake_up(&nl_table_wait);
238}
239
240static struct sock *netlink_lookup(struct net *net, int protocol, u32 pid)
241{
242 struct nl_pid_hash *hash = &nl_table[protocol].hash;
243 struct hlist_head *head;
244 struct sock *sk;
245 struct hlist_node *node;
246
247 read_lock(&nl_table_lock);
248 head = nl_pid_hashfn(hash, pid);
249 sk_for_each(sk, node, head) {
250 if (net_eq(sock_net(sk), net) && (nlk_sk(sk)->pid == pid)) {
251 sock_hold(sk);
252 goto found;
253 }
254 }
255 sk = NULL;
256found:
257 read_unlock(&nl_table_lock);
258 return sk;
259}
260
261static struct hlist_head *nl_pid_hash_zalloc(size_t size)
262{
263 if (size <= PAGE_SIZE)
264 return kzalloc(size, GFP_ATOMIC);
265 else
266 return (struct hlist_head *)
267 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
268 get_order(size));
269}
270
271static void nl_pid_hash_free(struct hlist_head *table, size_t size)
272{
273 if (size <= PAGE_SIZE)
274 kfree(table);
275 else
276 free_pages((unsigned long)table, get_order(size));
277}
278
279static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
280{
281 unsigned int omask, mask, shift;
282 size_t osize, size;
283 struct hlist_head *otable, *table;
284 int i;
285
286 omask = mask = hash->mask;
287 osize = size = (mask + 1) * sizeof(*table);
288 shift = hash->shift;
289
290 if (grow) {
291 if (++shift > hash->max_shift)
292 return 0;
293 mask = mask * 2 + 1;
294 size *= 2;
295 }
296
297 table = nl_pid_hash_zalloc(size);
298 if (!table)
299 return 0;
300
301 otable = hash->table;
302 hash->table = table;
303 hash->mask = mask;
304 hash->shift = shift;
305 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
306
307 for (i = 0; i <= omask; i++) {
308 struct sock *sk;
309 struct hlist_node *node, *tmp;
310
311 sk_for_each_safe(sk, node, tmp, &otable[i])
312 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
313 }
314
315 nl_pid_hash_free(otable, osize);
316 hash->rehash_time = jiffies + 10 * 60 * HZ;
317 return 1;
318}
319
320static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
321{
322 int avg = hash->entries >> hash->shift;
323
324 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
325 return 1;
326
327 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
328 nl_pid_hash_rehash(hash, 0);
329 return 1;
330 }
331
332 return 0;
333}
334
335static const struct proto_ops netlink_ops;
336
337static void
338netlink_update_listeners(struct sock *sk)
339{
340 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
341 struct hlist_node *node;
342 unsigned long mask;
343 unsigned int i;
344
345 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
346 mask = 0;
347 sk_for_each_bound(sk, node, &tbl->mc_list) {
348 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
349 mask |= nlk_sk(sk)->groups[i];
350 }
351 tbl->listeners->masks[i] = mask;
352 }
353 /* this function is only called with the netlink table "grabbed", which
354 * makes sure updates are visible before bind or setsockopt return. */
355}
356
357static int netlink_insert(struct sock *sk, struct net *net, u32 pid)
358{
359 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
360 struct hlist_head *head;
361 int err = -EADDRINUSE;
362 struct sock *osk;
363 struct hlist_node *node;
364 int len;
365
366 netlink_table_grab();
367 head = nl_pid_hashfn(hash, pid);
368 len = 0;
369 sk_for_each(osk, node, head) {
370 if (net_eq(sock_net(osk), net) && (nlk_sk(osk)->pid == pid))
371 break;
372 len++;
373 }
374 if (node)
375 goto err;
376
377 err = -EBUSY;
378 if (nlk_sk(sk)->pid)
379 goto err;
380
381 err = -ENOMEM;
382 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
383 goto err;
384
385 if (len && nl_pid_hash_dilute(hash, len))
386 head = nl_pid_hashfn(hash, pid);
387 hash->entries++;
388 nlk_sk(sk)->pid = pid;
389 sk_add_node(sk, head);
390 err = 0;
391
392err:
393 netlink_table_ungrab();
394 return err;
395}
396
397static void netlink_remove(struct sock *sk)
398{
399 netlink_table_grab();
400 if (sk_del_node_init(sk))
401 nl_table[sk->sk_protocol].hash.entries--;
402 if (nlk_sk(sk)->subscriptions)
403 __sk_del_bind_node(sk);
404 netlink_table_ungrab();
405}
406
407static struct proto netlink_proto = {
408 .name = "NETLINK",
409 .owner = THIS_MODULE,
410 .obj_size = sizeof(struct netlink_sock),
411};
412
413static int __netlink_create(struct net *net, struct socket *sock,
414 struct mutex *cb_mutex, int protocol)
415{
416 struct sock *sk;
417 struct netlink_sock *nlk;
418
419 sock->ops = &netlink_ops;
420
421 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
422 if (!sk)
423 return -ENOMEM;
424
425 sock_init_data(sock, sk);
426
427 nlk = nlk_sk(sk);
428 if (cb_mutex) {
429 nlk->cb_mutex = cb_mutex;
430 } else {
431 nlk->cb_mutex = &nlk->cb_def_mutex;
432 mutex_init(nlk->cb_mutex);
433 }
434 init_waitqueue_head(&nlk->wait);
435
436 sk->sk_destruct = netlink_sock_destruct;
437 sk->sk_protocol = protocol;
438 return 0;
439}
440
441static int netlink_create(struct net *net, struct socket *sock, int protocol,
442 int kern)
443{
444 struct module *module = NULL;
445 struct mutex *cb_mutex;
446 struct netlink_sock *nlk;
447 int err = 0;
448
449 sock->state = SS_UNCONNECTED;
450
451 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
452 return -ESOCKTNOSUPPORT;
453
454 if (protocol < 0 || protocol >= MAX_LINKS)
455 return -EPROTONOSUPPORT;
456
457 netlink_lock_table();
458#ifdef CONFIG_MODULES
459 if (!nl_table[protocol].registered) {
460 netlink_unlock_table();
461 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
462 netlink_lock_table();
463 }
464#endif
465 if (nl_table[protocol].registered &&
466 try_module_get(nl_table[protocol].module))
467 module = nl_table[protocol].module;
468 else
469 err = -EPROTONOSUPPORT;
470 cb_mutex = nl_table[protocol].cb_mutex;
471 netlink_unlock_table();
472
473 if (err < 0)
474 goto out;
475
476 err = __netlink_create(net, sock, cb_mutex, protocol);
477 if (err < 0)
478 goto out_module;
479
480 local_bh_disable();
481 sock_prot_inuse_add(net, &netlink_proto, 1);
482 local_bh_enable();
483
484 nlk = nlk_sk(sock->sk);
485 nlk->module = module;
486out:
487 return err;
488
489out_module:
490 module_put(module);
491 goto out;
492}
493
494static int netlink_release(struct socket *sock)
495{
496 struct sock *sk = sock->sk;
497 struct netlink_sock *nlk;
498
499 if (!sk)
500 return 0;
501
502 netlink_remove(sk);
503 sock_orphan(sk);
504 nlk = nlk_sk(sk);
505
506 /*
507 * OK. Socket is unlinked, any packets that arrive now
508 * will be purged.
509 */
510
511 sock->sk = NULL;
512 wake_up_interruptible_all(&nlk->wait);
513
514 skb_queue_purge(&sk->sk_write_queue);
515
516 if (nlk->pid) {
517 struct netlink_notify n = {
518 .net = sock_net(sk),
519 .protocol = sk->sk_protocol,
520 .pid = nlk->pid,
521 };
522 atomic_notifier_call_chain(&netlink_chain,
523 NETLINK_URELEASE, &n);
524 }
525
526 module_put(nlk->module);
527
528 netlink_table_grab();
529 if (netlink_is_kernel(sk)) {
530 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
531 if (--nl_table[sk->sk_protocol].registered == 0) {
532 kfree(nl_table[sk->sk_protocol].listeners);
533 nl_table[sk->sk_protocol].module = NULL;
534 nl_table[sk->sk_protocol].registered = 0;
535 }
536 } else if (nlk->subscriptions) {
537 netlink_update_listeners(sk);
538 }
539 netlink_table_ungrab();
540
541 kfree(nlk->groups);
542 nlk->groups = NULL;
543
544 local_bh_disable();
545 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
546 local_bh_enable();
547 sock_put(sk);
548 return 0;
549}
550
551static int netlink_autobind(struct socket *sock)
552{
553 struct sock *sk = sock->sk;
554 struct net *net = sock_net(sk);
555 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
556 struct hlist_head *head;
557 struct sock *osk;
558 struct hlist_node *node;
559 s32 pid = task_tgid_vnr(current);
560 int err;
561 static s32 rover = -4097;
562
563retry:
564 cond_resched();
565 netlink_table_grab();
566 head = nl_pid_hashfn(hash, pid);
567 sk_for_each(osk, node, head) {
568 if (!net_eq(sock_net(osk), net))
569 continue;
570 if (nlk_sk(osk)->pid == pid) {
571 /* Bind collision, search negative pid values. */
572 pid = rover--;
573 if (rover > -4097)
574 rover = -4097;
575 netlink_table_ungrab();
576 goto retry;
577 }
578 }
579 netlink_table_ungrab();
580
581 err = netlink_insert(sk, net, pid);
582 if (err == -EADDRINUSE)
583 goto retry;
584
585 /* If 2 threads race to autobind, that is fine. */
586 if (err == -EBUSY)
587 err = 0;
588
589 return err;
590}
591
592static inline int netlink_capable(const struct socket *sock, unsigned int flag)
593{
594 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
595 capable(CAP_NET_ADMIN);
596}
597
598static void
599netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
600{
601 struct netlink_sock *nlk = nlk_sk(sk);
602
603 if (nlk->subscriptions && !subscriptions)
604 __sk_del_bind_node(sk);
605 else if (!nlk->subscriptions && subscriptions)
606 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
607 nlk->subscriptions = subscriptions;
608}
609
610static int netlink_realloc_groups(struct sock *sk)
611{
612 struct netlink_sock *nlk = nlk_sk(sk);
613 unsigned int groups;
614 unsigned long *new_groups;
615 int err = 0;
616
617 netlink_table_grab();
618
619 groups = nl_table[sk->sk_protocol].groups;
620 if (!nl_table[sk->sk_protocol].registered) {
621 err = -ENOENT;
622 goto out_unlock;
623 }
624
625 if (nlk->ngroups >= groups)
626 goto out_unlock;
627
628 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
629 if (new_groups == NULL) {
630 err = -ENOMEM;
631 goto out_unlock;
632 }
633 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
634 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
635
636 nlk->groups = new_groups;
637 nlk->ngroups = groups;
638 out_unlock:
639 netlink_table_ungrab();
640 return err;
641}
642
643static int netlink_bind(struct socket *sock, struct sockaddr *addr,
644 int addr_len)
645{
646 struct sock *sk = sock->sk;
647 struct net *net = sock_net(sk);
648 struct netlink_sock *nlk = nlk_sk(sk);
649 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
650 int err;
651
652 if (nladdr->nl_family != AF_NETLINK)
653 return -EINVAL;
654
655 /* Only superuser is allowed to listen multicasts */
656 if (nladdr->nl_groups) {
657 if (!netlink_capable(sock, NL_NONROOT_RECV))
658 return -EPERM;
659 err = netlink_realloc_groups(sk);
660 if (err)
661 return err;
662 }
663
664 if (nlk->pid) {
665 if (nladdr->nl_pid != nlk->pid)
666 return -EINVAL;
667 } else {
668 err = nladdr->nl_pid ?
669 netlink_insert(sk, net, nladdr->nl_pid) :
670 netlink_autobind(sock);
671 if (err)
672 return err;
673 }
674
675 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
676 return 0;
677
678 netlink_table_grab();
679 netlink_update_subscriptions(sk, nlk->subscriptions +
680 hweight32(nladdr->nl_groups) -
681 hweight32(nlk->groups[0]));
682 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
683 netlink_update_listeners(sk);
684 netlink_table_ungrab();
685
686 return 0;
687}
688
689static int netlink_connect(struct socket *sock, struct sockaddr *addr,
690 int alen, int flags)
691{
692 int err = 0;
693 struct sock *sk = sock->sk;
694 struct netlink_sock *nlk = nlk_sk(sk);
695 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
696
697 if (alen < sizeof(addr->sa_family))
698 return -EINVAL;
699
700 if (addr->sa_family == AF_UNSPEC) {
701 sk->sk_state = NETLINK_UNCONNECTED;
702 nlk->dst_pid = 0;
703 nlk->dst_group = 0;
704 return 0;
705 }
706 if (addr->sa_family != AF_NETLINK)
707 return -EINVAL;
708
709 /* Only superuser is allowed to send multicasts */
710 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
711 return -EPERM;
712
713 if (!nlk->pid)
714 err = netlink_autobind(sock);
715
716 if (err == 0) {
717 sk->sk_state = NETLINK_CONNECTED;
718 nlk->dst_pid = nladdr->nl_pid;
719 nlk->dst_group = ffs(nladdr->nl_groups);
720 }
721
722 return err;
723}
724
725static int netlink_getname(struct socket *sock, struct sockaddr *addr,
726 int *addr_len, int peer)
727{
728 struct sock *sk = sock->sk;
729 struct netlink_sock *nlk = nlk_sk(sk);
730 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
731
732 nladdr->nl_family = AF_NETLINK;
733 nladdr->nl_pad = 0;
734 *addr_len = sizeof(*nladdr);
735
736 if (peer) {
737 nladdr->nl_pid = nlk->dst_pid;
738 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
739 } else {
740 nladdr->nl_pid = nlk->pid;
741 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
742 }
743 return 0;
744}
745
746static void netlink_overrun(struct sock *sk)
747{
748 struct netlink_sock *nlk = nlk_sk(sk);
749
750 if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) {
751 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
752 sk->sk_err = ENOBUFS;
753 sk->sk_error_report(sk);
754 }
755 }
756 atomic_inc(&sk->sk_drops);
757}
758
759static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
760{
761 struct sock *sock;
762 struct netlink_sock *nlk;
763
764 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, pid);
765 if (!sock)
766 return ERR_PTR(-ECONNREFUSED);
767
768 /* Don't bother queuing skb if kernel socket has no input function */
769 nlk = nlk_sk(sock);
770 if (sock->sk_state == NETLINK_CONNECTED &&
771 nlk->dst_pid != nlk_sk(ssk)->pid) {
772 sock_put(sock);
773 return ERR_PTR(-ECONNREFUSED);
774 }
775 return sock;
776}
777
778struct sock *netlink_getsockbyfilp(struct file *filp)
779{
780 struct inode *inode = filp->f_path.dentry->d_inode;
781 struct sock *sock;
782
783 if (!S_ISSOCK(inode->i_mode))
784 return ERR_PTR(-ENOTSOCK);
785
786 sock = SOCKET_I(inode)->sk;
787 if (sock->sk_family != AF_NETLINK)
788 return ERR_PTR(-EINVAL);
789
790 sock_hold(sock);
791 return sock;
792}
793
794/*
795 * Attach a skb to a netlink socket.
796 * The caller must hold a reference to the destination socket. On error, the
797 * reference is dropped. The skb is not send to the destination, just all
798 * all error checks are performed and memory in the queue is reserved.
799 * Return values:
800 * < 0: error. skb freed, reference to sock dropped.
801 * 0: continue
802 * 1: repeat lookup - reference dropped while waiting for socket memory.
803 */
804int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
805 long *timeo, struct sock *ssk)
806{
807 struct netlink_sock *nlk;
808
809 nlk = nlk_sk(sk);
810
811 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
812 test_bit(0, &nlk->state)) {
813 DECLARE_WAITQUEUE(wait, current);
814 if (!*timeo) {
815 if (!ssk || netlink_is_kernel(ssk))
816 netlink_overrun(sk);
817 sock_put(sk);
818 kfree_skb(skb);
819 return -EAGAIN;
820 }
821
822 __set_current_state(TASK_INTERRUPTIBLE);
823 add_wait_queue(&nlk->wait, &wait);
824
825 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
826 test_bit(0, &nlk->state)) &&
827 !sock_flag(sk, SOCK_DEAD))
828 *timeo = schedule_timeout(*timeo);
829
830 __set_current_state(TASK_RUNNING);
831 remove_wait_queue(&nlk->wait, &wait);
832 sock_put(sk);
833
834 if (signal_pending(current)) {
835 kfree_skb(skb);
836 return sock_intr_errno(*timeo);
837 }
838 return 1;
839 }
840 skb_set_owner_r(skb, sk);
841 return 0;
842}
843
844static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
845{
846 int len = skb->len;
847
848 skb_queue_tail(&sk->sk_receive_queue, skb);
849 sk->sk_data_ready(sk, len);
850 return len;
851}
852
853int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
854{
855 int len = __netlink_sendskb(sk, skb);
856
857 sock_put(sk);
858 return len;
859}
860
861void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
862{
863 kfree_skb(skb);
864 sock_put(sk);
865}
866
867static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
868{
869 int delta;
870
871 skb_orphan(skb);
872
873 delta = skb->end - skb->tail;
874 if (delta * 2 < skb->truesize)
875 return skb;
876
877 if (skb_shared(skb)) {
878 struct sk_buff *nskb = skb_clone(skb, allocation);
879 if (!nskb)
880 return skb;
881 consume_skb(skb);
882 skb = nskb;
883 }
884
885 if (!pskb_expand_head(skb, 0, -delta, allocation))
886 skb->truesize -= delta;
887
888 return skb;
889}
890
891static void netlink_rcv_wake(struct sock *sk)
892{
893 struct netlink_sock *nlk = nlk_sk(sk);
894
895 if (skb_queue_empty(&sk->sk_receive_queue))
896 clear_bit(0, &nlk->state);
897 if (!test_bit(0, &nlk->state))
898 wake_up_interruptible(&nlk->wait);
899}
900
901static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb)
902{
903 int ret;
904 struct netlink_sock *nlk = nlk_sk(sk);
905
906 ret = -ECONNREFUSED;
907 if (nlk->netlink_rcv != NULL) {
908 ret = skb->len;
909 skb_set_owner_r(skb, sk);
910 nlk->netlink_rcv(skb);
911 consume_skb(skb);
912 } else {
913 kfree_skb(skb);
914 }
915 sock_put(sk);
916 return ret;
917}
918
919int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
920 u32 pid, int nonblock)
921{
922 struct sock *sk;
923 int err;
924 long timeo;
925
926 skb = netlink_trim(skb, gfp_any());
927
928 timeo = sock_sndtimeo(ssk, nonblock);
929retry:
930 sk = netlink_getsockbypid(ssk, pid);
931 if (IS_ERR(sk)) {
932 kfree_skb(skb);
933 return PTR_ERR(sk);
934 }
935 if (netlink_is_kernel(sk))
936 return netlink_unicast_kernel(sk, skb);
937
938 if (sk_filter(sk, skb)) {
939 err = skb->len;
940 kfree_skb(skb);
941 sock_put(sk);
942 return err;
943 }
944
945 err = netlink_attachskb(sk, skb, &timeo, ssk);
946 if (err == 1)
947 goto retry;
948 if (err)
949 return err;
950
951 return netlink_sendskb(sk, skb);
952}
953EXPORT_SYMBOL(netlink_unicast);
954
955int netlink_has_listeners(struct sock *sk, unsigned int group)
956{
957 int res = 0;
958 struct listeners *listeners;
959
960 BUG_ON(!netlink_is_kernel(sk));
961
962 rcu_read_lock();
963 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
964
965 if (group - 1 < nl_table[sk->sk_protocol].groups)
966 res = test_bit(group - 1, listeners->masks);
967
968 rcu_read_unlock();
969
970 return res;
971}
972EXPORT_SYMBOL_GPL(netlink_has_listeners);
973
974static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
975{
976 struct netlink_sock *nlk = nlk_sk(sk);
977
978 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
979 !test_bit(0, &nlk->state)) {
980 skb_set_owner_r(skb, sk);
981 __netlink_sendskb(sk, skb);
982 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
983 }
984 return -1;
985}
986
987struct netlink_broadcast_data {
988 struct sock *exclude_sk;
989 struct net *net;
990 u32 pid;
991 u32 group;
992 int failure;
993 int delivery_failure;
994 int congested;
995 int delivered;
996 gfp_t allocation;
997 struct sk_buff *skb, *skb2;
998 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
999 void *tx_data;
1000};
1001
1002static int do_one_broadcast(struct sock *sk,
1003 struct netlink_broadcast_data *p)
1004{
1005 struct netlink_sock *nlk = nlk_sk(sk);
1006 int val;
1007
1008 if (p->exclude_sk == sk)
1009 goto out;
1010
1011 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1012 !test_bit(p->group - 1, nlk->groups))
1013 goto out;
1014
1015 if (!net_eq(sock_net(sk), p->net))
1016 goto out;
1017
1018 if (p->failure) {
1019 netlink_overrun(sk);
1020 goto out;
1021 }
1022
1023 sock_hold(sk);
1024 if (p->skb2 == NULL) {
1025 if (skb_shared(p->skb)) {
1026 p->skb2 = skb_clone(p->skb, p->allocation);
1027 } else {
1028 p->skb2 = skb_get(p->skb);
1029 /*
1030 * skb ownership may have been set when
1031 * delivered to a previous socket.
1032 */
1033 skb_orphan(p->skb2);
1034 }
1035 }
1036 if (p->skb2 == NULL) {
1037 netlink_overrun(sk);
1038 /* Clone failed. Notify ALL listeners. */
1039 p->failure = 1;
1040 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1041 p->delivery_failure = 1;
1042 } else if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1043 kfree_skb(p->skb2);
1044 p->skb2 = NULL;
1045 } else if (sk_filter(sk, p->skb2)) {
1046 kfree_skb(p->skb2);
1047 p->skb2 = NULL;
1048 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1049 netlink_overrun(sk);
1050 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1051 p->delivery_failure = 1;
1052 } else {
1053 p->congested |= val;
1054 p->delivered = 1;
1055 p->skb2 = NULL;
1056 }
1057 sock_put(sk);
1058
1059out:
1060 return 0;
1061}
1062
1063int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 pid,
1064 u32 group, gfp_t allocation,
1065 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1066 void *filter_data)
1067{
1068 struct net *net = sock_net(ssk);
1069 struct netlink_broadcast_data info;
1070 struct hlist_node *node;
1071 struct sock *sk;
1072
1073 skb = netlink_trim(skb, allocation);
1074
1075 info.exclude_sk = ssk;
1076 info.net = net;
1077 info.pid = pid;
1078 info.group = group;
1079 info.failure = 0;
1080 info.delivery_failure = 0;
1081 info.congested = 0;
1082 info.delivered = 0;
1083 info.allocation = allocation;
1084 info.skb = skb;
1085 info.skb2 = NULL;
1086 info.tx_filter = filter;
1087 info.tx_data = filter_data;
1088
1089 /* While we sleep in clone, do not allow to change socket list */
1090
1091 netlink_lock_table();
1092
1093 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1094 do_one_broadcast(sk, &info);
1095
1096 consume_skb(skb);
1097
1098 netlink_unlock_table();
1099
1100 if (info.delivery_failure) {
1101 kfree_skb(info.skb2);
1102 return -ENOBUFS;
1103 }
1104 consume_skb(info.skb2);
1105
1106 if (info.delivered) {
1107 if (info.congested && (allocation & __GFP_WAIT))
1108 yield();
1109 return 0;
1110 }
1111 return -ESRCH;
1112}
1113EXPORT_SYMBOL(netlink_broadcast_filtered);
1114
1115int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
1116 u32 group, gfp_t allocation)
1117{
1118 return netlink_broadcast_filtered(ssk, skb, pid, group, allocation,
1119 NULL, NULL);
1120}
1121EXPORT_SYMBOL(netlink_broadcast);
1122
1123struct netlink_set_err_data {
1124 struct sock *exclude_sk;
1125 u32 pid;
1126 u32 group;
1127 int code;
1128};
1129
1130static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1131{
1132 struct netlink_sock *nlk = nlk_sk(sk);
1133 int ret = 0;
1134
1135 if (sk == p->exclude_sk)
1136 goto out;
1137
1138 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1139 goto out;
1140
1141 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1142 !test_bit(p->group - 1, nlk->groups))
1143 goto out;
1144
1145 if (p->code == ENOBUFS && nlk->flags & NETLINK_RECV_NO_ENOBUFS) {
1146 ret = 1;
1147 goto out;
1148 }
1149
1150 sk->sk_err = p->code;
1151 sk->sk_error_report(sk);
1152out:
1153 return ret;
1154}
1155
1156/**
1157 * netlink_set_err - report error to broadcast listeners
1158 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1159 * @pid: the PID of a process that we want to skip (if any)
1160 * @groups: the broadcast group that will notice the error
1161 * @code: error code, must be negative (as usual in kernelspace)
1162 *
1163 * This function returns the number of broadcast listeners that have set the
1164 * NETLINK_RECV_NO_ENOBUFS socket option.
1165 */
1166int netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
1167{
1168 struct netlink_set_err_data info;
1169 struct hlist_node *node;
1170 struct sock *sk;
1171 int ret = 0;
1172
1173 info.exclude_sk = ssk;
1174 info.pid = pid;
1175 info.group = group;
1176 /* sk->sk_err wants a positive error value */
1177 info.code = -code;
1178
1179 read_lock(&nl_table_lock);
1180
1181 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1182 ret += do_one_set_err(sk, &info);
1183
1184 read_unlock(&nl_table_lock);
1185 return ret;
1186}
1187EXPORT_SYMBOL(netlink_set_err);
1188
1189/* must be called with netlink table grabbed */
1190static void netlink_update_socket_mc(struct netlink_sock *nlk,
1191 unsigned int group,
1192 int is_new)
1193{
1194 int old, new = !!is_new, subscriptions;
1195
1196 old = test_bit(group - 1, nlk->groups);
1197 subscriptions = nlk->subscriptions - old + new;
1198 if (new)
1199 __set_bit(group - 1, nlk->groups);
1200 else
1201 __clear_bit(group - 1, nlk->groups);
1202 netlink_update_subscriptions(&nlk->sk, subscriptions);
1203 netlink_update_listeners(&nlk->sk);
1204}
1205
1206static int netlink_setsockopt(struct socket *sock, int level, int optname,
1207 char __user *optval, unsigned int optlen)
1208{
1209 struct sock *sk = sock->sk;
1210 struct netlink_sock *nlk = nlk_sk(sk);
1211 unsigned int val = 0;
1212 int err;
1213
1214 if (level != SOL_NETLINK)
1215 return -ENOPROTOOPT;
1216
1217 if (optlen >= sizeof(int) &&
1218 get_user(val, (unsigned int __user *)optval))
1219 return -EFAULT;
1220
1221 switch (optname) {
1222 case NETLINK_PKTINFO:
1223 if (val)
1224 nlk->flags |= NETLINK_RECV_PKTINFO;
1225 else
1226 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1227 err = 0;
1228 break;
1229 case NETLINK_ADD_MEMBERSHIP:
1230 case NETLINK_DROP_MEMBERSHIP: {
1231 if (!netlink_capable(sock, NL_NONROOT_RECV))
1232 return -EPERM;
1233 err = netlink_realloc_groups(sk);
1234 if (err)
1235 return err;
1236 if (!val || val - 1 >= nlk->ngroups)
1237 return -EINVAL;
1238 netlink_table_grab();
1239 netlink_update_socket_mc(nlk, val,
1240 optname == NETLINK_ADD_MEMBERSHIP);
1241 netlink_table_ungrab();
1242 err = 0;
1243 break;
1244 }
1245 case NETLINK_BROADCAST_ERROR:
1246 if (val)
1247 nlk->flags |= NETLINK_BROADCAST_SEND_ERROR;
1248 else
1249 nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR;
1250 err = 0;
1251 break;
1252 case NETLINK_NO_ENOBUFS:
1253 if (val) {
1254 nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
1255 clear_bit(0, &nlk->state);
1256 wake_up_interruptible(&nlk->wait);
1257 } else {
1258 nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
1259 }
1260 err = 0;
1261 break;
1262 default:
1263 err = -ENOPROTOOPT;
1264 }
1265 return err;
1266}
1267
1268static int netlink_getsockopt(struct socket *sock, int level, int optname,
1269 char __user *optval, int __user *optlen)
1270{
1271 struct sock *sk = sock->sk;
1272 struct netlink_sock *nlk = nlk_sk(sk);
1273 int len, val, err;
1274
1275 if (level != SOL_NETLINK)
1276 return -ENOPROTOOPT;
1277
1278 if (get_user(len, optlen))
1279 return -EFAULT;
1280 if (len < 0)
1281 return -EINVAL;
1282
1283 switch (optname) {
1284 case NETLINK_PKTINFO:
1285 if (len < sizeof(int))
1286 return -EINVAL;
1287 len = sizeof(int);
1288 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1289 if (put_user(len, optlen) ||
1290 put_user(val, optval))
1291 return -EFAULT;
1292 err = 0;
1293 break;
1294 case NETLINK_BROADCAST_ERROR:
1295 if (len < sizeof(int))
1296 return -EINVAL;
1297 len = sizeof(int);
1298 val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0;
1299 if (put_user(len, optlen) ||
1300 put_user(val, optval))
1301 return -EFAULT;
1302 err = 0;
1303 break;
1304 case NETLINK_NO_ENOBUFS:
1305 if (len < sizeof(int))
1306 return -EINVAL;
1307 len = sizeof(int);
1308 val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0;
1309 if (put_user(len, optlen) ||
1310 put_user(val, optval))
1311 return -EFAULT;
1312 err = 0;
1313 break;
1314 default:
1315 err = -ENOPROTOOPT;
1316 }
1317 return err;
1318}
1319
1320static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1321{
1322 struct nl_pktinfo info;
1323
1324 info.group = NETLINK_CB(skb).dst_group;
1325 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1326}
1327
1328static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1329 struct msghdr *msg, size_t len)
1330{
1331 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1332 struct sock *sk = sock->sk;
1333 struct netlink_sock *nlk = nlk_sk(sk);
1334 struct sockaddr_nl *addr = msg->msg_name;
1335 u32 dst_pid;
1336 u32 dst_group;
1337 struct sk_buff *skb;
1338 int err;
1339 struct scm_cookie scm;
1340
1341 if (msg->msg_flags&MSG_OOB)
1342 return -EOPNOTSUPP;
1343
1344 if (NULL == siocb->scm)
1345 siocb->scm = &scm;
1346
1347 err = scm_send(sock, msg, siocb->scm, true);
1348 if (err < 0)
1349 return err;
1350
1351 if (msg->msg_namelen) {
1352 err = -EINVAL;
1353 if (addr->nl_family != AF_NETLINK)
1354 goto out;
1355 dst_pid = addr->nl_pid;
1356 dst_group = ffs(addr->nl_groups);
1357 err = -EPERM;
1358 if ((dst_group || dst_pid) &&
1359 !netlink_capable(sock, NL_NONROOT_SEND))
1360 goto out;
1361 } else {
1362 dst_pid = nlk->dst_pid;
1363 dst_group = nlk->dst_group;
1364 }
1365
1366 if (!nlk->pid) {
1367 err = netlink_autobind(sock);
1368 if (err)
1369 goto out;
1370 }
1371
1372 err = -EMSGSIZE;
1373 if (len > sk->sk_sndbuf - 32)
1374 goto out;
1375 err = -ENOBUFS;
1376 skb = alloc_skb(len, GFP_KERNEL);
1377 if (skb == NULL)
1378 goto out;
1379
1380 NETLINK_CB(skb).pid = nlk->pid;
1381 NETLINK_CB(skb).dst_group = dst_group;
1382 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1383
1384 err = -EFAULT;
1385 if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
1386 kfree_skb(skb);
1387 goto out;
1388 }
1389
1390 err = security_netlink_send(sk, skb);
1391 if (err) {
1392 kfree_skb(skb);
1393 goto out;
1394 }
1395
1396 if (dst_group) {
1397 atomic_inc(&skb->users);
1398 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1399 }
1400 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1401
1402out:
1403 scm_destroy(siocb->scm);
1404 return err;
1405}
1406
1407static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1408 struct msghdr *msg, size_t len,
1409 int flags)
1410{
1411 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1412 struct scm_cookie scm;
1413 struct sock *sk = sock->sk;
1414 struct netlink_sock *nlk = nlk_sk(sk);
1415 int noblock = flags&MSG_DONTWAIT;
1416 size_t copied;
1417 struct sk_buff *skb, *data_skb;
1418 int err, ret;
1419
1420 if (flags&MSG_OOB)
1421 return -EOPNOTSUPP;
1422
1423 copied = 0;
1424
1425 skb = skb_recv_datagram(sk, flags, noblock, &err);
1426 if (skb == NULL)
1427 goto out;
1428
1429 data_skb = skb;
1430
1431#ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1432 if (unlikely(skb_shinfo(skb)->frag_list)) {
1433 /*
1434 * If this skb has a frag_list, then here that means that we
1435 * will have to use the frag_list skb's data for compat tasks
1436 * and the regular skb's data for normal (non-compat) tasks.
1437 *
1438 * If we need to send the compat skb, assign it to the
1439 * 'data_skb' variable so that it will be used below for data
1440 * copying. We keep 'skb' for everything else, including
1441 * freeing both later.
1442 */
1443 if (flags & MSG_CMSG_COMPAT)
1444 data_skb = skb_shinfo(skb)->frag_list;
1445 }
1446#endif
1447
1448 msg->msg_namelen = 0;
1449
1450 copied = data_skb->len;
1451 if (len < copied) {
1452 msg->msg_flags |= MSG_TRUNC;
1453 copied = len;
1454 }
1455
1456 skb_reset_transport_header(data_skb);
1457 err = skb_copy_datagram_iovec(data_skb, 0, msg->msg_iov, copied);
1458
1459 if (msg->msg_name) {
1460 struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
1461 addr->nl_family = AF_NETLINK;
1462 addr->nl_pad = 0;
1463 addr->nl_pid = NETLINK_CB(skb).pid;
1464 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1465 msg->msg_namelen = sizeof(*addr);
1466 }
1467
1468 if (nlk->flags & NETLINK_RECV_PKTINFO)
1469 netlink_cmsg_recv_pktinfo(msg, skb);
1470
1471 if (NULL == siocb->scm) {
1472 memset(&scm, 0, sizeof(scm));
1473 siocb->scm = &scm;
1474 }
1475 siocb->scm->creds = *NETLINK_CREDS(skb);
1476 if (flags & MSG_TRUNC)
1477 copied = data_skb->len;
1478
1479 skb_free_datagram(sk, skb);
1480
1481 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1482 ret = netlink_dump(sk);
1483 if (ret) {
1484 sk->sk_err = ret;
1485 sk->sk_error_report(sk);
1486 }
1487 }
1488
1489 scm_recv(sock, msg, siocb->scm, flags);
1490out:
1491 netlink_rcv_wake(sk);
1492 return err ? : copied;
1493}
1494
1495static void netlink_data_ready(struct sock *sk, int len)
1496{
1497 BUG();
1498}
1499
1500/*
1501 * We export these functions to other modules. They provide a
1502 * complete set of kernel non-blocking support for message
1503 * queueing.
1504 */
1505
1506struct sock *
1507netlink_kernel_create(struct net *net, int unit, unsigned int groups,
1508 void (*input)(struct sk_buff *skb),
1509 struct mutex *cb_mutex, struct module *module)
1510{
1511 struct socket *sock;
1512 struct sock *sk;
1513 struct netlink_sock *nlk;
1514 struct listeners *listeners = NULL;
1515
1516 BUG_ON(!nl_table);
1517
1518 if (unit < 0 || unit >= MAX_LINKS)
1519 return NULL;
1520
1521 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1522 return NULL;
1523
1524 /*
1525 * We have to just have a reference on the net from sk, but don't
1526 * get_net it. Besides, we cannot get and then put the net here.
1527 * So we create one inside init_net and the move it to net.
1528 */
1529
1530 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
1531 goto out_sock_release_nosk;
1532
1533 sk = sock->sk;
1534 sk_change_net(sk, net);
1535
1536 if (groups < 32)
1537 groups = 32;
1538
1539 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
1540 if (!listeners)
1541 goto out_sock_release;
1542
1543 sk->sk_data_ready = netlink_data_ready;
1544 if (input)
1545 nlk_sk(sk)->netlink_rcv = input;
1546
1547 if (netlink_insert(sk, net, 0))
1548 goto out_sock_release;
1549
1550 nlk = nlk_sk(sk);
1551 nlk->flags |= NETLINK_KERNEL_SOCKET;
1552
1553 netlink_table_grab();
1554 if (!nl_table[unit].registered) {
1555 nl_table[unit].groups = groups;
1556 rcu_assign_pointer(nl_table[unit].listeners, listeners);
1557 nl_table[unit].cb_mutex = cb_mutex;
1558 nl_table[unit].module = module;
1559 nl_table[unit].registered = 1;
1560 } else {
1561 kfree(listeners);
1562 nl_table[unit].registered++;
1563 }
1564 netlink_table_ungrab();
1565 return sk;
1566
1567out_sock_release:
1568 kfree(listeners);
1569 netlink_kernel_release(sk);
1570 return NULL;
1571
1572out_sock_release_nosk:
1573 sock_release(sock);
1574 return NULL;
1575}
1576EXPORT_SYMBOL(netlink_kernel_create);
1577
1578
1579void
1580netlink_kernel_release(struct sock *sk)
1581{
1582 sk_release_kernel(sk);
1583}
1584EXPORT_SYMBOL(netlink_kernel_release);
1585
1586int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
1587{
1588 struct listeners *new, *old;
1589 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1590
1591 if (groups < 32)
1592 groups = 32;
1593
1594 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1595 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
1596 if (!new)
1597 return -ENOMEM;
1598 old = rcu_dereference_protected(tbl->listeners, 1);
1599 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
1600 rcu_assign_pointer(tbl->listeners, new);
1601
1602 kfree_rcu(old, rcu);
1603 }
1604 tbl->groups = groups;
1605
1606 return 0;
1607}
1608
1609/**
1610 * netlink_change_ngroups - change number of multicast groups
1611 *
1612 * This changes the number of multicast groups that are available
1613 * on a certain netlink family. Note that it is not possible to
1614 * change the number of groups to below 32. Also note that it does
1615 * not implicitly call netlink_clear_multicast_users() when the
1616 * number of groups is reduced.
1617 *
1618 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1619 * @groups: The new number of groups.
1620 */
1621int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1622{
1623 int err;
1624
1625 netlink_table_grab();
1626 err = __netlink_change_ngroups(sk, groups);
1627 netlink_table_ungrab();
1628
1629 return err;
1630}
1631
1632void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1633{
1634 struct sock *sk;
1635 struct hlist_node *node;
1636 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1637
1638 sk_for_each_bound(sk, node, &tbl->mc_list)
1639 netlink_update_socket_mc(nlk_sk(sk), group, 0);
1640}
1641
1642/**
1643 * netlink_clear_multicast_users - kick off multicast listeners
1644 *
1645 * This function removes all listeners from the given group.
1646 * @ksk: The kernel netlink socket, as returned by
1647 * netlink_kernel_create().
1648 * @group: The multicast group to clear.
1649 */
1650void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1651{
1652 netlink_table_grab();
1653 __netlink_clear_multicast_users(ksk, group);
1654 netlink_table_ungrab();
1655}
1656
1657void netlink_set_nonroot(int protocol, unsigned int flags)
1658{
1659 if ((unsigned int)protocol < MAX_LINKS)
1660 nl_table[protocol].nl_nonroot = flags;
1661}
1662EXPORT_SYMBOL(netlink_set_nonroot);
1663
1664struct nlmsghdr *
1665__nlmsg_put(struct sk_buff *skb, u32 pid, u32 seq, int type, int len, int flags)
1666{
1667 struct nlmsghdr *nlh;
1668 int size = NLMSG_LENGTH(len);
1669
1670 nlh = (struct nlmsghdr*)skb_put(skb, NLMSG_ALIGN(size));
1671 nlh->nlmsg_type = type;
1672 nlh->nlmsg_len = size;
1673 nlh->nlmsg_flags = flags;
1674 nlh->nlmsg_pid = pid;
1675 nlh->nlmsg_seq = seq;
1676 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
1677 memset(NLMSG_DATA(nlh) + len, 0, NLMSG_ALIGN(size) - size);
1678 return nlh;
1679}
1680EXPORT_SYMBOL(__nlmsg_put);
1681
1682/*
1683 * It looks a bit ugly.
1684 * It would be better to create kernel thread.
1685 */
1686
1687static int netlink_dump(struct sock *sk)
1688{
1689 struct netlink_sock *nlk = nlk_sk(sk);
1690 struct netlink_callback *cb;
1691 struct sk_buff *skb = NULL;
1692 struct nlmsghdr *nlh;
1693 int len, err = -ENOBUFS;
1694 int alloc_size;
1695
1696 mutex_lock(nlk->cb_mutex);
1697
1698 cb = nlk->cb;
1699 if (cb == NULL) {
1700 err = -EINVAL;
1701 goto errout_skb;
1702 }
1703
1704 alloc_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
1705
1706 skb = sock_rmalloc(sk, alloc_size, 0, GFP_KERNEL);
1707 if (!skb)
1708 goto errout_skb;
1709
1710 len = cb->dump(skb, cb);
1711
1712 if (len > 0) {
1713 mutex_unlock(nlk->cb_mutex);
1714
1715 if (sk_filter(sk, skb))
1716 kfree_skb(skb);
1717 else
1718 __netlink_sendskb(sk, skb);
1719 return 0;
1720 }
1721
1722 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1723 if (!nlh)
1724 goto errout_skb;
1725
1726 nl_dump_check_consistent(cb, nlh);
1727
1728 memcpy(nlmsg_data(nlh), &len, sizeof(len));
1729
1730 if (sk_filter(sk, skb))
1731 kfree_skb(skb);
1732 else
1733 __netlink_sendskb(sk, skb);
1734
1735 if (cb->done)
1736 cb->done(cb);
1737 nlk->cb = NULL;
1738 mutex_unlock(nlk->cb_mutex);
1739
1740 netlink_consume_callback(cb);
1741 return 0;
1742
1743errout_skb:
1744 mutex_unlock(nlk->cb_mutex);
1745 kfree_skb(skb);
1746 return err;
1747}
1748
1749int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1750 const struct nlmsghdr *nlh,
1751 struct netlink_dump_control *control)
1752{
1753 struct netlink_callback *cb;
1754 struct sock *sk;
1755 struct netlink_sock *nlk;
1756 int ret;
1757
1758 cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1759 if (cb == NULL)
1760 return -ENOBUFS;
1761
1762 cb->dump = control->dump;
1763 cb->done = control->done;
1764 cb->nlh = nlh;
1765 cb->data = control->data;
1766 cb->min_dump_alloc = control->min_dump_alloc;
1767 atomic_inc(&skb->users);
1768 cb->skb = skb;
1769
1770 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).pid);
1771 if (sk == NULL) {
1772 netlink_destroy_callback(cb);
1773 return -ECONNREFUSED;
1774 }
1775 nlk = nlk_sk(sk);
1776 /* A dump is in progress... */
1777 mutex_lock(nlk->cb_mutex);
1778 if (nlk->cb) {
1779 mutex_unlock(nlk->cb_mutex);
1780 netlink_destroy_callback(cb);
1781 sock_put(sk);
1782 return -EBUSY;
1783 }
1784 nlk->cb = cb;
1785 mutex_unlock(nlk->cb_mutex);
1786
1787 ret = netlink_dump(sk);
1788
1789 sock_put(sk);
1790
1791 if (ret)
1792 return ret;
1793
1794 /* We successfully started a dump, by returning -EINTR we
1795 * signal not to send ACK even if it was requested.
1796 */
1797 return -EINTR;
1798}
1799EXPORT_SYMBOL(netlink_dump_start);
1800
1801void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1802{
1803 struct sk_buff *skb;
1804 struct nlmsghdr *rep;
1805 struct nlmsgerr *errmsg;
1806 size_t payload = sizeof(*errmsg);
1807
1808 /* error messages get the original request appened */
1809 if (err)
1810 payload += nlmsg_len(nlh);
1811
1812 skb = nlmsg_new(payload, GFP_KERNEL);
1813 if (!skb) {
1814 struct sock *sk;
1815
1816 sk = netlink_lookup(sock_net(in_skb->sk),
1817 in_skb->sk->sk_protocol,
1818 NETLINK_CB(in_skb).pid);
1819 if (sk) {
1820 sk->sk_err = ENOBUFS;
1821 sk->sk_error_report(sk);
1822 sock_put(sk);
1823 }
1824 return;
1825 }
1826
1827 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1828 NLMSG_ERROR, payload, 0);
1829 errmsg = nlmsg_data(rep);
1830 errmsg->error = err;
1831 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1832 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1833}
1834EXPORT_SYMBOL(netlink_ack);
1835
1836int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1837 struct nlmsghdr *))
1838{
1839 struct nlmsghdr *nlh;
1840 int err;
1841
1842 while (skb->len >= nlmsg_total_size(0)) {
1843 int msglen;
1844
1845 nlh = nlmsg_hdr(skb);
1846 err = 0;
1847
1848 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1849 return 0;
1850
1851 /* Only requests are handled by the kernel */
1852 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1853 goto ack;
1854
1855 /* Skip control messages */
1856 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1857 goto ack;
1858
1859 err = cb(skb, nlh);
1860 if (err == -EINTR)
1861 goto skip;
1862
1863ack:
1864 if (nlh->nlmsg_flags & NLM_F_ACK || err)
1865 netlink_ack(skb, nlh, err);
1866
1867skip:
1868 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1869 if (msglen > skb->len)
1870 msglen = skb->len;
1871 skb_pull(skb, msglen);
1872 }
1873
1874 return 0;
1875}
1876EXPORT_SYMBOL(netlink_rcv_skb);
1877
1878/**
1879 * nlmsg_notify - send a notification netlink message
1880 * @sk: netlink socket to use
1881 * @skb: notification message
1882 * @pid: destination netlink pid for reports or 0
1883 * @group: destination multicast group or 0
1884 * @report: 1 to report back, 0 to disable
1885 * @flags: allocation flags
1886 */
1887int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1888 unsigned int group, int report, gfp_t flags)
1889{
1890 int err = 0;
1891
1892 if (group) {
1893 int exclude_pid = 0;
1894
1895 if (report) {
1896 atomic_inc(&skb->users);
1897 exclude_pid = pid;
1898 }
1899
1900 /* errors reported via destination sk->sk_err, but propagate
1901 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
1902 err = nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1903 }
1904
1905 if (report) {
1906 int err2;
1907
1908 err2 = nlmsg_unicast(sk, skb, pid);
1909 if (!err || err == -ESRCH)
1910 err = err2;
1911 }
1912
1913 return err;
1914}
1915EXPORT_SYMBOL(nlmsg_notify);
1916
1917#ifdef CONFIG_PROC_FS
1918struct nl_seq_iter {
1919 struct seq_net_private p;
1920 int link;
1921 int hash_idx;
1922};
1923
1924static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1925{
1926 struct nl_seq_iter *iter = seq->private;
1927 int i, j;
1928 struct sock *s;
1929 struct hlist_node *node;
1930 loff_t off = 0;
1931
1932 for (i = 0; i < MAX_LINKS; i++) {
1933 struct nl_pid_hash *hash = &nl_table[i].hash;
1934
1935 for (j = 0; j <= hash->mask; j++) {
1936 sk_for_each(s, node, &hash->table[j]) {
1937 if (sock_net(s) != seq_file_net(seq))
1938 continue;
1939 if (off == pos) {
1940 iter->link = i;
1941 iter->hash_idx = j;
1942 return s;
1943 }
1944 ++off;
1945 }
1946 }
1947 }
1948 return NULL;
1949}
1950
1951static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1952 __acquires(nl_table_lock)
1953{
1954 read_lock(&nl_table_lock);
1955 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1956}
1957
1958static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1959{
1960 struct sock *s;
1961 struct nl_seq_iter *iter;
1962 int i, j;
1963
1964 ++*pos;
1965
1966 if (v == SEQ_START_TOKEN)
1967 return netlink_seq_socket_idx(seq, 0);
1968
1969 iter = seq->private;
1970 s = v;
1971 do {
1972 s = sk_next(s);
1973 } while (s && sock_net(s) != seq_file_net(seq));
1974 if (s)
1975 return s;
1976
1977 i = iter->link;
1978 j = iter->hash_idx + 1;
1979
1980 do {
1981 struct nl_pid_hash *hash = &nl_table[i].hash;
1982
1983 for (; j <= hash->mask; j++) {
1984 s = sk_head(&hash->table[j]);
1985 while (s && sock_net(s) != seq_file_net(seq))
1986 s = sk_next(s);
1987 if (s) {
1988 iter->link = i;
1989 iter->hash_idx = j;
1990 return s;
1991 }
1992 }
1993
1994 j = 0;
1995 } while (++i < MAX_LINKS);
1996
1997 return NULL;
1998}
1999
2000static void netlink_seq_stop(struct seq_file *seq, void *v)
2001 __releases(nl_table_lock)
2002{
2003 read_unlock(&nl_table_lock);
2004}
2005
2006
2007static int netlink_seq_show(struct seq_file *seq, void *v)
2008{
2009 if (v == SEQ_START_TOKEN) {
2010 seq_puts(seq,
2011 "sk Eth Pid Groups "
2012 "Rmem Wmem Dump Locks Drops Inode\n");
2013 } else {
2014 struct sock *s = v;
2015 struct netlink_sock *nlk = nlk_sk(s);
2016
2017 seq_printf(seq, "%pK %-3d %-6d %08x %-8d %-8d %pK %-8d %-8d %-8lu\n",
2018 s,
2019 s->sk_protocol,
2020 nlk->pid,
2021 nlk->groups ? (u32)nlk->groups[0] : 0,
2022 sk_rmem_alloc_get(s),
2023 sk_wmem_alloc_get(s),
2024 nlk->cb,
2025 atomic_read(&s->sk_refcnt),
2026 atomic_read(&s->sk_drops),
2027 sock_i_ino(s)
2028 );
2029
2030 }
2031 return 0;
2032}
2033
2034static const struct seq_operations netlink_seq_ops = {
2035 .start = netlink_seq_start,
2036 .next = netlink_seq_next,
2037 .stop = netlink_seq_stop,
2038 .show = netlink_seq_show,
2039};
2040
2041
2042static int netlink_seq_open(struct inode *inode, struct file *file)
2043{
2044 return seq_open_net(inode, file, &netlink_seq_ops,
2045 sizeof(struct nl_seq_iter));
2046}
2047
2048static const struct file_operations netlink_seq_fops = {
2049 .owner = THIS_MODULE,
2050 .open = netlink_seq_open,
2051 .read = seq_read,
2052 .llseek = seq_lseek,
2053 .release = seq_release_net,
2054};
2055
2056#endif
2057
2058int netlink_register_notifier(struct notifier_block *nb)
2059{
2060 return atomic_notifier_chain_register(&netlink_chain, nb);
2061}
2062EXPORT_SYMBOL(netlink_register_notifier);
2063
2064int netlink_unregister_notifier(struct notifier_block *nb)
2065{
2066 return atomic_notifier_chain_unregister(&netlink_chain, nb);
2067}
2068EXPORT_SYMBOL(netlink_unregister_notifier);
2069
2070static const struct proto_ops netlink_ops = {
2071 .family = PF_NETLINK,
2072 .owner = THIS_MODULE,
2073 .release = netlink_release,
2074 .bind = netlink_bind,
2075 .connect = netlink_connect,
2076 .socketpair = sock_no_socketpair,
2077 .accept = sock_no_accept,
2078 .getname = netlink_getname,
2079 .poll = datagram_poll,
2080 .ioctl = sock_no_ioctl,
2081 .listen = sock_no_listen,
2082 .shutdown = sock_no_shutdown,
2083 .setsockopt = netlink_setsockopt,
2084 .getsockopt = netlink_getsockopt,
2085 .sendmsg = netlink_sendmsg,
2086 .recvmsg = netlink_recvmsg,
2087 .mmap = sock_no_mmap,
2088 .sendpage = sock_no_sendpage,
2089};
2090
2091static const struct net_proto_family netlink_family_ops = {
2092 .family = PF_NETLINK,
2093 .create = netlink_create,
2094 .owner = THIS_MODULE, /* for consistency 8) */
2095};
2096
2097static int __net_init netlink_net_init(struct net *net)
2098{
2099#ifdef CONFIG_PROC_FS
2100 if (!proc_net_fops_create(net, "netlink", 0, &netlink_seq_fops))
2101 return -ENOMEM;
2102#endif
2103 return 0;
2104}
2105
2106static void __net_exit netlink_net_exit(struct net *net)
2107{
2108#ifdef CONFIG_PROC_FS
2109 proc_net_remove(net, "netlink");
2110#endif
2111}
2112
2113static void __init netlink_add_usersock_entry(void)
2114{
2115 struct listeners *listeners;
2116 int groups = 32;
2117
2118 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2119 if (!listeners)
2120 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2121
2122 netlink_table_grab();
2123
2124 nl_table[NETLINK_USERSOCK].groups = groups;
2125 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2126 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2127 nl_table[NETLINK_USERSOCK].registered = 1;
2128 nl_table[NETLINK_USERSOCK].nl_nonroot = NL_NONROOT_SEND;
2129
2130 netlink_table_ungrab();
2131}
2132
2133static struct pernet_operations __net_initdata netlink_net_ops = {
2134 .init = netlink_net_init,
2135 .exit = netlink_net_exit,
2136};
2137
2138static int __init netlink_proto_init(void)
2139{
2140 struct sk_buff *dummy_skb;
2141 int i;
2142 unsigned long limit;
2143 unsigned int order;
2144 int err = proto_register(&netlink_proto, 0);
2145
2146 if (err != 0)
2147 goto out;
2148
2149 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
2150
2151 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2152 if (!nl_table)
2153 goto panic;
2154
2155 if (totalram_pages >= (128 * 1024))
2156 limit = totalram_pages >> (21 - PAGE_SHIFT);
2157 else
2158 limit = totalram_pages >> (23 - PAGE_SHIFT);
2159
2160 order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
2161 limit = (1UL << order) / sizeof(struct hlist_head);
2162 order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
2163
2164 for (i = 0; i < MAX_LINKS; i++) {
2165 struct nl_pid_hash *hash = &nl_table[i].hash;
2166
2167 hash->table = nl_pid_hash_zalloc(1 * sizeof(*hash->table));
2168 if (!hash->table) {
2169 while (i-- > 0)
2170 nl_pid_hash_free(nl_table[i].hash.table,
2171 1 * sizeof(*hash->table));
2172 kfree(nl_table);
2173 goto panic;
2174 }
2175 hash->max_shift = order;
2176 hash->shift = 0;
2177 hash->mask = 0;
2178 hash->rehash_time = jiffies;
2179 }
2180
2181 netlink_add_usersock_entry();
2182
2183 sock_register(&netlink_family_ops);
2184 register_pernet_subsys(&netlink_net_ops);
2185 /* The netlink device handler may be needed early. */
2186 rtnetlink_init();
2187out:
2188 return err;
2189panic:
2190 panic("netlink_init: Cannot allocate nl_table\n");
2191}
2192
2193core_initcall(netlink_proto_init);