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#include <linux/btf_ids.h>
64
65#include <net/net_namespace.h>
66#include <net/netns/generic.h>
67#include <net/sock.h>
68#include <net/scm.h>
69#include <net/netlink.h>
70#define CREATE_TRACE_POINTS
71#include <trace/events/netlink.h>
72
73#include "af_netlink.h"
74
75struct listeners {
76 struct rcu_head rcu;
77 unsigned long masks[];
78};
79
80/* state bits */
81#define NETLINK_S_CONGESTED 0x0
82
83static inline int netlink_is_kernel(struct sock *sk)
84{
85 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
86}
87
88struct netlink_table *nl_table __read_mostly;
89EXPORT_SYMBOL_GPL(nl_table);
90
91static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
92
93static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
94
95static const char *const nlk_cb_mutex_key_strings[MAX_LINKS + 1] = {
96 "nlk_cb_mutex-ROUTE",
97 "nlk_cb_mutex-1",
98 "nlk_cb_mutex-USERSOCK",
99 "nlk_cb_mutex-FIREWALL",
100 "nlk_cb_mutex-SOCK_DIAG",
101 "nlk_cb_mutex-NFLOG",
102 "nlk_cb_mutex-XFRM",
103 "nlk_cb_mutex-SELINUX",
104 "nlk_cb_mutex-ISCSI",
105 "nlk_cb_mutex-AUDIT",
106 "nlk_cb_mutex-FIB_LOOKUP",
107 "nlk_cb_mutex-CONNECTOR",
108 "nlk_cb_mutex-NETFILTER",
109 "nlk_cb_mutex-IP6_FW",
110 "nlk_cb_mutex-DNRTMSG",
111 "nlk_cb_mutex-KOBJECT_UEVENT",
112 "nlk_cb_mutex-GENERIC",
113 "nlk_cb_mutex-17",
114 "nlk_cb_mutex-SCSITRANSPORT",
115 "nlk_cb_mutex-ECRYPTFS",
116 "nlk_cb_mutex-RDMA",
117 "nlk_cb_mutex-CRYPTO",
118 "nlk_cb_mutex-SMC",
119 "nlk_cb_mutex-23",
120 "nlk_cb_mutex-24",
121 "nlk_cb_mutex-25",
122 "nlk_cb_mutex-26",
123 "nlk_cb_mutex-27",
124 "nlk_cb_mutex-28",
125 "nlk_cb_mutex-29",
126 "nlk_cb_mutex-30",
127 "nlk_cb_mutex-31",
128 "nlk_cb_mutex-MAX_LINKS"
129};
130
131static int netlink_dump(struct sock *sk);
132
133/* nl_table locking explained:
134 * Lookup and traversal are protected with an RCU read-side lock. Insertion
135 * and removal are protected with per bucket lock while using RCU list
136 * modification primitives and may run in parallel to RCU protected lookups.
137 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
138 * been acquired * either during or after the socket has been removed from
139 * the list and after an RCU grace period.
140 */
141DEFINE_RWLOCK(nl_table_lock);
142EXPORT_SYMBOL_GPL(nl_table_lock);
143static atomic_t nl_table_users = ATOMIC_INIT(0);
144
145#define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
146
147static BLOCKING_NOTIFIER_HEAD(netlink_chain);
148
149
150static const struct rhashtable_params netlink_rhashtable_params;
151
152void do_trace_netlink_extack(const char *msg)
153{
154 trace_netlink_extack(msg);
155}
156EXPORT_SYMBOL(do_trace_netlink_extack);
157
158static inline u32 netlink_group_mask(u32 group)
159{
160 return group ? 1 << (group - 1) : 0;
161}
162
163static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
164 gfp_t gfp_mask)
165{
166 unsigned int len = skb_end_offset(skb);
167 struct sk_buff *new;
168
169 new = alloc_skb(len, gfp_mask);
170 if (new == NULL)
171 return NULL;
172
173 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
174 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
175 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
176
177 skb_put_data(new, skb->data, len);
178 return new;
179}
180
181static unsigned int netlink_tap_net_id;
182
183struct netlink_tap_net {
184 struct list_head netlink_tap_all;
185 struct mutex netlink_tap_lock;
186};
187
188int netlink_add_tap(struct netlink_tap *nt)
189{
190 struct net *net = dev_net(nt->dev);
191 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
192
193 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
194 return -EINVAL;
195
196 mutex_lock(&nn->netlink_tap_lock);
197 list_add_rcu(&nt->list, &nn->netlink_tap_all);
198 mutex_unlock(&nn->netlink_tap_lock);
199
200 __module_get(nt->module);
201
202 return 0;
203}
204EXPORT_SYMBOL_GPL(netlink_add_tap);
205
206static int __netlink_remove_tap(struct netlink_tap *nt)
207{
208 struct net *net = dev_net(nt->dev);
209 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
210 bool found = false;
211 struct netlink_tap *tmp;
212
213 mutex_lock(&nn->netlink_tap_lock);
214
215 list_for_each_entry(tmp, &nn->netlink_tap_all, list) {
216 if (nt == tmp) {
217 list_del_rcu(&nt->list);
218 found = true;
219 goto out;
220 }
221 }
222
223 pr_warn("__netlink_remove_tap: %p not found\n", nt);
224out:
225 mutex_unlock(&nn->netlink_tap_lock);
226
227 if (found)
228 module_put(nt->module);
229
230 return found ? 0 : -ENODEV;
231}
232
233int netlink_remove_tap(struct netlink_tap *nt)
234{
235 int ret;
236
237 ret = __netlink_remove_tap(nt);
238 synchronize_net();
239
240 return ret;
241}
242EXPORT_SYMBOL_GPL(netlink_remove_tap);
243
244static __net_init int netlink_tap_init_net(struct net *net)
245{
246 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
247
248 INIT_LIST_HEAD(&nn->netlink_tap_all);
249 mutex_init(&nn->netlink_tap_lock);
250 return 0;
251}
252
253static struct pernet_operations netlink_tap_net_ops = {
254 .init = netlink_tap_init_net,
255 .id = &netlink_tap_net_id,
256 .size = sizeof(struct netlink_tap_net),
257};
258
259static bool netlink_filter_tap(const struct sk_buff *skb)
260{
261 struct sock *sk = skb->sk;
262
263 /* We take the more conservative approach and
264 * whitelist socket protocols that may pass.
265 */
266 switch (sk->sk_protocol) {
267 case NETLINK_ROUTE:
268 case NETLINK_USERSOCK:
269 case NETLINK_SOCK_DIAG:
270 case NETLINK_NFLOG:
271 case NETLINK_XFRM:
272 case NETLINK_FIB_LOOKUP:
273 case NETLINK_NETFILTER:
274 case NETLINK_GENERIC:
275 return true;
276 }
277
278 return false;
279}
280
281static int __netlink_deliver_tap_skb(struct sk_buff *skb,
282 struct net_device *dev)
283{
284 struct sk_buff *nskb;
285 struct sock *sk = skb->sk;
286 int ret = -ENOMEM;
287
288 if (!net_eq(dev_net(dev), sock_net(sk)))
289 return 0;
290
291 dev_hold(dev);
292
293 if (is_vmalloc_addr(skb->head))
294 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
295 else
296 nskb = skb_clone(skb, GFP_ATOMIC);
297 if (nskb) {
298 nskb->dev = dev;
299 nskb->protocol = htons((u16) sk->sk_protocol);
300 nskb->pkt_type = netlink_is_kernel(sk) ?
301 PACKET_KERNEL : PACKET_USER;
302 skb_reset_network_header(nskb);
303 ret = dev_queue_xmit(nskb);
304 if (unlikely(ret > 0))
305 ret = net_xmit_errno(ret);
306 }
307
308 dev_put(dev);
309 return ret;
310}
311
312static void __netlink_deliver_tap(struct sk_buff *skb, struct netlink_tap_net *nn)
313{
314 int ret;
315 struct netlink_tap *tmp;
316
317 if (!netlink_filter_tap(skb))
318 return;
319
320 list_for_each_entry_rcu(tmp, &nn->netlink_tap_all, list) {
321 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
322 if (unlikely(ret))
323 break;
324 }
325}
326
327static void netlink_deliver_tap(struct net *net, struct sk_buff *skb)
328{
329 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
330
331 rcu_read_lock();
332
333 if (unlikely(!list_empty(&nn->netlink_tap_all)))
334 __netlink_deliver_tap(skb, nn);
335
336 rcu_read_unlock();
337}
338
339static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
340 struct sk_buff *skb)
341{
342 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
343 netlink_deliver_tap(sock_net(dst), skb);
344}
345
346static void netlink_overrun(struct sock *sk)
347{
348 struct netlink_sock *nlk = nlk_sk(sk);
349
350 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
351 if (!test_and_set_bit(NETLINK_S_CONGESTED,
352 &nlk_sk(sk)->state)) {
353 sk->sk_err = ENOBUFS;
354 sk_error_report(sk);
355 }
356 }
357 atomic_inc(&sk->sk_drops);
358}
359
360static void netlink_rcv_wake(struct sock *sk)
361{
362 struct netlink_sock *nlk = nlk_sk(sk);
363
364 if (skb_queue_empty_lockless(&sk->sk_receive_queue))
365 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
366 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
367 wake_up_interruptible(&nlk->wait);
368}
369
370static void netlink_skb_destructor(struct sk_buff *skb)
371{
372 if (is_vmalloc_addr(skb->head)) {
373 if (!skb->cloned ||
374 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
375 vfree(skb->head);
376
377 skb->head = NULL;
378 }
379 if (skb->sk != NULL)
380 sock_rfree(skb);
381}
382
383static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
384{
385 WARN_ON(skb->sk != NULL);
386 skb->sk = sk;
387 skb->destructor = netlink_skb_destructor;
388 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
389 sk_mem_charge(sk, skb->truesize);
390}
391
392static void netlink_sock_destruct(struct sock *sk)
393{
394 struct netlink_sock *nlk = nlk_sk(sk);
395
396 if (nlk->cb_running) {
397 if (nlk->cb.done)
398 nlk->cb.done(&nlk->cb);
399 module_put(nlk->cb.module);
400 kfree_skb(nlk->cb.skb);
401 }
402
403 skb_queue_purge(&sk->sk_receive_queue);
404
405 if (!sock_flag(sk, SOCK_DEAD)) {
406 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
407 return;
408 }
409
410 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
411 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
412 WARN_ON(nlk_sk(sk)->groups);
413}
414
415static void netlink_sock_destruct_work(struct work_struct *work)
416{
417 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
418 work);
419
420 sk_free(&nlk->sk);
421}
422
423/* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
424 * SMP. Look, when several writers sleep and reader wakes them up, all but one
425 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
426 * this, _but_ remember, it adds useless work on UP machines.
427 */
428
429void netlink_table_grab(void)
430 __acquires(nl_table_lock)
431{
432 might_sleep();
433
434 write_lock_irq(&nl_table_lock);
435
436 if (atomic_read(&nl_table_users)) {
437 DECLARE_WAITQUEUE(wait, current);
438
439 add_wait_queue_exclusive(&nl_table_wait, &wait);
440 for (;;) {
441 set_current_state(TASK_UNINTERRUPTIBLE);
442 if (atomic_read(&nl_table_users) == 0)
443 break;
444 write_unlock_irq(&nl_table_lock);
445 schedule();
446 write_lock_irq(&nl_table_lock);
447 }
448
449 __set_current_state(TASK_RUNNING);
450 remove_wait_queue(&nl_table_wait, &wait);
451 }
452}
453
454void netlink_table_ungrab(void)
455 __releases(nl_table_lock)
456{
457 write_unlock_irq(&nl_table_lock);
458 wake_up(&nl_table_wait);
459}
460
461static inline void
462netlink_lock_table(void)
463{
464 unsigned long flags;
465
466 /* read_lock() synchronizes us to netlink_table_grab */
467
468 read_lock_irqsave(&nl_table_lock, flags);
469 atomic_inc(&nl_table_users);
470 read_unlock_irqrestore(&nl_table_lock, flags);
471}
472
473static inline void
474netlink_unlock_table(void)
475{
476 if (atomic_dec_and_test(&nl_table_users))
477 wake_up(&nl_table_wait);
478}
479
480struct netlink_compare_arg
481{
482 possible_net_t pnet;
483 u32 portid;
484};
485
486/* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
487#define netlink_compare_arg_len \
488 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
489
490static inline int netlink_compare(struct rhashtable_compare_arg *arg,
491 const void *ptr)
492{
493 const struct netlink_compare_arg *x = arg->key;
494 const struct netlink_sock *nlk = ptr;
495
496 return nlk->portid != x->portid ||
497 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
498}
499
500static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
501 struct net *net, u32 portid)
502{
503 memset(arg, 0, sizeof(*arg));
504 write_pnet(&arg->pnet, net);
505 arg->portid = portid;
506}
507
508static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
509 struct net *net)
510{
511 struct netlink_compare_arg arg;
512
513 netlink_compare_arg_init(&arg, net, portid);
514 return rhashtable_lookup_fast(&table->hash, &arg,
515 netlink_rhashtable_params);
516}
517
518static int __netlink_insert(struct netlink_table *table, struct sock *sk)
519{
520 struct netlink_compare_arg arg;
521
522 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
523 return rhashtable_lookup_insert_key(&table->hash, &arg,
524 &nlk_sk(sk)->node,
525 netlink_rhashtable_params);
526}
527
528static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
529{
530 struct netlink_table *table = &nl_table[protocol];
531 struct sock *sk;
532
533 rcu_read_lock();
534 sk = __netlink_lookup(table, portid, net);
535 if (sk)
536 sock_hold(sk);
537 rcu_read_unlock();
538
539 return sk;
540}
541
542static const struct proto_ops netlink_ops;
543
544static void
545netlink_update_listeners(struct sock *sk)
546{
547 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
548 unsigned long mask;
549 unsigned int i;
550 struct listeners *listeners;
551
552 listeners = nl_deref_protected(tbl->listeners);
553 if (!listeners)
554 return;
555
556 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
557 mask = 0;
558 sk_for_each_bound(sk, &tbl->mc_list) {
559 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
560 mask |= nlk_sk(sk)->groups[i];
561 }
562 listeners->masks[i] = mask;
563 }
564 /* this function is only called with the netlink table "grabbed", which
565 * makes sure updates are visible before bind or setsockopt return. */
566}
567
568static int netlink_insert(struct sock *sk, u32 portid)
569{
570 struct netlink_table *table = &nl_table[sk->sk_protocol];
571 int err;
572
573 lock_sock(sk);
574
575 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
576 if (nlk_sk(sk)->bound)
577 goto err;
578
579 nlk_sk(sk)->portid = portid;
580 sock_hold(sk);
581
582 err = __netlink_insert(table, sk);
583 if (err) {
584 /* In case the hashtable backend returns with -EBUSY
585 * from here, it must not escape to the caller.
586 */
587 if (unlikely(err == -EBUSY))
588 err = -EOVERFLOW;
589 if (err == -EEXIST)
590 err = -EADDRINUSE;
591 sock_put(sk);
592 goto err;
593 }
594
595 /* We need to ensure that the socket is hashed and visible. */
596 smp_wmb();
597 /* Paired with lockless reads from netlink_bind(),
598 * netlink_connect() and netlink_sendmsg().
599 */
600 WRITE_ONCE(nlk_sk(sk)->bound, portid);
601
602err:
603 release_sock(sk);
604 return err;
605}
606
607static void netlink_remove(struct sock *sk)
608{
609 struct netlink_table *table;
610
611 table = &nl_table[sk->sk_protocol];
612 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
613 netlink_rhashtable_params)) {
614 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
615 __sock_put(sk);
616 }
617
618 netlink_table_grab();
619 if (nlk_sk(sk)->subscriptions) {
620 __sk_del_bind_node(sk);
621 netlink_update_listeners(sk);
622 }
623 if (sk->sk_protocol == NETLINK_GENERIC)
624 atomic_inc(&genl_sk_destructing_cnt);
625 netlink_table_ungrab();
626}
627
628static struct proto netlink_proto = {
629 .name = "NETLINK",
630 .owner = THIS_MODULE,
631 .obj_size = sizeof(struct netlink_sock),
632};
633
634static int __netlink_create(struct net *net, struct socket *sock,
635 struct mutex *cb_mutex, int protocol,
636 int kern)
637{
638 struct sock *sk;
639 struct netlink_sock *nlk;
640
641 sock->ops = &netlink_ops;
642
643 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
644 if (!sk)
645 return -ENOMEM;
646
647 sock_init_data(sock, sk);
648
649 nlk = nlk_sk(sk);
650 if (cb_mutex) {
651 nlk->cb_mutex = cb_mutex;
652 } else {
653 nlk->cb_mutex = &nlk->cb_def_mutex;
654 mutex_init(nlk->cb_mutex);
655 lockdep_set_class_and_name(nlk->cb_mutex,
656 nlk_cb_mutex_keys + protocol,
657 nlk_cb_mutex_key_strings[protocol]);
658 }
659 init_waitqueue_head(&nlk->wait);
660
661 sk->sk_destruct = netlink_sock_destruct;
662 sk->sk_protocol = protocol;
663 return 0;
664}
665
666static int netlink_create(struct net *net, struct socket *sock, int protocol,
667 int kern)
668{
669 struct module *module = NULL;
670 struct mutex *cb_mutex;
671 struct netlink_sock *nlk;
672 int (*bind)(struct net *net, int group);
673 void (*unbind)(struct net *net, int group);
674 int err = 0;
675
676 sock->state = SS_UNCONNECTED;
677
678 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
679 return -ESOCKTNOSUPPORT;
680
681 if (protocol < 0 || protocol >= MAX_LINKS)
682 return -EPROTONOSUPPORT;
683 protocol = array_index_nospec(protocol, MAX_LINKS);
684
685 netlink_lock_table();
686#ifdef CONFIG_MODULES
687 if (!nl_table[protocol].registered) {
688 netlink_unlock_table();
689 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
690 netlink_lock_table();
691 }
692#endif
693 if (nl_table[protocol].registered &&
694 try_module_get(nl_table[protocol].module))
695 module = nl_table[protocol].module;
696 else
697 err = -EPROTONOSUPPORT;
698 cb_mutex = nl_table[protocol].cb_mutex;
699 bind = nl_table[protocol].bind;
700 unbind = nl_table[protocol].unbind;
701 netlink_unlock_table();
702
703 if (err < 0)
704 goto out;
705
706 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
707 if (err < 0)
708 goto out_module;
709
710 local_bh_disable();
711 sock_prot_inuse_add(net, &netlink_proto, 1);
712 local_bh_enable();
713
714 nlk = nlk_sk(sock->sk);
715 nlk->module = module;
716 nlk->netlink_bind = bind;
717 nlk->netlink_unbind = unbind;
718out:
719 return err;
720
721out_module:
722 module_put(module);
723 goto out;
724}
725
726static void deferred_put_nlk_sk(struct rcu_head *head)
727{
728 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
729 struct sock *sk = &nlk->sk;
730
731 kfree(nlk->groups);
732 nlk->groups = NULL;
733
734 if (!refcount_dec_and_test(&sk->sk_refcnt))
735 return;
736
737 if (nlk->cb_running && nlk->cb.done) {
738 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
739 schedule_work(&nlk->work);
740 return;
741 }
742
743 sk_free(sk);
744}
745
746static int netlink_release(struct socket *sock)
747{
748 struct sock *sk = sock->sk;
749 struct netlink_sock *nlk;
750
751 if (!sk)
752 return 0;
753
754 netlink_remove(sk);
755 sock_orphan(sk);
756 nlk = nlk_sk(sk);
757
758 /*
759 * OK. Socket is unlinked, any packets that arrive now
760 * will be purged.
761 */
762
763 /* must not acquire netlink_table_lock in any way again before unbind
764 * and notifying genetlink is done as otherwise it might deadlock
765 */
766 if (nlk->netlink_unbind) {
767 int i;
768
769 for (i = 0; i < nlk->ngroups; i++)
770 if (test_bit(i, nlk->groups))
771 nlk->netlink_unbind(sock_net(sk), i + 1);
772 }
773 if (sk->sk_protocol == NETLINK_GENERIC &&
774 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
775 wake_up(&genl_sk_destructing_waitq);
776
777 sock->sk = NULL;
778 wake_up_interruptible_all(&nlk->wait);
779
780 skb_queue_purge(&sk->sk_write_queue);
781
782 if (nlk->portid && nlk->bound) {
783 struct netlink_notify n = {
784 .net = sock_net(sk),
785 .protocol = sk->sk_protocol,
786 .portid = nlk->portid,
787 };
788 blocking_notifier_call_chain(&netlink_chain,
789 NETLINK_URELEASE, &n);
790 }
791
792 module_put(nlk->module);
793
794 if (netlink_is_kernel(sk)) {
795 netlink_table_grab();
796 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
797 if (--nl_table[sk->sk_protocol].registered == 0) {
798 struct listeners *old;
799
800 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
801 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
802 kfree_rcu(old, rcu);
803 nl_table[sk->sk_protocol].module = NULL;
804 nl_table[sk->sk_protocol].bind = NULL;
805 nl_table[sk->sk_protocol].unbind = NULL;
806 nl_table[sk->sk_protocol].flags = 0;
807 nl_table[sk->sk_protocol].registered = 0;
808 }
809 netlink_table_ungrab();
810 }
811
812 local_bh_disable();
813 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
814 local_bh_enable();
815 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
816 return 0;
817}
818
819static int netlink_autobind(struct socket *sock)
820{
821 struct sock *sk = sock->sk;
822 struct net *net = sock_net(sk);
823 struct netlink_table *table = &nl_table[sk->sk_protocol];
824 s32 portid = task_tgid_vnr(current);
825 int err;
826 s32 rover = -4096;
827 bool ok;
828
829retry:
830 cond_resched();
831 rcu_read_lock();
832 ok = !__netlink_lookup(table, portid, net);
833 rcu_read_unlock();
834 if (!ok) {
835 /* Bind collision, search negative portid values. */
836 if (rover == -4096)
837 /* rover will be in range [S32_MIN, -4097] */
838 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
839 else if (rover >= -4096)
840 rover = -4097;
841 portid = rover--;
842 goto retry;
843 }
844
845 err = netlink_insert(sk, portid);
846 if (err == -EADDRINUSE)
847 goto retry;
848
849 /* If 2 threads race to autobind, that is fine. */
850 if (err == -EBUSY)
851 err = 0;
852
853 return err;
854}
855
856/**
857 * __netlink_ns_capable - General netlink message capability test
858 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
859 * @user_ns: The user namespace of the capability to use
860 * @cap: The capability to use
861 *
862 * Test to see if the opener of the socket we received the message
863 * from had when the netlink socket was created and the sender of the
864 * message has the capability @cap in the user namespace @user_ns.
865 */
866bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
867 struct user_namespace *user_ns, int cap)
868{
869 return ((nsp->flags & NETLINK_SKB_DST) ||
870 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
871 ns_capable(user_ns, cap);
872}
873EXPORT_SYMBOL(__netlink_ns_capable);
874
875/**
876 * netlink_ns_capable - General netlink message capability test
877 * @skb: socket buffer holding a netlink command from userspace
878 * @user_ns: The user namespace of the capability to use
879 * @cap: The capability to use
880 *
881 * Test to see if the opener of the socket we received the message
882 * from had when the netlink socket was created and the sender of the
883 * message has the capability @cap in the user namespace @user_ns.
884 */
885bool netlink_ns_capable(const struct sk_buff *skb,
886 struct user_namespace *user_ns, int cap)
887{
888 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
889}
890EXPORT_SYMBOL(netlink_ns_capable);
891
892/**
893 * netlink_capable - Netlink global message capability test
894 * @skb: socket buffer holding a netlink command from userspace
895 * @cap: The capability to use
896 *
897 * Test to see if the opener of the socket we received the message
898 * from had when the netlink socket was created and the sender of the
899 * message has the capability @cap in all user namespaces.
900 */
901bool netlink_capable(const struct sk_buff *skb, int cap)
902{
903 return netlink_ns_capable(skb, &init_user_ns, cap);
904}
905EXPORT_SYMBOL(netlink_capable);
906
907/**
908 * netlink_net_capable - Netlink network namespace message capability test
909 * @skb: socket buffer holding a netlink command from userspace
910 * @cap: The capability to use
911 *
912 * Test to see if the opener of the socket we received the message
913 * from had when the netlink socket was created and the sender of the
914 * message has the capability @cap over the network namespace of
915 * the socket we received the message from.
916 */
917bool netlink_net_capable(const struct sk_buff *skb, int cap)
918{
919 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
920}
921EXPORT_SYMBOL(netlink_net_capable);
922
923static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
924{
925 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
926 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
927}
928
929static void
930netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
931{
932 struct netlink_sock *nlk = nlk_sk(sk);
933
934 if (nlk->subscriptions && !subscriptions)
935 __sk_del_bind_node(sk);
936 else if (!nlk->subscriptions && subscriptions)
937 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
938 nlk->subscriptions = subscriptions;
939}
940
941static int netlink_realloc_groups(struct sock *sk)
942{
943 struct netlink_sock *nlk = nlk_sk(sk);
944 unsigned int groups;
945 unsigned long *new_groups;
946 int err = 0;
947
948 netlink_table_grab();
949
950 groups = nl_table[sk->sk_protocol].groups;
951 if (!nl_table[sk->sk_protocol].registered) {
952 err = -ENOENT;
953 goto out_unlock;
954 }
955
956 if (nlk->ngroups >= groups)
957 goto out_unlock;
958
959 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
960 if (new_groups == NULL) {
961 err = -ENOMEM;
962 goto out_unlock;
963 }
964 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
965 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
966
967 nlk->groups = new_groups;
968 nlk->ngroups = groups;
969 out_unlock:
970 netlink_table_ungrab();
971 return err;
972}
973
974static void netlink_undo_bind(int group, long unsigned int groups,
975 struct sock *sk)
976{
977 struct netlink_sock *nlk = nlk_sk(sk);
978 int undo;
979
980 if (!nlk->netlink_unbind)
981 return;
982
983 for (undo = 0; undo < group; undo++)
984 if (test_bit(undo, &groups))
985 nlk->netlink_unbind(sock_net(sk), undo + 1);
986}
987
988static int netlink_bind(struct socket *sock, struct sockaddr *addr,
989 int addr_len)
990{
991 struct sock *sk = sock->sk;
992 struct net *net = sock_net(sk);
993 struct netlink_sock *nlk = nlk_sk(sk);
994 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
995 int err = 0;
996 unsigned long groups;
997 bool bound;
998
999 if (addr_len < sizeof(struct sockaddr_nl))
1000 return -EINVAL;
1001
1002 if (nladdr->nl_family != AF_NETLINK)
1003 return -EINVAL;
1004 groups = nladdr->nl_groups;
1005
1006 /* Only superuser is allowed to listen multicasts */
1007 if (groups) {
1008 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1009 return -EPERM;
1010 err = netlink_realloc_groups(sk);
1011 if (err)
1012 return err;
1013 }
1014
1015 if (nlk->ngroups < BITS_PER_LONG)
1016 groups &= (1UL << nlk->ngroups) - 1;
1017
1018 /* Paired with WRITE_ONCE() in netlink_insert() */
1019 bound = READ_ONCE(nlk->bound);
1020 if (bound) {
1021 /* Ensure nlk->portid is up-to-date. */
1022 smp_rmb();
1023
1024 if (nladdr->nl_pid != nlk->portid)
1025 return -EINVAL;
1026 }
1027
1028 if (nlk->netlink_bind && groups) {
1029 int group;
1030
1031 /* nl_groups is a u32, so cap the maximum groups we can bind */
1032 for (group = 0; group < BITS_PER_TYPE(u32); group++) {
1033 if (!test_bit(group, &groups))
1034 continue;
1035 err = nlk->netlink_bind(net, group + 1);
1036 if (!err)
1037 continue;
1038 netlink_undo_bind(group, groups, sk);
1039 return err;
1040 }
1041 }
1042
1043 /* No need for barriers here as we return to user-space without
1044 * using any of the bound attributes.
1045 */
1046 netlink_lock_table();
1047 if (!bound) {
1048 err = nladdr->nl_pid ?
1049 netlink_insert(sk, nladdr->nl_pid) :
1050 netlink_autobind(sock);
1051 if (err) {
1052 netlink_undo_bind(BITS_PER_TYPE(u32), groups, sk);
1053 goto unlock;
1054 }
1055 }
1056
1057 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1058 goto unlock;
1059 netlink_unlock_table();
1060
1061 netlink_table_grab();
1062 netlink_update_subscriptions(sk, nlk->subscriptions +
1063 hweight32(groups) -
1064 hweight32(nlk->groups[0]));
1065 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1066 netlink_update_listeners(sk);
1067 netlink_table_ungrab();
1068
1069 return 0;
1070
1071unlock:
1072 netlink_unlock_table();
1073 return err;
1074}
1075
1076static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1077 int alen, int flags)
1078{
1079 int err = 0;
1080 struct sock *sk = sock->sk;
1081 struct netlink_sock *nlk = nlk_sk(sk);
1082 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1083
1084 if (alen < sizeof(addr->sa_family))
1085 return -EINVAL;
1086
1087 if (addr->sa_family == AF_UNSPEC) {
1088 sk->sk_state = NETLINK_UNCONNECTED;
1089 nlk->dst_portid = 0;
1090 nlk->dst_group = 0;
1091 return 0;
1092 }
1093 if (addr->sa_family != AF_NETLINK)
1094 return -EINVAL;
1095
1096 if (alen < sizeof(struct sockaddr_nl))
1097 return -EINVAL;
1098
1099 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1100 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1101 return -EPERM;
1102
1103 /* No need for barriers here as we return to user-space without
1104 * using any of the bound attributes.
1105 * Paired with WRITE_ONCE() in netlink_insert().
1106 */
1107 if (!READ_ONCE(nlk->bound))
1108 err = netlink_autobind(sock);
1109
1110 if (err == 0) {
1111 sk->sk_state = NETLINK_CONNECTED;
1112 nlk->dst_portid = nladdr->nl_pid;
1113 nlk->dst_group = ffs(nladdr->nl_groups);
1114 }
1115
1116 return err;
1117}
1118
1119static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1120 int peer)
1121{
1122 struct sock *sk = sock->sk;
1123 struct netlink_sock *nlk = nlk_sk(sk);
1124 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1125
1126 nladdr->nl_family = AF_NETLINK;
1127 nladdr->nl_pad = 0;
1128
1129 if (peer) {
1130 nladdr->nl_pid = nlk->dst_portid;
1131 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1132 } else {
1133 nladdr->nl_pid = nlk->portid;
1134 netlink_lock_table();
1135 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1136 netlink_unlock_table();
1137 }
1138 return sizeof(*nladdr);
1139}
1140
1141static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1142 unsigned long arg)
1143{
1144 /* try to hand this ioctl down to the NIC drivers.
1145 */
1146 return -ENOIOCTLCMD;
1147}
1148
1149static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1150{
1151 struct sock *sock;
1152 struct netlink_sock *nlk;
1153
1154 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1155 if (!sock)
1156 return ERR_PTR(-ECONNREFUSED);
1157
1158 /* Don't bother queuing skb if kernel socket has no input function */
1159 nlk = nlk_sk(sock);
1160 if (sock->sk_state == NETLINK_CONNECTED &&
1161 nlk->dst_portid != nlk_sk(ssk)->portid) {
1162 sock_put(sock);
1163 return ERR_PTR(-ECONNREFUSED);
1164 }
1165 return sock;
1166}
1167
1168struct sock *netlink_getsockbyfilp(struct file *filp)
1169{
1170 struct inode *inode = file_inode(filp);
1171 struct sock *sock;
1172
1173 if (!S_ISSOCK(inode->i_mode))
1174 return ERR_PTR(-ENOTSOCK);
1175
1176 sock = SOCKET_I(inode)->sk;
1177 if (sock->sk_family != AF_NETLINK)
1178 return ERR_PTR(-EINVAL);
1179
1180 sock_hold(sock);
1181 return sock;
1182}
1183
1184static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1185 int broadcast)
1186{
1187 struct sk_buff *skb;
1188 void *data;
1189
1190 if (size <= NLMSG_GOODSIZE || broadcast)
1191 return alloc_skb(size, GFP_KERNEL);
1192
1193 size = SKB_DATA_ALIGN(size) +
1194 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1195
1196 data = vmalloc(size);
1197 if (data == NULL)
1198 return NULL;
1199
1200 skb = __build_skb(data, size);
1201 if (skb == NULL)
1202 vfree(data);
1203 else
1204 skb->destructor = netlink_skb_destructor;
1205
1206 return skb;
1207}
1208
1209/*
1210 * Attach a skb to a netlink socket.
1211 * The caller must hold a reference to the destination socket. On error, the
1212 * reference is dropped. The skb is not send to the destination, just all
1213 * all error checks are performed and memory in the queue is reserved.
1214 * Return values:
1215 * < 0: error. skb freed, reference to sock dropped.
1216 * 0: continue
1217 * 1: repeat lookup - reference dropped while waiting for socket memory.
1218 */
1219int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1220 long *timeo, struct sock *ssk)
1221{
1222 struct netlink_sock *nlk;
1223
1224 nlk = nlk_sk(sk);
1225
1226 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1227 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1228 DECLARE_WAITQUEUE(wait, current);
1229 if (!*timeo) {
1230 if (!ssk || netlink_is_kernel(ssk))
1231 netlink_overrun(sk);
1232 sock_put(sk);
1233 kfree_skb(skb);
1234 return -EAGAIN;
1235 }
1236
1237 __set_current_state(TASK_INTERRUPTIBLE);
1238 add_wait_queue(&nlk->wait, &wait);
1239
1240 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1241 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1242 !sock_flag(sk, SOCK_DEAD))
1243 *timeo = schedule_timeout(*timeo);
1244
1245 __set_current_state(TASK_RUNNING);
1246 remove_wait_queue(&nlk->wait, &wait);
1247 sock_put(sk);
1248
1249 if (signal_pending(current)) {
1250 kfree_skb(skb);
1251 return sock_intr_errno(*timeo);
1252 }
1253 return 1;
1254 }
1255 netlink_skb_set_owner_r(skb, sk);
1256 return 0;
1257}
1258
1259static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1260{
1261 int len = skb->len;
1262
1263 netlink_deliver_tap(sock_net(sk), skb);
1264
1265 skb_queue_tail(&sk->sk_receive_queue, skb);
1266 sk->sk_data_ready(sk);
1267 return len;
1268}
1269
1270int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1271{
1272 int len = __netlink_sendskb(sk, skb);
1273
1274 sock_put(sk);
1275 return len;
1276}
1277
1278void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1279{
1280 kfree_skb(skb);
1281 sock_put(sk);
1282}
1283
1284static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1285{
1286 int delta;
1287
1288 WARN_ON(skb->sk != NULL);
1289 delta = skb->end - skb->tail;
1290 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1291 return skb;
1292
1293 if (skb_shared(skb)) {
1294 struct sk_buff *nskb = skb_clone(skb, allocation);
1295 if (!nskb)
1296 return skb;
1297 consume_skb(skb);
1298 skb = nskb;
1299 }
1300
1301 pskb_expand_head(skb, 0, -delta,
1302 (allocation & ~__GFP_DIRECT_RECLAIM) |
1303 __GFP_NOWARN | __GFP_NORETRY);
1304 return skb;
1305}
1306
1307static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1308 struct sock *ssk)
1309{
1310 int ret;
1311 struct netlink_sock *nlk = nlk_sk(sk);
1312
1313 ret = -ECONNREFUSED;
1314 if (nlk->netlink_rcv != NULL) {
1315 ret = skb->len;
1316 netlink_skb_set_owner_r(skb, sk);
1317 NETLINK_CB(skb).sk = ssk;
1318 netlink_deliver_tap_kernel(sk, ssk, skb);
1319 nlk->netlink_rcv(skb);
1320 consume_skb(skb);
1321 } else {
1322 kfree_skb(skb);
1323 }
1324 sock_put(sk);
1325 return ret;
1326}
1327
1328int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1329 u32 portid, int nonblock)
1330{
1331 struct sock *sk;
1332 int err;
1333 long timeo;
1334
1335 skb = netlink_trim(skb, gfp_any());
1336
1337 timeo = sock_sndtimeo(ssk, nonblock);
1338retry:
1339 sk = netlink_getsockbyportid(ssk, portid);
1340 if (IS_ERR(sk)) {
1341 kfree_skb(skb);
1342 return PTR_ERR(sk);
1343 }
1344 if (netlink_is_kernel(sk))
1345 return netlink_unicast_kernel(sk, skb, ssk);
1346
1347 if (sk_filter(sk, skb)) {
1348 err = skb->len;
1349 kfree_skb(skb);
1350 sock_put(sk);
1351 return err;
1352 }
1353
1354 err = netlink_attachskb(sk, skb, &timeo, ssk);
1355 if (err == 1)
1356 goto retry;
1357 if (err)
1358 return err;
1359
1360 return netlink_sendskb(sk, skb);
1361}
1362EXPORT_SYMBOL(netlink_unicast);
1363
1364int netlink_has_listeners(struct sock *sk, unsigned int group)
1365{
1366 int res = 0;
1367 struct listeners *listeners;
1368
1369 BUG_ON(!netlink_is_kernel(sk));
1370
1371 rcu_read_lock();
1372 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1373
1374 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1375 res = test_bit(group - 1, listeners->masks);
1376
1377 rcu_read_unlock();
1378
1379 return res;
1380}
1381EXPORT_SYMBOL_GPL(netlink_has_listeners);
1382
1383bool netlink_strict_get_check(struct sk_buff *skb)
1384{
1385 const struct netlink_sock *nlk = nlk_sk(NETLINK_CB(skb).sk);
1386
1387 return nlk->flags & NETLINK_F_STRICT_CHK;
1388}
1389EXPORT_SYMBOL_GPL(netlink_strict_get_check);
1390
1391static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1392{
1393 struct netlink_sock *nlk = nlk_sk(sk);
1394
1395 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1396 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1397 netlink_skb_set_owner_r(skb, sk);
1398 __netlink_sendskb(sk, skb);
1399 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1400 }
1401 return -1;
1402}
1403
1404struct netlink_broadcast_data {
1405 struct sock *exclude_sk;
1406 struct net *net;
1407 u32 portid;
1408 u32 group;
1409 int failure;
1410 int delivery_failure;
1411 int congested;
1412 int delivered;
1413 gfp_t allocation;
1414 struct sk_buff *skb, *skb2;
1415 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1416 void *tx_data;
1417};
1418
1419static void do_one_broadcast(struct sock *sk,
1420 struct netlink_broadcast_data *p)
1421{
1422 struct netlink_sock *nlk = nlk_sk(sk);
1423 int val;
1424
1425 if (p->exclude_sk == sk)
1426 return;
1427
1428 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1429 !test_bit(p->group - 1, nlk->groups))
1430 return;
1431
1432 if (!net_eq(sock_net(sk), p->net)) {
1433 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1434 return;
1435
1436 if (!peernet_has_id(sock_net(sk), p->net))
1437 return;
1438
1439 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1440 CAP_NET_BROADCAST))
1441 return;
1442 }
1443
1444 if (p->failure) {
1445 netlink_overrun(sk);
1446 return;
1447 }
1448
1449 sock_hold(sk);
1450 if (p->skb2 == NULL) {
1451 if (skb_shared(p->skb)) {
1452 p->skb2 = skb_clone(p->skb, p->allocation);
1453 } else {
1454 p->skb2 = skb_get(p->skb);
1455 /*
1456 * skb ownership may have been set when
1457 * delivered to a previous socket.
1458 */
1459 skb_orphan(p->skb2);
1460 }
1461 }
1462 if (p->skb2 == NULL) {
1463 netlink_overrun(sk);
1464 /* Clone failed. Notify ALL listeners. */
1465 p->failure = 1;
1466 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1467 p->delivery_failure = 1;
1468 goto out;
1469 }
1470 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1471 kfree_skb(p->skb2);
1472 p->skb2 = NULL;
1473 goto out;
1474 }
1475 if (sk_filter(sk, p->skb2)) {
1476 kfree_skb(p->skb2);
1477 p->skb2 = NULL;
1478 goto out;
1479 }
1480 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1481 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1482 NETLINK_CB(p->skb2).nsid_is_set = true;
1483 val = netlink_broadcast_deliver(sk, p->skb2);
1484 if (val < 0) {
1485 netlink_overrun(sk);
1486 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1487 p->delivery_failure = 1;
1488 } else {
1489 p->congested |= val;
1490 p->delivered = 1;
1491 p->skb2 = NULL;
1492 }
1493out:
1494 sock_put(sk);
1495}
1496
1497int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1498 u32 group, gfp_t allocation,
1499 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1500 void *filter_data)
1501{
1502 struct net *net = sock_net(ssk);
1503 struct netlink_broadcast_data info;
1504 struct sock *sk;
1505
1506 skb = netlink_trim(skb, allocation);
1507
1508 info.exclude_sk = ssk;
1509 info.net = net;
1510 info.portid = portid;
1511 info.group = group;
1512 info.failure = 0;
1513 info.delivery_failure = 0;
1514 info.congested = 0;
1515 info.delivered = 0;
1516 info.allocation = allocation;
1517 info.skb = skb;
1518 info.skb2 = NULL;
1519 info.tx_filter = filter;
1520 info.tx_data = filter_data;
1521
1522 /* While we sleep in clone, do not allow to change socket list */
1523
1524 netlink_lock_table();
1525
1526 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1527 do_one_broadcast(sk, &info);
1528
1529 consume_skb(skb);
1530
1531 netlink_unlock_table();
1532
1533 if (info.delivery_failure) {
1534 kfree_skb(info.skb2);
1535 return -ENOBUFS;
1536 }
1537 consume_skb(info.skb2);
1538
1539 if (info.delivered) {
1540 if (info.congested && gfpflags_allow_blocking(allocation))
1541 yield();
1542 return 0;
1543 }
1544 return -ESRCH;
1545}
1546EXPORT_SYMBOL(netlink_broadcast_filtered);
1547
1548int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1549 u32 group, gfp_t allocation)
1550{
1551 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1552 NULL, NULL);
1553}
1554EXPORT_SYMBOL(netlink_broadcast);
1555
1556struct netlink_set_err_data {
1557 struct sock *exclude_sk;
1558 u32 portid;
1559 u32 group;
1560 int code;
1561};
1562
1563static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1564{
1565 struct netlink_sock *nlk = nlk_sk(sk);
1566 int ret = 0;
1567
1568 if (sk == p->exclude_sk)
1569 goto out;
1570
1571 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1572 goto out;
1573
1574 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1575 !test_bit(p->group - 1, nlk->groups))
1576 goto out;
1577
1578 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1579 ret = 1;
1580 goto out;
1581 }
1582
1583 sk->sk_err = p->code;
1584 sk_error_report(sk);
1585out:
1586 return ret;
1587}
1588
1589/**
1590 * netlink_set_err - report error to broadcast listeners
1591 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1592 * @portid: the PORTID of a process that we want to skip (if any)
1593 * @group: the broadcast group that will notice the error
1594 * @code: error code, must be negative (as usual in kernelspace)
1595 *
1596 * This function returns the number of broadcast listeners that have set the
1597 * NETLINK_NO_ENOBUFS socket option.
1598 */
1599int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1600{
1601 struct netlink_set_err_data info;
1602 struct sock *sk;
1603 int ret = 0;
1604
1605 info.exclude_sk = ssk;
1606 info.portid = portid;
1607 info.group = group;
1608 /* sk->sk_err wants a positive error value */
1609 info.code = -code;
1610
1611 read_lock(&nl_table_lock);
1612
1613 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1614 ret += do_one_set_err(sk, &info);
1615
1616 read_unlock(&nl_table_lock);
1617 return ret;
1618}
1619EXPORT_SYMBOL(netlink_set_err);
1620
1621/* must be called with netlink table grabbed */
1622static void netlink_update_socket_mc(struct netlink_sock *nlk,
1623 unsigned int group,
1624 int is_new)
1625{
1626 int old, new = !!is_new, subscriptions;
1627
1628 old = test_bit(group - 1, nlk->groups);
1629 subscriptions = nlk->subscriptions - old + new;
1630 if (new)
1631 __set_bit(group - 1, nlk->groups);
1632 else
1633 __clear_bit(group - 1, nlk->groups);
1634 netlink_update_subscriptions(&nlk->sk, subscriptions);
1635 netlink_update_listeners(&nlk->sk);
1636}
1637
1638static int netlink_setsockopt(struct socket *sock, int level, int optname,
1639 sockptr_t optval, unsigned int optlen)
1640{
1641 struct sock *sk = sock->sk;
1642 struct netlink_sock *nlk = nlk_sk(sk);
1643 unsigned int val = 0;
1644 int err;
1645
1646 if (level != SOL_NETLINK)
1647 return -ENOPROTOOPT;
1648
1649 if (optlen >= sizeof(int) &&
1650 copy_from_sockptr(&val, optval, sizeof(val)))
1651 return -EFAULT;
1652
1653 switch (optname) {
1654 case NETLINK_PKTINFO:
1655 if (val)
1656 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1657 else
1658 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1659 err = 0;
1660 break;
1661 case NETLINK_ADD_MEMBERSHIP:
1662 case NETLINK_DROP_MEMBERSHIP: {
1663 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1664 return -EPERM;
1665 err = netlink_realloc_groups(sk);
1666 if (err)
1667 return err;
1668 if (!val || val - 1 >= nlk->ngroups)
1669 return -EINVAL;
1670 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1671 err = nlk->netlink_bind(sock_net(sk), val);
1672 if (err)
1673 return err;
1674 }
1675 netlink_table_grab();
1676 netlink_update_socket_mc(nlk, val,
1677 optname == NETLINK_ADD_MEMBERSHIP);
1678 netlink_table_ungrab();
1679 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1680 nlk->netlink_unbind(sock_net(sk), val);
1681
1682 err = 0;
1683 break;
1684 }
1685 case NETLINK_BROADCAST_ERROR:
1686 if (val)
1687 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1688 else
1689 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1690 err = 0;
1691 break;
1692 case NETLINK_NO_ENOBUFS:
1693 if (val) {
1694 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1695 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1696 wake_up_interruptible(&nlk->wait);
1697 } else {
1698 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1699 }
1700 err = 0;
1701 break;
1702 case NETLINK_LISTEN_ALL_NSID:
1703 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1704 return -EPERM;
1705
1706 if (val)
1707 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1708 else
1709 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1710 err = 0;
1711 break;
1712 case NETLINK_CAP_ACK:
1713 if (val)
1714 nlk->flags |= NETLINK_F_CAP_ACK;
1715 else
1716 nlk->flags &= ~NETLINK_F_CAP_ACK;
1717 err = 0;
1718 break;
1719 case NETLINK_EXT_ACK:
1720 if (val)
1721 nlk->flags |= NETLINK_F_EXT_ACK;
1722 else
1723 nlk->flags &= ~NETLINK_F_EXT_ACK;
1724 err = 0;
1725 break;
1726 case NETLINK_GET_STRICT_CHK:
1727 if (val)
1728 nlk->flags |= NETLINK_F_STRICT_CHK;
1729 else
1730 nlk->flags &= ~NETLINK_F_STRICT_CHK;
1731 err = 0;
1732 break;
1733 default:
1734 err = -ENOPROTOOPT;
1735 }
1736 return err;
1737}
1738
1739static int netlink_getsockopt(struct socket *sock, int level, int optname,
1740 char __user *optval, int __user *optlen)
1741{
1742 struct sock *sk = sock->sk;
1743 struct netlink_sock *nlk = nlk_sk(sk);
1744 int len, val, err;
1745
1746 if (level != SOL_NETLINK)
1747 return -ENOPROTOOPT;
1748
1749 if (get_user(len, optlen))
1750 return -EFAULT;
1751 if (len < 0)
1752 return -EINVAL;
1753
1754 switch (optname) {
1755 case NETLINK_PKTINFO:
1756 if (len < sizeof(int))
1757 return -EINVAL;
1758 len = sizeof(int);
1759 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1760 if (put_user(len, optlen) ||
1761 put_user(val, optval))
1762 return -EFAULT;
1763 err = 0;
1764 break;
1765 case NETLINK_BROADCAST_ERROR:
1766 if (len < sizeof(int))
1767 return -EINVAL;
1768 len = sizeof(int);
1769 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1770 if (put_user(len, optlen) ||
1771 put_user(val, optval))
1772 return -EFAULT;
1773 err = 0;
1774 break;
1775 case NETLINK_NO_ENOBUFS:
1776 if (len < sizeof(int))
1777 return -EINVAL;
1778 len = sizeof(int);
1779 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1780 if (put_user(len, optlen) ||
1781 put_user(val, optval))
1782 return -EFAULT;
1783 err = 0;
1784 break;
1785 case NETLINK_LIST_MEMBERSHIPS: {
1786 int pos, idx, shift;
1787
1788 err = 0;
1789 netlink_lock_table();
1790 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1791 if (len - pos < sizeof(u32))
1792 break;
1793
1794 idx = pos / sizeof(unsigned long);
1795 shift = (pos % sizeof(unsigned long)) * 8;
1796 if (put_user((u32)(nlk->groups[idx] >> shift),
1797 (u32 __user *)(optval + pos))) {
1798 err = -EFAULT;
1799 break;
1800 }
1801 }
1802 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1803 err = -EFAULT;
1804 netlink_unlock_table();
1805 break;
1806 }
1807 case NETLINK_CAP_ACK:
1808 if (len < sizeof(int))
1809 return -EINVAL;
1810 len = sizeof(int);
1811 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1812 if (put_user(len, optlen) ||
1813 put_user(val, optval))
1814 return -EFAULT;
1815 err = 0;
1816 break;
1817 case NETLINK_EXT_ACK:
1818 if (len < sizeof(int))
1819 return -EINVAL;
1820 len = sizeof(int);
1821 val = nlk->flags & NETLINK_F_EXT_ACK ? 1 : 0;
1822 if (put_user(len, optlen) || put_user(val, optval))
1823 return -EFAULT;
1824 err = 0;
1825 break;
1826 case NETLINK_GET_STRICT_CHK:
1827 if (len < sizeof(int))
1828 return -EINVAL;
1829 len = sizeof(int);
1830 val = nlk->flags & NETLINK_F_STRICT_CHK ? 1 : 0;
1831 if (put_user(len, optlen) || put_user(val, optval))
1832 return -EFAULT;
1833 err = 0;
1834 break;
1835 default:
1836 err = -ENOPROTOOPT;
1837 }
1838 return err;
1839}
1840
1841static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1842{
1843 struct nl_pktinfo info;
1844
1845 info.group = NETLINK_CB(skb).dst_group;
1846 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1847}
1848
1849static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1850 struct sk_buff *skb)
1851{
1852 if (!NETLINK_CB(skb).nsid_is_set)
1853 return;
1854
1855 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1856 &NETLINK_CB(skb).nsid);
1857}
1858
1859static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1860{
1861 struct sock *sk = sock->sk;
1862 struct netlink_sock *nlk = nlk_sk(sk);
1863 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1864 u32 dst_portid;
1865 u32 dst_group;
1866 struct sk_buff *skb;
1867 int err;
1868 struct scm_cookie scm;
1869 u32 netlink_skb_flags = 0;
1870
1871 if (msg->msg_flags & MSG_OOB)
1872 return -EOPNOTSUPP;
1873
1874 err = scm_send(sock, msg, &scm, true);
1875 if (err < 0)
1876 return err;
1877
1878 if (msg->msg_namelen) {
1879 err = -EINVAL;
1880 if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1881 goto out;
1882 if (addr->nl_family != AF_NETLINK)
1883 goto out;
1884 dst_portid = addr->nl_pid;
1885 dst_group = ffs(addr->nl_groups);
1886 err = -EPERM;
1887 if ((dst_group || dst_portid) &&
1888 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1889 goto out;
1890 netlink_skb_flags |= NETLINK_SKB_DST;
1891 } else {
1892 dst_portid = nlk->dst_portid;
1893 dst_group = nlk->dst_group;
1894 }
1895
1896 /* Paired with WRITE_ONCE() in netlink_insert() */
1897 if (!READ_ONCE(nlk->bound)) {
1898 err = netlink_autobind(sock);
1899 if (err)
1900 goto out;
1901 } else {
1902 /* Ensure nlk is hashed and visible. */
1903 smp_rmb();
1904 }
1905
1906 err = -EMSGSIZE;
1907 if (len > sk->sk_sndbuf - 32)
1908 goto out;
1909 err = -ENOBUFS;
1910 skb = netlink_alloc_large_skb(len, dst_group);
1911 if (skb == NULL)
1912 goto out;
1913
1914 NETLINK_CB(skb).portid = nlk->portid;
1915 NETLINK_CB(skb).dst_group = dst_group;
1916 NETLINK_CB(skb).creds = scm.creds;
1917 NETLINK_CB(skb).flags = netlink_skb_flags;
1918
1919 err = -EFAULT;
1920 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1921 kfree_skb(skb);
1922 goto out;
1923 }
1924
1925 err = security_netlink_send(sk, skb);
1926 if (err) {
1927 kfree_skb(skb);
1928 goto out;
1929 }
1930
1931 if (dst_group) {
1932 refcount_inc(&skb->users);
1933 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1934 }
1935 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags & MSG_DONTWAIT);
1936
1937out:
1938 scm_destroy(&scm);
1939 return err;
1940}
1941
1942static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1943 int flags)
1944{
1945 struct scm_cookie scm;
1946 struct sock *sk = sock->sk;
1947 struct netlink_sock *nlk = nlk_sk(sk);
1948 int noblock = flags & MSG_DONTWAIT;
1949 size_t copied;
1950 struct sk_buff *skb, *data_skb;
1951 int err, ret;
1952
1953 if (flags & MSG_OOB)
1954 return -EOPNOTSUPP;
1955
1956 copied = 0;
1957
1958 skb = skb_recv_datagram(sk, flags, noblock, &err);
1959 if (skb == NULL)
1960 goto out;
1961
1962 data_skb = skb;
1963
1964#ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1965 if (unlikely(skb_shinfo(skb)->frag_list)) {
1966 /*
1967 * If this skb has a frag_list, then here that means that we
1968 * will have to use the frag_list skb's data for compat tasks
1969 * and the regular skb's data for normal (non-compat) tasks.
1970 *
1971 * If we need to send the compat skb, assign it to the
1972 * 'data_skb' variable so that it will be used below for data
1973 * copying. We keep 'skb' for everything else, including
1974 * freeing both later.
1975 */
1976 if (flags & MSG_CMSG_COMPAT)
1977 data_skb = skb_shinfo(skb)->frag_list;
1978 }
1979#endif
1980
1981 /* Record the max length of recvmsg() calls for future allocations */
1982 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1983 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1984 SKB_WITH_OVERHEAD(32768));
1985
1986 copied = data_skb->len;
1987 if (len < copied) {
1988 msg->msg_flags |= MSG_TRUNC;
1989 copied = len;
1990 }
1991
1992 skb_reset_transport_header(data_skb);
1993 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1994
1995 if (msg->msg_name) {
1996 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1997 addr->nl_family = AF_NETLINK;
1998 addr->nl_pad = 0;
1999 addr->nl_pid = NETLINK_CB(skb).portid;
2000 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
2001 msg->msg_namelen = sizeof(*addr);
2002 }
2003
2004 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
2005 netlink_cmsg_recv_pktinfo(msg, skb);
2006 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
2007 netlink_cmsg_listen_all_nsid(sk, msg, skb);
2008
2009 memset(&scm, 0, sizeof(scm));
2010 scm.creds = *NETLINK_CREDS(skb);
2011 if (flags & MSG_TRUNC)
2012 copied = data_skb->len;
2013
2014 skb_free_datagram(sk, skb);
2015
2016 if (nlk->cb_running &&
2017 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2018 ret = netlink_dump(sk);
2019 if (ret) {
2020 sk->sk_err = -ret;
2021 sk_error_report(sk);
2022 }
2023 }
2024
2025 scm_recv(sock, msg, &scm, flags);
2026out:
2027 netlink_rcv_wake(sk);
2028 return err ? : copied;
2029}
2030
2031static void netlink_data_ready(struct sock *sk)
2032{
2033 BUG();
2034}
2035
2036/*
2037 * We export these functions to other modules. They provide a
2038 * complete set of kernel non-blocking support for message
2039 * queueing.
2040 */
2041
2042struct sock *
2043__netlink_kernel_create(struct net *net, int unit, struct module *module,
2044 struct netlink_kernel_cfg *cfg)
2045{
2046 struct socket *sock;
2047 struct sock *sk;
2048 struct netlink_sock *nlk;
2049 struct listeners *listeners = NULL;
2050 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2051 unsigned int groups;
2052
2053 BUG_ON(!nl_table);
2054
2055 if (unit < 0 || unit >= MAX_LINKS)
2056 return NULL;
2057
2058 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2059 return NULL;
2060
2061 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2062 goto out_sock_release_nosk;
2063
2064 sk = sock->sk;
2065
2066 if (!cfg || cfg->groups < 32)
2067 groups = 32;
2068 else
2069 groups = cfg->groups;
2070
2071 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2072 if (!listeners)
2073 goto out_sock_release;
2074
2075 sk->sk_data_ready = netlink_data_ready;
2076 if (cfg && cfg->input)
2077 nlk_sk(sk)->netlink_rcv = cfg->input;
2078
2079 if (netlink_insert(sk, 0))
2080 goto out_sock_release;
2081
2082 nlk = nlk_sk(sk);
2083 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2084
2085 netlink_table_grab();
2086 if (!nl_table[unit].registered) {
2087 nl_table[unit].groups = groups;
2088 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2089 nl_table[unit].cb_mutex = cb_mutex;
2090 nl_table[unit].module = module;
2091 if (cfg) {
2092 nl_table[unit].bind = cfg->bind;
2093 nl_table[unit].unbind = cfg->unbind;
2094 nl_table[unit].flags = cfg->flags;
2095 if (cfg->compare)
2096 nl_table[unit].compare = cfg->compare;
2097 }
2098 nl_table[unit].registered = 1;
2099 } else {
2100 kfree(listeners);
2101 nl_table[unit].registered++;
2102 }
2103 netlink_table_ungrab();
2104 return sk;
2105
2106out_sock_release:
2107 kfree(listeners);
2108 netlink_kernel_release(sk);
2109 return NULL;
2110
2111out_sock_release_nosk:
2112 sock_release(sock);
2113 return NULL;
2114}
2115EXPORT_SYMBOL(__netlink_kernel_create);
2116
2117void
2118netlink_kernel_release(struct sock *sk)
2119{
2120 if (sk == NULL || sk->sk_socket == NULL)
2121 return;
2122
2123 sock_release(sk->sk_socket);
2124}
2125EXPORT_SYMBOL(netlink_kernel_release);
2126
2127int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2128{
2129 struct listeners *new, *old;
2130 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2131
2132 if (groups < 32)
2133 groups = 32;
2134
2135 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2136 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2137 if (!new)
2138 return -ENOMEM;
2139 old = nl_deref_protected(tbl->listeners);
2140 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2141 rcu_assign_pointer(tbl->listeners, new);
2142
2143 kfree_rcu(old, rcu);
2144 }
2145 tbl->groups = groups;
2146
2147 return 0;
2148}
2149
2150/**
2151 * netlink_change_ngroups - change number of multicast groups
2152 *
2153 * This changes the number of multicast groups that are available
2154 * on a certain netlink family. Note that it is not possible to
2155 * change the number of groups to below 32. Also note that it does
2156 * not implicitly call netlink_clear_multicast_users() when the
2157 * number of groups is reduced.
2158 *
2159 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2160 * @groups: The new number of groups.
2161 */
2162int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2163{
2164 int err;
2165
2166 netlink_table_grab();
2167 err = __netlink_change_ngroups(sk, groups);
2168 netlink_table_ungrab();
2169
2170 return err;
2171}
2172
2173void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2174{
2175 struct sock *sk;
2176 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2177
2178 sk_for_each_bound(sk, &tbl->mc_list)
2179 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2180}
2181
2182struct nlmsghdr *
2183__nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2184{
2185 struct nlmsghdr *nlh;
2186 int size = nlmsg_msg_size(len);
2187
2188 nlh = skb_put(skb, NLMSG_ALIGN(size));
2189 nlh->nlmsg_type = type;
2190 nlh->nlmsg_len = size;
2191 nlh->nlmsg_flags = flags;
2192 nlh->nlmsg_pid = portid;
2193 nlh->nlmsg_seq = seq;
2194 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2195 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2196 return nlh;
2197}
2198EXPORT_SYMBOL(__nlmsg_put);
2199
2200/*
2201 * It looks a bit ugly.
2202 * It would be better to create kernel thread.
2203 */
2204
2205static int netlink_dump_done(struct netlink_sock *nlk, struct sk_buff *skb,
2206 struct netlink_callback *cb,
2207 struct netlink_ext_ack *extack)
2208{
2209 struct nlmsghdr *nlh;
2210
2211 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(nlk->dump_done_errno),
2212 NLM_F_MULTI | cb->answer_flags);
2213 if (WARN_ON(!nlh))
2214 return -ENOBUFS;
2215
2216 nl_dump_check_consistent(cb, nlh);
2217 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno, sizeof(nlk->dump_done_errno));
2218
2219 if (extack->_msg && nlk->flags & NETLINK_F_EXT_ACK) {
2220 nlh->nlmsg_flags |= NLM_F_ACK_TLVS;
2221 if (!nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg))
2222 nlmsg_end(skb, nlh);
2223 }
2224
2225 return 0;
2226}
2227
2228static int netlink_dump(struct sock *sk)
2229{
2230 struct netlink_sock *nlk = nlk_sk(sk);
2231 struct netlink_ext_ack extack = {};
2232 struct netlink_callback *cb;
2233 struct sk_buff *skb = NULL;
2234 struct module *module;
2235 int err = -ENOBUFS;
2236 int alloc_min_size;
2237 int alloc_size;
2238
2239 mutex_lock(nlk->cb_mutex);
2240 if (!nlk->cb_running) {
2241 err = -EINVAL;
2242 goto errout_skb;
2243 }
2244
2245 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2246 goto errout_skb;
2247
2248 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2249 * required, but it makes sense to _attempt_ a 16K bytes allocation
2250 * to reduce number of system calls on dump operations, if user
2251 * ever provided a big enough buffer.
2252 */
2253 cb = &nlk->cb;
2254 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2255
2256 if (alloc_min_size < nlk->max_recvmsg_len) {
2257 alloc_size = nlk->max_recvmsg_len;
2258 skb = alloc_skb(alloc_size,
2259 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2260 __GFP_NOWARN | __GFP_NORETRY);
2261 }
2262 if (!skb) {
2263 alloc_size = alloc_min_size;
2264 skb = alloc_skb(alloc_size, GFP_KERNEL);
2265 }
2266 if (!skb)
2267 goto errout_skb;
2268
2269 /* Trim skb to allocated size. User is expected to provide buffer as
2270 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2271 * netlink_recvmsg())). dump will pack as many smaller messages as
2272 * could fit within the allocated skb. skb is typically allocated
2273 * with larger space than required (could be as much as near 2x the
2274 * requested size with align to next power of 2 approach). Allowing
2275 * dump to use the excess space makes it difficult for a user to have a
2276 * reasonable static buffer based on the expected largest dump of a
2277 * single netdev. The outcome is MSG_TRUNC error.
2278 */
2279 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2280 netlink_skb_set_owner_r(skb, sk);
2281
2282 if (nlk->dump_done_errno > 0) {
2283 cb->extack = &extack;
2284 nlk->dump_done_errno = cb->dump(skb, cb);
2285 cb->extack = NULL;
2286 }
2287
2288 if (nlk->dump_done_errno > 0 ||
2289 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2290 mutex_unlock(nlk->cb_mutex);
2291
2292 if (sk_filter(sk, skb))
2293 kfree_skb(skb);
2294 else
2295 __netlink_sendskb(sk, skb);
2296 return 0;
2297 }
2298
2299 if (netlink_dump_done(nlk, skb, cb, &extack))
2300 goto errout_skb;
2301
2302#ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2303 /* frag_list skb's data is used for compat tasks
2304 * and the regular skb's data for normal (non-compat) tasks.
2305 * See netlink_recvmsg().
2306 */
2307 if (unlikely(skb_shinfo(skb)->frag_list)) {
2308 if (netlink_dump_done(nlk, skb_shinfo(skb)->frag_list, cb, &extack))
2309 goto errout_skb;
2310 }
2311#endif
2312
2313 if (sk_filter(sk, skb))
2314 kfree_skb(skb);
2315 else
2316 __netlink_sendskb(sk, skb);
2317
2318 if (cb->done)
2319 cb->done(cb);
2320
2321 nlk->cb_running = false;
2322 module = cb->module;
2323 skb = cb->skb;
2324 mutex_unlock(nlk->cb_mutex);
2325 module_put(module);
2326 consume_skb(skb);
2327 return 0;
2328
2329errout_skb:
2330 mutex_unlock(nlk->cb_mutex);
2331 kfree_skb(skb);
2332 return err;
2333}
2334
2335int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2336 const struct nlmsghdr *nlh,
2337 struct netlink_dump_control *control)
2338{
2339 struct netlink_sock *nlk, *nlk2;
2340 struct netlink_callback *cb;
2341 struct sock *sk;
2342 int ret;
2343
2344 refcount_inc(&skb->users);
2345
2346 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2347 if (sk == NULL) {
2348 ret = -ECONNREFUSED;
2349 goto error_free;
2350 }
2351
2352 nlk = nlk_sk(sk);
2353 mutex_lock(nlk->cb_mutex);
2354 /* A dump is in progress... */
2355 if (nlk->cb_running) {
2356 ret = -EBUSY;
2357 goto error_unlock;
2358 }
2359 /* add reference of module which cb->dump belongs to */
2360 if (!try_module_get(control->module)) {
2361 ret = -EPROTONOSUPPORT;
2362 goto error_unlock;
2363 }
2364
2365 cb = &nlk->cb;
2366 memset(cb, 0, sizeof(*cb));
2367 cb->dump = control->dump;
2368 cb->done = control->done;
2369 cb->nlh = nlh;
2370 cb->data = control->data;
2371 cb->module = control->module;
2372 cb->min_dump_alloc = control->min_dump_alloc;
2373 cb->skb = skb;
2374
2375 nlk2 = nlk_sk(NETLINK_CB(skb).sk);
2376 cb->strict_check = !!(nlk2->flags & NETLINK_F_STRICT_CHK);
2377
2378 if (control->start) {
2379 ret = control->start(cb);
2380 if (ret)
2381 goto error_put;
2382 }
2383
2384 nlk->cb_running = true;
2385 nlk->dump_done_errno = INT_MAX;
2386
2387 mutex_unlock(nlk->cb_mutex);
2388
2389 ret = netlink_dump(sk);
2390
2391 sock_put(sk);
2392
2393 if (ret)
2394 return ret;
2395
2396 /* We successfully started a dump, by returning -EINTR we
2397 * signal not to send ACK even if it was requested.
2398 */
2399 return -EINTR;
2400
2401error_put:
2402 module_put(control->module);
2403error_unlock:
2404 sock_put(sk);
2405 mutex_unlock(nlk->cb_mutex);
2406error_free:
2407 kfree_skb(skb);
2408 return ret;
2409}
2410EXPORT_SYMBOL(__netlink_dump_start);
2411
2412void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2413 const struct netlink_ext_ack *extack)
2414{
2415 struct sk_buff *skb;
2416 struct nlmsghdr *rep;
2417 struct nlmsgerr *errmsg;
2418 size_t payload = sizeof(*errmsg);
2419 size_t tlvlen = 0;
2420 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2421 unsigned int flags = 0;
2422 bool nlk_has_extack = nlk->flags & NETLINK_F_EXT_ACK;
2423
2424 /* Error messages get the original request appened, unless the user
2425 * requests to cap the error message, and get extra error data if
2426 * requested.
2427 */
2428 if (nlk_has_extack && extack && extack->_msg)
2429 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2430
2431 if (err && !(nlk->flags & NETLINK_F_CAP_ACK))
2432 payload += nlmsg_len(nlh);
2433 else
2434 flags |= NLM_F_CAPPED;
2435 if (err && nlk_has_extack && extack && extack->bad_attr)
2436 tlvlen += nla_total_size(sizeof(u32));
2437 if (nlk_has_extack && extack && extack->cookie_len)
2438 tlvlen += nla_total_size(extack->cookie_len);
2439 if (err && nlk_has_extack && extack && extack->policy)
2440 tlvlen += netlink_policy_dump_attr_size_estimate(extack->policy);
2441
2442 if (tlvlen)
2443 flags |= NLM_F_ACK_TLVS;
2444
2445 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2446 if (!skb) {
2447 NETLINK_CB(in_skb).sk->sk_err = ENOBUFS;
2448 sk_error_report(NETLINK_CB(in_skb).sk);
2449 return;
2450 }
2451
2452 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2453 NLMSG_ERROR, payload, flags);
2454 errmsg = nlmsg_data(rep);
2455 errmsg->error = err;
2456 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2457
2458 if (nlk_has_extack && extack) {
2459 if (extack->_msg) {
2460 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG,
2461 extack->_msg));
2462 }
2463 if (err && extack->bad_attr &&
2464 !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2465 (u8 *)extack->bad_attr >= in_skb->data +
2466 in_skb->len))
2467 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2468 (u8 *)extack->bad_attr -
2469 (u8 *)nlh));
2470 if (extack->cookie_len)
2471 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2472 extack->cookie_len, extack->cookie));
2473 if (extack->policy)
2474 netlink_policy_dump_write_attr(skb, extack->policy,
2475 NLMSGERR_ATTR_POLICY);
2476 }
2477
2478 nlmsg_end(skb, rep);
2479
2480 nlmsg_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid);
2481}
2482EXPORT_SYMBOL(netlink_ack);
2483
2484int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2485 struct nlmsghdr *,
2486 struct netlink_ext_ack *))
2487{
2488 struct netlink_ext_ack extack;
2489 struct nlmsghdr *nlh;
2490 int err;
2491
2492 while (skb->len >= nlmsg_total_size(0)) {
2493 int msglen;
2494
2495 memset(&extack, 0, sizeof(extack));
2496 nlh = nlmsg_hdr(skb);
2497 err = 0;
2498
2499 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2500 return 0;
2501
2502 /* Only requests are handled by the kernel */
2503 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2504 goto ack;
2505
2506 /* Skip control messages */
2507 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2508 goto ack;
2509
2510 err = cb(skb, nlh, &extack);
2511 if (err == -EINTR)
2512 goto skip;
2513
2514ack:
2515 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2516 netlink_ack(skb, nlh, err, &extack);
2517
2518skip:
2519 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2520 if (msglen > skb->len)
2521 msglen = skb->len;
2522 skb_pull(skb, msglen);
2523 }
2524
2525 return 0;
2526}
2527EXPORT_SYMBOL(netlink_rcv_skb);
2528
2529/**
2530 * nlmsg_notify - send a notification netlink message
2531 * @sk: netlink socket to use
2532 * @skb: notification message
2533 * @portid: destination netlink portid for reports or 0
2534 * @group: destination multicast group or 0
2535 * @report: 1 to report back, 0 to disable
2536 * @flags: allocation flags
2537 */
2538int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2539 unsigned int group, int report, gfp_t flags)
2540{
2541 int err = 0;
2542
2543 if (group) {
2544 int exclude_portid = 0;
2545
2546 if (report) {
2547 refcount_inc(&skb->users);
2548 exclude_portid = portid;
2549 }
2550
2551 /* errors reported via destination sk->sk_err, but propagate
2552 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2553 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2554 if (err == -ESRCH)
2555 err = 0;
2556 }
2557
2558 if (report) {
2559 int err2;
2560
2561 err2 = nlmsg_unicast(sk, skb, portid);
2562 if (!err)
2563 err = err2;
2564 }
2565
2566 return err;
2567}
2568EXPORT_SYMBOL(nlmsg_notify);
2569
2570#ifdef CONFIG_PROC_FS
2571struct nl_seq_iter {
2572 struct seq_net_private p;
2573 struct rhashtable_iter hti;
2574 int link;
2575};
2576
2577static void netlink_walk_start(struct nl_seq_iter *iter)
2578{
2579 rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti);
2580 rhashtable_walk_start(&iter->hti);
2581}
2582
2583static void netlink_walk_stop(struct nl_seq_iter *iter)
2584{
2585 rhashtable_walk_stop(&iter->hti);
2586 rhashtable_walk_exit(&iter->hti);
2587}
2588
2589static void *__netlink_seq_next(struct seq_file *seq)
2590{
2591 struct nl_seq_iter *iter = seq->private;
2592 struct netlink_sock *nlk;
2593
2594 do {
2595 for (;;) {
2596 nlk = rhashtable_walk_next(&iter->hti);
2597
2598 if (IS_ERR(nlk)) {
2599 if (PTR_ERR(nlk) == -EAGAIN)
2600 continue;
2601
2602 return nlk;
2603 }
2604
2605 if (nlk)
2606 break;
2607
2608 netlink_walk_stop(iter);
2609 if (++iter->link >= MAX_LINKS)
2610 return NULL;
2611
2612 netlink_walk_start(iter);
2613 }
2614 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2615
2616 return nlk;
2617}
2618
2619static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2620 __acquires(RCU)
2621{
2622 struct nl_seq_iter *iter = seq->private;
2623 void *obj = SEQ_START_TOKEN;
2624 loff_t pos;
2625
2626 iter->link = 0;
2627
2628 netlink_walk_start(iter);
2629
2630 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2631 obj = __netlink_seq_next(seq);
2632
2633 return obj;
2634}
2635
2636static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2637{
2638 ++*pos;
2639 return __netlink_seq_next(seq);
2640}
2641
2642static void netlink_native_seq_stop(struct seq_file *seq, void *v)
2643{
2644 struct nl_seq_iter *iter = seq->private;
2645
2646 if (iter->link >= MAX_LINKS)
2647 return;
2648
2649 netlink_walk_stop(iter);
2650}
2651
2652
2653static int netlink_native_seq_show(struct seq_file *seq, void *v)
2654{
2655 if (v == SEQ_START_TOKEN) {
2656 seq_puts(seq,
2657 "sk Eth Pid Groups "
2658 "Rmem Wmem Dump Locks Drops Inode\n");
2659 } else {
2660 struct sock *s = v;
2661 struct netlink_sock *nlk = nlk_sk(s);
2662
2663 seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n",
2664 s,
2665 s->sk_protocol,
2666 nlk->portid,
2667 nlk->groups ? (u32)nlk->groups[0] : 0,
2668 sk_rmem_alloc_get(s),
2669 sk_wmem_alloc_get(s),
2670 nlk->cb_running,
2671 refcount_read(&s->sk_refcnt),
2672 atomic_read(&s->sk_drops),
2673 sock_i_ino(s)
2674 );
2675
2676 }
2677 return 0;
2678}
2679
2680#ifdef CONFIG_BPF_SYSCALL
2681struct bpf_iter__netlink {
2682 __bpf_md_ptr(struct bpf_iter_meta *, meta);
2683 __bpf_md_ptr(struct netlink_sock *, sk);
2684};
2685
2686DEFINE_BPF_ITER_FUNC(netlink, struct bpf_iter_meta *meta, struct netlink_sock *sk)
2687
2688static int netlink_prog_seq_show(struct bpf_prog *prog,
2689 struct bpf_iter_meta *meta,
2690 void *v)
2691{
2692 struct bpf_iter__netlink ctx;
2693
2694 meta->seq_num--; /* skip SEQ_START_TOKEN */
2695 ctx.meta = meta;
2696 ctx.sk = nlk_sk((struct sock *)v);
2697 return bpf_iter_run_prog(prog, &ctx);
2698}
2699
2700static int netlink_seq_show(struct seq_file *seq, void *v)
2701{
2702 struct bpf_iter_meta meta;
2703 struct bpf_prog *prog;
2704
2705 meta.seq = seq;
2706 prog = bpf_iter_get_info(&meta, false);
2707 if (!prog)
2708 return netlink_native_seq_show(seq, v);
2709
2710 if (v != SEQ_START_TOKEN)
2711 return netlink_prog_seq_show(prog, &meta, v);
2712
2713 return 0;
2714}
2715
2716static void netlink_seq_stop(struct seq_file *seq, void *v)
2717{
2718 struct bpf_iter_meta meta;
2719 struct bpf_prog *prog;
2720
2721 if (!v) {
2722 meta.seq = seq;
2723 prog = bpf_iter_get_info(&meta, true);
2724 if (prog)
2725 (void)netlink_prog_seq_show(prog, &meta, v);
2726 }
2727
2728 netlink_native_seq_stop(seq, v);
2729}
2730#else
2731static int netlink_seq_show(struct seq_file *seq, void *v)
2732{
2733 return netlink_native_seq_show(seq, v);
2734}
2735
2736static void netlink_seq_stop(struct seq_file *seq, void *v)
2737{
2738 netlink_native_seq_stop(seq, v);
2739}
2740#endif
2741
2742static const struct seq_operations netlink_seq_ops = {
2743 .start = netlink_seq_start,
2744 .next = netlink_seq_next,
2745 .stop = netlink_seq_stop,
2746 .show = netlink_seq_show,
2747};
2748#endif
2749
2750int netlink_register_notifier(struct notifier_block *nb)
2751{
2752 return blocking_notifier_chain_register(&netlink_chain, nb);
2753}
2754EXPORT_SYMBOL(netlink_register_notifier);
2755
2756int netlink_unregister_notifier(struct notifier_block *nb)
2757{
2758 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2759}
2760EXPORT_SYMBOL(netlink_unregister_notifier);
2761
2762static const struct proto_ops netlink_ops = {
2763 .family = PF_NETLINK,
2764 .owner = THIS_MODULE,
2765 .release = netlink_release,
2766 .bind = netlink_bind,
2767 .connect = netlink_connect,
2768 .socketpair = sock_no_socketpair,
2769 .accept = sock_no_accept,
2770 .getname = netlink_getname,
2771 .poll = datagram_poll,
2772 .ioctl = netlink_ioctl,
2773 .listen = sock_no_listen,
2774 .shutdown = sock_no_shutdown,
2775 .setsockopt = netlink_setsockopt,
2776 .getsockopt = netlink_getsockopt,
2777 .sendmsg = netlink_sendmsg,
2778 .recvmsg = netlink_recvmsg,
2779 .mmap = sock_no_mmap,
2780 .sendpage = sock_no_sendpage,
2781};
2782
2783static const struct net_proto_family netlink_family_ops = {
2784 .family = PF_NETLINK,
2785 .create = netlink_create,
2786 .owner = THIS_MODULE, /* for consistency 8) */
2787};
2788
2789static int __net_init netlink_net_init(struct net *net)
2790{
2791#ifdef CONFIG_PROC_FS
2792 if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2793 sizeof(struct nl_seq_iter)))
2794 return -ENOMEM;
2795#endif
2796 return 0;
2797}
2798
2799static void __net_exit netlink_net_exit(struct net *net)
2800{
2801#ifdef CONFIG_PROC_FS
2802 remove_proc_entry("netlink", net->proc_net);
2803#endif
2804}
2805
2806static void __init netlink_add_usersock_entry(void)
2807{
2808 struct listeners *listeners;
2809 int groups = 32;
2810
2811 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2812 if (!listeners)
2813 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2814
2815 netlink_table_grab();
2816
2817 nl_table[NETLINK_USERSOCK].groups = groups;
2818 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2819 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2820 nl_table[NETLINK_USERSOCK].registered = 1;
2821 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2822
2823 netlink_table_ungrab();
2824}
2825
2826static struct pernet_operations __net_initdata netlink_net_ops = {
2827 .init = netlink_net_init,
2828 .exit = netlink_net_exit,
2829};
2830
2831static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2832{
2833 const struct netlink_sock *nlk = data;
2834 struct netlink_compare_arg arg;
2835
2836 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2837 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2838}
2839
2840static const struct rhashtable_params netlink_rhashtable_params = {
2841 .head_offset = offsetof(struct netlink_sock, node),
2842 .key_len = netlink_compare_arg_len,
2843 .obj_hashfn = netlink_hash,
2844 .obj_cmpfn = netlink_compare,
2845 .automatic_shrinking = true,
2846};
2847
2848#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2849BTF_ID_LIST(btf_netlink_sock_id)
2850BTF_ID(struct, netlink_sock)
2851
2852static const struct bpf_iter_seq_info netlink_seq_info = {
2853 .seq_ops = &netlink_seq_ops,
2854 .init_seq_private = bpf_iter_init_seq_net,
2855 .fini_seq_private = bpf_iter_fini_seq_net,
2856 .seq_priv_size = sizeof(struct nl_seq_iter),
2857};
2858
2859static struct bpf_iter_reg netlink_reg_info = {
2860 .target = "netlink",
2861 .ctx_arg_info_size = 1,
2862 .ctx_arg_info = {
2863 { offsetof(struct bpf_iter__netlink, sk),
2864 PTR_TO_BTF_ID_OR_NULL },
2865 },
2866 .seq_info = &netlink_seq_info,
2867};
2868
2869static int __init bpf_iter_register(void)
2870{
2871 netlink_reg_info.ctx_arg_info[0].btf_id = *btf_netlink_sock_id;
2872 return bpf_iter_reg_target(&netlink_reg_info);
2873}
2874#endif
2875
2876static int __init netlink_proto_init(void)
2877{
2878 int i;
2879 int err = proto_register(&netlink_proto, 0);
2880
2881 if (err != 0)
2882 goto out;
2883
2884#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2885 err = bpf_iter_register();
2886 if (err)
2887 goto out;
2888#endif
2889
2890 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof_field(struct sk_buff, cb));
2891
2892 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2893 if (!nl_table)
2894 goto panic;
2895
2896 for (i = 0; i < MAX_LINKS; i++) {
2897 if (rhashtable_init(&nl_table[i].hash,
2898 &netlink_rhashtable_params) < 0) {
2899 while (--i > 0)
2900 rhashtable_destroy(&nl_table[i].hash);
2901 kfree(nl_table);
2902 goto panic;
2903 }
2904 }
2905
2906 netlink_add_usersock_entry();
2907
2908 sock_register(&netlink_family_ops);
2909 register_pernet_subsys(&netlink_net_ops);
2910 register_pernet_subsys(&netlink_tap_net_ops);
2911 /* The netlink device handler may be needed early. */
2912 rtnetlink_init();
2913out:
2914 return err;
2915panic:
2916 panic("netlink_init: Cannot allocate nl_table\n");
2917}
2918
2919core_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);
135static void netlink_destroy_callback(struct netlink_callback *cb);
136
137static DEFINE_RWLOCK(nl_table_lock);
138static atomic_t nl_table_users = ATOMIC_INIT(0);
139
140static ATOMIC_NOTIFIER_HEAD(netlink_chain);
141
142static u32 netlink_group_mask(u32 group)
143{
144 return group ? 1 << (group - 1) : 0;
145}
146
147static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
148{
149 return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
150}
151
152static void netlink_sock_destruct(struct sock *sk)
153{
154 struct netlink_sock *nlk = nlk_sk(sk);
155
156 if (nlk->cb) {
157 if (nlk->cb->done)
158 nlk->cb->done(nlk->cb);
159 netlink_destroy_callback(nlk->cb);
160 }
161
162 skb_queue_purge(&sk->sk_receive_queue);
163
164 if (!sock_flag(sk, SOCK_DEAD)) {
165 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
166 return;
167 }
168
169 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
170 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
171 WARN_ON(nlk_sk(sk)->groups);
172}
173
174/* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
175 * SMP. Look, when several writers sleep and reader wakes them up, all but one
176 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
177 * this, _but_ remember, it adds useless work on UP machines.
178 */
179
180void netlink_table_grab(void)
181 __acquires(nl_table_lock)
182{
183 might_sleep();
184
185 write_lock_irq(&nl_table_lock);
186
187 if (atomic_read(&nl_table_users)) {
188 DECLARE_WAITQUEUE(wait, current);
189
190 add_wait_queue_exclusive(&nl_table_wait, &wait);
191 for (;;) {
192 set_current_state(TASK_UNINTERRUPTIBLE);
193 if (atomic_read(&nl_table_users) == 0)
194 break;
195 write_unlock_irq(&nl_table_lock);
196 schedule();
197 write_lock_irq(&nl_table_lock);
198 }
199
200 __set_current_state(TASK_RUNNING);
201 remove_wait_queue(&nl_table_wait, &wait);
202 }
203}
204
205void netlink_table_ungrab(void)
206 __releases(nl_table_lock)
207{
208 write_unlock_irq(&nl_table_lock);
209 wake_up(&nl_table_wait);
210}
211
212static inline void
213netlink_lock_table(void)
214{
215 /* read_lock() synchronizes us to netlink_table_grab */
216
217 read_lock(&nl_table_lock);
218 atomic_inc(&nl_table_users);
219 read_unlock(&nl_table_lock);
220}
221
222static inline void
223netlink_unlock_table(void)
224{
225 if (atomic_dec_and_test(&nl_table_users))
226 wake_up(&nl_table_wait);
227}
228
229static inline struct sock *netlink_lookup(struct net *net, int protocol,
230 u32 pid)
231{
232 struct nl_pid_hash *hash = &nl_table[protocol].hash;
233 struct hlist_head *head;
234 struct sock *sk;
235 struct hlist_node *node;
236
237 read_lock(&nl_table_lock);
238 head = nl_pid_hashfn(hash, pid);
239 sk_for_each(sk, node, head) {
240 if (net_eq(sock_net(sk), net) && (nlk_sk(sk)->pid == pid)) {
241 sock_hold(sk);
242 goto found;
243 }
244 }
245 sk = NULL;
246found:
247 read_unlock(&nl_table_lock);
248 return sk;
249}
250
251static inline struct hlist_head *nl_pid_hash_zalloc(size_t size)
252{
253 if (size <= PAGE_SIZE)
254 return kzalloc(size, GFP_ATOMIC);
255 else
256 return (struct hlist_head *)
257 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
258 get_order(size));
259}
260
261static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
262{
263 if (size <= PAGE_SIZE)
264 kfree(table);
265 else
266 free_pages((unsigned long)table, get_order(size));
267}
268
269static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
270{
271 unsigned int omask, mask, shift;
272 size_t osize, size;
273 struct hlist_head *otable, *table;
274 int i;
275
276 omask = mask = hash->mask;
277 osize = size = (mask + 1) * sizeof(*table);
278 shift = hash->shift;
279
280 if (grow) {
281 if (++shift > hash->max_shift)
282 return 0;
283 mask = mask * 2 + 1;
284 size *= 2;
285 }
286
287 table = nl_pid_hash_zalloc(size);
288 if (!table)
289 return 0;
290
291 otable = hash->table;
292 hash->table = table;
293 hash->mask = mask;
294 hash->shift = shift;
295 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
296
297 for (i = 0; i <= omask; i++) {
298 struct sock *sk;
299 struct hlist_node *node, *tmp;
300
301 sk_for_each_safe(sk, node, tmp, &otable[i])
302 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
303 }
304
305 nl_pid_hash_free(otable, osize);
306 hash->rehash_time = jiffies + 10 * 60 * HZ;
307 return 1;
308}
309
310static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
311{
312 int avg = hash->entries >> hash->shift;
313
314 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
315 return 1;
316
317 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
318 nl_pid_hash_rehash(hash, 0);
319 return 1;
320 }
321
322 return 0;
323}
324
325static const struct proto_ops netlink_ops;
326
327static void
328netlink_update_listeners(struct sock *sk)
329{
330 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
331 struct hlist_node *node;
332 unsigned long mask;
333 unsigned int i;
334
335 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
336 mask = 0;
337 sk_for_each_bound(sk, node, &tbl->mc_list) {
338 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
339 mask |= nlk_sk(sk)->groups[i];
340 }
341 tbl->listeners->masks[i] = mask;
342 }
343 /* this function is only called with the netlink table "grabbed", which
344 * makes sure updates are visible before bind or setsockopt return. */
345}
346
347static int netlink_insert(struct sock *sk, struct net *net, u32 pid)
348{
349 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
350 struct hlist_head *head;
351 int err = -EADDRINUSE;
352 struct sock *osk;
353 struct hlist_node *node;
354 int len;
355
356 netlink_table_grab();
357 head = nl_pid_hashfn(hash, pid);
358 len = 0;
359 sk_for_each(osk, node, head) {
360 if (net_eq(sock_net(osk), net) && (nlk_sk(osk)->pid == pid))
361 break;
362 len++;
363 }
364 if (node)
365 goto err;
366
367 err = -EBUSY;
368 if (nlk_sk(sk)->pid)
369 goto err;
370
371 err = -ENOMEM;
372 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
373 goto err;
374
375 if (len && nl_pid_hash_dilute(hash, len))
376 head = nl_pid_hashfn(hash, pid);
377 hash->entries++;
378 nlk_sk(sk)->pid = pid;
379 sk_add_node(sk, head);
380 err = 0;
381
382err:
383 netlink_table_ungrab();
384 return err;
385}
386
387static void netlink_remove(struct sock *sk)
388{
389 netlink_table_grab();
390 if (sk_del_node_init(sk))
391 nl_table[sk->sk_protocol].hash.entries--;
392 if (nlk_sk(sk)->subscriptions)
393 __sk_del_bind_node(sk);
394 netlink_table_ungrab();
395}
396
397static struct proto netlink_proto = {
398 .name = "NETLINK",
399 .owner = THIS_MODULE,
400 .obj_size = sizeof(struct netlink_sock),
401};
402
403static int __netlink_create(struct net *net, struct socket *sock,
404 struct mutex *cb_mutex, int protocol)
405{
406 struct sock *sk;
407 struct netlink_sock *nlk;
408
409 sock->ops = &netlink_ops;
410
411 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
412 if (!sk)
413 return -ENOMEM;
414
415 sock_init_data(sock, sk);
416
417 nlk = nlk_sk(sk);
418 if (cb_mutex)
419 nlk->cb_mutex = cb_mutex;
420 else {
421 nlk->cb_mutex = &nlk->cb_def_mutex;
422 mutex_init(nlk->cb_mutex);
423 }
424 init_waitqueue_head(&nlk->wait);
425
426 sk->sk_destruct = netlink_sock_destruct;
427 sk->sk_protocol = protocol;
428 return 0;
429}
430
431static int netlink_create(struct net *net, struct socket *sock, int protocol,
432 int kern)
433{
434 struct module *module = NULL;
435 struct mutex *cb_mutex;
436 struct netlink_sock *nlk;
437 int err = 0;
438
439 sock->state = SS_UNCONNECTED;
440
441 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
442 return -ESOCKTNOSUPPORT;
443
444 if (protocol < 0 || protocol >= MAX_LINKS)
445 return -EPROTONOSUPPORT;
446
447 netlink_lock_table();
448#ifdef CONFIG_MODULES
449 if (!nl_table[protocol].registered) {
450 netlink_unlock_table();
451 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
452 netlink_lock_table();
453 }
454#endif
455 if (nl_table[protocol].registered &&
456 try_module_get(nl_table[protocol].module))
457 module = nl_table[protocol].module;
458 else
459 err = -EPROTONOSUPPORT;
460 cb_mutex = nl_table[protocol].cb_mutex;
461 netlink_unlock_table();
462
463 if (err < 0)
464 goto out;
465
466 err = __netlink_create(net, sock, cb_mutex, protocol);
467 if (err < 0)
468 goto out_module;
469
470 local_bh_disable();
471 sock_prot_inuse_add(net, &netlink_proto, 1);
472 local_bh_enable();
473
474 nlk = nlk_sk(sock->sk);
475 nlk->module = module;
476out:
477 return err;
478
479out_module:
480 module_put(module);
481 goto out;
482}
483
484static int netlink_release(struct socket *sock)
485{
486 struct sock *sk = sock->sk;
487 struct netlink_sock *nlk;
488
489 if (!sk)
490 return 0;
491
492 netlink_remove(sk);
493 sock_orphan(sk);
494 nlk = nlk_sk(sk);
495
496 /*
497 * OK. Socket is unlinked, any packets that arrive now
498 * will be purged.
499 */
500
501 sock->sk = NULL;
502 wake_up_interruptible_all(&nlk->wait);
503
504 skb_queue_purge(&sk->sk_write_queue);
505
506 if (nlk->pid) {
507 struct netlink_notify n = {
508 .net = sock_net(sk),
509 .protocol = sk->sk_protocol,
510 .pid = nlk->pid,
511 };
512 atomic_notifier_call_chain(&netlink_chain,
513 NETLINK_URELEASE, &n);
514 }
515
516 module_put(nlk->module);
517
518 netlink_table_grab();
519 if (netlink_is_kernel(sk)) {
520 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
521 if (--nl_table[sk->sk_protocol].registered == 0) {
522 kfree(nl_table[sk->sk_protocol].listeners);
523 nl_table[sk->sk_protocol].module = NULL;
524 nl_table[sk->sk_protocol].registered = 0;
525 }
526 } else if (nlk->subscriptions)
527 netlink_update_listeners(sk);
528 netlink_table_ungrab();
529
530 kfree(nlk->groups);
531 nlk->groups = NULL;
532
533 local_bh_disable();
534 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
535 local_bh_enable();
536 sock_put(sk);
537 return 0;
538}
539
540static int netlink_autobind(struct socket *sock)
541{
542 struct sock *sk = sock->sk;
543 struct net *net = sock_net(sk);
544 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
545 struct hlist_head *head;
546 struct sock *osk;
547 struct hlist_node *node;
548 s32 pid = task_tgid_vnr(current);
549 int err;
550 static s32 rover = -4097;
551
552retry:
553 cond_resched();
554 netlink_table_grab();
555 head = nl_pid_hashfn(hash, pid);
556 sk_for_each(osk, node, head) {
557 if (!net_eq(sock_net(osk), net))
558 continue;
559 if (nlk_sk(osk)->pid == pid) {
560 /* Bind collision, search negative pid values. */
561 pid = rover--;
562 if (rover > -4097)
563 rover = -4097;
564 netlink_table_ungrab();
565 goto retry;
566 }
567 }
568 netlink_table_ungrab();
569
570 err = netlink_insert(sk, net, pid);
571 if (err == -EADDRINUSE)
572 goto retry;
573
574 /* If 2 threads race to autobind, that is fine. */
575 if (err == -EBUSY)
576 err = 0;
577
578 return err;
579}
580
581static inline int netlink_capable(struct socket *sock, unsigned int flag)
582{
583 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
584 capable(CAP_NET_ADMIN);
585}
586
587static void
588netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
589{
590 struct netlink_sock *nlk = nlk_sk(sk);
591
592 if (nlk->subscriptions && !subscriptions)
593 __sk_del_bind_node(sk);
594 else if (!nlk->subscriptions && subscriptions)
595 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
596 nlk->subscriptions = subscriptions;
597}
598
599static int netlink_realloc_groups(struct sock *sk)
600{
601 struct netlink_sock *nlk = nlk_sk(sk);
602 unsigned int groups;
603 unsigned long *new_groups;
604 int err = 0;
605
606 netlink_table_grab();
607
608 groups = nl_table[sk->sk_protocol].groups;
609 if (!nl_table[sk->sk_protocol].registered) {
610 err = -ENOENT;
611 goto out_unlock;
612 }
613
614 if (nlk->ngroups >= groups)
615 goto out_unlock;
616
617 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
618 if (new_groups == NULL) {
619 err = -ENOMEM;
620 goto out_unlock;
621 }
622 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
623 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
624
625 nlk->groups = new_groups;
626 nlk->ngroups = groups;
627 out_unlock:
628 netlink_table_ungrab();
629 return err;
630}
631
632static int netlink_bind(struct socket *sock, struct sockaddr *addr,
633 int addr_len)
634{
635 struct sock *sk = sock->sk;
636 struct net *net = sock_net(sk);
637 struct netlink_sock *nlk = nlk_sk(sk);
638 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
639 int err;
640
641 if (nladdr->nl_family != AF_NETLINK)
642 return -EINVAL;
643
644 /* Only superuser is allowed to listen multicasts */
645 if (nladdr->nl_groups) {
646 if (!netlink_capable(sock, NL_NONROOT_RECV))
647 return -EPERM;
648 err = netlink_realloc_groups(sk);
649 if (err)
650 return err;
651 }
652
653 if (nlk->pid) {
654 if (nladdr->nl_pid != nlk->pid)
655 return -EINVAL;
656 } else {
657 err = nladdr->nl_pid ?
658 netlink_insert(sk, net, nladdr->nl_pid) :
659 netlink_autobind(sock);
660 if (err)
661 return err;
662 }
663
664 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
665 return 0;
666
667 netlink_table_grab();
668 netlink_update_subscriptions(sk, nlk->subscriptions +
669 hweight32(nladdr->nl_groups) -
670 hweight32(nlk->groups[0]));
671 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
672 netlink_update_listeners(sk);
673 netlink_table_ungrab();
674
675 return 0;
676}
677
678static int netlink_connect(struct socket *sock, struct sockaddr *addr,
679 int alen, int flags)
680{
681 int err = 0;
682 struct sock *sk = sock->sk;
683 struct netlink_sock *nlk = nlk_sk(sk);
684 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
685
686 if (alen < sizeof(addr->sa_family))
687 return -EINVAL;
688
689 if (addr->sa_family == AF_UNSPEC) {
690 sk->sk_state = NETLINK_UNCONNECTED;
691 nlk->dst_pid = 0;
692 nlk->dst_group = 0;
693 return 0;
694 }
695 if (addr->sa_family != AF_NETLINK)
696 return -EINVAL;
697
698 /* Only superuser is allowed to send multicasts */
699 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
700 return -EPERM;
701
702 if (!nlk->pid)
703 err = netlink_autobind(sock);
704
705 if (err == 0) {
706 sk->sk_state = NETLINK_CONNECTED;
707 nlk->dst_pid = nladdr->nl_pid;
708 nlk->dst_group = ffs(nladdr->nl_groups);
709 }
710
711 return err;
712}
713
714static int netlink_getname(struct socket *sock, struct sockaddr *addr,
715 int *addr_len, int peer)
716{
717 struct sock *sk = sock->sk;
718 struct netlink_sock *nlk = nlk_sk(sk);
719 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
720
721 nladdr->nl_family = AF_NETLINK;
722 nladdr->nl_pad = 0;
723 *addr_len = sizeof(*nladdr);
724
725 if (peer) {
726 nladdr->nl_pid = nlk->dst_pid;
727 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
728 } else {
729 nladdr->nl_pid = nlk->pid;
730 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
731 }
732 return 0;
733}
734
735static void netlink_overrun(struct sock *sk)
736{
737 struct netlink_sock *nlk = nlk_sk(sk);
738
739 if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) {
740 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
741 sk->sk_err = ENOBUFS;
742 sk->sk_error_report(sk);
743 }
744 }
745 atomic_inc(&sk->sk_drops);
746}
747
748static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
749{
750 struct sock *sock;
751 struct netlink_sock *nlk;
752
753 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, pid);
754 if (!sock)
755 return ERR_PTR(-ECONNREFUSED);
756
757 /* Don't bother queuing skb if kernel socket has no input function */
758 nlk = nlk_sk(sock);
759 if (sock->sk_state == NETLINK_CONNECTED &&
760 nlk->dst_pid != nlk_sk(ssk)->pid) {
761 sock_put(sock);
762 return ERR_PTR(-ECONNREFUSED);
763 }
764 return sock;
765}
766
767struct sock *netlink_getsockbyfilp(struct file *filp)
768{
769 struct inode *inode = filp->f_path.dentry->d_inode;
770 struct sock *sock;
771
772 if (!S_ISSOCK(inode->i_mode))
773 return ERR_PTR(-ENOTSOCK);
774
775 sock = SOCKET_I(inode)->sk;
776 if (sock->sk_family != AF_NETLINK)
777 return ERR_PTR(-EINVAL);
778
779 sock_hold(sock);
780 return sock;
781}
782
783/*
784 * Attach a skb to a netlink socket.
785 * The caller must hold a reference to the destination socket. On error, the
786 * reference is dropped. The skb is not send to the destination, just all
787 * all error checks are performed and memory in the queue is reserved.
788 * Return values:
789 * < 0: error. skb freed, reference to sock dropped.
790 * 0: continue
791 * 1: repeat lookup - reference dropped while waiting for socket memory.
792 */
793int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
794 long *timeo, struct sock *ssk)
795{
796 struct netlink_sock *nlk;
797
798 nlk = nlk_sk(sk);
799
800 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
801 test_bit(0, &nlk->state)) {
802 DECLARE_WAITQUEUE(wait, current);
803 if (!*timeo) {
804 if (!ssk || netlink_is_kernel(ssk))
805 netlink_overrun(sk);
806 sock_put(sk);
807 kfree_skb(skb);
808 return -EAGAIN;
809 }
810
811 __set_current_state(TASK_INTERRUPTIBLE);
812 add_wait_queue(&nlk->wait, &wait);
813
814 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
815 test_bit(0, &nlk->state)) &&
816 !sock_flag(sk, SOCK_DEAD))
817 *timeo = schedule_timeout(*timeo);
818
819 __set_current_state(TASK_RUNNING);
820 remove_wait_queue(&nlk->wait, &wait);
821 sock_put(sk);
822
823 if (signal_pending(current)) {
824 kfree_skb(skb);
825 return sock_intr_errno(*timeo);
826 }
827 return 1;
828 }
829 skb_set_owner_r(skb, sk);
830 return 0;
831}
832
833int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
834{
835 int len = skb->len;
836
837 skb_queue_tail(&sk->sk_receive_queue, skb);
838 sk->sk_data_ready(sk, len);
839 sock_put(sk);
840 return len;
841}
842
843void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
844{
845 kfree_skb(skb);
846 sock_put(sk);
847}
848
849static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
850 gfp_t allocation)
851{
852 int delta;
853
854 skb_orphan(skb);
855
856 delta = skb->end - skb->tail;
857 if (delta * 2 < skb->truesize)
858 return skb;
859
860 if (skb_shared(skb)) {
861 struct sk_buff *nskb = skb_clone(skb, allocation);
862 if (!nskb)
863 return skb;
864 kfree_skb(skb);
865 skb = nskb;
866 }
867
868 if (!pskb_expand_head(skb, 0, -delta, allocation))
869 skb->truesize -= delta;
870
871 return skb;
872}
873
874static inline void netlink_rcv_wake(struct sock *sk)
875{
876 struct netlink_sock *nlk = nlk_sk(sk);
877
878 if (skb_queue_empty(&sk->sk_receive_queue))
879 clear_bit(0, &nlk->state);
880 if (!test_bit(0, &nlk->state))
881 wake_up_interruptible(&nlk->wait);
882}
883
884static inline int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb)
885{
886 int ret;
887 struct netlink_sock *nlk = nlk_sk(sk);
888
889 ret = -ECONNREFUSED;
890 if (nlk->netlink_rcv != NULL) {
891 ret = skb->len;
892 skb_set_owner_r(skb, sk);
893 nlk->netlink_rcv(skb);
894 }
895 kfree_skb(skb);
896 sock_put(sk);
897 return ret;
898}
899
900int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
901 u32 pid, int nonblock)
902{
903 struct sock *sk;
904 int err;
905 long timeo;
906
907 skb = netlink_trim(skb, gfp_any());
908
909 timeo = sock_sndtimeo(ssk, nonblock);
910retry:
911 sk = netlink_getsockbypid(ssk, pid);
912 if (IS_ERR(sk)) {
913 kfree_skb(skb);
914 return PTR_ERR(sk);
915 }
916 if (netlink_is_kernel(sk))
917 return netlink_unicast_kernel(sk, skb);
918
919 if (sk_filter(sk, skb)) {
920 err = skb->len;
921 kfree_skb(skb);
922 sock_put(sk);
923 return err;
924 }
925
926 err = netlink_attachskb(sk, skb, &timeo, ssk);
927 if (err == 1)
928 goto retry;
929 if (err)
930 return err;
931
932 return netlink_sendskb(sk, skb);
933}
934EXPORT_SYMBOL(netlink_unicast);
935
936int netlink_has_listeners(struct sock *sk, unsigned int group)
937{
938 int res = 0;
939 struct listeners *listeners;
940
941 BUG_ON(!netlink_is_kernel(sk));
942
943 rcu_read_lock();
944 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
945
946 if (group - 1 < nl_table[sk->sk_protocol].groups)
947 res = test_bit(group - 1, listeners->masks);
948
949 rcu_read_unlock();
950
951 return res;
952}
953EXPORT_SYMBOL_GPL(netlink_has_listeners);
954
955static inline int netlink_broadcast_deliver(struct sock *sk,
956 struct sk_buff *skb)
957{
958 struct netlink_sock *nlk = nlk_sk(sk);
959
960 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
961 !test_bit(0, &nlk->state)) {
962 skb_set_owner_r(skb, sk);
963 skb_queue_tail(&sk->sk_receive_queue, skb);
964 sk->sk_data_ready(sk, skb->len);
965 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
966 }
967 return -1;
968}
969
970struct netlink_broadcast_data {
971 struct sock *exclude_sk;
972 struct net *net;
973 u32 pid;
974 u32 group;
975 int failure;
976 int delivery_failure;
977 int congested;
978 int delivered;
979 gfp_t allocation;
980 struct sk_buff *skb, *skb2;
981 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
982 void *tx_data;
983};
984
985static inline int do_one_broadcast(struct sock *sk,
986 struct netlink_broadcast_data *p)
987{
988 struct netlink_sock *nlk = nlk_sk(sk);
989 int val;
990
991 if (p->exclude_sk == sk)
992 goto out;
993
994 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
995 !test_bit(p->group - 1, nlk->groups))
996 goto out;
997
998 if (!net_eq(sock_net(sk), p->net))
999 goto out;
1000
1001 if (p->failure) {
1002 netlink_overrun(sk);
1003 goto out;
1004 }
1005
1006 sock_hold(sk);
1007 if (p->skb2 == NULL) {
1008 if (skb_shared(p->skb)) {
1009 p->skb2 = skb_clone(p->skb, p->allocation);
1010 } else {
1011 p->skb2 = skb_get(p->skb);
1012 /*
1013 * skb ownership may have been set when
1014 * delivered to a previous socket.
1015 */
1016 skb_orphan(p->skb2);
1017 }
1018 }
1019 if (p->skb2 == NULL) {
1020 netlink_overrun(sk);
1021 /* Clone failed. Notify ALL listeners. */
1022 p->failure = 1;
1023 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1024 p->delivery_failure = 1;
1025 } else if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1026 kfree_skb(p->skb2);
1027 p->skb2 = NULL;
1028 } else if (sk_filter(sk, p->skb2)) {
1029 kfree_skb(p->skb2);
1030 p->skb2 = NULL;
1031 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1032 netlink_overrun(sk);
1033 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1034 p->delivery_failure = 1;
1035 } else {
1036 p->congested |= val;
1037 p->delivered = 1;
1038 p->skb2 = NULL;
1039 }
1040 sock_put(sk);
1041
1042out:
1043 return 0;
1044}
1045
1046int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 pid,
1047 u32 group, gfp_t allocation,
1048 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1049 void *filter_data)
1050{
1051 struct net *net = sock_net(ssk);
1052 struct netlink_broadcast_data info;
1053 struct hlist_node *node;
1054 struct sock *sk;
1055
1056 skb = netlink_trim(skb, allocation);
1057
1058 info.exclude_sk = ssk;
1059 info.net = net;
1060 info.pid = pid;
1061 info.group = group;
1062 info.failure = 0;
1063 info.delivery_failure = 0;
1064 info.congested = 0;
1065 info.delivered = 0;
1066 info.allocation = allocation;
1067 info.skb = skb;
1068 info.skb2 = NULL;
1069 info.tx_filter = filter;
1070 info.tx_data = filter_data;
1071
1072 /* While we sleep in clone, do not allow to change socket list */
1073
1074 netlink_lock_table();
1075
1076 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1077 do_one_broadcast(sk, &info);
1078
1079 consume_skb(skb);
1080
1081 netlink_unlock_table();
1082
1083 if (info.delivery_failure) {
1084 kfree_skb(info.skb2);
1085 return -ENOBUFS;
1086 } else
1087 consume_skb(info.skb2);
1088
1089 if (info.delivered) {
1090 if (info.congested && (allocation & __GFP_WAIT))
1091 yield();
1092 return 0;
1093 }
1094 return -ESRCH;
1095}
1096EXPORT_SYMBOL(netlink_broadcast_filtered);
1097
1098int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
1099 u32 group, gfp_t allocation)
1100{
1101 return netlink_broadcast_filtered(ssk, skb, pid, group, allocation,
1102 NULL, NULL);
1103}
1104EXPORT_SYMBOL(netlink_broadcast);
1105
1106struct netlink_set_err_data {
1107 struct sock *exclude_sk;
1108 u32 pid;
1109 u32 group;
1110 int code;
1111};
1112
1113static inline int do_one_set_err(struct sock *sk,
1114 struct netlink_set_err_data *p)
1115{
1116 struct netlink_sock *nlk = nlk_sk(sk);
1117 int ret = 0;
1118
1119 if (sk == p->exclude_sk)
1120 goto out;
1121
1122 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1123 goto out;
1124
1125 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1126 !test_bit(p->group - 1, nlk->groups))
1127 goto out;
1128
1129 if (p->code == ENOBUFS && nlk->flags & NETLINK_RECV_NO_ENOBUFS) {
1130 ret = 1;
1131 goto out;
1132 }
1133
1134 sk->sk_err = p->code;
1135 sk->sk_error_report(sk);
1136out:
1137 return ret;
1138}
1139
1140/**
1141 * netlink_set_err - report error to broadcast listeners
1142 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1143 * @pid: the PID of a process that we want to skip (if any)
1144 * @groups: the broadcast group that will notice the error
1145 * @code: error code, must be negative (as usual in kernelspace)
1146 *
1147 * This function returns the number of broadcast listeners that have set the
1148 * NETLINK_RECV_NO_ENOBUFS socket option.
1149 */
1150int netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
1151{
1152 struct netlink_set_err_data info;
1153 struct hlist_node *node;
1154 struct sock *sk;
1155 int ret = 0;
1156
1157 info.exclude_sk = ssk;
1158 info.pid = pid;
1159 info.group = group;
1160 /* sk->sk_err wants a positive error value */
1161 info.code = -code;
1162
1163 read_lock(&nl_table_lock);
1164
1165 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1166 ret += do_one_set_err(sk, &info);
1167
1168 read_unlock(&nl_table_lock);
1169 return ret;
1170}
1171EXPORT_SYMBOL(netlink_set_err);
1172
1173/* must be called with netlink table grabbed */
1174static void netlink_update_socket_mc(struct netlink_sock *nlk,
1175 unsigned int group,
1176 int is_new)
1177{
1178 int old, new = !!is_new, subscriptions;
1179
1180 old = test_bit(group - 1, nlk->groups);
1181 subscriptions = nlk->subscriptions - old + new;
1182 if (new)
1183 __set_bit(group - 1, nlk->groups);
1184 else
1185 __clear_bit(group - 1, nlk->groups);
1186 netlink_update_subscriptions(&nlk->sk, subscriptions);
1187 netlink_update_listeners(&nlk->sk);
1188}
1189
1190static int netlink_setsockopt(struct socket *sock, int level, int optname,
1191 char __user *optval, unsigned int optlen)
1192{
1193 struct sock *sk = sock->sk;
1194 struct netlink_sock *nlk = nlk_sk(sk);
1195 unsigned int val = 0;
1196 int err;
1197
1198 if (level != SOL_NETLINK)
1199 return -ENOPROTOOPT;
1200
1201 if (optlen >= sizeof(int) &&
1202 get_user(val, (unsigned int __user *)optval))
1203 return -EFAULT;
1204
1205 switch (optname) {
1206 case NETLINK_PKTINFO:
1207 if (val)
1208 nlk->flags |= NETLINK_RECV_PKTINFO;
1209 else
1210 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1211 err = 0;
1212 break;
1213 case NETLINK_ADD_MEMBERSHIP:
1214 case NETLINK_DROP_MEMBERSHIP: {
1215 if (!netlink_capable(sock, NL_NONROOT_RECV))
1216 return -EPERM;
1217 err = netlink_realloc_groups(sk);
1218 if (err)
1219 return err;
1220 if (!val || val - 1 >= nlk->ngroups)
1221 return -EINVAL;
1222 netlink_table_grab();
1223 netlink_update_socket_mc(nlk, val,
1224 optname == NETLINK_ADD_MEMBERSHIP);
1225 netlink_table_ungrab();
1226 err = 0;
1227 break;
1228 }
1229 case NETLINK_BROADCAST_ERROR:
1230 if (val)
1231 nlk->flags |= NETLINK_BROADCAST_SEND_ERROR;
1232 else
1233 nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR;
1234 err = 0;
1235 break;
1236 case NETLINK_NO_ENOBUFS:
1237 if (val) {
1238 nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
1239 clear_bit(0, &nlk->state);
1240 wake_up_interruptible(&nlk->wait);
1241 } else
1242 nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
1243 err = 0;
1244 break;
1245 default:
1246 err = -ENOPROTOOPT;
1247 }
1248 return err;
1249}
1250
1251static int netlink_getsockopt(struct socket *sock, int level, int optname,
1252 char __user *optval, int __user *optlen)
1253{
1254 struct sock *sk = sock->sk;
1255 struct netlink_sock *nlk = nlk_sk(sk);
1256 int len, val, err;
1257
1258 if (level != SOL_NETLINK)
1259 return -ENOPROTOOPT;
1260
1261 if (get_user(len, optlen))
1262 return -EFAULT;
1263 if (len < 0)
1264 return -EINVAL;
1265
1266 switch (optname) {
1267 case NETLINK_PKTINFO:
1268 if (len < sizeof(int))
1269 return -EINVAL;
1270 len = sizeof(int);
1271 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1272 if (put_user(len, optlen) ||
1273 put_user(val, optval))
1274 return -EFAULT;
1275 err = 0;
1276 break;
1277 case NETLINK_BROADCAST_ERROR:
1278 if (len < sizeof(int))
1279 return -EINVAL;
1280 len = sizeof(int);
1281 val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0;
1282 if (put_user(len, optlen) ||
1283 put_user(val, optval))
1284 return -EFAULT;
1285 err = 0;
1286 break;
1287 case NETLINK_NO_ENOBUFS:
1288 if (len < sizeof(int))
1289 return -EINVAL;
1290 len = sizeof(int);
1291 val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0;
1292 if (put_user(len, optlen) ||
1293 put_user(val, optval))
1294 return -EFAULT;
1295 err = 0;
1296 break;
1297 default:
1298 err = -ENOPROTOOPT;
1299 }
1300 return err;
1301}
1302
1303static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1304{
1305 struct nl_pktinfo info;
1306
1307 info.group = NETLINK_CB(skb).dst_group;
1308 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1309}
1310
1311static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1312 struct msghdr *msg, size_t len)
1313{
1314 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1315 struct sock *sk = sock->sk;
1316 struct netlink_sock *nlk = nlk_sk(sk);
1317 struct sockaddr_nl *addr = msg->msg_name;
1318 u32 dst_pid;
1319 u32 dst_group;
1320 struct sk_buff *skb;
1321 int err;
1322 struct scm_cookie scm;
1323
1324 if (msg->msg_flags&MSG_OOB)
1325 return -EOPNOTSUPP;
1326
1327 if (NULL == siocb->scm) {
1328 siocb->scm = &scm;
1329 memset(&scm, 0, sizeof(scm));
1330 }
1331 err = scm_send(sock, msg, siocb->scm);
1332 if (err < 0)
1333 return err;
1334
1335 if (msg->msg_namelen) {
1336 err = -EINVAL;
1337 if (addr->nl_family != AF_NETLINK)
1338 goto out;
1339 dst_pid = addr->nl_pid;
1340 dst_group = ffs(addr->nl_groups);
1341 err = -EPERM;
1342 if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1343 goto out;
1344 } else {
1345 dst_pid = nlk->dst_pid;
1346 dst_group = nlk->dst_group;
1347 }
1348
1349 if (!nlk->pid) {
1350 err = netlink_autobind(sock);
1351 if (err)
1352 goto out;
1353 }
1354
1355 err = -EMSGSIZE;
1356 if (len > sk->sk_sndbuf - 32)
1357 goto out;
1358 err = -ENOBUFS;
1359 skb = alloc_skb(len, GFP_KERNEL);
1360 if (skb == NULL)
1361 goto out;
1362
1363 NETLINK_CB(skb).pid = nlk->pid;
1364 NETLINK_CB(skb).dst_group = dst_group;
1365 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1366
1367 err = -EFAULT;
1368 if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
1369 kfree_skb(skb);
1370 goto out;
1371 }
1372
1373 err = security_netlink_send(sk, skb);
1374 if (err) {
1375 kfree_skb(skb);
1376 goto out;
1377 }
1378
1379 if (dst_group) {
1380 atomic_inc(&skb->users);
1381 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1382 }
1383 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1384
1385out:
1386 scm_destroy(siocb->scm);
1387 return err;
1388}
1389
1390static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1391 struct msghdr *msg, size_t len,
1392 int flags)
1393{
1394 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1395 struct scm_cookie scm;
1396 struct sock *sk = sock->sk;
1397 struct netlink_sock *nlk = nlk_sk(sk);
1398 int noblock = flags&MSG_DONTWAIT;
1399 size_t copied;
1400 struct sk_buff *skb, *data_skb;
1401 int err, ret;
1402
1403 if (flags&MSG_OOB)
1404 return -EOPNOTSUPP;
1405
1406 copied = 0;
1407
1408 skb = skb_recv_datagram(sk, flags, noblock, &err);
1409 if (skb == NULL)
1410 goto out;
1411
1412 data_skb = skb;
1413
1414#ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1415 if (unlikely(skb_shinfo(skb)->frag_list)) {
1416 /*
1417 * If this skb has a frag_list, then here that means that we
1418 * will have to use the frag_list skb's data for compat tasks
1419 * and the regular skb's data for normal (non-compat) tasks.
1420 *
1421 * If we need to send the compat skb, assign it to the
1422 * 'data_skb' variable so that it will be used below for data
1423 * copying. We keep 'skb' for everything else, including
1424 * freeing both later.
1425 */
1426 if (flags & MSG_CMSG_COMPAT)
1427 data_skb = skb_shinfo(skb)->frag_list;
1428 }
1429#endif
1430
1431 msg->msg_namelen = 0;
1432
1433 copied = data_skb->len;
1434 if (len < copied) {
1435 msg->msg_flags |= MSG_TRUNC;
1436 copied = len;
1437 }
1438
1439 skb_reset_transport_header(data_skb);
1440 err = skb_copy_datagram_iovec(data_skb, 0, msg->msg_iov, copied);
1441
1442 if (msg->msg_name) {
1443 struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
1444 addr->nl_family = AF_NETLINK;
1445 addr->nl_pad = 0;
1446 addr->nl_pid = NETLINK_CB(skb).pid;
1447 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1448 msg->msg_namelen = sizeof(*addr);
1449 }
1450
1451 if (nlk->flags & NETLINK_RECV_PKTINFO)
1452 netlink_cmsg_recv_pktinfo(msg, skb);
1453
1454 if (NULL == siocb->scm) {
1455 memset(&scm, 0, sizeof(scm));
1456 siocb->scm = &scm;
1457 }
1458 siocb->scm->creds = *NETLINK_CREDS(skb);
1459 if (flags & MSG_TRUNC)
1460 copied = data_skb->len;
1461
1462 skb_free_datagram(sk, skb);
1463
1464 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1465 ret = netlink_dump(sk);
1466 if (ret) {
1467 sk->sk_err = ret;
1468 sk->sk_error_report(sk);
1469 }
1470 }
1471
1472 scm_recv(sock, msg, siocb->scm, flags);
1473out:
1474 netlink_rcv_wake(sk);
1475 return err ? : copied;
1476}
1477
1478static void netlink_data_ready(struct sock *sk, int len)
1479{
1480 BUG();
1481}
1482
1483/*
1484 * We export these functions to other modules. They provide a
1485 * complete set of kernel non-blocking support for message
1486 * queueing.
1487 */
1488
1489struct sock *
1490netlink_kernel_create(struct net *net, int unit, unsigned int groups,
1491 void (*input)(struct sk_buff *skb),
1492 struct mutex *cb_mutex, struct module *module)
1493{
1494 struct socket *sock;
1495 struct sock *sk;
1496 struct netlink_sock *nlk;
1497 struct listeners *listeners = NULL;
1498
1499 BUG_ON(!nl_table);
1500
1501 if (unit < 0 || unit >= MAX_LINKS)
1502 return NULL;
1503
1504 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1505 return NULL;
1506
1507 /*
1508 * We have to just have a reference on the net from sk, but don't
1509 * get_net it. Besides, we cannot get and then put the net here.
1510 * So we create one inside init_net and the move it to net.
1511 */
1512
1513 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
1514 goto out_sock_release_nosk;
1515
1516 sk = sock->sk;
1517 sk_change_net(sk, net);
1518
1519 if (groups < 32)
1520 groups = 32;
1521
1522 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
1523 if (!listeners)
1524 goto out_sock_release;
1525
1526 sk->sk_data_ready = netlink_data_ready;
1527 if (input)
1528 nlk_sk(sk)->netlink_rcv = input;
1529
1530 if (netlink_insert(sk, net, 0))
1531 goto out_sock_release;
1532
1533 nlk = nlk_sk(sk);
1534 nlk->flags |= NETLINK_KERNEL_SOCKET;
1535
1536 netlink_table_grab();
1537 if (!nl_table[unit].registered) {
1538 nl_table[unit].groups = groups;
1539 rcu_assign_pointer(nl_table[unit].listeners, listeners);
1540 nl_table[unit].cb_mutex = cb_mutex;
1541 nl_table[unit].module = module;
1542 nl_table[unit].registered = 1;
1543 } else {
1544 kfree(listeners);
1545 nl_table[unit].registered++;
1546 }
1547 netlink_table_ungrab();
1548 return sk;
1549
1550out_sock_release:
1551 kfree(listeners);
1552 netlink_kernel_release(sk);
1553 return NULL;
1554
1555out_sock_release_nosk:
1556 sock_release(sock);
1557 return NULL;
1558}
1559EXPORT_SYMBOL(netlink_kernel_create);
1560
1561
1562void
1563netlink_kernel_release(struct sock *sk)
1564{
1565 sk_release_kernel(sk);
1566}
1567EXPORT_SYMBOL(netlink_kernel_release);
1568
1569int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
1570{
1571 struct listeners *new, *old;
1572 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1573
1574 if (groups < 32)
1575 groups = 32;
1576
1577 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1578 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
1579 if (!new)
1580 return -ENOMEM;
1581 old = rcu_dereference_raw(tbl->listeners);
1582 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
1583 rcu_assign_pointer(tbl->listeners, new);
1584
1585 kfree_rcu(old, rcu);
1586 }
1587 tbl->groups = groups;
1588
1589 return 0;
1590}
1591
1592/**
1593 * netlink_change_ngroups - change number of multicast groups
1594 *
1595 * This changes the number of multicast groups that are available
1596 * on a certain netlink family. Note that it is not possible to
1597 * change the number of groups to below 32. Also note that it does
1598 * not implicitly call netlink_clear_multicast_users() when the
1599 * number of groups is reduced.
1600 *
1601 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1602 * @groups: The new number of groups.
1603 */
1604int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1605{
1606 int err;
1607
1608 netlink_table_grab();
1609 err = __netlink_change_ngroups(sk, groups);
1610 netlink_table_ungrab();
1611
1612 return err;
1613}
1614
1615void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1616{
1617 struct sock *sk;
1618 struct hlist_node *node;
1619 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1620
1621 sk_for_each_bound(sk, node, &tbl->mc_list)
1622 netlink_update_socket_mc(nlk_sk(sk), group, 0);
1623}
1624
1625/**
1626 * netlink_clear_multicast_users - kick off multicast listeners
1627 *
1628 * This function removes all listeners from the given group.
1629 * @ksk: The kernel netlink socket, as returned by
1630 * netlink_kernel_create().
1631 * @group: The multicast group to clear.
1632 */
1633void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1634{
1635 netlink_table_grab();
1636 __netlink_clear_multicast_users(ksk, group);
1637 netlink_table_ungrab();
1638}
1639
1640void netlink_set_nonroot(int protocol, unsigned int flags)
1641{
1642 if ((unsigned int)protocol < MAX_LINKS)
1643 nl_table[protocol].nl_nonroot = flags;
1644}
1645EXPORT_SYMBOL(netlink_set_nonroot);
1646
1647static void netlink_destroy_callback(struct netlink_callback *cb)
1648{
1649 kfree_skb(cb->skb);
1650 kfree(cb);
1651}
1652
1653/*
1654 * It looks a bit ugly.
1655 * It would be better to create kernel thread.
1656 */
1657
1658static int netlink_dump(struct sock *sk)
1659{
1660 struct netlink_sock *nlk = nlk_sk(sk);
1661 struct netlink_callback *cb;
1662 struct sk_buff *skb = NULL;
1663 struct nlmsghdr *nlh;
1664 int len, err = -ENOBUFS;
1665 int alloc_size;
1666
1667 mutex_lock(nlk->cb_mutex);
1668
1669 cb = nlk->cb;
1670 if (cb == NULL) {
1671 err = -EINVAL;
1672 goto errout_skb;
1673 }
1674
1675 alloc_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
1676
1677 skb = sock_rmalloc(sk, alloc_size, 0, GFP_KERNEL);
1678 if (!skb)
1679 goto errout_skb;
1680
1681 len = cb->dump(skb, cb);
1682
1683 if (len > 0) {
1684 mutex_unlock(nlk->cb_mutex);
1685
1686 if (sk_filter(sk, skb))
1687 kfree_skb(skb);
1688 else {
1689 skb_queue_tail(&sk->sk_receive_queue, skb);
1690 sk->sk_data_ready(sk, skb->len);
1691 }
1692 return 0;
1693 }
1694
1695 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1696 if (!nlh)
1697 goto errout_skb;
1698
1699 nl_dump_check_consistent(cb, nlh);
1700
1701 memcpy(nlmsg_data(nlh), &len, sizeof(len));
1702
1703 if (sk_filter(sk, skb))
1704 kfree_skb(skb);
1705 else {
1706 skb_queue_tail(&sk->sk_receive_queue, skb);
1707 sk->sk_data_ready(sk, skb->len);
1708 }
1709
1710 if (cb->done)
1711 cb->done(cb);
1712 nlk->cb = NULL;
1713 mutex_unlock(nlk->cb_mutex);
1714
1715 netlink_destroy_callback(cb);
1716 return 0;
1717
1718errout_skb:
1719 mutex_unlock(nlk->cb_mutex);
1720 kfree_skb(skb);
1721 return err;
1722}
1723
1724int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1725 const struct nlmsghdr *nlh,
1726 int (*dump)(struct sk_buff *skb,
1727 struct netlink_callback *),
1728 int (*done)(struct netlink_callback *),
1729 u16 min_dump_alloc)
1730{
1731 struct netlink_callback *cb;
1732 struct sock *sk;
1733 struct netlink_sock *nlk;
1734 int ret;
1735
1736 cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1737 if (cb == NULL)
1738 return -ENOBUFS;
1739
1740 cb->dump = dump;
1741 cb->done = done;
1742 cb->nlh = nlh;
1743 cb->min_dump_alloc = min_dump_alloc;
1744 atomic_inc(&skb->users);
1745 cb->skb = skb;
1746
1747 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).pid);
1748 if (sk == NULL) {
1749 netlink_destroy_callback(cb);
1750 return -ECONNREFUSED;
1751 }
1752 nlk = nlk_sk(sk);
1753 /* A dump is in progress... */
1754 mutex_lock(nlk->cb_mutex);
1755 if (nlk->cb) {
1756 mutex_unlock(nlk->cb_mutex);
1757 netlink_destroy_callback(cb);
1758 sock_put(sk);
1759 return -EBUSY;
1760 }
1761 nlk->cb = cb;
1762 mutex_unlock(nlk->cb_mutex);
1763
1764 ret = netlink_dump(sk);
1765
1766 sock_put(sk);
1767
1768 if (ret)
1769 return ret;
1770
1771 /* We successfully started a dump, by returning -EINTR we
1772 * signal not to send ACK even if it was requested.
1773 */
1774 return -EINTR;
1775}
1776EXPORT_SYMBOL(netlink_dump_start);
1777
1778void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1779{
1780 struct sk_buff *skb;
1781 struct nlmsghdr *rep;
1782 struct nlmsgerr *errmsg;
1783 size_t payload = sizeof(*errmsg);
1784
1785 /* error messages get the original request appened */
1786 if (err)
1787 payload += nlmsg_len(nlh);
1788
1789 skb = nlmsg_new(payload, GFP_KERNEL);
1790 if (!skb) {
1791 struct sock *sk;
1792
1793 sk = netlink_lookup(sock_net(in_skb->sk),
1794 in_skb->sk->sk_protocol,
1795 NETLINK_CB(in_skb).pid);
1796 if (sk) {
1797 sk->sk_err = ENOBUFS;
1798 sk->sk_error_report(sk);
1799 sock_put(sk);
1800 }
1801 return;
1802 }
1803
1804 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1805 NLMSG_ERROR, payload, 0);
1806 errmsg = nlmsg_data(rep);
1807 errmsg->error = err;
1808 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1809 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1810}
1811EXPORT_SYMBOL(netlink_ack);
1812
1813int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1814 struct nlmsghdr *))
1815{
1816 struct nlmsghdr *nlh;
1817 int err;
1818
1819 while (skb->len >= nlmsg_total_size(0)) {
1820 int msglen;
1821
1822 nlh = nlmsg_hdr(skb);
1823 err = 0;
1824
1825 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1826 return 0;
1827
1828 /* Only requests are handled by the kernel */
1829 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1830 goto ack;
1831
1832 /* Skip control messages */
1833 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1834 goto ack;
1835
1836 err = cb(skb, nlh);
1837 if (err == -EINTR)
1838 goto skip;
1839
1840ack:
1841 if (nlh->nlmsg_flags & NLM_F_ACK || err)
1842 netlink_ack(skb, nlh, err);
1843
1844skip:
1845 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1846 if (msglen > skb->len)
1847 msglen = skb->len;
1848 skb_pull(skb, msglen);
1849 }
1850
1851 return 0;
1852}
1853EXPORT_SYMBOL(netlink_rcv_skb);
1854
1855/**
1856 * nlmsg_notify - send a notification netlink message
1857 * @sk: netlink socket to use
1858 * @skb: notification message
1859 * @pid: destination netlink pid for reports or 0
1860 * @group: destination multicast group or 0
1861 * @report: 1 to report back, 0 to disable
1862 * @flags: allocation flags
1863 */
1864int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1865 unsigned int group, int report, gfp_t flags)
1866{
1867 int err = 0;
1868
1869 if (group) {
1870 int exclude_pid = 0;
1871
1872 if (report) {
1873 atomic_inc(&skb->users);
1874 exclude_pid = pid;
1875 }
1876
1877 /* errors reported via destination sk->sk_err, but propagate
1878 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
1879 err = nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1880 }
1881
1882 if (report) {
1883 int err2;
1884
1885 err2 = nlmsg_unicast(sk, skb, pid);
1886 if (!err || err == -ESRCH)
1887 err = err2;
1888 }
1889
1890 return err;
1891}
1892EXPORT_SYMBOL(nlmsg_notify);
1893
1894#ifdef CONFIG_PROC_FS
1895struct nl_seq_iter {
1896 struct seq_net_private p;
1897 int link;
1898 int hash_idx;
1899};
1900
1901static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1902{
1903 struct nl_seq_iter *iter = seq->private;
1904 int i, j;
1905 struct sock *s;
1906 struct hlist_node *node;
1907 loff_t off = 0;
1908
1909 for (i = 0; i < MAX_LINKS; i++) {
1910 struct nl_pid_hash *hash = &nl_table[i].hash;
1911
1912 for (j = 0; j <= hash->mask; j++) {
1913 sk_for_each(s, node, &hash->table[j]) {
1914 if (sock_net(s) != seq_file_net(seq))
1915 continue;
1916 if (off == pos) {
1917 iter->link = i;
1918 iter->hash_idx = j;
1919 return s;
1920 }
1921 ++off;
1922 }
1923 }
1924 }
1925 return NULL;
1926}
1927
1928static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1929 __acquires(nl_table_lock)
1930{
1931 read_lock(&nl_table_lock);
1932 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1933}
1934
1935static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1936{
1937 struct sock *s;
1938 struct nl_seq_iter *iter;
1939 int i, j;
1940
1941 ++*pos;
1942
1943 if (v == SEQ_START_TOKEN)
1944 return netlink_seq_socket_idx(seq, 0);
1945
1946 iter = seq->private;
1947 s = v;
1948 do {
1949 s = sk_next(s);
1950 } while (s && sock_net(s) != seq_file_net(seq));
1951 if (s)
1952 return s;
1953
1954 i = iter->link;
1955 j = iter->hash_idx + 1;
1956
1957 do {
1958 struct nl_pid_hash *hash = &nl_table[i].hash;
1959
1960 for (; j <= hash->mask; j++) {
1961 s = sk_head(&hash->table[j]);
1962 while (s && sock_net(s) != seq_file_net(seq))
1963 s = sk_next(s);
1964 if (s) {
1965 iter->link = i;
1966 iter->hash_idx = j;
1967 return s;
1968 }
1969 }
1970
1971 j = 0;
1972 } while (++i < MAX_LINKS);
1973
1974 return NULL;
1975}
1976
1977static void netlink_seq_stop(struct seq_file *seq, void *v)
1978 __releases(nl_table_lock)
1979{
1980 read_unlock(&nl_table_lock);
1981}
1982
1983
1984static int netlink_seq_show(struct seq_file *seq, void *v)
1985{
1986 if (v == SEQ_START_TOKEN)
1987 seq_puts(seq,
1988 "sk Eth Pid Groups "
1989 "Rmem Wmem Dump Locks Drops Inode\n");
1990 else {
1991 struct sock *s = v;
1992 struct netlink_sock *nlk = nlk_sk(s);
1993
1994 seq_printf(seq, "%pK %-3d %-6d %08x %-8d %-8d %pK %-8d %-8d %-8lu\n",
1995 s,
1996 s->sk_protocol,
1997 nlk->pid,
1998 nlk->groups ? (u32)nlk->groups[0] : 0,
1999 sk_rmem_alloc_get(s),
2000 sk_wmem_alloc_get(s),
2001 nlk->cb,
2002 atomic_read(&s->sk_refcnt),
2003 atomic_read(&s->sk_drops),
2004 sock_i_ino(s)
2005 );
2006
2007 }
2008 return 0;
2009}
2010
2011static const struct seq_operations netlink_seq_ops = {
2012 .start = netlink_seq_start,
2013 .next = netlink_seq_next,
2014 .stop = netlink_seq_stop,
2015 .show = netlink_seq_show,
2016};
2017
2018
2019static int netlink_seq_open(struct inode *inode, struct file *file)
2020{
2021 return seq_open_net(inode, file, &netlink_seq_ops,
2022 sizeof(struct nl_seq_iter));
2023}
2024
2025static const struct file_operations netlink_seq_fops = {
2026 .owner = THIS_MODULE,
2027 .open = netlink_seq_open,
2028 .read = seq_read,
2029 .llseek = seq_lseek,
2030 .release = seq_release_net,
2031};
2032
2033#endif
2034
2035int netlink_register_notifier(struct notifier_block *nb)
2036{
2037 return atomic_notifier_chain_register(&netlink_chain, nb);
2038}
2039EXPORT_SYMBOL(netlink_register_notifier);
2040
2041int netlink_unregister_notifier(struct notifier_block *nb)
2042{
2043 return atomic_notifier_chain_unregister(&netlink_chain, nb);
2044}
2045EXPORT_SYMBOL(netlink_unregister_notifier);
2046
2047static const struct proto_ops netlink_ops = {
2048 .family = PF_NETLINK,
2049 .owner = THIS_MODULE,
2050 .release = netlink_release,
2051 .bind = netlink_bind,
2052 .connect = netlink_connect,
2053 .socketpair = sock_no_socketpair,
2054 .accept = sock_no_accept,
2055 .getname = netlink_getname,
2056 .poll = datagram_poll,
2057 .ioctl = sock_no_ioctl,
2058 .listen = sock_no_listen,
2059 .shutdown = sock_no_shutdown,
2060 .setsockopt = netlink_setsockopt,
2061 .getsockopt = netlink_getsockopt,
2062 .sendmsg = netlink_sendmsg,
2063 .recvmsg = netlink_recvmsg,
2064 .mmap = sock_no_mmap,
2065 .sendpage = sock_no_sendpage,
2066};
2067
2068static const struct net_proto_family netlink_family_ops = {
2069 .family = PF_NETLINK,
2070 .create = netlink_create,
2071 .owner = THIS_MODULE, /* for consistency 8) */
2072};
2073
2074static int __net_init netlink_net_init(struct net *net)
2075{
2076#ifdef CONFIG_PROC_FS
2077 if (!proc_net_fops_create(net, "netlink", 0, &netlink_seq_fops))
2078 return -ENOMEM;
2079#endif
2080 return 0;
2081}
2082
2083static void __net_exit netlink_net_exit(struct net *net)
2084{
2085#ifdef CONFIG_PROC_FS
2086 proc_net_remove(net, "netlink");
2087#endif
2088}
2089
2090static void __init netlink_add_usersock_entry(void)
2091{
2092 struct listeners *listeners;
2093 int groups = 32;
2094
2095 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2096 if (!listeners)
2097 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2098
2099 netlink_table_grab();
2100
2101 nl_table[NETLINK_USERSOCK].groups = groups;
2102 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2103 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2104 nl_table[NETLINK_USERSOCK].registered = 1;
2105
2106 netlink_table_ungrab();
2107}
2108
2109static struct pernet_operations __net_initdata netlink_net_ops = {
2110 .init = netlink_net_init,
2111 .exit = netlink_net_exit,
2112};
2113
2114static int __init netlink_proto_init(void)
2115{
2116 struct sk_buff *dummy_skb;
2117 int i;
2118 unsigned long limit;
2119 unsigned int order;
2120 int err = proto_register(&netlink_proto, 0);
2121
2122 if (err != 0)
2123 goto out;
2124
2125 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
2126
2127 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2128 if (!nl_table)
2129 goto panic;
2130
2131 if (totalram_pages >= (128 * 1024))
2132 limit = totalram_pages >> (21 - PAGE_SHIFT);
2133 else
2134 limit = totalram_pages >> (23 - PAGE_SHIFT);
2135
2136 order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
2137 limit = (1UL << order) / sizeof(struct hlist_head);
2138 order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
2139
2140 for (i = 0; i < MAX_LINKS; i++) {
2141 struct nl_pid_hash *hash = &nl_table[i].hash;
2142
2143 hash->table = nl_pid_hash_zalloc(1 * sizeof(*hash->table));
2144 if (!hash->table) {
2145 while (i-- > 0)
2146 nl_pid_hash_free(nl_table[i].hash.table,
2147 1 * sizeof(*hash->table));
2148 kfree(nl_table);
2149 goto panic;
2150 }
2151 hash->max_shift = order;
2152 hash->shift = 0;
2153 hash->mask = 0;
2154 hash->rehash_time = jiffies;
2155 }
2156
2157 netlink_add_usersock_entry();
2158
2159 sock_register(&netlink_family_ops);
2160 register_pernet_subsys(&netlink_net_ops);
2161 /* The netlink device handler may be needed early. */
2162 rtnetlink_init();
2163out:
2164 return err;
2165panic:
2166 panic("netlink_init: Cannot allocate nl_table\n");
2167}
2168
2169core_initcall(netlink_proto_init);