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