Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * NETLINK Kernel-user communication protocol.
4 *
5 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
6 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
7 * Patrick McHardy <kaber@trash.net>
8 *
9 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
10 * added netlink_proto_exit
11 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
12 * use nlk_sk, as sk->protinfo is on a diet 8)
13 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
14 * - inc module use count of module that owns
15 * the kernel socket in case userspace opens
16 * socket of same protocol
17 * - remove all module support, since netlink is
18 * mandatory if CONFIG_NET=y these days
19 */
20
21#include <linux/module.h>
22
23#include <linux/bpf.h>
24#include <linux/capability.h>
25#include <linux/kernel.h>
26#include <linux/filter.h>
27#include <linux/init.h>
28#include <linux/signal.h>
29#include <linux/sched.h>
30#include <linux/errno.h>
31#include <linux/string.h>
32#include <linux/stat.h>
33#include <linux/socket.h>
34#include <linux/un.h>
35#include <linux/fcntl.h>
36#include <linux/termios.h>
37#include <linux/sockios.h>
38#include <linux/net.h>
39#include <linux/fs.h>
40#include <linux/slab.h>
41#include <linux/uaccess.h>
42#include <linux/skbuff.h>
43#include <linux/netdevice.h>
44#include <linux/rtnetlink.h>
45#include <linux/proc_fs.h>
46#include <linux/seq_file.h>
47#include <linux/notifier.h>
48#include <linux/security.h>
49#include <linux/jhash.h>
50#include <linux/jiffies.h>
51#include <linux/random.h>
52#include <linux/bitops.h>
53#include <linux/mm.h>
54#include <linux/types.h>
55#include <linux/audit.h>
56#include <linux/mutex.h>
57#include <linux/vmalloc.h>
58#include <linux/if_arp.h>
59#include <linux/rhashtable.h>
60#include <asm/cacheflush.h>
61#include <linux/hash.h>
62#include <linux/genetlink.h>
63#include <linux/net_namespace.h>
64#include <linux/nospec.h>
65#include <linux/btf_ids.h>
66
67#include <net/net_namespace.h>
68#include <net/netns/generic.h>
69#include <net/sock.h>
70#include <net/scm.h>
71#include <net/netlink.h>
72#define CREATE_TRACE_POINTS
73#include <trace/events/netlink.h>
74
75#include "af_netlink.h"
76
77struct listeners {
78 struct rcu_head rcu;
79 unsigned long masks[];
80};
81
82/* state bits */
83#define NETLINK_S_CONGESTED 0x0
84
85static inline int netlink_is_kernel(struct sock *sk)
86{
87 return nlk_test_bit(KERNEL_SOCKET, sk);
88}
89
90struct netlink_table *nl_table __read_mostly;
91EXPORT_SYMBOL_GPL(nl_table);
92
93static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
94
95static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
96
97static const char *const nlk_cb_mutex_key_strings[MAX_LINKS + 1] = {
98 "nlk_cb_mutex-ROUTE",
99 "nlk_cb_mutex-1",
100 "nlk_cb_mutex-USERSOCK",
101 "nlk_cb_mutex-FIREWALL",
102 "nlk_cb_mutex-SOCK_DIAG",
103 "nlk_cb_mutex-NFLOG",
104 "nlk_cb_mutex-XFRM",
105 "nlk_cb_mutex-SELINUX",
106 "nlk_cb_mutex-ISCSI",
107 "nlk_cb_mutex-AUDIT",
108 "nlk_cb_mutex-FIB_LOOKUP",
109 "nlk_cb_mutex-CONNECTOR",
110 "nlk_cb_mutex-NETFILTER",
111 "nlk_cb_mutex-IP6_FW",
112 "nlk_cb_mutex-DNRTMSG",
113 "nlk_cb_mutex-KOBJECT_UEVENT",
114 "nlk_cb_mutex-GENERIC",
115 "nlk_cb_mutex-17",
116 "nlk_cb_mutex-SCSITRANSPORT",
117 "nlk_cb_mutex-ECRYPTFS",
118 "nlk_cb_mutex-RDMA",
119 "nlk_cb_mutex-CRYPTO",
120 "nlk_cb_mutex-SMC",
121 "nlk_cb_mutex-23",
122 "nlk_cb_mutex-24",
123 "nlk_cb_mutex-25",
124 "nlk_cb_mutex-26",
125 "nlk_cb_mutex-27",
126 "nlk_cb_mutex-28",
127 "nlk_cb_mutex-29",
128 "nlk_cb_mutex-30",
129 "nlk_cb_mutex-31",
130 "nlk_cb_mutex-MAX_LINKS"
131};
132
133static int netlink_dump(struct sock *sk);
134
135/* nl_table locking explained:
136 * Lookup and traversal are protected with an RCU read-side lock. Insertion
137 * and removal are protected with per bucket lock while using RCU list
138 * modification primitives and may run in parallel to RCU protected lookups.
139 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
140 * been acquired * either during or after the socket has been removed from
141 * the list and after an RCU grace period.
142 */
143DEFINE_RWLOCK(nl_table_lock);
144EXPORT_SYMBOL_GPL(nl_table_lock);
145static atomic_t nl_table_users = ATOMIC_INIT(0);
146
147#define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
148
149static BLOCKING_NOTIFIER_HEAD(netlink_chain);
150
151
152static const struct rhashtable_params netlink_rhashtable_params;
153
154void do_trace_netlink_extack(const char *msg)
155{
156 trace_netlink_extack(msg);
157}
158EXPORT_SYMBOL(do_trace_netlink_extack);
159
160static inline u32 netlink_group_mask(u32 group)
161{
162 if (group > 32)
163 return 0;
164 return group ? 1 << (group - 1) : 0;
165}
166
167static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
168 gfp_t gfp_mask)
169{
170 unsigned int len = skb->len;
171 struct sk_buff *new;
172
173 new = alloc_skb(len, gfp_mask);
174 if (new == NULL)
175 return NULL;
176
177 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
178 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
179 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
180
181 skb_put_data(new, skb->data, len);
182 return new;
183}
184
185static unsigned int netlink_tap_net_id;
186
187struct netlink_tap_net {
188 struct list_head netlink_tap_all;
189 struct mutex netlink_tap_lock;
190};
191
192int netlink_add_tap(struct netlink_tap *nt)
193{
194 struct net *net = dev_net(nt->dev);
195 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
196
197 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
198 return -EINVAL;
199
200 mutex_lock(&nn->netlink_tap_lock);
201 list_add_rcu(&nt->list, &nn->netlink_tap_all);
202 mutex_unlock(&nn->netlink_tap_lock);
203
204 __module_get(nt->module);
205
206 return 0;
207}
208EXPORT_SYMBOL_GPL(netlink_add_tap);
209
210static int __netlink_remove_tap(struct netlink_tap *nt)
211{
212 struct net *net = dev_net(nt->dev);
213 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
214 bool found = false;
215 struct netlink_tap *tmp;
216
217 mutex_lock(&nn->netlink_tap_lock);
218
219 list_for_each_entry(tmp, &nn->netlink_tap_all, list) {
220 if (nt == tmp) {
221 list_del_rcu(&nt->list);
222 found = true;
223 goto out;
224 }
225 }
226
227 pr_warn("__netlink_remove_tap: %p not found\n", nt);
228out:
229 mutex_unlock(&nn->netlink_tap_lock);
230
231 if (found)
232 module_put(nt->module);
233
234 return found ? 0 : -ENODEV;
235}
236
237int netlink_remove_tap(struct netlink_tap *nt)
238{
239 int ret;
240
241 ret = __netlink_remove_tap(nt);
242 synchronize_net();
243
244 return ret;
245}
246EXPORT_SYMBOL_GPL(netlink_remove_tap);
247
248static __net_init int netlink_tap_init_net(struct net *net)
249{
250 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
251
252 INIT_LIST_HEAD(&nn->netlink_tap_all);
253 mutex_init(&nn->netlink_tap_lock);
254 return 0;
255}
256
257static struct pernet_operations netlink_tap_net_ops = {
258 .init = netlink_tap_init_net,
259 .id = &netlink_tap_net_id,
260 .size = sizeof(struct netlink_tap_net),
261};
262
263static bool netlink_filter_tap(const struct sk_buff *skb)
264{
265 struct sock *sk = skb->sk;
266
267 /* We take the more conservative approach and
268 * whitelist socket protocols that may pass.
269 */
270 switch (sk->sk_protocol) {
271 case NETLINK_ROUTE:
272 case NETLINK_USERSOCK:
273 case NETLINK_SOCK_DIAG:
274 case NETLINK_NFLOG:
275 case NETLINK_XFRM:
276 case NETLINK_FIB_LOOKUP:
277 case NETLINK_NETFILTER:
278 case NETLINK_GENERIC:
279 return true;
280 }
281
282 return false;
283}
284
285static int __netlink_deliver_tap_skb(struct sk_buff *skb,
286 struct net_device *dev)
287{
288 struct sk_buff *nskb;
289 struct sock *sk = skb->sk;
290 int ret = -ENOMEM;
291
292 if (!net_eq(dev_net(dev), sock_net(sk)))
293 return 0;
294
295 dev_hold(dev);
296
297 if (is_vmalloc_addr(skb->head))
298 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
299 else
300 nskb = skb_clone(skb, GFP_ATOMIC);
301 if (nskb) {
302 nskb->dev = dev;
303 nskb->protocol = htons((u16) sk->sk_protocol);
304 nskb->pkt_type = netlink_is_kernel(sk) ?
305 PACKET_KERNEL : PACKET_USER;
306 skb_reset_network_header(nskb);
307 ret = dev_queue_xmit(nskb);
308 if (unlikely(ret > 0))
309 ret = net_xmit_errno(ret);
310 }
311
312 dev_put(dev);
313 return ret;
314}
315
316static void __netlink_deliver_tap(struct sk_buff *skb, struct netlink_tap_net *nn)
317{
318 int ret;
319 struct netlink_tap *tmp;
320
321 if (!netlink_filter_tap(skb))
322 return;
323
324 list_for_each_entry_rcu(tmp, &nn->netlink_tap_all, list) {
325 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
326 if (unlikely(ret))
327 break;
328 }
329}
330
331static void netlink_deliver_tap(struct net *net, struct sk_buff *skb)
332{
333 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
334
335 rcu_read_lock();
336
337 if (unlikely(!list_empty(&nn->netlink_tap_all)))
338 __netlink_deliver_tap(skb, nn);
339
340 rcu_read_unlock();
341}
342
343static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
344 struct sk_buff *skb)
345{
346 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
347 netlink_deliver_tap(sock_net(dst), skb);
348}
349
350static void netlink_overrun(struct sock *sk)
351{
352 if (!nlk_test_bit(RECV_NO_ENOBUFS, sk)) {
353 if (!test_and_set_bit(NETLINK_S_CONGESTED,
354 &nlk_sk(sk)->state)) {
355 WRITE_ONCE(sk->sk_err, ENOBUFS);
356 sk_error_report(sk);
357 }
358 }
359 atomic_inc(&sk->sk_drops);
360}
361
362static void netlink_rcv_wake(struct sock *sk)
363{
364 struct netlink_sock *nlk = nlk_sk(sk);
365
366 if (skb_queue_empty_lockless(&sk->sk_receive_queue))
367 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
368 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
369 wake_up_interruptible(&nlk->wait);
370}
371
372static void netlink_skb_destructor(struct sk_buff *skb)
373{
374 if (is_vmalloc_addr(skb->head)) {
375 if (!skb->cloned ||
376 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
377 vfree_atomic(skb->head);
378
379 skb->head = NULL;
380 }
381 if (skb->sk != NULL)
382 sock_rfree(skb);
383}
384
385static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
386{
387 WARN_ON(skb->sk != NULL);
388 skb->sk = sk;
389 skb->destructor = netlink_skb_destructor;
390 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
391 sk_mem_charge(sk, skb->truesize);
392}
393
394static void netlink_sock_destruct(struct sock *sk)
395{
396 struct netlink_sock *nlk = nlk_sk(sk);
397
398 if (nlk->cb_running) {
399 if (nlk->cb.done)
400 nlk->cb.done(&nlk->cb);
401 module_put(nlk->cb.module);
402 kfree_skb(nlk->cb.skb);
403 }
404
405 skb_queue_purge(&sk->sk_receive_queue);
406
407 if (!sock_flag(sk, SOCK_DEAD)) {
408 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
409 return;
410 }
411
412 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
413 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
414 WARN_ON(nlk_sk(sk)->groups);
415}
416
417static void netlink_sock_destruct_work(struct work_struct *work)
418{
419 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
420 work);
421
422 sk_free(&nlk->sk);
423}
424
425/* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
426 * SMP. Look, when several writers sleep and reader wakes them up, all but one
427 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
428 * this, _but_ remember, it adds useless work on UP machines.
429 */
430
431void netlink_table_grab(void)
432 __acquires(nl_table_lock)
433{
434 might_sleep();
435
436 write_lock_irq(&nl_table_lock);
437
438 if (atomic_read(&nl_table_users)) {
439 DECLARE_WAITQUEUE(wait, current);
440
441 add_wait_queue_exclusive(&nl_table_wait, &wait);
442 for (;;) {
443 set_current_state(TASK_UNINTERRUPTIBLE);
444 if (atomic_read(&nl_table_users) == 0)
445 break;
446 write_unlock_irq(&nl_table_lock);
447 schedule();
448 write_lock_irq(&nl_table_lock);
449 }
450
451 __set_current_state(TASK_RUNNING);
452 remove_wait_queue(&nl_table_wait, &wait);
453 }
454}
455
456void netlink_table_ungrab(void)
457 __releases(nl_table_lock)
458{
459 write_unlock_irq(&nl_table_lock);
460 wake_up(&nl_table_wait);
461}
462
463static inline void
464netlink_lock_table(void)
465{
466 unsigned long flags;
467
468 /* read_lock() synchronizes us to netlink_table_grab */
469
470 read_lock_irqsave(&nl_table_lock, flags);
471 atomic_inc(&nl_table_users);
472 read_unlock_irqrestore(&nl_table_lock, flags);
473}
474
475static inline void
476netlink_unlock_table(void)
477{
478 if (atomic_dec_and_test(&nl_table_users))
479 wake_up(&nl_table_wait);
480}
481
482struct netlink_compare_arg
483{
484 possible_net_t pnet;
485 u32 portid;
486};
487
488/* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
489#define netlink_compare_arg_len \
490 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
491
492static inline int netlink_compare(struct rhashtable_compare_arg *arg,
493 const void *ptr)
494{
495 const struct netlink_compare_arg *x = arg->key;
496 const struct netlink_sock *nlk = ptr;
497
498 return nlk->portid != x->portid ||
499 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
500}
501
502static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
503 struct net *net, u32 portid)
504{
505 memset(arg, 0, sizeof(*arg));
506 write_pnet(&arg->pnet, net);
507 arg->portid = portid;
508}
509
510static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
511 struct net *net)
512{
513 struct netlink_compare_arg arg;
514
515 netlink_compare_arg_init(&arg, net, portid);
516 return rhashtable_lookup_fast(&table->hash, &arg,
517 netlink_rhashtable_params);
518}
519
520static int __netlink_insert(struct netlink_table *table, struct sock *sk)
521{
522 struct netlink_compare_arg arg;
523
524 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
525 return rhashtable_lookup_insert_key(&table->hash, &arg,
526 &nlk_sk(sk)->node,
527 netlink_rhashtable_params);
528}
529
530static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
531{
532 struct netlink_table *table = &nl_table[protocol];
533 struct sock *sk;
534
535 rcu_read_lock();
536 sk = __netlink_lookup(table, portid, net);
537 if (sk)
538 sock_hold(sk);
539 rcu_read_unlock();
540
541 return sk;
542}
543
544static const struct proto_ops netlink_ops;
545
546static void
547netlink_update_listeners(struct sock *sk)
548{
549 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
550 unsigned long mask;
551 unsigned int i;
552 struct listeners *listeners;
553
554 listeners = nl_deref_protected(tbl->listeners);
555 if (!listeners)
556 return;
557
558 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
559 mask = 0;
560 sk_for_each_bound(sk, &tbl->mc_list) {
561 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
562 mask |= nlk_sk(sk)->groups[i];
563 }
564 listeners->masks[i] = mask;
565 }
566 /* this function is only called with the netlink table "grabbed", which
567 * makes sure updates are visible before bind or setsockopt return. */
568}
569
570static int netlink_insert(struct sock *sk, u32 portid)
571{
572 struct netlink_table *table = &nl_table[sk->sk_protocol];
573 int err;
574
575 lock_sock(sk);
576
577 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
578 if (nlk_sk(sk)->bound)
579 goto err;
580
581 /* portid can be read locklessly from netlink_getname(). */
582 WRITE_ONCE(nlk_sk(sk)->portid, portid);
583
584 sock_hold(sk);
585
586 err = __netlink_insert(table, sk);
587 if (err) {
588 /* In case the hashtable backend returns with -EBUSY
589 * from here, it must not escape to the caller.
590 */
591 if (unlikely(err == -EBUSY))
592 err = -EOVERFLOW;
593 if (err == -EEXIST)
594 err = -EADDRINUSE;
595 sock_put(sk);
596 goto err;
597 }
598
599 /* We need to ensure that the socket is hashed and visible. */
600 smp_wmb();
601 /* Paired with lockless reads from netlink_bind(),
602 * netlink_connect() and netlink_sendmsg().
603 */
604 WRITE_ONCE(nlk_sk(sk)->bound, portid);
605
606err:
607 release_sock(sk);
608 return err;
609}
610
611static void netlink_remove(struct sock *sk)
612{
613 struct netlink_table *table;
614
615 table = &nl_table[sk->sk_protocol];
616 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
617 netlink_rhashtable_params)) {
618 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
619 __sock_put(sk);
620 }
621
622 netlink_table_grab();
623 if (nlk_sk(sk)->subscriptions) {
624 __sk_del_bind_node(sk);
625 netlink_update_listeners(sk);
626 }
627 if (sk->sk_protocol == NETLINK_GENERIC)
628 atomic_inc(&genl_sk_destructing_cnt);
629 netlink_table_ungrab();
630}
631
632static struct proto netlink_proto = {
633 .name = "NETLINK",
634 .owner = THIS_MODULE,
635 .obj_size = sizeof(struct netlink_sock),
636};
637
638static int __netlink_create(struct net *net, struct socket *sock,
639 struct mutex *cb_mutex, int protocol,
640 int kern)
641{
642 struct sock *sk;
643 struct netlink_sock *nlk;
644
645 sock->ops = &netlink_ops;
646
647 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
648 if (!sk)
649 return -ENOMEM;
650
651 sock_init_data(sock, sk);
652
653 nlk = nlk_sk(sk);
654 if (cb_mutex) {
655 nlk->cb_mutex = cb_mutex;
656 } else {
657 nlk->cb_mutex = &nlk->cb_def_mutex;
658 mutex_init(nlk->cb_mutex);
659 lockdep_set_class_and_name(nlk->cb_mutex,
660 nlk_cb_mutex_keys + protocol,
661 nlk_cb_mutex_key_strings[protocol]);
662 }
663 init_waitqueue_head(&nlk->wait);
664
665 sk->sk_destruct = netlink_sock_destruct;
666 sk->sk_protocol = protocol;
667 return 0;
668}
669
670static int netlink_create(struct net *net, struct socket *sock, int protocol,
671 int kern)
672{
673 struct module *module = NULL;
674 struct mutex *cb_mutex;
675 struct netlink_sock *nlk;
676 int (*bind)(struct net *net, int group);
677 void (*unbind)(struct net *net, int group);
678 void (*release)(struct sock *sock, unsigned long *groups);
679 int err = 0;
680
681 sock->state = SS_UNCONNECTED;
682
683 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
684 return -ESOCKTNOSUPPORT;
685
686 if (protocol < 0 || protocol >= MAX_LINKS)
687 return -EPROTONOSUPPORT;
688 protocol = array_index_nospec(protocol, MAX_LINKS);
689
690 netlink_lock_table();
691#ifdef CONFIG_MODULES
692 if (!nl_table[protocol].registered) {
693 netlink_unlock_table();
694 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
695 netlink_lock_table();
696 }
697#endif
698 if (nl_table[protocol].registered &&
699 try_module_get(nl_table[protocol].module))
700 module = nl_table[protocol].module;
701 else
702 err = -EPROTONOSUPPORT;
703 cb_mutex = nl_table[protocol].cb_mutex;
704 bind = nl_table[protocol].bind;
705 unbind = nl_table[protocol].unbind;
706 release = nl_table[protocol].release;
707 netlink_unlock_table();
708
709 if (err < 0)
710 goto out;
711
712 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
713 if (err < 0)
714 goto out_module;
715
716 sock_prot_inuse_add(net, &netlink_proto, 1);
717
718 nlk = nlk_sk(sock->sk);
719 nlk->module = module;
720 nlk->netlink_bind = bind;
721 nlk->netlink_unbind = unbind;
722 nlk->netlink_release = release;
723out:
724 return err;
725
726out_module:
727 module_put(module);
728 goto out;
729}
730
731static void deferred_put_nlk_sk(struct rcu_head *head)
732{
733 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
734 struct sock *sk = &nlk->sk;
735
736 kfree(nlk->groups);
737 nlk->groups = NULL;
738
739 if (!refcount_dec_and_test(&sk->sk_refcnt))
740 return;
741
742 if (nlk->cb_running && nlk->cb.done) {
743 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
744 schedule_work(&nlk->work);
745 return;
746 }
747
748 sk_free(sk);
749}
750
751static int netlink_release(struct socket *sock)
752{
753 struct sock *sk = sock->sk;
754 struct netlink_sock *nlk;
755
756 if (!sk)
757 return 0;
758
759 netlink_remove(sk);
760 sock_orphan(sk);
761 nlk = nlk_sk(sk);
762
763 /*
764 * OK. Socket is unlinked, any packets that arrive now
765 * will be purged.
766 */
767 if (nlk->netlink_release)
768 nlk->netlink_release(sk, nlk->groups);
769
770 /* must not acquire netlink_table_lock in any way again before unbind
771 * and notifying genetlink is done as otherwise it might deadlock
772 */
773 if (nlk->netlink_unbind) {
774 int i;
775
776 for (i = 0; i < nlk->ngroups; i++)
777 if (test_bit(i, nlk->groups))
778 nlk->netlink_unbind(sock_net(sk), i + 1);
779 }
780 if (sk->sk_protocol == NETLINK_GENERIC &&
781 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
782 wake_up(&genl_sk_destructing_waitq);
783
784 sock->sk = NULL;
785 wake_up_interruptible_all(&nlk->wait);
786
787 skb_queue_purge(&sk->sk_write_queue);
788
789 if (nlk->portid && nlk->bound) {
790 struct netlink_notify n = {
791 .net = sock_net(sk),
792 .protocol = sk->sk_protocol,
793 .portid = nlk->portid,
794 };
795 blocking_notifier_call_chain(&netlink_chain,
796 NETLINK_URELEASE, &n);
797 }
798
799 module_put(nlk->module);
800
801 if (netlink_is_kernel(sk)) {
802 netlink_table_grab();
803 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
804 if (--nl_table[sk->sk_protocol].registered == 0) {
805 struct listeners *old;
806
807 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
808 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
809 kfree_rcu(old, rcu);
810 nl_table[sk->sk_protocol].module = NULL;
811 nl_table[sk->sk_protocol].bind = NULL;
812 nl_table[sk->sk_protocol].unbind = NULL;
813 nl_table[sk->sk_protocol].flags = 0;
814 nl_table[sk->sk_protocol].registered = 0;
815 }
816 netlink_table_ungrab();
817 }
818
819 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
820
821 /* Because struct net might disappear soon, do not keep a pointer. */
822 if (!sk->sk_net_refcnt && sock_net(sk) != &init_net) {
823 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false);
824 /* Because of deferred_put_nlk_sk and use of work queue,
825 * it is possible netns will be freed before this socket.
826 */
827 sock_net_set(sk, &init_net);
828 __netns_tracker_alloc(&init_net, &sk->ns_tracker,
829 false, GFP_KERNEL);
830 }
831 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
832 return 0;
833}
834
835static int netlink_autobind(struct socket *sock)
836{
837 struct sock *sk = sock->sk;
838 struct net *net = sock_net(sk);
839 struct netlink_table *table = &nl_table[sk->sk_protocol];
840 s32 portid = task_tgid_vnr(current);
841 int err;
842 s32 rover = -4096;
843 bool ok;
844
845retry:
846 cond_resched();
847 rcu_read_lock();
848 ok = !__netlink_lookup(table, portid, net);
849 rcu_read_unlock();
850 if (!ok) {
851 /* Bind collision, search negative portid values. */
852 if (rover == -4096)
853 /* rover will be in range [S32_MIN, -4097] */
854 rover = S32_MIN + get_random_u32_below(-4096 - S32_MIN);
855 else if (rover >= -4096)
856 rover = -4097;
857 portid = rover--;
858 goto retry;
859 }
860
861 err = netlink_insert(sk, portid);
862 if (err == -EADDRINUSE)
863 goto retry;
864
865 /* If 2 threads race to autobind, that is fine. */
866 if (err == -EBUSY)
867 err = 0;
868
869 return err;
870}
871
872/**
873 * __netlink_ns_capable - General netlink message capability test
874 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
875 * @user_ns: The user namespace of the capability to use
876 * @cap: The capability to use
877 *
878 * Test to see if the opener of the socket we received the message
879 * from had when the netlink socket was created and the sender of the
880 * message has the capability @cap in the user namespace @user_ns.
881 */
882bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
883 struct user_namespace *user_ns, int cap)
884{
885 return ((nsp->flags & NETLINK_SKB_DST) ||
886 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
887 ns_capable(user_ns, cap);
888}
889EXPORT_SYMBOL(__netlink_ns_capable);
890
891/**
892 * netlink_ns_capable - General netlink message capability test
893 * @skb: socket buffer holding a netlink command from userspace
894 * @user_ns: The user namespace of the capability to use
895 * @cap: The capability to use
896 *
897 * Test to see if the opener of the socket we received the message
898 * from had when the netlink socket was created and the sender of the
899 * message has the capability @cap in the user namespace @user_ns.
900 */
901bool netlink_ns_capable(const struct sk_buff *skb,
902 struct user_namespace *user_ns, int cap)
903{
904 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
905}
906EXPORT_SYMBOL(netlink_ns_capable);
907
908/**
909 * netlink_capable - Netlink global message capability test
910 * @skb: socket buffer holding a netlink command from userspace
911 * @cap: The capability to use
912 *
913 * Test to see if the opener of the socket we received the message
914 * from had when the netlink socket was created and the sender of the
915 * message has the capability @cap in all user namespaces.
916 */
917bool netlink_capable(const struct sk_buff *skb, int cap)
918{
919 return netlink_ns_capable(skb, &init_user_ns, cap);
920}
921EXPORT_SYMBOL(netlink_capable);
922
923/**
924 * netlink_net_capable - Netlink network namespace message capability test
925 * @skb: socket buffer holding a netlink command from userspace
926 * @cap: The capability to use
927 *
928 * Test to see if the opener of the socket we received the message
929 * from had when the netlink socket was created and the sender of the
930 * message has the capability @cap over the network namespace of
931 * the socket we received the message from.
932 */
933bool netlink_net_capable(const struct sk_buff *skb, int cap)
934{
935 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
936}
937EXPORT_SYMBOL(netlink_net_capable);
938
939static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
940{
941 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
942 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
943}
944
945static void
946netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
947{
948 struct netlink_sock *nlk = nlk_sk(sk);
949
950 if (nlk->subscriptions && !subscriptions)
951 __sk_del_bind_node(sk);
952 else if (!nlk->subscriptions && subscriptions)
953 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
954 nlk->subscriptions = subscriptions;
955}
956
957static int netlink_realloc_groups(struct sock *sk)
958{
959 struct netlink_sock *nlk = nlk_sk(sk);
960 unsigned int groups;
961 unsigned long *new_groups;
962 int err = 0;
963
964 netlink_table_grab();
965
966 groups = nl_table[sk->sk_protocol].groups;
967 if (!nl_table[sk->sk_protocol].registered) {
968 err = -ENOENT;
969 goto out_unlock;
970 }
971
972 if (nlk->ngroups >= groups)
973 goto out_unlock;
974
975 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
976 if (new_groups == NULL) {
977 err = -ENOMEM;
978 goto out_unlock;
979 }
980 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
981 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
982
983 nlk->groups = new_groups;
984 nlk->ngroups = groups;
985 out_unlock:
986 netlink_table_ungrab();
987 return err;
988}
989
990static void netlink_undo_bind(int group, long unsigned int groups,
991 struct sock *sk)
992{
993 struct netlink_sock *nlk = nlk_sk(sk);
994 int undo;
995
996 if (!nlk->netlink_unbind)
997 return;
998
999 for (undo = 0; undo < group; undo++)
1000 if (test_bit(undo, &groups))
1001 nlk->netlink_unbind(sock_net(sk), undo + 1);
1002}
1003
1004static int netlink_bind(struct socket *sock, struct sockaddr *addr,
1005 int addr_len)
1006{
1007 struct sock *sk = sock->sk;
1008 struct net *net = sock_net(sk);
1009 struct netlink_sock *nlk = nlk_sk(sk);
1010 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1011 int err = 0;
1012 unsigned long groups;
1013 bool bound;
1014
1015 if (addr_len < sizeof(struct sockaddr_nl))
1016 return -EINVAL;
1017
1018 if (nladdr->nl_family != AF_NETLINK)
1019 return -EINVAL;
1020 groups = nladdr->nl_groups;
1021
1022 /* Only superuser is allowed to listen multicasts */
1023 if (groups) {
1024 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1025 return -EPERM;
1026 err = netlink_realloc_groups(sk);
1027 if (err)
1028 return err;
1029 }
1030
1031 if (nlk->ngroups < BITS_PER_LONG)
1032 groups &= (1UL << nlk->ngroups) - 1;
1033
1034 /* Paired with WRITE_ONCE() in netlink_insert() */
1035 bound = READ_ONCE(nlk->bound);
1036 if (bound) {
1037 /* Ensure nlk->portid is up-to-date. */
1038 smp_rmb();
1039
1040 if (nladdr->nl_pid != nlk->portid)
1041 return -EINVAL;
1042 }
1043
1044 if (nlk->netlink_bind && groups) {
1045 int group;
1046
1047 /* nl_groups is a u32, so cap the maximum groups we can bind */
1048 for (group = 0; group < BITS_PER_TYPE(u32); group++) {
1049 if (!test_bit(group, &groups))
1050 continue;
1051 err = nlk->netlink_bind(net, group + 1);
1052 if (!err)
1053 continue;
1054 netlink_undo_bind(group, groups, sk);
1055 return err;
1056 }
1057 }
1058
1059 /* No need for barriers here as we return to user-space without
1060 * using any of the bound attributes.
1061 */
1062 netlink_lock_table();
1063 if (!bound) {
1064 err = nladdr->nl_pid ?
1065 netlink_insert(sk, nladdr->nl_pid) :
1066 netlink_autobind(sock);
1067 if (err) {
1068 netlink_undo_bind(BITS_PER_TYPE(u32), groups, sk);
1069 goto unlock;
1070 }
1071 }
1072
1073 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1074 goto unlock;
1075 netlink_unlock_table();
1076
1077 netlink_table_grab();
1078 netlink_update_subscriptions(sk, nlk->subscriptions +
1079 hweight32(groups) -
1080 hweight32(nlk->groups[0]));
1081 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1082 netlink_update_listeners(sk);
1083 netlink_table_ungrab();
1084
1085 return 0;
1086
1087unlock:
1088 netlink_unlock_table();
1089 return err;
1090}
1091
1092static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1093 int alen, int flags)
1094{
1095 int err = 0;
1096 struct sock *sk = sock->sk;
1097 struct netlink_sock *nlk = nlk_sk(sk);
1098 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1099
1100 if (alen < sizeof(addr->sa_family))
1101 return -EINVAL;
1102
1103 if (addr->sa_family == AF_UNSPEC) {
1104 /* paired with READ_ONCE() in netlink_getsockbyportid() */
1105 WRITE_ONCE(sk->sk_state, NETLINK_UNCONNECTED);
1106 /* dst_portid and dst_group can be read locklessly */
1107 WRITE_ONCE(nlk->dst_portid, 0);
1108 WRITE_ONCE(nlk->dst_group, 0);
1109 return 0;
1110 }
1111 if (addr->sa_family != AF_NETLINK)
1112 return -EINVAL;
1113
1114 if (alen < sizeof(struct sockaddr_nl))
1115 return -EINVAL;
1116
1117 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1118 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1119 return -EPERM;
1120
1121 /* No need for barriers here as we return to user-space without
1122 * using any of the bound attributes.
1123 * Paired with WRITE_ONCE() in netlink_insert().
1124 */
1125 if (!READ_ONCE(nlk->bound))
1126 err = netlink_autobind(sock);
1127
1128 if (err == 0) {
1129 /* paired with READ_ONCE() in netlink_getsockbyportid() */
1130 WRITE_ONCE(sk->sk_state, NETLINK_CONNECTED);
1131 /* dst_portid and dst_group can be read locklessly */
1132 WRITE_ONCE(nlk->dst_portid, nladdr->nl_pid);
1133 WRITE_ONCE(nlk->dst_group, ffs(nladdr->nl_groups));
1134 }
1135
1136 return err;
1137}
1138
1139static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1140 int peer)
1141{
1142 struct sock *sk = sock->sk;
1143 struct netlink_sock *nlk = nlk_sk(sk);
1144 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1145
1146 nladdr->nl_family = AF_NETLINK;
1147 nladdr->nl_pad = 0;
1148
1149 if (peer) {
1150 /* Paired with WRITE_ONCE() in netlink_connect() */
1151 nladdr->nl_pid = READ_ONCE(nlk->dst_portid);
1152 nladdr->nl_groups = netlink_group_mask(READ_ONCE(nlk->dst_group));
1153 } else {
1154 /* Paired with WRITE_ONCE() in netlink_insert() */
1155 nladdr->nl_pid = READ_ONCE(nlk->portid);
1156 netlink_lock_table();
1157 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1158 netlink_unlock_table();
1159 }
1160 return sizeof(*nladdr);
1161}
1162
1163static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1164 unsigned long arg)
1165{
1166 /* try to hand this ioctl down to the NIC drivers.
1167 */
1168 return -ENOIOCTLCMD;
1169}
1170
1171static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1172{
1173 struct sock *sock;
1174 struct netlink_sock *nlk;
1175
1176 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1177 if (!sock)
1178 return ERR_PTR(-ECONNREFUSED);
1179
1180 /* Don't bother queuing skb if kernel socket has no input function */
1181 nlk = nlk_sk(sock);
1182 /* dst_portid and sk_state can be changed in netlink_connect() */
1183 if (READ_ONCE(sock->sk_state) == NETLINK_CONNECTED &&
1184 READ_ONCE(nlk->dst_portid) != nlk_sk(ssk)->portid) {
1185 sock_put(sock);
1186 return ERR_PTR(-ECONNREFUSED);
1187 }
1188 return sock;
1189}
1190
1191struct sock *netlink_getsockbyfilp(struct file *filp)
1192{
1193 struct inode *inode = file_inode(filp);
1194 struct sock *sock;
1195
1196 if (!S_ISSOCK(inode->i_mode))
1197 return ERR_PTR(-ENOTSOCK);
1198
1199 sock = SOCKET_I(inode)->sk;
1200 if (sock->sk_family != AF_NETLINK)
1201 return ERR_PTR(-EINVAL);
1202
1203 sock_hold(sock);
1204 return sock;
1205}
1206
1207struct sk_buff *netlink_alloc_large_skb(unsigned int size, int broadcast)
1208{
1209 struct sk_buff *skb;
1210 void *data;
1211
1212 if (size <= NLMSG_GOODSIZE || broadcast)
1213 return alloc_skb(size, GFP_KERNEL);
1214
1215 size = SKB_DATA_ALIGN(size) +
1216 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1217
1218 data = vmalloc(size);
1219 if (data == NULL)
1220 return NULL;
1221
1222 skb = __build_skb(data, size);
1223 if (skb == NULL)
1224 vfree(data);
1225 else
1226 skb->destructor = netlink_skb_destructor;
1227
1228 return skb;
1229}
1230
1231/*
1232 * Attach a skb to a netlink socket.
1233 * The caller must hold a reference to the destination socket. On error, the
1234 * reference is dropped. The skb is not send to the destination, just all
1235 * all error checks are performed and memory in the queue is reserved.
1236 * Return values:
1237 * < 0: error. skb freed, reference to sock dropped.
1238 * 0: continue
1239 * 1: repeat lookup - reference dropped while waiting for socket memory.
1240 */
1241int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1242 long *timeo, struct sock *ssk)
1243{
1244 struct netlink_sock *nlk;
1245
1246 nlk = nlk_sk(sk);
1247
1248 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1249 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1250 DECLARE_WAITQUEUE(wait, current);
1251 if (!*timeo) {
1252 if (!ssk || netlink_is_kernel(ssk))
1253 netlink_overrun(sk);
1254 sock_put(sk);
1255 kfree_skb(skb);
1256 return -EAGAIN;
1257 }
1258
1259 __set_current_state(TASK_INTERRUPTIBLE);
1260 add_wait_queue(&nlk->wait, &wait);
1261
1262 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1263 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1264 !sock_flag(sk, SOCK_DEAD))
1265 *timeo = schedule_timeout(*timeo);
1266
1267 __set_current_state(TASK_RUNNING);
1268 remove_wait_queue(&nlk->wait, &wait);
1269 sock_put(sk);
1270
1271 if (signal_pending(current)) {
1272 kfree_skb(skb);
1273 return sock_intr_errno(*timeo);
1274 }
1275 return 1;
1276 }
1277 netlink_skb_set_owner_r(skb, sk);
1278 return 0;
1279}
1280
1281static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1282{
1283 int len = skb->len;
1284
1285 netlink_deliver_tap(sock_net(sk), skb);
1286
1287 skb_queue_tail(&sk->sk_receive_queue, skb);
1288 sk->sk_data_ready(sk);
1289 return len;
1290}
1291
1292int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1293{
1294 int len = __netlink_sendskb(sk, skb);
1295
1296 sock_put(sk);
1297 return len;
1298}
1299
1300void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1301{
1302 kfree_skb(skb);
1303 sock_put(sk);
1304}
1305
1306static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1307{
1308 int delta;
1309
1310 WARN_ON(skb->sk != NULL);
1311 delta = skb->end - skb->tail;
1312 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1313 return skb;
1314
1315 if (skb_shared(skb)) {
1316 struct sk_buff *nskb = skb_clone(skb, allocation);
1317 if (!nskb)
1318 return skb;
1319 consume_skb(skb);
1320 skb = nskb;
1321 }
1322
1323 pskb_expand_head(skb, 0, -delta,
1324 (allocation & ~__GFP_DIRECT_RECLAIM) |
1325 __GFP_NOWARN | __GFP_NORETRY);
1326 return skb;
1327}
1328
1329static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1330 struct sock *ssk)
1331{
1332 int ret;
1333 struct netlink_sock *nlk = nlk_sk(sk);
1334
1335 ret = -ECONNREFUSED;
1336 if (nlk->netlink_rcv != NULL) {
1337 ret = skb->len;
1338 netlink_skb_set_owner_r(skb, sk);
1339 NETLINK_CB(skb).sk = ssk;
1340 netlink_deliver_tap_kernel(sk, ssk, skb);
1341 nlk->netlink_rcv(skb);
1342 consume_skb(skb);
1343 } else {
1344 kfree_skb(skb);
1345 }
1346 sock_put(sk);
1347 return ret;
1348}
1349
1350int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1351 u32 portid, int nonblock)
1352{
1353 struct sock *sk;
1354 int err;
1355 long timeo;
1356
1357 skb = netlink_trim(skb, gfp_any());
1358
1359 timeo = sock_sndtimeo(ssk, nonblock);
1360retry:
1361 sk = netlink_getsockbyportid(ssk, portid);
1362 if (IS_ERR(sk)) {
1363 kfree_skb(skb);
1364 return PTR_ERR(sk);
1365 }
1366 if (netlink_is_kernel(sk))
1367 return netlink_unicast_kernel(sk, skb, ssk);
1368
1369 if (sk_filter(sk, skb)) {
1370 err = skb->len;
1371 kfree_skb(skb);
1372 sock_put(sk);
1373 return err;
1374 }
1375
1376 err = netlink_attachskb(sk, skb, &timeo, ssk);
1377 if (err == 1)
1378 goto retry;
1379 if (err)
1380 return err;
1381
1382 return netlink_sendskb(sk, skb);
1383}
1384EXPORT_SYMBOL(netlink_unicast);
1385
1386int netlink_has_listeners(struct sock *sk, unsigned int group)
1387{
1388 int res = 0;
1389 struct listeners *listeners;
1390
1391 BUG_ON(!netlink_is_kernel(sk));
1392
1393 rcu_read_lock();
1394 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1395
1396 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1397 res = test_bit(group - 1, listeners->masks);
1398
1399 rcu_read_unlock();
1400
1401 return res;
1402}
1403EXPORT_SYMBOL_GPL(netlink_has_listeners);
1404
1405bool netlink_strict_get_check(struct sk_buff *skb)
1406{
1407 return nlk_test_bit(STRICT_CHK, NETLINK_CB(skb).sk);
1408}
1409EXPORT_SYMBOL_GPL(netlink_strict_get_check);
1410
1411static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1412{
1413 struct netlink_sock *nlk = nlk_sk(sk);
1414
1415 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1416 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1417 netlink_skb_set_owner_r(skb, sk);
1418 __netlink_sendskb(sk, skb);
1419 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1420 }
1421 return -1;
1422}
1423
1424struct netlink_broadcast_data {
1425 struct sock *exclude_sk;
1426 struct net *net;
1427 u32 portid;
1428 u32 group;
1429 int failure;
1430 int delivery_failure;
1431 int congested;
1432 int delivered;
1433 gfp_t allocation;
1434 struct sk_buff *skb, *skb2;
1435 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1436 void *tx_data;
1437};
1438
1439static void do_one_broadcast(struct sock *sk,
1440 struct netlink_broadcast_data *p)
1441{
1442 struct netlink_sock *nlk = nlk_sk(sk);
1443 int val;
1444
1445 if (p->exclude_sk == sk)
1446 return;
1447
1448 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1449 !test_bit(p->group - 1, nlk->groups))
1450 return;
1451
1452 if (!net_eq(sock_net(sk), p->net)) {
1453 if (!nlk_test_bit(LISTEN_ALL_NSID, sk))
1454 return;
1455
1456 if (!peernet_has_id(sock_net(sk), p->net))
1457 return;
1458
1459 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1460 CAP_NET_BROADCAST))
1461 return;
1462 }
1463
1464 if (p->failure) {
1465 netlink_overrun(sk);
1466 return;
1467 }
1468
1469 sock_hold(sk);
1470 if (p->skb2 == NULL) {
1471 if (skb_shared(p->skb)) {
1472 p->skb2 = skb_clone(p->skb, p->allocation);
1473 } else {
1474 p->skb2 = skb_get(p->skb);
1475 /*
1476 * skb ownership may have been set when
1477 * delivered to a previous socket.
1478 */
1479 skb_orphan(p->skb2);
1480 }
1481 }
1482 if (p->skb2 == NULL) {
1483 netlink_overrun(sk);
1484 /* Clone failed. Notify ALL listeners. */
1485 p->failure = 1;
1486 if (nlk_test_bit(BROADCAST_SEND_ERROR, sk))
1487 p->delivery_failure = 1;
1488 goto out;
1489 }
1490
1491 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1492 kfree_skb(p->skb2);
1493 p->skb2 = NULL;
1494 goto out;
1495 }
1496
1497 if (sk_filter(sk, p->skb2)) {
1498 kfree_skb(p->skb2);
1499 p->skb2 = NULL;
1500 goto out;
1501 }
1502 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1503 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1504 NETLINK_CB(p->skb2).nsid_is_set = true;
1505 val = netlink_broadcast_deliver(sk, p->skb2);
1506 if (val < 0) {
1507 netlink_overrun(sk);
1508 if (nlk_test_bit(BROADCAST_SEND_ERROR, sk))
1509 p->delivery_failure = 1;
1510 } else {
1511 p->congested |= val;
1512 p->delivered = 1;
1513 p->skb2 = NULL;
1514 }
1515out:
1516 sock_put(sk);
1517}
1518
1519int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb,
1520 u32 portid,
1521 u32 group, gfp_t allocation,
1522 netlink_filter_fn filter,
1523 void *filter_data)
1524{
1525 struct net *net = sock_net(ssk);
1526 struct netlink_broadcast_data info;
1527 struct sock *sk;
1528
1529 skb = netlink_trim(skb, allocation);
1530
1531 info.exclude_sk = ssk;
1532 info.net = net;
1533 info.portid = portid;
1534 info.group = group;
1535 info.failure = 0;
1536 info.delivery_failure = 0;
1537 info.congested = 0;
1538 info.delivered = 0;
1539 info.allocation = allocation;
1540 info.skb = skb;
1541 info.skb2 = NULL;
1542 info.tx_filter = filter;
1543 info.tx_data = filter_data;
1544
1545 /* While we sleep in clone, do not allow to change socket list */
1546
1547 netlink_lock_table();
1548
1549 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1550 do_one_broadcast(sk, &info);
1551
1552 consume_skb(skb);
1553
1554 netlink_unlock_table();
1555
1556 if (info.delivery_failure) {
1557 kfree_skb(info.skb2);
1558 return -ENOBUFS;
1559 }
1560 consume_skb(info.skb2);
1561
1562 if (info.delivered) {
1563 if (info.congested && gfpflags_allow_blocking(allocation))
1564 yield();
1565 return 0;
1566 }
1567 return -ESRCH;
1568}
1569EXPORT_SYMBOL(netlink_broadcast_filtered);
1570
1571int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1572 u32 group, gfp_t allocation)
1573{
1574 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1575 NULL, NULL);
1576}
1577EXPORT_SYMBOL(netlink_broadcast);
1578
1579struct netlink_set_err_data {
1580 struct sock *exclude_sk;
1581 u32 portid;
1582 u32 group;
1583 int code;
1584};
1585
1586static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1587{
1588 struct netlink_sock *nlk = nlk_sk(sk);
1589 int ret = 0;
1590
1591 if (sk == p->exclude_sk)
1592 goto out;
1593
1594 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1595 goto out;
1596
1597 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1598 !test_bit(p->group - 1, nlk->groups))
1599 goto out;
1600
1601 if (p->code == ENOBUFS && nlk_test_bit(RECV_NO_ENOBUFS, sk)) {
1602 ret = 1;
1603 goto out;
1604 }
1605
1606 WRITE_ONCE(sk->sk_err, p->code);
1607 sk_error_report(sk);
1608out:
1609 return ret;
1610}
1611
1612/**
1613 * netlink_set_err - report error to broadcast listeners
1614 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1615 * @portid: the PORTID of a process that we want to skip (if any)
1616 * @group: the broadcast group that will notice the error
1617 * @code: error code, must be negative (as usual in kernelspace)
1618 *
1619 * This function returns the number of broadcast listeners that have set the
1620 * NETLINK_NO_ENOBUFS socket option.
1621 */
1622int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1623{
1624 struct netlink_set_err_data info;
1625 unsigned long flags;
1626 struct sock *sk;
1627 int ret = 0;
1628
1629 info.exclude_sk = ssk;
1630 info.portid = portid;
1631 info.group = group;
1632 /* sk->sk_err wants a positive error value */
1633 info.code = -code;
1634
1635 read_lock_irqsave(&nl_table_lock, flags);
1636
1637 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1638 ret += do_one_set_err(sk, &info);
1639
1640 read_unlock_irqrestore(&nl_table_lock, flags);
1641 return ret;
1642}
1643EXPORT_SYMBOL(netlink_set_err);
1644
1645/* must be called with netlink table grabbed */
1646static void netlink_update_socket_mc(struct netlink_sock *nlk,
1647 unsigned int group,
1648 int is_new)
1649{
1650 int old, new = !!is_new, subscriptions;
1651
1652 old = test_bit(group - 1, nlk->groups);
1653 subscriptions = nlk->subscriptions - old + new;
1654 __assign_bit(group - 1, nlk->groups, new);
1655 netlink_update_subscriptions(&nlk->sk, subscriptions);
1656 netlink_update_listeners(&nlk->sk);
1657}
1658
1659static int netlink_setsockopt(struct socket *sock, int level, int optname,
1660 sockptr_t optval, unsigned int optlen)
1661{
1662 struct sock *sk = sock->sk;
1663 struct netlink_sock *nlk = nlk_sk(sk);
1664 unsigned int val = 0;
1665 int nr = -1;
1666
1667 if (level != SOL_NETLINK)
1668 return -ENOPROTOOPT;
1669
1670 if (optlen >= sizeof(int) &&
1671 copy_from_sockptr(&val, optval, sizeof(val)))
1672 return -EFAULT;
1673
1674 switch (optname) {
1675 case NETLINK_PKTINFO:
1676 nr = NETLINK_F_RECV_PKTINFO;
1677 break;
1678 case NETLINK_ADD_MEMBERSHIP:
1679 case NETLINK_DROP_MEMBERSHIP: {
1680 int err;
1681
1682 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1683 return -EPERM;
1684 err = netlink_realloc_groups(sk);
1685 if (err)
1686 return err;
1687 if (!val || val - 1 >= nlk->ngroups)
1688 return -EINVAL;
1689 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1690 err = nlk->netlink_bind(sock_net(sk), val);
1691 if (err)
1692 return err;
1693 }
1694 netlink_table_grab();
1695 netlink_update_socket_mc(nlk, val,
1696 optname == NETLINK_ADD_MEMBERSHIP);
1697 netlink_table_ungrab();
1698 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1699 nlk->netlink_unbind(sock_net(sk), val);
1700
1701 break;
1702 }
1703 case NETLINK_BROADCAST_ERROR:
1704 nr = NETLINK_F_BROADCAST_SEND_ERROR;
1705 break;
1706 case NETLINK_NO_ENOBUFS:
1707 assign_bit(NETLINK_F_RECV_NO_ENOBUFS, &nlk->flags, val);
1708 if (val) {
1709 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1710 wake_up_interruptible(&nlk->wait);
1711 }
1712 break;
1713 case NETLINK_LISTEN_ALL_NSID:
1714 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1715 return -EPERM;
1716 nr = NETLINK_F_LISTEN_ALL_NSID;
1717 break;
1718 case NETLINK_CAP_ACK:
1719 nr = NETLINK_F_CAP_ACK;
1720 break;
1721 case NETLINK_EXT_ACK:
1722 nr = NETLINK_F_EXT_ACK;
1723 break;
1724 case NETLINK_GET_STRICT_CHK:
1725 nr = NETLINK_F_STRICT_CHK;
1726 break;
1727 default:
1728 return -ENOPROTOOPT;
1729 }
1730 if (nr >= 0)
1731 assign_bit(nr, &nlk->flags, val);
1732 return 0;
1733}
1734
1735static int netlink_getsockopt(struct socket *sock, int level, int optname,
1736 char __user *optval, int __user *optlen)
1737{
1738 struct sock *sk = sock->sk;
1739 struct netlink_sock *nlk = nlk_sk(sk);
1740 unsigned int flag;
1741 int len, val;
1742
1743 if (level != SOL_NETLINK)
1744 return -ENOPROTOOPT;
1745
1746 if (get_user(len, optlen))
1747 return -EFAULT;
1748 if (len < 0)
1749 return -EINVAL;
1750
1751 switch (optname) {
1752 case NETLINK_PKTINFO:
1753 flag = NETLINK_F_RECV_PKTINFO;
1754 break;
1755 case NETLINK_BROADCAST_ERROR:
1756 flag = NETLINK_F_BROADCAST_SEND_ERROR;
1757 break;
1758 case NETLINK_NO_ENOBUFS:
1759 flag = NETLINK_F_RECV_NO_ENOBUFS;
1760 break;
1761 case NETLINK_LIST_MEMBERSHIPS: {
1762 int pos, idx, shift, err = 0;
1763
1764 netlink_lock_table();
1765 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1766 if (len - pos < sizeof(u32))
1767 break;
1768
1769 idx = pos / sizeof(unsigned long);
1770 shift = (pos % sizeof(unsigned long)) * 8;
1771 if (put_user((u32)(nlk->groups[idx] >> shift),
1772 (u32 __user *)(optval + pos))) {
1773 err = -EFAULT;
1774 break;
1775 }
1776 }
1777 if (put_user(ALIGN(BITS_TO_BYTES(nlk->ngroups), sizeof(u32)), optlen))
1778 err = -EFAULT;
1779 netlink_unlock_table();
1780 return err;
1781 }
1782 case NETLINK_CAP_ACK:
1783 flag = NETLINK_F_CAP_ACK;
1784 break;
1785 case NETLINK_EXT_ACK:
1786 flag = NETLINK_F_EXT_ACK;
1787 break;
1788 case NETLINK_GET_STRICT_CHK:
1789 flag = NETLINK_F_STRICT_CHK;
1790 break;
1791 default:
1792 return -ENOPROTOOPT;
1793 }
1794
1795 if (len < sizeof(int))
1796 return -EINVAL;
1797
1798 len = sizeof(int);
1799 val = test_bit(flag, &nlk->flags);
1800
1801 if (put_user(len, optlen) ||
1802 copy_to_user(optval, &val, len))
1803 return -EFAULT;
1804
1805 return 0;
1806}
1807
1808static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1809{
1810 struct nl_pktinfo info;
1811
1812 info.group = NETLINK_CB(skb).dst_group;
1813 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1814}
1815
1816static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1817 struct sk_buff *skb)
1818{
1819 if (!NETLINK_CB(skb).nsid_is_set)
1820 return;
1821
1822 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1823 &NETLINK_CB(skb).nsid);
1824}
1825
1826static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1827{
1828 struct sock *sk = sock->sk;
1829 struct netlink_sock *nlk = nlk_sk(sk);
1830 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1831 u32 dst_portid;
1832 u32 dst_group;
1833 struct sk_buff *skb;
1834 int err;
1835 struct scm_cookie scm;
1836 u32 netlink_skb_flags = 0;
1837
1838 if (msg->msg_flags & MSG_OOB)
1839 return -EOPNOTSUPP;
1840
1841 if (len == 0) {
1842 pr_warn_once("Zero length message leads to an empty skb\n");
1843 return -ENODATA;
1844 }
1845
1846 err = scm_send(sock, msg, &scm, true);
1847 if (err < 0)
1848 return err;
1849
1850 if (msg->msg_namelen) {
1851 err = -EINVAL;
1852 if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1853 goto out;
1854 if (addr->nl_family != AF_NETLINK)
1855 goto out;
1856 dst_portid = addr->nl_pid;
1857 dst_group = ffs(addr->nl_groups);
1858 err = -EPERM;
1859 if ((dst_group || dst_portid) &&
1860 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1861 goto out;
1862 netlink_skb_flags |= NETLINK_SKB_DST;
1863 } else {
1864 /* Paired with WRITE_ONCE() in netlink_connect() */
1865 dst_portid = READ_ONCE(nlk->dst_portid);
1866 dst_group = READ_ONCE(nlk->dst_group);
1867 }
1868
1869 /* Paired with WRITE_ONCE() in netlink_insert() */
1870 if (!READ_ONCE(nlk->bound)) {
1871 err = netlink_autobind(sock);
1872 if (err)
1873 goto out;
1874 } else {
1875 /* Ensure nlk is hashed and visible. */
1876 smp_rmb();
1877 }
1878
1879 err = -EMSGSIZE;
1880 if (len > sk->sk_sndbuf - 32)
1881 goto out;
1882 err = -ENOBUFS;
1883 skb = netlink_alloc_large_skb(len, dst_group);
1884 if (skb == NULL)
1885 goto out;
1886
1887 NETLINK_CB(skb).portid = nlk->portid;
1888 NETLINK_CB(skb).dst_group = dst_group;
1889 NETLINK_CB(skb).creds = scm.creds;
1890 NETLINK_CB(skb).flags = netlink_skb_flags;
1891
1892 err = -EFAULT;
1893 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1894 kfree_skb(skb);
1895 goto out;
1896 }
1897
1898 err = security_netlink_send(sk, skb);
1899 if (err) {
1900 kfree_skb(skb);
1901 goto out;
1902 }
1903
1904 if (dst_group) {
1905 refcount_inc(&skb->users);
1906 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1907 }
1908 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags & MSG_DONTWAIT);
1909
1910out:
1911 scm_destroy(&scm);
1912 return err;
1913}
1914
1915static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1916 int flags)
1917{
1918 struct scm_cookie scm;
1919 struct sock *sk = sock->sk;
1920 struct netlink_sock *nlk = nlk_sk(sk);
1921 size_t copied, max_recvmsg_len;
1922 struct sk_buff *skb, *data_skb;
1923 int err, ret;
1924
1925 if (flags & MSG_OOB)
1926 return -EOPNOTSUPP;
1927
1928 copied = 0;
1929
1930 skb = skb_recv_datagram(sk, flags, &err);
1931 if (skb == NULL)
1932 goto out;
1933
1934 data_skb = skb;
1935
1936#ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1937 if (unlikely(skb_shinfo(skb)->frag_list)) {
1938 /*
1939 * If this skb has a frag_list, then here that means that we
1940 * will have to use the frag_list skb's data for compat tasks
1941 * and the regular skb's data for normal (non-compat) tasks.
1942 *
1943 * If we need to send the compat skb, assign it to the
1944 * 'data_skb' variable so that it will be used below for data
1945 * copying. We keep 'skb' for everything else, including
1946 * freeing both later.
1947 */
1948 if (flags & MSG_CMSG_COMPAT)
1949 data_skb = skb_shinfo(skb)->frag_list;
1950 }
1951#endif
1952
1953 /* Record the max length of recvmsg() calls for future allocations */
1954 max_recvmsg_len = max(READ_ONCE(nlk->max_recvmsg_len), len);
1955 max_recvmsg_len = min_t(size_t, max_recvmsg_len,
1956 SKB_WITH_OVERHEAD(32768));
1957 WRITE_ONCE(nlk->max_recvmsg_len, max_recvmsg_len);
1958
1959 copied = data_skb->len;
1960 if (len < copied) {
1961 msg->msg_flags |= MSG_TRUNC;
1962 copied = len;
1963 }
1964
1965 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1966
1967 if (msg->msg_name) {
1968 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1969 addr->nl_family = AF_NETLINK;
1970 addr->nl_pad = 0;
1971 addr->nl_pid = NETLINK_CB(skb).portid;
1972 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1973 msg->msg_namelen = sizeof(*addr);
1974 }
1975
1976 if (nlk_test_bit(RECV_PKTINFO, sk))
1977 netlink_cmsg_recv_pktinfo(msg, skb);
1978 if (nlk_test_bit(LISTEN_ALL_NSID, sk))
1979 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1980
1981 memset(&scm, 0, sizeof(scm));
1982 scm.creds = *NETLINK_CREDS(skb);
1983 if (flags & MSG_TRUNC)
1984 copied = data_skb->len;
1985
1986 skb_free_datagram(sk, skb);
1987
1988 if (READ_ONCE(nlk->cb_running) &&
1989 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1990 ret = netlink_dump(sk);
1991 if (ret) {
1992 WRITE_ONCE(sk->sk_err, -ret);
1993 sk_error_report(sk);
1994 }
1995 }
1996
1997 scm_recv(sock, msg, &scm, flags);
1998out:
1999 netlink_rcv_wake(sk);
2000 return err ? : copied;
2001}
2002
2003static void netlink_data_ready(struct sock *sk)
2004{
2005 BUG();
2006}
2007
2008/*
2009 * We export these functions to other modules. They provide a
2010 * complete set of kernel non-blocking support for message
2011 * queueing.
2012 */
2013
2014struct sock *
2015__netlink_kernel_create(struct net *net, int unit, struct module *module,
2016 struct netlink_kernel_cfg *cfg)
2017{
2018 struct socket *sock;
2019 struct sock *sk;
2020 struct netlink_sock *nlk;
2021 struct listeners *listeners = NULL;
2022 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2023 unsigned int groups;
2024
2025 BUG_ON(!nl_table);
2026
2027 if (unit < 0 || unit >= MAX_LINKS)
2028 return NULL;
2029
2030 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2031 return NULL;
2032
2033 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2034 goto out_sock_release_nosk;
2035
2036 sk = sock->sk;
2037
2038 if (!cfg || cfg->groups < 32)
2039 groups = 32;
2040 else
2041 groups = cfg->groups;
2042
2043 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2044 if (!listeners)
2045 goto out_sock_release;
2046
2047 sk->sk_data_ready = netlink_data_ready;
2048 if (cfg && cfg->input)
2049 nlk_sk(sk)->netlink_rcv = cfg->input;
2050
2051 if (netlink_insert(sk, 0))
2052 goto out_sock_release;
2053
2054 nlk = nlk_sk(sk);
2055 set_bit(NETLINK_F_KERNEL_SOCKET, &nlk->flags);
2056
2057 netlink_table_grab();
2058 if (!nl_table[unit].registered) {
2059 nl_table[unit].groups = groups;
2060 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2061 nl_table[unit].cb_mutex = cb_mutex;
2062 nl_table[unit].module = module;
2063 if (cfg) {
2064 nl_table[unit].bind = cfg->bind;
2065 nl_table[unit].unbind = cfg->unbind;
2066 nl_table[unit].release = cfg->release;
2067 nl_table[unit].flags = cfg->flags;
2068 }
2069 nl_table[unit].registered = 1;
2070 } else {
2071 kfree(listeners);
2072 nl_table[unit].registered++;
2073 }
2074 netlink_table_ungrab();
2075 return sk;
2076
2077out_sock_release:
2078 kfree(listeners);
2079 netlink_kernel_release(sk);
2080 return NULL;
2081
2082out_sock_release_nosk:
2083 sock_release(sock);
2084 return NULL;
2085}
2086EXPORT_SYMBOL(__netlink_kernel_create);
2087
2088void
2089netlink_kernel_release(struct sock *sk)
2090{
2091 if (sk == NULL || sk->sk_socket == NULL)
2092 return;
2093
2094 sock_release(sk->sk_socket);
2095}
2096EXPORT_SYMBOL(netlink_kernel_release);
2097
2098int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2099{
2100 struct listeners *new, *old;
2101 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2102
2103 if (groups < 32)
2104 groups = 32;
2105
2106 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2107 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2108 if (!new)
2109 return -ENOMEM;
2110 old = nl_deref_protected(tbl->listeners);
2111 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2112 rcu_assign_pointer(tbl->listeners, new);
2113
2114 kfree_rcu(old, rcu);
2115 }
2116 tbl->groups = groups;
2117
2118 return 0;
2119}
2120
2121/**
2122 * netlink_change_ngroups - change number of multicast groups
2123 *
2124 * This changes the number of multicast groups that are available
2125 * on a certain netlink family. Note that it is not possible to
2126 * change the number of groups to below 32. Also note that it does
2127 * not implicitly call netlink_clear_multicast_users() when the
2128 * number of groups is reduced.
2129 *
2130 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2131 * @groups: The new number of groups.
2132 */
2133int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2134{
2135 int err;
2136
2137 netlink_table_grab();
2138 err = __netlink_change_ngroups(sk, groups);
2139 netlink_table_ungrab();
2140
2141 return err;
2142}
2143
2144void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2145{
2146 struct sock *sk;
2147 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2148
2149 sk_for_each_bound(sk, &tbl->mc_list)
2150 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2151}
2152
2153struct nlmsghdr *
2154__nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2155{
2156 struct nlmsghdr *nlh;
2157 int size = nlmsg_msg_size(len);
2158
2159 nlh = skb_put(skb, NLMSG_ALIGN(size));
2160 nlh->nlmsg_type = type;
2161 nlh->nlmsg_len = size;
2162 nlh->nlmsg_flags = flags;
2163 nlh->nlmsg_pid = portid;
2164 nlh->nlmsg_seq = seq;
2165 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2166 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2167 return nlh;
2168}
2169EXPORT_SYMBOL(__nlmsg_put);
2170
2171/*
2172 * It looks a bit ugly.
2173 * It would be better to create kernel thread.
2174 */
2175
2176static int netlink_dump_done(struct netlink_sock *nlk, struct sk_buff *skb,
2177 struct netlink_callback *cb,
2178 struct netlink_ext_ack *extack)
2179{
2180 struct nlmsghdr *nlh;
2181
2182 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(nlk->dump_done_errno),
2183 NLM_F_MULTI | cb->answer_flags);
2184 if (WARN_ON(!nlh))
2185 return -ENOBUFS;
2186
2187 nl_dump_check_consistent(cb, nlh);
2188 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno, sizeof(nlk->dump_done_errno));
2189
2190 if (extack->_msg && test_bit(NETLINK_F_EXT_ACK, &nlk->flags)) {
2191 nlh->nlmsg_flags |= NLM_F_ACK_TLVS;
2192 if (!nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg))
2193 nlmsg_end(skb, nlh);
2194 }
2195
2196 return 0;
2197}
2198
2199static int netlink_dump(struct sock *sk)
2200{
2201 struct netlink_sock *nlk = nlk_sk(sk);
2202 struct netlink_ext_ack extack = {};
2203 struct netlink_callback *cb;
2204 struct sk_buff *skb = NULL;
2205 size_t max_recvmsg_len;
2206 struct module *module;
2207 int err = -ENOBUFS;
2208 int alloc_min_size;
2209 int alloc_size;
2210
2211 mutex_lock(nlk->cb_mutex);
2212 if (!nlk->cb_running) {
2213 err = -EINVAL;
2214 goto errout_skb;
2215 }
2216
2217 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2218 goto errout_skb;
2219
2220 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2221 * required, but it makes sense to _attempt_ a 16K bytes allocation
2222 * to reduce number of system calls on dump operations, if user
2223 * ever provided a big enough buffer.
2224 */
2225 cb = &nlk->cb;
2226 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2227
2228 max_recvmsg_len = READ_ONCE(nlk->max_recvmsg_len);
2229 if (alloc_min_size < max_recvmsg_len) {
2230 alloc_size = max_recvmsg_len;
2231 skb = alloc_skb(alloc_size,
2232 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2233 __GFP_NOWARN | __GFP_NORETRY);
2234 }
2235 if (!skb) {
2236 alloc_size = alloc_min_size;
2237 skb = alloc_skb(alloc_size, GFP_KERNEL);
2238 }
2239 if (!skb)
2240 goto errout_skb;
2241
2242 /* Trim skb to allocated size. User is expected to provide buffer as
2243 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2244 * netlink_recvmsg())). dump will pack as many smaller messages as
2245 * could fit within the allocated skb. skb is typically allocated
2246 * with larger space than required (could be as much as near 2x the
2247 * requested size with align to next power of 2 approach). Allowing
2248 * dump to use the excess space makes it difficult for a user to have a
2249 * reasonable static buffer based on the expected largest dump of a
2250 * single netdev. The outcome is MSG_TRUNC error.
2251 */
2252 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2253
2254 /* Make sure malicious BPF programs can not read unitialized memory
2255 * from skb->head -> skb->data
2256 */
2257 skb_reset_network_header(skb);
2258 skb_reset_mac_header(skb);
2259
2260 netlink_skb_set_owner_r(skb, sk);
2261
2262 if (nlk->dump_done_errno > 0) {
2263 cb->extack = &extack;
2264 nlk->dump_done_errno = cb->dump(skb, cb);
2265 cb->extack = NULL;
2266 }
2267
2268 if (nlk->dump_done_errno > 0 ||
2269 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2270 mutex_unlock(nlk->cb_mutex);
2271
2272 if (sk_filter(sk, skb))
2273 kfree_skb(skb);
2274 else
2275 __netlink_sendskb(sk, skb);
2276 return 0;
2277 }
2278
2279 if (netlink_dump_done(nlk, skb, cb, &extack))
2280 goto errout_skb;
2281
2282#ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2283 /* frag_list skb's data is used for compat tasks
2284 * and the regular skb's data for normal (non-compat) tasks.
2285 * See netlink_recvmsg().
2286 */
2287 if (unlikely(skb_shinfo(skb)->frag_list)) {
2288 if (netlink_dump_done(nlk, skb_shinfo(skb)->frag_list, cb, &extack))
2289 goto errout_skb;
2290 }
2291#endif
2292
2293 if (sk_filter(sk, skb))
2294 kfree_skb(skb);
2295 else
2296 __netlink_sendskb(sk, skb);
2297
2298 if (cb->done)
2299 cb->done(cb);
2300
2301 WRITE_ONCE(nlk->cb_running, false);
2302 module = cb->module;
2303 skb = cb->skb;
2304 mutex_unlock(nlk->cb_mutex);
2305 module_put(module);
2306 consume_skb(skb);
2307 return 0;
2308
2309errout_skb:
2310 mutex_unlock(nlk->cb_mutex);
2311 kfree_skb(skb);
2312 return err;
2313}
2314
2315int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2316 const struct nlmsghdr *nlh,
2317 struct netlink_dump_control *control)
2318{
2319 struct netlink_callback *cb;
2320 struct netlink_sock *nlk;
2321 struct sock *sk;
2322 int ret;
2323
2324 refcount_inc(&skb->users);
2325
2326 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2327 if (sk == NULL) {
2328 ret = -ECONNREFUSED;
2329 goto error_free;
2330 }
2331
2332 nlk = nlk_sk(sk);
2333 mutex_lock(nlk->cb_mutex);
2334 /* A dump is in progress... */
2335 if (nlk->cb_running) {
2336 ret = -EBUSY;
2337 goto error_unlock;
2338 }
2339 /* add reference of module which cb->dump belongs to */
2340 if (!try_module_get(control->module)) {
2341 ret = -EPROTONOSUPPORT;
2342 goto error_unlock;
2343 }
2344
2345 cb = &nlk->cb;
2346 memset(cb, 0, sizeof(*cb));
2347 cb->dump = control->dump;
2348 cb->done = control->done;
2349 cb->nlh = nlh;
2350 cb->data = control->data;
2351 cb->module = control->module;
2352 cb->min_dump_alloc = control->min_dump_alloc;
2353 cb->skb = skb;
2354
2355 cb->strict_check = nlk_test_bit(STRICT_CHK, NETLINK_CB(skb).sk);
2356
2357 if (control->start) {
2358 cb->extack = control->extack;
2359 ret = control->start(cb);
2360 cb->extack = NULL;
2361 if (ret)
2362 goto error_put;
2363 }
2364
2365 WRITE_ONCE(nlk->cb_running, true);
2366 nlk->dump_done_errno = INT_MAX;
2367
2368 mutex_unlock(nlk->cb_mutex);
2369
2370 ret = netlink_dump(sk);
2371
2372 sock_put(sk);
2373
2374 if (ret)
2375 return ret;
2376
2377 /* We successfully started a dump, by returning -EINTR we
2378 * signal not to send ACK even if it was requested.
2379 */
2380 return -EINTR;
2381
2382error_put:
2383 module_put(control->module);
2384error_unlock:
2385 sock_put(sk);
2386 mutex_unlock(nlk->cb_mutex);
2387error_free:
2388 kfree_skb(skb);
2389 return ret;
2390}
2391EXPORT_SYMBOL(__netlink_dump_start);
2392
2393static size_t
2394netlink_ack_tlv_len(struct netlink_sock *nlk, int err,
2395 const struct netlink_ext_ack *extack)
2396{
2397 size_t tlvlen;
2398
2399 if (!extack || !test_bit(NETLINK_F_EXT_ACK, &nlk->flags))
2400 return 0;
2401
2402 tlvlen = 0;
2403 if (extack->_msg)
2404 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2405 if (extack->cookie_len)
2406 tlvlen += nla_total_size(extack->cookie_len);
2407
2408 /* Following attributes are only reported as error (not warning) */
2409 if (!err)
2410 return tlvlen;
2411
2412 if (extack->bad_attr)
2413 tlvlen += nla_total_size(sizeof(u32));
2414 if (extack->policy)
2415 tlvlen += netlink_policy_dump_attr_size_estimate(extack->policy);
2416 if (extack->miss_type)
2417 tlvlen += nla_total_size(sizeof(u32));
2418 if (extack->miss_nest)
2419 tlvlen += nla_total_size(sizeof(u32));
2420
2421 return tlvlen;
2422}
2423
2424static void
2425netlink_ack_tlv_fill(struct sk_buff *in_skb, struct sk_buff *skb,
2426 struct nlmsghdr *nlh, int err,
2427 const struct netlink_ext_ack *extack)
2428{
2429 if (extack->_msg)
2430 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg));
2431 if (extack->cookie_len)
2432 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2433 extack->cookie_len, extack->cookie));
2434
2435 if (!err)
2436 return;
2437
2438 if (extack->bad_attr &&
2439 !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2440 (u8 *)extack->bad_attr >= in_skb->data + in_skb->len))
2441 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2442 (u8 *)extack->bad_attr - (u8 *)nlh));
2443 if (extack->policy)
2444 netlink_policy_dump_write_attr(skb, extack->policy,
2445 NLMSGERR_ATTR_POLICY);
2446 if (extack->miss_type)
2447 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_TYPE,
2448 extack->miss_type));
2449 if (extack->miss_nest &&
2450 !WARN_ON((u8 *)extack->miss_nest < in_skb->data ||
2451 (u8 *)extack->miss_nest > in_skb->data + in_skb->len))
2452 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_NEST,
2453 (u8 *)extack->miss_nest - (u8 *)nlh));
2454}
2455
2456void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2457 const struct netlink_ext_ack *extack)
2458{
2459 struct sk_buff *skb;
2460 struct nlmsghdr *rep;
2461 struct nlmsgerr *errmsg;
2462 size_t payload = sizeof(*errmsg);
2463 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2464 unsigned int flags = 0;
2465 size_t tlvlen;
2466
2467 /* Error messages get the original request appened, unless the user
2468 * requests to cap the error message, and get extra error data if
2469 * requested.
2470 */
2471 if (err && !test_bit(NETLINK_F_CAP_ACK, &nlk->flags))
2472 payload += nlmsg_len(nlh);
2473 else
2474 flags |= NLM_F_CAPPED;
2475
2476 tlvlen = netlink_ack_tlv_len(nlk, err, extack);
2477 if (tlvlen)
2478 flags |= NLM_F_ACK_TLVS;
2479
2480 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2481 if (!skb)
2482 goto err_skb;
2483
2484 rep = nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2485 NLMSG_ERROR, sizeof(*errmsg), flags);
2486 if (!rep)
2487 goto err_bad_put;
2488 errmsg = nlmsg_data(rep);
2489 errmsg->error = err;
2490 errmsg->msg = *nlh;
2491
2492 if (!(flags & NLM_F_CAPPED)) {
2493 if (!nlmsg_append(skb, nlmsg_len(nlh)))
2494 goto err_bad_put;
2495
2496 memcpy(nlmsg_data(&errmsg->msg), nlmsg_data(nlh),
2497 nlmsg_len(nlh));
2498 }
2499
2500 if (tlvlen)
2501 netlink_ack_tlv_fill(in_skb, skb, nlh, err, extack);
2502
2503 nlmsg_end(skb, rep);
2504
2505 nlmsg_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid);
2506
2507 return;
2508
2509err_bad_put:
2510 nlmsg_free(skb);
2511err_skb:
2512 WRITE_ONCE(NETLINK_CB(in_skb).sk->sk_err, ENOBUFS);
2513 sk_error_report(NETLINK_CB(in_skb).sk);
2514}
2515EXPORT_SYMBOL(netlink_ack);
2516
2517int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2518 struct nlmsghdr *,
2519 struct netlink_ext_ack *))
2520{
2521 struct netlink_ext_ack extack;
2522 struct nlmsghdr *nlh;
2523 int err;
2524
2525 while (skb->len >= nlmsg_total_size(0)) {
2526 int msglen;
2527
2528 memset(&extack, 0, sizeof(extack));
2529 nlh = nlmsg_hdr(skb);
2530 err = 0;
2531
2532 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2533 return 0;
2534
2535 /* Only requests are handled by the kernel */
2536 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2537 goto ack;
2538
2539 /* Skip control messages */
2540 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2541 goto ack;
2542
2543 err = cb(skb, nlh, &extack);
2544 if (err == -EINTR)
2545 goto skip;
2546
2547ack:
2548 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2549 netlink_ack(skb, nlh, err, &extack);
2550
2551skip:
2552 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2553 if (msglen > skb->len)
2554 msglen = skb->len;
2555 skb_pull(skb, msglen);
2556 }
2557
2558 return 0;
2559}
2560EXPORT_SYMBOL(netlink_rcv_skb);
2561
2562/**
2563 * nlmsg_notify - send a notification netlink message
2564 * @sk: netlink socket to use
2565 * @skb: notification message
2566 * @portid: destination netlink portid for reports or 0
2567 * @group: destination multicast group or 0
2568 * @report: 1 to report back, 0 to disable
2569 * @flags: allocation flags
2570 */
2571int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2572 unsigned int group, int report, gfp_t flags)
2573{
2574 int err = 0;
2575
2576 if (group) {
2577 int exclude_portid = 0;
2578
2579 if (report) {
2580 refcount_inc(&skb->users);
2581 exclude_portid = portid;
2582 }
2583
2584 /* errors reported via destination sk->sk_err, but propagate
2585 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2586 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2587 if (err == -ESRCH)
2588 err = 0;
2589 }
2590
2591 if (report) {
2592 int err2;
2593
2594 err2 = nlmsg_unicast(sk, skb, portid);
2595 if (!err)
2596 err = err2;
2597 }
2598
2599 return err;
2600}
2601EXPORT_SYMBOL(nlmsg_notify);
2602
2603#ifdef CONFIG_PROC_FS
2604struct nl_seq_iter {
2605 struct seq_net_private p;
2606 struct rhashtable_iter hti;
2607 int link;
2608};
2609
2610static void netlink_walk_start(struct nl_seq_iter *iter)
2611{
2612 rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti);
2613 rhashtable_walk_start(&iter->hti);
2614}
2615
2616static void netlink_walk_stop(struct nl_seq_iter *iter)
2617{
2618 rhashtable_walk_stop(&iter->hti);
2619 rhashtable_walk_exit(&iter->hti);
2620}
2621
2622static void *__netlink_seq_next(struct seq_file *seq)
2623{
2624 struct nl_seq_iter *iter = seq->private;
2625 struct netlink_sock *nlk;
2626
2627 do {
2628 for (;;) {
2629 nlk = rhashtable_walk_next(&iter->hti);
2630
2631 if (IS_ERR(nlk)) {
2632 if (PTR_ERR(nlk) == -EAGAIN)
2633 continue;
2634
2635 return nlk;
2636 }
2637
2638 if (nlk)
2639 break;
2640
2641 netlink_walk_stop(iter);
2642 if (++iter->link >= MAX_LINKS)
2643 return NULL;
2644
2645 netlink_walk_start(iter);
2646 }
2647 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2648
2649 return nlk;
2650}
2651
2652static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2653 __acquires(RCU)
2654{
2655 struct nl_seq_iter *iter = seq->private;
2656 void *obj = SEQ_START_TOKEN;
2657 loff_t pos;
2658
2659 iter->link = 0;
2660
2661 netlink_walk_start(iter);
2662
2663 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2664 obj = __netlink_seq_next(seq);
2665
2666 return obj;
2667}
2668
2669static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2670{
2671 ++*pos;
2672 return __netlink_seq_next(seq);
2673}
2674
2675static void netlink_native_seq_stop(struct seq_file *seq, void *v)
2676{
2677 struct nl_seq_iter *iter = seq->private;
2678
2679 if (iter->link >= MAX_LINKS)
2680 return;
2681
2682 netlink_walk_stop(iter);
2683}
2684
2685
2686static int netlink_native_seq_show(struct seq_file *seq, void *v)
2687{
2688 if (v == SEQ_START_TOKEN) {
2689 seq_puts(seq,
2690 "sk Eth Pid Groups "
2691 "Rmem Wmem Dump Locks Drops Inode\n");
2692 } else {
2693 struct sock *s = v;
2694 struct netlink_sock *nlk = nlk_sk(s);
2695
2696 seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n",
2697 s,
2698 s->sk_protocol,
2699 nlk->portid,
2700 nlk->groups ? (u32)nlk->groups[0] : 0,
2701 sk_rmem_alloc_get(s),
2702 sk_wmem_alloc_get(s),
2703 READ_ONCE(nlk->cb_running),
2704 refcount_read(&s->sk_refcnt),
2705 atomic_read(&s->sk_drops),
2706 sock_i_ino(s)
2707 );
2708
2709 }
2710 return 0;
2711}
2712
2713#ifdef CONFIG_BPF_SYSCALL
2714struct bpf_iter__netlink {
2715 __bpf_md_ptr(struct bpf_iter_meta *, meta);
2716 __bpf_md_ptr(struct netlink_sock *, sk);
2717};
2718
2719DEFINE_BPF_ITER_FUNC(netlink, struct bpf_iter_meta *meta, struct netlink_sock *sk)
2720
2721static int netlink_prog_seq_show(struct bpf_prog *prog,
2722 struct bpf_iter_meta *meta,
2723 void *v)
2724{
2725 struct bpf_iter__netlink ctx;
2726
2727 meta->seq_num--; /* skip SEQ_START_TOKEN */
2728 ctx.meta = meta;
2729 ctx.sk = nlk_sk((struct sock *)v);
2730 return bpf_iter_run_prog(prog, &ctx);
2731}
2732
2733static int netlink_seq_show(struct seq_file *seq, void *v)
2734{
2735 struct bpf_iter_meta meta;
2736 struct bpf_prog *prog;
2737
2738 meta.seq = seq;
2739 prog = bpf_iter_get_info(&meta, false);
2740 if (!prog)
2741 return netlink_native_seq_show(seq, v);
2742
2743 if (v != SEQ_START_TOKEN)
2744 return netlink_prog_seq_show(prog, &meta, v);
2745
2746 return 0;
2747}
2748
2749static void netlink_seq_stop(struct seq_file *seq, void *v)
2750{
2751 struct bpf_iter_meta meta;
2752 struct bpf_prog *prog;
2753
2754 if (!v) {
2755 meta.seq = seq;
2756 prog = bpf_iter_get_info(&meta, true);
2757 if (prog)
2758 (void)netlink_prog_seq_show(prog, &meta, v);
2759 }
2760
2761 netlink_native_seq_stop(seq, v);
2762}
2763#else
2764static int netlink_seq_show(struct seq_file *seq, void *v)
2765{
2766 return netlink_native_seq_show(seq, v);
2767}
2768
2769static void netlink_seq_stop(struct seq_file *seq, void *v)
2770{
2771 netlink_native_seq_stop(seq, v);
2772}
2773#endif
2774
2775static const struct seq_operations netlink_seq_ops = {
2776 .start = netlink_seq_start,
2777 .next = netlink_seq_next,
2778 .stop = netlink_seq_stop,
2779 .show = netlink_seq_show,
2780};
2781#endif
2782
2783int netlink_register_notifier(struct notifier_block *nb)
2784{
2785 return blocking_notifier_chain_register(&netlink_chain, nb);
2786}
2787EXPORT_SYMBOL(netlink_register_notifier);
2788
2789int netlink_unregister_notifier(struct notifier_block *nb)
2790{
2791 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2792}
2793EXPORT_SYMBOL(netlink_unregister_notifier);
2794
2795static const struct proto_ops netlink_ops = {
2796 .family = PF_NETLINK,
2797 .owner = THIS_MODULE,
2798 .release = netlink_release,
2799 .bind = netlink_bind,
2800 .connect = netlink_connect,
2801 .socketpair = sock_no_socketpair,
2802 .accept = sock_no_accept,
2803 .getname = netlink_getname,
2804 .poll = datagram_poll,
2805 .ioctl = netlink_ioctl,
2806 .listen = sock_no_listen,
2807 .shutdown = sock_no_shutdown,
2808 .setsockopt = netlink_setsockopt,
2809 .getsockopt = netlink_getsockopt,
2810 .sendmsg = netlink_sendmsg,
2811 .recvmsg = netlink_recvmsg,
2812 .mmap = sock_no_mmap,
2813};
2814
2815static const struct net_proto_family netlink_family_ops = {
2816 .family = PF_NETLINK,
2817 .create = netlink_create,
2818 .owner = THIS_MODULE, /* for consistency 8) */
2819};
2820
2821static int __net_init netlink_net_init(struct net *net)
2822{
2823#ifdef CONFIG_PROC_FS
2824 if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2825 sizeof(struct nl_seq_iter)))
2826 return -ENOMEM;
2827#endif
2828 return 0;
2829}
2830
2831static void __net_exit netlink_net_exit(struct net *net)
2832{
2833#ifdef CONFIG_PROC_FS
2834 remove_proc_entry("netlink", net->proc_net);
2835#endif
2836}
2837
2838static void __init netlink_add_usersock_entry(void)
2839{
2840 struct listeners *listeners;
2841 int groups = 32;
2842
2843 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2844 if (!listeners)
2845 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2846
2847 netlink_table_grab();
2848
2849 nl_table[NETLINK_USERSOCK].groups = groups;
2850 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2851 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2852 nl_table[NETLINK_USERSOCK].registered = 1;
2853 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2854
2855 netlink_table_ungrab();
2856}
2857
2858static struct pernet_operations __net_initdata netlink_net_ops = {
2859 .init = netlink_net_init,
2860 .exit = netlink_net_exit,
2861};
2862
2863static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2864{
2865 const struct netlink_sock *nlk = data;
2866 struct netlink_compare_arg arg;
2867
2868 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2869 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2870}
2871
2872static const struct rhashtable_params netlink_rhashtable_params = {
2873 .head_offset = offsetof(struct netlink_sock, node),
2874 .key_len = netlink_compare_arg_len,
2875 .obj_hashfn = netlink_hash,
2876 .obj_cmpfn = netlink_compare,
2877 .automatic_shrinking = true,
2878};
2879
2880#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2881BTF_ID_LIST(btf_netlink_sock_id)
2882BTF_ID(struct, netlink_sock)
2883
2884static const struct bpf_iter_seq_info netlink_seq_info = {
2885 .seq_ops = &netlink_seq_ops,
2886 .init_seq_private = bpf_iter_init_seq_net,
2887 .fini_seq_private = bpf_iter_fini_seq_net,
2888 .seq_priv_size = sizeof(struct nl_seq_iter),
2889};
2890
2891static struct bpf_iter_reg netlink_reg_info = {
2892 .target = "netlink",
2893 .ctx_arg_info_size = 1,
2894 .ctx_arg_info = {
2895 { offsetof(struct bpf_iter__netlink, sk),
2896 PTR_TO_BTF_ID_OR_NULL },
2897 },
2898 .seq_info = &netlink_seq_info,
2899};
2900
2901static int __init bpf_iter_register(void)
2902{
2903 netlink_reg_info.ctx_arg_info[0].btf_id = *btf_netlink_sock_id;
2904 return bpf_iter_reg_target(&netlink_reg_info);
2905}
2906#endif
2907
2908static int __init netlink_proto_init(void)
2909{
2910 int i;
2911 int err = proto_register(&netlink_proto, 0);
2912
2913 if (err != 0)
2914 goto out;
2915
2916#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2917 err = bpf_iter_register();
2918 if (err)
2919 goto out;
2920#endif
2921
2922 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof_field(struct sk_buff, cb));
2923
2924 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2925 if (!nl_table)
2926 goto panic;
2927
2928 for (i = 0; i < MAX_LINKS; i++) {
2929 if (rhashtable_init(&nl_table[i].hash,
2930 &netlink_rhashtable_params) < 0) {
2931 while (--i > 0)
2932 rhashtable_destroy(&nl_table[i].hash);
2933 kfree(nl_table);
2934 goto panic;
2935 }
2936 }
2937
2938 netlink_add_usersock_entry();
2939
2940 sock_register(&netlink_family_ops);
2941 register_pernet_subsys(&netlink_net_ops);
2942 register_pernet_subsys(&netlink_tap_net_ops);
2943 /* The netlink device handler may be needed early. */
2944 rtnetlink_init();
2945out:
2946 return err;
2947panic:
2948 panic("netlink_init: Cannot allocate nl_table\n");
2949}
2950
2951core_initcall(netlink_proto_init);
1/*
2 * NETLINK Kernel-user communication protocol.
3 *
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
6 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
11 *
12 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
13 * added netlink_proto_exit
14 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
15 * use nlk_sk, as sk->protinfo is on a diet 8)
16 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
17 * - inc module use count of module that owns
18 * the kernel socket in case userspace opens
19 * socket of same protocol
20 * - remove all module support, since netlink is
21 * mandatory if CONFIG_NET=y these days
22 */
23
24#include <linux/module.h>
25
26#include <linux/capability.h>
27#include <linux/kernel.h>
28#include <linux/init.h>
29#include <linux/signal.h>
30#include <linux/sched.h>
31#include <linux/errno.h>
32#include <linux/string.h>
33#include <linux/stat.h>
34#include <linux/socket.h>
35#include <linux/un.h>
36#include <linux/fcntl.h>
37#include <linux/termios.h>
38#include <linux/sockios.h>
39#include <linux/net.h>
40#include <linux/fs.h>
41#include <linux/slab.h>
42#include <asm/uaccess.h>
43#include <linux/skbuff.h>
44#include <linux/netdevice.h>
45#include <linux/rtnetlink.h>
46#include <linux/proc_fs.h>
47#include <linux/seq_file.h>
48#include <linux/notifier.h>
49#include <linux/security.h>
50#include <linux/jhash.h>
51#include <linux/jiffies.h>
52#include <linux/random.h>
53#include <linux/bitops.h>
54#include <linux/mm.h>
55#include <linux/types.h>
56#include <linux/audit.h>
57#include <linux/mutex.h>
58
59#include <net/net_namespace.h>
60#include <net/sock.h>
61#include <net/scm.h>
62#include <net/netlink.h>
63
64#define NLGRPSZ(x) (ALIGN(x, sizeof(unsigned long) * 8) / 8)
65#define NLGRPLONGS(x) (NLGRPSZ(x)/sizeof(unsigned long))
66
67struct netlink_sock {
68 /* struct sock has to be the first member of netlink_sock */
69 struct sock sk;
70 u32 pid;
71 u32 dst_pid;
72 u32 dst_group;
73 u32 flags;
74 u32 subscriptions;
75 u32 ngroups;
76 unsigned long *groups;
77 unsigned long state;
78 wait_queue_head_t wait;
79 struct netlink_callback *cb;
80 struct mutex *cb_mutex;
81 struct mutex cb_def_mutex;
82 void (*netlink_rcv)(struct sk_buff *skb);
83 struct module *module;
84};
85
86struct listeners {
87 struct rcu_head rcu;
88 unsigned long masks[0];
89};
90
91#define NETLINK_KERNEL_SOCKET 0x1
92#define NETLINK_RECV_PKTINFO 0x2
93#define NETLINK_BROADCAST_SEND_ERROR 0x4
94#define NETLINK_RECV_NO_ENOBUFS 0x8
95
96static inline struct netlink_sock *nlk_sk(struct sock *sk)
97{
98 return container_of(sk, struct netlink_sock, sk);
99}
100
101static inline int netlink_is_kernel(struct sock *sk)
102{
103 return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET;
104}
105
106struct nl_pid_hash {
107 struct hlist_head *table;
108 unsigned long rehash_time;
109
110 unsigned int mask;
111 unsigned int shift;
112
113 unsigned int entries;
114 unsigned int max_shift;
115
116 u32 rnd;
117};
118
119struct netlink_table {
120 struct nl_pid_hash hash;
121 struct hlist_head mc_list;
122 struct listeners __rcu *listeners;
123 unsigned int nl_nonroot;
124 unsigned int groups;
125 struct mutex *cb_mutex;
126 struct module *module;
127 int registered;
128};
129
130static struct netlink_table *nl_table;
131
132static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
133
134static int netlink_dump(struct sock *sk);
135
136static DEFINE_RWLOCK(nl_table_lock);
137static atomic_t nl_table_users = ATOMIC_INIT(0);
138
139static ATOMIC_NOTIFIER_HEAD(netlink_chain);
140
141static inline u32 netlink_group_mask(u32 group)
142{
143 return group ? 1 << (group - 1) : 0;
144}
145
146static inline struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
147{
148 return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
149}
150
151static void netlink_destroy_callback(struct netlink_callback *cb)
152{
153 kfree_skb(cb->skb);
154 kfree(cb);
155}
156
157static void netlink_consume_callback(struct netlink_callback *cb)
158{
159 consume_skb(cb->skb);
160 kfree(cb);
161}
162
163static void netlink_sock_destruct(struct sock *sk)
164{
165 struct netlink_sock *nlk = nlk_sk(sk);
166
167 if (nlk->cb) {
168 if (nlk->cb->done)
169 nlk->cb->done(nlk->cb);
170 netlink_destroy_callback(nlk->cb);
171 }
172
173 skb_queue_purge(&sk->sk_receive_queue);
174
175 if (!sock_flag(sk, SOCK_DEAD)) {
176 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
177 return;
178 }
179
180 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
181 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
182 WARN_ON(nlk_sk(sk)->groups);
183}
184
185/* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
186 * SMP. Look, when several writers sleep and reader wakes them up, all but one
187 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
188 * this, _but_ remember, it adds useless work on UP machines.
189 */
190
191void netlink_table_grab(void)
192 __acquires(nl_table_lock)
193{
194 might_sleep();
195
196 write_lock_irq(&nl_table_lock);
197
198 if (atomic_read(&nl_table_users)) {
199 DECLARE_WAITQUEUE(wait, current);
200
201 add_wait_queue_exclusive(&nl_table_wait, &wait);
202 for (;;) {
203 set_current_state(TASK_UNINTERRUPTIBLE);
204 if (atomic_read(&nl_table_users) == 0)
205 break;
206 write_unlock_irq(&nl_table_lock);
207 schedule();
208 write_lock_irq(&nl_table_lock);
209 }
210
211 __set_current_state(TASK_RUNNING);
212 remove_wait_queue(&nl_table_wait, &wait);
213 }
214}
215
216void netlink_table_ungrab(void)
217 __releases(nl_table_lock)
218{
219 write_unlock_irq(&nl_table_lock);
220 wake_up(&nl_table_wait);
221}
222
223static inline void
224netlink_lock_table(void)
225{
226 /* read_lock() synchronizes us to netlink_table_grab */
227
228 read_lock(&nl_table_lock);
229 atomic_inc(&nl_table_users);
230 read_unlock(&nl_table_lock);
231}
232
233static inline void
234netlink_unlock_table(void)
235{
236 if (atomic_dec_and_test(&nl_table_users))
237 wake_up(&nl_table_wait);
238}
239
240static struct sock *netlink_lookup(struct net *net, int protocol, u32 pid)
241{
242 struct nl_pid_hash *hash = &nl_table[protocol].hash;
243 struct hlist_head *head;
244 struct sock *sk;
245 struct hlist_node *node;
246
247 read_lock(&nl_table_lock);
248 head = nl_pid_hashfn(hash, pid);
249 sk_for_each(sk, node, head) {
250 if (net_eq(sock_net(sk), net) && (nlk_sk(sk)->pid == pid)) {
251 sock_hold(sk);
252 goto found;
253 }
254 }
255 sk = NULL;
256found:
257 read_unlock(&nl_table_lock);
258 return sk;
259}
260
261static struct hlist_head *nl_pid_hash_zalloc(size_t size)
262{
263 if (size <= PAGE_SIZE)
264 return kzalloc(size, GFP_ATOMIC);
265 else
266 return (struct hlist_head *)
267 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
268 get_order(size));
269}
270
271static void nl_pid_hash_free(struct hlist_head *table, size_t size)
272{
273 if (size <= PAGE_SIZE)
274 kfree(table);
275 else
276 free_pages((unsigned long)table, get_order(size));
277}
278
279static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
280{
281 unsigned int omask, mask, shift;
282 size_t osize, size;
283 struct hlist_head *otable, *table;
284 int i;
285
286 omask = mask = hash->mask;
287 osize = size = (mask + 1) * sizeof(*table);
288 shift = hash->shift;
289
290 if (grow) {
291 if (++shift > hash->max_shift)
292 return 0;
293 mask = mask * 2 + 1;
294 size *= 2;
295 }
296
297 table = nl_pid_hash_zalloc(size);
298 if (!table)
299 return 0;
300
301 otable = hash->table;
302 hash->table = table;
303 hash->mask = mask;
304 hash->shift = shift;
305 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
306
307 for (i = 0; i <= omask; i++) {
308 struct sock *sk;
309 struct hlist_node *node, *tmp;
310
311 sk_for_each_safe(sk, node, tmp, &otable[i])
312 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
313 }
314
315 nl_pid_hash_free(otable, osize);
316 hash->rehash_time = jiffies + 10 * 60 * HZ;
317 return 1;
318}
319
320static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
321{
322 int avg = hash->entries >> hash->shift;
323
324 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
325 return 1;
326
327 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
328 nl_pid_hash_rehash(hash, 0);
329 return 1;
330 }
331
332 return 0;
333}
334
335static const struct proto_ops netlink_ops;
336
337static void
338netlink_update_listeners(struct sock *sk)
339{
340 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
341 struct hlist_node *node;
342 unsigned long mask;
343 unsigned int i;
344
345 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
346 mask = 0;
347 sk_for_each_bound(sk, node, &tbl->mc_list) {
348 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
349 mask |= nlk_sk(sk)->groups[i];
350 }
351 tbl->listeners->masks[i] = mask;
352 }
353 /* this function is only called with the netlink table "grabbed", which
354 * makes sure updates are visible before bind or setsockopt return. */
355}
356
357static int netlink_insert(struct sock *sk, struct net *net, u32 pid)
358{
359 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
360 struct hlist_head *head;
361 int err = -EADDRINUSE;
362 struct sock *osk;
363 struct hlist_node *node;
364 int len;
365
366 netlink_table_grab();
367 head = nl_pid_hashfn(hash, pid);
368 len = 0;
369 sk_for_each(osk, node, head) {
370 if (net_eq(sock_net(osk), net) && (nlk_sk(osk)->pid == pid))
371 break;
372 len++;
373 }
374 if (node)
375 goto err;
376
377 err = -EBUSY;
378 if (nlk_sk(sk)->pid)
379 goto err;
380
381 err = -ENOMEM;
382 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
383 goto err;
384
385 if (len && nl_pid_hash_dilute(hash, len))
386 head = nl_pid_hashfn(hash, pid);
387 hash->entries++;
388 nlk_sk(sk)->pid = pid;
389 sk_add_node(sk, head);
390 err = 0;
391
392err:
393 netlink_table_ungrab();
394 return err;
395}
396
397static void netlink_remove(struct sock *sk)
398{
399 netlink_table_grab();
400 if (sk_del_node_init(sk))
401 nl_table[sk->sk_protocol].hash.entries--;
402 if (nlk_sk(sk)->subscriptions)
403 __sk_del_bind_node(sk);
404 netlink_table_ungrab();
405}
406
407static struct proto netlink_proto = {
408 .name = "NETLINK",
409 .owner = THIS_MODULE,
410 .obj_size = sizeof(struct netlink_sock),
411};
412
413static int __netlink_create(struct net *net, struct socket *sock,
414 struct mutex *cb_mutex, int protocol)
415{
416 struct sock *sk;
417 struct netlink_sock *nlk;
418
419 sock->ops = &netlink_ops;
420
421 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
422 if (!sk)
423 return -ENOMEM;
424
425 sock_init_data(sock, sk);
426
427 nlk = nlk_sk(sk);
428 if (cb_mutex) {
429 nlk->cb_mutex = cb_mutex;
430 } else {
431 nlk->cb_mutex = &nlk->cb_def_mutex;
432 mutex_init(nlk->cb_mutex);
433 }
434 init_waitqueue_head(&nlk->wait);
435
436 sk->sk_destruct = netlink_sock_destruct;
437 sk->sk_protocol = protocol;
438 return 0;
439}
440
441static int netlink_create(struct net *net, struct socket *sock, int protocol,
442 int kern)
443{
444 struct module *module = NULL;
445 struct mutex *cb_mutex;
446 struct netlink_sock *nlk;
447 int err = 0;
448
449 sock->state = SS_UNCONNECTED;
450
451 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
452 return -ESOCKTNOSUPPORT;
453
454 if (protocol < 0 || protocol >= MAX_LINKS)
455 return -EPROTONOSUPPORT;
456
457 netlink_lock_table();
458#ifdef CONFIG_MODULES
459 if (!nl_table[protocol].registered) {
460 netlink_unlock_table();
461 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
462 netlink_lock_table();
463 }
464#endif
465 if (nl_table[protocol].registered &&
466 try_module_get(nl_table[protocol].module))
467 module = nl_table[protocol].module;
468 else
469 err = -EPROTONOSUPPORT;
470 cb_mutex = nl_table[protocol].cb_mutex;
471 netlink_unlock_table();
472
473 if (err < 0)
474 goto out;
475
476 err = __netlink_create(net, sock, cb_mutex, protocol);
477 if (err < 0)
478 goto out_module;
479
480 local_bh_disable();
481 sock_prot_inuse_add(net, &netlink_proto, 1);
482 local_bh_enable();
483
484 nlk = nlk_sk(sock->sk);
485 nlk->module = module;
486out:
487 return err;
488
489out_module:
490 module_put(module);
491 goto out;
492}
493
494static int netlink_release(struct socket *sock)
495{
496 struct sock *sk = sock->sk;
497 struct netlink_sock *nlk;
498
499 if (!sk)
500 return 0;
501
502 netlink_remove(sk);
503 sock_orphan(sk);
504 nlk = nlk_sk(sk);
505
506 /*
507 * OK. Socket is unlinked, any packets that arrive now
508 * will be purged.
509 */
510
511 sock->sk = NULL;
512 wake_up_interruptible_all(&nlk->wait);
513
514 skb_queue_purge(&sk->sk_write_queue);
515
516 if (nlk->pid) {
517 struct netlink_notify n = {
518 .net = sock_net(sk),
519 .protocol = sk->sk_protocol,
520 .pid = nlk->pid,
521 };
522 atomic_notifier_call_chain(&netlink_chain,
523 NETLINK_URELEASE, &n);
524 }
525
526 module_put(nlk->module);
527
528 netlink_table_grab();
529 if (netlink_is_kernel(sk)) {
530 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
531 if (--nl_table[sk->sk_protocol].registered == 0) {
532 kfree(nl_table[sk->sk_protocol].listeners);
533 nl_table[sk->sk_protocol].module = NULL;
534 nl_table[sk->sk_protocol].registered = 0;
535 }
536 } else if (nlk->subscriptions) {
537 netlink_update_listeners(sk);
538 }
539 netlink_table_ungrab();
540
541 kfree(nlk->groups);
542 nlk->groups = NULL;
543
544 local_bh_disable();
545 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
546 local_bh_enable();
547 sock_put(sk);
548 return 0;
549}
550
551static int netlink_autobind(struct socket *sock)
552{
553 struct sock *sk = sock->sk;
554 struct net *net = sock_net(sk);
555 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
556 struct hlist_head *head;
557 struct sock *osk;
558 struct hlist_node *node;
559 s32 pid = task_tgid_vnr(current);
560 int err;
561 static s32 rover = -4097;
562
563retry:
564 cond_resched();
565 netlink_table_grab();
566 head = nl_pid_hashfn(hash, pid);
567 sk_for_each(osk, node, head) {
568 if (!net_eq(sock_net(osk), net))
569 continue;
570 if (nlk_sk(osk)->pid == pid) {
571 /* Bind collision, search negative pid values. */
572 pid = rover--;
573 if (rover > -4097)
574 rover = -4097;
575 netlink_table_ungrab();
576 goto retry;
577 }
578 }
579 netlink_table_ungrab();
580
581 err = netlink_insert(sk, net, pid);
582 if (err == -EADDRINUSE)
583 goto retry;
584
585 /* If 2 threads race to autobind, that is fine. */
586 if (err == -EBUSY)
587 err = 0;
588
589 return err;
590}
591
592static inline int netlink_capable(const struct socket *sock, unsigned int flag)
593{
594 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
595 capable(CAP_NET_ADMIN);
596}
597
598static void
599netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
600{
601 struct netlink_sock *nlk = nlk_sk(sk);
602
603 if (nlk->subscriptions && !subscriptions)
604 __sk_del_bind_node(sk);
605 else if (!nlk->subscriptions && subscriptions)
606 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
607 nlk->subscriptions = subscriptions;
608}
609
610static int netlink_realloc_groups(struct sock *sk)
611{
612 struct netlink_sock *nlk = nlk_sk(sk);
613 unsigned int groups;
614 unsigned long *new_groups;
615 int err = 0;
616
617 netlink_table_grab();
618
619 groups = nl_table[sk->sk_protocol].groups;
620 if (!nl_table[sk->sk_protocol].registered) {
621 err = -ENOENT;
622 goto out_unlock;
623 }
624
625 if (nlk->ngroups >= groups)
626 goto out_unlock;
627
628 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
629 if (new_groups == NULL) {
630 err = -ENOMEM;
631 goto out_unlock;
632 }
633 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
634 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
635
636 nlk->groups = new_groups;
637 nlk->ngroups = groups;
638 out_unlock:
639 netlink_table_ungrab();
640 return err;
641}
642
643static int netlink_bind(struct socket *sock, struct sockaddr *addr,
644 int addr_len)
645{
646 struct sock *sk = sock->sk;
647 struct net *net = sock_net(sk);
648 struct netlink_sock *nlk = nlk_sk(sk);
649 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
650 int err;
651
652 if (nladdr->nl_family != AF_NETLINK)
653 return -EINVAL;
654
655 /* Only superuser is allowed to listen multicasts */
656 if (nladdr->nl_groups) {
657 if (!netlink_capable(sock, NL_NONROOT_RECV))
658 return -EPERM;
659 err = netlink_realloc_groups(sk);
660 if (err)
661 return err;
662 }
663
664 if (nlk->pid) {
665 if (nladdr->nl_pid != nlk->pid)
666 return -EINVAL;
667 } else {
668 err = nladdr->nl_pid ?
669 netlink_insert(sk, net, nladdr->nl_pid) :
670 netlink_autobind(sock);
671 if (err)
672 return err;
673 }
674
675 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
676 return 0;
677
678 netlink_table_grab();
679 netlink_update_subscriptions(sk, nlk->subscriptions +
680 hweight32(nladdr->nl_groups) -
681 hweight32(nlk->groups[0]));
682 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
683 netlink_update_listeners(sk);
684 netlink_table_ungrab();
685
686 return 0;
687}
688
689static int netlink_connect(struct socket *sock, struct sockaddr *addr,
690 int alen, int flags)
691{
692 int err = 0;
693 struct sock *sk = sock->sk;
694 struct netlink_sock *nlk = nlk_sk(sk);
695 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
696
697 if (alen < sizeof(addr->sa_family))
698 return -EINVAL;
699
700 if (addr->sa_family == AF_UNSPEC) {
701 sk->sk_state = NETLINK_UNCONNECTED;
702 nlk->dst_pid = 0;
703 nlk->dst_group = 0;
704 return 0;
705 }
706 if (addr->sa_family != AF_NETLINK)
707 return -EINVAL;
708
709 /* Only superuser is allowed to send multicasts */
710 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
711 return -EPERM;
712
713 if (!nlk->pid)
714 err = netlink_autobind(sock);
715
716 if (err == 0) {
717 sk->sk_state = NETLINK_CONNECTED;
718 nlk->dst_pid = nladdr->nl_pid;
719 nlk->dst_group = ffs(nladdr->nl_groups);
720 }
721
722 return err;
723}
724
725static int netlink_getname(struct socket *sock, struct sockaddr *addr,
726 int *addr_len, int peer)
727{
728 struct sock *sk = sock->sk;
729 struct netlink_sock *nlk = nlk_sk(sk);
730 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
731
732 nladdr->nl_family = AF_NETLINK;
733 nladdr->nl_pad = 0;
734 *addr_len = sizeof(*nladdr);
735
736 if (peer) {
737 nladdr->nl_pid = nlk->dst_pid;
738 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
739 } else {
740 nladdr->nl_pid = nlk->pid;
741 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
742 }
743 return 0;
744}
745
746static void netlink_overrun(struct sock *sk)
747{
748 struct netlink_sock *nlk = nlk_sk(sk);
749
750 if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) {
751 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
752 sk->sk_err = ENOBUFS;
753 sk->sk_error_report(sk);
754 }
755 }
756 atomic_inc(&sk->sk_drops);
757}
758
759static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
760{
761 struct sock *sock;
762 struct netlink_sock *nlk;
763
764 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, pid);
765 if (!sock)
766 return ERR_PTR(-ECONNREFUSED);
767
768 /* Don't bother queuing skb if kernel socket has no input function */
769 nlk = nlk_sk(sock);
770 if (sock->sk_state == NETLINK_CONNECTED &&
771 nlk->dst_pid != nlk_sk(ssk)->pid) {
772 sock_put(sock);
773 return ERR_PTR(-ECONNREFUSED);
774 }
775 return sock;
776}
777
778struct sock *netlink_getsockbyfilp(struct file *filp)
779{
780 struct inode *inode = filp->f_path.dentry->d_inode;
781 struct sock *sock;
782
783 if (!S_ISSOCK(inode->i_mode))
784 return ERR_PTR(-ENOTSOCK);
785
786 sock = SOCKET_I(inode)->sk;
787 if (sock->sk_family != AF_NETLINK)
788 return ERR_PTR(-EINVAL);
789
790 sock_hold(sock);
791 return sock;
792}
793
794/*
795 * Attach a skb to a netlink socket.
796 * The caller must hold a reference to the destination socket. On error, the
797 * reference is dropped. The skb is not send to the destination, just all
798 * all error checks are performed and memory in the queue is reserved.
799 * Return values:
800 * < 0: error. skb freed, reference to sock dropped.
801 * 0: continue
802 * 1: repeat lookup - reference dropped while waiting for socket memory.
803 */
804int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
805 long *timeo, struct sock *ssk)
806{
807 struct netlink_sock *nlk;
808
809 nlk = nlk_sk(sk);
810
811 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
812 test_bit(0, &nlk->state)) {
813 DECLARE_WAITQUEUE(wait, current);
814 if (!*timeo) {
815 if (!ssk || netlink_is_kernel(ssk))
816 netlink_overrun(sk);
817 sock_put(sk);
818 kfree_skb(skb);
819 return -EAGAIN;
820 }
821
822 __set_current_state(TASK_INTERRUPTIBLE);
823 add_wait_queue(&nlk->wait, &wait);
824
825 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
826 test_bit(0, &nlk->state)) &&
827 !sock_flag(sk, SOCK_DEAD))
828 *timeo = schedule_timeout(*timeo);
829
830 __set_current_state(TASK_RUNNING);
831 remove_wait_queue(&nlk->wait, &wait);
832 sock_put(sk);
833
834 if (signal_pending(current)) {
835 kfree_skb(skb);
836 return sock_intr_errno(*timeo);
837 }
838 return 1;
839 }
840 skb_set_owner_r(skb, sk);
841 return 0;
842}
843
844static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
845{
846 int len = skb->len;
847
848 skb_queue_tail(&sk->sk_receive_queue, skb);
849 sk->sk_data_ready(sk, len);
850 return len;
851}
852
853int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
854{
855 int len = __netlink_sendskb(sk, skb);
856
857 sock_put(sk);
858 return len;
859}
860
861void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
862{
863 kfree_skb(skb);
864 sock_put(sk);
865}
866
867static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
868{
869 int delta;
870
871 skb_orphan(skb);
872
873 delta = skb->end - skb->tail;
874 if (delta * 2 < skb->truesize)
875 return skb;
876
877 if (skb_shared(skb)) {
878 struct sk_buff *nskb = skb_clone(skb, allocation);
879 if (!nskb)
880 return skb;
881 consume_skb(skb);
882 skb = nskb;
883 }
884
885 if (!pskb_expand_head(skb, 0, -delta, allocation))
886 skb->truesize -= delta;
887
888 return skb;
889}
890
891static void netlink_rcv_wake(struct sock *sk)
892{
893 struct netlink_sock *nlk = nlk_sk(sk);
894
895 if (skb_queue_empty(&sk->sk_receive_queue))
896 clear_bit(0, &nlk->state);
897 if (!test_bit(0, &nlk->state))
898 wake_up_interruptible(&nlk->wait);
899}
900
901static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb)
902{
903 int ret;
904 struct netlink_sock *nlk = nlk_sk(sk);
905
906 ret = -ECONNREFUSED;
907 if (nlk->netlink_rcv != NULL) {
908 ret = skb->len;
909 skb_set_owner_r(skb, sk);
910 nlk->netlink_rcv(skb);
911 consume_skb(skb);
912 } else {
913 kfree_skb(skb);
914 }
915 sock_put(sk);
916 return ret;
917}
918
919int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
920 u32 pid, int nonblock)
921{
922 struct sock *sk;
923 int err;
924 long timeo;
925
926 skb = netlink_trim(skb, gfp_any());
927
928 timeo = sock_sndtimeo(ssk, nonblock);
929retry:
930 sk = netlink_getsockbypid(ssk, pid);
931 if (IS_ERR(sk)) {
932 kfree_skb(skb);
933 return PTR_ERR(sk);
934 }
935 if (netlink_is_kernel(sk))
936 return netlink_unicast_kernel(sk, skb);
937
938 if (sk_filter(sk, skb)) {
939 err = skb->len;
940 kfree_skb(skb);
941 sock_put(sk);
942 return err;
943 }
944
945 err = netlink_attachskb(sk, skb, &timeo, ssk);
946 if (err == 1)
947 goto retry;
948 if (err)
949 return err;
950
951 return netlink_sendskb(sk, skb);
952}
953EXPORT_SYMBOL(netlink_unicast);
954
955int netlink_has_listeners(struct sock *sk, unsigned int group)
956{
957 int res = 0;
958 struct listeners *listeners;
959
960 BUG_ON(!netlink_is_kernel(sk));
961
962 rcu_read_lock();
963 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
964
965 if (group - 1 < nl_table[sk->sk_protocol].groups)
966 res = test_bit(group - 1, listeners->masks);
967
968 rcu_read_unlock();
969
970 return res;
971}
972EXPORT_SYMBOL_GPL(netlink_has_listeners);
973
974static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
975{
976 struct netlink_sock *nlk = nlk_sk(sk);
977
978 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
979 !test_bit(0, &nlk->state)) {
980 skb_set_owner_r(skb, sk);
981 __netlink_sendskb(sk, skb);
982 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
983 }
984 return -1;
985}
986
987struct netlink_broadcast_data {
988 struct sock *exclude_sk;
989 struct net *net;
990 u32 pid;
991 u32 group;
992 int failure;
993 int delivery_failure;
994 int congested;
995 int delivered;
996 gfp_t allocation;
997 struct sk_buff *skb, *skb2;
998 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
999 void *tx_data;
1000};
1001
1002static int do_one_broadcast(struct sock *sk,
1003 struct netlink_broadcast_data *p)
1004{
1005 struct netlink_sock *nlk = nlk_sk(sk);
1006 int val;
1007
1008 if (p->exclude_sk == sk)
1009 goto out;
1010
1011 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1012 !test_bit(p->group - 1, nlk->groups))
1013 goto out;
1014
1015 if (!net_eq(sock_net(sk), p->net))
1016 goto out;
1017
1018 if (p->failure) {
1019 netlink_overrun(sk);
1020 goto out;
1021 }
1022
1023 sock_hold(sk);
1024 if (p->skb2 == NULL) {
1025 if (skb_shared(p->skb)) {
1026 p->skb2 = skb_clone(p->skb, p->allocation);
1027 } else {
1028 p->skb2 = skb_get(p->skb);
1029 /*
1030 * skb ownership may have been set when
1031 * delivered to a previous socket.
1032 */
1033 skb_orphan(p->skb2);
1034 }
1035 }
1036 if (p->skb2 == NULL) {
1037 netlink_overrun(sk);
1038 /* Clone failed. Notify ALL listeners. */
1039 p->failure = 1;
1040 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1041 p->delivery_failure = 1;
1042 } else if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1043 kfree_skb(p->skb2);
1044 p->skb2 = NULL;
1045 } else if (sk_filter(sk, p->skb2)) {
1046 kfree_skb(p->skb2);
1047 p->skb2 = NULL;
1048 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1049 netlink_overrun(sk);
1050 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1051 p->delivery_failure = 1;
1052 } else {
1053 p->congested |= val;
1054 p->delivered = 1;
1055 p->skb2 = NULL;
1056 }
1057 sock_put(sk);
1058
1059out:
1060 return 0;
1061}
1062
1063int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 pid,
1064 u32 group, gfp_t allocation,
1065 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1066 void *filter_data)
1067{
1068 struct net *net = sock_net(ssk);
1069 struct netlink_broadcast_data info;
1070 struct hlist_node *node;
1071 struct sock *sk;
1072
1073 skb = netlink_trim(skb, allocation);
1074
1075 info.exclude_sk = ssk;
1076 info.net = net;
1077 info.pid = pid;
1078 info.group = group;
1079 info.failure = 0;
1080 info.delivery_failure = 0;
1081 info.congested = 0;
1082 info.delivered = 0;
1083 info.allocation = allocation;
1084 info.skb = skb;
1085 info.skb2 = NULL;
1086 info.tx_filter = filter;
1087 info.tx_data = filter_data;
1088
1089 /* While we sleep in clone, do not allow to change socket list */
1090
1091 netlink_lock_table();
1092
1093 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1094 do_one_broadcast(sk, &info);
1095
1096 consume_skb(skb);
1097
1098 netlink_unlock_table();
1099
1100 if (info.delivery_failure) {
1101 kfree_skb(info.skb2);
1102 return -ENOBUFS;
1103 }
1104 consume_skb(info.skb2);
1105
1106 if (info.delivered) {
1107 if (info.congested && (allocation & __GFP_WAIT))
1108 yield();
1109 return 0;
1110 }
1111 return -ESRCH;
1112}
1113EXPORT_SYMBOL(netlink_broadcast_filtered);
1114
1115int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
1116 u32 group, gfp_t allocation)
1117{
1118 return netlink_broadcast_filtered(ssk, skb, pid, group, allocation,
1119 NULL, NULL);
1120}
1121EXPORT_SYMBOL(netlink_broadcast);
1122
1123struct netlink_set_err_data {
1124 struct sock *exclude_sk;
1125 u32 pid;
1126 u32 group;
1127 int code;
1128};
1129
1130static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1131{
1132 struct netlink_sock *nlk = nlk_sk(sk);
1133 int ret = 0;
1134
1135 if (sk == p->exclude_sk)
1136 goto out;
1137
1138 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1139 goto out;
1140
1141 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1142 !test_bit(p->group - 1, nlk->groups))
1143 goto out;
1144
1145 if (p->code == ENOBUFS && nlk->flags & NETLINK_RECV_NO_ENOBUFS) {
1146 ret = 1;
1147 goto out;
1148 }
1149
1150 sk->sk_err = p->code;
1151 sk->sk_error_report(sk);
1152out:
1153 return ret;
1154}
1155
1156/**
1157 * netlink_set_err - report error to broadcast listeners
1158 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1159 * @pid: the PID of a process that we want to skip (if any)
1160 * @groups: the broadcast group that will notice the error
1161 * @code: error code, must be negative (as usual in kernelspace)
1162 *
1163 * This function returns the number of broadcast listeners that have set the
1164 * NETLINK_RECV_NO_ENOBUFS socket option.
1165 */
1166int netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
1167{
1168 struct netlink_set_err_data info;
1169 struct hlist_node *node;
1170 struct sock *sk;
1171 int ret = 0;
1172
1173 info.exclude_sk = ssk;
1174 info.pid = pid;
1175 info.group = group;
1176 /* sk->sk_err wants a positive error value */
1177 info.code = -code;
1178
1179 read_lock(&nl_table_lock);
1180
1181 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1182 ret += do_one_set_err(sk, &info);
1183
1184 read_unlock(&nl_table_lock);
1185 return ret;
1186}
1187EXPORT_SYMBOL(netlink_set_err);
1188
1189/* must be called with netlink table grabbed */
1190static void netlink_update_socket_mc(struct netlink_sock *nlk,
1191 unsigned int group,
1192 int is_new)
1193{
1194 int old, new = !!is_new, subscriptions;
1195
1196 old = test_bit(group - 1, nlk->groups);
1197 subscriptions = nlk->subscriptions - old + new;
1198 if (new)
1199 __set_bit(group - 1, nlk->groups);
1200 else
1201 __clear_bit(group - 1, nlk->groups);
1202 netlink_update_subscriptions(&nlk->sk, subscriptions);
1203 netlink_update_listeners(&nlk->sk);
1204}
1205
1206static int netlink_setsockopt(struct socket *sock, int level, int optname,
1207 char __user *optval, unsigned int optlen)
1208{
1209 struct sock *sk = sock->sk;
1210 struct netlink_sock *nlk = nlk_sk(sk);
1211 unsigned int val = 0;
1212 int err;
1213
1214 if (level != SOL_NETLINK)
1215 return -ENOPROTOOPT;
1216
1217 if (optlen >= sizeof(int) &&
1218 get_user(val, (unsigned int __user *)optval))
1219 return -EFAULT;
1220
1221 switch (optname) {
1222 case NETLINK_PKTINFO:
1223 if (val)
1224 nlk->flags |= NETLINK_RECV_PKTINFO;
1225 else
1226 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1227 err = 0;
1228 break;
1229 case NETLINK_ADD_MEMBERSHIP:
1230 case NETLINK_DROP_MEMBERSHIP: {
1231 if (!netlink_capable(sock, NL_NONROOT_RECV))
1232 return -EPERM;
1233 err = netlink_realloc_groups(sk);
1234 if (err)
1235 return err;
1236 if (!val || val - 1 >= nlk->ngroups)
1237 return -EINVAL;
1238 netlink_table_grab();
1239 netlink_update_socket_mc(nlk, val,
1240 optname == NETLINK_ADD_MEMBERSHIP);
1241 netlink_table_ungrab();
1242 err = 0;
1243 break;
1244 }
1245 case NETLINK_BROADCAST_ERROR:
1246 if (val)
1247 nlk->flags |= NETLINK_BROADCAST_SEND_ERROR;
1248 else
1249 nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR;
1250 err = 0;
1251 break;
1252 case NETLINK_NO_ENOBUFS:
1253 if (val) {
1254 nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
1255 clear_bit(0, &nlk->state);
1256 wake_up_interruptible(&nlk->wait);
1257 } else {
1258 nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
1259 }
1260 err = 0;
1261 break;
1262 default:
1263 err = -ENOPROTOOPT;
1264 }
1265 return err;
1266}
1267
1268static int netlink_getsockopt(struct socket *sock, int level, int optname,
1269 char __user *optval, int __user *optlen)
1270{
1271 struct sock *sk = sock->sk;
1272 struct netlink_sock *nlk = nlk_sk(sk);
1273 int len, val, err;
1274
1275 if (level != SOL_NETLINK)
1276 return -ENOPROTOOPT;
1277
1278 if (get_user(len, optlen))
1279 return -EFAULT;
1280 if (len < 0)
1281 return -EINVAL;
1282
1283 switch (optname) {
1284 case NETLINK_PKTINFO:
1285 if (len < sizeof(int))
1286 return -EINVAL;
1287 len = sizeof(int);
1288 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1289 if (put_user(len, optlen) ||
1290 put_user(val, optval))
1291 return -EFAULT;
1292 err = 0;
1293 break;
1294 case NETLINK_BROADCAST_ERROR:
1295 if (len < sizeof(int))
1296 return -EINVAL;
1297 len = sizeof(int);
1298 val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0;
1299 if (put_user(len, optlen) ||
1300 put_user(val, optval))
1301 return -EFAULT;
1302 err = 0;
1303 break;
1304 case NETLINK_NO_ENOBUFS:
1305 if (len < sizeof(int))
1306 return -EINVAL;
1307 len = sizeof(int);
1308 val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0;
1309 if (put_user(len, optlen) ||
1310 put_user(val, optval))
1311 return -EFAULT;
1312 err = 0;
1313 break;
1314 default:
1315 err = -ENOPROTOOPT;
1316 }
1317 return err;
1318}
1319
1320static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1321{
1322 struct nl_pktinfo info;
1323
1324 info.group = NETLINK_CB(skb).dst_group;
1325 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1326}
1327
1328static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1329 struct msghdr *msg, size_t len)
1330{
1331 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1332 struct sock *sk = sock->sk;
1333 struct netlink_sock *nlk = nlk_sk(sk);
1334 struct sockaddr_nl *addr = msg->msg_name;
1335 u32 dst_pid;
1336 u32 dst_group;
1337 struct sk_buff *skb;
1338 int err;
1339 struct scm_cookie scm;
1340
1341 if (msg->msg_flags&MSG_OOB)
1342 return -EOPNOTSUPP;
1343
1344 if (NULL == siocb->scm)
1345 siocb->scm = &scm;
1346
1347 err = scm_send(sock, msg, siocb->scm, true);
1348 if (err < 0)
1349 return err;
1350
1351 if (msg->msg_namelen) {
1352 err = -EINVAL;
1353 if (addr->nl_family != AF_NETLINK)
1354 goto out;
1355 dst_pid = addr->nl_pid;
1356 dst_group = ffs(addr->nl_groups);
1357 err = -EPERM;
1358 if ((dst_group || dst_pid) &&
1359 !netlink_capable(sock, NL_NONROOT_SEND))
1360 goto out;
1361 } else {
1362 dst_pid = nlk->dst_pid;
1363 dst_group = nlk->dst_group;
1364 }
1365
1366 if (!nlk->pid) {
1367 err = netlink_autobind(sock);
1368 if (err)
1369 goto out;
1370 }
1371
1372 err = -EMSGSIZE;
1373 if (len > sk->sk_sndbuf - 32)
1374 goto out;
1375 err = -ENOBUFS;
1376 skb = alloc_skb(len, GFP_KERNEL);
1377 if (skb == NULL)
1378 goto out;
1379
1380 NETLINK_CB(skb).pid = nlk->pid;
1381 NETLINK_CB(skb).dst_group = dst_group;
1382 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1383
1384 err = -EFAULT;
1385 if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
1386 kfree_skb(skb);
1387 goto out;
1388 }
1389
1390 err = security_netlink_send(sk, skb);
1391 if (err) {
1392 kfree_skb(skb);
1393 goto out;
1394 }
1395
1396 if (dst_group) {
1397 atomic_inc(&skb->users);
1398 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1399 }
1400 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1401
1402out:
1403 scm_destroy(siocb->scm);
1404 return err;
1405}
1406
1407static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1408 struct msghdr *msg, size_t len,
1409 int flags)
1410{
1411 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1412 struct scm_cookie scm;
1413 struct sock *sk = sock->sk;
1414 struct netlink_sock *nlk = nlk_sk(sk);
1415 int noblock = flags&MSG_DONTWAIT;
1416 size_t copied;
1417 struct sk_buff *skb, *data_skb;
1418 int err, ret;
1419
1420 if (flags&MSG_OOB)
1421 return -EOPNOTSUPP;
1422
1423 copied = 0;
1424
1425 skb = skb_recv_datagram(sk, flags, noblock, &err);
1426 if (skb == NULL)
1427 goto out;
1428
1429 data_skb = skb;
1430
1431#ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1432 if (unlikely(skb_shinfo(skb)->frag_list)) {
1433 /*
1434 * If this skb has a frag_list, then here that means that we
1435 * will have to use the frag_list skb's data for compat tasks
1436 * and the regular skb's data for normal (non-compat) tasks.
1437 *
1438 * If we need to send the compat skb, assign it to the
1439 * 'data_skb' variable so that it will be used below for data
1440 * copying. We keep 'skb' for everything else, including
1441 * freeing both later.
1442 */
1443 if (flags & MSG_CMSG_COMPAT)
1444 data_skb = skb_shinfo(skb)->frag_list;
1445 }
1446#endif
1447
1448 msg->msg_namelen = 0;
1449
1450 copied = data_skb->len;
1451 if (len < copied) {
1452 msg->msg_flags |= MSG_TRUNC;
1453 copied = len;
1454 }
1455
1456 skb_reset_transport_header(data_skb);
1457 err = skb_copy_datagram_iovec(data_skb, 0, msg->msg_iov, copied);
1458
1459 if (msg->msg_name) {
1460 struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
1461 addr->nl_family = AF_NETLINK;
1462 addr->nl_pad = 0;
1463 addr->nl_pid = NETLINK_CB(skb).pid;
1464 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1465 msg->msg_namelen = sizeof(*addr);
1466 }
1467
1468 if (nlk->flags & NETLINK_RECV_PKTINFO)
1469 netlink_cmsg_recv_pktinfo(msg, skb);
1470
1471 if (NULL == siocb->scm) {
1472 memset(&scm, 0, sizeof(scm));
1473 siocb->scm = &scm;
1474 }
1475 siocb->scm->creds = *NETLINK_CREDS(skb);
1476 if (flags & MSG_TRUNC)
1477 copied = data_skb->len;
1478
1479 skb_free_datagram(sk, skb);
1480
1481 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1482 ret = netlink_dump(sk);
1483 if (ret) {
1484 sk->sk_err = ret;
1485 sk->sk_error_report(sk);
1486 }
1487 }
1488
1489 scm_recv(sock, msg, siocb->scm, flags);
1490out:
1491 netlink_rcv_wake(sk);
1492 return err ? : copied;
1493}
1494
1495static void netlink_data_ready(struct sock *sk, int len)
1496{
1497 BUG();
1498}
1499
1500/*
1501 * We export these functions to other modules. They provide a
1502 * complete set of kernel non-blocking support for message
1503 * queueing.
1504 */
1505
1506struct sock *
1507netlink_kernel_create(struct net *net, int unit, unsigned int groups,
1508 void (*input)(struct sk_buff *skb),
1509 struct mutex *cb_mutex, struct module *module)
1510{
1511 struct socket *sock;
1512 struct sock *sk;
1513 struct netlink_sock *nlk;
1514 struct listeners *listeners = NULL;
1515
1516 BUG_ON(!nl_table);
1517
1518 if (unit < 0 || unit >= MAX_LINKS)
1519 return NULL;
1520
1521 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1522 return NULL;
1523
1524 /*
1525 * We have to just have a reference on the net from sk, but don't
1526 * get_net it. Besides, we cannot get and then put the net here.
1527 * So we create one inside init_net and the move it to net.
1528 */
1529
1530 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
1531 goto out_sock_release_nosk;
1532
1533 sk = sock->sk;
1534 sk_change_net(sk, net);
1535
1536 if (groups < 32)
1537 groups = 32;
1538
1539 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
1540 if (!listeners)
1541 goto out_sock_release;
1542
1543 sk->sk_data_ready = netlink_data_ready;
1544 if (input)
1545 nlk_sk(sk)->netlink_rcv = input;
1546
1547 if (netlink_insert(sk, net, 0))
1548 goto out_sock_release;
1549
1550 nlk = nlk_sk(sk);
1551 nlk->flags |= NETLINK_KERNEL_SOCKET;
1552
1553 netlink_table_grab();
1554 if (!nl_table[unit].registered) {
1555 nl_table[unit].groups = groups;
1556 rcu_assign_pointer(nl_table[unit].listeners, listeners);
1557 nl_table[unit].cb_mutex = cb_mutex;
1558 nl_table[unit].module = module;
1559 nl_table[unit].registered = 1;
1560 } else {
1561 kfree(listeners);
1562 nl_table[unit].registered++;
1563 }
1564 netlink_table_ungrab();
1565 return sk;
1566
1567out_sock_release:
1568 kfree(listeners);
1569 netlink_kernel_release(sk);
1570 return NULL;
1571
1572out_sock_release_nosk:
1573 sock_release(sock);
1574 return NULL;
1575}
1576EXPORT_SYMBOL(netlink_kernel_create);
1577
1578
1579void
1580netlink_kernel_release(struct sock *sk)
1581{
1582 sk_release_kernel(sk);
1583}
1584EXPORT_SYMBOL(netlink_kernel_release);
1585
1586int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
1587{
1588 struct listeners *new, *old;
1589 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1590
1591 if (groups < 32)
1592 groups = 32;
1593
1594 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1595 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
1596 if (!new)
1597 return -ENOMEM;
1598 old = rcu_dereference_protected(tbl->listeners, 1);
1599 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
1600 rcu_assign_pointer(tbl->listeners, new);
1601
1602 kfree_rcu(old, rcu);
1603 }
1604 tbl->groups = groups;
1605
1606 return 0;
1607}
1608
1609/**
1610 * netlink_change_ngroups - change number of multicast groups
1611 *
1612 * This changes the number of multicast groups that are available
1613 * on a certain netlink family. Note that it is not possible to
1614 * change the number of groups to below 32. Also note that it does
1615 * not implicitly call netlink_clear_multicast_users() when the
1616 * number of groups is reduced.
1617 *
1618 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1619 * @groups: The new number of groups.
1620 */
1621int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1622{
1623 int err;
1624
1625 netlink_table_grab();
1626 err = __netlink_change_ngroups(sk, groups);
1627 netlink_table_ungrab();
1628
1629 return err;
1630}
1631
1632void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1633{
1634 struct sock *sk;
1635 struct hlist_node *node;
1636 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1637
1638 sk_for_each_bound(sk, node, &tbl->mc_list)
1639 netlink_update_socket_mc(nlk_sk(sk), group, 0);
1640}
1641
1642/**
1643 * netlink_clear_multicast_users - kick off multicast listeners
1644 *
1645 * This function removes all listeners from the given group.
1646 * @ksk: The kernel netlink socket, as returned by
1647 * netlink_kernel_create().
1648 * @group: The multicast group to clear.
1649 */
1650void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1651{
1652 netlink_table_grab();
1653 __netlink_clear_multicast_users(ksk, group);
1654 netlink_table_ungrab();
1655}
1656
1657void netlink_set_nonroot(int protocol, unsigned int flags)
1658{
1659 if ((unsigned int)protocol < MAX_LINKS)
1660 nl_table[protocol].nl_nonroot = flags;
1661}
1662EXPORT_SYMBOL(netlink_set_nonroot);
1663
1664struct nlmsghdr *
1665__nlmsg_put(struct sk_buff *skb, u32 pid, u32 seq, int type, int len, int flags)
1666{
1667 struct nlmsghdr *nlh;
1668 int size = NLMSG_LENGTH(len);
1669
1670 nlh = (struct nlmsghdr*)skb_put(skb, NLMSG_ALIGN(size));
1671 nlh->nlmsg_type = type;
1672 nlh->nlmsg_len = size;
1673 nlh->nlmsg_flags = flags;
1674 nlh->nlmsg_pid = pid;
1675 nlh->nlmsg_seq = seq;
1676 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
1677 memset(NLMSG_DATA(nlh) + len, 0, NLMSG_ALIGN(size) - size);
1678 return nlh;
1679}
1680EXPORT_SYMBOL(__nlmsg_put);
1681
1682/*
1683 * It looks a bit ugly.
1684 * It would be better to create kernel thread.
1685 */
1686
1687static int netlink_dump(struct sock *sk)
1688{
1689 struct netlink_sock *nlk = nlk_sk(sk);
1690 struct netlink_callback *cb;
1691 struct sk_buff *skb = NULL;
1692 struct nlmsghdr *nlh;
1693 int len, err = -ENOBUFS;
1694 int alloc_size;
1695
1696 mutex_lock(nlk->cb_mutex);
1697
1698 cb = nlk->cb;
1699 if (cb == NULL) {
1700 err = -EINVAL;
1701 goto errout_skb;
1702 }
1703
1704 alloc_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
1705
1706 skb = sock_rmalloc(sk, alloc_size, 0, GFP_KERNEL);
1707 if (!skb)
1708 goto errout_skb;
1709
1710 len = cb->dump(skb, cb);
1711
1712 if (len > 0) {
1713 mutex_unlock(nlk->cb_mutex);
1714
1715 if (sk_filter(sk, skb))
1716 kfree_skb(skb);
1717 else
1718 __netlink_sendskb(sk, skb);
1719 return 0;
1720 }
1721
1722 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1723 if (!nlh)
1724 goto errout_skb;
1725
1726 nl_dump_check_consistent(cb, nlh);
1727
1728 memcpy(nlmsg_data(nlh), &len, sizeof(len));
1729
1730 if (sk_filter(sk, skb))
1731 kfree_skb(skb);
1732 else
1733 __netlink_sendskb(sk, skb);
1734
1735 if (cb->done)
1736 cb->done(cb);
1737 nlk->cb = NULL;
1738 mutex_unlock(nlk->cb_mutex);
1739
1740 netlink_consume_callback(cb);
1741 return 0;
1742
1743errout_skb:
1744 mutex_unlock(nlk->cb_mutex);
1745 kfree_skb(skb);
1746 return err;
1747}
1748
1749int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1750 const struct nlmsghdr *nlh,
1751 struct netlink_dump_control *control)
1752{
1753 struct netlink_callback *cb;
1754 struct sock *sk;
1755 struct netlink_sock *nlk;
1756 int ret;
1757
1758 cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1759 if (cb == NULL)
1760 return -ENOBUFS;
1761
1762 cb->dump = control->dump;
1763 cb->done = control->done;
1764 cb->nlh = nlh;
1765 cb->data = control->data;
1766 cb->min_dump_alloc = control->min_dump_alloc;
1767 atomic_inc(&skb->users);
1768 cb->skb = skb;
1769
1770 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).pid);
1771 if (sk == NULL) {
1772 netlink_destroy_callback(cb);
1773 return -ECONNREFUSED;
1774 }
1775 nlk = nlk_sk(sk);
1776 /* A dump is in progress... */
1777 mutex_lock(nlk->cb_mutex);
1778 if (nlk->cb) {
1779 mutex_unlock(nlk->cb_mutex);
1780 netlink_destroy_callback(cb);
1781 sock_put(sk);
1782 return -EBUSY;
1783 }
1784 nlk->cb = cb;
1785 mutex_unlock(nlk->cb_mutex);
1786
1787 ret = netlink_dump(sk);
1788
1789 sock_put(sk);
1790
1791 if (ret)
1792 return ret;
1793
1794 /* We successfully started a dump, by returning -EINTR we
1795 * signal not to send ACK even if it was requested.
1796 */
1797 return -EINTR;
1798}
1799EXPORT_SYMBOL(netlink_dump_start);
1800
1801void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1802{
1803 struct sk_buff *skb;
1804 struct nlmsghdr *rep;
1805 struct nlmsgerr *errmsg;
1806 size_t payload = sizeof(*errmsg);
1807
1808 /* error messages get the original request appened */
1809 if (err)
1810 payload += nlmsg_len(nlh);
1811
1812 skb = nlmsg_new(payload, GFP_KERNEL);
1813 if (!skb) {
1814 struct sock *sk;
1815
1816 sk = netlink_lookup(sock_net(in_skb->sk),
1817 in_skb->sk->sk_protocol,
1818 NETLINK_CB(in_skb).pid);
1819 if (sk) {
1820 sk->sk_err = ENOBUFS;
1821 sk->sk_error_report(sk);
1822 sock_put(sk);
1823 }
1824 return;
1825 }
1826
1827 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1828 NLMSG_ERROR, payload, 0);
1829 errmsg = nlmsg_data(rep);
1830 errmsg->error = err;
1831 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1832 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1833}
1834EXPORT_SYMBOL(netlink_ack);
1835
1836int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1837 struct nlmsghdr *))
1838{
1839 struct nlmsghdr *nlh;
1840 int err;
1841
1842 while (skb->len >= nlmsg_total_size(0)) {
1843 int msglen;
1844
1845 nlh = nlmsg_hdr(skb);
1846 err = 0;
1847
1848 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1849 return 0;
1850
1851 /* Only requests are handled by the kernel */
1852 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1853 goto ack;
1854
1855 /* Skip control messages */
1856 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1857 goto ack;
1858
1859 err = cb(skb, nlh);
1860 if (err == -EINTR)
1861 goto skip;
1862
1863ack:
1864 if (nlh->nlmsg_flags & NLM_F_ACK || err)
1865 netlink_ack(skb, nlh, err);
1866
1867skip:
1868 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1869 if (msglen > skb->len)
1870 msglen = skb->len;
1871 skb_pull(skb, msglen);
1872 }
1873
1874 return 0;
1875}
1876EXPORT_SYMBOL(netlink_rcv_skb);
1877
1878/**
1879 * nlmsg_notify - send a notification netlink message
1880 * @sk: netlink socket to use
1881 * @skb: notification message
1882 * @pid: destination netlink pid for reports or 0
1883 * @group: destination multicast group or 0
1884 * @report: 1 to report back, 0 to disable
1885 * @flags: allocation flags
1886 */
1887int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1888 unsigned int group, int report, gfp_t flags)
1889{
1890 int err = 0;
1891
1892 if (group) {
1893 int exclude_pid = 0;
1894
1895 if (report) {
1896 atomic_inc(&skb->users);
1897 exclude_pid = pid;
1898 }
1899
1900 /* errors reported via destination sk->sk_err, but propagate
1901 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
1902 err = nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1903 }
1904
1905 if (report) {
1906 int err2;
1907
1908 err2 = nlmsg_unicast(sk, skb, pid);
1909 if (!err || err == -ESRCH)
1910 err = err2;
1911 }
1912
1913 return err;
1914}
1915EXPORT_SYMBOL(nlmsg_notify);
1916
1917#ifdef CONFIG_PROC_FS
1918struct nl_seq_iter {
1919 struct seq_net_private p;
1920 int link;
1921 int hash_idx;
1922};
1923
1924static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1925{
1926 struct nl_seq_iter *iter = seq->private;
1927 int i, j;
1928 struct sock *s;
1929 struct hlist_node *node;
1930 loff_t off = 0;
1931
1932 for (i = 0; i < MAX_LINKS; i++) {
1933 struct nl_pid_hash *hash = &nl_table[i].hash;
1934
1935 for (j = 0; j <= hash->mask; j++) {
1936 sk_for_each(s, node, &hash->table[j]) {
1937 if (sock_net(s) != seq_file_net(seq))
1938 continue;
1939 if (off == pos) {
1940 iter->link = i;
1941 iter->hash_idx = j;
1942 return s;
1943 }
1944 ++off;
1945 }
1946 }
1947 }
1948 return NULL;
1949}
1950
1951static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1952 __acquires(nl_table_lock)
1953{
1954 read_lock(&nl_table_lock);
1955 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1956}
1957
1958static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1959{
1960 struct sock *s;
1961 struct nl_seq_iter *iter;
1962 int i, j;
1963
1964 ++*pos;
1965
1966 if (v == SEQ_START_TOKEN)
1967 return netlink_seq_socket_idx(seq, 0);
1968
1969 iter = seq->private;
1970 s = v;
1971 do {
1972 s = sk_next(s);
1973 } while (s && sock_net(s) != seq_file_net(seq));
1974 if (s)
1975 return s;
1976
1977 i = iter->link;
1978 j = iter->hash_idx + 1;
1979
1980 do {
1981 struct nl_pid_hash *hash = &nl_table[i].hash;
1982
1983 for (; j <= hash->mask; j++) {
1984 s = sk_head(&hash->table[j]);
1985 while (s && sock_net(s) != seq_file_net(seq))
1986 s = sk_next(s);
1987 if (s) {
1988 iter->link = i;
1989 iter->hash_idx = j;
1990 return s;
1991 }
1992 }
1993
1994 j = 0;
1995 } while (++i < MAX_LINKS);
1996
1997 return NULL;
1998}
1999
2000static void netlink_seq_stop(struct seq_file *seq, void *v)
2001 __releases(nl_table_lock)
2002{
2003 read_unlock(&nl_table_lock);
2004}
2005
2006
2007static int netlink_seq_show(struct seq_file *seq, void *v)
2008{
2009 if (v == SEQ_START_TOKEN) {
2010 seq_puts(seq,
2011 "sk Eth Pid Groups "
2012 "Rmem Wmem Dump Locks Drops Inode\n");
2013 } else {
2014 struct sock *s = v;
2015 struct netlink_sock *nlk = nlk_sk(s);
2016
2017 seq_printf(seq, "%pK %-3d %-6d %08x %-8d %-8d %pK %-8d %-8d %-8lu\n",
2018 s,
2019 s->sk_protocol,
2020 nlk->pid,
2021 nlk->groups ? (u32)nlk->groups[0] : 0,
2022 sk_rmem_alloc_get(s),
2023 sk_wmem_alloc_get(s),
2024 nlk->cb,
2025 atomic_read(&s->sk_refcnt),
2026 atomic_read(&s->sk_drops),
2027 sock_i_ino(s)
2028 );
2029
2030 }
2031 return 0;
2032}
2033
2034static const struct seq_operations netlink_seq_ops = {
2035 .start = netlink_seq_start,
2036 .next = netlink_seq_next,
2037 .stop = netlink_seq_stop,
2038 .show = netlink_seq_show,
2039};
2040
2041
2042static int netlink_seq_open(struct inode *inode, struct file *file)
2043{
2044 return seq_open_net(inode, file, &netlink_seq_ops,
2045 sizeof(struct nl_seq_iter));
2046}
2047
2048static const struct file_operations netlink_seq_fops = {
2049 .owner = THIS_MODULE,
2050 .open = netlink_seq_open,
2051 .read = seq_read,
2052 .llseek = seq_lseek,
2053 .release = seq_release_net,
2054};
2055
2056#endif
2057
2058int netlink_register_notifier(struct notifier_block *nb)
2059{
2060 return atomic_notifier_chain_register(&netlink_chain, nb);
2061}
2062EXPORT_SYMBOL(netlink_register_notifier);
2063
2064int netlink_unregister_notifier(struct notifier_block *nb)
2065{
2066 return atomic_notifier_chain_unregister(&netlink_chain, nb);
2067}
2068EXPORT_SYMBOL(netlink_unregister_notifier);
2069
2070static const struct proto_ops netlink_ops = {
2071 .family = PF_NETLINK,
2072 .owner = THIS_MODULE,
2073 .release = netlink_release,
2074 .bind = netlink_bind,
2075 .connect = netlink_connect,
2076 .socketpair = sock_no_socketpair,
2077 .accept = sock_no_accept,
2078 .getname = netlink_getname,
2079 .poll = datagram_poll,
2080 .ioctl = sock_no_ioctl,
2081 .listen = sock_no_listen,
2082 .shutdown = sock_no_shutdown,
2083 .setsockopt = netlink_setsockopt,
2084 .getsockopt = netlink_getsockopt,
2085 .sendmsg = netlink_sendmsg,
2086 .recvmsg = netlink_recvmsg,
2087 .mmap = sock_no_mmap,
2088 .sendpage = sock_no_sendpage,
2089};
2090
2091static const struct net_proto_family netlink_family_ops = {
2092 .family = PF_NETLINK,
2093 .create = netlink_create,
2094 .owner = THIS_MODULE, /* for consistency 8) */
2095};
2096
2097static int __net_init netlink_net_init(struct net *net)
2098{
2099#ifdef CONFIG_PROC_FS
2100 if (!proc_net_fops_create(net, "netlink", 0, &netlink_seq_fops))
2101 return -ENOMEM;
2102#endif
2103 return 0;
2104}
2105
2106static void __net_exit netlink_net_exit(struct net *net)
2107{
2108#ifdef CONFIG_PROC_FS
2109 proc_net_remove(net, "netlink");
2110#endif
2111}
2112
2113static void __init netlink_add_usersock_entry(void)
2114{
2115 struct listeners *listeners;
2116 int groups = 32;
2117
2118 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2119 if (!listeners)
2120 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2121
2122 netlink_table_grab();
2123
2124 nl_table[NETLINK_USERSOCK].groups = groups;
2125 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2126 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2127 nl_table[NETLINK_USERSOCK].registered = 1;
2128 nl_table[NETLINK_USERSOCK].nl_nonroot = NL_NONROOT_SEND;
2129
2130 netlink_table_ungrab();
2131}
2132
2133static struct pernet_operations __net_initdata netlink_net_ops = {
2134 .init = netlink_net_init,
2135 .exit = netlink_net_exit,
2136};
2137
2138static int __init netlink_proto_init(void)
2139{
2140 struct sk_buff *dummy_skb;
2141 int i;
2142 unsigned long limit;
2143 unsigned int order;
2144 int err = proto_register(&netlink_proto, 0);
2145
2146 if (err != 0)
2147 goto out;
2148
2149 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
2150
2151 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2152 if (!nl_table)
2153 goto panic;
2154
2155 if (totalram_pages >= (128 * 1024))
2156 limit = totalram_pages >> (21 - PAGE_SHIFT);
2157 else
2158 limit = totalram_pages >> (23 - PAGE_SHIFT);
2159
2160 order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
2161 limit = (1UL << order) / sizeof(struct hlist_head);
2162 order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
2163
2164 for (i = 0; i < MAX_LINKS; i++) {
2165 struct nl_pid_hash *hash = &nl_table[i].hash;
2166
2167 hash->table = nl_pid_hash_zalloc(1 * sizeof(*hash->table));
2168 if (!hash->table) {
2169 while (i-- > 0)
2170 nl_pid_hash_free(nl_table[i].hash.table,
2171 1 * sizeof(*hash->table));
2172 kfree(nl_table);
2173 goto panic;
2174 }
2175 hash->max_shift = order;
2176 hash->shift = 0;
2177 hash->mask = 0;
2178 hash->rehash_time = jiffies;
2179 }
2180
2181 netlink_add_usersock_entry();
2182
2183 sock_register(&netlink_family_ops);
2184 register_pernet_subsys(&netlink_net_ops);
2185 /* The netlink device handler may be needed early. */
2186 rtnetlink_init();
2187out:
2188 return err;
2189panic:
2190 panic("netlink_init: Cannot allocate nl_table\n");
2191}
2192
2193core_initcall(netlink_proto_init);